mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-11-25 03:01:03 +00:00
a4c0d0b8c3
Original commit message from CVS: * gst/playback/gstdecodebin.c: (type_found): Special-case the text/plain media type: we only want to recognise it as a 'raw' decoded media type if it comes from a demuxer or subtitle parser, but not if the entire stream is of text/plain type. If the entire stream is text/plain, we should just error out. This fixes playback of audio files with lyrics in totem. Totem can't distinguish between text files and subtitle files and passes any .txt file with the same basename as the main file to playbin as suburi, and playbin will then throw a 'subtitle found, but no video stream' error, which isn't entirely helpful. See #380342. Also, with this change we'll show a slightly more correct error message in case totem passes a playlist file to us (although a custom error message wording instead of the default text would probably not be a bad idea either). Same problem also needs to be fixed for playbin+decodebin2. * tests/check/Makefile.am: * tests/check/elements/decodebin.c: (src_handoff_cb), (decodebin_new_decoded_pad_cb), (GST_START_TEST), (decodebin_suite): Add simple unit test for decodebin for the above. |
||
---|---|---|
.. | ||
.gitignore | ||
decodetest.c | ||
gstdecodebin.c | ||
gstdecodebin2.c | ||
gstplay-marshal.list | ||
gstplaybasebin.c | ||
gstplaybasebin.h | ||
gstplaybin.c | ||
gststreaminfo.c | ||
gststreaminfo.h | ||
gststreamselector.c | ||
gststreamselector.h | ||
Makefile.am | ||
README | ||
test.c | ||
test2.c | ||
test3.c | ||
test4.c | ||
test5.c | ||
test6.c |
decoderbin: A bin with a sinkpad that decodes the data into raw formats. It works by sending the input data through a typefind element and then recursively autoplugs elements from the registry until a raw format is obtained. It will then create a new ghostpad on itself to signal the app of the new pad. Decodebin will also remove pads when they are removed from the stream. TODO - reuse of decoderbin, cleanup in READY state - threading after demuxing? - new_media events should be handled. - caching of elements. - abstract more elements, pads (typefind, ...); The autoplugging happens as follows: 1) typefind is added internally to the bin. 2) the have_type signal is connected to typefind. 3) in the have_type callback the close_pad_link function is called 4) close_pad_link checks the type on the pad, if it is raw, a ghostpad is created and autoplugging for that pad stops. 5) if the type of the pad is not raw, a list of possible elements that can connect to this type is generated in find_compatibles. 6) try_to_link_1 with the element list is called. The function will loop over the element list and will try to connect one of the elements to the pad. If the link works, a call is made to close_link. 7) close_link loops over all the source pads of the element and recursively calls 4) for any ALWAYS pad. For elements with a SOMETIMES pad, a structure is set up and is passed to the callback of the new_pad signal. 8) in the new_pad callback, 4) is called to try to autoplug the new pad. playbasebin: A bin with an uri property. It will find the right source element from the registry and connect a decoderbin to it. When going to the PAUSED state, it will iterate the decoderbin and listen for new pad signals from it. It will connect a queue to each new pad and will iterate the decoderbin until one of the queues is filled. It is assumed that by that time all the streams will be found so that when leaving the PAUSED state, one can query the number of streams in the media file with the given uri. Playbasebin internally groups related streams together in a GstPlayBaseGroup. This is particulary important for chained oggs. Initially, a new group is created in the 'building' state. All new streams will be added to the building group until no-more-pads is signaled or one of the preroll queues overflows. When this happens, the group is commited to a list of groups ready for playback. PlaybaseBin will then attach a padprobe to each stream to figure out when it finished. It will remove the current group and install the next playable group, then. Before going to the PLAYING state, it is possible to connect a custom element to each of the streams. To do that, you have to add the element to the bin and then connect the pad(s) from the stream(s). You do not have to add the elements in a thread, the bin will take care of then when it's needed. You are allowed to use threads inside the elements, of course. The bin tries to be smart and doesn't add a queue when there is only one possible stream. TODO - reuse, cleanup in ready state - when the first pad is closed, it's possible that another dynamic element is added somewhere so that we need a queue for the first pad as well. playbin: Extends playbasebin, sets up default audiosink and videosink for first audio/video stream detected. implements seeking and querying on the configured sinks. It also waits for new notifications from playbasebin about any new groups that are becomming active. It then disconnects the sinks and reconnects them to the new pads in the group. TODO - reuse, refcounting, cleanup in READY state - be smarter about replugging the sinks instead of removing them and readding them. - Do not crap out when the audio device is in use. general TODO - playlist support. maybe use a playlist bin that streams the contents of the playlist on a pad, interleaved with new_media events. Also add a tuner interface while we're at it. - plug the many memleaks.