Strange difference in FD between E*E and E**2

[Fred Bapst]

Hello,

With the following code, strangePower1(A,B) gives one solution, where the other 
versions give two:

strangePower1(A,B) :-
        fd_domain([A,B],0,5),
        (A-B)**2 #> 16,          % <-------
        fd_labeling([A,B]).

strangePower2(A,B) :-
        fd_domain([A,B],0,5),
        (A-B)*(A-B) #> 16,       % <-------
        fd_labeling([A,B]).

strangePower3(A,B) :-
        fd_domain([A,B],0,5),
        A**2 -2*A*B +B**2 #> 16, % <-------
        fd_labeling([A,B]).

| ?- strangePower2(A,B).
  A = 0  B = 5 ? a
  A = 5  B = 0
  (15 ms) no
| ?- strangePower1(A,B).
  A = 5 B = 0
  yes

Can you explain why? Do you think this difference is reasonable from the 
programmer's point of view?

Observed in gprolog-1.4.0, both in Win32 and Linux32.

Best regards.