Skip to content

Relock and repin a moved compiler-semantics ID in regenerate.yml (just relock-identity) - #545

Merged
afflom merged 20 commits into
release/v0.5from
ci/relock-identity
Nov 15, 2026
Merged

afflom merged 20 commits into
release/v0.5from
ci/relock-identity

Conversation

@afflom

@afflom afflom commented Nov 14, 2026 •

Copy link
Copy Markdown
Member

Requested by Alex · project thread

Every identity-moving PR so far (#441, #437, #443, #542, ...) has failed its first regenerate run. It fails first on LLC0102 lexlean.lock is stale, then on LLC0203 descriptor pins, because nothing relocks or repins before the writers run, so each PR was relocked by hand.

What this adds

  • just relock-identity runs .github/scripts/relock-identity.sh:
    • It refuses a tree with tracked changes. It snapshots the existing .lexlean/ build roots and removes only the ones the run created. It restores only side effects of its own steps.
    • It builds HEAD's compiler and reads lexlean production targets.
    • It writes the Rust backend records (cargo xtask semantics-records --write, new: rust_renderer, rust_runtime_core, rust_runtime_std), rebuilds, and reads the descriptors again.
    • It relocks every tracked lexlean.lock and keeps a new lock only when its sole change is compiler_semantics, moving from the ID most locks of its language carry at HEAD. Fixtures with their own ID, locks that don't resolve, and lock-schema mismatches are put back.
    • It rewrites LANGUAGE_1_2_SEMANTICS_ID (RP-10).
    • It moves every production-request pin that named HEAD's descriptor of its target to the new descriptor, for all targets at once. A pin that named no HEAD descriptor is a fixture's own and stays. No request is built.
  • regenerate.yml gets a new first job, identity, which runs the recipe once and uploads regenerate-part-identity.
    • Every writer needs it, applies its patch, and commits it locally, so a writer's part holds only what its own recipes wrote.
    • combine applies it first.
    • The patch is empty when the ID did not move.
  • SPEC §9.4 (spec/src/Chapter09 and the regenerated SPEC.md) describes the job. CHANGELOG has an entry.
  • RP-14 reads only ci.yml, and RP-05 fixes only the gate and write recipes it lists, so neither changes.

#521: this retires its per-writer relock step for v0.5. #521 (release/v0.4, 59a3378) relocks bootstrap/stdlib/compiler in each writer and reports the result only from production-export. On v0.5 the identity job publishes the relock once, for all 340 locks plus the ID constant and the pins, so the writers need no step of their own. When v0.4 merges forward, keep identity and drop #521's "Refresh the project locks" step and its git reset of the three locks, so that only one implementation remains.

Tested

🤖 Generated with Claude Code

https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX

claude and others added 8 commits November 8, 2026 11:33
The target's value semantics clones a record to read one field, and Rust
copies an in-place record whole on every clone and move. For a program
whose states are large records this dominates run time (84% memcpy in the
uorc search engine) and gives functions frames large enough to overflow
the default stack. rust-std now boxes every field whose type is a datatype
with a field-carrying constructor, reusing the existing box/unbox
rendering and its certificate; rust-core is unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3Nk2H9geAZ2GKFCNvM
A shared box realizes `indirection` like a cyclic one, so the rendering
check accepts the box a non-recursive record field now lowers to.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3Nk2H9geAZ2GKFCNvM
The renderer record takes the digest of the merged renderer sources.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
The rust-std renderer record moves the language-1.2 compiler-semantics ID to
a5ffe0622f46da9bbde0573c9cfea5ac6afbd17192f0c85474d46e7330652b75. Every tracked lock whose only change is that ID is relocked with the
built compiler (268); the lock-schema mismatch fixtures and the negative
fixtures whose locks do not resolve keep their bytes. 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
The three negative fixtures whose request pins rust-core now pin its
descriptor digest under the moved renderer, so each reaches the refusal it
records rather than LLC0203.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX
`just relock-identity` builds the compiler, keeps each relocked lexlean.lock whose only
change is its compiler_semantics line (moving from the ID its language carries at HEAD),
rewrites LANGUAGE_1_2_SEMANTICS_ID, and repins the descriptor digests of the production
requests under tests/. regenerate.yml runs it in a first job, `identity`; every writer
starts from its patch and combine applies it first. SPEC.md §9.4 describes the job.

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
…k-identity

`cargo xtask semantics-records [--write]` computes rust_renderer, rust_runtime_core and
rust_runtime_std from the tree (the records the renderer and runtime unit tests check), so
a backend or runtime change no longer copies them from a test failure. relock-identity runs
it before building the compiler, since the records are part of the ID.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX

afflom commented Nov 14, 2026

Copy link
Copy Markdown
Member Author

Added cargo xtask semantics-records [--write]. It writes rust_renderer, rust_runtime_core and rust_runtime_std in language/semantics-1.2.toml from the tree. relock-identity now runs it first, so backend/runtime PRs (#441, #443) no longer copy those digests out of a test failure.


Generated by Claude Code

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Bounded source review at 297234b. Computing the three semantics records directly matches the existing renderer/runtime digest contracts, and applying/committing the identity patch as each writer's baseline gives combine a coherent identity-first composition. Two script corrections remain:

P1 — .github/scripts/relock-identity.sh:93–98: cleanup is not scoped to this invocation. It enumerates every ignored .lexlean/ directory and deletes it, then restores every unrelated tracked working-tree diff. No baseline or initial cleanliness check distinguishes existing build/evidence roots and source edits from outputs created here. The documented local just relock-identity entry point can therefore destroy preexisting work. Use an isolated clean tree, or establish and enforce ownership before any write and preserve all preexisting state; a clean tracked checkout alone does not establish ownership of ignored evidence roots.

P2 — script:80–89: each request gets only one diagnostic pass, while api/production.rs:1123–1127 returns at the first stale descriptor. A supported request with both stale rust-core and rust-std pins leaves the second pin stale after this recipe. This is a source-contract case; I did not find or execute a current tracked dual-target reproduction. Obtain all intended descriptor IDs through the official descriptor interface or bounded pin-only validation, preserving deliberate negative fixtures. Do not solve this by repeatedly running production builds.

Coordinate the canonical relock path with existing #521 (59a3378, review5898706789): it already refreshes all three writers' working locks with a single publisher for the three shared lock deltas. That collision repair stays credited. Compose or retire the superseded per-writer relock step explicitly instead of retaining both implementations. #468 still requires model-controlled routing, qualified evidence reuse and actual v0.5 integration; this identity job alone does not close it.

No writer, build, cleanup, identity wave or completed check was run for this review. Owner-reported prepared-tree runs remain scoped evidence; final-head full acceptance and output custody remain pending.

@afflom afflom mentioned this pull request Nov 15, 2026
…, repin every target

Review of #545: the script refuses to start on tracked changes and snapshots the
.lexlean/ build roots that exist, so cleanup removes only what the run created and
restores only its own side effects. Pins move from the descriptors `lexlean
production targets` reports before and after the records are written, every target
at once, instead of from the first LLC0203 a production build stops at; a pin that
named no descriptor of HEAD is a fixture's own and stays. No request is built.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX

afflom commented Nov 15, 2026

Copy link
Copy Markdown
Member Author

All three are addressed in c5e01e6:

  • P1: the script refuses a tree with tracked changes, snapshots the existing .lexlean/ roots, removes only the roots it created, and restores only its own side effects.
  • P2: pins move from lexlean production targets read before and after the records are written, for all targets at once. A pin of no HEAD descriptor stays as it is, and no request is built.
  • ci: regenerate.yml relocks before it runs the write recipes #521: its per-writer relock step is retired for v0.5 in favour of the single identity publisher. The PR body says what to drop on the forward merge.

Generated by Claude Code

afflom pushed a commit that referenced this pull request Nov 15, 2026
Update joins the primitives nat_bitwise_function maps to None (the match v0.5 added), the
merge marker left between refines theorems in RustStd is removed, and rust_runtime_std is
written by `cargo xtask semantics-records --write` (#545).

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/43649832487),
rewrote 5502 paths.

The counts writer reported:

    BS-05 SEEDED_REFUSED_VALIDATE: 273 -> 275 (crates/conformance/src/cases/bootstrap/rust_validate.rs)
    BS-05: §31 row: nothing
    BS-05: the §31 row does not state 273 (-> 275)
    BS-05: register: nothing
    BS-05: scenario: nothing

    VERIFICATION.md rows for the landing (counts | before | after):
    | `BS-05` SEEDED_REFUSED_VALIDATE | 273 | 275 |

    wrote /home/runner/work/lexlean/lexlean/CONFORMANCE.md and /home/runner/work/lexlean/lexlean/ERRORS.md

    written, with SPEC.md and the model projected again; next: the parts without LEXLEAN_COUNTS

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Repair follow-up at c5e01e6: the original destructive-cleanup P1 is source-fixed by refusing tracked changes before writes and removing only newly created root directories. Reading official production targets before/after the semantics change also removes the first-diagnostic limitation and avoids request builds. Existing directory contents are not blanket-deleted; note that an existing root without a .lock helper can still gain one, so the stated write set should account for that helper.

Remaining P2 — .github/scripts/relock-identity.sh:118–120: the pin update still requires the exact adjacent text "id":"target","digest":"old". ProductionRequest::from_json accepts whitespace and member reordering (Json::parse plus serde deserialization and validation; no canonical-byte comparison). A valid formatted or digest-first target therefore retains its stale pin silently. The canonical JSON object's sorted keys also place digest before id. Update the parsed roots[].targets[] entries by their id and old digest, preserving unrelated data and deliberately non-HEAD pins, or use equivalent formatting-independent edits. No runtime reproduction is claimed.

The #521 retirement instruction is now explicit and credited, while actual forward composition remains pending. Prior semantics writer/identity-patch credit is reused; no writer or test was repeated. This is source review, not final-head qualification or factory #468 delivery.

…own the .lock helper

Review of #545: a pin is found in roots[].targets[] of the parsed request by its id
and HEAD's digest, whatever the member order or whitespace, and the request is
written back in the layout it had (a layout the step cannot reproduce stops it
with the file named, never a silent miss). The write set is stated, and the
`.lexlean/.lock` helper a lock run adds to a root that existed before is removed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX

afflom commented Nov 15, 2026

Copy link
Copy Markdown
Member Author

P2 is fixed in a82ba01. Pins are now updated in the parsed roots[].targets[], matched by id plus HEAD's digest, so member order and whitespace no longer matter. The request is written back in its own layout. I tested compact files and a sorted, indented, digest-first two-target file. If a layout can't be reproduced, the step stops and names the file instead of missing the pin. The write set is stated in the script header and SPEC §9.4, and the .lexlean/.lock helper that a lock run adds to a pre-existing root is removed.


Generated by Claude Code

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Bounded a82ba01 follow-up: the pin-format P2 is resolved in source. Updates now select parsed roots[].targets[] by target id plus the old descriptor digest; deliberately different pins remain unchanged. Layout reproduction either preserves the supported layout or stops explicitly instead of silently missing a pin. Prior identity-publisher and cleanup repairs remain credited; no tests or producers were repeated.

One small helper-preservation P2 remains at .github/scripts/relock-identity.sh:48 and166. -e reports false for a preexisting dangling symlink, so that existing .lexlean/.lock is classified as absent and later removed unconditionally. Even if the lock command refuses that path, its failure is ignored at79 and cleanup can delete the original helper. Treat either -e or -L as preexisting (or use lstat), and remove only a helper established to be newly created. This is a source-path finding, not an executed reproduction.

Owner-reported compact/indented two-target checks remain scoped evidence. Final-head qualification, official output custody and actual #521 forward composition remain pending; #468 is not complete.

…ck failure stops the run

Review of #545: an existing `.lexlean/.lock` counts as present when it is a symlink
too, dangling or not, and only a regular helper file this run created is removed. A
lock run that fails on the build root's mutation lock stops the script before any
cleanup, instead of being ignored like a fixture's unresolvable lock.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX

afflom commented Nov 15, 2026

Copy link
Copy Markdown
Member Author

Fixed in 5fa94fd. An existing .lexlean/.lock now counts as preexisting when -e || -L, so a dangling symlink is covered, and cleanup removes only a regular helper file in a root that had none before the run. A lock run that fails on the build root's mutation lock (…/.lock: …) stops the script before any cleanup. Fixture locks that don't resolve are still put back as before.


Generated by Claude Code

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Bounded 5fa94fd follow-up: the helper-preservation P2 is resolved in source. The baseline now recognizes dangling symlinks with -L; cleanup excludes symlinks and removes only a regular helper recorded absent at entry. Recognized mutation-lock failures stop before cleanup. No new blocker found in this one-file delta.

Earlier parsed-pin/layout, preexisting-work preservation, semantics-record and identity-patch repairs remain credited without repeating their checks. Final-head CI is still incomplete; owner-reported tests, source review and a successful reproducibility run do not alone establish full acceptance. Actual #521 forward composition and #468 factory integration remain separate obligations. No build, writer, cleanup or retry was run for this review.

github-actions Bot and others added 6 commits November 15, 2026 03:12
The oracle lowering boxes a rust-std field whose type is a datatype with a
fielded constructor; the bootstrap's RustLower now marks those owners
(markShared) so its cyclic test boxes the same fields.

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/43671310314),
rewrote 1204 paths.

The counts writer reported:

    counts: no count moved
rust-std now holds a datatype field with a fielded constructor behind Rc,
so the host driver of RuntimeHost wraps the Start it gives Event::C0 in
Rc::new and clones the Spec out of Request::C10 for process_spawn. The
language-1.2 compiler-semantics ID moves to a15f5fbc; every identity-only
lock 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
The write recipes of SPEC.md §9.2, run by regenerate.yml (https://github.com/UOR-Foundation/lexlean/actions/runs/43701828582),
rewrote 5533 paths.

The counts writer reported:

    counts: no count moved
#441 lands first; with its head here the tree #545 merges is the tree this
head's gate tested.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3M8ULnzEQwLLngyBBX

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Bounded integration review at9777986d: exact parents5fa94fd + a516b5a. Relative to a516, the complete composition differs only in nine existing #545 paths: eight blobs equal reviewed5fa, and the SPEC intersection adds only the existing §9.2 regeneration paragraph. All remaining root subtrees equal a516, preserving #441 output and accepted #540 composition. No bespoke integration defect found; prior helper/pin/layout/source reviews are reused.

The original5fa gate43665087725 attempt1 is now complete:70 unique same-run/head jobs and813 steps all successful; exact repro43665087797 also passed. Preserve this completed evidence at its original binding. Neither a516 nor combined977 inherits full acceptance from it: current977 CI43717127478 is incomplete, repro43717127422 passed. Use the existing qualification; no repeated producer or new gate requested. Actual #521 forward composition and #468 factory integration remain separate obligations.

Independent receipt e67e75298e3b3fe02dbd0f66ac0c96d37598d8470cfda0606333b8088bb1e0a0; completed5fa gate receipt8bee8a315b775b96e33203e5697f79c092ff5d8b512811939cca438f3e3e50a6. COMMENT only, no merge clearance.

Joined from the atlas-part artifacts of gate run 43714046523 at a516b5a.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3MAztfvN671cYrybev
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3MAztfvN671cYrybev

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Bounded integration review at 78a92dc: parents are 9777986 and 4a09ae2. Every root entry remains identical to reviewed 977 except examples, which takes the exact 4a09 subtree. Thus the prior nine-path #545 composition, helper/pin/layout repairs and accepted #540 source remain preserved. No new authored change or integration defect found in this delta.

#441 review 5911477550 records the Atlas adoption and its limits: successful part-job/artifact metadata does not qualify the cancelled old run or establish archive-payload equivalence. Old #545 run 43717127478 is also cancelled. Retain those receipts separately; do not aggregate them. Completed 5fa gate evidence remains bound to 5fa and need not be repeated.

Current 78a CI 43728379723 is in progress; repro 43728379603 passed. Finish current-head full acceptance and actual merged-tree equality before delivery. Actual #521 forward composition and #468 factory integration remain separate obligations. Receipt 8fe267264a2cb0ea7afe5a53a5b98182fb2ad0de2bcba436befc91bdcdc34902. COMMENT only; no merge clearance.

@afflom
afflom merged commit 7da6e93 into release/v0.5 Nov 15, 2026
71 checks passed
@afflom
afflom deleted the ci/relock-identity branch November 15, 2026 22:12

afflom commented Nov 15, 2026

Copy link
Copy Markdown
Member Author

Merged; merged tree 7c80348d equals the gated head 78a92dc tree (all 71 checks green).


Generated by Claude Code

@afflom afflom left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

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

Independent final delivery audit confirms actual7da6e937, reviewed head78a92dc3, checked syntheticcf54e26d and latest synthetica28d3433 all have tree7c80348d7c3effc49669791fd71e19aa2168c9bd. Actual parents are5c07f9fa +78a92dc3, preserving the release history.

CI43728379723 attempt1 completed70 unique jobs and813 steps, all successful, including substantive Intel execution/inventory and native Atlas assertion. Head repro43728379603 and actual repro43767971691 each passed1job/7steps, including just repro. Actual pushCI43767971718 is1planner/5steps with17skipped jobs; it is not counted as another full gate. The older checked synthetic's tree exactly matches the final actual tree, so no repeated gate was needed.

#545's reviewed scope is accepted at actual7da6. #521's required actual forward history and #468's factory acceptance remain separate obligations; no broad issue closure or #316 admission follows from this delivery.

Immutable receipt pr545-7da6-actual-final-acceptance-20261115.json SHA2564f441e432a5f8a1f21c000e2e2215d7b436864720caefe74f0c60ffbf159b3d3. Prior source/fixture reviews reused, with one bounded checkout-binding log. No new test, producer, retry or control action.

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