Re: Segfault when using readline and dynamically linked code

[rm]

On Fri, Oct 05, 2001 at 08:29:04PM +0100, Neil Jerram wrote:
> >>>>> "ralf" == ralf  <address@hidden> writes:
> 
>     ralf> Hello everyone, i just encountered a strange problem: i
>     ralf> wrote some small wrapper code that creates guile bindings
>     ralf> for some external library. Everything works fine (thank's
>     ralf> for the nice documentation) _unless_ i activate readline
>     ralf> support.  As soon as i use the tab key guile segfaults.
>     ralf> Using readline without my custom module works without any
>     ralf> problems, so i assume the segfault is caused by my code. As
>     ralf> a starting point i compiled guile and my module with
>     ralf> debuging support, but the backtraces i get aren't very
>     ralf> helpfull:
> 
>     ralf> smobs. Any pointers to where i might start looking?
> 
> The backtrace shows the GC coming to a null pointer.  My guess would
> be that you've written a SMOB mark function that is telling the GC to
> mark something that is actually NULL.

Thank you for our fast reply! Yes, that was indeed the case (as i figured
out after a bit more thinking and gdb-ing ...). I switched from 'heavy' smobs
(smobs containing references to other SCM values) to 'light' ones (where the
smob is just a wrapper arround a pointer to a C struct but missed to change
one of the smob-mark functions. Darn! Sorry for this unneccessary work :-/
   
Just out of curiosity: I find it rather interessting that the gc behaves
so different if readline is activated. I wasn't able to produce a coredump
without readline, even under heavy memory load.

Thanks again for your help (and the great work on guile)

  Ralf Mattes