Skip to content

docs(0.10): record W0 baseline evidence and SQLite blocker - #144

Merged
bazer merged 1 commit into
v0.10from
codex/0.10-w0-baseline
Sep 17, 2026
Merged

bazer merged 1 commit into
v0.10from
codex/0.10-w0-baseline

Conversation

@bazer

@bazer bazer commented Sep 16, 2026

Copy link
Copy Markdown
Owner

W0 needs a reproducible before-state before shared async runtime changes. This PR records the completed capture from clean commit 7e36614b5f26f1dd199ff93bc60620f4b0714d5f, adds a source-linked I/O and ownership map, and updates the implementation checkpoints.

W0 remains open. The first quick run failed a SQLite parallel-transaction test. Although the full matrix and repeats passed, a standalone public-API probe independently reproduced two distinct, still-open Microsoft.Data.Sqlite 10.0.11 connections receiving the same native handle. A controlled internal probe supports a pool activation/reclamation race. The relationship to the original DataLinq failure and a safe correction remain unproven; neither failure is waived, and W1 has not started.

Captured evidence:

  • Full provider matrix: 5,855 passed, zero failed/skipped, all 17 expected entries, valid for evidence. The initial quick failure, passing quick repeat and eleven focused repeats are all retained.
  • Six heavy, strict .NET 10 benchmark lanes: 90/90 rows, exact target sets, complete telemetry and valid provenance. Timing warnings remain visible; these are baseline measurements, not a regression comparison.
  • Forced build, PackOnly, package inspection, API comparison against locked published 0.9.2, and packed consumers. Zero new API breaks; the two accepted inherited loadLock differences remain explicit.
  • Consumer builds/generated-source checks on .NET 8/9/10; execution on .NET 10. An additional forced .NET 10 model build produced 45 generated files.
  • Existing WebAssembly warnings and source-audit ownership gaps have explicit dispositions and owners.

The tracked index records a local 672,635,108-byte ZIP containing 1,340 files and its manifest. All 1,339 artifact hashes, the archived manifest, ZIP integrity and the full ZIP hash were independently verified. Raw packages/logs remain outside Git; this is not an external backup or a GitHub artifact upload.

Documentation verification: 108 local/pinned-source link targets checked and git diff --cached --check passed. These planning pages are excluded from DocFX, so no site-build result is claimed. No runtime, dependency, workflow or website changes; no packages published.

Next gate: a focused W0-F1 provider-lifetime correction/investigation, with any current-consumer fix landing on master and merging forward, followed by recapture of affected evidence under its new identity.

Record the clean pre-async baseline, all 5,855 full-matrix tests, 90 strict .NET 10 benchmark rows, package/API checks and generated-source snapshots. Pin the local evidence archive and manifest hashes and map current I/O ownership to the implementation waves.

Preserve the initial SQLite transaction failure and the separately reproduced driver pool ownership defect. Keep W0 and the W1 handoff open pending a verified correction; update the roadmap checkpoints without changing runtime behavior.
@bazer
bazer merged commit 6bf3732 into v0.10 Sep 17, 2026
12 checks passed
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.

1 participant