When using transport configured via Media API sample spec needs to be
updated since codec configuration may affect it when e.g. headset
configure a different frequency or number of channels from default.
Commit 65ef80b fixed building with xcb-util >= 0.3.8, but the reply is never
checked (possible SIGSEGV if the reply is NULL) nor freed (memory leak at each
call of the functions).
Also, remove include and dependencies on xcb-atom, as it was only meaningful
for xcb_atom_get() and STRING, and depend instead on xcb >= 1.6 for
XCB_ATOM_STRING.
If u->connections isn't empty when module-dbus-protocol is
unloaded, then connection_free() is called for the
remaining connections when the idxset is freed.
connection_free() tries to remove the connection from the
idxset, but that fails, because the item has already been
removed from the idxset in this scenario.
The problem is solved by not trying to remove the connection
from the idxset in connection_free(). Instead, whoever wants
to delete connections, has to remove the connection from the
idxset in addition to calling connection_free().
This patch adds a new update mode specifier that can be optionally
given in match rules after the regexp. Property list updates triggered
by the rule will honour the given mode. The two allowed modes are 'merge'
and 'replace', corresponding to PA_UPDATE_MERGE and PA_UPDATE_REPLACE
respectively. If omitted, the mode defaults to PA_UPDATE_MERGE, ie. to
the original behavior.
For example, to force 'media.role' to be overwritten with 'bar' for
streams matching foo you can use an entry like this:
foo replace "bar"
This will really overwrite media.role to bar even if it has already been
set to something else by the application.
Thanks to Krisztian Litkey for the original patch and the description
above. In addition to implementing the new feature, this patch fixes
a number of bugs in the parsing code.
It's possible that the memblockq of a sink input is rewound to a negative read
index if the sink input is moved between sinks shortly after its creation. When
this happens, pa_memblockq_peek() returns a memchunk whose 'memblock' field is
NULL and whose 'length' field indicates the length of the gap caused by the
negative read index. This will trigger an assert in play-memblockq.c.
If the memblockq had a silence memchunk, pa_memblockq_peek() would return
silence for the duration of the gap and the assert would be avoided. However,
this approach would prevent the sink input from being drained and is thus not
possible. Instead, we handle the aforementioned situation by dropping the gap
indicated by the 'length' field of the memchunk and by peeking the actual data
that comes after the gap.
This scenario seems to be quite rare in everyday use, but it causes a severe
bug in the handheld world. The assert can be triggered e.g. by loading two null
sinks, playing a sample from the cache to one of them and then moving the
created sink input between the two sinks. The rewinds done by the null sinks
seem to be quite long (I don't know if this is normal behaviour or something
fishy in module-null-sink).
See also:
6bd34156b1
virtual-sink: Fix a crash when moving the sink to a new master right after setup.
https://tango.0pointer.de/pipermail/pulseaudio-discuss/2011-February/009105.html
Reproduce:
This problem can be reproduced with the following script:
SAMPLE_PATH="/usr/share/sounds/alsa/"
SAMPLE="Front_Left"
pactl remove-sample $SAMPLE 2> /dev/null
pactl upload-sample $SAMPLE_PATH$SAMPLE.wav
mod1=`pactl load-module module-null-sink sink_name=null1`
mod2=`pactl load-module module-null-sink sink_name=null2`
pactl play-sample $SAMPLE null1
input=`pactl list | grep "Sink Input #" | tail -n 1 | cut -d# -f2`
echo "Sample $SAMPLE playing as Sink Input #$input"
pactl move-sink-input $input null2
pactl move-sink-input $input null1
pactl unload-module $mod1
pactl unload-module $mod2
With automaticl filter loading by module-filter-apply, setting the
virtual sink/source to have the "phone" intended role will break routing
when you first connect a phone stream to an ALSA device and then turn on
your Bluetooth headset. This happens because module-intended-roles
doesn't move a stream if it is already on a device that provides the
required role.
This patch introduces a "manual_load" parameter that is meant to be used
when not using module-filter-apply for loading the AEC module. If this
parameter is set, the virtual devices are given the "phone" role, else
we count on module-filter-heuristics to do the right thing.
This makes the core code in the filter-* modules generic enough to be
used on sources or sinks. We need special handling for modules that
introduce more than one sink (for now echo-cancel only).
This makes sure that we handle streams moving between sinks properly. To
do this, we change the way the filter.* properties are handled a little
bit.
Firstly, this splits up the "filter.apply" property into two properties
- "filter.want" and "filter.apply". "filter.apply" acts as before - it
bypasses module-filter-heuristics and directly tells module-filter-apply
what filters are to be applied.
"filter.want" is used to tell module-filter-heuristics what filters the
client wants. The module then decides whether to actually apply the
filter or not (for now, this makes sure we don't apply echo-cancellation
even if requested on phone sinks (where it is assumed AEC is taken care
of or is not required).
Next, we also make sure that we track whether the client set
"filter.apply" or module-filter-heuristics did - and in the latter case,
we recalculate "filter.apply" and then have module-filter-apply apply
the filter if required. This introduces some evil in the form of causing
the move_finish callback to possibly trigger another move, but we
protect for this case (with a property) to be doubly sure of not causing
an infinite loop.
This makes sure that we don't apply AEC on sinks that are already
connected to a "phone" device, the assumptiong being that anything
marked as such either doesn't have need it, or handles it itself.
When running two connect-stress tests at the same time the liklihood of >32 streams
per sink increases. All it takes is for an event sound to fire to trigger an abort of
the test.
This leaves just a little bit of wriggle room for a couple external streams.
Of course the overall problem is still there but this just makes it
slightly less likely without really affecting the test itself.
There were several memory leaks. In addition to those,
pa_dbus_protocol_add_interface() used a string from the
caller as a key to a hashmap, instead of a copy of the
string. This caused trouble when the caller freed the
string while the key was still in use in the hashmap.
Defining this macro on a global level is disadvantageous for other APIs,
and as we need it for clock_gettime() only on Mac OS X, define it
locally in pulsecore/core-rtclock.c only.
This value is passed on to the instances of module-coreaudio-device that
are loaded upon device detection. The value is purely optional, as the
device module will fall back to to its default if it's not given.
Adds an autoclean option (defaults to TRUE) that controls whether
module-filter-apply cleans up unused modules or not. This is useful in
cases where you know that a filter will be used often and thus can avoid
overhead from repeated module load/unload.
This makes the volume tests run in two loops and print the minimum,
maximum and standard deviation of readings from the inner loop. This
makes it easier to reason out performance drops (i.e. algorithmic
problems vs. other system issues such as processor contention).
For systems which have a fcntl() implementation, we can simplify the
code which determines whether a file selector is valid in pa_poll().
The old code, which is harder to read and more expensive, stays around
for all platforms we need to emulate poll() for using select(), and
which don't provide fcntl(). IOW, for Windows.
On Mac OS X, however, the detection for bad fds via more select() calls
doesn't work, resulting in hung main loops, so the patch fixes a real
bug there.
This brings more uniformity to arguments to match module-echo-cancel
(which needs both sink and source masters, hence the disambiguation).
This will allow other modules to load filters in a more uniform way
in the future without kludges to deal with variation in arguments.
When volume changes in bluetooth device PulseAudio volume is rounded
one too low, so if bluetooth headset changes volume and that volume
is immediately set again for bluetooth device, bluetooth step drifts
lower all the time. Volume is incremented by one in the conversion so
that we get right bluetooth step when re-applying volume.
Signed-off-by: Juho Hämäläinen <ext-juho.hamalainen@nokia.com>
