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.
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.
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 new fill virtual method to basesrc. The purpose of this method is to fill
a provided buffer with data.
Add a default implementation of the create method that allocates a buffer and
calls the fill method on it. This would allow the base class to implement
bufferpool and allocator negotiation on behalf of the subclasses.
Fix the blocksize property.
Make filesrc use the new fill method.
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
This allows subclass to indicate that size reported by src may not be static
and should as such be updated regularly, rather than only when really
needed.
Particular examples are filesrc or fdsrc reading from a file that is still
growing (e.g. being downloaded).
Fixes#652037.
This reverts commit 934faf163c.
Original commit leads to possibly sending newsegment event downstream
in pull mode. In push mode, quite some downstream elements
are likely to only expect newsegment event following a seek they performed
and as such may have their state messed up.
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.
Instead of passing it structure by structure. This allows
better optimized transform_caps functions and allows better
transformation decisions.
See bug #619844.
Don't error out when the allocation query returns success.
Do bufferpool query after we pushed the caps event downstream so that we can get
a good bufferpool suggestion.
Also proxy the bufferpool query downstream when we operate in in_place mode.
Avoid installing a setcaps function on the srcpad and calling the setcaps
function, we can do more efficiently with sending the event ourself and calling
our vmethod.
While some formats allow subclass to determine a specific subsequent
needed frame size, others may to need to scan for markers and can only
request 'additional data' by whatever reasonable available step.
In push mode, trying to minimize additional latency leads to step size
being the next input buffer. In pull mode, any reasonable step size
(such as already used by buffer caching) can be applied.
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.
Doing so avoids a large timestamp gap between first and second buffer
for live sources which take time to start up.
The first buffer now has a "live" timestamp based on the running time,
as other buffers do.
https://bugzilla.gnome.org/show_bug.cgi?id=649369
Use the caps event to configure basetransform.
Remove force_alloc hack, we don't need this in 0.11 with new upstream
negotiation.
Avoid getting some pad caps.
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.
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.
Protect index with its own lock. gst_index_get_writer_id() may take
the object lock internally (the default resolver, GST_INDEX_RESOLVER_PATH,
will anyway), so if we're using that to protect the index as well,
we'll deadlock.
https://bugzilla.gnome.org/show_bug.cgi?id=646811
Change semantics of gst_base_parse_push_frame() and make it take
ownership of the whole frame, not just the frame contents. This
is more in line with how gst_pad_push() etc. work. Just transfering
the content, but not the container of something that's not really
known to be a container is hard to annotate properly and probably
won't work. We mark frames allocated on the stack now with a private
flag in gst_base_parse_frame_init(), so gst_base_parse_frame_free()
only frees the contents in that case but not the frame struct itself.
https://bugzilla.gnome.org/show_bug.cgi?id=518857
API: gst_base_parse_frame_new()
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.
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
Seems like the best fit to what it does, and is shorter than
set_frame_properties() which might also have been confusing
because of GstBaseParseFrame.
https://bugzilla.gnome.org/show_bug.cgi?id=518857
This is more in line with e.g. GstBaseTransform's API, and makes for nicer
to read code. No getters for now since I don't see any use case for them,
the API is for subclasses, which usually know what format they're
dealing with already and hence know what they've set.
https://bugzilla.gnome.org/show_bug.cgi?id=518857
The first because it seems a better fit conceptually, the second
to express booleanness. Also change the accessor macros for subclasses
to GST_BASE_PARSE_DRAINING and GST_BASE_PARSE_LOST_SYNC.
https://bugzilla.gnome.org/show_bug.cgi?id=518857
This is useful for parser like flacparse or h264parse which may need to process
some buffers before they can construct the final caps, in which case they may
want to delay pushing the initial buffers until the full and proper caps are
known.
https://bugzilla.gnome.org/show_bug.cgi?id=646341
This makes more sense conceptually, since the bitrate may be used
to estimate a seek position if there's no seek table or just for
duration reporting/estimation if we can't seek. Also, even if the
format is not syncable, we could still seek by pushing data from the
start and using the segment to make downstream clip.
https://bugzilla.gnome.org/show_bug.cgi?id=518857
Also change gst_base_parse_set_format(parse,flags,switch_on) to
gst_base_parse_set_format_flags(parse,flags) which is more in line
with the rest of our API and how the function is used.
Especially drop tag events when flushing to not send them over
and over again.
Should've been in the last commit already but I forgot to call
git rebase --continue...
basesrc's default event handler returns TRUE regardless of whether the
event is handled or not. This fixes the handler to conform with the
expected behaviour (which is to only return TRUE when the event has
actually benn handled). gst_bin_do_latency_func() depended on this
(incorrect) behaviour, and is now modified as well.
(Remaining 1-liner change in gstbasesrc.c is to keep gst-indent happy)
Deal with the hints from gtk-doc and fix the xrefs. Apply a work-around for ()
precedence over @. Move "MT Safe" text to doc body in many places. Trim eol
whitespaces.
If the element gave us caps in a specific order, let's retain that
by intersecting against the template but retaining the order given
by the element.
https://bugzilla.gnome.org/show_bug.cgi?id=617045
gstbytereader.h: In function ‘guint8* gst_byte_reader_dup_data_unchecked(GstByteReader*, guint)’:
gstbytereader.h:249:75: error: invalid conversion from ‘void*’ to ‘guint8*’
gstbytewriter.h: In function ‘gboolean _gst_byte_writer_ensure_free_space_inline(GstByteWriter*, guint)’:
gstbytewriter.h:196:75: error: invalid conversion from ‘void*’ to ‘guint8*’
https://bugzilla.gnome.org/show_bug.cgi?id=645595
Avoid doing unnecessary pad-allocs when on passthrough mode.
If multiple basetransform elements are on a pipeline, they
would do a pad-alloc for each received buffer, each element
would do this, so we would have lots of pad allocs on the
pipeline for a single buffer being pushed through it.
This patch attempts to reduce this amount by avoiding
doing pad-allocs if the element has already done it
after the last pushed buffer. So it will only be allowed
to do a new pad-alloc after it has pushed a buffer, so we get
1x1 pad-alloc and buffer ratio
https://bugzilla.gnome.org/show_bug.cgi?id=642373
If after computing the suggestion with downstream caps we still have
a non-fixed suggestion caps try to intersect with the input caps
of the pad alloc to avoid useless renegotiations.
https://bugzilla.gnome.org/show_bug.cgi?id=642130
Improve the calculation of the duration. When we have no input duration set on
the input buffers stop is set to start and then we end up using a 0 duration in
the average calculation.
Keep track of the earliest allowed timestamp according to the latest
QoS report and drop buffers before that time. Activate this filter
when throttling is enabled. We could later also activate this in the
other QoS cases.
See #638891
Only go into LIVE_WAIT when the are not live_running and only stop waiting when
live_running is TRUE. If we don't loop, we could deadlock when called from
outside of basesrc, such as baseaudiosrc.
Fixes#635785
This can happen for example when downstream proposed new caps, later proposed
the previous caps again which in turn enables passthrough mode in upstream
elements and the wrong-sized buffer appears in an element where the caps
change never happened. Simply allocate a new buffer in this case.
See bug #635461.
Only update the last_stop value when we had a valid stop position for the
clipping or else the clipping code assumes the stop position extends to the end
of the segment, which makes the position reporting return weird values.
Make the _get_caps functions behave like the _get_caps_reffed variants and
remove the _reffed variants. This means that _get_caps doesn't return a writable
caps anymore and an explicit _make_writable() is needed before modifying the
caps.
Unify the different position reporting code paths to make it more
understandable.
Use start_time to get more accurate position reporting in paused.
Fix unit tests for more accurate reporting.
This reverts commit 80727c1177.
This doesn't make sense. gst_data_queue_new_full() is already
documented above. And we need the doc blurb for _new() here.
Use atomic ops to read and write more properties. Taking the preroll lock in get_property
can lock up applications reading the property during preroll.
Because of the awkward refcounting in prepare_output_buffer, we might end up
with writable buffers that point to the same data. Check for those cases so that
we avoid a useless memcpy and keep valgrind quiet.
Fixes#628176
Add a function to retrieve a list of buffers containing the first N bytes from
the adapter. This can be done without a memcpy and should make it possible to
transfer the list to a GstBufferList later.
Make code including GStreamer headers compile with -Wcast-qual by
maintaining const-ness when casting. Also fix function signature of
gst_byte_writer_set_pos(): the byte writer should not be marked as
const.
https://bugzilla.gnome.org/show_bug.cgi?id=627910
Sets up a GST_PKG_CONFIG_PATH variable for use in Makefile.am
(avoids trailing ':' in PKG_CONFIG_PATH used). A useful side
effect of this is also that the PKG_CONFIG_PATH environment
is now logged in the configure output.
When we are handling a buffer and need to allocate an output buffer, handle the
case when downstream suggests us a format that we can't convert the input buffer
to. In that case, check if there is another format available downstream instead
of failing.
Fixes#621332 and see also #614296
If initially pass-through caps are negotiated between a transform element's
sink and src pads, but then the downstream element returns different caps
on a buffer from pad_alloc(), basetransform gets stuck with proxy_alloc=TRUE
even though the upstream peer doesn't accept the caps, causing
gst_pad_peer_accept_caps() to be called on each buffer in _buffer_alloc():
if (!gst_caps_is_equal (newcaps, caps)) {
GST_DEBUG_OBJECT (trans, "caps are new");
/* we have new caps, see if we can proxy downstream */
>> if (gst_pad_peer_accept_caps (pad, newcaps)) {
/* peer accepts the caps, return a buffer in this format */
GST_DEBUG_OBJECT (trans, "peer accepted new caps");
which is taking ~40ms/frame.
This patch does two things. (1) if the buffer returned from pad_alloc() has
new caps, trigger the decision whether to proxy the buffer-alloc to be
revisited, and (2) disable proxy if peer does not accept new caps. (The first
part may not be strictly needed, but seemed like a good idea.)
Note that this issue would not arise except in case of downstream elements
who have on their template-caps, some that would be suitable for pass-through,
but at runtime pick more restrictive caps (for ex, after querying a driver for
what formats it actually supports).
When basetransform received an unsupported caps on pad_alloc
it just returned not-negotiated. This patch makes it query
the allowed caps between his sinkpad and upstream's srcpad
to find a caps to suggest.
This happens when dinamically switching pipeline elements
and upstream pad_allocs with the previous caps that was
being used.
Fixes#614296
Add a new enable-last-buffer property. When false, it disables storing the last
received buffer in basesink::last-buffer. This can be useful in cases where
buffers need to be released asap.
API: GstBaseSink::enable-last-buffer
Retain the last scanned buffer entry and offset, so we can resume buffer
scanning there in case of a typical progressive scan.
Also potentially optimize _copy subsequently occurring in that area.
Allow subclasses to override the acceptcaps function because in some cases a
custom implementation can be much much faster than the default one.
See #621190
Point g-ir-scanner to the .la file of our library, which hopefully
makes it find the right dependencies in all cases (ie. our locally
built libgstreamer and not the system-installed one). This is also
how it's done in Gtk+ and how it's documented in the wiki, see
http://live.gnome.org/GObjectIntrospection/AutotoolsIntegration
Based on patches by Vincent Untz and Alan Knowles.
Fixes#603710.
Our own pkgconfig directory should come first, so that pkg-config uses
the in-tree libgstreamer and not some external one when --pkg=gstreamer-0.10
is passed to g-ir-scanner.
See #603710.
As the headers were broken in 0.10.26 the functions weren't really
usable back then, so we should advertise them as being there only
since 0.10.27.
Spotted by Mart Raudsepp.
When doing pad_allocs, use non-fixed caps suggestions and
try to fixate them before using. This makes possible to
have suggested buffer size with 0 in basetransform just
to signal upstream a renegotiation is needed
Fixes#576234Fixes#609046
Adds that warning to configure.ac
Includes a tiny change of the GST_BOILERPLATE_FULL() macro:
The get_type() function is no longer declared before being defined.
https://bugzilla.gnome.org/show_bug.cgi?id=611692
The functions are called gst_byte_writer_put_{float32|float64}_*() and not
gst_byte_writer_put_{float|double}_*().
Spotted by: Benjamin Otte <otte@redhat.com>
_get_range() is a pad function set by ourselves, therefore we're certain that
the parent is a GstBaseSrc.
Speeds up _get_range by 38%, and the total call by 30%. (valgrind instruction
calls measurements).
Fixes#610246
Adds a new function to GstByteWriter that writes
a constant value to a memory area (aka memset).
Useful for adding padding to buffers.
Also updates .def file and docs.
API: gst_byte_writer_fill()
Updating the segment values must only be done while holding the
STREAM_LOCK and OBJECT_LOCK. This means, reading can be done as
long as one of them is held, not both, which removes some lock-unlock
blocks from performance critical code paths.
Also document, that gst_base_src_set_format() *must* be called in
states <= READY and add an assertion for this. Changing the format
later will completely mess up the segment information.
gst_byte_writer_reset_and_get_buffer wasn't declared
in .h, instead there was _reset_and_get_data_as_buffer.
Replace it with the real function name, that is smaller
and matches gst_byte_writer_free_and_get_buffer
https://bugzilla.gnome.org/show_bug.cgi?id=608726
For the reason outlined at the beginning of gst_private.h (inline
functions in glib may need the g_log_domain variable). Also include
gst_private.h before using any G_OS_* defines, esp. in plugin loader.
Add a method to perform get_range typefinding that also uses the
uri/location extension as an extra hint. It will first try to call the
typefind functions of the factories that handle the given extension. The result
is that in the common case, we only call one typefind function, which speeds up
the typefinding a lot.
Make sure we take ownership of the buffer early without increasing its refcount
when we go in the collect function. This reduces the amount of copies needed in
order to make the buffer writable in most cases.
In some cases we can avoid allocating a subbuffer and instead simply ref
the buffer. Callers should perform _make_metadata_writable() in all
cases now.
Add a method to install a clipping function that is called when a buffer is
received. Users of collectpads can then perform clipping on the incomming
buffers.
Also retab the header file a little.
See #590265
This allows demuxers to update the segment stop of an already
finished stream. This might be needed if some stream goes to
EOS before the duration of the longest stream is known to properly
set the segment stop of all streams to the same value in the end.
Set the pad flushing and stop the pad task when the initial seek fails
during activation. Avoids racy calls into the _create() function when
BaseSrc::stop() has already run.
Fixes: #603059
Also, fix some misspelled comments.
Rounding errors with the floating point rate could make it so that we
don't end up exactly at the required stepping duration.
Use the segment clipping boundaries, which are not subject to rate
adjustements, instead to detect when we reached the stepping duration.
Add some debug info related to going to the PAUSED state.
Some gcc versions warn about bytewriter writing to memory accessed
via a const guint8 pointer, despite our explicit cast to guint8 *.
Work around that by using an intermediary variable.
Fixes#598526.
When clamping the base time, correctly use 'now', instead of
'-now' - the intent is to prevent 'now-base' ever being
negative, which would cause a position report outside the segment.
Fixes: #602419
Element base_time is a signed quantity, which leads to basesink returning
a position of 0 when dealing with a negative base time - which are quite
legal when clocks (such as the audio clock) are close to 0.
This doesn't manifest in normal pipelines, of course - but can happen
(at least) when manually setting the base time on a pipeline.
This avoids:
* triple-checking for the GType when type-checking is enabled (see #597260)
* Avoids going through an expensive no-argument checking which landed in
glib-2.22
* Avoids going through 2 extrac functions (g_object_new -> g_object_new_valist)
There's not much point in using GST_DEBUG_FUNCPTR with GObject
virtual functions such as get_property, set_propery, finalize and
dispose, since they'll never be used by anyone anyway. Saves a
few bytes and possibly a tenth of a polar bear.
This avoids having to do the sorting everytime we use typefind
The behaviour of gst_type_find_factory_get_list has subtlely changed
in the sense that the order was previously undefined, whereas now
it returns them sorted by rank and then by name.
The list against which we run the comparefunc will only contain
GstPluginFeature, therefore remove the 6 expensive type checks we do
for every single comparision.
When we quickly switch from PLAYING to PAUSED and back to PLAYING it's possible
in some cases that the task refuses to start, This is because when we go to
PAUSED, we unschedule the clock timeout, which could return UNSCHEDULED when
we're back to PLAYING, causing the task to PAUSE again with a wrong-state.
This patch checks if we are running when we return with an UNSCHEDULED return
value and if we are, try to create a new buffer.
Fixes#597550
The code was previously:
* checking if ret was != OK
* .. but if it was FLOW_STEP, swith it to OK
* .. and then not using ret
Instead we just make it more compact by checking if it's OK or STEP.
In most places in core and baseclasses we just need the caps to do caps-
intersections. In that case ref'ed caps are enough (no need to copy).
This patch also switches the code to use the new functions.
API: gst_pad_get_caps_refed(), gst_pad_peer_get_caps_refed()
The normal functions are always useful to have for bindings, especially
runtime-created bindings like Seed or new GObject-Introspection based
Python bindings.
Check when we need to touch the metadata of the output buffer after selecting
the output buffer so that we have everything in one place.
Also take flags and timestamp modifications into account.
When we have the same input as output caps, reuse the input caps object. After
the caps refcounting has been sorted out now, we can finally enable this
optimisation.
Without this, we risked:
* Checking the flushing state on an unexisting list
* Not setting the flushing state on pads that had just been added
Partially fixes#590056
Return FALSE in basesrc's default query handler when we get a SEEKING query for
a format that's not the one the source operates in. Previously (ie. before, in
the git version) we would return TRUE in that case and seekable=FALSE, which
is more correct, but causes backwards compatibility problems. (Before that
we would change the format of the query when answering, which was completely
broken since callers don't expect that or check for it). Since the SEEKING
query is a fairly recent addition, not all demuxers, parsers and decoders
implement it yet, in which case any SEEKING query by an application will
just be passed upstream where it will then be handled by basesrc. Now, if
e.g. totem does a SEEKING query for TIME format and we have a demuxer that
doesn't implement the query, basesrc would answer it with seekable=FALSE in
most cases, and totem can only take that as authoritative answer, not knowing
that the demuxer doesn't implement the SEEKING query. To avoid this, we make
basesrc return FALSE to SEEKING queries in unhandled formats. That way
applications like totem can fall back on assuming seekability depending on
whether a duration is available, or somesuch. Downstream elements doing
such queries are likely to equate an unhandled query with a non-seekable
response as well, so this should be an acceptable fix for the time being.
See #584838, #588944, #589423 and #589424.
Clarify byte reader docs a bit: offset is relative to the current
position of the reader, not to the start of the data. Also, the
examples in both the adapter docs and the byte reader docs have
the mask and pattern arguments swapped (see #587561). Spotted
by Carl-Anton Ingmarsson.
