Re: extreme behavior of nfsubfields()

[Ilya Zakharevich <ilya@Math.Berkeley.EDU>]

On Sun, Sep 29, 2002 at 03:27:34AM -0400, Igor Schein wrote:
> Hi,
> 
> I need to find a degree-27 subfield of a non-Galois degree-81
> polynomial.  When I run nfsubfields(), it returns "too many blocks".
> So I increased the value of oldllist inside choose_prime() from 10^6
> to 5*10^7, and ran gp with 2048MB stack on 32bit Linux.  In a few
> seconds, I saw the screen filled with "***   bug in  ***   bug in  ***
> bug in  *** bug in  ***   bug in  ***   bug in  ***".   If I reduce
> stack size to 1900MB, I'm getting a nicer "not enough memory".  When I
> tried the same on 32bit Solaris with 2048MB stack, the session just
> died with SEGV.  I'll see what happens with 64bit binary, but I'm
> skeptical.

An idea: at some moment PARI may compare two memory addresses (e.g.,
avma with bot).  If it compares them as long's (as opposed to
ulong's), and they are on different sides of 2G limit, a disaster will
follow.

Hope this helps,
Ilya