Add a per-Document cache that maps selector strings to their parsed
SelectorList results. This avoids re-tokenizing and re-parsing the
same selector strings on every call, which is a huge win for pages
that repeatedly call querySelectorAll with the same selectors.
On microsoft.com, where they spam querySelectorAll with the same two
selectors, profiling showed 62% of main thread time was spent
re-parsing selectors. This cache eliminates that entirely for
repeated queries.
Previously, destroyed-document tasks were forced to be runnable to
prevent them from leaking in the task queue. Instead, discard them
during task selection so their callbacks never run with stale state.
This used to cause issues with a couple of `spin_until()`s in the past,
but since we've removed some of them that had to do with the document
lifecycle, let's see if we can stick closer to the spec now.
Instead of immediately firing fullscreenchange, defer that until
WebContent's client has confirmed that it is in fullscreen for the
content. The fullscreenchange is fired by the viewport change, so in
cases where the fullscreen transition is instantaneous (i.e. the
fullscreen state is entered at the exact moment the viewport expands),
the resize event should precede the fullscreenchange event, as the spec
requires.
This fixes the WPT element-request-fullscreen-timing.html test, which
was previously succeeding by accident because we were immediately
fullscreenchange upon requestFullscreen() being called, instead of
following spec and doing the viewport (window) resize in parallel. The
WPT test was actually initially intended to assert that the
fullscreenchange event follows the resize event, but the WPT runner
didn't actually have a different resolution for normal vs fullscreen
viewports, so the resize event doesn't actually fire in their setup. In
our headless mode, the default viewport is 800x600, and the fullscreen
viewport is 1920x1080, so we do fire a resize event when entering
fullscreen. Therefore, that imported test is reverted to assert that
the resize precedes the fullscreenchange.
We were conflating elements being the active element and elements being
activated. The :active pseudo class is supposed to be based on whether
an element will have its activation behavior run upon a button being
released.
Store whether an element is being activated as a flag that is set/reset
by EventHandler.
Doing this allows label elements to visually activate their control
without doing a weird paintable hack, so the Labelable classes have
been yeeted.
Rename DocumentDestructionState into a more generic
DocumentLifecycleState so it can be reused for the document unloading
logic. Instead of using ::spin_until(), use a callback mechanism to run
code once all child navigables are unloaded.
1. unload_a_document_and_its_descendants() now follows the spec
algorithm structure: recursively unload child navigables via queued
tasks (step 4), wait for them (step 5), then queue this document's
own unload as a separate task (step 6). Previously, this was
flattened into a single spin that unloaded all descendants and the
document together, followed by a non-spec call to
destroy_a_document_and_its_descendants().
2. unload() step 19 now calls destroy() when the document is not
salvageable, as the spec requires. Previously this was a no-op with a
comment deferring to unload_a_document_and_its_descendants().
3. destroy_top_level_traversable() step 2 now calls
destroy_a_document_and_its_descendants() instead of destroy().
The iframe-unloading-order test, which exercises named iframe access
during unload handlers (the scenario the previous logic was designed to
protect), still passes.
Fixes#7825
Replace per-node heap-allocated AtomicRefCounted
AccumulatedVisualContext objects with a single contiguous Vector inside
AccumulatedVisualContextTree. All nodes for a frame are now stored in
one allocation, using type-safe VisualContextIndex instead of RefPtr
pointers.
This reduces allocation churn, improves cache locality, and opens the
door for future snapshotting of visual context state — similar to how
scroll offsets are snapshotted today.
The style of an element depends on it's navigable's viewport size which
in turn depends on the navigable's container's style - so if requires a
style update then so does the original element.
Introduce Document::update_layout_if_needed_for_node() which only calls
update_layout() when the node is connected. Use it at all call sites
that query layout metrics (offsets, client dimensions, image size, SVG
bounding box, etc.) so disconnected elements no longer trigger an
unnecessary layout.
With per-paintable display list command caching now in place, the
separate paint-only properties resolution phase is no longer needed.
Resolution now happens inline during painting and its cost is amortized
since it only runs on cache miss.
Move all property resolution to point of consumption:
- is_visible() and visible_for_hit_testing() compute on the fly
- Filter resolution moved to assign_accumulated_visual_contexts()
- Border radii, outlines computed on access
- Box shadows, backdrop filter resolved inline during painting
- Background resolution moved into paint_background()
- Mask resolution moved to StackingContext::paint()
- Text fragment and SVG stroke properties resolved during painting
Cache the display list commands produced by each PaintableBox's paint()
on a per-phase basis. On subsequent display list rebuilds, if a
paintable's cache is still valid, replay the recorded commands directly
— skipping paint() and all the property resolution it entails.
Besides saving time on property resolution, this also enables Skia to
reuse path tessellation results across frames — e.g. border paths are
preserved in the cache and don't need to be re-tessellated on every
repaint.
Rename Document::set_needs_display() to set_needs_repaint() and make it
private. External callers must now go through Node/Paintable which
route the request to the document internally.
Fix one existing misuse in AnimationEffect that was calling
document-level set_needs_display() instead of routing through the
target element's paintable.
This is preparation for per-paintable display list command caching:
repaint requests must go through specific paintables so their cached
command lists can be invalidated.
* Don't publicly expose methods that are only used internally.
* Modernize comment style (wrap at 120 chars, use NB for our notes).
* Change the stringifier for RequestFullscreenError to return a UTF-16
string. We were allocating a String, then transcoding it to UTF-16.
* Give helper methods clearer names, e.g.:
fullscreen_has_error_check -> is_element_allowed_to_enter_fullscreen
Throwaway LayoutState instances used for intrinsic sizing should not
access nodes outside the laid-out subtree. Make this explicit by setting
a subtree root on each throwaway state, pre-populating only the
immediate containing block, and using try_get() for any ancestor
lookups — treating unavailable ancestors as indefinite rather than
silently populating them with incorrectly-resolved values.
Add unsafe_layout_node(), unsafe_paintable(), and unsafe_paintable_box()
accessors that skip layout-staleness verification. These are for use in
contexts where accessing layout/paintable data is legitimate despite
layout not being up to date: tree construction, style recalculation,
painting, animation interpolation, DOM mutation, and invalidation
propagation.
Also add wrapper APIs on Node to centralize common patterns:
- set_needs_display() wraps if (unsafe_paintable()) ...set_needs_display
- set_needs_paint_only_properties_update() wraps similar
- set_needs_layout_update() wraps if (unsafe_layout_node()) ...
And add Document::layout_is_up_to_date() which checks whether layout
tree update flags are all clear.
Replace the Window::scroll_by(0, 0) call at the end of
Document::update_layout() with a dedicated
Navigable::clamp_viewport_scroll_offset() that directly clamps the
viewport scroll offset to valid bounds.
The old approach re-entered layout from within layout, since scroll_by()
would trigger another layout update. The new approach is called from the
event loop's rendering steps, after layout is complete.
Stop converting between CSS and device pixels as part of rendering - the
display list should be as simple as possible, so convert to DevicePixels
once when constructing the display list.
During SVG subtree relayout, position:fixed elements inside
<foreignObject> use the viewport as their containing block. Since the
viewport is outside the SVG subtree, it was not pre-populated in the
LayoutState, causing a VERIFY failure in ensure_used_values_for().
Fix this by unconditionally pre-populating the viewport node from its
paintable in relayout_svg_root().
Replace the unsafe HashTable<GC::Weak<Animation>> with
GC::WeakHashSet<Animation>, and update all callers to use reference
syntax instead of pointer syntax since the iterator now yields T&.
In practice, the event loop queries Document::fonts(), so we don't gain
anything by delaying the allocation, apart from making it unclear when
it happens.
Previously, PaintNestedDisplayList was treated as an opaque command,
printing only its name and rect without showing the nested display
list's contents. This made it impossible to debug painting issues
involving SVG masks/clips, CSS background-clip: text, and iframe
content through display list dumps.
Refactor the command dump loop into a recursive lambda that expands
nested display lists inline with increased indentation.
relayout_svg_root() clears individual stacking contexts via
reset_for_relayout() but didn't call invalidate_stacking_context_tree().
The viewport's stacking context remained non-null, so
build_stacking_context_tree_if_needed() skipped the rebuild. This caused
foreignObject to lose its stacking context after relayout, breaking SVG
mask application.
There are some predefined counter styles (such as the longhand east
asian ones) which are too complex to be defined here and will need to be
implemented ad-hoc, this remains as a FIXME for now.
The tricky bit of this is resolving cycles in extending rules and
ensuring that counter styles are registered in the required order for
extension (i.e. for any pair of extended/extending rules the extended
one should be registered first).
Each NodeWithStyle is assigned a sequential layout index during the
pre-layout tree traversal. LayoutState stores UsedValues in a
PagedStore — a two-level page table indexed by layout_index that
gives O(1) lookup via two array accesses, with pages allocated
lazily on first write. UsedValues are stored directly in pages
(Optional<T>) rather than behind heap pointers, eliminating
per-entry malloc/free calls and improving cache locality.
This cuts ensure_used_values_for() from ~14% to ~7% in profiles
on https://www.nyan.cat/.
The cursor blink timer fires every 500ms and only needs to toggle
the blink state and mark the paintable as needing display. If the
paintable doesn't exist yet, we can simply skip the blink -- the
cursor will appear after the next natural rendering update.
This avoids a potentially expensive synchronous layout every 500ms
for what is a purely cosmetic operation.
This is a targeted version of update_style() that only performs a
style update if the given element (or its ancestors) actually have
dirty style. Useful when we only need up-to-date computed properties
for a specific element.
Eliminate O(n) tree-order sorted insert on every add() by simply
appending elements and deferring order resolution to lookup time. Cache
the first-in-tree-order element in get() so repeated getElementById()
calls avoid repeated subtree traversal.
Improves performance on YouTube where add() was hot in profiles while
scrolling through comments.
