abrt wishlist

[Adam Jackson]

On Mon, 2010-12-13 at 10:50 -0500, James Laska wrote:
> On Thu, 2010-12-09 at 10:51 -0500, Adam Jackson wrote:
> > 3) Reporting to bugzilla is a mistake.
> 
> Not discounting the idea, but just looking for more detail.  What
> alternatives would you want to see?  More kerneloops-like aggregate data
> collection, or something else?

Basically, yes.  For a few reasons:

- Crash analysis requires more semantic knowledge of the content of the
report than bugzilla is really designed for. I can have 17 reports with
different backtraces all in different applications, but where the bug is
"this exclusive section in this library is being called without the lock
held".  It's not reasonable to expect to add that kind of content
awareness to bugzilla, and doing it from bz clients is clumsy at best.

- Bugzilla forces you to frame the discussion in terms of a component.
That's probably right for applications.  It's usually wrong for
libraries, drivers, servers, or kernels.  You want to start from the
report as a gestalt, and not assign blame to a component until you
actually know what's going on.  (This is a condemnation of bugzilla in
general, but it's made worse by the next point.)

- Separating machine-generated content from human-generated content is
valuable for the developer.  The two require different mental processes
to handle.  I have a much stronger guarantee that the abrt bug contains
facts, but I also know there's no point in asking for more information.
Reading a crash report is looking at structured data and divining
patterns.  Reading a human's bug report is listening to a story.  Left
brain, right brain.

- ajax