We are meant to store a weak reference to the element indicated by this
attribute, rather than a GC-protected strong reference. This also hoists
the "get the attr-associated element" AO into its own function, rather
than being hidden in IDL, to match "get the attr-associated elements".
There are ARIA attributes, e.g. ariaControlsElements, which refer to a
list of elements by their ID. For example:
<div aria-controls="item1 item2">
The div.ariaControlsElements attribute would be a list of elements whose
ID matches the values in the aria-controls attribute.
Math functions like abs(), clamp(), round(), etc, can be used by
themselves in property values, without wrapping them in calc().
Before this change, we were neglecting to run calc simplification on the
generated calculation node trees. By doing that manually after parsing a
standalone math function, we score at least a couple hundred WPT points.
Whenever we introduce a block element in a container that at that point
has only had inline children, we create an anonymous wrapper for all the
inline elements so we can keep the invariant that each container
contains either inline or non-inline children. For some reason, we
ignore all the out-of-flow nodes since they are layed out separately and
it was thought that this shouldn't matter.
However, if we are dealing with inline blocks and floating blocks, the
order of the inline contents _including_ out-of-flow nodes becomes very
important: floating blocks need to take the order of nodes into account
when positioning themselves.
Fix this by simply hoisting the out-of-flow nodes into the anonymous
wrapper as well.
Fixes the order of blocks in #4212. The gap is still not present.
Instead, use the generic create_independent_formatting_context_if_needed
so that unusual situations like image-as-table-caption don't crash.
This logic clearly needs more work, but let's at least do better than
crashing. This gives us 26 new subtest passes on WPT.
We were incorrectly deciding that abspos elements shouldn't treat many
max-width and max-height values as `none`. My best understanding is that
this was a hack in 2023 for an issue that has been solved since then.
By removing the incorrect short-circuit, we stop at least one WPT test
from crashing due to infinite recursion and get ourselves +34 passes.
By doing that we avoid doing separate allocation for each such vector,
which was really expensive on js heavy websites. For example this change
helps to get EC allocation down from ~17% to ~2% on Google Maps. This
comes at cost of adding extra complexity to custom execution context
allocator, because EC no longer has fixed size and we need to maintain
a list of buckets.
This is better because:
- Better data locality
- Allocate vector for registers+constants+locals+arguments in one go
instead of allocating two vectors separately
Fixes a bunch of websites breaking because we now verify jump offsets by
trying to remove 0-offset jumps.
This has been broken for a good while, it was just rare to see Repeat
inside alternatives that lended themselves well to tree alts.
We had a bit of an awkward setup where we want the Application to be a
SettingsObserver, but neither the Settings object nor the Application
itself was fully initialized by the time the observer was created. So
we invented a deferred observer initializer specifically for the
Application.
Instead, let's just create a dedicated observer subclass that is owned
by the Application. We can then create it once we have the singleton
Application appropriately set up.
JavaScript module requests (in a non-worker context) always have CORS
enabled. However, CORS requests are only allowed for same-origin or
HTTP/S requests. This patch extends this to allow resource:// requests
from opaque origins (e.g. about: URLs).
We must also set the Access-Control-Allow-Origin header to "null" to
ensure that the response is accepted by the CORS checks. This does not
affect requesting resource:// URLs from resource:// URLs as those are
same-origin and skip CORS checks.
This ultimately enables requesting resource:// JS modules from the
about:settings page.
The special case for anonymous table wrappers actually ended up hurting
correctness by preventing the full ancestor chain from being marked for
for intrinsic size cache invalidation.
Caused Layout/input/table/propagate-style-update-to-wrapper.html to
flake on CI, and was easy to reproduce locally with sanitizers.
The fix here is simply to remove the special handling of anonymous table
wrapper parents, since *all* parents are invalidated automatically
anyway!
Began flaking in fa9c463ffd.
Previously we were counting the total number of *nodes* in the tree for
the chain cost, which greatly underestimated its cost when large
bytecode entries were present,
This commit switches to estimating it using the total bytecode *size*,
which is a closer value to the true cost than the tree node count.
This corresponds to a ~4x perf improvement on /<script|<style|<link/ in
speedometer.
Previously, if a transition property was not wrapped in a list, it
would be replaced with the default value in
`StyleComputer::compute_transitioned_properties`.
