Dup of #20025 and #18490?

Oh no, not a dup. Sorry.

I just faced some confusing related to this ambiguous text message myself.

My colleague thought it applied to the entire test run. I thought he was wrong, as the usage message clearly says a test. He did however a test with the following setup:

package main

import (
	"testing"
	"time"
)

func Test1(t *testing.T) {
	time.Sleep(3 * time.Second)
}

func Test2(t *testing.T) {
	time.Sleep(3 * time.Second)
}

Which you can then run with go test -timeout=5s, which will indeed fail. OK this made me understand that clearly the timeout is not applied on an individual test level. But it still didn't define if it's applied on the entire test run, or on a package level. So I did some testing myself, inspired by my colleague, using the following setup:

// foo_test.go

package main

import (
	"testing"
	"time"
)

func Test1(t *testing.T) {
	time.Sleep(3 * time.Second)
}

// bar/bar_test.go

package bar

import (
	"testing"
	"time"
)

func Test1(t *testing.T) {
	time.Sleep(3 * time.Second)
}

And then run: go test -timeout=4s ./...

You'll see that this DOES run. So clearly the timeout value IS on a per package level. It is NOT applied on the entire test run, and it is not applied per individual level.

Which is why I propose that we change the message to something like:

	-timeout t
	    If a package runs longer than t, panic.
	    The default is 10 minutes (10m).

Or anything similar that mentions the word "package", as "a test" is really way too ambiguous, especially when left on its own.

CL https://golang.org/cl/45816 mentions this issue.