Repository navigation
Slow macOS Intel build times in CI #90
Description
The macos-15-intel matrix job in .github/workflows/build.yml takes significantly longer than all other platform builds, taking 16m44s on cold/updated runs and serving as the primary bottleneck for CI completion.
Comparison across recent runs
| Job | Cold run (PR #89) | Warm cache (dev) |
|---|---|---|
macOS ARM64 (macos-latest) |
3m46s | 2m56s |
Windows (windows-latest) |
8m46s | 4m02s |
| Performance benchmark | 7m35s | 3m57s |
Linux (ubuntu-22.04) |
8m46s | ~8m (packaging + smoke tests) |
macOS Intel (macos-15-intel) |
16m44s | ~9–16m |
The slowness stems from macos-15-intel running on virtualized x86 hardware with lower IPC and fewer vCPUs than Apple Silicon runners, combined with release-mode Rust compilation and DMG disk image creation (hdiutil).
Potential solutions
- Cross-compile the
x86_64-apple-darwintarget directly on the faster Apple Silicon (macos-latest) runner withrustup target add x86_64-apple-darwin. - Package a universal binary on
macos-latestusingtauri build --target universal-apple-darwin. - Cache or streamline DMG generation for ordinary non-release CI runs if feasible.
Activity
DoodlesEpic commentedon Oct 7, 2026
Historical Context
In commit 216b224, both
aarch64-apple-darwinandx86_64-apple-darwinwere built onmacos-latest(Apple Silicon) via Rust cross-compilation (targets: aarch64-apple-darwin,x86_64-apple-darwin).Commit f2813d1 separated the targets, moving
x86_64-apple-darwintomacos-15-intel. As noted indocs/releases/v2.0.0.md, the intent was to ensure each architecture built on its respective runner and produced dedicated download assets.Universal Binary Trade-offs
Switching to
universal-apple-darwinbuilt onmacos-latestwould resolve the Intel runner bottleneck, with the following considerations:.dmgwithout needing to pick between Intel and Apple Silicon.