Re: [lmi] Segfault in lmi with wx-2.9.5

[Vadim Zeitlin]

On Sat, 13 Jul 2013 13:51:20 +0000 Greg Chicares <address@hidden> wrote:

GC> I guess all I can do is try somehow to reproduce it.

 This is almost certainly not the same problem but I just ran into a
perfectly reproducible crash: after starting LMI, press "Ctrl-N|I" and
Escape when the illustration dialog is shown. The program immediately
crashes. Unfortunately the version I built (using the official install
script) doesn't seem to be debuggable so I don't know why exactly does this
happen, nor do I know if it is new.

GC> Would it be feasible to log all events, so that once a
GC> random segfault is observed, a log could be played back
GC> with a fair possibility of reproducing the problem?

 This is an interesting idea but I'm afraid it's going to be difficult to
realize in practice because the user events are not the only external
factors affecting the program. In any case, there is nothing like this
right now.

GC> Should the planned GUI-testing automation have a "random"
GC> mode that simulates a thousand monkeys entering arbitrary
GC> (logged) input to find latent problems like segfaults?

 Again, this would be certainly worthwhile to have and I did think about
this but it's not as simple to implement as that because even a thousand
monkeys risk to not be enough to exercise a non-trivial part of all the
possible entries of a decently sized GUI if they stupidly try all the
possible variants -- at least not unless you also give them a thousand
years to do it. So I think we'd need smarter monkeys (the difference in
practice between Haskell QuickCheck and fuzzing testers for normal C
programs is that the former is really useful for everyday testing because
it's smart and uses its knowledge of the program provided by the Haskell
type system while fuzzers are only attractive to people really dedicated to
using them) but building smarter monkeys is hard...

VZ