Original commit message from CVS:
2005-03-07 Andy Wingo <wingo@pobox.com>
* check/gst/gstobject.c (test_fake_object_name): The object *does*
have a name after g_object_new.
* gst/base/gstbasesink.c: Change "arguments" to "properties" in
the code. Die GtkObject die!
(GstBaseSink::preroll-queue-len): New object property, the number
of buffers to queue in preroll. By default, does not queue any
buffers. Set to a higher number if you have a one-threaded
demuxer.
(gst_basesink_preroll_queue_push)
(gst_basesink_preroll_queue_empty)
(gst_basesink_preroll_queue_flush): Implement a queue of buffers
for preroll. All must be called with PREROLL_LOCK.
(gst_basesink_finish_preroll): Instead of always blocking, push
the buffer onto the queue. preroll_queue_push will block if the
queue is full.
(gst_basesink_event): Make sure the preroll queue is emptied on
eos and flushed on flush.
(gst_basesink_handle_buffer): Does the work of chain_unlocked, but
without going into finish_preroll.
(gst_basesink_change_state): Handle setting up the queue and
flushing it in READY<->PAUSED transitions.
Original commit message from CVS:
2005-03-03 Andy Wingo <wingo@pobox.com>
* gst/gstiterator.c (gst_iterator_find_custom)
(gst_iterator_foreach, gst_iterator_fold): Never free the
iterator.
* gst/elements/gsttee.c (gst_tee_handle_buffer):
* gst/gstutils.c (gst_pad_proxy_setcaps, gst_pad_proxy_getcaps):
Always free the iterator.
* check/gst/gstbus.c (pull_messages): Use public
gst_message_get_structure().
Original commit message from CVS:
2005-03-03 Andy Wingo <wingo@pobox.com>
* tools/gst-launch.c: Update to use the new message-parsing API.
* tools/gst-md5sum.c: Update to new spiffy poll API. Not that
md5sum is ported though!
* gst/gstmessage.c:
* gst/gstmessage.h (struct _GstMessage): Changed so the message
data is just a GstStructure, not a union.
(gst_message_get_structure): New function, gets the structure
backing the message. Owned by the message.
(gst_message_parse_tag, gst_message_parse_state_changed)
(gst_message_parse_error, gst_message_parse_warning): New
functions, parse the message into their components. Return values
are copies, and so must be freed by the caller.
Original commit message from CVS:
2005-03-01 Andy Wingo <wingo@pobox.com>
* tests/complexity.c: Adjust to lack of gst_bin_iterate, boolean
link return values.
* gst/elements/gsttee.c, gst/elements/gsttee.h: I am the master of
do-nothing plugins! Fear the dev-zero!
* gst/elements/gstelements.c
* gst/elements/Makefile.am: Add tee back to gstelements.
* gst/gstutils.h
* gst/gstutils.c (gst_pad_proxy_getcaps, gst_pad_proxy_setcaps):
Resurrect from the dead. Use gst_iterator_fold to be threadsafe.
* gst/gstiterator.h (GstIteratorFoldFunction): Return a bool.
* gst/gstiterator.c (gst_iterator_fold): Only continue folding as
long as the fold function returns TRUE. Add a bunch o docs.
(gst_iterator_foreach): Add docs about when the iterator will be
freed.
(gst_iterator_find_custom): Make more efficient because of the
bool-return-value thing.
* check/gst/gstiterator.c (add_fold_func): Adapt to new
fold-funcs-returning-bool policy.
* gst/gstutils.h:
* gst/gstutils.c (gst_element_link, gst_element_link_many)
(gst_element_link_filtered, gst_element_link_pads)
(gst_element_link_pads_filtered): It seems Wim changed the return
type to booleans internally. Assume he knows what he's doing and
change the prototypes as well.
Original commit message from CVS:
2005-03-01 Andy Wingo <wingo@pobox.com>
* gst/gstpipeline.c: Fix element details.
(gst_pipeline_set_property, gst_pipeline_get_property): Lock
around the whole get/set properties.
* gst/gstpad.c (gst_real_pad_set_property): Add a FIXME, the
::active property doesn't make sense any more.
(gst_pad_set_active): Check to see if the pad has the right
functions to be activated in this mode.
(gst_pad_event_default): Handle EOS specially, pausing the task on
the pad if necessary.
* gst/gstbin.c: Adapt callers of gst_iterator_foreach and
gst_iterator_filter to new argument order.
* gst/gstiterator.c (gst_iterator_find_custom)
(gst_iterator_foreach): Implement on top of gst_iterator_fold
instead of using the filter_next internals. A bit cleaner this
way.
* gst/gstiterator.h:
(gst_iterator_filter, gst_iterator_find_custom): Switch the
argument order so user_data is last.
(gst_iterator_foreach): Return the GstIteratorResult so the caller
knows if all elements were called, or if an error or resync
happened.
(gst_iterator_fold): New procedure.
* check/Makefile.am (TESTS):
* check/gst/gstiterator.c: New test suite for GstIterator. Checks
that iterating through a list hits all members in order, that
resync works correctly, and that fold works.
* gst/base/gstbasesink.c (gst_basesink_event): Fix Waymans bug.
Original commit message from CVS:
* gst/base/README:
* gst/base/gstbasesink.c: (gst_basesink_get_template),
(gst_basesink_base_init), (gst_basesink_pad_getcaps),
(gst_basesink_pad_setcaps), (gst_basesink_pad_buffer_alloc),
(gst_basesink_init), (gst_base_sink_get_template),
(gst_base_sink_get_caps), (gst_base_sink_set_caps),
(gst_base_sink_buffer_alloc), (gst_basesink_finish_preroll),
(gst_basesink_event), (gst_basesink_get_times),
(gst_basesink_do_sync), (gst_basesink_change_state):
Added small README for base classes.
Make default getcaps function in the base class return a
copy of the padtemplate caps if nothing else.
Original commit message from CVS:
* gst/elements/gstfakesrc.c: (gst_fakesrc_get_range_unlocked):
* gst/elements/gstidentity.c: (gst_identity_handle_buffer):
* gst/gstelement.c: (gst_element_get_state_func),
(gst_element_lost_state):
* gst/gstelement.h:
* gst/gstpipeline.c: (gst_pipeline_class_init),
(gst_pipeline_init), (gst_pipeline_set_property),
(gst_pipeline_get_property), (is_eos), (pipeline_bus_handler),
(gst_pipeline_change_state):
* gst/gstpipeline.h:
Convert timeout value in _get_state() to absolute time as
required by the _cond_wait().
Add _lost_state() to GstElement when elements in PAUSED loose
the preroll after a flush.
Add properties to GstPipeline, one to adjust clock times and
another to configure a timeout for a possibly blocking operation.
Removed some old methods.
Original commit message from CVS:
* gst/base/gstbasesink.c: (gst_basesink_get_template),
(gst_basesink_base_init), (gst_basesink_class_init),
(gst_basesink_pad_getcaps), (gst_basesink_pad_setcaps),
(gst_basesink_pad_buffer_alloc), (gst_basesink_init),
(gst_base_sink_get_template), (gst_base_sink_get_caps),
(gst_base_sink_set_caps), (gst_base_sink_buffer_alloc),
(gst_basesink_finish_preroll), (gst_basesink_event),
(gst_basesink_get_times), (gst_basesink_do_sync),
(gst_basesink_change_state):
* gst/base/gstbasesink.h:
* gst/gstiterator.c: (gst_iterator_init), (gst_iterator_new),
(gst_list_iterator_next), (gst_list_iterator_free),
(gst_iterator_new_list), (gst_iterator_pop), (gst_iterator_next),
(gst_iterator_push):
Only call the item function in the iterator if there is an item.
Add capsnego stuff and buffer_alloc to the basesink class.
Cleanups in the preroll code.
Original commit message from CVS:
Benjamin's mass_elements works now, but for other reasons. It's a lot faster than 0.8 :-)
2005-02-23 Andy Wingo <wingo@pobox.com>
* gst/elements/gstidentity.c (gst_identity_event): Pause the sink
task, if any, during EOS. Don't need to take the stream lock, we
already have it.
Original commit message from CVS:
2005-02-23 Andy Wingo <wingo@pobox.com>
* gst/gstmessage.h
* gst/gstmessage.c (_gst_message_free)
(gst_message_new_application): New message type, APPLICATION,
explicitly for use by applications.
* gst/gstbus.c (gst_bus_post): Only write a byte to the wakeup
socket if the queue is coming from an empty state. Fixes a bug
where posting a message could block, waiting for someone to read
out bytes from the socket.
(bus_watch_callback): Renamed from bus_callback, let gst_bus_pop
handle the socket read.
(bus_watch_destroy): Renamed from bus_destroy.
* check/Makefile.am: Re-enable the gst-register test so we can
deal with elements here. Add the gstbus tests.
* check/gst/gstbus.c: New check, spawns off a bunch of threads all
posting messages to a bus, then reads to see if they are in the
right order.
* check/pipelines/simple_launch_lines.c: Add some tests, and make
sure to pop the message off the bus after the poll.
* gst/gstbus.c (gst_bus_pop): Read off the control byte if the
queue becomes empty.
Original commit message from CVS:
2005-02-22 Andy Wingo <wingo@pobox.com>
* gst/elements/gstidentity.h
* gst/elements/gstidentity.c (gst_identity_handle_buffer): New
proc, factored out of the old chain func.
(gst_identity_chain): Use handle_buffer. Lock the stream.
(gst_identity_getrange): Lock the stream. Still doesn't do any
reporting tho.
(gst_identity_event): Handle flush events in the loop-based and
decoupled cases.
(identity_queue_flush, identity_queue_pop, identity_queue_push):
New procs, implement a 1-data buffer pen between threads in
decoupled operation.
(gst_identity_class_init, gst_identity_get_property)
(gst_identity_set_property): Use PROP_FOO instead of ARG_FOO. It's
not null if we get it, but I might remove five year old code. Add
has-src-loop, has-sink-loop, has-chain, has-getrange properties,
remove loop-based.
(gst_identity_finalize): Free mutex and cond.
(gst_identity_init): Alloc mutex and cond.
(gst_identity_sink_loop, gst_identity_src_loop): New procs.
(gst_identity_set_dataflow_funcs): New proc.
Original commit message from CVS:
Add finalize method to RealPad.
Add new lock to correctly do the preroll, should probably not
be put here.
Make fakesink do preroll correctly. Emit a message when the
preroll sample is queued.
Add more info in gst-launch regarding state changes.
Original commit message from CVS:
2005-02-21 Andy Wingo <wingo@pobox.com>
* gst/elements/gstfakesink.c:
* gst/elements/gstfakesrc.c:
(gst_fakesink_loop, gst_fakesrc_loop, gst_fakesink_chain)
(gst_fakesrc_get_range): Assert the pad has been activated in the
proper mode. This will fail right now for fakesrc ! identity !
fakesink.
(gst_fakesrc_activate, gst_fakesink_activate): Record the pad
activation mode.
* gst/elements/gstfakesrc.h: Add a pad_mode instance variable.
Original commit message from CVS:
2005-02-21 Andy Wingo <wingo@pobox.com>
* gst/gstpad.c (gst_real_pad_dispose): Free the stream lock (it
wasn't being freed before).
(gst_real_pad_init): Allocate and initialize the stream lock.
* gst/gstpad.h (GstRealPad): Change the stream lock to be
recursive, so that if an event comes while a loop-based sink is
pulling, the stream lock can be had by both the sink's loop
function and the sink's event function.
(GST_STREAM_GET_LOCK, GST_STREAM_LOCK, GST_STREAM_TRYLOCK)
(GST_STREAM_UNLOCK): Updated accordingly.
Original commit message from CVS:
2005-02-21 Andy Wingo <wingo@pobox.com>
* gst/gstmessage.h (GstMessageType): Turned into flags. Added
GST_MESSAGE_ANY as an OR of all flags.
(GST_MESSAGE_PARSE_STATE_CHANGED): New terrible macro. Will be
made into a function soon.
(GstMessage): Add a state_changed structure to the union. The
union will die soon in favor of a single GstStructure tho.
(gst_message_new_state_changed): New API.
* gst/gstmessage.c (gst_message_new_state_changed): New API.
* tools/gst-launch.c (check_intr): Set the state of the pipeline
to PAUSED here; the poll will catch the state change.
(event_loop): New function, polls the pipeline bus for events. Can
block until eos/error/state change, or just handle the pending
events.
(main): Changed to use event_loop instead of running a main loop.
* gst/gstbus.h (gst_bus_poll): Added.
(gst_bus_peek): Return non-const; the message is refcounted
anyway.
* gst/gstbus.c (gst_bus_init): Replace the GAsyncQueue with a
GQueue+mutex to allow for _peek. (The wake-up functionality
provided by GAsyncQueue is already done by our socketpair.) All
queue users changed to lock, operate, and unlock.
(gst_bus_post): Check the retval of write(2) and handle errno.
(gst_bus_peek): Implemented.
(gst_bus_pop, gst_bus_peek, bus_callback): Because the socketpair
is used to wake up the GSource, read off the character in the
GSource handler and not in pop/peek. This is because a peek will
require a pop in the future, and you can't read off the char
twice. Deal with errno in the read.
(bus_callback): Interpret the handler return value as whether or
not to pop the message from the bus.
(poll_handler, poll_timeout, gst_bus_poll): New API. gst_bus_poll
is meant to replace the while(gst_bin_iterate()) idiom.
Original commit message from CVS:
2005-02-18 Andy Wingo <wingo@pobox.com>
* gst/elements/gstfakesrc.c (gst_fakesrc_class_init): Add HAS_LOOP
and HAS_GETRANGE properties, readwrite + construct.
(gst_fakesrc_set_pad_functions)
(gst_fakesrc_set_all_pad_functions): New procs, like in fakesrc.
(gst_fakesrc_set_property, gst_fakesrc_get_property): Same as
fakesrc.
(gst_fakesrc_get_range_unlocked): New proc, factored out of the
core of _loop.
(gst_fakesrc_get_range): Getrange implementation for fakesink.
(gst_fakesrc_loop): Interface with get_range_unlocked.
(gst_fakesrc_activate): Add support for both activation modes,
with assertions.
* gst/elements/gstfakesink.h:
* gst/elements/gstfakesrc.h: Add has_loop and has_chain instance
variables.
* gst/elements/gstfakesink.c (gst_fakesink_class_init): Add
HAS_LOOP and HAS_CHAIN object properties, readwrite but also set
on construction.
(gst_fakesink_set_pad_functions): New proc, sets the functions on
one pad, setting the loop and chain functions properly.
(gst_fakesink_set_all_pad_functions): New proc, does
set_pad_functions on all pads.
(gst_fakesink_request_new_pad): Call set_pad_functions.
(gst_fakesink_set_property, gst_fakesink_get_property): When the
HAS_LOOP and HAS_CHAIN properties are set, reset the functions on
all pads.
(gst_fakesink_activate): Assert that the appropriate functions are
available for the activation mode.
(gst_fakesink_loop): Don't unref the buffer, chain_unlocked does
it for us.
Original commit message from CVS:
2005-02-16 Andy Wingo <wingo@pobox.com>
* gst/elements/gstfakesink.c (gst_fakesink_loop)
(gst_fakesink_chain_unlocked, gst_fakesink_activate): New
functions.
(gst_fakesink_chain): Use chain_unlock.
Original commit message from CVS:
* gst/gstevent.c: (gst_event_new_discontinuous_valist),
(gst_event_discont_get_value), (gst_event_new_segment_seek):
stop time can be -1.
Original commit message from CVS:
2005-02-10 Andy Wingo <wingo@pobox.com>
* testsuite/caps/value_serialize.c: merge from HEAD.
* testsuite/caps/subtract.c:
* testsuite/caps/filtercaps.c:
* testsuite/caps/enumcaps.c:
* testsuite/caps/deserialize.c:
* testsuite/caps/caps.c:
* testsuite/caps/audioscale.c: Unref caps, not free.
* testsuite/caps/caps_strings: Merged from HEAD.
* gst/elements/gstidentity.c (gst_identity_proxy_getcaps): Getcaps
implementation for identity.
* gst/gststructure.h
* gst/gststructure.c
(gst_caps_structure_fixate_field_nearest_double):
(gst_caps_structure_fixate_field_nearest_int): Moved from
gstcaps.c because we need the private IS_MUTABLE macro.
(IS_MUTABLE): New macro, determines if a
structure is mutable or not.
(gst_structure_free): Don't allow free while holding a pointer to
a parent's refcount.
(gst_structure_set_name): Check for writability.
(gst_structure_id_set_value): Same.
(gst_structure_set_value): Same.
(gst_structure_set_valist): Same.
(gst_structure_remove_field): Same.
(gst_structure_remove_all_fields): Same.
(gst_structure_map_in_place): New routine, like _foreach but
allows the function to mutate the value (via a non-const
prototype). Check for writability.
(gst_structure_foreach): Redefine to only take non-mutating
functions.
* gst/gstcaps.c (IS_WRITABLE): New macro, returns TRUE if the caps
are writable.
(gst_caps_copy): Add some docs about mutability, etc.
(_gst_caps_free): Set the parent refcount pointer on structures to
NULL before freeing them.
(gst_caps_ref): Document.
(gst_caps_make_writable): Doc.
(gst_caps_make_writable): Changed to make_writable, get_writable
sounds too much like retrieving a property.
(gst_caps_copy_1): Removed, not very useful.
(gst_caps_ref_by_count): Removed, no need to have something GLib
doesn't.
(gst_caps_copy_conditional): Ref instead of copying.
(gst_static_caps_get): Retain a reference to the returned caps, so
that the static caps will remain immutable.
(gst_caps_remove_and_get_structure): Set the parent refcount to
NULL when removing the structure.
(gst_caps_append): Fix to work with structure parent refcounts.
Check for writability.
(gst_caps_append_structure): Check for writability, work with
parent refcounts.
(gst_caps_remove_structure): Check for writability.
(gst_caps_set_simple): Check for writability.
(gst_caps_set_simple_valist): Check for writability.
(gst_caps_is_fixed_foreach): Update for non-mutating foreach.
(gst_structure_is_equal_foreach): Same.
(gst_caps_structure_intersect_field): Same.
(gst_caps_structure_subtract_field): Same. Make static.
(gst_caps_normalize_foreach): Same.
(gst_caps_structure_figure_out_union): Same.
(gst_caps_switch_structures): New static function, switches out
structures inside a caps, taking care of parent_refcount setting.
* gst/gstpad.c (gst_pad_get_allowed_caps): Remove the restriction
on only src pads, wim von masterhack claims it was bogus.
* testsuite/caps/Makefile.am
* testsuite/caps/app_fixate.c: Removed app_fixate test, things are
done a bit differently in THREADED.
Original commit message from CVS:
2005-02-10 Andy Wingo <wingo@pobox.com>
* gst/gstcaps.h:
* gst/gstcaps.c (gst_caps_get_writable): Change from
copy_on_write.
Original commit message from CVS:
Removed schedulers and compat stuff.
Use iterators in bin for the various lookup functions.
Optimized the clock a bit.
Added some more docs.
Fixed scheduling in the ()-(l) case.
Fixed and added some testcases.
Removed old code.
Original commit message from CVS:
* gst/gstelement.c: (gst_element_add_pad),
(gst_element_remove_pad), (gst_element_pads_activate),
(gst_element_dispose):
Small cleanups, free the state mutex.
Original commit message from CVS:
Clock fixes. Added async callbacks and clock unscheduling.
Threading fixes. Fixed race condition in GstTask and possible
deadlock in _pad_get_caps(). Made various subsystems (query,
format,..) threadsafe.
Lots of cleanups and documentation.
Original commit message from CVS:
* gst/gstevent.c: (gst_event_new_discontinuous_valist),
(gst_event_discont_get_value):
* gst/gstevent.h:
We'll need the end offset in discont soon.
