Before committing a new layout (and thus building a new paint tree)
we now go through both the old paint tree and the layout tree and detach
them from each other.
This is a little extra work, but it ensures that there are no lingering
references across the trees, which we were apparently accumulating in
some cases on Discord, causing GC leaks.
The `RecordedNodeValue` struct contains a `GC::Ref` to a DOM node, which
might disappear as a result of a garbage collection. For example, during
the "outdent" command, we record nodes, split the parent of those nodes
potentially resulting in all kinds of DOM changes, and then try to
restore the nodes' values. This caused a crash in the
`editing/run/outdent.html` WPT subtests.
By returning a `ConservativeVector`, we make sure the `GC::Ref` gets
marked during sweeps and nodes do not suddenly disappear.
This parses `anchor-size(..)` functions in CSS, but does not yet result
in a useful `Size`: we need style & layout interleaving similar to
container queries for this, since the resulting value depends on layout
results.
Not supported yet: `anchor-size()` appearing inside a `calc()` node.
Adds 4280 WPT subtest passes in `css/css-anchor-position`.
This allows authors to check if Trusted Type is required for a given
Element's content attribute. This allows to pass the correct type
when they call to Element.setAttribute.
Many wpt test on trusted-types relay on this class being defined to even
begin the test as it declares some event handlers.
This is not really an implementation but the most basic setup needed to
run the tests.
This allows them to keep style sheets alive while loading fonts for
them. Fixes some GC crashes seen on the WPT WOFF2 tests after
66a19b8550 stopped FetchRecord leaks from
keeping various other things alive.
Our previous implementation kept track of an AnimationTimeline being
monotonically increasing, by looking at new time values coming in and
setting `m_monotonically_increasing` to `false` whenever a new value
is before the previous known time value.
As far as I can tell, the spec doesn't really ask us to do so: it just
defines 'monotonically increasing' as a property of a timeline, i.e. it
guarantees that returned time values from `::current_time()` are always
greater than or equal to the last returned value.
This fixes a common crash seen when the last render opportunity lies
before the document's origin time, and `::set_current_time()` was
invoked with a negative value. This was especially visible in the
`Text/input/wpt-import/css/cssom/CSSStyleSheet-constructable.html` test.
If class doesn't have any private fields, we could avoid allocating
PrivateEnvironment for it.
This allows us to skip thousands of unnecessary PrivateEnvironment
allocations on Discord.
Before this change, whenever element's attributes changed, we would add
a flag to "pending invalidation", indicating that all descendants whose
style uses CSS custom properties needed to be recomputed. This resulted
in severe overinvalidation, because we would run invalidation regardless
of whether any custom property on affected element actually changed.
This change takes another approach, and now we decide whether
descendant's style needs to be recomputed based on whether ancestor's
style recomputation results in a change of custom properties, though
this approach adds a little overhead to style computation as now we have
to compare old vs new hashmap of custom properties.
This brings substantial improvement on discord and x.com where, before
this change, advantage of using invalidation sets was lost and we had
to recompute all descendants, because almost all of them use custom
properties.
Compare `Vector<Parser::ComponentValue>` directly instead of
serializing them into strings first.
This is required for the upcoming changes where we would compare
previous and new sets of custom properties to figure out whether we need
to invalidate descendant elements. Without this change `equals()` would
show up being hot in profiles.
According to the spec, `ResizeObserver` needs to live for as long as
it's referenced from script or has observation targets. With this change
we make sure that `ResizeObserver` is unregistered from the `Document`
when it has no target.
Fixes GC leak that caused us to keep all resize observers alive until
document they belong to is destroyed.
`ShadowRoot` register itself in Document` from constructor and
unregister itself from `finalize()`. The problem is that `finalize()`
won't be invoked for as long as `ShadowRoot` is visited by
`Document`, leading to GC leaks.
`DocumentObserver` register itself in Document` from constructor and
unregister itself from `finalize()`. The problem is that `finalize()`
won't be invoked for as long as `DocumentObserver` is visited by
`Document`. By not visiting registered observers from `Document` we
move this responsibility to object that allocated observer, which is
always exactly what we want, e.g. once `SVGUseElement` that uses
observer is gone, observer won't be visited anymore which will lead to
`finalize()` being called.
This patch expands our generated content support beyond single strings
to lists of strings and/or images.
Pseudo-elements like ::before and ::after can now use content:url(...)
to insert anonymous image boxes into the layout tree.
This is heavily used in Google Docs for UI elements.
The faux position we created here is adjusted by the device pixel ratio
later on, which would invoke integer overflow on screens with a DPR
greater than 1.
Instead of creating special data for a mouse move event, let's just add
an explicit leave event handler.
Not accounting for opcode size when calculating incoming jump edges
meant that we were merging nodes where we otherwise shouldn't have been,
for example /.*a|.*b/.
This porting effort makes it pretty clear we will want a UTF-16-aware
GenericLexer. But for now, we can actually make ASCII assumptions about
what we are parsing, and act accordingly.
