Repository navigation
Conversation
|
Python 3.10/3.13 and Windows gates are green. Both real-Lean jobs failed only in the unchanged #65 detached-child PID-file fixture before reaching the generated-project build. #67 now fixes that test and its full Python/Windows/real-Lean CI is green, but it remains blocked on contributor CLA. Keeping this stack awaiting that dependency rather than weakening or retrying the flaky gate. |
…okresearch/autoform-bot into codex/pr26-clean-20261003
|
Restacked additively onto final #85 head |
|
Superseded by #96, which rebuilds project creation on main without the provenance dependency and loosens the single-release lock. |
Depends on #85 and is based directly on its upstream branch.
Summary
autoform project newfor an absent target and an explicit bundledLean/Mathlib release
nine-package Lake lock, and optional verified workflows without running
project code during planning
exact bytes/types/modes/ownership/links through retained descriptors, and
publish with a native atomic no-replace rename
commit-uncertain state accurately
Review hardening
creation descriptor that cross-checks public release identity, complete Lake
manifest, and all reserved dependency roots (including
Archive,Counterexamples, andMathlibTest)overlong or surrogate paths before staging
publication and again after parent fsync; preserve the stage on prepublish
drift and report exact published/synced state on postpublish drift
than mocking the primitive
PYTHONIOENCODING=asciiThis is a clean reconstruction of the project-creation feature delta on #85,
without the stale merge history or removed #14 private APIs from old #26.
Validation
108 full create; 35 final root/skill tests
git diff --check, andmake check-example: passeddependency, and began compilation; local cache extraction then hit shared
ENOSPC, so fresh CI owns the complete-build proof
lake buildunder explicit step bounds with a 75-minute job ceiling
Independent review found no remaining correctness blocker.