Repository navigation
Conversation
Owner, 2026-11-01. The board drops about 300 job-minutes (engine-cost 180, evidence run about 199 per release). - native-platforms.yml has no push trigger; the release gate reads its reports from the main CI run. - template-engine-cost runs on the nightly only; a plan states `nightly`. - auto-exposure cold boot runs when exposure code changes and on the nightly. - Scenarios of 500+ scripted ticks on shooter, collectathon, rain, adventure, platformer and kart run on the nightly. - The starter leg keeps its pristine check; its scenarios and parity pose run on the nightly. - benchmark keeps count-loc --check; the rest runs on the nightly. - The Linux ARM64 starter JS frame comparison runs on the nightly.
…re the compiled workspace with integration Owner, 2026-11-01: "cache whatever else that can be cached". Estimates from the cache audit, not measured: - native-rust-cache: about 12 job-minutes a board. Only a push saves. - integration legs take build-artifacts' workspace: about 8-12 job-minutes. - playwright-debs: about 3-5 job-minutes, and it avoids a slow apt mirror for those packages.
Run 41345897601 test-native: the spec still pinned the trigger PRD-776 removed.
Push board 41347117996: every integration leg failed at 'git rev-parse <candidate>^1'. The cold-boot narrowing from PRD-776 read a parent that a depth-1 checkout does not have, so the main push board could not have qualified. A missing parent now keeps cold boot selected for the writer only; the paths job and both joins fetch full history and still decide exactly.
…tion keeps the cause Run 41348117470: the Windows startup sample died on an error redaction withheld because it held a Windows path. Mailbox read and remove retried a lock; the rename did not. The write now retries for two seconds, the collector's screenshot request uses the same write, and a withheld failure keeps its error code and syscall so the next one names itself.
…eads promotion as the verdict The tag run passed --bundle-id dev.threenative.runtime while the packager stamps the consumer's app id, so the simulator never listed the app and clean-consumer-ios failed the same way on 0.3.6 and 0.3.7. That job does not gate promotion (owner, 2026-09-23), yet it failed the run, and release:ship stopped on the run's conclusion after finalize had already made the release public. Ship now waits on the public release, and still stops on a failed run that never promoted.
…ough an upgrade Release proof TN_RELEASE_UPGRADE_RED: every 0.3.6 scaffold pins maxFrameMsP95 33 with no reason, and the PRD-735 rule made that scenario invalid, so pnpm test failed on upgrade. The exact legacy pin now loads, and the report's budget reason says it passes at 30 fps. Any other loose number still needs a written reason, and templates still pin none.
…as unused 0.3.7's tag holds a public GitHub release with no npm cohort, and ship refuses to move such a tag.
… main pass goes negative Field 0 answers for the renderer that drew last and field 27 for the world renderer, so a loading cover over a shadowed world read 4 - 98 = -94 main draws (measured on the platformer kit). The negative summed into the frame main pass, and when it won one sample failed isRuntimeDiagnosticsSample and the whole series with it: TN_PLAYTEST_PERFORMANCE_SAMPLES_MISSING, intermittent because it needs shadow draws above the frame main draws. The overlay also re-reported the world shadow pass, so shadow counts doubled on those frames.
This branch has not been deployed
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
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.
What is this PR about?
It promotes develop to main with the owner's CI cuts and new caches, so 0.3.7 can ship.
Why does it matter?
Main CI on 57bc89e stays red because the rain engine-cost row hits its 45-minute limit on slow runners. This head moves that row to the nightly, and the release no longer needs a second native evidence run.
Done
template-engine-cost, starter scenarios, long scenarios on six kits, benchmark extras and the ARM64 frame comparison run on the nightly.auto-exposure cold bootruns when exposure code changes and on the nightly.native-platforms.ymlhas no push trigger. The release gate reads its reports from the main CI run.Checks
pnpm typecheck: exit 0.