Hello,
I'd like to point out a couple of questionable situations in GnuProlog when
dealing with FD variables with large domains (eg default domain).
- A*B #= C. no
- fd_domain([A,B,C],0,1000000),
A*B #= C. no
- fd_domain([A,B,C],0,10000000),
A*B #= C. yes
- X#<1000000, 4321 rem X #= 1. no
- X#<10000000, 4321 rem X #= 1. yes
- X*X #= Y. yes
- X*X*X #= Y. no
- X**3 #= Y. yes
Some of these issues might be due to integer overflow. My "numerical problem
sniffer" tool suggested for instance to double-check this code snippet from
BipsFD/fd_math_fd.fd:
pl_xy_eq_z(fdv X, fdv Y, fdv Z) {
...
start Z in min(X)*min(Y)..max(X)*max(Y)
}
Depending how it gets translated, the integer multiplication might overflow.
Am I off-topic? If not, do you think it is desirable to provide a "mode" where
it is ensured that propagation rules avoids overflows that could crash a domain?
Could it be proposed as a student project (for how many hours of work)?
Best regards.
_______________________________________________
Users-prolog mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/users-prolog
