345618 - "GetPrimaryFrameFor() called while nsFrameManager is being destroyed!" reloading page with misspelled words

Status: RESOLVED DUPLICATE of bug 459613

(Core :: Spelling checker, defect)

Description

[Jesse Ruderman]

Comment 1

[Jesse Ruderman]

Attachment: testcase (reload to reproduce)

Comment 2

[Martijn Wargers (dead)]

This is worksforme with my debug build, so I guess this could be fixed by one of the latest spellchecker patches. Or this might be a Mac specific issue, I guess.

Comment 3

[Jesse Ruderman]

Still happens for me using an updated trunk build on Mac.

Comment 4

[Ryan VanderMeulen [:RyanVM]]

This appears to be a Mac-only bug. I can't reproduce on Windows XP SP2, but I got vlad to reproduce on a Mac trunk build.

Comment 5

[Jesse Ruderman]

CCing Aaron, who added this printf in bug 129628.

Comment 6

[Aaron Leventhal]

Well I guess if it wasn't for that fix this would probably be crashing :)

It should be an NS_WARNING I think, not a printf. Anyway look into who's calling GetPrimaryFrameFor and whether it should be avoided at that point. They're calling it during teardown.

Comment 7

[Jesse Ruderman]

This still happens on trunk, and it causes a lot of console noise when browsing around Bugzilla.

Comment 8

[Uri Bernstein]

This is a regression from bug 339434, where a call to spellCheckSelection->RemoveAllRanges() was added to mozInlineSpellChecker::Cleanup().

Comment 9

[Uri Bernstein]

Attachment: patch

Don't call RemoveAllRanges() when destroying the editor - only when disabling real time spellchecking.

Comment 10

[Boris Zbarsky [:bzbarsky]]

How does that work when the editor is destroyed but the frame tree is not (e.g. contentEditable/designMode being toggled)?  Or do those cases not destroy the editor?

Comment 11

[Uri Bernstein]

Comment on attachment 343838 [details] [diff] [review]
patch

It doesn't work in that case, the dotted underlines remain shown.

Comment 12

[Uri Bernstein]

So, I don't know how to fix this.

Comment 13

[Boris Zbarsky [:bzbarsky]]

One way seems to be to have a way to ask preshell whether it's destroying frames, and if so to not call this code in that case...  or something.  Whatever "this code" is.

Comment 14

[Jesse Ruderman]

WFM.  Uri/bz, do you happen to know what fixed this?

Comment 15

[Boris Zbarsky [:bzbarsky]]

I have no idea...  roc might know, maybe.  That said, if we still do the RemoveAllRanges during teardown, that's a bug that we should still fix, even if it's not asserting on this particular assert for some reason...

Comment 16

[Timothy Nikkel (:tnikkel)]

Bug 459613 would have fix this (which is access denied for me, but I can tell based on the changeset http://hg.mozilla.org/mozilla-central/rev/2cbcaca3fcdb).

Comment 17

[Boris Zbarsky [:bzbarsky]]

Ah, indeed.  See, I _knew_ roc would have known!  ;)