GitLab

and is replaced by two members : i_physical_channels and i_original_channels.
i_physical_channels describes the channels which are effectively present
in the stream (at the current point of the aout pipeline), whereas
i_original_channels represents the channels we used to constitute the
stream. For instance a mono stream (FRONT_CENTER) may emanate from a
FRONT_LEFT channel of a STEREO stream.

Additionally, this contains hooks to dynamically select the audio device
and the channels configuration. In order to do that, all aout plugins,
and all interface plug-ins need to be adapted. Currently only SDL
(partially OSS) and the rc interface have been adapted, please use them
as a guide. Other aout plug-ins have been temporarily disabled.

IMPORTANT : if you need a working sound output, DO NOT UPDATE YOUR TREE
for several days until we fix that.

* modules/misc/network/ipv4.c: Fixed a long-standing segfault when saving
  preferences and asking for multicast.

* src/audio_output/intf.c: fixed a bug affecting audio volume initialization.

    it simply outputs a usage message.
  * ./src/*: fixed a bunch of (legitimate) warnings under VC++.
  * ./evc/*, ./msvc/*: split the workspaces into three projects (libvlc, vlc
    and plugins -- plugins doesn't build anything yet).

  * ./modules/access/http.c: fixed a segfault (unchecked strstr return value).
  * ./src/input/input_ext-plugins.c: removed platform-specific code and put
    it in plugins. This leads to some code duplication but I have an evil
    plan to get rid of that, too.

* src/audio_output/filters.c, src/audio_output/input.c, include/aout_internal.h,
   modules/audio_filter/resampler/*: Changes that allow the resamplers to set
   the start and end date of the frame it outputs.
   This allows us for instance to output a smaller frame than what we should, and
   keep the rest of the data to compute the resampling of the next frame.
   In short, we can implement much more complex resampling algorithms than before.

* modules/audio_filter/resampler/linear.c: rewrote the linear resampler. The audio
   quality doesn't seem to be better than the ugly resampler, maybe I shouldn't
   have wasted my time on this...

* src/audio_output/output.c: fixed another quality affecting bug.

* include/aout_internal.h src/audio_output/input.c: major change to the
   resampling algorithm. When resampling is requested to keep the audio stream
   synchronised to the main clock, we trigger it but we change the resampling
   rate only progressively so it doesn't get noticed too much by the listener.

* ALL: using "%ll" in printf format strings is not portable (notably on win32) so
   we now use the I64Fx familly of macros to print 64 bits numbers (x being
   replaced by d,i,o,u,x or X).
   eg: msg_Dbg( p_this, "sample is "I64Fi" late", ll_time );

* src/misc/messages.c: got rid of ConvertPrintfFormatString().

* src/audio_output/common.c, include/aout_internal.h: added a new function
   aout_FifoFirstDate() that allows the aout plugin to schedule the first
   play.
* src/audio_output/output.c: fixed another problem affecting the audio quality.
* modules/audio_output/directx.c: improvements. The buffering scheme has been
   changed to be less affected by temporary starving. We also schedule the first   sample to play.

* modules/codec/spudec/spudec.c, modules/codec/spudec/parse.c: fixed a couple
   of problems introduced with the recent changes to the bitstream facility.
* src/audio_output/output.c: fixed a quite annoying bug in aout3 that was
   triggering unnecessary trashing of audio frames.

* Miscellaneous small fixes.

  messages,
* Updated BeOS and QNX outputs with respect to i_channels policy.

* include/vlc_threads.h: Avoid locking the interface,
* ALL: reworked all audio output plug-ins to correctly take into account
  i_channels. At present this is known to break OSS, probably others.

* Finally fixed the audio problem on most platforms.

  * ./src/misc/objects.c: structure_lock is now local.
  * ./src/misc/threads.c: implemented named mutexes which provide a handy
    way to protect data across plugins which might be sharing the same
    resources. Thread-unsafe libraries come to mind, but we can imagine
    using a named mutex for configuration files, or special devices.
  * ./include/main.h: removed global_lock because of vlc_mutex_need, removed
    p_global_data because it was never used, removed structure_lock because
    it did not belong here.

    locales breakage.
  * ./src/libvlc.c: libvlc understands the VLC_VERBOSE environment variable,
    to be set between 0 and 4. Default value is 0 for a program using libvlc,
    but vlc sets it to 1 by default.
  * ./src/misc/configuration.c: -v now works the old way (-v, -vv, -vvv) as
    well as the new way (-v0, -v1, -v4). -v1 is the same as -v, and -v0 is
    the same as -q (quiet). Hope it's all understandable.

new ones.

* ALL: Fewer overhead when resampling.

* Fixed a compile issue in s16tofloat32swab.
* Fixed a typo in demux/mpeg/ts.c.

VLS's.
* Fixed a bug with non-ASCII characters in the OS X playlist.
* Fixed a warning on start-up.

* First hooks for changing the audio output plug-in and its parameters
  on-the-fly.

  for information on how to use it in your interface plug-ins. In the
  SDL vout, b and n are mapped to sound down/sound up.
* Fixed a major in the mad plug-in with wrong dates.
* Fixed a compilation bug.

* configure.in: Fixed a typo.

* modules/codec/a52.c: Fixed detection of A/52 sound.
* modules/audio_filter/converter/a52tofloat32.c: Fixed a bug related to
  downmixing.
* ALL: Added hooks for audio volume management.

autodetect S/PDIF hardware support and the audio output plug-in chooses
the best one. Of course it probably breaks existing output modules, see
oss.c for hints :).

Also include minor tweaks which will entitle us to decode and output 5.1
audio. It may break existing plug-ins, too. See
audio_output.c:aout_FormatNbChannels for hints.

pf_setformat disappears (it was never called independantly of Open). If
the format needs to be changed, we will have Close/Open. This is much
simpler than before. Please check that I didn't break some plug-ins.

  (see my last mail).
* aout: Rewrote our whole lock policy. The output thread now doesn't require
  the mixer_lock, which might avoid delays. We will also be able to change
  the filter pipelines at runtime.

* Reactivated the A/52 demux.
* Wrote a real full-featured float32 mixer.

* New NextPTS() function, to replace CurrentPTS() (gives the PTS which will
  be valid for NEXT byte).
* Fixed mono mode in the builtin mpeg audio decoder.
* Various fixes related to PTS calculation in audio decoders.

    sound synchronization.
  * ./src/audio_output/output.c: reverted another change I previously did :-)

    prototype change; it looks like we won't need it after all.
  * ./src/audio_output/output.c: if audio output is too slow, but we
    know there is no next packet, then we play it anyway. Rationale behind
    this is: better play a slightly wrongly dated packet than play silence,
    because it might be wrongly dated but there are chances it won't break
    audio continuity.
  * ./modules/audio_output/sdl.c: workaround for SDL's crap audio output
    system (callback gets called at random times, and no way to know the
    latency of the samples being played). Audio output is now rather good
    here, but please give feedback.
  * ./modules/audio_output/oss.c: we delay a bit before asking for a new
    buffer to avoid starvation.

    which lets the audio output module give a timeout value for buffer
    delivery. This fixes a few issues with aout modules which were calling
    aout_OutputNextBuffer way too early.

* Audio and video should now _always_ be perfectly in sync.
* Changes in a52.c (end_date management) must be ported to other codecs,
  otherwise mixer errors will appear.
* Audio output plug-in can now tell whether they can handle a little
  clock drift (for instance, inserting zeroes after a pause).

pause/play and change of audio channel.

* Fixed a few boolean_t.

* Major changes in the handling of dates in the audio output ; still buggy,
but differently.
* Fixed a major bug in the aout pre-filters, preventing to run any other
decoder than liba52.
* p_aout->output.pf_play now has only one argument.

  * ./src/audio_output/audio_output.c, ./src/audio_output/mixer.c: fixed a
    horrible sound issue by making sure internal buffers have a size multiple
    of 4 bytes. Waiting for a real fix made in Meuuh, of course.

  received does not start exactly at the given date, it indicates if the
  output plug-in is able to compensate for the drift (for instance on
  startup, or with S/PDIF packets), or if we need the aout core to
  resample the coming buffers. It is currently unimplemented.