GitLab

This change introduces two main optimizations for move-by-format. Both
of these optimizations combined end up improving performance of Narrator
heading moves by 25% on NYTimes.com and 40% on a reduced test page.
I'm now seeing the format move cost dwarfed by the GetAttributeValue
cost for Narrator Scan Mode move-by-heading, which I'll look into next.

The first category of improvements in this change is from replacing
AsLeafTextPosition calls with AsLeafTreePosition. When we're moving
by format in intermediate steps, it doesn't matter whether the node
is a tree position or a text position, so this avoids expensive
MaxTextOffset calls.

The second category involves returning a reason why a format boundary
was determined. I saw that we were calling
Create(Next/Previous)TreePosition several times just to see if a move is
at the end of a document. So a more complex return value that provides
the reason of why a format boundary exists allows us to not have to do
that work multiple times. The AXBoundaryType enum class added with this
change provides this information, with the new method
'GetFormatEndBoundaryType' doing the work of determining format
boundaries and returning the appropriate AXBoundaryType.

Bug: 1029867
Change-Id: I12151ab2b4000227e1d5e2726223a7545d73bc2e
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1965865Reviewed-by: Kevin Babbitt <kbabbitt@microsoft.com>
Reviewed-by: Nektarios Paisios <nektar@chromium.org>
Commit-Queue: Nektarios Paisios <nektar@chromium.org>
Cr-Commit-Position: refs/heads/master@{#727375}