Skip to content

Bootstrap count cases: report every moved count in one pass, and write them - #442

Merged
humuhumu33 merged 4 commits into
release/v0.4from
bs-count-writer
Nov 12, 2026
Merged

humuhumu33 merged 4 commits into
release/v0.4from
bs-count-writer

Conversation

@humuhumu33

Copy link
Copy Markdown
Contributor

Why

Each standard-library landing moves the counts that the bootstrap count cases assert: BS-05 to BS-08 and BS-11 to BS-15. The cases check those counts in stages, cheapest first. So a landing fails at the first stage, gets mirrored by hand into seven files, is run again, and fails at the next stage. SL-14 (#414) took four rounds like that.

What

  • crates/conformance/src/counts.rs. Each count check is now crate::counts!(id, what, { CONSTANT: observed }). With LEXLEAN_COUNTS unset, a moved count fails the case as before. With LEXLEAN_COUNTS=report, every count goes to target/count-report/<id>.jsonl and the case runs on. With fast, the case also stops before its evaluator and rustc runs. Constants that drive the computation, such as mutant draws and bounds, stay plain assertions.
  • cargo xtask counts [--write]. It reads the report and rewrites each moved constant. It also updates the case's §31 row, register statement and scenario, but only for numbers no other constant of the case holds, plus sums of exactly one pair of its counts. In §17.23 prose, it rewrites a number when the row's words follow it there too, or when every constant across the bootstrap cases that holds it moves alike. It prints the VERIFICATION.md rows for the landing. Anything it cannot place unambiguously is listed for a reader, never guessed.

Checked

  • release/v0.4: the nine parts under LEXLEAN_COUNTS=fast take about four minutes and record 58 counts. None moved, and the writer reports nothing to do.
  • SL-13: reading canonical JSON in the standard library (#276) #415 (SL-13): one fast pass and one write produced its whole count mirror. That was 23 constants, nine rows, statements and scenarios, and 18 prose numbers, including the derived 2,927 of BS-06. It left four numbers to a reader: 273 is also a seeded-mutant count of BS-05, and 117 is a timing in seconds. The nine parts are running whole on that tree now.
  • Gates: validate-model, validate-spec-links, cargo fmt --check, clippy with -D warnings on xtask and repo-conformance, the bdd meta-gate, check-test-inventory, and the writer's unit tests.

No register row is added: this is test and xtask tooling, and the default behavior of every case is unchanged. If you want it registered, say so and I'll add the row.

🤖 Generated with Claude Code

humuhumu33 and others added 3 commits November 8, 2026 23:57
…iter

The count cases BS-05 to BS-08 and BS-11 to BS-15 assert their counts in
stages, the cheapest first, so a landing that moves several of them
failed at the first stage, was mirrored by hand, run again, and failed
at the next: four rounds for SL-14, each a run of the nine parts and a
hand edit of seven files.

Each count check is now `crate::counts!(id, what, { CONSTANT: observed })`
(crates/conformance/src/counts.rs). Unset, `LEXLEAN_COUNTS` keeps the
assertion: a moved count fails the case with both tuples. Under
`report` every count is recorded in target/count-report/<id>.jsonl and
the case runs on; under `fast` it also stops before its evaluator and
rustc runs. Constants that drive the computation (the mutant draws of
BS-05, BS-07 and BS-08, the bounds) stay plain assertions, so they are
never reported as counts.

`cargo xtask counts [--write]` reads the report and rewrites each moved
constant whose literal holds the reported value; in the case's §31 row,
register statement and scenario, every number whose value no other
constant of the case holds, and each sum of one pair of its counts; and
in the prose of §17.23, a number the row's words follow there too, or
one every constant of every bootstrap case that holds it moves alike. It
prints the VERIFICATION.md rows of the landing and lists every number it
cannot place unambiguously, never guessing.

Checked: on release/v0.4 the nine parts under LEXLEAN_COUNTS=fast take
about four minutes and record 58 counts, none moved. On #415 (SL-13) one
fast pass and one write produced its whole count mirror: 23 constants,
the nine rows, statements and scenarios, 18 prose numbers including the
derived 2,927 of BS-06, with four numbers left to a reader (273 is also
a seeded-mutant count of BS-05; 117 is a timing).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
RP-05 holds the description of xtask/Cargo.toml to every task main.rs
dispatches; `counts` is one now.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The one conflict is the CHANGELOG, where both entries stay, the
writer's first as the later landing.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
humuhumu33 added a commit that referenced this pull request Nov 9, 2026
…d BS-13 to BS-15

The committed export gains one package, so the committed manifests, packages,
programs and crates each grow by one and the corpora the cases draw from them
follow: 18 constants in seven cases, their §31 rows, register statements and
scenarios, and the §17.23 prose that repeats them. Measured under
LEXLEAN_COUNTS=fast with the count writer of #442 applied and then removed
again; the writer placed every number. BS-11, BS-12 and every mutant draw are
unchanged. VERIFICATION.md carries the landing's table.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
@afflom afflom mentioned this pull request Nov 9, 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.

Source review of exact head0eff75c3: the default assertions remain intact, and explicit fast/report modes are documented diagnostics. This developer tooling does not need a new shipped-capability row merely for adding a count writer. Two writer decisions still permit ambiguous edits:

  1. xtask/src/counts.rs:478: number_map excludes unchanged candidate pairs before deciding that a sum has one destination. For example, A10→11, B20→20, C12→12 and D18→18 give two explanations for a row's30: A+B moves to31, while C+D stays30. The code considers only31 and rewrites automatically. Include unchanged explanations, then flag conflicting destinations instead of guessing.
  2. xtask/src/counts.rs:728: prose matching stops at the first old-number/context anchor. If two cases share the same numeral and following context but map it differently, case order selects the edit. Collect all matching destinations and require agreement; a known conflict must not fall through to the shared-number fallback.

These are source-traced synthetic counterexamples, not executed test results or claims that today's specific landing was miswritten. Focused planner tests for both ambiguities would demonstrate the repairs; the current helper tests do not exercise these decisions. No test, writer, regeneration or full gate was repeated for this review. Existing ownership remains with this PR.

afflom commented Nov 9, 2026

Copy link
Copy Markdown
Member

Coordination note for the BS-07 assertion hunk: draft #448 (release/v0.4; issue #447) moves unused evaluator/rustc mutant preparation below BS-07's existing normative-host return. It keeps the base 22/350/292 counts and four native differential roots before the return; mutant seeds/counts/consumers are unchanged. This does not alter this PR's count-writer design. When reconciling the overlap, keep the base-count reporting before the host return and mutant construction/count assertions after it. #448 is source-reviewed only; Rust and final-host qualification are still pending.

afflom commented Nov 10, 2026

Copy link
Copy Markdown
Member

Coordination for BS-15: #457 owns a narrow test-only optimization on release/v0.4. It will assemble the same ordered evaluator/native render vectors directly instead of constructing full-library Runs that neither consumer uses. The 16,384-byte evaluator membership, existing 4,096-byte native prefix, seeds, corpus/count assertions, recipe assertions and all plants stay unchanged. Please preserve your BS-15 count reporting and fast return when reconciling the nearby construction hunk. No count-writer redesign or work on #272 is included; no speedup is claimed before final hosted evidence.

afflom commented Nov 10, 2026

Copy link
Copy Markdown
Member

Upcoming-release integration update: #450 has landed on release/v0.4, including #448, and the accepted-source checkpoint for #405 is now published at 2fd4265. Its generated products and combined-corpus counts remain unqualified; this is where your all-nine-case reporter/writer will be useful once accepted. We are not duplicating that tooling.

The two ambiguity findings in review #442 (review) remain unresolved at 0eff75c. Please address them with focused planner tests, preserving default assertion behavior, so this reporter can be admitted without guessing numeric substitutions. Keep the earlier BS-07 coordination and the #459 BS-15 materialization optimization when reconciling overlaps. #459 is now in full qualification for v0.4.1; it is not merged yet. #405's official model-projection preparation can proceed independently while the count writer is corrected.

afflom added a commit that referenced this pull request Nov 10, 2026
Add independently reviewed model-projection-only driver and scoped read-only hosted workflow. Use official pinned xtask, exact admitted source, protected input custody, two-pass model fixed point, and same-run patches/logs/hashes with original failure exits. Actions never commits or pushes.

Twelve Python behavioral plants passed; these use fake subprocess data and are not Rust qualification. No renderer/corpus observation, generated-product acceptance, full gate, release merge or #442 implementation is included. Historical counts remain unqualified. Existing #405 and all release/feature branches are unchanged.
afflom added a commit that referenced this pull request Nov 10, 2026
Add only six independently reviewed temporary infrastructure files, patch SHA256 f855f08ebf50cf2fcfe5e5dfa9aa4c8e90ac520fd2d51912ad9fa124f4720547. Fifteen Python custody tests pass, including failure-path finalization plants.

Observe renderer through owning APIs, narrowly reconcile the recorded scalar, rebuild and regenerate 28 admitted ordinary locks, run the official calculus writer once, then rebuild, observe, relock and check without writing. Preserve original failures and stop before production-export, golden, fixtures and examples. Workflow has read-only permissions and never commits or pushes.

Qualified: false. Native compilation/execution, product review, full fixed-point closure, real accepted-parent ancestry and final qualification remain outstanding. No #442 or #316 source imported; all fixture inputs stay frozen.

afflom commented Nov 10, 2026

Copy link
Copy Markdown
Member

The two planner findings now have a narrow draft stacked repair: #462 (issue #461), based on this PR's exact 0eff75c head. It includes unchanged pair explanations before ambiguity detection and requires all matching prose anchors to agree, blocking shared fallback on conflict. Four tests call the pair/prose planners for collisions, agreement, unchanged sums, order independence and fallback behavior.

Only xtask/src/counts.rs changes. Your bs-count-writer branch and reporter/fast-mode design remain unchanged; this is available for review and integration into your existing work. Source review and independent reference checks passed, but Rust/rustfmt/Clippy were unavailable locally and hosted native checks are pending. #462 remains draft; no full release gate or merge is being requested yet.

afflom commented Nov 10, 2026

Copy link
Copy Markdown
Member

The coordinated #462 planner repair at 810cf9df has now completed its ready-event stacked run 42763429738: plan, honesty and rest passed; seven release groups were skipped because the target remains this feature branch. This is completed development evidence, not full release acceptance.

Exact source review5836936933 and native-evidence comment6760032401 retain the four real-planner regressions, all eight count tests, pinned formatting and Clippy results. No tests or logs were repeated for this terminal check. Your reporter/fast-mode design and integration ownership remain intact; #461/#442 stay open pending reviewed integration and required final qualification. Preserve the reconciled #459 optimization and the actual #395 corpus additions when composing the eventual release candidate.

humuhumu33 added a commit that referenced this pull request Nov 11, 2026
… tooling packages

Written by the count writer of #442 from LEXLEAN_COUNTS=fast runs of the nine
parts on the regenerated tree, then validate-model --write.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
maurathat pushed a commit to maurathat/LexLean that referenced this pull request Nov 11, 2026
…S-13 to BS-15

The committed export gains six packages (those of BS-21 to BS-24 on this
tree), so the committed manifests, packages, programs and crates each grow by
six and the corpora the cases draw from them follow: 20 constants in seven
cases, their §31 rows, register statements and scenarios, and the §17.23
prose that repeats them. Measured under LEXLEAN_COUNTS=fast with the count
writer of UOR-Foundation#442 applied and removed again; the writer placed every number.
BS-11, BS-12 and every mutant draw are unchanged. VERIFICATION.md carries
the landing's table.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts:
#	CHANGELOG.md
#	crates/conformance/src/cases/bootstrap/rust_lower.rs

@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 integration readback at18def: xtask/src/counts.rs is byte-identical to0eff (bloba194f608), so both findings in review5828996717 remain: unchanged summand explanations are discarded, and prose planning stops at the first matching anchor without checking later destinations.

The existing #462 implementation addresses precisely these two decisions and already has scoped native planner/format/Clippy evidence. The smallest next step is owner source/history reconciliation of that repair with current18def, preserving its valid evidence; no competing implementation or automatic repeat of those checks is requested. #462 currently reports its old0eff base and is not mergeable, so this is not an instruction to merge it as-is or treat its old feature-base gate as current release acceptance.

This review reused the exact source blob and prior repair receipt. It does not claim a full audit of the70 inherited commits or all generated products, and no build, writer, test or CI action was run.

afflom commented Nov 11, 2026

Copy link
Copy Markdown
Member

Follow-up to review5866295792: refreshed GitHub metadata now provides a clean current #462 integration, so no conflict resolution or rebase work is demonstrated. Synthetic e050a49 has ordered parents [current #44218defa1b, #462810cf9df]. Against18def it changes only xtask/src/counts.rs, and that blob d8ee566c is exactly the already reviewed #462 repair. The earlier non-mergeable observation was transient and is not a current blocker.

The owner can review this existing one-file integration and decide adoption/qualification without reimplementing the two fixes. It is not yet adopted into #442; the old native planner/format/Clippy evidence retains its recorded scope and does not establish full current release acceptance. No branch, merge, test, writer or CI action was performed here.

@humuhumu33
humuhumu33 merged commit 84647cc into release/v0.4 Nov 12, 2026
16 checks passed
afflom added a commit that referenced this pull request Nov 13, 2026
…guities

Fix ambiguous count rewrites (stacked on #442)
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