When the FontFaceSet gets unlinked, we remove the strong pointer it holds to to
the UserFontSet.
This is not strictly necessary, since that object will no longer have any
reference to any other cycle collected object.
In any case, the loaders keep alive the user font entries, which _don't_ keep
alive the user font set (they have a weak reference instead). So if the user
font set is gone, all is bad.
Ensure we cancel all loads when unlinking rather than just when the object is
destroyed, and that the font face loader doesn't keep a reference to the user
font entry anymore after being canceled (this shouldn't be necessary either, but
it's better IMO).
Differential Revision: https://phabricator.services.mozilla.com/D18256
--HG--
extra : moz-landing-system : lando
Without the check that I'm adding in this patch, we'd violate the
"parallel writing mode" expectation of some baseline accessors
that we use in the now-guarded code. And we'd produce bogus layout
as a result.
The added assertions are just for good measure. The included testcase
causes us to fail both assertions, in a build that's missing the fix.
Differential Revision: https://phabricator.services.mozilla.com/D18715
--HG--
extra : moz-landing-system : lando
After this I can pass the document from the caller to ResolveSameStructsAs, and
get rid of the pres context pointer.
Differential Revision: https://phabricator.services.mozilla.com/D18600
--HG--
extra : moz-landing-system : lando
Summary:
Flushing it at a bad time can cancel loads whose timer / completion
handler is in progress, which makes no sense.
Reviewers: jfkthame, jwatt, heycam
Tags: #secure-revision
Bug #: 1523181
Differential Revision: https://phabricator.services.mozilla.com/D17856
This makes our behavior a bit closer to Blink / WebKit.
This patch fixes multiple issues:
First, fixes the caret movement getting stuck on a <select> element inside an
editor. This is because of the IsRootOfAnonymousSubtree() check that I'm
removing. Instead of that, consider NAC unselectable in UsedUserSelect, just
like generated content. This makes us jump across it correctly, and doesn't
regress the test-case that was added in bug 989012.
Second, it allows to select nodes with user-select: none as long as you're on an
editor. This matches WebKit and Blink. It's something you could do earlier
regardless with user-select: all on the parent, which is why the reporter's
test-case worked before my patch. I think being able to jump across these and
delete them on an editor is the right thing to do.
It adds tests for all this plus the same thing working for non-editable contents
(there was no pre-existing test for that).
Differential Revision: https://phabricator.services.mozilla.com/D18494
This makes our behavior a bit closer to Blink / WebKit.
This patch fixes multiple issues:
First, fixes the caret movement getting stuck on a <select> element inside an
editor. This is because of the IsRootOfAnonymousSubtree() check that I'm
removing. Instead of that, consider NAC unselectable in UsedUserSelect, just
like generated content. This makes us jump across it correctly, and doesn't
regress the test-case that was added in bug 989012.
Second, it allows to select nodes with user-select: none as long as you're on an
editor. This matches WebKit and Blink. It's something you could do earlier
regardless with user-select: all on the parent, which is why the reporter's
test-case worked before my patch. I think being able to jump across these and
delete them on an editor is the right thing to do.
It adds tests for all this plus the same thing working for non-editable contents
(there was no pre-existing test for that).
Differential Revision: https://phabricator.services.mozilla.com/D18494
--HG--
extra : moz-landing-system : lando
It may not be safe to handle events even when
PresShell::EventHandler::HandleEvent(). In such case, we need to discard
received events with notifying somebody. This patch move this rare case
jobs into the new method, MaybeDiscardEvent(). Then, the caller, HandleEvnet(),
becomes easier to read.
Differential Revision: https://phabricator.services.mozilla.com/D16960
--HG--
extra : moz-landing-system : lando
Doing it during layout instead. This also has the nice side-effect of
no longer needing to do a full restyle when counter-style rules are inserted.
Differential Revision: https://phabricator.services.mozilla.com/D18343
Let's move the redirection of coming event in
PresShell::EventHandler::HandleEvent() into a method. This makes the caller
easier to read.
Differential Revision: https://phabricator.services.mozilla.com/D16959
--HG--
extra : moz-landing-system : lando
It seems like we intermittently get fuzz on the clip-path-inset tests.
It's better for us to accept that fuzz than intermittently fail.
Differential Revision: https://phabricator.services.mozilla.com/D18277
--HG--
extra : moz-landing-system : lando
This patch also makes a couple of changes related to clipping:
- The composition bounds clip is applied to the async zoom container but
not its contents.
- The clip applied to the async zoom container is not divided by the
resolution. This clip is applied after the resolution, so dividing
by the resolution clips content away when zoomed in.
Differential Revision: https://phabricator.services.mozilla.com/D17176
--HG--
extra : moz-landing-system : lando
When doing bidi resolution for column-content blocks, we may still
traverse the parent chain up in RemoveBidiContinuation, reach
nsColumnSetFrame, and accidentally convert nsColumnSetFrame's
continuation into fluid ones.
Differential Revision: https://phabricator.services.mozilla.com/D17551
--HG--
extra : moz-landing-system : lando
Calling nsBlockFrame::ResolveBidi() on ColumnSpanWrapperFrame may cause
nsColumnSetFrame's non-fluid continuations being converted into fluid
ones in JoinInlineAncestors().
Since ColumnSpanWrapperFrame can only have nsColumnSetFrame and
column-span wrappers kids, no need to perform bidi resolution. This
doesn't affect column contents because they'll inherit "unicode-bidi"
from ColumnSetWrapperFrame in ua.css.
Differential Revision: https://phabricator.services.mozilla.com/D17550
--HG--
extra : moz-landing-system : lando
Next, we need to look for a frame for first parameter of calling
PresShell::HandleEvent() of another PresShell instance. This patch creates
PresShell::EventHandler::GetFrameForHandlingEventWith() to do it.
Unfortunately, the result is used in 3 patterns. One is, the caller should
stop handling the event. Another one is, the caller should keep handling
the event by itself. The other is, the caller should call
PresShell::HandleEvent() of different PresShell instance. Therefore, this
patch makes the method take aFrame of the caller. Then, the caller can check
the last 2 patterns with check the result is same as aFrame. This is not so
smart approach, but I have no better idea without adding a bool argument or
making the return type bool and adding out argument of nsIFrame.
Differential Revision: https://phabricator.services.mozilla.com/D16957
--HG--
extra : moz-landing-system : lando
When some of a border's corners have a border-radius, and that radius
is larger than the sum of the border width and element size, then it
results in the corners of the border overlapping. Webrender draws
borders by rasterizing each segment individually in to the cache, then
compositing them together. In this overlapping case, this has 2
problems:
a) we composite overlapping segments on top of eachother
b) corner segments are not correctly clipped to the curve of the
overlapping adjacent corners
This patch allows corner segments to be clipped by their adjacent
corners. We provide the outer corner position and radii of the
adjacent corners to the border shader, which then applies those clips,
if required, along with the segment's own corner clip when rasterizing
the segment.
As the adjacent corners now affect the result of the cached segment,
they are added to the cache key.
We continue to rasterize the entire segment in to the cache as before,
but now modify the local rect and texel rect of the BrushSegment so
that it only composites the subportion of the corner segment which
does not overlap with the opposite edges of the border.
Differential Revision: https://phabricator.services.mozilla.com/D16872
--HG--
extra : moz-landing-system : lando
In some cases, PresShell::EventHandler::HandleEvent() needs to call
HandleEvent() of another instance.
For retrieving the instance, we need to compute retarget document first.
This patch makes new method to retrieve it. The following patch will clean
up it.
Differential Revision: https://phabricator.services.mozilla.com/D16955
--HG--
extra : moz-landing-system : lando
There are two mochitests need to be changed. Both of contents have very large
element (5000px, 5000px), to avoid expanding the layout viewport to the large
size we restrict the minimum scale to 1.0 so that we can still check the layout
scroll range.
Also with this minimum scale size usage change, no-zoom-ref.html doesn't render
the horizontal scrollbar on _desktops_ for some reasons (presumably
reftest-async-zoom affects it, and possibly the reasons are the same as bug
1385145 or bug 1269739). Instead of fixing the issue on desktops, I am going to
take a workaround to add explicit minimum-scale value here, it somehow renders
the scrollbar on desktops too.
Note that the reftest added in this commit fails without this fix.
Depends on D18041
Differential Revision: https://phabricator.services.mozilla.com/D18042
--HG--
extra : moz-landing-system : lando
Those reftests assume that the content is automatically zoomed in/out but
unfortunately apz.allow_zooming has set to false on Android since bug 1180267.
With the preference value all of the reftests marked as 'fail' in bug 1504659
now succeed.
Though 1133905-{5,6}-v.html and 1133905-{5,6}-v-rtl.html still fail (they've
been failing before bug 1504659), they will be hopefully fixed in bug 1308702.
Depends on D18038
Differential Revision: https://phabricator.services.mozilla.com/D18041
--HG--
extra : moz-landing-system : lando
The wrench reftest suite runs reftests on the standalone webrender
project, and emits output that is compatible with the reftest analyzer.
However, the output takes a slightly different path through taskcluster
(the job runs via generic-worker rather than mozharness) and so doesn't
have the "(INFO|ERROR) -" prefix that the analyzer expects. This patch
loosens that restriction so that these logs can be parsed properly.
Differential Revision: https://phabricator.services.mozilla.com/D15577
--HG--
extra : moz-landing-system : lando
This code has gotten tweaked a number of times over the years to support
new log formats, but there's always the risk that a tweak will break old
log formats. So this patch adds a small test function that can be run by
anybody making changes to ensure old logs still parse fine. A few
current log samples are tested thusly.
Differential Revision: https://phabricator.services.mozilla.com/D15576
--HG--
extra : moz-landing-system : lando
The API to create a sticky spatial node doesn't allow us to set a clip
chain that will apply to all the contents of the sticky node. This means
that when the ClipManager sets up the clip state for the sticky node,
the clip chain for it is dropped. Everything still works currently
because the contents of the sticky node have their own clip chains whose
parent link will include the sticky node's clip chain. However, in the
next patch we're going to sever that parent link to fix other issues,
and so we will break this mechanism. This patch fixes it up by
explicitly applying the sticky node's clip chain to the stacking context
that contains all the sticky contents. This ensures all those items pick
up the clip chain.
Differential Revision: https://phabricator.services.mozilla.com/D18100
--HG--
extra : moz-landing-system : lando
But enable it in all tests because a lot of them rely on using it in the
style="" attribute for example, or in inline stylesheets, which will no longer
parse this (even in chrome documents), and we don't want to rewrite all the XUL
and XBL tests.
Differential Revision: https://phabricator.services.mozilla.com/D18027
--HG--
extra : moz-landing-system : lando
The text in the <th> element was causing intermittent fuzz due to
antialiasing. This patch removes the text to eliminate the problem.
Differential Revision: https://phabricator.services.mozilla.com/D18092
--HG--
extra : moz-landing-system : lando
When visiting a text frame with many continuations, traversing ancestors to compute the
transform to the ancestor scroll frame can become very hot. This commit changes the
algorithm to translate all the text continuations to an ancestor that can then be
transformed just once.
Differential Revision: https://phabricator.services.mozilla.com/D17730
--HG--
extra : moz-landing-system : lando
A continuation text frame's rect will be considered when visiting the primary
frame via 'FindScrollAnchoringBoundingRect', so we have no reason to compute
the same rect again if for some reason we have excluded the primary text frame.
Differential Revision: https://phabricator.services.mozilla.com/D17729
--HG--
extra : moz-landing-system : lando
PresShell::HandleEvent() treats capturing content only when received event is
related to pointing device. And it's used in 2 purposes. One is for computing
to target document of coming event. The other is for handling events using
coordinates. Therefore, if we create a helper method to retrieve it, we can
move the variable into smaller blocks.
Differential Revision: https://phabricator.services.mozilla.com/D16954
--HG--
extra : moz-landing-system : lando
Because of spinning out from PresShell::EventHandler::HandleEvent(), we can use
early-return style in MaybeHandleEventWithAccessibleCaret(). This patch
rewrites MaybeHandleEventWithAccessibleCaret() with the style.
Differential Revision: https://phabricator.services.mozilla.com/D16953
--HG--
extra : moz-landing-system : lando
We can replace it by a simple for-loop. If we want to print not only the
tag, but the detailed frame information, we can use nsFrameList::List().
Differential Revision: https://phabricator.services.mozilla.com/D17734
--HG--
extra : moz-landing-system : lando
Many of the modifications are guarded by #ifdefs. I verify them locally
by manually define them in nsBlockDebugFlags.h and nsLinelayout.cpp.
Note that I replace "mFrame" with "frame" in lines guarded by
NOISY_BLOCK_DIR_MARGINS in nsBlockFrame.cpp because they were
incorrectly renamed in Bug 1277129 Part 6a.
https://hg.mozilla.org/mozilla-central/rev/a70b04f074fc
Differential Revision: https://phabricator.services.mozilla.com/D17733
--HG--
extra : moz-landing-system : lando
There are three different APIs that serve similar purpose. Keeping only
one is sufficient.
Differential Revision: https://phabricator.services.mozilla.com/D17732
--HG--
extra : moz-landing-system : lando
Would be pretty surprising if a perspective transform scrolled stuff in an
iframe for example.
Differential Revision: https://phabricator.services.mozilla.com/D17905
--HG--
extra : moz-landing-system : lando
PresShell::HandleEvent() and PresShell::HandleEventInternal() are too big.
Additionally, we have a lot of methods used only by them. So, if we'll
split those big methods, PresShell will have a lot of small methods which
are not grouped as a part of event handling. That's too bad because some
of them may depend on the calling order, etc.
So, for grouping them, PresShell should create a stack class instance to handle
each event. Then, we can store shared information in it only while we're
handling an event.
This patch creates PresShell::EventHandler and PresShell methods become
wrappers of the stack class, but this patch does not change any logic in the
code, i.e., just reorganizing existing methods.
Note that HandleEventWithTarget() and HandleEventInternal() need to take
WidgetEvent rather than WidgetGUIEvent. Additionally, some other methods
require WidgetGUIEvent to refer WidgetGUIEvent::mWidget. Therefore, this
patch does not make the new class store the event as a member.
Differential Revision: https://phabricator.services.mozilla.com/D16951
--HG--
extra : moz-landing-system : lando
The GeckoView test app doesn't handle visited link history, so disable
a couple tests that rely on that.
Differential Revision: https://phabricator.services.mozilla.com/D17698
--HG--
extra : moz-landing-system : lando