With this commit, multi-player support becomes possible... it's just
not wired to the frontend yet.
This is based on massive amounts of refactoring work I did over the
past 9 years.
During UnlockActivate() while the mutex is unlocked, the IOThread can
set a new error condition, and will never again wake up the
OutputThread. This race condition can cause a deadlock in the
OutputThread.
Use SND_PCM_NONBLOCK, and perform all snd_pcm_writei() calls in the
IOThread. Use a lockless queue to copy data from the OutputThread to
the IOThread.
This rather major change aims to improve MPD's internal latency. All
waits are now under MPD's control, instead of blocking inside
libasound2.
As a side effect, an output's filter is now decoupled from the actual
device I/O, which solves a major latency problem with the conversion
filter on slow CPUs and small period buffers. See:
https://bugs.musicpd.org/view.php?id=3900
The main EventLoop can block for a long time while a client's command
runs, and is therefore inappropriate for internal engine I/O. This
fixes a serious regression for at least the "httpd" output, which used
to be hard-coded for the IOThread, but now receives the main EventLoop
as an initialization parameter.
For the mixers, this doesn't make much of a difference. They are not
latency critical.
Now ClearSocketList() may only be called from PrepareSockets().
Calling it before destroying the object doesn't work properly, because
it doesn't unregister the TimeoutMonitor and the IdleMonitor. Some of
its callers need to be fixed.
Change EventLoop::IsInside() call to EventLoop::IsInsideOrNull().
This means that BlockingCall() may be used during shutdown, after the
main EventLoop::Run() has finished. This is important because mixers
are currently registered in the main EventLoop.
An explicit destructor prevents usage of implicit move operators, even
if it's empty. Therefore, declaring a defaulted destructor with GCC
attribute "noinline" does what we want without preventing those
implicit operators.
