Skip to content

ci(release): cut hosted macOS minutes and keep the iOS build cache warm - #30

Merged
getBoolean merged 4 commits into
mainfrom
t3/disable-unneeded-actions
Nov 17, 2026
Merged

getBoolean merged 4 commits into
mainfrom
t3/disable-unneeded-actions

Conversation

@getBoolean

@getBoolean getBoolean commented Nov 17, 2026 •

Copy link
Copy Markdown
Owner

Release runs were the fork's main hosted macOS cost: the iOS IPA (18–98 min), macOS x64 (6–13 min), and macOS arm64 (~6 min) builds ran on every release. Nightlies could publish every six hours. Because these builds ran beside the release tests, failing runs still used 38–117 macOS minutes.

  • T3CODE_NIGHTLY_GAP_HOURS sets the minimum gap between automatic nightlies. Unset keeps upstream's 6 hours; the fork now sets 24.
  • The macOS desktop builds and the unsigned IPA now wait for the release quality and test jobs, on every trigger (nightly, workflow dispatch, tag push). A failing release spends no macOS time; each release gains roughly 7 minutes of wall time.
  • The unsigned IPA's Xcode compilation cache is keyed by dependency inputs instead of the commit, and restores any earlier entry for the same Xcode build. The per-commit key saved a new 1.6 GB entry every release. That filled the repository's 10 GB cache and evicted the entry, and a lockfile change had no fallback, so builds went cold: about 80 minutes against 13 warm (CompileC 11,742 s cold vs 94 s warm in the build timing summary). A restored cache over 3 GB is dropped so it can't crowd out the other caches.
  • Added a test for the configurable gap, updated docs/operations/fork-releases.md, and updated the fork manifest.

After merge: re-enable Release, Release desktop build, and Release iOS (unsigned) (Publish AUR package stays disabled), and set T3CODE_RELEASES=true and PAUSE_NIGHTLY=false.

Claude Opus 5.5 via Claude Code in T3 Code

Nightlies could publish every six hours, and the macOS desktop and iOS
builds ran beside the release tests, so failing runs still spent 40-117
macOS minutes. T3CODE_NIGHTLY_GAP_HOURS now sets the nightly gap (fork
uses 24), and macOS/iOS builds wait for the release tests.

Claude Opus 5.5 via Claude Code in T3 Code
…file changes

The cache key included the commit SHA, so every release saved a new
1.6 GB entry. That filled the 10 GB repository cache and evicted it, and
a lockfile change had no fallback, so builds went cold (~80 min against
~13 warm). Key by dependency inputs, restore any earlier entry for the
same Xcode, and drop a restored cache over 3 GB.

Claude Opus 5.5 via Claude Code in T3 Code
@getBoolean getBoolean changed the title ci(release): cut hosted macOS minutes so fork releases stay affordable ci(release): cut hosted macOS minutes and keep the iOS build cache warm Nov 17, 2026
@getBoolean
getBoolean merged commit 0a72a34 into main Nov 17, 2026
19 checks passed
@getBoolean
getBoolean deleted the t3/disable-unneeded-actions branch November 17, 2026 22:25
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