Debugging must be configuring first (before any parsing), and then
the types are initialized at the end.
Fixes issues with debugging categories not being available at the
start
Before this patch, NLE and GES did not support NleOperations (respectively
GESEffects) that changed the speed/tempo/rate at which the source plays. For
example, the 'pitch' element can make audio play faster or slower. In GES 1.5.90
and before, an NleOperation containing the pitch element to change the rate (or
tempo) would cause a pipeline state change to PAUSED after that stack; that has
been fixed in 1.5.91 (see #755012 [0]). But even then, in 1.5.91 and later,
NleComposition would send segment events to its NleSources assuming that one
source second is equal to one pipeline second. The resulting early EOS event
(in the case of a source rate higher than 1.0) would cause it to switch stacks
too early, causing confusion in the timeline and spectacularly messed up
output.
This patch fixes that by searching for rate-changing elements in
GESTrackElements such as GESEffects. If such rate-changing elements are found,
their final effect on the playing rate is stored in the corresponding NleObject
as the 'media duration factor', named like this because the 'media duration',
or source duration, of an NleObject can be computed by multiplying the duration
with the media duration factor of that object and its parents (this is called
the 'recursive media duration factor'). For example, a 4-second NleSource with
an NleOperation with a media duration factor of 2.0 will have an 8-second media
duration, which means that for playing 4 seconds in the pipeline, the seek
event sent to it must span 8 seconds of media. (So, the 'duration' of an
NleObject or GES object always refers to its duration in the timeline, not the
media duration.)
To summarize:
* Rate-changing elements are registered in the GESEffectClass (pitch::tempo and
pitch::rate are registered by default);
* GESTimelineElement is responsible for detecting rate-changing elements and
computing the media_duration_factor;
* GESTrackElement is responsible for storing the media_duration_factor in
NleObject;
* NleComposition is responsible for the recursive_media_duration_factor;
* The latter property finally fixes media time computations in NleObject.
NLE and GES tests are included.
[0] https://bugzilla.gnome.org/show_bug.cgi?id=755012
Differential Revision: https://phabricator.freedesktop.org/D276
Read only properties will throw a GLib warning like this
when accessed with "set_child_property":
Warning: g_object_set_property: property 'text-x' of object class 'GstTextOverlay' is not writable
Avoiding all the pending_xx dance and making the code simpler.
This is now possible thanks to the various recent refactoring.
Thanks to that the user is able to set_child_property on objects
that are not in GESTrack yet, as expected.
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D739
Making it possible to create the nleobject right at the creation
of the element.
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D738
In get_property we should return the default values if
we have not created any GESTitleSource yet
(instead of segfaulting).
And fix GESTitleSource default values!
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D737
This way we have informations about the content of the
children as soon as possible.
Most code paths where already ready to handle that as we use it for
copying clips.
Fix framepositionner to properly handle that (it would have broke
with copied clips before).
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D736
Allowing pasting groups paste exactly what had been copied
And not the new version of the contained objects
This technically breaks the C API but this is a new API and I believe
and hope nobody is using it right now.
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D616
it should always have been private
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D617
This commit was causing issue where we were reporting the toplevel
element as an element but that element was actually in another
not serialized group. That is very tricky to handle for end users
as they are not guaranteed the toplevel clips were actually not
contained in another element.
This reverts commit ceb82ba302.
Reviewed-by: Thibault Saunier <thibault.saunier@collabora.com>
Differential Revision: https://phabricator.freedesktop.org/D627
ges-validate.c:237:22: error: implicit conversion from enumeration type
'GESEdge' to different enumeration type 'GESEditMode'
[-Werror,-Wenum-conversion]
GESEditMode edge = GES_EDGE_NONE;
~~~~ ^~~~~~~~~~~~~
ges-validate.c:277:41: error: implicit conversion from enumeration type
'GESEditMode' to different enumeration type 'GESEdge'
[-Werror,-Wenum-conversion]
new_layer_priority, mode, edge, position))) {
^~~~
https://bugzilla.gnome.org/show_bug.cgi?id=759758
Allowing application to force the asset system to recheck if an
asset has been "fixed" and can be used again
API:
+ ges_asset_needs_reload
Differential Revision: https://phabricator.freedesktop.org/D584
Otherwise we could have not negotiated errors in audiomixer when
the channel/channel-mask do not match
Differential Revision: https://phabricator.freedesktop.org/D493
Reviewed-by: Mathieu Duponchelle <mathieu.duponchelle@opencreed.com>
We just need to make sure they are always serialized in the right
order (which is the case) and de serializing them will lead to the
right behaviour.
We should not serialize the priority as the priority of the source
itself depends on the action having been done on the parent clip,
and we do not serialize the source priorities (and should not, GES
should just do the right thing).
Differential Revision: https://phabricator.freedesktop.org/D491
Summary:
Normally, mixing_operation is created and added to nlecomposition
as a child element so it will be freed when nlecomposition is removed
from a track.
Reviewers: thiblahute
Projects: #gstreamer_editing_services
Differential Revision: https://phabricator.freedesktop.org/D319
Summary:
Manual iteration can be replaced with foreach function.
In addition, this patch fixes mismatched GFunc type for
g_list_foreach and adds debug cateory for gst-asset for
convenient debugging.
Reviewers: thiblahute
Reviewed By: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D312
Summary: A passed GError is re-allocated when discoverer has no information.
Reviewers: thiblahute
Projects: #gstreamer_editing_services
Reviewed By: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D302
Summary:
To dispose properly, a child object should call same function
of parent class.
Reviewers: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D311
Summary:
g_file_new_for_uri never fails so GFile always has valid pointer.
And fix a bug of double unref from D303.
Reviewers: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D310
Summary:
Valgrind reports trivial leakages related to handling
objects and their converted strings.
Reviewers: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D303
Summary:
The backend commits itself automatically in these cases, so track
needs to do so too.
Reviewers: thiblahute
Reviewed By: thiblahute
Differential Revision: https://phabricator.freedesktop.org/D94
In the GESTimeline, TrackElement addition to a clip might get cancelled
(and thus the element gets removed), we need to make sure users do not
get wrong signals.
Also document the fact that user should connect to container::child-added
with g_signal_connect_after.
And fix a computation bug where we would be having mixing order
reversed between layers.
And make sure that the positionner does not mix up Transition handling
of the zorder
Summary:
Making the code simpler and handling the transition case
where elements are in the same layer (which was failing
/setting same zorders until now).
Reviewers: Mathieu_Du
Differential Revision: https://phabricator.freedesktop.org/D237
We should never let 3 objects to overlap at a same position, for that
we introduce a "rollback" feature and whenever such an editing happens,
we rollback object position to whatever it was before the move.
In case of groups, we can have track elements that do not belong
directly to the moved_trackelements but will be moved as others. Never
create transition to all object that have a start > moving group start.
Instead of trying to compute it ourself which might lead to wrong
behaviour when moving between layer.
+ Make sure that when we reset clip children priority (to make space
for effects,) we update the container knowledge of priority offsets
We were computing the priority offset taking the global MIN_NLE_PRIO
(which is a constant == 2 to make space for the mixing elements) instead
of the layer 'track element' relative priority, leading to very big
offsets on layer with a prio > 0. In the end it leaded to effects having
the same priority as the sources which leads to an undefined behaviour
in NLE.
Summary:
The user might want to render only some media type of the timeline,
for example he wants to only render the audio part of the timeline.
It was failing as we were not connecting the track but were still trying
to 'render' it.
Depends on D153
Reviewers: Mathieu_Du
Reviewed By: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D154
This means we need to properly track the layer a clip was in. We now
keep track of the various signal IDs in a dedicated structure and
keep a ref on the layer an object is in.
http://phabricator.freedesktop.org/T88
Summary:
No need to recheck if error exists since it has already been checked by the
conditional above.
Coverity CID #1302832
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D200
Summary:
We use to end up removing the nleobject when the following case happened:
* add an object
* remove that object
* re add the object
* commit the composition
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D193
Summary:
Handle the fact that some new features can be added and that means
generated files will not be fully understandable by older versions of
the formatter.
Make sure that we set the format version to 0.2 when we serialize the
GstEncodingProfile.enabled property.
Add some tests around that.
+ Fix a minor bug in the test-utils
+ Add a meta on the projects to tell in what format version a project
has been serialized/parsed back
API:
GES_META_FORMAT_VERSION
Depends on D178
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D184
Summary:
In case we just removed it from its layer, make sure to
just use the first layer when none specified.
Depends on D177
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D178
Summary:
Giving more details about the issue to the user
Depends on D151
Reviewers: Mathieu_Du
Differential Revision: http://phabricator.freedesktop.org/D176
Summary: It must only be done when the object is commited.
We can do that in constructed though, as the changes will
anyway be commited when the object is added to a composition.
Also update the tests, as we set properties spearately then
check the stop, we can commit the source at its creation without
removing meaning from the tests.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D84
Summary:
+ [API] GESTrack::commited signal.
+ [API] ges_track_commit_sync
We were emitting commited when timeline_commit was called, which
wasn't very helpful. This commit makes it so we emit commited once
all the compositions have actually been commited.
We also add a synchronous commit method to spare the user
the need to connect to the signal and wait, and update the
documentation.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D83
Summary:
Otherwise if there was still a reference to the layer when it
is removed from the timeline, it fails when the last reference
is released, because timeline_element_set_timeline calls
timeline_remove_element, which tries to remove the element from
an already disposed hashtable.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D82
Summary:
+ And freeze notifies while doing so.
We had a race with GstController which isn't MT safe, we can
fix it by propertly disconnecting signals, and making sure
no notifies are emitted while doing so.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D64
Summary: Before checking if we have a specific constructor for a track type.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D63
Summary: g-ir-scanner was erroring like crazy on the generated sources.
Reviewers: thiblahute
Differential Revision: http://phabricator.freedesktop.org/D57
And add the new element to the same layer as the last clip that
was added, insted of adding to the last layer of the timeline
(and with the current code, actually adding a new layer each time)
I got some trouble with
arc land
and I wanted to push the 3 commit coming after this revert as 3
different commits but they ended up being all squash into one single
commit, which is clearly not cool for later bisecting and blaming.
Reverting that commit and re pushing those 3 commits as they were
supposed to be.
This reverts commit 9fe15ef435.
GstStructureForeachFunc has a gboolean return value,
and the foreach function will stop unless we return
TRUE here. This meant it was potluck whether all
properties in the structure got set or not.
Fixes setting of text overlay clip text property
in particular.
https://bugzilla.gnome.org/show_bug.cgi?id=743874
Summary:
We should not restrict the CapsFeatures on the track caps.
If someone want to do such a restriction he should add it to
the restriction caps directly
Test Plan: Run testsuite
Reviewers: mathieu.duponchelle
If priv->timeline is False the function does not set any value for
timeline_duration before using it in gap_new (). Initialize the value to aviod
unexpected behaviour.
CID #1268405
The strategy here is to seek at the new end of the composition. And in
GES we always add a 1ns long gap at the end of the tracks so that all
track have the exact same duration, and we have black frames when the
timeline is empty
gpointer useless is indeed useless since we can use GST_DEBUG_REGISTER_FUNCPTR
to avoid having to store the return of the GST_DEBUG_FUNCPTR registration.
CID #1265771
They were not serialized until now.
That implies several changes:
* Override GESTimelineElement [start, inpoint, duration] properties in
GESGroup to ensure that those properties are not serialized as they
should not be.
* Rename GESBaseXmlContainer->clips field to
GESBaseXmlContainer->containers as the hashtable now contains Groups
https://bugzilla.gnome.org/show_bug.cgi?id=709148
For example if the size has been serialized in a file, but the user has
not personalized the size, we want that whenever the restriction caps
change the size, the video should take the size of the track
restriction caps.
We know need to keep track of the current positionner.size even if
setting through caps size changes.
https://bugzilla.gnome.org/show_bug.cgi?id=739527
This method can be used for subclass to override the default behaviour
for child lookup. This vmethod can be used for example in the case where
you want the name of a child property to be 'overridden'.
As an example in the GESTitleSource where we have a videotestsrc
which has a 'foreground-color' property that is used in the TitleSource
to set the background color of the title, this vmethod is now used to
tweak the name passed as parameter to rename "background" to
"foreground-backend" making our API understandable.
API:
GESTrackElement::lookup_child
https://bugzilla.gnome.org/show_bug.cgi?id=727880
+ Make sure that the TitleClip properties are not serialized anymore as
they are serialized through children properties now.
+ Enhance debugging for not serialized properties in GESXmlFormatter.
There could be a race where the new segments were pushed after
a seek on some / all pads before the operation had had its basetime
updated, and thus incoming segments were tweaked wrongly.
Reproducible with 3 clips composited and multiple seeks,
FIXME hard to validate.
Avoiding races where we would launch a seek right after a FLUSH_STOP and
before we get a Buffer which would possibly lead to ERROR message when upstream
elements try to push a buffer and check_sticky fails because downstream
is flushing.
In case we are not in a PLAYING state and the project is loaded, the
only thing that should be done is to fill the gaps and this way when the
composition get to PLAYING, their initialization will be enough to get
everything on track.
* In the action closure invokation we were alway leaking the composition.
* gst_bin_add will actually take an extra ref since we already gst_object_ref_sink so we
own the object, other call to that method will increase the refcount which means we do
not need to pass an extra ref to the bin.
* We want to ref_sink right when the object is added to the composition, making things
cleaner and simpler to follow in the tests.
Since commit 060b16ac75
"pad: don't accept flush-stop on inactive pads" in -core, the flush_stop event will not be
fowarded downstream in case the pad is not activated. In our case the element is in
READY state, so pads are deactivated. In that commit we simply make sure that the
event can be fowarded downstream
It means stop using a dedicated probe to restart task so that the main probe does not
drop the FLUSH_STOP event before we have a chance to restart the task. (and this is
for sure cleaner/and simpler to read).
Unref pads_info hash table in dispose instead of
finalize, i.e. before gst_bin_dispose runs and
destroys pads_info->bin (to which the pads_info
does not hold a ref).
