IndexedDB: Closing connection in upgradeneeded should result in error event
Created attachment 168991 [details] Patch
This is a weird one. I think the behavior is more correct with this patch (and the updated test looks more correct), but I'm still feeling iffy about whether logic like this should go in enqueue vs. dispatch.
Comment on attachment 168991 [details] Patch LGTM In the chromium bug you mentioned that this might just be postponing the flakiness but I can't see how failures would occur after this patch.
(In reply to comment #3) > In the chromium bug you mentioned that this might just be postponing the flakiness but I can't see how failures would occur after this patch. Yeah, I agree - I don't recall what edge cases I might have been thinking of. As discussed offline, a more "elegant" approach than this patch would be to have the "complete" event dispatch send a message to the browser to unblock the success/error event, but this patch should have the equivalent from user script.
Comment on attachment 168991 [details] Patch Actually, seeing some tests fail locally that may do an open and never see a success or an error.
Created attachment 169279 [details] Patch
(In reply to comment #5) > Actually, seeing some tests fail locally that may do an open and never see a success or an error. Whoops, that was a false alarm - I was running NRWT with a manually set and very low timeout. tony@ - r?
Comment on attachment 169279 [details] Patch Clearing flags on attachment: 169279 Committed r131668: <http://trac.webkit.org/changeset/131668>
All reviewed patches have been landed. Closing bug.