net/http: TestLinuxSendfile failing on mips64

[bradfitz]

TestLinuxSendfile is failing on mips64 (but not mips64le) with:

--- FAIL: TestLinuxSendfile (0.25s)
	fs_test.go:1122: no sendfile system call found in:
		[pid  2578] +++ exited with 0 +++
		[pid  2577] +++ exited with 0 +++
		[pid  2576] +++ exited with 0 +++
		+++ exited with 0 +++
FAIL
FAIL	net/http	96.972s

Not sure why.

[gopherbot]

CL https://golang.org/cl/33426 mentions this issue.

[cherrymui]

One possibility is that strace on the machine is either too old or not fully compatible with MIPS64 ABI. On some distros strace by default uses a different ABI, even the kernel is 64-bit. Not sure trace=n64_sendfile works or not.

[bradfitz]

Good theory.

[bradfitz]

That last CL didn't help. New output is:

--- FAIL: TestLinuxSendfile (0.16s)
	fs_test.go:1126: no sendfile system call found in:
		execve("/tmp/go-build627887128/net/http/_test/http.test", ["/tmp/go-build627887128/net/http/"..., "-test.run=TestLinuxSendfileChild"], [/* 33 vars */]) = 0
		[pid 20665] +++ exited with 0 +++
		[pid 20664] +++ exited with 0 +++
		[pid 20663] +++ exited with 0 +++
		+++ exited with 0 +++

[cherrymui]

I tried to trace n64_sendfile on linux-mips64 gomote machine. Now it sees the syscall,

=== RUN   TestLinuxSendfile
--- FAIL: TestLinuxSendfile (0.22s)
	fs_test.go:1122: no sendfile system call found in:
		[pid  4003] n64_sendfile()              = 22
		[pid  4006] +++ exited with 0 +++
		[pid  4005] +++ exited with 0 +++
		[pid  4004] +++ exited with 0 +++
		+++ exited with 0 +++

but it fails to get the arguments as it uses different ABI, so the regexp match later in the test function fails.

Also changing to trace n64_sendfile would break systems on which strace is using the 64-nit ABI (because the syscall is named sendfile there...). (One possible way is to try both...)

[vstefanovic]

This test is passing on a 64bit rootfs. We might want to setup mips64 buildbot to run on a 64bit rootfs (especially because of cgo), rather than adapting this test for 32bit rootfs.

[bradfitz]

@vstefanovic, what's the latest here? Can the test be re-enabled or be made robust to whichever environment it's running in?