The fact that Scenario.pipeline was not accessible in a thread way lead
to the fact that all users had to take the unref the last pipeline ref
in the main thread, otherwise we were crying. This was an ugly
restriction which lead to issue when using scenario on gst-rtsp-server.
This break the API as this commit remove the GstValidateScenario.pipeline
field but it is worth it.
Since glib version 2.54, g_object_newv() is deprecated.
This patch changes that function with a simpler g_object_new(),
since no properties are set.
https://bugzilla.gnome.org/show_bug.cgi?id=782860
When replacing an action structure, also update the action name with
the (new) name from the new structure. Otherwise we end up with
a bogus name from the previous (deleted) structure.
The name of the action comes directly (i.e. not copied) from the
contained GstStructure field. Therefore make sure to take that
name from the proper structure field (copied just before) and
not from an outside one.
And let following actions to be executed (setting the action as
INTERLACED) which will make sure the track switch happened at some
point. It means the user has to set the pipeline to PLAYING so we can
make it works but we do not have choice here I think
https://bugzilla.gnome.org/show_bug.cgi?id=781213
Protect the expected seek values with the same lock as the one
that will be used to read/validate the resulting segments and flush
values.
Avoids races with duplicated seeks (i.e. a seek that was already
sent and handled via another pad, such as in demuxers).
https://bugzilla.gnome.org/show_bug.cgi?id=781112
We want to be able to do GstValidate.Monitor and not
GstValidate.ValidateMonitor.
And do not pass header to the list of sources to build libraries as
it is not needed.
In the case of h264 the stream might very well be in `nal` format but the decoder
might not accept it thus the parser converts to `byte-stream`, leading
to a correct stream detection but a failure in the validate-media-check
tool.
gst-validate-runner.c:856:7: error: implicit conversion from enumeration type 'GstValidateReportLevel' to different enumeration type 'GstValidateReportingDetails' [-Werror,-Wenum-conversion]
GST_VALIDATE_REPORT_LEVEL_UNKNOWN);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This issue was most likely introduced by the refactoring of the
position querying into a standalone function.
In execute_next_action() the rate variable was never replaced by
the current rate of the pipeline, this would result in all reverse
playback actions to trigger immediately instead of waiting for
the actual target time.
https://bugzilla.gnome.org/show_bug.cgi?id=776280
The underlying integer type for enums are implementation defined and may
not be the same size as gint/guint. So implicitly casting from pointers-
to-enum-types to pointers-to-int-types is unsafe. MSVC warns on these.
https://bugzilla.gnome.org/show_bug.cgi?id=774638
The unw_word_t type has different sizes for 32-bit and 64-bit, so using the
%lx format specifier on a 32-bit CPU leads to the following compile warning:
CC libgstvalidate_1.0_la-gst-validate-report.lo
gst-validate-report.c: In function 'generate_unwind_trace':
gst-validate-report.c:137:36: error: format '%lx' expects argument of type 'long unsigned int', but argument 4 has type 'unw_word_t {aka unsigned int}' [-Werror=format=]
g_string_append_printf (trace, "%s (0x%lx)\n", name, offset);
Cast to long so the %lx fomart specifier can be always used.
Instead of trying to parsing stdout, generate json messages and
send them over a socket so that gst-validate-launcher can properly
have informations about gst-validate subprocess execution.
Keeping negotation information around and trying to figure
out precisely why the elements could not negotied the caps
when we get a NOT_NEGOTIATED error on the bus giving the
user details about it.
My patch fixing monitor leak (15e7f1bbfd)
introduced a ref cycle between GstValidateReporter and
GstValidateReport.
The reports uses its reporter so it needs a ref on it
to ensure it's stay alive. But reports are owned by
GstValidateReporter and/or GstValidateRunner.
Fix this by not taking a reference on the reporter but instead caching
its name.
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D1029
My patch fixing monitor leak (15e7f1bbfd)
introduced a ref cycle between GstValidateReporter and
GstValidateReport.
The reports uses its reporter so it needs a ref on it
to ensure it's stay alive. But reports are owned by GstValidateReporter and/or
GstValidateRunner.
The best way I found to break this cycle is to introduce this purge
method. It's not great but the design is a bit tricky.
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D1029
A GHashTableIter is invalided if the hash table is modified while we are
iterating. Prevent this by taking the runner lock.
Fix assertion warnings with
validate.file.transcode.to_vorbis_and_vp8_in_webm.Sintel_2010_720p_mkv_srt
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D1026
Only expect fully identical stream-id from URI which are not local files
nor from our local http server.
Fixes issues with non-default http server port
gst-validate-scenario.c:183:7: error: '_gst_validate_action_type' redeclared without dllimport attribute: previous dllimport ignore
This is also declared in gst-validate-internal.h
As this is what user will expect in this case.
For example with this scenario:
set-state, state=null; playback-time=5
set-property, target-element-name=dvbsrc0, property-name=delsys, property-value=11
play;
The monitor returned by gst_validate_monitor_factory_create() was never
unreffed.
Report instances now have to keep a ref, as suggested by the TODO, as
the reporter is no longer leaked.
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D1012
The gstvalidate_debug may not be initialized like with the
validate/reporting which was crashing when run with GST_DEBUG=5.
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D1004
- the pad returned by gst_element_get_static_pad() was leaked.
- unref the pad from snode when updating it, not the pad passed as
callback to pad_added_cb()
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D958
_add_override_from_struct() could, in theory, register more than once
the same override so we should not transfer the ref.
Reviewed-by: Thibault Saunier <tsaunier@gnome.org>
Differential Revision: https://phabricator.freedesktop.org/D956
* Some SEGMENT might be updates caused by calling gst_pad_set_offset(),
which will send the same segment but with an updated offset and/or
based field. For those segments, we don't require a DISCONT on the
following buffer.
* Ignore differences in flags, they aren't relevant for now to figure
out whether the segment is an update or not
* Ignore difference in 'position', it's only meant for internal usage
by elements.
* Changes in the end position (stop in forward playback and start in
reverse playback) are considering updates
Furthermore, also expect a DISCONT flag on the first buffer following
a STREAM_START.
After a seek we need to wait for the right segment (meaning the segment
with seqnum == last seek/flush stop seqnum) to check whether the segment.time
has been properly set.
In the case and element is in READY or is going to READY state, it can
always return GST_FLOW_FLUSHING.
Avoid a race where a demuxer sinkpad has not been set to FLUSHING when we are
still processing a buffer but downstream is already FLUSHING and thus
the demuxer is already returning FLUSHING.
MAX(0, ((gint64) priv->segment_start - priv->seek_pos_tol) will be a high
positive number thanks to being interpreted as unsigned values if
segment_start < seek_pos_tol. Fix this by explicitly checking for this case
and only doing the subtraction otherwise.
This fixes the problem from fdccffbb2e
completely now.
https://bugzilla.gnome.org/show_bug.cgi?id=763602
This way we do not need the LD_PRELOAD hack anymore
Add a new libgstvalidateplugin GStreamer plugin, making sure it shares
the exact same code as the library (exposing only the wanted symbols).
Fix the way we set where to install GstValidate plugins
Try to keep backward compatibility even if tracers should never be instantiated
after an GstElement has been instantiated.
Differential Revision: https://phabricator.freedesktop.org/D459
This ensures our sink pad event wrapper is properly called if the
element implement a GstPadEventFullFunc instead of a regular one.
Removes all stray "buffer received before segment" issues with
queue/multiqueue
PTS and DTS can be deceiving as a change in segment can dramatically change
playback synchronization. Track the running-time as well to properly
get any change in synchronization
Having a default value of 0 meant that a g_idle_add loop was constantly
running, causing each test to use 100% cpu.
This is no longer required. Using a 10ms interval brings down cpu usage
to a sane value
When a file does not contain any stream info, then there is no need
to create the media info file as, it is not considered to be a valid file
and no validate checks are done for the same.
This skips unnecessary files like .txt, .dump files
https://bugzilla.gnome.org/show_bug.cgi?id=754006
The user might have scenarios specific to a particular pipeline, and the
application might have several pipelines running and scenarios that
apply on specific pipeline. We have to handle that valid use case.
Summary:
Move variable declarations in the for block so we won't try re-free
tldir in case of early short circuiting of the 'for' code.
Depends on D348
Reviewers: thiblahute
Reviewed By: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D349
Summary:
We were checking if the path was a full one but was using the
scenario_name instead of this path when trying to load the scenario.
Depends on D346
Reviewers: thiblahute
Reviewed By: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D348
writer_new_discover() API should be able to accept NULL GError and in case of
error, if GError is passed on as parameter, it should be propagated, else it
should be free'd.
https://bugzilla.gnome.org/show_bug.cgi?id=753340
If the scenario handles the states and wants to stay in PAUSED, it's not a
good idea to change the state to PLAYING when receiving BUFFERING=100%. This
caused a race condition in varios seeking tests, most often in the dash scrub
seeking test.
When message_async is not called during error cases, needs_parsing GList is
not being freed resulting in leak. Hence free'ing the same in finalize.
https://bugzilla.gnome.org/show_bug.cgi?id=753339
There is no check to see if stream info is available. This leads to
assertion error. Adding proper error messages for the same and reported
the same as a validate warning message.
https://bugzilla.gnome.org/show_bug.cgi?id=752758
When discovering the files, there will be different kind of errors. If we print
the exact message, then it will be more helpful for user. Especially in the case
of missing plugins, displaying which plugin is missing as error message
https://bugzilla.gnome.org/show_bug.cgi?id=752758
file:// base stream-id will vary depending on the file path. As we
don't expect everyone to use the same absolute path to place the
validate testsuite, the resulting stream-id changes. Because of that,
we can't match the stream-id in the recorded file, hence cannot do
further check. We work around this by doing what filesink would do,
which is compute a SHA256 of the URI which we can use to first
validate the ID is prefixed like expected, and decide if we should
consider the stream IDs the same or not.
https://bugzilla.gnome.org/show_bug.cgi?id=753079
while comparing the media descriptor with --expected-results, the return
values are not being handled properly, which results in wrong comparision
https://bugzilla.gnome.org/show_bug.cgi?id=748390
As stated in the bug, this comparison failing is not a critical
error, warning is enough. Add a comment so nobody thinks it's a
coding error.
https://bugzilla.gnome.org/review?bug=748390
when comparing tags, two conditions in if an else if are same
the correct way is to first check if both are NULL and return.
changed the condition accordingly.
https://bugzilla.gnome.org/show_bug.cgi?id=748390
Since _set_done() is meant to be thread safe,
it can not be used with g_object_add_weak_pointer(),
instead, one must use GWeakRef. But since it is in the API,
document that fact and add a couple assertions to make sure
it doesn't get broken in the future.
In case of files, which don't have duration in header, baseparse
estimates the duration only after 1.5 seconds. But Async_done event
is sent before the duration is estimated, which results in error.
If duration query fails, getting the duration from the media-info being
passed through --set-media-info. If media-info is also not set,
printing an error message and throwing error.
https://bugzilla.gnome.org/show_bug.cgi?id=752521
Summary:
When running valgrind we'll have 2 scenarios loaded (the normal one and
"setup_sink_props_max_lateness.scenario"). The loading code shouldn't assume
which one will contain the description it actually care about and so just look
for the fields it actually needs.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D199
Allowing users to decide the time between which the action should be
executed. In some cases executing on idle might lead to action not
being executed fast enough so the user might want to force an interval
in that case.
Summary:
Properly handling the different types that can represent ClockTime
Make use of it in gst_validate_action_get_clocktime
API: gst_validate_utils_get_clocktime
Depends on D209
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D210
Summary:
+ Fix a minor mixup bug between klass_overrides and name_overrides
Depends on D205
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D206
Summary:
This way we can subclass them getting a proper
context in the various override methods.
Depends on D204
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D205
Summary:
Otherwise we end up with circular / complicated dependencies between
Validate, its libraries, and the plugins
Depends on D203
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D204
Otherwise we end up with rounding error and instead of
seeking to 0.1 we seek to 0.09999999999 for example
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D203
This method is similar to runner_printf() but can be used
only once. The user needs to make sure all the pipeline
are in NULL state when this is called.
The method emits a "STOPPING" signal and at that point
overrides or monitors should do extra processing/checks if
needed.
+ Make use of it everywhere where it makes sense.
API:
gst_validate_runner_exit
GstValidateRunner::stopping signal
Summary:
Before returning GST_FLOW_ERROR, an element must post an ERROR GstMessage,
enforce that.
Reviewers: thiblahute, Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D201
Using sqrt of -1 is not valid and leads to undefined results.
When comparing the return value of the fucntion in validate-scenario,
it is being checked with ret == -1, so it makes sense to just return -1 in error case.
https://bugzilla.gnome.org/show_bug.cgi?id=748389
Summary:
It is possible to keep executing actions after the pipeline
has been destroyed.
API:
GST_VALIDATE_ACTION_TYPE_DOESNT_NEED_PIPELINE
Depends on D171
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D172
