Original commit message from CVS:
- Documentation updates, renamed some _ methods because gtkdoc didn't
want to generate docs for them.
- Add some more events for future use.
Original commit message from CVS:
Make sizeof(GstBuffer) independent of config.h and HAVE_ATOMIC_H. This still
needs a more fundamental solution though.
Original commit message from CVS:
- Removed bufferpool code and move that to gstbuffer.c
- implemented refcounting on GstData
- implemented new buffer code based on Company's work in the EVENTS2 branch
- added boxed types for GstData/GstEvent/GstBuffer/GstBufferPool
- added refcounting to bufferpools and events
- use lockfree allocation for buffers
- simplified the clock, use lockfree allocation
- use GQueue in GstQueue for faster access to the tail element
- update core plugins to the new event API
Original commit message from CVS:
- Added a bunch of fast atomic operations on x86 with C fallbacks
- Added the code for refcounting in GstData
- Added a gstmemchunk allocator using atomic operations
Original commit message from CVS:
- Removed unused locking from the cothreads
- use G_*_DECLS in .h files
- remove gstlog.h include from gstobject.h, add to .c files
- removed unused refcounting code from gstobject
- small fixes in #includes
- Added Scheduling policy and priority properties to gstthread so that
SCHED_FIFO and SCHED_RR threads can be constructed.
Original commit message from CVS:
if you can't write to the registry (ie it is root-owned and you are user) and it
is out of date, just ignore it; plugin state will get written to the user reg
Original commit message from CVS:
more protective programming
cothread 0 cleans up higher cothreads if they're still around
use a magic number to protect against mem corruption in the struct
flames welcome
Original commit message from CVS:
ok, my latest added error checking and stuff
this reverts uraeus's patch but should tell us what goes wrong with munmap
Original commit message from CVS:
Fixes suggested by thomasvs to get rhythmbox working with basic threads
- int res;
+ int res = 0;
- res = munmap ((void *) thread, COTHREAD_STACKSIZE);
+/* res = munmap ((void *) thread, COTHREAD_STACKSIZE); */
Original commit message from CVS:
some debug info
remove the preventive munmap
reinstate the other munmap for cothread stocks which really should be there IMO
Original commit message from CVS:
ok, this seems to be the right fix for the basic scheduler.
I'd appreciate it if :
a) someone looks over these minimal changes and tells me if they're done
in the right way (especially the alignment argument ;))
b) others run lots of pipes with basic and tell me if they still have
issues
If all goes well I want to revert to the basic scheduler and do a release
Original commit message from CVS:
Don't seek past the end of the file
Fix a bug where the mapend was incorrectly calculated when we read past
the end of the file.
Original commit message from CVS:
Added a runtime option to use a dummy threading implementation that uses
NOPs for all synchronisation and threading operations.
Original commit message from CVS:
remove the "iterate_started" signal and replace it with pre_iterate and post_iterate callbacks.
Apps can now put a lock around an iterate so they can do stuff on the bin from another thread. much nicer than the signal approach.
Original commit message from CVS:
Whan the buffer is filled, the buffer is kept and added to the queue
when there is room again. Don't do this after a flush.
Original commit message from CVS:
put upstream events in an async queue and resend them at the start of the chain. not working as expected yet but a start
Original commit message from CVS:
Add an event function so that seek events are not forwarded when we
have a cache, seeking in the cache still needs to be implemented though..
Original commit message from CVS:
a couple of fixes here:
- forward eos events to all src pads, even if spider didn't connect them. this ensures that the pipeline will stay in a consistant state even when not all pads are sent data
- don't check for nonnull spider until it is actually needed. it really is null when it has been disposed and this method is called.
Original commit message from CVS:
added a "iterate_started" signal which fires before the iterate of a bin. useful for non-threadsafe operations like seek in threaded pipelines
Original commit message from CVS:
add an optional timeout for when queue blocks. when timeout is reached a GST_EVENT_FILLER is sent downstream
This is needed for gst-player since videosink has to be in the same thread as the ui, and spider blocks when there is no video (thus freezing the ui).
Original commit message from CVS:
only use ranked element factories for autoplugging. This is the only commit that needs to be backed out to reinstate previous behaviour.
Before your spider will work again, you will have to update/rebuild plugins and run gst-register.
Original commit message from CVS:
Created an overridable method for seeking on an element, along with a default
implementation.
make queue flush on discont (not entirely correct)
removed the new_media from the discont event.
Original commit message from CVS:
* fixups in the prop view/controller
* compilation fixes in the player
* add gst-editor to gst-all
* fixes to adder to comply with new osssink sync issues
* alsa fixes, although still 100% cpu is used, yum
* reenable locking of threaded elements, seems to work fine here
* fix a makefile in examples/plugins
Original commit message from CVS:
- The clock_wait now returns the difference between requested time and
unlock time.
- Misc defines like GST_SECOND in gstclock.h
- remove pre/post in gstelement.c until fixed.
- added release_locks to gstelement so that the element can unlock itself
- added some more predefined events.
- added folowing functions to gstpad:
- convert function: get the relation between formats on this pad
- query function: get stats about the pad (position/total/latency)
- internal connect function: find out how this pad connects to other
pad internally to the element.
- generic pad_dispatcher.
- removed the last bits of pullregion
- use release_locks on the queue.
- added some events to queue
- make gstthread use the new release_locks function
- make the scheduler use the new clock_wait functions
- added events to fakesink
- added query functions to filesrc
- swap type and offset in the bytestream seek API to match fseek
- added some event handling in bytestream.
Original commit message from CVS:
* implemented threadsafe property set/get system as discussed in
docs/random/wingo/threadsafe-properties
* some cleanups
* this change will cause binary incompatibilities, better rebuild them plugins
now, off to drink :-)
Original commit message from CVS:
* add gerror to gstplugin.c -- not fully propagated to periphery apis
* fix recursive dir creation in the xml registry
* comment identity props
* fix dependency resolution in lib loading
Original commit message from CVS:
* fix refcounting tests so that they compile and run, but they fail currently:
gst leaks obscene amounts of memory ;)
* fix plugin loading test so that it only refers to plugins within the gstreamer/
tree
* store gst plugin paths in the registry
* is GST_REGISTRY is set, only use the user registry with the PLUGIN_PATH explictly
specified by the user
* all tests should pass now except refcounting
Original commit message from CVS:
* added a get_perms_func to gstxmlregistry that will set _WRITABLE and _READABLE
as appropriate
* added an object property for location so that we can do some cleanup and initialization
Original commit message from CVS:
* GST_PLUGIN_PATH gets split into the user registry
* some debugging output in registry rebuilding
* don't go into =build, =inst, etc
* i really don't know what the current idiom is for the plugin test suites, disabling for now
still pending issues: what to do when other plugin paths are passed on the command
line for existing registries. if the existing registries were built against those
paths, the time checks work, but if not they will need to be rebuilt. i have a feeling
they should be rebuilt in any case, but it's a tricky issue.
Original commit message from CVS:
Check to see if the registry is out of date, and if so try to rebuild it.
WIERD THING: trying to run gst-register using system() fails. Apparently child processes
can't link to libs parent processes use? It's wierd, and it could be a libtool issue.
That's probably just as well, it would be better just to use -register's code from within
core and not rely on another binary.
Original commit message from CVS:
Totally rewritten registry handling.
- move the registry save/load code into a gstregistry subclass, this
will make it possible to use other registries (flat file, web based,
RDBMS type, etc..)
- a simple GMarkup xml registry is implemented
- use standard statically linked plugins for core elements.
- GstPlugin has a very well defined set of functions now
A little bytestream hack..
Added more info to -inspect.
Some more debugging info for clocking.
Small cleanups
I use ./gst-register --gst-plugin-path=/opt/src/sourceforge/gst-plugins/gst-libs:/opt/src/sourceforge/gst-plugins/
to register core and gst-plugins now.
Original commit message from CVS:
merged schedulers
- new file cothreads_compat.h to provide linking to desired cothreads package
- changes in basic and fast scheduler to use cothreads_compat.h
- updated Makefile to build the basic and the fast scheduler from both packages
- removed gststandardscheduler.c - it is not needed anymore
Available schedulers are now 'basic', 'standard', 'fastbasic' and 'faststandard' where the basic ones are built with old cothreads and the standard ones with the new cothreads.
Original commit message from CVS:
* gst_scheduler_factory_make () can accept NULL as the first arg now
* _create () sets the element sched and refcounts the scheduler
* corresponding touchups to gstpipeline and gstthread
Original commit message from CVS:
Header file fix (these ## are needed for multi-arguments) and a fix for gst_props_get() if safe = FALSE (we still need three arguments per fetch)
