Repository navigation
Conversation
webkitgtk takes 3h05 to 5h00 of the 6h GitHub-hosted job limit (runs 35815959556..35966980787), and a job killed at the limit fails the whole atomic run. Run 36073571248 had it at 4h20 and still compiling. Reuse the sccache setup mozjs140 already has: set it up for webkitgtk too and pass it to both cmake configurations as the C/C++ compiler launcher. The Actions-cache backend stores each object as it compiles, so an attempt cut off at the limit still leaves the next attempt mostly cache hits. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QMTLsnqW2jtz5QsNC3W3qv
|
Holding as planned. If this is needed: it widens the |
castrojo
left a comment
There was a problem hiding this comment.
Standards: Conformant. Appropriately adds webkitgtk to sccache setup step in .github/workflows/build-stage.yml and configures optional CMAKE_C_COMPILER_LAUNCHER/CMAKE_CXX_COMPILER_LAUNCHER in packages/webkitgtk/webkitgtk.spec with clean xtrace preservation and fallback when sccache is absent.
Spec: Meets objective of enabling sccache for webkitgtk builds to prevent timeouts on long multi-hour builds. Factory configuration and unit tests pass.
Reviewer status: Needs a second independent review (+2 requirement) before merging.
|
Verification:
Outstanding approvals: 0 approvals currently recorded. Requires +2 independent approving reviews before merging. |
castrojo
left a comment
There was a problem hiding this comment.
Reviewed at cd47064. Following PR #264, the hermetic mock lane is now default for all package builds. Hermetic mock builds are network-isolated via unshare --net (tools/hermetic_build.sh:153), making GHA sccache unreachable and causing sccache to miss on every compile (docs/architecture.md:346). Furthermore, mounting sccache.env into /work exposes ACTIONS_RUNTIME_TOKEN to untrusted build scripts (per docs/skills/repeated-mistakes.md rule on sccache token scoping). Re-enabling sccache effectively would require moving webkitgtk to the container lane (build_lane: container in config/upstream-sources.json), which requires maintainer architectural consensus.
|
Closing this held fallback. Its condition for merging was webkitgtk hitting the 6h job limit in run 36073571248; that run built webkitgtk successfully in 4h55m (rebuild5 / build (webkitgtk), 01:42 -> 06:37 UTC on 09-25), under the limit. Since then #264 made the hermetic mock lane the default, and that lane is network-isolated (tools/hermetic_build.sh, unshare --net), so the Actions-cache sccache backend would miss on every compile (docs/architecture.md, Compiler cache row). The change would add the ACTIONS_RUNTIME_TOKEN exposure without the speedup. If webkitgtk approaches the limit again, the fix belongs in the hermetic lane (mock's ccache plugin + actions/cache, per the agreed direction). |
Fallback, held as a draft. I'll merge it only if webkitgtk hits the 6h job limit in run 36073571248. Merging it while that run can still publish would force yet another cold webkitgtk rebuild next run, because touching a recipe schedules its rebuild.
webkitgtk takes between 3h05 and 5h00 of the 6h GitHub-hosted job limit (runs 35815959556, 35824179148, 35829110579, 35966980787). It builds two full flavors, gtk4 and gtk3. There is no
%check, and%limit_build -m 3072already caps parallelism, so the time is compile time, not tests or swap. A job killed at the limit fails the whole atomic run.This reuses the pinned sccache client that mozjs140 already uses:
setup-sccachenow also runs forwebkitgtk.CMAKE_{C,CXX}_COMPILER_LAUNCHER=/work/tools/sccachewhen the client is mounted. Outside CI the recipe builds unchanged.Stats are printed after the build.
🤖 Generated with Claude Code
https://claude.ai/code/session_01QMTLsnqW2jtz5QsNC3W3qv