Instead of collecting all documents in a big vector and then filtering
the vector (twice!) with remove_all_matching(), we now pass a filter
callback to documents_in_this_event_loop_matching() and avoid all the
extra shuffling work.
Saw this stuff hogging ~20% of CPU time when profiling a WebContent
process in the middle of a WPT run.
Rather than accumulating margins into a vector, and then looping through
them when resolving the margin, it's much simpler to just update two
fields, and sum them when resolving.
When inserting a new utf-16 surrogate next to an existing surrogate
with replaceData, the surrogates would not get merged correctly into a
single code point. This is because internally the text data is stored
as utf-8, and the two surrogates would be converted seperately. This
has now been fixed by first recreating the whole string in utf-16 and
then converting it back to utf-8.
It's not the most efficient solution, but this fixes at least 6 WPT
subtests.
This isn't directly in the spec, but since replaceChild is implemented
in terms of remove + insert, the removal step may cause arbitrary code
to execute, and so we have to verify that the replaceChild inputs still
make sense afterwards, before doing the insertion.
This roughly matches what WebKit does, and makes a bunch of HTML parsing
tests in WPT stop asserting.
In particular, input character lookahead now knows how to stop at the
insertion point marker if needed.
This makes it possible to do amazing things like having document.write()
insert doctypes one character at a time.
This is necessary to avoid a circular reference when including
Serializable.h in DOMException.h.
This moves the definition of SerializationRecord, SerializationMemory,
and DeserializationMemory into LibWeb/Forward.h so that Serializable.h
only needs to include LibWeb/Forward.h.
Elements with transforms were tested on their pre-transformed
positions, causing incorrect hits.
Copy the position transformation done in `StackingContext::hit_test`
to ensure that hit tests are done on the _actual_ position.
If we reach the insertion point at the same time as we switch to another
tokenizer state, we have to bail immediately without proceeding with the
next code point. Otherwise we'd fetch the next token, get an EOF marker,
and then proceed as if we're at the end of the input stream, even though
more data may be coming (with more calls to document.write()..)
The repository being in static storage is a bit of a hodgepodge, but in
line with how our current storage partitioning is done. We should
eventually move this, along with other across browsing context APIs to a
proper location at a later stage. But for now, this makes progress on
the meat of the BroadcastChannel API.
In the case where we had a preferred aspect ratio and a natural height
but no natural width, we'd get into ping-ponging infinite recursion by
trying to find the width to resolve the height to resolve the width to
resolve the height...
This one is particularly weird as there's a priority order, and we even
have to look at attributes from the container element if we're inside a
subframe.
This is an ad-hoc hack papering over the fact that we can apparently
end up in these places without an active window, and proceeding without
one leads to assertions on WPT.
