Repository navigation
test(cache): isolate measurement fixtures from host disk capacity #1753
Description
The tiny fixtures added in #1744 consult real host free space, but Snapshot correctly reserves the full bounded production handoff (about 5 GiB) even for tiny fixture contents. On the shared host with 4,488,278,016 free bytes, the default reserve of 5,368,709,176 bytes rejects fixture setup; the full script suite then has two failures and seven errors in measurement tests. This was found while validating #1730; the measurement files there are unchanged from main.
Make only the test fixture deterministic: default Fixture.setUp should mock ample disk capacity with cleanup, while the explicit insufficient-space test retains its own smaller override. Do not change production resource limits, delete caches or activate the study. A source-only probe with this fixture adjustment passes all 24 measurement tests on the constrained host.
Completion: actual constrained-host measurement tests and the full Python suite pass; all six default checks pass on the frozen revision; independent test review confirms that production code and the explicit budget-rejection case are unchanged. No library declaration, selected abstract or rendering impact. Keep this test-only repair separate from the six-file #1730 retry implementation; #1197 remains open for actual study and net-cost evidence.
Worker /root/mathlib_cache_retry_1730 will claim this bounded follow-up under coordinator /root. AI assistance: Codex diagnosed the fixture precondition and scoped the repair.
Activity
SamuelSchlesinger commentedon Oct 30, 2026
Worker
/root/mathlib_cache_retry_1730(Codex, session 2026-10-30): active on this test-only follow-up. Branchwork/measurement-fixture-disk-codex-20261030from maindf2279f9; source-only checkout. I will mock default fixture capacity with cleanup, retain the explicit low-space override and leave production resource checks unchanged. Coordinator/rootowns serialized full validation and independent review/integration.SamuelSchlesinger commentedon Oct 30, 2026
Worker
/root/mathlib_cache_retry_1730: frozene7ec6966passes all 24 measurement tests and the full 435-test Python suite (17 optional skips) on the constrained host. The diff is one test file; production driver/protocol and the explicit low-budget case are unchanged. Awaiting the coordinator's serialized six-gate validation before opening the focused PR for independent review.SamuelSchlesinger commentedon Oct 30, 2026
Worker
/root/mathlib_cache_retry_1730: awaiting review in #1755 ate7ec69666f24faa80f0cae062951e9fd340221d6. The PR owns the validation record: 24 measurement tests, 435 script tests and all six default checks pass; production code and limits are unchanged. Coordinator/rootowns the separate-context review record and integration after hosted CI.test(cache): isolate measurement fixtures from host free space (cslib…
feat(polytime): certify fixed-list flatMap and filtered templates (cs…