Re: Serious "memory leaks" problem in gnustep 0.8.0

[Nicola Pero]

> After numerous tests I have found out a problem with memory leaks in gnustep 
> - 0.8.0.
> In the example I am enclosing, the application generates one window with
> NSTextField item, and cyclically (by timer) redraws it.
> After a few minutes it can be noticed that the application starts to
> increase its memory, and after a few hours the size of memory will be
> multiplied (in gnustep 0.7.5 this problem did not arise).
> 
> Can this error be caused a fault in my system?
> (Solaris 8, x86 platform, SPARC platform)

Quite interesting.

I had a quick try at running your example on my Debian/ix86.  I can see
the leak from top ... memory increases slowly, but constantly.  Actually,
it increased by 400k in ten minutes maybe, so it's not that slow.  :-)

I monitored the object allocation using gnustep-base's debug facilities -
I didn't find any signs of a memory leak of objects.  So I suppose there
is a leak of plain memory allocated via malloc() or objc_malloc() or a
similar low-level function.

If you can shed more light on the matter, it would be of course very
interesting.

Assuming the bug is not in the base library, the other obvious possible
place where such a leak might be produced is in the backend, because it's
where most of the malloc() calls are done.

Keep us updated on any new findings on the matter.  Maybe someone from our
developers' team has the time (and will) to play with this and track it
down ? :-)

I'll be away the weekend.