There seem race conditions that we do a paint process when we started observing
the refresh driver but the first tick hasn't happened yet.
MozReview-Commit-ID: KRP8WR644q1
Normally the refresh driver's time changes in nsRefreshDriver::Tick, and then
DocumentTimeline::WillRefresh is called for the change. But nsRefreshDriver
sometimes updates its own time when their timer changes to the active one.
This patch lets DocumentTimeline update animations in response to the refresh
driver's time updates. And thus this patch prevents animation state and
relevant stuff inconsistencies such as an animation has been finished without
proper processes, e.g. without invalidating frame for the animation.
MozReview-Commit-ID: 42p5BWITQN0
nsDOMNavigationTiming is not thread-safe ref-counted, so it's better not to
put it into RefPtr because we might call those functions from servo side.
Actually, nsDOMNavigationTiming is held by the document, and the document
must still live in these functions.
MozReview-Commit-ID: AyZ4k749btm
The bulk of this commit was generated with a script, executed at the top
level of a typical source code checkout. The only non-machine-generated
part was modifying MFBT's moz.build to reflect the new naming.
CLOSED TREE makes big refactorings like this a piece of cake.
# The main substitution.
find . -name '*.cpp' -o -name '*.cc' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \
xargs perl -p -i -e '
s/nsRefPtr\.h/RefPtr\.h/g; # handle includes
s/nsRefPtr ?</RefPtr</g; # handle declarations and variables
'
# Handle a special friend declaration in gfx/layers/AtomicRefCountedWithFinalize.h.
perl -p -i -e 's/::nsRefPtr;/::RefPtr;/' gfx/layers/AtomicRefCountedWithFinalize.h
# Handle nsRefPtr.h itself, a couple places that define constructors
# from nsRefPtr, and code generators specially. We do this here, rather
# than indiscriminantly s/nsRefPtr/RefPtr/, because that would rename
# things like nsRefPtrHashtable.
perl -p -i -e 's/nsRefPtr/RefPtr/g' \
mfbt/nsRefPtr.h \
xpcom/glue/nsCOMPtr.h \
xpcom/base/OwningNonNull.h \
ipc/ipdl/ipdl/lower.py \
ipc/ipdl/ipdl/builtin.py \
dom/bindings/Codegen.py \
python/lldbutils/lldbutils/utils.py
# In our indiscriminate substitution above, we renamed
# nsRefPtrGetterAddRefs, the class behind getter_AddRefs. Fix that up.
find . -name '*.cpp' -o -name '*.h' -o -name '*.idl' | \
xargs perl -p -i -e 's/nsRefPtrGetterAddRefs/RefPtrGetterAddRefs/g'
if [ -d .git ]; then
git mv mfbt/nsRefPtr.h mfbt/RefPtr.h
else
hg mv mfbt/nsRefPtr.h mfbt/RefPtr.h
fi
Currently AnimationTimeline stores animations in a hashmap which means that
when we go to iterate over those animations to tick them we will visit them
in an order that is non-deterministic.
Although many of the observable effects of ticking an animation (e.g. CSS
animation/transition events, mutation observer events) are later sorted so that
the result does not depend on the order in which animations are ticked, this is
not true for in all cases. In particular, the order in which Animation.finished
promises are resolved will vary depending on the order in which animations are
ticked. Likewise, for Animation finish events.
Furthermore, it seems generally desirable to have a deterministic order for
visiting animations in order to aid reproducing bugs.
To achieve this, this patch switches the storage of animations in
AnimationTimeline to use an array instead. However, when adding animations
we need to determine if the animation to add already exists. To this end we
also maintain a hashmap of the animations so we can quickly determine if
the animation to add is a duplicate or not.
Now that DocumentTimeline observes the refresh driver we can use regular
ticks to remove unnecessary animations.
We do this because in a subsequent patch, in order to provide deterministic
enumeration order when ticking animations, we will store animations in an array.
Removing an arbitrary element from an nsTArray is O(n) since we have to search
for the array index first, or O(log n) if we keep the array sorted. If we
destroy a subtree containing n animations, the operation effectively becomes
O(n^2), or, if we keep the array sorted, O(n log n). By destroying during a
tick when we are already iterating over the array, however, we will be able
to do this much more efficiently.
Whether an animation is newly associated with a timeline, or is disassociated
from a timeline, or if it merely has its timing updated, the behavior
implemented in this patch is to simply make sure we are observing the refresh
driver and deal with the animation on the next tick.
It might seem that we could be a lot more clever about this and, for example, if
an animation reports NeedsTicks() == false, not start observing the refresh
driver. There are various edge cases however that need to be taken into account.
For example, if a CSS animation is finished (IsRelevant() == false so that
animation will have been removed from the timeline), and paused
(NeedsTicks() == false), and we seek it back to the point where it is relevant
again, we actually need to observe the refresh driver so that it can dispatch an
animationstart event on the next tick. A test case in a subsequent patch tests
this specific situation.
We could possibly add logic to detect if we need to fire events on the next tick
but the complexity does not seem warranted given that even if we unnecessarily
start observing the refresh driver, we will stop watching it on the next tick.
This patch removes some rather lengthy comments from
AnimationTiming::UpdateTiming. This is, in part, because of the behavior
described above that makes these comments no longer relevant. Other parts are
removed because the Web Animations specification has been updated such that a
timeline becoming inactive now pauses the animation[1] so that the issue
regarding detecting timelines becoming active/inactive no longer applies
since animations attached to an inactive timeline remain "relevant".
[1] https://w3c.github.io/web-animations/#responding-to-a-newly-inactive-timeline
