net: TestReadTimeout failures with "EOF"

[bcmills]

#!watchflakes
post <- pkg == "net" && test == "TestReadTimeout" && `timeout_test.go:\d+: .*EOF`

--- FAIL: TestReadTimeout (0.31s)
    timeout_test.go:348: #1/1: EOF
FAIL
FAIL	net	47.085s

2021-01-13T23:37:31-7eb31d9/dragonfly-amd64-5_8
2020-11-06T19:42:05-362d25f/freebsd-arm-paulzhol

These failures seem to be quite rare, but the failure mode is not exactly useful. It's not clear to me what's going on here.

CC @bradfitz @ianlancetaylor

[bcmills]

Recently seen in https://build.golang.org/log/2a86707544423f7ac6bc8efe0d52c8e6ac424915; added a watchflakes pattern.

(CC @neild)

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net" && test == "TestReadTimeout" && `timeout_test.go:\d+: .*EOF`

2023-10-23 22:56 linux-s390x-ibm go@f09db2bb net.TestReadTimeout (log)

2023/10/23 20:37:47 killing splice client after 5 second shutdown timeout
--- FAIL: TestReadTimeout (0.35s)
    timeout_test.go:364: #1/1: EOF

— watchflakes

[gopherbot]

Found new dashboard test flakes for:

#!watchflakes
post <- pkg == "net" && test == "TestReadTimeout" && `timeout_test.go:\d+: .*EOF`

2024-07-15 17:08 gotip-plan9-arm go@8f1ec59b net.TestReadTimeout (log)

=== RUN   TestReadTimeout
    timeout_test.go:369: #1/1: EOF
    timeout_test.go:338: accept tcp 127.0.0.1:51179: open /net/tcp/140/listen: '/net/tcp/140/listen' listen hungup
--- FAIL: TestReadTimeout (0.05s)

— watchflakes