os/signal: "timeout waiting for hangup" in TestSignal

[bcmills]

Lots of failures in the builders of the form:

--- FAIL: TestSignal (0.91s)
    signal_test.go:80: sighup...
    signal_test.go:89: sigwinch...
    signal_test.go:96: sighup...
    signal_test.go:69: timeout waiting for hangup
FAIL
FAIL	os/signal	5.449s

2019-11-08T19:24:30-e6c12c3/freebsd-386-11_2
2019-11-08T19:24:30-e6c12c3/linux-386-387
2019-11-08T16:46:24-47232f0/linux-mips64le-mengzhuo
2019-11-07T21:59:16-4cde749/freebsd-amd64-11_2
2019-11-07T20:34:27-4751db9/linux-amd64-jessie
2019-11-07T19:20:35-ceca99b/linux-amd64
2019-11-07T19:14:38-e1ddf05/freebsd-386-11_2
2019-11-06T15:08:19-cf3be9b/freebsd-amd64-11_2
2019-11-06T02:49:10-de70de6/dragonfly-amd64
2019-11-05T20:56:05-81559af/netbsd-arm-bsiegert
2019-11-05T16:29:31-f0390ff/openbsd-386-62
2019-11-05T14:44:56-e457cc3/openbsd-386-62
2019-11-05T03:42:29-9a0a824/dragonfly-amd64
2019-11-05T03:42:00-11db7e4/freebsd-amd64-11_2
2019-11-04T21:30:29-5a7c571/linux-amd64-noopt
2019-11-04T17:19:36-1da575a/linux-amd64-fedora
2019-11-04T15:18:34-d3660e8/openbsd-386-64

CC @aclements @ianlancetaylor @mknyszek @cherrymui

[ianlancetaylor]

This is a problem with the test, exposed by using signals to preempt goroutines. The test uses a channel with a buffer size of 1 and passes it to signal.Notify with no arguments, meaning to accept all signals. Then it sends a SIGHUP to the process and expects to see a SIGHUP on the channel. But if the runtime sends itself a SIGURG to preempt a goroutine, then it also reports that signal to the channel. So if we get both a SIGURG and a SIGHUP, before the test reads from the channel, there isn't room in the buffer to hold the signals.

[gopherbot]

Change https://golang.org/cl/206257 mentions this issue: os/signal: use a larger channel buffer for all signals in TestSignal