Re: 1000 bugreports in one day, every day

[Brent Smith <gnome nextreality net>]

Olav Vitters wrote:
> On Sat, Jun 03, 2006 at 08:46:16AM +0200, Vincent Untz wrote:
> > Le jeudi 01 juin 2006 à 22:39 +0200, Olav Vitters a écrit :
> [..]
> > > This will remove a lot of the reasons I can think of why we are not
> > > getting a 1000 bugreports a day. So I have to wonder what happens when
> > > 2.16.0 is released with the perfect Bug-Buddy.
> > > There are a few more ideas to make filing bugreports easier. What
> > > happens if those are implemented? Will Bugzilla still be usable?
> > While 1000 bugs per day would be nice for my bugzilla points (most of
> > the bugs should be duplicates for crashers), there's no way for us to
> > manually deal with this number of bugs.
>
> I agree, so what are our options? The XML-RPC Bug-Buddy will likely be
> released soon. That would only be for 2.15.x (only a slight increase).
> The 1000 bugreports a day would only be if the new Bug-Buddy avoids all
> the reasons why we are not getting these bugreports today. And also it
> would be less than 1000/day initially. It will go up as more stable
> GNOME versions have it (2.16.x, 2.18.x, etc).

Here is my two cents:

1) I think we need to figure out some client side checks we can do to
automate what the bugsquad is already doing... That is, we need to parse
the stacktraces and if there are no debugging symbols, tell the user to
to install debug packages, or point them at the relevant URLs.  This
will both reduce the number of bugreports and the load on the bugzilla
server.  I would hope that we could come up with some regular
expressions to look for ?? for the first 5 or so stack frames before the
SEGV signal handler is called.  This would probably only be a couple
days of work at most to code and test this in bug-buddy.

2) Presenting the user with an interface for simple-dup-finder would be
great, but we should only do this if the stacktrace passes the debugging
symbols check _AND_ if we perform the simple-dup-finder query and it
returns 5 or fewer hits with the same five frames before the SEGV
handler is called.  This is tricky, since often it's hard for bugsquad
members to notice subtle difference between two similar stacktraces.
Also tricky because it sounds like someone would have to write some
XML-RPC client/server code to get this functionality.

I looked at the bug-buddy-xmlrpc branch today; The new bug buddy is dead
simple to use and a massive improvement over the old one IMO.  I'm
excited to see it included in the next release.

--
Brent Smith <gnome nextreality net>
IRC: smitten