[gst-devel] GStreamer status, 13 October 2005

[Andy Wingo]

Goeiemorgen GStreamer hackers,


Another day, another dollar, another 1000 lines of ChangeLog. You are
tuned into Radio GStreamer, operating at a frequency of 0.131 summaries
per day, which, when the DJs are not sipping piña coladas and letting
the
radio software play some chilled-out dub, is all about some crucial
GStreamer news. Represent.


The Coming Winter
-----------------

As our plan for 0.10 [0] indicates, the frost is coming. After Monday 17
October, GStreamer will enter an API freeze. Much of the frenetic
hacking this last week has focused on dealing with the things we
normally ignore, like dirt under the fridge.

Command-line argument parsing is now done with GOption instead of the
byzantine relic known as popt. The GST_FLAG_IS_SET macro and friends
were renamed to GST_OBJECT_FLAG_*, and the flags themselves are now
declared as flags instead of enums. gst_element_get_state now takes a
GstClockTime instead of a GTimeVal for the timeout, the newsegment
events got a new boolean flag only to be used by Wim Taymans, and
iterators now declare which GType they are iterating over so as to be
more bindings-friendly. Clock distribution got the Taymans treatment,
including a new provide_clock vmethod in GstElement. Tim Müller sped up
typefinding while fixing some bugs. Also, there's a new type of
messages, TYPE_ELEMENT, which is designed for element-specific
notifications. GStreamer will not post TYPE_APPLICATION messages
anymore; those are reserved for you.

But by far the largest change of the past week was Wim's state change
patch. Before this week, applications had to keep a rather complicated
model of GStreamer's state in order to deal with all possible pipeline
behaviors. Setting a pipeline to PLAYING could block or not depending on
an object property on the GstPipeline object; it could succeed all the
way up to PAUSED but not try to go to PLAYING in some cases; error
conditions were poorly defined; sometimes messages would be posted and
sometimes not. It was more complex than it should be.

The core of the patch was rather simple. The state lock was made into a
recursive mutex that can be waited on. Setting the state on a bin simply
sets the state on all elements and returns even if some elements are
changing state asynchronously, instead of blocking like GstPipeline used
to do. Elements record the state that they are trying to go to; that is,
if you set an element to PLAYING that is in READY, its _current_ state
will be READY, the _next_ state is PAUSED, but the _pending_ state is
PLAYING. Then when it commits its state either by returning SUCCESS from
the change_state function or manually in the case of ASYNC elements
(sinks mainly), it pulls itself up recursively into the _pending_ state
(PLAYING in this case).

After the patch, things are much more simple. Setting a pipeline to a
given state will either result in success or error. Messages will always
be posted. There are a couple of bugs still but we are confident they
will be ironed out soon. That or we send Wim to go work on MPlayer.

[0] http://article.gmane.org/gmane.comp.video.gstreamer.devel/13768


Plugin Activity
---------------

Our wire services indicate the following changes in the plugins modules:

  base: multifdsink bugfixes, audiosink bugfixes, typefind leaks
        plugged, adder querying added, addition of a most rocking
        audiotestsrc by Stefan Kost, tcpserversrc state change fix,
        unsigned audio in alsaink, playbin seeking fix

  good: cairo timeoverlay ported to cairo 1.0 API, speexenc ported by
        Edgard Lima, error handling in oss elements, dv1394src made
        nonblocking and interruptible (and it posts a message when the
        cable is plugged/unplugged), many debugging plugins ported by
        Tim Müller (progressreport, navseek, navigationtest, testsink,
        breakmydata)

   bad: faac bugfixes

  ugly: mad bugfix, amrnbdec bugfix


Build Status
------------

We are pleased to announce that Thomas Vander Stichele now the proud
godfather of a new buildslave running FreeBSD, thanks to Koop Mast for
naming Thomas to this honored position. Also the Fedora Core 3
buildslaves have been removed, as they do not support GLib 2.6, and the
OS X buildslaves have been reenabled thanks to Stephen Thorne.

For a while now we have also had a monster of a Power 5 machine running
builds from Augsburg, Germany. /proc/cpuinfo says it has 32
processors!!!!! Thanks to Thomas Morpe and the debian-ppc folks over at
http://tuxppc.rz.uni-augsburg.de/ for the use of this machine.


Vital statistics
----------------

Core files changed this week                            : 166
Lines added this week                                   : 10082
Lines removed                                           : 2947
Net change in core value according to sloccount(1)      : +$88,294
Total number of GStreamer bugs as of today              : 209 [1]
Number opened in the last week                          : 23
Number closed in the last week                          : 20
Net change                                              : -3
Rank among most buggy projects hosted in gnome bugzilla : 8

[1] Not counting enhancements; this discounts the
divine-beings-are-munging-our-bug-counts theory, but leaves open the
possibility of malicious gremlins.


Current Releases
----------------

No new tarballs were dropped on the world this week. Monday, however,
will see the release into the wild of the 0.9.4 development versions.

Latest development versions:
    gstreamer:        0.9.3
    gst-plugins-base: 0.9.3
    gst-plugins-good: 0.9.3
    gst-plugins-ugly: 0.9.3
    gst-plugins-bad:  0.9.3
    gst-ffmpeg:       0.9.3
    gst-python:       0.9.3

Latest stable versions:
    gstreamer:        0.8.11
    gst-plugins:      0.8.11
    gst-ffmpeg:       0.8.6
    gst-python:       0.8.2


Featured Application
--------------------

Christian Schaller, our foreign correspondent in Norwegia, was too busy
researching a hot lead to send us a featured application this week. Stay
tuned next week for his report.


That brings this piece to a close. My fellow DJ's appear to already be
in dub-and-piña-colada mode, so I am taking that as a sign from the most
high. From our broadcasting office in Barcelona this is Radio GStreamer
signing off. Tune in again in 1/0.131 days!
-- 
Andy Wingo
http://wingolog.org/