Avoid playing with the refcount to decide when a buffer has been recycled by the
dispose function. The problem is that we then temporarily can have a buffer with
a refcount > 1 being acquired from the pool, which is not writable. Instead use
a simple boolean return value from the dispose function to inform the called
that the object was recycled or not.
After we allocated a new buffer, call the release_buffer vmethod to put the new
buffer in the pool instead of assuming that the pool uses the default
release_method implementation.
Don't remove memory blocks from the buffer when we clip and resize, instead set
the memory offset and size to 0. This allows us to make the buffer larger again
later.
Add more debug.
Alow resize to 0 bytes.
Do clipping correctly.
Add more unit tests. Also add a failing test: when we resize to 0 and then
try to resize back to the original size it fails because the memory was
removed.
Allow for negative offsets when doing memory copy and share.
Add fast path in the _get_sizes() function.
Fix resize for negative offset and expanding the buffer.
Add some unit tests.
Also return the offset in a GstMemory block with the get_sizes() method. This
allows us to figure out how much prefix there is unused.
Change the resize function so that a negative offset can be given. This would
make it possible to resize the buffer so that the prefix becomes available.
Add gst_buffer_get_sizes() to return the offset and maxsize as well as the size.
Also change the buffer resize method so that we can specify a negative offset
to remove prefix bytes.
It was a bit too clever, and didn't really work as an API,
confusing people to no end. Better implement specific methods
whether an interface is usable/available/ready on the interface
itself, or even add GError arguments, rather than try to have
per-instance interfaces.
Don't mix messages and pads and tags.
Make the sink post tag messages when a tag event is received.
Since tags are sticky on pads now, they can be retrieved from there
when needed.
Make the Array of structures private. This should allow us to implement
the array more efficiently or with some preallocated structures when
we want to later.
Add a new method to clean up a static structure so that we can remove some code
that pokes into the private bits of the caps.
Add a function to retrieve an array of supported metadata apis from the the
bufferpool.
Add functions to configure and query the configured metadata apis in a
bufferpool configuration.
This reverts commit de29ae7b92.
Re-adds GFLOAT_TO_LE, GFLOAT_TO_BE, GDOUBLE_TO_LE, and GDOUBLE_TO_BE.
Turns out these aren't in GLib yet afer all (since we didn't
actually open a bug to get them added..)
Add an index to gst_buffer_take_memory() so that we can also insert memory at a
certain offset. This is mostly interesting to prepend a header memory block to
the buffer.
Fix the code to support allocating the buffer and memory in one memory block.
Add an extra variable to store the memory of the buffer.
This code is disabled still because of complications.
Make a new method to allocate a buffer + memory that takes the allocator and the
alignment as parameters. Provide a macro for the old method but prefer to use
the new method to encourage plugins to negotiate the allocator properly.
Add a boolean to the flush_stop event to make it possible to implement flushes
that don't reset_time.
Make basesink post async_done with the reset_time property from the flush stop
event.
Fix some unit tests
Add a new method to steal the miniobject stored at a location.
Add a new method to store a miniobject in a location and taking ownership
of the miniobject.
Always forward all events in the default handler. Previously it used to not
forward caps events by default. It makes more sense to forward the caps events,
if the element is interested in the caps, it will implement an event handler to
retrieve the caps and then it can decide to forward or not. If the element has
no event handler, it probably just doesn't care about caps and it probably is
also not going to modify the data in a way that needs a caps change.
Move the flag to indicate that a new_base_time should be distributed to the
pipeline, from the async_start to the async_done message. This would allow us to
decide when to reset the pipeline time based on other reasons than the
FLUSH_START event.
The main goal eventually is to make the FLUSH events not reset time at all but
reset the time based on the first buffer or segment that prerolls the pipeline
again.
Require the memory implementations to implement a share operation. This allows
us to remove the fallback share implementation which uses a different allocator
implementation and complicates things too much.
Update design doc a bit.
Make the fallback copy use the same memory allocator as the original object.
Improve some docs.
Require an alloc function when registering an allocator.
Remove gst_memory_allocator_get_default() and merge the feature in
gst_memory_allocator_find()
Fix locks on the hashtable.
Remove defined but not-implemented gst_memory_span() method.
Rename the GstMemoryImpl to GstMemoryAllocator because that's really what it is.
Add an alloc vmethod to the allocator members.
Improve registration of allocators.
Add methods to get and set the default allocator
Always use an allocator to allocate memory, use the default allocator when NULL
is passed.
Add user_data to the allocator Info so that we can pass extra info to the
allocator new method.
Rename gst_pad_dispatcher() to gst_pad_forward() and make it more useful by
iterating the internal links of a pad and handling resync properly.
Add a method gst_pad_event_forward() that unconditionally forwards an event to
all internally linked pads.
Update some pad code to use the new forward function.
See #651514 for details. It's apparently impossible to write code
that avoids both type punning warnings with old g_atomic headers and
assertions in the new. Thus, macros and a version check.
The most common case is to not specify any flags when doing the allocation. Make
the allocation from a pool with a maximum amount of buffers block by default for
this reason.
Optimize linking by only releasing the pad locks when there are link functions
installed on the pads.
Add some G_LIKELY here and there.
Move error paths out of the main code flow.
Keep track of installed number of probes to shortcut emission.
Allow NULL callbacks, this is useful for blocking probes.
Improve probe selection based on the mask, an empty mask for the data or the
scheduling flags equals that all probes match.
Add some more debug info.
Don't check the flushing flag in the probe callback handler, this needs to be
done before calling the handler.
Fix blocking probes.
Fix unit tests
Make the PadBlock callback take a GstBlockType parameter to handle the different
kind of stages in the pad block. This provides for more backwards compatibility
in the pad block API.
Separate blocking and unblocking into different methods, only blocking can do a
callback, unblock is always immediately. Also removed synchronous blocking, it
can always be implemented with a callback.
Previously this was only done in the is_subset() check but
having it only there brings us into definition-hell where
"1" and "{1}" are subset of each other but not equal.
Make pad block call the callback as soon as the pad is not in use. This makes it
possible to make sure that when the callback is called, no activity is happening
on the pad and that no activity will ever happen until the pad is unblocked
again. This makes pad blocking work when there is no dataflow or after EOS and
greatly helps dynamic pipelines.
Move the probe handling right where we wait on the pad block. The two are
related but not the same and the probe can eventually influence the pad
blocking as we'll se later.
Fix up some broken unit tests or tests that fail with the new behaviour.
..and as a result gst_caps_is_equal() and others.
This now only checks if for every subset structure there is
a superset structure in the superset caps. Previously we were
subtracting one from another, creating completely new caps
and then even simplified them.
The new implemention now is about 1.27 times faster and doesn't
break the -base unit tests are anything anymore.
..and as a result gst_caps_is_equal() and others.
This now only checks if for every subset structure there is
a superset structure in the superset caps. Previously we were
subtracting one from another, creating completely new caps
and then even simplified them.
The new implemention now is about 1.27 times faster.
Rework _push_event() a little so that it drops events on blocking pads.
Make sure that events are forwarded when we unblock.
Add counter on the pad to keep track of busy pads.
Implement a default buffer-list function in case the element doesn't implement
one.
Also pass buffer-lists to the have-data signal, this allows us to remove some
backward compatibility code.
Change the sticky event array so that it contains a pending and an active event.
Events on the sinkpad are copied to the pending array and after the eventfunc
returned TRUE, moved to the active event. This allows us to queue new events
like when we do per-pad offsets without removing the currently active event.
Remove the active argument from the gst_pad_get_sticky_event() method, the
pending events are not something we want to expose.
When linking pads and when copying a segment event from the sourc pad to the
sinkpad, apply the src and sinkpad offsets to the segment base. Make sure that
we only modify the event stored on the sinkpad and never the one on the source
pad.
When changing the pad offset, perform the segment copy with the updated offsets.
When pushing a segment event, apply the srcpad offset before sending the event
to the peer pad.
This part is missing the adjustment of the segment event on the sinkpad, which
is for a later patch.
Add methods to adjust the offset. This will be used to change the segment events
with an offset so that we can tweak the timing of the stream on a per-pad base.
First store the sticky event on the sinkpad in the inactive state, then check
for the flushing flag. We want to have the events on sinkpads at all times,
ready to be activated when the pad becomes active.
Make a function to call the eventfunc and perform a caps check when we are
dispatching a caps event.
This makes sure that all code paths correctly check that the caps are
acceptable before sending the caps to the eventfunction.
Update the design docs with some clear rules for how sticky events are
handled.
Reimplement the sticky tags, use a small structure to hold the event and its
current state (active or inactive).
Events on sinkpads only become active when the event function returned success
for the event.
When linking, only update events that are different.
Avoid making a copy of the event array, use the object lock to protect the event
array and release it only to call the event function. This will need to check
if something changed, later.
Disable a test in the unit test, it can't work yet.
The feature name is not supposed to change over time anyway. In order to enforce
this parentize features to the registry and make the feature->name pointing to
GstObject:name. In 0.11 we could consider of removing the feature->name variable
(FIXME comment added).
Fixes: #459466
gst_pad_template_get_caps(), gst_pad_get_pad_template_caps()
and gst_pad_get_pad_template() return a new reference of the
caps or template now and the return value needs to be
unreffed after usage.
This reverts commit cf4fbc005c.
This change did not improve the situation for bindings because
queries are usually created, then directly passed to a function
and not stored elsewhere, and the writability problem with
miniobjects usually happens with buffers or caps instead.
Improve GstSegment, rename some fields. The idea is to have the GstSegment
structure represent the timing structure of the buffers as they are generated by
the source or demuxer element.
gst_segment_set_seek() -> gst_segment_do_seek()
Rename the NEWSEGMENT event to SEGMENT.
Make parsing of the SEGMENT event into a GstSegment structure.
Pass a GstSegment structure when making a new SEGMENT event. This allows us to
pass the timing info directly to the next element. No accumulation is needed in
the receiving element, all the info is inside the element.
Remove gst_segment_set_newsegment(): This function as used to accumulate
segments received from upstream, which is now not needed anymore because the
segment event contains the complete timing information.
Some tests (e.g. elements/capsfilter) have pipelines with dangling
sinkpads and without a sink element. These pipelines can never post
an EOS message (because this is only valid by a sink) and as such
should never get an EOS message posted by the bin.
API: gst_mini_object_weak_ref()
API: gst_mini_object_weak_unref()
Add weak referencing functionality to GstMiniObject, which
allows to get notifications when an mini object is destroyed
but doesn't increase the real refcount. This is mostly
useful for bindings.
Fixes bug #609473.
API: GstElement::state_changed
This is always called when the state of an element has changed and
before the corresponding state-changed message is posted on the bus.
This allows to get the internal pad of ghostpads and
proxypads without using gst_pad_iterate_internal_links()
and is much more convenient.
The internal pad of a ghostpad is the pad of the opposite direction
that is used to link to the ghostpad target.
This prevents adding duplicates over and over again to the resulting
caps if they already describe the new intersection result.
While this changes intersection from O(n*m) to O(n^2*m), it results in
smaller caps, which in the end will decrease further processing times.
For example in an audioconvert ! audioconvert ! audioconvert pipeline,
when forwarding the downstream caps preference in basetransform
(see e26da72de25a91c3eaad9f7c8b2f53ba888a0394) this results in
16 instead of 191 caps structures.
Resetting the result is not necessary when resyncing because
pads that previously got the event will be skipped and we
need to consider the results of the previous pushes.
Add a GType to the metadata to identify the GstMetaInfo.
We can remove the (de)serialize functions for the metadata because we can
register GTtype transform functions between various types to implement
serialization later.
gst_structure_get_type() -> _gst_structure_type to avoid method calls for
getting the GType that initialized at the start.
Hide some structure fields in private data so that we can change the
implementation.
Move structure equality check from caps.c to structure.c where it belongs.
When we get the structure of an event, make sure it also contains the fields
that we keep in fast variables, this way we can easily serialize and debug
the events. We would probably later simply prefer to register a transform
function to G_TYPE_STRING and G_TYPE_BYTEARRAY etc..
Hide the GstStructure of the event in the implementation specific part so that
we can change it.
Add methods to check and make the event writable.
Add a new method to get a writable GstStructure of the element.
Avoid directly accising the event structure.
Fix replace of caps events when linking: we need to unref the old ones.
Make sure we pass error values around.
Move backward compat code into the default handler for now.
Don't use the buffer caps for negotiation anymore but use the CAPS events.
Make the _set_caps method produce the CAPS event, add some backward
compatibility code to trigger the setcaps functions on src and sinkpads.
Remove all negotiation code from the chain functions.
Don't use the GST_PAD_CAPS variable anymore to store the caps but retrieve the
caps from the sticky event array.
Remove the context again, adding an extra layer of refcounting and object
creation to manage an array is too complicated and inefficient. Use a simple
array again.
Also implement event updates when calling gst_pad_chain() and
gst_event_send_event() directly.
The refactoring of gst_debug_add_log_function() now causes build failure when
debug-logging is turned off. Just move it to the conditional part of the header.
Pass the context downstream when it got updated.
Have two ways of informing downstream of events, do a full context update when
the CONTEXT_PENDING flag is set and simply forward the event otherwise.
Set the CONTENT_PENDING flag when linking pads.
We don't need to old context anymore when updating the context of a pad.
This prevents adding duplicates over and over again to the resulting
caps if they already describe the new intersection result.
While this changes intersection from O(n*m) to O(n^2*m), it results in
smaller caps, which in the end will decrease further processing times.
For example in an audioconvert ! audioconvert ! audioconvert pipeline,
when forwarding the downstream caps preference in basetransform
(see e26da72de25a91c3eaad9f7c8b2f53ba888a0394) this results in
16 instead of 191 caps structures.
If there is no custom getcaps function on a sink pad, then changes in
downstream caps will never be propagated, so there is no point in trying to
renegotiate the capabilities.
This reverts commit 9ef1346b1f.
Way to much for one commit and I'm not sure we want to get rid of the pad caps
just like that. It's nice to have the buffer and its type in onw nice bundle
without having to drag the complete context with it.
Resetting the result is not necessary when resyncing because
pads that previously got the event will be skipped and we
need to consider the results of the previous pushes.
Add a new CAPS event that will be used to negotiate downstream elements. It'll
also stick on pad so that we can remove the GstCaps field on pads and the
GstCaps field on buffers.
Copy the sticky events from the srcpad to the sinkpad when linking pads. Set the
STICKY_PENDING flag to make sure that the sticky events are dispatched before
pushing the next buffer to the element.
Add the sticky flag to events and a sticky index.
Keep sticky events in an array on each pad.
Remove GST_EVENT_SRC(), it is causing refcycles with sticky events, was not used
and is not very interesting anyway.
Remove pad_alloc and all references. This can now be done more efficiently and
more flexible with the ALLOCATION query and the bufferpool objects. There is no
reverse negotiation yet but that will be done with an event later.
Add a query to request allocation parameters and optionally a bufferpool as
well. This should allow elements to discover downstream capabilities and also
use the downstream allocators.
Drop in old GstBus code for the release to play it safe, since
regressions that are apparently hard to track down and reproduce
have been reported (on windows/OSX mostly) against the lockfree
version, and more time is needed to fix them.
This reverts commit 03391a8970.
This reverts commit 43cdbc17e6.
This reverts commit 80eb160e0f.
This reverts commit c41b0ade28.
This reverts commit 874d60e589.
This reverts commit 79370d4b17.
This reverts commit 2cb3e52351.
This reverts commit bd1c400114.
This reverts commit 4bf8f1524f.
This reverts commit 14d7db1b52.
https://bugzilla.gnome.org/show_bug.cgi?id=647493
When a plugin file no longer exists, e.g. because it's been removed or
renamed, don't remove all features in the registry based on the *name*
of the plugin they belong to, but only remove those who actually belong
to that particular plugin (object/pointer).
This fixes issues of plugin features disappearing when a plugin .so file
is renamed.
https://bugzilla.gnome.org/show_bug.cgi?id=604094
... which happens in particular flushing a bus, possibly as part
of a state change, e.g. when having a pipeline in a pipeline
and then changing state back to NULL. The interior pipeline
will/might then flush the bus, which is a child bus from the
parent which does not have a poll anymore these days.
https://bugzilla.gnome.org/show_bug.cgi?id=648297
This allows to add pad templates and set metadata in class_init instead of
base_init. base_init is a concept that is not supported by almost all
languages and copying the templates/metadata for subclasses is the more
intuitive way of doing things.
Subclasses can override pad templates of parent classes by adding a new
template with the same now.
Also gst_element_class_add_pad_template() now takes ownership of the
pad template, which was assumed by all code before anyway.
Fixes bug #491501.
Based on patch by: Daniel Macks <dmacks@netspace.org>
Earlier versions of OSX don't support proper multiarch and
trying to use /usr/bin/arch -foo with those versions would
just break things.
https://bugzilla.gnome.org/show_bug.cgi?id=615357
1) We need to lock and get a strong ref to the parent, if still there.
2) If it has gone away, we need to handle that gracefully.
This is necessary in order to safely modify a running pipeline. Has been
observed when a streaming thread is doing a buffer_alloc() while an
application thread sends an event on a pad further downstream, and from
within a pad probe (holding STREAM_LOCK) carries out the pipeline plumbing
while the streaming thread has its buffer_alloc() in progress.
On OSX, GStreamer might be built as a 'fat/universal' binary containing
both 32-bit and 64-bit code. We must take care that gst-plugin-scanner
is executed with the same architecture as the GStreamer core, otherwise
bad things may happen and core/scanner will not be able to communicate
properly.
Should fix issues with (32-bit) firefox using a 32-bit GStreamer core
which then spawns a 'universal' gst-plugin-scanner binary which gets
run in 64-bit mode, causing 100% cpu usage / busy loops or just hanging
firefox until killed.
https://bugzilla.gnome.org/show_bug.cgi?id=615357
As GST_SCHEDULING reports when buffers pass through pads due to
gst_pad_push calls, they are a good way of tracking the progress of
buffers through pipelines. As such, adding output of the buffer pointers
to these messages allows tracking of specific buffers, easing debugging.
Remove the android/ top dir
Fixe the Makefile.am to be androgenized
To build gstreamer for android we are now using androgenizer which generates the needed Android.mk files.
Androgenizer can be found here: http://git.collabora.co.uk/?p=user/derek/androgenizer.git
Even if we currently do not have a duration yet, assume seekable if
it looks like we'll likely be able to determine it later on
(which coincides with needed information to perform seeking).
Fixes#641047.
Rather than a fixed default frame count, estimate frame count to aim for
an interval duration depending on fps if available, otherwise use old
fixed default.
Also add a format flag to signal baseparse that subclass/format can provide
(parsed) timestamp rather than an estimated one. In particular, such "strong"
timestamp then allows to e.g. determine duration.
Don't unref the event if it hasn't been handled, because the caller
assumes it is still valid and might reuse it.
I ran into this problem when transcoding an AVI (with mp3 inside)
to gpp.
https://bugzilla.gnome.org/show_bug.cgi?id=639555
That is, as such formats allow subclass to extract position from frame,
it is possible to extract duration (if not otherwise provided)
from (near) last frame, and a seek can fairly accurately target the required
position.
Fixes#631389.
Arrange for upstream as well as downstream flushing when seeking.
Also determine upstream size as well as seekability. Adjust some comments
to reality and employ debug statement in proper order.
This reverts commit b5a3d60363.
Reverting this for now, since no one really seems to remember why this
property exists or what it could possibly be good for. It seems to have
been in the original mp3parse since the beginning of time and was back-
ported from there.
Seekability, like duration, etc is unlikely to change (frequently), and
the default assumption covers most cases, so let subclass set when needed.
At the same time, allow subclass to indicate if it has seek-metadata (table)
available, and possibly have it provide an average bitrate.
This allows the child class to chain its event handler with
GstBaseParse, so that subclasses don't have to duplicate all the default
event handling logic.
https://bugzilla.gnome.org/show_bug.cgi?id=622276
We wait to parse a minimum number of frames (10, arbitrarily) before
emiting bitrate tags so that our early estimates are not wildly
inaccurate for streams that start with a silence. If the stream ends
before that, we just emit the tags anyway.
While it _would_ be nicer to be specify the threshold to start pushing
the tags in terms of duration, this would introduce more complexity than
this merits.
https://bugzilla.gnome.org/show_bug.cgi?id=614991
This makes baseparse keep a running average of the stream bitrate, as
well as the minimum and maximum bitrates. Subclasses can override a
vfunc to make sure that per-frame overhead from the container is not
accounted for in the bitrate calculation.
We take care not to override the bitrate, minimum-bitrate, and
maximum-bitrate tags if they have been posted upstream. We also
rate-limit the emission of bitrate so that it is only triggered by a
change of >10 kbps.
Perform sanity check on type of seek, and only perform one that is
appropriately supported. Adjust downstream newsegment event
to first buffer timestamp that is sent downstream.
In particular, consider DISCONT == !sync, and allow subclass to query
sync state, as it may want to perform additional checks depending
on whether sync was achieved earlier on.
Also arrange for subclass to query whether leftover data is being drained.
In particular, (optionally) provide baseparse with a notion of frames per second
(and therefore also frame duration) and have it track frame and byte counts.
This way, subclass can provide baseparse with fps and have it provide default
buffer time metadata and conversions, though subclass can still install
callbacks to handle such itself.
After all, stream is as-is, and there is little molding to downstream's
taste that can be done. If subclass can and wants to do so, it can
still override as such.
Also handle the case gracefully where the subclass decides to drop
the first buffers and has no caps set yet. It's still required to
have valid caps set when the first buffer should be passed downstream.
Sending the flush-start event forward before taking the stream lock actually
works, in contrast to deadlocking in downstream preroll_wait (hunk 1).
After that we get the chain function being stuck in a busy loop. This is fixed
by updating the minimum frame size inside the synchronization loop because the
subclass asks for more data in this way (hunk 2).
Finally, this leads to a very probable crash because the subclass can find a
valid frame with a size greater than the currently available data in the
adapter. This makes the subsequent gst_adapter_take_buffer call return NULL,
which is not expected (hunk 3).
Baseparse internaly breaks the semantics of a _chain function by calling it with
buffer==NULL. The reson I belived it was okay to remove it was that there is
also an unchecked access to buffer later in _chain. Actually that code is wrong,
as it most probably wants to set discont on the outgoing buffer.
This allows to only create the socketpair when it is really required instead
of always creating it and immediately destroying it again for child buses.
https://bugzilla.gnome.org/show_bug.cgi?id=647005
This is used by GstBin to create a child bus without
a socketpair because child buses will always work
synchronous. Otherwise too many sockets could be
created and the limit of file descriptors for the
process could be reached.
Fixes bug #646624.
