Should this block 1.5 final as it is stopping GreaseMonkey from working properly? See: http://greaseblog.blogspot.com/2005/09/firefox-15-compatible-greasemonkey.html
sounds like a major regression that impacts a popular extension. Pulling into the blocking radar, but time is really short and we don't yet have anyone signed up to investigate and fix. Aaron, if you can help us find resources, or figure out exactly what broke this, that'd be a big help.
Bad timing: Aaron's in Vietnam until the end of this week, I think.
I'll try to pull this in. I'm hoping there's going to be a simple fix for this. I really wish that I'd known about this weeks ago :-(.
I'm hoping that this fix will be evalInSandbox-specific (and thus low-risk).
Created attachment 199235 [details] [diff] [review] Something like this This, sadly, isn't evalInSandbox-specific. It gets the current window off of the docshell if it can't get it off of the context. I think this can only happen in the "extensions" case. jst, what do you think?
Comment on attachment 199235 [details] [diff] [review] Something like this r- based on discussion with mrbkap. This would make us use the wrong URL when as the base when resolving relative URLs on location objects from a window other than the callers.
Created attachment 199607 [details] [diff] [review] updated to comments
Comment on attachment 199607 [details] [diff] [review] updated to comments *sigh*, the iterator needs to go _backwards_ (iterating over a stack, and all).
Created attachment 199619 [details] [diff] [review] with those changes This has r=jst looking over my shoulder.
Comment on attachment 199619 [details] [diff] [review] with those changes I nagged mrbkap on IRC about not sharing the common return from GetContextFromStack, allowing the break to become a return NS_OK and the final return to be immediately preceded by an unconditional *aContext = nsnull. /be
Blake, we need this landed and verified on the trunk ASAP. Thanks.
Fix checked into the trunk (with another small tweak to make the iterator handle an empty deque).
Neither of my depends builds register the new iterator, which busts all sorts of things :-( BTW, why is ContextStackIteror not spelled correctly?
FYI, I fixed the misspelling (thanks copy/paste) on the trunk today. Neil, I'm not sure why your builds are failing. I can't reproduce that problem either on my Linux build or my Windows build...
Seems to have been some weird compreg.dat issue, it didn't reregister properly. Deleting compreg.dat fixed that and also found some obsolete components ;-)
Neil, is that a verification? Have you got a working build yet?
Checked into MOZILLA_1_8_BRANCH.