runtime: TestRaceSignal failures

[gopherbot]

#!watchflakes
post <- pkg == "runtime" && test == "TestRaceSignal"

Issue created automatically to collect these failures.

Example (log):

--- FAIL: TestRaceSignal (1.69s)
    crash_cgo_test.go:431: 
    crash_cgo_test.go:432: signal: child exited

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "runtime" && test == "TestRaceSignal"

2023-05-17 14:45 darwin-amd64-longtest go@1bb2f732 runtime.TestRaceSignal (log)

--- FAIL: TestRaceSignal (1.69s)
    crash_cgo_test.go:431: 
    crash_cgo_test.go:432: signal: child exited

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "runtime" && test == "TestRaceSignal"

2023-05-20 21:12 darwin-amd64-longtest go@2ec648aa runtime.TestRaceSignal (log)

--- FAIL: TestRaceSignal (2.43s)
    crash_cgo_test.go:431: 
    crash_cgo_test.go:432: signal: child exited

— watchflakes

[bcmills]

(attn @ianlancetaylor)

I'm not sure what's going on here. The docs for os/signal indicate that the Go runtime unblocks SIGCHLD, but as far as I can tell the default behavior of SIGCHLD is still to ignore the signal. I'm not sure where the non-default behavior (of terminating the process) is coming from.

[ianlancetaylor]

I agree that this seems impossible. Sending CL 497055.

[gopherbot]

Change https://go.dev/cl/497055 mentions this issue: runtime: treat TestRaceSignal as flaky on Darwin