Repository navigation
Revert #6304 ancestry cache-sharing — fixes String→refinement coercion regression (main-red) - #6310
Merged
Conversation
…coercion regression) #6304 replaced, on single-import chains in build_type_env/build_type_env_unresolved: map_merge(parent.type_env.ancestry_str_bindings, parent.type_env.str_bindings) with: parent.type_env_cache.str_bindings claiming equivalence. It is NOT equivalent: type_env_cache.str_bindings omits the transitive ancestry, so type bindings defined several hops up an import chain (e.g. `ContentHash = NonEmptyStr where brand(...)` and `CommitSha = String` in std.types) become invisible to a single-import descendant. The compiler then inserts a coercion whose target type node has no resolvable name, and the interpreter fails at runtime: "cannot cast String to " (empty target). This reddened BOTH CI gate programs that coerce git-output Strings into brand/ refinement fields — merge_admission_stamp (base_tree_hash: ContentHash) and rust_gates_ci — turning the ci AND rust_tests jobs red on main. Definitive bisect (local, this container): - aa4399e (#6304 parent): merge_admission_stamp -> ExitSuccess - 552477d / d48d5b1 (#6304): -> cast error - this revert: merge_admission_stamp AND rust_gates_ci -> ExitSuccess Restores the proven-correct map_merge. #6304's receipt had scaling tests but no ancestry-completeness test; the Rc-share optimization can re-land once type_env_cache.str_bindings provably equals the full ancestry union (a completeness witness, not just a scaling one). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
Author
|
CI analysis — the
The revert's correctness is verified by execution: with this change, both — sent from lively-raven-355 |
This was referenced Jul 5, 2026
briansrls
added a commit
that referenced
this pull request
Jul 6, 2026
…isted import uses
Closes the §5 fail-open where 'import M { A }' exposed ALL of M: a module could
reference a name it never imported (dag_collect used ErrorNode in 8 fn sigs
without importing it), and the emitter then emitted Rust referencing an un-used
symbol (the batch-4 leak).
Mechanism (hybrid mask, lively-raven-355 reviewed the grounding boundary):
- TypeEnv.source_visible_names = locals + kernel + selective specific_names
+ is_all-module exports + type-params. Computed once in build_type_env.
- resolve_node_bounded threads a 'masked' bit: TRUE at the resolve_node source
entry, FALSE into grounding recursions (peel/resolve_alias/param-alias/:561),
so use-site type-refs are masked but defining-module internals stay available
(no #6310 drop). Invariant recorded as a data-doc at the resolver.
- Leaf mask check emits UnlistedImportUse when masked && name not in
source_visible_names; keeps 'resolved' intact.
Rollout: DIAGNOSTIC-COLLECT — UnlistedImportUse is advisory (non-fatal,
reported via discovery-corpus advisory), so nothing breaks. Measured blast
radius: 15 violations across 6 files (ErrorNode/Symbol/FmaContractionPolicy/
NodeOccurrenceId) — all missing-import fixes. Follow-up fixes those + flips to
hard UnresolvedType for full fail-closed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls
added a commit
that referenced
this pull request
Jul 6, 2026
…tic (diagnostic-collect) (#6321) * WIP: regen * WIP: regen * WIP: regen * WIP: regen * resolve: selective-import fail-closed (diagnostic-collect) — mask unlisted import uses Closes the §5 fail-open where 'import M { A }' exposed ALL of M: a module could reference a name it never imported (dag_collect used ErrorNode in 8 fn sigs without importing it), and the emitter then emitted Rust referencing an un-used symbol (the batch-4 leak). Mechanism (hybrid mask, lively-raven-355 reviewed the grounding boundary): - TypeEnv.source_visible_names = locals + kernel + selective specific_names + is_all-module exports + type-params. Computed once in build_type_env. - resolve_node_bounded threads a 'masked' bit: TRUE at the resolve_node source entry, FALSE into grounding recursions (peel/resolve_alias/param-alias/:561), so use-site type-refs are masked but defining-module internals stay available (no #6310 drop). Invariant recorded as a data-doc at the resolver. - Leaf mask check emits UnlistedImportUse when masked && name not in source_visible_names; keeps 'resolved' intact. Rollout: DIAGNOSTIC-COLLECT — UnlistedImportUse is advisory (non-fatal, reported via discovery-corpus advisory), so nothing breaks. Measured blast radius: 15 violations across 6 files (ErrorNode/Symbol/FmaContractionPolicy/ NodeOccurrenceId) — all missing-import fixes. Follow-up fixes those + flips to hard UnresolvedType for full fail-closed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: regen * WIP: regen * resolve: wire UnlistedImportUse advisory reporting + add discriminating witness Addresses cursor/composer-2.5 REQUEST_CHANGES on #6321: 1. (inert reporting) UnlistedImportUse was emitted into resolved.diagnostics but never surfaced: the discovery-advisory logger (cli_run.rs) required BOTH is_discovery_corpus_advisory AND is_interpreter_blocking, and UnlistedImportUse is advisory-but-non-blocking, so it printed nowhere — §5 spec-without-execution, deficit frequency zeroed by classification. Fix: the logger now surfaces ALL discovery-advisory diagnostics. Every other advisory diagnostic is already interpreter-blocking, so this is a no-op for them and wires only UnlistedImportUse into the reporting path. 2. (no witness) Added unlisted_import_use_witness — an executing consumer with a discriminating input: module_b references Widget while importing only Gadget -> asserts UnlistedImportUse 'Widget' emitted; module_c imports Widget -> asserts NOT flagged (red control against both an inert mask and an over-firing one). Authored in the .dag authority (compiler_tests_rust.dag) + emitted .rs. Passes: 'test compiler_tests::unlisted_import_use_witness ... ok'. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: regen * resolve: add source_visible_names to TypeEnv ctors in v1-compiler-tests Fixes CI rust_tests E0063 on #6321: the Part A TypeEnv field ripple missed 4 constructions in src/v1/tests/src/type_env_scope_chain_test.rs. That file is the separate v1-compiler-tests crate, which -p v1-compiler --tests does not build, so my local check passed while CI (cargo check --workspace) failed. These tests predate the mask and exercise lookup/scope-chain behavior, not masking, so an empty source_visible_names keeps the mask inert (empty-SVN no-op) and all 14 scope-chain tests stay green. Verified: cargo check --workspace --tests clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * WIP: regen * WIP: regen * resolve: add §6 dissolve-on trigger to masked_boundary scaffold carrier Addresses claude-opus-4-7 review on #6321 (DESIGN §6: every scaffold lands with a named dissolution trigger). The resolve_node_bounded_masked_boundary data-doc now names when the diagnostic-collect posture dissolves into §5's fail-closed wall: (a) family-closure SVN lands (imported name → whole coproduct family, owner + all siblings, lively-raven-355 ruling — removes the variant-owner-reverse false-positive class), AND (b) corpus burndown of genuine unlisted uses reaches zero; at which point UnlistedImportUse promotes to a hard UnresolvedType and the empty-SVN guard (:345 count>0, the §3 dual-signal claude flagged) collapses so 'masked' is the sole authority. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
briansrls
added a commit
that referenced
this pull request
Jul 14, 2026
…mary instead of full parent TypeEnv chains (resolver-graph-major-design.md ~L143) — owed transitive-binding RED (#6304/#6310 class: A imports B imports C, C-surface name through B's export); 04_infer.dag change + seed regen in ONE (#6543) * WIP: Inc B interface/body split: build_type_env consumes std.interface_summar * WIP: Inc B interface/body split: build_type_env consumes std.interface_summar * WIP: Inc B interface/body split: build_type_env consumes std.interface_summar * Sync regen after rewire Rc-identity fix for cache decode path. rewire_type_env_parent_links keeps parent.type_env as the canonical binding authority on cache hit; build_type_env consumption still reads interface grain. Co-authored-by: Cursor <cursoragent@cursor.com> * WIP: Inc B interface/body split: build_type_env consumes std.interface_summar * Land explicit dissolution rows for increment-B review nits. Document TypedModule body/interface dual-field transitional shape and the interim v0 signature fingerprint (name + resolved authored-name, not yet structural type hash) with pointers to std.interface_summary dissolution. Co-authored-by: Cursor <cursoragent@cursor.com> * WIP: Inc B interface/body split: build_type_env consumes std.interface_summar * Drop unused get_exported_names import from 04_infer.dag. Only referenced in a docstring data row, not called in this module. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Cursor <cursoragent@cursor.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Reverts the two
build_type_envsingle-import hunks from #6304. Restoresmap_merge(parent.type_env.ancestry_str_bindings, parent.type_env.str_bindings)in place ofparent.type_env_cache.str_bindings.Why — correctness regression breaking main
#6304 claimed
type_env_cache.str_bindingsis equivalent to the ancestrymap_merge. It is not: the cache'sstr_bindingsomits the transitive ancestry, so a type defined several hops up an import chain (e.g.ContentHash = NonEmptyStr where brand(...),CommitSha = Stringinstd.types) is invisible to a single-import descendant. The compiler then inserts a coercion whose target type node has no resolvable name, and the interpreter dies at runtime:This broke both CI gate programs that coerce git-output
Strings into brand/refinement fields:merge_admission_stamp(base_tree_hash: ContentHash) → reddened the ci jobrust_gates_ci→ reddened the rust_tests jobDefinitive bisect (local, this container)
merge_admission_stampaa4399e1d(#6304 parent)552477d0a/d48d5b188(#6304)Perf note
#6304's Rc-share was a real optimization (kills O(M²) map_merge churn on import chains), but its receipt had only a scaling test, no completeness test. It can re-land once
type_env_cache.str_bindingsis proven to equal the full ancestry union — an ancestry-completeness witness that would have caught this. Recommend routing that to the compile-wall burndown lane.🤖 Generated with Claude Code