gstreamer/docs/manual/appendix-porting.xml

313 lines
13 KiB
XML
Raw Normal View History

<chapter id="chapter-porting">
<title>Porting 0.8 applications to 0.10</title>
<para>
This section of the appendix will discuss shortly what changes to
applications will be needed to quickly and conveniently port most
applications from &GStreamer;-0.8 to &GStreamer;-0.10, with references
to the relevant sections in this Application Development Manual
where needed. With this list, it should be possible to port simple
applications to &GStreamer;-0.10 in less than a day.
</para>
<sect1 id="section-porting-objects">
<title>List of changes</title>
<itemizedlist>
<listitem>
<para>
Most functions returning an object or an object property have
been changed to return its own reference rather than a constant
reference of the one owned by the object itself. The reason for
this change is primarily thread safety. This means, effectively,
that return values of functions such as
<function>gst_element_get_pad ()</function>,
<function>gst_pad_get_name ()</function> and many more like these
have to be free'ed or unreferenced after use. Check the API
references of each function to know for sure whether return
values should be free'ed or not. It is important that all objects
derived from GstObject are ref'ed/unref'ed using gst_object_ref()
and gst_object_unref() respectively (instead of g_object_ref/unref).
</para>
</listitem>
<listitem>
<para>
Applications should no longer use signal handlers to be notified
of errors, end-of-stream and other similar pipeline events.
Instead, they should use the <classname>GstBus</classname>, which
has been discussed in <xref linkend="chapter-bus"/>. The bus will
take care that the messages will be delivered in the context of a
main loop, which is almost certainly the application's main thread.
The big advantage of this is that applications no longer need to
be thread-aware; they don't need to use <function>g_idle_add
()</function> in the signal handler and do the actual real work
in the idle-callback. &GStreamer; now does all that internally.
</para>
</listitem>
<listitem>
<para>
Related to this, <function>gst_bin_iterate ()</function> has been
removed. Pipelines will iterate in their own thread, and applications
can simply run a <classname>GMainLoop</classname> (or call the
mainloop of their UI toolkit, such as <function>gtk_main
()</function>).
</para>
</listitem>
<listitem>
<para>
State changes can be delayed (ASYNC). Due to the new fully threaded
nature of GStreamer-0.10, state changes are not always immediate,
in particular changes including the transition from READY to PAUSED
state. This means two things in the context of porting applications:
first of all, it is no longer always possible to do
<function>gst_element_set_state ()</function> and check for a return
value of GST_STATE_CHANGE_SUCCESS, as the state change might be
delayed (ASYNC) and the result will not be known until later. You
should still check for GST_STATE_CHANGE_FAILURE right away, it is
just no longer possible to assume that everything that is not SUCCESS
means failure. Secondly, state changes might not be immediate, so
your code needs to take that into account. You can wait for a state
change to complete if you use GST_CLOCK_TIME_NONE as timeout interval
with <function>gst_element_get_state ()</function>.
</para>
</listitem>
<listitem>
<para>
In 0.8, events and queries had to manually be sent to sinks in
pipelines (unless you were using playbin). This is no longer
the case in 0.10. In 0.10, queries and events can be sent to
toplevel pipelines, and the pipeline will do the dispatching
internally for you. This means less bookkeeping in your
application. For a short code example, see <xref
linkend="chapter-queryevents"/>. Related, seeking is now
threadsafe, and your video output will show the new video
position's frame while seeking, providing a better user
experience.
</para>
</listitem>
<listitem>
<para>
The <classname>GstThread</classname> object has been removed.
Applications can now simply put elements in a pipeline with
optionally some <quote>queue</quote> elements in between for
buffering, and &GStreamer; will take care of creating threads
internally. It is still possible to have parts of a pipeline
run in different threads than others, by using the
<quote>queue</quote> element. See <xref linkend="chapter-threads"/>
for details.
</para>
</listitem>
<listitem>
<para>
Filtered caps -> capsfilter element (the pipeline syntax for
gst-launch has not changed though).
</para>
</listitem>
<listitem>
<para>
libgstgconf-0.10.la does not exist. Use the
<quote>gconfvideosink</quote> and <quote>gconfaudiosink</quote>
elements instead, which will do live-updates and require no library
linking.
</para>
</listitem>
<listitem>
<para>
The <quote>new-pad</quote> and <quote>state-change</quote> signals on
<classname>GstElement</classname> were renamed to
<quote>pad-added</quote> and <quote>state-changed</quote>.
</para>
</listitem>
<listitem>
<para>
<function>gst_init_get_popt_table ()</function> has been removed
in favour of the new GOption command line option API that was
added to GLib 2.6. <function>gst_init_get_option_group ()</function>
is the new GOption-based equivalent to
<function>gst_init_get_ptop_table ()</function>.
</para>
</listitem>
</itemizedlist>
</sect1>
</chapter>
<chapter id="chapter-porting-1.0">
<title>Porting 0.10 applications to 1.0</title>
<para>This section outlines some of the changes necessary to port
applications from &GStreamer;-0.10 to &GStreamer;-1.0. For a
comprehensive and up-to-date list, see the separate <ulink
type="http"
url="http://cgit.freedesktop.org/gstreamer/gstreamer/plain/docs/random/porting-to-1.0.txt">
Porting to 1.0</ulink> document.
</para>
<para>
It should be possible to port simple applications to
&GStreamer;-1.0 in less than a day.
</para>
<sect1 id="section-porting-objects-1.0">
<title>List of changes</title>
<itemizedlist>
<listitem>
<para>
All deprecated methods were removed. Recompile against 0.10 with
2013-02-17 20:50:03 +00:00
GST_DISABLE_DEPRECATED defined (such as by adding
-DGST_DISABLE_DEPRECATED to the compiler flags) and fix issues
before attempting to port to 1.0.
</para>
</listitem>
<listitem>
<para>
"playbin2" has been renamed to "playbin", with similar API
</para>
</listitem>
<listitem>
<para>
"decodebin2" has been renamed to "decodebin", with similar API. Note
that there is no longer a "new-decoded-pad" signal, just use GstElement's
"pad-added" signal instead (but don't forget to remove the 'gboolean last'
argument from your old signal callback functino signature).
</para>
</listitem>
<listitem>
<para>
the names of some "formatted" pad templates has been changed from e.g.
"src%d" to "src%u" or "src_%u" or similar, since we don't want to see
negative numbers in pad names. This mostly affects applications that
create request pads from elements.
</para>
</listitem>
<listitem>
<para>
some elements that used to have a single dynamic source pad have a
source pad now. Example: wavparse, id3demux, iceydemux, apedemux.
(This does not affect applications using decodebin or playbin).
</para>
</listitem>
<listitem>
<para>
playbin now proxies the GstVideoOverlay (former GstXOverlay) interface,
so most applications can just remove the sync bus handler where they
would set the window ID, and instead just set the window ID on playbin
from the application thread before starting playback.
</para>
<para>
playbin also proxies the GstColorBalance and GstNavigation interfaces,
so applications that use this don't need to go fishing for elements
2013-02-17 20:50:03 +00:00
that may implement those any more, but can just use on playbin
unconditionally.
</para>
</listitem>
<listitem>
<para>
multifdsink, tcpclientsink, tcpclientsrc, tcpserversrc the protocol property
is removed, use gdppay and gdpdepay.
</para>
</listitem>
<listitem>
<para>
XML serialization was removed.
</para>
</listitem>
<listitem>
<para>
Probes and pad blocking was merged into new pad probes.
</para>
</listitem>
<listitem>
<para>
Position, duration and convert functions no longer use an inout parameter
for the destination format.
</para>
</listitem>
<listitem>
<para>
Video and audio caps were simplified. audio/x-raw-int and audio/x-raw-float
are now all under the audio/x-raw media type. Similarly, video/x-raw-rgb
and video/x-raw-yuv are now video/x-raw.
</para>
</listitem>
<listitem>
<para>
ffmpegcolorspace was removed and replaced with videoconvert.
</para>
</listitem>
<listitem>
<para>
GstMixerInterface / GstTunerInterface were removed without replacement.
</para>
</listitem>
<listitem>
<para>
The GstXOverlay interface was renamed to GstVideoOverlay, and now part
of the video library in gst-plugins-base, as the interfaces library
no longer exists.
</para>
<para>
The name of the GstXOverlay "prepare-xwindow-id" message has changed
to "prepare-window-handle" (and GstXOverlay has been renamed to
GstVideoOverlay). Code that checks for the string directly should be
changed to use gst_is_video_overlay_prepare_window_handle_message(message)
instead.
</para>
</listitem>
<listitem>
<para>
The GstPropertyProbe interface was removed. the is no replacement yet,
but a more featureful replacement for device discovery and feature
querying is planned, see https://bugzilla.gnome.org/show_bug.cgi?id=678402
</para>
</listitem>
<listitem>
<para>
gst_uri_handler_get_uri() and the get_uri vfunc now return a copy of
the URI string
</para>
<para>
gst_uri_handler_set_uri() and the set_uri vfunc now take an additional
GError argument so the handler can notify the caller why it didn't
accept a particular URI.
</para>
<para>
gst_uri_handler_set_uri() now checks if the protocol of the URI passed
is one of the protocols advertised by the uri handler, so set_uri vfunc
implementations no longer need to check that as well.
</para>
</listitem>
<listitem>
<para>
GstTagList is now an opaque mini object instead of being typedefed to a
GstStructure. While it was previously okay (and in some cases required because of
missing taglist API) to cast a GstTagList to a GstStructure or use
gst_structure_* API on taglists, you can no longer do that. Doing so will
cause crashes.
</para>
<para>
Also, tag lists are refcounted now, and can therefore not be freely
modified any longer. Make sure to call gst_tag_list_make_writable (taglist)
before adding, removing or changing tags in the taglist.
</para>
<para>
GST_TAG_IMAGE, GST_TAG_PREVIEW_IMAGE, GST_TAG_ATTACHMENT: many tags that
used to be of type GstBuffer are now of type GstSample (which is basically
a struct containing a buffer alongside caps and some other info).
</para>
</listitem>
<listitem>
<para>
GstController has now been merged into GstObject. It does not exists as an
individual object anymore. In addition core contains a GstControlSource base
class and the GstControlBinding. The actual control sources are in the controller
library as before. The 2nd big change is that control sources generate
a sequence of gdouble values and those are mapped to the property type and
value range by GstControlBindings.
</para>
<para>
The whole gst_controller_* API is gone and now available in simplified form
under gst_object_*. ControlSources are now attached via GstControlBinding
to properties. There are no GValue arguments used anymore when programming
control sources.
</para>
</listitem>
</itemizedlist>
</sect1>
</chapter>