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
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().
For GST_EXPORT define and also things like GST_DISABLE_REGISTRY.
Hopefully fixes the following build failure on cerbero-cross-mingw32:
helpers/gst-plugin-scanner.c:50: undefined reference to `_imp___gst_disable_registry_cache'
This static library gets included directly into libgstreamer-1.0.so, so it needs
the same GST_EXPORTS definition as the rest of the code that's compiled into
that otherwise it will try to find the constants it uses from gstinfo via DLL
importing (__declspec(dllimport)).
Fixes https://ci.gstreamer.net/job/cerbero-cross-mingw32/4393/
__declspec(dllexport/import) are supported by GCC and are needed for
properly generating code that fetches the values of constants from DLLs
built with __declspec(dllexport) which happens when anything using
GST_EXPORT is built with MSVC.
See: https://msdn.microsoft.com/en-us/library/619w14ds.aspx
Essentially, if you built gstreamer with MSVC and then tried to use
constants from it (such as GST_TYPE_CAPS) in a plugin, GCC would
retrieve the address of the value instead of the value itself.
This means applications and bin sub-classes can easily track when
a new child element is added to the pipeline sub-hierarchy or
removed.
Currently doesn't signal deep added/removed for elements inside
a bin if a bin is added/removed.
https://bugzilla.gnome.org/show_bug.cgi?id=578933
Make it explicit that the pad is only blocked while the callback is running,
and the pad will be unblocked again once the callback returned.
If BLOCK and IDLE behaviour is needed, both need to be used.
https://bugzilla.gnome.org/show_bug.cgi?id=766002
When doing a transition from PLAYING to PLAYING, we will fail
to forward an EOS message on the bus, and noone else will ever
send it because there'll be no actual state changed message.
Allow EOS through directly in that case.
If there is only one pad in the internal pads, when folding for
LATENCY queries it will just drop the response if it's not live.
This is maybe not the proper fix, but it will just accept the first
peer responses, and if there are any other pads, it will only take
them into account if the response is live.
This *should* properly handle the aggregation/folding behaviour of
multiple live peer responses, while at the same time handling the
simple one-pad-only-and-forward use-case
https://bugzilla.gnome.org/show_bug.cgi?id=766360
A lot of debug categories are declared in element class_init
functions, which don't get run until the element is first created
(not just registered in the plugin load function). This means
that --gst-debug-help doesn't print out a lot of categories.
Creating an instance of each element from the element factory
makes them visible, at some extra cost - 2-3 times longer, which can
be a full second or two of extra waiting. Yikes!
https://bugzilla.gnome.org/show_bug.cgi?id=741001
When activating a pad in PULL mode, it might already be in PUSH mode. We now
first try to deactivate it from PUSH mode and then try to activate it in PULL
mode. If the activation fails, we would set the pad to flushing and set it
back to its old mode. However the old mode is wrong, the pad is not in PUSH
mode anymore but in NONE mode.
This fixes e.g. typefind in decodebin reactivating PUSH/PULL mode if upstream
actually fails to go into PULL mode after first PUSHING data to typefind.
This calls a function from another thread, asynchronously. This is to be
used for cases when a state change has to be performed from a streaming
thread, directly via gst_element_set_state() or indirectly e.g. via SEEK
events.
Calling those functions directly from the streaming thread will cause
deadlocks in many situations, as they might involve waiting for the
streaming thread to shut down from this very streaming thread.
This is mostly a convenience function around a GThreadPool and is for example
used by GstBin to continue asynchronous state changes.
https://bugzilla.gnome.org/show_bug.cgi?id=760532
Passing years > 9999, months > 12 or days > 31 to gst_date_time_new() will
cause an assertion and generally does not make much sense. Instead consider it
as a parsing error like hours > 24 and return NULL.
This previously caused uninitialized memory unless something else was
initializing all the fields explicitly to something.
To be on the safe side, we also allocate metas without init function to all
zeroes now as it was relatively common.
https://bugzilla.gnome.org/show_bug.cgi?id=764902
Be notified in the application thread via bus messages about
notify::* and deep-notify::* property changes, instead of
having to deal with it in a non-application thread.
API: gst_element_add_property_notify_watch()
API: gst_element_add_property_deep_notify_watch()
API: gst_element_remove_property_notify_watch()
API: gst_message_new_property_notify()
API: gst_message_parse_property_notify()
API: GST_MESSAGE_PROPERTY_NOTIFY
https://bugzilla.gnome.org/show_bug.cgi?id=763142
Checking the current element's state when we're adding pads to
the parent element is checking the wrong thing.
Silences a 'attempting to add an inactive pad to a running element'
warning when adding a ghost pad to a running parent bin of the parent
bin of the element.
https://bugzilla.gnome.org/show_bug.cgi?id=764176
This is a useful function to automatically add ghost pads when linking
two elements across bin boundaries without know their exact parentage.
e.g. when using gst_parse_bin_from_description (with or without it ghosting pads),
one can simply retreive the src/sink pads from the bin to link to another pad.
Similar functionality is provided by gst_element_link_pads{_full}() however only
by pad name rather than by actual pads.
API: gst_pad_link_maybe_ghosting_full
https://bugzilla.gnome.org/show_bug.cgi?id=764176
Updated the GST_REFCOUNTING logging so that it includes the pointer
address of the object that is being disposed or finalized.
With this change is is then possible to match up GST_REFCOUNTING log messages
for object allocation/disposal/finalization. This can help with diagnosing
"memory leaks" in applications that have not correctly disposed of all the
GStreamer objects it creates.
https://bugzilla.gnome.org/show_bug.cgi?id=749427
It returned TRUE when regression failed, while not setting any of the out
parameters. This caused uninitialized data from the stack to be used for
setting the clock calibration.
PUSH and PULL mode have opposite scenarios for IDLE and BLOCK
probes.
For PUSH it will BLOCK with some data type and IDLE won't have a type.
For PULL it will BLOCK before getting some data and will be IDLE when
some data is obtained.
The check in hook_marshall was specific for PUSH mode and would cause
PULL probes to fail to be called. Adding different checks for the mode
to fix this issue.
https://bugzilla.gnome.org/show_bug.cgi?id=761211
We want to use the flag/enum nicks here, not only because they
are shorter but also because in case of element-specific enums
and flags we abuse the enum/flag name field for the description,
and we don't want that printed in the dot file.
https://bugzilla.gnome.org/show_bug.cgi?id=763814
Make compiler issue a warning for common beginner mistakes such as:
...
gst_buffer_make_writable (buf);
gst_buffer_map (buf, &map, GST_MAP_WRITE);
...
and similar. Only do this for some functions for now.
Don't keep the registry locked whilst iterating over the plugins
or features with a filter function. This would deadlock if the
callback tried to access the registry from the function. Instead,
make a copy of the feature/plugin list and then filter it without
holding the registry lock. This is still considerably faster than
the alternative which would be to use a GstIterator.
https://bugzilla.gnome.org/show_bug.cgi?id=756738
By default, gst_parse_launch_full() creates a GstPipeline if there's more
than one toplevel element. Add a flag to let it use a GstBin instead.
Also fix the parser to let it use this flag for GST_TYPE_ELEMENT property
values, to avoid having GstPipelines inside other GstPipelines.
https://bugzilla.gnome.org/show_bug.cgi?id=763457
When going from READY to NULL all element pads are deactivated. If
simultaneously the pad is being removed from the element with
gst_element_remove_pad() and the pad is unparented, there is a race
where the deactivation will assert (g_critical) if the parent is lost at
the wrong time.
The proposed fix will check parent only once and retain it to avoid the
race.
https://bugzilla.gnome.org/show_bug.cgi?id=761912
First load all system presets, then all from the environment variable, then
from the app directory, then from the user directory. Any one in the chain
with the highest version completely replaces all previous ones, later ones
with lower versions are merged in without replacing existing presets.
This is basically the same behaviour as before, just that GST_PRESET_PATH is
inserted as another source of directories between the system and app presets.
It was added in ca08af1f17, but was
accidentially overriding the user preset path there. Which caused inconsistent
behaviour as new presets were still stored in the system path, just not loaded
from there. Meaning you could store a new preset (in the user path), just for
GstPreset to not find it anymore later (because it only looked in the
GST_PRESET_PATH instead of the user path).
https://bugzilla.gnome.org/show_bug.cgi?id=764034
The alias define GST_TYPE_PARENT_BUFFER_META_API_TYPE is wrong and
breaks the usage of gst_buffer_get_parent_buffer_meta().
This patch fixes the GType alias and make another alias to keep the API
compatibility guarded by GST_DISABLE_DEPRECATED.
Also added a unit test.
https://bugzilla.gnome.org/show_bug.cgi?id=763112
gst_structure_new_empty() is not returning NULL in any valid scenarios,
checking for NULL here is useless. Especially because we would dereference any
NULL right after the NULL check again.
CID 1352037.
We previously check if the string ends on .class, as such strrchr() should
return something non-NULL. Add an assertion for that.
CID 1349642.
Pretty much every single element does
gst_element_class_add_pad_template (element_class,
gst_static_pad_template_get (&some_templ));
which is both confusing and unnecessary. We might just
as well add a function to do that in one step.
https://bugzilla.gnome.org/show_bug.cgi?id=762778
When holding a regular ref it will cause the GstBus to never
reach 0 references and it won't be destroyed unless the application
explicitly calls gst_bus_remove_signal_watch().
Switching to weakref will allow the GstBus to be destroyed.
The application is still responsible for destroying the
GSource.
https://bugzilla.gnome.org/show_bug.cgi?id=762552
This reverts commit b427997119.
It breaks things that used to work before, even if the change by itself is
correct and the previous code is just working around deeper bugs in the async
state change code. Let's go back to what previously worked and then fix async
state changes in general.
https://bugzilla.gnome.org/show_bug.cgi?id=760532
When calling gst_pad_activate_mode() on a ghostpad
gst_ghost_pad_activate_push_default() will be called. This will call
gst_pad_activate_mode() on the proxypad (which is internal of the
ghostpad), calling gst_ghost_pad_internal_activate_push_default(), which
again will call gst_pad_activate_mode() on the original ghostpad.
By simply returning TRUE in
gst_ghost_pad_internal_activate_push_default() the redundant call to
gst_pad_activate_mode() (for the same pad) is avoided.
https://bugzilla.gnome.org/show_bug.cgi?id=761913
When allocating a new buffer in the pool, both the do_alloc_buffer() and the
vmethod, alloc_buffer(), receive the parameter GstBufferPoolAcquireParams.
Nonetheless, when default_acquire_buffer() calls the do_alloc_buffer() it does
not pass the received GstBufferPoolAcquireParams, so when the user pass those
parameters they are ignored by alloc_buffer() vmethod.
This one-liner patch pass the received acquire params to do_alloc_buffer().
https://bugzilla.gnome.org/show_bug.cgi?id=761824
Fixes a race where an entry is set to BUSY in
gst_system_clock_id_wait_jitter() and is UNSCHEDULED before
gst_system_clock_id_wait_jitter_unlocked() starts processing it. The
wakeup added by gst_system_clock_id_unschedule() must be cleaned up.
Two stress tests are added. One test that triggers the specific issue
described above. The second stresses the code path where a wait is
rescheduled because the poll returned early.
https://bugzilla.gnome.org/show_bug.cgi?id=761586
If the clockentry is too late and is unscheduled before it gets
a change to detect its lateness the wakeup count and the poll are
used but never properly cleaned up. This leaves it in a dirty state
that is going to mess with the next clock entry waiting requests.
https://bugzilla.gnome.org/show_bug.cgi?id=761586
Change the gst_tracer_record_new() api to take the parameters the make the
spec structure directly. This allows us to own the top-level structure and
also collect the args so that we can take ownership of the sub-structures.
https://bugzilla.gnome.org/show_bug.cgi?id=760821
We're comparing a pointer type with '\0' here, which
probably isn't right, and the loop condition made sure
that classes[0] is != NULL already, so it's pointless.
Was probaby meant to check if the string pointed to is
not empty, so make it do that instead.
When choosing the first entry from the list, gst_system_clock_async_thread
must set the entry state to busy before releasing the clock lock. Otherwise
a new entry could be added to the beginning of the list and
gst_system_clock_async_thread will be unaware and keep waiting on the entry
it has already chosen.
Also improved messages about expected state and bumped them to ERROR level
to detect unexpected state changes.
https://bugzilla.gnome.org/show_bug.cgi?id=760757
self->spec is got using g_value_get_boxed(), which is a transfer none function.
So the same should not be freed, which is resulting in wrong behavior.
https://bugzilla.gnome.org/show_bug.cgi?id=760821
GstTracerValueFlags is not being initialized and the same could result in wrong
comparision and behavior. Hence initializing it to GST_TRACER_VALUE_FLAGS_NONE.
https://bugzilla.gnome.org/show_bug.cgi?id=760821
Only hide GstTracer and GstTracerRecord API behind GST_USE_UNSTABLE_API,
but don't spew any warnings, otherwise everyone has to define this
to avoid compiler warnings.
This reverts parts of commit 89ee5d948d.
We use this class to register tracer log entry metadata and build a log
template. With the log template we can serialize log data very efficiently.
This also simplifies the logging code, since that is now a simple varargs
function that is not exposing the implementation details.
Add docs for the new class and basic tests.
Remove the previous log handler.
Fixes#760267
If no providers for a particular class could be found, then removing unmatched
filters would cause all devices to be returned instead which is not at all what
the user intended. We still return 0 for unmatched filters.
Other gst libraries and/or elements may want to add some debug logging to an
external debug system or implement delayed debugging for performance reasons.
Exposes the internal __gst_vasprintf as gst_info_vasprintf which has a fallback
to g_vasprintf if the debug system is disabled.
API: gst_info_vasprintf
API: gst_info_strdup_vprintf
API: gst_info_strdup_printf
https://bugzilla.gnome.org/show_bug.cgi?id=760421
GstControlSourceGetValue() value paramater is a gdouble, not a GValue
and GstControlSourceGetValueArray doesn't return a GstValueArray but
an array of double.
https://bugzilla.gnome.org/show_bug.cgi?id=758668
Previously we used the latter one still for the tracer utility code, causing
undefined references in the resulting binary if the debugging system was
disabled but the tracer system not.
It's used by the debugging and tracer subsystem and in various files, make it
a central thing that is initialized independ of the existence of those
subsystems.
The parse-launch API automagically handles dynamic pads and performs delayed
linking as needed, without any feedback about whether the linking succeeded or
not however. If a delayed dynamic link can't be completed for whatever reason,
parse-launch will simply wait in case a suitable pad appears later. This may
never happen though, in which case the pipeline may just hang forever.
Try to improve this by connecting to the "no-more-pads" signal of any element
with dynamic pads and posting a warning message for the related outstanding
dynamic links when "no-more-pads" is emitted.
Fixes#760003
'gst_element_post_message' takes the ownership of the message, so it
shall unref it when there is no post_message implementation. Otherwise
message is leaked.
https://bugzilla.gnome.org/show_bug.cgi?id=759300
