Other than saving an immense amount of 4 bytes of memory this
prevents clang from complaining and keeps the ring buffer state
in a single variable instead of two.
If downstream is operating in pull mode, short-circuit any pulls beyond
the end of the file and return FLOW_UNEXPECTED immediately instead of
sending a seek beyond the end of the file upstream, since this might
confuse upstream elements (and/or http servers, for example). Fixes
playback of apple trailers in totem and youtube/html5 clips in
WebkitGTK+.
https://bugzilla.gnome.org/show_bug.cgi?id=632977
Make sure property names passed to g_object_notify() are in the canonical form
(ie. "last-message" not "last_message"), so that g_param_spec_pool_lookup()
doesn't have to do strdup/canonicalize/free for every single notify call.
This only applies when building against older GLib versions (< 2.26).
Use more efficient g_object_notify_by_pspec() if we're compiling against
GLib >= 2.26, and also remove work-around for g_object_notify() thread-
safety issues with older GLib versions if it's not needed any more.
Use more efficient g_object_notify_by_pspec() if we're compiling against
GLib >= 2.26, and also remove work-around for g_object_notify() thread-
safety issues with older GLib versions if it's not needed any more.
Instead return after every iteration, which makes sure that the
stream lock is released for a short time after every iteration,
task state changes are checked, etc and this allows the task
to be stopped properly.
- Set reading_pos correctly in _create_read ()
- Seek to data if it is further than QUEUE_MAX_BYTES (queue) -
cur_level.bytes away. This should avoid a situation where the ring
buffer is full but the data offset from which we shall read is not in
the ring buffer.
- Only update the max_reading_pos to a lower value to protect data when
necessary
- Always signal an ADD in _locked_enqueue () so that an EOS unlocks the
reader
- More useful debug output
update_buffering () needs to be called every time we write to the ring
buffer so that applications don't get stuck waiting for a 100% buffered
message while queue2 is waiting for space
_create_write () must only be called for temp file/ring buffer cases
Cached data could have been overwritten so it is now protected until
it is read. Similarly data was overread as _have_data () was always
looking for the originally requested data even if part of it had been
read already.
Use cur_level.bytes to see how much space is free in the ringbuffer.
Simplyfy the write function, avoid taking subbuffers, move waiting for free
space in one spot, use simply counter to write data of a buffer.
- make _get_range () emit the del signal once a buffer has been read
- use do {} while (); for wait code as queue is locked and no data could
have been read in the mean time so it makes no sense to check before
waiting
- make _is_filled () more robust
Current range was being updated in the thread performing seek, but as
no locks were kept for a short section, data flow could resume before
current range updated, so data for the new range would be accepted as
from the previous range.
Rather, range should be updated in serialized manner based on
newsegment event.
Don't override the default get_times vmethod so that we can use the sync
property.
Set the default sync property to FALSE. It used to be set to TRUE but because
the get_times was NULL, it always behaved like FALSE.
Fixes#621530
Implement a custom acceptcaps function. We can simply check if there is an
intersection with the new caps. This makes the accept caps function much faster.
See #621190
When in download mode and we need to provide data for an offset that we don't
have, also perform a seek to the requested location when we are EOS. The reason
why we shouldn't wait for more data is because after EOS, there simply will be
no more data and we end up waiting forever.
Fixes#620500
Parse a size=value from the query string to specify a size. This is interesting
when reading from a file descriptor that actually has a size (and is not
stat-able, such as the socket of an http connection)
Maintain a separate variable to control src and sink flowreturn values so that
we can unlock the src part without shutting down the sink part.
Add flushing for upstream pull based elements that unblocks our getrange
function. This implements seeking when blocking for more data.
Add some arbitrary threshold before attempting a seek. Add a FIXME for this
because we need to find a sensible threshold based on the input rate.
When in download mode and the requested offset is too far away, attempt to do a
seek request to fetch the data.
Keep track of all downloaded parts and merge ranges when needed.
Fixes#600877
This changes some APIs in compatible ways:
- Some functions now take "const char *" arguments, not "char *"
- Some structs now have "conts char *" members, not "char *"
The changes may cause warnings when compiling with the right warning
flags. You've been warned.
Also adds -Wwrite-strings as a warning flag in configure.ac.
https://bugzilla.gnome.org/show_bug.cgi?id=611692
_create() is a pad function set by ourselves, therefore we're sure basesrc
is a GstFileSrc.
Speeds up _create() by 17% and the total call by 8% (instruction calls measurements
done with valgrind).
Fixes#610246
Initialize new_fd with DEFAULT_FD and fd with -1. Setting the property will set
new_fd and in _update_fd() we cehck fd against -1. Also add a coment about the
warning we get in the log from gst_poll_remove_fd(). We could get rid of the
warning if we want by tracking if fd has been added to fdset.
Typefind might mess up pads modes (pull/push) if a
downstream element is plugged and its pads activated
in 'step 2' of typefind pads activation.
This happens because the following steps don't check
if we already emitted typefound due to upstream setting
caps on buffers being pulled in the typefind helpers.
Avoid that by checking if typefound is already emmited.
Fixes#608036
When a downstream element returns GST_FLOW_UNEXPECTED we want to:
* let the dataqueue task running
* forward the flow return upstream.
This allows upstream elements to push EOS, and have that EOS event come
downstream.
Fixes#609274
When we receive an UNEXPECTED flowreturn from downstream, we must not shutdown
the pushing thread because upstream will at some point push an EOS that we still
need to push further downstream.
To achieve this, convert the UNEXPECTED return value to OK. Add a fixme so that
we implement the right logic to propagate the flowreturn upstream at some point.
Also clean up the unit test a little.
Fixes#608136
Add an option to automatically remove the temp file (TRUE by default). This
should make it possible for the application to keep the temp file by other means
than hardlinking or holding an fd open.
Fixes#607739
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.
Protect the ->removed field with the object lock as well. Take the DYN lock
earlier so that we can mark the pad removed and avoid a race in pad_alloc.
Fixes#606435
gstfilesink.c:399: error: format ‘%d’ expects type ‘int’, but argument 8 has type ‘size_t’
gstfilesink.c:399: error: format ‘%d’ expects type ‘int’, but argument 9 has type ‘gsize’
gstfilesrc.c:588: error: format ‘%08llx’ expects type ‘long long unsigned int’, but argument 8 has type ‘off_t’
It is not easy to setup a tee on the fly, thus apps need to add them always if
they might need them. This changes the code so, that if only one src-pad is
active, we push buffers directly. In the normal code path all buffers are pushed
with an extra ref, that forces followup inplace elements to copy the data.
Cast the variable to gint to conform to the printf format used.
It is casted rather than changing the format because the
message is created with a cast to gint too.
queue2 would crash when using small buffer sizes because
it would overflow when calculating the percentage, resulting
in the buffering GstMessage not being created and trying to be
used. This patch uses a gint64 instead of a gint to do the
percentage math, making it harder to overflow.
Using the current fill level of the queue, and the average input
rate, we can determine how long it will take to finish downloading
the whole stream to the temporary file.
Fixes#600726
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.
Add support for buffering mode where we post BUFFERING messages based on the
level of the queues. It currently operates on the first queue that goes over or
under the high/low thresholds.
In buffering mode we want to ignore the max visible items to decide when the
queue is filled. Instead, we only look at the number of bytes and/or time in the
queue.
Don't shadow the sq argument in the underrun_cb function but use
a different variable name to iterate the other queues.
Use the same variable name in the overrun_cb function.
Split gst_queue_locked_enqueue() into variant for buffer and event to get rid of
the if() and make the code more readable (constant boolean parameters are never
nice). Removes the if (item) checks as we dereference the pointer before anyway.
Also apply the same idea of reusing the previous knowledge in
gst_queue_locked_dequeue to remove more type checks.
We know earlier on in the code whether we're handling an event or a buffer,
just pass that information through.
This commit and the previous commit reduce instruction fetch:
* when pushing buffer (_chain) by 10%
* when popping buffer (_loop) by 3%
The task will always exist as long as its owner (i.e. the pad) and that
owner's owner (i.e. multiqueue) exist.
Reduces the number of instruction fetches by 36%.
We know whether we have a buffer or an event, use that instead of going
trough the expensive GLib typecheck.
The overall instruction fetch reduction introduced by this commit and the
2 previous commits:
* receiving a buffer (_chain) by 20%
* popping a buffer (_loop) by 14%
Numbers acquired through callgrind passing 100000 buffers through queue.
Pads have their GstSingleQueue stored as element private data
so there's no need to iterate over the list of single queues
every time. Also every pad only has a single internal link so
use a single iterator instead of a complex custom iterator.
Set the element private data of the pad to NULL when freeing the
single queue.
If downstream returns error and upstream has already delivered
everything (including EOS) and will no longer be around to find
out that we paused (and why), post error message. Fixes#589991.
When we have an input buffer with caps and when those caps are different from
the caps we want, only then make a writable copy of the input buffer as the
output buffer and set the caps on that output buffer. This avoids some cases
where we took a subbuffer for setting caps that were the same.
Users should never see the term 'file descriptor', much less a file
descriptor number, in an error message. Put that into the debug
string instead and use the default error message.
The compiler suggests to add some () to indicate if the && or the || takes
priority, so reflow code a bit so we don't have to add yet another layer
of (). Hopefully this was the intended meaning of the code.
When min-threshold is set on a queue, it is possible that one of
the minima remains unsatisfied while one of the maxima is already
reached. Therefore, always consider the queue non-empty if it is full.
Fixes#585433.
Out-of-band events might lead to us calling g_object_notify() from a
non-streaming thread, which can cause crashes if g_object_notify() is
being called from the streaming thread at the same time. See #554460.
GObject may crash if two threads do concurrent g_object_notify() on the same
object. This may happen if fakesink receives an out-of-band event such as
FLUSH_START while processing a buffer or serialised event in the streaming
thread. Since this may happen with the default settings during a common
operation like a seek, and there seems to be little chance of a timely fix
in GObject (see #166020), we should hack around this issue by protecting all
of fakesink's direct g_object_notify() calls with a lock.
Also add unit test for the above.
Fixes#554460.
Original commit message from CVS:
* gst/gstdebugutils.c:
Add some ideas, how to make the graph smaller.
* gst/gstutils.c:
Add a comment from a debug session.
* libs/gst/base/gstbasetransform.c:
Log more context.
* libs/gst/controller/gstinterpolationcontrolsource.c:
Indet.
* plugins/elements/gstcapsfilter.c:
Fix typo in docs.
Original commit message from CVS:
* docs/gst/gstreamer-sections.txt:
* gst/gstquark.c:
* gst/gstquark.h:
* gst/gstquery.c: (gst_query_new_uri), (gst_query_set_uri),
(gst_query_parse_uri):
* gst/gstquery.h:
API: Add URI query type. This is useful to query the URI
of a sink/source element and can be used by demuxers that
need to get data from other files.
This query should go upstream by default.
Fixes bug #562949.
* plugins/elements/gstfdsink.c: (gst_fd_sink_query):
* plugins/elements/gstfdsrc.c: (gst_fd_src_class_init),
(gst_fd_src_query):
* plugins/elements/gstfilesink.c: (gst_file_sink_query):
* plugins/elements/gstfilesrc.c: (gst_file_src_class_init),
(gst_file_src_query):
Implement URI query.
Original commit message from CVS:
* plugins/elements/gsttee.c: (gst_tee_handle_buffer):
Check for changed pads-list before checking the last returned
GstFlowReturn because the pad could have been removed and we
need to ignore the value in that case.
