gstreamer/docs/manual/advanced-autoplugging.xml

622 lines
21 KiB
XML

<chapter id="chapter-autoplugging">
<title>Autoplugging</title>
<para>
In <xref linkend="chapter-helloworld"/>, you've learned to build a
simple media player for Ogg/Vorbis files. By using alternative elements,
you are able to build media players for other media types, such as
Ogg/Speex, MP3 or even video formats. However, you would rather want
to build an application that can automatically detect the media type
of a stream and automatically generate the best possible pipeline
by looking at all available elements in a system. This process is called
autoplugging, and &GStreamer; contains high-quality autopluggers. If
you're looking for an autoplugger, don't read any further and go to
<xref linkend="chapter-components"/>. This chapter will explain the
<emphasis>concept</emphasis> of autoplugging and typefinding. It will
explain what systems &GStreamer; includes to dynamically detect the
type of a media stream, and how to generate a pipeline of decoder
elements to playback this media. The same principles can also be used
for transcoding. Because of the full dynamicity of this concept,
&GStreamer; can be automatically extended to support new media types
without needing any adaptations to its autopluggers.
</para>
<para>
We will first introduce the concept of MIME types as a dynamic and
extendible way of identifying media streams. After that, we will introduce
the concept of typefinding to find the type of a media stream. Lastly,
we will explain how autoplugging and the &GStreamer; registry can be
used to setup a pipeline that will convert media from one mimetype to
another, for example for media decoding.
</para>
<sect1 id="section-mime">
<title>MIME-types as a way to identify streams</title>
<para>
We have previously introduced the concept of capabilities as a way
for elements (or, rather, pads) to agree on a media type when
streaming data from one element to the next (see <xref
linkend="section-caps"/>). We have explained that a capability is
a combination of a mimetype and a set of properties. For most
container formats (those are the files that you will find on your
hard disk; Ogg, for example, is a container format), no properties
are needed to describe the stream. Only a MIME-type is needed. A
full list of MIME-types and accompanying properties can be found
in <ulink type="http"
url="http://gstreamer.freedesktop.org/data/doc/gstreamer/head/pwg/html/section-types-definitions.html">the
Plugin Writer's Guide</ulink>.
</para>
<para>
An element must associate a MIME-type to its source and sink pads
when it is loaded into the system. &GStreamer; knows about the
different elements and what type of data they expect and emit through
the &GStreamer; registry. This allows for very dynamic and extensible
element creation as we will see.
</para>
<para>
In <xref linkend="chapter-helloworld"/>, we've learned to build a
music player for Ogg/Vorbis files. Let's look at the MIME-types
associated with each pad in this pipeline. <xref
linkend="section-mime-img"/> shows what MIME-type belongs to each
pad in this pipeline.
</para>
<figure float="1" id="section-mime-img">
<title>The Hello world pipeline with MIME types</title>
<mediaobject>
<imageobject>
<imagedata scale="75" fileref="images/mime-world.&image;" format="&IMAGE;"/>
</imageobject>
</mediaobject>
</figure>
<para>
Now that we have an idea how &GStreamer; identifies known media
streams, we can look at methods &GStreamer; uses to setup pipelines
for media handling and for media type detection.
</para>
</sect1>
<sect1 id="section-typefinding">
<title>Media stream type detection</title>
<para>
Usually, when loading a media stream, the type of the stream is not
known. This means that before we can choose a pipeline to decode the
stream, we first need to detect the stream type. &GStreamer; uses the
concept of typefinding for this. Typefinding is a normal part of a
pipeline, it will read data for as long as the type of a stream is
unknown. During this period, it will provide data to all plugins
that implement a typefinder. When one of the typefinders recognizes
the stream, the typefind element will emit a signal and act as a
passthrough module from that point on. If no type was found, it will
emit an error and further media processing will stop.
</para>
<para>
Once the typefind element has found a type, the application can
use this to plug together a pipeline to decode the media stream.
This will be discussed in the next section.
</para>
<para>
Plugins in &GStreamer; can, as mentioned before, implement typefinder
functionality. A plugin implementing this functionality will submit
a mimetype, optionally a set of file extensions commonly used for this
media type, and a typefind function. Once this typefind function inside
the plugin is called, the plugin will see if the data in this media
stream matches a specific pattern that marks the media type identified
by that mimetype. If it does, it will notify the typefind element of
this fact, telling which mediatype was recognized and how certain we
are that this stream is indeed that mediatype. Once this run has been
completed for all plugins implementing a typefind functionality, the
typefind element will tell the application what kind of media stream
it thinks to have recognized.
</para>
<para>
The following code should explain how to use the typefind element.
It will print the detected media type, or tell that the media type
was not found. The next section will introduce more useful behaviours,
such as plugging together a decoding pipeline.
</para>
<programlisting><!-- example-begin typefind.c a -->
#include &lt;gst/gst.h&gt;
<!-- example-end typefind.c a -->
[.. my_bus_callback goes here ..]<!-- example-begin typefind.c b --><!--
static gboolean
my_bus_callback (GstBus *bus,
GstMessage *message,
gpointer data)
{
GMainLoop *loop = data;
switch (GST_MESSAGE_TYPE (message)) {
case GST_MESSAGE_ERROR: {
GError *err;
gchar *debug;
gst_message_parse_error (message, &amp;err, &amp;debug);
g_print ("Error: %s\n", err-&gt;message);
g_error_free (err);
g_free (debug);
g_main_loop_quit (loop);
break;
}
case GST_MESSAGE_EOS:
/* end-of-stream */
g_main_loop_quit (loop);
break;
default:
break;
}
/* remove from queue */
return TRUE;
}
--><!-- example-end typefind.c b -->
<!-- example-begin typefind.c c -->
static gboolean
idle_exit_loop (gpointer data)
{
g_main_loop_quit ((GMainLoop *) data);
/* once */
return FALSE;
}
static void
cb_typefound (GstElement *typefind,
guint probability,
GstCaps *caps,
gpointer data)
{
GMainLoop *loop = data;
gchar *type;
type = gst_caps_to_string (caps);
g_print ("Media type %s found, probability %d%%\n", type, probability);
g_free (type);
/* since we connect to a signal in the pipeline thread context, we need
* to set an idle handler to exit the main loop in the mainloop context.
* Normally, your app should not need to worry about such things. */
g_idle_add (idle_exit_loop, loop);
}
gint
main (gint argc,
gchar *argv[])
{
GMainLoop *loop;
GstElement *pipeline, *filesrc, *typefind, *fakesink;
GstBus *bus;
/* init GStreamer */
gst_init (&amp;argc, &amp;argv);
loop = g_main_loop_new (NULL, FALSE);
/* check args */
if (argc != 2) {
g_print ("Usage: %s &lt;filename&gt;\n", argv[0]);
return -1;
}
/* create a new pipeline to hold the elements */
pipeline = gst_pipeline_new ("pipe");
bus = gst_pipeline_get_bus (GST_PIPELINE (pipeline));
gst_bus_add_watch (bus, my_bus_callback, NULL);
gst_object_unref (bus);
/* create file source and typefind element */
filesrc = gst_element_factory_make ("filesrc", "source");
g_object_set (G_OBJECT (filesrc), "location", argv[1], NULL);
typefind = gst_element_factory_make ("typefind", "typefinder");
g_signal_connect (typefind, "have-type", G_CALLBACK (cb_typefound), loop);
fakesink = gst_element_factory_make ("fakesink", "sink");
/* setup */
gst_bin_add_many (GST_BIN (pipeline), filesrc, typefind, fakesink, NULL);
gst_element_link_many (filesrc, typefind, fakesink, NULL);
gst_element_set_state (GST_ELEMENT (pipeline), GST_STATE_PLAYING);
g_main_loop_run (loop);
/* unset */
gst_element_set_state (GST_ELEMENT (pipeline), GST_STATE_NULL);
gst_object_unref (GST_OBJECT (pipeline));
return 0;
}
<!-- example-end typefind.c c --></programlisting>
<para>
Once a media type has been detected, you can plug an element (e.g. a
demuxer or decoder) to the source pad of the typefind element, and
decoding of the media stream will start right after.
</para>
</sect1>
<sect1 id="section-dynamic">
<title>Plugging together dynamic pipelines</title>
<warning><para>
The code in this section is broken, outdated and overly complicated.
Also, you should use decodebin, playbin or uridecodebin to get
decoders plugged automatically.
</para></warning>
<para>
In this chapter we will see how you can create a dynamic pipeline. A
dynamic pipeline is a pipeline that is updated or created while data
is flowing through it. We will create a partial pipeline first and add
more elements while the pipeline is playing. The basis of this player
will be the application that we wrote in the previous section (<xref
linkend="section-typefinding"/>) to identify unknown media streams.
</para>
<!-- example-begin dynamic.c a --><!--
#include &lt;gst/gst.h&gt;
GstElement *pipeline;
--><!-- example-end dynamic.c a -->
<para>
Once the type of the media has been found, we will find elements in
the registry that can decode this streamtype. For this, we will get
all element factories (which we've seen before in <xref
linkend="section-elements-create"/>) and find the ones with the
given MIME-type and capabilities on their sinkpad. Note that we will
only use parsers, demuxers and decoders. We will not use factories for
any other element types, or we might get into a loop of encoders and
decoders. For this, we will want to build a list of <quote>allowed</quote>
factories right after initializing &GStreamer;.
</para>
<programlisting><!-- example-begin dynamic.c b -->
static GList *factories;
/*
* This function is called by the registry loader. Its return value
* (TRUE or FALSE) decides whether the given feature will be included
* in the list that we're generating further down.
*/
static gboolean
cb_feature_filter (GstPluginFeature *feature,
gpointer data)
{
const gchar *klass;
guint rank;
/* we only care about element factories */
if (!GST_IS_ELEMENT_FACTORY (feature))
return FALSE;
/* only parsers, demuxers and decoders */
klass = gst_element_factory_get_klass (GST_ELEMENT_FACTORY (feature));
if (g_strrstr (klass, "Demux") == NULL &amp;&amp;
g_strrstr (klass, "Decoder") == NULL &amp;&amp;
g_strrstr (klass, "Parse") == NULL)
return FALSE;
/* only select elements with autoplugging rank */
rank = gst_plugin_feature_get_rank (feature);
if (rank &lt; GST_RANK_MARGINAL)
return FALSE;
return TRUE;
}
/*
* This function is called to sort features by rank.
*/
static gint
cb_compare_ranks (GstPluginFeature *f1,
GstPluginFeature *f2)
{
return gst_plugin_feature_get_rank (f2) - gst_plugin_feature_get_rank (f1);
}
static void
init_factories (void)
{
/* first filter out the interesting element factories */
factories = gst_registry_feature_filter (
gst_registry_get_default (),
(GstPluginFeatureFilter) cb_feature_filter, FALSE, NULL);
/* sort them according to their ranks */
factories = g_list_sort (factories, (GCompareFunc) cb_compare_ranks);
}
<!-- example-end dynamic.c b --></programlisting>
<para>
From this list of element factories, we will select the one that most
likely will help us decoding a media stream to a given output type.
For each newly created element, we will again try to autoplug new
elements to its source pad(s). Also, if the element has dynamic pads
(which we've seen before in <xref linkend="section-pads-dynamic"/>),
we will listen for newly created source pads and handle those, too.
The following code replaces the <function>cb_type_found</function>
from the previous section with a function to initiate autoplugging,
which will continue with the above approach.
</para>
<programlisting><!-- example-begin dynamic.c c -->
static void try_to_plug (GstPad *pad, const GstCaps *caps);
static GstElement *audiosink;
static void
cb_newpad (GstElement *element,
GstPad *pad,
gpointer data)
{
GstCaps *caps;
caps = gst_pad_get_caps (pad);
try_to_plug (pad, caps);
gst_caps_unref (caps);
}
static void
close_link (GstPad *srcpad,
GstElement *sinkelement,
const gchar *padname,
const GList *templlist)
{
GstPad *pad;
gboolean has_dynamic_pads = FALSE;
g_print ("Plugging pad %s:%s to newly created %s:%s\n",
gst_object_get_name (GST_OBJECT (gst_pad_get_parent (srcpad))),
gst_pad_get_name (srcpad),
gst_object_get_name (GST_OBJECT (sinkelement)), padname);
/* add the element to the pipeline and set correct state */
if (sinkelement != audiosink) {
gst_bin_add (GST_BIN (pipeline), sinkelement);
gst_element_set_state (sinkelement, GST_STATE_READY);
}
pad = gst_element_get_static_pad (sinkelement, padname);
gst_pad_link (srcpad, pad);
if (sinkelement != audiosink) {
gst_element_set_state (sinkelement, GST_STATE_PAUSED);
}
gst_object_unref (GST_OBJECT (pad));
/* if we have static source pads, link those. If we have dynamic
* source pads, listen for pad-added signals on the element */
for ( ; templlist != NULL; templlist = templlist->next) {
GstStaticPadTemplate *templ = templlist->data;
/* only sourcepads, no request pads */
if (templ->direction != GST_PAD_SRC ||
templ->presence == GST_PAD_REQUEST) {
continue;
}
switch (templ->presence) {
case GST_PAD_ALWAYS: {
GstPad *pad = gst_element_get_static_pad (sinkelement, templ->name_template);
GstCaps *caps = gst_pad_get_caps (pad);
/* link */
try_to_plug (pad, caps);
gst_object_unref (GST_OBJECT (pad));
gst_caps_unref (caps);
break;
}
case GST_PAD_SOMETIMES:
has_dynamic_pads = TRUE;
break;
default:
break;
}
}
/* listen for newly created pads if this element supports that */
if (has_dynamic_pads) {
g_signal_connect (sinkelement, "pad-added", G_CALLBACK (cb_newpad), NULL);
}
}
static void
try_to_plug (GstPad *pad,
const GstCaps *caps)
{
GstObject *parent = GST_OBJECT (GST_OBJECT_PARENT (pad));
const gchar *mime;
const GList *item;
GstCaps *res, *audiocaps;
/* don't plug if we're already plugged - FIXME: memleak for pad */
if (GST_PAD_IS_LINKED (gst_element_get_static_pad (audiosink, "sink"))) {
g_print ("Omitting link for pad %s:%s because we're already linked\n",
GST_OBJECT_NAME (parent), GST_OBJECT_NAME (pad));
return;
}
/* as said above, we only try to plug audio... Omit video */
mime = gst_structure_get_name (gst_caps_get_structure (caps, 0));
if (g_strrstr (mime, "video")) {
g_print ("Omitting link for pad %s:%s because mimetype %s is non-audio\n",
GST_OBJECT_NAME (parent), GST_OBJECT_NAME (pad), mime);
return;
}
/* can it link to the audiopad? */
audiocaps = gst_pad_get_caps (gst_element_get_static_pad (audiosink, "sink"));
res = gst_caps_intersect (caps, audiocaps);
if (res &amp;&amp; !gst_caps_is_empty (res)) {
g_print ("Found pad to link to audiosink - plugging is now done\n");
close_link (pad, audiosink, "sink", NULL);
gst_caps_unref (audiocaps);
gst_caps_unref (res);
return;
}
gst_caps_unref (audiocaps);
gst_caps_unref (res);
/* try to plug from our list */
for (item = factories; item != NULL; item = item->next) {
GstElementFactory *factory = GST_ELEMENT_FACTORY (item->data);
const GList *pads;
for (pads = gst_element_factory_get_static_pad_templates (factory);
pads != NULL; pads = pads->next) {
GstStaticPadTemplate *templ = pads->data;
/* find the sink template - need an always pad*/
if (templ->direction != GST_PAD_SINK ||
templ->presence != GST_PAD_ALWAYS) {
continue;
}
/* can it link? */
res = gst_caps_intersect (caps,
gst_static_caps_get (&amp;templ->static_caps));
if (res &amp;&amp; !gst_caps_is_empty (res)) {
GstElement *element;
gchar *name_template = g_strdup (templ->name_template);
/* close link and return */
gst_caps_unref (res);
element = gst_element_factory_create (factory, NULL);
close_link (pad, element, name_template,
gst_element_factory_get_static_pad_templates (factory));
g_free (name_template);
return;
}
gst_caps_unref (res);
/* we only check one sink template per factory, so move on to the
* next factory now */
break;
}
}
/* if we get here, no item was found */
g_print ("No compatible pad found to decode %s on %s:%s\n",
mime, GST_OBJECT_NAME (parent), GST_OBJECT_NAME (pad));
}
static void
cb_typefound (GstElement *typefind,
guint probability,
GstCaps *caps,
gpointer data)
{
gchar *s;
GstPad *pad;
s = gst_caps_to_string (caps);
g_print ("Detected media type %s\n", s);
g_free (s);
/* actually plug now */
pad = gst_element_get_static_pad (typefind, "src");
try_to_plug (pad, caps);
gst_object_unref (GST_OBJECT (pad));
}
<!-- example-end dynamic.c c --></programlisting>
<para>
By doing all this, we will be able to make a simple autoplugger that
can automatically setup a pipeline for any media type. In the example
above, we did this for audio only. However, we can also do this
for video to create a player that plays both audio and video.
</para>
<!-- example-begin dynamic.c d --><!--
static gboolean
my_bus_callback (GstBus *bus,
GstMessage *message,
gpointer data)
{
GMainLoop *loop = data;
switch (GST_MESSAGE_TYPE (message)) {
case GST_MESSAGE_ERROR: {
GError *err;
gchar *debug;
gst_message_parse_error (message, &amp;err, &amp;debug);
g_print ("Error: %s\n", err-&gt;message);
g_error_free (err);
g_free (debug);
g_main_loop_quit (loop);
break;
}
case GST_MESSAGE_EOS:
/* end-of-stream */
g_main_loop_quit (loop);
break;
default:
break;
}
/* remove from queue */
return TRUE;
}
gint
main (gint argc,
gchar *argv[])
{
GMainLoop *loop;
GstElement *typefind, *realsink;
GstBus *bus;
GError *err = NULL;
gchar *p;
/* init GStreamer and ourselves */
gst_init (&amp;argc, &amp;argv);
loop = g_main_loop_new (NULL, FALSE);
init_factories ();
/* args */
if (argc != 2) {
g_print ("Usage: %s &lt;filename&gt;\n", argv[0]);
return -1;
}
/* pipeline */
p = g_strdup_printf ("filesrc location=\"%s\" ! typefind name=tf", argv[1]);
pipeline = gst_parse_launch (p, &amp;err);
g_free (p);
if (err) {
g_error ("Could not construct pipeline: %s", err-&gt;message);
g_error_free (err);
return -1;
}
bus = gst_pipeline_get_bus (GST_PIPELINE (pipeline));
gst_bus_add_watch (bus, my_bus_callback, NULL);
gst_object_unref (bus);
typefind = gst_bin_get_by_name (GST_BIN (pipeline), "tf");
g_signal_connect (typefind, "have-type", G_CALLBACK (cb_typefound), NULL);
gst_object_unref (GST_OBJECT (typefind));
audiosink = gst_element_factory_make ("audioconvert", "aconv");
realsink = gst_element_factory_make ("alsasink", "audiosink");
gst_bin_add_many (GST_BIN (pipeline), audiosink, realsink, NULL);
gst_element_link (audiosink, realsink);
gst_element_set_state (pipeline, GST_STATE_PLAYING);
/* run */
g_main_loop_run (loop);
/* exit */
gst_element_set_state (pipeline, GST_STATE_NULL);
gst_object_unref (GST_OBJECT (pipeline));
return 0;
}
--><!-- example-end dynamic.c d -->
<para>
The example above is a good first try for an autoplugger. Next steps
would be to listen for <quote>pad-removed</quote> signals, so we
can dynamically change the plugged pipeline if the stream changes
(this happens for DVB or Ogg radio). Also, you might want special-case
code for input with known content (such as a DVD or an audio-CD),
and much, much more. Moreover, you'll want many checks to prevent
infinite loops during autoplugging, maybe you'll want to implement
shortest-path-finding to make sure the most optimal pipeline is chosen,
and so on. Basically, the features that you implement in an autoplugger
depend on what you want to use it for. For full-blown implementations,
see the <quote>playbin</quote> and <quote>decodebin</quote> elements in
<xref linkend="chapter-components"/>.
</para>
</sect1>
</chapter>