Re: Crash when launched -nw on Mac OS X, compiled with gcc 4.0.1

[Peter Dyballa]

Am 26.07.2006 um 17:40 schrieb Richard Stallman:

> 	    Run till exit from #0  extract_float (num=4216608) at  
> floatfns.c:220
>             0x001f27dc in Fsqrt (arg=4216608) at floatfns.c:556
>             Value returned is $7 = 527076
>
> That value seems correct.  It has converted the integer to a double.
>
> I suggest you step thru the failing line
> instruction by instruction, with the `si' command,
> and see which instruction fails.

The short version of the session:

        Breakpoint 1, Fsqrt (arg=4216608) at floatfns.c:556
        extract_float (num=2) at floatfns.c:219
        0x001f9894 in extract_float (num=2) at floatfns.c:219
        extract_float (num=4216608) at floatfns.c:220
        0x001f993c in extract_float (num=4216608) at floatfns.c:225
        0x001fb248 in Fsqrt (arg=4216608) at floatfns.c:556
        (gdb) print d
        $1 = -1.9828643798828138
        (gdb) si
        ...
        (gdb) si
        (gdb) print d
        $2 = 527076
        (gdb) si
        ...
        (gdb) si
        0x0028c000 in dyld_stub___error ()
        (gdb) fin
        Run till exit from #0  0x0028c004 in dyld_stub___error ()
        0x001fb284 in Fsqrt (arg=4216608) at floatfns.c:561

Can this dyld_stub___error be a call to a library that does not  
exist? Apple documents that it does not provide a library that  
emulates floating point arithmetics ...

> Then you can determine whether it was correct compilation output
> for the GCC options that you specified.


Last night and this evening I made a large number of tests à la 'make  
distclean ; ./configure ... ; make' with different compiler options.  
It seems that the Metrowerks CodeWarrior and MrC compiler and old  
Apple gcc supporting option -faltivec changes some things, including  
setting the -mpim-altivec switch. If I do not use that switch ld  
complains with some files that incompatible options are given (/sw/ 
lib/odcctools/bin/ld: warning /var/tmp//ccMZzWzD.o cpusubtype (100,  
architecture ppc970) does not combine with cpusubtype (11) for  
specified -arch flag: ppc7450 and -force_cpusubtype_ALL not specified  
(file not loaded)). I presume it's among these tripletts '-mcpu=7450 - 
mpowerpc -mpowerpc-gpopt' or '-mcpu=G4 -mpowerpc -mpowerpc-gpopt' --  
in the end temacs is not executable.

Since -mpim-altivec works with another set of switches fine, it seems  
to be that I started to use incompatible options instead of relying  
that a simple -mcpu=7450 would activate all working optimisations at  
the specified level. And version of gcc. The warning I take from this  
is not to use -faltivec!

Do you want me to do some more debugging?

--
Greetings

  Pete

Windows, c'est un peu comme le beaujolais nouveau: à chaque nouvelle  
cuvée on sait que ce sera dégueulasse, mais on en prend quand même,  
par masochisme.