Le dimanche 17 janvier 2010 à 12:53 +0100, Michael Schwendt a écrit :
On Sun, 17 Jan 2010 12:36:03 +0100, Nicolas wrote:
> Le samedi 16 janvier 2010 à 15:09 -0500, Tom Lane a écrit :
> > Users have to provide information
> > about what they were doing, copies of input files, etc etc just
the
> > same as in a manually-initiated bug report.
>
> IMHO the big plus of abrt is it triggers even when the user is not
> giving his full attention to the app and not checking what it does
> exactly when it crashes (typical example is multitasking and doing
stuff
> in 3-4 apps when one dies). There is a huge class of crashes that
were
> not reported before because the user had no idea what the app was
doing
> exactly when it crashed and could not reproduce it with debuginfo
later.
A downside is that ABRT is triggered for all sorts of weird
memory/heap
corruption that isn't reproducible. Stability problems with RAM chips
are widespread.
A bugzilla stock response that points at "memtester" and
"memtest86+"
will likely be needed more often.
That seems totally unecessary and counter-productive to me. You can
distinguish between local memory problems and actual hard-to-trigger
bugs without bothering users by checking if the trace is reported by
abrt for other systems.
I know it's very human to shoot the messenger but packagers &
developpers should resist the urge to make tester life miserable to
punish them from reporting inconvenient problems.
--
Nicolas Mailhot