Moc caching (moccache) landed on master 2026-07-23/24 (e4e3701, ed7fb53, 1fed23e, 9a5fea9). Comparing CI build-step medians before vs. after shows large wins on Android/Windows/iOS, but Linux and macOS are essentially flat despite healthy moccache hit rates (88–95%).
| Job |
Before |
After |
Δ |
| Android (windows) |
37.4m |
12.1m |
−68% |
| Android (linux-emulator) |
20.1m |
6.9m |
−66% |
| Android (linux) |
16.6m |
7.8m |
−53% |
| Windows msvc2022_64 |
17.8m |
4.6m |
−74% |
| Windows arm64_cross |
17.8m |
4.8m |
−73% |
| iOS |
8.9m |
4.9m |
−45% |
| macOS |
9.2m |
8.5m |
~flat |
| Linux gcc_64 |
7.8m |
7.6m |
~flat |
| Linux Test+Coverage |
8.0m |
7.9m |
~flat |
Notes:
- Medians of Build step durations from successful first-attempt runs (n=6 before, n=2 after — post-landing sample is small).
- Likely explanation to verify: Linux/macOS builds were already well-served by ccache, so moc caching adds little; whereas on Android/Windows/iOS deterministic moc output unlocks downstream ccache hits.
- Outlier to watch: Windows msvc2022_arm64 (native) went 11.8m → 21.2m (n=2, possibly cold cache/runner variance).
Follow-up: dig into Linux/macOS build logs to see where the time goes and whether moc/ccache interaction can be improved there.
Moc caching (moccache) landed on master 2026-07-23/24 (e4e3701, ed7fb53, 1fed23e, 9a5fea9). Comparing CI build-step medians before vs. after shows large wins on Android/Windows/iOS, but Linux and macOS are essentially flat despite healthy moccache hit rates (88–95%).
Notes:
Follow-up: dig into Linux/macOS build logs to see where the time goes and whether moc/ccache interaction can be improved there.