proposal: testing: expose `-test.shuffle` seed in a public function.

[ucirello]

Proposal Details

#28592 added support for go test -shuffle off|on|seed.

#61256 debates the need to redo the seed when executing tests -count times.

This proposal would expose to the tests the seed used on invocation so as to enable reproducible tests.

Currently, this workaround is possible:

func TestMain(m *testing.M) {
	flag.Parse()
	shuffleFlag := flag.Lookup("test.shuffle")
	if shuffleFlag != nil {
		shuffle := shuffleFlag.Value.String()
		if shuffle != "on" && shuffle != "off" && shuffle != ""  {
			n, err = strconv.ParseInt(shuffle, 10, 64)
			if err != nil {
				fmt.Fprintln(os.Stderr, `testing: -shuffle should be "off", "on", or a valid integer:`, err)
				os.Exit(2)
				return
			}
			rand.Seed(n)
		}
	}
	os.Exit(m.Run())
}

This workaround has two problems. First problem, it relies on a deprecated interface (math/rand.Seed). Second problem, it will not share the same seed with -test.shuffle on.

Thus this proposal modifies testing to add this call.

// ShuffleSeed reports the environment random seed.
func ShuffleSeed() int64

With this interface, one can use:

$ go test -shuffle 12389761 -run=TestRandomnessSensitive # specific seed that causes a problem to happen

Code:

func TestRandomnessSensitive(t *testing.T) {
        rng := rand.New(rand.NewSource(testing.ShuffleSeed()))
        _ = rng // use RNG that will cause the issue.
}

[seankhliao]

This appears to overload the use of the -shuffle flag for not just randomizing the order of tests, but also some other application specific use cases.
I think applications would be better off defining their own flag for use in seeding their RNG.

[ucirello]

I based my decision on issuing this proposal based off of the help text:

        -shuffle off,on,N
            Randomize the execution order of tests and benchmarks.
            It is off by default. If -shuffle is set to on, then it will seed
            the randomizer using the system clock. If -shuffle is set to an
            integer N, then N will be used as the seed value. In both cases,
            the seed will be reported for reproducibility.

Perhaps I misinterpreted it?

[seankhliao]

The help text clearly states that it is only for shuffling the order of tests, not for any other purposes. and for reproducibility, the seed value is reported.

go/src/testing/testing.go

Line 2011 in d49a14c

rng := rand.New(rand.NewSource(n))

[ucirello]

I think I misunderstood of the spirit of the shuffle flag.

I am going to retract this proposal.