Compiler optimisations have been seen to initialise
m->n_waiting_for_accept to a positive non-zero value, so the while() in
pa_threaded_mainloop_signal() never proceeds. Fix this by properly
initializing m->n_waiting_for_accept in pa_threaded_mainloop_new().
Patch from Iain Bucław.
https://bugs.launchpad.net/bugs/502992
Fix missing argument to pa_read(), and be consistent with declaration of
state variable in pa_cpu_init_arm().
Signed-off-by: Daniel T Chen <crimsun@ubuntu.com>
pulsecore/cpu-arm.c: In function 'get_cpuinfo':
pulsecore/cpu-arm.c:70: warning: implicit declaration of function 'pa_read' [-Wimplicit-function-declaration]
pulsecore/cpu-arm.c:72: warning: implicit declaration of function 'pa_close' [-Wimplicit-function-declaration]
pulsecore/cpu-arm.c: In function 'pa_cpu_init_arm':
pulsecore/cpu-arm.c:110: warning: implicit declaration of function 'pa_split_spaces' [-Wimplicit-function-declaration]
pulsecore/cpu-arm.c:110: warning: assignment makes pointer from integer without a cast
Function `pa_split_spaces' implicitly converted to pointer at pulsecore/cpu-arm.c:110
Signed-off-by: Daniel T Chen <crimsun@ubuntu.com>
Avahi and dbus is too heavy for OSX just for the sake of publishing our
services via mDNS/Zeroconf. Apple has its own Zeroconf implementation
called Bonjour, and this patch adds a module that implements service
announcement with that API.
All data gathering is copied from module-zeroconf-publish.c, but
unfortunately the code there is too specifically made for avahi, so I
couldn't factor it out to reuse it.
This patch adds support for CoreAudio driven devices under Mac OS X. It
is typically instanciated by the CoreAudio device detection module and
handles all available streams on a specific device.
Sinks are created according to the reported stream configuration.
Float32 is used as default audio sample format at it is the only format
CoreAudio speaks natively.
Hardware volume control is not implemented yet.
This adds a new module for CoreAudio device detection. It registers a
callback to detect hotplugged devices and creates/destroys modules named
'module-coreaudio-device'. Devices are identified via a system-wide
unique AudioDeviceID.
At least for pipes, recv() with MSG_PEEK does actually eat up data from
file descriptors. Hence, this can't be used for PULLHUP emulation.
Use another ioctl hack for that.
Even on 10.5.8, poll() does not do the right thing. Haven't checked on
newer versions. Hence, wrap all occurences of poll() to pa_poll and
emulate that call with select() on OSX. This is totally embarassing.
This patch serves two purposes:
1) Allows something other than the de-facto standard udev rules dir or
/lib/udev/rules.d to be used (the udev build system allows you to
customise this)
2) Allows a prefixed, non-root install (right now, the /lib/... path
is hard-coded into the build system
'n_waiting' and 'n_waiting_for_accept' may be accessed from mulitple
threads, and thus need to be marked as volatile to suppres certain
compiler optimisations. All uses are protected by a mutex, so we don't
need to worry about cache issues (added documentation for this as well).
This addresses bug #738.
Since the stream identifiers (channels) are monotonically growing integer, it
isn't a good idea to use them as index to a dynamic array, because the array
will grow all the time. This is not a problem with client connections that
don't create many streams, but, for example, long-running clients that use
libcanberra for playing event sounds, this means that the client connection
effectively leaks memory.
Instead of declaring extra variables for the ALSA profiles and PATHS, and
using EXTRA_DIST for the udev rule, use the dist_ prefix when declaring
them for conditional installation.
This relies on the fact that automake *is* smart enough to know that there
exist *no* conditional dist, and will always distribute all of them.
Instead of having a parameter to enable/disable them (which was also
ignored by the source code side), simply identify NetBSD 5 (when the atomic
ops were implemented the first time) in a $host switch and ignore checking
for libatomic_ops.
The type -p check is not compatible with POSIX shells, and indeed fails
when used with the FreeBSD 8 default shell. Instead the AC_CHECK_PROG macro
is written properly to look for the command.
FIONREAD on OpenSolaris is declared in sys/filio.h. The RTP modules already
rely on the presence of FIONREAD, and check for the header, so this brings
the pipe source and sink modules on par with them.
Without this change, on OpenSolaris, it's possible that the compiler will
be reported, by default, as failing to produce executables, since
_XOPEN_SOURCE=600 requires a C99 compiler to compile even the simples of
the programs.
