Repository navigation
Testing: restore a green full macOS hosted regression baseline #299
Description
Review context
Found during the requested GPT-6 Astra medium sweep of the published branch at 51a91a5e, reconciled against open issues and the last 200 commits. This issue tracks a specific follow-up; it does not reopen unrelated release or playback scope.
H1 — Re-establish a green full macOS hosted regression baseline
- Priority/type: P2 test-debt / release validation; confidence high that the tracking/acceptance gap exists, unknown root cause and user impact.
- Evidence: [Bug]: Direct Play and Maximum Transcode can converge on an unplayable HEVC copy lane on macOS #288 explicitly records a reproducible DownloadStore lookup assertion and two transport-presentation timing assertions in a 605-test suite.
b8fd0bd8b4carecords later full-suite failures indocs/plans/2026-09-06-client-first-playback.md:97-104: download keepalive cancellation and cross-kind poster retirement still fail isolated reruns; transport presentation passes isolated. Later journal entries at lines 175, 208, 230 preserve these as open release blockers.51a91a5e746creports focused hosted passes, not a new full-suite green result. - Precise test anchors:
LabstreamTests/DownloadStoreLookupTests.swift:97(startupRetirementPreservesCrossKindMainMediaAlias) models a cached poster name aliasing another record's main media.LabstreamTests/DownloadKeepaliveCoordinatorTests.swift:9onward covers cancellation and backend registries; itswaitUntiluses boundedTask.yield()iterations, so distinguishing scheduling assumptions from product cancellation behavior matters. Do not assume this is a media-deletion bug merely because a test fails. - Why not already tracked: [Bug]: Direct Play and Maximum Transcode can converge on an unplayable HEVC copy lane on macOS #288 owns Plex playback and Emby: investigate native MP4 starvation and fragmented-MP4 recovery stalls #291 owns Emby delivery. Both acknowledge unrelated failures without acceptance ownership. App Store shippability: release-readiness prereqs, pre-ship audits, CI, and submission mechanics #92 has a broad release correctness checkbox; macOS: add an isolated AX-driven agent validation harness #262 targets the AX harness, not hosted test failures. No supplied open issue names this repeatable baseline failure cluster.
- Concrete acceptance: Capture one bounded clean full-suite result at the exact candidate SHA and isolated repeats for named failures. Classify each as product defect, stale expectation, or scheduling flake. Fix root causes without globally skipping tests; use deterministic synchronization where required. Verify stale poster references clear while shared valid media remains readable; exact-attempt cancellation must not cancel replacements or unrelated backends. Repeat full suite and affected host lanes, preserving clean result bundles and reporting residual failures explicitly.
- Routing: Prefer one focused regression-baseline issue linked from App Store shippability: release-readiness prereqs, pre-ship audits, CI, and submission mechanics #92, with separately scoped fixes only after reproduction. Coordinate with download sweep rather than duplicate any root-cause finding it supplies.
Activity
jlipworth commentedon Sep 6, 2026
Fresh independent baseline at source 518e286: clean isolated macOS hosted build/run completed with xcresult summary 687 tests total, 686 passed, 1 failed, 0 skipped. The sole failure was DownloadStoreLookupTests/startupRetirementPreservesCrossKindMainMediaAlias(): it expected posterRelativePath to be nil. Previously noted keepalive and transport failures did not recur in this run. Independent source challenge suggests the poster fixture may predate the intentional legacy side-asset ownership adoption in 997cbca; classify stale expectation versus product defect before changing behavior. No runtime media-deletion defect is inferred from this assertion. The result bundle is retained locally; no raw private logs are published.
Client-first playback and bounded agent troubleshooting (#290) (#300)
jlipworth commentedon Sep 7, 2026
Merged and published in #301, reachable on origin/main at
6a5fa9be.Restored the full hosted baseline: explicit stale side-asset ownership replaces an obsolete legacy-adoption expectation; coalesced transport enqueue order is synchronized; the 140-durable-write retention test has a bounded 30-second drain allowance. Final clean isolated run passed all 693 tests with no skips. Earlier keepalive/transport failures did not recur.
Validation: 693 macOS hosted tests, 1,695 PMSKit hermetic tests, 320 tooling tests, four native builds, and passive iPhone/tvOS/visionOS fixture smoke checks passed. All five PR CI checks passed. Current testing guidance and the active playback plan were aligned; live/physical/release gates remain separate.