gstreamer/docs/random/autoplug1
Wim Taymans 1e3492f0b3 Added explanation for current autoplugging
Original commit message from CVS:
Added explanation for current autoplugging
2000-12-20 00:55:45 +00:00

202 lines
6.5 KiB
Text

A little explanation of the first autoplugger in GStreamer:
Autoplugging is implemented in the following places:
gstpipeline.c : construction of the pipeline
gstautoplug.c : selection of the elementfactories needed for autoplugging
1) pipeline setup
-----------------
before any autoplugging will take place, a new GstPipeline has to be created.
The autoplugger needs to have a src element and one or more sink elements. the
autoplugger will try to find the elements needed to connect the src element
to the sinks.
using:
gst_pipeline_add_src (GstPipeline *pipeline, GstElement *element);
a source element is added to the pipeline. only one src element can be added
for now.
using:
gst_pipeline_add_sink (GstPipeline *pipeline, GstElement *element);
a sink element can be added to the pipeline.
2) starting autoplug
--------------------
when the pipeline has been set up as above, you will call
gst_pipeline_autoplug (GstPipeline *pipeline);
to start the autoplugger. this will be done in four phases
ex. we are going to autoplug an mpeg1 system stream.
2a) phase1: figure out the type (GstCaps) of the src element.
-------------------------------------------------------------
the gsttypefind element is connected to the "src" pad of the source
element. gst_bin_iterate is called in a loop until gsttypefind
signals "have_type". the gst_bin_iterate is stopped and the GstCaps
is retrieved from the gsttypefind element.
gsttypefind is disconnected from the src element and removed from the
bin.
the GstCaps of the source element is called src_caps later on.
ex. all typefind functions are tried and the one in mpeg1types will
return a GstCaps:
video/mpeg,
"systemstream", GST_PROPS_BOOLEAN (TRUE),
"mpegversion", GST_PROPS_INT (1),
NULL
2b) phase2: create lists of factories.
---------------------------------------
for each sink:
{
sinkpad = take the first sinkpad of the sink (HACK)
call
list[i] = gst_autoplug_caps (src_caps, sinkpad->caps);
I++;
}
gst_autoplug_caps will figure out (based on the padtemplates)
which elementfactories are needed to connect src_caps to sinkpad->caps
and will return them in a list.
ex. we have two sinks with following caps:
video/raw audio/raw
"...." "...."
gst_autoplug_caps will figure out that for the first sink the following
elements are needed:
mpeg1parse, mp1videoparse, mpeg_play
for the second sink the following is needed:
mpeg1parse, mp3parse, mpg123
We now have two lists of elementfactories.
2c) phase3: collect common elements from the lists.
---------------------------------------------------
the rationale is that from the lists we have created in phase2, there
must be some element that is a splitter and that it has to come first (HACK)
We try to find that element by comparing the lists until an element differs.
we add the common elements to the bin and run gst_pipeline_pads_autoplug. this
function will loop over the pads of the previous element and the one we
just added, and tries to connect src to sink if possible.
If a connection between the two elements could not be made, a signal "new_pad"
is connected to the element so that pad connection can occur later on when
the pad is actually created.
ex. when we compare the two lists we see that we have common element: mpeg1parse.
we add this element to the bin and try to connect it to the previous element in
the bin, the disksrc.
we see that the src pad of the disksrc and the sinkpad of the mpeg1parse element
can be connected because they are compatible. We have a pipeline like:
---------) (--------
disksrc ! ! mpeg1parse
src --- sink
---------) (--------
2d) phase4: add remaining elements
----------------------------------
now we loop over all the list and try to add the remaining elements
(HACK) we always use a new thread for the elements when there is a common
element found.
if a new thread is needed (either bacuase the previous element is a common
element or the object flag of the next element is set to GST_SUGGEST_THREAD)
we add a queue to the bin and we add a new thread. We add the elements to
the bin and connect them using gst_pipeline_pads_autoplug.
If we add a queue, we have to copy the caps of the sink element of the queue
to the src pad of the queue (else they won't connect)
we finally arrive at the sink element and we're done.
ex.
we have just found our mpeg1parse common element, so we start a thread.
We add a queue to the bin and a new thread, we add the elements
mp1videoparse and mpeg_play to the thread. We arrive at the videosink, we
see that the SUGGEST_THREAD flag is set, we add a queue and a thread and
add the videosink in the thread.
the same procedure happens for the audio part. We are now left with the
following pipeline:
We will also have set a signal "new_pad" on the mpeg1parse element bacause
the element mp1videoparse could not be connected to the element just yet.
(------------------------------------) (----------
!thread ! ! thread
! (-------------) (---------) ! ! (---------)
! !mp1videoparse! !mpeg_play! ! ! !videosink!
videoqueue--sink src -- sink src -- queue --- sink !
---------) (-----------) ! (-------------) (---------) ! ! (---------)
disksrc ! ! mpeg1parse! (------------------------------------) (-------------
src --- sink !
---------) (-----------)
queue----- same for audio
then we play, create_plan happens, data is flowing and the "new_pad" signal is called
from mpeg1parse, gst_pipeline_pad_autoplug is called and the connection between
mpeg1parse and the videoqueue is made. same for audio.
voila. smame procedure for mp3/vorbis/avi/qt/mpeg2 etc...
Problems:
---------
this is obviously a very naive solution. the creation of the elements actually happens
beforehand. MPEG2, for one, fails bacause there are multiple possibilities to go
from the mpeg demuxer to audio/raw (ac3, mp3)
Also any intermedia elements like mixers (subtitles) are not possible because we
assume that after the common elements, the streams to not converge anymore.