Use of uninitialized variables after commit 099256f about airspyone_host HOT 8 CLOSED

Did you read the actual code and go through the "rabbit hole" or do you just "think" it should be "better" this way?
More precisely, did you read this line?

https://github.com/airspy/airspyone_host/blob/master/libairspy/src/airspy.c#L832

from airspyone_host.

Did you read the actual code and go through the "rabbit hole" or do you just "think" it should be "better" this way? More precisely, did you read this line?

https://github.com/airspy/airspyone_host/blob/master/libairspy/src/airspy.c#L832

Realized the wording "uninitialized" is wrong on my side. I should have expressed it as "calling ptherad_jonin() with a thread id of zero causes segfault".

from airspyone_host.

I am yet to see a segfault with that code. pthread_join returns ESRCH with thread_id 0, which is perfectly fine.

from airspyone_host.

The following program segfaults on CentOS 6, CentOS 7 and Fedora 39 but works on Raspberry Pi OS (Bookworm and Buster). So not all pthread implementations on all platforms handles a thread id of 0 to pthread_join() as ESRCH.

#include <pthread.h>

int main(int argc, char **argv) {
    pthread_t thread = 0;

    pthread_join(thread, NULL);

    return 0;
}

Compiled with:

$ gcc -o out segfault.c -lpthread

from airspyone_host.

Can't reproduce on any up to date Linux on any CPU I tested (x86, x86_64, ARMHF, AARCH64.) Maybe you just highlighted a problem in your distro? I'm sure you can find more broken Linux stuff if you bring a distro from the 90s.
Yet, the "fix" you proposed still doesn't work because of the many possible ways libusb can fail. Someone has to go through that rabbit hole and do the actual investigation work. Maybe a simple check can work for the fossilized distros. Or maybe we should exclude them to keep a sane code base.

Try it: https://onlinegdb.com/VQarzj8G-

from airspyone_host.

The reference to the old CentOS distros was just to show that the behavior of pthread_join() has been consistent for a very long time in the RHEL-line of distros. Fedora 39 is a recent distro and the issue I'm raising is not about fossilized distros.

As far as I can read, passing anything other than a joinable thread id as the first argument to pthread_join() is undefined behavior. Distros/implementations have apparently made different choices how to handle this (Debian/Ubuntu: return ESRCH, RHEL-line: segfault) . Nothing odd about that but code using pthread should be written with this in mind. In this case not calling pthread_join() with a invalid/NULL thread id or a thread id that was previously joined.

I'll create a PR with a check.

from airspyone_host.

This should be it.

from airspyone_host.

Tnx.

from airspyone_host.