RE: 6.2.1 vs 6.0.1

On 21 June 2004 10:34, Arjan van IJzendoorn wrote:

> Thanks for all your replies to my previous message about the
> "overflown relocation field".
> 
> I was trying to find a bug that makes my application crash. I failed
> at that but when I switched from using GHC 6.2.1 to version 6.0.1 (on
> Windows XP) the bug disappeared! That makes me happy because now I
> can release my software to my "customers" again. Still, it should
> work with 6.2.1, too. 
> 
> The bug only shows up now and then, completely impredictable, it
> seems. It complete crashes my program and I get a Windows dialogue
> asking me whether I want to send an error report to Microsoft :-) If
> I debug in Visual Studio the stack trace points into my code and not
> into the libraries I use (wxHaskell and SMILE (a C++ library for
> probabilistic networks)). 
> 
> So, I have a quite big application (a graphical editor for
> probabilistic networks) that uses two complicated libraries that
> crashes every once in a while under GHC 6.2.1 and not under 6.0.1. I
> was wondering how to report this problem to you. Has there been a
> significant change in the code generator since 6.0.1? I know that
> Martijn Schrage had exactly the same problem (random crashes) with
> his (huge) Proxima editor. That application also uses wxHaskell so it
> could be an interaction between GHC's code and wxHaskell...

<grumble>
Urgh.  I hate these kind of bugs, because it takes me half a day to get
up to the point where I can reproduce it, let alone debug it... 
</grumble>

Are you using any FFI stuff?  Any chance there could be a problem with
that?

Is there anything you can do to try to make the bug deterministic?  Is
your app scriptable, for example?  Does changing the heap settings make
it crash more often?  Try a small allocation area: +RTS -A16k.

Try compiling with -debug, and run with +RTS -DS (sanity checking).  Do
you get any assertion failures?

When you've exhausted all the ways you can think of to make the bug more
predictable and the code smaller, could you try to package up the code
with instructions and send it to us?

Cheers,
        Simon