Bignum support still broken on powerpc (ping: Jerry James)

[robert delius royar]

Mon, 10 Apr 2006 (16:38 +0900 UTC) Stephen J. Turnbull wrote:

>>>>>> "robert" == robert delius royar <xemacs at frinabulax.org> writes:
>
>    robert> OK, I simply removed the 'if (__GMP_ULONG_MAX <=
>    robert> GMP_NUMB_MASK)' from line 1595 of gmp.h.  That stopped the
>    robert> warnings.  Now, my loadup/dump fails with the same type
>    robert> error
>
> What GMP are you using?  I'm pretty sure I just built (on my home
> machine, will check details tonight) with GMP on Mac OS X 10.4.6 with
> the default gcc (I assume 4.0x), with no such problems.  XEmacs
> 21.5.26 with one unrelated patch (already in CVS).  I'm using the GMP
> from DarwinPorts, updated April 5 or 6.

% port installed gmp
The following ports are currently installed:
   gmp @4.1.4_2 (active)

I don't think the error is in gmp itself but in some part of the loadup 
that is getting a value it does not expect as a result of bignum 
support.  I think the backtrace might lead to finding that place in the 
code.  On the G5 (iMac) the problem triggers when --enable-union-type 
is configured in.  On my older powerbook, removing that directive did 
not fix the problems with the earlier version of gmp. 
If I get time, I will try a compile on the G4 machine with the new gmp 
version installed to see if --enable-union-type is the problem. 
However, if it is, then does that mean that the real problem is an 
alignment error that gets triggered when an object is referenced that is 
not as it is expected?  If so, could that point to something lower 
level?

-- 
Dr. Robert Delius Royar                   Associate Professor of English
Morehead State University                             Morehead, Kentucky