Re: var_range changes behaviour if eplex loaded

"Jonathan M. Lever" wrote:
> 
> Hi,
> 
> You may well know already, but var_range reports zero differently according to whether eplex is loaded.
> 
> E.g. 1       lib(range).
> 2       lib(ria).
> 3       ria : (X :: 0 .. 10), var_range(X, L, H).
> 4       lib(eplex).
> 5       ria : (X :: 0 .. 10), var_range(X, L, H).
> 
> 3 gives L = 0.0 whereas 5 gives L = -0.0
> 
> Doing something like ria:(L >= 0) is a workaround.
> 
> Ran into this this week, as a test VarMin == 0.0 started failing.
> Think it's the same in 5.4, 5.5 and prerelease.

Probably not in 5.4.

Since 5.5, "syntactic" comparisons (=/2, ==/2 etc) consider
negative and positive zeros as different. This was a deliberate
change to clean up the semantics.

Since arithmetic computations will sometimes result in negative zeros,
(here it is probably something like ceiling(-1.0e-6) giving -0.0)
you should not use code that relies on the sign.

Change the test to VarMin =:= 0.0 (the arithmetic comparisons treat
the zeros as equal).  Oh, and be aware that you are violating the
old rule "never test floating point numbers for equality" :-)

Cheers,
Joachim