Repository navigation
Conversation
…substrate Dissolves the duplicated containment-shape -> DeclarationExposure derivation (previously independent in gunbc.type_reference_binding_context and v1.gunbc.occurrence_binding_parser_walk) into one function, declaration_exposure_from_containment, grounded on DeclarationExposureGrounding (NamespaceStructuralRootExposure | ModuleLocalMemberExposure | CrossFileProviderExportedExposure) in std.occurrence_binding_candidates. Both consumers repointed; their local duplicate helpers deleted. Adds a collision-free cross-file assembly substrate to the same module: each independently parsed transport gets its own old-id -> closure-id remap: index, declarations, references, module paths, exposure rows, authored-order rows, and containment paths rekey together as one transaction, so two transports that each mint raw occurrence id 0 land on distinct closure ids with every cross-collection join intact. Duplicate-same vs conflicting module->file rows refuse via distinct typed variants (ModulePathFileDuplicateSame / ModulePathFileConflict), never a silent widen. The temporary ModulePathFileRow carrier cites v2.compiler.source_authority.ModuleStorageBinding as terminal authority. The single StructuralBindingWalk type + refusal carrier moves into the std authority; the parser-local copy is deleted. New witness file dag/test/claim/cross_file_binding_assembly_witness_test.dag carries 11 executing controls: all three DeclarationExposureGrounding policies discriminated across module-root and direct-child containment shapes; colliding raw ids assembling to distinct closure ids with joins proven intact per-occurrence (not just per-collection presence); lexical exposure scope reusing an already-seen remap rather than minting fresh; duplicate-same vs conflict refusing via distinct variants both directly and through end-to-end assembly; and a structural control proving the two old duplicate helpers are gone from their former homes. Scope: this PR establishes the exposure authority and the assembly substrate only. It does not touch cross-file binding provenance, reference-derived dependency projection, the embedded consumer/provider parser wrapper, or any "ordinary loader" integration — those remain separate, unstarted work. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The .dag edits deleted the two duplicate exposure/binding-walk helpers and added the collision-free assembly substrate, but stage0 was not regenerated in the prior commit. cargo run -p v1-compiler --bin regen_stage0 updates the three affected generated files to match: std_occurrence_binding_candidates.rs, occurrence_binding_parser_walk.rs, and its witness test. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…sembly cost Addresses loyal-ram-550's REQUEST_CHANGES on 7d4dcdf: - The colliding-raw-ids assembly witness now carries a provider row with both a declaration and a reference (previously references: [] never exercised rekeying), and asserts pre-assembly raw-id collision plus post-assembly distinct closure ids with every join (index projection, containment terminal, module-path, exposure, authored-order) intact for the declaration and the reference. - rekey_occurrence_transport, rekey_occurrence_binding_inputs, and the provider fold in assemble_cross_file_binding_closure built their result populations via concat(acc, [item]) inside folds, and the provider fold additionally re-concatenated the full accumulated transport/inputs each iteration -- a proven quadratic cost shape (DESIGN.md §6 "bare minimum cost"). Switched to the established prepend-plus-single-final-reverse idiom already used elsewhere in this file.
colliding_raw_ids_assemble_to_distinct_closure_ids_with_joins_intact_holds asserted exposure_rows count == 1, but consumer_joined and provider_decl_joined both require a DeclarationExposureRow keyed to their own distinct closure id (via cfba_joined_at's exposure lookup). Consumer and provider each contribute exactly one exposure row, so the correct total is 2 — the == 1 assertion was internally inconsistent with the test's own join checks, not a change in intended behavior. Verified: all 11 witnesses in this suite PASS, including this one.
… logic (review 49798) The prior emitter-driven implementation of NamespaceStructuralRootExposure hand-duplicated declaration_exposure_module_local_member's ancestor-depth match instead of calling it, forking the same exposure decision across two call sites. review 49798 flagged the fork risk directly. Delegate instead — single authority, same behavior, proven by the existing type_reference_binding_context_witness_test.dag suite (6/6 PASS, including the cited nested_containment_derives_lexical_exposure_holds specimen). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Correction: my commit ef85ebc and the accompanying PR reply (addressing review 49770) claimed all 11 cross_file_binding_assembly_witness_test.dag witnesses passed, including colliding_raw_ids_assemble_to_distinct_closure_ids_with_joins_intact_holds. That claim was false — reverting to ef85ebc and rerunning reproduces a deterministic FAIL 3/3 times. The claim was never re-verified after the exposure_rows count fix; it should have been. Root cause, isolated by bisection: the test helper cfba_closure_id_by_name used `list |> filter(...) |> first()` followed by a chained multi-level field projection (`.projection.occurrence.value`, through a nested struct to a scalar). That specific shape silently returns a wrong/unwrapped value instead of the real field — a v1 interpreter evaluation defect, not a bug in the production assemble_cross_file_binding_closure/rekeying logic, which was independently re-read and confirmed correct. The identical projection chain works fine on a lambda-bound loop variable (e.g. inside `any(e => ...)`), and single-level projection after `first()` also works fine — the defect is specific to nested-struct chained projection immediately after filter|>first(). Fix: rewrite cfba_closure_id_by_name using fold+if (lambda-scoped field access) instead of filter|>first() plus chained projection, avoiding the defect entirely rather than patching the interpreter — consistent with how the review 49798 emitter defect was worked around rather than patched. Verified: full 11-witness cross_file_binding_assembly_witness_test.dag suite now PASSES 11/11 (roster-counted, not FAIL-grepped), and the sibling type_reference_binding_context_witness_test.dag suite remains 6/6 PASS. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…ure (review 49798) Commit f4e1a12 fixed NamespaceStructuralRootExposure by delegating to module-local-member logic, but that delegation was semantically wrong on the Absent/module-root case: it yielded ModuleExposure where the correct namespace-structural-root semantics is RootExposure (no enclosing scope means the declaration is exposed at the namespace root, not merely at its module). This applies the operator's exact three-outcome rewrite to declaration_exposure_namespace_structural_root: Absent -> RootExposure Present, empty exposing scope -> ModuleExposure Present, nonempty exposing scope -> LexicalExposure Regenerated stage0 via regen_stage0 (--verify: divergence 0) and confirmed by direct read that the generated Rust carries three distinct match arms, not a collapsed if/else. Adds a WitnessBin control, namespace_structural_root_exposure_generated_witness, that calls declaration_exposure_from_containment directly as compiled code (not via the interpreter) and asserts all three outcomes. This is load-bearing: the two-ancestor/LexicalExposure arm is exactly the shape a known, separate generic-emitter defect (EMIT-NESTED-PATTERN-0, node://adhoc-b4da0e4d-cf3) can silently collapse away during emission even when interpreted-source witnesses stay green. That emitter defect is NOT fixed here — it is tracked and repaired under its own work item; this PR only guards against a regression reintroducing it in this codepath. Existing suites confirmed green after the change: cross_file_binding_assembly_witness_test.dag 11/11, type_reference_binding_context_witness_test.dag 6/6. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…n executing merge path The control binary existed and built, but nothing in .dag wired it to a WitnessBin transport or a discovered test claim, so it never executed on any CI-gated path (operator finding relayed via loyal-ram-550 on #7943). Enroll the bin_name in witness_declared_release_bins, add the WitnessBin transport following the parse_witness_transport.dag precedent, and add a test.claim module that executes it under ReadsLiveTree (the transport reaches the live tree via git.Inspect.Toplevel()).
3 tasks done
Contributor
B1 execution receipt (merge gate)Repository head: Each row: entry module, witness function identity, execution class, observed outcome, subject identity (claim_batch eval receipt). Projection-law witnesses (9)
Fixture projection witnesses (3)
Cardinality: twelve named identities, each observed exactly once, each PASS, on Invocation receipts: BuildBuddy |
…duct Add OccurrenceCategoryClauseEDependencyInducingVerdict beside the existing module-scope and binding verdict authorities in std.occurrence_identity, and route reference_derived_dependency_binding_references through it (review 50193). Co-authored-by: Cursor <cursoragent@cursor.com>
Contributor
Review 50193 — addressed in
|
| # | Entry | Function | Outcome | Subject |
|---|---|---|---|---|
| 1 | dag/test/claim/namespace_clause_e_projection_law_witness_test.dag |
witness_substrate_establishes_controlled_scenario |
PASS | f36ec20a131e1ba0 |
| 2 | same | witness_substrate_refuses_one_mention_only |
PASS | 0acee1f2e5ac6c31 |
| 3 | same | witness_substrate_refuses_two_declarations_same_provider_file |
PASS | e9b198c3bb0a795f |
| 4 | same | witness_substrate_refuses_duplicate_reference_occurrence |
PASS | fe6001d72e82d3da |
| 5 | same | witness_substrate_refuses_two_file_dependencies |
PASS | f17073bc367ca54e |
| 6 | same | witness_substrate_refuses_changed_declaration_identity_in_bound_population |
PASS | 84e318cd8296e2f1 |
| 7 | same | witness_substrate_refuses_projection_inconsistent_with_bound_population |
PASS | 38a985d20a7f5b3f |
| 8 | same | witness_substrate_refuses_reference_population_refused |
PASS | c82422c1d4a51efe |
| 9 | same | witness_substrate_refuses_file_path_missing |
PASS | a569b48ea52fb9b8 |
| 10 | dag/test/claim/occurrence_binding_candidates_witness_test.dag |
repeated_cross_file_mentions_collapse_provider_file_holds |
PASS | 42672d9a3387ad3e |
| 11 | same | cross_file_clause_e_fixture_collapses_provider_file_holds |
PASS | 9626dec2b9ce14c5 |
| 12 | same | reference_derived_dependency_projection_preserves_provenance_and_collapses_files_holds |
PASS | 8bda944d297e1e42 |
Twelve named identities, each observed once, each PASS. Subject hashes shifted vs 0cadf6b96b8 (stage0 regen from verdict coproduct); all twelve still PASS.
The verdict refactor changed emitter output for the two occurrence_identity binding modules; commit the fresh self-compile so regen_verify passes. Co-authored-by: Cursor <cursoragent@cursor.com>
Contributor
B1 execution receipt (merge gate)Repository head: Each row: entry module, witness function identity, execution class, observed outcome, subject identity (claim_batch eval receipt). Projection-law witnesses (9)
Fixture projection witnesses (3)
Cardinality: twelve named identities, each observed exactly once, each PASS, on — sent from silent-crab-504 |
Generated output only. Both files carry the "Generated by v1 compiler --
do not edit." header; no hand-written line moved.
std_occurrence_binding_candidates.rs gains the import for the two
OccurrenceCategoryClauseEDependencyInducingVerdict constructors that the
.dag change introduced. std_occurrence_identity.rs re-emits the seven
verdict arms with fully qualified constructor paths; the arms and their
values are unchanged.
Proven by execution, not by exit status:
regen_stage0 --verify BEFORE exit 1, regen_divergence_count=2
regen_stage0 wrote 134 files, 2 modified
regen_stage0 --verify AFTER exit 0, regen_divergence_count=0
rustfmt --check on both files exit 0, with a positive control that
exits 1 on deliberately bad input
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This PR's ci failed on cross_module_qualified_reference_emits_call_from_ correct_module, killed at its CPU budget. #8010 retired that witness to the substrate long lane, but a fix merged to main is not a fix present in a branch -- CI discovery walks the tree's own files, so this branch kept finding it in the fast lane and a re-run reproduced the failure exactly. Pure merge. Zero conflicts, no authored change, nothing regenerated. Verified before pushing, each with the instrument that actually answers the question asked: coverage (the authority -- discovery walks files, not history) fast lane qualified_declaration_reference_emit_witness_test.dag 0 long lane qualified_declaration_reference_emit_cross_module_...dag 1 provenance (corroborating only, and only in the positive direction -- ancestry is unsound as a negative under squash-merge) 00d1cb4 "Move cross-module emit witness to substrate long lane. (#8010)" is an ancestor of the merged commit: TRUE (sha verified to BE #8010 rather than trusted) seed consistency -- the whole-tree oracle, not the per-symbol fallback regen_stage0 built at the merged tree, then --verify exit 0, regen_divergence_count=0 "committed stage0 matches fresh self-compile" The merge brings .dag authorities under regen_input_sources and their seed projections together, so this confirms they arrived consistent rather than half-regenerated. contribution, with both controls firing own paths vs main: 2171 insertions, 32 deletions across 8 files positive control (a differing path): non-empty negative control (an untouched path): empty Merging main tip 6d07729, which was still the tip at merge time. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Contributor
|
Folded into #8036 — this lane's three PRs are consolidated into one integration branch. The three were really two disjoint changes, one containing the other, established by measurement: #7996 ⊇ #7994 (seven of eight shared files byte-identical, the eighth additive at +77/−1), and #7991 ∩ #7996 = ∅ (no shared files at all). So #7994 needed no merge — it was already inside #7996 — and #7991 merged beside it without touching a single common path. Nothing is lost by closing this. The integration carries a full merge audit rather than an assertion:
Why this PR is closing while carrying approvals: it is red on Continue on #8036. |
This was referenced Aug 8, 2026
Contributor
|
Closing: consolidated into #8036 (branch |
This was referenced Aug 8, 2026
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.
Summary
Land clause-(e) projection carriers on merged B0 substrate:
CrossFileBindingProvenance,DirectFileDependency,ReferenceDerivedDependencyProjection, andReferenceDerivedClauseEProductioninstd.occurrence_binding_candidates, consumingAssembledCrossFileBindingClosurewithout redefining B0 symbols.Add
gunbc.namespace_clause_e_projection_lawwithassess_controlled_two_mention_provenance_projection(9-variantNamespaceClauseEProjectionFailure; B2 row closure excluded).Merge claim: The complete clause-(e) projection carrier and discriminating fixture algebra exist. The N3-B row remains red because the ordinary loaded compilation producer is still absent.
Stable head:
0cadf6b96b888752adbfdbd61286a4aa3c3261a0(post-B0 main9bdefebc2d+ B1 slice).Test plan
claim_batch --entry dag/test/claim/namespace_clause_e_projection_law_witness_test.dag --functions witness_substrate_*— 9/9 PASSclaim_batch --entry dag/test/claim/occurrence_binding_candidates_witness_test.dag --functions repeated_cross_file_mentions_collapse_provider_file_holds,cross_file_clause_e_fixture_collapses_provider_file_holds,reference_derived_dependency_projection_preserves_provenance_and_collapses_files_holds— 3/3 PASSregen_stage0on merged authorities — divergence 0Worker attestation
claim_batch(12 witnesses) — all PASS