Re: please check this

Hi,
I discovered with git-bisect that the regression has been probably intoduced here:
https://pari.math.u-bordeaux.fr/cgi-bin/sgitweb.cgi?p=pari.git;a=commitdiff;h=a4113027d04815ddc04d111db7a6620f4a8ab7aa

a4113027d04815ddc04d111db7a6620f4a8ab7aa gives (2*a^2 - 2*b^2)/(4*b*a)
183213fc2fc9e2798b1023790ed455ee17eb3c07 gives (a^2 - b^2)/(2*b*a)

I let people more familiar with the source code to discuss the potential fix.

Best regards,
R. Suwalski

On Sunday, March 30th, 2025 at 00:01, American Citizen <[email protected]> wrote:

> owner@localhost:~/math/cuboids> gp
> 
> Reading GPRC: /home/owner/.gprc
> GPRC Done.
> 
> GP/PARI CALCULATOR Version
> 2.18.1 (development git-1fa1ee045f)
> amd64 running linux
> (x86-64/GMP-6.3.0 kernel) 64-bit version
> compiled: Mar 29 2025, gcc
> version 13.3.0 (SUSE Linux)
> threading engine: single
> (readline v7.0
> enabled, extended help enabled)
> 
> Copyright (C)
> 2000-2025 The PARI Group
> 
> PARI/GP is free software, covered by the GNU General Public License, and
> comes WITHOUT ANY WARRANTY WHATSOEVER.
> 
> Type ? for help, \q to quit.
> Type ?18 for how to get moral (and possibly technical) support.
> 
> parisizemax = 10000003072, primelimit = 1048576, factorlimit = 1048576
> ? chkr1=(a^2-b^2)/(2ab)
> %1 = (2a^2 - 2b^2)/(4ba)
> ?
> What??? where did that extra factor of 2 come from?
> 
> While %1 is correct, in extracting the numerator and denominator, things
> quickly fall apart.
>