- From: Steve Kobes <skobes@google.com>
- Date: Sat, 15 Aug 2020 19:24:20 -0400
- To: public-web-perf@w3.org
- Message-ID: <CAAjAU=bx4zfKvy49SK-55kNXqq7oAsQRc33mEXTP31MDSWH_Jw@mail.gmail.com>
> Mozilla is interested in prototyping layout instability, but is concerned > about the potential overhead in making this a part of a standard and > required for every page load. Google also encountered challenges making layout instability performant. In the spec and explainer for layout instability, we tried to emphasize the possibility of trading off precision for computational efficiency. For example, Chrome's initial implementation snapped visual representations to a low-resolution grid to calculate the impact region. We also added a note in spec section 2.3 with links to resources about the Klee measure problem and the O(n lg n) algorithm we ended up with in LayoutShiftRegion <https://source.chromium.org/chromium/chromium/src/+/master:third_party/blink/renderer/core/layout/layout_shift_region.h> .. There has also been some recent discussion about simplifying bounding rects for inlines, in https://crbug.com/1108622#c2. However, the spec has not yet been updated to reflect that proposal. In my opinion, for a feature like layout instability, user agents ought to have some flexibility to make implementation decisions that best fit their rendering architecture, even at the cost of minor cross-browser divergences in reported scores. See https://github.com/WICG/layout-instability/blob/master/README.md#precision-variance-and-evolution ..
Received on Saturday, 15 August 2020 23:25:16 UTC