Repository navigation
Mac push: macOS reads process groups without cgo (from #2039) #2055
Description
This draft was closed, unmerged, when the 2026-10-29 Mac stability and performance push was stopped early to save usage. Its branch is kept: reopen the PR or start from the branch, rebase onto the latest base, and finish the steps below.
PR: #2039 (processgroup: macOS reads process groups, so left-running services and the job bound work there)
Branch: feat/darwin-process-inspection
What it fixes. Release builds use CGO_ENABLED=0, so on macOS process-group inspection fell back to nothing: left-running services were not reported and the job bound could not see a subtree's CPU. The branch reads process groups, descendants, CPU, argv and listening ports through system calls without cgo.
Done. Implementation, the two R4 tests passing on a Mac, CGO_ENABLED=0 darwin builds.
Left. Rebase onto the latest app/v3; full Mac go test ./... vs base; CI comparison; a Codex round; a manual check in the app (a left-running service is reported with argv and port; a bounded CPU-heavy job shows subtree usage); the benchmark's ns/op.
🤖 Generated with Claude Code
Activity
AbirAbbas commentedon Nov 3, 2026
Merged via #2039 (app/v3 2c7f108). On macOS 15.6 arm64 the darwinproc, executor, processgroup and session tests fail on the branch only where base fails too (9 failures, all among base's 11). With CGO_ENABLED=0, as release builds use, the branch passes executor's TestLocalRecordsAListeningServiceFully and TestAShellCallTellsTheObserverWhatItLeftRunning, which base fails. CGO_ENABLED=0 darwin builds, Linux and darwin/amd64 vet, and the Windows build are all clean. CI has nothing beyond the app/v3 baseline except flakes that pass on a rerun of the same SHA or also fail on unrelated branches. Codex's only finding, a PID-reuse race in the service description that Linux already had, is filed as #2260. Not done: the in-app manual check and the benchmark.