the patch changes the interface of the (internal) fit_buf() function:
fit_buf() manages the memblock of the buf chunk, it reallocates the memblock
if the requested number of bytes ('len') if larger than the memblock's size ('size')
and optionally preserves 'copy' bytes
the code should be in line with the comment now
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
rindex() appears to be "non-standard" to an extent, and it caused a
build failure on mingw32.
From the man page of rindex(): "POSIX.1-2008 removes the
specifications of index() and rindex(), recommending strchr(3) and
strrchr(3) instead."
This fixes an assertion crash:
[pulseaudio] source.c: Assertion 'PA_SOURCE_IS_LINKED(s->state)' failed at pulsecore/source.c:734, function pa_source_update_status(). Aborting.
The crash happened when a Bluetooth headset profile was changed from
a2dp to hsp. During the profile change three devices are created:
a sink, a monitor source for the sink, and a regular source. First
pa_sink/source_new() are called for each device, and that puts the
devices to u->core->sinks/sources. Then, pa_sink_put() is called for
the sink, and that in turn calls pa_source_put() for the source. At
that point module-device-manager decides to reroute all source
outputs. The non-monitor source that the Bluetooth card created hasn't
been linked yet at this stage, because it will only be linked after
the sink and the monitor source have been linked. So,
module-device-manager should take into account during the rerouting
that not all sinks and sources are necessarily linked. This patch does
that.
Reported-By: Iskren Hadzhinedev <i.hadzhinedev@gmail.com>
Not having ORC_SOURCE defined results in different tarballs depending on
whether the dev issuing 'make dist' has orc support enabled or disabled.
Specifying ORC_SOURCE unconditionally addresses that, without causing
negative effects on users not having orc in the end.
The bigger than usual bump in libpulse-simple was warranted by the
change in pa_simple_flush() that allows also record streams to be
flushed. There are no changes to the function signature, but it's in
practice a change in the ABI anyway, because new clients using the new
possibility won't work with older versions of the library.
libpulse-mainloop-glib got a bug fix in commit
68156d3f79.
A crash was observed that was caused by pa_memblockq_peek() returning
a NULL memblock in sink_input_pop_cb(). The scenario where this was
happening was
1. Delete 2 rtp-recv's connected to a ladspa-sink
2. Delete ladspa-sink
3. Delete alsa-sink
4. Create alsa-sink
5. Create ladspa-sink
6. Create 2 rtp-recv's connected to the ladspa-sink
The crash was probably caused by a rewind that made the read index go
negative while the write index was at least zero, causing there to be
a gap in the memblockq. The problematic rewind might have been caused
by adding the rtp-recv stream to the ladspa-sink. That has not been
proven, but this looks very similar to the bug that was fixed in
module-virtual-sink in commit 6bd34156b1.
pulsecore/core-format.c was recently added to libpulsecommon, and
core-format.c depends on functions in libpulse, which libpulsecommon
doesn't link to. That broke building with --as-needed. This patch adds
pulse/format.c to libpulsecommon, so that core-format.c doesn't need
to depend on libpulse any more. format.c pulls in also the dependency
to json-c.
Reported-By: Jan Steffens <jan.steffens@gmail.com>
The modargs are in both cases (a succesfull as well as a failed module
initialization) freed already in pa__done().
To avoid leaking modargs memory before they are assigned to u->modargs, the
code is reorganized to first allocate userdata, and then allocate the modargs.
Local variable ma is not needed anymore.
discussion here
http://lists.freedesktop.org/archives/pulseaudio-discuss/2013-December/019661.html
Signed-off-by: Peter Meerwald <pmeerw@pmeerw.net>
Reported-by: poljar (Damir Jelić) <poljarinho@gmail.com>
The "fix flags" (PA_SINK_INPUT_FIX_FORMAT etc.) don't work properly
with the pa_stream_new_extended() interface. This patch fixes it so
that the same effect can be achieved by leaving some of the PCM
parameters unspecified in format info objects. Also, when converting
a sample spec to a format info when using the old pa_stream_new()
interface, the "fix flags" are taken into account in that conversion.
BugLink: https://bugs.freedesktop.org/show_bug.cgi?id=68952
The check is done for clients that use pa_stream_new() but not for
clients that use pa_stream_new_extended(). This is inconsistent. We
could check that the volume channels match the channels set in the
format info struct that is passed to pa_stream_new_extended(), but
that doesn't work if the format info doesn't contain the channel
information (that can happen if the client wants the server to choose
the channel count for the stream). And it should also be possible to
pass a mono volume for a multi-channel stream. The check could be
extended to handle all these cases, but I don't see much point in
wasting time on that. The server will anyway validate the stream
parameters, it's not particularly important to fail already when the
stream is being created at the client side.
The function will be used in pa_sink_input_new() and
pa_source_output_new() to convert the sample spec given by the client
to a format info object. The set_format, set_rate and set_channels
will be set according to the stream flags (PA_SINK_INPUT_FIX_FORMAT
etc.).
The function will be used in pa_sink_input_new() and
pa_source_output_new(). The fallback parameters are used to merge the
data in the format info with the sink/source sample spec and channel
map, when the format info is lacking some information.
This also fixes an issue in pa_format_info_to_sample_spec(): it did
no validation for the channels value. Now the validation is taken care
of in pa_format_info_get_channels().
This also fixes an issue in pa_format_info_to_sample_spec(): it did
no validation for the rate value. Now the validation is taken care of
in pa_format_info_get_rate().
I will need to use the function from outside libpulse.
I added the channel map argument, because the function will be called
from another function that is expected to initialize the channel map.
I don't know if it's in practice necessary, but it shouldn't do any
harm either.
Quoting Ryan Lortie from [1]:
I assumed from my reading of the Linux code ("cap_clear()...") that it
was clearing all capabilities of the process when in fact it is only
clearing the "special to root" capabilities.
The FreeBSD version of the code indeed clears _all_ capabilities beyond
ones that the process already has (ie: cannot open any new files, create
sockets, etc.)
This has a pretty obvious adverse effect on pulseaudio's ability to do
what it needs to do -- indeed, it bombs out pretty quickly due to an
inability to read its own config file.
[1] https://bugs.freedesktop.org/show_bug.cgi?id=72580#c11
Hello.
Over time, I became aware of several instances of tempting but
semantically incorrect usage of PulseAudio API (one from my own bad
proposal of "improving" Wine, one from Parole media player and one
from Webkit-GTK). I want to document these gotchas so that other
developers don't fall for that. See the attached patch.
I have checked that the rendered HTML is correct, but need someone to
confirm the factual accuracy of the proposed changes and, possibly, to
improve the wording.
--
Alexander E. Patrakov
I don't like the expression "this Pulseaudio" (even though that's
originally written by me), just "PulseAudio" is enough. Also, on
FreeBSD there's no libcap, so let's refer only to "capabilities".
cap_init() and friends are Linux-specific, so only use them if we're on
Linux.
Add support for FreeBSD capabilities if we find <sys/capability.h> to be
available there.
Add an #else (not Linux or FreeBSD) case with an #error requesting
contributions for other platforms.
This patch keeps the cap_init check in configure.ac but removes the
error if it fails. This will ensure we link to -lcap if needed, but
won't fail for the case that capabilities are part of the core system
(as on FreeBSD).
We do however, modify the header check to ensure we fail if there is no
<sys/capability.h> at all and we are on a system where it could be
installed. The logic here is that it is better to give the user the
chance to install it than it is to proceed silently with a disabled
security feature on a system where it could easily be supported.
--without-caps remains an option if the user wants to force it.
https://bugs.freedesktop.org/show_bug.cgi?id=72580
