[Kanti]

Problem

Follow-up to #159. Its scroll listener recomputed all holes on every scroll frame, on every page, also without any sticky or fixed target: getBoundingClientRect for all targets plus unionRects, which grows quadratically. Measured on a fast machine: about 0.4 ms per frame with 50 targets, 1.2 ms with 300 and 3.7 ms with 1000. The overlay path was not rewritten, but #144 had made scrolling run no script at all, and that was lost. The existing test only counted writes of the path and passed with and without this work.

Implemented behavior

• Whether a target or one of its ancestors has the computed position fixed or sticky is checked when the targets are selected: in highlight() and in refreshTargets() (resize, mutation, toggle), not while scrolling. Across shadow roots the host counts as ancestor. Ancestors shared by several targets are checked once.
• Class and style attribute changes of the ancestors of the targets are observed as well, so a header that a script makes fixed while scrolling is followed, as on main. They only recompute the holes when the scroll listener is added or removed.
• The scroll listener is only registered while such a target exists. Without one, scrolling runs no code again, as after [BUGFIX] Fix spotlight overlay stutter on scroll #144.
• The rAF coalescing and updateHoles() without measuring the document are unchanged.
• Known limit: an ancestor that becomes sticky or fixed without a class or style change on one of the ancestors, e.g. by a class on a sibling that a selector depends on, is only noticed with the next selection of the targets (resize, mutation of a target).

Tests

• spotlight-overlay.test.js (browser): the test for static targets counts the getBoundingClientRect calls while scrolling three times and expects none, instead of counting writes of the path. It fails without the fix (6 calls).
• The test for sticky and fixed targets passes unchanged.
• A header that a script makes fixed while scrolling (class toggle, a placeholder keeps its height) stays clickable. This test fails without the observation of the ancestors and passes on main.
• npm run test: 93 passing, npm run lint