[Debian GNUstep maintainers] Bug#592751: Broken on hppa; programs abort with malloc assertion failure

[dann frazier]

On Mon, Aug 16, 2010 at 09:18:51PM +0300, Yavor Doganov wrote:
> dann frazier wrote:
> > Same thing:
> 
> OK.  Another try with 1.20.0 this time?
> (http://snapshot.debian.org/archive/debian/20100524T154820Z/pool/main/g/gnustep-base/gnustep-base_1.20.0-1.dsc)

locally built 1.20.0:

root at c3700:/tmp# gdnc --help
gdnc: malloc.c:3097: sYSMALLOc: Assertion `(old_top == (((mbinptr) (((char *) &((av)->bins[((1) - 1) * 2])) - __builtin_offsetof (struct malloc_chunk, fd)))) && old_size == 0) || ((unsigned long) (old_size) >= (unsigned long)((((__builtin_offsetof (struct malloc_chunk, fd_nextsize))+((2 * (sizeof(size_t))) - 1)) & ~((2 * (sizeof(size_t))) - 1))) && ((old_top)->size & 0x1) && ((unsigned long)old_end & pagemask) == 0)' failed.
Aborted

I would've tried the buildd-generated binaries as well, but it looks like
this version never built on hppa.

     -dann

> 
> Upstream thinks this is either a GCC/libobjc issue exposed by the new
> -base code, or a serious problem in gnustep-base itself.
> 
> As I can't reproduce the nefarious behavior on gcc61.fsffrance.org
> with a manually built GCC 4.4.4 from pristine source, I'm now building
> a new gcc with Debian patches.
> 
> If I still can't reproduce, and 1.20.0 fails on Debian machines in the
> same way (which I suspect so), I'm afraid you or someone else from the
> hppa folks gotta take the hard road of bisecting...
> 
> Upstream said:
> 
> | My only new idea on how to proceed is to try a bisection of the code
> | changes since 1.19.3. You should start of by testing 1.20.0 and then
> | use SVN versions of the code in between.
> |
> | We have too much that changed since 1.19.3 to tell otherwise. The
> | whole thread and lock handling was completely rewritten, the
> | interaction with the Objective-C runtime was reimplemented, so were
> | NSInvocation and NSNumber. We enabled 64 bit NSInteger support and so
> | much more.
> 

-- 
dann frazier