Fix clippy and doc test errors

develop changed every sequence-based `keylet::` factory to take
`SeqProxy const&` instead of `std::uint32_t`, and removed the
two-argument `mptokenIssuance(seq, issuer)` overload. The wasm host
functions and their tests still passed raw sequences, so the branch
merged cleanly but did not compile.

Wrap the raw sequences at the call sites, matching the idiom develop
adopted in its own tests:

- `SeqProxy::rawSequence` for check, escrow, nftokenOffer, offer,
  payChannel, permissionedDomain and vault
- `SeqProxy::rawTicket` for ticket
- `keylet::mptokenIssuance(makeMptID(seq, issuer))` for the removed
  overload

No computed keylet changes: the factories only read `seq.value()`, and
the removed overload was itself defined as `mptokenIssuance(makeMptID(
seq, issuer))`.

.clang-tidy

@@ -9,130 +9,64 @@ Checks: "-*,
 
   cppcoreguidelines-*,
   -cppcoreguidelines-avoid-c-arrays,
-  -cppcoreguidelines-avoid-capturing-lambda-coroutines,
   -cppcoreguidelines-avoid-const-or-ref-data-members,
   -cppcoreguidelines-avoid-do-while,
-  -cppcoreguidelines-avoid-goto,
   -cppcoreguidelines-avoid-magic-numbers,
   -cppcoreguidelines-avoid-non-const-global-variables,
-  -cppcoreguidelines-avoid-reference-coroutine-parameters,
   -cppcoreguidelines-c-copy-assignment-signature,
-  -cppcoreguidelines-explicit-virtual-functions,
   -cppcoreguidelines-interfaces-global-init,
-  -cppcoreguidelines-macro-to-enum,
   -cppcoreguidelines-macro-usage,
   -cppcoreguidelines-missing-std-forward,
   -cppcoreguidelines-narrowing-conversions,
-  -cppcoreguidelines-no-malloc,
-  -cppcoreguidelines-noexcept-destructor,
   -cppcoreguidelines-noexcept-move-operations,
-  -cppcoreguidelines-noexcept-swap,
   -cppcoreguidelines-non-private-member-variables-in-classes,
   -cppcoreguidelines-owning-memory,
-  -cppcoreguidelines-prefer-member-initializer,
   -cppcoreguidelines-pro-bounds-array-to-pointer-decay,
   -cppcoreguidelines-pro-bounds-avoid-unchecked-container-access,
   -cppcoreguidelines-pro-bounds-constant-array-index,
   -cppcoreguidelines-pro-bounds-pointer-arithmetic,
-  -cppcoreguidelines-pro-type-const-cast,
-  -cppcoreguidelines-pro-type-cstyle-cast,
   -cppcoreguidelines-pro-type-reinterpret-cast,
   -cppcoreguidelines-pro-type-union-access,
-  -cppcoreguidelines-pro-type-vararg,
-  -cppcoreguidelines-slicing,
   -cppcoreguidelines-special-member-functions,
 
   llvm-namespace-comment,
 
   misc-*,
-  -misc-anonymous-namespace-in-header,
-  -misc-confusable-identifiers,
-  -misc-coroutine-hostile-raii,
-  -misc-misleading-bidirectional,
-  -misc-misleading-identifier,
   -misc-multiple-inheritance,
-  -misc-new-delete-overloads,
   -misc-no-recursion,
-  -misc-non-copyable-objects,
   -misc-non-private-member-variables-in-classes,
   -misc-override-with-different-visibility,
-  -misc-predictable-rand,
-  -misc-unconventional-assign-operator,
-  -misc-uniqueptr-reset-release,
   -misc-unused-parameters,
   -misc-use-anonymous-namespace,
   -misc-use-internal-linkage,
 
   modernize-*,
-  -modernize-avoid-bind,
   -modernize-avoid-c-arrays,
   -modernize-avoid-c-style-cast,
-  -modernize-avoid-setjmp-longjmp,
-  -modernize-avoid-variadic-functions,
-  -modernize-deprecated-ios-base-aliases,
-  -modernize-loop-convert,
-  -modernize-macro-to-enum,
-  -modernize-min-max-use-initializer-list,
-  -modernize-raw-string-literal,
-  -modernize-redundant-void-arg,
-  -modernize-replace-auto-ptr,
-  -modernize-replace-disallow-copy-and-assign-macro,
-  -modernize-replace-random-shuffle,
   -modernize-return-braced-init-list,
-  -modernize-shrink-to-fit,
-  -modernize-use-bool-literals,
-  -modernize-use-constraints,
-  -modernize-use-default-member-init,
   -modernize-use-integer-sign-comparison,
-  -modernize-use-noexcept,
-  -modernize-use-nullptr,
-  -modernize-use-std-format,
-  -modernize-use-std-print,
   -modernize-use-trailing-return-type,
-  -modernize-use-transparent-functors,
-  -modernize-use-uncaught-exceptions,
 
   performance-*,
   -performance-avoid-endl,
   -performance-enum-size,
-  -performance-inefficient-algorithm,
-  -performance-inefficient-string-concatenation,
-  -performance-no-int-to-ptr,
-  -performance-noexcept-destructor,
   -performance-noexcept-move-constructor,
-  -performance-noexcept-swap,
-  -performance-type-promotion-in-math-fn,
   -performance-unnecessary-copy-initialization,
   -performance-unnecessary-value-param,
 
   readability-*,
   -readability-avoid-const-params-in-decls,
-  -readability-avoid-unconditional-preprocessor-if,
   -readability-container-data-pointer,
-  -readability-delete-null-pointer,
   -readability-function-cognitive-complexity,
-  -readability-function-size,
   -readability-identifier-length,
   -readability-inconsistent-declaration-parameter-name,
   -readability-isolate-declaration,
   -readability-magic-numbers,
-  -readability-misplaced-array-index,
   -readability-named-parameter,
-  -readability-operators-representation,
   -readability-qualified-auto,
   -readability-redundant-access-specifiers,
-  -readability-redundant-control-flow,
-  -readability-redundant-function-ptr-dereference,
-  -readability-redundant-preprocessor,
-  -readability-redundant-smartptr-get,
-  -readability-redundant-string-cstr,
-  -readability-simplify-subscript-expr,
   -readability-static-accessed-through-instance,
-  -readability-string-compare,
-  -readability-uniqueptr-delete-release,
-  -readability-uppercase-literal-suffix,
-  -readability-use-anyofallof,
-  -readability-use-concise-preprocessor-directives
+  -readability-uppercase-literal-suffix
   "
 # ---
 # bugprone-narrowing-conversions,                      # This will break a lot of code but we should enable it in the future because it can eliminate a lot of bugs
@@ -141,6 +75,8 @@ Checks: "-*,
 # readability-static-accessed-through-instance,        # this check is probably unnecessary. It makes the code less readable
 # ---
 
+FormatStyle: file
+
 CheckOptions:
   bugprone-unsafe-functions.ReportMoreUnsafeFunctions: true
   bugprone-unused-return-value.CheckedReturnTypes: ::std::error_code;::std::error_condition;::std::errc
@@ -149,6 +85,8 @@ CheckOptions:
 
   readability-braces-around-statements.ShortStatementLines: 2
   readability-identifier-naming.MacroDefinitionCase: UPPER_CASE
+  readability-identifier-naming.NamespaceCase: lower_case
+  readability-identifier-naming.InlineNamespaceCase: lower_case
   readability-identifier-naming.ClassCase: CamelCase
   readability-identifier-naming.StructCase: CamelCase
   readability-identifier-naming.UnionCase: CamelCase

.codecov.yml

@@ -1,10 +1,32 @@
 codecov:
   require_ci_to_pass: true
+  # The C++ and Rust uploads land minutes apart; without this gate Codecov
+  # publishes a near-zero total from whichever one arrives first.
+  notify:
+    after_n_builds: 2
+    wait_for_ci: true
 
 comment:
   behavior: default
   layout: reach,diff,flags,tree,reach
-  show_carryforward_flags: false
+  show_carryforward_flags: true
+  after_n_builds: 2
+
+# C++ and Rust coverage upload from independent workflows under the `cpp` and
+# `rust` flags; carryforward keeps one language's total when only the other reran.
+flag_management:
+  default_rules:
+    carryforward: true
+  individual_flags:
+    - name: cpp
+      carryforward: true
+      paths:
+        - include/
+        - src/
+    - name: rust
+      carryforward: true
+      paths:
+        - crates/
 
 coverage:
   range: "70..85"

.cspell.config.yaml

@@ -30,7 +30,9 @@ ignoreRegExpList:
   - ABCDEFGHIJKLMNOPQRSTUVWXYZ
   - ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz
 overrides:
-  - filename: "**/*_test.cpp" # all test files
+  - filename:
+      - "**/*_test.cpp" # legacy boost.test files
+      - "src/tests/**/*.cpp" # gtest test files
     ignoreRegExpList:
       - /"[^"]*"/g # double-quoted strings
       - /'[^']*'/g # single-quoted strings
@@ -65,6 +67,8 @@ words:
   - Btrfs
   - Buildx
   - canonicality
+  - cdylib
+  - canonicalised
   - changespq
   - checkme
   - choco
@@ -72,6 +76,7 @@ words:
   - citardauq
   - clawback
   - clawbacks
+  - clippy
   - cmaketoolchain
   - coeffs
   - coldwallet
@@ -99,6 +104,7 @@ words:
   - deleteme
   - demultiplexer
   - deserializaton
+  - desugars
   - desync
   - desynced
   - determ
@@ -124,15 +130,19 @@ words:
   - gcov
   - gcovr
   - ghead
+  - gmock
   - Gnutella
   - godexsoft
   - gpgcheck
   - gpgkey
+  - Hinnant
   - hotwallet
   - hwaddress
   - hwrap
   - ifndef
+  - impls
   - inequation
+  - initialiser
   - insuf
   - insuff
   - invasively
@@ -161,6 +171,7 @@ words:
   - llection
   - LOCALGOOD
   - logwstream
+  - Lombrozo
   - lseq
   - lsmf
   - ltype
@@ -170,7 +181,6 @@ words:
   - MPTAMM
   - MPTDEX
   - Merkle
-  - Metafuncton
   - misprediction
   - missingok
   - mptbalance
@@ -198,6 +208,7 @@ words:
   - nftokens
   - nftpage
   - nikb
+  - Nikolaos
   - nixfmt
   - nixos
   - nixpkgs
@@ -214,6 +225,7 @@ words:
   - Nyffenegger
   - onlatest
   - ostr
+  - oxalica
   - pargs
   - partitioner
   - paychan
@@ -238,6 +250,7 @@ words:
   - pyparsing
   - qalloc
   - qbsprofile
+  - qself
   - queuable
   - Raphson
   - rcflags
@@ -259,6 +272,9 @@ words:
   - rocksdb
   - Rohrs
   - roundings
+  - rustc
+  - rustfmt
+  - rustup
   - sahyadri
   - Satoshi
   - scons
@@ -279,12 +295,15 @@ words:
   - sles
   - soci
   - socidb
+  - sponsee
+  - sponsees
   - SRPMS
   - sslws
   - statsd
   - STATSDCOLLECTOR
   - stissue
   - stnum
+  - stnumber
   - stobj
   - stobject
   - stpath
@@ -296,10 +315,13 @@ words:
   - summands
   - superpeer
   - superpeers
+  - Swatinem
   - takergets
   - takerpays
   - ters
   - TMEndpointv2
+  - toolchain
+  - tparam
   - trixie
   - tx
   - txid
@@ -322,14 +344,17 @@ words:
   - unflatten
   - unfund
   - unimpair
+  - unmetered
   - unroutable
   - unscalable
   - unserviced
   - unshareable
   - unshares
+  - unsponsored
   - unsquelch
   - unsquelched
   - unsquelching
+  - unsuffixed
   - unvalidated
   - unveto
   - unvetoed
@@ -340,6 +365,7 @@ words:
   - vfalco
   - vinnie
   - wasmi
+  - Werror
   - wextra
   - wptr
   - writeme
@@ -347,6 +373,7 @@ words:
   - wthread
   - xbridge
   - xchain
+  - xfloat
   - ximinez
   - XMACRO
   - xrpkuwait

.envrc

@@ -0,0 +1,3 @@
+watch_file nix/*.nix
+
+use flake

.gersemi/definitions.cmake

@@ -11,6 +11,9 @@ endfunction()
 function(create_symbolic_link target link)
 endfunction()
 
+function(xrpl_add_benchmark name)
+endfunction()
+
 macro(exclude_from_default target_)
 endmacro()
 

.github/dependabot.yml

@@ -15,3 +15,24 @@ updates:
     commit-message:
       prefix: "ci: [DEPENDABOT] "
     target-branch: develop
+    groups:
+      github-actions:
+        patterns:
+          - "*"
+
+  - package-ecosystem: cargo
+    directory: /crates
+    schedule:
+      interval: weekly
+      day: monday
+      time: "04:00"
+      timezone: Etc/GMT
+    commit-message:
+      prefix: "ci: [DEPENDABOT] "
+    target-branch: develop
+    open-pull-requests-limit: 10
+    # Bundle all Rust dependency bumps into a single PR per run to reduce noise.
+    groups:
+      rust-dependencies:
+        patterns:
+          - "*"

.github/pull_request_template.md

@@ -1,10 +1,10 @@
 <!--
-This PR template helps you to write a good pull request description.
+This PR template helps you write a good pull request description.
 Please feel free to include additional useful information even beyond what is requested below.
 
 If your branch is on a personal fork and has a name that allows it to
 run CI build/test jobs (e.g. "ci/foo"), remember to rename it BEFORE
-opening the PR.  This avoids unnecessary redundant test runs. Renaming
+opening the PR. This avoids redundant test runs. Renaming
 the branch after opening the PR will close the PR.
 https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/managing-branches-in-your-repository/renaming-a-branch
 -->
@@ -15,7 +15,7 @@ https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-
 Please include a summary of the changes.
 This may be a direct input to the release notes.
 If too broad, please consider splitting into multiple PRs.
-If a relevant task or issue, please link it here.
+If there is a relevant task or issue, please link it here.
 -->
 
 ### Context of Change
@@ -65,5 +65,5 @@ This section may not be needed if your change includes thoroughly commented unit
 
 <!--
 ## Future Tasks
-For future tasks related to PR.
+For future tasks related to this PR.
 -->

.github/scripts/levelization/README.md

@@ -40,18 +40,18 @@ listed later.
 | 04           | xrpl/protocol                       |
 | 05           | xrpl/core xrpl/resource xrpl/server |
 | 06           | xrpl/ledger xrpl/nodestore xrpl/net |
-| 07           | xrpl/shamap                         |
+| 07           | xrpl/shamap xrpl/consensus          |
 
 ## xrpld Modules (Application Implementation)
 
-| Level / Tier | Module(s)                        |
-| ------------ | -------------------------------- |
-| 05           | xrpld/conditions xrpld/consensus |
-| 06           | xrpld/core xrpld/peerfinder      |
-| 07           | xrpld/shamap xrpld/overlay       |
-| 08           | xrpld/app                        |
-| 09           | xrpld/rpc                        |
-| 10           | xrpld/perflog                    |
+| Level / Tier | Module(s)                   |
+| ------------ | --------------------------- |
+| 05           | xrpld/conditions            |
+| 06           | xrpld/core xrpld/peerfinder |
+| 07           | xrpld/shamap xrpld/overlay  |
+| 08           | xrpld/app                   |
+| 09           | xrpld/rpc                   |
+| 10           | xrpld/perflog               |
 
 ## Test Modules
 

.github/scripts/levelization/results/loops.txt

@@ -1,9 +1,6 @@
 Loop: xrpld.app xrpld.overlay
   xrpld.app > xrpld.overlay
 
-Loop: xrpld.app xrpld.peerfinder
-  xrpld.peerfinder ~= xrpld.app
-
 Loop: xrpld.app xrpld.rpc
   xrpld.rpc > xrpld.app
 

.github/scripts/levelization/results/ordering.txt

@@ -1,8 +1,13 @@
+benchmarks.libxrpl > xrpl.basics
+benchmarks.libxrpl > xrpl.config
+benchmarks.libxrpl > xrpl.nodestore
 libxrpl.basics > xrpl.basics
 libxrpl.conditions > xrpl.basics
 libxrpl.conditions > xrpl.conditions
 libxrpl.config > xrpl.basics
 libxrpl.config > xrpl.config
+libxrpl.consensus > xrpl.basics
+libxrpl.consensus > xrpl.consensus
 libxrpl.core > xrpl.basics
 libxrpl.core > xrpl.core
 libxrpl.core > xrpl.json
@@ -22,6 +27,9 @@ libxrpl.nodestore > xrpl.config
 libxrpl.nodestore > xrpl.json
 libxrpl.nodestore > xrpl.nodestore
 libxrpl.nodestore > xrpl.protocol
+libxrpl.peerfinder > xrpl.basics
+libxrpl.peerfinder > xrpl.peerfinder
+libxrpl.peerfinder > xrpl.protocol
 libxrpl.protocol > xrpl.basics
 libxrpl.protocol > xrpl.json
 libxrpl.protocol > xrpl.protocol
@@ -57,9 +65,9 @@ test.app > test.jtx
 test.app > test.unit_test
 test.app > xrpl.basics
 test.app > xrpl.config
+test.app > xrpl.consensus
 test.app > xrpl.core
 test.app > xrpld.app
-test.app > xrpld.consensus
 test.app > xrpld.core
 test.app > xrpld.overlay
 test.app > xrpld.rpc
@@ -72,7 +80,6 @@ test.app > xrpl.server
 test.app > xrpl.shamap
 test.app > xrpl.tx
 test.basics > test.jtx
-test.basics > test.unit_test
 test.basics > xrpl.basics
 test.basics > xrpl.core
 test.basics > xrpld.rpc
@@ -81,12 +88,9 @@ test.basics > xrpl.protocol
 test.beast > xrpl.basics
 test.conditions > xrpl.basics
 test.conditions > xrpl.conditions
-test.consensus > test.csf
 test.consensus > test.jtx
-test.consensus > test.unit_test
 test.consensus > xrpl.basics
 test.consensus > xrpld.app
-test.consensus > xrpld.consensus
 test.consensus > xrpl.ledger
 test.consensus > xrpl.protocol
 test.consensus > xrpl.shamap
@@ -101,10 +105,6 @@ test.core > xrpl.json
 test.core > xrpl.protocol
 test.core > xrpl.rdb
 test.core > xrpl.server
-test.csf > xrpl.basics
-test.csf > xrpld.consensus
-test.csf > xrpl.json
-test.csf > xrpl.ledger
 test.json > test.jtx
 test.json > xrpl.json
 test.jtx > test.unit_test
@@ -130,12 +130,9 @@ test.ledger > xrpl.json
 test.ledger > xrpl.ledger
 test.ledger > xrpl.protocol
 test.nodestore > test.jtx
-test.nodestore > test.unit_test
 test.nodestore > xrpl.basics
-test.nodestore > xrpl.config
 test.nodestore > xrpld.core
 test.nodestore > xrpl.nodestore
-test.nodestore > xrpl.protocol
 test.nodestore > xrpl.rdb
 test.overlay > test.jtx
 test.overlay > test.unit_test
@@ -144,27 +141,19 @@ test.overlay > xrpl.config
 test.overlay > xrpld.app
 test.overlay > xrpld.core
 test.overlay > xrpld.overlay
-test.overlay > xrpld.peerfinder
 test.overlay > xrpl.json
 test.overlay > xrpl.nodestore
+test.overlay > xrpl.peerfinder
 test.overlay > xrpl.protocol
 test.overlay > xrpl.resource
 test.overlay > xrpl.server
 test.overlay > xrpl.shamap
-test.peerfinder > test.beast
-test.peerfinder > test.unit_test
-test.peerfinder > xrpl.basics
-test.peerfinder > xrpld.core
-test.peerfinder > xrpld.peerfinder
-test.peerfinder > xrpl.protocol
 test.protocol > test.jtx
 test.protocol > test.unit_test
 test.protocol > xrpl.basics
+test.protocol > xrpld.core
 test.protocol > xrpl.json
 test.protocol > xrpl.protocol
-test.resource > test.unit_test
-test.resource > xrpl.basics
-test.resource > xrpl.resource
 test.rpc > test.jtx
 test.rpc > xrpl.basics
 test.rpc > xrpl.config
@@ -188,29 +177,30 @@ test.server > xrpld.core
 test.server > xrpl.json
 test.server > xrpl.protocol
 test.server > xrpl.server
-test.shamap > test.unit_test
-test.shamap > xrpl.basics
-test.shamap > xrpl.config
-test.shamap > xrpl.nodestore
-test.shamap > xrpl.protocol
-test.shamap > xrpl.shamap
 test.unit_test > xrpl.basics
 test.unit_test > xrpl.protocol
 tests.libxrpl > xrpl.basics
 tests.libxrpl > xrpl.config
+tests.libxrpl > xrpl.consensus
 tests.libxrpl > xrpl.core
 tests.libxrpl > xrpl.json
 tests.libxrpl > xrpl.ledger
 tests.libxrpl > xrpl.net
 tests.libxrpl > xrpl.nodestore
+tests.libxrpl > xrpl.peerfinder
 tests.libxrpl > xrpl.protocol
 tests.libxrpl > xrpl.protocol_autogen
+tests.libxrpl > xrpl.resource
 tests.libxrpl > xrpl.server
 tests.libxrpl > xrpl.shamap
 tests.libxrpl > xrpl.tx
 xrpl.conditions > xrpl.basics
 xrpl.conditions > xrpl.protocol
 xrpl.config > xrpl.basics
+xrpl.consensus > xrpl.basics
+xrpl.consensus > xrpl.json
+xrpl.consensus > xrpl.ledger
+xrpl.consensus > xrpl.protocol
 xrpl.core > xrpl.basics
 xrpl.core > xrpl.json
 xrpl.core > xrpl.protocol
@@ -225,6 +215,8 @@ xrpl.nodestore > xrpl.basics
 xrpl.nodestore > xrpl.config
 xrpl.nodestore > xrpl.json
 xrpl.nodestore > xrpl.protocol
+xrpl.peerfinder > xrpl.basics
+xrpl.peerfinder > xrpl.protocol
 xrpl.protocol > xrpl.basics
 xrpl.protocol > xrpl.json
 xrpl.protocol_autogen > xrpl.json
@@ -251,23 +243,20 @@ xrpl.tx > xrpl.protocol
 xrpld.app > test.unit_test
 xrpld.app > xrpl.basics
 xrpld.app > xrpl.config
+xrpld.app > xrpl.consensus
 xrpld.app > xrpl.core
-xrpld.app > xrpld.consensus
 xrpld.app > xrpld.core
 xrpld.app > xrpl.json
 xrpld.app > xrpl.ledger
 xrpld.app > xrpl.net
 xrpld.app > xrpl.nodestore
+xrpld.app > xrpl.peerfinder
 xrpld.app > xrpl.protocol
 xrpld.app > xrpl.rdb
 xrpld.app > xrpl.resource
 xrpld.app > xrpl.server
 xrpld.app > xrpl.shamap
 xrpld.app > xrpl.tx
-xrpld.consensus > xrpl.basics
-xrpld.consensus > xrpl.json
-xrpld.consensus > xrpl.ledger
-xrpld.consensus > xrpl.protocol
 xrpld.core > xrpl.basics
 xrpld.core > xrpl.config
 xrpld.core > xrpl.core
@@ -276,21 +265,22 @@ xrpld.core > xrpl.protocol
 xrpld.core > xrpl.rdb
 xrpld.overlay > xrpl.basics
 xrpld.overlay > xrpl.config
+xrpld.overlay > xrpl.consensus
 xrpld.overlay > xrpl.core
-xrpld.overlay > xrpld.consensus
 xrpld.overlay > xrpld.core
 xrpld.overlay > xrpld.peerfinder
 xrpld.overlay > xrpl.json
 xrpld.overlay > xrpl.ledger
+xrpld.overlay > xrpl.peerfinder
 xrpld.overlay > xrpl.protocol
 xrpld.overlay > xrpl.resource
 xrpld.overlay > xrpl.server
 xrpld.overlay > xrpl.shamap
 xrpld.overlay > xrpl.tx
 xrpld.peerfinder > xrpl.basics
-xrpld.peerfinder > xrpl.config
+xrpld.peerfinder > xrpld.app
 xrpld.peerfinder > xrpld.core
-xrpld.peerfinder > xrpl.protocol
+xrpld.peerfinder > xrpl.peerfinder
 xrpld.peerfinder > xrpl.rdb
 xrpld.perflog > xrpl.basics
 xrpld.perflog > xrpl.config

.github/scripts/rename/namespace.sh

@@ -46,9 +46,6 @@ for DIRECTORY in "${DIRECTORIES[@]}"; do
     done
 done
 
-# Special case for NuDBFactory that has ripple twice in the test suite name.
-${SED_COMMAND} -i -E 's/(BEAST_DEFINE_TESTSUITE.+)ripple(.+)/\1xrpl\2/g' src/test/nodestore/NuDBFactory_test.cpp
-
 DIRECTORY=$1
 find "${DIRECTORY}" -type f -name "*.md" | while read -r FILE; do
     echo "Processing file: ${FILE}"

.github/scripts/strategy-matrix/generate.py

@@ -25,24 +25,20 @@ def get_cmake_args(build_type: str, extra_args: str) -> str:
     return " ".join(args)
 
 
-def runs_on_event(exclude_event_types: list[str], event: str | None) -> bool:
-    """Whether a config should run for the current event.
-
-    'exclude_event_types' is a list of GitHub event names (e.g.
-    ["pull_request"]) on which the config should NOT run; an empty list means
-    the config runs on every event. When no event is given (event is None), no
-    filtering is applied.
-    """
-    if event is None:
-        return True
-    return event not in exclude_event_types
-
-
 # ---------------------------------------------------------------------------
 # Input types — shapes of the JSON config files
 # ---------------------------------------------------------------------------
 
 
+# Every config must declare 'minimal'. Minimal configs form the reduced matrix
+# built for pull requests by default; the full matrix adds the rest. Packaging
+# configs declare it too, but packaging is gated in the workflow, not by it.
+#
+# Configs may also opt into 'benchmark' to smoke-run the benchmarks. Note that
+# the flag applies to every entry a config expands into, so only set it on
+# configs that expand to a single combination.
+
+
 @dataclasses.dataclass
 class LinuxConfig:
     """One entry in linux.json's 'configs' or 'package_configs' arrays."""
@@ -50,13 +46,12 @@ class LinuxConfig:
     compiler: list[str]
     build_type: list[str]
     arch: list[str]
+    minimal: bool
+    benchmark: bool = False  # if true, smoke-run the benchmarks after testing
     sanitizers: list[str] = dataclasses.field(default_factory=list)
     suffix: str = ""
     extra_cmake_args: str = ""
     image: str = ""  # only used by package_configs entries
-    # List of GitHub event names (e.g. "pull_request") on which this config
-    # should NOT run. Empty means it runs on every event.
-    exclude_event_types: list[str] = dataclasses.field(default_factory=list)
 
 
 @dataclasses.dataclass
@@ -89,11 +84,10 @@ class PlatformConfig:
     """One entry in macos.json's or windows.json's 'configs' array."""
 
     build_type: list[str]
+    minimal: bool
     build_only: bool = False  # if true, skip tests (e.g. macos/Windows Debug)
+    benchmark: bool = False  # if true, smoke-run the benchmarks after testing
     extra_cmake_args: str = ""
-    # List of GitHub event names (e.g. "pull_request") on which this config
-    # should NOT run. Empty means it runs on every event.
-    exclude_event_types: list[str] = dataclasses.field(default_factory=list)
 
     def __post_init__(self) -> None:
         if isinstance(self.build_type, str):
@@ -137,6 +131,7 @@ class MatrixEntry:
     cmake_args: str
     cmake_target: str
     build_only: bool
+    benchmark: bool
     build_type: str
     architecture: Architecture
     sanitizers: str
@@ -148,7 +143,8 @@ class MatrixEntry:
 class PackagingEntry:
     """One entry in the generated packaging strategy matrix."""
 
-    artifact_name: str
+    xrpld_artifact_name: str
+    validator_keys_artifact_name: str
     image: str
     distro: str  # e.g. "debian" or "rhel"; drives package-format-specific steps
 
@@ -168,20 +164,18 @@ _ARCHS: dict[str, Architecture] = {
 }
 
 
-def expand_linux_matrix(
-    linux: LinuxFile, event: str | None = None
-) -> list[MatrixEntry]:
+def expand_linux_matrix(linux: LinuxFile, minimal: bool) -> list[MatrixEntry]:
     """Expand a LinuxFile into a flat list of matrix entries.
 
     Each config entry is expanded over the cross-product of its
-    compiler, build_type, sanitizers, and architecture lists. Configs that
-    exclude the current event are skipped.
+    compiler, build_type, sanitizers, and architecture lists. When 'minimal' is
+    true, only configs flagged as minimal are included.
     """
     entries: list[MatrixEntry] = []
 
     for distro, configs in linux.configs.items():
         for cfg in configs:
-            if not runs_on_event(cfg.exclude_event_types, event):
+            if minimal and not cfg.minimal:
                 continue
             # An empty sanitizers list means "one entry with no sanitizer".
             effective_sanitizers = cfg.sanitizers or [""]
@@ -207,6 +201,7 @@ def expand_linux_matrix(
                         cmake_args=get_cmake_args(build_type, cfg.extra_cmake_args),
                         cmake_target="all",
                         build_only=False,
+                        benchmark=cfg.benchmark,
                         build_type=build_type,
                         architecture=arch_info,
                         sanitizers=sanitizer,
@@ -224,14 +219,19 @@ def expand_linux_packaging(linux: LinuxFile) -> list[PackagingEntry]:
     the nix-based build images, because deb/rpm tooling (debhelper, rpm-build)
     is taken from the distro's archive rather than from nixpkgs. Each config
     entry carries its own 'image'.
+
+    The artifact names must match what the build job uploads: one artifact per
+    binary, each named after the build config.
     """
     entries = []
     for distro, configs in linux.package_configs.items():
         for cfg in configs:
             for compiler, build_type in itertools.product(cfg.compiler, cfg.build_type):
+                config_name = f"{distro}-{compiler}-{build_type.lower()}-amd64"
                 entries.append(
                     PackagingEntry(
-                        artifact_name=f"xrpld-{distro}-{compiler}-{build_type.lower()}-amd64",
+                        xrpld_artifact_name=f"xrpld-{config_name}",
+                        validator_keys_artifact_name=f"validator-keys-{config_name}",
                         image=cfg.image,
                         distro=distro,
                     )
@@ -240,19 +240,17 @@ def expand_linux_packaging(linux: LinuxFile) -> list[PackagingEntry]:
     return entries
 
 
-def expand_platform_matrix(
-    pf: PlatformFile, event: str | None = None
-) -> list[MatrixEntry]:
+def expand_platform_matrix(pf: PlatformFile, minimal: bool) -> list[MatrixEntry]:
     """Expand a PlatformFile (macOS or Windows) into matrix entries.
 
-    Configs that exclude the current event are skipped.
+    When 'minimal' is true, only configs flagged as minimal are included.
     """
     platform_name, arch = pf.platform.split("/")
     is_windows = platform_name == "windows"
 
     entries: list[MatrixEntry] = []
     for cfg in pf.configs:
-        if not runs_on_event(cfg.exclude_event_types, event):
+        if minimal and not cfg.minimal:
             continue
         for build_type in cfg.build_type:
             entries.append(
@@ -261,6 +259,7 @@ def expand_platform_matrix(
                     cmake_args=get_cmake_args(build_type, cfg.extra_cmake_args),
                     cmake_target="install" if is_windows else "all",
                     build_only=cfg.build_only,
+                    benchmark=cfg.benchmark,
                     build_type=build_type,
                     architecture=Architecture(platform=pf.platform, runner=pf.runner),
                     sanitizers="",
@@ -292,12 +291,12 @@ if __name__ == "__main__":
         action="store_true",
     )
     parser.add_argument(
-        "-e",
-        "--event",
-        help="The GitHub event name that triggered the workflow (e.g. 'push', "
-        "'pull_request'). Configs are filtered by their 'event_type'. If "
-        "omitted, no filtering is applied.",
-        default=None,
+        "-m",
+        "--minimal",
+        help="Emit only the minimal matrix (the configs flagged 'minimal'), "
+        "used for pull requests by default. If omitted, the full matrix is "
+        "emitted.",
+        action="store_true",
     )
     args = parser.parse_args()
 
@@ -308,15 +307,15 @@ if __name__ == "__main__":
     else:
         if args.config in ("linux", None):
             matrix += expand_linux_matrix(
-                LinuxFile.load(THIS_DIR / "linux.json"), args.event
+                LinuxFile.load(THIS_DIR / "linux.json"), args.minimal
             )
         if args.config in ("macos", None):
             matrix += expand_platform_matrix(
-                PlatformFile.load(THIS_DIR / "macos.json"), args.event
+                PlatformFile.load(THIS_DIR / "macos.json"), args.minimal
             )
         if args.config in ("windows", None):
             matrix += expand_platform_matrix(
-                PlatformFile.load(THIS_DIR / "windows.json"), args.event
+                PlatformFile.load(THIS_DIR / "windows.json"), args.minimal
             )
 
     print(f"matrix={json.dumps({'include': [dataclasses.asdict(e) for e in matrix]})}")

.github/scripts/strategy-matrix/linux.json

@@ -1,31 +1,49 @@
 {
-  "image_tag": "sha-e29b523",
+  "image_tag": "sha-fecfc0c",
   "configs": {
     "ubuntu": [
-      {
-        "compiler": ["gcc", "clang"],
-        "build_type": ["Debug", "Release"],
-        "arch": ["amd64", "arm64"]
-      },
-
-      {
-        "compiler": ["gcc", "clang"],
-        "build_type": ["Debug", "Release"],
-        "arch": ["amd64"],
-        "sanitizers": ["address", "undefinedbehavior"]
-      },
-
       {
         "compiler": ["gcc"],
         "build_type": ["Debug"],
         "arch": ["amd64"],
+        "minimal": true,
         "suffix": "coverage",
         "extra_cmake_args": "-DUNIT_TEST_REFERENCE_FEE=500 -Dcoverage=ON -Dcoverage_format=xml -DCODE_COVERAGE_VERBOSE=ON -DCMAKE_C_FLAGS=-O0 -DCMAKE_CXX_FLAGS=-O0"
       },
+      {
+        "compiler": ["clang"],
+        "build_type": ["Release"],
+        "arch": ["amd64"],
+        "minimal": true,
+        "benchmark": true
+      },
+
+      {
+        "compiler": ["gcc"],
+        "build_type": ["Release"],
+        "arch": ["amd64"],
+        "minimal": false
+      },
+      {
+        "compiler": ["gcc", "clang"],
+        "build_type": ["Debug", "Release"],
+        "arch": ["arm64"],
+        "minimal": false
+      },
+
+      {
+        "compiler": ["gcc", "clang"],
+        "build_type": ["Debug", "Release"],
+        "arch": ["amd64"],
+        "minimal": false,
+        "sanitizers": ["address", "undefinedbehavior"]
+      },
+
       {
         "compiler": ["clang"],
         "build_type": ["Debug"],
         "arch": ["amd64"],
+        "minimal": false,
         "suffix": "voidstar",
         "extra_cmake_args": "-Dvoidstar=ON"
       },
@@ -33,6 +51,7 @@
         "compiler": ["clang"],
         "build_type": ["Release"],
         "arch": ["amd64"],
+        "minimal": false,
         "suffix": "reffee",
         "extra_cmake_args": "-DUNIT_TEST_REFERENCE_FEE=1000"
       },
@@ -40,9 +59,9 @@
         "compiler": ["gcc"],
         "build_type": ["Debug"],
         "arch": ["amd64"],
+        "minimal": false,
         "suffix": "unity",
-        "extra_cmake_args": "-Dunity=ON",
-        "exclude_event_types": ["pull_request"]
+        "extra_cmake_args": "-Dunity=ON"
       }
     ],
 
@@ -50,7 +69,9 @@
       {
         "compiler": ["gcc"],
         "build_type": ["Release"],
-        "arch": ["amd64"]
+        "arch": ["amd64"],
+        "minimal": false,
+        "extra_cmake_args": "-Dvalidator_keys=ON"
       }
     ],
 
@@ -58,7 +79,9 @@
       {
         "compiler": ["gcc"],
         "build_type": ["Release"],
-        "arch": ["amd64"]
+        "arch": ["amd64"],
+        "minimal": false,
+        "extra_cmake_args": "-Dvalidator_keys=ON"
       }
     ]
   },
@@ -68,6 +91,7 @@
         "compiler": ["gcc"],
         "build_type": ["Release"],
         "arch": ["amd64"],
+        "minimal": false,
         "image": "ghcr.io/xrplf/xrpld/packaging-debian:sha-577d745"
       }
     ],
@@ -77,6 +101,7 @@
         "compiler": ["gcc"],
         "build_type": ["Release"],
         "arch": ["amd64"],
+        "minimal": false,
         "image": "ghcr.io/xrplf/xrpld/packaging-rhel:sha-577d745"
       }
     ]

.github/scripts/strategy-matrix/macos.json

@@ -4,13 +4,14 @@
   "configs": [
     {
       "build_type": "Release",
-      "extra_cmake_args": "-DCMAKE_POLICY_VERSION_MINIMUM=3.5"
+      "extra_cmake_args": "-DCMAKE_POLICY_VERSION_MINIMUM=3.5",
+      "minimal": true
     },
     {
       "build_type": "Debug",
       "extra_cmake_args": "-DCMAKE_POLICY_VERSION_MINIMUM=3.5",
       "build_only": true,
-      "exclude_event_types": ["pull_request"]
+      "minimal": false
     }
   ]
 }

.github/scripts/strategy-matrix/windows.json

@@ -2,11 +2,11 @@
   "platform": "windows/amd64",
   "runner": ["self-hosted", "Windows", "dev-box-windows-2026"],
   "configs": [
-    { "build_type": "Release" },
+    { "build_type": "Release", "minimal": true },
     {
       "build_type": "Debug",
       "build_only": true,
-      "exclude_event_types": ["pull_request"]
+      "minimal": false
     }
   ]
 }

.github/workflows/build-nix-images.yml

@@ -1,4 +1,4 @@
-name: Build Nix Docker images
+name: Build `nix` Docker images
 
 on:
   push:
@@ -8,20 +8,24 @@ on:
       - ".github/workflows/build-nix-images.yml"
       - "flake.nix"
       - "flake.lock"
+      - "rust-toolchain.toml"
       - "nix/**"
       - "!nix/docker/README.md"
       - "!nix/devshell.nix"
       - "bin/check-tools.sh"
+      - "bin/default-loader-path.sh"
       - "bin/install-sanitizer-libs.sh"
   pull_request:
     paths:
       - ".github/workflows/build-nix-images.yml"
       - "flake.nix"
       - "flake.lock"
+      - "rust-toolchain.toml"
       - "nix/**"
       - "!nix/docker/README.md"
       - "!nix/devshell.nix"
       - "bin/check-tools.sh"
+      - "bin/default-loader-path.sh"
       - "bin/install-sanitizer-libs.sh"
   workflow_dispatch:
 
@@ -36,7 +40,7 @@ defaults:
 
 jobs:
   build-merge:
-    name: Build and push nix-${{ matrix.distro.name }}
+    name: Build and push `nix-${{ matrix.distro.name }}` image
     permissions:
       contents: read
       packages: write
@@ -54,7 +58,7 @@ jobs:
             base_image: debian:bookworm
           - name: rhel
             base_image: registry.access.redhat.com/ubi9/ubi:latest
-    uses: XRPLF/actions/.github/workflows/build-multiarch-image.yml@ee03d31bcc4501d7599dc1b1ecd7a34af582ad1c
+    uses: XRPLF/actions/.github/workflows/build-multiarch-image.yml@9e7e4e80af9e684c116b38369add8eea64451f32
     with:
       image_name: xrpld/nix-${{ matrix.distro.name }}
       dockerfile: nix/docker/Dockerfile

.github/workflows/build-packaging-images.yml

@@ -1,4 +1,4 @@
-name: Build packaging Docker images
+name: Build `packaging` Docker images
 
 on:
   push:
@@ -26,7 +26,7 @@ defaults:
 
 jobs:
   build-merge:
-    name: Build and push packaging-${{ matrix.distro.name }}
+    name: Build and push `packaging-${{ matrix.distro.name }}` image
     permissions:
       contents: read
       packages: write
@@ -38,7 +38,7 @@ jobs:
             base_image: debian:bookworm
           - name: rhel
             base_image: registry.access.redhat.com/ubi9/ubi:latest
-    uses: XRPLF/actions/.github/workflows/build-multiarch-image.yml@ee03d31bcc4501d7599dc1b1ecd7a34af582ad1c
+    uses: XRPLF/actions/.github/workflows/build-multiarch-image.yml@9e7e4e80af9e684c116b38369add8eea64451f32
     with:
       image_name: xrpld/packaging-${{ matrix.distro.name }}
       dockerfile: package/Dockerfile

.github/workflows/build-pre-commit-image.yml

@@ -0,0 +1,38 @@
+name: Build `pre-commit` Docker image
+
+on:
+  push:
+    branches:
+      - develop
+    paths:
+      - ".github/workflows/build-pre-commit-image.yml"
+      - "bin/pre-commit/Dockerfile"
+      - "rust-toolchain.toml"
+  pull_request:
+    paths:
+      - ".github/workflows/build-pre-commit-image.yml"
+      - "bin/pre-commit/Dockerfile"
+      - "rust-toolchain.toml"
+  workflow_dispatch:
+
+concurrency:
+  # Read `on-trigger.yml` for the rationale behind this concurrency group name.
+  group: ${{ github.workflow }}-${{ github.event_name == 'push' && github.ref == 'refs/heads/develop' && github.sha || github.ref }}
+  cancel-in-progress: true
+
+defaults:
+  run:
+    shell: bash
+
+jobs:
+  build-merge:
+    name: Build and push `pre-commit` image
+    permissions:
+      contents: read
+      packages: write
+    uses: XRPLF/actions/.github/workflows/build-multiarch-image.yml@9e7e4e80af9e684c116b38369add8eea64451f32
+    with:
+      image_name: xrpld/pre-commit
+      dockerfile: bin/pre-commit/Dockerfile
+      base_image: ubuntu:26.04
+      push: ${{ github.event_name == 'push' }}

.github/workflows/cargo-audit.yml

@@ -0,0 +1,84 @@
+# This workflow audits the Rust dependencies in crates/ for known security
+# advisories using cargo-audit. It runs on a weekly schedule, whenever the
+# dependency graph changes (Cargo.lock / Cargo.toml), and on demand. On a
+# scheduled run, a failure opens a tracking issue (matching the clang-tidy
+# workflow's behavior); on push/PR it simply fails the check.
+name: Cargo audit
+
+on:
+  schedule:
+    # 06:32 UTC every Monday.
+    - cron: "32 6 * * 1"
+  push:
+    branches:
+      - "develop"
+      - "release*"
+    paths:
+      - "crates/**/Cargo.toml"
+      - "crates/Cargo.lock"
+      - ".github/workflows/cargo-audit.yml"
+  pull_request:
+    paths:
+      - "crates/**/Cargo.toml"
+      - "crates/Cargo.lock"
+      - ".github/workflows/cargo-audit.yml"
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+defaults:
+  run:
+    shell: bash
+    working-directory: crates
+
+permissions:
+  contents: read
+
+jobs:
+  audit:
+    runs-on: ubuntu-latest
+    container: ghcr.io/xrplf/xrpld/nix-debian:sha-2e25435
+    permissions:
+      contents: read
+      # Needed to open an issue on scheduled failures.
+      issues: write
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+
+      - name: Run cargo audit
+        id: audit
+        continue-on-error: true
+        run: |
+          set -o pipefail
+          cargo audit | tee /tmp/cargo-audit.txt
+
+      - name: Prepare issue body
+        if: ${{ steps.audit.outcome != 'success' && github.event_name == 'schedule' }}
+        run: |
+          {
+              echo "## \`cargo audit\` found advisories"
+              echo
+              echo '```'
+              cat /tmp/cargo-audit.txt
+              echo '```'
+              echo
+              echo "---"
+              echo "*This issue was automatically created by the cargo-audit workflow.*"
+          } >/tmp/cargo-audit-issue.md
+
+      - name: Create issue
+        if: ${{ steps.audit.outcome != 'success' && github.event_name == 'schedule' }}
+        uses: XRPLF/actions/create-issue@2b8bc36af85b88bca0dd7bfac2e2dc05f94ad712
+        with:
+          title: "cargo audit found vulnerabilities"
+          body_file: /tmp/cargo-audit-issue.md
+          labels: "Bug,Security"
+
+      - name: Fail if advisories were found
+        if: ${{ steps.audit.outcome != 'success' }}
+        run: |
+          echo "cargo audit found advisories!"
+          exit 1

.github/workflows/check-pr-description.yml

@@ -23,7 +23,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Write PR body to file
         env:

.github/workflows/check-pr-title.yml

@@ -20,4 +20,4 @@ on:
 jobs:
   check_title:
     if: ${{ github.event.pull_request.draft != true }}
-    uses: XRPLF/actions/.github/workflows/check-pr-title.yml@cba1f0891650baf1a9c88624dc2d72573be2eb81
+    uses: XRPLF/actions/.github/workflows/check-pr-title.yml@d7c65e49225a38f6d8010eacf017bb5a98d7476c

.github/workflows/check-tools.yml

@@ -0,0 +1,114 @@
+# Verifies the committed snapshots of `bin/check-tools.sh` output for each Nix
+# environment (see nix/check-tools/). If the environment changes — a new image
+# tag, an updated flake.lock, a different tool list — without the matching
+# snapshot being regenerated and committed, this workflow fails so the drift is
+# caught in review.
+#
+# To regenerate the snapshots, see nix/check-tools/README.md.
+name: Check tools
+
+on:
+  pull_request:
+    paths:
+      - ".github/workflows/check-tools.yml"
+      - ".github/scripts/strategy-matrix/linux.json"
+      - "bin/check-tools.sh"
+      - "nix/**"
+      - "flake.nix"
+      - "flake.lock"
+      - "rust-toolchain.toml"
+  push:
+    branches:
+      - "develop"
+    paths:
+      - ".github/workflows/check-tools.yml"
+      - ".github/scripts/strategy-matrix/linux.json"
+      - "bin/check-tools.sh"
+      - "nix/**"
+      - "flake.nix"
+      - "flake.lock"
+      - "rust-toolchain.toml"
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+defaults:
+  run:
+    shell: bash
+
+jobs:
+  # The nix-nixos image tag is pinned alongside the build matrix in linux.json,
+  # so snapshots are checked against the exact image CI builds against.
+  linux-image-tag:
+    runs-on: ubuntu-latest
+    outputs:
+      tag: ${{ steps.tag.outputs.tag }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Read nix image tag
+        id: tag
+        run: echo "tag=$(jq -r .image_tag .github/scripts/strategy-matrix/linux.json)" >>"${GITHUB_OUTPUT}"
+
+  # One job for all environments; they differ only in whether the tools come
+  # from the nix-nixos container (Linux) or `nix develop` (macOS).
+  check-tools:
+    needs: linux-image-tag
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - runner: ubuntu-latest
+            snapshot: nix/check-tools/nix-ubuntu-amd64.txt
+            nix_develop: false
+          - runner: ubuntu-24.04-arm
+            snapshot: nix/check-tools/nix-ubuntu-arm64.txt
+            nix_develop: false
+          - runner: macos-26-apple-clang-21
+            snapshot: nix/check-tools/macos.txt
+            nix_develop: true
+    runs-on: ${{ matrix.runner }}
+    # Linux runs inside the pinned nix-nixos image; macOS runs natively and uses
+    # the flake's dev shell instead (see the run step below).
+    container: ${{ !matrix.nix_develop && format('ghcr.io/xrplf/xrpld/nix-ubuntu:{0}', needs.linux-image-tag.outputs.tag) || null }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Prepare runner
+        uses: XRPLF/actions/prepare-runner@c00c22ada3bd6bcda48fcb0d62fbbab49fec8a0f
+        with:
+          enable_ccache: false
+
+      - name: Regenerate snapshot
+        env:
+          CHECK_TOOLS_SKIP_CLONE: "1"
+          # check-tools.sh skips some macOS tools when CI is set; the snapshots
+          # capture the full `nix develop` environment, so unset it here.
+          CI: ""
+        run: |
+          if [ "${{ matrix.nix_develop }}" = "true" ]; then
+              # `nix develop` prints the dev-shell greeting first; keep only the
+              # check-tools.sh output (from the "Detected OS:" line onward).
+              nix --extra-experimental-features "nix-command flakes" develop \
+                  -c bash bin/check-tools.sh | sed -n '/^Detected OS:/,$p' >"${{ matrix.snapshot }}"
+          else
+              bash bin/check-tools.sh >"${{ matrix.snapshot }}"
+          fi
+
+      - name: Verify snapshot is up to date
+        run: |
+          if ! git diff --exit-code -- "${{ matrix.snapshot }}"; then
+              echo "::error::${{ matrix.snapshot }} is out of date. Regenerate it (see nix/check-tools/README.md) and commit the result."
+              exit 1
+          fi
+
+      - name: Upload regenerated snapshot
+        if: failure()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: check-tools-${{ runner.os }}-${{ runner.arch }}
+          path: ${{ matrix.snapshot }}

.github/workflows/conflicting-pr.yml

@@ -14,6 +14,7 @@ permissions:
 
 jobs:
   main:
+    if: ${{ !contains(github.event.pull_request.labels.*.name, 'IgnoreConflicts') }}
     runs-on: ubuntu-latest
     steps:
       - name: Check if PRs are dirty

.github/workflows/on-pr.yml

@@ -1,7 +1,11 @@
-# This workflow runs all workflows to check, build and test the project on
-# various Linux flavors, as well as on MacOS and Windows, on every push to a
-# user branch. However, it will not run if the pull request is a draft unless it
-# has the 'DraftRunCI' label. For commits to PRs that target a release branch,
+# This workflow runs workflows to check, build and test the project
+# on every meaningful change on pull_request.
+# However, it will not run if the PR is a draft
+# unless it has the 'DraftRunCI' or 'Full CI build' label.
+#
+# By default a PR builds only a minimal matrix.
+# The full matrix runs once the PR is labeled "Ready to merge" or "Full CI build".
+# For commits to PRs that target a release branch,
 # it also uploads the libxrpl recipe to the Conan remote.
 name: PR
 
@@ -15,8 +19,16 @@ on:
       - reopened
       - synchronize
       - ready_for_review
+      # Trigger on label changes so toggling "Ready to merge" or "Full CI build"
+      # switches between the minimal and full matrix without needing a new push.
+      - labeled
+      - unlabeled
 
 concurrency:
+  # A single per-ref group with cancel-in-progress means any newer run (a push
+  # or a label change) supersedes the in-progress one for that ref. Keeping
+  # exactly one authoritative run per ref ensures a fast do-nothing run can never
+  # mask a real build's checks.
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
@@ -25,15 +37,22 @@ defaults:
     shell: bash
 
 jobs:
-  # This job determines whether the rest of the workflow should run. It runs
-  # when the PR is not a draft (which should also cover merge-group) or
-  # has the 'DraftRunCI' label.
+  # This job determines whether the rest of the workflow should run at all,
+  # based on the current set of labels: it runs when the PR is not a draft
+  # (which should also cover merge-group) or has the 'DraftRunCI' or
+  # 'Full CI build' label. Whether a build then happens, and whether it is the
+  # minimal or full matrix, is decided further below and in the strategy matrix.
   should-run:
-    if: ${{ !github.event.pull_request.draft || contains(github.event.pull_request.labels.*.name, 'DraftRunCI') }}
+    if: >-
+      ${{
+        !github.event.pull_request.draft
+        || contains(github.event.pull_request.labels.*.name, 'DraftRunCI')
+        || contains(github.event.pull_request.labels.*.name, 'Full CI build')
+      }}
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       - name: Determine changed files
         # This step checks whether any files have changed that should
         # cause the next jobs to run. We do it this way rather than
@@ -63,17 +82,21 @@ jobs:
             .github/scripts/strategy-matrix/**
             .github/workflows/reusable-build-test-config.yml
             .github/workflows/reusable-build-test.yml
+            .github/workflows/reusable-check-autogen.yml
             .github/workflows/reusable-clang-tidy.yml
             .github/workflows/reusable-package.yml
+            .github/workflows/reusable-rust.yml
             .github/workflows/reusable-strategy-matrix.yml
             .github/workflows/reusable-test.yml
             .github/workflows/reusable-upload-recipe.yml
             .clang-tidy
             .codecov.yml
             bin/check-tools.sh
+            bin/default-loader-path.sh
             cfg/**
             cmake/**
             conan/**
+            crates/**
             external/**
             include/**
             src/**
@@ -91,19 +114,26 @@ jobs:
         # least one of:
         # * Any of the files checked in the `changes` step were modified
         # * The PR is NOT a draft and is labeled "Ready to merge"
+        # * The PR is labeled "Full CI build" (draft or not)
         # * The workflow is running from the merge queue
         id: go
         env:
           FILES: ${{ steps.changes.outputs.any_changed }}
           DRAFT: ${{ github.event.pull_request.draft }}
           READY: ${{ contains(github.event.pull_request.labels.*.name, 'Ready to merge') }}
+          FULL: ${{ contains(github.event.pull_request.labels.*.name, 'Full CI build') }}
           MERGE: ${{ github.event_name == 'merge_group' }}
         run: |
-          echo "go=${{ (env.DRAFT != 'true' && env.READY == 'true') || env.FILES == 'true' || env.MERGE == 'true' }}" >>"${GITHUB_OUTPUT}"
+          echo "go=${{ (env.DRAFT != 'true' && env.READY == 'true') || env.FULL == 'true' || env.FILES == 'true' || env.MERGE == 'true' }}" >>"${GITHUB_OUTPUT}"
           cat "${GITHUB_OUTPUT}"
     outputs:
       go: ${{ steps.go.outputs.go == 'true' }}
 
+  check-autogen:
+    needs: should-run
+    if: ${{ needs.should-run.outputs.go == 'true' }}
+    uses: ./.github/workflows/reusable-check-autogen.yml
+
   check-levelization:
     needs: should-run
     if: ${{ needs.should-run.outputs.go == 'true' }}
@@ -140,9 +170,19 @@ jobs:
     secrets:
       CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
 
+  rust:
+    needs: should-run
+    if: ${{ needs.should-run.outputs.go == 'true' }}
+    uses: ./.github/workflows/reusable-rust.yml
+    secrets:
+      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+
   package:
     needs: [should-run, build-test]
-    if: ${{ needs.should-run.outputs.go == 'true' }}
+    # Packaging consumes the debian/rhel release binaries, which are only built
+    # by the full matrix. Skip it for pull requests that ran only the minimal
+    # matrix (i.e. not yet labeled "Ready to merge" or "Full CI build").
+    if: ${{ needs.should-run.outputs.go == 'true' && (github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'Ready to merge') || contains(github.event.pull_request.labels.*.name, 'Full CI build')) }}
     uses: ./.github/workflows/reusable-package.yml
 
   upload-recipe:
@@ -175,10 +215,12 @@ jobs:
   passed:
     if: failure() || cancelled()
     needs:
+      - check-autogen
       - check-levelization
       - check-rename
       - clang-tidy
       - build-test
+      - rust
       - package
       - upload-recipe
       - notify-clio

.github/workflows/on-trigger.yml

@@ -20,17 +20,21 @@ on:
       - ".github/scripts/strategy-matrix/**"
       - ".github/workflows/reusable-build-test-config.yml"
       - ".github/workflows/reusable-build-test.yml"
+      - ".github/workflows/reusable-check-autogen.yml"
       - ".github/workflows/reusable-clang-tidy.yml"
       - ".github/workflows/reusable-package.yml"
+      - ".github/workflows/reusable-rust.yml"
       - ".github/workflows/reusable-strategy-matrix.yml"
       - ".github/workflows/reusable-test.yml"
       - ".github/workflows/reusable-upload-recipe.yml"
       - ".clang-tidy"
       - ".codecov.yml"
       - "bin/check-tools.sh"
+      - "bin/default-loader-path.sh"
       - "cfg/**"
       - "cmake/**"
       - "conan/**"
+      - "crates/**"
       - "external/**"
       - "include/**"
       - "src/**"
@@ -66,6 +70,9 @@ defaults:
     shell: bash
 
 jobs:
+  check-autogen:
+    uses: ./.github/workflows/reusable-check-autogen.yml
+
   clang-tidy:
     uses: ./.github/workflows/reusable-clang-tidy.yml
     permissions:
@@ -91,6 +98,11 @@ jobs:
     secrets:
       CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
 
+  rust:
+    uses: ./.github/workflows/reusable-rust.yml
+    secrets:
+      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+
   upload-recipe:
     needs: build-test
     # Only run when pushing to the develop branch.

.github/workflows/pre-commit.yml

@@ -14,7 +14,7 @@ on:
 jobs:
   # Call the workflow in the XRPLF/actions repo that runs the pre-commit hooks.
   run-hooks:
-    uses: XRPLF/actions/.github/workflows/pre-commit.yml@1bde119a1ab71305ba5d3716e7a82cea1c7bdede
+    uses: XRPLF/actions/.github/workflows/pre-commit.yml@3ba08d6ddf114092891d48491fc2e26c3ba15552
     with:
       runs_on: ubuntu-latest
-      container: '{ "image": "ghcr.io/xrplf/ci/tools-rippled-pre-commit:sha-41ec7c1" }'
+      container: '{ "image": "ghcr.io/xrplf/xrpld/pre-commit:sha-f56b79f" }'

.github/workflows/publish-docs.yml

@@ -41,13 +41,13 @@ env:
 jobs:
   build:
     runs-on: ubuntu-latest
-    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-e29b523
+    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-fecfc0c
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Prepare runner
-        uses: XRPLF/actions/prepare-runner@64ec3cf3b152b4444638f470bbd6df7a7a30c81c
+        uses: XRPLF/actions/prepare-runner@c00c22ada3bd6bcda48fcb0d62fbbab49fec8a0f
         with:
           enable_ccache: false
 

.github/workflows/reusable-build-test-config.yml

@@ -3,6 +3,12 @@ name: Build and test configuration
 on:
   workflow_call:
     inputs:
+      benchmark:
+        description: "Whether to smoke-run the benchmarks after testing."
+        required: false
+        type: boolean
+        default: false
+
       build_only:
         description: 'Whether to only build or to build and test the code ("true", "false").'
         required: true
@@ -100,9 +106,10 @@ jobs:
       # header files are copied into separate directories by CMake, which will
       # otherwise result in cache misses.
       CCACHE_SLOPPINESS: include_file_ctime,include_file_mtime
-      # Determine if coverage and voidstar should be enabled.
+      # Determine if coverage, voidstar and validator-keys should be enabled.
       COVERAGE_ENABLED: ${{ contains(inputs.cmake_args, '-Dcoverage=ON') }}
       VOIDSTAR_ENABLED: ${{ contains(inputs.cmake_args, '-Dvoidstar=ON') }}
+      VALIDATOR_KEYS_ENABLED: ${{ contains(inputs.cmake_args, '-Dvalidator_keys=ON') }}
       SANITIZERS_ENABLED: ${{ inputs.sanitizers != '' }}
     steps:
       - name: Cleanup workspace (macOS and Windows)
@@ -110,10 +117,10 @@ jobs:
         uses: XRPLF/actions/cleanup-workspace@c7d9ce5ebb03c752a354889ecd870cadfc2b1cd4
 
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Prepare runner
-        uses: XRPLF/actions/prepare-runner@64ec3cf3b152b4444638f470bbd6df7a7a30c81c
+        uses: XRPLF/actions/prepare-runner@c00c22ada3bd6bcda48fcb0d62fbbab49fec8a0f
         with:
           enable_ccache: ${{ inputs.ccache_enabled }}
 
@@ -124,7 +131,7 @@ jobs:
       - name: Check tools
         env:
           CHECK_TOOLS_SKIP_CLONE: "1"
-        run: ./bin/check-tools.sh
+        run: ./bin/check-tools.sh || true
 
       - name: Print build environment
         uses: XRPLF/actions/print-build-env@59dec886e4afb05a1724443af08baccbc045b574
@@ -170,9 +177,9 @@ jobs:
               ..
 
       # Export the sanitizer options before any instrumented binary runs. The
-      # protocol code-gen and build steps below invoke instrumented dependency
-      # tools (protoc, grpc), so setting UBSAN_OPTIONS here lets the UBSan
-      # suppression list silence their diagnostics too, not just at test time.
+      # build step below invokes instrumented dependency tools (protoc, grpc),
+      # so setting UBSAN_OPTIONS here lets the UBSan suppression list silence
+      # their diagnostics too, not just at test time.
       # GITHUB_WORKSPACE (not the github.workspace context) is used so the path
       # resolves correctly inside the container job.
       - name: Set sanitizer options
@@ -190,32 +197,6 @@ jobs:
           echo "UBSAN_OPTIONS=include=${SUPP}/runtime-ubsan-options.txt:suppressions=${SUPP}/ubsan.supp" >>${GITHUB_ENV}
           echo "LSAN_OPTIONS=include=${SUPP}/runtime-lsan-options.txt:suppressions=${SUPP}/lsan.supp" >>${GITHUB_ENV}
 
-      - name: Check protocol autogen files are up-to-date
-        working-directory: ${{ env.BUILD_DIR }}
-        env:
-          MESSAGE: |
-
-            The generated protocol wrapper classes are out of date.
-
-            This typically happens when the macro files or generator scripts
-            have changed but the generated files were not regenerated.
-
-            To fix this:
-              1. Run: cmake --build . --target setup_code_gen
-              2. Run: cmake --build . --target code_gen
-              3. Commit and push the regenerated files
-        run: |
-          set -e
-          cmake --build . --target setup_code_gen
-          cmake --build . --target code_gen
-          DIFF=$(git -C .. status --porcelain -- include/xrpl/protocol_autogen src/tests/libxrpl/protocol_autogen)
-          if [ -n "${DIFF}" ]; then
-              echo "::error::Generated protocol files are out of date"
-              git -C .. diff -- include/xrpl/protocol_autogen src/tests/libxrpl/protocol_autogen
-              echo "${MESSAGE}"
-              exit 1
-          fi
-
       - name: Build the binary
         working-directory: ${{ env.BUILD_DIR }}
         env:
@@ -223,11 +204,13 @@ jobs:
           BUILD_TYPE: ${{ inputs.build_type }}
           CMAKE_TARGET: ${{ inputs.cmake_target }}
         run: |
+          set -o pipefail
           cmake \
               --build . \
               --config "${BUILD_TYPE}" \
               --parallel "${BUILD_NPROC}" \
-              --target "${CMAKE_TARGET}"
+              --target "${CMAKE_TARGET}" \
+              2>&1 | tee "${GITHUB_WORKSPACE}/build.log"
 
       - name: Show ccache statistics
         if: ${{ inputs.ccache_enabled }}
@@ -247,6 +230,22 @@ jobs:
           retention-days: 3
           if-no-files-found: error
 
+      - name: Run the validator-keys tests
+        if: ${{ env.VALIDATOR_KEYS_ENABLED == 'true' }}
+        working-directory: ${{ env.BUILD_DIR }}
+        run: ./validator-keys --unittest
+
+      - name: Upload the validator-keys binary
+        if: ${{ github.event.repository.visibility == 'public' && env.VALIDATOR_KEYS_ENABLED == 'true' }}
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: validator-keys-${{ inputs.config_name }}
+          path: |
+            ${{ env.BUILD_DIR }}/validator-keys
+            ${{ env.BUILD_DIR }}/validator-keys-LICENSE
+          retention-days: 3
+          if-no-files-found: error
+
       - name: Upload the test binary (Linux)
         if: ${{ github.event.repository.visibility == 'public' && runner.os == 'Linux' }}
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
@@ -264,7 +263,7 @@ jobs:
           ./xrpld --definitions | python3 -m json.tool >server_definitions.json
 
       - name: Upload server definitions
-        if: ${{ github.event.repository.visibility == 'public' && inputs.config_name == 'debian-gcc-release-amd64' }}
+        if: ${{ github.event.repository.visibility == 'public' && inputs.config_name == 'ubuntu-gcc-debug-amd64-coverage' }}
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: server-definitions
@@ -277,7 +276,7 @@ jobs:
         working-directory: ${{ env.BUILD_DIR }}
         run: |
           ldd ./xrpld
-          if [ "$(ldd ./xrpld | grep -E '(libstdc\+\+|libgcc)' | wc -l)" -eq 0 ]; then
+          if [ "$(ldd ./xrpld | grep -E '(libstdc\+\+)' | wc -l)" -eq 0 ]; then
               echo 'The binary is statically linked.'
           else
               echo 'The binary is dynamically linked.'
@@ -290,6 +289,14 @@ jobs:
         run: |
           ./xrpld --version | grep libvoidstar
 
+      - name: Run Rust tests
+        if: ${{ !inputs.build_only }}
+        working-directory: crates
+        # `xrpl-wasm-vm-ffi` is left out on Windows: its tests link as an executable, and
+        # MSVC - unlike the Unix linkers - will not dead-strip the never-called cxx wrappers
+        # whose C++ shims only the CMake build defines. The other runners cover these tests.
+        run: cargo nextest run --workspace --all-features --locked --no-tests=warn ${{ runner.os == 'Windows' && '--exclude xrpl-wasm-vm-ffi' || '' }}
+
       - name: Run the separate tests
         if: ${{ !inputs.build_only }}
         working-directory: ${{ runner.os == 'Windows' && format('{0}/{1}', env.BUILD_DIR, inputs.build_type) || env.BUILD_DIR }}
@@ -322,27 +329,49 @@ jobs:
               PRELOAD=""
           fi
 
-          LD_PRELOAD="$PRELOAD" ./xrpld --unittest --unittest-jobs "${BUILD_NPROC}" 2>&1 | tee unittest.log
+          LD_PRELOAD="$PRELOAD" ./xrpld --unittest --unittest-jobs "${BUILD_NPROC}" 2>&1 | tee "${GITHUB_WORKSPACE}/unittest.log"
 
-      - name: Show test failure summary
-        if: ${{ failure() && !inputs.build_only }}
-        env:
-          WORKING_DIR: ${{ runner.os == 'Windows' && format('{0}\{1}', env.BUILD_DIR, inputs.build_type) || env.BUILD_DIR }}
+      # Smoke-run every benchmark module with a single repetition to confirm the
+      # benchmarks still build and execute. This is a correctness check, not a
+      # performance measurement, so there is nothing to gain from repeating it
+      # across configurations: it is opted into by a single config in the
+      # strategy matrix (see the 'benchmark' flag in the JSON files), which
+      # keeps it off instrumented builds (sanitizers/coverage/voidstar), where
+      # it would be slow and meaningless, off Debug builds, where it is much
+      # slower, and off Windows, where the `install` target does not build them.
+      - name: Run the benchmarks
+        if: ${{ inputs.benchmark }}
+        working-directory: ${{ env.BUILD_DIR }}
         run: |
-          if [ ! -d "${WORKING_DIR}" ]; then
-              echo "Working directory '${WORKING_DIR}' does not exist."
-              exit 0
-          fi
+          rc=0
+          while IFS= read -r bench; do
+              echo "::group::${bench}"
+              "./${bench}" --benchmark_repetitions=1 || rc=1
+              echo "::endgroup::"
+          done < <(find src/benchmarks -type f -perm -u+x -name 'xrpl.bench.*')
+          exit "${rc}"
 
-          cd "${WORKING_DIR}"
+      - name: Show build/test failure summary
+        if: ${{ failure() }}
+        run: |
+          cd "${GITHUB_WORKSPACE}"
 
-          if [ ! -f unittest.log ]; then
-              echo "unittest.log not found; embedded tests may not have run."
-              exit 0
-          fi
-
-          if ! grep -E "failed" unittest.log; then
-              echo "Log present but no failure lines found in unittest.log."
+          if [ -f unittest.log ]; then
+              if ! grep -E "failed" unittest.log | grep -vE "^I[0-9]|^[0-9]+> (ERR:|FTL:)"; then
+                  echo "unittest.log present but no failure lines found."
+              fi
+          elif [ -f build.log ]; then
+              # GCC/Clang emit "error:" (covers "fatal error:"); MSVC emits
+              # "error C####:", "error LNK####:", and "fatal error LNK####:".
+              # -A6 prints the lines that follow each match (source line, caret,
+              # notes, and the "N errors generated" tally) to capture the whole
+              # diagnostic block.
+              if ! grep -E -A6 "error:|error C[0-9]{4}|error LNK[0-9]{4}|fatal error" build.log; then
+                  echo "build.log present but no compile errors found."
+              fi
+          else
+              echo "unittest.log/build.log not found; something went wrong."
+              exit 1
           fi
       - name: Debug failure (Linux)
         if: ${{ failure() && runner.os == 'Linux' && !inputs.build_only }}
@@ -366,13 +395,14 @@ jobs:
               --target coverage
 
       - name: Upload coverage report
-        if: ${{ github.repository == 'XRPLF/rippled' && !inputs.build_only && env.COVERAGE_ENABLED == 'true' }}
+        if: ${{ github.repository_owner == 'XRPLF' && !inputs.build_only && env.COVERAGE_ENABLED == 'true' }}
         uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
         with:
           disable_search: true
           disable_telem: true
           fail_ci_if_error: true
           files: ${{ env.BUILD_DIR }}/coverage.xml
+          flags: cpp
           plugins: noop
           token: ${{ secrets.CODECOV_TOKEN }}
           verbose: true

.github/workflows/reusable-build-test.yml

@@ -40,6 +40,7 @@ jobs:
       fail-fast: ${{ github.event_name == 'merge_group' }}
       matrix: ${{ fromJson(needs.generate-matrix.outputs.matrix) }}
     with:
+      benchmark: ${{ matrix.benchmark }}
       build_only: ${{ matrix.build_only }}
       build_type: ${{ matrix.build_type }}
       ccache_enabled: ${{ inputs.ccache_enabled }}

.github/workflows/reusable-check-autogen.yml

@@ -0,0 +1,76 @@
+# This workflow checks that the generated protocol wrapper classes are
+# up-to-date with the macro files and generator scripts they are produced from,
+# see more info in include/xrpl/protocol_autogen/README.md.
+name: Check autogen
+
+# This workflow can only be triggered by other workflows.
+on: workflow_call
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}-autogen
+  cancel-in-progress: true
+
+defaults:
+  run:
+    shell: bash
+
+env:
+  BUILD_DIR: build/codegen
+
+jobs:
+  autogen:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Set up Python
+        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
+        with:
+          python-version: "3.13"
+
+      # Code generation is pure Python, so the standalone project below offers
+      # the same targets as the main build without needing its dependencies or
+      # a compiler, which keeps this job down to a few seconds.
+      - name: Configure CMake
+        run: cmake -S cmake/codegen -B "${BUILD_DIR}"
+
+      - name: Install code generation dependencies
+        run: cmake --build "${BUILD_DIR}" --target setup_code_gen
+
+      - name: Generate code
+        run: cmake --build "${BUILD_DIR}" --target code_gen
+
+      - name: Check for differences
+        env:
+          MESSAGE: |
+
+            The generated protocol wrapper classes are out of date.
+
+            This typically happens when the macro files or generator scripts
+            have changed but the generated files were not regenerated.
+
+            Run the following from the repository root, then commit and push
+            the regenerated files. This needs neither the dependencies nor a
+            compiler. See include/xrpl/protocol_autogen/README.md for more info.
+
+              cmake -S cmake/codegen -B build/codegen
+              cmake --build build/codegen --target setup_code_gen
+              cmake --build build/codegen --target code_gen
+
+            In an already configured build directory, the 'setup_code_gen' and
+            'code_gen' targets do the same thing.
+        run: |
+          # Record untracked files in the index without staging their contents,
+          # so that classes generated for a newly added transaction or ledger
+          # entry type show up in the diff below rather than silently as an
+          # empty one.
+          git add --intent-to-add .
+          DIFF=$(git status --porcelain)
+          if [ -n "${DIFF}" ]; then
+              # Print the differences to give the contributor a hint about what to
+              # expect when running code generation on their own machine.
+              git diff
+              echo "${MESSAGE}"
+              exit 1
+          fi

.github/workflows/reusable-check-levelization.yml

@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       - name: Check levelization
         run: python .github/scripts/levelization/generate.py
       - name: Check for differences

.github/workflows/reusable-check-rename.yml

@@ -18,7 +18,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
       - name: Check definitions
         run: .github/scripts/rename/definitions.sh .
       - name: Check copyright notices

.github/workflows/reusable-clang-tidy.yml

@@ -34,16 +34,16 @@ jobs:
     needs: [determine-files]
     if: ${{ needs.determine-files.outputs.cpp_changed_files != '' || needs.determine-files.outputs.need_full_run == 'true' }}
     runs-on: ["self-hosted", "Linux", "X64", "heavy"]
-    container: "ghcr.io/xrplf/xrpld/nix-debian:sha-e29b523"
+    container: "ghcr.io/xrplf/xrpld/nix-debian:sha-fecfc0c"
     permissions:
       contents: read
       issues: write
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Prepare runner
-        uses: XRPLF/actions/prepare-runner@64ec3cf3b152b4444638f470bbd6df7a7a30c81c
+        uses: XRPLF/actions/prepare-runner@c00c22ada3bd6bcda48fcb0d62fbbab49fec8a0f
         with:
           enable_ccache: false
 
@@ -88,6 +88,12 @@ jobs:
         run: |
           ninja -j ${{ steps.nproc.outputs.nproc }} xrpl.libpb
 
+      # clang-tidy needs cxxbridge headers generated from Rust crates
+      - name: Build xrpl_crates
+        working-directory: ${{ env.BUILD_DIR }}
+        run: |
+          ninja -j ${{ steps.nproc.outputs.nproc }} xrpl_crates
+
       - name: Run clang tidy
         id: run_clang_tidy
         continue-on-error: true
@@ -95,7 +101,7 @@ jobs:
           TARGETS: ${{ needs.determine-files.outputs.need_full_run != 'true' && needs.determine-files.outputs.cpp_changed_files || 'include src tests' }}
         run: |
           set -o pipefail
-          run-clang-tidy -j ${{ steps.nproc.outputs.nproc }} -p "${BUILD_DIR}" -quiet -fix -allow-no-checks ${TARGETS} 2>&1 | tee "${OUTPUT_FILE}"
+          run-clang-tidy -j ${{ steps.nproc.outputs.nproc }} -p "${BUILD_DIR}" -quiet -fix -format -allow-no-checks ${TARGETS} 2>&1 | tee "${OUTPUT_FILE}"
 
       - name: Print filtered clang-tidy errors
         if: ${{ steps.run_clang_tidy.outcome != 'success' }}

.github/workflows/reusable-package.yml

@@ -1,7 +1,7 @@
-# Build Linux packages (DEB and RPM) from pre-built binary artifacts.
-# Discovers which configurations to package from linux.json (configs in
-# "package_configs") and fans out one job per distro. Only linux/amd64 is
-# supported; the runner is hardcoded in the job below.
+# Build Linux packages (DEB and RPM) from pre-built binary artifacts (xrpld and
+# validator-keys). Discovers which configurations to package from linux.json
+# (configs in "package_configs") and fans out one job per distro. Only
+# linux/amd64 is supported; the runner is hardcoded in the job below.
 name: Package
 
 on:
@@ -27,10 +27,10 @@ jobs:
       matrix: ${{ steps.generate.outputs.matrix }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Set up Python
-        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
+        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
         with:
           python-version: "3.13"
 
@@ -45,7 +45,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix: ${{ fromJson(needs.generate-matrix.outputs.matrix) }}
-    name: "${{ matrix.artifact_name }}"
+    name: "${{ matrix.xrpld_artifact_name }}"
     permissions:
       contents: read
     runs-on: ["self-hosted", "Linux", "X64", "heavy"]
@@ -54,16 +54,22 @@ jobs:
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
-      - name: Download pre-built binary
+      - name: Download pre-built xrpld binary
         uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
         with:
-          name: ${{ matrix.artifact_name }}
+          name: ${{ matrix.xrpld_artifact_name }}
           path: ${{ env.BUILD_DIR }}
 
-      - name: Make binary executable
-        run: chmod +x "${BUILD_DIR}/xrpld"
+      - name: Download pre-built validator-keys binary
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: ${{ matrix.validator_keys_artifact_name }}
+          path: ${{ env.BUILD_DIR }}
+
+      - name: Make binaries executable
+        run: chmod +x "${BUILD_DIR}/xrpld" "${BUILD_DIR}/validator-keys"
 
       - name: Build package
         env:
@@ -73,7 +79,7 @@ jobs:
       - name: Upload package artifact
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
-          name: ${{ matrix.artifact_name }}-pkg
+          name: ${{ matrix.xrpld_artifact_name }}-pkg
           path: |
             ${{ env.BUILD_DIR }}/debbuild/*.deb
             ${{ env.BUILD_DIR }}/debbuild/*.ddeb

.github/workflows/reusable-rust.yml

@@ -0,0 +1,86 @@
+# Clippy, coverage and documentation for the Rust crates in crates/. Each runs
+# as an independent job on a GitHub-hosted runner, but inside the same container
+# image used to build the crates in the C++/Corrosion path, so the toolchain
+# (and therefore the lints, coverage instrumentation and the cargo cache) matches
+# what production builds use.
+#
+# Rust unit tests are deliberately NOT run here. They run as part of the C++
+# build (reusable-build-test-config.yml), which already compiles the crates on a
+# self-hosted runner, so there is no need to provision a toolchain again.
+name: Rust
+
+on:
+  workflow_call:
+    secrets:
+      CODECOV_TOKEN:
+        description: "The Codecov token to use for uploading coverage reports."
+        required: false
+
+defaults:
+  run:
+    shell: bash
+    working-directory: crates
+
+permissions:
+  contents: read
+
+jobs:
+  clippy:
+    runs-on: ubuntu-latest
+    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-fecfc0c
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+
+      - name: Cache cargo artifacts
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
+        with:
+          workspaces: crates
+
+      - name: Run clippy
+        run: cargo clippy --workspace --all-targets --all-features --locked -- -D warnings
+
+  coverage:
+    runs-on: ubuntu-latest
+    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-fecfc0c
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+
+      - name: Cache cargo artifacts
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
+        with:
+          workspaces: crates
+
+      - name: Generate coverage report
+        run: cargo llvm-cov nextest --workspace --all-features --locked --no-tests=warn --lcov --output-path lcov.info
+
+      - name: Upload coverage report
+        if: ${{ github.repository == 'XRPLF/rippled' }}
+        uses: codecov/codecov-action@fb8b3582c8e4def4969c97caa2f19720cb33a72f # v7.0.0
+        with:
+          disable_search: true
+          disable_telem: true
+          fail_ci_if_error: true
+          files: crates/lcov.info
+          flags: rust
+          plugins: noop
+          token: ${{ secrets.CODECOV_TOKEN }}
+          verbose: true
+
+  doc:
+    runs-on: ubuntu-latest
+    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-fecfc0c
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6.0.3
+
+      - name: Cache cargo artifacts
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
+        with:
+          workspaces: crates
+
+      - name: Build documentation
+        env:
+          RUSTDOCFLAGS: "-D warnings"
+        run: cargo doc --workspace --no-deps --all-features --locked

.github/workflows/reusable-strategy-matrix.yml

@@ -23,10 +23,10 @@ jobs:
       matrix: ${{ steps.generate.outputs.matrix }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Set up Python
-        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
+        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
         with:
           python-version: "3.13"
 
@@ -35,5 +35,8 @@ jobs:
         id: generate
         env:
           GENERATE_CONFIG: ${{ inputs.os != '' && format('--config={0}', inputs.os) || '' }}
-          GENERATE_EVENT: ${{ github.event_name }}
-        run: ./generate.py ${GENERATE_CONFIG} --event="${GENERATE_EVENT}" >>"${GITHUB_OUTPUT}"
+          # Run only the minimal matrix for pull requests that are not yet
+          # labeled "Ready to merge" or "Full CI build". Any other event (merge
+          # queue, push, schedule, manual dispatch) runs the full matrix.
+          GENERATE_MINIMAL: ${{ (github.event_name == 'pull_request' && !contains(github.event.pull_request.labels.*.name, 'Ready to merge') && !contains(github.event.pull_request.labels.*.name, 'Full CI build')) && '--minimal' || '' }}
+        run: ./generate.py ${GENERATE_CONFIG} ${GENERATE_MINIMAL} >>"${GITHUB_OUTPUT}"

.github/workflows/reusable-upload-recipe.yml

@@ -40,14 +40,14 @@ defaults:
 jobs:
   upload:
     runs-on: ubuntu-latest
-    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-e29b523
+    container: ghcr.io/xrplf/xrpld/nix-ubuntu:sha-fecfc0c
     env:
       REMOTE_NAME: ${{ inputs.remote_name }}
       CONAN_LOGIN_USERNAME_XRPLF: ${{ secrets.remote_username }}
       CONAN_PASSWORD_XRPLF: ${{ secrets.remote_password }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Generate build version number
         id: version

.github/workflows/upload-conan-deps.yml

@@ -65,10 +65,10 @@ jobs:
         uses: XRPLF/actions/cleanup-workspace@c7d9ce5ebb03c752a354889ecd870cadfc2b1cd4
 
       - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Prepare runner
-        uses: XRPLF/actions/prepare-runner@64ec3cf3b152b4444638f470bbd6df7a7a30c81c
+        uses: XRPLF/actions/prepare-runner@c00c22ada3bd6bcda48fcb0d62fbbab49fec8a0f
         with:
           enable_ccache: false
 

.gitignore

@@ -81,8 +81,14 @@ DerivedData
 # Python
 __pycache__
 
+# Rust build artifacts.
+target/
+
 # Direnv's directory
 /.direnv
 
 # clangd cache
 /.cache
+
+# Rust build directory
+crates/target

.pre-commit-config.yaml

@@ -32,6 +32,11 @@ repos:
         # as standalone translation units, so they have no compile_commands.json
         # entry to lint (verify_headers checks them transitively).
         exclude: '^include/xrpl/protocol_autogen|\.ipp$'
+        # run-clang-tidy --fix may edit headers included by files it is not run on,
+        # so pre-commit must not split the files across parallel hook invocations.
+        # The script determines the staged files itself and lets run-clang-tidy
+        # handle parallelism internally.
+        pass_filenames: false
       - id: fix-include-style
         name: fix include style
         entry: ./bin/pre-commit/fix_include_style.py
@@ -43,22 +48,36 @@ repos:
         language: python
         entry: ./bin/pre-commit/fix_pragma_once.py
         files: \.(h|hpp)$
+      - id: check-doxygen-style
+        name: check Doxygen comment style
+        entry: ./bin/pre-commit/check_doxygen_style.py
+        language: python
+        types_or: [c++, c]
 
   - repo: https://github.com/pre-commit/mirrors-clang-format
-    rev: dd18dad857d6133e90bbe478f4f2f22ec0030269 # frozen: v22.1.5
+    rev: f4d7745e17a28aad7eed2f4874ca8d1568c11c4c # frozen: v22.1.8
     hooks:
       - id: clang-format
         args: [--style=file]
         types_or: [c++, c, proto]
         exclude: ^include/xrpl/protocol_autogen/(transactions|ledger_entries)/
 
+  - repo: local
+    hooks:
+      - id: cargo-fmt
+        name: cargo fmt
+        entry: cargo fmt --manifest-path crates/Cargo.toml --all
+        language: system
+        types: [rust]
+        pass_filenames: false # rustfmt formats the whole workspace
+
   - repo: https://github.com/BlankSpruce/gersemi-pre-commit
     rev: e98930bdc210d3387007f9252d8c1694ea7e410f # frozen: 0.27.7
     hooks:
       - id: gersemi
 
   - repo: https://github.com/rbubley/mirrors-prettier
-    rev: 39e2973981e6d2f9b6c543b0086a2d2393abdc89 # frozen: v3.9.4
+    rev: 9337a74165b178ae2c766f60bee7252a0f06f3e8 # frozen: v3.9.5
     hooks:
       - id: prettier
         args: [--end-of-line=auto]

API-CHANGELOG.md

@@ -28,6 +28,9 @@ This section contains changes targeting a future version.
 
 ### Additions
 
+- `account_tx`: Added an optional `delegate` request object to filter delegated transactions. The object requires `delegate_filter`, which must be either `actor` for transactions owned by the requested account but signed by another account, or `authorizer` for transactions signed by the requested account on behalf of another account. The optional `counter_party` account narrows the results to a specific signer/delegate for `actor` or a specific owner/delegator for `authorizer`. Malformed `delegate`, `delegate_filter`, and `counter_party` values return standard invalid field errors, and invalid account IDs return `actMalformed`.
+  When paginating delegate-filtered queries, a marker from a delegate-filtered query includes a `delegate` flag and is only valid for follow-up requests that also supply `delegate` (mixing marker conventions returns `invalidParams`). Because filtering is applied after the ledger scan, a page may contain fewer results than `limit` (possibly zero) while still returning a marker, so callers must continue until no marker is present.
+
 - `ledger_entry`, `account_objects`: The `Delegate` ledger entry now includes an optional `DestinationNode` field, which stores the index into the authorized account's owner directory. This field is present on entries created after bidirectional directory tracking was introduced and may appear in RPC responses for those entries. ([#6681](https://github.com/XRPLF/rippled/pull/6681))
 
 - `server_definitions`: Added the following new sections to the response ([#6321](https://github.com/XRPLF/rippled/pull/6321)):
@@ -39,6 +42,7 @@ This section contains changes targeting a future version.
 
 ### Bugfixes
 
+- `get_aggregate_price`: Duplicate entries in the `oracles` request array are now ignored. [#6586](https://github.com/XRPLF/rippled/pull/6586)
 - Peer Crawler: The `port` field in `overlay.active[]` now consistently returns an integer instead of a string for outbound peers. [#6318](https://github.com/XRPLF/rippled/pull/6318)
 - `ping`: The `ip` field is no longer returned as an empty string for proxied connections without a forwarded-for header. It is now omitted, consistent with the behavior for identified connections. [#6730](https://github.com/XRPLF/rippled/pull/6730)
 - gRPC `GetLedgerDiff`: Fixed error message that incorrectly said "base ledger not validated" when the desired ledger was not validated. [#6730](https://github.com/XRPLF/rippled/pull/6730)

BUILD.md

@@ -42,6 +42,8 @@ Our Linux CI tooling is distro-independent and uses a Nix-based environment, so
 ### macOS
 
 Many `xrpld` engineers use macOS for development.
+The minimum supported version is macOS 15 (Sequoia).
+CI testing is done in macOS 26 (Tahoe), but the build defaults `CMAKE_OSX_DEPLOYMENT_TARGET` to 15.
 
 ### Windows
 
@@ -245,7 +247,17 @@ cmake --build . --target setup_code_gen  # create venv and install dependencies
 cmake --build . --target code_gen        # regenerate code
 ```
 
-The regenerated files should be committed alongside your changes.
+The same targets are also available as a standalone project, which does not
+need the dependencies to be configured first:
+
+```
+cmake -S cmake/codegen -B build/codegen
+cmake --build build/codegen --target setup_code_gen
+cmake --build build/codegen --target code_gen
+```
+
+The regenerated files should be committed alongside your changes. CI verifies
+that they are up-to-date.
 
 ## Coverage report
 

CMakeLists.txt

@@ -13,6 +13,23 @@ if(DEFINED CMAKE_MODULE_PATH)
 endif()
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
 
+# Must be set before project() because project() consumes it when configuring the compiler and SDK.
+# A user-provided -DCMAKE_OSX_DEPLOYMENT_TARGET still takes precedence.
+#
+# CMAKE_SYSTEM_NAME can't be used before project(), so CMAKE_HOST_SYSTEM_NAME is used instead.
+#
+# When CMAKE_OSX_DEPLOYMENT_TARGET is bumped to >=26.0, FastFloat dependency won't be needed anymore
+if(
+    CMAKE_HOST_SYSTEM_NAME STREQUAL "Darwin"
+    AND NOT DEFINED CMAKE_OSX_DEPLOYMENT_TARGET
+)
+    set(CMAKE_OSX_DEPLOYMENT_TARGET
+        "15.0"
+        CACHE STRING
+        "Minimum macOS deployment version"
+    )
+endif()
+
 project(xrpl)
 set(CMAKE_CXX_EXTENSIONS OFF)
 set(CMAKE_CXX_STANDARD 23)
@@ -87,6 +104,7 @@ include(deps/Boost)
 add_subdirectory(external/antithesis-sdk)
 find_package(date REQUIRED)
 find_package(ed25519 REQUIRED)
+find_package(FastFloat REQUIRED)
 find_package(gRPC REQUIRED)
 find_package(LibArchive REQUIRED)
 find_package(lz4 REQUIRED)
@@ -102,6 +120,7 @@ target_link_libraries(
     xrpl_libs
     INTERFACE
         ed25519::ed25519
+        FastFloat::fast_float
         lz4::lz4
         mpt-crypto::mpt-crypto
         OpenSSL::Crypto
@@ -131,17 +150,28 @@ else()
 endif()
 target_link_libraries(xrpl_libs INTERFACE ${nudb})
 
+if(benchmark)
+    find_package(benchmark REQUIRED)
+endif()
+
 if(coverage)
     include(XrplCov)
 endif()
 
+add_subdirectory(crates)
 include(XrplCore)
 include(XrplProtocolAutogen)
 include(XrplInstall)
-include(XrplPackaging)
 include(XrplValidatorKeys)
+# Must come after XrplValidatorKeys: the 'package' target depends on the
+# validator-keys target existing.
+include(XrplPackaging)
 
 if(tests)
     include(CTest)
     add_subdirectory(src/tests/libxrpl)
 endif()
+
+if(benchmark)
+    add_subdirectory(src/benchmarks/libxrpl)
+endif()

CODEOWNERS

@@ -0,0 +1,4 @@
+# By default, anyone can review changes.
+
+# The CI tooling team should review changes to the CI configuration.
+/.github/ @XRPLF/ci-tooling

CONTRIBUTING.md

@@ -83,8 +83,11 @@ If you create new source files, they must be organized as follows:
   `src/libxrpl`.
 - All other non-test files must go under `src/xrpld`.
 - All test source files must go under `src/test`.
+- All benchmark source files must go under `src/benchmarks`.
 
-The source must be formatted according to the style guide below.
+The source must be formatted according to the style guide below. The easiest
+way to satisfy this is to install the [`pre-commit`](#pre-commit-hooks) hooks,
+which format and lint your changes automatically on every commit.
 
 Header includes must be [levelized](.github/scripts/levelization).
 
@@ -212,13 +215,61 @@ This is a non-exhaustive list of recommended style guidelines. These are
 not always strictly enforced and serve as a way to keep the codebase
 coherent rather than a set of _thou shalt not_ commandments.
 
+## Pre-commit hooks
+
+We use the [`pre-commit`](https://pre-commit.com/) framework to run the
+formatting and linting tools that keep the codebase consistent. `pre-commit`
+runs each tool configured in
+[`.pre-commit-config.yaml`](./.pre-commit-config.yaml) in its own isolated
+environment, so you don't need to install most of the individual tools
+yourself. The version of each hook sourced from an external repository
+(`clang-format`, `gersemi`, etc.) is pinned in that file, so running the hooks
+locally uses exactly the same versions as CI. A few `local` hooks — most notably
+`clang-tidy` — run tools from your own environment; see
+[Installing clang-tidy](#installing-clang-tidy) for how to get those.
+
+To get started, install `pre-commit` and enable the git hook scripts:
+
+```bash
+pip install pre-commit
+pre-commit install
+```
+
+Once installed, the hooks run automatically on your staged files every time you
+`git commit`. You can also run them on demand:
+
+```bash
+# Run all hooks against only the staged files
+pre-commit run
+
+# Run all hooks against every file in the repository
+pre-commit run --all-files
+
+# Run a single hook (e.g. clang-format) against all files
+pre-commit run clang-format --all-files
+```
+
+The hooks configured in this repository include, among others:
+
+- `clang-format` — C++/proto formatting (see [Formatting](#formatting))
+- `clang-tidy` — C++ static analysis (see [Clang-tidy](#clang-tidy)); opt in with `TIDY=1`
+- `fix-include-style`, `fix-pragma-once`, `check-doxygen-style` — C++ hygiene
+- `gersemi` — CMake formatting
+- `prettier`, `black`, `shfmt` — formatting for JavaScript/JSON/Markdown, Python, and shell
+- `cspell` — spell checking
+
+The same hooks run in CI on every pull request, so running them locally before
+you push helps you avoid CI failures.
+
 ## Formatting
 
-All code must conform to `clang-format` version 22,
-according to the settings in [`.clang-format`](./.clang-format),
-unless the result would be unreasonably difficult to read or maintain.
-To demarcate lines that should be left as-is, surround them with comments like
-this:
+All code must conform to `clang-format`, according to the settings in
+[`.clang-format`](./.clang-format), unless the result would be unreasonably
+difficult to read or maintain. The `clang-format` version is pinned in
+[`.pre-commit-config.yaml`](./.pre-commit-config.yaml), so the
+[`pre-commit`](#pre-commit-hooks) hook always formats with the same version as
+CI. To demarcate lines that should be left as-is, surround them with comments
+like this:
 
 ```
 // clang-format off
@@ -226,9 +277,21 @@ this:
 // clang-format on
 ```
 
-You can format individual files in place by running `clang-format -i <file>...`
+The easiest way to format your changes is to let the `pre-commit` hook run
+automatically on commit, or to run it manually:
+
+```bash
+pre-commit run clang-format --all-files
+```
+
+You can also format individual files in place by running `clang-format -i <file>...`
 from any directory within this project.
 
+> [!NOTE]
+> This uses whatever `clang-format` version is installed locally, which may
+> differ from the pinned version used by `pre-commit` and CI, so the results
+> can vary.
+
 There is a Continuous Integration job that runs clang-format on pull requests. If the code doesn't comply, a patch file that corrects auto-fixable formatting issues is generated.
 
 To download the patch file:
@@ -239,13 +302,6 @@ To download the patch file:
 4. Download the zip file and extract it to your local git repository. Run `git apply [patch-file-name]`.
 5. Commit and push.
 
-You can install a pre-commit hook to automatically run `clang-format` before every commit:
-
-```
-pip3 install pre-commit
-pre-commit install
-```
-
 ## Clang-tidy
 
 All code must pass `clang-tidy` checks according to the settings in [`.clang-tidy`](./.clang-tidy).
@@ -267,7 +323,7 @@ Before running clang-tidy, you must build the project to generate required files
 
 #### Via pre-commit (recommended)
 
-If you have already installed the pre-commit hooks (see above), you can run clang-tidy on your staged files using:
+If you have already installed the [`pre-commit`](#pre-commit-hooks) hooks, you can run clang-tidy on your staged files using:
 
 ```
 TIDY=1 pre-commit run clang-tidy
@@ -292,12 +348,14 @@ run-clang-tidy -p build -allow-no-checks src tests
 ```
 
 This will check all source files in the `src`, `include` and `tests` directories using the compile commands from your `build` directory.
-If you wish to automatically fix whatever clang-tidy finds _and_ is capable of fixing, add `-fix` to the above command:
+If you wish to automatically fix whatever clang-tidy finds _and_ is capable of fixing, add `-fix -format` to the above command:
 
 ```
-run-clang-tidy -p build -quiet -fix -allow-no-checks src tests
+run-clang-tidy -p build -quiet -fix -format -allow-no-checks src tests
 ```
 
+`-format` reformats the fixed code with [`.clang-format`](./.clang-format); without it the fixes are inserted in LLVM style and the `clang-format` hook rewrites them afterwards.
+
 ## Contracts and instrumentation
 
 We are using [Antithesis](https://antithesis.com/) for continuous fuzzing,

bin/check-tools.sh

@@ -30,8 +30,10 @@ missing=()
 checked=0
 
 # check <name> [probe-command...]
-# Runs the probe (default: "<name> --version") quietly. Records <name> as
-# missing if the command is not found or exits non-zero.
+# Runs the probe (default: "<name> --version"), capturing both stdout and
+# stderr, and prints one aligned line: the status, the name, and the first
+# non-blank line of the probe output (its version). Records <name> as missing
+# if the command is not found or exits non-zero.
 check() {
     local name="$1"
     shift
@@ -40,10 +42,11 @@ check() {
         probe=("${name}" --version)
     fi
 
-    echo "Checking ${name}..."
     checked=$((checked + 1))
-    if "${probe[@]}" | head -n 1; then
-        printf '  [ ok ] %s\n' "${name}"
+    local output version
+    if output="$("${probe[@]}" 2>&1)"; then
+        version="$(printf '%s\n' "${output}" | grep -m1 '[^[:space:]]' || true)"
+        printf '  [ ok ] %-20s %s\n' "${name}" "${version}"
     else
         printf '  [MISS] %s\n' "${name}"
         missing+=("${name}")
@@ -85,12 +88,14 @@ if [ "${os}" = "linux" ] || [ "${os}" = "macos" ]; then
     check file
     check less
     check make
-    check netstat which netstat
+    # net-tools netstat reports "net-tools X.Y"; macOS ships BSD netstat with no
+    # version flag, so fall back to a presence marker there.
+    check netstat sh -c 'command -v netstat >/dev/null && { netstat --version 2>&1 | grep -m1 -oE "net-tools [0-9.]+" || echo present; }'
     check ninja
-    check perl
+    check perl perl -e 'print "$^V\n"'
     check pkg-config
     check vim
-    check zip
+    check zip bash -c 'zip --version 2>&1 | grep -m1 -oE "Zip [0-9.]+"'
 
     # These tools are present in our Linux CI images and in local development
     # setups, but not in the macOS CI environment. So check them everywhere
@@ -110,6 +115,23 @@ if [ "${os}" = "linux" ] || [ "${os}" = "macos" ]; then
     fi
 fi
 
+# Rust toolchain. Part of the Nix commonPackages, so available on both Linux
+# and macOS. The cargo plugins are invoked through cargo (`cargo <sub>`), which
+# resolves the matching `cargo-<sub>` binary on PATH; `--version` is offline and
+# does not need a Cargo project.
+if [ "${os}" = "linux" ] || [ "${os}" = "macos" ]; then
+    echo
+    echo "Rust toolchain:"
+    check cargo
+    check cargo-audit cargo audit --version
+    check cargo-llvm-cov cargo llvm-cov --version
+    check cargo-nextest cargo nextest --version
+    check clippy clippy-driver --version
+    check rust-analyzer
+    check rustc
+    check rustfmt
+fi
+
 # GCC is the default compiler on Linux. macOS uses the system Apple Clang
 # instead, so GCC/g++/gcov are not expected there.
 if [ "${os}" = "linux" ]; then

bin/pre-commit/Dockerfile

@@ -0,0 +1,55 @@
+ARG BASE_IMAGE=ubuntu:26.04
+
+FROM ${BASE_IMAGE}
+
+SHELL ["/bin/bash", "-e", "-o", "pipefail", "-c"]
+ENTRYPOINT ["/bin/bash"]
+
+ARG DEBIAN_FRONTEND=noninteractive
+
+RUN <<EOF
+pkgs=()
+pkgs+=(curl)         # Required to install nix.
+pkgs+=(doxygen)      # Needed for Clio's check-doxygen-docs.sh.
+pkgs+=(git)          # Required for prepare-runner.
+pkgs+=(libatomic1)   # Required to run pre-commit provided `node`.
+pkgs+=(python3)      # Python 3 interpreter.
+pkgs+=(python3-pip)  # Package manager for Python applications.
+pkgs+=(xz-utils)     # Required to install nix
+
+apt-get update
+apt-get install -y --no-install-recommends "${pkgs[@]}"
+apt-get clean
+rm -rf /var/lib/apt/lists/*
+EOF
+
+ARG PRE_COMMIT_VERSION=4.6.0
+RUN pip install --no-cache --break-system-packages \
+    pre-commit==${PRE_COMMIT_VERSION}
+
+RUN sh <(curl --proto '=https' --tlsv1.2 -L https://nixos.org/nix/install) --daemon --yes
+
+# Add nix to PATH and set NIX environment variables,
+# so nix is available in all shells including non-interactive shells (e.g., GitHub Actions).
+ENV PATH="/nix/var/nix/profiles/default/bin:${PATH}"
+ENV NIX_PROFILES="/nix/var/nix/profiles/default"
+ENV NIX_SSL_CERT_FILE="/nix/var/nix/profiles/default/etc/ssl/certs/ca-bundle.crt"
+
+# Verify nix installation
+RUN nix --version
+
+ENV RUSTUP_HOME="/opt/rust/rustup"
+ENV CARGO_HOME="/opt/rust/cargo"
+ENV PATH="/opt/rust/cargo/bin:${PATH}"
+
+WORKDIR /tmp
+COPY rust-toolchain.toml /tmp/rust-toolchain.toml
+RUN <<EOF
+curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs \
+    | sh -s -- -y --no-modify-path --profile minimal --default-toolchain none
+rustup toolchain install
+rustup show
+cargo fmt --version
+EOF
+
+WORKDIR /

bin/pre-commit/check_doxygen_style.py

@@ -0,0 +1,440 @@
+#!/usr/bin/env python3
+"""
+Check C++ Doxygen comment style.
+
+Enforces the house convention for documentation comments:
+
+  * Use ``/** ... */`` blocks, not ``///``, ``//!`` or ``/*! ... */``; a plain
+    ``/* ... */`` that contains Doxygen commands is a doc comment missing its
+    second star. Trailing member-after comments use ``///<`` (not ``//!<``,
+    ``/*!< ... */`` or ``/**< ... */`` -- the block forms get reflowed and
+    mis-attached by clang-format on packed enum values, the line form does not).
+  * ``/**`` sits alone on its line; the closing ``*/`` sits alone on its line.
+  * Every content line is prefixed with `` * `` (no bare-indented continuation).
+  * The first content line is flush (not over-indented).
+  * Doxygen commands use the ``@cmd`` form, not ``\\cmd``.
+  * Use ``@return`` / ``@throws`` rather than prose ``Returns:`` / ``Throws:``.
+  * A plain ``//`` comment carrying a block-level ``@command`` (``@param``,
+    ``@return``, ``@see``, ...) is documentation and must be a ``/** ... */``
+    block (Doxygen ignores ``//``).
+  * Use canonical command spellings: ``@return`` (not ``@returns``),
+    ``@throws`` (not ``@throw``), ``@see`` (not ``@sa``).
+  * Order block tags ``@tparam`` -> ``@param`` -> ``@return``. (Whether
+    ``@param`` order matches the signature is not checked here -- too fragile to
+    parse; Doxygen's WARN_IF_DOC_ERROR covers name mismatches.)
+  * One-liners are expanded to three lines, EXCEPT bare markers ``@{`` / ``@}``
+    / ``@cond [label]`` / ``@endcond`` / ``@file [name]`` which stay on one line.
+
+Left intentionally alone (recognized, valid Doxygen that is not this style's
+concern):
+
+  * ``///<`` trailing "member-after" comments (the house form).
+  * Divider lines made only of slashes (``//////////``).
+  * Plain ``/* ... */`` (non-Doxygen) comments.
+
+Usage:
+    check_doxygen_style.py [FILE ...]      # explicit files
+    check_doxygen_style.py                 # default: src/ and include/ trees
+
+Exit status is non-zero if any violation is found.
+"""
+
+import argparse
+import re
+import sys
+from collections.abc import Iterable, Iterator
+from dataclasses import dataclass
+from enum import Enum
+from pathlib import Path
+
+
+class Category(Enum):
+    """A kind of style violation: a printed ``label`` and its ``description``.
+
+    The description is the default message; a few categories whose wording
+    depends on the offending text (see ``Finding.detail``) override it.
+    """
+
+    def __init__(self, label: str, description: str) -> None:
+        self.label = label
+        self.description = description
+
+    BACKSLASH_COMMAND = ("backslash-command", "use the @cmd form, not \\cmd")
+    WRONG_COMMAND = ("wrong-command", "use the canonical command spelling")
+    TRIPLE_SLASH = ("triple-slash", "use a /** ... */ block instead of ///")
+    QT_MEMBER = ("qt-member", "use ///< instead of //!<")
+    QT_LINE = ("qt-line", "use a /** ... */ block instead of //!")
+    BLOCK_MEMBER = ("block-member", "use ///< instead of /**<")
+    QT_BLOCK_MEMBER = ("qt-block-member", "use ///< instead of /*!<")
+    DOC_IN_LINE_COMMENT = (
+        "doc-in-line-comment",
+        "use a /** ... */ block for documentation, not //",
+    )
+    QT_COMMENT = ("qt-comment", "use /** instead of /*!")
+    SINGLE_LINE_BLOCK = (
+        "single-line-block",
+        "expand one-line /** ... */ to a multi-line block "
+        "(markers @{ @} @cond @endcond @file may stay)",
+    )
+    TEXT_ON_OPENER = ("text-on-opener", "move text off the /** opener line")
+    BARE_CONTINUATION = ("bare-continuation", 'prefix continuation lines with " * "')
+    OVER_INDENTED = ("over-indented", "first content line is over-indented")
+    OVER_INDENTED_TAG = (
+        "over-indented-tag",
+        'Doxygen tag over-indented; use a single space after "*"',
+    )
+    COMBINED_MARKER = (
+        "combined-marker",
+        "scope marker @{ / @} should be its own single-line /** @{ */ block",
+    )
+    PROSE_LABEL = ("prose-label", "use a Doxygen tag instead of a prose label")
+    CONTENT_ON_CLOSER = ("content-on-closer", "move content off the closing */ line")
+    PLAIN_BLOCK_DOC = (
+        "plain-block-doc",
+        "documentation comment must open with /** not /*",
+    )
+    TAG_ORDER = (
+        "tag-order",
+        "block tags out of order; expected @tparam, then @param, then @return",
+    )
+
+
+@dataclass(frozen=True)
+class Finding:
+    """A single style violation at a 1-based line number.
+
+    ``detail`` overrides the category's default description when the message
+    depends on the offending text (e.g. which command was misspelled).
+    """
+
+    line: int
+    category: Category
+    detail: str | None = None
+
+    @property
+    def message(self) -> str:
+        return self.detail if self.detail is not None else self.category.description
+
+
+DEFAULT_ROOTS = ("src", "include")
+EXTS = {".h", ".hpp", ".cpp", ".ipp", ".cxx", ".cc"}
+
+# Every Doxygen command we recognize when written with a backslash (\cmd).
+_ALL_COMMANDS = (
+    "brief|param|tparam|return|returns|retval|note|warning|pre|post|see|sa|ref|"
+    "throw|throws|exception|deprecated|details|code|endcode|verbatim|endverbatim|"
+    "li|arg|c|internal|since|todo|attention|remark|remarks|ingroup|defgroup"
+)
+# Block-level tags whose over-indentation we flag inside a block body.
+_BLOCK_TAGS = (
+    "param|tparam|returns?|retval|brief|throws?|note|warning|"
+    "pre|post|see|sa|details|deprecated"
+)
+# Tags that, appearing anywhere in a comment, mark it as documentation.
+_ANY_DOC_TAGS = (
+    "param|tparam|returns?|retval|brief|throws?|note|warning|pre|post|see|sa"
+)
+# Tags that make a plain // comment a mis-styled doc comment.
+_LINE_DOC_TAGS = "brief|param|tparam|returns?|retval|throws?|note|see|pre|post"
+
+# \cmd that should be @cmd.
+RE_BACKSLASH_CMD = re.compile(r"\\(" + _ALL_COMMANDS + r")\b")
+# Bare markers that may legitimately stay on a single line.
+RE_MARKER = re.compile(r"^@(\{|\}|cond(\s.*)?|endcond|file(\s.*)?)$")
+# Prose section labels that should be Doxygen tags.
+RE_PROSE_LABEL = re.compile(r"^\*\s(Returns|Throws|Exceptions):\s*$")
+# An over-indented block tag: "*" followed by 2+ spaces then the tag.
+RE_OVERINDENTED_TAG = re.compile(r"^\*\s{2,}@(" + _BLOCK_TAGS + r")\b")
+# Any documentation tag (used to spot a doc comment hiding in a plain /* */).
+RE_ANY_DOC_TAG = re.compile(r"@(" + _ANY_DOC_TAGS + r")\b")
+# A documentation tag inside a // comment.
+RE_LINE_DOC_TAG = re.compile(r"@(" + _LINE_DOC_TAGS + r")\b")
+# Order-relevant tags, for the @tparam -> @param -> @return ordering check.
+RE_ORDER_TAG = re.compile(r"^\*\s*@(param|tparam|returns?|retval)\b")
+# First content line indented by 2+ spaces after the "*".
+RE_FIRST_OVERINDENT = re.compile(r"^\s*\*\s{2,}\S")
+# A scope marker @{ / @} sharing a comment with other text.
+RE_COMBINED_MARKER = re.compile(r"^\*\s*@[{}]\s*$")
+
+# Non-canonical command spellings -> the house spelling (bare command names).
+# Used both to flag a wrong @form and to suggest the right @form for a \wrong.
+CANONICAL_COMMAND = {"returns": "return", "throw": "throws", "sa": "see"}
+WRONG_SPELLINGS = [
+    (re.compile(rf"@{wrong}\b"), f"@{right}")
+    for wrong, right in CANONICAL_COMMAND.items()
+]
+
+# Order block tags should appear in; a body out of this order is a violation.
+EXPECTED_TAG_ORDER = ("tparam", "param", "return")
+
+
+def is_doxy_open(stripped: str) -> bool:
+    """True for a line-start Doxygen block opener we should normalize."""
+    if stripped.startswith("/*!"):  # Qt-style Doxygen
+        return not stripped.startswith("/*!<")  # member-after, leave inline
+    return (
+        stripped.startswith("/**")
+        and not stripped.startswith("/***")
+        and not stripped.startswith("/**/")
+        and not stripped.startswith("/**<")
+    )
+
+
+def _flag_commands(raw_line: str, stripped: str, index: int) -> list[Finding]:
+    """Flag \\cmd and misspelled @cmd on a comment line (opener, body, or closer)."""
+    if not stripped.startswith(("*", "//", "/*")):
+        return []
+    findings: list[Finding] = []
+    backslash = RE_BACKSLASH_CMD.search(raw_line)
+    if backslash:
+        command = backslash.group(1)
+        canonical = CANONICAL_COMMAND.get(command, command)
+        findings.append(
+            Finding(
+                index + 1,
+                Category.BACKSLASH_COMMAND,
+                f"use @{canonical} instead of \\{command}",
+            )
+        )
+    for pattern, replacement in WRONG_SPELLINGS:
+        wrong = pattern.search(raw_line)
+        if wrong:
+            findings.append(
+                Finding(
+                    index + 1,
+                    Category.WRONG_COMMAND,
+                    f"use {replacement} instead of {wrong.group(0)}",
+                )
+            )
+    return findings
+
+
+def _flag_line_comment(raw_line: str, stripped: str, index: int) -> Finding | None:
+    """Return the finding for a single-line comment form (///, //!, /**<, ...), else None."""
+    if stripped.startswith("///") and not stripped.startswith(("////", "///<")):
+        return Finding(index + 1, Category.TRIPLE_SLASH)
+    if "//!<" in raw_line:
+        return Finding(index + 1, Category.QT_MEMBER)
+    if stripped.startswith("//!"):
+        return Finding(index + 1, Category.QT_LINE)
+    if "/**<" in raw_line:
+        return Finding(index + 1, Category.BLOCK_MEMBER)
+    if "/*!<" in raw_line:
+        return Finding(index + 1, Category.QT_BLOCK_MEMBER)
+    if stripped.startswith("//") and RE_LINE_DOC_TAG.search(stripped):
+        return Finding(index + 1, Category.DOC_IN_LINE_COMMENT)
+    return None
+
+
+def _flag_single_line_block(stripped: str, line_no: int, is_qt: bool) -> list[Finding]:
+    """Findings for a whole /** ... */ or /*! ... */ block on one line."""
+    inner = re.sub(r"^/\*[*!]", "", stripped)
+    inner = re.sub(r"\*/\s*$", "", inner).strip()
+    findings: list[Finding] = []
+    if is_qt:
+        findings.append(Finding(line_no, Category.QT_COMMENT))
+    if inner and not RE_MARKER.match(inner):
+        findings.append(Finding(line_no, Category.SINGLE_LINE_BLOCK))
+    return findings
+
+
+def _canonical_order_tag(body: str) -> str | None:
+    """The order-relevant tag (tparam/param/return) a body line opens with, if any."""
+    match = RE_ORDER_TAG.match(body)
+    if match is None:
+        return None
+    command = match.group(1)
+    return "return" if command in ("return", "returns", "retval") else command
+
+
+def _flag_body_line(
+    body_line: str, line_no: int, is_first_content: bool
+) -> list[Finding]:
+    """Findings for one interior line of a multi-line block."""
+    body = body_line.strip()
+    findings: list[Finding] = []
+    if body and not body.startswith("*"):
+        findings.append(Finding(line_no, Category.BARE_CONTINUATION))
+    if body.startswith("*"):
+        if is_first_content and RE_FIRST_OVERINDENT.match(body_line):
+            findings.append(Finding(line_no, Category.OVER_INDENTED))
+        if RE_OVERINDENTED_TAG.match(body):
+            findings.append(Finding(line_no, Category.OVER_INDENTED_TAG))
+        if RE_COMBINED_MARKER.match(body):
+            findings.append(Finding(line_no, Category.COMBINED_MARKER))
+        label = RE_PROSE_LABEL.match(body)
+        if label:
+            suggested_tag = "@return" if label.group(1) == "Returns" else "@throws"
+            findings.append(
+                Finding(
+                    line_no,
+                    Category.PROSE_LABEL,
+                    f'use {suggested_tag} instead of prose "{label.group(1)}:"',
+                )
+            )
+    return findings
+
+
+def _flag_closer(closer_line: str, line_no: int) -> list[Finding]:
+    """Findings for content sharing the closing */ line."""
+    before = closer_line[: closer_line.index("*/")].strip()
+    if before and before != "*":
+        return [Finding(line_no, Category.CONTENT_ON_CLOSER)]
+    return []
+
+
+def _flag_tag_order(first_tag_line: dict[str, int]) -> list[Finding]:
+    """One finding if the present block tags are not in EXPECTED_TAG_ORDER."""
+    tag_lines = [
+        first_tag_line[tag] for tag in EXPECTED_TAG_ORDER if tag in first_tag_line
+    ]
+    if tag_lines != sorted(tag_lines):
+        return [Finding(min(tag_lines), Category.TAG_ORDER)]
+    return []
+
+
+def _flag_doxy_block(lines: list[str], start: int) -> tuple[int, list[Finding]]:
+    """Handle a /** or /*! block opening at ``start``; return (next index, findings)."""
+    raw_line = lines[start]
+    stripped = raw_line.lstrip()
+    open_pos = raw_line.index("/*")
+    is_qt = stripped.startswith("/*!")
+
+    # A whole block on one line: /** ... */.
+    if "*/" in raw_line[open_pos + 2 :]:
+        return start + 1, _flag_single_line_block(stripped, start + 1, is_qt)
+
+    # Multi-line block: opener, then scan the body to the closer.
+    findings: list[Finding] = []
+    if is_qt:
+        findings.append(Finding(start + 1, Category.QT_COMMENT))
+    if raw_line[open_pos + 3 :].strip():
+        findings.append(Finding(start + 1, Category.TEXT_ON_OPENER))
+
+    line_count = len(lines)
+    cursor = start + 1
+    is_first_content = True
+    first_tag_line: dict[str, int] = {}  # canonical tag -> 1-based first line
+    while cursor < line_count and "*/" not in lines[cursor]:
+        body_line = lines[cursor]
+        body = body_line.strip()
+        findings.extend(_flag_commands(body_line, body, cursor))
+        tag = _canonical_order_tag(body)
+        if tag is not None:
+            first_tag_line.setdefault(tag, cursor + 1)
+        findings.extend(_flag_body_line(body_line, cursor + 1, is_first_content))
+        if body.startswith("*"):
+            is_first_content = False
+        cursor += 1
+
+    if cursor < line_count:
+        closer_line = lines[cursor]
+        findings.extend(_flag_commands(closer_line, closer_line.strip(), cursor))
+        findings.extend(_flag_closer(closer_line, cursor + 1))
+    findings.extend(_flag_tag_order(first_tag_line))
+
+    return cursor + 1, findings
+
+
+def _flag_plain_block(lines: list[str], start: int) -> tuple[int, list[Finding]]:
+    """Handle a line-start plain /* ... */ block; return (next index, findings).
+
+    Only flagged when it hides a documentation command (a missing second star).
+    """
+    line_count = len(lines)
+    cursor = start
+    while cursor < line_count and "*/" not in lines[cursor]:
+        cursor += 1
+    findings: list[Finding] = []
+    # The opener (start) is command-checked by check_file; check the rest here.
+    for i in range(start + 1, min(cursor + 1, line_count)):
+        findings.extend(_flag_commands(lines[i], lines[i].strip(), i))
+    block_text = "\n".join(
+        lines[start : cursor + 1] if cursor < line_count else lines[start:]
+    )
+    if RE_ANY_DOC_TAG.search(block_text):
+        findings.append(Finding(start + 1, Category.PLAIN_BLOCK_DOC))
+    next_index = cursor + 1 if cursor < line_count else line_count
+    return next_index, findings
+
+
+def check_source(text: str) -> list[Finding]:
+    """Return all style violations found in the given source text."""
+    lines = text.split("\n")
+    findings: list[Finding] = []
+    line_count = len(lines)
+    index = 0
+    in_plain_block = False  # inside a mid-line, non-Doxygen /* ... */
+    while index < line_count:
+        raw_line = lines[index]
+        stripped = raw_line.lstrip()
+
+        # Skip the interior of a plain block opened on an earlier line.
+        if in_plain_block:
+            in_plain_block = "*/" not in raw_line
+            index += 1
+            continue
+
+        findings.extend(_flag_commands(raw_line, stripped, index))
+
+        line_finding = _flag_line_comment(raw_line, stripped, index)
+        if line_finding is not None:
+            findings.append(line_finding)
+            index += 1
+        elif is_doxy_open(stripped):
+            index, block_findings = _flag_doxy_block(lines, index)
+            findings.extend(block_findings)
+        elif stripped.startswith("/*"):
+            index, block_findings = _flag_plain_block(lines, index)
+            findings.extend(block_findings)
+        else:
+            # A /* that opens mid-line without closing starts a plain block.
+            if "/*" in raw_line and not stripped.startswith("//"):
+                if "*/" not in raw_line[raw_line.index("/*") + 2 :]:
+                    in_plain_block = True
+            index += 1
+    return findings
+
+
+def check_file(path: Path) -> list[Finding]:
+    """Return all style violations found in one file."""
+    return check_source(path.read_text(encoding="utf-8"))
+
+
+def iter_files(paths: Iterable[str]) -> Iterator[Path]:
+    """Yield every C++ source file among the given files and directories."""
+    for raw_path in paths:
+        path = Path(raw_path)
+        if path.is_dir():
+            for candidate in path.rglob("*"):
+                if candidate.is_file() and candidate.suffix in EXTS:
+                    yield candidate
+        elif path.suffix in EXTS:
+            yield path
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Check Doxygen comment style.")
+    parser.add_argument(
+        "files", nargs="*", help="files or directories (default: src/ include/)"
+    )
+    parser.add_argument(
+        "-q", "--quiet", action="store_true", help="only print the summary count"
+    )
+    args = parser.parse_args()
+    roots = args.files or [root for root in DEFAULT_ROOTS if Path(root).is_dir()]
+
+    total = 0
+    for path in sorted(set(iter_files(roots)), key=str):
+        for finding in check_file(path):
+            total += 1
+            if not args.quiet:
+                print(
+                    f"{path}:{finding.line}: {finding.category.label}: {finding.message}"
+                )
+    print(f"\n{total} doxygen-style violation(s)", file=sys.stderr)
+    return 1 if total else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

bin/pre-commit/clang_tidy_check.py

@@ -1,27 +1,46 @@
 #!/usr/bin/env python3
-"""Pre-commit hook that runs clang-tidy on changed files using run-clang-tidy.
+"""Pre-commit hook that runs clang-tidy on staged files using run-clang-tidy.
 
-The set of files is chosen by pre-commit (see .pre-commit-config.yaml), which
-filters to C/C++ sources and excludes `.ipp` fragments. Headers are linted
-directly: the `verify_headers` build option (ON by default) compiles every
-`.h`/`.hpp` on its own, so each header is the main file of its own
-compile_commands.json entry and run-clang-tidy can analyse it just like a
-`.cpp`.
+The script determines the staged files itself (see `pass_filenames: false` in
+.pre-commit-config.yaml) so run-clang-tidy is run once and handles parallelism
+internally: pre-commit would otherwise split the files across parallel hook
+invocations that race when fixes edit a shared header.
+
+Fixes are collected with `-export-fixes` and applied by clang-apply-replacements
+in a separate step rather than with run-clang-tidy's `-fix`. The `add_module`
+build isolates each module's headers behind a per-module symlink directory
+(build/modules/<module>/...), so a header reachable from several translation
+units is referenced through different paths that all resolve to the same source
+file. clang-apply-replacements deduplicates identical replacements by their
+literal path, so those paths must be canonicalised to the real source path
+first; otherwise the same fix is applied once per path and corrupts the header.
 """
 
 from __future__ import annotations
 
 import os
+import re
 import shutil
 import subprocess
 import sys
+import tempfile
 from pathlib import Path
 
 CLANG_TIDY_VERSION = 22
 
+# Extensions run-clang-tidy can analyse: `.cpp` translation units and, thanks to
+# the `verify_headers` build option, `.h`/`.hpp` headers (each has its own
+# compile_commands.json entry). `.ipp` fragments have no entry and are skipped.
+TIDY_EXTENSIONS = {".cpp", ".h", ".hpp"}
 
-def find_run_clang_tidy() -> str | None:
-    for candidate in (f"run-clang-tidy-{CLANG_TIDY_VERSION}", "run-clang-tidy"):
+# A single-quoted `FilePath:` entry in an -export-fixes YAML file, allowing the
+# `- ` marker that precedes it inside a `Replacements:` sequence. clang-tidy
+# emits paths single-quoted and doubles any embedded quote per YAML rules.
+FILEPATH_RE = re.compile(r"^(\s*(?:-\s+)?FilePath:\s*)'((?:[^']|'')*)'\s*$")
+
+
+def find_tool(name: str) -> str | None:
+    for candidate in (f"{name}-{CLANG_TIDY_VERSION}", name):
         if path := shutil.which(candidate):
             return path
     return None
@@ -35,23 +54,43 @@ def find_build_dir(repo_root: Path) -> Path | None:
     return None
 
 
+def staged_files(repo_root: Path) -> list[Path]:
+    """Return absolute paths of staged, lint-able C/C++ files.
+
+    `--diff-filter=d` excludes deletions so we never lint a removed file.
+    """
+    output = subprocess.check_output(
+        ["git", "diff", "--staged", "--name-only", "--diff-filter=d", "--"]
+        + [f"*{ext}" for ext in TIDY_EXTENSIONS],
+        text=True,
+        cwd=repo_root,
+    )
+    return [repo_root / rel for rel in output.splitlines() if rel]
+
+
+def canonicalize_fix_paths(fixes_dir: Path) -> None:
+    """Rewrite every `FilePath` in the exported fixes to its real source path.
+
+    A header included through a module's isolation symlink is recorded under that
+    symlink's path; collapsing all paths to the same real file lets
+    clang-apply-replacements recognise the per-translation-unit duplicates and
+    apply each fix once.
+    """
+    for yaml in fixes_dir.glob("*.yaml"):
+        lines = []
+        for line in yaml.read_text().splitlines():
+            if m := FILEPATH_RE.match(line):
+                path = m.group(2).replace("''", "'")
+                real = os.path.realpath(path).replace("'", "''")
+                line = f"{m.group(1)}'{real}'"
+            lines.append(line)
+        yaml.write_text("\n".join(lines) + "\n")
+
+
 def main():
     if not os.environ.get("TIDY"):
         return 0
 
-    files = sys.argv[1:]
-    if not files:
-        return 0
-
-    run_clang_tidy = find_run_clang_tidy()
-    if not run_clang_tidy:
-        print(
-            f"clang-tidy check failed: TIDY is enabled but neither "
-            f"'run-clang-tidy-{CLANG_TIDY_VERSION}' nor 'run-clang-tidy' was found in PATH.",
-            file=sys.stderr,
-        )
-        return 1
-
     repo_root = Path(
         subprocess.check_output(
             ["git", "rev-parse", "--show-toplevel"],
@@ -59,6 +98,29 @@ def main():
             text=True,
         ).strip()
     )
+
+    files = staged_files(repo_root)
+    if not files:
+        return 0
+
+    run_clang_tidy = find_tool("run-clang-tidy")
+    clang_apply_replacements = find_tool("clang-apply-replacements")
+    missing = [
+        name
+        for name, path in (
+            ("run-clang-tidy", run_clang_tidy),
+            ("clang-apply-replacements", clang_apply_replacements),
+        )
+        if not path
+    ]
+    if missing:
+        print(
+            f"clang-tidy check failed: TIDY is enabled but {' and '.join(missing)} "
+            f"was not found in PATH (tried the '-{CLANG_TIDY_VERSION}' suffix too).",
+            file=sys.stderr,
+        )
+        return 1
+
     build_dir = find_build_dir(repo_root)
     if not build_dir:
         print(
@@ -68,11 +130,27 @@ def main():
         )
         return 1
 
-    result = subprocess.run(
-        [run_clang_tidy, "-quiet", "-p", str(build_dir), "-fix", "-allow-no-checks"]
-        + files
-    )
-    return result.returncode
+    with tempfile.TemporaryDirectory() as fixes_dir:
+        result = subprocess.run(
+            [
+                run_clang_tidy,
+                "-quiet",
+                "-p",
+                build_dir,
+                "-export-fixes",
+                fixes_dir,
+                "-allow-no-checks",
+            ]
+            + files
+        )
+        canonicalize_fix_paths(Path(fixes_dir))
+        # `FormatStyle` in .clang-tidy does not reach this path,
+        # so ask for the repository style here.
+        applied = subprocess.run(
+            [clang_apply_replacements, "--format", "--style=file", fixes_dir]
+        )
+
+    return result.returncode or applied.returncode
 
 
 if __name__ == "__main__":

bin/pre-commit/test_check_doxygen_style.py

@@ -0,0 +1,406 @@
+#!/usr/bin/env python3
+"""
+Tests for check_doxygen_style.py.
+
+Run directly (no test framework needed):
+    ./bin/pre-commit/test_check_doxygen_style.py
+or under pytest:
+    pytest bin/pre-commit/test_check_doxygen_style.py
+"""
+
+import sys
+import textwrap
+
+from check_doxygen_style import Finding, check_source
+
+
+def findings_for(text: str) -> list[Finding]:
+    """Return the style violations for the given source text.
+
+    The text is dedented and its leading newline stripped, so fixtures can be
+    written as indented triple-quoted here-docs while keeping honest 1-based
+    line numbers.
+    """
+    text = textwrap.dedent(text).lstrip("\n")
+    return check_source(text)
+
+
+def labels_for(text: str) -> list[str]:
+    return [f.category.label for f in findings_for(text)]
+
+
+def messages_for(text: str) -> list[str]:
+    return [f.message for f in findings_for(text)]
+
+
+# --- well-formed input produces nothing -------------------------------------
+
+
+def test_clean_block_ok() -> None:
+    code = """
+    /**
+     * Brief.
+     *
+     * @tparam T a type
+     * @param x the x
+     * @return the result
+     */
+    """
+    assert findings_for(code) == []
+
+
+def test_blank_lines_inside_block_ok() -> None:
+    code = """
+    /**
+     * a
+     *
+     * b
+     */
+    """
+    assert findings_for(code) == []
+
+
+def test_member_and_divider_allowed() -> None:
+    assert findings_for("int x;  ///< ok member\n") == []
+    assert findings_for("//////////\n") == []
+    assert findings_for("//// text\n") == []
+
+
+# --- line-comment forms ------------------------------------------------------
+
+
+def test_triple_slash() -> None:
+    code = "/// doc\n"
+    assert labels_for(code) == ["triple-slash"]
+
+
+def test_qt_line() -> None:
+    code = "//! doc\n"
+    assert labels_for(code) == ["qt-line"]
+
+
+def test_qt_member() -> None:
+    code = "int x;  //!< doc\n"
+    assert labels_for(code) == ["qt-member"]
+
+
+def test_block_member() -> None:
+    code = "int x;  /**< doc */\n"
+    assert labels_for(code) == ["block-member"]
+
+
+def test_qt_block_member() -> None:
+    code = "int x;  /*!< doc */\n"
+    assert labels_for(code) == ["qt-block-member"]
+
+
+def test_doc_in_line_comment() -> None:
+    code = "// @param x\n"
+    assert labels_for(code) == ["doc-in-line-comment"]
+
+
+# --- block forms -------------------------------------------------------------
+
+
+def test_qt_comment() -> None:
+    code = """
+    /*!
+     * brief
+     */
+    """
+    assert labels_for(code) == ["qt-comment"]
+
+
+def test_qt_comment_single_line() -> None:
+    # /*! ... */ on one line -> qt-comment (plus single-line-block)
+    code = "/*! brief */\n"
+    assert labels_for(code) == ["qt-comment", "single-line-block"]
+
+
+def test_single_line_block() -> None:
+    code = "/** brief */\n"
+    assert labels_for(code) == ["single-line-block"]
+
+
+def test_single_line_markers_allowed() -> None:
+    for marker in ("@{", "@}", "@cond LABEL", "@endcond", "@file foo.h"):
+        code = f"/** {marker} */\n"
+        assert findings_for(code) == [], marker
+
+
+def test_text_on_opener() -> None:
+    code = """
+    /** text here
+     * more
+     */
+    """
+    assert labels_for(code) == ["text-on-opener"]
+
+
+def test_bare_continuation() -> None:
+    code = """
+    /**
+     * a
+       bare line
+     */
+    """
+    assert labels_for(code) == ["bare-continuation"]
+
+
+def test_over_indented_first_line() -> None:
+    code = """
+    /**
+     *   over
+     */
+    """
+    assert labels_for(code) == ["over-indented"]
+
+
+def test_over_indented_tag() -> None:
+    # a flush first line consumes "first content", isolating the tag check
+    code = """
+    /**
+     * brief
+     *   @param x
+     */
+    """
+    assert labels_for(code) == ["over-indented-tag"]
+
+
+def test_combined_marker() -> None:
+    code = """
+    /**
+     * @{
+     */
+    """
+    assert labels_for(code) == ["combined-marker"]
+
+
+def test_prose_label() -> None:
+    for word in ("Returns", "Throws", "Exceptions"):
+        code = f"""
+        /**
+         * {word}:
+         */
+        """
+        assert labels_for(code) == ["prose-label"], word
+
+
+def test_content_on_closer() -> None:
+    code = """
+    /**
+     * a
+     * b */
+    """
+    assert labels_for(code) == ["content-on-closer"]
+
+
+def test_plain_block_doc() -> None:
+    assert labels_for("/* @param x */\n") == ["plain-block-doc"]
+    assert findings_for("/* just an ordinary note */\n") == []
+
+
+def test_tag_order() -> None:
+    out_of_order = """
+    /**
+     * @param x
+     * @tparam T
+     */
+    """
+    assert labels_for(out_of_order) == ["tag-order"]
+
+    correct = """
+    /**
+     * @tparam T
+     * @param x
+     * @return r
+     */
+    """
+    assert findings_for(correct) == []
+
+    single = """
+    /**
+     * @param x
+     */
+    """
+    assert findings_for(single) == []  # single tag: never out of order
+
+
+# --- command spelling (must work on body/closer lines, not just the opener) --
+
+
+def test_backslash_command_on_body_line() -> None:
+    code = r"""
+    /**
+     * \brief x
+     */
+    """
+    assert labels_for(code) == ["backslash-command"]
+
+
+def test_backslash_command_suggests_canonical_spelling() -> None:
+    # a backslash + non-canonical spelling is fixed in one pass, not two:
+    # \sa -> @see (not @sa), \returns -> @return (not @returns)
+    sa = r"""
+    /**
+     * \sa other
+     */
+    """
+    assert messages_for(sa) == [r"use @see instead of \sa"]
+
+    returns = r"""
+    /**
+     * \returns x
+     */
+    """
+    assert messages_for(returns) == [r"use @return instead of \returns"]
+
+
+def test_wrong_command_on_body_line() -> None:
+    code = """
+    /**
+     * @returns x
+     */
+    """
+    assert labels_for(code) == ["wrong-command"]
+
+
+def test_body_line_commands_regression() -> None:
+    # regression: these live on body lines of a multi-line block
+    code = r"""
+    /**
+     * @returns bad
+     * @throw ex
+     * @sa other
+     * \param y
+     */
+    """
+    assert labels_for(code) == [
+        "wrong-command",
+        "wrong-command",
+        "wrong-command",
+        "backslash-command",
+    ]
+
+
+def test_command_on_closer_line() -> None:
+    code = """
+    /**
+     * a
+     * @sa b */
+    """
+    assert labels_for(code) == ["wrong-command", "content-on-closer"]
+
+
+def test_no_double_count_across_opener_body_closer() -> None:
+    code = """
+    /** @returns opener
+     * @throw body
+     * @sa closer */
+    """
+    assert labels_for(code).count("wrong-command") == 3
+
+
+def test_code_with_word_allowed() -> None:
+    # @code{.cpp} is valid Doxygen and must not be flagged
+    code = """
+    /**
+     * @code{.cpp}
+     * int x;
+     * @endcode
+     */
+    """
+    assert findings_for(code) == []
+
+
+# --- rendered message text ---------------------------------------------------
+
+
+def test_message_uses_category_description() -> None:
+    # a static category renders its default description
+    code = "/// doc\n"
+    assert messages_for(code) == ["use a /** ... */ block instead of ///"]
+
+
+def test_message_detail_overrides() -> None:
+    # dynamic categories render the offending text via Finding.detail
+    backslash = r"""
+    /**
+     * \param y
+     */
+    """
+    assert messages_for(backslash) == [r"use @param instead of \param"]
+
+    wrong = """
+    /**
+     * @returns x
+     */
+    """
+    assert messages_for(wrong) == ["use @return instead of @returns"]
+
+    prose = """
+    /**
+     * Throws:
+     */
+    """
+    assert messages_for(prose) == ['use @throws instead of prose "Throws:"']
+
+
+# --- robustness --------------------------------------------------------------
+
+
+def test_empty_file_no_crash() -> None:
+    assert findings_for("") == []
+
+
+def test_mid_line_plain_block_skipped() -> None:
+    # a /* opened mid-line (after code) and spanning lines is skipped, so its
+    # comment-like contents are not analyzed
+    code = """
+    int x = 0;  /* note: @returns is not a real tag here
+     * @param also not real
+     */
+    int y = 0;
+    """
+    assert findings_for(code) == []
+
+
+def test_unclosed_block_scanned_to_eof() -> None:
+    # an unterminated /** block is still scanned to EOF (no crash, body checked)
+    code = """
+    /**
+     * @returns x
+    """
+    assert labels_for(code) == ["wrong-command"]
+
+
+def test_banner_and_empty_comment_not_flagged() -> None:
+    code = """
+    /***
+     * banner
+     ***/
+    """
+    assert findings_for(code) == []
+    assert findings_for("/**/\n") == []
+
+
+def main() -> int:
+    tests = sorted(
+        (name, fn)
+        for name, fn in globals().items()
+        if name.startswith("test_") and callable(fn)
+    )
+    failed = 0
+    for name, fn in tests:
+        try:
+            fn()
+            print(f"PASS {name}")
+        except AssertionError as exc:
+            failed += 1
+            print(f"FAIL {name}: {exc!r}")
+    print(f"\n{len(tests) - failed}/{len(tests)} passed")
+    return 1 if failed else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

cfg/xrpld-example.cfg

@@ -488,6 +488,17 @@
 #   Must be a number between 100 and 1000, defaults to 250
 #
 #
+# [max_subscriptions_per_connection]
+#
+#   Maximum number of account, real-time account, and account-history
+#   subscriptions a single client connection may hold at once. Bounds the
+#   per-connection state torn down when the connection disconnects. Book
+#   subscriptions are tracked separately and are not counted here.
+#
+#   Defaults to 100000 if not set; large enough for legitimate power users
+#   such as block explorers.
+#
+#
 # [overlay]
 #
 #   Controls settings related to the peer to peer overlay.
@@ -538,6 +549,45 @@
 #       only be used for local testing and debugging. Do not disable
 #       on mainnet.
 #
+#   max_untrusted_count = <number>
+#
+#       The number of manifests the server keeps for validators it does not
+#       list, and the number it sends and processes in a single peer protocol
+#       message. Once the server holds this many, a manifest for a new
+#       unlisted validator is rejected, so peer gossip cannot grow the cache
+#       without end.
+#
+#       This option can take any value between 50 and 1000, inclusive. If
+#       the option is not present the server uses its built-in value.
+#
+#       The current default (which is subject to change) is 300.
+#
+#   max_trusted_count = <number>
+#
+#       The number of manifests for listed validators to allow for when
+#       sizing peer protocol messages. Manifests for listed validators are
+#       never dropped, whether sending or receiving, because doing so would
+#       delay a validator key change reaching this server. Set this above the
+#       number of validators the server lists.
+#
+#       Together the two counts above set the largest manifest message the
+#       server accepts: bigger messages are discarded without reading them,
+#       and without penalising the sender. Raising either means the server
+#       accepts and sends bigger messages than a peer using the defaults, and
+#       those peers will discard what this server sends. Lowering either below
+#       what peers send makes this server discard their manifest messages,
+#       which it does without recording anything.
+#
+#       This option can take any value between 50 and 1000, inclusive. If
+#       the option is not present the server uses its built-in value.
+#
+#       The current default (which is subject to change) is 300.
+#
+#       NOTE: These two options (max_untrusted_count and max_trusted_count)
+#       are transitional. They exist to bound manifest-message size and cache
+#       growth during the network upgrade. They may be removed in a future
+#       release once the fleet has upgraded, and should not be relied upon as
+#       stable configuration.
 #
 # [transaction_queue] EXPERIMENTAL
 #

cmake/CompilationEnv.cmake

@@ -29,6 +29,27 @@ if(CMAKE_GENERATOR STREQUAL "Xcode")
     set(is_xcode TRUE)
 endif()
 
+# --------------------------------------------------------------------
+# Nix toolchain detection
+# --------------------------------------------------------------------
+# True when the C++ compiler resolves into the Nix store. CMAKE_CXX_COMPILER may
+# be referenced through a symlink outside the store (a Nix profile, a /usr/bin
+# alternative, ...), so resolve the real path before matching.
+set(is_nix_compiler FALSE)
+get_filename_component(_cxx_real "${CMAKE_CXX_COMPILER}" REALPATH)
+if(_cxx_real MATCHES "^/nix/store/")
+    set(is_nix_compiler TRUE)
+endif()
+unset(_cxx_real)
+
+# True inside the Nix CI Docker image, identified by the /nix/ci-env tree it
+# ships (see nix/docker/Dockerfile). The dev shell and bare systems don't have
+# it, so it distinguishes the CI image from other Nix-compiler environments.
+set(is_ci_image FALSE)
+if(EXISTS "/nix/ci-env/bin")
+    set(is_ci_image TRUE)
+endif()
+
 # --------------------------------------------------------------------
 # Operating system detection
 # --------------------------------------------------------------------

cmake/PatchNixBinary.cmake

@@ -1,26 +1,38 @@
 #[===================================================================[
    Patch executables to run in non-Nix environments.
 
-   The Nix-based CI image links binaries against an ELF interpreter (loader)
+   The Nix toolchain links binaries against an ELF interpreter (loader)
    that lives in the Nix store, so the resulting binaries don't run elsewhere
-   (including once installed from the .deb package). `patch_nix_binary` adds a
-   POST_BUILD step that resets the interpreter to the system default loader and
-   drops the rpath.
+   (including once installed from the .deb package). `patch_nix_binary` resets
+   the interpreter to the system default loader and drops the rpath, once the
+   binary has been linked.
 
-   This is only active inside the Nix-based image, detected by the presence of
-   /tmp/loader-path.sh (shipped by that image, resolves the default loader). It
-   is skipped for sanitizer builds, whose runtime libraries are resolved through
-   the rpath. Everywhere else `patch_nix_binary` is a no-op.
+   This runs by default for Nix-toolchain builds (determined by whether the compiler resolves under /nix/store/).
+   Those builds are where binaries get a Nix-store loader.
+   It is opted out of by setting the XRPLD_NO_PATCH_NIX_BINARY environment variable —
+   the plain Nix dev shells set it, since their binaries link a newer glibc
+   and must not be retargeted to the system loader.
+
+   Non-Nix builds (a system compiler, already using the system loader) and sanitizer builds
+   (runtime libraries resolved through the rpath) are skipped too.
+   Everywhere else `patch_nix_binary` is a no-op.
+
+   The default loader is resolved by bin/default-loader-path.sh.
 #]===================================================================]
 
 include_guard(GLOBAL)
 
 include(CompilationEnv)
 
-# Provided by the Nix-based CI image; prints the system default ELF loader path.
-set(_loader_path_script "/tmp/loader-path.sh")
+# Resolves the system default ELF loader path for the current architecture.
+set(_loader_path_script "${CMAKE_SOURCE_DIR}/bin/default-loader-path.sh")
 
-if(is_linux AND NOT SANITIZERS_ENABLED AND EXISTS "${_loader_path_script}")
+if(
+    is_linux
+    AND NOT SANITIZERS_ENABLED
+    AND is_nix_compiler
+    AND NOT DEFINED ENV{XRPLD_NO_PATCH_NIX_BINARY}
+)
     execute_process(
         COMMAND "${_loader_path_script}"
         OUTPUT_VARIABLE DEFAULT_LOADER_PATH
@@ -41,13 +53,38 @@ function(patch_nix_binary target)
     if(NOT PATCH_NIX_BINARIES)
         return()
     endif()
-    add_custom_command(
-        TARGET ${target}
-        POST_BUILD
-        COMMAND
-            "${PATCHELF_COMMAND}" --set-interpreter "${DEFAULT_LOADER_PATH}"
-            --remove-rpath "$<TARGET_FILE:${target}>"
-        COMMENT "Patching ${target}: set default loader, remove rpath"
-        VERBATIM
+
+    set(patch_command
+        "${PATCHELF_COMMAND}"
+        --set-interpreter
+        "${DEFAULT_LOADER_PATH}"
+        --remove-rpath
+        "$<TARGET_FILE:${target}>"
     )
+    set(comment "Patching ${target}: set default loader, remove rpath")
+
+    # POST_BUILD is the cheap way to do this: it runs only when the binary is
+    # relinked. It is also only available in the directory that defined the
+    # target, so for a target from elsewhere (e.g. a FetchContent subproject)
+    # fall back to a custom target that runs after the binary is linked. That
+    # one runs on every build, which is harmless because patchelf is idempotent.
+    get_target_property(target_source_dir ${target} SOURCE_DIR)
+    if("${target_source_dir}" STREQUAL "${CMAKE_CURRENT_SOURCE_DIR}")
+        add_custom_command(
+            TARGET ${target}
+            POST_BUILD
+            COMMAND ${patch_command}
+            COMMENT "${comment}"
+            VERBATIM
+        )
+    else()
+        add_custom_target(
+            ${target}-patch-nix
+            ALL
+            COMMAND ${patch_command}
+            COMMENT "${comment}"
+            VERBATIM
+        )
+        add_dependencies(${target}-patch-nix ${target})
+    endif()
 endfunction()

cmake/XrplAddBenchmark.cmake

@@ -0,0 +1,38 @@
+include_guard()
+
+include(isolate_headers)
+
+# Define a benchmark executable for the module `name`.
+#
+# This follows the same general pattern as other build helpers in this repo
+# (e.g. `add_module`): create a target and isolate headers, but here the target
+# is a benchmark executable and no `add_test(...)` is registered.
+#
+# `isolate_headers` exposes only `${CMAKE_CURRENT_SOURCE_DIR}/${name}` on the
+# include path, rooted at `src`, so a benchmark's own headers are reached as
+# `<benchmarks/.../${name}/...>` and nothing else in the tree leaks in.
+function(xrpl_add_benchmark name)
+    set(target ${PROJECT_NAME}.bench.${name})
+
+    file(
+        GLOB_RECURSE sources
+        CONFIGURE_DEPENDS
+        "${CMAKE_CURRENT_SOURCE_DIR}/${name}/*.cpp"
+        "${CMAKE_CURRENT_SOURCE_DIR}/${name}.cpp"
+    )
+    add_executable(${target} ${ARGN} ${sources})
+
+    # Benchmark sources register cases through Google Benchmark's static
+    # registrars (anonymous-namespace lambdas). Merging several such files into
+    # one unity translation unit collides those internal-linkage entities, so
+    # keep benchmarks out of the unity build - mirroring xrpl.libpb in
+    # XrplCore.cmake. Each file compiles fine on its own.
+    set_target_properties(${target} PROPERTIES UNITY_BUILD OFF)
+
+    isolate_headers(
+        ${target}
+        "${CMAKE_SOURCE_DIR}/src"
+        "${CMAKE_CURRENT_SOURCE_DIR}/${name}"
+        PRIVATE
+    )
+endfunction()

cmake/XrplCompiler.cmake

@@ -171,9 +171,8 @@ else()
     # Clang wrapper supplies those paths itself (via -nostdinc++), so at compile time the
     # flag is unused -> Clang errors under our -Werror. At link time the flag IS consumed
     # (it selects the C++ runtime), so we move it there instead of dropping it entirely.
-    get_filename_component(_cxx_real "${CMAKE_CXX_COMPILER}" REALPATH)
     if(
-        _cxx_real MATCHES "^/nix/store/"
+        is_nix_compiler
         AND is_linux
         AND is_clang
         AND CMAKE_CXX_FLAGS MATCHES "stdlib=libstdc"

cmake/XrplCore.cmake

@@ -133,6 +133,12 @@ target_link_libraries(
 add_module(xrpl resource)
 target_link_libraries(xrpl.libxrpl.resource PUBLIC xrpl.libxrpl.protocol)
 
+add_module(xrpl peerfinder)
+target_link_libraries(
+    xrpl.libxrpl.peerfinder
+    PUBLIC xrpl.libxrpl.basics xrpl.libxrpl.protocol
+)
+
 # Level 08
 add_module(xrpl net)
 target_link_libraries(
@@ -199,7 +205,27 @@ target_link_libraries(
 )
 
 add_module(xrpl tx)
-target_link_libraries(xrpl.libxrpl.tx PUBLIC xrpl.libxrpl.ledger)
+# The wasm engine is a Rust crate reached over cxx: the bridge target supplies the
+# generated `lib.h` and `rust/cxx.h` that `tx/wasm` compiles against, and the Rust
+# static library everything downstream links. PUBLIC because the include path travels
+# with the module's own public headers.
+target_link_libraries(
+    xrpl.libxrpl.tx
+    PUBLIC xrpl.libxrpl.ledger xrpl_wasm_vm_ffi_cxxbridge
+)
+# Those headers do not exist at configure time, and the header-verification target
+# compiles this module's headers on their own, so both need the crates built first.
+add_dependencies(xrpl.libxrpl.tx xrpl_crates)
+
+add_module(xrpl consensus)
+target_link_libraries(
+    xrpl.libxrpl.consensus
+    PUBLIC
+        xrpl.libxrpl.basics
+        xrpl.libxrpl.json
+        xrpl.libxrpl.protocol
+        xrpl.libxrpl.ledger
+)
 
 add_library(xrpl.libxrpl)
 set_target_properties(xrpl.libxrpl PROPERTIES OUTPUT_NAME xrpl)
@@ -220,6 +246,7 @@ target_link_modules(
     beast
     conditions
     config
+    consensus
     core
     crypto
     git
@@ -227,6 +254,7 @@ target_link_modules(
     ledger
     net
     nodestore
+    peerfinder
     protocol
     protocol_autogen
     rdb

cmake/XrplPackaging.cmake

@@ -25,6 +25,19 @@ if(NOT (RPMBUILD_EXECUTABLE OR DPKG_BUILDPACKAGE_EXECUTABLE))
     return()
 endif()
 
+if(NOT TARGET xrpld)
+    message(STATUS "xrpld=ON is required; 'package' target not available")
+    return()
+endif()
+
+if(NOT TARGET validator-keys)
+    message(
+        STATUS
+        "validator_keys=ON is required; 'package' target not available"
+    )
+    return()
+endif()
+
 set(package_env
     SRC_DIR=${CMAKE_SOURCE_DIR}
     BUILD_DIR=${CMAKE_BINARY_DIR}
@@ -37,7 +50,7 @@ add_custom_target(
         ${CMAKE_COMMAND} -E env ${package_env}
         ${CMAKE_SOURCE_DIR}/package/build_pkg.sh
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR}
-    DEPENDS xrpld
+    DEPENDS xrpld validator-keys
     COMMENT "Building Linux package (deb/rpm inferred from host tooling)"
     VERBATIM
 )

cmake/XrplProtocolAutogen.cmake

@@ -2,21 +2,22 @@
    Protocol Autogen - Code generation for protocol wrapper classes
 #]===================================================================]
 
+# The repository root, derived from the location of this file rather than from
+# the including project, so that the targets below can also be offered on their
+# own by cmake/codegen/CMakeLists.txt.
+get_filename_component(XRPL_ROOT "${CMAKE_CURRENT_LIST_DIR}/.." ABSOLUTE)
+
 set(CODEGEN_VENV_DIR
-    "${CMAKE_CURRENT_SOURCE_DIR}/.venv"
+    "${XRPL_ROOT}/.venv"
     CACHE PATH
     "Path to a Python virtual environment for code generation. A venv will be created here by setup_code_gen and used to run generation scripts."
 )
 
 # Directory paths
-set(MACRO_DIR "${CMAKE_CURRENT_SOURCE_DIR}/include/xrpl/protocol/detail")
-set(AUTOGEN_HEADER_DIR
-    "${CMAKE_CURRENT_SOURCE_DIR}/include/xrpl/protocol_autogen"
-)
-set(AUTOGEN_TEST_DIR
-    "${CMAKE_CURRENT_SOURCE_DIR}/src/tests/libxrpl/protocol_autogen"
-)
-set(SCRIPTS_DIR "${CMAKE_CURRENT_SOURCE_DIR}/cmake/scripts/codegen")
+set(MACRO_DIR "${XRPL_ROOT}/include/xrpl/protocol/detail")
+set(AUTOGEN_HEADER_DIR "${XRPL_ROOT}/include/xrpl/protocol_autogen")
+set(AUTOGEN_TEST_DIR "${XRPL_ROOT}/src/tests/libxrpl/protocol_autogen")
+set(SCRIPTS_DIR "${XRPL_ROOT}/cmake/scripts/codegen")
 
 # Input macro files
 set(TRANSACTIONS_MACRO "${MACRO_DIR}/transactions.macro")
@@ -114,14 +115,14 @@ if(CODEGEN_VENV_DIR)
         setup_code_gen
         COMMAND ${Python3_EXECUTABLE} -m venv "${CODEGEN_VENV_DIR}"
         COMMAND ${CODEGEN_PYTHON} -m pip install -r "${REQUIREMENTS_FILE}"
-        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}"
+        WORKING_DIRECTORY "${XRPL_ROOT}"
         COMMENT "Creating venv and installing code generation dependencies..."
     )
 else()
     add_custom_target(
         setup_code_gen
         COMMAND ${Python3_EXECUTABLE} -m pip install -r "${REQUIREMENTS_FILE}"
-        WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}"
+        WORKING_DIRECTORY "${XRPL_ROOT}"
         COMMENT "Installing code generation dependencies..."
     )
 endif()
@@ -139,8 +140,8 @@ add_custom_target(
         -DSFIELDS_MACRO=${SFIELDS_MACRO}
         -DAUTOGEN_HEADER_DIR=${AUTOGEN_HEADER_DIR}
         -DAUTOGEN_TEST_DIR=${AUTOGEN_TEST_DIR} -P
-        "${CMAKE_CURRENT_SOURCE_DIR}/cmake/XrplProtocolAutogenRun.cmake"
-    WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}"
+        "${CMAKE_CURRENT_LIST_DIR}/XrplProtocolAutogenRun.cmake"
+    WORKING_DIRECTORY "${XRPL_ROOT}"
     COMMENT "Running protocol code generation..."
     SOURCES ${ALL_INPUT_FILES}
 )

cmake/XrplSanity.cmake

@@ -36,6 +36,19 @@ elseif(is_gcc)
     endif()
 endif()
 
+# A Nix compiler is only meant to be used from a managed environment: the xrpld
+# dev shell (which exports XRPL_DEVSHELL) or the CI image. Using one from a bare
+# shell usually means a leaked toolchain (picked up via PATH or a Conan profile)
+# and leads to confusing breakage, so fail early with guidance.
+if(is_nix_compiler AND NOT is_ci_image AND NOT DEFINED ENV{XRPL_DEVSHELL})
+    message(
+        FATAL_ERROR
+        "A Nix compiler (${CMAKE_CXX_COMPILER}) is being used outside the xrpld "
+        "dev shell. Enter it with `nix develop` (see docs/build/nix.md) before "
+        "configuring the build."
+    )
+endif()
+
 # check for in-source build and fail
 if("${CMAKE_CURRENT_SOURCE_DIR}" STREQUAL "${CMAKE_BINARY_DIR}")
     message(

cmake/XrplSettings.cmake

@@ -30,6 +30,8 @@ if(tests)
     endif()
 endif()
 
+option(benchmark "Build benchmarks" ON)
+
 # Enabled by default so every header is compiled on its own as the main file of
 # its own compile_commands.json entry - this is what lets clang-tidy (and clangd
 # and IDEs) analyse a header's own includes directly. The per-header objects are

cmake/XrplValidatorKeys.cmake

@@ -5,22 +5,39 @@ option(
 )
 
 if(validator_keys)
-    git_branch(current_branch)
-    # default to tracking VK master branch unless we are on release
-    if(NOT (current_branch STREQUAL "release"))
-        set(current_branch "master")
-    endif()
-    message(STATUS "Tracking ValidatorKeys branch: ${current_branch}")
+    # Own the install destination below rather than relying on another module
+    # having pulled this in first.
+    include(GNUInstallDirs)
+
+    # Pinned to an exact commit, not a branch: the tool ships inside our
+    # packages, so the same xrpld version must always package the same
+    # validator-keys. Bump this deliberately.
+    set(validator_keys_commit "4c0fb75eec9601c711645998c904507e87e910ae")
+    message(STATUS "Using ValidatorKeys commit: ${validator_keys_commit}")
 
     FetchContent_Declare(
         validator_keys
         GIT_REPOSITORY https://github.com/ripple/validator-keys-tool.git
-        GIT_TAG "${current_branch}"
+        GIT_TAG "${validator_keys_commit}"
     )
     FetchContent_MakeAvailable(validator_keys)
+    # The tool's own CMakeLists excludes the target from 'all' when it is built
+    # as a subproject. Undo that, so validator_keys=ON really does build it.
     set_target_properties(
         validator-keys
-        PROPERTIES RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}"
+        PROPERTIES
+            RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}"
+            EXCLUDE_FROM_ALL OFF
+            EXCLUDE_FROM_DEFAULT_BUILD OFF
+    )
+    # We ship this binary, so like xrpld it must not keep the Nix store's ELF
+    # loader, or it cannot run on the target distro at all.
+    patch_nix_binary(validator-keys)
+
+    configure_file(
+        "${validator_keys_SOURCE_DIR}/LICENSE"
+        "${CMAKE_BINARY_DIR}/validator-keys-LICENSE"
+        COPYONLY
     )
     install(TARGETS validator-keys RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
 endif()

cmake/codegen/CMakeLists.txt

@@ -0,0 +1,21 @@
+#[===================================================================[
+   Protocol Autogen - Standalone project
+
+   Exposes the 'setup_code_gen' and 'code_gen' targets on their own, without
+   configuring the rest of xrpl. Code generation is pure Python, so this needs
+   neither the dependencies nor a compiler, which makes it usable in CI and by
+   contributors who only want to regenerate the protocol wrapper classes:
+
+     cmake -S cmake/codegen -B build/codegen
+     cmake --build build/codegen --target setup_code_gen
+     cmake --build build/codegen --target code_gen
+
+   The targets are identical to the ones offered by the top-level build, since
+   both come from cmake/XrplProtocolAutogen.cmake.
+#]===================================================================]
+
+cmake_minimum_required(VERSION 3.16)
+
+project(xrpl_codegen LANGUAGES NONE)
+
+include("${CMAKE_CURRENT_LIST_DIR}/../XrplProtocolAutogen.cmake")

cmake/scripts/codegen/templates/LedgerEntry.h.mako

@@ -177,7 +177,9 @@ ${field['typeData']['setter_type']} ${field['paramName']}${',' if i < len(requir
         object_ = *sle;
     }
 
-    /** @brief Ledger entry-specific field setters */
+    /**
+     * @brief Ledger entry-specific field setters
+     */
 % for field in fields:
 
     /**

cmake/scripts/codegen/templates/Transaction.h.mako

@@ -185,7 +185,9 @@ public:
         object_ = *tx;
     }
 
-    /** @brief Transaction-specific field setters */
+    /**
+     * @brief Transaction-specific field setters
+     */
 % for field in fields:
 
     /**

conan.lock

@@ -10,22 +10,24 @@
         "rocksdb/10.5.1#4a197eca381a3e5ae8adf8cffa5aacd0%1782392413.075713",
         "re2/20251105#8579cfd0bda4daf0683f9e3898f964b4%1782392402.431897",
         "protobuf/6.33.5#ff253ead763bd8d9904a52979cd21e81%1782392410.233933",
-        "openssl/3.6.3#1163d4ddc603907084d08a6a0c6e580f%1782307150.583886",
+        "openssl/3.6.3#f806de8933e3bf6f01016c6a888cee2e%1783945160.863288",
         "nudb/2.0.9#11149c73f8f2baff9a0198fe25971fc7%1782392402.297166",
-        "mpt-crypto/0.4.0-rc2#a580f2f9ad0e795de696aa62d54fb9af%1782425834.488828",
+        "mpt-crypto/1.0.2#b313cef0c1a493eb970ad185b2e9bab7%1784285108.866483",
         "lz4/1.10.0#982d9b673900f665a1da109e09c17cab%1782392402.164188",
         "libiconv/1.17#9923bc6dc6f106646d6967e0039a5ada%1782392792.775744",
         "libbacktrace/cci.20210118#a7691bfccd8caaf66309df196790a5a1%1782392402.420732",
         "libarchive/3.8.7#c446109bd1f1d8ba7936c94189bc50e6%1782392403.066892",
         "jemalloc/5.3.1#1fc58d55316041f10fbc1e8a2eae632a%1776700028.228",
         "gtest/1.17.0#5224b3b3ff3b4ce1133cbdd27d53ee7d%1782392402.791979",
-        "grpc/1.81.1#5217e6ef0544c42b46f4af35d5e7f649%1782307148.845616",
+        "grpc/1.81.1#f729f6d75992d20f9c72828e9142d62f%1783945160.094135",
+        "fast_float/8.2.10#f6f28d6bb22112078e7dbda611caf681%1782494504.298",
         "ed25519/2015.03#ae761bdc52730a843f0809bdf6c1b1f6%1782307148.15562",
         "date/3.0.4#862e11e80030356b53c2c38599ceb32b%1782392402.538492",
         "c-ares/1.34.6#545240bb1c40e2cacd4362d6b8967650%1782392402.681654",
         "bzip2/1.0.8#c470882369c2d95c5c77e970c0c7e321%1782392402.296732",
         "boost/1.91.0#ea540ca2133d831b560036aa24dece3c%1782392419.475605",
-        "abseil/20250127.0#bb0baf1f362bc4a725a24eddd419b8f7%1782307147.395833"
+        "benchmark/1.9.5#b885dc73ad67b40a55d45684d1c88ad1%1782736613.864841",
+        "abseil/20250127.0#9ef01c1451a8340f9022e46238c0fbb6%1783945159.651047"
     ],
     "build_requires": [
         "zlib/1.3.2#1cb806da49011867778ffb6ac7190fcb%1782392402.122708",
@@ -33,12 +35,12 @@
         "protobuf/6.33.5#ff253ead763bd8d9904a52979cd21e81%1782392410.233933",
         "nasm/2.16.01#31e26f2ee3c4346ecd347911bd126904%1782395690.33162",
         "msys2/cci.latest#d22fe7b2808f5fd34d0a7923ace9c54f%1770657326.649",
-        "m4/1.4.19#34c4bbc3eeebe98ca6edf2f52d602e7d%1777282960.259",
+        "m4/1.4.19#1727f439cf74e83826ec96d0b4904eee%1784541921.659",
         "cmake/4.3.3#840cf00ea09777e05c2050a50a82c722%1782392418.696091",
         "b2/5.4.2#ffd6084a119587e70f11cd45d1a386e2%1782392402.624226",
         "automake/1.16.5#b91b7c384c3deaa9d535be02da14d04f%1755524470.56",
         "autoconf/2.71#51077f068e61700d65bb05541ea1e4b0%1731054366.86",
-        "abseil/20250127.0#bb0baf1f362bc4a725a24eddd419b8f7%1782307147.395833"
+        "abseil/20250127.0#9ef01c1451a8340f9022e46238c0fbb6%1783945159.651047"
     ],
     "python_requires": [],
     "overrides": {

conan/profiles/default

@@ -5,6 +5,13 @@
 {% if os == "Linux" %}
 {% set compiler_version = detect_api.default_compiler_version(compiler, version) %}
 {% endif %}
+{% if os == "Macos" %}
+{# Minimum macOS the dependencies target. #}
+{# Without this, Conan builds each dependency against the (possibly newer) host SDK, so the #}
+{# dependency objects target a newer macOS than the binary and the linker warns. #}
+{# Keep at or below CMAKE_OSX_DEPLOYMENT_TARGET in CMakeLists.txt. #}
+{% set min_macos_version = "15.0" %}
+{% endif %}
 
 [settings]
 os={{ os }}
@@ -18,10 +25,39 @@ compiler.runtime=static
 {% else %}
 compiler.libcxx={{ detect_api.detect_libcxx(compiler, version, compiler_exe) }}
 {% endif %}
+{% if os == "Macos" %}
+os.version={{ min_macos_version }}
+{% endif %}
 
 [conf]
+{# The Boost recipe builds with b2, which doesn't use Conan's toolchain files. #}
+{# Instead it hand-rolls the compiler for user-config.jam, #}
+{# and its fallback probes a version-suffixed binary (e.g. `g++-15`) before plain `g++`. #}
+{# Inside the Nix shell the wrapper only provides `g++`/`gcc` (no `-15` suffix), #}
+{# so on a host that also has a system `g++-15` the probe escapes Nix #}
+{# and picks the system compiler, which is mismatched with the Nix libraries #}
+{# and breaks the build (e.g. Boost.Stacktrace link checks fail). #}
+{# Pinning the executables here short-circuits that probe so Boost (and the rest of the toolchain) #}
+{# resolve the same compiler. #}
+{# Not part of the package ID, so binaries stay shareable. #}
+{% if os != "Windows" %}
+{% set cc_exe  = {"gcc": "gcc",  "clang": "clang",   "apple-clang": "clang"}.get(compiler) %}
+{% set cxx_exe = {"gcc": "g++",  "clang": "clang++", "apple-clang": "clang++"}.get(compiler) %}
+tools.build:compiler_executables={'c':'{{ cc_exe }}','cpp':'{{ cxx_exe }}'}
+{% endif %}
+
 {# By default, Conan tries to reuse binaries built with different cppstd versions. #}
 {# We want to avoid that to improve reproduceability, so we add the cppstd version to the package ID. #}
 {# More info: https://docs.conan.io/2/reference/extensions/binary_compatibility.html #}
 user.package:cppstd_version=23
 tools.info.package_id:confs+=["user.package:cppstd_version"]
+
+{% if os == "Macos" %}
+[buildenv]
+{# os.version adds -mmacosx-version-min to compiler command lines, #}
+{# but Boost.Context's b2 assembly (.S) rule ignores it, #}
+{# so those objects keep the host SDK version and still warn at link time. #}
+{# clang's assembler honors this env var regardless, pinning them. #}
+{# Scoped to boost/* since it is the only gap. #}
+boost/*:MACOSX_DEPLOYMENT_TARGET={{ min_macos_version }}
+{% endif %}

conanfile.py

@@ -15,6 +15,7 @@ class Xrpl(ConanFile):
     settings = "os", "compiler", "build_type", "arch"
     options = {
         "assertions": [True, False],
+        "benchmark": [True, False],
         "coverage": [True, False],
         "fPIC": [True, False],
         "jemalloc": [True, False],
@@ -28,6 +29,7 @@ class Xrpl(ConanFile):
 
     requires = [
         "ed25519/2015.03",
+        "fast_float/8.2.10",
         "grpc/1.81.1",
         "libarchive/3.8.7",
         "nudb/2.0.9",
@@ -46,6 +48,7 @@ class Xrpl(ConanFile):
 
     default_options = {
         "assertions": False,
+        "benchmark": True,
         "coverage": False,
         "fPIC": True,
         "jemalloc": False,
@@ -129,12 +132,14 @@ class Xrpl(ConanFile):
             self.options["boost"].without_cobalt = True
 
     def requirements(self):
+        if self.options.benchmark:
+            self.requires("benchmark/1.9.5")
         self.requires("boost/1.91.0", force=True, transitive_headers=True)
         self.requires("date/3.0.4", transitive_headers=True)
         if self.options.jemalloc:
             self.requires("jemalloc/5.3.1")
         self.requires("lz4/1.10.0", force=True)
-        self.requires("mpt-crypto/0.4.0-rc2", transitive_headers=True)
+        self.requires("mpt-crypto/1.0.2", transitive_headers=True)
         self.requires("protobuf/6.33.5", force=True)
         if self.options.rocksdb:
             self.requires("rocksdb/10.5.1")
@@ -162,6 +167,7 @@ class Xrpl(ConanFile):
     def generate(self):
         tc = CMakeToolchain(self)
         tc.variables["tests"] = self.options.tests
+        tc.variables["benchmark"] = self.options.benchmark
         tc.variables["assert"] = self.options.assertions
         tc.variables["coverage"] = self.options.coverage
         tc.variables["jemalloc"] = self.options.jemalloc
@@ -206,6 +212,7 @@ class Xrpl(ConanFile):
             "boost::thread",
             "date::date",
             "ed25519::ed25519",
+            "fast_float::fast_float",
             "grpc::grpc++",
             "libarchive::libarchive",
             "lz4::lz4",

crates/.cargo/config.toml

@@ -0,0 +1,5 @@
+[target.x86_64-unknown-linux-gnu]
+rustflags = ["-C", "link-args=-static-libgcc"]
+
+[target.x86_64-pc-windows-msvc]
+rustflags = ["-C", "target-feature=+crt-static"]

crates/CMakeLists.txt

@@ -0,0 +1,59 @@
+set(CORROSION_VERSION 0.6.1)
+
+find_package(Corrosion ${CORROSION_VERSION} QUIET)
+if(NOT Corrosion_FOUND)
+    include(FetchContent)
+    FetchContent_Declare(
+        Corrosion
+        GIT_REPOSITORY https://github.com/corrosion-rs/corrosion.git
+        GIT_TAG v${CORROSION_VERSION}
+    )
+    FetchContent_MakeAvailable(Corrosion)
+endif()
+
+corrosion_import_crate(MANIFEST_PATH ${CMAKE_CURRENT_SOURCE_DIR}/Cargo.toml)
+
+file(
+    WRITE "${CMAKE_CURRENT_BINARY_DIR}/.clang-tidy"
+    "# Auto-generated by crates/CMakeLists.txt. Do not edit.\n"
+    "# Neutralizes clang-tidy for corrosion/cxxbridge-generated C++.\n"
+    "# One check kept enabled to avoid clang-tidy's \"no checks enabled\" error.\n"
+    "Checks: '-*,google-readability-todo'\n"
+    "WarningsAsErrors: ''\n"
+    "HeaderFilterRegex: ''\n"
+    "InheritParentConfig: false\n"
+)
+
+# Umbrella target that aggregates all crate-generated code (cxxbridge headers,
+# etc.). Build this before running clang-tidy so generated headers are present.
+add_custom_target(xrpl_crates)
+
+# add_xrpl_crate(<name> CRATE <crate> FILES <file>...) Creates a cxxbridge
+# target <name>_cxxbridge and registers it with xrpl_crates.
+function(add_xrpl_crate name)
+    cmake_parse_arguments(ARG "" "CRATE" "FILES" ${ARGN})
+    corrosion_add_cxxbridge(${name}_cxxbridge CRATE ${ARG_CRATE} FILES
+                            ${ARG_FILES}
+    )
+    # Generated cxxbridge headers don't exist at configure time; CMake 3.28+
+    # validates INTERFACE_SOURCES on consuming targets. Clear it to skip the
+    # existence check — build-time ordering is enforced by the custom commands.
+    set_target_properties(${name}_cxxbridge PROPERTIES INTERFACE_SOURCES "")
+    add_dependencies(xrpl_crates ${name}_cxxbridge)
+endfunction()
+
+add_xrpl_crate(xrpl_wasm_vm_ffi CRATE xrpl_wasm_vm_ffi FILES lib.rs)
+
+# Test-only, and deliberately not part of xrpl_wasm_vm_ffi: it carries the `wat` assembler,
+# which the engine's `wasmi default-features = false` exists to keep out of the consensus
+# path. Linked from src/tests/libxrpl only, so the shipped node cannot contain it.
+add_xrpl_crate(xrpl_wasm_testkit CRATE xrpl_wasm_testkit FILES lib.rs)
+
+# The wasm bridge `include!`s a project header, so its generated translation unit needs
+# the project's include root. Deliberately only that: a header reached from here must
+# stay light enough to compile without the Boost paths this target does not get, which
+# is why `HostContext.h` forward-declares `xrpl::HostFunctions` instead of including it.
+target_include_directories(
+    xrpl_wasm_vm_ffi_cxxbridge
+    PRIVATE ${CMAKE_SOURCE_DIR}/include
+)

crates/Cargo.lock

@@ -0,0 +1,497 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 4
+
+[[package]]
+name = "anstyle"
+version = "1.0.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "940b3a0ca603d1eade50a4846a2afffd5ef57a9feac2c0e2ec2e14f9ead76000"
+
+[[package]]
+name = "bitflags"
+version = "2.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b588b76d00fde79687d7646a9b5bdf3cc0f655e0bbd080335a95d7e96f3587da"
+
+[[package]]
+name = "bumpalo"
+version = "3.20.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "72f5acc6cb2ba439de613abc23857ec3d78374d8ed5ac84e9d11336e87da8649"
+
+[[package]]
+name = "cc"
+version = "1.2.61"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d16d90359e986641506914ba71350897565610e87ce0ad9e6f28569db3dd5c6d"
+dependencies = [
+ "find-msvc-tools",
+ "shlex",
+]
+
+[[package]]
+name = "clap"
+version = "4.6.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1ddb117e43bbf7dacf0a4190fef4d345b9bad68dfc649cb349e7d17d28428e51"
+dependencies = [
+ "clap_builder",
+]
+
+[[package]]
+name = "clap_builder"
+version = "4.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "714a53001bf66416adb0e2ef5ac857140e7dc3a0c48fb28b2f10762fc4b5069f"
+dependencies = [
+ "anstyle",
+ "clap_lex",
+ "strsim",
+]
+
+[[package]]
+name = "clap_lex"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c8d4a3bb8b1e0c1050499d1815f5ab16d04f0959b233085fb31653fbfc9d98f9"
+
+[[package]]
+name = "codespan-reporting"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "af491d569909a7e4dee0ad7db7f5341fef5c614d5b8ec8cf765732aba3cff681"
+dependencies = [
+ "serde",
+ "termcolor",
+ "unicode-width",
+]
+
+[[package]]
+name = "cxx"
+version = "1.0.198"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6fe442a792c7c736eea18b32a7f8a3b63cf8aafabda6760042dc2fdeda456291"
+dependencies = [
+ "cc",
+ "cxx-build",
+ "cxxbridge-cmd",
+ "cxxbridge-flags",
+ "cxxbridge-macro",
+ "foldhash 0.2.0",
+ "link-cplusplus",
+]
+
+[[package]]
+name = "cxx-build"
+version = "1.0.198"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e3184a94384c663718698311a78a51ac00c484c10b4eeac06fb0a068c5f64fa2"
+dependencies = [
+ "cc",
+ "codespan-reporting",
+ "indexmap",
+ "proc-macro2",
+ "quote",
+ "scratch",
+ "syn 3.0.3",
+]
+
+[[package]]
+name = "cxxbridge-cmd"
+version = "1.0.198"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0148d8fd1199329ddf1d157a5e134e51ceff37c6a7ddd38615c399d81cb05d8d"
+dependencies = [
+ "clap",
+ "codespan-reporting",
+ "indexmap",
+ "proc-macro2",
+ "quote",
+ "syn 3.0.3",
+]
+
+[[package]]
+name = "cxxbridge-flags"
+version = "1.0.198"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "52850339faed2eaadd24e286dc1d8268cc6f8a7bd9524d713adc9099566b4c89"
+
+[[package]]
+name = "cxxbridge-macro"
+version = "1.0.198"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2c77c856545d886c9bd5215409ebb63b925e262135248b50c79e5a5f194ee47c"
+dependencies = [
+ "indexmap",
+ "proc-macro2",
+ "quote",
+ "syn 3.0.3",
+]
+
+[[package]]
+name = "equivalent"
+version = "1.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "877a4ace8713b0bcf2a4e7eec82529c029f1d0619886d18145fea96c3ffe5c0f"
+
+[[package]]
+name = "find-msvc-tools"
+version = "0.1.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5baebc0774151f905a1a2cc41989300b1e6fbb29aff0ceffa1064fdd3088d582"
+
+[[package]]
+name = "foldhash"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2"
+
+[[package]]
+name = "foldhash"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "77ce24cb58228fbb8aa041425bb1050850ac19177686ea6e0f41a70416f56fdb"
+
+[[package]]
+name = "hashbrown"
+version = "0.15.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9229cfe53dfd69f0609a49f65461bd93001ea1ef889cd5529dd176593f5338a1"
+dependencies = [
+ "foldhash 0.1.5",
+]
+
+[[package]]
+name = "hashbrown"
+version = "0.17.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4f467dd6dccf739c208452f8014c75c18bb8301b050ad1cfb27153803edb0f51"
+
+[[package]]
+name = "indexmap"
+version = "2.14.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d466e9454f08e4a911e14806c24e16fba1b4c121d1ea474396f396069cf949d9"
+dependencies = [
+ "equivalent",
+ "hashbrown 0.17.0",
+]
+
+[[package]]
+name = "leb128fmt"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09edd9e8b54e49e587e4f6295a7d29c3ea94d469cb40ab8ca70b288248a81db2"
+
+[[package]]
+name = "libm"
+version = "0.2.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6d2cec3eae94f9f509c767b45932f1ada8350c4bdb85af2fcab4a3c14807981"
+
+[[package]]
+name = "link-cplusplus"
+version = "1.0.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7f78c730aaa7d0b9336a299029ea49f9ee53b0ed06e9202e8cb7db9bae7b8c82"
+dependencies = [
+ "cc",
+]
+
+[[package]]
+name = "memchr"
+version = "2.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cf8baf1c55e62ffcace7a9f06f4bd9cd3f0c4beb022d3b367256b91b87513d98"
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.106"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8fd00f0bb2e90d81d1044c2b32617f68fcb9fa3bb7640c23e9c748e53fb30934"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.45"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "41f2619966050689382d2b44f664f4bc593e129785a36d6ee376ddf37259b924"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "scratch"
+version = "1.0.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d68f2ec51b097e4c1a75b681a8bec621909b5e91f15bb7b840c4f2f7b01148b2"
+
+[[package]]
+name = "semver"
+version = "1.0.28"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8a7852d02fc848982e0c167ef163aaff9cd91dc640ba85e263cb1ce46fae51cd"
+
+[[package]]
+name = "serde"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9a8e94ea7f378bd32cbbd37198a4a91436180c5bb472411e48b5ec2e2124ae9e"
+dependencies = [
+ "serde_core",
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_core"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "41d385c7d4ca58e59fc732af25c3983b67ac852c1a25000afe1175de458b67ad"
+dependencies = [
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_derive"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d540f220d3187173da220f885ab66608367b6574e925011a9353e4badda91d79"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "shlex"
+version = "1.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0fda2ff0d084019ba4d7c6f371c95d8fd75ce3524c3cb8fb653a3023f6323e64"
+
+[[package]]
+name = "spin"
+version = "0.9.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3763264f6b73151db08c50ff20d7d8a0b8796e021cdea7ceedad07b80155fa0e"
+
+[[package]]
+name = "string-interner"
+version = "0.19.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "23de088478b31c349c9ba67816fa55d9355232d63c3afea8bf513e31f0f1d2c0"
+dependencies = [
+ "hashbrown 0.15.5",
+ "serde",
+]
+
+[[package]]
+name = "strsim"
+version = "0.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7da8b5736845d9f2fcb837ea5d9e2628564b3b043a70948a3f0b778838c5fb4f"
+
+[[package]]
+name = "syn"
+version = "2.0.117"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e665b8803e7b1d2a727f4023456bbbbe74da67099c585258af0ad9c5013b9b99"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "syn"
+version = "3.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "53e9bae58849f64dfa4f5d5ae372c8341f7305f82a3868709269343628b659a3"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "termcolor"
+version = "1.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06794f8f6c5c898b3275aebefa6b8a1cb24cd2c6c79397ab15774837a0bc5755"
+dependencies = [
+ "winapi-util",
+]
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e6e4313cd5fcd3dad5cafa179702e2b244f760991f45397d14d4ebf38247da75"
+
+[[package]]
+name = "unicode-width"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b4ac048d71ede7ee76d585517add45da530660ef4390e49b098733c6e897f254"
+
+[[package]]
+name = "wasm-encoder"
+version = "0.254.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09480d646178e5fdd12bb06e812d0af9a3a191dbc9cd697fdc86687beade7393"
+dependencies = [
+ "leb128fmt",
+ "wasmparser 0.254.0",
+]
+
+[[package]]
+name = "wasmi"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2300d0f78cba12f14e29e8dd157ea64050c0a688179aefdb2050105805594a0c"
+dependencies = [
+ "spin",
+ "wasmi_collections",
+ "wasmi_core",
+ "wasmi_ir",
+ "wasmparser 0.239.0",
+]
+
+[[package]]
+name = "wasmi_collections"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f8a8c42a2a76148d43097b1d7cc2a5bf33d5c23bd4dd69015fc887e311767884"
+dependencies = [
+ "string-interner",
+]
+
+[[package]]
+name = "wasmi_core"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9013136083d988725953390bf668b64b7a218fabf26f8b913bbc59546b97ee27"
+dependencies = [
+ "libm",
+]
+
+[[package]]
+name = "wasmi_ir"
+version = "1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba1fa003f79156f406d62ef0e1464dc03e11ace37170e9fa7524299a75ad8f68"
+dependencies = [
+ "wasmi_core",
+]
+
+[[package]]
+name = "wasmparser"
+version = "0.239.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8c9d90bb93e764f6beabf1d02028c70a2156a6583e63ac4218dd07ef733368b0"
+dependencies = [
+ "bitflags",
+ "indexmap",
+]
+
+[[package]]
+name = "wasmparser"
+version = "0.254.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d5769a29f799fbab136aaf65b4fe5384cd7d93fe6fc9ba0dcb6c8382a1f16e27"
+dependencies = [
+ "bitflags",
+ "indexmap",
+ "semver",
+]
+
+[[package]]
+name = "wast"
+version = "254.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e7ed4dfc8f6b9fc38b231065e2cdfbf7359af5ab945990abf09658dcc63c3e32"
+dependencies = [
+ "bumpalo",
+ "leb128fmt",
+ "memchr",
+ "unicode-width",
+ "wasm-encoder",
+]
+
+[[package]]
+name = "wat"
+version = "1.254.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7127f7f9b8f127c879991cecd35f494e4628bae1b0874c681414d8d8831e952c"
+dependencies = [
+ "wast",
+]
+
+[[package]]
+name = "winapi-util"
+version = "0.1.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c2a7b1c03c876122aa43f3020e6c3c3ee5c05081c9a00739faf7503aeba10d22"
+dependencies = [
+ "windows-sys",
+]
+
+[[package]]
+name = "windows-link"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f0805222e57f7521d6a62e36fa9163bc891acd422f971defe97d64e70d0a4fe5"
+
+[[package]]
+name = "windows-sys"
+version = "0.61.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ae137229bcbd6cdf0f7b80a31df61766145077ddf49416a728b02cb3921ff3fc"
+dependencies = [
+ "windows-link",
+]
+
+[[package]]
+name = "xrpl-host-functions"
+version = "0.1.0"
+dependencies = [
+ "xrpl-host-functions-macros",
+]
+
+[[package]]
+name = "xrpl-host-functions-macros"
+version = "0.1.0"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 3.0.3",
+ "xrpl-host-functions",
+]
+
+[[package]]
+name = "xrpl-wasm-testkit"
+version = "0.1.0"
+dependencies = [
+ "cxx",
+ "wat",
+]
+
+[[package]]
+name = "xrpl-wasm-vm"
+version = "0.1.0"
+dependencies = [
+ "wasmi",
+ "wat",
+ "xrpl-host-functions",
+]
+
+[[package]]
+name = "xrpl-wasm-vm-ffi"
+version = "0.1.0"
+dependencies = [
+ "cxx",
+ "xrpl-host-functions",
+ "xrpl-wasm-vm",
+]

crates/Cargo.toml

@@ -0,0 +1,15 @@
+[workspace]
+members = ["xrpl-wasm-vm-ffi", "xrpl-wasm-vm", "xrpl-wasm-testkit", "xrpl-host-functions", "xrpl-host-functions-macros"]
+resolver = "3"
+
+[workspace.dependencies]
+cxx = { version = "1.0.198", features = ["c++20"] }
+
+[workspace.package]
+edition = "2024"
+
+[profile.release]
+opt-level = 3
+overflow-checks = true
+lto = true
+debug = true

crates/xrpl-host-functions-macros/Cargo.toml

@@ -0,0 +1,18 @@
+[package]
+name = "xrpl-host-functions-macros"
+version = "0.1.0"
+edition.workspace = true
+
+[lib]
+proc-macro = true
+
+[dependencies]
+syn = { version = "3", features = ["full"] }
+quote = "1"
+proc-macro2 = "1"
+
+# The expansion names `::xrpl_host_functions::HostFnSpec`, so the doctest needs the
+# facade crate. Cargo allows this cycle because dev-dependencies are outside the
+# library build graph.
+[dev-dependencies]
+xrpl-host-functions.path = "../xrpl-host-functions"

crates/xrpl-host-functions-macros/src/errors.rs

@@ -0,0 +1,12 @@
+/// Folds accumulated diagnostics into the single error a macro can return.
+///
+/// `syn::Error` is itself a collection: `combine` appends, and
+/// `into_compile_error` emits one `compile_error!` per recorded span. Folding
+/// instead of returning the first error means every mistake in a
+/// `host_functions!` block surfaces in one build rather than one per rebuild.
+pub(crate) fn combine(errors: Vec<syn::Error>) -> Option<syn::Error> {
+    errors.into_iter().reduce(|mut first, next| {
+        first.combine(next);
+        first
+    })
+}

crates/xrpl-host-functions-macros/src/lib.rs

@@ -0,0 +1,384 @@
+mod errors;
+mod parsed_host_function;
+
+use std::collections::HashSet;
+
+use proc_macro2::TokenStream;
+use quote::quote;
+use syn::{
+    TraitItemFn,
+    parse::{Parse, ParseStream},
+    parse2,
+};
+
+use parsed_host_function::ParsedHostFunction;
+
+/// Declares the wasm host ABI once, and generates everything that follows from it.
+///
+/// The input is a block of `fn` declarations, each carrying the gas cost the host
+/// charges before the call and the name the guest imports it under. Doc comments
+/// are kept and appear on the generated items.
+///
+/// This crate is an implementation detail of `xrpl-host-functions`, which
+/// hand-writes the types the expansion refers to and holds the one declaration
+/// block. The expansion names those types by absolute path, so a call site needs
+/// `xrpl-host-functions` as a dependency but no imports from it.
+///
+/// ```
+/// use xrpl_host_functions::HostResult;
+/// use xrpl_host_functions_macros::host_functions;
+///
+/// host_functions! {
+///     /// The sequence number of the ledger being built, as 4 little-endian bytes.
+///     #[gas = 60]
+///     #[wasm_name = "ldgr_index"]
+///     fn get_ledger_sqn(&self, out: &mut [u8]) -> HostResult<usize>;
+///
+///     /// Writes `msg` to the trace log.
+///     #[gas = 500]
+///     #[wasm_name = "trace_num"]
+///     fn trace_num(&self, msg: &str, number: i64) -> HostResult<()>;
+/// }
+///
+/// // A `HostFunctions` trait, holding the declarations verbatim:
+/// struct Host;
+/// impl HostFunctions for Host {
+///     fn get_ledger_sqn(&self, out: &mut [u8]) -> HostResult<usize> {
+///         out[..4].copy_from_slice(&7u32.to_le_bytes());
+///         Ok(4)
+///     }
+///     fn trace_num(&self, _msg: &str, _number: i64) -> HostResult<()> { Ok(()) }
+/// }
+///
+/// // A `HostFunctionSpec` enum carrying the ABI metadata as a `const` table:
+/// assert_eq!(HostFunctionSpec::GetLedgerSqn.gas(), 60);
+/// assert_eq!(HostFunctionSpec::TraceNum.wasm_name(), "trace_num");
+/// assert_eq!(HostFunctionSpec::ALL.len(), 2);
+/// ```
+///
+/// A declaration must be a plain `fn` taking `&self` and returning
+/// `HostResult<T>`, with no body and no generics: it maps to exactly one wasm
+/// import signature. Two declarations may not share a `wasm_name`, nor collapse to
+/// the same PascalCase variant.
+#[proc_macro]
+pub fn host_functions(input: proc_macro::TokenStream) -> proc_macro::TokenStream {
+    expand(input.into())
+        .unwrap_or_else(syn::Error::into_compile_error)
+        .into()
+}
+
+fn expand(input: TokenStream) -> syn::Result<TokenStream> {
+    let HostFunctionsInput { functions } = parse2(input)?;
+
+    let mut parsed = Vec::with_capacity(functions.len());
+    let mut errors = Vec::new();
+    for function in functions {
+        match ParsedHostFunction::parse(function) {
+            Ok(function) => parsed.push(function),
+            Err(error) => errors.push(error),
+        }
+    }
+    if let Some(error) = errors::combine(errors) {
+        return Err(error);
+    }
+    if let Some(error) = errors::combine(collisions(&parsed)) {
+        return Err(error);
+    }
+
+    Ok(generate(&parsed))
+}
+
+/// Names two declarations may not share, because the generated code would then
+/// fail to compile at a span the caller cannot see.
+fn collisions(functions: &[ParsedHostFunction]) -> Vec<syn::Error> {
+    let mut errors = Vec::new();
+    let mut variants = HashSet::new();
+    let mut wasm_names = HashSet::new();
+
+    for function in functions {
+        if !variants.insert(function.variant.to_string()) {
+            errors.push(syn::Error::new_spanned(
+                &function.variant,
+                format!(
+                    "another host function already becomes the `{}` variant",
+                    function.variant
+                ),
+            ));
+        }
+        if !wasm_names.insert(function.wasm_name.value()) {
+            errors.push(syn::Error::new_spanned(
+                &function.wasm_name,
+                format!(
+                    "another host function is already imported as `{}`",
+                    function.wasm_name.value()
+                ),
+            ));
+        }
+    }
+
+    errors
+}
+
+fn generate(functions: &[ParsedHostFunction]) -> TokenStream {
+    let trait_methods = functions.iter().map(ParsedHostFunction::trait_method);
+    let variants = functions
+        .iter()
+        .map(ParsedHostFunction::variant_declaration);
+    let spec_arms = functions.iter().map(ParsedHostFunction::spec_arm);
+    let all = functions.iter().map(|function| &function.variant);
+
+    quote! {
+        /// The host side of the wasm ABI: one method per function a guest may
+        /// import.
+        ///
+        /// Implement it once per execution environment — the ledger host, a test
+        /// double, a benchmark fake — and a guest module cannot tell them apart.
+        /// Each method is one declaration from the `host_functions!` block, as
+        /// written; its `&self` receiver is not part of the ABI the guest sees,
+        /// so a host that must mutate does so behind interior mutability.
+        ///
+        /// # The output contract
+        ///
+        /// A method handed an `out` buffer **writes into it only when the whole
+        /// value fits, and returns the value's true length whether it fitted or
+        /// not.**
+        ///
+        /// The length is the value's, not the number of bytes written, because it
+        /// is how a guest that asked with too small a buffer learns the size to
+        /// ask for next time. The engine turns a length past the buffer into
+        /// `BufferTooSmall`, and one past the field cap into `DataFieldTooLarge`,
+        /// so a host needs to know neither.
+        ///
+        /// Writing nothing unless the value fits is the half only a host can hold
+        /// up. An engine can bound how many bytes are *writable* — and does, by
+        /// handing over a region clamped to the field cap — but it cannot take
+        /// back what a method already put there. A host that wrote a truncated
+        /// prefix and then reported the larger length would leave those bytes in
+        /// guest memory behind a refusal the guest is told to ignore. C++'s
+        /// `setData` is the reference point: it wrote only on a value that fit.
+        pub trait HostFunctions {
+            #(#trait_methods)*
+        }
+
+        /// One row of the ABI table: what [`HostFunctionSpec::wasm_name`] and
+        /// [`HostFunctionSpec::gas`] read from.
+        ///
+        /// Private, and the only reason it exists is to keep both of them fed
+        /// from a single `match` over the declarations.
+        struct HostFnSpec {
+            name: &'static str,
+            gas: u64,
+        }
+
+        /// Identifies one host function, and is the compile-time source of its
+        /// ABI metadata.
+        ///
+        /// One variant per `host_functions!` declaration, named by converting the
+        /// function name to PascalCase. [`Self::ALL`] is the whole ABI, which is
+        /// what a wasm engine iterates to build its import table.
+        #[derive(Debug, Clone, Copy, PartialEq, Eq)]
+        pub enum HostFunctionSpec {
+            #(#variants,)*
+        }
+
+        impl HostFunctionSpec {
+            /// Every host function, in the order declared.
+            ///
+            /// This is the complete import surface a guest may link against: a
+            /// function absent here cannot be called, and one present here must
+            /// be registered for a module that imports it to instantiate.
+            pub const ALL: &'static [Self] = &[#(Self::#all,)*];
+
+            /// This function's row of the ABI table.
+            const fn spec(self) -> HostFnSpec {
+                match self {
+                    #(#spec_arms,)*
+                }
+            }
+
+            /// The name a guest imports this function under.
+            ///
+            /// A guest's import name must match this exactly, or the module
+            /// fails to instantiate. Usable in `const` context, so import lists
+            /// can be built at compile time.
+            pub const fn wasm_name(self) -> &'static str {
+                self.spec().name
+            }
+
+            /// Gas charged before the call runs, independent of its arguments.
+            ///
+            /// Consensus-relevant: two nodes that disagree on this value
+            /// disagree on transaction outcomes. Usable in `const` context, so
+            /// gas tables can be built at compile time.
+            pub const fn gas(self) -> u64 {
+                self.spec().gas
+            }
+        }
+    }
+}
+
+struct HostFunctionsInput {
+    functions: Vec<TraitItemFn>,
+}
+
+impl Parse for HostFunctionsInput {
+    fn parse(input: ParseStream) -> syn::Result<Self> {
+        let mut functions = Vec::new();
+        while !input.is_empty() {
+            functions.push(input.parse()?);
+        }
+        Ok(HostFunctionsInput { functions })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn accepts_an_empty_block() {
+        expand(quote! {}).unwrap();
+    }
+
+    #[test]
+    fn reports_mistakes_from_every_function() {
+        let error = expand(quote! {
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+
+            #[gas = 2000]
+            fn sha512_half(&self, data: &[u8]) -> HostResult<[u8; 32]>;
+        })
+        .expect_err("expected parsing to fail");
+
+        let messages: Vec<_> = error.into_iter().map(|error| error.to_string()).collect();
+        assert_eq!(messages.len(), 2, "{messages:?}");
+        assert!(messages[0].contains("missing `#[gas"), "{messages:?}");
+        assert!(messages[1].contains("missing `#[wasm_name"), "{messages:?}");
+    }
+
+    #[test]
+    fn propagates_syntax_errors() {
+        let error = expand(quote! { fn missing_semicolon() }).expect_err("expected a syntax error");
+        assert!(!error.to_string().is_empty());
+    }
+
+    /// The messages of every diagnostic recorded by one failed `expand`.
+    fn messages(input: TokenStream) -> Vec<String> {
+        let Err(error) = expand(input) else {
+            panic!("expected expansion to fail");
+        };
+        error.into_iter().map(|error| error.to_string()).collect()
+    }
+
+    #[test]
+    fn generates_the_trait_the_enum_and_the_table() {
+        let generated = expand(quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+
+            #[gas = 500]
+            #[wasm_name = "trace_num"]
+            fn trace_num(&self, msg: &str, number: i64) -> HostResult<()>;
+        })
+        .unwrap()
+        .to_string();
+
+        for expected in [
+            "pub trait HostFunctions",
+            "fn get_ledger_sqn (& self) -> HostResult < [u8 ; 4] > ;",
+            "fn trace_num (& self , msg : & str , number : i64) -> HostResult < () > ;",
+            "pub enum HostFunctionSpec { GetLedgerSqn , TraceNum , }",
+            "pub const ALL : & 'static [Self] = & [Self :: GetLedgerSqn , Self :: TraceNum ,]",
+            // The table's row type is generated too, and stays private.
+            "struct HostFnSpec { name : & 'static str , gas : u64 , }",
+            "const fn spec (self) -> HostFnSpec",
+            "Self :: GetLedgerSqn => HostFnSpec { name : \"ldgr_index\" , gas : 60u64 }",
+            "pub const fn wasm_name (self) -> & 'static str",
+            "pub const fn gas (self) -> u64",
+        ] {
+            assert!(generated.contains(expected), "missing {expected:?}");
+        }
+    }
+
+    /// The expansion stands alone: every name in it is either generated here or
+    /// written in the declarations, so it cannot depend on the crate it lands in.
+    #[test]
+    fn names_no_crate_of_its_own() {
+        let generated = expand(quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        })
+        .unwrap()
+        .to_string();
+
+        assert!(!generated.contains("xrpl_host_functions"), "{generated}");
+
+        // `Self::Variant` is the only path the expansion may build: anything else
+        // would reach out of the generated code. Doc comments spell paths without
+        // spaces (`Self::ALL`), so they do not match.
+        for (index, _) in generated.match_indices(" :: ") {
+            assert!(
+                generated[..index].ends_with("Self"),
+                "path out of the expansion at {index}: {generated}"
+            );
+        }
+    }
+
+    /// `spec` is an implementation detail of the two accessors, so it must not
+    /// become part of the ABI crate's public surface.
+    #[test]
+    fn keeps_the_table_row_private() {
+        let generated = expand(quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        })
+        .unwrap()
+        .to_string();
+
+        assert!(!generated.contains("pub struct HostFnSpec"), "{generated}");
+        assert!(!generated.contains("pub const fn spec"), "{generated}");
+    }
+
+    #[test]
+    fn rejects_two_functions_that_share_a_wasm_name() {
+        let messages = messages(quote! {
+            #[gas = 60]
+            #[wasm_name = "trace"]
+            fn trace(&self, msg: &str) -> HostResult<()>;
+
+            #[gas = 70]
+            #[wasm_name = "trace"]
+            fn trace_num(&self, msg: &str, number: i64) -> HostResult<()>;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(
+            messages[0].contains("already imported as `trace`"),
+            "{messages:?}"
+        );
+    }
+
+    /// Names that differ only in underscores collapse to one enum variant.
+    #[test]
+    fn rejects_two_functions_that_share_a_variant() {
+        let messages = messages(quote! {
+            #[gas = 60]
+            #[wasm_name = "a"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+
+            #[gas = 70]
+            #[wasm_name = "b"]
+            fn get_ledger__sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(
+            messages[0].contains("`GetLedgerSqn` variant"),
+            "{messages:?}"
+        );
+    }
+}

crates/xrpl-host-functions-macros/src/parsed_host_function.rs

@@ -0,0 +1,871 @@
+use proc_macro2::TokenStream;
+use quote::{format_ident, quote};
+use syn::{
+    Attribute, Expr, ExprLit, Ident, Lit, LitStr, PathArguments, ReceiverKind, ReturnType, Safety,
+    Signature, TraitItemFn, Type, TypePath,
+};
+
+use crate::errors;
+
+/// `#[gas = N]`: the base gas charged before the call runs.
+const GAS: &str = "gas";
+/// `#[wasm_name = "..."]`: the name the guest imports the function under.
+const WASM_NAME: &str = "wasm_name";
+/// `///` desugars to `#[doc = "..."]` before macro expansion.
+const DOC: &str = "doc";
+/// The alias every declaration returns its success type through.
+const HOST_RESULT: &str = "HostResult";
+
+/// One entry of a `host_functions!` block: its ABI metadata and its signature.
+pub(crate) struct ParsedHostFunction {
+    pub(crate) gas: u64,
+    /// Kept as the literal the user wrote, so diagnostics and the generated
+    /// string both carry that span.
+    pub(crate) wasm_name: LitStr,
+    /// Doc comments, in source order, to re-emit on the generated items.
+    pub(crate) docs: Vec<Attribute>,
+    /// The enum variant this declaration becomes, spanned at the function name.
+    pub(crate) variant: Ident,
+    pub(crate) signature: Signature,
+}
+
+impl ParsedHostFunction {
+    /// `#[doc …] fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;`
+    pub(crate) fn trait_method(&self) -> TokenStream {
+        let docs = &self.docs;
+        // The declaration is already a trait method: emitted verbatim, so what
+        // the block reads like is what the trait is.
+        let signature = &self.signature;
+
+        quote! {
+            #(#docs)*
+            #signature;
+        }
+    }
+
+    /// `#[doc …] GetLedgerSqn`
+    pub(crate) fn variant_declaration(&self) -> TokenStream {
+        let docs = &self.docs;
+        let variant = &self.variant;
+        quote! {
+            #(#docs)*
+            #variant
+        }
+    }
+
+    /// `Self::GetLedgerSqn => HostFnSpec { name: "ldgr_index", gas: 60u64 }`
+    pub(crate) fn spec_arm(&self) -> TokenStream {
+        let Self {
+            gas,
+            wasm_name,
+            variant,
+            ..
+        } = self;
+        quote! {
+            Self::#variant => HostFnSpec { name: #wasm_name, gas: #gas }
+        }
+    }
+
+    pub(crate) fn parse(function: TraitItemFn) -> syn::Result<Self> {
+        let mut gas = None;
+        let mut wasm_name = None;
+        let mut docs = Vec::new();
+        let mut errors = Vec::new();
+
+        // Tracked separately from `gas`/`wasm_name` so a malformed attribute is
+        // not also reported as a missing one.
+        let mut saw_gas = false;
+        let mut saw_wasm_name = false;
+
+        for attr in function.attrs {
+            if attr.path().is_ident(GAS) {
+                saw_gas = true;
+                if let Err(error) = int_value(&attr).and_then(|v| set_once(&mut gas, v, &attr)) {
+                    errors.push(error);
+                }
+            } else if attr.path().is_ident(WASM_NAME) {
+                saw_wasm_name = true;
+                if let Err(error) =
+                    string_value(&attr).and_then(|v| set_once(&mut wasm_name, v, &attr))
+                {
+                    errors.push(error);
+                }
+            } else if attr.path().is_ident(DOC) {
+                docs.push(attr);
+            } else {
+                errors.push(syn::Error::new_spanned(
+                    &attr,
+                    format!("unexpected attribute `{}`", path_name(&attr)),
+                ));
+            }
+        }
+
+        if !saw_gas {
+            errors.push(syn::Error::new_spanned(
+                &function.sig.ident,
+                format!("missing `#[{GAS} = ...]` attribute"),
+            ));
+        }
+        if !saw_wasm_name {
+            errors.push(syn::Error::new_spanned(
+                &function.sig.ident,
+                format!("missing `#[{WASM_NAME} = \"...\"]` attribute"),
+            ));
+        }
+        if let Some(body) = &function.default {
+            errors.push(syn::Error::new_spanned(
+                body,
+                "a host function is implemented by the host, so it must not have a body",
+            ));
+        }
+        if !function.sig.generics.params.is_empty() || function.sig.generics.where_clause.is_some()
+        {
+            errors.push(syn::Error::new_spanned(
+                &function.sig.ident,
+                "a host function must not be generic: it maps to one wasm import signature",
+            ));
+        }
+        errors.extend(check_receiver(&function.sig).err());
+        errors.extend(check_return_type(&function.sig).err());
+        if let Some(name) = &wasm_name {
+            errors.extend(check_wasm_name(name).err());
+        }
+        reject_modifiers(&function.sig, &mut errors);
+
+        // A name whose PascalCase form is not a legal variant is reported here
+        // rather than emitted, which would either panic or fail downstream.
+        let variant = match variant_ident(&function.sig.ident) {
+            Ok(variant) => Some(variant),
+            Err(error) => {
+                errors.push(error);
+                None
+            }
+        };
+
+        if let Some(error) = errors::combine(errors) {
+            return Err(error);
+        }
+
+        let (Some(gas), Some(wasm_name), Some(variant)) = (gas, wasm_name, variant) else {
+            unreachable!("every absent field is reported above");
+        };
+
+        Ok(Self {
+            gas,
+            wasm_name,
+            docs,
+            variant,
+            signature: function.sig,
+        })
+    }
+}
+
+/// Every declaration carries a receiver, and it is always `&self`.
+///
+/// `&self` is the only receiver that can work: the VM reaches the host through a
+/// shared `&dyn HostFunctions` stored in the wasmi `Store`, and a host that needs
+/// to mutate does so behind interior mutability. The receiver is not part of the
+/// wasm ABI — the guest passes no `self` — so it is uniform across the block.
+fn check_receiver(signature: &Signature) -> syn::Result<()> {
+    let Some(receiver) = signature.receiver() else {
+        return Err(syn::Error::new_spanned(
+            &signature.ident,
+            format!(
+                "a host function must declare its receiver: `fn {}(&self, ...)`",
+                signature.ident
+            ),
+        ));
+    };
+
+    // `&self` and nothing else: not `&mut self`, not `self`/`mut self`, not a
+    // typed `self: Box<Self>`, and not a spelled-out lifetime.
+    if !matches!(receiver.kind, ReceiverKind::Reference(_, None, None)) {
+        return Err(syn::Error::new_spanned(
+            receiver,
+            "a host function's receiver must be exactly `&self`: the VM calls the host \
+             through a shared `&dyn HostFunctions`",
+        ));
+    }
+    Ok(())
+}
+
+/// Every declaration returns `HostResult<T>`, including the ones that yield
+/// nothing (`HostResult<()>`).
+///
+/// One shape for every function is what lets a single dispatch adapter lower them
+/// all: lift the arguments out of guest memory, call the host, then turn `Ok(T)`
+/// into the wire's non-negative `i32` and `Err(e)` into a negative code or a trap.
+/// A function returning a bare `T` would need its own arm.
+fn check_return_type(signature: &Signature) -> syn::Result<()> {
+    const SHAPE: &str = "a host function must return `HostResult<T>` — \
+                         `HostResult<()>` if it yields nothing";
+
+    let ReturnType::Type(_, returned) = &signature.output else {
+        return Err(syn::Error::new_spanned(&signature.ident, SHAPE));
+    };
+
+    let Type::Path(TypePath {
+        qself: None, path, ..
+    }) = &**returned
+    else {
+        return Err(syn::Error::new_spanned(returned, SHAPE));
+    };
+    // The last segment only, so `HostResult<T>` may be written qualified.
+    let Some(last) = path.segments.last() else {
+        return Err(syn::Error::new_spanned(returned, SHAPE));
+    };
+    if last.ident != HOST_RESULT {
+        return Err(syn::Error::new_spanned(returned, SHAPE));
+    }
+
+    // `HostResult` without its success type is `HostResult` the alias, which names
+    // no type; rustc's own message for that is unhelpfully far from the cause.
+    let PathArguments::AngleBracketed(arguments) = &last.arguments else {
+        return Err(syn::Error::new_spanned(
+            returned,
+            format!("`{HOST_RESULT}` needs its success type: `{HOST_RESULT}<T>`"),
+        ));
+    };
+    if arguments.args.len() != 1 {
+        return Err(syn::Error::new_spanned(
+            arguments,
+            format!("`{HOST_RESULT}` takes exactly one type: `{HOST_RESULT}<T>`"),
+        ));
+    }
+    Ok(())
+}
+
+/// `const`, `async`, `unsafe`/`safe` and `extern "…"` have no meaning in the
+/// wasm ABI, and would otherwise pass silently into the generated trait.
+fn reject_modifiers(signature: &Signature, errors: &mut Vec<syn::Error>) {
+    const PLAIN: &str =
+        "a host function must be a plain `fn`: this modifier is not part of the wasm ABI";
+
+    if let Some(constness) = &signature.constness {
+        errors.push(syn::Error::new_spanned(constness, PLAIN));
+    }
+    if let Some(asyncness) = &signature.asyncness {
+        errors.push(syn::Error::new_spanned(asyncness, PLAIN));
+    }
+    match &signature.safety {
+        Safety::Default => {}
+        Safety::Safe(token) => errors.push(syn::Error::new_spanned(token, PLAIN)),
+        Safety::Unsafe(token) => errors.push(syn::Error::new_spanned(token, PLAIN)),
+    }
+    if let Some(abi) = &signature.abi {
+        errors.push(syn::Error::new_spanned(abi, PLAIN));
+    }
+}
+
+/// The wasm import name reaches the engine's import table verbatim, so it is
+/// held to what an import name can sanely be rather than to any string.
+fn check_wasm_name(name: &LitStr) -> syn::Result<()> {
+    let value = name.value();
+    if value.is_empty() {
+        return Err(syn::Error::new_spanned(
+            name,
+            "the wasm name must not be empty",
+        ));
+    }
+    if let Some(character) = value
+        .chars()
+        .find(|c| !c.is_ascii_alphanumeric() && *c != '_')
+    {
+        return Err(syn::Error::new_spanned(
+            name,
+            format!(
+                "a wasm name may only contain `A-Za-z0-9_`, but this one contains {character:?}"
+            ),
+        ));
+    }
+    Ok(())
+}
+
+/// The enum variant a declaration becomes: `get_ledger_sqn` -> `GetLedgerSqn`.
+///
+/// The result carries `ident`'s span, so anything the compiler says about the
+/// variant points at the declaration that produced it.
+fn variant_ident(ident: &Ident) -> syn::Result<Ident> {
+    // `to_string` spells raw identifiers `r#type`; the `r#` is not part of the name.
+    let name = ident.to_string();
+    let name = name.strip_prefix("r#").unwrap_or(&name);
+
+    let mut pascal = String::with_capacity(name.len());
+    let mut capitalize = true;
+    for character in name.chars() {
+        if character == '_' {
+            capitalize = true;
+        } else if capitalize {
+            pascal.extend(character.to_uppercase());
+            capitalize = false;
+        } else {
+            pascal.push(character);
+        }
+    }
+
+    // A name of nothing but underscores leaves `pascal` empty; the original is
+    // already a legal identifier, so keep it.
+    if pascal.is_empty() {
+        return Ok(ident.clone());
+    }
+
+    // `Ident::new` panics on a leading digit (`_2fa` -> `2fa`) and silently
+    // accepts keyword spellings (`self_` -> `Self`), which then fails to parse
+    // where the variant is emitted. Parsing rejects both, without panicking.
+    if let Err(error) = syn::parse_str::<Ident>(&pascal) {
+        return Err(syn::Error::new_spanned(
+            ident,
+            format!(
+                "this name becomes the enum variant `{pascal}`, which is not a valid \
+                 variant name ({error}); rename the host function"
+            ),
+        ));
+    }
+    Ok(format_ident!("{pascal}", span = ident.span()))
+}
+
+/// Records `value`, or reports that the attribute appeared more than once.
+fn set_once<T>(slot: &mut Option<T>, value: T, attr: &Attribute) -> syn::Result<()> {
+    if slot.replace(value).is_some() {
+        return Err(syn::Error::new_spanned(
+            attr,
+            format!("duplicate `{}` attribute", path_name(attr)),
+        ));
+    }
+    Ok(())
+}
+
+fn int_value(attr: &Attribute) -> syn::Result<u64> {
+    match &attr.meta.require_name_value()?.value {
+        Expr::Lit(ExprLit {
+            lit: Lit::Int(int), ..
+        }) => {
+            // `LitInt` keeps the sign in its digits, so `base10_parse::<u64>`
+            // would report a negative value as "invalid digit found in string".
+            if int.base10_digits().starts_with('-') {
+                return Err(syn::Error::new_spanned(
+                    int,
+                    format!("`{}` must not be negative", path_name(attr)),
+                ));
+            }
+            int.base10_parse()
+        }
+        other => Err(syn::Error::new_spanned(
+            other,
+            format!("`{}` expects an integer literal", path_name(attr)),
+        )),
+    }
+}
+
+fn string_value(attr: &Attribute) -> syn::Result<LitStr> {
+    match &attr.meta.require_name_value()?.value {
+        Expr::Lit(ExprLit {
+            lit: Lit::Str(string),
+            ..
+        }) => Ok(string.clone()),
+        other => Err(syn::Error::new_spanned(
+            other,
+            format!("`{}` expects a string literal", path_name(attr)),
+        )),
+    }
+}
+
+/// The attribute's path as written, for diagnostics: `gas`, or `foo::bar`.
+fn path_name(attr: &Attribute) -> String {
+    attr.path()
+        .segments
+        .iter()
+        .map(|segment| segment.ident.to_string())
+        .collect::<Vec<_>>()
+        .join("::")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use quote::ToTokens;
+    use syn::parse_quote;
+
+    /// The message of every diagnostic recorded by one failed `parse`.
+    ///
+    /// `expect_err` is unavailable here: it needs `T: Debug`, and syn only
+    /// implements `Debug` for its AST types under the `extra-traits` feature.
+    fn messages(function: TraitItemFn) -> Vec<String> {
+        let Err(error) = ParsedHostFunction::parse(function) else {
+            panic!("expected parsing to fail");
+        };
+        error.into_iter().map(|error| error.to_string()).collect()
+    }
+
+    fn doc_text(attr: &Attribute) -> String {
+        match &attr.meta.require_name_value().unwrap().value {
+            Expr::Lit(ExprLit {
+                lit: Lit::Str(text),
+                ..
+            }) => text.value(),
+            _ => panic!("doc attribute is not a string literal"),
+        }
+    }
+
+    #[test]
+    fn reads_gas_and_wasm_name() {
+        let parsed = ParsedHostFunction::parse(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        })
+        .unwrap();
+
+        assert_eq!(parsed.gas, 60);
+        assert_eq!(parsed.wasm_name.value(), "ldgr_index");
+        assert_eq!(parsed.signature.ident.to_string(), "get_ledger_sqn");
+        assert_eq!(parsed.variant.to_string(), "GetLedgerSqn");
+        assert!(parsed.docs.is_empty());
+    }
+
+    #[test]
+    fn derives_variant_names_from_function_names() {
+        for (function, variant) in [
+            ("get_ledger_sqn", "GetLedgerSqn"),
+            ("sha512_half", "Sha512Half"),
+            ("trace", "Trace"),
+            ("get_current_ledger_obj_field", "GetCurrentLedgerObjField"),
+            ("r#type", "Type"),
+            ("trace2", "Trace2"),
+            // Pathological, but must not panic: no letters to capitalize.
+            ("__", "__"),
+        ] {
+            let ident = format_ident!("{function}");
+            assert_eq!(
+                variant_ident(&ident).map(|v| v.to_string()).ok(),
+                Some(variant.to_owned()),
+                "{function}"
+            );
+        }
+    }
+
+    /// `_2fa` would PascalCase to `2fa`; building that `Ident` panics, and a
+    /// panic in a proc macro is reported with no useful span at all.
+    #[test]
+    fn rejects_a_name_that_becomes_a_leading_digit() {
+        let messages = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "two_factor"]
+            fn _2fa(&self) -> HostResult<()>;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(
+            messages[0].contains("becomes the enum variant `2fa`"),
+            "{messages:?}"
+        );
+    }
+
+    /// `self_` PascalCases to `Self`, which `Ident::new` accepts and rustc then
+    /// rejects where the variant is emitted. `r#Self` is not a legal escape.
+    #[test]
+    fn rejects_a_name_that_becomes_a_keyword() {
+        for function in ["self_", "_self"] {
+            let ident = format_ident!("{function}");
+            let Err(error) = variant_ident(&ident) else {
+                panic!("expected `{function}` to be rejected");
+            };
+            assert!(
+                error.to_string().contains("variant `Self`"),
+                "{}",
+                error.to_string()
+            );
+        }
+    }
+
+    #[test]
+    fn rejects_negative_gas() {
+        let messages = messages(parse_quote! {
+            #[gas = -5]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert_eq!(messages[0], "`gas` must not be negative");
+    }
+
+    #[test]
+    fn rejects_unusable_wasm_names() {
+        let empty = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = ""]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+        assert_eq!(empty.len(), 1, "{empty:?}");
+        assert_eq!(empty[0], "the wasm name must not be empty");
+
+        let spaced = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+        assert_eq!(spaced.len(), 1, "{spaced:?}");
+        assert!(spaced[0].contains("may only contain"), "{spaced:?}");
+    }
+
+    #[test]
+    fn rejects_signature_modifiers() {
+        for declaration in [
+            quote! { unsafe fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>; },
+            quote! { async fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>; },
+            quote! { const fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>; },
+            quote! { extern "C" fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>; },
+        ] {
+            let function: TraitItemFn = syn::parse2(quote! {
+                #[gas = 60]
+                #[wasm_name = "ldgr_index"]
+                #declaration
+            })
+            .unwrap();
+
+            let messages = messages(function);
+            assert_eq!(messages.len(), 1, "{messages:?}");
+            assert!(messages[0].contains("must be a plain `fn`"), "{messages:?}");
+        }
+    }
+
+    #[test]
+    fn trait_method_keeps_the_declared_receiver_and_ends_in_a_semicolon() {
+        let parsed = ParsedHostFunction::parse(parse_quote! {
+            /// Hashes `data`.
+            #[gas = 2000]
+            #[wasm_name = "sha512_half"]
+            fn sha512_half(&self, data: &[u8]) -> HostResult<[u8; 32]>;
+        })
+        .unwrap();
+
+        // `///` reaches the macro as `#[doc = r"..."]`: rustc's lexer spells doc
+        // comments as raw string literals.
+        let method = parsed.trait_method().to_string();
+        assert!(
+            method.starts_with("# [doc = r\" Hashes `data`.\"]"),
+            "{method}"
+        );
+        assert!(
+            method
+                .contains("fn sha512_half (& self , data : & [u8]) -> HostResult < [u8 ; 32] > ;"),
+            "{method}"
+        );
+    }
+
+    #[test]
+    fn spec_arm_carries_the_name_and_the_gas() {
+        let parsed = ParsedHostFunction::parse(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        })
+        .unwrap();
+
+        assert_eq!(
+            parsed.spec_arm().to_string(),
+            "Self :: GetLedgerSqn => HostFnSpec { name : \"ldgr_index\" , gas : 60u64 }"
+        );
+    }
+
+    #[test]
+    fn keeps_doc_comments_in_source_order() {
+        let parsed = ParsedHostFunction::parse(parse_quote! {
+            /// First line.
+            ///
+            /// Third line.
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        })
+        .unwrap();
+
+        let docs: Vec<_> = parsed.docs.iter().map(doc_text).collect();
+        assert_eq!(docs, vec![" First line.", "", " Third line."]);
+    }
+
+    #[test]
+    fn preserves_parameters_and_return_type() {
+        let traced = ParsedHostFunction::parse(parse_quote! {
+            #[gas = 500]
+            #[wasm_name = "trace"]
+            fn trace(&self, msg: &str, data: &[u8], as_hex: bool) -> HostResult<()>;
+        })
+        .unwrap();
+        // The receiver is `inputs[0]`; the three wasm parameters follow it.
+        assert_eq!(traced.signature.inputs.len(), 4);
+        assert_eq!(
+            traced.signature.output.to_token_stream().to_string(),
+            "-> HostResult < () >"
+        );
+
+        let hashed = ParsedHostFunction::parse(parse_quote! {
+            #[gas = 2000]
+            #[wasm_name = "sha512_half"]
+            fn sha512_half(&self, data: &[u8]) -> HostResult<[u8; HASH_LEN]>;
+        })
+        .unwrap();
+        assert_eq!(
+            hashed.signature.output.to_token_stream().to_string(),
+            "-> HostResult < [u8 ; HASH_LEN] >"
+        );
+    }
+
+    #[test]
+    fn reports_both_missing_attributes_at_once() {
+        let messages = messages(parse_quote! {
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 2);
+        assert!(messages[0].contains("missing `#[gas"), "{messages:?}");
+        assert!(messages[1].contains("missing `#[wasm_name"), "{messages:?}");
+    }
+
+    #[test]
+    fn names_the_unexpected_attribute() {
+        let messages = messages(parse_quote! {
+            #[gas = 60]
+            #[wsam_name = "typo"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        // The typo'd attribute, plus the `wasm_name` it failed to be.
+        assert_eq!(messages.len(), 2);
+        assert!(
+            messages.iter().any(|m| m.contains("`wsam_name`")),
+            "{messages:?}"
+        );
+    }
+
+    #[test]
+    fn rejects_wrong_literal_types() {
+        let gas = messages(parse_quote! {
+            #[gas = "60"]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+        assert_eq!(gas.len(), 1, "{gas:?}");
+        assert!(
+            gas[0].contains("`gas` expects an integer literal"),
+            "{gas:?}"
+        );
+
+        let name = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = 7]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+        assert_eq!(name.len(), 1, "{name:?}");
+        assert!(
+            name[0].contains("`wasm_name` expects a string literal"),
+            "{name:?}"
+        );
+    }
+
+    #[test]
+    fn rejects_gas_that_does_not_fit_in_u64() {
+        let messages = messages(parse_quote! {
+            #[gas = 99999999999999999999999]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(messages[0].contains("number too large"), "{messages:?}");
+    }
+
+    #[test]
+    fn rejects_attribute_shapes_other_than_name_value() {
+        let bare = messages(parse_quote! {
+            #[gas]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+        assert_eq!(bare.len(), 1, "{bare:?}");
+        assert!(bare[0].contains("gas = ..."), "{bare:?}");
+
+        let list = messages(parse_quote! {
+            #[gas(60)]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+        assert_eq!(list.len(), 1, "{list:?}");
+    }
+
+    #[test]
+    fn rejects_duplicate_attributes() {
+        let messages = messages(parse_quote! {
+            #[gas = 60]
+            #[gas = 70]
+            #[wasm_name = "ldgr_index"]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 2, "{messages:?}");
+        assert!(messages[0].contains("duplicate `gas`"), "{messages:?}");
+        assert!(
+            messages[1].contains("duplicate `wasm_name`"),
+            "{messages:?}"
+        );
+    }
+
+    /// A malformed attribute must not also be reported as an absent one.
+    #[test]
+    fn does_not_report_a_malformed_attribute_as_missing() {
+        let messages = messages(parse_quote! {
+            #[gas = "60"]
+            #[wasm_name = 7]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 2, "{messages:?}");
+        assert!(
+            !messages.iter().any(|m| m.contains("missing")),
+            "{messages:?}"
+        );
+    }
+
+    #[test]
+    fn rejects_a_body() {
+        let messages = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]> { Ok([0; 4]) }
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(messages[0].contains("must not have a body"), "{messages:?}");
+    }
+
+    #[test]
+    fn rejects_generics() {
+        let parameter = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn<T>(&self) -> HostResult<T>;
+        });
+        assert_eq!(parameter.len(), 1, "{parameter:?}");
+        assert!(
+            parameter[0].contains("must not be generic"),
+            "{parameter:?}"
+        );
+
+        let clause = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult<[u8; 4]> where Self: Sized;
+        });
+        assert_eq!(clause.len(), 1, "{clause:?}");
+    }
+
+    #[test]
+    fn requires_a_receiver() {
+        let messages = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn() -> HostResult<[u8; 4]>;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(
+            messages[0].contains("must declare its receiver: `fn get_ledger_sqn(&self, ...)`"),
+            "{messages:?}"
+        );
+    }
+
+    /// Anything but `&self` would need a host the VM cannot hand out: it holds
+    /// one shared `&dyn HostFunctions` for the whole run.
+    #[test]
+    fn rejects_receivers_other_than_shared_self() {
+        for receiver in [
+            quote! { &mut self },
+            quote! { self },
+            quote! { mut self },
+            quote! { self: Box<Self> },
+            quote! { &'a self },
+        ] {
+            let function: TraitItemFn = syn::parse2(quote! {
+                #[gas = 60]
+                #[wasm_name = "ldgr_index"]
+                fn get_ledger_sqn(#receiver) -> HostResult<[u8; 4]>;
+            })
+            .unwrap_or_else(|_| panic!("`{receiver}` should parse"));
+
+            let messages = messages(function);
+            assert_eq!(messages.len(), 1, "`{receiver}`: {messages:?}");
+            assert!(
+                messages[0].contains("must be exactly `&self`"),
+                "`{receiver}`: {messages:?}"
+            );
+        }
+    }
+
+    /// A bare `T` return would need its own lowering arm, so the uniform shape is
+    /// required rather than inferred.
+    #[test]
+    fn rejects_returns_that_are_not_host_result() {
+        for output in [
+            quote! {},
+            quote! { -> () },
+            quote! { -> [u8; 4] },
+            quote! { -> i32 },
+            quote! { -> Result<[u8; 4], HostError> },
+            quote! { -> impl Iterator<Item = u8> },
+        ] {
+            let function: TraitItemFn = syn::parse2(quote! {
+                #[gas = 60]
+                #[wasm_name = "ldgr_index"]
+                fn get_ledger_sqn(&self) #output;
+            })
+            .unwrap_or_else(|_| panic!("`{output}` should parse"));
+
+            let messages = messages(function);
+            assert_eq!(messages.len(), 1, "`{output}`: {messages:?}");
+            assert!(
+                messages[0].contains("must return `HostResult<T>`"),
+                "`{output}`: {messages:?}"
+            );
+        }
+    }
+
+    /// `HostResult` may be written qualified, since the trait method keeps whatever
+    /// path resolves where the block is written.
+    #[test]
+    fn accepts_a_qualified_host_result() {
+        let parsed = ParsedHostFunction::parse(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> xrpl_host_functions::HostResult<[u8; 4]>;
+        })
+        .unwrap();
+
+        assert!(
+            parsed
+                .trait_method()
+                .to_string()
+                .contains("xrpl_host_functions :: HostResult < [u8 ; 4] >"),
+            "{}",
+            parsed.trait_method()
+        );
+    }
+
+    /// `HostResult` with no success type names no type at all; rustc's own error
+    /// for that lands on the generated trait, far from the declaration.
+    #[test]
+    fn rejects_host_result_without_a_success_type() {
+        let messages = messages(parse_quote! {
+            #[gas = 60]
+            #[wasm_name = "ldgr_index"]
+            fn get_ledger_sqn(&self) -> HostResult;
+        });
+
+        assert_eq!(messages.len(), 1, "{messages:?}");
+        assert!(
+            messages[0].contains("needs its success type"),
+            "{messages:?}"
+        );
+    }
+}

crates/xrpl-host-functions/Cargo.toml

@@ -0,0 +1,7 @@
+[package]
+name = "xrpl-host-functions"
+version = "0.1.0"
+edition.workspace = true
+
+[dependencies]
+xrpl-host-functions-macros.path = "../xrpl-host-functions-macros"

crates/xrpl-host-functions/src/lib.rs

@@ -0,0 +1,539 @@
+//! The wasm host ABI: the one place it is declared.
+//!
+//! `host_functions!` turns the declaration block at the bottom of this file into the
+//! [`HostFunctions`] trait a host implements and the [`HostFunctionSpec`] table a
+//! wasm engine registers from.
+//!
+//! The split: hand-written here is the vocabulary the declarations are written in —
+//! [`HostError`], [`TraceDataType`], [`HostResult`], [`HASH_LEN`] — and everything
+//! derived from the declarations is generated. The expansion names nothing this file
+//! does not, so the two sides meet only in the block below.
+//!
+//! So this file is lists — error codes, trace data types, functions. The `macro_rules!`
+//! that expand the first two into enums live in `macros.rs`.
+
+#![no_std]
+
+#[macro_use]
+mod macros;
+
+// Not re-exported: the ABI is declared once, here, and this is the only call site.
+use xrpl_host_functions_macros::host_functions;
+
+host_errors! {
+    Unimplemented = -1,
+    FieldNotFound = -2,
+    BufferTooSmall = -3,
+    NoArray = -4,
+    NotLeafField = -5,
+    LocatorMalformed = -6,
+    SlotOutRange = -7,
+    SlotsFull = -8,
+    EmptySlot = -9,
+    LedgerObjNotFound = -10,
+    OutOfTransferLimit = -11,
+    DataFieldTooLarge = -12,
+    PointerOutOfBounds = -13,
+    NoMemExported = -14,
+    InvalidParams = -15,
+    InvalidAccount = -16,
+    InvalidField = -17,
+    IndexOutOfBounds = -18,
+    FloatInputMalformed = -19,
+    FloatComputationError = -20,
+    /// Internal fatal error.
+    /// User code will never see this error but keep it reserved to not rely on the value.
+    InternalFatal = -2147483648,
+}
+
+/// Convenience alias for the trait's fallible returns.
+pub type HostResult<T> = Result<T, HostError>;
+
+/// A `sha512Half` digest: the first 32 bytes of a SHA-512, as XRPL uses it.
+pub const HASH_LEN: usize = 32;
+
+trace_data_types! {
+    /// 8 little-endian bytes, rendered as a signed decimal.
+    Int64 = 1,
+    /// 8 little-endian bytes, rendered as an unsigned decimal.
+    Uint64 = 2,
+    /// A serialized XRPL float: 12 bytes, mantissa then exponent.
+    Xfloat = 3,
+    /// A 20-byte account ID, rendered as base58.
+    Account = 4,
+    /// A serialized `STAmount`.
+    Amount = 5,
+    /// Raw bytes, hex-encoded.
+    AsHex = 6,
+    /// Bytes rendered verbatim as text.
+    AsText = 7,
+}
+
+host_functions! {
+    /// The sequence number of the ledger being built, as 4 little-endian bytes.
+    #[gas = 60]
+    #[wasm_name = "ldgr_index"]
+    fn get_ledger_sqn(&self, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The close time of the parent (last-closed) ledger, as 4 little-endian bytes.
+    #[gas = 60]
+    #[wasm_name = "parent_ldgr_time"]
+    fn get_parent_ledger_time(&self, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The hash of the parent (last-closed) ledger, as 32 bytes.
+    #[gas = 60]
+    #[wasm_name = "parent_ldgr_hash"]
+    fn get_parent_ledger_hash(&self, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The base fee of the ledger being built, in drops, as 4 little-endian bytes.
+    #[gas = 60]
+    #[wasm_name = "base_fee"]
+    fn get_base_fee(&self, out: &mut [u8]) -> HostResult<usize>;
+
+    /// Whether an amendment is enabled. The input is either its 32-byte id or its
+    /// name; the answer is `1` if enabled and `0` if not. Unlike the getters, this
+    /// reads an input region and returns the flag directly rather than writing bytes.
+    #[gas = 100]
+    #[wasm_name = "amendment_enabled"]
+    fn is_amendment_enabled(&self, amendment: &[u8]) -> HostResult<i32>;
+
+    /// Load the ledger object with the given 32-byte id into a cache slot, so later
+    /// calls can read its fields. `cache_idx` selects the slot (1-based); `0` asks the
+    /// host to assign a free one. Returns the slot used, or a negative error.
+    #[gas = 5000]
+    #[wasm_name = "cache_le"]
+    fn cache_ledger_obj(&self, obj_id: &[u8], cache_idx: i32) -> HostResult<i32>;
+
+    /// The serialized bytes of one field of the transaction being executed, selected
+    /// by its `SField` code.
+    #[gas = 70]
+    #[wasm_name = "tx_field"]
+    fn get_tx_field(&self, field: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The serialized bytes of one field of the current (escrow) ledger object.
+    #[gas = 70]
+    #[wasm_name = "home_le_field"]
+    fn get_current_ledger_obj_field(&self, field: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The serialized bytes of one field of a previously cached ledger object,
+    /// selected by its cache slot and the field's `SField` code.
+    #[gas = 70]
+    #[wasm_name = "le_field"]
+    fn get_ledger_obj_field(&self, cache_idx: i32, field: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The serialized bytes of a nested field of the transaction, reached by a
+    /// `locator`: a path of little-endian `i32` steps (so its byte length is a
+    /// non-zero multiple of 4). Reads the locator region and writes the field bytes.
+    #[gas = 110]
+    #[wasm_name = "tx_inner"]
+    fn get_tx_nested_field(&self, locator: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The serialized bytes of a nested field of the current (escrow) ledger object,
+    /// reached by a `locator`, as with [`HostFunctions::get_tx_nested_field`].
+    #[gas = 110]
+    #[wasm_name = "home_le_inner"]
+    fn get_current_ledger_obj_nested_field(
+        &self,
+        locator: &[u8],
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The serialized bytes of a nested field of a previously cached ledger object,
+    /// selected by its cache slot and reached by a `locator`.
+    #[gas = 110]
+    #[wasm_name = "le_inner"]
+    fn get_ledger_obj_nested_field(
+        &self,
+        cache_idx: i32,
+        locator: &[u8],
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The number of elements in an array field of the transaction, selected by its
+    /// `SField` code. Answers the count directly, or a negative error (`NoArray` if
+    /// the field is not an array). Reads and writes no memory.
+    #[gas = 40]
+    #[wasm_name = "tx_arr_len"]
+    fn get_tx_array_len(&self, field: i32) -> HostResult<i32>;
+
+    /// The number of elements in an array field of the current (escrow) ledger
+    /// object, as with [`HostFunctions::get_tx_array_len`].
+    #[gas = 40]
+    #[wasm_name = "home_le_arr_len"]
+    fn get_current_ledger_obj_array_len(&self, field: i32) -> HostResult<i32>;
+
+    /// The number of elements in an array field of a previously cached ledger object,
+    /// selected by its cache slot and `SField` code.
+    #[gas = 40]
+    #[wasm_name = "le_arr_len"]
+    fn get_ledger_obj_array_len(&self, cache_idx: i32, field: i32) -> HostResult<i32>;
+
+    /// The number of elements in a nested array field of the transaction, reached by a
+    /// `locator`. Reads the locator region and answers the count directly.
+    #[gas = 70]
+    #[wasm_name = "tx_inner_arr_len"]
+    fn get_tx_nested_array_len(&self, locator: &[u8]) -> HostResult<i32>;
+
+    /// The number of elements in a nested array field of the current (escrow) ledger
+    /// object, reached by a `locator`, as with [`HostFunctions::get_tx_nested_array_len`].
+    #[gas = 70]
+    #[wasm_name = "home_le_inner_arr_len"]
+    fn get_current_ledger_obj_nested_array_len(&self, locator: &[u8]) -> HostResult<i32>;
+
+    /// The number of elements in a nested array field of a previously cached ledger
+    /// object, selected by its cache slot and reached by a `locator`.
+    #[gas = 70]
+    #[wasm_name = "le_inner_arr_len"]
+    fn get_ledger_obj_nested_array_len(&self, cache_idx: i32, locator: &[u8]) -> HostResult<i32>;
+
+    /// Verify `signature` over `message` under `pubkey`. Reads the three regions and
+    /// answers `1` if the signature is valid, `0` if not, or a negative error.
+    ///
+    /// GAS DISCREPANCY: this 300 is the value the C-ABI fork registered
+    /// (`rippled-wasm-host-functions`, WasmVM.cpp), which this port follows. The
+    /// prior C++ integration in this tree charged 35000 for the same call — 100x
+    /// more, and closer to the real cost of signature verification. The value is
+    /// consensus-critical, so confirm which is intended before this ships.
+    #[gas = 300]
+    #[wasm_name = "check_sig"]
+    fn check_signature(
+        &self,
+        message: &[u8],
+        signature: &[u8],
+        pubkey: &[u8],
+    ) -> HostResult<i32>;
+
+    /// The 32-byte ledger key (keylet) of an account's `AccountRoot`, computed from a
+    /// 20-byte account id. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "accountroot_id"]
+    fn account_keylet(&self, account: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an AMM, computed from its two assets. Each asset is a
+    /// byte slice whose length selects its kind (24 = MPT, 20 = XRP, 40 = issued
+    /// currency + issuer). Reads both asset regions and writes the keylet.
+    #[gas = 450]
+    #[wasm_name = "amm_id"]
+    fn amm_keylet(&self, asset1: &[u8], asset2: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `Check`, computed from a 20-byte account id and its
+    /// sequence number. `seq` is the guest's `u32` carried as its `i32` bit pattern.
+    /// Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "check_id"]
+    fn check_keylet(&self, account: &[u8], seq: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `Credential`, computed from the 20-byte subject and
+    /// issuer account ids and a credential-type byte string. Reads all three regions
+    /// and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "credential_id"]
+    fn credential_keylet(
+        &self,
+        subject: &[u8],
+        issuer: &[u8],
+        credential_type: &[u8],
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `Delegate` object, computed from the 20-byte account
+    /// and the account it authorizes. Reads both account regions and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "delegate_id"]
+    fn delegate_keylet(
+        &self,
+        account: &[u8],
+        authorize: &[u8],
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `DepositPreauth`, computed from the 20-byte account and
+    /// the account it authorizes to deposit. Reads both account regions and writes the
+    /// keylet.
+    #[gas = 350]
+    #[wasm_name = "deposit_preauth_id"]
+    fn deposit_preauth_keylet(
+        &self,
+        account: &[u8],
+        authorize: &[u8],
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an account's `DID`, computed from its 20-byte account id.
+    /// Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "did_id"]
+    fn did_keylet(&self, account: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an `Escrow`, computed from the 20-byte owner account and
+    /// its sequence number. `seq` is the guest's `u32` carried as its `i32` bit
+    /// pattern. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "escrow_id"]
+    fn escrow_keylet(&self, account: &[u8], seq: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `RippleState` (trust line), computed from two 20-byte
+    /// account ids and a 20-byte currency. Reads all three regions and writes the
+    /// keylet.
+    #[gas = 400]
+    #[wasm_name = "trustline_id"]
+    fn trust_line_keylet(
+        &self,
+        account1: &[u8],
+        account2: &[u8],
+        currency: &[u8],
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an `MPTokenIssuance`, computed from the 20-byte issuer
+    /// account and its sequence number. `seq` is the guest's `u32` carried as its
+    /// `i32` bit pattern. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "mpt_issuance_id"]
+    fn mptoken_issuance_keylet(
+        &self,
+        issuer: &[u8],
+        seq: i32,
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an `MPToken`, computed from a 24-byte MPT issuance id and
+    /// the 20-byte holder account. Reads both regions and writes the keylet.
+    #[gas = 500]
+    #[wasm_name = "mptoken_id"]
+    fn mptoken_keylet(&self, mptid: &[u8], holder: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an `NFTokenOffer`, computed from the 20-byte owner account
+    /// and its sequence number. `seq` is the guest's `u32` carried as its `i32` bit
+    /// pattern. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "nft_offer_id"]
+    fn nftoken_offer_keylet(
+        &self,
+        account: &[u8],
+        seq: i32,
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an `Offer`, computed from the 20-byte owner account and
+    /// its sequence number. `seq` is the guest's `u32` carried as its `i32` bit
+    /// pattern. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "offer_id"]
+    fn offer_keylet(&self, account: &[u8], seq: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of an `Oracle`, computed from the 20-byte owner account and
+    /// its document id. `doc_id` is the guest's `u32` carried as its `i32` bit pattern.
+    /// Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "oracle_id"]
+    fn oracle_keylet(&self, account: &[u8], doc_id: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `PayChannel`, computed from the 20-byte source account,
+    /// the 20-byte destination account, and the channel's sequence number. `seq` is the
+    /// guest's `u32` carried as its `i32` bit pattern. Reads both account regions and
+    /// writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "paychan_id"]
+    fn paychannel_keylet(
+        &self,
+        account: &[u8],
+        destination: &[u8],
+        seq: i32,
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `PermissionedDomain`, computed from the 20-byte owner
+    /// account and its sequence number. `seq` is the guest's `u32` carried as its `i32`
+    /// bit pattern. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "permissioned_domain_id"]
+    fn permissioned_domain_keylet(
+        &self,
+        account: &[u8],
+        seq: i32,
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `SignerList`, computed from its 20-byte owner account.
+    /// Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "signers_id"]
+    fn signer_list_keylet(&self, account: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `Ticket`, computed from the 20-byte owner account and its
+    /// ticket sequence number. `seq` is the guest's `u32` carried as its `i32` bit
+    /// pattern. Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "ticket_id"]
+    fn ticket_keylet(&self, account: &[u8], seq: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 32-byte keylet of a `Vault`, computed from the 20-byte owner account and its
+    /// sequence number. `seq` is the guest's `u32` carried as its `i32` bit pattern.
+    /// Reads the account region and writes the keylet.
+    #[gas = 350]
+    #[wasm_name = "vault_id"]
+    fn vault_keylet(&self, account: &[u8], seq: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The XRPL `sha512Half` of `data`: the first [`HASH_LEN`] bytes of its SHA-512.
+    #[gas = 2000]
+    #[wasm_name = "sha512_half"]
+    fn sha512_half(&self, data: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// Writes `msg` to the trace log, followed by `data` rendered as `data_type` says.
+    ///
+    /// The one declaration whose wasm function has **no result**: this node's own log
+    /// is its only effect, so a guest is told nothing. An `Err` from a host therefore
+    /// reaches it in no form, and only the host-fatal ones do anything at all.
+    ///
+    /// It is also the one declaration that is **not** the wasm parameter order.
+    /// `data_type` is the third wasm parameter, between the two regions, because that
+    /// is where xrpld's `trace_proto` and the guest stdlib put it; `register.rs` takes
+    /// the arguments in wasm order and calls this in declaration order.
+    #[gas = 30]
+    #[wasm_name = "trace"]
+    fn trace(&self, msg: &str, data: &[u8], data_type: TraceDataType) -> HostResult<()>;
+
+    /// Stores `data` as the current object's data field, replacing whatever was there,
+    /// and returns the number of bytes stored. Reads the data region; `DataFieldTooLarge`
+    /// if it exceeds the host's limit.
+    #[gas = 1000]
+    #[wasm_name = "set_data"]
+    fn update_data(&self, data: &[u8]) -> HostResult<i32>;
+
+    /// The URI of the `NFToken` with id `nft_id` (32 bytes) held by the 20-byte
+    /// `account`. Reads both regions and writes the URI bytes.
+    #[gas = 5000]
+    #[wasm_name = "nft_uri"]
+    fn get_nft(&self, account: &[u8], nft_id: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The 20-byte issuer account encoded in the `NFToken` id `nft_id` (32 bytes).
+    /// Reads the id region and writes the issuer bytes.
+    #[gas = 70]
+    #[wasm_name = "nft_issuer"]
+    fn get_nft_issuer(&self, nft_id: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The taxon encoded in the `NFToken` id `nft_id` (32 bytes). Reads the id region
+    /// and writes the taxon as its four little-endian bytes.
+    #[gas = 60]
+    #[wasm_name = "nft_taxon"]
+    fn get_nft_taxon(&self, nft_id: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    /// The flags encoded in the `NFToken` id `nft_id` (32 bytes). Reads the id region
+    /// and returns the flags as the call's scalar result.
+    #[gas = 60]
+    #[wasm_name = "nft_flags"]
+    fn get_nft_flags(&self, nft_id: &[u8]) -> HostResult<i32>;
+
+    /// The transfer fee encoded in the `NFToken` id `nft_id` (32 bytes). Reads the id
+    /// region and returns the fee as the call's scalar result.
+    #[gas = 60]
+    #[wasm_name = "nft_xfer_fee"]
+    fn get_nft_transfer_fee(&self, nft_id: &[u8]) -> HostResult<i32>;
+
+    /// The sequence number encoded in the `NFToken` id `nft_id` (32 bytes). Reads the
+    /// id region and writes the sequence as its four little-endian bytes.
+    #[gas = 60]
+    #[wasm_name = "nft_serial"]
+    fn get_nft_sequence(&self, nft_id: &[u8], out: &mut [u8]) -> HostResult<usize>;
+
+    // A "float" here is an XRPL `Number` in its serialized form: a byte blob the guest
+    // holds opaquely and hands back to these functions. Inputs and outputs that are
+    // floats are byte regions; `mode` is the rounding mode, a scalar the guest chooses.
+
+    /// A float built from the signed integer `x` under rounding `mode`. Writes the
+    /// float bytes; no input region.
+    #[gas = 100]
+    #[wasm_name = "float_from_int"]
+    fn float_from_int(&self, x: i64, mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// A float built from the unsigned integer in the 8-byte region `x` under rounding
+    /// `mode`. Reads the integer region and writes the float bytes.
+    #[gas = 130]
+    #[wasm_name = "float_from_uint"]
+    fn float_from_uint(&self, x: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// A float built from the serialized `STAmount` in `amount` under rounding `mode`.
+    /// Reads the amount region and writes the float bytes.
+    #[gas = 150]
+    #[wasm_name = "float_from_stamount"]
+    fn float_from_stamount(&self, amount: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// A float built from the serialized `STNumber` in `number` under rounding `mode`.
+    /// Reads the number region and writes the float bytes.
+    #[gas = 150]
+    #[wasm_name = "float_from_stnumber"]
+    fn float_from_stnumber(&self, number: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The float `x` rounded to a signed integer under rounding `mode`. Reads the float
+    /// region and writes the integer as its eight little-endian bytes.
+    #[gas = 130]
+    #[wasm_name = "float_to_int"]
+    fn float_to_int(&self, x: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The float `x` split into its mantissa and exponent. Reads the float region and
+    /// writes the mantissa (eight little-endian bytes) and the exponent (four little-
+    /// endian bytes) to two separate output regions.
+    #[gas = 130]
+    #[wasm_name = "float_to_mant_exp"]
+    fn float_to_mant_exp(
+        &self,
+        x: &[u8],
+        mantissa_out: &mut [u8],
+        exponent_out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// A float built from `mantissa` and `exponent` under rounding `mode`. Writes the
+    /// float bytes; no input region.
+    #[gas = 100]
+    #[wasm_name = "float_from_mant_exp"]
+    fn float_from_mant_exp(
+        &self,
+        mantissa: i64,
+        exponent: i32,
+        mode: i32,
+        out: &mut [u8],
+    ) -> HostResult<usize>;
+
+    /// Compares floats `x` and `y`, returning a negative, zero, or positive scalar as
+    /// `x` is less than, equal to, or greater than `y`. Reads both float regions.
+    #[gas = 80]
+    #[wasm_name = "float_cmp"]
+    fn float_compare(&self, x: &[u8], y: &[u8]) -> HostResult<i32>;
+
+    /// The float sum `x + y` under rounding `mode`. Reads both float regions and writes
+    /// the result bytes.
+    #[gas = 160]
+    #[wasm_name = "float_add"]
+    fn float_add(&self, x: &[u8], y: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The float difference `x - y` under rounding `mode`. Reads both float regions and
+    /// writes the result bytes.
+    #[gas = 160]
+    #[wasm_name = "float_sub"]
+    fn float_subtract(&self, x: &[u8], y: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The float product `x * y` under rounding `mode`. Reads both float regions and
+    /// writes the result bytes.
+    #[gas = 300]
+    #[wasm_name = "float_mult"]
+    fn float_multiply(&self, x: &[u8], y: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The float quotient `x / y` under rounding `mode`. Reads both float regions and
+    /// writes the result bytes.
+    #[gas = 300]
+    #[wasm_name = "float_div"]
+    fn float_divide(&self, x: &[u8], y: &[u8], mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The `n`-th root of the float `x` under rounding `mode`. Reads the float region
+    /// and writes the result bytes.
+    #[gas = 5500]
+    #[wasm_name = "float_root"]
+    fn float_root(&self, x: &[u8], n: i32, mode: i32, out: &mut [u8]) -> HostResult<usize>;
+
+    /// The float `x` raised to the power `n` under rounding `mode`. Reads the float
+    /// region and writes the result bytes.
+    #[gas = 5500]
+    #[wasm_name = "float_pow"]
+    fn float_power(&self, x: &[u8], n: i32, mode: i32, out: &mut [u8]) -> HostResult<usize>;
+}

crates/xrpl-host-functions/src/macros.rs

@@ -0,0 +1,102 @@
+//! The `macro_rules!` behind the two hand-listed enums, [`crate::HostError`] and
+//! [`crate::TraceDataType`].
+//!
+//! Each takes one list of `Variant = code,` and expands the enum together with the
+//! `ALL`/`code`/`from_code` set that must not fall behind it. The lists themselves stay
+//! in `lib.rs`, beside the `host_functions!` block.
+
+/// Declares [`crate::HostError`] from one list: the variants, `HostError::ALL` and
+/// `HostError::from_code`'s table all expand from the codes given.
+///
+/// One list is what makes `ALL` complete. Rust cannot enumerate an enum's
+/// variants — an exhaustive `match` forces an arm per variant but gives nothing to
+/// iterate — so a hand-written `ALL` beside a hand-written enum could only be kept
+/// in step by review, and `ALL`'s whole purpose is to be the set a test can trust.
+/// A code added to the list gains its `ALL` entry and its `from_code` arm by
+/// construction. `HostFunctionSpec::ALL` is complete the same way, from the
+/// `host_functions!` block.
+macro_rules! host_errors {
+    ($($(#[$doc:meta])* $variant:ident = $code:literal,)+) => {
+        /// Error codes a host function may return.
+        ///
+        #[derive(Debug, Clone, Copy, PartialEq, Eq)]
+        #[repr(i32)]
+        pub enum HostError {
+            $($(#[$doc])* $variant = $code,)+
+        }
+
+        impl HostError {
+            /// Every error a host function may return, in code order.
+            ///
+            /// The complete set, and complete by construction: a wasm engine's
+            /// split between the codes it hands the guest and the conditions it
+            /// traps on is a decision per variant, so the test that checks the
+            /// split iterates this and a code added to the ABI cannot slip past it.
+            pub const ALL: &'static [HostError] = &[$(HostError::$variant,)+];
+
+            /// The negative wire value a failed call returns. Every code but
+            /// `InternalFatal` is one a guest reads off that value.
+            #[inline]
+            pub const fn code(self) -> i32 {
+                self as i32
+            }
+
+            /// Reconstruct a `HostError` from its wire code.
+            ///
+            /// A code this ABI does not define is `InternalFatal`: an answer the
+            /// caller cannot act on is the call not having been served, and that is
+            /// the variant which says so. Positive values are not errors at all and go
+            /// the same way, since this is reached only once a negative return has
+            /// been read as a failure.
+            pub const fn from_code(code: i32) -> HostError {
+                match code {
+                    $($code => HostError::$variant,)+
+                    _ => HostError::InternalFatal,
+                }
+            }
+        }
+    };
+}
+
+/// Declares [`crate::TraceDataType`] from one list, so `TraceDataType::ALL`,
+/// `TraceDataType::code` and `TraceDataType::from_code` cannot fall behind the
+/// variants — the reason `host_errors!` above is written this way.
+macro_rules! trace_data_types {
+    ($($(#[$doc:meta])* $variant:ident = $code:literal,)+) => {
+        /// How [`HostFunctions::trace`] is to read its data buffer.
+        ///
+        /// The discriminants are wire values shared with the guest stdlib: append only,
+        /// never renumber. They start at 1, so a zeroed argument names no type rather
+        /// than the first one.
+        ///
+        /// This is the declaration a guest and a host both compile against. The host
+        /// side needs a second one — `cxx` cannot be a dependency here, since this
+        /// crate also links into the guest — so `xrpl-wasm-vm-ffi` declares a shared
+        /// enum for C++ and converts, exhaustively, from this.
+        #[derive(Debug, Clone, Copy, PartialEq, Eq)]
+        #[repr(i32)]
+        pub enum TraceDataType {
+            $($(#[$doc])* $variant = $code,)+
+        }
+
+        impl TraceDataType {
+            /// Every data type a guest may name, in code order.
+            pub const ALL: &'static [TraceDataType] = &[$(TraceDataType::$variant,)+];
+
+            /// The wire value a guest passes to name this type.
+            #[inline]
+            pub const fn code(self) -> i32 {
+                self as i32
+            }
+
+            /// The type `code` names, or `None`: the engine drops a call it cannot
+            /// read rather than guessing at a rendering the guest did not ask for.
+            pub const fn from_code(code: i32) -> Option<TraceDataType> {
+                match code {
+                    $($code => Some(TraceDataType::$variant),)+
+                    _ => None,
+                }
+            }
+        }
+    };
+}

crates/xrpl-host-functions/tests/expansion_hygiene.rs

@@ -0,0 +1,34 @@
+//! `host_functions!` must work outside the crate that declares the ABI: the only
+//! names its expansion needs are the ones the declarations themselves spell.
+
+use xrpl_host_functions::HostResult;
+use xrpl_host_functions_macros::host_functions;
+
+host_functions! {
+    /// Answers with the number it was given.
+    #[gas = 7]
+    #[wasm_name = "ping"]
+    fn ping(&self, number: i32) -> HostResult<i32>;
+}
+
+struct Host;
+
+impl HostFunctions for Host {
+    fn ping(&self, number: i32) -> HostResult<i32> {
+        Ok(number)
+    }
+}
+
+#[test]
+fn the_generated_table_stands_on_its_own() {
+    assert_eq!(HostFunctionSpec::ALL.len(), 1);
+    assert_eq!(HostFunctionSpec::Ping.wasm_name(), "ping");
+    assert_eq!(HostFunctionSpec::Ping.gas(), 7);
+}
+
+/// The generated trait is implementable from another crate, which is the point of
+/// declaring the ABI in a library at all.
+#[test]
+fn the_generated_trait_is_implementable_here() {
+    assert_eq!(Host.ping(3), Ok(3));
+}

crates/xrpl-host-functions/tests/host_errors.rs

@@ -0,0 +1,102 @@
+//! Exercises what `host_errors!` generates: the wire codes, the set
+//! [`HostError::ALL`] names, and the round trip between them.
+//!
+//! The codes are consensus input — they are what a guest reads off a failed host
+//! call — so they are pinned here as literals and derived everywhere else.
+
+use xrpl_host_functions::HostError;
+
+/// The whole set, written out in the order `ALL` gives it: the one place the wire
+/// codes appear as literals, and a deliberate change-detector, since a code that
+/// moves changes what every deployed guest is told.
+#[test]
+fn the_error_table_matches_the_declarations() {
+    let table: Vec<(HostError, i32)> = HostError::ALL
+        .iter()
+        .map(|&error| (error, error.code()))
+        .collect();
+
+    assert_eq!(
+        table,
+        [
+            (HostError::Unimplemented, -1),
+            (HostError::FieldNotFound, -2),
+            (HostError::BufferTooSmall, -3),
+            (HostError::NoArray, -4),
+            (HostError::NotLeafField, -5),
+            (HostError::LocatorMalformed, -6),
+            (HostError::SlotOutRange, -7),
+            (HostError::SlotsFull, -8),
+            (HostError::EmptySlot, -9),
+            (HostError::LedgerObjNotFound, -10),
+            (HostError::OutOfTransferLimit, -11),
+            (HostError::DataFieldTooLarge, -12),
+            (HostError::PointerOutOfBounds, -13),
+            (HostError::NoMemExported, -14),
+            (HostError::InvalidParams, -15),
+            (HostError::InvalidAccount, -16),
+            (HostError::InvalidField, -17),
+            (HostError::IndexOutOfBounds, -18),
+            (HostError::FloatInputMalformed, -19),
+            (HostError::FloatComputationError, -20),
+            (HostError::InternalFatal, i32::MIN),
+        ]
+    );
+}
+
+/// The guest-facing set is `-1 ..= -20` and nothing else: those entries are xrpld's
+/// `HostFunctionError`, and each is a code some contract may read.
+///
+/// `InternalFatal` is the one deliberate exception, exempted by name rather than by
+/// widening the range: a condition with no number a contract can act on needs no number
+/// in the range a contract reads, and holding it at `i32::MIN` is what keeps it from
+/// ever colliding with a code appended to xrpld's list.
+#[test]
+fn every_code_but_the_sentinel_is_in_the_shared_range() {
+    let shared: Vec<HostError> = HostError::ALL
+        .iter()
+        .copied()
+        .filter(|&error| error != HostError::InternalFatal)
+        .collect();
+
+    let outside: Vec<HostError> = shared
+        .iter()
+        .copied()
+        .filter(|error| !(-20..=-1).contains(&error.code()))
+        .collect();
+
+    assert!(outside.is_empty(), "outside -1..=-20: {outside:?}");
+    assert_eq!(shared.len(), 20);
+    assert_eq!(HostError::InternalFatal.code(), i32::MIN);
+    assert_eq!(HostError::ALL.len(), 21);
+}
+
+/// Every code a guest can be handed comes back as the error that produced it, so a
+/// caller reading a negative return value recovers the condition and not a
+/// neighbouring one. The table above pins the numbers; this adds only the round
+/// trip.
+#[test]
+fn every_wire_code_round_trips_back_to_its_error() {
+    for &error in HostError::ALL {
+        assert_eq!(HostError::from_code(error.code()), error, "{error:?}");
+    }
+}
+
+/// A code from outside the set is `InternalFatal`: a host answering something this ABI
+/// does not define has not served the call, whatever it meant by it, and success is not
+/// an error at all.
+///
+/// `-21` is the code xrpld would append next, so it is the one that decides whether a
+/// list this crate has not caught up with reaches a guest or stops the run. `i32::MIN +
+/// 1` is next to the sentinel and unassigned, which is what makes the sentinel a value
+/// rather than a range.
+#[test]
+fn a_code_outside_the_set_is_internal_fatal() {
+    for code in [-21, i32::MIN + 1, 0, 1, i32::MAX] {
+        assert_eq!(
+            HostError::from_code(code),
+            HostError::InternalFatal,
+            "{code}"
+        );
+    }
+}

crates/xrpl-wasm-testkit/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "xrpl-wasm-testkit"
+version = "0.1.0"
+edition.workspace = true
+
+[lib]
+crate-type = ["staticlib", "rlib"]
+
+[dependencies]
+cxx.workspace = true
+wat = "1"

crates/xrpl-wasm-testkit/src/lib.rs

@@ -0,0 +1,49 @@
+//! Assembles WebAssembly text for the C++ test suite. **Test-only.**
+//!
+//! A crate of its own rather than an entry on `xrpl-wasm-vm-ffi`, and the separation is the
+//! point. The engine pins `wasmi = { default-features = false }` precisely so a text
+//! assembler cannot reach the consensus path — wasmi's `wat` feature is on by default and
+//! makes `Module::new` accept text as readily as binary, which would make a transaction's
+//! validity a build flag (review finding A5). Putting `compile_wat` on the production bridge
+//! would link `wat` into xrpld even if nothing called it.
+//!
+//! Linked only into `xrpl_tests`, never into `libxrpl` or `xrpld`, so "no assembler in the
+//! shipped node" is a property of the link graph rather than a flag someone can flip.
+#![deny(rustdoc::broken_intra_doc_links)]
+
+#[cxx::bridge(namespace = "rs::wasm_testkit")]
+mod ffi {
+    extern "Rust" {
+        /// Assemble `wat` to a wasm module.
+        ///
+        /// Throws `rust::Error` on invalid input, which is what a test wants: a typo in a
+        /// fixture should fail the test that holds it, at the line that holds it.
+        fn compile_wat(wat: &str) -> Result<Vec<u8>>;
+    }
+}
+
+fn compile_wat(wat: &str) -> Result<Vec<u8>, wat::Error> {
+    wat::parse_str(wat)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::compile_wat;
+
+    #[test]
+    fn a_module_assembles_to_something_beginning_with_the_wasm_magic() {
+        let wasm = compile_wat("(module)").expect("assembles");
+
+        assert_eq!(&wasm[..4], b"\0asm");
+    }
+
+    #[test]
+    fn a_typo_is_an_error_rather_than_a_module() {
+        let error = compile_wat("(module (func (export").expect_err("must not assemble");
+
+        assert!(
+            !error.to_string().is_empty(),
+            "the error has to say something"
+        );
+    }
+}

crates/xrpl-wasm-vm-ffi/Cargo.toml

@@ -0,0 +1,12 @@
+[package]
+name = "xrpl-wasm-vm-ffi"
+version = "0.1.0"
+edition.workspace = true
+
+[lib]
+crate-type = ["staticlib", "rlib"]
+
+[dependencies]
+cxx.workspace = true
+xrpl-host-functions = { path = "../xrpl-host-functions" }
+xrpl-wasm-vm = { path = "../xrpl-wasm-vm" }

crates/xrpl-wasm-vm/Cargo.toml

@@ -0,0 +1,11 @@
+[package]
+name = "xrpl-wasm-vm"
+version = "0.1.0"
+edition.workspace = true
+
+[dependencies]
+wasmi = { version = "1.1.0", default-features = false, features = ["std"] }
+xrpl-host-functions = { path = "../xrpl-host-functions" }
+
+[dev-dependencies]
+wat = "1"

crates/xrpl-wasm-vm/src/abi.rs

@@ -0,0 +1,824 @@
+use crate::region::Region;
+use crate::vm::{MAX_FIELD_BYTES, VmState};
+use wasmi::{Caller, Memory};
+use xrpl_host_functions::{HostError, HostFunctionSpec, HostFunctions, HostResult};
+
+/// A condition that stops the run. It is a property of the run rather than an answer
+/// to a call, so it reaches no guest and carries no wire code — which is why it is
+/// not a [`HostError`]: no host can report one and no contract can read one.
+///
+/// The three are the outcomes a host call can end a run with, and
+/// `From<Fault> for RunError` in `vm.rs` is where each gets its name.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum Fault {
+    /// This call's charge would take the meter below zero. The guest exhausting the
+    /// meter with its own instructions reaches [`crate::vm::RunError::OutOfGas`] by
+    /// wasmi's `OutOfFuel` trap instead, never through here.
+    OutOfGas,
+    /// The call could not be served: either the host said so, or this engine's own
+    /// fuel meter did not answer.
+    Internal,
+    /// There is no linear memory to work in — the module exports none, or the call
+    /// came from a start section, which runs before there is an instance.
+    NoMemory,
+}
+
+/// How a host call fails: with a code the guest reads off the return value, or with a
+/// [`Fault`] that stops the run.
+///
+/// **The variant picks the channel.** [`to_wire`] reads it rather than asking a
+/// predicate, so the two cannot disagree, and a [`FatalHostError`] cannot be built
+/// around something a guest was supposed to see.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum CallError {
+    Code(HostError),
+    Fatal(Fault),
+}
+
+/// A host call's result inside the engine: [`HostResult`] plus the faults only the
+/// engine can raise.
+pub(crate) type CallResult<T> = Result<T, CallError>;
+
+/// Which channel a host's answer takes, decided once, here.
+///
+/// Three codes stop the run instead of reaching the contract that asked. Each says the
+/// call was not served at all — the host could not do it, it has not been wired, or
+/// there is nowhere to put the answer — and a contract has no business interpreting
+/// any of them, so it is told nothing and the run ends. Every other code is the
+/// contract's to read.
+impl From<HostError> for CallError {
+    fn from(error: HostError) -> CallError {
+        match error {
+            HostError::InternalFatal => CallError::Fatal(Fault::Internal),
+            HostError::Unimplemented => CallError::Fatal(Fault::Internal),
+            HostError::NoMemExported => CallError::Fatal(Fault::NoMemory),
+            code => CallError::Code(code),
+        }
+    }
+}
+
+/// The payload a trap carries so [`crate::vm::run`] can name the outcome without
+/// parsing a message. Holds a [`Fault`], so by construction no guest-visible code can
+/// leave through this channel.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) struct FatalHostError(pub(crate) Fault);
+
+impl wasmi::errors::HostError for FatalHostError {}
+
+impl core::fmt::Display for FatalHostError {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        write!(f, "host call refused: {:?}", self.0)
+    }
+}
+
+/// Charge the call's gas, run its body, put the result on the wire. The one path
+/// every registered closure takes, so gas cannot be forgotten.
+pub(crate) fn charged(
+    caller: &mut Caller<'_, VmState<'_>>,
+    op: HostFunctionSpec,
+    body: impl FnOnce(&mut Caller<'_, VmState<'_>>) -> CallResult<i32>,
+) -> Result<i32, wasmi::Error> {
+    to_wire(charge(caller, op.gas()).and_then(|()| body(caller)))
+}
+
+/// [`charged`] for a call the guest gets no answer from: its wasm function has no
+/// result, so a soft error has nowhere to go and is dropped. The gas is charged first
+/// and charged whatever happens after, so the cost is all such a call leaves behind.
+///
+/// Only `trace` takes this path.
+pub(crate) fn charged_unreported(
+    caller: &mut Caller<'_, VmState<'_>>,
+    op: HostFunctionSpec,
+    body: impl FnOnce(&mut Caller<'_, VmState<'_>>) -> CallResult<()>,
+) -> Result<(), wasmi::Error> {
+    dropped(charge(caller, op.gas()).and_then(|()| body(caller)))
+}
+
+/// [`to_wire`] for a call with no result: there is no return value to encode a code
+/// in, so it is dropped. A [`Fault`] still stops the run — that is a property of the
+/// run, not an answer to the call.
+fn dropped(result: CallResult<()>) -> Result<(), wasmi::Error> {
+    match result {
+        Err(CallError::Fatal(fault)) => Err(wasmi::Error::host(FatalHostError(fault))),
+        _ => Ok(()),
+    }
+}
+
+fn to_wire(result: CallResult<i32>) -> Result<i32, wasmi::Error> {
+    match result {
+        Ok(value) => Ok(value),
+        Err(CallError::Code(error)) => Ok(error.code()),
+        Err(CallError::Fatal(fault)) => Err(wasmi::Error::host(FatalHostError(fault))),
+    }
+}
+
+/// Deduct `cost` fuel; [`Fault::OutOfGas`] if it would go negative.
+///
+/// A meter that will not answer is this crate's own defect, not the contract's, so it
+/// is [`Fault::Internal`] rather than a number a guest could act on.
+fn charge<T>(caller: &mut Caller<'_, T>, cost: u64) -> CallResult<()> {
+    let remaining = caller
+        .get_fuel()
+        .map_err(|_| CallError::Fatal(Fault::Internal))?;
+    match remaining.checked_sub(cost) {
+        Some(left) => caller
+            .set_fuel(left)
+            .map_err(|_| CallError::Fatal(Fault::Internal)),
+        None => {
+            let _ = caller.set_fuel(0);
+            Err(CallError::Fatal(Fault::OutOfGas))
+        }
+    }
+}
+
+fn charge_transfer(state: &VmState<'_>, n: usize) -> Result<(), HostError> {
+    let n = n as u64;
+    let remaining = state.transfer_budget.get();
+    match remaining.checked_sub(n) {
+        Some(left) => {
+            state.transfer_budget.set(left);
+            Ok(())
+        }
+        None => Err(HostError::OutOfTransferLimit),
+    }
+}
+
+fn memory(caller: &Caller<'_, VmState<'_>>) -> CallResult<Memory> {
+    caller
+        .data()
+        .memory
+        .ok_or(CallError::Fatal(Fault::NoMemory))
+}
+
+/// [`Region::read`] of the guest's memory, for a call that reads and writes nothing
+/// back (`trace`).
+pub(crate) fn read_borrowed<'a>(
+    caller: &'a Caller<'_, VmState<'_>>,
+    input: Region,
+) -> CallResult<&'a [u8]> {
+    let mem = memory(caller)?;
+    Ok(input.read(mem.data(caller))?)
+}
+
+/// Decode a guest `u32` argument — a keylet's sequence number or document id — from
+/// its four little-endian bytes, carried on to the host as its `i32` bit pattern.
+///
+/// The ABI transports these as a 4-byte region rather than a wasm scalar (the guest
+/// SDK passes `seq.to_le_bytes()`), so the region must be exactly four bytes;
+/// `InvalidParams` otherwise, matching the C-ABI wrapper's `getDataUInt32`.
+pub(crate) fn read_u32_arg(bytes: &[u8]) -> HostResult<i32> {
+    let arr: [u8; 4] = bytes.try_into().map_err(|_| HostError::InvalidParams)?;
+    Ok(i32::from_le_bytes(arr))
+}
+
+/// Service a call whose answer is bytes, written straight into the guest's output
+/// region.
+///
+/// **`fill` returns the value's true length, not what it wrote**: a host holding 64
+/// bytes and offered room for 4 writes nothing and answers `64`, which is how the
+/// guest learns the size to ask for. So `n` is bounded by neither the region nor the
+/// cap, and both checks below are reachable.
+pub(crate) fn write_into(
+    caller: &mut Caller<'_, VmState<'_>>,
+    out: Region,
+    fill: impl FnOnce(&dyn HostFunctions, &mut [u8]) -> HostResult<usize>,
+) -> CallResult<i32> {
+    let range = out.range()?;
+    let cap = range.len();
+    let mem = memory(caller)?;
+    let host: &dyn HostFunctions = caller.data().host;
+    // Bounds-checked over the guest's whole declared region, so a buffer running
+    // past memory is a wrong pointer rather than a truncated prefix being served…
+    let buf = mem
+        .data_mut(&mut *caller)
+        .get_mut(range)
+        .ok_or(HostError::PointerOutOfBounds)?;
+    // …of which only the field cap is writable, so no call can exceed it whatever
+    // the guest declared.
+    let buf = &mut buf[..cap.min(MAX_FIELD_BYTES)];
+
+    let n = fill(host, buf)?;
+
+    if n > MAX_FIELD_BYTES {
+        return Err(HostError::DataFieldTooLarge.into());
+    }
+    if n > cap {
+        return Err(HostError::BufferTooSmall.into());
+    }
+    charge_transfer(caller.data(), n)?;
+    #[expect(
+        clippy::cast_possible_truncation,
+        clippy::cast_possible_wrap,
+        reason = "`n > MAX_FIELD_BYTES` returned above, and the cap is far inside i32"
+    )]
+    let n = n as i32;
+    Ok(n)
+}
+
+/// Service a call that reads guest memory and writes bytes back to it: the host
+/// fills the run's output buffer, which is copied to the guest once every rule has
+/// passed.
+///
+/// `call` gets the guest's whole memory, so it can borrow any number of input
+/// regions with [`Region::read`] — which a `&mut` view of that memory would forbid.
+/// That is why the answer goes through a buffer instead of straight into the guest
+/// as [`write_into`]'s does.
+///
+/// **The host is never told the guest's capacity**: it is offered the whole buffer
+/// and reports the value's true length, so the fit is decided here, with nothing yet
+/// in guest memory. A refused value therefore reaches it in no part.
+///
+/// The output is judged after the inputs, so a call with both bad reports the
+/// input's verdict. `NoMemExported` precedes both: there is no memory to validate a
+/// region against.
+pub(crate) fn write_buffered(
+    caller: &mut Caller<'_, VmState<'_>>,
+    out: Region,
+    call: impl FnOnce(&dyn HostFunctions, &[u8], &mut [u8]) -> HostResult<usize>,
+) -> CallResult<i32> {
+    let mem = memory(caller)?;
+    // One borrow split in two: the guest's bytes for the inputs, the store data for
+    // the output buffer. Taking them together is what keeps the inputs borrowed
+    // rather than copied out.
+    let (data, state) = mem.data_and_store_mut(&mut *caller);
+    let host: &dyn HostFunctions = state.host;
+
+    let n = call(host, data, &mut state.out_buffer[..])?;
+
+    // `out` is checked here rather than before the call: the inputs are judged
+    // first, so a call with both malformed reports the input's verdict.
+    let range = out.range()?;
+    let cap = range.len();
+    if n > MAX_FIELD_BYTES {
+        return Err(HostError::DataFieldTooLarge.into());
+    }
+    let buf = data.get_mut(range).ok_or(HostError::PointerOutOfBounds)?;
+    if n > cap {
+        return Err(HostError::BufferTooSmall.into());
+    }
+    charge_transfer(state, n)?;
+    buf[..n].copy_from_slice(&state.out_buffer[..n]);
+    #[expect(
+        clippy::cast_possible_truncation,
+        clippy::cast_possible_wrap,
+        reason = "`n > MAX_FIELD_BYTES` returned above, and the cap is far inside i32"
+    )]
+    let n = n as i32;
+    Ok(n)
+}
+
+/// The mantissa and exponent widths `float_to_mant_exp` writes: an `i64` and an `i32`.
+/// Fixed by the ABI, not the guest, so the split is a constant rather than a reported
+/// length.
+const MANTISSA_BYTES: usize = 8;
+const EXPONENT_BYTES: usize = 4;
+
+/// Service `float_to_mant_exp`, the one call that writes two output regions: the host
+/// fills the run's output buffer with the mantissa followed by the exponent, and each
+/// is copied to its own guest region once every rule has passed.
+///
+/// Like [`write_buffered`], the host reads its input from the guest's memory and writes
+/// to a scratch buffer, so the input stays borrowed rather than copied. The two output
+/// regions are judged after the input, and the mantissa's region before the exponent's,
+/// so the first fault reported is the leftmost.
+///
+/// The two widths are the ABI's rather than the guest's, so the length the host reports
+/// is checked against their sum for equality rather than as a bound, and ahead of the
+/// output regions: a wrong total means there is no answer to place, whatever the guest
+/// declared. That is a fatal error and not a status, since the guest asked for nothing
+/// wrong.
+pub(crate) fn write_mant_exp(
+    caller: &mut Caller<'_, VmState<'_>>,
+    mantissa_out: Region,
+    exponent_out: Region,
+    call: impl FnOnce(&dyn HostFunctions, &[u8], &mut [u8], &mut [u8]) -> HostResult<usize>,
+) -> CallResult<i32> {
+    let mem = memory(caller)?;
+    let (data, state) = mem.data_and_store_mut(&mut *caller);
+    let host: &dyn HostFunctions = state.host;
+
+    // The scratch buffer is split at the fixed mantissa width: the host fills the first
+    // eight bytes with the mantissa and the next four with the exponent.
+    let (mant_buf, exp_buf) = state.out_buffer.split_at_mut(MANTISSA_BYTES);
+    let mant_buf = &mut mant_buf[..MANTISSA_BYTES];
+    let exp_buf = &mut exp_buf[..EXPONENT_BYTES];
+
+    let total = call(host, data, mant_buf, exp_buf)?;
+
+    // Both buffers are fixed-width and were offered whole, so the only length the host
+    // can correctly report is their sum. Anything else is the host contradicting the
+    // ABI: with the widths in doubt, part of what would be copied out is whatever the
+    // previous call left in the buffer, so none of it is copied.
+    if total != MANTISSA_BYTES + EXPONENT_BYTES {
+        return Err(HostError::InternalFatal.into());
+    }
+
+    // Copy the mantissa, then the exponent, each only if its whole value fits its
+    // region — a region too small is `BufferTooSmall`, with nothing written.
+    let mant_range = mantissa_out.range()?;
+    let mant_dst = data
+        .get_mut(mant_range)
+        .ok_or(HostError::PointerOutOfBounds)?;
+    if mant_dst.len() < MANTISSA_BYTES {
+        return Err(HostError::BufferTooSmall.into());
+    }
+    mant_dst[..MANTISSA_BYTES].copy_from_slice(&state.out_buffer[..MANTISSA_BYTES]);
+
+    let exp_range = exponent_out.range()?;
+    let exp_dst = data
+        .get_mut(exp_range)
+        .ok_or(HostError::PointerOutOfBounds)?;
+    if exp_dst.len() < EXPONENT_BYTES {
+        return Err(HostError::BufferTooSmall.into());
+    }
+    exp_dst[..EXPONENT_BYTES]
+        .copy_from_slice(&state.out_buffer[MANTISSA_BYTES..MANTISSA_BYTES + EXPONENT_BYTES]);
+
+    charge_transfer(state, MANTISSA_BYTES + EXPONENT_BYTES)?;
+    #[expect(
+        clippy::cast_possible_truncation,
+        clippy::cast_possible_wrap,
+        reason = "a total other than 12 returned above, and 12 is far inside i32"
+    )]
+    let total = total as i32;
+    Ok(total)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::vm::TRANSFER_LIMIT_BYTES;
+    use std::cell::Cell;
+    use wasmi::StoreLimitsBuilder;
+    use xrpl_host_functions::TraceDataType;
+
+    /// `charge_transfer` takes the store data, which has to hold a host.
+    struct UncalledHost;
+
+    impl HostFunctions for UncalledHost {
+        fn get_ledger_sqn(&self, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_parent_ledger_time(&self, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_parent_ledger_hash(&self, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_base_fee(&self, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn is_amendment_enabled(&self, _amendment: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn cache_ledger_obj(&self, _obj_id: &[u8], _cache_idx: i32) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_tx_field(&self, _field: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_current_ledger_obj_field(&self, _field: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_ledger_obj_field(
+            &self,
+            _cache_idx: i32,
+            _field: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_tx_nested_field(&self, _locator: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_current_ledger_obj_nested_field(
+            &self,
+            _locator: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_ledger_obj_nested_field(
+            &self,
+            _cache_idx: i32,
+            _locator: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_tx_array_len(&self, _field: i32) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_current_ledger_obj_array_len(&self, _field: i32) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_ledger_obj_array_len(&self, _cache_idx: i32, _field: i32) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_tx_nested_array_len(&self, _locator: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_current_ledger_obj_nested_array_len(&self, _locator: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_ledger_obj_nested_array_len(
+            &self,
+            _cache_idx: i32,
+            _locator: &[u8],
+        ) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn check_signature(
+            &self,
+            _message: &[u8],
+            _signature: &[u8],
+            _pubkey: &[u8],
+        ) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn account_keylet(&self, _account: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn amm_keylet(&self, _asset1: &[u8], _asset2: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn check_keylet(&self, _account: &[u8], _seq: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn credential_keylet(
+            &self,
+            _subject: &[u8],
+            _issuer: &[u8],
+            _credential_type: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn delegate_keylet(
+            &self,
+            _account: &[u8],
+            _authorize: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn deposit_preauth_keylet(
+            &self,
+            _account: &[u8],
+            _authorize: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn did_keylet(&self, _account: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn escrow_keylet(&self, _account: &[u8], _seq: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn trust_line_keylet(
+            &self,
+            _account1: &[u8],
+            _account2: &[u8],
+            _currency: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn mptoken_issuance_keylet(
+            &self,
+            _issuer: &[u8],
+            _seq: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn mptoken_keylet(
+            &self,
+            _mptid: &[u8],
+            _holder: &[u8],
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn nftoken_offer_keylet(
+            &self,
+            _account: &[u8],
+            _seq: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn offer_keylet(&self, _account: &[u8], _seq: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn oracle_keylet(
+            &self,
+            _account: &[u8],
+            _doc_id: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn paychannel_keylet(
+            &self,
+            _account: &[u8],
+            _destination: &[u8],
+            _seq: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn permissioned_domain_keylet(
+            &self,
+            _account: &[u8],
+            _seq: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn signer_list_keylet(&self, _account: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn ticket_keylet(&self, _account: &[u8], _seq: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn vault_keylet(&self, _account: &[u8], _seq: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn sha512_half(&self, _data: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn trace(&self, _msg: &str, _data: &[u8], _data_type: TraceDataType) -> HostResult<()> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn update_data(&self, _data: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_nft(&self, _account: &[u8], _nft_id: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_nft_issuer(&self, _nft_id: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_nft_taxon(&self, _nft_id: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_nft_flags(&self, _nft_id: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_nft_transfer_fee(&self, _nft_id: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn get_nft_sequence(&self, _nft_id: &[u8], _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_from_int(&self, _x: i64, _mode: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_from_uint(&self, _x: &[u8], _mode: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_from_stamount(
+            &self,
+            _amount: &[u8],
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_from_stnumber(
+            &self,
+            _number: &[u8],
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_to_int(&self, _x: &[u8], _mode: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_to_mant_exp(
+            &self,
+            _x: &[u8],
+            _mantissa_out: &mut [u8],
+            _exponent_out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_from_mant_exp(
+            &self,
+            _mantissa: i64,
+            _exponent: i32,
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_compare(&self, _x: &[u8], _y: &[u8]) -> HostResult<i32> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_add(
+            &self,
+            _x: &[u8],
+            _y: &[u8],
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_subtract(
+            &self,
+            _x: &[u8],
+            _y: &[u8],
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_multiply(
+            &self,
+            _x: &[u8],
+            _y: &[u8],
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_divide(
+            &self,
+            _x: &[u8],
+            _y: &[u8],
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_root(&self, _x: &[u8], _n: i32, _mode: i32, _out: &mut [u8]) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+        fn float_power(
+            &self,
+            _x: &[u8],
+            _n: i32,
+            _mode: i32,
+            _out: &mut [u8],
+        ) -> HostResult<usize> {
+            unreachable!("no unit test in this module calls the host")
+        }
+    }
+
+    fn state(budget: u64) -> VmState<'static> {
+        VmState {
+            host: &UncalledHost,
+            mem_limits: StoreLimitsBuilder::new().build(),
+            transfer_budget: Cell::new(budget),
+            memory: None,
+            out_buffer: [0u8; MAX_FIELD_BYTES],
+        }
+    }
+
+    /// `wasmi::Error` is not `PartialEq`, so a test expecting the guest-visible
+    /// channel says so by going through here.
+    fn wire(result: CallResult<i32>) -> i32 {
+        to_wire(result)
+            .unwrap_or_else(|trap| panic!("expected a guest-visible status, got a trap: {trap}"))
+    }
+
+    #[test]
+    fn a_success_becomes_the_value_and_an_error_becomes_its_code() {
+        assert_eq!(wire(Ok(0)), 0);
+        assert_eq!(wire(Ok(32)), 32);
+        assert_eq!(wire(Err(HostError::BufferTooSmall.into())), -3);
+    }
+
+    /// The codes a host may answer that a contract must not see, and the fault each
+    /// becomes. Written out rather than derived from `From<HostError>`, which is what
+    /// they are asserting.
+    const STOPS_THE_RUN: [(HostError, Fault); 3] = [
+        (HostError::InternalFatal, Fault::Internal),
+        (HostError::Unimplemented, Fault::Internal),
+        (HostError::NoMemExported, Fault::NoMemory),
+    ];
+
+    /// Every fault, so the two tests below are the whole set and not a sample.
+    /// `From<Fault> for RunError` is what forces a fault added later to be
+    /// considered; this is what forces it to be tested.
+    const ALL_FAULTS: [Fault; 3] = [Fault::OutOfGas, Fault::Internal, Fault::NoMemory];
+
+    #[test]
+    fn a_code_that_stops_the_run_converts_to_its_fault() {
+        for (error, fault) in STOPS_THE_RUN {
+            assert_eq!(CallError::from(error), CallError::Fatal(fault), "{error:?}");
+        }
+    }
+
+    /// Over `HostError::ALL`, so it is the whole ABI and not a sample: a code added
+    /// to the ABI arrives already asserted to reach the guest as itself, and stopping
+    /// the run on it is then a change someone has to come and make.
+    ///
+    /// `OutOfTransferLimit` is the row worth reading twice: the one budget a
+    /// contract can be expected to handle, so it is told no rather than killed.
+    #[test]
+    fn every_other_code_reaches_the_guest_as_itself() {
+        for &error in HostError::ALL {
+            if STOPS_THE_RUN.iter().any(|&(stops, _)| stops == error) {
+                continue;
+            }
+            assert_eq!(CallError::from(error), CallError::Code(error), "{error:?}");
+            assert_eq!(wire(Err(error.into())), error.code(), "{error:?}");
+        }
+    }
+
+    /// The trap carries the fault, so `run` can name the outcome without parsing a
+    /// message.
+    #[test]
+    fn a_fault_becomes_a_trap_carrying_it() {
+        for fault in ALL_FAULTS {
+            let trap = to_wire(Err(CallError::Fatal(fault)))
+                .expect_err("a fault must not reach the guest as a code");
+            let payload = trap.downcast_ref::<FatalHostError>().unwrap_or_else(|| {
+                panic!("{fault:?}: expected a FatalHostError payload, got: {trap}")
+            });
+            assert_eq!(*payload, FatalHostError(fault));
+        }
+    }
+
+    /// The result-less path splits the same two channels differently: a fault still
+    /// stops the run, and every code is dropped, since `trace` has no return value to
+    /// carry it. Over `HostError::ALL` for the reason above — a code added to the ABI
+    /// arrives asserted against both paths.
+    #[test]
+    fn a_call_with_no_result_drops_a_code_and_traps_on_a_fault() {
+        assert!(dropped(Ok(())).is_ok());
+
+        for &error in HostError::ALL {
+            if let CallError::Code(code) = CallError::from(error) {
+                assert!(
+                    dropped(Err(CallError::Code(code))).is_ok(),
+                    "{error:?} has no channel to the guest and must be dropped"
+                );
+            }
+        }
+
+        for fault in ALL_FAULTS {
+            let trap =
+                dropped(Err(CallError::Fatal(fault))).expect_err("a fault must stop the run");
+            let payload = trap.downcast_ref::<FatalHostError>().unwrap_or_else(|| {
+                panic!("{fault:?}: expected a FatalHostError payload, got: {trap}")
+            });
+            assert_eq!(*payload, FatalHostError(fault));
+        }
+    }
+
+    #[test]
+    fn a_transfer_spends_the_budget() {
+        let state = state(100);
+
+        assert_eq!(charge_transfer(&state, 30), Ok(()));
+        assert_eq!(state.transfer_budget.get(), 70);
+        assert_eq!(charge_transfer(&state, 70), Ok(()));
+        assert_eq!(state.transfer_budget.get(), 0);
+    }
+
+    /// The budget bounds the total, so the transfer that would overrun it is
+    /// refused whole rather than partially charged.
+    #[test]
+    fn a_transfer_past_the_budget_is_refused_and_charges_nothing() {
+        let state = state(100);
+
+        assert_eq!(
+            charge_transfer(&state, 101),
+            Err(HostError::OutOfTransferLimit)
+        );
+        assert_eq!(
+            state.transfer_budget.get(),
+            100,
+            "a refusal must not charge"
+        );
+        assert_eq!(charge_transfer(&state, 100), Ok(()));
+        assert_eq!(
+            charge_transfer(&state, 1),
+            Err(HostError::OutOfTransferLimit)
+        );
+    }
+
+    #[test]
+    fn transferring_nothing_costs_nothing() {
+        let state = state(0);
+
+        assert_eq!(charge_transfer(&state, 0), Ok(()));
+        assert_eq!(state.transfer_budget.get(), 0);
+    }
+
+    /// The field cap holds one call to a small share of the run's budget, so the
+    /// budget bounds a run rather than a call. An inequality, not the two values:
+    /// those are pinned in `vm.rs`.
+    #[test]
+    fn no_single_value_can_exhaust_the_run_budget() {
+        assert!(
+            (MAX_FIELD_BYTES as u64) * 64 <= TRANSFER_LIMIT_BYTES,
+            "one {MAX_FIELD_BYTES}-byte value against a {TRANSFER_LIMIT_BYTES}-byte budget"
+        );
+    }
+}

crates/xrpl-wasm-vm/src/lib.rs

@@ -0,0 +1,28 @@
+//! The escrow wasm VM: compile a contract, meter it, and serve its host calls.
+//!
+//! Every guest access goes through `abi.rs` and reaches linear memory only by
+//! wasmi's bounds-checked slice operations; `forbid(unsafe_code)` makes that a
+//! property rather than a claim. The cast lints are on for the same reason — on a
+//! consensus path a truncating or sign-losing cast changes what a contract is
+//! charged or told, so each one is argued for at its site.
+#![forbid(unsafe_code)]
+#![deny(rustdoc::broken_intra_doc_links)]
+#![deny(unreachable_pub)]
+#![deny(
+    clippy::cast_possible_truncation,
+    clippy::cast_possible_wrap,
+    clippy::cast_sign_loss,
+    clippy::cast_lossless
+)]
+
+mod abi;
+mod preflight;
+mod region;
+mod register;
+mod vm;
+
+pub use preflight::{CheckError, check};
+pub use vm::{
+    MAX_FIELD_BYTES, MAX_MEMORY_BYTES, MAX_MEMORY_PAGES, RunError, RunFailure, RunOutcome,
+    TRANSFER_LIMIT_BYTES, run,
+};

crates/xrpl-wasm-vm/src/preflight.rs

@@ -0,0 +1,351 @@
+//! Screening a contract before it reaches the ledger.
+//!
+//! [`check`] answers whether [`crate::run`] would refuse a module before the
+//! guest's first instruction — the three stages a caller maps to a malformed
+//! transaction rather than to a failed one. It needs **no host, no store and no
+//! gas**: everything it reads is a property of the compiled module. That is what
+//! makes it callable from a transaction's preflight, which has no ledger to serve
+//! host calls from.
+//!
+//! Two things it deliberately does not screen. A module exporting **no** linear
+//! memory passes: a contract that makes no host call needs none, and one that
+//! does is refused at the call and charged for what it burned. A start section
+//! passes: it is guest code, and executing it is the one thing a check must not do
+//! — a trap in one is charged to the contract like any other trap.
+//!
+//! One thing it screens that a run can only discover: an exported memory larger
+//! than the engine grants. See [`check_memory`] for what stays invisible.
+
+use std::fmt;
+use wasmi::{ExternType, FuncType, Module, ValType};
+use xrpl_host_functions::HostFunctionSpec;
+
+use crate::register::HOST_MODULE;
+use crate::vm::{MAX_MEMORY_PAGES, compile};
+
+/// Why a module cannot be run. One variant per stage, since the caller maps the
+/// stages separately.
+#[derive(Debug)]
+pub enum CheckError {
+    /// `wasm` is not a valid module under this engine's configuration.
+    Compile(String),
+    /// An import no engine of this ABI defines: another module namespace, a name
+    /// that is not a host function, or one imported as something other than a
+    /// function.
+    Import(String),
+    /// No export named `function_name` with signature `() -> i32`.
+    EntryPoint(String),
+    /// The module asks for more linear memory than the engine grants.
+    Memory(String),
+}
+
+impl fmt::Display for CheckError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            CheckError::Compile(detail) => write!(f, "compile: {detail}"),
+            CheckError::Import(detail) => write!(f, "import: {detail}"),
+            // The detail says which of the entry point's failures this is, since
+            // "no entry point" would be wrong for an export of the wrong type.
+            CheckError::EntryPoint(detail) => write!(f, "{detail}"),
+            CheckError::Memory(detail) => write!(f, "memory: {detail}"),
+        }
+    }
+}
+
+/// Screen `wasm`: it must compile, import only what the engine serves, export
+/// `function_name` as `() -> i32`, and ask for no more memory than it may have.
+///
+/// The stages are ordered by how much of the module each explains. An import fault
+/// is reported before a missing entry point because the imports are what the rest of
+/// the module is built on; memory comes last, being a resource request rather than a
+/// mistake about the ABI.
+pub fn check(wasm: &[u8], function_name: &str) -> Result<(), CheckError> {
+    let module = compile(wasm).map_err(CheckError::Compile)?;
+    check_imports(&module)?;
+    check_entry_point(&module, function_name)?;
+    check_memory(&module)
+}
+
+/// Every import must be one the linker defines. The first that is not ends the
+/// check, so a module with several faults reports the earliest.
+fn check_imports(module: &Module) -> Result<(), CheckError> {
+    for import in module.imports() {
+        check_import(import.module(), import.name(), import.ty()).map_err(CheckError::Import)?;
+    }
+    Ok(())
+}
+
+/// Whether the engine defines this one import.
+///
+/// The set of names is [`HostFunctionSpec::ALL`], which is also what
+/// [`crate::register::register_host_functions`] iterates — so a check and a run
+/// cannot disagree about which names exist, and adding a host function extends
+/// both at once. The one thing this does not compare is `ty`'s *signature*, which
+/// still parts a module from the engine at instantiation; the kind is compared
+/// because the engine defines these names as functions and as nothing else.
+///
+/// The rules are ordered, not merely alternatives: a guest importing `env::malloc`
+/// is told about the namespace rather than that `malloc` is not a host function,
+/// because the namespace is the one that explains every other import it has too.
+fn check_import(module: &str, name: &str, ty: &ExternType) -> Result<(), String> {
+    if module != HOST_MODULE {
+        return Err(format!("'{module}::{name}' is not from '{HOST_MODULE}'"));
+    }
+    if !HostFunctionSpec::ALL
+        .iter()
+        .any(|op| op.wasm_name() == name)
+    {
+        return Err(format!("no host function '{name}'"));
+    }
+    if !matches!(ty, ExternType::Func(_)) {
+        return Err(format!("'{HOST_MODULE}::{name}' is not a function"));
+    }
+    Ok(())
+}
+
+fn check_entry_point(module: &Module, name: &str) -> Result<(), CheckError> {
+    match module.get_export(name) {
+        Some(ExternType::Func(ty)) if is_entry_point(&ty) => Ok(()),
+        found => Err(CheckError::EntryPoint(entry_point_fault(found, name))),
+    }
+}
+
+/// The entry point's type: nothing in, one `i32` out — what [`crate::run`]'s
+/// `get_typed_func::<(), i32>` accepts.
+fn is_entry_point(ty: &FuncType) -> bool {
+    ty.params().is_empty() && matches!(ty.results(), [ValType::I32])
+}
+
+/// A module may not declare more linear memory than the engine grants.
+///
+/// Only what it *exports* is visible here. A memory a module keeps to itself is not
+/// in its exports, and the store's limiter is what refuses that one — at
+/// instantiation, where the run is charged nothing and the caller cannot tell it
+/// from any other resource failure. Screening the exported case covers every
+/// contract built against the guest SDK, since a contract needs an exported memory
+/// to make a host call at all.
+fn check_memory(module: &Module) -> Result<(), CheckError> {
+    for export in module.exports() {
+        if let ExternType::Memory(ty) = export.ty() {
+            check_initial_pages(ty.minimum()).map_err(CheckError::Memory)?;
+        }
+    }
+    Ok(())
+}
+
+/// Whether the engine will grant a memory of this declared initial size.
+///
+/// The *minimum* only: a declared maximum past the cap is legal and simply
+/// unreachable, which `vm_limits::a_declared_maximum_past_the_cap_is_allowed_but_
+/// unreachable` pins on the run side. Refusing it here would turn a runnable
+/// contract away.
+fn check_initial_pages(pages: u64) -> Result<(), String> {
+    if pages > u64::from(MAX_MEMORY_PAGES) {
+        return Err(format!(
+            "initial memory of {pages} pages is past the {MAX_MEMORY_PAGES}-page cap"
+        ));
+    }
+    Ok(())
+}
+
+/// How an entry-point lookup failed, in the words both stages use: a check and a
+/// run describe the same module the same way, and "no entry point" would send a
+/// contract author looking for a function they already have.
+pub(crate) fn entry_point_fault(found: Option<ExternType>, name: &str) -> String {
+    match found {
+        Some(ExternType::Func(_)) => {
+            format!("entry point '{name}' has the wrong signature, expected '() -> i32'")
+        }
+        Some(_) => format!("export '{name}' is not a function"),
+        None => format!("no entry point '{name}'"),
+    }
+}
+
+/// The rules, one by one, on inputs built directly rather than parsed out of a
+/// module. `tests/preflight.rs` runs real modules through [`check`]; what is here is
+/// what a module cannot state precisely — which rule fires, in which order, and in
+/// what words the caller logs it.
+///
+/// `wat` is a dev-dependency, so the one test here that does need a module writes it
+/// as text like every other test in the crate. What the library must not gain is a
+/// text *entry point* — `check` and `run` take binaries — and a `cfg(test)` caller
+/// cannot give it one.
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use wasmi::{GlobalType, MemoryType, Mutability};
+
+    /// A host function as a guest declares it. Any function type will do: the
+    /// signature is not what [`check_import`] compares.
+    fn a_function() -> ExternType {
+        ExternType::Func(FuncType::new([ValType::I32], [ValType::I32]))
+    }
+
+    /// A name every one of these tests can use, taken from the ABI rather than
+    /// spelled, so it stays a real host function as the ABI changes.
+    fn a_host_function_name() -> &'static str {
+        HostFunctionSpec::ALL[0].wasm_name()
+    }
+
+    // -----------------------------------------------------------------------
+    // Imports
+    // -----------------------------------------------------------------------
+
+    /// Every name the ABI declares is served. Derived from `ALL` rather than
+    /// listed, so a host function added to the ABI is covered the day it lands.
+    #[test]
+    fn every_declared_host_function_is_served() {
+        for op in HostFunctionSpec::ALL {
+            assert_eq!(
+                check_import(HOST_MODULE, op.wasm_name(), &a_function()),
+                Ok(()),
+                "{}",
+                op.wasm_name()
+            );
+        }
+    }
+
+    #[test]
+    fn an_import_from_another_namespace_is_refused() {
+        for namespace in ["env", "host", "host_lib2", ""] {
+            let refusal = check_import(namespace, a_host_function_name(), &a_function())
+                .expect_err(namespace);
+            assert!(
+                refusal.contains("is not from 'host_lib'"),
+                "{namespace}: {refusal}"
+            );
+        }
+    }
+
+    #[test]
+    fn an_unknown_name_is_refused() {
+        let refusal =
+            check_import(HOST_MODULE, "no_such_function", &a_function()).expect_err("unknown name");
+        assert_eq!(refusal, "no host function 'no_such_function'");
+    }
+
+    /// The engine defines these names as functions and as nothing else, so a module
+    /// importing one as a global or a memory does not link either.
+    #[test]
+    fn a_host_function_imported_as_anything_else_is_refused() {
+        for ty in [
+            ExternType::Global(GlobalType::new(ValType::I32, Mutability::Const)),
+            ExternType::Memory(MemoryType::new(1, None)),
+        ] {
+            let name = a_host_function_name();
+            let refusal = check_import(HOST_MODULE, name, &ty).expect_err("not a function");
+            assert_eq!(refusal, format!("'host_lib::{name}' is not a function"));
+        }
+    }
+
+    /// The rules are ordered. An import that breaks two of them is reported by the
+    /// first, so the message a contract author reads is the one that explains the
+    /// rest of their imports too.
+    #[test]
+    fn the_namespace_is_reported_before_the_name() {
+        let refusal = check_import("env", "no_such_function", &a_function())
+            .expect_err("neither the namespace nor the name is served");
+
+        assert!(refusal.contains("is not from 'host_lib'"), "{refusal}");
+        assert!(
+            !refusal.contains("no host function"),
+            "the namespace explains it: {refusal}"
+        );
+    }
+
+    /// Both halves of the type are load-bearing, and neither is checked anywhere
+    /// a module cannot reach.
+    #[test]
+    fn the_entry_point_type_is_nothing_in_and_one_i32_out() {
+        assert!(is_entry_point(&FuncType::new([], [ValType::I32])));
+
+        for wrong in [
+            FuncType::new([], []),
+            FuncType::new([], [ValType::I64]),
+            FuncType::new([ValType::I32], [ValType::I32]),
+            FuncType::new([], [ValType::I32, ValType::I32]),
+        ] {
+            assert!(!is_entry_point(&wrong), "{wrong:?}");
+        }
+    }
+
+    /// Three faults, three descriptions. A run reports these too, with wasmi's own
+    /// error appended, so a swapped arm would mislead at both stages at once.
+    #[test]
+    fn each_entry_point_fault_is_described_as_itself() {
+        assert_eq!(
+            entry_point_fault(Some(a_function()), "finish"),
+            "entry point 'finish' has the wrong signature, expected '() -> i32'"
+        );
+        assert_eq!(
+            entry_point_fault(
+                Some(ExternType::Global(GlobalType::new(
+                    ValType::I32,
+                    Mutability::Const
+                ))),
+                "finish"
+            ),
+            "export 'finish' is not a function"
+        );
+        assert_eq!(
+            entry_point_fault(None, "finish"),
+            "no entry point 'finish'",
+            "an absent export must not be reported as a wrong signature"
+        );
+    }
+
+    /// The cap itself is granted; one page past it is not. The boundary is the whole
+    /// rule, and it is the same boundary the store's limiter applies at
+    /// instantiation.
+    #[test]
+    fn the_initial_memory_may_reach_the_cap_but_not_pass_it() {
+        assert_eq!(check_initial_pages(0), Ok(()));
+        assert_eq!(check_initial_pages(u64::from(MAX_MEMORY_PAGES)), Ok(()));
+
+        let past = u64::from(MAX_MEMORY_PAGES) + 1;
+        let refusal = check_initial_pages(past).expect_err("one page past the cap");
+        assert_eq!(
+            refusal,
+            format!("initial memory of {past} pages is past the {MAX_MEMORY_PAGES}-page cap")
+        );
+    }
+
+    /// The bridge logs this string and the C++ tests match on it, so the stage's
+    /// prefix is part of the interface rather than a debugging aid.
+    #[test]
+    fn a_refusal_names_its_stage() {
+        assert_eq!(
+            CheckError::Compile("bad magic".to_string()).to_string(),
+            "compile: bad magic"
+        );
+        assert_eq!(
+            CheckError::Memory("initial memory of 129 pages".to_string()).to_string(),
+            "memory: initial memory of 129 pages"
+        );
+        assert_eq!(
+            CheckError::Import("no host function 'x'".to_string()).to_string(),
+            "import: no host function 'x'"
+        );
+        // The entry point's detail already says which of its three faults it is,
+        // so a prefix would only repeat it.
+        assert_eq!(
+            CheckError::EntryPoint("no entry point 'finish'".to_string()).to_string(),
+            "no entry point 'finish'"
+        );
+    }
+
+    #[test]
+    fn the_stages_run_in_order() {
+        assert!(
+            matches!(check(b"not wasm", "finish"), Err(CheckError::Compile(_))),
+            "nothing is screened until the module compiles"
+        );
+
+        // A module that compiles and imports nothing, so it reaches the entry point.
+        let empty = wat::parse_str("(module)").expect("assembles");
+        assert!(
+            matches!(check(&empty, "finish"), Err(CheckError::EntryPoint(_))),
+            "a module that compiles and imports nothing reaches the entry point"
+        );
+    }
+}

crates/xrpl-wasm-vm/src/region.rs

@@ -0,0 +1,50 @@
+use crate::vm::MAX_FIELD_BYTES;
+use core::ops::Range;
+use xrpl_host_functions::{HostError, HostResult};
+
+/// A byte region as the guest declared it: the `(ptr, len)` pair off the wire, not
+/// yet checked.
+///
+/// Every byte parameter in this ABI is such a pair, so pairing them once at the wire
+/// boundary is what keeps the helpers in `abi.rs` from each taking two loose integers
+/// they could be handed in either order.
+///
+/// It lives in a module of its own so that the fields are out of reach and
+/// [`range`](Region::range) is the *only* way to indices — the check cannot be
+/// skipped, only deferred. Construction is infallible for that reason: a call whose
+/// output region is malformed is then refused in the order its own helper chooses,
+/// rather than at the moment the pair happened to be formed.
+#[derive(Copy, Clone)]
+pub(crate) struct Region {
+    ptr: i32,
+    len: i32,
+}
+
+impl Region {
+    pub(crate) fn new(ptr: i32, len: i32) -> Region {
+        Region { ptr, len }
+    }
+
+    /// `start..end` as indices. The conversion is the negativity check — it fails on
+    /// exactly the negative values — and the addition guards a 32-bit `usize`, where
+    /// two `i32`s can sum past the end.
+    pub(crate) fn range(self) -> HostResult<Range<usize>> {
+        let (Ok(start), Ok(len)) = (usize::try_from(self.ptr), usize::try_from(self.len)) else {
+            return Err(HostError::InvalidParams);
+        };
+        let end = start
+            .checked_add(len)
+            .ok_or(HostError::PointerOutOfBounds)?;
+        Ok(start..end)
+    }
+
+    /// The region's bytes, refused past the field cap. No copy: the slice aliases
+    /// `data`.
+    pub(crate) fn read(self, data: &[u8]) -> HostResult<&[u8]> {
+        let range = self.range()?;
+        if range.len() > MAX_FIELD_BYTES {
+            return Err(HostError::DataFieldTooLarge);
+        }
+        data.get(range).ok_or(HostError::PointerOutOfBounds)
+    }
+}

crates/xrpl-wasm-vm/src/vm.rs

@@ -0,0 +1,353 @@
+use std::cell::Cell;
+use std::fmt;
+use std::sync::LazyLock;
+use wasmi::{
+    Config, Engine, Export, Linker, Memory, Module, Store, StoreLimits, StoreLimitsBuilder,
+    TrapCode,
+};
+use xrpl_host_functions::HostFunctions;
+
+use crate::abi::{FatalHostError, Fault};
+use crate::preflight::entry_point_fault;
+use crate::register::register_host_functions;
+
+/// wasm linear-memory page size, fixed by the wasm spec (64 KiB).
+const WASM_PAGE_BYTES: u32 = 64 * 1024;
+
+/// Linear-memory page cap.
+pub const MAX_MEMORY_PAGES: u32 = 128;
+
+/// [`MAX_MEMORY_PAGES`] in bytes: 8 MiB.
+pub const MAX_MEMORY_BYTES: usize = (MAX_MEMORY_PAGES * WASM_PAGE_BYTES) as usize;
+
+/// Total bytes that may cross the host/guest boundary in one [`run`], separate
+/// from gas.
+pub const TRANSFER_LIMIT_BYTES: u64 = 1 << 20;
+
+/// Size cap on any single value crossing the boundary, in either direction; over
+/// it is `DataFieldTooLarge`.
+///
+/// A protocol limit: `kMaxWasmDataLength` in `include/xrpl/protocol/Protocol.h`.
+pub const MAX_FIELD_BYTES: usize = 1024;
+
+/// State threaded through every host call, stored in the wasmi [`Store`].
+pub(crate) struct VmState<'h> {
+    pub(crate) host: &'h dyn HostFunctions,
+    /// Enforces [`MAX_MEMORY_BYTES`] via `Store::limiter`, which needs a `&mut`
+    /// into it from `&mut VmState` — hence a field rather than a local.
+    pub(crate) mem_limits: StoreLimits,
+    /// Remaining transfer budget for this run ([`TRANSFER_LIMIT_BYTES`]).
+    ///
+    /// A `Cell` because it is decremented from a shared `&Caller`. One thread per
+    /// invocation touches the store, so the lack of `Sync` costs nothing.
+    ///
+    /// TODO: the extra charge for an unaligned field copy has nothing to attach to
+    /// until this ABI gains a `FieldLocator` host function.
+    pub(crate) transfer_budget: Cell<u64>,
+    /// The guest's linear memory, resolved once by [`run`] after instantiation so
+    /// no host call pays for an export lookup.
+    ///
+    /// Caching the handle is sound because a [`Memory`] is an arena index, not a
+    /// pointer to the bytes: it survives `memory.grow`, and `data`/`data_mut`
+    /// re-derive the slice per call.
+    ///
+    /// The handle is scoped to one store, so this assumes **one module, one
+    /// instance, one store per `run`**. Module linking or nested execution would
+    /// have to resolve per instance: a cached handle would serve a call against the
+    /// wrong instance's memory, which is a wrong answer rather than an error.
+    pub(crate) memory: Option<Memory>,
+    /// Where a host writes a value before [`crate::abi::write_buffered`] copies it
+    /// to the guest. One buffer per run, so no call zero-fills one of its own.
+    ///
+    /// Inline rather than boxed: the store's data is built once and then only
+    /// borrowed, so a kilobyte in it costs a move where a `Box` costs an
+    /// allocation. A local would cost neither, but `forbid(unsafe_code)` means a
+    /// stack buffer is zero-filled — per call, which is the cost this removes.
+    pub(crate) out_buffer: [u8; MAX_FIELD_BYTES],
+}
+
+/// Outcome of running an escrow contract to completion.
+#[derive(Debug)]
+pub struct RunOutcome {
+    /// The value returned by the exported entry point (`finish`): `> 0` means
+    /// allow the escrow to finish.
+    pub result: i32,
+    /// Fuel (gas) consumed by the whole invocation — guest instructions plus
+    /// the per-call host charges.
+    pub fuel_used: u64,
+}
+
+/// Why a run produced no result. Each variant is one outcome for the caller to
+/// map to a TER.
+#[derive(Debug)]
+pub enum RunError {
+    /// `wasm` is not a valid module under this engine's configuration.
+    Compile(String),
+    /// The module compiled but the engine would not accept it: an import the
+    /// linker does not define, or an initial memory past the page cap. Not guest
+    /// code failing — a start section that traps is [`RunError::Trap`].
+    Instantiate(String),
+    /// No export named `function_name` with signature `() -> i32`: absent, not a
+    /// function, or a function of another type — which the detail tells apart.
+    EntryPoint(String),
+    /// Gas exhausted — by the guest's own instructions or by a host call's
+    /// charge. [`RunFailure::fuel_used`] is the whole limit.
+    OutOfGas,
+    /// The host could not serve a call.
+    Internal,
+    /// A host call had no linear memory to work in: the module exports none, or
+    /// the call came from a start section, which runs before there is an instance
+    /// to resolve the memory from.
+    NoMemory,
+    /// The guest trapped: `unreachable`, division by zero, an out-of-bounds
+    /// access, or `memory.grow` past the page cap. Wherever the guest was
+    /// executing, including a start section during instantiation.
+    Trap(String),
+}
+
+impl fmt::Display for RunError {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            RunError::Compile(detail) => write!(f, "compile: {detail}"),
+            RunError::Instantiate(detail) => write!(f, "instantiate: {detail}"),
+            // The detail says which of the entry point's failures this is, since
+            // "no entry point" would be wrong for an export of the wrong type.
+            RunError::EntryPoint(detail) => write!(f, "{detail}"),
+            RunError::OutOfGas => write!(f, "out of gas"),
+            RunError::Internal => write!(f, "internal error"),
+            RunError::NoMemory => write!(f, "no exported memory"),
+            RunError::Trap(detail) => write!(f, "trap: {detail}"),
+        }
+    }
+}
+
+/// A failed run, with the gas it still owes: a contract that traps or exhausts
+/// its gas is charged for what it burned.
+#[derive(Debug)]
+pub struct RunFailure {
+    pub error: RunError,
+    /// Fuel consumed before the failure. The whole limit when gas ran out; `0`
+    /// when the module never ran.
+    pub fuel_used: u64,
+}
+
+impl fmt::Display for RunFailure {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "{} (fuel used: {})", self.error, self.fuel_used)
+    }
+}
+
+impl RunFailure {
+    /// A failure with no fuel accounted: it stopped the run at or before the guest's
+    /// first instruction, or under a store with no meter to read.
+    fn owing_nothing(error: RunError) -> RunFailure {
+        RunFailure {
+            error,
+            fuel_used: 0,
+        }
+    }
+}
+
+/// Fuel spent out of `gas`: the one place a run's cost is measured, so success,
+/// trap and refusal all report it the same way.
+///
+/// `Store::get_fuel` fails only on a store without fuel metering, which
+/// [`build_wasm_engine`] rules out and `run`'s `set_fuel` would already have
+/// caught — so a failure here is a defect in this crate. It must not become a
+/// number: `0` forgives a run its whole cost, `gas` charges an untouched one for
+/// everything. [`RunError::Internal`] instead.
+fn fuel_used(store: &Store<VmState<'_>>, gas: u64) -> Result<u64, RunError> {
+    store
+        .get_fuel()
+        .map(|remaining| gas.saturating_sub(remaining))
+        .map_err(|_| RunError::Internal)
+}
+
+/// Report `error` with the run's cost attached. A cost that cannot be read replaces
+/// the outcome rather than being invented — see [`fuel_used`].
+fn failed(store: &Store<VmState<'_>>, gas: u64, error: RunError) -> RunFailure {
+    match fuel_used(store, gas) {
+        Ok(fuel_used) => RunFailure { error, fuel_used },
+        Err(unmetered) => RunFailure::owing_nothing(unmetered),
+    }
+}
+
+/// The outcome a `wasmi::Error` names for itself, if any, rather than leaving it to
+/// the stage that raised it.
+///
+/// Two ways a run halts mid-flight: a host call that could not be served, which
+/// carries a [`FatalHostError`] saying which condition it was, and the guest's own
+/// instructions exhausting the meter, which wasmi raises as `OutOfFuel`.
+///
+/// Both can happen anywhere the guest executes — including a start section, which
+/// is guest code running during instantiation — so every stage from there on asks
+/// this before naming a failure after itself.
+fn guest_halted(error: &wasmi::Error) -> Option<RunError> {
+    if let Some(fatal) = error.downcast_ref::<FatalHostError>() {
+        return Some(fatal.0.into());
+    }
+    (error.as_trap_code() == Some(TrapCode::OutOfFuel)).then_some(RunError::OutOfGas)
+}
+
+/// Why instantiation failed, once [`guest_halted`] has ruled out the two conditions
+/// that can arise anywhere.
+///
+/// A start section is guest code, so it can trap on its own — `unreachable`, a
+/// division by zero, an out-of-bounds access — and a trap is the guest's fault
+/// wherever it happens. Naming that after the *stage* would file it beside the
+/// module faults a caller treats as its own defect, and charge nothing for
+/// instructions the contract burned. What is left for [`RunError::Instantiate`] is a
+/// module the linker or the store would not accept at all.
+fn instantiation_failure(error: &wasmi::Error) -> RunError {
+    match error.as_trap_code() {
+        Some(_) => RunError::Trap(error.to_string()),
+        None => RunError::Instantiate(error.to_string()),
+    }
+}
+
+/// The outcome a [`Fault`] is: the one place a stopped call becomes a stopped run.
+///
+/// Total and one arm each, because a `Fault` is only ever a condition that stops the
+/// run — the guest-visible codes cannot reach here, which is what
+/// [`crate::abi::CallError`] buys. A fault added later has no arm and does not
+/// compile.
+impl From<Fault> for RunError {
+    fn from(fault: Fault) -> RunError {
+        match fault {
+            Fault::OutOfGas => RunError::OutOfGas,
+            Fault::Internal => RunError::Internal,
+            Fault::NoMemory => RunError::NoMemory,
+        }
+    }
+}
+
+/// The process-wide wasmi engine, built once on first use.
+///
+/// The configuration is consensus-fixed and identical for every invocation, and an
+/// [`Engine`] is an internally `Arc`ed `Send + Sync` handle, so one shared engine
+/// serves concurrent [`run`] calls.
+pub(crate) fn wasm_engine() -> &'static Engine {
+    static ENGINE: LazyLock<Engine> = LazyLock::new(build_wasm_engine);
+    &ENGINE
+}
+
+/// Build the wasmi engine the escrow VM requires: deterministic, minimal
+/// features, fuel metering on.
+fn build_wasm_engine() -> Engine {
+    let mut config = Config::default();
+    config.consume_fuel(true);
+    config.ignore_custom_sections(true);
+    config.wasm_mutable_global(false);
+    config.wasm_multi_value(false);
+    config.wasm_sign_extension(false);
+    config.wasm_saturating_float_to_int(false);
+    config.wasm_bulk_memory(false);
+    config.wasm_reference_types(false);
+    config.wasm_tail_call(false);
+    config.wasm_extended_const(false);
+    config.floats(false);
+    config.wasm_multi_memory(false);
+    config.wasm_custom_page_sizes(false);
+    config.wasm_memory64(false);
+    config.wasm_wide_arithmetic(false);
+    // TODO: enable option to reject wasm code containing start section after wasmi 2.0 release
+    Engine::new(&config)
+}
+
+/// Compile `wasm` for this engine.
+///
+/// The one path to a [`Module`]: the configuration is what decides whether a
+/// contract is valid at all, so [`run`] and [`crate::check`] must not be able to
+/// compile against different ones.
+pub(crate) fn compile(wasm: &[u8]) -> Result<Module, String> {
+    Module::new(wasm_engine(), wasm).map_err(|e| e.to_string())
+}
+
+/// Run a contract: compile `wasm`, give it `gas` fuel, service its host
+/// calls through `host`, and call the exported `function_name`.
+pub fn run<'h>(
+    wasm: &[u8],
+    gas: u64,
+    host: &'h dyn HostFunctions,
+    function_name: &str,
+) -> Result<RunOutcome, RunFailure> {
+    let engine = wasm_engine();
+    let module =
+        compile(wasm).map_err(|detail| RunFailure::owing_nothing(RunError::Compile(detail)))?;
+
+    let mem_limits = StoreLimitsBuilder::new()
+        .memory_size(MAX_MEMORY_BYTES)
+        .trap_on_grow_failure(true)
+        .build();
+    let mut store = Store::new(
+        engine,
+        VmState {
+            host,
+            mem_limits,
+            transfer_budget: Cell::new(TRANSFER_LIMIT_BYTES),
+            memory: None,
+            out_buffer: [0u8; MAX_FIELD_BYTES],
+        },
+    );
+
+    store
+        .set_fuel(gas)
+        .map_err(|_| RunFailure::owing_nothing(RunError::Internal))?;
+    store.limiter(|state| &mut state.mem_limits);
+
+    let mut linker = Linker::<VmState<'h>>::new(engine);
+    register_host_functions(&mut linker)
+        .map_err(|_| RunFailure::owing_nothing(RunError::Internal))?;
+
+    let instance = match linker.instantiate_and_start(&mut store, &module) {
+        Ok(instance) => instance,
+        Err(e) => {
+            let error = guest_halted(&e).unwrap_or_else(|| instantiation_failure(&e));
+            return Err(failed(&store, gas, error));
+        }
+    };
+    store.data_mut().memory = instance.exports(&store).find_map(Export::into_memory);
+
+    let function = match instance.get_typed_func::<(), i32>(&store, function_name) {
+        Ok(function) => function,
+        Err(e) => {
+            let found = instance
+                .get_export(&store, function_name)
+                .map(|export| export.ty(&store));
+            let error =
+                RunError::EntryPoint(format!("{}: {e}", entry_point_fault(found, function_name)));
+            return Err(failed(&store, gas, error));
+        }
+    };
+
+    let result = match function.call(&mut store, ()) {
+        Ok(result) => result,
+        Err(e) => {
+            let error = guest_halted(&e).unwrap_or_else(|| RunError::Trap(e.to_string()));
+            return Err(failed(&store, gas, error));
+        }
+    };
+
+    let fuel_used = fuel_used(&store, gas).map_err(RunFailure::owing_nothing)?;
+    Ok(RunOutcome { result, fuel_used })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn the_engine_is_one_engine() {
+        assert!(Engine::same(wasm_engine(), wasm_engine()));
+    }
+
+    /// The only place these numbers appear as literals; every other test derives
+    /// them from the constants.
+    #[test]
+    fn the_limits_are_the_protocol_limits() {
+        assert_eq!(MAX_MEMORY_PAGES, 128, "linear-memory page cap");
+        assert_eq!(MAX_MEMORY_BYTES, 8 * 1024 * 1024, "page cap in bytes");
+        assert_eq!(MAX_FIELD_BYTES, 1024, "kMaxWasmDataLength");
+        assert_eq!(TRANSFER_LIMIT_BYTES, 1 << 20, "kWasmTransferLimit");
+    }
+}

crates/xrpl-wasm-vm/tests/budgets.rs

@@ -0,0 +1,750 @@
+//! The two budgets a run spends: gas (fuel), and the transfer limit on bytes
+//! crossing the boundary. Both are consensus input, so several of these tests
+//! assert exact numbers.
+
+mod support;
+
+use support::{
+    Answer, EMPTY_REGION, FakeHost, ONE_PAGE, PLENTY_OF_GAS, code, import, module, run,
+    run_with_gas, trace_call,
+};
+use xrpl_host_functions::{HASH_LEN, HostError, HostFunctionSpec, TraceDataType};
+use xrpl_wasm_vm::{MAX_FIELD_BYTES, RunError, TRANSFER_LIMIT_BYTES};
+
+// ---------------------------------------------------------------------------
+// Gas
+// ---------------------------------------------------------------------------
+
+/// The fuel a module of `body` burns, given gas to spare.
+fn fuel_for(body: &str, parts: &[&str], host: &FakeHost) -> u64 {
+    let wat = module(parts, body);
+    run(&wat, host).expect("the module should run").fuel_used
+}
+
+/// The fuel a module burns doing nothing but returning a constant; every figure
+/// below builds on it. wasmi's number, pinned deliberately because wasmi's fuel
+/// table is consensus input.
+const EMPTY_MODULE_FUEL: u64 = 30;
+
+/// wasmi's own fuel for a host call whose operands are all constants under 64: 14
+/// per `*.const`, plus 1 for the call. Our gas sits on top.
+///
+/// The formula holds only under 64, because wasmi widens a constant's encoding
+/// above that, each tier costing 7 more. Every call in [`call_for`] keeps its
+/// operands small for that reason; one with a larger constant fails here by a
+/// multiple of 7.
+fn wasmi_call_fuel(small_const_operands: u64) -> u64 {
+    14 * small_const_operands + 1
+}
+
+/// What wasmi charges on top of that for a call to a function with no result —
+/// `trace`'s shape, and nothing else in the ABI. Per call, not per module. Measured
+/// and pinned like the figures above.
+const WASMI_NO_RESULT_FUEL: u64 = 14;
+
+/// wasmi's fuel for one `(drop …)`, which is how a module makes more than one call
+/// and keeps only the last result. Pinned like the two above.
+const WASMI_DROP_FUEL: u64 = 21;
+
+/// The wasm a test needs in order to call one host function: the `(import …)`
+/// declaration, a call with small-constant operands, and how many it pushes.
+struct Call {
+    import: &'static str,
+    call: &'static str,
+    operands: u64,
+    /// Whether the call leaves an `i32` behind. `trace` does not, which is why
+    /// [`Call::body`] ends every module with a constant instead of the call.
+    yields: bool,
+}
+
+impl Call {
+    /// `n` calls in a row, leaving one `i32` for the module to return: the last
+    /// answer where there is one, and a constant where the call has none.
+    fn body(&self, n: usize) -> String {
+        if self.yields {
+            format!(
+                "{}{}",
+                format!("(drop {}) ", self.call).repeat(n - 1),
+                self.call
+            )
+        } else {
+            format!("{}(i32.const 0)", format!("{} ", self.call).repeat(n))
+        }
+    }
+
+    /// What [`Call::body`] burns beside the calls' own gas and the module's floor:
+    /// one `drop` between consecutive answers, or wasmi's own surcharge on a call
+    /// that has none.
+    fn overhead(&self, n: u64) -> u64 {
+        if self.yields {
+            (n - 1) * WASMI_DROP_FUEL
+        } else {
+            n * WASMI_NO_RESULT_FUEL
+        }
+    }
+}
+
+/// The test wasm for each host function. The `match` is exhaustive, so a function
+/// added to the ABI fails to compile until it has wasm here, and iterating
+/// [`HostFunctionSpec::ALL`] then covers the whole ABI.
+fn call_for(op: HostFunctionSpec) -> Call {
+    let (import, call, operands) = match op {
+        HostFunctionSpec::GetLedgerSqn => (
+            import::LDGR_INDEX,
+            "(call $ldgr_index (i32.const 0) (i32.const 4))",
+            2,
+        ),
+        HostFunctionSpec::GetParentLedgerTime => (
+            import::PARENT_LDGR_TIME,
+            "(call $parent_ldgr_time (i32.const 0) (i32.const 4))",
+            2,
+        ),
+        HostFunctionSpec::GetParentLedgerHash => (
+            import::PARENT_LDGR_HASH,
+            "(call $parent_ldgr_hash (i32.const 0) (i32.const 32))",
+            2,
+        ),
+        HostFunctionSpec::GetBaseFee => (
+            import::BASE_FEE,
+            "(call $base_fee (i32.const 0) (i32.const 4))",
+            2,
+        ),
+        HostFunctionSpec::IsAmendmentEnabled => (
+            import::AMENDMENT_ENABLED,
+            "(call $amendment_enabled (i32.const 0) (i32.const 32))",
+            2,
+        ),
+        HostFunctionSpec::CacheLedgerObj => (
+            import::CACHE_LE,
+            "(call $cache_le (i32.const 0) (i32.const 32) (i32.const 0))",
+            3,
+        ),
+        HostFunctionSpec::GetTxField => (
+            import::TX_FIELD,
+            "(call $tx_field (i32.const 1) (i32.const 0) (i32.const 4))",
+            3,
+        ),
+        HostFunctionSpec::GetCurrentLedgerObjField => (
+            import::HOME_LE_FIELD,
+            "(call $home_le_field (i32.const 1) (i32.const 0) (i32.const 4))",
+            3,
+        ),
+        HostFunctionSpec::GetLedgerObjField => (
+            import::LE_FIELD,
+            "(call $le_field (i32.const 1) (i32.const 1) (i32.const 0) (i32.const 4))",
+            4,
+        ),
+        HostFunctionSpec::GetTxNestedField => (
+            import::TX_INNER,
+            "(call $tx_inner (i32.const 0) (i32.const 4) (i32.const 8) (i32.const 4))",
+            4,
+        ),
+        HostFunctionSpec::GetCurrentLedgerObjNestedField => (
+            import::HOME_LE_INNER,
+            "(call $home_le_inner (i32.const 0) (i32.const 4) (i32.const 8) (i32.const 4))",
+            4,
+        ),
+        HostFunctionSpec::GetLedgerObjNestedField => (
+            import::LE_INNER,
+            "(call $le_inner (i32.const 1) (i32.const 0) (i32.const 4) (i32.const 8) (i32.const 4))",
+            5,
+        ),
+        HostFunctionSpec::GetTxArrayLen => {
+            (import::TX_ARR_LEN, "(call $tx_arr_len (i32.const 1))", 1)
+        }
+        HostFunctionSpec::GetCurrentLedgerObjArrayLen => (
+            import::HOME_LE_ARR_LEN,
+            "(call $home_le_arr_len (i32.const 1))",
+            1,
+        ),
+        HostFunctionSpec::GetLedgerObjArrayLen => (
+            import::LE_ARR_LEN,
+            "(call $le_arr_len (i32.const 1) (i32.const 1))",
+            2,
+        ),
+        HostFunctionSpec::GetTxNestedArrayLen => (
+            import::TX_INNER_ARR_LEN,
+            "(call $tx_inner_arr_len (i32.const 0) (i32.const 4))",
+            2,
+        ),
+        HostFunctionSpec::GetCurrentLedgerObjNestedArrayLen => (
+            import::HOME_LE_INNER_ARR_LEN,
+            "(call $home_le_inner_arr_len (i32.const 0) (i32.const 4))",
+            2,
+        ),
+        HostFunctionSpec::GetLedgerObjNestedArrayLen => (
+            import::LE_INNER_ARR_LEN,
+            "(call $le_inner_arr_len (i32.const 1) (i32.const 0) (i32.const 4))",
+            3,
+        ),
+        HostFunctionSpec::CheckSignature => (
+            import::CHECK_SIG,
+            "(call $check_sig (i32.const 0) (i32.const 0) (i32.const 0) (i32.const 0) (i32.const 0) (i32.const 0))",
+            6,
+        ),
+        HostFunctionSpec::AccountKeylet => (
+            import::ACCOUNTROOT_ID,
+            "(call $accountroot_id (i32.const 0) (i32.const 20) (i32.const 32) (i32.const 32))",
+            4,
+        ),
+        HostFunctionSpec::AmmKeylet => (
+            import::AMM_ID,
+            "(call $amm_id (i32.const 0) (i32.const 20) (i32.const 24) (i32.const 40) (i32.const 0) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::CheckKeylet => (
+            import::CHECK_ID,
+            "(call $check_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::CredentialKeylet => (
+            import::CREDENTIAL_ID,
+            "(call $credential_id (i32.const 0) (i32.const 20) (i32.const 20) (i32.const 20) (i32.const 40) (i32.const 4) (i32.const 44) (i32.const 20))",
+            8,
+        ),
+        HostFunctionSpec::DelegateKeylet => (
+            import::DELEGATE_ID,
+            "(call $delegate_id (i32.const 0) (i32.const 20) (i32.const 20) (i32.const 20) (i32.const 40) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::DepositPreauthKeylet => (
+            import::DEPOSIT_PREAUTH_ID,
+            "(call $deposit_preauth_id (i32.const 0) (i32.const 20) (i32.const 20) (i32.const 20) (i32.const 40) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::DidKeylet => (
+            import::DID_ID,
+            "(call $did_id (i32.const 0) (i32.const 20) (i32.const 32) (i32.const 32))",
+            4,
+        ),
+        HostFunctionSpec::EscrowKeylet => (
+            import::ESCROW_ID,
+            "(call $escrow_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::TrustLineKeylet => (
+            import::TRUSTLINE_ID,
+            "(call $trustline_id (i32.const 0) (i32.const 20) (i32.const 20) (i32.const 20) (i32.const 40) (i32.const 20) (i32.const 60) (i32.const 32))",
+            8,
+        ),
+        HostFunctionSpec::MptokenIssuanceKeylet => (
+            import::MPT_ISSUANCE_ID,
+            "(call $mpt_issuance_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::MptokenKeylet => (
+            import::MPTOKEN_ID,
+            "(call $mptoken_id (i32.const 0) (i32.const 24) (i32.const 24) (i32.const 20) (i32.const 44) (i32.const 20))",
+            6,
+        ),
+        HostFunctionSpec::NftokenOfferKeylet => (
+            import::NFT_OFFER_ID,
+            "(call $nft_offer_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::OfferKeylet => (
+            import::OFFER_ID,
+            "(call $offer_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::OracleKeylet => (
+            import::ORACLE_ID,
+            "(call $oracle_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::PaychannelKeylet => (
+            import::PAYCHAN_ID,
+            "(call $paychan_id (i32.const 0) (i32.const 20) (i32.const 20) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 40) (i32.const 20))",
+            8,
+        ),
+        HostFunctionSpec::PermissionedDomainKeylet => (
+            import::PERMISSIONED_DOMAIN_ID,
+            "(call $permissioned_domain_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::SignerListKeylet => (
+            import::SIGNERS_ID,
+            "(call $signers_id (i32.const 0) (i32.const 20) (i32.const 32) (i32.const 32))",
+            4,
+        ),
+        HostFunctionSpec::TicketKeylet => (
+            import::TICKET_ID,
+            "(call $ticket_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::VaultKeylet => (
+            import::VAULT_ID,
+            "(call $vault_id (i32.const 0) (i32.const 20) (i32.const 0) (i32.const 4) (i32.const 32) (i32.const 32))",
+            6,
+        ),
+        HostFunctionSpec::Sha512Half => (
+            import::SHA512_HALF,
+            "(call $sha512_half (i32.const 0) (i32.const 4) (i32.const 0) (i32.const 32))",
+            4,
+        ),
+        HostFunctionSpec::Trace => (
+            import::TRACE,
+            "(call $trace (i32.const 0) (i32.const 0) (i32.const 1) (i32.const 0) (i32.const 0))",
+            5,
+        ),
+        HostFunctionSpec::UpdateData => (
+            import::SET_DATA,
+            "(call $set_data (i32.const 0) (i32.const 8))",
+            2,
+        ),
+        HostFunctionSpec::GetNft => (
+            import::NFT_URI,
+            "(call $nft_uri (i32.const 0) (i32.const 20) (i32.const 20) (i32.const 32) (i32.const 52) (i32.const 12))",
+            6,
+        ),
+        HostFunctionSpec::GetNftIssuer => (
+            import::NFT_ISSUER,
+            "(call $nft_issuer (i32.const 0) (i32.const 32) (i32.const 32) (i32.const 20))",
+            4,
+        ),
+        HostFunctionSpec::GetNftTaxon => (
+            import::NFT_TAXON,
+            "(call $nft_taxon (i32.const 0) (i32.const 32) (i32.const 32) (i32.const 4))",
+            4,
+        ),
+        HostFunctionSpec::GetNftFlags => (
+            import::NFT_FLAGS,
+            "(call $nft_flags (i32.const 0) (i32.const 32))",
+            2,
+        ),
+        HostFunctionSpec::GetNftTransferFee => (
+            import::NFT_XFER_FEE,
+            "(call $nft_xfer_fee (i32.const 0) (i32.const 32))",
+            2,
+        ),
+        HostFunctionSpec::GetNftSequence => (
+            import::NFT_SERIAL,
+            "(call $nft_serial (i32.const 0) (i32.const 32) (i32.const 32) (i32.const 4))",
+            4,
+        ),
+        HostFunctionSpec::FloatFromInt => (
+            import::FLOAT_FROM_INT,
+            "(call $float_from_int (i64.const 0) (i32.const 0) (i32.const 8) (i32.const 0))",
+            4,
+        ),
+        HostFunctionSpec::FloatFromUint => (
+            import::FLOAT_FROM_UINT,
+            "(call $float_from_uint (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 0))",
+            5,
+        ),
+        HostFunctionSpec::FloatFromStamount => (
+            import::FLOAT_FROM_STAMOUNT,
+            "(call $float_from_stamount (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 0))",
+            5,
+        ),
+        HostFunctionSpec::FloatFromStnumber => (
+            import::FLOAT_FROM_STNUMBER,
+            "(call $float_from_stnumber (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 0))",
+            5,
+        ),
+        HostFunctionSpec::FloatToInt => (
+            import::FLOAT_TO_INT,
+            "(call $float_to_int (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 0))",
+            5,
+        ),
+        HostFunctionSpec::FloatToMantExp => (
+            import::FLOAT_TO_MANT_EXP,
+            "(call $float_to_mant_exp (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 16) (i32.const 4))",
+            6,
+        ),
+        HostFunctionSpec::FloatFromMantExp => (
+            import::FLOAT_FROM_MANT_EXP,
+            "(call $float_from_mant_exp (i64.const 0) (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 0))",
+            5,
+        ),
+        HostFunctionSpec::FloatCompare => (
+            import::FLOAT_CMP,
+            "(call $float_cmp (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8))",
+            4,
+        ),
+        HostFunctionSpec::FloatAdd => (
+            import::FLOAT_ADD,
+            "(call $float_add (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 16) (i32.const 8) (i32.const 0))",
+            7,
+        ),
+        HostFunctionSpec::FloatSubtract => (
+            import::FLOAT_SUB,
+            "(call $float_sub (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 16) (i32.const 8) (i32.const 0))",
+            7,
+        ),
+        HostFunctionSpec::FloatMultiply => (
+            import::FLOAT_MULT,
+            "(call $float_mult (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 16) (i32.const 8) (i32.const 0))",
+            7,
+        ),
+        HostFunctionSpec::FloatDivide => (
+            import::FLOAT_DIV,
+            "(call $float_div (i32.const 0) (i32.const 8) (i32.const 8) (i32.const 8) (i32.const 16) (i32.const 8) (i32.const 0))",
+            7,
+        ),
+        HostFunctionSpec::FloatRoot => (
+            import::FLOAT_ROOT,
+            "(call $float_root (i32.const 0) (i32.const 8) (i32.const 2) (i32.const 8) (i32.const 8) (i32.const 0))",
+            6,
+        ),
+        HostFunctionSpec::FloatPower => (
+            import::FLOAT_POW,
+            "(call $float_pow (i32.const 0) (i32.const 8) (i32.const 2) (i32.const 8) (i32.const 8) (i32.const 0))",
+            6,
+        ),
+    };
+    Call {
+        import,
+        call,
+        operands,
+        yields: !matches!(op, HostFunctionSpec::Trace),
+    }
+}
+
+#[test]
+fn an_empty_module_burns_a_fixed_amount_of_fuel() {
+    let fuel = fuel_for("(i32.const 0)", &[ONE_PAGE], &FakeHost::new());
+    assert_eq!(fuel, EMPTY_MODULE_FUEL);
+}
+
+/// Calling a host function `n` times costs `n` times its gas, to the unit. Every
+/// other term is known — the module's floor, wasmi's fuel per call, one `drop` per
+/// answered call — so the total is a closed form, with the gas read from the spec
+/// table rather than restated. `n = 1` pins the charge, `n > 1` pins that it lands
+/// on every call rather than once per run.
+#[test]
+fn a_host_call_costs_its_gas_every_time_it_is_called() {
+    let host = FakeHost::new().answering_field(1, Answer::bytes([0xaa]));
+
+    for &op in HostFunctionSpec::ALL {
+        let call = call_for(op);
+        let per_call = wasmi_call_fuel(call.operands) + op.gas();
+
+        for n in 1..=3 {
+            let body = call.body(n);
+            let n = n as u64;
+
+            assert_eq!(
+                fuel_for(&body, &[call.import, ONE_PAGE], &host),
+                EMPTY_MODULE_FUEL + n * per_call + call.overhead(n),
+                "{n} x {}",
+                call.call
+            );
+        }
+    }
+}
+
+/// The gas charge precedes the call's body, so a failing call costs exactly what a
+/// successful one costs. Field 1 is answered and field 7 is not; the two modules
+/// are otherwise identical, so their totals are comparable.
+#[test]
+fn a_failing_host_call_costs_exactly_what_a_successful_one_costs() {
+    let host = FakeHost::new().answering_field(1, Answer::bytes([0xaa]));
+    let call = |field: i32| {
+        module(
+            &[import::HOME_LE_FIELD, ONE_PAGE],
+            &format!("(call $home_le_field (i32.const {field}) (i32.const 0) (i32.const 4))"),
+        )
+    };
+
+    let answered = run(&call(1), &host).expect("the module should run");
+    let refused = run(&call(7), &host).expect("the module should run");
+
+    assert_eq!(answered.result, 1);
+    assert_eq!(refused.result, code(HostError::FieldNotFound));
+    assert_eq!(refused.fuel_used, answered.fuel_used);
+}
+
+/// `fuel_used` is `gas - remaining`: what the run spent, not what was left or what
+/// it was handed. The gas figures are derived from the run's cost, so the boundary
+/// — exactly enough, and one short — is among the cases.
+#[test]
+fn fuel_used_is_what_was_spent_not_what_was_supplied() {
+    let host = FakeHost::new();
+    let op = HostFunctionSpec::GetLedgerSqn;
+    let call = call_for(op);
+    let wat = module(&[call.import, ONE_PAGE], call.call);
+    let cost = EMPTY_MODULE_FUEL + wasmi_call_fuel(call.operands) + op.gas();
+
+    // Exactly its cost is enough, and no amount above it changes the figure. The
+    // result is checked too, so the figure belongs to a run that did the work
+    // rather than to one that was cut short.
+    for gas in [cost, cost + 1, cost * 100, PLENTY_OF_GAS] {
+        let outcome = run_with_gas(&wat, gas, &host).expect("should run");
+        assert_eq!(
+            outcome.result, 4,
+            "gas {gas}: the call should have succeeded"
+        );
+        assert_eq!(outcome.fuel_used, cost, "gas {gas}");
+    }
+
+    // One fuel short: the run ends at the call it cannot pay for and still owes the
+    // whole limit, because `charge` spends what is left.
+    let short = run_with_gas(&wat, cost - 1, &host).expect_err("one fuel short must not complete");
+    assert!(
+        matches!(short.error, RunError::OutOfGas),
+        "expected the run to end out of gas, got: {short}"
+    );
+    assert_eq!(short.fuel_used, cost - 1);
+}
+
+/// Fuel is metered, so the same module burns the same fuel every time — a
+/// property consensus depends on.
+#[test]
+fn the_same_run_burns_the_same_fuel() {
+    let call = call_for(HostFunctionSpec::Trace);
+    let wat = module(&[call.import, ONE_PAGE], &call.body(1));
+
+    let first = run(&wat, &FakeHost::new()).expect("should run").fuel_used;
+    for _ in 0..4 {
+        assert_eq!(
+            run(&wat, &FakeHost::new()).expect("should run").fuel_used,
+            first
+        );
+    }
+    assert!(first > HostFunctionSpec::Trace.gas());
+}
+
+/// Too little gas to finish stops the run: the meter refuses the guest's own
+/// instructions before it ever reaches the host call.
+#[test]
+fn a_run_that_cannot_afford_itself_fails() {
+    let host = FakeHost::new();
+    let wat = module(
+        &[import::LDGR_INDEX, ONE_PAGE],
+        "(call $ldgr_index (i32.const 0) (i32.const 4))",
+    );
+
+    for gas in [0, 1, 10] {
+        let Err(failure) = run_with_gas(&wat, gas, &host) else {
+            panic!("gas {gas} should not have completed");
+        };
+        assert!(
+            matches!(failure.error, RunError::OutOfGas),
+            "gas {gas}: expected the run to end out of gas, got: {failure}"
+        );
+    }
+}
+
+/// A guest looping forever is stopped by gas rather than running away, and owes
+/// the gas it burned doing it.
+#[test]
+fn an_endless_loop_is_stopped_by_gas() {
+    const GAS: u64 = 100_000;
+
+    let host = FakeHost::new();
+    let wat = module(&[ONE_PAGE], "(loop $l (br $l)) (i32.const 0)");
+
+    let failure = run_with_gas(&wat, GAS, &host).expect_err("an endless loop must not complete");
+    assert!(
+        matches!(failure.error, RunError::OutOfGas),
+        "expected the meter to stop it, got: {failure}"
+    );
+    assert_eq!(
+        failure.fuel_used, GAS,
+        "a runaway guest burns the whole limit"
+    );
+}
+
+/// A host call refused its gas stops the run: the guest never gets a chance to
+/// ignore the refusal and carry on, and it is charged the whole limit.
+///
+/// The gas range is every amount that reaches the call and cannot pay for it, so
+/// the case is the whole boundary rather than one number. `trace` is the call under
+/// it because it is the one that could not report a refusal even if it wanted to:
+/// stopping the run is the whole of what the guest sees.
+#[test]
+fn a_host_call_refused_its_gas_stops_the_run() {
+    let host = FakeHost::new();
+    let op = HostFunctionSpec::Trace;
+    let call = call_for(op);
+    let wat = module(&[call.import, ONE_PAGE], &call.body(1));
+    // Measured rather than derived: the whole run's cost, less the call's own gas,
+    // is the least a guest can be given and still reach the call. Below that the
+    // meter stops the guest's own instructions instead, which is
+    // `a_run_that_cannot_afford_itself_fails`'s case, not this one.
+    let cost = run(&wat, &FakeHost::new())
+        .expect("the module should run")
+        .fuel_used;
+
+    for gas in cost - op.gas()..cost {
+        let Err(failure) = run_with_gas(&wat, gas, &host) else {
+            panic!("gas {gas}: the run completed, so the guest was handed the refusal");
+        };
+        assert!(
+            matches!(failure.error, RunError::OutOfGas),
+            "gas {gas}: expected the run to end out of gas, got: {failure}"
+        );
+        assert_eq!(
+            failure.fuel_used, gas,
+            "gas {gas}: a call it cannot afford burns the whole limit"
+        );
+    }
+    assert!(host.traces().is_empty(), "the host body must not have run");
+}
+
+// ---------------------------------------------------------------------------
+// The transfer limit
+// ---------------------------------------------------------------------------
+
+/// A module that repeats `call` while `keep_going` holds, then returns the last
+/// status, so a budget can be run to exhaustion inside one invocation.
+fn until_refused(imports: &str, call: &str, keep_going: &str) -> String {
+    module(
+        &[imports, ONE_PAGE],
+        &format!(
+            "(local $r i32)
+             (loop $l
+               (local.set $r {call})
+               (br_if $l {keep_going}))
+             (local.get $r)"
+        ),
+    )
+}
+
+/// For a call whose success is a positive byte count.
+const WHILE_POSITIVE: &str = "(i32.gt_s (local.get $r) (i32.const 0))";
+
+/// Bytes written into guest memory are charged against the run's budget, and the
+/// budget is a per-run total: 1 MiB of 1 KiB values exhausts it.
+#[test]
+fn writes_spend_the_transfer_budget() {
+    let host = FakeHost::new().answering_field(1, Answer::filler(MAX_FIELD_BYTES));
+    let wat = until_refused(
+        import::HOME_LE_FIELD,
+        &format!("(call $home_le_field (i32.const 1) (i32.const 0) (i32.const {MAX_FIELD_BYTES}))"),
+        WHILE_POSITIVE,
+    );
+
+    let outcome = run(&wat, &host).expect("the module should run");
+    assert_eq!(outcome.result, code(HostError::OutOfTransferLimit));
+    assert_eq!(
+        host.fields_asked.borrow().len() as u64,
+        TRANSFER_LIMIT_BYTES / MAX_FIELD_BYTES as u64 + 1,
+        "one call per 1 KiB of budget, plus the one that was refused"
+    );
+}
+
+/// The budget is per run, not per call: a fresh run starts with a full budget.
+#[test]
+fn each_run_gets_its_own_budget() {
+    let wat = until_refused(
+        import::HOME_LE_FIELD,
+        &format!("(call $home_le_field (i32.const 1) (i32.const 0) (i32.const {MAX_FIELD_BYTES}))"),
+        WHILE_POSITIVE,
+    );
+
+    for _ in 0..2 {
+        let host = FakeHost::new().answering_field(1, Answer::filler(MAX_FIELD_BYTES));
+        let outcome = run(&wat, &host).expect("the module should run");
+        assert_eq!(outcome.result, code(HostError::OutOfTransferLimit));
+        assert_eq!(
+            host.fields_asked.borrow().len() as u64,
+            TRANSFER_LIMIT_BYTES / MAX_FIELD_BYTES as u64 + 1
+        );
+    }
+}
+
+/// A run well inside the budget never sees it.
+#[test]
+fn a_modest_run_never_meets_the_budget() {
+    let host = FakeHost::new().answering_field(1, Answer::filler(MAX_FIELD_BYTES));
+    let wat = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        &format!("(call $home_le_field (i32.const 1) (i32.const 0) (i32.const {MAX_FIELD_BYTES}))"),
+    );
+
+    let outcome = run(&wat, &host).expect("the module should run");
+    assert_eq!(outcome.result, MAX_FIELD_BYTES as i32);
+}
+
+/// Reads leave the budget alone: `read_borrowed` hands the host a slice *aliasing*
+/// guest memory, so there are no copied bytes to charge. What bounds how many reads
+/// a run can make is gas, which every host call pays before its body runs.
+///
+/// The observation is the write at the end, not the reads: the module reads four
+/// times the whole budget first, so a rule that charged reads would have nothing
+/// left, and the write would answer `OutOfTransferLimit` instead of a byte count.
+#[test]
+fn reads_do_not_spend_the_transfer_budget() {
+    /// 1 KiB reads, four times over the budget.
+    const READS: u64 = 4 * TRANSFER_LIMIT_BYTES / MAX_FIELD_BYTES as u64;
+
+    let host = FakeHost::new().answering_field(1, Answer::filler(MAX_FIELD_BYTES));
+    let read = trace_call(
+        TraceDataType::AsHex,
+        EMPTY_REGION,
+        &format!("(i32.const 0) (i32.const {MAX_FIELD_BYTES})"),
+    );
+    let wat = module(
+        &[import::TRACE, import::HOME_LE_FIELD, ONE_PAGE],
+        &format!(
+            "(local $i i32)
+             (loop $l
+               {read}
+               (local.set $i (i32.add (local.get $i) (i32.const 1)))
+               (br_if $l (i32.lt_u (local.get $i) (i32.const {READS}))))
+             (call $home_le_field (i32.const 1) (i32.const 0) (i32.const {MAX_FIELD_BYTES}))"
+        ),
+    );
+
+    let outcome = run(&wat, &host).expect("the module should run");
+    assert_eq!(
+        host.traces().len() as u64,
+        READS,
+        "every read should have been served"
+    );
+    assert_eq!(
+        outcome.result, MAX_FIELD_BYTES as i32,
+        "the write after {READS} reads of {MAX_FIELD_BYTES} bytes should still have its budget"
+    );
+}
+
+/// Only the output half of a read-write call spends the budget. `sha512_half`'s
+/// input is a borrowed read like any other, aliasing guest memory rather than
+/// crossing the boundary, so a run may hash far more bytes than the budget holds as
+/// long as the digests it writes fit inside it.
+///
+/// The two totals are asserted, so the arithmetic that makes the case is in the
+/// test rather than in a comment: the inputs alone would overrun the budget, the
+/// digests alone are a small fraction of it.
+#[test]
+fn only_the_output_half_of_a_read_write_spends_the_budget() {
+    /// Enough 1 KiB inputs to overrun the budget twice over.
+    const CALLS: u64 = 2 * TRANSFER_LIMIT_BYTES / MAX_FIELD_BYTES as u64;
+
+    assert!(
+        CALLS * MAX_FIELD_BYTES as u64 > TRANSFER_LIMIT_BYTES,
+        "the inputs alone must overrun the budget"
+    );
+    assert!(
+        CALLS * HASH_LEN as u64 <= TRANSFER_LIMIT_BYTES / 2,
+        "the digests alone must stay well inside it"
+    );
+
+    let host = FakeHost::new().answering_digest(Answer::filler(HASH_LEN));
+    let wat = module(
+        &[import::SHA512_HALF, ONE_PAGE],
+        &format!(
+            "(local $i i32)
+             (local $r i32)
+             (loop $l
+               (local.set $r (call $sha512_half (i32.const 0) (i32.const {MAX_FIELD_BYTES})
+                                                (i32.const 0) (i32.const {HASH_LEN})))
+               (local.set $i (i32.add (local.get $i) (i32.const 1)))
+               (br_if $l (i32.lt_u (local.get $i) (i32.const {CALLS}))))
+             (local.get $r)"
+        ),
+    );
+
+    let outcome = run(&wat, &host).expect("the module should run");
+    assert_eq!(
+        host.digested.borrow().len() as u64,
+        CALLS,
+        "every call should have been served"
+    );
+    assert_eq!(
+        outcome.result, HASH_LEN as i32,
+        "only the digests are charged, and they fit"
+    );
+}

crates/xrpl-wasm-vm/tests/memory_policy.rs

@@ -0,0 +1,621 @@
+//! The bounds, field-cap and buffer-fit rules `abi.rs` enforces on every region
+//! crossing the boundary. This is the policy the guest observes, so each rule is
+//! pinned to the code it answers with.
+
+mod support;
+
+use support::{
+    Answer, COMPLETED, EMPTY_REGION, FakeHost, ONE_PAGE, code, failure, import, module, status,
+    traced,
+};
+use xrpl_host_functions::{HASH_LEN, HostError, TraceDataType};
+use xrpl_wasm_vm::{MAX_FIELD_BYTES, RunError};
+
+/// One page, so anything at or past 65536 is out of bounds.
+const PAGE: i64 = 64 * 1024;
+
+/// The per-field size cap, as a wasm operand.
+const CAP: i64 = MAX_FIELD_BYTES as i64;
+/// One byte over the cap: the smallest value the engine must refuse.
+const OVER_CAP: i64 = CAP + 1;
+
+// ---------------------------------------------------------------------------
+// Output regions (`write_into`)
+// ---------------------------------------------------------------------------
+
+/// The whole output region must be in bounds, not merely its start — the engine
+/// checks `[dst, dst + cap)` before the host is allowed to write.
+#[test]
+fn an_output_region_running_past_memory_is_refused() {
+    let host = FakeHost::new();
+
+    for (dst, cap) in [(PAGE, 4), (PAGE - 3, 4), (PAGE + 1024, 4), (0, PAGE + 1)] {
+        let wat = module(
+            &[import::LDGR_INDEX, ONE_PAGE],
+            &format!("(call $ldgr_index (i32.const {dst}) (i32.const {cap}))"),
+        );
+        assert_eq!(
+            status(&wat, &host),
+            code(HostError::PointerOutOfBounds),
+            "dst {dst} cap {cap}"
+        );
+    }
+}
+
+/// A region ending exactly at the last byte of memory is in bounds.
+#[test]
+fn an_output_region_ending_at_the_last_byte_is_allowed() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[import::LDGR_INDEX, ONE_PAGE],
+        &format!("(call $ldgr_index (i32.const {}) (i32.const 4))", PAGE - 4),
+    );
+    assert_eq!(status(&wat, &host), 4);
+}
+
+/// The wire carries `i32`, so a guest can present a negative pointer or length.
+#[test]
+fn a_negative_output_pointer_or_length_is_refused() {
+    let host = FakeHost::new();
+
+    for (dst, cap) in [(-1, 4), (0, -1), (-1, -1), (i32::MIN, 4)] {
+        let wat = module(
+            &[import::LDGR_INDEX, ONE_PAGE],
+            &format!("(call $ldgr_index (i32.const {dst}) (i32.const {cap}))"),
+        );
+        assert_eq!(
+            status(&wat, &host),
+            code(HostError::InvalidParams),
+            "dst {dst} cap {cap}"
+        );
+    }
+}
+
+/// The host reports a value's true length whether or not it fitted; a value that
+/// did not fit is the guest's error, not the host's.
+#[test]
+fn a_value_larger_than_the_buffer_is_refused() {
+    let host = FakeHost::new().answering_field(1, Answer::filler(64));
+
+    let wat = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        "(call $home_le_field (i32.const 1) (i32.const 0) (i32.const 63))",
+    );
+    assert_eq!(status(&wat, &host), code(HostError::BufferTooSmall));
+
+    let wat = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        "(call $home_le_field (i32.const 1) (i32.const 0) (i32.const 64))",
+    );
+    assert_eq!(status(&wat, &host), 64, "exactly enough room is enough");
+}
+
+/// A zero-length output region is in bounds and simply cannot hold anything.
+#[test]
+fn a_zero_length_output_region_is_in_bounds_but_too_small() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[import::LDGR_INDEX, ONE_PAGE],
+        "(call $ldgr_index (i32.const 0) (i32.const 0))",
+    );
+    assert_eq!(status(&wat, &host), code(HostError::BufferTooSmall));
+}
+
+/// A host that reports more than the per-field cap is refused even when the
+/// guest offered room for it: the cap is the engine's rule, not the buffer's.
+#[test]
+fn a_value_past_the_field_cap_is_refused() {
+    let host = FakeHost::new()
+        .answering_field(1, Answer::claiming(OVER_CAP as usize))
+        .answering_field(2, Answer::claiming(MAX_FIELD_BYTES));
+
+    let wat = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        "(call $home_le_field (i32.const 1) (i32.const 0) (i32.const 4096))",
+    );
+    assert_eq!(status(&wat, &host), code(HostError::DataFieldTooLarge));
+
+    let wat = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        "(call $home_le_field (i32.const 2) (i32.const 0) (i32.const 4096))",
+    );
+    assert_eq!(status(&wat, &host), CAP as i32, "the cap itself is allowed");
+}
+
+/// A refused over-cap value leaves nothing behind. `write_into` hands the host at
+/// most [`MAX_FIELD_BYTES`] of the guest's buffer however much room the guest
+/// declared, so a value past the cap does not fit the region it is offered and no
+/// prefix of it can reach guest memory either.
+///
+/// The host answers with a real over-cap value: [`Answer::claiming`] writes
+/// nothing whatever the engine does, so it could not tell the two apart. The
+/// second module folds the *whole* declared buffer rather than one byte, so the
+/// claim is about the region and not about its first byte.
+#[test]
+fn an_over_cap_value_is_refused_without_reaching_guest_memory() {
+    /// The buffer the guest declares: well over the cap, so the clamp bites.
+    const BUFFER: usize = 4096;
+
+    let over_cap = vec![0xff; MAX_FIELD_BYTES + 1];
+    let host = FakeHost::new().answering_field(1, Answer::bytes(over_cap));
+    let call = format!("(call $home_le_field (i32.const 1) (i32.const 0) (i32.const {BUFFER}))");
+
+    // The status the guest sees, from a module that returns it directly.
+    let refusing = module(&[import::HOME_LE_FIELD, ONE_PAGE], &call);
+    assert_eq!(
+        status(&refusing, &host),
+        code(HostError::DataFieldTooLarge),
+        "the value is refused"
+    );
+
+    // Every byte of the buffer, or-ed together: guest memory starts zero-filled,
+    // so any byte the host wrote shows up here.
+    let reading = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        &format!(
+            "(local $i i32)
+             (local $seen i32)
+             (drop {call})
+             (loop $l
+               (local.set $seen (i32.or (local.get $seen) (i32.load8_u (local.get $i))))
+               (local.set $i (i32.add (local.get $i) (i32.const 1)))
+               (br_if $l (i32.lt_u (local.get $i) (i32.const {BUFFER}))))
+             (local.get $seen)"
+        ),
+    );
+    assert_eq!(
+        status(&reading, &host),
+        0,
+        "and not one of its bytes is in the guest's buffer"
+    );
+}
+
+/// The field cap is checked before the buffer-fit rule, so a value that breaks both
+/// is reported as over-cap. The guest branches on the code, and the two rules
+/// answer different questions, so the order is worth pinning.
+#[test]
+fn the_field_cap_precedes_the_buffer_fit_check() {
+    let host = FakeHost::new().answering_field(1, Answer::claiming(MAX_FIELD_BYTES + 1));
+
+    // A 63-byte buffer: the value is both over the cap and far too big to fit.
+    let wat = module(
+        &[import::HOME_LE_FIELD, ONE_PAGE],
+        "(call $home_le_field (i32.const 1) (i32.const 0) (i32.const 63))",
+    );
+    assert_eq!(status(&wat, &host), code(HostError::DataFieldTooLarge));
+}
+
+// ---------------------------------------------------------------------------
+// Input regions (`Region::read`, via `sha512_half`)
+//
+// `sha512_half`'s first pair is an input region like any other, and it is the
+// input the guest gets a status back from: `trace`, the other reader, answers
+// nothing at all. So the codes are pinned here and the silence below.
+// ---------------------------------------------------------------------------
+
+/// An input region is bounds-checked the same way an output region is. Every case
+/// here stays within the field cap, which on an input is checked first.
+#[test]
+fn an_input_region_running_past_memory_is_refused() {
+    let host = FakeHost::new();
+
+    for (ptr, len) in [(PAGE, 1), (PAGE - 3, 4), (PAGE - 1, CAP)] {
+        let wat = module(
+            &[import::SHA512_HALF, ONE_PAGE],
+            &format!(
+                "(call $sha512_half (i32.const {ptr}) (i32.const {len})
+                                    (i32.const 0) (i32.const {HASH_LEN}))"
+            ),
+        );
+        assert_eq!(
+            status(&wat, &host),
+            code(HostError::PointerOutOfBounds),
+            "ptr {ptr} len {len}"
+        );
+        assert!(host.digested.borrow().is_empty(), "the host is not called");
+    }
+}
+
+#[test]
+fn a_negative_input_pointer_or_length_is_refused() {
+    let host = FakeHost::new();
+
+    for (ptr, len) in [(-1, 1), (0, -1), (i32::MIN, 1)] {
+        let wat = module(
+            &[import::SHA512_HALF, ONE_PAGE],
+            &format!(
+                "(call $sha512_half (i32.const {ptr}) (i32.const {len})
+                                    (i32.const 0) (i32.const {HASH_LEN}))"
+            ),
+        );
+        assert_eq!(
+            status(&wat, &host),
+            code(HostError::InvalidParams),
+            "ptr {ptr} len {len}"
+        );
+    }
+}
+
+/// The field cap bounds what the guest may hand *in*, too.
+#[test]
+fn an_input_past_the_field_cap_is_refused() {
+    let host = FakeHost::new();
+    let digest = |len: i64| {
+        module(
+            &[import::SHA512_HALF, ONE_PAGE],
+            &format!(
+                "(call $sha512_half (i32.const 0) (i32.const {len})
+                                    (i32.const 2048) (i32.const {HASH_LEN}))"
+            ),
+        )
+    };
+
+    assert_eq!(
+        status(&digest(OVER_CAP), &host),
+        code(HostError::DataFieldTooLarge)
+    );
+    assert!(host.digested.borrow().is_empty());
+
+    assert_eq!(
+        status(&digest(CAP), &host),
+        HASH_LEN as i32,
+        "the cap itself is allowed"
+    );
+}
+
+/// The two directions check in opposite orders: an input's length is known before
+/// the read, so the cap comes first, while an output's region has to be resolved
+/// before the host can produce a value, so bounds come first there.
+#[test]
+fn the_field_cap_precedes_the_bounds_check_on_an_input() {
+    let host = FakeHost::new();
+
+    let reading = module(
+        &[import::SHA512_HALF, ONE_PAGE],
+        &format!(
+            "(call $sha512_half (i32.const 0) (i32.const {})
+                                (i32.const 0) (i32.const {HASH_LEN}))",
+            PAGE + 1
+        ),
+    );
+    assert_eq!(status(&reading, &host), code(HostError::DataFieldTooLarge));
+
+    let writing = module(
+        &[import::LDGR_INDEX, ONE_PAGE],
+        &format!("(call $ldgr_index (i32.const 0) (i32.const {}))", PAGE + 1),
+    );
+    assert_eq!(status(&writing, &host), code(HostError::PointerOutOfBounds));
+}
+
+// ---------------------------------------------------------------------------
+// The reader with no result (`read_borrowed`, via `trace`)
+// ---------------------------------------------------------------------------
+
+/// `trace` reads two regions and either one being bad refuses the call. The same
+/// rule as above, and the guest is told nothing: the refusal is the host not being
+/// called, and the run carries on to the constant that follows.
+#[test]
+fn both_of_traces_regions_are_checked_silently() {
+    let host = FakeHost::new();
+    let regions = [
+        (
+            format!("(i32.const {PAGE}) (i32.const 1)"),
+            EMPTY_REGION.to_owned(),
+        ),
+        (
+            EMPTY_REGION.to_owned(),
+            format!("(i32.const {PAGE}) (i32.const 1)"),
+        ),
+        (
+            EMPTY_REGION.to_owned(),
+            format!("(i32.const 0) (i32.const {OVER_CAP})"),
+        ),
+        (
+            "(i32.const -1) (i32.const 1)".to_owned(),
+            EMPTY_REGION.to_owned(),
+        ),
+    ];
+
+    for (msg, data) in regions {
+        let wat = module(
+            &[import::TRACE, ONE_PAGE],
+            &traced(TraceDataType::AsHex, &msg, &data),
+        );
+        assert_eq!(status(&wat, &host), COMPLETED, "msg {msg} data {data}");
+        assert!(
+            host.traces().is_empty(),
+            "msg {msg} data {data}: the host must not be called"
+        );
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Both at once (`write_buffered`, via `sha512_half`)
+// ---------------------------------------------------------------------------
+
+/// A call with an input and an output region decides everything about the input
+/// before anything about the output, so a bad input is reported however the output
+/// region is wrong — out of bounds, or a pointer that is not one at all.
+///
+/// The whole output region, params included, is judged after the host has answered.
+/// Hoisting any part of that above the call would put the output's verdict first for
+/// these cases, and there is no half of it that can be hoisted on a principle the
+/// other half shares.
+#[test]
+fn a_read_write_checks_its_input_before_its_output() {
+    let host = FakeHost::new();
+    let digest = |src: i64, src_len: i64, dst: i64| {
+        module(
+            &[import::SHA512_HALF, ONE_PAGE],
+            &format!(
+                "(call $sha512_half (i32.const {src}) (i32.const {src_len})
+                                    (i32.const {dst}) (i32.const {HASH_LEN}))"
+            ),
+        )
+    };
+
+    let over_cap = digest(0, OVER_CAP, 0);
+    assert_eq!(status(&over_cap, &host), code(HostError::DataFieldTooLarge));
+
+    let out_of_bounds = digest(PAGE, 4, 0);
+    assert_eq!(
+        status(&out_of_bounds, &host),
+        code(HostError::PointerOutOfBounds)
+    );
+
+    // A bad input against each way the output can be wrong: the input's verdict is
+    // the one reported, and the host is never asked for a value nobody can take.
+    for dst in [PAGE, -1] {
+        let both_bad = digest(0, OVER_CAP, dst);
+        assert_eq!(
+            status(&both_bad, &host),
+            code(HostError::DataFieldTooLarge),
+            "dst {dst}"
+        );
+    }
+    assert!(host.digested.borrow().is_empty(), "the host is not reached");
+}
+
+/// The output half of a read-write call obeys the same rules as a plain write.
+#[test]
+fn a_read_write_output_obeys_the_write_rules() {
+    let host = FakeHost::new().answering_digest(Answer::filler(32));
+
+    let wat = module(
+        &[import::SHA512_HALF, ONE_PAGE],
+        "(call $sha512_half (i32.const 0) (i32.const 4) (i32.const 0) (i32.const 31))",
+    );
+    assert_eq!(status(&wat, &host), code(HostError::BufferTooSmall));
+
+    let wat = module(
+        &[import::SHA512_HALF, ONE_PAGE],
+        &format!(
+            "(call $sha512_half (i32.const 0) (i32.const 4) (i32.const {PAGE}) (i32.const 32))"
+        ),
+    );
+    assert_eq!(status(&wat, &host), code(HostError::PointerOutOfBounds));
+}
+
+/// A refused value reaches guest memory in no part, however much of it the host
+/// wrote. The host answers with 32 bytes it did write and a length it did not, so
+/// the refusal happens with the value sitting in the run's output buffer — and the
+/// guest's buffer has to come back untouched.
+///
+/// Stronger than the contract asks for: a guest must not read its buffer on a
+/// negative status. It holds because the buffer is copied to the guest only after
+/// the length, the bounds, the fit and the budget have all passed, so there is no
+/// window in which a refused value is in guest memory.
+#[test]
+fn a_refused_value_leaves_nothing_in_guest_memory() {
+    const MARKER: u8 = 77;
+
+    // The two refusals a value can meet after the host has produced it: longer
+    // than the field cap, and longer than the buffer the guest offered.
+    let refusals = [
+        (MAX_FIELD_BYTES + 1, HASH_LEN, HostError::DataFieldTooLarge),
+        (HASH_LEN, HASH_LEN - 1, HostError::BufferTooSmall),
+    ];
+
+    for (claimed, cap, expected) in refusals {
+        let host =
+            FakeHost::new().answering_digest(Answer::writing_but_claiming([MARKER; 32], claimed));
+        let call = format!(
+            "(call $sha512_half (i32.const 0) (i32.const 4) (i32.const 64) (i32.const {cap}))"
+        );
+
+        let refused = module(&[import::SHA512_HALF, ONE_PAGE], &call);
+        assert_eq!(
+            status(&refused, &host),
+            code(expected),
+            "claiming {claimed}"
+        );
+
+        // The same call, reporting what is at the output region afterwards.
+        let inspect = module(
+            &[import::SHA512_HALF, ONE_PAGE],
+            &format!("(drop {call}) (i32.load8_u (i32.const 64))"),
+        );
+        assert_eq!(
+            status(&inspect, &host),
+            0,
+            "claiming {claimed}: the refused value must not have been written"
+        );
+    }
+}
+
+/// An input region may overlap the output region: the host is served the input as
+/// it stands and its answer lands afterwards, so the two cannot interfere. The
+/// marker is any byte distinct from the input's first (`a`), so `finish` returning
+/// it proves the write landed.
+#[test]
+fn an_input_may_overlap_the_output() {
+    const MARKER: u8 = 99;
+
+    let host = FakeHost::new().answering_digest(Answer::bytes([MARKER; HASH_LEN]));
+
+    let wat = module(
+        &[
+            import::SHA512_HALF,
+            ONE_PAGE,
+            r#"(data (i32.const 0) "abcd")"#,
+        ],
+        &format!(
+            "(drop (call $sha512_half (i32.const 0) (i32.const 4)
+                                      (i32.const 0) (i32.const {HASH_LEN})))
+             (i32.load8_u (i32.const 0))"
+        ),
+    );
+    assert_eq!(
+        status(&wat, &host),
+        i32::from(MARKER),
+        "the output overwrote the input"
+    );
+    assert_eq!(
+        *host.digested.borrow(),
+        vec![b"abcd".to_vec()],
+        "the host saw the input as it was"
+    );
+}
+
+// ---------------------------------------------------------------------------
+// The memory export itself
+// ---------------------------------------------------------------------------
+
+/// A host call with no memory to work in ends the run instead of answering the
+/// guest: there is no buffer for a status to describe, and nothing the guest could
+/// do about the answer — which is what puts this beside out-of-gas on the fatal
+/// channel. What the guest burned getting there is still charged.
+fn assert_no_memory(wat: &str, host: &FakeHost) {
+    let failure = failure(wat, host);
+    assert!(
+        matches!(failure.error, RunError::NoMemory),
+        "expected the run to end for want of a memory export, got: {failure}"
+    );
+    assert!(failure.fuel_used > 0, "{failure}");
+}
+
+/// Every region is relative to the guest's exported memory, so a module without
+/// one cannot make a host call at all.
+#[test]
+fn a_module_that_exports_no_memory_cannot_call_the_host() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[import::LDGR_INDEX, "(memory 1)"],
+        "(call $ldgr_index (i32.const 0) (i32.const 4))",
+    );
+    assert_no_memory(&wat, &host);
+}
+
+/// Having no memory is answered before anything about a call's arguments, so a
+/// module without one ends the run even when its arguments would have earned a
+/// guest-visible code of their own (here an input over the field cap).
+///
+/// The order is deliberate: no memory is a fact about the instance, not about this
+/// call, and a region cannot be validated against a memory that is not there. It
+/// costs the guest nothing — every call such a module makes ends the run anyway.
+#[test]
+fn no_memory_is_answered_before_a_calls_arguments_are() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[import::SHA512_HALF, "(memory 1)"],
+        &format!(
+            "(call $sha512_half (i32.const 0) (i32.const {OVER_CAP})
+                                (i32.const 0) (i32.const {HASH_LEN}))"
+        ),
+    );
+    assert_no_memory(&wat, &host);
+}
+
+/// The memory's export *name* is not part of the contract: the engine takes the
+/// module's memory whatever it is called. Nothing in the wasm spec attaches meaning
+/// to `"memory"` — it is a toolchain convention, so the kind decides.
+#[test]
+fn a_memory_exported_under_any_name_is_the_guests_memory() {
+    let host = FakeHost::new();
+
+    for name in ["mem", "linear", "the memory"] {
+        let wat = module(
+            &[
+                import::LDGR_INDEX,
+                &format!(r#"(memory (export "{name}") 1)"#),
+            ],
+            "(drop (call $ldgr_index (i32.const 64) (i32.const 4)))
+             (i32.load (i32.const 64))",
+        );
+        assert_eq!(
+            status(&wat, &host),
+            7,
+            "the host wrote into the memory exported as '{name}'"
+        );
+    }
+}
+
+/// One memory exported under several names is one memory. The engine resolves the
+/// first export of kind memory, and with at most one memory per module every such
+/// export is that memory, so the order the exports are walked in cannot change the
+/// answer.
+#[test]
+fn one_memory_exported_under_several_names_is_still_that_memory() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[
+            import::LDGR_INDEX,
+            r#"(memory (export "memory") (export "mem") (export "linear") 1)"#,
+        ],
+        "(drop (call $ldgr_index (i32.const 64) (i32.const 4)))
+         (i32.load (i32.const 64))",
+    );
+    assert_eq!(status(&wat, &host), 7);
+}
+
+/// The export has to *be* a memory: a global named `memory` is not one, and it
+/// neither serves as the guest's memory nor hides the memory the module really
+/// exports. The kind decides, so the conventional name carries no weight on
+/// either side.
+#[test]
+fn an_export_named_memory_that_is_not_a_memory_is_not_the_guests_memory() {
+    let host = FakeHost::new();
+
+    let call = "(call $ldgr_index (i32.const 0) (i32.const 4))";
+
+    let wrong_kind = module(
+        &[
+            import::LDGR_INDEX,
+            "(memory 1)",
+            r#"(global (export "memory") i32 (i32.const 0))"#,
+        ],
+        call,
+    );
+    assert_no_memory(&wrong_kind, &host);
+
+    let shadowed = module(
+        &[
+            import::LDGR_INDEX,
+            r#"(memory (export "mem") 1)"#,
+            r#"(global (export "memory") i32 (i32.const 0))"#,
+        ],
+        call,
+    );
+    assert_eq!(
+        status(&shadowed, &host),
+        4,
+        "the real memory is found past the global that took its name"
+    );
+}
+
+/// Bounds follow the memory the module actually declared, not a fixed page.
+#[test]
+fn bounds_follow_the_declared_memory_size() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[import::LDGR_INDEX, r#"(memory (export "memory") 2)"#],
+        &format!("(call $ldgr_index (i32.const {PAGE}) (i32.const 4))"),
+    );
+    assert_eq!(status(&wat, &host), 4, "the second page is in bounds");
+}

crates/xrpl-wasm-vm/tests/preflight.rs

@@ -0,0 +1,516 @@
+//! What screening refuses, and that it refuses nothing a run would have served.
+//!
+//! `check` reaches its verdict from the compiled module alone, so these tests take
+//! no host — except the ones that put the same module through `run` to compare the
+//! two.
+
+mod support;
+
+use support::{ENTRY, FakeHost, ONE_PAGE, PLENTY_OF_GAS, assemble, import, module};
+use xrpl_host_functions::HostFunctionSpec;
+use xrpl_wasm_vm::{CheckError, MAX_MEMORY_PAGES, RunError};
+
+/// Assert which stage screening refused a module at, because the caller maps the
+/// stages separately. The error comes back out for the tests that also read its
+/// message.
+macro_rules! assert_stage {
+    ($refusal:expr, $stage:pat) => {{
+        let refusal = $refusal;
+        assert!(
+            matches!(refusal, $stage),
+            concat!("expected a ", stringify!($stage), " refusal, got: {}"),
+            refusal
+        );
+        refusal
+    }};
+}
+
+/// Screens `wat`, which must assemble.
+fn check(wat: &str) -> Result<(), CheckError> {
+    xrpl_wasm_vm::check(&assemble(wat), ENTRY)
+}
+
+fn refusal(wat: &str) -> CheckError {
+    check(wat).expect_err(&format!("expected this module to be refused:\n{wat}"))
+}
+
+fn passes(wat: &str) {
+    if let Err(refusal) = check(wat) {
+        panic!("expected this module to pass, but: {refusal}\n{wat}");
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Compiling
+// ---------------------------------------------------------------------------
+
+/// A contract that imports a host function, exports its memory and exports the
+/// entry point is what screening is looking for.
+#[test]
+fn a_runnable_contract_passes() {
+    passes(&module(
+        &[import::LDGR_INDEX, ONE_PAGE],
+        "(call $ldgr_index (i32.const 0) (i32.const 4))",
+    ));
+}
+
+/// Bytes that are not a wasm module at all.
+#[test]
+fn garbage_does_not_pass() {
+    for bytes in [b"".as_slice(), b"not wasm", &[0x00, 0x61, 0x73, 0x6d]] {
+        let refusal = xrpl_wasm_vm::check(bytes, ENTRY).expect_err("garbage must not pass");
+        assert_stage!(refusal, CheckError::Compile(_));
+    }
+}
+
+/// Screening takes wasm binaries, and text is not one — the same rule the VM
+/// applies, from the same `wasmi` built without its `wat` feature. Turning that
+/// feature on would make this transaction blob valid at both ends.
+#[test]
+fn a_text_format_module_does_not_pass() {
+    let text = module(&[ONE_PAGE], "(i32.const 0)");
+
+    let refusal =
+        xrpl_wasm_vm::check(text.as_bytes(), ENTRY).expect_err("text must not pass as a module");
+    assert_stage!(refusal, CheckError::Compile(_));
+
+    // The same module, assembled first, passes: the text is sound and only the
+    // format was refused.
+    passes(&text);
+}
+
+/// A feature the engine disables is refused here too, because both stages compile
+/// against the one engine. `vm_limits.rs` walks every disabled feature; this pins
+/// that screening sees the same configuration.
+#[test]
+fn a_disabled_feature_does_not_pass() {
+    let refusal = refusal(&module(
+        &[ONE_PAGE],
+        "(drop (f64.add (f64.const 1) (f64.const 2))) (i32.const 0)",
+    ));
+    let refusal = assert_stage!(refusal, CheckError::Compile(_)).to_string();
+    assert!(refusal.contains("floating-point"), "{refusal}");
+}
+
+// ---------------------------------------------------------------------------
+// Imports
+// ---------------------------------------------------------------------------
+
+/// Every host function the ABI declares, spelled as a guest imports it. The count
+/// is asserted against the ABI so a function added to it cannot be left out here.
+const ALL_IMPORTS: [&str; 61] = [
+    import::LDGR_INDEX,
+    import::PARENT_LDGR_TIME,
+    import::PARENT_LDGR_HASH,
+    import::BASE_FEE,
+    import::AMENDMENT_ENABLED,
+    import::CACHE_LE,
+    import::TX_FIELD,
+    import::HOME_LE_FIELD,
+    import::LE_FIELD,
+    import::TX_INNER,
+    import::HOME_LE_INNER,
+    import::LE_INNER,
+    import::TX_ARR_LEN,
+    import::HOME_LE_ARR_LEN,
+    import::LE_ARR_LEN,
+    import::TX_INNER_ARR_LEN,
+    import::HOME_LE_INNER_ARR_LEN,
+    import::LE_INNER_ARR_LEN,
+    import::CHECK_SIG,
+    import::ACCOUNTROOT_ID,
+    import::AMM_ID,
+    import::CHECK_ID,
+    import::CREDENTIAL_ID,
+    import::DELEGATE_ID,
+    import::DEPOSIT_PREAUTH_ID,
+    import::DID_ID,
+    import::ESCROW_ID,
+    import::TRUSTLINE_ID,
+    import::MPT_ISSUANCE_ID,
+    import::MPTOKEN_ID,
+    import::NFT_OFFER_ID,
+    import::OFFER_ID,
+    import::ORACLE_ID,
+    import::PAYCHAN_ID,
+    import::PERMISSIONED_DOMAIN_ID,
+    import::SIGNERS_ID,
+    import::TICKET_ID,
+    import::VAULT_ID,
+    import::SHA512_HALF,
+    import::TRACE,
+    import::SET_DATA,
+    import::NFT_URI,
+    import::NFT_ISSUER,
+    import::NFT_TAXON,
+    import::NFT_FLAGS,
+    import::NFT_XFER_FEE,
+    import::NFT_SERIAL,
+    import::FLOAT_FROM_INT,
+    import::FLOAT_FROM_UINT,
+    import::FLOAT_FROM_STAMOUNT,
+    import::FLOAT_FROM_STNUMBER,
+    import::FLOAT_TO_INT,
+    import::FLOAT_TO_MANT_EXP,
+    import::FLOAT_FROM_MANT_EXP,
+    import::FLOAT_CMP,
+    import::FLOAT_ADD,
+    import::FLOAT_SUB,
+    import::FLOAT_MULT,
+    import::FLOAT_DIV,
+    import::FLOAT_ROOT,
+    import::FLOAT_POW,
+];
+
+#[test]
+fn every_declared_host_function_may_be_imported() {
+    assert_eq!(
+        ALL_IMPORTS.len(),
+        HostFunctionSpec::ALL.len(),
+        "the ABI gained a host function with no import declaration in this test"
+    );
+
+    let mut parts = ALL_IMPORTS.to_vec();
+    parts.push(ONE_PAGE);
+    passes(&module(&parts, "(i32.const 0)"));
+}
+
+/// A module may import fewer host functions than are registered, but not more.
+#[test]
+fn an_unknown_host_function_does_not_pass() {
+    let refusal = refusal(&module(
+        &[
+            r#"(import "host_lib" "no_such_function" (func $f (param i32) (result i32)))"#,
+            ONE_PAGE,
+        ],
+        "(call $f (i32.const 0))",
+    ));
+    let refusal = assert_stage!(refusal, CheckError::Import(_)).to_string();
+    assert!(
+        refusal.contains("no host function 'no_such_function'"),
+        "{refusal}"
+    );
+}
+
+/// Host functions live under one module name — `host_lib` — and an import naming
+/// another is refused even when the function name is real. `env` is in the list
+/// because that is what plain clang emits.
+#[test]
+fn an_import_from_another_module_does_not_pass() {
+    for module_name in ["host", "env", ""] {
+        let refusal = refusal(&module(
+            &[
+                &format!(
+                    r#"(import "{module_name}" "ldgr_index" (func $f (param i32 i32) (result i32)))"#
+                ),
+                ONE_PAGE,
+            ],
+            "(call $f (i32.const 0) (i32.const 4))",
+        ));
+        let refusal = assert_stage!(refusal, CheckError::Import(_)).to_string();
+        assert!(refusal.contains("is not from 'host_lib'"), "{refusal}");
+    }
+}
+
+/// A host function's name imported as something other than a function. The engine
+/// defines it as a function and nothing else, so this does not link either.
+#[test]
+fn a_host_function_imported_as_a_global_does_not_pass() {
+    let refusal = refusal(&module(
+        &[
+            r#"(import "host_lib" "ldgr_index" (global $g i32))"#,
+            ONE_PAGE,
+        ],
+        "(global.get $g)",
+    ));
+    let refusal = assert_stage!(refusal, CheckError::Import(_)).to_string();
+    assert!(
+        refusal.contains("'host_lib::ldgr_index' is not a function"),
+        "{refusal}"
+    );
+}
+
+/// A module faulty at two stages is refused by the earlier one — it imports what no
+/// engine serves *and* exports no entry point. The imports are what the rest of the
+/// module depends on, so that is the message worth having.
+#[test]
+fn the_earlier_stage_is_the_one_reported() {
+    let refusal = refusal(
+        r#"(module
+             (import "host_lib" "no_such_function" (func $f (result i32)))
+             (memory (export "memory") 1)
+             (func (export "not_the_entry_point") (result i32) (call $f)))"#,
+    );
+
+    assert_stage!(refusal, CheckError::Import(_));
+}
+
+/// The signature is the one part of an import screening does not compare, so a
+/// module that will not link can still pass. Recorded here because it is the gap
+/// this stage leaves, not because it is wanted.
+#[test]
+fn an_import_with_the_wrong_signature_still_passes() {
+    let wat = module(
+        &[
+            r#"(import "host_lib" "ldgr_index" (func $f (param i64 i64) (result i32)))"#,
+            ONE_PAGE,
+        ],
+        "(i32.const 0)",
+    );
+    passes(&wat);
+
+    let host = FakeHost::new();
+    let failure = xrpl_wasm_vm::run(&assemble(&wat), PLENTY_OF_GAS, &host, ENTRY)
+        .expect_err("a mistyped import must not link");
+    assert!(
+        matches!(failure.error, RunError::Instantiate(_)),
+        "{failure}"
+    );
+}
+
+// ---------------------------------------------------------------------------
+// The entry point
+// ---------------------------------------------------------------------------
+
+#[test]
+fn a_missing_entry_point_does_not_pass() {
+    let refusal = refusal(
+        r#"(module (memory (export "memory") 1)
+                              (func (export "other") (result i32) (i32.const 0)))"#,
+    );
+    let refusal = assert_stage!(refusal, CheckError::EntryPoint(_)).to_string();
+    assert_eq!(refusal, "no entry point 'finish'");
+}
+
+/// The entry point is looked up by the name the caller asks for, as a run looks it
+/// up: screening a contract for one entry point says nothing about another.
+#[test]
+fn the_entry_point_is_the_name_the_caller_gives() {
+    let wasm = assemble(
+        r#"(module (memory (export "memory") 1)
+             (func (export "other") (result i32) (i32.const 0)))"#,
+    );
+
+    assert!(xrpl_wasm_vm::check(&wasm, "other").is_ok());
+    assert!(xrpl_wasm_vm::check(&wasm, ENTRY).is_err());
+}
+
+/// Both halves of the entry point's type are screened: a module returning the
+/// wrong thing, or taking anything at all, would fail the run's typed lookup.
+#[test]
+fn an_entry_point_of_the_wrong_type_does_not_pass() {
+    for (signature, body) in [
+        ("(result i64)", "(i64.const 0)"),
+        ("(param i32) (result i32)", "(i32.const 0)"),
+        ("", "(nop)"),
+    ] {
+        let refusal = refusal(&format!(
+            r#"(module (memory (export "memory") 1)
+                 (func (export "finish") {signature} {body}))"#
+        ));
+        let refusal = assert_stage!(refusal, CheckError::EntryPoint(_)).to_string();
+        assert_eq!(
+            refusal, "entry point 'finish' has the wrong signature, expected '() -> i32'",
+            "{signature}"
+        );
+    }
+}
+
+/// An export of the entry point's name that is not a function at all is a third
+/// case, and named as such: nothing is missing and no signature is wrong.
+#[test]
+fn an_entry_point_that_is_not_a_function_does_not_pass() {
+    let refusal = refusal(
+        r#"(module (memory (export "memory") 1) (global (export "finish") i32 (i32.const 0)))"#,
+    );
+    let refusal = assert_stage!(refusal, CheckError::EntryPoint(_)).to_string();
+    assert_eq!(refusal, "export 'finish' is not a function");
+}
+
+// ---------------------------------------------------------------------------
+// Agreement with a run
+// ---------------------------------------------------------------------------
+
+/// A module with no linear memory to export passes. A contract that makes no host
+/// call needs none, and one that does is refused at the call and charged — a
+/// runtime fault, not a malformed module.
+#[test]
+fn a_module_exporting_no_memory_passes() {
+    let wat = r#"(module (func (export "finish") (result i32) (i32.const 0)))"#;
+    passes(wat);
+
+    let host = FakeHost::new();
+    assert_eq!(
+        xrpl_wasm_vm::run(&assemble(wat), PLENTY_OF_GAS, &host, ENTRY)
+            .expect("a module that calls no host function needs no memory")
+            .result,
+        0
+    );
+}
+
+/// Modules spanning what screening decides, each also put through a run.
+fn modules() -> Vec<(&'static str, String)> {
+    vec![
+        (
+            "a runnable contract",
+            module(&[import::LDGR_INDEX, ONE_PAGE], "(i32.const 0)"),
+        ),
+        (
+            "a contract that traps",
+            module(&[ONE_PAGE], "(unreachable)"),
+        ),
+        (
+            "a disabled feature",
+            module(&[ONE_PAGE], "(i32.extend8_s (i32.const 1))"),
+        ),
+        (
+            "an unknown host function",
+            module(
+                &[
+                    r#"(import "host_lib" "nope" (func $f (result i32)))"#,
+                    ONE_PAGE,
+                ],
+                "(call $f)",
+            ),
+        ),
+        (
+            "an import from another module",
+            module(
+                &[
+                    r#"(import "env" "ldgr_index" (func $f (param i32 i32) (result i32)))"#,
+                    ONE_PAGE,
+                ],
+                "(i32.const 0)",
+            ),
+        ),
+        (
+            "a host function imported as a global",
+            module(
+                &[r#"(import "host_lib" "trace" (global $g i32))"#, ONE_PAGE],
+                "(global.get $g)",
+            ),
+        ),
+        (
+            "no entry point",
+            r#"(module (memory (export "memory") 1)
+                 (func (export "other") (result i32) (i32.const 0)))"#
+                .to_string(),
+        ),
+        (
+            "an entry point of the wrong type",
+            r#"(module (memory (export "memory") 1)
+                 (func (export "finish") (result i64) (i64.const 0)))"#
+                .to_string(),
+        ),
+    ]
+}
+
+/// Screening refuses a module exactly when a run would refuse it at one of the
+/// three stages screening covers — nothing it rejects would have run, and nothing
+/// it passes stops before the entry point is called. The exceptions are the ones
+/// [`what_static_screening_cannot_see`] lists.
+#[test]
+fn screening_and_a_run_agree() {
+    let host = FakeHost::new();
+
+    for (label, wat) in modules() {
+        let wasm = assemble(&wat);
+        let refused_early = match xrpl_wasm_vm::run(&wasm, PLENTY_OF_GAS, &host, ENTRY) {
+            Err(failure) => matches!(
+                failure.error,
+                RunError::Compile(_) | RunError::Instantiate(_) | RunError::EntryPoint(_)
+            ),
+            Ok(_) => false,
+        };
+
+        assert_eq!(
+            xrpl_wasm_vm::check(&wasm, ENTRY).is_err(),
+            refused_early,
+            "{label}"
+        );
+    }
+}
+
+/// A module asking for more memory than the engine grants is refused, so the
+/// contract that could never run does not reach the ledger. The cap itself passes.
+#[test]
+fn an_exported_memory_past_the_cap_does_not_pass() {
+    let wat = module(
+        &[&format!(
+            r#"(memory (export "memory") {})"#,
+            MAX_MEMORY_PAGES + 1
+        )],
+        "(i32.const 0)",
+    );
+    let refusal = assert_stage!(refusal(&wat), CheckError::Memory(_)).to_string();
+    assert!(refusal.contains("past the 128-page cap"), "{refusal}");
+
+    passes(&module(
+        &[&format!(r#"(memory (export "memory") {MAX_MEMORY_PAGES})"#)],
+        "(i32.const 0)",
+    ));
+}
+
+/// A declared *maximum* past the cap is legal and simply unreachable, so screening
+/// must not turn it away: `vm_limits` runs this very module to completion.
+#[test]
+fn a_declared_maximum_past_the_cap_still_passes() {
+    passes(&module(
+        &[&format!(
+            r#"(memory (export "memory") 1 {})"#,
+            MAX_MEMORY_PAGES + 1
+        )],
+        "(i32.const 0)",
+    ));
+}
+
+/// The gap, listed rather than described, and now one entry long. A memory a module
+/// keeps to itself is not in its exports, so this is the one module that passes
+/// screening and then fails to *instantiate* — which is why a run's refusal at that
+/// stage cannot be read as the node's fault.
+///
+/// A contract needs an exported memory to make any host call, so a module of this
+/// shape can do nothing but compute; the SDK does not produce one.
+#[test]
+fn what_static_screening_cannot_see() {
+    let host = FakeHost::new();
+    let wat = format!(
+        r#"(module (memory {})
+             (func (export "finish") (result i32) (i32.const 0)))"#,
+        MAX_MEMORY_PAGES + 1
+    );
+
+    passes(&wat);
+
+    let failure = xrpl_wasm_vm::run(&assemble(&wat), PLENTY_OF_GAS, &host, ENTRY)
+        .expect_err("the store's limiter must refuse the memory");
+    assert!(
+        matches!(failure.error, RunError::Instantiate(_)),
+        "{failure}"
+    );
+}
+
+/// A start section is guest code, so screening cannot see whether it traps — but it
+/// no longer has to. A trap is the guest's fault wherever it happens, so the run
+/// charges the contract for what it burned instead of reporting a module the node
+/// should have screened.
+#[test]
+fn a_start_section_screening_cannot_see_is_charged_as_a_trap() {
+    let host = FakeHost::new();
+    let wat = format!(
+        r#"(module {ONE_PAGE}
+             (func $init (unreachable))
+             (start $init)
+             (func (export "finish") (result i32) (i32.const 0)))"#
+    );
+
+    passes(&wat);
+
+    let failure = xrpl_wasm_vm::run(&assemble(&wat), PLENTY_OF_GAS, &host, ENTRY)
+        .expect_err("a start section that traps must not complete the run");
+    assert!(matches!(failure.error, RunError::Trap(_)), "{failure}");
+    assert!(
+        failure.fuel_used > 0,
+        "charged for what it burned: {failure}"
+    );
+}

crates/xrpl-wasm-vm/tests/vm_limits.rs

@@ -0,0 +1,580 @@
+//! What the engine refuses outright: modules it will not compile, will not
+//! instantiate, or cannot find an entry point in — plus the linear-memory cap.
+//!
+//! These are the sandbox's outer wall. Everything here fails the run rather than
+//! returning a code to the guest, so each test reads the failure's message.
+
+mod support;
+
+use support::{
+    FakeHost, ONE_PAGE, PLENTY_OF_GAS, failure, import, module, run, run_entry, run_with_gas,
+};
+use xrpl_wasm_vm::{MAX_MEMORY_PAGES, RunError};
+
+/// Assert which stage a run failed at, because the caller maps the stages to
+/// different outcomes. A stage is one `RunError` variant, so the expectation is a
+/// pattern; the failure comes back out for the tests that also read its message.
+macro_rules! assert_stage {
+    ($failure:expr, $stage:pat) => {{
+        let failure = $failure;
+        assert!(
+            matches!(failure.error, $stage),
+            concat!("expected a ", stringify!($stage), " failure, got: {}"),
+            failure
+        );
+        failure
+    }};
+}
+
+// ---------------------------------------------------------------------------
+// Linear memory
+// ---------------------------------------------------------------------------
+
+/// A module declaring more than the cap fails to instantiate — the limit applies
+/// to the initial memory, not only to growth.
+#[test]
+fn an_initial_memory_past_the_cap_is_refused() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[&format!(
+            r#"(memory (export "memory") {})"#,
+            MAX_MEMORY_PAGES + 1
+        )],
+        "(i32.const 0)",
+    );
+    assert_stage!(failure(&wat, &host), RunError::Instantiate(_));
+}
+
+/// The cap itself is allowed.
+#[test]
+fn an_initial_memory_at_the_cap_is_allowed() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[&format!(r#"(memory (export "memory") {MAX_MEMORY_PAGES})"#)],
+        "(i32.const 0)",
+    );
+    assert_eq!(run(&wat, &host).expect("should run").result, 0);
+}
+
+/// Growth up to the cap succeeds; growth past it traps rather than answering -1 as
+/// `memory.grow` otherwise would, because the engine's limiter sets
+/// `trap_on_grow_failure(true)`.
+#[test]
+fn growth_stops_at_the_cap() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[ONE_PAGE],
+        &format!("(memory.grow (i32.const {}))", MAX_MEMORY_PAGES - 1),
+    );
+    assert_eq!(
+        run(&wat, &host).expect("should run").result,
+        1,
+        "growing to exactly the cap answers the previous size"
+    );
+
+    let wat = module(
+        &[ONE_PAGE],
+        &format!("(memory.grow (i32.const {MAX_MEMORY_PAGES}))"),
+    );
+    assert_stage!(failure(&wat, &host), RunError::Trap(_));
+}
+
+/// A module may declare a maximum above the cap: the cap is enforced on the initial
+/// memory and on growth, not on the memory type's declared bound.
+#[test]
+fn a_declared_maximum_past_the_cap_is_allowed_but_unreachable() {
+    let host = FakeHost::new();
+    let memory = format!(r#"(memory (export "memory") 1 {})"#, MAX_MEMORY_PAGES + 1);
+
+    let wat = module(&[&memory], "(i32.const 0)");
+    assert_eq!(run(&wat, &host).expect("should run").result, 0);
+
+    let wat = module(
+        &[&memory],
+        &format!("(memory.grow (i32.const {MAX_MEMORY_PAGES}))"),
+    );
+    assert_stage!(failure(&wat, &host), RunError::Trap(_));
+}
+
+// ---------------------------------------------------------------------------
+// Engine configuration
+// ---------------------------------------------------------------------------
+
+/// One row per feature `build_wasm_engine` turns off: the smallest module that uses
+/// it, and the fragment of wasmi's refusal that names the feature. A row declaring
+/// its own memory omits [`ONE_PAGE`], or it is refused for having two memories
+/// instead.
+fn disabled_features() -> Vec<(&'static str, Vec<&'static str>, &'static str, &'static str)> {
+    vec![
+        (
+            "wasm_multi_value",
+            vec![
+                ONE_PAGE,
+                "(func $two (result i32 i32) (i32.const 1) (i32.const 2))",
+            ],
+            "(call $two) (drop) (drop) (i32.const 0)",
+            "multi-value",
+        ),
+        (
+            "wasm_sign_extension",
+            vec![ONE_PAGE],
+            "(i32.extend8_s (i32.const 1))",
+            "sign extension",
+        ),
+        (
+            "wasm_bulk_memory",
+            vec![ONE_PAGE],
+            "(memory.fill (i32.const 0) (i32.const 0) (i32.const 1)) (i32.const 0)",
+            "bulk memory",
+        ),
+        (
+            "wasm_reference_types",
+            vec![ONE_PAGE, "(table 1 externref)"],
+            "(i32.const 0)",
+            "reference types",
+        ),
+        // The proposal covers mutable globals crossing the module boundary; an
+        // internal one is core wasm and stays allowed — see the test below.
+        (
+            "wasm_mutable_global",
+            vec![ONE_PAGE, r#"(global (export "g") (mut i32) (i32.const 0))"#],
+            "(i32.const 0)",
+            "mutable global",
+        ),
+        (
+            "wasm_tail_call",
+            vec![ONE_PAGE, "(func $f (result i32) (i32.const 0))"],
+            "(return_call $f)",
+            "tail call",
+        ),
+        // Arithmetic in a constant initialiser. wasmi names the operator rather
+        // than the proposal here.
+        (
+            "wasm_extended_const",
+            vec![
+                ONE_PAGE,
+                "(global $g i32 (i32.add (i32.const 1) (i32.const 2)))",
+            ],
+            "(global.get $g)",
+            "non-constant operator",
+        ),
+        (
+            "wasm_multi_memory",
+            vec![ONE_PAGE, "(memory 1)"],
+            "(i32.const 0)",
+            "multiple memories",
+        ),
+        (
+            "wasm_memory64",
+            vec![r#"(memory (export "memory") i64 1)"#],
+            "(i32.const 0)",
+            "memory64",
+        ),
+        (
+            "wasm_custom_page_sizes",
+            vec![r#"(memory (export "memory") 1 (pagesize 1))"#],
+            "(i32.const 0)",
+            "custom page sizes",
+        ),
+        (
+            "wasm_wide_arithmetic",
+            vec![ONE_PAGE],
+            "(drop (i64.add128 (i64.const 1) (i64.const 2) (i64.const 3) (i64.const 4)))
+             (i32.const 0)",
+            "wide arithmetic",
+        ),
+        // Determinism across nodes is the reason floats are off.
+        (
+            "floats",
+            vec![ONE_PAGE],
+            "(drop (f64.add (f64.const 1) (f64.const 2))) (i32.const 0)",
+            "floating-point",
+        ),
+    ]
+}
+
+/// Every feature the engine disables is refused, and refused for that reason.
+///
+/// `wasm_custom_page_sizes` and `wasm_wide_arithmetic` are off by default in wasmi
+/// 1.1 (`engine/config.rs:72,74`), so their rows guard against wasmi changing that
+/// default rather than against our own config.
+#[test]
+fn every_disabled_feature_is_refused_by_name() {
+    let host = FakeHost::new();
+
+    for (knob, parts, body, expected) in disabled_features() {
+        let wat = module(&parts, body);
+        let failure = assert_stage!(failure(&wat, &host), RunError::Compile(_)).to_string();
+
+        assert!(
+            failure.contains(expected),
+            "{knob}: expected a refusal mentioning {expected:?}, got: {failure}"
+        );
+    }
+}
+
+/// The three knobs [`every_disabled_feature_is_refused_by_name`] cannot cover. The
+/// engine is a process-wide `LazyLock`, so a test observes the one configuration we
+/// build: a knob masked by another, or with no caller-visible effect, has no
+/// distinguishing module.
+#[test]
+fn the_knobs_without_a_module_of_their_own() {
+    let host = FakeHost::new();
+
+    // `wasm_saturating_float_to_int(false)`: every saturating conversion takes a
+    // float operand, so `floats(false)` refuses it first, as the message shows.
+    let wat = module(&[ONE_PAGE], "(i32.trunc_sat_f32_s (f32.const 1))");
+    let refusal = failure(&wat, &host).to_string();
+    assert!(refusal.contains("floating-point"), "{refusal}");
+    assert!(!refusal.contains("saturating"), "{refusal}");
+
+    // `ignore_custom_sections(true)`: governs whether wasmi retains custom
+    // sections, not accept/reject, so this pins only that one is harmless.
+    let wat = module(
+        &[ONE_PAGE, r#"(@custom "note" "ignored")"#],
+        "(i32.const 0)",
+    );
+    assert_eq!(run(&wat, &host).expect("should run").result, 0);
+
+    // `consume_fuel(true)`: with it off, `Store::set_fuel` fails and `run` returns
+    // before instantiating, so every test in the suite fails.
+    let wat = module(&[ONE_PAGE], "(i32.const 0)");
+    assert!(run(&wat, &host).expect("should run").fuel_used > 0);
+}
+
+/// A mutable global the module keeps to itself is core wasm, so the disabled
+/// proposal does not reach it: a guest can still have mutable state.
+#[test]
+fn an_internal_mutable_global_is_still_allowed() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[ONE_PAGE, "(global $g (mut i32) (i32.const 0))"],
+        "(global.set $g (i32.const 7)) (global.get $g)",
+    );
+    assert_eq!(run(&wat, &host).expect("should run").result, 7);
+}
+
+/// Bytes that are not a wasm module at all.
+#[test]
+fn garbage_does_not_compile() {
+    let host = FakeHost::new();
+
+    for bytes in [b"".as_slice(), b"not wasm", &[0x00, 0x61, 0x73, 0x6d]] {
+        let failure = xrpl_wasm_vm::run(bytes, PLENTY_OF_GAS, &host, support::ENTRY)
+            .expect_err("garbage must not compile");
+        assert_stage!(failure, RunError::Compile(_));
+    }
+}
+
+/// The VM takes wasm binaries, and text is not one. wasmi's `wat` feature is on by
+/// default and would have `Module::new` assemble text too, so the crate builds
+/// wasmi without it; turning it back on would make this transaction blob valid.
+#[test]
+fn the_vm_refuses_a_text_format_module() {
+    let host = FakeHost::new();
+    let text = module(&[ONE_PAGE], "(i32.const 0)");
+
+    let failure = xrpl_wasm_vm::run(text.as_bytes(), PLENTY_OF_GAS, &host, support::ENTRY)
+        .expect_err("text must not compile as a module");
+    assert_stage!(failure, RunError::Compile(_));
+
+    // The same module, assembled first, runs: the text is sound and only the
+    // format was refused.
+    assert_eq!(run(&text, &host).expect("should run").result, 0);
+}
+
+// ---------------------------------------------------------------------------
+// Imports
+// ---------------------------------------------------------------------------
+
+/// A module may import fewer host functions than are registered, but not more:
+/// an import the linker does not define fails instantiation.
+#[test]
+fn an_unknown_import_fails_instantiation() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[
+            r#"(import "host_lib" "no_such_function" (func $f (param i32) (result i32)))"#,
+            ONE_PAGE,
+        ],
+        "(call $f (i32.const 0))",
+    );
+    assert_stage!(failure(&wat, &host), RunError::Instantiate(_));
+}
+
+/// Host functions are registered under one module name — `host_lib`, the name the
+/// guest SDK and this repo's fixtures import from — and a guest naming a different
+/// one does not link. `env` is in the list because that is what plain clang emits.
+#[test]
+fn the_import_module_name_must_match() {
+    let host = FakeHost::new();
+
+    for module_name in ["host", "env", ""] {
+        let wat = module(
+            &[
+                &format!(
+                    r#"(import "{module_name}" "ldgr_index" (func $f (param i32 i32) (result i32)))"#
+                ),
+                ONE_PAGE,
+            ],
+            "(call $f (i32.const 0) (i32.const 4))",
+        );
+        assert_stage!(failure(&wat, &host), RunError::Instantiate(_));
+    }
+}
+
+/// An import spelled with the wrong signature does not link even under the right
+/// name, which is what makes the registered signatures load-bearing.
+#[test]
+fn an_import_with_the_wrong_signature_fails_instantiation() {
+    let host = FakeHost::new();
+
+    for signature in [
+        "(param i32) (result i32)",         // too few parameters
+        "(param i32 i32 i32) (result i32)", // too many
+        "(param i64 i64) (result i32)",     // wrong parameter types
+        "(param i32 i32) (result i64)",     // wrong result type
+        "(param i32 i32)",                  // no result
+    ] {
+        let wat = module(
+            &[
+                &format!(r#"(import "host_lib" "ldgr_index" (func $f {signature}))"#),
+                ONE_PAGE,
+            ],
+            "(i32.const 0)",
+        );
+        assert_stage!(failure(&wat, &host), RunError::Instantiate(_));
+    }
+}
+
+/// A module that imports a host function it never calls still has to link.
+#[test]
+fn an_unused_import_is_still_linked() {
+    let host = FakeHost::new();
+
+    let wat = module(
+        &[import::LDGR_INDEX, import::TRACE, ONE_PAGE],
+        "(i32.const 0)",
+    );
+    assert_eq!(run(&wat, &host).expect("should run").result, 0);
+}
+
+// ---------------------------------------------------------------------------
+// The start section
+// ---------------------------------------------------------------------------
+
+/// A start section runs guest code during instantiation, before the entry point
+/// is even looked up, and `set_fuel` and the memory limiter are both installed by
+/// then — so it is metered like any other guest code, and a run it stops is
+/// charged for what it burned.
+///
+/// Reported as a **trap**, not as a module that would not instantiate: a trap is the
+/// guest's fault wherever it happens, and the stage a run stopped at is not what the
+/// caller maps. Filing it under the stage would put a contract's own defect among the
+/// faults a caller treats as the node's, and charge nothing for the instructions the
+/// contract burned reaching it.
+#[test]
+fn a_trapping_start_section_is_a_guest_trap_and_is_charged() {
+    let host = FakeHost::new();
+
+    let wat = format!(
+        r#"(module {ONE_PAGE}
+             (func $init (unreachable))
+             (start $init)
+             (func (export "finish") (result i32) (i32.const 0)))"#
+    );
+    let failure = assert_stage!(
+        run_with_gas(&wat, PLENTY_OF_GAS, &host)
+            .expect_err("a start section that traps must not complete the run"),
+        RunError::Trap(_)
+    );
+    assert!(
+        failure.fuel_used > 0,
+        "the start section's instructions are metered: {failure}"
+    );
+}
+
+/// What `RunError::Instantiate` is left to mean: a module the linker or the store
+/// would not accept, rather than one whose guest code failed. Its two shapes, so the
+/// variant is not left standing for nothing.
+#[test]
+fn instantiation_failure_is_a_module_the_engine_will_not_accept() {
+    let host = FakeHost::new();
+
+    // The linker defines no such import.
+    let wat = module(
+        &[
+            r#"(import "host_lib" "no_such_function" (func $f (result i32)))"#,
+            ONE_PAGE,
+        ],
+        "(call $f)",
+    );
+    assert_stage!(failure(&wat, &host), RunError::Instantiate(_));
+
+    // The store's limiter will not grant the memory, and does not trap to say so.
+    let wat = module(
+        &[&format!("(memory {})", MAX_MEMORY_PAGES + 1)],
+        "(i32.const 0)",
+    );
+    assert_stage!(failure(&wat, &host), RunError::Instantiate(_));
+}
+
+/// A start section that runs out of gas is reported as out of gas, not as a module
+/// that would not instantiate. The stage a run stopped at is not what the caller
+/// maps — the reason is — and gas exhaustion is one outcome wherever the guest
+/// reaches it.
+#[test]
+fn a_start_section_that_exhausts_gas_is_out_of_gas_not_an_instantiation_failure() {
+    const GAS: u64 = 10_000;
+
+    let host = FakeHost::new();
+    let wat = format!(
+        r#"(module {ONE_PAGE}
+             (func $init (loop $l (br $l)))
+             (start $init)
+             (func (export "finish") (result i32) (i32.const 0)))"#
+    );
+
+    let failure = assert_stage!(
+        run_with_gas(&wat, GAS, &host).expect_err("an endless start section must not instantiate"),
+        RunError::OutOfGas
+    );
+    assert_eq!(
+        failure.fuel_used, GAS,
+        "a runaway start section burns the whole limit"
+    );
+}
+
+/// A start section cannot make a host call that needs guest memory, even in a
+/// module that exports one: the memory is resolved from the *instance's* exports,
+/// and instantiation is what produces the instance, so a call made while it is
+/// still running has no memory to work in and ends the run.
+///
+/// Not a choice: `Module::instantiate` is `pub(crate)` in wasmi, so instantiation
+/// cannot be split from the start section to resolve the memory in between.
+#[test]
+fn a_start_section_cannot_make_a_host_call() {
+    let host = FakeHost::new();
+
+    let wat = format!(
+        r#"(module {ldgr_index} {ONE_PAGE}
+             (func $init (drop (call $ldgr_index (i32.const 0) (i32.const 4))))
+             (start $init)
+             (func (export "finish") (result i32) (i32.const 0)))"#,
+        ldgr_index = import::LDGR_INDEX
+    );
+
+    let failure = assert_stage!(
+        run_with_gas(&wat, PLENTY_OF_GAS, &host)
+            .expect_err("a host call from a start section must not be served"),
+        RunError::NoMemory
+    );
+    assert!(
+        failure.fuel_used > 0,
+        "the start section is metered up to the refused call: {failure}"
+    );
+}
+
+// ---------------------------------------------------------------------------
+// The entry point
+// ---------------------------------------------------------------------------
+
+#[test]
+fn a_missing_entry_point_fails() {
+    let host = FakeHost::new();
+
+    let wat = r#"(module (memory (export "memory") 1) (func (export "other") (result i32) (i32.const 0)))"#;
+    let failure = assert_stage!(
+        run_with_gas(wat, PLENTY_OF_GAS, &host)
+            .expect_err("a module without the entry point must not run"),
+        RunError::EntryPoint(_)
+    );
+    assert!(
+        failure.to_string().contains("no entry point 'finish'"),
+        "{failure}"
+    );
+}
+
+/// The entry point is looked up by the name the caller asks for.
+#[test]
+fn the_entry_point_is_the_name_the_caller_gives() {
+    let host = FakeHost::new();
+
+    let wat = r#"(module (memory (export "memory") 1) (func (export "other") (result i32) (i32.const 9)))"#;
+    let outcome = run_entry(wat, &host, "other").expect("should run");
+    assert_eq!(outcome.result, 9);
+}
+
+/// The entry point must take nothing and return an `i32`. A module that exports the
+/// name with another signature is told so, rather than being told the export is
+/// missing: wasmi answers both cases with one error, and "no entry point" would send
+/// a contract author looking for a function they already have.
+#[test]
+fn an_entry_point_of_the_wrong_type_fails() {
+    let host = FakeHost::new();
+
+    for signature in ["(result i64)", "(param i32) (result i32)", ""] {
+        let body = if signature.contains("result i64") {
+            "(i64.const 0)"
+        } else if signature.is_empty() {
+            "(nop)"
+        } else {
+            "(i32.const 0)"
+        };
+        let wat = format!(
+            r#"(module (memory (export "memory") 1) (func (export "finish") {signature} {body}))"#
+        );
+        let failure = assert_stage!(
+            run_with_gas(&wat, PLENTY_OF_GAS, &host)
+                .expect_err("a wrongly-typed entry point must not run"),
+            RunError::EntryPoint(_)
+        )
+        .to_string();
+        assert!(
+            failure.contains("entry point 'finish' has the wrong signature"),
+            "{signature}: {failure}"
+        );
+        assert!(
+            !failure.contains("no entry point"),
+            "a present export must not be reported as absent — {signature}: {failure}"
+        );
+    }
+}
+
+/// An export of the entry point's name that is not a function at all is a third
+/// case, and named as such: nothing is missing and no signature is wrong.
+#[test]
+fn an_entry_point_that_is_not_a_function_fails() {
+    let host = FakeHost::new();
+
+    let wat =
+        r#"(module (memory (export "memory") 1) (global (export "finish") i32 (i32.const 0)))"#;
+    let failure = assert_stage!(
+        run_with_gas(wat, PLENTY_OF_GAS, &host).expect_err("a non-function export must not run"),
+        RunError::EntryPoint(_)
+    )
+    .to_string();
+    assert!(
+        failure.contains("export 'finish' is not a function"),
+        "{failure}"
+    );
+}
+
+/// A guest that traps fails the run rather than returning a value.
+#[test]
+fn a_trapping_guest_fails_the_run() {
+    let host = FakeHost::new();
+
+    let wat = module(&[ONE_PAGE], "(unreachable)");
+    assert_stage!(failure(&wat, &host), RunError::Trap(_));
+
+    // An out-of-bounds guest access is a trap too, caught by the engine rather
+    // than anything the host is asked about.
+    let wat = module(&[ONE_PAGE], "(i32.load (i32.const 100000))");
+    assert_stage!(failure(&wat, &host), RunError::Trap(_));
+}

docs/build/environment.md

@@ -33,9 +33,10 @@ with a single command and without installing anything system-wide:
 nix --experimental-features 'nix-command flakes' develop
 ```
 
-On **Linux**, Nix also provides the compiler (GCC). On **macOS**, the shell uses
-your **system-wide Apple Clang** as the compiler, so you still need to manage
-its version (see below).
+On **Linux**, Nix also provides the compiler (GCC); on **macOS**, it provides
+Clang. If you instead opt to use your system-wide Apple Clang (via
+`nix develop .#apple-clang`), you need to manage its version yourself (see
+below).
 
 See [Using the Nix development shell](./nix.md) for installation and usage
 details, including how to select a different compiler.
@@ -48,10 +49,10 @@ details, including how to select a different compiler.
 
 ### macOS: managing the Apple Clang version
 
-Because the Nix shell uses the system-wide Apple Clang on macOS, the compiler
-version is whatever your installed Xcode (or Command Line Tools) provides. The
-following command should return a version greater than or equal to the
-[minimum required](#tested-compiler-versions):
+If you use your system-wide Apple Clang on macOS (via `nix develop .#apple-clang`),
+the compiler version is whatever your installed Xcode (or Command Line Tools)
+provides. The following command should return a version greater than or equal to
+the [minimum required](#tested-compiler-versions):
 
 ```bash
 clang --version