Skip to content

BS-70: the semantic declarations and modules as source (SemanticDecls) - #491

Draft
humuhumu33 wants to merge 16 commits into
claude/issue-269-semantic-termsfrom
claude/issue-269-semantic-decls
Draft

humuhumu33 wants to merge 16 commits into
claude/issue-269-semantic-termsfrom
claude/issue-269-semantic-decls

Conversation

@humuhumu33

Copy link
Copy Markdown
Contributor

BS-70 is the sixth slice of the typed IR of #269, stacked on #490 (BS-69). It covers the six ordinary declaration kinds of the semantic-module language (structure, class, instance, inductive, definition, theorem) and the module that holds them, in the serde JSON the linked IR embeds. They are language-1.2 source in bootstrap/src/SemanticDecls.lex.tex, over SemanticTerms and SemanticTypes.

Root Oracle Effects
declarationText canonical JSON of the serde form of SemanticDeclaration allocation, overflow, recursion
moduleText canonical JSON of the serde form of SemanticModule allocation, overflow, recursion

Every bootstrap/ and stdlib module declares only these kinds, so moduleText writes LexLean's own semantic modules. The model, reasoning and text declaration kinds follow in BS-71.

The case is crates/conformance/src/cases/bootstrap/semantic_decls.rs. Its corpus:

  • every distinct ordinary declaration of every committed semantic module that links;
  • every such module whose declarations are all ordinary;
  • hand-built vectors with every optional member both present and absent;
  • seeded declarations.

It also has 13 planted defects.

State: draft, for the parent's batch.

  • lexlean check, build and fmt --check pass, and both roots realize exactly their declared effects.
  • The gates and the cases run on case/ir-bs65-70.

🤖 Generated with Claude Code

The ordinary declarations of the semantic module language (structures,
classes, instances, inductives, definitions, theorems; SPEC.md §17.11-§17.12)
and the module holding them as language-1.2 source in bootstrap/, sixth
slice of the typed IR (#269): declarationText and moduleText of
SemanticDecls, held to the serde form of SemanticDeclaration and
SemanticModule in canonical JSON, the latter for every bootstrap and
stdlib module (13 planted defects).

Register row (499), SPEC §31 row and §17.23 prose, scenario, bdd part and
lane, README count and range, CHANGELOG, VERIFICATION record, roots, the
register-count mirrors.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@afflom afflom mentioned this pull request Nov 12, 2026

@afflom afflom left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Bounded diagnostic for the existing BS65–70 owner batch: case-ir run43220787074 uses f694763, which differs from this exact7df356f head only by the added diagnostic workflow. It is separate-run source evidence, not PR qualification.

Two concrete prerequisites are demonstrated:

  • Gates job131191470472 fails R1/RP06 because CONFORMANCE.md is stale against model/*.toml. The official model projection is needed; the spec/register table already reports499 bijective rows, and the five BDD meta-tests and25 Intel-script tests pass.
  • BS65 job131191470609 passes all five evaluator roots (246 runs each), detects23 planted defects and reports3,630 rustc comparisons agreeing, then stops at stack_bounds.rs349: bootstrap/term_text is no package of committed exports. This is missing committed export inventory/provenance, not a stack-exhaustion result or an IR differential disagreement.

Please retain these successful scoped results and reconcile the stale projection plus the affected official export products before another temporary case run. Do not repeat native comparisons merely to rediscover the same prerequisite. Later conformance assertions and the required final-head gate still need acceptance; this does not claim BS65 complete. The other failed case causes have not been diagnosed, and BS67 was still active at the snapshot. Existing owner retains the work; no competing patch, gate action or producer was started.

afflom commented Nov 12, 2026

Copy link
Copy Markdown
Member

Terminal supplement to review5875482730 for separate case-ir43220787074/f694: BS67 job131191470956 ends at14:12:17 with “The runner has received a shutdown signal” and “The operation was canceled.” Metadata reports failure, but the log shows no conformance assertion or explicit timeout. Roughly87 minutes elapsed against the workflow's180-minute bound; the shutdown cause is unproven. Do not classify this as a source defect, OOM, stall or timeout without further evidence.

Preserve the completed scoped results:13 vectors,49 linked projects/212 modules; evaluator agreement on moduleText98, linkedText39 and projectSemanticId38 runs; all14 plants detected. No rustc-agreement, export-acceptance or final case-success summary was reached in the log, so those remain owed.

All seven diagnostic jobs are terminal failures. Existing R1/projection and BS65 export-inventory prerequisites still stand; causes of BS66/68/69/70 were not inferred. Please reconcile those known official products before deciding the necessary next qualification step, retaining these successful portions. No automatic rerun, timeout increase or duplicate producer was initiated. Only the previously uncaptured BS67 log was read; the earlier logs were reused.

humuhumu33 and others added 3 commits November 12, 2026 14:59
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

afflom commented Nov 12, 2026 •

Copy link
Copy Markdown
Member

Delivery coordination for the owner IR batch: six regeneration runs (43243757762, 43243768130, 43243773373, 43243779005, 43243783808, 43243790502) currently use the same broad production-export / records / oracle recipes across different heads and bootstrap configurations. Their recipe scopes overlap, but the differing entrypoint sets/configuration bindings mean the resulting outputs are not automatically interchangeable; shared compiler semantics alone does not prove reuse.

Please identify one canonical combined batch for output adoption and the next qualification, with the concrete remaining blocker and next action. Retain each existing run's exact-head receipts and reuse outputs only where their complete source/configuration/product bindings apply. Resolve the newly reported BS-69 evaluator-coverage mismatch in #490 before treating a limited case as full promised acceptance; the prior #487 lock-error fallback finding also remains open. New #496/#497 should be included in the owner's explicit batch disposition rather than creating another implicit whole regeneration cycle.

These are coordination requests for the existing implementing owner. No running job was canceled or rerun, no competing source/product wave was started, and no partial/canceled run is aggregated into acceptance.

Terminal update (2026-11-12 17:43 UTC): separate case-ir43251954712 atcd0ca (source matches #49730ad plus only its diagnostic workflow) ended with all9 jobs failed. BS72 failed at pinned Lean installation; that metadata does not identify a source defect. BS67 job131299826666 ended on runner shutdown/exit143, with no observed assertion or explicit timeout:49 linked projects/214 modules, evaluator agreement98/39/38 and14 detected plants were reached, but rustc/export completion was not. Retain those scoped results without treating them as full acceptance or reusing them across unmatched configurations. One previously unread BS67 log was archived; no old log or test was repeated. The canonical adoption/blocker/next-action request above remains the next owner checkpoint.

humuhumu33 and others added 12 commits November 12, 2026 15:46
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The stack-bound step of the regeneration maps a package to its root by name,
so a root is the camel case of its package and unique in the project:
semanticTermText, semanticProofText, semanticDeclarationText,
semanticModuleText, and wholeSemanticModuleText.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants