[pulseaudio-discuss] Is mainloop thread safe?

[Lennart Poettering]

On Thu, 16.04.09 17:59, Zhang, Xing Z (xing.z.zhang at intel.com) wrote:

> Hi Lennart:

> It seems mainloop is not thread safe. I'd like to cite time_free as
> example, pls correct me if I miss something.
> 
> cleanup_time_events() simply iterates time_events list to free dead
> events and decrease time_events_please_scan. If there are three time
> events in list, #1, #2, #3, and cleanup_time_events() currently at
> #2 to check whether it is dead.
> 
> Now another thread call time_free() on #1 which will increase
> time_events_please_scan. Then pa_assert(m->time_events_please_scan
> == 0); will fail when iteration ends in cleanup_time_events().
> 
> I didn't encounter this case but from code it exists in theory.  

Quoting the doxygen generated docs:

<snip>
The PulseAudio client libraries are not designed to be directly
thread-safe. They are however designed to be reentrant and
threads-aware.
</snip>

So, nothing in PA is thread safe on its own. However the whole thing
is written with threads in mind. It is your job however to place the
appropriate locks.

If you use the pa_threaded_mainloop mainloop implementation there are
a couple of functions that ease development with threads. However,
it's always your job to put the appropriate
pa_threaded_mainloop_lock() and pa_threaded_mainloop_unlock() each
time you access one of the PA client objects.

Lennart

-- 
Lennart Poettering                        Red Hat, Inc.
lennart [at] poettering [dot] net         ICQ# 11060553
http://0pointer.net/lennart/           GnuPG 0x1A015CC4