A previous patch moved CanPerformOnCompositorThread to
CommonElementAnimationData including a FIXME saying that active layer
notification should happen at call sites. Now that the code for
GetAnimationsForCompositor is common, we can do the active layer notification
there.
This patch still leaves ElementAnimations|
ElementTransitions::GetAnimationsForCompositor as shortcuts
for the method now defined on CommonElementAnimationData.
This patch moves HasAnimationOfProperty to CommonElementAnimationData. It also
takes the chance to start removing some redundancy from nsLayoutUtils
/ ActiveLayerTracker. Some of this should never have been added in the first
place and some could have been removed earlier on but while we're fixing up
HasAnimationOfProperty it seems like an appropriate time to fix up its call
sites too.
Also, since HasAnimationOrTransition actually returns an object, not a bool, we
this patch renames it to GetAnimationsOrTransitions.
In a number of places in nsAnimationManager we have the following sequence of
calls:
CommonElementAnimationData::EnsureStyleRuleFor
ElementAnimations::GetEventsAt
nsAnimationManager::CheckNeedsRefresh
nsAnimationManager::EnsureStyleRuleFor already does exactly that so we should
just reuse it.
At the same time we rename EnsureStyleRuleFor to UpdateStyleAndEvents since
that's a bit more accurate. It's also confusing to have two methods of the same
name (but on different objects) that don't exactly correspond in terms of
the scope of what they do.
Both ElementAnimations and ElementTransitions have an EnsureStyleRuleFor method.
The ElementAnimations version is a more general of the ElementTransitions one
with the exception that the ElementTransitions version checks for finished
transitions. This patch moves the code from ElementAnimations to
CommonElementAnimationData with one minor change: adding the checks for finished
transitions. The ElementTransitions version is removed.
Since the ElementAnimations version contains a second parameter, aIsThrottled,
callers of ElementTransitions must include this extra parameter. In
a subsequent patch we add an enum for this parameter to make call sites easier
to read.
The ElementAnimations version also sets the mNeedsRefreshes member so at the
same time we move mNeedsRefreshes to CommonElementAnimationData. Furthermore,
since the ElementAnimations version which we have adopted returns early if
mNeedsRefreshes is false, this patch ensures that when we call
EnsureStyleRuleFor from ElementTransitions::WalkTransitionRule, we set
mNeedsRefreshes to true first.
Another difference to account for is that the ElementTransitions version of
EnsureStyleRuleFor *always* sets mStyleRule (even if it doesn't add anything to
it) where as the ElementAnimations version only creates the rule when necessary
so we need to add a check to ElementTransitions::WalkTransitionRule that
mStyleRule is actually set before using it.
Now that an animation's delay is part of AnimationTiming--the struct we pass to
GetComputedTimingAt--it makes sense to act on it in GetComputedTimingAt.
This also happens to bring the procedures here closer to the algorithm
definitions in Web Animations.
As part of this refactoring, this patch converts ElementAnimation::IsRunningAt
to use GetComputedTiming since the previous approach no longer works now that
GetLocalTimeAt (nee ElapsedDurationAt) no longer handles delays. This also
removes duplicated logic.
Also, previously ElapsedDurationAt would assert if called on a finished
transition since TimeDuration's - operator wouldn't like the null mStartTime.
This patch adds an assertion for this case to GetLocalTimeAt to ease debugging.
One of the main differences in handling a list of transitions vs a list of
regular animations is that when we are dealing with a list of transitions we
need to check for transitions that have finished and are about to be discarded
but need to be retained temporarily to provide correct triggering of subsequent
transitions. Such transitions are marked as "removed sentinels" and are ignored
for most operations.
This patch moves the methods for setting and checking such transitions to the
base class ElementAnimation so that we can treat animations and transitions
alike without having to downcast or do obscure checks for mStartTime.IsNull()
(which equates to checking if the animation is a "removed sentinel" but is not
particularly clear).
In the process, this patch renames said methods to Is/SetFinishedTransition
since hopefully that is a little easier to understand at a glance.
This patch is the first part in preparing the way to merge ElementTransitions
with CommonElementAnimationData (which we'll eventually rename to something
nicer).
Here we move mTiming from CommonElementAnimationData to the AnimationTiming
struct. While this is not strictly necessary in order to do the later
refactoring it makes it simpler since it:
- Divides time calculation into calculation based on dynamic play state (the
responsibility of animation players in Web Animations terms) and static
author-specified timing parameters (a property of animations in Web Animations
terms).
- In future we will probably put animations on the compositor during their
delay phase so we will want the delay to be present in the AnimationTiming
struct then.
- Makes AnimationTiming line up with the dictionary of the same name in Web
Animations.
This is in preparation for nsStyleCoords becoming non-POD types. The
use of memcpy is a bit of a trap when modifying the style structs, and
it likely has marginal value now that anything with an nsStyleCoord
can't be memcpy-initialized.
In addition to moving the code, this patch also:
* changes the square character from U+25AA to U+25FE to match the spec
* changes all the generating functions from using append to using assign
* removes negative handling from CJKIdeographicToText since it is no longer used there
* fixes indentation and naming conventions
* changes buffer size in DecimalToText
This patch removes the check that skipped queueing events for animations
without keyframes since the spec indicates such animations should dispatch
events.
There is a further correctness fix here for the case where a keyframes rule
is modified using the CSSOM so that it becomes empty. Previously when we
came to create the new animation rules we would end up setting
ElementAnimations::mNeedRefreshes to false since we check if the keyframes
rule is empty and if it is we would skip all further processing (including
setting mNeedsRefreshes).
That means that:
(a) We may end up unregistering from the refresh observer so we would never
dispatch the end event for such an animation.
(b) If the animation was running on the compositor we may never remove it from
the compositor or may not do it in a timely fashion.
To fix both these problems, this patch removes the check for an empty keyframes
rule so that mNeedsRefreshes is set in this case.
This patch adjusts the tests for transform transitions that run on the
compositor to handle transitions that begin with a non-invertible transform.
In this case the first sample at the start of the animation won't create
a layer because in nsDisplayTransform::BuildLayer
/ FrameLayerBuilder::BuildContainerLayerFor we'll skip creating the layer once
we notice the equivalent matrix is singular.
In this patch we detect that case and force an extra sample betwee 0 and 200s at
100s. This means the layer will be created at t=100s and be available for
querying at the next sample.
Similar issues could occur, for example, if the transforms at both t=0s
and t=100s are not invertible but currently that doesn't occur. We can add
handling for that if and when it becomes necessary.
This patch takes the existing tests for transitions running on the compositor
and makes them re-use the same test utility methods as used for testing CSS
Animations that run on the compositor. This means these tests now also check
that the transition is in fact running on the compositor when it is expected to.
It seems the big_omta_round_error is no longer needed so I've removed that.
The test that begins with "skew(45deg, 45deg)" currently fails so it is skipped
here. This is addressed in the next patch in the series.
This patch simply re-arranges the order of arguments to omta_is_approx so that
the delta sits along side the values being compared.
This, I think, makes more sense and also is more consistent when converting
tests from test_animations.html to test_animations_omta.html since the
"RunningOn" parameter is consistently inserted in the second-to-last position
just before the description for both omta_is and omta_is_approx.
This patch removes the line from new_div that forced a style flush. This was
very confusing because:
* It behaved differently to new_div in test_animations.html so copying tests
over was more complex (particularly when registering for events is involved).
* It meant after setting up initial style using new_div you could just call
waitForPaints but if you updated style using elem.style you'd need to call
waitForPaintsFlushed.
In adjusting test_animations_omta.html we are able to simplify the tests
somewhat. This patch also adds a few additional checks that waiting to update
the compositor does not produce different results.
This patch moves some test utility methods from test_animations_omta.html to
animation_utils.js. It also renames addAsyncTest to addAsyncAnimTest and
likewise for a few other methods.
This patch adjusts GetComputedTimingAt to set the time fraction and current
iteration fields of the output computed timing correctly for animations with
zero iteration duration. Care must be taken to handle cases such as animations
that have zero duration but repeat infinitely.
The code is significantly re-arranged to more closely align with the naming and
algorithms defined in Web Animations.
A couple of tests in test_animations.html have been tweaked to account for
floating-point error. This is not because the new code is less precise but
actually the opposite. These tests fall on the transition point of step-timing
functions. The new code uses the closest possible floating-point representation
of these times which happens to cause them to fall on the opposite side of the
transition point.
For example, in evaluating a point 3s into a reversed interval the old code
would give us an intermediate time fraction of:
0.29999999999999982
When we reverse that by subtracting from 1.0 we get: 0.70000000000000018
With the code in this patch we get an intermediate time fraction of:
0.29999999999999999
When we reverse that by subtracting from 1.0 we get: 0.69999999999999996
Hence we fall on the opposite side of the transition boundary.
This patch also makes ElementAnimation::ActiveDuration a static method that
takes timing parameters as an argument. This is so that this method can be
used within ElementAnimations::GetComputedTimingAt (a static method) in a
future patch.
We could also make ActiveDuration() a method of AnimationTiming. I suspect
this logic belongs together in ElementAnimation however.
In a future patch we could also add the active duration to the ComputedTiming
struct which would simplify the only other place this is currently used
which is ElementAnimations::GetEventsAt.
viewport-units-rounding-1.html fails without the patch and passes with
the patch.
viewport-units-rounding-2.html fails with an early version of the patch
but not without the patch or with the final version.
Also shuffle the initialization of members in
nsAnimationManager::BuildAnimations to roughly match the order in which they
are declared (with the exception that mPlayState needs to be set before calling
IsPaused() which is used to set mPauseStart).
This was only needed when we were inspecting the returned time fraction but now
that we inspect the phase it's not necessary to force the fill mode to "both".
This patch simply moves the code from ElementAnimations to ElementAnimation so
that it can later be used in transitions code and so we can later move
EnsureStyleRuleFor to ElementAnimation.
This patch shuffles the code in ElementAnimations::GetEventsAt to make it easier
to follow.
It also removes a check for whether or not the animation is paused.
Previously we would not dispatch events if the animation was paused and in its
active phase (but we would if the animation had finished). There doesn't seem to
be any reason for this. If the animation was paused between the last sample and
the current sample and the boundary of an iteration also occurred in that time
then I expect we should dispatch that event. Removing this check for the pause
state does not cause any tests fail.
Separating out the event logic here makes it clear that we do not dispatch start
events in the situation where one sample falls before the active interval and
one sample falls after it (filed as bug 1004361). This patch adds a comment to
this effect.
This patch simply shifts the event-related code from GetPositionInIteration to
GetEventsAt. Although there are simplifications that could be done to
GetEventsAt, they are deferred to a subsequent patch so as not to obscure the
translation of code from one function to another.
As a result of moving event-related handling from GetPositionInIteration it no
longer needs to support different main-thread vs compositor modes.
This patch makes ElementAnimations::GetPositionInIteration return
a ComputedTiming object instead of just a time portion (time fraction).
Since the ComputedTiming object includes phase information, we can fix those
parts of EnsureStyleRule and GetEventsAt that were temporarily using the time
portion to guess if the animation might have finished or not.
This patch moves the FillsForwards/FillsBackwards methods previously defined on
ElementAnimations to the structure contain the fill mode: AnimationTiming. It
also changes GetPositionInIteration to use these methods.
Introduces a struct to store timing parameters for passing to
GetPositionInIteration. In future this struct is expected to be expanded to
include other timing parameters as well (based roughly on Web Animations'
"Timing" interface, hence the name AnimationTiming).
This patch moves event queuing out of EnsureStyleRuleFor into a separate method.
This is a preparatory step towards making GetPositionInIteration into a more
generic method for calculating the current time fraction.
In order to achieve this, GetPositionInIteration needs to be able to calculate
the correct time portion for times outside the range [0, 1] even when it is not
passed a ElementAnimation object. Specifically, it needs the fill mode of the
animation to be passed in.
(Rather than using FillForwards/FillBackwards this patch just compares the
NS_STYLE_ANIMATION_FILL_MODE_* values directly but FillForwards/FillBackwards
are restored in a subsequent patch when they are added to the struct used to
lump the timing parameters together.)
There are a number of places where positionInIteration is used to determine if
the current sample occurs in the active phase or after. This is sub-optimal but
is fixed in a subsequent patch in this series.
The actual work of removing event queuing from GetPositionInIteration is
deferred to a subsequent patch in order to keep the changes as small as
possible. This patch simply makes separate calls to GetPositionInIteration for
interpolating and for event queuing.
This patch adds tests for triggering animations based on the length of the
animation-name property as well as tests for dynamic changes to style rules.
These tests are based on tests in test_animations.html but for directed at
animations that run on the compositor thread.
This patch adds tests for the cascanding of keyframes rules based on those in
test_animations.html but for animations that run on the compositor thread.
This patch adds tests for the interaction of animation and restyling (Bug
686656) based on those in test_animations.html but for animations that run on
the compositor thread.
The implementation here current expects BOTH the following to fail:
- The comparison between the OMTA value and the expected value
- The comparison between the OMTA value and the computed value
This generally tends to be the case since the computed value and expected value
normally match unless we have a bug that affects all CSS animations. If we need
to mark tests where the computed value is also wrong we'll need to modify the
behavior here at that time.
This patch also applies this new function to the author !important test that was
previously commented-out because it currently fails.
This patch also ensures that when we have an animation running on the compositor
when it should not that we still compare the values produced on the compositor
and on the main thread so that the visual result is correct even if the
performance characteristics are not.
This patch adds tests for the handling of author !important rules and animations
based on similar tests in test_animations.html but for animations that run on
the compositor thread.
Due to bug 847287, these tests don't pass and are partly disabled. Subsequent
patches add todo_is tests for this.
This patch adds a version of tests in test_animations.html for bug 651456 which
covers multiple samples with the same timestamp. The version here, however,
tests animations that run on the compositor thread.
This patch adds a version of tests in test_animations.html for keyframe
animations with multiple properties where some properties are present in only
some keyframes. The version here, however, tests animations that run on the
compositor thread.
This patch adds a version of tests in test_animations.html under the heading,
"css3-animation: 3.8 The 'animation-delay' Property", for animations
that run on the compositor thread.
This patch adds a version of tests in test_animations.html under the heading,
"css3-animation: 3.7 The 'animation-play-state' Property", for animations
that run on the compositor thread.
This patch adds a version of tests in test_animations.html under the heading,
"css3-animation: 3.6 The 'animation-direction' Property, for animations
that run on the compositor thread.
This patch addresses and issue where the OMTA style and computed style were not
comparing equal in one particular case.
In this case AddTransformTranslate in nsStyleAnimation would give us
a translate-y value of 94.331673 in both cases (i.e. when calculating the
animated value on the compositor thread or when fetching computed style).
For the OMTA case, however, after we apply additional layer transformations and
then reverse them (so we can query the CSS value) we'd end up with 94.331642,
a difference of 0.000031. The reversing procedure is only used for testing so
the actual error introduced here by the additional layer transformations is
probably less.
Unfortunately, when we pass 94.331642 this along to MatrixToCSSValue we get back
matrix(1, 0, 0, 1, 94.3316) since it only outputs 6 digits of precision.
On the other hand, on the computed style end we'd pass 94.331673 to
MatrixToCSSValue which gives us matrix(1, 0, 0, 1, 94.3317), so the error swells
from 0.000031 to 0.0001.
Then when we subtract 94.3316 from 94.3317 in Javascript we get
0.00010000000000331966 due to floating-point precision issues which compares
greater than the default tolerance of 0.0001.
This patch simply adjusts the default tolerance to 0.00011 to accommodate
these floating-point differences.
This patch adds a version of tests in test_animations.html under the heading,
"css3-animation: 3.5 The 'animation-iteration-count' Property" for animations
that run on the compositor thread.
These tests surface an issue where in some cases precision errors lead to
discrepencies between the OMTA style and computed style. This is fixed in
a subsequent patch in this series.
This patch adds a version of the tests in test_animations.html under the
heading, "css3-animations: 3.2. The 'animation-name' Property" that tests the
same features when animations are running on the compositor thread.
The test harness code for normalizing transform inputs to a standard form for
comparison fails to detect the case where the input is a string such as
{ tx: "20px" }
instead of:
{ tx: 20 }
When we go to compare matrix components we fail if:
Math.abs(a.comp - b.comp) > tolerance
But if a.comp or b.comp is a string, we'll get NaN on the LHS and
"NaN > tolerance" will return false so we'll skip the failure handling and
continue onto the next component. That means if we have input { tx: "30px" } and
we get "20" as the x-translation component we'll pass the test.
This patch fixes this condition to check for isNaN.
We *could* also just drop a few .map(parseFloat) calls into
convertObjectTo3dMatrix and convertArrayTo3dMatrix to ensure "20px" becomes 20
but there may be situations where that masks bugs (since "20px" and "20em" turn
into the same thing) so for now this minimal fix should be enough.
This patch converts the tests in test_animations.html under the heading,
"css3-animations: 3.1. Timing functions for keyframes" to an equivalent version
for testing animations that run on the compositor thread.
These functions get invoked as event listeners, so we'll automatically get the
proper |this|. The reason for the existing shenanigans was to work around
bug 872772, which has now been fixed.
As a result, transitions are now stored using a pointer to the base class,
mozilla::ElementAnimation. We downcast to a transition only when necessary. No
error-checking of the result of AsTransition is performed since we only ever
call it on the mAnimations member of ElementTransitions.
Add a method for downcasting from an ElementAnimation to an
ElementPropertyTransition (when the underlying object is an
ElementPropertyTransition).
This, unfortunately, adds a vtable to ElementAnimation but in the long term
I hope we will be able to isolate transition-specific code to a specific kind of
TransitionEffect that hangs off ElementAnimation and put the vtable on
AnimationEffect instead. (The AnimationEffect concept is part of the Web
Animations API.)
We currently have mozilla::StyleAnimation as well as nsStyleAnimation. This
patch renames StyleAnimation back to ElementAnimation.
Although ElementAnimation is very similar to ElementAnimations, in the near
future we expect to retire ElementAnimations and replace it with a common
AnimationSet-like structure that is covers the features of ElementAnimations and
ElementTransitions.
This patch takes StyleAnimation and makes it ref-counted heap object. This
should allow us to store StyleAnimation and its subclasses (transitions only
currently) in a consistent fashion (an array of base-class pointers).
Furthermore, this will be helpful if we want these things to be pointed to
from Javascript objects that may, for example, preserve their lifetime beyond
that of the element that currently owns them.
This patch also introduces a typedef for an array of refptrs to StyleAnimation
objects (and similarly for the subclass ElementPropertyTransition) to simplify
the code somewhat.
- OverflowChangedTracker::AddFrame now accepts an enumerated type parameter to
indicate if the overflow areas of children have changed (CHILDREN_CHANGED),
the overflow areas of the children have changed and the parent have changed
(CHILDREN_AND_PARENT_CHANGED), or if only the transform has changed
(TRANSFORM_CHANGED).
- OverflowChangedTracker::Flush no longer falls back to calling
nsIFrame::UpdateOverflow when a frame lacks a PreTransformOverflowAreas
property.
- Added an additional change hint, nsChangeHint_ChildrenOnlyTransform, which
results in TRANSFORM_CHANGED being passed in to
OverflowChangedTracker::AddFrame.
- In nsIFrame::FinishAndStoreOverflow, the passed in overflow is now stored as
the InitialTransformProperty for elements that are IsTransformed().
- Partially corrected Bug 926155, by only calling
OverflowChangedTracker::AddFrame on parents of the sticky element during
StickyScrollContainer::UpdatePositions, using CHILDREN_CHANGED.
This patch was mostly generated with the following command:
find . -name "*.h" -o -name "*.cpp" | xargs sed -e '/WrapObject(JSContext/ {; N; s/\(WrapObject(JSContext *\* *a\{0,1\}[Cc]x\),\n\{0,1\} *JS::Handle<JSObject\*> a\{0,1\}[sS]cope/\1/ ; }' -i ""
and then reverting the changes that made to
dom/bindings/BindingUtils.h, since those WrapObject methods are not
the ones we're trying to change here, plus a bunch of manual fixups
for cases that this command did not catch (including all the callsites
of WrapObject()).
This patch was mostly generated with this command:
find . -name "*.h" -o -name "*.cpp" | xargs sed -e 's/Binding::Wrap(aCx, aScope, this/Binding::Wrap(aCx, this/' -e 's/Binding_workers::Wrap(aCx, aScope, this/Binding_workers::Wrap(aCx, this/' -e 's/Binding::Wrap(cx, scope, this/Binding::Wrap(cx, this/' -i ""
plus a few manual fixes to dom/bindings/Codegen.py, js/xpconnect/src/event_impl_gen.py, and a few C++ files that were not caught in the search-and-replace above.
- OverflowChangedTracker::AddFrame now accepts an enumerated type parameter to
indicate if the overflow areas of children have changed (CHILDREN_CHANGED) or
if the transform has changed (TRANSFORM_CHANGED).
- OverflowChangedTracker::Flush no longer falls back to calling
nsIFrame::UpdateOverflow when a frame lacks a PreTransformOverflowAreas
property.
- Added an additional change hint, nsChangeHint_ChildrenOnlyTransform, which
results in TRANSFORM_CHANGED being passed in to
OverflowChangedTracker::AddFrame.
- In nsIFrame::FinishAndStoreOverflow, the passed in overflow is now stored as
the InitialTransformProperty for elements that are IsTransformed().
- Partially corrected Bug 926155, by only calling
OverflowChangedTracker::AddFrame on parents of the sticky element during
StickyScrollContainer::UpdatePositions, using CHILDREN_CHANGED.
Tests for animation list handling on the compositor thread. Some checks are
currently marked todo_is because they depend on Bug 980769 in order to pass.
This patch takes the compareTransform utility methods and makes them more
generic so they can be used for testing opacity too (the other property
currently animated on the compositor thread).
The naming omta_is and omta_is_approx is intended to mirror is and is_approx in
test_animations.html.
This patch adds an additional mochitest for specifically targetting CSS
Animations that run on the compositor thread. The content of the test mimicks
test_animations.html but using properties whose animations are expected to run
on the compositor thread.
Since off-main thread animation (OMTA) is not available on all platforms we
define a common wrapper function that runs OMTA tests only when available. This
patch further performs an internal check of basic OMTA operation so that
only a single error is produced if OMTA is unexpectedly unavailable.
Typical usage is:
SimpleTest.waitForExplicitFinish();
runOMTATest(function() {
... test code ...
SimpleTest.finish();
},
SimpleTest.finish);
This can be easily wrapped with promises if needed but does not require using
promises.
The calls to waitForExplicitFinish and finish are not performed automatically
since this function may be integrated with test suites that do other work
outside the call to runOMTATest.
Two comments in AnimationCommon.h refer to 'mFlushCount' which was presumably
the old name for mAnimationGeneration. Also, one comment says
nsCSSFrameConstructor tracks this. This patch adjusts the comments to refer
to mAnimationGeneration and RestyleManager.
(The reference to nsTransitionManager::UpdateAllThrottleStyles() is still valid
since there is useful documentation accompanying that method despite the fact
that the relevant code is mostly contained in AnimationCommon.h since bug
914847. Eventually we will unify the structures of transitions and
animations to the the point that we can replace the
IMPL_UPDATE_ALL_THROTTLED_STYLES_INTERNAL macro in AnimationCommon.h with an
actual method. At that point we can move the documentation accompanying
nsTransitionManager::UpdateAllThrottleStyles and its references to
AnimationCommon.)
We need a basic representation of animations from which we can derive subclasses
to represent specific cases such as transitions. For now we will retrofit
ElementAnimation for that purpose hence renaming it to StyleAnimation.
This patch removes the "using namespace mozilla::layers" line from
AnimationCommon.cpp since the unified build system concatenates several files
together before compiling making using declarations like this leak into other
files potentially creating ambiguities. Previously, when we were calling
ElementAnimation, 'Animation', there were ambiguities between
mozilla::layers::Animation and this new 'Animation' class. In general, it is
probably a good idea to limit the scope of these using declarations so I've kept
that change.
