In M59 the ChromeVox focus ring regressed, it used to correctly outline the bounds of an object, but this change broke it:

r460412
https://codereview.chromium.org/2762373002
"Fix computation of Automation API location offsets in WebViews"

The fix was supposed to be for  bug 658947 , which was that the focus ring was incorrect in the sign-in screen.

The fix worked correctly for the sign-in screen, and was technically an improvement, but it exposed other bugs and the overall result is *worse*, not better.

Since then we've fixed the issue on trunk, however the most recent fix is larger and I'm not sure we want to merge it.

I think the safest solution is to *revert* the above change, on the M59 branch. That brings M59 back to the previous state where it was mostly working fine but off on the sign-in screen. Then in M60 we'll fix the remaining issues so that it works correctly everywhere.

I just did some manual testing on the M59 branch.

Currently, the focus ring is way off. Turn on ChromeVox and tab around any web page to see.

If I revert r460412, it reverts completely cleanly, and virtually all windows and web pages now work fine. The focus ring goes back to being wrong on the sign-in screen, but everywhere else it's correct.

Reverting r460412 on trunk would not be easy because other changes depend on it. Plus, the situation has already been fixed properly on trunk - the focus ring is correct everywhere including the sign-in screen.