feat: canonical type IDs, P3 component support, 54 WAST fixes (90→36) by avrabe · Pull Request #220 · pulseengine/kiln

avrabe · 2026-04-06T21:34:12Z

Summary

Major push on WebAssembly spec conformance and P3 component model support.

WAST Conformance: 90 → 36 failures (65,597/65,633 = 99.945%)

Runtime fixes

Canonical type IDs with finality + iso-recursive rec-group canonicalization
GcTypeInfo::Func extended with params/results for function type distinction
GcFieldStorage::Ref/RefNull variants preserving concrete type indices
call_indirect canonical type matching with subtype chain walking
any.convert_extern / extern.convert_any round-trip with thread-local side table
br_on_non_null multi-value branch preservation
Element segment GC instructions (array.new, array.new_fixed, struct.new)
array.new_elem / array.init_elem using resolved_elem_items + dropped checks

Validator fixes

Unreachable code type checking (br, br_if, select, local.tee)
br_on_cast nullability + label type validation + fall-through types
table.copy element type compatibility + mixed table32/table64 length
Import type subtyping: PreciseTypeInfo, rec-group matching, finality

Decoder fixes

Empty custom section rejection
Array field mutability byte validation
Custom-descriptors proposal markers (0x4D/0x4C)

P3 Component Support

Library mode for nested components (no _start required)
Lenient import linking for WAC-composed components
Deferred InlineExports resolution
Nested engine discovery and swap for entry point
Inter-component call dispatch via InterComponentHandler
P3 CLI (p3_cli.wasm) runs and prints help text

Linker fixes

Struct field type index relativization in cross-module rec-group comparison
Binary-level global ref type parsing preserving nullability

Test plan

All v0.5.0 components: hello_rust, calculator, datetime, hello_c_cli, hello_cpp_cli, yolo
P3 CLI: runs help text, nested components instantiate
WAST: 65,597 of 65,633 assertions pass (99.945%)
CI pipeline

🤖 Generated with Claude Code

…onversion Implement canonical type IDs for structural type equivalence in ref.test, ref.cast, and br_on_cast with concrete types. Add proper any.convert_extern / extern.convert_any semantics using Value::Ref for internalized externs. Support GC instructions (array.new, array.new_fixed, struct.new) in element segment expression evaluation. Fix array.new_elem to use resolved_elem_items. WAST conformance: 90 → 82 failures (65,551/65,633 = 99.875%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Custom sections must contain at least a valid name per the WebAssembly spec. A custom section with size 0 was incorrectly accepted. Now properly returns 'unexpected end' error. WAST conformance: 82 → 80 failures (65,553/65,633 = 99.878%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ved items - Validate array field mutability byte in decoder (must be 0 or 1) - Check element segment dropped status in array.init_elem and array.new_elem - Use resolved_elem_items in array.init_elem for pre-evaluated GC values WAST conformance: 80 → 77 failures (65,556/65,633 = 99.883%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

BrOnNonNull was only preserving 1 value when branching, but multi-value blocks require preserving all result values. Now uses the same values_to_preserve logic as the Br handler. WAST conformance: 77 → 76 failures (65,557/65,633 = 99.884%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…tity Implement a thread-local side table for preserving original values through the externalize→internalize cycle. Uses high-bit sentinel (0x80000000) in ExternRef.index to distinguish table entries from host-provided externrefs. Host externrefs become opaque anyref (Value::Ref), while round-tripped values recover their original I31Ref/StructRef/ArrayRef identity. WAST conformance: 76 → 73 failures (65,560/65,633 = 99.889%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…mponents Add deferred resolution for InlineExports that reference source instances not yet instantiated. Resolves after all core instances are created. This is needed for WAC-composed P3 components where the section ordering puts InlineExports before their source instance. Note: P3 CLI still fails at nested component instantiation (issue #211) because from_parsed requires _start for library components. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add from_parsed_library() that instantiates components without requiring _start export. Needed for WAC-composed P3 components where nested library components provide functions but don't have a command entry point. Also passes parent's host_registry to nested components for WASI resolution. P3 CLI progress: gets past _start check, now fails on arg_ref resolution for inter-component imports (text:processor/analyzer, etc.). The arg_refs loop is currently unimplemented — this is the next step for full P3 support. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Enable execution of WAC-composed P3 (Preview 3) components with nested library components. The implementation: - Lenient import linking for nested library components: non-WASI imports that can't be resolved (inter-component interfaces) are stub-resolved so the component can still instantiate - Nested engine discovery: when the parent has no core instances, search nested components for _start or wasi:cli/run entry points - Engine swap: take the nested component's engine as the parent's runtime engine, storing it AFTER the main_handle search - Runtime entry point search: scan all instance handles in the engine to find the correct one exporting the entry point - WASI handler re-setup for P3 components with nested instances Tested with p3_cli.wasm from wasm-component-examples v0.5.0. Note: inter-component function calls (fibonacci, analyze) require full arg_ref resolution which is not yet implemented. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Support descriptor (0x4D) and describes (0x4C) type markers from the WebAssembly custom-descriptors proposal. Each clause can appear at most once per type definition; duplicates are rejected as malformed. WAST conformance: 73 → 72 failures (65,561/65,633 = 99.890%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Fix 4 issues in the validator that allowed invalid types in unreachable code: 1. validate_branch_with_module: check concrete values against branch target types even when unreachable (polymorphic underflow still OK) 2. br_if: properly pop branch values and push label types in unreachable code (br_if defines concrete output types regardless of reachability) 3. select: check concrete operands match even in unreachable code 4. local.tee: model [t] -> [t] stack effect so polymorphic underflow pushes the local's type correctly WAST conformance: 72 → 65 failures (65,568/65,633 = 99.901%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Import subtyping (12 of 13 fixed): - PreciseTypeInfo for global ref nullability, rec group boundaries, finality - Binary-level global type parser preserving nullable vs non-nullable - Rec-group-aware type matching for function/tag import validation - Finality checking in cross-module type comparison br_on_cast/fail validation (6 fixed): - Nullability check: rt2 nullable + rt1 non-nullable is type mismatch - Branch label type validation for cast result types - Proper fall-through type computation (diff type rt1\rt2) WAST conformance: 65 → 47 failures (65,586/65,633 = 99.928%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ypes Extend GcTypeInfo::Func to carry param/result types so canonical type IDs distinguish different function signatures. Add GcFieldStorage::Ref/RefNull variants to preserve concrete type indices in struct/array fields. Add call_indirect_type_matches() that uses canonical type IDs with subtype chain walking for proper GC type matching, falling back to structural comparison for pre-GC modules. Update ref_test_value_with_module to prefer canonical IDs when available. WAST conformance: 47 → 36 failures (65,597/65,633 = 99.945%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Per the spec, table.copy's length operand follows the destination table's index type, not dst64 || src64. This fixes mixed table32/table64 copy validation. WAST conformance: 36 → 35 failures (65,598/65,633 = 99.947%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…table64 Corrects the table.copy length operand type for mixed table32/table64 copies. Length is i32 unless BOTH tables are table64. WAST conformance: 35 → 34 failures (65,599/65,633 = 99.948%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ison When comparing struct field types across modules, reference type indices must be relativized to their rec-group positions. Fields containing Ref(idx)/RefNull(idx) now check if the referenced type is inside or outside the rec group and compare accordingly. WAST conformance: 34 → 31 failures (65,602/65,633 = 99.953%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ents Add InterComponentHandler that chains WASI dispatch with inter-component routing for WAC-composed P3 components. Non-WASI function calls are routed to nested component engines based on interface name → component index mapping. The handler tries multiple export name formats: 1. Plain function name 2. Interface-qualified: "interface@version#function" 3. Async-lift: "[async-lift]interface@version#function" The dispatch reaches the correct nested component but P3 async functions use the async protocol (task.return, subtask.drop) which requires a coroutine runtime not yet implemented. Synchronous P3 functions would work; async ones hit call stack exhaustion from the async-lift wrapper. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Add validation that the source table's element type is a subtype of the destination table's element type for table.copy instructions. WAST conformance: 31 → 30 failures (65,603/65,633 = 99.954%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

github-actions · 2026-04-06T21:40:13Z

🔍 Build Diagnostics Report

Summary

Metric	Base Branch	This PR	Change
Errors	0	0	0
Warnings	5	5	0

🎯 Impact Analysis

Issues in Files You Modified

0 new errors introduced by your changes
0 new warnings introduced by your changes
0 total errors in modified files
0 total warnings in modified files
0 files you modified

Cascading Issues (Your Changes Breaking Other Files)

0 new errors in unchanged files
0 new warnings in unchanged files
0 unchanged files now affected

Note: "Cascading issues" are errors in files you didn't modify, caused by your changes (e.g., breaking API changes, dependency issues).

✅ No Issues Detected

Perfect! Your changes don't introduce any new errors or warnings, and don't break any existing code.

📊 Full diagnostic data available in workflow artifacts

🔧 To reproduce locally:

# Install cargo-kiln
cargo install --path cargo-kiln

# Analyze your changes
cargo-kiln build --output json --filter-severity error
cargo-kiln check --output json --filter-severity warning

Implement the P3 async task protocol builtins that Meld-fused components import from the $root namespace: - [context-get-N] / [context-set-N]: per-instance task-local context slots - [task-return]N: capture async function return values (variadic sigs) - [waitable-set-new]: allocate waitable set handles - [waitable-set-poll]: synchronous no-op (returns 0 in sync mode) - [waitable-set-drop] / [waitable-join] / [task-cancel]: no-ops for sync Builtins are dispatched in both import resolution paths (initial call and Instruction::Call handler) before falling through to WASI/default dispatch. Handles Meld's $N instance suffix stripping for multi-instance components. Also adds unlinked import → host handler fallback so fused modules can resolve WASI imports without explicit import links. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

codecov · 2026-04-06T22:02:11Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

Add P3 async builtin dispatch in the call_indirect import handler, which is the path used by Meld-fused modules (they use table-based dispatch for all imports). Also adds host_handler fallback for unlinked WASI imports. Fused P3 modules execute successfully with --function "0" flag. All P3 builtins (context-get/set, task-return, waitable-set-*) are intercepted before WASI dispatch. Adds automatic entry point detection for fused modules: tries numbered exports ("0", "1") when _start is not found. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

This reverts commit 948d0b0.

Add non-nullable StackType variants (NonNullFuncRef, NonNullAnyRef, etc.) and update heap_type_to_stack_type to respect the nullable flag from br_on_cast instruction encoding. Fixes 4 additional assert_invalid tests where the validator accepted modules with nullable/non-nullable type mismatches. WAST conformance: 34 → 30 failures (65,603/65,633 = 99.954%) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

avrabe and others added 17 commits April 5, 2026 23:43

This was referenced Apr 6, 2026

fix(validator): detect excess values on stack in unreachable code #146

Open

fix(runtime): linking, call_indirect, and exception handling execution failures #149

Open

feat: support P3 library component instantiation #211

Open

avrabe and others added 4 commits April 6, 2026 17:07

chore: remove debug logging from validator

948d0b0

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Revert "chore: remove debug logging from validator"

b8090ff

This reverts commit 948d0b0.

avrabe merged commit 9c831f0 into main Apr 7, 2026
17 of 18 checks passed

avrabe deleted the fix/canonical-type-ids branch April 7, 2026 01:15

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: canonical type IDs, P3 component support, 54 WAST fixes (90→36)#220

feat: canonical type IDs, P3 component support, 54 WAST fixes (90→36)#220
avrabe merged 22 commits intomainfrom
fix/canonical-type-ids

avrabe commented Apr 6, 2026

Uh oh!

github-actions bot commented Apr 6, 2026 •

edited

Loading

Uh oh!

codecov bot commented Apr 6, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

avrabe commented Apr 6, 2026

Summary

Runtime fixes

Validator fixes

Decoder fixes

P3 Component Support

Linker fixes

Test plan

Uh oh!

github-actions bot commented Apr 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔍 Build Diagnostics Report

Summary

🎯 Impact Analysis

Issues in Files You Modified

Cascading Issues (Your Changes Breaking Other Files)

✅ No Issues Detected

Uh oh!

codecov bot commented Apr 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

github-actions bot commented Apr 6, 2026 •

edited

Loading

codecov bot commented Apr 6, 2026 •

edited

Loading