The problem lies in the fact that multiqueue will now operate somewhat
similarly to the flow aggregation logic of demuxers and therefore
will stopp whenever all downstream pads return NOT_LINKED and/or
UNEXPECTED and there's no more buffers to push.
The latest commits should not affect any regular use-case, but the bug
report will be kept open so the previous behaviour can be re-established
if needed.
Fixes#609486
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
When we unblock a pad with the same user_data, the destroy callback is not
called. This leads to refcounting leaks that cannot be avoided. Instead always
call the destroy notify whenever we install a new pad block.
In particular, this fixes a nasty pad leak in decodebin2.
Also update the unit test to have more accurate comments and test the required
behaviour.
and install into a different directory $(libexecdir/gstreamer-0.10) so that
everything is versioned properly.
NOTE: run 'make clean' after updating; if you are running an uninstalled setup,
you will need to update your gst-uninstalled script (unless it's symlinked
to gstreamer core master) and exit/enter your uninstalled environment to get
the updated environment. If you are running an installed setup, you should
run 'make uninstall' before merging this change or remove the old
plugin-scanner binary manually.
Fixes#601698.
Avoid a race where a miniobject is recycled and quickly freed, which causes the
g_type_free_instance() to be called on the same object twice.
Ref the object before calling the finalize method and check if we still need to
free it afterward.
Also add a unit test for this case.
Fixes#601587
This test used to SIGBUS on OS/X but now SIGSEGV's instead on
Snow Leopard. It's not worth the effort to figure out which platform
should produce which error for what is fundamentally a pretty silly
test, so just disable it on OS/X
Reflow the code to move error handling to the end of the functions. Initialize
gvalue like we do in the setter. Add a unit-test module with two simple tests
the catche this bug.
Init variable to avoid compiler warning and make the build bot happy
(the compiler most likely complains about this because it doesn't know
here that fail_unless will abort/exit in the path where it fails).
Init variable to avoid compiler warning and make the build bot happy
(the compiler most likely complains about this because it doesn't know
here that fail_unless will abort/exit in the path where it fails).
GMP only uses "unsigned long int", which is 32 bit
on 32 bit architectures and can't hold a guint64.
This resulted in false unit test failures on 32 bit architectures.
Fixes bug #595133.
This tests 100000 random multiplications/divisions of all scaling
function variants and compares the result with the result that is
generated by GMP on the same input.
For this check for GSL and GMP during configure but only use
it for this single unit test.
Testing functions were provided by Kipp Cannon <kcannon@ligo.caltech.edu>
Make a downstream element return an error after upstream has already
put all data into queue (including EOS). As such, upstream
will not be around to pick up the error, so it is up to queue to
act appropriately. See #589991.
Note there may be downstream fatal errors (e.g. negotiation) that do
not warrant an error message already having been posted.
For now, don't show a g_warning() for empty tag strings and NULL
tags with non-git versions; we should wait for the fixes in our
plugin modules to make it into a release before we enable this
unconditionally.
