Waiting before posting calculated bitrates seems to be the
intent of the code, so avoid adding them to the tag list
pushed with the first frame.
When the threshold is reached, gst_base_parse_update_bitrates
sets tags_changed, so this posts the calculated ones right
that moment.
This prevents an insane average calculated from just the
first (key) frame from getting posted.
https://bugzilla.gnome.org/show_bug.cgi?id=768439
There must be a SEGMENT event before the GAP event, and SEGMENT events must
come after any CAPS event. We however did not produce any CAPS yet, so we need
to ensure to insert the CAPS event before the SEGMENT event into the pending
events list.
https://bugzilla.gnome.org/show_bug.cgi?id=766970
gcc 6 has problems detecting and avoiding throwing
a warning for tautological compares in macros (they
should only trigger for compares outside macros).
Avoid them with a nasty cast of one parameter to void *
https://bugzilla.gnome.org/show_bug.cgi?id=764526
This is an update on c9b6848885
multiqueue: Fix not-linked pad handling at EOS
While that commit did fix the behaviour if upstream sent a GST_EVENT_EOS,
it would break the same issue when *downstream* returns GST_FLOW_EOS
(which can happen for example when downstream decoders receive data
from after the segment stop).
GST_PAD_IS_EOS() is only TRUE when a GST_EVENT_EOS has flown through it
and not when a GST_EVENT_EOS has gone through it.
In order to handle both cases, also take into account the last flow
return.
https://bugzilla.gnome.org/show_bug.cgi?id=763770
When syncing by running time, multiqueue will throttle unlinked streams
based on a global "high-time" and the pending "next_time" of a stream.
The idea is that we don't want unlinked streams to be "behind" the global
running time of linked streams, so that if/when they get linked (like when
switching tracks) decoding/playback can resume from the same position as
the other streams.
The problem is that it assumes elements downstream will have a more or less
equal buffering/latency ... which isn't the case for streams of different
type. Video decoders tend to have higher latency (and therefore consume more
from upstream to output a given decoded frame) compared to audio ones, resulting
in the computed "high_time" being at the position of the video stream,
much further than the audio streams.
This means the unlinked audio streams end up being quite a bit after the linked
audio streams, resulting in gaps when switching streams.
In order to mitigate this issue, this patch adds a new "group-id" pad property
which allows users to "group" streams together. Calculating the high-time will
now be done not only globally, but also per group. This ensures that within
a given group unlinked streams will be throttled by that group's high-time
instead.
This fixes gaps when switching downstream elements (like switching audio tracks).
Especially if multiple threads are waiting for buffers to be available again,
the current code was wrong. Fix this and document clearly how the GstPoll is
supposed to be used.
Also fix some potential races with reading from the GstPoll before writing
actually happened.
https://bugzilla.gnome.org/show_bug.cgi?id=767979
It might happen that we popped the message before writing of the control
happened. In this case we just have to retry again a bit later, and failure to
do so will cause an additional byte in the control and the GSource /
gst_poll_wait() to always wake up again immediately.
https://bugzilla.gnome.org/show_bug.cgi?id=750397
And also mention what the expected values of errno are going to be.
write_control() will only ever return FALSE if there was a critical error. It
will never return because of EINTR, EAGAIN or EWOULDBLOCK.
read_control() will return FALSE if there was no byte to read, in which case
errno would be EWOULDBLOCK.
In all other cases there was a critical error.
https://bugzilla.gnome.org/show_bug.cgi?id=750397
On timer GstPolls it will cause the control socket state to become
inconsistent as now one less read_control() than write_control() be would
needed.
Similarly, read_control() and write_control() are only valid on timer
GstPolls.
https://bugzilla.gnome.org/show_bug.cgi?id=750397
This might fail even under correct usage, e.g. if read_control() is called
from another thread before write_control() finished in another. It has to be
retried then, or other measures have to be taken, depending on how it is used
by the surrounding code.
https://bugzilla.gnome.org/show_bug.cgi?id=750397
This addresses slightly different race conditions on Linux and Windows, and
fixes gst_poll_read_control() when control_pending == 0.
On Linux, the socketpair() used for control should not be made O_NONBLOCK.
If there's any propagation delay between set->control_write_fd.fd and
set->control_read_fd.fd, even the mutex now held will not be sufficient to
prevent a race condition. There's no benefit to using O_NONBLOCK, here.
Only liabilities.
For Windows, it's necessary to fix the race condition between testing
set->control_pending and performing WAKE_EVENT()/RELEASE_EVENT(). This is
accomplished by acquiring and holding set->lock, for both of these operations.
We could optimize the Linux version by making this Windows-specific.
For consistency with the Linux implementation, Windows' RELEASE_EVENT()
has also been made to block, although it should never happen.
Also, changed release_wakeup() to return TRUE and decrement control_pending
only when > 0. Furthermore, RELEASE_EVENT() is called only when
control_pending == 1.
Finally, changed control_pending to use normal, non-atomic arithmetic
operations, since it's now protected by set->lock.
Note: even though the underlying signaling mechanisms are blocking,
release_wakeup() is effectively non-blocking, as it will only attempt to read
from control_read_fd.fd after a byte has been written to control_write_fd.fd
or WaitForSingleObject() after it's been signaled.
https://bugzilla.gnome.org/show_bug.cgi?id=750397
GCC emits an error for this with -Werror:
plugin.c:22:1: error: 'gst_plugin_desc' initialized and declared 'extern' [-Werror]
This matches how glib does symbol exporting.
https://bugzilla.gnome.org/show_bug.cgi?id=767463
If the prototypes in the public API have dllimport in them when building
statically on Windows, the compiler will look for symbols with symbol
mangling and indirection corresponding to a DLL. This will cause a build
failure when trying to link tests/examples/etc.
External users of GStreamer also need to define -DGST_STATIC_COMPILATION
if they want to link to static gstreamer libraries on Windows.
A similar version of this patch has been committed to all gstreamer
repositories.
https://bugzilla.gnome.org/show_bug.cgi?id=767463
Ensure we do not attempt to destroy the current range. Doing so
causes the current one to be left dangling, and it may be dereferenced
later, leading to a crash.
This can happen with a very small queue2 ring buffer (10000 bytes)
and 4 kB buffers.
repro case:
gst-launch-1.0 fakesrc sizetype=2 sizemax=4096 ! \
queue2 ring-buffer-max-size=1000 ! fakesink sync=true
https://bugzilla.gnome.org/show_bug.cgi?id=767688
This patch handle the case when you have 1 pad (so the fast path is
being used) but this pad is removed. If we are in allow-not-linked, we
should return GST_FLOW_OK, otherwise, we should return GST_FLOW_UNLINKED
and ignore the meaningless return value obtained from pushing.
https://bugzilla.gnome.org/show_bug.cgi?id=767413
If we were in PAUSED, the current clock time and base time don't have much to
do with the running time anymore as the clock might have advanced while we
were PAUSED. The system clock does that for example, audio clocks often don't.
Updating the start time in PAUSED will cause a) the wrong position to be
reported, b) step events to step not just the requested amount but the amount
of time we spent in PAUSED. The start time should only ever be updated when
going from PLAYING to PAUSED to remember the current running time (to be able
to compensate later when going to PLAYING for the clock time advancing while
PAUSED), not when we are already in PAUSED.
Based on a patch by Kishore Arepalli <kishore.arepalli@gmail.com>
The updating of the start time when the state is lost was added in commit
ba943a82c0 to fix the position reporting when
the state is lost. This still works correctly after this change.
https://bugzilla.gnome.org/show_bug.cgi?id=739289
We don't do calculations with different units (buffer offsets and bytes)
anymore but have functions for:
1) getting the number of bytes since the last discont
2) getting the offset (and pts/dts) at the last discont
and the previously added function to get the last offset and its distance from
the current adapter position.
https://bugzilla.gnome.org/show_bug.cgi?id=766647
API: gst_buffer_prev_offset
API: gst_buffer_get_offset_from_discont
The gst_buffer_get_offset_from_discont() method allows retrieving the current
offset based on the GST_BUFFER_OFFSET of the buffers that were pushed in.
The offset will be set initially by the GST_BUFFER_OFFSET of
DISCONT buffers, and then incremented by the sizes of the following
buffers.
The gst_buffer_prev_offset() method allows retrievent the previous
GST_BUFFER_OFFSET regardless of flags. It works in the same way as
the other gst_buffer_prev_*() methods.
https://bugzilla.gnome.org/show_bug.cgi?id=766647
We already had a _full() version, but having that alone seems inconsistent.
Add a non-full version that mirrors the behaviour of gst_pad_link() vs
gst_pad_link_full().
- we know number of filter items is not going to change,
but compiler doesn't
- only do GST_IS_TRACER check for GObjects, not mini objects
- use non-type check cast macros in performance critical paths
It is better to have separate tests:
1) the test name will tell what is broekn when the test fails
2) we still run the other tests when one assert fails
3) the tests are easier to understand
4) we don't rely on sie effect of previous actions
5) ...
Also ix the assertion message for the name checks (Gst -> fakeobject).
