Always keep probing the pipeline for the current position and compare
with the latest requested seek segment to detect if the seek boundaries
are being respected
The "switch-track" command can be used to switch tracks. The "type"
argument selects which track type to change (can be "audio", "video",
or "text"). The "index" argument selects which track of this type
to use: it can be either a number, which will be the Nth track of
the given type, or a number with a "+" or "-" prefix, which means
a relative change (eg, "+1" means "next track", "-1" means "previous
track").
Conflicts:
validate/gst/validate/gst-validate-scenario.c
When used with LD_PRELOAD, the application might use various pipelines
for several different thing, we need to make it possible to spcify a
specific pipeline (or set of pipelines) on which to run the scenario.
The format is in the form of:
scenario_name:pipelinename_pattern*
For instance, "fakesrc" will return a fakesrc, not a pipeline.
This is similar to what gst-launch does, and avoids calling
pipeline API on a non pipeline object (and thus asserting).
Improve buffer timestamp range check:
* Only do it for encoders or decoders
* Audio has an acceptable tolerance of 100ms
To do this, keep track of the caps on the pad and store
if it is dealing with audio or video
Allows both src and sink pad to keep track of the current caps, but
the duplicated caps check is still only applied to sink pads as
src pads can push the same caps multiple times when it isn't linked
Makes the result a bit more readable than a compact multi-line list.
FIXME: Figure out how to print the description of the issues (which can
spawn multiple lines) in a nice way.
Tag events are hard to track and check if properly serialized because
they mutate too much inside elements. There is no reliable way currently
to match a tag event pushed into an element and another tag event
leaving the element (other than if the pointers are actually the same).
This test will find the first input selector with more than one
sink pad, and cycle through them till it gets back to the original
one. Five seconds between switches. The test checks that some data
was sent from the input selector when each of the sink pads was
selected.
We know are sequential so whenever the wanted position is passed we
should execute the action.
This avoid issue with the tolerance when we have high rate playback
Testing on source pads can lead to false positives when pads are
unlinked. The caps event is sticky and will be pushed again later
when another buffer/event is pushed, leading to an acceptable
situation to push the caps twice.
When seeking we might want to execute seeks at a playback time inferior than previous
seek, so we need to be able to define the order in which actions have to be
executed, the simplest way is to just concider that actions are always
order in the XML files.
+ Add some more debugs
Conflicts:
gst/validate/gst-validate-scenario.c
Update to have the same features as gst-validate.
1) Handle interrupts properly, with the additional of having the
'eos-on-shutdown' argument that sends EOS to the pipeline. This is
very useful for transcoding processes to finish correctly.
2) Print issues on the end of application
Adds a new expected-results argument to receive a file that is used
as a base for comparison with the new results. In case differences are
found, the application will print those issues.
They weren't waiting for the pipeline to properly change state
before sending seek events, that would cause some events to
return TRUE even if they were not handled
The tests are very simple as they only write the first error they
found during playback. If no error is set, an empty string is
printed.
The playback pipeline isn't monitored with validate monitors for now
This struct stores information about a media and tests run on it. It
also has a few helper functions that allows storing the results to a
file and loading it back.
Instead of having the file-checker object that would compare the
extracted values from the file to expected results set to its properties,
the media-info will store the values and it will be possible to compare
old media-info with new media-info from the same file. This allows
tracking improvements and regressions on different gstreamer versions.
Right now, the media-info is very tiny and doesn't store much info, only
the uri and the file size in bytes, but it will receive more additions in
the upcoming commits for storing duration, media topology, seekability and
playback information.
The GstQaRunner is now a simple aggregator of reports that it receives
from monitors and filechecker. This allows it to be used in both
scenarios without APIs that expect GstElement or Monitors, that are
only used on the pipeline monitoring QA tests.
Adds a property that triggers the file playback tests on
GstQaFileCheker. Also enable it in the gst-transcoding post file checks.
The implementation is simple, just create a playbin2 and use fakesinks
as sinks, set it to playing and wait for either EOS or ERROR messages.
Use our own custom comparison to allow to add more fine grained error
reporting. Also the encoding profile is_equal function is too strict as
it also compares profiles names, that doesn't matter to us.
This commit implementation is still initial and needs improvements as it
isn't using the restriction caps, which includes information that might not be
on the profile format caps.
It is an object that is capable to run a few file checks. The
implemented tests are: file size, duration, if the file is seekable and
comparing the file stream types with a encoding profile
The pause action instructs the pipeline to go to paused state and then
return to playing. It has the argument 'duration', that indicates the
duration for which the pipeline will remain in paused
expose gst_qa_issue_register and gst_qa_issue_new to allow applications
to register their own custom issues.
Issues IDs should use Areas higher than GST_QA_AREA_OTHER for custom
areas. And to add more issues to existing areas, the IDs should be
higher than GST_QA_ISSUE_ID_CUSTOM_FIRST.
Custom issues registering should be done at startup and from the same
thread as there is no locking around the issues hashtable
Shared objects listed in GST_QA_OVERRIDE are loaded on startup,
and the symbol gst_qa_create_overrides is run. It should create
any override needed. While it can do anything it wants, this
is discouraged.
GST_QA_OVERRIDE should be a comma separated list of shared objects,
any relative paths should be from the current working directory
at the time they are loaded (ie, if the process to be traced
changes cwd, use absolute paths).
No attempt whatsoever is made at not running what was not meant.
Includes a sample shared object for illustration purposes.
Overrides can now be registerd by gtype, meaning that they will
be attached to monitors that the target is of the requested type.
Also by element klass, that will check that the element has the
selected class in its details
It is used to iterate over overrides and modify the report level if
the overrides wants to do so.
Also adds a new mutex only for the overrides to avoid deadlocks when
reporting if we used the same lock for iterating the overrides
Reports now point to Issues, that are uniquely identified and have
translatable descriptions. This way we are going to be able to uniquely
identify the issues and applications can enable/disable checks for
specific elements.
That was never set, but it is needed for the reporter to properly add
reports to the runner.
We still keep a reference on the monitor to make things simpler
This way we can report issues from a scenario
Also add a Seek aread to the known areas list
We now need to pass the runner to the scenario instead of the
pipeline as the GstQaReporter interface needs it.
Various type of object should be able to do some reporting, so we have
to make sure all the code to do that is in one place. Creating an interface
makes it simple to share information and it avoid to have a baseclass for
something that is not actually important enough to create a baseclass.
Conflicts:
gst/qa/gst-qa-pad-monitor.c
A scenario correspond to a suite of action to execute on a pipeline,
for the time being, we only support seeking the pipeline, but in the
future we can imagine doing some queries, setting pipeline state, etc...
The scenario can be loaded thanks to the GST_QA_SCENARIO environment
variable, making it usable with any existant application, in case, the
application can be used interactively, the user should either, not load
any scenario or let the application run without interacting with it.
When handling elements that spawn multiple threads (hardware
enc/decoders), the pad monitor has to protect its variables specially
because some checks involve iterating over internally linked pads to
add/get some data for comparison (expected events, timestamp ranges,
caps).
Aside from locking its own mutex, the pad monitor can also lock the
parent's mutex when it needs to use data from its internally linked
pads. The locking order should always be parent and then individual
pad-monitor mutexes. This should prevent deadlocks when multiple
pad-monitors from the same element start doing checks at the same time
from different threads.
Only check if fields are proxied for sink getcaps as it is when
downstream restrictions should be proxied. Also improve the
fields comparison to handle single value x multi value
(list/array/range) contain relations.
Replace the macro with a more powerful variadic function that can
check for more acceptable types for the same caps.
This removes a few more false positives
flush events shouldn't fail, so we don't need to rollback when it
returns false from downstream (this is common when downstream is still
not-linked) and it would cause gst-qa to spit false positives.
Also refactor the common event handling for both sink and src event
functions into a common place. Currently we handle flushes the same
for both pad's directions
Keep the full range stored by the element in the monitor and check
if outgoing timestamps are within that range. It is simple and
should generally work.
