Repository navigation
Patch the single Lean pin to 4.32.2 for kernel soundness - #518
Conversation
Exact tree of locally reviewed source commit c62a9e739e91a07eaa3dee07d6686b51280a5160. Published through the connected GitHub API; commit metadata differs. Generated products and fixed-kernel full acceptance remain pending.
Exact tree of locally reviewed source commit d8be05d3e2e266a8d10ba8e2fa4c6f4ca95e0514. Published through the connected GitHub API; commit metadata differs. Generated products and fixed-kernel full acceptance remain pending.
Exact tree of locally reviewed source commit ba93ce4885f8adf505df284cdf33252f663a319f. Published through the connected GitHub API; commit metadata differs. Generated products and fixed-kernel full acceptance remain pending.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
…he moved identities Relocked with the built compiler every project whose lock was current (324); the two lock-schema mismatch fixtures and the five negative fixtures whose locks do not resolve stay as they are. The three compiler-semantics IDs are carried to their new values in provenance and manifests (never *.md), and LANGUAGE_1_2_SEMANTICS_ID (RP-10, RP-19 from release/v0.4) takes the recomputed language-1.2 ID 7ba1ed3d. The regenerate workflow rewrites the records next. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
afflom
left a comment
There was a problem hiding this comment.
Independent bounded security-scope review confirms the official release binding: leanprover/lean4 v4.32.2~1 is8be817b3f6310f62f220861b0c92dbabb951115d, explicitly closing upstream14576. The release backport preserves the projection-parameter local context and performs final type checking, with regression refusals. v4.32.1→v4.32.2 is exactly that kernel/regression change plus the release-version commit f3b06c7. This supports prioritizing the fixed-kernel migration; no LexLean exploitation or invalid downstream proof is established here.
Current a6aa57d has also published generated identity/lock/provenance/manifest updates and the checked1.2 constant. Please update the source-only/local-writer-results description and old fd7 base description to distinguish the retained authored-source checkpoint from current composed products on actual6a02. The API's first300 changed-file limit is not a complete1469-file product audit; no constant recomputation or complete product acceptance is claimed by this review.
Keep draft until the existing owner finishes coherent fixed-kernel product review, exact refusal/axiom-policy acceptance, full resulting-head CI and reproducibility, followed by actual merged-tree equality. Preserve historical4.32.1 evidence and frozen #316 artifacts in their original scope. #433's dormant4.33.1 work and Rule30's native v0.5 composition remain separate owner obligations; this head is not a drop-in qualification for them. No local tools, producers, tests or CI controls were changed or repeated.
language/lcnf-1.2/tooling is the generator's output (lcnf_registry::files) over the committed registry, whose 53 Lean source digests are unchanged in the 4.32.2 toolchain (checked against its src/lean). Its pin, toolchain and authority were still 4.32.1, which the regenerate production-export writer refused (LLC0101). The language-1.2 identity moves to 4b3ed237; every current lock, record and LANGUAGE_1_2_SEMANTICS_ID follow it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
afflom
left a comment
There was a problem hiding this comment.
Bounded a6aa57→f0e130 review credits the remaining LCNF tooling pin/authority repair: tooling lean-toolchain and lexlean.toml now agree on4.32.2; Extract.lex.tex names LEAN-LCNF-4-32-2 and the retained official f3b06c7 revision. No blocking defect found in those small source patches. The checked1.2 constant also advances to4b3ed237; this review does not independently recompute it or certify all propagated products.
The commit's prior LLC0101 cause, generator provenance and53 unchanged Lean-source digests remain owner-reported; the first300 returned changed files are not an exhaustive product audit. Reuse the already-established upstream8be817 fix binding. Current exact repro43502855845 passed; CI43502855833 is queued and regeneration43502851857 active. Finish this existing product operation and publish its coherent output disposition, update the stale source-only/local-results body, then review/qualify the exact resulting head. Preserve historical4.32.1/#316 receipts and separate native-v0.5 integration. No producer, toolchain mutation or check was repeated.
|
Current fixed-kernel regeneration43502851857 at unchanged f0e130 has completed all three writers and combine successfully. Combined artifact14083892565 is present and nonexpired, digest9511432b17ddd5bd3d61b35e5a79843b4fa70c7dd4f793e0f491b8c7ffc56669; all three part artifacts are retained. The existing counts-and-stack-bounds job132224592583 is still queued, so this is not a completed final patch, adoption or full product acceptance. Retain these exact-source outputs and finish/disposition the remaining phase of this run; no replacement producer is needed merely because publication is pending. Review/adopt the coherent final result, update the source-only/local-results description, and qualify that resulting head with the existing refusal/axiom/full CI/repro guarantees. Current pre-adoption CI43502855833 failed, while exact repro43502855845 passed; neither artifact presence nor partial regeneration substitutes for acceptance. No artifact contents, tests, logs or producers were replayed. |
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/LexLean/actions/runs/43502851857), rewrote 4826 paths. The counts writer reported: counts: no count moved
Their locks kept leanprover/lean4:v4.32.1, so the regenerated expectations gained a second diagnostic (LLC0102, the toolchain) beside the schema mismatch (LLC0103) each fixture exists to show. The locks now pin 4.32.2 with its lean-toolchain digest, keeping their mismatched schemas, and the expectations are again the single LLC0103 (equal to the built compiler's output). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
Ready: merged release/v0.4, relocked, fixed the stale LCNF tooling pin and Generated by Claude Code |
afflom
left a comment
There was a problem hiding this comment.
Bounded adoption review found no new blocking source defect. Official regeneration43502851857 completed all six jobs, including counts/stack bounds and delivery. The chain is reviewed f0e130 → bot adoption689fcc0 → focused fixture repair3576e16. A complete changed-subtree inventory confirms adoption is confined to generated products, records, fixtures and projection; authored compiler/language/model/schema/workflow subtrees are unchanged by adoption. This is lineage/path evidence, not a byte-for-byte artifact audit or universal product proof.
The six-file fixture repair preserves both intentionally mismatched lock schemas/language pairings. It updates their toolchain bindings to4.32.2 and removes the unrelated LLC0102 while retaining single LLC0103, exit2 and failure status. Sampled root/tooling/authority/product pins agree with the reviewed official4.32.2 backport binding. No proof/refusal/axiom check is weakened, and historical4.32.1 evidence retains its original scope.
Current full43535898392 is active; exact repro43535876750 passed. Keep canceled43535876804 excluded. Actual release has advanced to f8854bf through #459 since this head integrated6a02. Review that narrow resulting-tree impact and bind final acceptance to the tree actually landed; no writer is indicated by #459's test-only optimization. Existing owner retains final qualification and landing. Receipt pr518-3576-adoption-source-review/receipt.json; no tests, logs, artifacts or producers replayed.
…inned Lean The gate's OR-01..07 failed: lean4-prod's pinned commit names leanprover/lean4:v4.32.1 in lean/lean-toolchain, which the digest covers, and no upstream commit names 4.32.2, so the oracle was unavailable. The pin gains the optional source_lean_toolchain: exactly the earlier patch release of the pinned Lean's version the oracle's file names, built with lean_toolchain all the same; any other value is a drift, and a pin naming another version or a later patch is malformed (SPEC.md §17.19, OR-01 tests both). The exporter builds with 4.32.2's lake from the pinned commit, and OR-01 passes on the normative host with the oracle acquired and built. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
…urity-patch Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
afflom
left a comment
There was a problem hiding this comment.
Bounded delta review of 3576→9a4d credits the five-file oracle repair bd1ec05 and actual69de integration. source_lean_toolchain explicitly permits a named earlier patch within the same numeric major/minor; absent configuration retains the matching-source requirement. Source4.32.1 is explicit while the actual oracle build still uses pinned4.32.2. Source revision/digest and check_toolchains/verify_checkout remain enforced; assertion coverage includes absent drift and invalid equal/future/different-version names. No additional blocking source defect demonstrated.
The complete21-path delta leaves adopted generated products and the six lock-schema fixture repairs unchanged, so no repeat producer is indicated by this change. The old OR01..07 failure attribution is from the owner commit, not independently established from another log fetch. Old failed/canceled43535898392 is not acceptance; exact-head43555811699 and repro remain required.
Incoming69de preserves actual459/498/492 history; #498's inherited source findings are tracked in6871651933 and are not resolved by merging them. Update the body with the oracle exception/current chronology, retain exact final-tree qualification and actual landing comparison. Receipt pr518-520-69de-integration-review/receipt.json; no tests, logs, or producers repeated.
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43569985914), rewrote 174 paths. The counts writer reported: counts: no count moved
…atch The repository rename makes every Intel test construction job refuse its run as foreign until #533 lands; carrying it here lets this gate pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
…patch Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Lands PF-26 on the Lean 4.32.2 identity (#518 lands first on release/v0.4). The only handwritten conflicts are the language-1.2 ID constant (recomputed next) and fold-laws' README (PF-26 kept, pin 4.32.2). Every generated path is taken from #518 and is regenerated on this tree. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Brings in #535 (Intel macOS on the line). The only conflict was the Unreleased CHANGELOG list; both entries are kept, the Lean 4.32.2 pin first. No file #535 touches names the Lean pin. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Brings in #431, #535, #519 and #518's merge. No conflicts. #507 already carried #518's change and its regenerated outputs (d954d7b), so the combined 1.2 identity is unchanged; #431 only widens the snapshot nesting the canonical writer accepts and does not change any written bytes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
Post-landing evidence audit: actual 4727a69 has parents11fe3d8 +25d5742 and tree4a363be2, whereas the final reviewed proposal tree is5778b739. The actual tree adds12 incoming #519 paths, so the proposal's evidence does not qualify the actual composition by equality. Final proposal run43649392428 attempt1 reports success but contains64 successful jobs/725 successful steps and3 skipped Intel obligations (construction/local targets, conformance, x86_64-apple-darwin build/non-Lean tests). Exact proposal reproducibility passed. Actual run43670267521 is pending with no jobs yet; actual reproducibility43670267454 passed. These are retained scoped facts, not full acceptance under the standing full-premerge instruction. The previously reported external Intel waiver remains in conflict with that instruction; it has not been adopted here. Existing owner should retain this evidence and obtain applicable complete actual-tree qualification plus policy/STOP disposition without synthesizing success across runs or blindly repeating producers. External merge is not retrospective approval. #431/oldTC03 imports received metadata/history inventory only; no stopped source/runtime adoption, #316 admission, retry or CI control occurred. |
…#535, #536) into the arrays branch The generated outputs that both sides rewrote take release/v0.4's copy, or are removed where the line no longer has them; regenerate.yml rewrites the whole set from this tree. Lean moves to 4.32.2 with #518, so the array authority follows the other Lean rows: LEAN-ARRAY-4-32-1 becomes LEAN-ARRAY-4-32-2, cited at f3b06c705e6c. The three files its statement rests on (src/include/lean/lean.h, src/Init/Prelude.lean, src/Init/Data/Array/Set.lean) are byte-identical at f054605aea4b and f3b06c705e6c, so the statement carries over unchanged. runtime-1.2/library.toml keeps the line's toolchain and the branch's Arrays module. LANGUAGE_1_2_SEMANTICS_ID is the merged tree's, as RP-10 recomputes it. CHANGELOG and VERIFICATION keep both sides' entries. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The merge moved the language-1.2 compiler-semantics ID from the line's 4b3ed237 to 946c112c, the ID of the tree with the arrays runtime module; it is carried to every lock, provenance record and manifest that named it (never *.md), as #518 did for 4.32.2. examples/arrays and the five array negative fixtures pinned leanprover/lean4:v4.32.1 and were refused with LLC0101 before reaching their case; they now pin v4.32.2 and are relocked with the built compiler. Every lock the line keeps current checks (324 of 337; the other 13 are negative fixtures whose locks are refused by design), and each array negative fixture reports its expected code. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Source-only security patch, regenerated
Moves the single active Lean pin from 4.32.1 to official 4.32.2, which fixes the kernel soundness defect in leanprover/lean4#14576 (fix: leanprover/lean4#14577). See the official release notes. No evidence here establishes exploitation or an invalid downstream result.
Scope
Updates exact compiler/config/schema pins, all three semantics inputs, authority revision/archive references, active project/toolchain files and generator inputs. No accepted-version range is widened. Official source revision
f3b06c705e6c85f5314019d5d3baab0fec5b580c; source archive SHA-256ab2322bc…45e9, checked against all five authority rows. Frozen older-language runtime bytes, native runtime source and dated evidence are unchanged. PX-07 gains prior-patch Lean/Lake banner and compiler-input-version tamper vectors; CF-14 refuses the 4.32.1 pin (LLC0101).Separate from dormant #420/#433's 4.33.1 migration. Rule30's native v0.5 composition (#439/#482) needs its own fixed-kernel integration on the v0.5 line (#405).
Landing on release/v0.4 (mechanized)
6a02d2c7d5(Releases and regeneration need no hand steps (§9.4) #517, Fix ambiguous count rewrites (stacked on #442) #462); CHANGELOG keeps both entries.a6aa57db1f,f0e130a882: every currentlexlean.lockrelocked with the built compiler (324; the two lock-schema mismatch fixtures and five negative fixtures whose locks do not resolve are kept);language/lcnf-1.2/tooling(generator output, missed by the source patch) now citesLEAN-LCNF-4-32-2— its 53 Lean source digests intests/lcnf/registry.jsonare unchanged in the 4.32.2 toolchain'ssrc/lean;LANGUAGE_1_2_SEMANTICS_ID(RP-10/RP-19) takes the recomputed language-1.2 ID4b3ed237…; the IDs are carried in provenance/manifests.689fcc0cc4: regenerate run 43502851857 (all writers, counts and stack bounds, deliver: success).3576e16ff9: the two lock-schema mismatch fixtures pin 4.32.2 so each shows only itsLLC0103.Fixed-kernel acceptance
elan toolchain install leanprover/lean4:v4.32.2reportsLean (version 4.32.2, x86_64-unknown-linux-gnu, commit f3b06c705e6c85f5314019d5d3baab0fec5b580c, Release), the pinned revision; verification preflight refuses any other banner (PX-07, CF-14).