When marking a part of the layout tree for rebuild, if the subtree root
that we're marking has an anonymous parent, we now mark from the nearest
non-anonymous ancestor instead.
This ensures that subtrees inside anonymous wrappers don't just get
duplicated (i.e recreated but inserted instead of replaced).
We were calling into `Range::set_start_or_end()` indirectly through
`::set_start()` and `::set_end()`, but that algorithm only calls for an
invocation whenever the start or end of a range needs to be set to a
boundary point. If an algorithm step calls for setting the node or
offset, we should directly modify the range.
The problem with calling into `::set_start_or_end()` is that this
algorithm potentially modifies _both_ the start and end of the range,
but algorithms trying to update a range's start or end often have
explicit steps to take both the start and end into account and end up
overcompensating for the start or end offset resulting in an invalid
range (e.g. with an end offset beyond a node's length).
This makes updating a range's start/end a bit more efficient and removes
a piece of ad-hoc code in CharacterData needed to make it work before.
The play_or_cancel_animations_after_display_property_change() helper
was being called by Node::inserted() and Node::removed_from() and then
recursing into the shadow-including subtree.
This had quadratic complexity since inserted() and removed_from() are
themselves already invoked recursively for everything in the
shadow-including subtree.
Only one caller of this API actually needed the recursive behavior,
so this patch moves that responsibility to the caller and puts the logic
in style recomputation instead.
1.02x speedup on Speedometer's TodoMVC-jQuery.
This was recently added to both the HTML and DOM specifications,
introducing the new moveBefore DOM API, as well as the new internal
'removing steps'.
See:
* 432e8fb
* eaf2ac7
The upcoming generated types will match those for pseudo-classes: A
PseudoElementSelector type, that then holds a PseudoElement enum
defining what it is. That enum will be at the top level in the Web::CSS
namespace.
In order to keep the diffs clearer, this commit renames and moves the
types, and then a following one will replace the handwritten enum with
a generated one.
If an element is affected only by selectors using the direct sibling
combinator `+`, we can calculate the maximum invalidation distance and
use it to limit style invalidation. For example, the selector
`.a + .b + .c` has a maximum invalidation distance of 2, meaning we can
skip invalidating any element affected by this selector if it's more
than two siblings away from the element that triggered the style
invalidation.
This change results in visible performance improvement when hovering
PR list on GitHub.
12c6ac78e2 with fixed mistake when cache
slot is copied instead of being referenced:
```cpp
auto cache =
box.cached_intrinsic_sizes().min_content_height.ensure(width);
```
while it should've been:
```cpp
auto& cache =
box.cached_intrinsic_sizes().min_content_height.ensure(width);
```
This change moves intrinsic sizes cache from
LayoutState, which is local to current layout run,
to layout nodes, so it could be reused between
layout runs. This optimization is possible because
we can guarantee that these measurements will
remain unchanged unless the style of the element
or any of its descendants changes.
For now, invalidation is implemented simply by
resetting cache on whole ancestors chain once we
figured that element needs layout update.
The case when layout is invalidated by DOM's
structural changes is covered by layout tree
invalidation that drops intrinsic sizes cache
along with layout nodes.
I measured improvement on couple websites:
- Mail list on GMail 28ms -> 6ms
- GitHub large code page 47ms -> 36ms
- Discord chat history 15ms -> 8ms
(Time does not include `commit()`)
This allows us to avoid a full layout tree rebuild after change of
"display" property, which happens frequently in practice. It also
allows us to avoid a full rebuild after DOM node insertion, since
previously, computing styles for newly inserted nodes would trigger a
complete layout tree rebuild.
We already have logic to play or cancel animations in an element's
subtree when the display property changes to or from none. However,
this was not sufficient to cover the case when an element starts/stops
being nested in display none after insertion.
The DOMParsing spec is in the process of being merged into the HTML one,
gradually. The linked spec change moves XMLSerializer, but many of the
algorithms are still in the DOMParsing spec so I've left the links to
those alone.
I've done my best to update the GN build but since I'm not actually
using it, I might have done that wrong.
Corresponds to 2edb8cc7ee
Instead of marking all nodes in the subtree for style recalculation,
including subtrees of subsequent siblings, we can fall back to the
default invalidation path, which is optimized to skip siblings
unaffected by sibling selectors.
Makes scrolling on https://frame.work/pl/en/about go a lot smoother.
This reduces the number of `.cpp` files that need to be recompiled when
one of the below header files changes as follows:
CSS/ComputedProperties.h: 1113 -> 49
CSS/ComputedValues.h: 1120 -> 209
Instead of checking all elements in a document for containment in
`:has()` invalidation set, we could narrow this down to ancestors and
ancestor siblings, like we already do for subject `:has()` invalidation.
This change brings great improvement on GitHub that has selectors with
non-subject `:has()` and sibling combinators (e.g., `.a:has(.b) ~ .c`)
which prior to this change meant style invalidation for whole document.
This commit changes the strategy for updating inherited styles. Instead
of marking all potentially affected nodes during style invalidation, the
decision is now made on-the-fly during style recalculation. Child nodes
will only have their inherited styles recalculated if their parent's
properties have changed.
On Discord this allows to 1000x reduce number of nodes with recalculated
inherited style.
The current implementation of `:has()` style invalidation is divided
into two cases:
- When used in subject position (e.g., `.a:has(.b)`).
- When in a non-subject position (e.g., `.a > .b:has(.c)`).
This change focuses on improving the first case. For non-subject usage,
we still perform a full tree traversal and invalidate all elements
affected by the `:has()` pseudo-class invalidation set.
We already optimize subject `:has()` invalidations by limiting
invalidated elements to ones that were tested against `has()` selectors
during selector matching. However, selectors like `div:has(.a)`
currently cause every div element in the document to be invalidated.
By modifying the invalidation traversal to consider only ancestor nodes
(and, optionally, their siblings), we can drastically reduce the number
of invalidated elements for broad selectors like the example above.
On Discord, when scrolling through message history, this change allows
to reduce number of invalidated elements from ~1k to ~5.
f7a3f78 made the layout tree invalidate only the inserted nodes
themselves, but it turned out that CSS containment invalidation relies
on the parent being invalidated as well.
There is no need for this invalidation because taking care of siblings
is already done by invalidation with `NodeInsertBefore` reason. Parent
element itself (without subtree) is always invalidated by
`Node::children_changed()` hook, so `:empty` pseudo-class invalidation
is already covered.
When checking whether an early return is possible because some ancestor
already has the whole subtree invalidation flag set, the check should
begin with the current node's parent rather than with the node itself.
Otherwise, if a node already has the whole subtree invalidation flag
set and is subsequently invalidated for the reason `NodeInsertBefore`
or `NodeRemove`, we will skip the sibling invalidation required for
these operations
This fix is required for optimizations in subsequent commits.
With this change, siblings of an inserted node are no longer invalidated
unless the insertion could potentially affect their style. By
"potentially affected," we mean elements that are evaluated against the
following selectors during matching:
- Sibling combinators (+ or ~)
- Pseudo-classes :first-child and :last-child
- Pseudo-classes :nth-child, :nth-last-child, :nth-of-type, and
:nth-last-of-type
Previous name for misleading because it checks if box could be scrolled
by user input event which is diffent from checking if box is scrollable.
For example box with `overflow: hidden` is scrollable but it can't be
scrolled by user input event.
Currently, this metadata is only provided on the insertion steps,
though I believe it would be useful to extend to the other cases
as well. This metadata can aid in making optimizations for these
steps by providing extra context into the type of change which
was made on the child.
...until Document::update_style(). This allows to avoid doing full
document DOM tree traversal on each Node::invalidate_style() call.
Fixes performance regression on wpt.fyi
Prior to this change, we invalidated all elements in the document if it
used any selectors with :has(). This change aims to improve that by
applying a combination of techniques:
- Collect metadata for each element if it was matched against a selector
with :has() in the subject position. This is needed to invalidate all
elements that could be affected by selectors like `div:has(.a:empty)`
because they are not covered by the invalidation sets.
- Use invalidation sets to invalidate elements that are affected by
selectors with :has() in a non-subject position.
Selectors like `.a:has(.b) + .c` still cause whole-document invalidation
because invalidation sets cover only descendants, not siblings. As a
result, there is no performance improvement on github.com due to this
limitation. However, youtube.com and discord.com benefit from this
change.
