Add support for accessible tables with missing/bad data.

Previously we relied on Blink to correctly fill in the
row and column index of every cell in a table. This
tends to be expensive to do in Blink because it
requires invalidating and reserializing the entire table
if any cell changes.

Instead, try to defer as much of those calculations to
downstream to be computed when AT actually requests
table attributes.

This is more efficient, because Blink accessibility code
needs to run any time the page is updated, like when a
table is being built dynamically using JS - whereas AT
typically only queries table attributes when users
explore a table explicitly, so it makes sense to defer
computation until then.

This is also more accurate and less error-prone, as it
allows us to completely validate the table in a single
pass and fix any errors in table indexes.

Finally, a potential added benefit is that this makes it
easier to handle tables from Views or Arc++ that
don't have all of the table attributes set, we can
compute them automatically.

A follow-up change will remove the code that computes
some of these table attributes from Blink.

Bug: 832289

Change-Id: I252a592c854e47cf6d6e5bd6b7c4bc5d51775e72
Reviewed-by: Aaron Leventhal <>
Commit-Queue: Dominic Mazzoni <>
Cr-Commit-Position: refs/heads/master@{#606940}
5 files changed