Repository navigation
Conversation
Source-only implementation for #434, preserving shared Decimal parse evidence, exact policies and frozen older-language files. Runtime/compiler identities and official products remain deliberately stale until the coordinated v0.5 wave. This known-incomplete source commit skips CI; final changed-head full acceptance and CI are mandatory before merge. No post-fix policy, output or certificate pass is claimed.
afflom
left a comment
There was a problem hiding this comment.
Source review recorded at exact head77f6dd2d/tree8e9e2dc7. The four-part implementation matches the independently reviewed patch75b21d14: only formatDecimal's Portable type/value roots change; nine Fixed formatting operands and nine RustLemmas formatting applications use the matching existing ToString instances; the constructor's FormatDecimal arm supplies that evidence. The shared Decimal parsing closures/operands, exact policy declarations, runtime-axiom table and all169 older-language files remain unchanged. The15 affected RustLemmas proof-value closures are accounted for; broader proofs sharing the primitive evaluator intentionally change while parse evidence is retained.
The source unit is coherent and has no remaining static finding. This COMMENT is not self-approval or merge clearance. The actual named pre-fix LLV7005 is recorded at3fd873; post-fix authoring/type checks, positive/negative SM39 and certificate evidence, coordinated identities/products and complete final-head CI are still owed. Keep draft, keep #434 and parent #381 open, and do not treat native skip-ci as passing acceptance.
|
Draft on Generated by Claude Code |
SM-39..SM-41 are claimed on release/v0.4 (#493), so decimal formatting isolation is registered as SM-43. The generated projections are release/v0.5's, rewritten by the landing regeneration. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
#441 lands first on release/v0.5; this head follows it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Decimal formatting's runtime and certificate sources move the language-1.2 compiler-semantics ID to 6558e37caeeccfcdd86a54c31db6143364f4c8ae4a0c409c4d07dfaec5c11df2. Every tracked lock whose only change is that ID is relocked with the built compiler (267); LANGUAGE_1_2_SEMANTICS_ID takes the recomputed ID. 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
The ToString operands of Portable, Fixed and RustLemmas were written under the node labels they replaced, but each new constant is a node of its own, so the labels after it named the wrong nodes (LLP2001: "this statement defines node 3602, not node 3600"). The edited statements are unchanged; `lexlean fmt` numbers the nodes again from symbolic labels. All 19 runtime and preservation modules are canonical. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
The canonical graphs move the ID to 8c54fe7b15d5ddf9ff9848559747031aaf161e3793ffa932e46c63e6df475972; written by `just relock-identity`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
Portable and RustLemmas carry both changes: #437's ToString operands for formatting and this branch's LexLeanRuntime.update and its cases, numbered again by `lexlean fmt`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
…olation 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
The isolated decimal formatting no longer reaches Classical.choice or Quot.sound, so the exact policy of the production-coverage root follows (LLV7005 on regeneration). 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
Its lock came from #441 under that change's ID and was skipped by the relock (LLC0102 in fixtures-write). 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
afflom
left a comment
There was a problem hiding this comment.
Bounded output-adoption review: fe53de1 is the direct child of producer input cdd6e45. Official run43685249965 attempt1 completed all six stages, and final artifact14153630597 is present/nonexpired (SHA256256804e7994ab04f9e34062f8dffc8c059944764ac4619c9a43038b822be87e3). Inspected authored crates, workflows, language, features, spec source, docs and bootstrap/src subtrees are unchanged. Generated SPEC/CONFORMANCE now project decimal formatting as SM43. This establishes source preservation and official producer custody, not an independent comparison of every artifact byte or clearance of the earlier cdd integration seam.
The body needs an owner update: it still names SM39, says draft/no post-fix products, and describes pre-405 coordination. Preserve and link the original failure under its historical identifier; do not recreate it. Update the current SM43 mapping, adopted output and remaining acceptance honestly.
Exact CI43693292535 and reproducibility43693292505 report action_required; existing dispatch43693280141 is queued. The authorized owner should resolve that workflow state and consume the existing qualification result. No additional producer is needed merely to publish this evidence. Native151-boundary/27-policy and A/B/E/refusal/export/full final-head obligations and actual merged-tree equality remain required. #434 and broad parent #381 remain open. COMMENT only; no merge clearance.
Independent bounded receipt eac550014f2bc38f14367446028ea4bfc86286442ae2e88c10b93a7ba90e923b.
|
Body updated: SM-43 mapping, historical SM-39 record kept in VERIFICATION.md, adopted output from run 43685249965, remaining acceptance. The two Generated by Claude Code |
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
afflom
left a comment
There was a problem hiding this comment.
Bounded two-file follow-up at1ecb40ae: the SM43 change is one canonical blank line. The certificate source consistently renames the private to_string_instance helper and its sole FormatDecimal call to tostring_instance; its body, operands and selected ToString instance are unchanged. No new source blocker found; prior fe53 adoption credit stays at its original scope.
The revised body correctly records historical SM39/current SM43, original failure and adopted output. Its remaining dispatch43693280141 is now canceled, as are current CI43697574966 and regeneration43697563923. Preserve those results and let the existing owner disposition current qualification; none is acceptance or permission for an automatic retry. Full final-head evidence and actual-tree equality remain owed. Receipt adc29330d39cf9e111f502a678c8b6189be6cb5f38f76edc58179ddc5d1c23a9; no runtime/check repeated.
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
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43704121680), rewrote 5533 paths. The counts writer reported: counts: no count moved
afflom
left a comment
There was a problem hiding this comment.
Bounded output adoption review at986e3b36: direct child of a905, bound to official regenerate43704121680 attempt1. All six stages passed; final artifact14161731255 is nonexpired and input-bound. Inspected authored objects remain unchanged and changed descendants classify as generated products. Prior source/SM43/R3 reviews remain credited. This is API/object custody, not a downloaded artifact byte comparison or complete5533-path byte audit.
Please refresh the PR body's adoption and qualification paragraphs: they still name historical fe53/run43685249965 and old gate43693280141. Retain that historical evidence, but bind current adoption to986e/run43704121680 and current existing dispatch43713178304. The bot CI/repro pair is action_required; the existing dispatch is queued. No full acceptance yet, no repeated writer or original failure requested.
Keep owner-acknowledged serialization6893439291; final-head full gate and actual-tree equality remain necessary. #434 stays open pending acceptance and broad #381 remains open. Receipt a66a4493243ac1976614d06b6b7a6d843e8c65185953b010c6bbb803c760f26a. COMMENT only.
b209368 renumbered Portable's nodes from 3600 on by two but kept the bare node numbers of listIndex's equational-lemma record, so its type and value named the nodes two before them (4316 for 4318, and the inner lambda 4354, with loose bound variables, for 4356); Lean then could not make listIndex.eq_1 when the certification library's Tail asked for it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
#437 formats a decimal from its ToString instance alone, so the equations TargetSemantics.primitive makes for formatDecimal now pass instToStringInt and instToString<width> where they passed instDecimalInt and instDecimalOfFixedOfToString; Factor's stored copies of the nine equations say so (parseDecimal keeps its Decimal evidence), renumbered by lexlean fmt. 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
afflom
left a comment
There was a problem hiding this comment.
Current gate43713178304 has concrete failures beyond its pending lanes. Three failed-job logs were retained once, all checking out986e; no previous producer or source check was repeated.
- rest job132810212143: conformance_sm_43 rejects eight signed formatting/call declarations (Int8/16/32/64 pairs) with LLV7005: observed [propext] under the none axiom policy. Trace the signed formatting dependencies and repair against the intended contract; this log does not identify the introducing dependency or justify weakening the policy.
- preservation job132810212223: conformance_gc_09 fails at Tail.lean661 generating listIndex.eq_1, cannot locate the expected .brecOn application, then Lean reports a loose-bvar panic. This independently confirms #552's reported GC09 symptom. Existing stale EqnInfo triage remains applicable; it still does not establish formatter renumbering as the cause or make EqnInfo.value==declaration.value a valid repair.
- Atlas join132849297282 reaches joined13-shard verification, then R10 fails for missing oracle LexLeanAudit.A979683999a41ac7214a57d69f3238a73.LexLeanExample.Atlas.lean. Disposition/adopt the applicable generated records with source binding and review; the skipped native assertion is not success, and this diagnosis requests no repeated producer.
Existing owner retains this repair batch and serialized coordination6893439291. Preserve all successful/failed/cancelled scopes separately; current full acceptance remains absent. Receipt f6d159596e91804d07aeb0f5e6f071791058311dc6cf8da50a6be0682d29249a contains full logs and SHA256s. COMMENT only; no retry or CI-control action.
|
Triage of gate 43713178304 (fixes are local, held for #437's serialized wave after #441/#545):
Generated by Claude Code |
#443 had merged the open #437 (decimal formatting through ToString, SM-43) several times. That change is not on release/v0.5 and the certification library is stated against the base formatting graph, so GC-09 refused LexLeanCertification.Factor (primitive.eq_3170 and match_29.eq_1 are not what the elaborator's construction makes once formatDecimal takes ToString). This branch now differs from release/v0.5 by the update primitive alone: - Portable is the base module plus LexLeanRuntime.update (112c5ee). - RustLemmas is the base module plus the update cases (42919b1). - Fixed, IdsSM, the semantic-ir feature, README and VERIFICATION are the base's. - The tostring_instance writer, the decimal_format case and the tests/decimal-format fixture are removed. #437 stays its own PR. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3PPDQti4rghmmPSZGy
Lean 4.32.1's ToString at Int8..Int64 reaches Int.repr, whose "-" ++ depends on propext. formatDecimal now takes the runtime class DecimalFormat: ToString at Int and the unsigned widths, and at the signed widths intDecimal of the value's Int, which writes the sign with String.ofList. Bytes are unchanged; parsing keeps its Decimal evidence. Fixed, RustLemmas, Factor and certificate A name the DecimalFormat instances at their nine format operands. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
…solation Generated outputs take the release side (ours where the release has none) for regeneration. SM-43 and SM-44 are both registered: 525 IDs. The compiler-semantics ID is the line's until regeneration relocks it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
The SM-43 repair is implemented (option b, per #434's acceptance) on head
Generated by Claude Code |
afflom
left a comment
There was a problem hiding this comment.
Reviewed the new source at db46237 (0a59099 plus the actual 361dc35 merge), reusing the 986e adoption and failure receipts. The new DecimalFormat graph implements signed fixed widths through intDecimal/toInt with String.ofList and Nat.toDigits, while Int retains exact [propext] and unsigned widths retain []. The nine format operands in Fixed/RustLemmas/Factor and certificate A select the matching instances; parsing keeps its existing Decimal/ToString evidence and the runtime axiom table accounts for Int. All six code/proof blobs match 0a59099 after the merge. The unchanged SM-43 suite still demands 151 boundary/parsing cases, 27 native policy observations, A/B/E, false-output and Int empty-policy controls. This is bounded source credit: the oversized Factor/RustLemmas full graphs, exporter custody, owner-reported local passes and completed final acceptance are not independently established here. The concrete listIndex metadata repair is visible, without establishing a general #552 fix or observed GC09 pass. Run 43952168559 attempt 1 is still in progress on db462, so no generated-output adoption or final gate is credited. Please refresh the body, which still describes ToString/current986e, when the final adoption is available. Existing owner retains exact current output custody, identity, full final-head gates and actual merged-tree equality; #434/#381 and all existing STOP boundaries remain unchanged.
afflom
left a comment
There was a problem hiding this comment.
Producer 43952168559 attempt 1 is now terminal FAILURE at exact db46237. The two newly failed logs were retained once. Production-export job 133608276593 fails at 21:06:40Z in check-production-export --write; records job 133608276705 fails at 21:23:36Z in verify-examples --write. Both stop at production-coverage with LLV7011: named-root extraction: Nat.repr is noncomputable or has no compiled code, so no compiler can realize it. Both checked out db462 and successfully applied the same relocked identity artifact. This identifies the common extraction blocker, not whether the declaration is noncomputable versus lacks compiled code, its introducing source, or a proven repair. Relock identity and oracles passed; combine, counts/stack and delivery skipped, so no completed output adoption or final gate is credited. Existing owner should inspect the production-coverage named-root/dependency extraction for Nat.repr under the pinned toolchain and propose the smallest source/export fix while preserving decimal output and exact axiom policies. Preserve successful identity/oracle outputs and all historical failures; no blind retry or additional wave is requested. The body still describes ToString/current986e and needs current provenance when disposition is available. Prior bounded source credit stays scoped; full SM43/GC09/Atlas/final-head and actual-tree obligations remain, with #434/#381 open and existing STOP boundaries unchanged.
Nat.repr is implemented_by Nat.reprFast, so named-root extraction finds no code for it (LLV7011 at production-coverage in regeneration 43952168559). String.ofList (Nat.toDigits 10 n) is the same bytes and admissible. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
|
Regeneration 43952168559 failed because of LLV7011 at production-coverage. The cause is in
bf13f10 spells the nonnegative case Regeneration has been re-pushed on Generated by Claude Code |
afflom
left a comment
There was a problem hiding this comment.
Bounded source review at bf13f10 (tree878d449e, direct child of reviewed db462) finds the intended concrete dependency repair: intDecimal's nonnegative case now uses String.ofList (Nat.toDigits 10 n), removing its named Nat.repr dependency. The negative case retains '-' followed by digits of n+1; signed toInt instances, Int/unsigned implementations, exact policies, parsing/certificate evidence and SM43 regression source are unchanged by this two-file delta. The production admissibility guard is unchanged. This supports the source repair without independently establishing replacement codegen attributes, exporter custody, universal byte equality or the owner's reported pinned-toolchain/local verify passes. Prior db462 reviews and the preserved failed producer remain scoped evidence. Producer43963500260 attempt1 is in progress on exact bf13, so no output adoption or final gate is credited. Existing owner retains current output/identity custody, full SM43/GC09/Atlas/final-head and actual-tree obligations, and the promised body refresh (still ToString/current986e). #434/#381 remain open and existing STOP boundaries remain unchanged. COMMENT only.
Native language-1.2 decimal formatters inherited parser axioms through the combined
Decimalinstance. This change gives formatting its own evidence and keeps the sharedDecimalparsing evidence and every exact policy declaration.LexLeanRuntime.formatDecimaltakes the runtime classDecimalFormat(0a59099, bf13f10). Its instances:Int:ToString, exactly[propext];ToString,[];intDecimal (x.toInt),[]. This spells both signs withString.ofListoverNat.toDigits, giving the same bytes asInt.repr.Int.repr's"-" ++depends onpropext, andNat.reprhas no code for extraction.instDecimalOfFixedOfToString.Case identifier. Authored as SM-39. After #405's forward merge the v0.5 line numbers it SM-43; generated SPEC/CONFORMANCE now project it as SM43. SM-43 keeps all 151 original boundary/parsing checks and 27 formatter/caller/parser policy observations, using the existing native runtime and A/B/E preservation machinery (explicit replay/audit, typed false-output lowering control, Int empty-policy refusal). Unbounded Int observations stay native. The earlier unmerged closed-decision proof method was revised after its 18 LLV7002 failures; no new proof interface was added.
Historical failure record (kept under its original identifier). Before implementation,
conformance_sm_39ran once on landed #396: 0 passed, 1 failed, 483 filtered, 7.15s, with the intended sole LLV7005 for formatUInt8's empty policy observing[Classical.choice, Quot.sound, propext]. This record stays as SM-39 in VERIFICATION.md; it is not recreated as SM-43.Adopted output (current). None yet. Source head bf13f10 includes the release/v0.5 merge db46237 (361dc35, #553 and #539; 525 IDs). Its regeneration, producer 43963500260, is running. Producer 43952168559 on db46237 failed with LLV7011 (
Nat.repr), which bf13f10 fixes, and that failure is kept as a historical record.Earlier adoption (superseded). 986e3b3 from run 43704121680 (artifact 14161731255), with gate dispatch 43713178304. That gate failed SM-43 on the signed widths'
[propext](comment 6898179750).Historical adoption (superseded, kept for the record). fe53de1 from run 43685249965 (artifact 14153630597). Its gate 43693280141 was cancelled after 1ecb40a fixed the SM-43 blank line and the
tostring_instancelint.Remaining acceptance. Ready (not draft). Required before merge: the full final-head gate on the head that regeneration delivers, the 151 native boundary and 27 policy observations, A/B/E, refusal and export obligations, and merged-tree equality. No positive SM-43 result is claimed until that gate passes.
Closes #434 only on accepted completion. Parent #381 stays open for parsing coupling and the remaining String/decoder work.
Generated by Claude Code