>  This smacks of weirdness in /usr/src/gnu/dist/gawk/builtin.c::format_tree(),
>  maybe in the "unsigned long" stuff in the handlers for "%d"/"%i"?  I tried
>  to run (g)awk under GDB, but attempts to print out variables at breakpoints
>  give me "Error accessing memory address 0x7NN: Invalid argument." errors.)

[Moving this to port-sparc64.  I thought both mailing lists were the same,
 obviously they aren't.  Sorry folks.  Mea culpa.  I've joined the port-sparc64
 list now, which I should've done as soon as I got the Ultra 5+.  Sorry!]

OK, I think this is the key problem.  It looks like pretty much exactly the
opposite of Andrei Petrov's problem with converting from u_int64_t to double:

(Reduced example from gawk's format_tree())

netbsd4me# uname -mr
1.5 sparc64

netbsd4me# cat unsignedlongcastbug.c

#include <stdio.h>

main()
{
        double tmpval = 1;
        int dval;
        long lval;
        unsigned long uval;

        dval = (int) tmpval;
        lval = (long) tmpval;
        uval = (unsigned long) tmpval;

        (void) printf("tmpval/dval/lval/uval == %e, %d, %ld, %lu\n",
                        tmpval, dval, lval, uval);
        (void) exit(0);
}
netbsd4me# cc -o unsignedlongcastbug unsignedlongcastbug.c

netbsd4me# unsignedlongcastbug
tmpval/dval/lval/uval == 1.000000e+00, 1, 1, 13830554455654793216

[Later on]
I went to the port-sparc64 mailing list archives and noticed that Matt said:

"there are known issues with 'float and 'double' support on sparc64.  you
 should compile with -msoft-quad-float."

Well now, lookie here:

netbsd4me# cc -msoft-quad-float -o unsignedfloatbug unsignedfloatbug.c
netbsd4me# ./unsignedfloatbug
tmpval/dval/lval/uval == 1.000000e+00, 1, 1, 1

I just looked at the -current "gawk" Makefile and see that this was added:

*** 19 ****
--- 20,23 ----
+ 
+ .if (${MACHINE_ARCH} == "sparc64")
+ CFLAGS+=      -msoft-quad-float
+ .endif

Looks like you damn guys are ahead of me already  :-)

(Wonder if there should be a code check in the tree to look for double/quad
 references, and flag 'em so that they'll all be built with "-msoft-quad-float"
 when being built under NetBSD/SPARC64?)

Whee!  Aren't 64-bit issues *fun*?  :-)

	- Greg