False likely-duplicates from firefox clue file.

Alexander Sack asac at jwsdot.com
Sat Jun 16 18:03:25 BST 2007


On Sat, Jun 16, 2007 at 05:09:32PM +0200, Hilario J. Montoliu wrote:
> Hi Freddy et al.,
> 
> I did today some work at dholbach firefox's bughelper file ([1]) and
> I've noticed that *most* (68) of the report listed are marked as
> possible duplicates of bug #96544 ([2]) "[edgy] firefox crash
> [@nsCOMPtr_base::assign_with_AddRef] [@nsCSSSelector::Reset]"; which is
> a report that I rejected some days ago for being a *one timer crash*.
> The clue for it includes "nsCOMPtr_base" and "nsProfileLock" which are
> quite common in the retraced stacktraces. I don't know (as I didn't
> follow the process) how bugs are added to the clue file, but I think
> that this one should be removed from the firefox clue file.

Yes right ... maybe we should duplicate the most common ambiguous
stack signatures somewhere, so clue file maintainers can take that in
consideration?

This might as well be beneficial for the ungoing work to improve
auto-retracers and auto-duplicate detection.

> 
> There is also another bug report that takes another high amount of
> possible duplicates (64), bug #14911 ([3]) "Flash plugin problem with
> ARGB visuals causes crash", and that happens because the search string
> in the clue file is just "ARGB". Since bughelper looks even inside full
> crash reports that short string matches almost every coredump attached
> to the full crash reports. IMHO the clue file for bug #14911 should be
> improved.

Can we constraint the clue match to not look into attachments? I hope
that this will reduce the match for ARGB a lot.


 - Alexander




More information about the Ubuntu-mozillateam mailing list