`cabal test` freezes but `cabal test --show-details=streaming` works

[jstolarek]

I'm running the testsuite for singletons package. When I run cabal test the testsuite freezes, but if I add --show-details=streaming it works perfectly and finishes after ~30seconds. This happens with cabal 1.20 and GHC 7.8.2.

[tibbe]

@ttuegel do you mind taking a look?

[ttuegel]

@jstolarek Which operating system are you testing on?

[jstolarek]

This is 64-bit Linux

[tibbe]

@jstolarek Could you strace it to get an idea of where it's frozen?

[jstolarek]

Sure. Log is about 1.5MB - where should I upload it?

[tibbe]

Could you just paste the last 10 lines?

On Thu, Apr 24, 2014 at 11:41 AM, Jan Stolarek notifications@github.aichem.orgwrote:

Sure. Log is about 1.5MB - where should I upload it?

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/1810#issuecomment-41261417
.

[jstolarek]

Sure. Last 2K lines or so look identical:

wait4(26241, 0x7fffe3a2c4cc, 0, NULL)   = ? ERESTARTSYS (To be restarted)
--- SIGVTALRM (Virtual timer expired) @ 0 (0) ---
rt_sigreturn(0x1a)                      = 61
wait4(26241, 0x7fffe3a2c4cc, 0, NULL)   = ? ERESTARTSYS (To be restarted)
--- SIGVTALRM (Virtual timer expired) @ 0 (0) ---
rt_sigreturn(0x1a)                      = 61
wait4(26241, 0x7fffe3a2c4cc, 0, NULL)   = ? ERESTARTSYS (To be restarted)
--- SIGVTALRM (Virtual timer expired) @ 0 (0) ---
rt_sigreturn(0x1a)                      = 61
wait4(26241, 0x7fffe3a2c4cc, 0, NULL)   = ? ERESTARTSYS (To be restarted)
--- SIGVTALRM (Virtual timer expired) @ 0 (0) ---
write(9, "\377\0\0\0\0\0\0\0", 8)       = 8
rt_sigreturn(0x2)                       = 61
wait4(26241,  <unfinished ...>

This is interrupted several times with

write(9, "\377\0\0\0\0\0\0\0", 8)       = 8
rt_sigreturn(0x2)                       = 61

And before all that start there's:

rt_sigprocmask(SIG_BLOCK, [INT], [], 8) = 0
rt_sigaction(SIGINT, {SIG_IGN, [], SA_RESTORER, 0x7fc530a1c2d0}, NULL, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
rt_sigprocmask(SIG_BLOCK, [QUIT], [], 8) = 0
rt_sigaction(SIGQUIT, {SIG_IGN, [], SA_RESTORER, 0x7fc530a1c2d0}, NULL, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
pipe([14, 15])                          = 0
rt_sigprocmask(SIG_BLOCK, [], [], 8)    = 0
timer_settime(0x1, 0, {it_interval={0, 0}, it_value={0, 0}}, NULL) = 0
vfork()                                 = 26241
close(15)                               = 0
fcntl(14, F_SETFD, FD_CLOEXEC)          = 0
read(14, "", 4)                         = 0
close(14)                               = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
timer_settime(0x1, 0, {it_interval={0, 10000000}, it_value={0, 10000000}}, NULL) = 0
wait4(26241, 0x7fffe3a2c4cc, 0, NULL)   = ? ERESTARTSYS (To be restarted)
--- SIGVTALRM (Virtual timer expired) @ 0 (0) ---
rt_sigreturn(0x1a)                      = 61

Hope that helps.

[nikita-volkov]

I'm experiencing the same issue here with Cabal 1.20. Here is a failed Travis CI build. Here is a source tree, which reproduces the issue.

[tibbe]

@ttuegel have you been able to reproduce the issue or do you need help with that?

[ttuegel]

@tibbe I have reproduced @jstolarek's problem, but I haven't checked if it's the same as @nikita-volkov's. The freeze happens because rawSystemIOWithEnv (which we use to invoke the test executable) never finishes. Based on CPU usage, I think the tests are running, but deadlocking at the end. For example, the singletons tests run for 13s on my machine after last output; then CPU usage drops off. I cannot reproduce the problem with cabal-install's test suite, either. I would say it's a bug in how some test frameworks are doing output, but the test framework can't even tell what the --show-details= setting is. Thoughts?

[tibbe]

So if this didn't happen in 1.18, where we didn't have streaming, then we must either be invoking the process in the same way, not waiting for it to finish in the same way, or not passing it handles that in the same state. Some printf style debugging might tell us.