STR: 1. With NVDA and Firefox, open http://whatsock.com/tsg/Coding%20Arena/Modals/Modal%20(Internal%20Content)/demo.htm. 2. Tab to the Login button and press Space. The login form appears. 3. Press tab once to give focus to the User Name field. 4. Press NVDA+Space to turn off focus mode. 5. UpArrow a few times until you land in a list of 3 items. Expected: This content should no longer be accessible to NVDA. Actual: It reads the contents of that explanatory text which is now hidden. Only if you uparrow more and focus reaches an element that is not supposed to have focus, focus gets redirected to the User Name field. Examining the tree, the elements outside the login group are all still there. The headings, the list etc. are still part of the main document. According to Bryan's report via e-mail, this used to work fine, with the content outside the login form being hidden from screen readers. Alex, can you take a look, please? Is thsi a code technique not covered by our tests and thus we missed that this should work?

Is it intermittent issue? Does anyone know whether we expose hidden object attribute in aria-hidden subtree? If we do, then it should be a screen reader issue.

(In reply to alexander :surkov from comment #1) > Is it intermittent issue? Does anyone know whether we expose hidden object > attribute in aria-hidden subtree? If we do, then it should be a screen > reader issue. It is not intermittent here, and no, I don't see that attribute on the accessibles that shouldn't be there. I can reproduce it 100%.

It probably doesn't make much sense to fix this bug on its own, if we plan to work on bug 1349223 (cut aria-hidden). In other words bug 1349223 should be a fix for this one.

This scenario is covered within the general testable statements within the archive at https://github.com/whatsock/w3c-alternative-text-computation And I believe has already been added to the W3C testable statements wiki, or soon will be.

This now works as expected after bug 1249223 was fixed.