This adds fallbacks for:
* synthetic bold
* synthetic italics
* text-writing-modes
This also removes an old hack to make synthetic italics less broken.
This also prevents special opacity handling for color fonts so that webrender
gets that information.
MozReview-Commit-ID: DKiTUBR6hzy
--HG--
extra : rebase_source : 22b445d40ee46bb09d4325828a8e959d3da7a9c7
By returning true from WebRenderLayerManager::EndEmptyTransaction, we
avoid doing a full paint in cases where the caller decides an empty
transaction would be sufficient. WebRenderLayerManager already rejects
attempts to set some forms of empty-transaction data (specifically
transform and scroll offset updates). This means that we will never get
a call to EndEmptyTransaction where the caller is expecting a transform
or scroll offset update to be sent over to the compositor. So if we have an
implementation of EndEmptyTransaction that ignores that data, it will not
break expectations.
There is still one piece of information that WebRenderLayerManager
doesn't reject in this manner, the APZ focus state. That is, if the
layout code sets a pending APZ focus state on the WRLM, followed by a
all to EndEmptyTransaction, it expects the focus state to get propagated
to the compositor. This patch makes sure that it does happen by using
the new API added in the previous patch.
MozReview-Commit-ID: 596UgW9ZWAF
--HG--
extra : rebase_source : e0f4f201a76747d6e29cde5da26fe760fd7f770b
Between this pair of gfxContext::Save/Load, only the draw calls of the
DrawTarget are called. Since draw calls do not alter the content of a AureState,
we should just remove this pair of Save/Load.
MozReview-Commit-ID: BexjGJkO77E
--HG--
extra : rebase_source : eda2795fb8993d578d6ecdca6df1c39fcc4073f0
Even if Stylo on Android, the following is successful.
- unit-vh-vw-overflow-scroll.html
- unit-vh-vw-overflow-scroll-x.html
- unit-vh-vw-overflow-scroll-y.html
- Although we use fails-if(stylo|styloVsGecko), since last condition is
fuzzy-if(Android,...), android's reftest uses fuzzy-if unfortunately.
We should modify fuzzy-if's order to pass reftest for Stylo on Android.
- input-transition-1.html
MozReview-Commit-ID: 12mRwJW67Nz
--HG--
extra : rebase_source : 1ffcfe7024846b65e3f0e5d9ad8a1053e66d9c86
This is for implementing shape-outside: <image> later.
MozReview-Commit-ID: 93TmLecRjRx
--HG--
extra : rebase_source : c2b39b1e12ef96e80a3ceb0a800767d379e6e6bb
After StyleBasicShape is set to StyleShapeSource, it's life cycle never go
beyond StyleShapeSource, so I make StyleBasicShape hold by a UniquePtr in
StyleShapeSource.
Also, replace all raw pointers to StyleBasicShape by UniquePtr in all APIs.
MozReview-Commit-ID: 1MfIFjP8TsQ
--HG--
extra : rebase_source : bdbbd47de57e0bc610e37913752ab8413f62588a
The original operator==() (implemented by EqualsInternal<true>) is not been
used. Therefore, I expand EqualsInternal<false> into it, and move it to
nsStyleStruct.cpp.
Also, use DefinitelyEqualURIs() in nsStyleStruct.cpp in operator==().
MozReview-Commit-ID: HccwKvzQHR
--HG--
extra : rebase_source : f63dac121e13fd9c6dccf7f0c6e870bd75d9e03a
We exclude these testcases from being tested under Gecko, because they fail some
mochitest checks for parse+serialize idempotence. The failures happen because of
a combination of these factors:
(a) Gecko's style system is inconsistent about whether an explicitly-specified
(but default) gradient direction should get serialized.
(b) Gecko's style system is inconsistent about whether the vendor prefix is
preserved, when parsing a prefixed gradient.
Since these prefixed gradients aren't super-important and since Gecko's style-system
is being replaced via stylo, we're not bothering to fix the gecko style-system for
this -- we'll just make these tests stylo-specific.
MozReview-Commit-ID: 8lL1aRnDO2H
--HG--
extra : rebase_source : 86d5eab7097ebf39238859debc5ff5539cb2429a
In nsCSSRendering::PaintStyleImageLayerWithSC, we reuse the same clip-state
object, clipState, for all bg/mask layers. We hit this assertion if the
border-radius of previous one is not 0, but the the border-radius of the next
one is 0. In this condition, clipState.mHasRoundedCorners is false, but
clipState.mClippedRadii still stores border-radius information of the previous
layer.
MozReview-Commit-ID: KyxD0vPnMb9
--HG--
extra : rebase_source : 198c916c35ecd51a72dc3df83df43d6dc5f6424c
extra : source : e630c34a8b6397a3e8a08e57d400accff392e25f
This will allow us to verify the entire detection pipeline in real nightly
builds, which will give us confidence that real heap corruption will be
detected and reported properly.
MozReview-Commit-ID: 43Fp2HT8RYy
This ensures the video frame's reflow status is not affect by its children's
reflow result.
MozReview-Commit-ID: 4WwLFyIyV4q
--HG--
extra : rebase_source : 1a2f4c4b4b6508dfda195bceef39a80f9093e117
Also, added MOZ_ASSERT to ensure all the video frame's children are fully
complete after reflowing (except for the mVideoControls special case on
Android).
This fixed aStatus.IsEmpty() assertion for the test case added in Part 2.
MozReview-Commit-ID: HKze4EZoaJY
--HG--
extra : rebase_source : d695ee3a7c884450ba7dd9da41e14ab3621c89ed
The main bug this fixes is that on reftest, the objdir needs to be added to the
whitelist on Windows. However, this only happens when running on Linux for some
reason.
Changing the --work-path and --obj-path arguments to --sandbox-read-whitelist
was more of a drive-by cleanup than anything necessary.
MozReview-Commit-ID: Dq8ZLETMzeM
--HG--
extra : rebase_source : 3d2cdda125205e76f86235eb373074899fe0789a
<applet> is not a thing anymore, and that selector in our UA sheet will never
match anyway, since an <applet> element will never have the BROKEN state.
MozReview-Commit-ID: 7UOMKOv55uJ
Signed-off-by: Emilio Cobos Álvarez <emilio@crisal.io>
Revise the pointer event implementation, includes
- Revise the implementation to fetch preference values.
- Separate the logic to get the pointer capturing frame to PointerEventHandler.
MozReview-Commit-ID: 7pdAr0XFNT2
There are two motivations for this change.
First, reftest.jsm has become very large and monolithic. It has lots of global
state and is hard to modify without breaking something. This change is a first
attempt at dividing reftest.jsm into multiple standalone(ish) modules. This
will make it easier to comprehend and extend.
Second, we'd like to implement "run-by-manifest" mode for reftest. This means
we'll restart the browser between each manifest run. This means much of the
state which is normally stored in global variables in reftest.jsm, will instead
need to be stored in python. A prerequisite to doing that is parsing the
manifests before starting the test suite. A prerequisite to that is moving the
manifest parsing code into a standalone JSM. This is the first step.
MozReview-Commit-ID: 42epTs8EU1O
--HG--
extra : rebase_source : 3db27445af0969867857fbfe41add13161a01cb8
nsFrame::SchedulePaint() invokes InvalidateRenderingObservers, and
InvalidateRenderingObservers ends up posting change hints during we are
processing sequential tasks for animations, but we don't allow posting any
change hints while we are in the middle of restyling process.
Though the change hints posted by InvalidateRenderingObservers are not harmful
in this case since the change hints will be processed in a second post
traversal without problems. That's said, InvalidateRenderingObservers stuff
should be processed in display list based invalidation anyway (bug 1284053).
MozReview-Commit-ID: GKVRZ98lvEN
In this patch, we add 3 tests:
1. test for blocked domain
2. test for blocked sub-domain
3. test for non-blocked domain
MozReview-Commit-ID: JzMImsbGoPr
--HG--
extra : rebase_source : 0eed42f6d2815bd8adbb9fb6be54b6feb3b1bcc8
In the current blocklist implementation, we read the stylo blocklist from the user
preferences very early and only once, even earlier than the test preferences
updating happens.
So, to be able to test the functionality of the stylo blocklist, we add these two
nsIDOMWindowUtils APIs to be able to add/remove a mock domain to the existing
blocklist.
MozReview-Commit-ID: K48ejLBcNbn
--HG--
extra : rebase_source : 354fe9871231f2a94f09bd2612c9d21ced4143e8
In this patch, we read the stylo blocklist into nsLayoutUtils's global static
variable during nsLayoutUtils::Initialize(). So, we can decide if we should
fallback to use Gecko backend while updating style backend for a document.
We add "layout.css.stylo-blocklist.blocked_domains" and
"layout.css.stylo-blocklist.enabled" to ContentPrefs.cpp because they are read
very early (during nsLayoutUtils::Initialize).
MozReview-Commit-ID: 8c4n6m9dYD8
--HG--
extra : rebase_source : ca81cd57980d98b9b73f18bb03a8035b161d23ef
It's easy to mess up the scoping so that (a) the label is pushed and then
immediately popped, and/or (b) the string doesn't live long enough. It's also
easy to do a utf16-to-utf8 conversion unnecessarily when the profiler is
inactive.
This patch splits that macro into three new ones that are harder to mess up.
- AUTO_PROFILER_LABEL_DYNAMIC_CSTR: same as current.
- AUTO_PROFILER_LABEL_DYNAMIC_NSCSTRING: for nsCStrings.
- AUTO_PROFILER_LABEL_DYNAMIC_LOSSY_NSSTRING: for nsStrings.
--HG--
extra : rebase_source : 53c8b43b6a1be06d00618a133e28bf95c46a3ba3
Summary:
The implementations of most SVGTextElement methods flush layout when they're
called. Any SVG <text> that is displayed normally will be reflowed by
SVGTextFrame::ReflowSVG, and any <text> that is in displayed SVG but is not
displayed directly (such as <text> under a <defs>) will be reflowed by
SVGTextFrame::ReflowSVGNonDisplayText. Prior to this fix the changed
code always assumed that one of these methods would reflow the text. However,
in the case that the SVG is under an HTML element that isn't itself ever
reflowed (such as if it's under an HTML <caption>) then the SVG text will
not be reflowed either and the invariants of the code are broken.
This patch modifies the functions that implement the SVGTextElement
methods to return an error code if the text has not been reflowed.
Note: I did try to reuse SVGTextFrame::ReflowSVGNonDisplayText in this
case, but I ran into problems. Given that this is an edge case and likely one
than normal authors will never run into I gave up trying to make that work for
now.
Reviewers: heycam
Bug #: 1402109
Differential Revision: https://phabricator.services.mozilla.com/D107
It's easy to mess up the scoping so that (a) the label is pushed and then
immediately popped, and/or (b) the string doesn't live long enough. It's also
easy to do a utf16-to-utf8 conversion unnecessarily when the profiler is
inactive.
This patch splits that macro into three new ones that are harder to mess up.
- AUTO_PROFILER_LABEL_DYNAMIC_CSTR: same as current.
- AUTO_PROFILER_LABEL_DYNAMIC_NSCSTRING: for nsCStrings.
- AUTO_PROFILER_LABEL_DYNAMIC_LOSSY_NSSTRING: for nsStrings.
--HG--
extra : rebase_source : 59f77df0124249bfd11fee3585420a17b4201d37
Fix some test failures on mobile for the new date/time controls,
* Because mobile does not use the built-in reset button, tests
involving the reset button are skipped.
* The mobile-only date/time tests in test_input_typing_sanitization.html
do not apply to the new date/time controls, so they are removed to make
mobile match desktop.
* test_input_sanitization.html now takes longer to run on Android and
may time out, so the timeout limit is increased.
* time-content-left-aligned.html may fail because of different drawing
behavior on Android, so the test is made fuzzy on Android.
MozReview-Commit-ID: EmfGf6JYdpW
Right now, date/time fields in Fennec appear as regular text fields,
which display the date/time values without formatting. This patch makes
the fields use the Gecko controls, which do support formatting. This
only changes the appearance of the fields; we still display the native
date/time pickers when the fields are tapped on. The reset button is
hidden in the controls because the Fennec date/time picker provides
a separate "clear" button.
MozReview-Commit-ID: EBE2U1zL74Q
When we generate the glyph mask for a transformed frame in
GenerateAndPushTextMask, the transform vector had been applied into aContext[1],
so we should find a way to prevent applying the vector again when painting the
glyph mask.
In bug 1299715, I tried to prevent double apply at [2], it caused two problems:
1. We only skip generating nsDisplayTransform, but we may still create a
nsDisplayPerspactive bellow. Since the parent of a nsDisplayPerspective must be
a nsDisplayTransform, which have been ignored, so we hit this assertion.
2. We skip all transform for all frames while painting the glyph mask, which is
not correct. We should only skip double applying transform vector of the root
frame.
This patch fixes both of these issues:
a. We will still create a nsDisplayTransform for the root frame if need. But
the transform matrix we apply into the target context will be an identity
matrix, so we fix#1 above.
b. In #a, we change the transform matrix to an identity matrix only for the root
frame of the glyph mask, so we fix#2.
[1]
https://hg.mozilla.org/mozilla-central/file/59e5ec5729db/layout/painting/nsDisplayList.cpp#l752
[2]
https://hg.mozilla.org/mozilla-central/file/ce2c129f0a87/layout/generic/nsFrame.cpp#l2806
MozReview-Commit-ID: 973lkQQxLB6
--HG--
extra : rebase_source : aef80444e94d3af7eca776c981f8faded03bc985
One remaining issue is that requesters of CSSRules might also want to know if those rules are complete, and to listen for completion if the rules are incomplete. Bug 1404538 will fix this up at least for devtools.
MozReview-Commit-ID: COnkS40Ooe2
--HG--
extra : rebase_source : 453bed12805fec668e90d5194fcd3df2d48af56e
We have too many layers-free things in WebRenderLayerManager. I create a new
class WebRenderCommandsBuilder and move some functions and variables from
WebRenderLayerManager to WebRenderCommandsBuilder.
MozReview-Commit-ID: BJi1E51W7ax
--HG--
extra : rebase_source : ddbfb044d467430403a3c480030ef9dec803c9f7
nsColorControlFrame::UpdateColor() looks up the color value from the
corresponding <input> element -- and it expects to receive a valid color
string, regardless of what the user/author has done (or whether they've done
anything), thanks to the HTMLInputElement sanitization code that gets run when
the value is set.
As a basic sanity-check, UpdateColor() has an assertion to verify that the
value it receives is non-empty. However, if it happens to be called while the
element is still being appended (e.g. due to greedy frame construction), then
it *can* legitimately get an empty value. So, the assertion isn't entirely valid!
Hence, this patch relaxes the assertion to only take effect after the frame has
been reflowed, and it also makes UpdateColor() a no-op in that case. This
is fine because we can expect that UpdateColor() will be called again (and will
see a non-empty color value at that point) before the frame gets
reflowed/painted.
MozReview-Commit-ID: LOymuwy6gIM
--HG--
extra : rebase_source : 57a2b58cc371ec82f1d2db0334b2236a33056e40
Fix some test failures on mobile for the new date/time controls,
* Because mobile does not use the built-in reset button, tests
involving the reset button are skipped.
* The mobile-only date/time tests in test_input_typing_sanitization.html
do not apply to the new date/time controls, so they are removed to make
mobile match desktop.
* test_input_sanitization.html now takes a lot longer to run on Android
debug and frequently times out, so it's disabled for debug Android builds.
* time-content-left-aligned.html may fail because of different drawing
behavior on Android, so the test is made fuzzy on Android.
MozReview-Commit-ID: FTzwHTJgVJe
Right now, date/time fields in Fennec appear as regular text fields,
which display the date/time values without formatting. This patch makes
the fields use the Gecko controls, which do support formatting. This
only changes the appearance of the fields; we still display the native
date/time pickers when the fields are tapped on. The reset button is
hidden in the controls because the Fennec date/time picker provides
a separate "clear" button.
MozReview-Commit-ID: 75QyKmolNuf
In this case, the caller doesn't go through all the PreTraverseSync stuff (we
don't really want it to do so), via GetComputedStyleNoFlush.
It makes sense to not flush the user font set that case, we'll schedule a
restyle properly if / when needed.
Since the set should be flushed otherwise, let's just not flush it from the
styling code.
MozReview-Commit-ID: LLHfxemJ8QQ
We don't follow this bit intentionally because we know that even if it's set,
when none of the other two bits are set there are no other restyle / change
hints down the tree.
We rely on the frame constructor to clean the mess up, though, and it doesn't
really do a good work about it.
In particular, the case we're hitting on the test-case is:
<body descendant-need-frames change=reconstruct style="display: table-column-group">
<div descendant-need-frames>
<div descendant-need-frames>
<span needs-frame></span>
</div>
</div>
</body>
When we see we need to reconstruct the body, we call
ClearRestyleStateFromSubtree, but that doesn't do much now, since we don't
follow the descendant-need-frames bits.
Then, when we reconstruct the content, we arrive at[1] when constructing the
first child <div>. The <div> flags have been cleared, but not the children's!
Then a text-node is inserted in a <div>, breaking all sorts of invariants.
This is the easiest fix. Other fixes include clearing the flags on
SetAsUndisplayedContent. But that implies not clearing them in
ShouldCreateItemsForChild, and doing that somewhere more sensible.
I suspect it's not too hard, but that's a slightly more risky change, will do it
if you prefer it.
[1]: http://searchfox.org/mozilla-central/rev/3dbb47302e114219c53e99ebaf50c5cb727358ab/layout/base/nsCSSFrameConstructor.cpp#6092
MozReview-Commit-ID: 7026wkQLQkz
If Firefox crashes while mochitests or reftests are running, Marionette
will trigger an IOError exception which currently gets logged immediately,
and causes no post-test checks to be performed. This results in missing
crash and leak checks, and an unclear failure message on Treeherder.
With this change only the IOError from Marionette gets deferred until
all post-test checks are done. This fixes the failure messages, and will
put PROCESS_CRASH or leak log output first.
MozReview-Commit-ID: JCYP5LlPE1m
--HG--
extra : rebase_source : a4a9455402b01db8ef1dbafccc7a726d2927ec03
Removes the XPCOM interface for nsIDOMHTMLAreaElement, replacing it
with binding class usage.
MozReview-Commit-ID: IaX4JFTPZn6
--HG--
extra : rebase_source : 79f9200c6ff9e081a5d9bc21eaa605f88caa99e9
mTargetFrame is not a correct one, we should check the SVGStyle of 'frame'
object.
MozReview-Commit-ID: 3UNfbXkICkz
--HG--
extra : amend_source : 9fa5f095881fc3b92329a7d187f4d5926c43fed7
We don't follow this bit intentionally because we know that even if it's set,
when none of the other two bits are set there are no other restyle / change
hints down the tree.
We rely on the frame constructor to clean the mess up, though, and it doesn't
really do a good work about it.
In particular, the case we're hitting on the test-case is:
<body descendant-need-frames change=reconstruct style="display: table-column-group">
<div descendant-need-frames>
<div descendant-need-frames>
<span needs-frame></span>
</div>
</div>
</body>
When we see we need to reconstruct the body, we call
ClearRestyleStateFromSubtree, but that doesn't do much now, since we don't
follow the descendant-need-frames bits.
Then, when we reconstruct the content, we arrive at[1] when constructing the
first child <div>. The <div> flags have been cleared, but not the children's!
Then a text-node is inserted in a <div>, breaking all sorts of invariants.
This is the easiest fix. Other fixes include clearing the flags on
SetAsUndisplayedContent. But that implies not clearing them in
ShouldCreateItemsForChild, and doing that somewhere more sensible.
I suspect it's not too hard, but that's a slightly more risky change, will do it
if you prefer it.
[1]: http://searchfox.org/mozilla-central/rev/3dbb47302e114219c53e99ebaf50c5cb727358ab/layout/base/nsCSSFrameConstructor.cpp#6092
MozReview-Commit-ID: 7026wkQLQkz
We had previously started running some tests in layers-free mode
already, either by setting the default-preferences on the folder to
turn on layers-free, or by duplicating an individual reftest annotation
so that it ran in both layers-full and layers-free mode. This patch
removes these changes since layers-free is now the default and we don't
need to run layers-full at all.
MozReview-Commit-ID: JJwB9iO5clU
--HG--
extra : rebase_source : ea20545945b825d7ff829526d4d263850e6b5b7f
This patch:
- adds fails-if annotations for all the reftests that were consistently failing
with layers-free turned on.
- removes fails-if or reduces the range on fuzzy-if annotations for all
the reftests that were producing UNEXPECTED-PASS results with
layers-free turned on.
- adds skip-if, random-if, or fuzzy-if annotations to the reftests that
were intermittently failing due to timeout, obvious incorrectness, or
slight pixel differences, respectively.
MozReview-Commit-ID: A0Aknn6rnjj
--HG--
extra : rebase_source : 420d9cf43f23a5d654fa36eec69138937d13c173
This adds new test verification steps for mochitest/reftest/xpcshell tests
with MOZ_CHAOSMODE=3 (thread scheduling and network thread scheduling).
Enabling all chaos mode features seems to destabilize test verification
so I am only enabling these features for now.