Re: Numerics

Dominique Colnet a écrit :

> josé bollo wrote:
>
>> here is a common expression (where n is 9 and m is INTEGER):
>>
>>   m := 10000*2^n
>>
>> so i apply SE rules: {INTEGER_16}*{INTEGER_8}^{INTEGER_8}
>>
>> and nothing does fit! : 2^9 is an INTEGER_16 and the 5120000 is 
>> obviously not an INTEGER_16.
>>
>> And me the programmer have to be aware of such a thing even when i 
>> know that the result is an INTEGER so INTEGER_32 that fit.
>
> You are right here. The compiler should infer that 10000*2^9 = 5120000
> and thus fit on INTEGER_32.
> Exactly as the compiler is currently able to see that 127+1 = 128
> So you are right and I consider it as a bug in the compiler.
> I have registered your example in our private test suite (no need for 
> you to fill a SmartZilla record).
> Thanks
> And as usual best regards to all,

good

but have tried to write the expression?

let try it again

    n: INTEGER

    check n.in_range(2,17) end
    m := 10_000 * 2^n

here ^^^^^ it is pretty and safe

    check n.fit_integer_8 end
    m := {INTEGER_32 10_000} * 2^(n.to_integer_8).to_integer_32

here ^^^ it is ugly and not safe, n>18 will give a deasaster

    m := {INTEGER_32 10_000} * ({INTEGER_32 2}^n).to_integer_32

here ^^^ it is ugly and not safe but maybe best

    m := 10_000 * ({INTEGER_32 2}^n).to_integer_32

trying the automatic conversion

You see, here except by using a log in preconditions of ^, the check 
n.in_range is the best solution for safety

cheers
josé