Sometimes, the screen location returned by accessible objects seems to be incorrect. Str: 1. Open Google Calendar. 2. Open the Create Event page. 3. Press alt+d to move to the Location Bar, then tab a few times until focus lands on the "Mail" link. 4. Press alt+d to move to the location bar, then shift+tab a few times until focus lands on the Discard button. 5. Check the location returned by the focused accessible. Expected: The location returned should be the location of the button on screen (inside the browser window). Actual: The location returned is outside the browser window. I've seen this issue in quite a few other places and it makes routing the mouse to and/or clicking these elements impossible, which is sometimes necessary. It has made some nasty sites literally unusable for me with NVDA. The example above is the easiest and most reliable to reproduce. Unfortunately, I have no idea at all what is causing this, so I haven't been able to come up with a simple, isolated test case. This works fine in Firefox 3.6.
Jamie, can you re-test this with latest nightlies? We just recently got a bug fix in that fixes a few child/deepest child at certain locations issues.
I can still reproduce as described using Mozilla/5.0 (Windows NT 6.1; WOW64; rv:6.0a1) Gecko/20110502 Firefox/6.0a1
Thanks, so bug 650235 didn't fix this one.
Can't reproduce with accprobe on trunk, it shows red rectangle around discard button after I do steps to reproduce. Any idea, how else I can try to see a bug?
Again the question: Is this maybe fixed by bug 656089?
(In reply to comment #5) > Again the question: Is this maybe fixed by bug 656089? I have doubts because bug 656089 is about caret rect, this one is about accessible object rect, they are calculated independently.
Created attachment 544724 [details] Test case. Okay. Finally figured out what's causing this particular case. 1. Open the attached test case. 2. Press control+0 to reset page zoom. 3. Check the top left of the location reported for the first paragraph accessible. Result (correct): The point is within the paragraph. 4. Press control+= (control+equals) twice to zoom the page in. 5. Check the top left of the location reported for the first paragraph accessible. Expected: The point should be the top left of the paragraph. Actual: The point is in the top part of the browser chrome.
regression from bug 528796, lack of testsuite, again.
Created attachment 599142 [details] [diff] [review] patch
Comment on attachment 599142 [details] [diff] [review] patch r=me with just one nit: >+ * Return the accessible coordinates and size relative the screen. "relative to the screen"
Created attachment 599495 [details] [diff] [review] improve testing
Created attachment 599506 [details] [diff] [review] improve testing v2 updated to trunk
Comment on attachment 599506 [details] [diff] [review] improve testing v2 r=me, veeery nice!
https://hg.mozilla.org/mozilla-central/rev/bde8c5a15988 (Leaving open for test to merge)
I'm confused, there was another cset with this bug number (comment 17), but it's not the test. Reopening; suspect wrong bug number in commit message for one of them - can I leave you to sort please?
(In reply to Ed Morley [:edmorley] from comment #18) > I'm confused, there was another cset with this bug number (comment 17), but > it's not the test. > > Reopening; suspect wrong bug number in commit message for one of them - can > I leave you to sort please? there are three landings here 1) original patch + mochitest (central) 2) disabled test because mochitest of original patch failed (central) 3) improved test (inbound) so this changeset was #2 (no patch is attached here).
(In reply to alexander :surkov from comment #19) > (no patch is attached here). Ah - thank you :-) https://hg.mozilla.org/mozilla-central/rev/499846502bf3