I have ran the test suite a couple of times with -count=10, no failures so far.

I have ran the test suite a couple of times with -count=10, no failures so far.

And I just did one more run, and it failed:

    executor_test.go:1117: Result did not match the golden fixture. Diff is below:

        --- Expected
        +++ Actual
        @@ -1,4 +1,3 @@
         dep1
        -dep2
        -dep3
        +dep3dep2


--- FAIL: TestFailfast (0.24s)
    --- PASS: TestFailfast/Task (13.85s)
        --- PASS: TestFailfast/Task/task (13.85s)
    --- PASS: TestFailfast/Option (14.11s)
        --- PASS: TestFailfast/Option/default (14.03s)
    --- FAIL: TestFailfast/Default (14.27s)
        --- FAIL: TestFailfast/Default/default (14.13s)

@Ikke that is a different flake. The executor tests already use a sync-buffer for the collection of stdout so I don't really understand why this would be observed, other than it might be OS weakness (Windows or Mac?). The more tasks that run in parallel, the more likely it is such a problem would occur.

Not exactly sure if anything can be done to mitigate it, on Linux I've run a few 100 cycles with no problem. Anyway, I've pushed a change, but I don't think it will help.