Andreas Kling 356a369aa6 LibWeb: Avoid descendant recomputes for same-parent moves ...

Moving a node within the same parent changes sibling and positional
relationships, but it does not make every descendant of the moved node
need a fresh computed style. Handle this as a structural mutation at
the old and new sibling edges and dirty only the moved root and the
affected ancestors, instead of marking the entire moved subtree.

Factor the existing previous- and next-sibling structural invalidation
out of Node::invalidate_style() into invalidate_structurally_affected_-
siblings(), and pull the ancestor child-needs-style-update walk into
mark_ancestors_as_having_child_needing_style_update(). The new
invalidate_style_after_same_parent_move() reuses both helpers.

Whether the moved root itself needs its own style recomputed depends
on whether any selector matched against it (or against a descendant)
relied on its position in the sibling list. Track that via two new
sticky bits on Element, set during selector matching:

  - m_affected_by_structural_pseudo_class_in_non_subject_position
  - m_affected_by_sibling_combinator_in_non_subject_position

Both are write-once (sticky) because matching descendants can set them
while we're not currently re-matching this element's own selectors;
keeping them set is conservative and avoids stale descendant style.

When neither bit is set and the element only carries subject-position
positional/sibling/has() involvement, we just dirty the root and skip
its descendants.

Rebaseline same-parent-move-root-only and the structural-feature filter
counters to reflect the new path. Matching behavior is unchanged.

2026-04-28 15:34:49 +02:00

90 KiB

Raw Blame History

View Raw