starseeker (Dec 06 2023 at 01:05): starseeker (Dec 06 2023 at 01:05):I have a rather vexing problem with the POSIX wait(&retcode) function, and I'm not sure how to zero in on it. I've defined a test for the libbu process wrappers to verify that I can detect a non-zero child exit code:
https://github.com/BRL-CAD/brlcad/blob/facetize/src/libbu/tests/process.c#L96
https://github.com/BRL-CAD/brlcad/blob/facetize/src/libbu/tests/subprocess.cpp#L62
It should return 2, but on Linux when I run this via bu_test I get a "0" return code. However, if I run the subprocess exec directly, "echo $?" does indeed report "2" for an exit code.
Even more vexing, if I put the bu_test call into gdb and break on bu_process_wait, the program behavior changes and it DOES end up with 2 as the return code.
Am I setting this up incorrectly somehow?
starseeker (Dec 06 2023 at 01:48):OK, some progress - closing the stdout and stderr of the child process before waiting was an issue, from the looks of things.
Alexis Naveros (Dec 06 2023 at 04:15):Perhaps not the issue, but you should only use WEXITSTATUS on a return code that passes the WIFEXITED() macro test
Alexis Naveros (Dec 06 2023 at 04:16):And when a program behaves differently when launched in gdb than standalone, what I usually do is to put a sleep() at the top of main(), launch it standalone and attach gdb to it before the sleep() finishes
starseeker (Dec 06 2023 at 15:17):It's looking like there are multiple issues at play... it never fails.
starseeker (Dec 06 2023 at 15:40):I'm commencing to think it's time to dissect CTest and identify the subset of their capabilities that we need for libbu process control - a cursory inspection suggests that they're leveraging libuv and put quite a bit of work into this, so trying to custom roll our own solution without reusing their code is probably going to be both a lot of work and error prone. The core of what CTest does is almost exactly what we need bu_process to do - launch processes, support timeouts and tracking return codes, and collect output.
starseeker (Dec 06 2023 at 18:43):Hmm. Looks like there are also a few small libraries for doing similar things. Experiment time
Christopher (Jan 10 2024 at 14:30):starseeker said:
Hmm. Looks like there are also a few small libraries for doing similar things. Experiment time
I think I'm running into something similar.. Did anything come of your experiments?
starseeker (Jan 10 2024 at 20:40):Yes, actually. One of them worked in the facetize branch.
starseeker (Jan 10 2024 at 20:43):The subprocess bit is here: https://github.com/BRL-CAD/brlcad/commit/c3eefb86430b7bb85e586412d0a26bf3e1dd585c
starseeker (Jan 10 2024 at 20:43):https://github.com/sheredom/subprocess.h
starseeker (Jan 10 2024 at 20:44):I was focused on facetize, so I haven't audited whether it can straight up replace libbu's process bits completely
Christopher (Jan 10 2024 at 20:46):Gotcha. I'll see if that helps the errors I'm seeing in rtweight.
Christopher (Jan 10 2024 at 20:47):starseeker said:
I was focused on facetize, so I haven't audited whether it can straight up replace libbu's process bits completely
Should probably set up a test suite for the process stuff's eventually
starseeker (Jan 10 2024 at 20:48):There is this: https://github.com/BRL-CAD/brlcad/blob/main/src/libbu/tests/process.c
Last updated: Oct 09 2024 at 00:44 UTC