Original commit message from CVS:
Several encoding changes:
* for video, double check the pix_fmt after opening the AVCodec. The pix_fmt
will be changed to the preferred pix_fmt instead of that opening the
codec fails. We fail when it has changed (then, ffmpeg doesn't like it).
* for video, set the correct timestamp.
* for audio, do *not* set the frame_size: most codecs don't like it. Instead,
keep a cache of buffers and always give data of size "frame_size". Maybe
I should have used bytestream here, I'm not sure. This works, though.
I can now create mpeg1 video, mp2 and ac3 audio. I didn't test any others yet.
I also didn't start working on integration with any of the muxers yet, that's
all one big TODO. One thing at a time, please. :).
Original commit message from CVS:
Make it compile against latest CVS of FFMPEG:
* Put gst/gst.h above avcodec.h because it needs some types defined in glib.h
* Remove HQ (removed from lavc) property
Also add some functions to make SVQ1 decoding work:
* Add YUV9 colorspace
* Let lavc do its own buffer allocation and copy from there instead of using
our own buffer allocation functions. Somehow, that breaks. I don't know
why.
Original commit message from CVS:
Typo (> instead of <) in codec selection (we don't want raw audio to raw audio encoders), and small improvement in naming of the FFMPEG encoders. This makes their naming look slightly better in apps using element_factory->details->longname.
Original commit message from CVS:
compatibility fix for new GST_DEBUG stuff.
Includes fixes for missing includes for config.h and unistd.h
I only ensured for plugins I can build that they work, so if some of them are still broken, you gotta fix them yourselves unfortunately.
Original commit message from CVS:
* remove gstffmpegdecall. This was a temporary hack to get around
some issues with our video/avi-centered mimetype system. Now that
we use proper identifiers, we don't need that piece of cruft anymore.
* update ffmpeg codec map for ffdec_*, plus make the uncompressed stream
caps also be formed by the codec mapper (this is easier to handle imo).
* add audio support (untested, though :P).
* remove the copying of buffer content and add a get_buffer() handler
instead.
* add an int to ffmpegenc to make it compile and resolve the symbol. I'll
truely fix that later on.
* make mimetypes in the ffmpeg codec mapper conform to docs/random/mimetypes
* etc.
Oh, and let's not forget that it can play back DIVX5 et all nicely. Support
for SVQ1/3 is also there, so if someone completes qtdemux, we can watch
these, too.
Original commit message from CVS:
another batch of connect->link fixes
please let me know about issues
and please refrain of making them yourself, so that I don't spend double
the time resolving conflicts
Original commit message from CVS:
Release early.. try to wrap more ffmpeg stuff, like muxers and demuxers.
This thing needs a fix in ffmpeg (which I will send to the ffmpeg authors
eventually)
Original commit message from CVS:
* removal of //-style comments
* don't link plugins to core libs -- the versioning is done internally to the plugins with the plugin_info struct,
and symbol resolution is lazy, so we can always know if a plugin can be loaded by the plugin_info data. in theory.
