commit 0b03969b330f4fe829119cb7c11d63988a14a591
author Aaron Krajeski <aaronhk@chromium.org> Mon May 15 18:34:46 2023
committer Chromium LUCI CQ <chromium-scoped@luci-project-accounts.iam.gserviceaccount.com> Mon May 15 18:34:46 2023
tree fb18aaf3b9e5d5703879f66ee624f7597e005a51
parent 02df6036d03376b078556ab41ee03c4970dda0f2

Use float for calculating opacity groups.

Before the blink::Color refactor there was the possibility that very
nearly transparent colors would register as fully transparent because
call sites were checking AlphaAsInteger() > 0 and any number less than
1/255 would fail this check, despite still being semi-opaque.

This caused a particular issue for BoxBorderPainter as edges were being
added to the visible_edge_set_ based on their float alpha, but
opaciy_groups were being built on the basis of integer alpha. Thus, a
very-nearly transparent edge would be added to the visible_edge_set_,
but then get skipped by the opacity_group, leading to an index out of
bounds crash in release builds.

This code was surrounded by DCHECKs to prevent this very kind of thing
from happening, but clearly the code is untested and this situation is
rare for dev builds, as the crashes only began appearing for canary
users without DCHECK.

Bug 1445351 is more straightforward, but has a similar cause. A DCHECK
was checking for integer alpha in a function that assumed float alpha.

This CL speculatively solves two release blockers, so it should be
merged asap. After that I will focus on resolving the missing test
coverage.

Bug: 1445288, 1445351
Change-Id: I03e61176e2c3e2e6f86a770793e3e6d2bb832941
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/4533018
Reviewed-by: Aaron Krajeski <aaronhk@chromium.org>
Reviewed-by: Justin Novosad <junov@chromium.org>
Commit-Queue: Aaron Krajeski <aaronhk@chromium.org>
Cr-Commit-Position: refs/heads/main@{#1144252}