[gst-devel] GStreamer summary, 16 December 2005
Andy Wingo
wingo at pobox.com
Fri Dec 16 08:51:04 CET 2005
Mwa pendukeni oohacker yo GStreamer,
I know it's been a long time since I rapped at ya, but we had some very
crucial details to take care of, such as releasing a stable GStreamer.
So amigos let's see what's been up over the past three weeks already.
A to the P to the I
-------------------
We are most totally and definitely stable! December 5 saw the 0.10.0
release of all modules, from core on out to python and ffmpeg. Check out
the hyped-up release notes [0] if you haven't already.
As the last edition of your favorite GStreamer news product was released
two weeks before 0.10.0, there were still a number of lines of ChangeLog
that made it into the core before we went stable. And that number is
1360.
There weren't all that many features added. The identity element got a
new property, "single-segment", that will cause it to re-stamp the
buffers passing through it, so that e.g. a looping segment can appear to
downstream elements as one continuous stream. This is especially useful
if you want to encode a stream whose source is a looping file. Wim got a
bit bored and implemented high-precision scaling of 64-bit integers.
GstBaseTransform got an event vmethod implemented, and Jan Schmidt added
seek support to fdsrc. Stefan added code to GstController so it could
control enums, and Edward Hervey made the gst-launch output slightly
less cryptic.
API-wise, functions in gstvalue.h were changed to use GLib-style types
only, e.g. gchar instead of char. Yes they are the same thing, but such
is our attention to consistency that we can't bear seeing non-GLib types
in the API. gst_pad_alloc_buffer was changed so it never sets caps, and
a new function was added: gst_pad_alloc_buffer_and_set_caps. I bet you
can guess what it does.
Do take some time today to mourn the removal of bash completion support
for gst-launch. gst-complete was a noble friend, but unmaintained, and
required its own XML cache to be kept up-to-date via running the
gst-compprep utility. A patch to port gst-complete to 0.10 would
certainly be accepted, if it can work without gst-compprep, for the same
reasons that we removed gst-register.
As a source package, GStreamer core itself saw a lot of loving from its
favorite consort, Thomas Vander Stichele. Directories moved around,
--disable-gst-debug and related options now work again, the queue
element is out of the core (finally), and the normal
dist/distcheck/visual studio fixes were applied. Aside from all that,
object padding was updated, leaks were plugged, and there were
documentation updates all around.
Also, welcome to our new translations: Bulgarian and Traditional
Chinese. Dobre doshli! Huan yin!
[0] http://gstreamer.freedesktop.org/documentation/gstreamer010.html
Post Oh Ten Oh
--------------
After a short continuation of the CVS freeze, in case a new release
needed to be rolled out quickly, the core opened again for commits. One
of the first was an ABI test that makes sure our structures do not
change in size. gstparse handling of the { and } characters (which make
GstThread bins) was removed, as GstThread is no more, and Wim fixed the
"sync" fakesrc property. Obsolete configure.ac checks for makecontext
were removed, and Mike Smith fixed a leak in the typefind element. Tim
spent some time updating the manual for 0.10, and Stefan ported some of
the old examples to the current API.
BaseSrc finally got support for seeks, position, and queries in any
format -- before, it only provided hooks for byte positions. The net
time provider can be deactivated, in which state it will not reply to
time messages. The typefind element was bugfixed some more, and Julien
fixed threadsafety bugs in GstCollectPads.
Other than that there wasn't anything to speak of -- a relaxed week and
a half. This is how it should be, minor bugfixes as we concentrate on
making applications users will enjoy using.
Seven Noses, Seven Cheeses
--------------------------
Leading up to the 0.10.0 releases, there was a lot of last-minute
polish being applied to the plugins. Here's the changes between 23
November and 5 December:
* base: multifdsink threadsafety fixes, oggdemux made more robust,
audiorate respects flow returns, ximagesink, xvimagesink, and the
xoverlay interface polished and documented by video hero Julien,
decodebin got some queues, audio clock supports operating in
master/slave mode, playbin polishes and support for multiple audio
streams, rtp plugin names made more consistent, xoverlay stress test
added, pango plugin ported nicely and moved from -good, audioresample
fixed (finally!)
* good: navigation, stride, and passthrough fixes in videoflip, quarktv
crasher fix, documentation added for autovideo, audioaudio, and
flacdec added, dvdec fixes for pixel aspect-ratio, matroska muxer
updates, video filters use a base class from plugins-base, event
support added to the navseek and progressreport debug plugins,
multipart muxer and demuxer ported, dvdemux seeking fixes, rtp plugin
renaming.
* ugly: mpegstream work, mad gracefully handles unlinked source pads,
a52dec supports streams with special DVD headers, real media demuxer
EOS fix, id3tag removed.
* bad: wavpack ported (no correction file support yet though), qtdemux
state change fixes, faad robustness fix, libmms plugin ported to 0.10
by Edgard
After the initial rest that everyone took after 0.10.0, the bug reports
started to pour in. Since then there have been a number of fixes that
will come out in our next release:
* base: ogm support added to the ogg plugin, playbin fixes for glib 2.6,
gst-launch-ext removed (it wasn't working and isn't useful), bugfixes
in the subtitle parser, cdparanoia ported to 0.10 (made possible by
basesrc changes), videorate bugfixes, plugged massive leak in
audioresample, audiotestsrc and videotestsrc seeking support,
GstPropertyProbe support added back to the alsa elements.
* good: Memleak and crasher fixes in wavparse and avidemux, docs added
to multipart muxer/demuxer, flacdec bugfixes, OSS portability fixes,
flxdec endianness fix, fix totally ironic crash in the electric fence
debugging element, auparse beautification and crasher fix,
mastroskademux timestamping fix, videomixer capsnego fixed, RTP
payloader/depayloader caps fixes, speex rtp elements set clock rate
properly, rtsp portability fix.
* ugly: more mpegstream work, mad query/position fixes, plug leak in
rmdemux
* bad: qtdemux EOS and 64-bit fixes, swfdec ported to 0.9 by Edgard,
mmsh support in libmms, dtsdec, xviddec and xvidenc ported to 0.10
also by Edgard.
Introducing the Media Test Suite
--------------------------------
Edward Hervey recently gave birth to a beautiful HTML baby. Tell us
about your baby, Edward!
Fixing bugs in GStreamer plugins has always been a tedious task.
Sometimes fixing playback for one file breaks it for other files.
Testing files playback/seeking by *hand* is repetitive and no
developer likes it. And finally, nobody (apart maybe for the main
developer of a plugin) ever knows at one given point how well some
media formats are supported, or how well the API is implemented in
each plugins.
In order to solve all these issues, a new testsuite is born :
gst-media-test. It currently contains two tests : typefindtest (Can
we read this file and all the media streams contained within ?) and
playtest (does playback/seeking work properly?). The tests check
that complete and correct API-compliance is respected for every
file. HTML reports of the tests are then generated, with complete
debug logs, exact reasons why the test failed, backtraces if the
test crashed, plugins used, etc...
We currently run the testsuite about everyday on over 20Gbytes of
test files with the latest cvs version of GStreamer and the results
are available online [1]. Pick your bug and fix it :)
All buggy media files that land in the GStreamer bugzilla should
also be added to the list of files tested here, so users can also
see if the current cvs version of GStreamer solved their issue.
[1] http://core.fluendo.com/gst-media-test/
We're all impressed, Edward. Nice work!
Um... So Like What Are We Doing And Stuff
-----------------------------------------
It seems we've lost our direction, inasmuch as we spent the last 10
months looking to the day of 0.10.0 and now it has passed. What now?
So your editor sent a clandestine reporter to the GStreamer IRC channel,
asking people that same question. Here's what they said:
OK so they didn't say anything, so given the pressure of the story's
deadline our clandestine reporter just left the editor with a link to a
mailing list discussion [2] started by Thomas.
[2]
http://news.gmane.org/find-root.php?group=gmane.comp.video.gstreamer.devel&article=14432
114 Gs Down
-----------
Thomas removed a bunch of outdated examples, which combined with his
previous thefts is probably enough to buy a large plasma TV. Have him
buy your drinks at the bar!
Core files changed since the last summary : 266
Lines added since the last summary : 9303
Lines removed : 7860
Net change in core value according to sloccount(1) : -$114,303
Total number of GStreamer bugs as of today : 228
Number opened in the last week : 12
Number closed in the last week : 2
Net change : +10
Rank among most buggy projects hosted in gnome bugzilla : Tied for 8
Our bug count is shooting back up. Your editor is waiting for the new
bugzilla to be installed before going back to looking at bugs. Of course
that means that after Monday, there will be no excuse.
Current Releases
----------------
No more 0.8, no more 0.9, 0.10 is the new stable series!
Latest stable versions:
gstreamer: 0.10.0
gst-plugins-base: 0.10.0
gst-plugins-good: 0.10.0
gst-plugins-ugly: 0.10.0
gst-plugins-bad: 0.10.0
gst-ffmpeg: 0.10.0
gst-python: 0.10.0
Featured Application: bmpX
--------------------------
Christian Schaller brings us this week's featured application.
This weeks featured application is bmpX, a music player with a
Winamp
and XMMS compatible user interface. The bmp project started its life
as a GTK+ 2 port of XMMS, but is today a fully separate project.
bmpX CVS has working GStreamer 0.10 support and will soon bring
bliss to the world among those who never fell in love with
iTunes-style user interfaces. A big thanks to Milosz Derezynski and
the rest of the bmpX team for this cool project.
You find more information about this project at:
http://beep-media-player.org/index.php/BMPx_Homepage
Well that brings this most lengthy newsletter to a close. Everyone
should pull out those dusty personal projects, update them for the new
API, and give the world a nice present for the new year. Word is bond.
Peace.
--
Andy Wingo
http://wingolog.org/
More information about the gstreamer-devel
mailing list