Re: bnfisnorm() bug

[Igor Schein <igor@txc.com>]

On Fri, Dec 17, 1999 at 01:17:14PM +0100, Bill Allombert wrote:
> Hello,
> 
> I test the bug on a Alpha/Osf1 here and I get 
> 
> parisize = 8000000, primelimit = 500000
> ? bnfisnorm(bnfinit(polsubcyclo(7,3)),2)
> %1 = [Mod(1, x^3 + x^2 - 2*x - 1), 2]               
> 
> moreover, the code is very clean from memory point of vue,
> so I don't guess where is the bug.
> 
> 
> So perhaps send us a stack trace.

Hi,

since Bordeaux mailer is having a problem, it appears, I'm replying to
the list. 

A little investigation made me believe that the bug is indeed in the
code, but Alpha native compiler fails to hit it, unlike gcc.  Here's a
way to reproduce the bug using native compiler:

Index: src/basemath/gen3.c
===================================================================
RCS file: /home/megrez/cvsroot/pari/src/basemath/gen3.c,v
retrieving revision 1.12
diff -r1.12 gen3.c
1663a1664
>       output(y);

Simply adding an output() statement inside ground() function
causes natively compiled binary to fail at the same point as
gcc-compiled, at least for me.

So please try it for yourself, and I'll do some further
investigation. 

Thanks

Igor