This fixes an issue with SSA/ASS subtitles, where subtitles
would fail to appear if there was already a subtitle on screen.
This was because `struct _GstAssRender` had a single
`GstBuffer *subtitle_pending` member. This meant that
the assrender context could only be aware of one subtitle
at a time.
This patch changes the subtitle_pending member to a
linked list of pending subtitles.
The `gst_ass_render_chain_text` function no longer needs
to care about whether there are already subtitles pending,
it simply appends new subtitles to the list.
The `gst_ass_render_chain_video` function has been modified
to handle the list of pending subtitles.
Finally, the `gst_ass_render_pop_text` function has been
modified to pop the entire list of pending subtitles.
https://bugzilla.gnome.org/show_bug.cgi?id=735944
This is mostly a copy/paste of the negotiation function in
basetextoverlay, which was improved recently to handle many more cases.
This will allow us to negotiate a window size with downstream.
https://bugzilla.gnome.org/show_bug.cgi?id=753824
Re-uses existing blitting code and also means we can support
a lot more video formats out of the box, hence avoid unnecessary
pixel format or colour space conversions.
https://bugzilla.gnome.org/show_bug.cgi?id=692012
Especially the combining of the text and video buffers is now
almost the same as in textoverlay.
We should really get a base class for this.
Conflicts:
ext/assrender/gstassrender.c
ext/assrender/gstassrender.h
This makes sure that libass already has all text buffers for the next frame
at the time when the next frame is processed. Without this text buffers
might be one frame late or not shown at all if they're only shown for a
single frame.
Use gst_element_klass_set_details_simple(), install properties
with static strings, create pads directly from the static pad
templates, directly put the segment into the instance struct,
use GST_MSECOND instead of numbers.
The PAR is a GstFraction and not a double, use gst_structure_get_fraction().
Simplify setcaps() functions.
