Roadmap follow-up: preserve E5 async-function declined-runner residue while widening route parity

[faktorial-ai]

User report of this issue

---

Context

Created by Roadmapper build issue autrun-dj1c8v8ke914-a39f3d3eec on 2026-06-08. ADR 0373 superseded the attempted async-function declined-runner bridge retirement because rejecting route-ineligible async functions broke current async semantics. docs/plans/bytecode-proof-manifest.json now keeps E5b/E5d async-function runner anchors as classified open residue.

Scope

Pick one bounded route-parity slice that reduces async-function declined-runner reliance without replacing fallback execution with rejection. Keep terminal dynamic residue and ordinary sync/class/sync-generator runner owners separate.

Proof Shape

Update docs/plans/bytecode-proof-manifest.json, docs/plans/bytecode-burndown-checklist.md, and focused async-function route/fallback tests together. The proof must show declined async bodies still complete through the classified legacy runner unless the exact shape is semantically owned by the resumable unified route.

Investigation

Blast Radius

• src/Asynkron.JsEngine/Ast/TypedAstEvaluator.AsyncFunctionInvoker.cs - owns async-function resumable route attempts, classified fallback runner construction, and should thread strictness into eligibility.
• src/Asynkron.JsEngine/Execution/UnifiedBytecode/UnifiedBytecodeProductionEligibility.cs - owns the descriptor-backed block-scoped function declaration decline gate and activation descriptor.
• src/Asynkron.JsEngine/Execution/UnifiedBytecode/UnifiedBytecodeVirtualMachine.cs - already has DeclareFunction handling; inspect only if the strict route exposes missing declaration-environment behavior.
• tests/Asynkron.JsEngine.Tests/AsyncAwaitTests.cs - contains the current strict block-function fallback pins and the nearby sloppy fallback completion pin.
• tests/Asynkron.JsEngine.Tests/BytecodeProofManifestTests.cs - validates E5b/E5d manifest shape and may need the new proof row accounted for.
• docs/plans/bytecode-proof-manifest.json - source of the E5b/E5d async declined-runner proof rows; add the new route proof without retiring the source-presence residue.
• docs/plans/bytecode-burndown-checklist.md - human checklist must match the manifest’s narrowed async-function route/fallback boundary.

Dependencies

• ADR 0373 is the controlling contract: route-ineligible async functions must complete through the classified legacy runner, not reject.
• The existing E5-async-function-declined-body-runner-still-present proof should remain open/source-present after this slice.
• No human decision is needed.

User Stories

Story 1: As a runtime maintainer, I want one exact async-function body shape to route through resumable unified bytecode, so that E5 async declined-runner residue shrinks without changing fallback semantics.

Story 2: As a JavaScript semantics reviewer, I want unsupported async-function bodies to keep completing through the classified legacy runner, so that currently valid async programs do not reject or hang.

Approach

• Use the narrow candidate slice: strict async-function block-scoped function declarations that do not need sloppy Annex B var-style hoisting or unowned activation-capture behavior.
• Compute/pass IsStrict into the async-function UnifiedBytecodeProductionActivationDescriptor before EvaluateResumable.
• Narrow the descriptor-backed FunctionDeclarationInstruction resumable decline so the strict async-owned shape can route, while sloppy/Annex B, captured-activation, dynamic, and terminal residue shapes stay declined.
• Convert or add focused async tests: one strict route-hit proof should assert unified-bytecode-resumable-async-fast-path func=run and no classified-async-function-declined-body-runner-residue; one nearby unsupported case should still resolve through the classified fallback.
• Add the corresponding runtime proof row to the manifest and update the checklist wording. Keep E5b/E5d runner source-presence/allowlist rows open.

Risks

• Broadly admitting all descriptor-backed block declarations would accidentally include sloppy Annex B or capture cases.
• Treating this as runner retirement would violate ADR 0373; the classified async fallback helper must remain until every declined shape is owned.
• Async settlement tests can pass incorrectly if promises are not deterministically drained; keep the focused async timeout pattern.

Additional quality gates

• Async fallback preservation - declined async bodies must still complete through classified-async-function-declined-body-runner-residue unless the exact source shape routes.
• No rejection-based retirement - do not replace fallback completion with explicit rejection.
• Owner separation - do not fold terminal dynamic residue or ordinary sync/class/sync-generator runner owners into this async-function slice.

Recommended Labels

• task
• bytecode
• area/runtime
• unified-bytecode
• priority: medium
• size:M

Build Reasoning Effort

High. The code change is narrow, but the correctness boundary crosses async promise settlement, strict block declaration semantics, and manifest source-presence ratchets. Build should avoid broad admission and prove both the new route and the preserved fallback neighbor.

Acceptance Criteria (Refined)

AC-1: A strict async-function block-scoped function declaration resolves correctly through the resumable async route and logs unified-bytecode-resumable-async-fast-path func=run.

AC-2: A nearby unsupported async-function block declaration shape still resolves through classified-async-function-declined-body-runner-residue and does not log the resumable async fast path.

AC-3: CreateClassifiedAsyncDeclinedBodyRunner remains source-present and classified in the E5b/E5d manifest rows.

AC-4: docs/plans/bytecode-proof-manifest.json, docs/plans/bytecode-burndown-checklist.md, and manifest tests are synchronized with the new route proof.

AC-5: Focused async and manifest proof commands pass with the repository async timeout settings.

[faktorial-ai]

Build Update

• Applied runtime notes: none were present for this issue.
• Current-base check: fast-forwarded to latest origin/main before committing; current branch is agent-go/task-gh3491 ahead by 1 commit.
• Scope narrowed: the requested strict async block-declaration route admission is superseded by ADR 0383/current main, which keeps descriptor-backed async block declarations classified until resumable declaration-environment and Annex B blocked-name ownership exists.
• Changed paths: docs/plans/bytecode-proof-manifest.json, docs/plans/bytecode-burndown-checklist.md, tests/Asynkron.JsEngine.Tests/BytecodeProofManifestTests.cs.
• Implementation: added E5-async-block-function-declaration-stays-classified as an executable eligibility proof for the strict async block declaration no-route boundary, synchronized E5d checklist wording, and ratcheted manifest tests around the ADR 0383 boundary.
• Verification: rtk git diff --check passed; focused rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~BytecodeProofManifestTests|FullyQualifiedName~AsyncAwaitTests.AsyncFunction_BlockScopedFunctionDeclaration_UsesClassifiedDeclinedBodyRunner|FullyQualifiedName~AsyncAwaitTests.AsyncFunction_StrictBlockScopedFunctionDeclaration_DoesNotLeakPastBlock|FullyQualifiedName~AsyncAwaitTests.AsyncFunction_SloppyBlockFunctionBlockedByParameter_DoesNotLeakGlobal|FullyQualifiedName~UnifiedBytecodeProductionEligibilityTests.EvaluateResumable_BlockScopedFunctionDeclaration_DeclinesUntilDeclarationEnvironmentOwned|FullyQualifiedName~UnifiedBytecodeProductionEligibilityTests.EvaluateResumable_BlockScopedFunctionDeclarationCapturingActivation_Declines" -- xUnit.MaxParallelThreads=1 -timeout 20000 passed: 254 tests, 7 existing nullable warnings.
• Commit: 46ff4b86d Ratchet async block declaration residue proof.

[faktorial-ai]

Build Update

• Runtime notes: no ## Faktorial Runtime Notes section was supplied.
• Rebased the existing ratchet commit onto latest origin/main after main advanced during the build re-entry.
• Commit: 79f65f03c (Ratchet async block declaration residue proof).
• Scope: kept the ADR 0383 boundary; did not admit descriptor-backed async block declarations to the resumable route.
• Changed paths: docs/plans/bytecode-proof-manifest.json, docs/plans/bytecode-burndown-checklist.md, tests/Asynkron.JsEngine.Tests/BytecodeProofManifestTests.cs.
• Diff stat: 3 files, 25 insertions, 1 deletion.
• Verification: rtk dotnet test tests/Asynkron.JsEngine.Tests/Asynkron.JsEngine.Tests.csproj -c Debug --filter "FullyQualifiedName~AsyncAwaitTests|FullyQualifiedName~BytecodeProofManifestTests" -- xUnit.MaxParallelThreads=1 -timeout 20000 passed: 278 tests.
• Cheap checks: rtk git diff --check passed; origin/main...HEAD shows only 79f65f03c ahead.
• Note: branch tracking still shows divergence from stale origin/agent-go/task-gh3491; no push performed per build-stage instructions.

[faktorial-ai]

Build Update

• Repaired the failed make quality guard by syncing the drift-checked compiler decline-template inventory in docs/unified-bytecode-expansion-contract.md.
• Added the three current UnifiedBytecodeCompiler.cs templates reported by the failing test: nested array RHS boundary, computed property key call-target context, and nested object RHS boundary.
• Preserved the intended scope: no route admission, no async fallback retirement, and no descriptor-backed async block declaration widening.
• Runtime notes: none found in the issue body/comments, so no runtime-note action was applicable.
• Verification: rtk dotnet test tests/Asynkron.JsEngine.Tests --filter "FullyQualifiedName~ExpressionProgramCoverageMapTests.UnifiedBytecodeCompiler_DeclineReasonTemplatesMatchExpansionContract" -- xUnit.MaxParallelThreads=1 -timeout 20000 passed; rtk git diff --check passed.
• Commit: c75413800 Sync unified bytecode decline contract templates. Branch is intentionally not pushed and is ahead 1.

[faktorial-ai]

Build Update

• Runtime notes: none found in the issue body/comments.
• Implementation: committed direct function-body async await using resumable routing. The compiler emits RegisterDisposable with an async-dispose operand, the VM registers async records, and AsyncFunctionInvoker drains async disposer promises before resolving or rejecting.
• Repair: 315c0ee55 fixes the environment ownership guard so function-body await using also forces a materialized resumable body environment; added a plan-level proof test for that guard.
• Boundary: block-scoped using / await using and async-generator await using remain declined; classified async declined-body runner residue is not retired.
• Artifacts: A42 manifest runtime proofs, bytecode burndown checklist, expansion contract, and UnifiedBytecodeResumableUsingDeclarationTests are synchronized.
• Verification: rtk git diff --check origin/main...HEAD passed; static source scan for A42 proofs/route guard passed. Did not run xUnit, build, typecheck, or make quality per build-stage guardrail. Rider file diagnostics were unavailable in this session.
• Commits: a9f815ea6 Route async await using disposal, 315c0ee55 Materialize async await using disposal environment. Branch is ahead of origin/main by 2 commits; no push performed.

[faktorial-ai]

Build Update

• Runtime notes: none found in issue body/comments; no runtime-note action needed.
• Repair focus: prior review did not reach quality commands because .faktorial/main-verify.json was missing in this worktree.
• Implementation state: existing implementation commits remain a9f815ea6 and 315c0ee55; added marker commit 6d2e2889b without committing ignored .faktorial/ runtime state.
• Config repair: recreated .faktorial/main-verify.json from /Users/rogerjohansson/git/asynkron/Asynkron.JsEngine/.faktorial/main-verify.json; python3 -m json.tool passed and cmp -s matched the parent config.
• Verification: rtk git diff --check origin/main...HEAD passed; tracked worktree is clean and one commit ahead. Did not run xUnit, build, typecheck, or make quality per build-stage guardrail.
• Next: requested canonical run-quality targeting review. If it fails after this config repair, use the prior source handoff for await-using/A42/AsyncFunctionInvoker investigation.

[faktorial-ai]

Build Update

• Runtime notes: none found in the issue body/comments; no runtime-note action was needed.
• Repair focus: the prior run-quality failure was UnifiedBytecodeProductionInvocationTests.AwaitUsingDeclaration_DoesNotUseUnifiedBytecodeProductionFastPath, which still expected classified-async-function-declined-body-runner-residue after A42 admitted direct async function-body await using on the resumable async route.
• Implementation: updated that test to preserve the no ordinary sync production fast-path assertion, prove async disposal happens before fulfillment, require unified-bytecode-resumable-async-fast-path func=disposeAsyncLater argc=1, and forbid the classified async fallback residue for this admitted shape.
• Changed path: tests/Asynkron.JsEngine.Tests/UnifiedBytecodeProductionInvocationTests.cs.
• Verification: rtk git diff --check passed; rtk git diff --check origin/main...HEAD passed; narrow rtk rg scans confirmed the repaired test and the existing A42 manifest/checklist route contract are aligned. Did not run xUnit, build, typecheck, or canonical quality commands per build-stage guardrail. Rider file diagnostics were unavailable; no Rider MCP tool was exposed in this session.
• Commit: 7c01f294e Align await using route proof. Branch is clean and ahead of origin/agent-go/task-gh3491 by 7 commits; no push performed.