fix(tests): the validation ledger writer refuses a ledger its own verify rejects - #383
Merged
Merged
Conversation
…ify rejects `tests/test_ownir_validation_fixtures.py --write` exited 0 after writing a ledger that the very next verify run rejected. The trigger is any change of `OWNIR_VERSION`: 207 controls spelled the version as the literal `0`, so once the constant moves every one of them is refused at the version gate before the rule it exists to exercise can run. 58 acceptance controls flip to rejections and the writer commits the result (accepted: 2 of 294). - the controls say `OWNIR_VERSION`, not a literal; at version 0 the committed ledger is byte-identical, so no fixture moves - a new control rebuilds the ledger with the version shifted on both sides (the controls and the strict door) and requires every verdict and category to hold - the invariants live in one function shared by verify and `--write`, and the writer refuses to commit a ledger that fails them Negative control: with the new controls and without the fixes, verify fails with `FAIL[version-agnostic]: 58 control(s) change their verdict`. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
9 of 10 tasks
PhysShell
added a commit
that referenced
this pull request
Oct 1, 2026
…sion `tests/check_fix_candidates_facts.py` asserted `ownir_version == 0` under the message "ownir_version must stay 0". What it meant is that `--fix-candidates` is additive metadata and does not move the vocabulary version; what it checked was a literal, which held only until the vocabulary first moved. OwnIR v1 moved it, and the `C# leak extractor (Roslyn) -> OwnIR -> core` job went red on that line — with the extractor doing exactly the right thing. It now checks the claim: the flag-on facts are stamped with the core's current OWNIR_VERSION, and with the same value as the flag-off facts. This is the third control of this shape the bump has found (the validation ledger in #383, the instrument's generated facts in this branch), and the only one that needed the server to find it: the script takes the extractor's output as its argument, so it is not part of `tests/run_tests.py` and no local run of that suite could reach it. The job's steps were then run locally as written, from this one to the end of the job, on a Linux clone of the committed tree. tests/goldens/README.md records the golden's second amendment: one line, `"ownir_version": 0` -> `1`, made in the vocabulary commit. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
PhysShell
pushed a commit
that referenced
this pull request
Oct 1, 2026
…rch promotion onto main, strict-door binding included) OWN053 "orphaned awaitable": a local initialised by an un-awaited Task / ValueTask invocation of an effectful operation (the unwrapped result is a real disposable, or the member is a connection / transaction lifecycle call) that is never referenced again in its member. Default-on advisory in both engines with an identical message, rendered as a warning, never the exit code, hidden at --verbosity quiet; no automatic fix. The scope is frozen as measured in ownership-semantics-lab H-29: discards, expression statements, stored / passed / returned awaitables and lambdas are untouched; new lifecycle members enter only by witnesses. Provenance: a re-port onto main 5323dd4 of research/ownership-semantics-lab-v1 commits ab2964f (promotion) and 298b305 (P-OWN053-DOOR), carrying ONLY the OWN053 hunks. Unlike a file-level copy of those two commits, this port does not bring the research branch's content along: the extractor gains one detector (CollectOrphanedAwaitables), one holder and the additive facts branch instead of 1,900 lines of env-gated research seams; ownlang/ownir.py gains the family set, the strict-door block, the check_facts call and the finding builder instead of the H-20 / resource-effects E3 / P-037-X seams; spec/OwnIR.md and the schema gain §9, the §4.2 note, the orphanedAwaitable / orphanFamily definitions and nothing of params[].ordinal or flag_var; tests/test_ownir_validation_fixtures.py keeps main's #383 writer contract and gains the orphaned_awaitables section on top of it; tests/test_p037_evidence.py is untouched because main has none of the research read sites; the Rust diagnostics catalogue and the diagnostics ledger (47 -> 48) are included, which the file-level copy had lost. Extractor: the sites are collected into the ADDITIVE top-level facts list orphaned_awaitables (absent when there is no site, so such a document is byte-identical to the pre-OWN053 shape; ownir_version unchanged). On the committed fixture corpus/ownership-lab/h29/fx/Orphan.cs the production build finds exactly the five frozen primary sites of the H-29 scan and none of the eleven twins; the facts equal the research build's modulo the research-only params[].ordinal field. Strict doors: both load() and the Rust strict door validate the list last in BR-D1 order (array of objects; non-empty local / callee as identity, file string, line in the §4.2 domain, column as every other column, method / result_type string-or-null, family from the closed set as vocabulary). cp1 ledger 294 -> 344 controls, appended insertion-stable on top of #383's writer; the Rust validation replay reports no permissive accept and no category mismatch. Schema bound to sourceLine / sourceColumn / orphanFamily (pinned to ownlang/ownir.py::_ORPHAN_FAMILIES); binding map BOUND; coordinate census door slots; checkpoint documents regenerated. Verification on this tree: the fixture through both CLIs (5 advisory OWN053, 0 findings, exit 0, Python == Rust); three malformed documents refused by both CLIs with exit 2 (corpus/ownership-lab/h29/door); CLI fixtures, verdict goldens (one new synthetic case), diagnostics ledger and repro digests regenerated; full Python suite and cargo test --release green; the extractor builds with main's pre-existing warning only. Not included (registered in Own.NET-paperwork): P-OWN053-WORDING (family-specific message tails) stays open and precedes any family A/B expansion; discards / expression statements / family-B expansion / autofix / H-26A are not earned. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Am9eQwzNfbugH72eVKetC2
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Что и зачем
python tests/test_ownir_validation_fixtures.py --writeзавершался с кодом 0 и записывал реестр, который следующая же проверка отвергает. Причина: в 207 контролях версия была зашита литералом0, и при любом измененииOWNIR_VERSIONвсе они отклоняются на воротах версии раньше, чем срабатывает проверяемое правило (58 принимаемых контролей становятся отказами,accepted: 2из 294). Теперь контроли берутOWNIR_VERSION, а писатель отказывается фиксировать реестр, не проходящий собственные инварианты.При версии 0 закоммиченный реестр побайтово прежний: фикстуры не меняются, словарь OwnIR не затронут.
Тип изменения
Как проверено
python tests/run_tests.pyruff check .иmypypython scripts/<...>.py --selftest)Linux (WSL),
ruff 0.15.8,mypy 1.19.1.FAIL[version-agnostic]: 58 control(s) change their verdict when OWNIR_VERSION moves.main: поднятьOWNIR_VERSIONдо 1,--write→ код 0, затем проверка →FAIL[ledger-category].cargo test -p own-ir(Rust-replay реестра) — зелёный: реестр не изменился.python tests/run_tests.pyгонялся до коммита на дереве с обоими исправлениями (этим и соседним, про repro-артефакты); единственное падение там — контроль P-037 «dirty tree», который на закоммиченном дереве проходит.Связанные issue
Нет. Найдено при пробном bump
OWNIR_VERSIONв исследовательской ветке; сам bump сюда не входит.Чеклист
feat:,fix:,docs:…)Документация не требуется: поведение
--writeпри годном реестре прежнее.🤖 Generated with Claude Code