abrt wishlist

[David Malcolm]

On Thu, 2010-12-09 at 10:51 -0500, Adam Jackson wrote:
> Here's some of my pet irritations with abrt.  Feel free to add your own,
> but please keep the gratuitous "me too"ing to a minimum.
> 
> 1) The generated reports contain far too little information for library
> owners.  Consider this report:
> 
> https://bugzilla.redhat.com/show_bug.cgi?id=658013
> 
> I get the EVRs for the kernel, and for the executable that happened to
> be running, but not for any of the libraries in between, even though the
> report very carefully tells me exactly which DSOs are loaded and where
> they are in memory.
Gratuitous me too!

But more seriously, I filed this a while back as:
  https://bugzilla.redhat.com/show_bug.cgi?id=555205
"RFE: include "rpm -qf" output for all memory-mapped DSOs"

This would be a great boon to things like python where you can have an
arbitrary amount of DSOs mapped into your process, from pretty much the
whole of the distribution (which is both a good and a bad thing; would
love to sandbox python modules, but that's an unsolved problem, AIUI).

> 2) I really dislike that "local trace generation" and "retrace server"
> are discussed as though they're the only options.  If nothing else, for
> many non-trivial apps where abrt is potentially of the most use the core
> you're uploading can easily be hundreds of megabytes; that's not really
> better than downloading hundreds of megabytes.  A network debuginfo
> service [1] would approach this problem in a completely different way,
> by letting the client download exactly as much debuginfo as it needs.

Another gratuitous me too, see:
  https://fedoraproject.org/wiki/Talk:Features/RetraceServer

I'm wondering why the network debuginfo service couldn't be simply wired
up to Koji's rpm via readonly NFS and scrape debugingo rpms on demand.

> 3) Reporting to bugzilla is a mistake.

FWIW, I'll mention again my ABRT triage scripts; they help me keep sane
whre dealing with incoming crash reports:
http://fedorapeople.org/gitweb?p=dmalcolm/public_git/triage.git;a=tree

(The heuristics there are targetting python crash reports, since that
what I get)

> 4) Reporting to bugzilla without being able to scrape the bz username
> and password out of the firefox credential store is just cruel.

Seems like this should in bugzilla.



Hope this is helpful
Dave