A lot of time is spent in wait4() when running a BATS test-suite

[brgl]

I'm the author of libgpiod which uses BATS for its command-line tools.

I noticed that these tests - even though they use the same kernel infrastructure in the form of the gpio-sim module - take much longer to complete than the unit tests for the core C library. We're talking several orders of magnitude: library tests: 0,28s on my laptop, bash tests: ~26 seconds(!). Some users have even reported that it takes over 2 minutes for the tests to complete. They cover pretty much the same functionality and while it's clear that bash tests will be much slower, I don't think this is entirely normal.

A quick run of strace -c -f reveals the following:

% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ------------------
 91,88   19,167164         944     20285      8653 wait4
  1,72    0,359402           0    814675           rt_sigprocmask
  1,30    0,271654           1    174942    145707 readlink
  1,22    0,254268           0    261146         8 read
  1,02    0,211915          18     11283           clone
  0,50    0,104316          16      6398       665 execve
  0,47    0,097529           0    128670     13692 newfstatat
  0,36    0,074594           0     95369     49559 ioctl
  0,25    0,052272           0     77944           mmap
  0,24    0,050331           2     22081           write
[...]

The majority of time is spent in wait4() and I'm seeing a lot of wait4(-1, 0x<snip>, WNOHANG, NULL) = -1 ECHILD (No child processes) calls in the complete strace output.

Any idea what could be causing this?

[martin-schulze-vireso]

I think we need more details. In essence this is a bug report so please provide the usual details:

• OS version
• bash version
• Bats version
• Test Code
• Bats invocation command line

[brgl]

I wasn't sure if this is a bug or a feature so I opened a discussion, but sure, I'll post a detailed bug report in the issues section.

[martin-schulze-vireso]

You are welcome to post this as a bug issue but I Just wanted to request the usual information that is necessary to reproduce this issue.