Re: Segmentation fault on OSX for intel

[Enrico Sersale]

When I've read "Program received signal EXC_BAD_ACCESS, Could not access 
memory." in the first mail, I've thougth that it was the same error I get. But the 
problem is a bit different.
On 2005-12-21 the following lines have been changed in 
make/Instance/bundle.make:

# On Apple, two-level namespaces require all symbols in bundles
# to be resolved at link time.
ifeq ($(FOUNDATION_LIB), apple)
   LINK_BUNDLE_AGAINST_ALL_LIBS = yes
 endif

has become:

# On Darwin, two-level namespaces require all symbols in bundles
# to be resolved at link time. Do this even if we don't think we need it.
ifeq ($(findstring darwin, $(GNUSTEP_TARGET_OS)), darwin)
   LINK_BUNDLE_AGAINST_ALL_LIBS = yes
 endif

After this change I get things like this:

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_PROTECTION_FAILURE at address: 0x00000007
0x000288f9 in objc_hash_string () at NSBezierPath.m:2120
2120          bleft[1].y = (coeff[0].y + coeff[1].y) / 2;
(gdb) bt
#0  0x000288f9 in objc_hash_string () at NSBezierPath.m:2120
#1  0x00250000 in _IcePaAuthDataEntries ()
#2  0xbfffe978 in ?? ()
#3  0x0002551a in objc_hash_value_for_key () at NSBezierPath.m:1434
#4  0x031271c0 in ?? ()
#5  0x00000007 in _mh_dylib_header ()
#6  0x0317eb80 in ?? ()
#7  0x0002908b in __sel_register_typed_name () at NSBitmapImageRep.m:202
#8  0x0317e1fc in ?? ()
#9  0x00000024 in _mh_dylib_header ()
#10 0xbfffe9b8 in ?? ()
#11 0x000290ab in __sel_register_typed_name () at NSBitmapImageRep.m:204
#12 0x031271c0 in ?? ()
#13 0x00000007 in _mh_dylib_header ()
#14 0x001100b6 in NSIsEmptyRect (aRect={origin = {x = 4.23249713e-37, y = 
4.30361224e-37}, size = {width = 0, height = 2.40030908e-38}}) at 
/usr/GNUstep/System/Library/Headers/Foundation/NSGeometry.h:289
#15 0x0002b6cb in __objc_thread_id () at NSBitmapImageRep.m:917
#16 0x03100640 in ?? ()
#17 0x031271c0 in ?? ()
#18 0x00000000 in _mh_dylib_header ()
#19 0x0102af6d in -[NSBezierPath elementCount] (self=0x31271c0, _cmd=0x7) at 
NSBezierPath.m:795
Previous frame inner to this frame (corrupt stack?)

reverting to the old bundle.make fixes the problem.

On 2006-01-04 22:25:08 +0200 David Ayers <d.ayers@inode.at> wrote:

> joekewoud_@joekewoud.nl schrieb:
>
> > When I start *any* gnustep program. Including most from the Tools directory
> > I get a segmentation fault
> [snip]
> > I use GCC4.02 FSF
> [snip]
> > Program received signal EXC_BAD_ACCESS, Could not access memory.
> > Reason: KERN_INVALID_ADDRESS at address: 0x746f7270
> > 0x90a418b0 in objc_msgSend ()
> > (gdb) bt
> > #0  0x90a418b0 in objc_msgSend ()
> > (gdb)
>
> Hello Ewoud,
>
> This is an indication that your binary is being linked against Apple's
> libobjc as this function is called objc_msg_send on FSF libobjc.
>
> > This is the crashlog:
> [snip]
> > Thread 0 Crashed:
> > 0   libobjc.A.dylib     0x90a418b0 objc_msgSend + 16
>
> Note the name.
>
> > 1   dyld                0x8fe0e2d6
> > ImageLoaderMachO::doNotification(dyld_image_mode, unsigned, dyld_image_info
> > const*) + 70
> > 2   dyld                0x8fe03409
> > dyld::notifyAdding(std::vector<ImageLoader*, std::allocator<ImageLoader*>
> >
> > > &) + 201
> >
> > 3   dyld                0x8fe0c4bd
> > ImageLoader::runInitializers(ImageLoader::LinkContext const&) + 75
> > 4   dyld                0x8fe02cfa dyld::initializeMainExecutable() + 112
> > 5   Affiche             0x0000214d _call_mod_init_funcs + 36 (crt.c:305)
> > 6   Affiche             0x00002079 _start + 79 (crt.c:240)
> > 7   Affiche             0x00002029 start + 41
> >
> > Thread 0 crashed with i386 Thread State:
> > eax: 0x746f7250    ebx: 0x90a404e8 ecx:0x90a4f33c edx: 0x00095470
> > edi: 0x00000000    esi: 0x00095470 ebp:0xbfffee58 esp: 0xbfffec98
> >   ss: 0x0000002f    efl: 0x00010282 eip:0x90a418b0  cs: 0x00000027
> >   ds: 0x0000002f     es: 0x0000002f  fs:0x00000000  gs: 0x00000037
> >
> > Binary Images Description:
> >      0x1000 -    0x17fff Affiche
> > /usr/GNUstep/Local/Applications/Affiche.app/Affiche
> >     0x81000 -    0x92fff libobjc-gnu.1.dylib
> > /opt/local/lib/gcc40/libobjc-gnu.1.dylib
>
> This is the libobjc you actually want.
>
>
> > 0x907f0000 - 0x908b9fff com.apple.CoreFoundation 6.4.4 (368.22)
> > /System/Library/Frameworks/CoreFoundation.framework/Versions/A/CoreFoundatio
>
> Not sure where CoreFoundation comes from...
>
> > 0x908fa000 - 0x909eefff libicucore.A.dylib      /usr/lib/libicucore.A.dylib
> > 0x90a3f000 - 0x90abffff libobjc.A.dylib         /usr/lib/libobjc.A.dylib
>
> And here is Apples libobjc
>
> > 0x90ae8000 - 0x90b4cfff libstdc++.6.dylib       /usr/lib/libstdc++.6.dylib
> > 0x90bbb000 - 0x90bc2fff libgcc_s.1.dylib        /usr/lib/libgcc_s.1.dylib
>
> and Apples gcc.
>
> So it seems you have linked code previously linked with something that
> pulls in CoreFoundation which is possibly responsible for also linking
> the NeXT runtime, so your code is currently linked with both runtimes
> which is causing the confusion.
>
> I'm no expert on Darwin but I think with otool you can figure out the
> dependencies of dylibs to determine what is pulling in the Apple code.
>
> Good luck!
> David
>
>
> _______________________________________________
> Discuss-gnustep mailing list
> Discuss-gnustep@gnu.org
> http://lists.gnu.org/mailman/listinfo/discuss-gnustep