Do not re-call avcodec_get_context_defaults3(), it is already
called by avformat_new_stream() and it leaks the codec
priv_data that was already allocated.
Use avformat_free_context() instead of freeing (not)
everything manually.
Fix trac ticket #2322.
This reverts commit 6cc12353a8.
Conflicts:
libavformat/version.h
Allowing to automatically select the concat demuxer raises
security concerns, as it allows a possibly hostile file to
access any file on the system. Guessing the format based on
the file name extension does not allow to enable the safe
mode designed to avoid it.
If the first "special" character in a filename is a comma,
it can introduce protocol options, but only if there is a
colon at the end. Otherwise, it is just a filename with a
comma.
Fix trac ticket #2303.
Fix trac ticket #2300 because the duration of the segments
was computed using the timestamp of the last packet plus its
duration using the 1/90000 default time base instead of using
the chained muxer time base.
Fix trac ticket #2242.
Note: under valid circumstances, when using -fix_sub_duration,
the last subtitle is a dummy termination packet, with no
allocated memory.
The following commit:
b97d61f avfilter/ff_merge_formats: only merge if doing so does not loose chroma or alpha
introduced an exception to avoid lossy conversions.
Add a comment to explain the logic.
Fix the call to avoid applying it on audio formats.
If buffersrc was configured for frames with an unknown layout,
the incoming frames will have channel_layout = 0.
If the format negotiation has selected a known (and compatible)
channel layout for the link, the frame is assumed to have
that layout, the field must be set before injecting the frame
in the filters.
The size of the AVFrame structure is not part of the ABI;
it can grow with later versions. Therefore, applications
are not supposed to allocate AVFrame directly, they are
supposed to use avcodec_alloc_frame() instead.
Add a field to the AVBufferSinkParams for a list of accepted
channel counts.
Change the default value for the lists in AVBufferSinkParams
to NULL, to let lavfi fill them with something useful.
The fail_enomem path is unneeded because filters are uninited
on failure.
The exact packing of Opus inside Matroska is not finalized.
Use A_OPUS/EXPERIMENTAL as codec name, like mkvtoolnix.
The A_OPUS name stays to let ffmpeg open files it has produced
until now, but newly produced file use the EXPERIMENTAL version.
Once the spec is stabilized it will be possible to consider
options to ensure compatibility with these files.
Some of the filters tests use globbing characters, especially
brackets for filter pad labels. While most of these strings
are way too complicated to ever match an existing file name
and are therefore kept unchanged in the command line (an old
misfeature of the shell language that happens to be convenient
here), at least one use is simple enough to match random files
lying in the current directory. If that happens, the string,
that was meant to be kept verbatim, is replaced by the file
name, and that causes the test to fail (or worse).
Consider the color space as an hypercone with apex alpha=0
and base alpha=1 instead of an hypercube.
Make the encoder consider very transparent colors more similar
even if the hue is very different.
This corresponds roughly to using the alpha as a weight for the
color difference.
Only 4 bits of alpha are used, because this is what dvdsub uses,
and it avoids overflows.
Fix trac ticket #2005.
This is a temporary workaround until all filters have been
upgraded to filter_frame and the framework can forget completely
about start_frame/draw_slice/end_frame.
Until now, the end_display_time was ignored, making single
packets subtitles (like dvdsub) stay indefinitely.
start_display_time was also ignored, but is it almost always 0.
The previous code computes the offset by reversing the growth
of the allocated buffer size: it is complex and did lead to
inconsistencies when the size limit is reached.
Fix trac ticket #1991.
The new expansion mechanism uses the %{...} notation.
For compatibility reasons, it must be enabled explicitly,
but a warning is printed if a conflict is likely to happen.
Nicolas George