steps to reproduce: 1) run a test case from bug 1347075 2) stop debugger at MOZ_ASSERT(origContainer == prevChild->Parent(), "Broken tree") in DocAccessible.cpp https://dxr.mozilla.org/mozilla-central/source/accessible/generic/DocAccessible.cpp#2193

aria-owns on a document puts back its aria owned children on next ValidateARIAOwned(), which eventually leads us to broken tree, perhaps through a series of other bugs. By fixing aria-owns on a document, I don't longer see any of assertions added in bug 1347075.

Comment on attachment 8863799 [details] [diff] [review] patch Review of attachment 8863799 [details] [diff] [review]: ----------------------------------------------------------------- Thanks! I don't really understand the commit message. Can you rephrase it or expand on it so that I can understand what is happening without debugging? Note in general we should only add e10s happy browser tests until we have a way forward for our mochitests surviving. ::: accessible/tests/mochitest/treeupdate/test_ariaowns.html @@ +586,5 @@ > + }) > + ]; > + > + this.invoke = () => { > + getNode('t9_container').src = nit: unecessary trailing whitespace.

(In reply to David Bolter [:davidb] from comment #3) > Thanks! I don't really understand the commit message. Can you rephrase it or > expand on it so that I can understand what is happening without debugging? technically there's no commit message to understand, right? comment #1 though has a description of the change, the first part is the most important: "aria-owns on a document puts back its aria owned children on next ValidateARIAOwned()". Eventually fixing aria-owns on a document, fixes this bug. Does this one need to be extended? > Note in general we should only add e10s happy browser tests until we have a > way forward for our mochitests surviving. can you tell more, what's wrong with mochitest in e10s environment?

(In reply to alexander :surkov from comment #4) > (In reply to David Bolter [:davidb] from comment #3) > > > Thanks! I don't really understand the commit message. Can you rephrase it or > > expand on it so that I can understand what is happening without debugging? > > technically there's no commit message to understand, right? comment #1 > though has a description of the change, the first part is the most > important: "aria-owns on a document puts back its aria owned children on > next ValidateARIAOwned()". Eventually fixing aria-owns on a document, fixes > this bug. Does this one need to be extended? Yes please. I think I need to understand the sequence in detail. I know you have tests but if you can describe here that would help me. > > > Note in general we should only add e10s happy browser tests until we have a > > way forward for our mochitests surviving. > > can you tell more, what's wrong with mochitest in e10s environment? A good question for RyanVM :)

(In reply to David Bolter [:davidb] from comment #5) > (In reply to alexander :surkov from comment #4) > > (In reply to David Bolter [:davidb] from comment #3) > > > > > Thanks! I don't really understand the commit message. Can you rephrase it or > > > expand on it so that I can understand what is happening without debugging? > > > > technically there's no commit message to understand, right? comment #1 > > though has a description of the change, the first part is the most > > important: "aria-owns on a document puts back its aria owned children on > > next ValidateARIAOwned()". Eventually fixing aria-owns on a document, fixes > > this bug. Does this one need to be extended? > > Yes please. I think I need to understand the sequence in detail. I know you > have tests but if you can describe here that would help me. * you put aria-owns on a document element * the aria-owned children are relocated under the document by DoARIAOwnsRelocation * then ValidateARIAOwned puts them back on next WillRefresh, which is a bug that the patch fixes that's basically it > > > Note in general we should only add e10s happy browser tests until we have a > > > way forward for our mochitests surviving. > > > > can you tell more, what's wrong with mochitest in e10s environment? > > A good question for RyanVM :) is there anything I should do for this bug?

Comment on attachment 8863799 [details] [diff] [review] patch Review of attachment 8863799 [details] [diff] [review]: ----------------------------------------------------------------- Thanks. For future tests, let's add them to /browser until we know more.

(In reply to David Bolter [:davidb] from comment #7) > For future tests, let's add them to /browser until we know more. I would prefer to know more before jumping into new area for sure. Afaik, a11y browser tests infrastructure is not that rich as mochitest one, and I wouldn't invest into that much at this stage without a strong reason.

https://hg.mozilla.org/integration/mozilla-inbound/rev/8a2b71463a1592a56c686009fb4d16a776e9c52c Bug 1355488 - fix aria-owns on a document, r=davidb

Pushed by surkov.alexander@gmail.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/8a2b71463a15 fix aria-owns on a document, r=davidb