There is a race going on somewhere when we attempt to remove elements
*while* the parent container is switching to PLAYING.
In order to avoid this issue with discoverer, make sure we never
remove elements while switching to PLAYING.
This would result in a lot of warnings regarding elements not being
in NULL state when removed, or even leaked elements.
Instead make sure we take the lock and check whether we are processing
or not before allocating or adding anything to the pipeline
And monitor no_more_pads.
With live sources such as rtsp, uridecodebin only creates its
child decodebins between PAUSED and PLAYING.
This means that the ASYNC_DONE it posts when getting NO_PREROLL
in its change_state method gets immediately propagated by the
GstBin parent class, as opposed to a situation where a
decodebin has been added to it already, and has posted ASYNC_START.
The proposed solution, instead of simply waiting for ASYNC_DONE,
and finishing prematurely in that case, waits for three conditions
to be true:
* the uridecodebin needs to have emitted no_more_pads
* its current state must be PAUSED if not live, PLAYING otherwise
* There must be no "pending subtitle pads", ie pads where we haven't
received tags yet.
All these conditions are checked in the message handler, as we
post custom messages on it when we get subtitle tags or no_more_pads.
https://bugzilla.gnome.org/show_bug.cgi?id=783257
Child streams could have more accurate width/height or various other
information added. If they have the same name, they are likely to be the
same streams.
https://bugzilla.gnome.org/show_bug.cgi?id=782697
The parser might do some conversion on a stream but the stream keeps
being the same, and we need to make sure GstDiscoverer detects it is the
case.
https://bugzilla.gnome.org/show_bug.cgi?id=778298
It is often usefull to make sure that you get a full copy of a profile.
For example you want to let the user modify it in the user interface
but still keep an unchanged version for later use.
API:
gst_encoding_profile_copy
encoding-profile.c: In function ‘get_profile_format_from_possible_factory_name’:
encoding-profile.c:1532:6: error: ‘fact’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
if (fact)
^
encoding-profile.c: In function ‘profile_from_string’:
encoding-profile.c:1720:6: error: ‘res’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
if (profile)
^
cc1: all warnings being treated as errors
Instead of enforcing the user to know and understand caps to describe
the encoding format, let him use element factory names directly.
This also makes it possible to ensure that a specific encodore/muxer
is used instead of letting the ranking system do it.
It is now possible to describe an encoding format simply specifying:
matroskamux:x264enc:vobisenc
Factor out functions in the parsing, cleaning up the whole thing.
Update documentation.
We used to only care about the name of the files even if the name
is defined in the encoding target serialized file.
That commit also allows user to define several names for a single
target file (using a ';' between the names) which allows us to have
a target for youtube that is called 'youtube;yt' or a target for
'ogg;ogv;oga' file extension.
The caps put into the stream topology by decodebin are the caps at the
moment the pads are exposed on it. This is usually before decoders
received any buffers.
In discoverer we however wait for pre-roll, which ensures that each
decoder handled buffers already. At this point, there might be more
information known about the caps already that we could make use of.
One example here is extra information stored in the SEI of H264, like
the multiview-mode. This will be known if there is a SEI before the
first keyframe, but decodebin won't put this into the topology as it
only waits for the initial caps of h264parse (which come directly after
SPS/PPS).
With this change, the multiview-mode is in the caps reported by
discoverer in many cases.
