GitLab

This is a partial revert of patch set 4 in CL:2515399 (to the initial
patch set). The change in patch set 4 was done in order to improve
readability, but with one oversight: We'd set a non-zero distance when
actually within range, since we'd always subtract one edge from another.

This helps fix 6 tests, but the tests themselves also had to be updated,
because legacy PositionForPoint() for multicol is broken when hitting
above or below columns, in that it assumed UNIX / Mac editing behavior,
which makes a difference when hitting just above the character at e.g.
offset 2 in the first column. UNIX / Mac would return 0 as startOffset
("above the first line... okay, we're before the text, then"), while
Windows should return 2 ("we're above the first line, but the nearest
character is the one at offset 2 - use that"). Force Windows editing
behavior in the tests, to fix this.

The PositionForPoint() implementation for LayoutNG multicol is now
behaving way better than the legacy engine, which allows us to add the
missing parts to those tests (what we tested for vertical-rl was really
limited due to legacy bugs). Legacy was also off-by-one for RTL columns,
when hitting near the middle of a column gap (assigning one pixel too
many to one column, and one too little to the other column).

Rebaseline the tests. Everything now passes in LayoutNG, but there are
obviously failures when running with the legacy engine now.

Bug: 829028
Change-Id: Iecdecf1454318e645fa91bac29119df0fc7ca377
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2526402Reviewed-by: Ian Kilpatrick <ikilpatrick@chromium.org>
Reviewed-by: Koji Ishii <kojii@chromium.org>
Reviewed-by: Yoshifumi Inoue <yosin@chromium.org>
Commit-Queue: Yoshifumi Inoue <yosin@chromium.org>
Cr-Commit-Position: refs/heads/master@{#825689}