testing: TestRaceDuringParallelFailsAllSubtests failures

[gopherbot]

#!watchflakes
default <- pkg == "testing" && test == "TestRaceDuringParallelFailsAllSubtests"

Issue created automatically to collect these failures.

Example (log):

=== RUN   TestRaceDuringParallelFailsAllSubtests
    testing_test.go:525: /home/swarming/.swarming/w/ir/x/t/go-build1832037510/b611/testing.test -test.run=^TestRaceDuringParallelFailsAllSubtests$ -test.bench=TestRaceDuringParallelFailsAllSubtests -test.v -test.parallel=2 -test.benchtime=2x: <nil>
        === RUN   TestRaceDuringParallelFailsAllSubtests
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        === PAUSE TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === PAUSE TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === CONT  TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === CONT  TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        --- PASS: TestRaceDuringParallelFailsAllSubtests (0.00s)
            --- PASS: TestRaceDuringParallelFailsAllSubtests/sub (0.00s)
                --- PASS: TestRaceDuringParallelFailsAllSubtests/sub/subsub1 (0.00s)
                --- PASS: TestRaceDuringParallelFailsAllSubtests/sub/subsub2 (0.00s)
        PASS
    testing_test.go:534: got 0 race reports, want 2
--- FAIL: TestRaceDuringParallelFailsAllSubtests (1.04s)

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "testing" && test == "TestRaceDuringParallelFailsAllSubtests"

2024-03-04 17:32 gotip-linux-amd64-longtest-race go@c2734187 testing.TestRaceDuringParallelFailsAllSubtests (log)

=== RUN   TestRaceDuringParallelFailsAllSubtests
    testing_test.go:525: /home/swarming/.swarming/w/ir/x/t/go-build1832037510/b611/testing.test -test.run=^TestRaceDuringParallelFailsAllSubtests$ -test.bench=TestRaceDuringParallelFailsAllSubtests -test.v -test.parallel=2 -test.benchtime=2x: <nil>
        === RUN   TestRaceDuringParallelFailsAllSubtests
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        === PAUSE TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === PAUSE TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === CONT  TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === CONT  TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        --- PASS: TestRaceDuringParallelFailsAllSubtests (0.00s)
            --- PASS: TestRaceDuringParallelFailsAllSubtests/sub (0.00s)
                --- PASS: TestRaceDuringParallelFailsAllSubtests/sub/subsub1 (0.00s)
                --- PASS: TestRaceDuringParallelFailsAllSubtests/sub/subsub2 (0.00s)
        PASS
    testing_test.go:534: got 0 race reports, want 2
--- FAIL: TestRaceDuringParallelFailsAllSubtests (1.04s)

— watchflakes

[bcmills]

@golang/compiler, this looks like a false-negative from the race detector. (I don't see any way that the test in question could fail to produce the intended race.)

[mknyszek]

Interesting, I'm not sure what we can do about that. The race detector, IIUC, is always potentially going to have false negatives. Perhaps someone who's familiar with the race detector internals could help.

CC @dvyukov maybe?

[dvyukov]

Positive race tests are executed with GOMAXPROCS=1 for this reason. Do testing package tests do it?

Another option is to repeat positive tests multiple tests, if it fails at least once (on N/M times), that's good enough.

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
default <- pkg == "testing" && test == "TestRaceDuringParallelFailsAllSubtests"

2024-11-01 20:11 gotip-linux-arm64-race go@47cd14f2 testing.TestRaceDuringParallelFailsAllSubtests (log)

=== RUN   TestRaceDuringParallelFailsAllSubtests
    testing_test.go:636: /home/swarming/.swarming/w/ir/x/t/go-build2363101731/b1167/testing.test -test.run=^TestRaceDuringParallelFailsAllSubtests$ -test.bench=TestRaceDuringParallelFailsAllSubtests -test.v -test.parallel=2 -test.benchtime=2x: <nil>
        === RUN   TestRaceDuringParallelFailsAllSubtests
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        === PAUSE TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        === RUN   TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === PAUSE TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === CONT  TestRaceDuringParallelFailsAllSubtests/sub/subsub2
        === CONT  TestRaceDuringParallelFailsAllSubtests/sub/subsub1
        --- PASS: TestRaceDuringParallelFailsAllSubtests (0.00s)
            --- PASS: TestRaceDuringParallelFailsAllSubtests/sub (0.00s)
                --- PASS: TestRaceDuringParallelFailsAllSubtests/sub/subsub1 (0.00s)
                --- PASS: TestRaceDuringParallelFailsAllSubtests/sub/subsub2 (0.00s)
        PASS
    testing_test.go:645: got 0 race reports, want 2
--- FAIL: TestRaceDuringParallelFailsAllSubtests (1.04s)

— watchflakes