[Gambas-user] Confusion in Data Types

[Doriano Blengino]

nando ha scritto:
> Float and Double is not as simple an explaination as the Int and others.
> ...
>
> Floating point numbers have two parts.
> Simplistically: One part is an integer which represents the actual digits in binary.
> The other is the value of the exponent (binary point and not decimal point).
>
>   
>>    ' Single (4-Byte C Float) 1.2e-38 to 3.4e38 
>>    mySingle = 3.4e+38
>>    PRINT "Max value in Single " & mySingle
>>    mySingle += 1 '1e+36    ' Issues
>>    PRINT "Min value in Single " & mySingle
>>
>>    ' Float (4-Byte C Double) 2.2e-308 to 1.8e308
>>    'myFloat = 1.8e+308  BUG: Cant hold this value
>>    myFloat = 1.79e+308
>>    PRINT "Max value in Float " & myFloat
>>    myFloat += 1 ' BUG: DoesNOT Wrap
>>    PRINT "Min value in Float " & myFloat
>>
>> END
>>
>> ----------------------------
>>
>> Program Output:
>>
>> ----------------------------
>>
>> RANGES
>> Max value in Single 3.399999952144E+38
>> Min value in Single 3.399999952144E+38
>> Max value in Float 1.79E+308
>> Min value in Float 1E+2147483647
>>
>> ----------------------------
>>     
Nando you are right but, anyway, there is something strange here. I 
think that the "single" behaviour is correct - if you add 1 to a very 
large number, this small added quantity should be so small to be ignored 
(rounded away), and this is what the "single" part does.

But the "float" behaviour scares me... floating point numbers should not 
overflow via an addition, I think. You can expect overflows and NaN's 
when more complex operations are involved, such as dividing by small 
numbers, or calculating tangents or other trascendentals (is this right 
english?). And, "1E+2147483647" is a strange number...

At this point, just for curiosity, I tried this program in freepascal:

    program test;

    var
      valsingle     : single;
      valdouble     : double;

    begin
      valsingle := 3.4e+38;
      writeln('MAX single: ', valsingle);
      valsingle := valsingle+1;
      writeln('MIN single: ', valsingle);

      valdouble := 1.79e+308;
      writeln('MAX double: ', valdouble);
      valdouble := valdouble+1;
      writeln('MIN double: ', valdouble);
    end.


Assigning 1.8e+308 to valdouble does not and give error, but prints out 
as "+Inf", so I resorted to use 1.79e+308; the result is:

    MAX single:  3.399999952E+38
    MIN single:  3.399999952E+38
    MAX double:  1.79000000000000E+308
    MIN double:  1.79000000000000E+308


Which confirms that there is something strange in KhurramM experiment. 
By the way, KhurramM: a double takes 8 bytes, not four.

Regards,

-- 
Doriano Blengino

"Listen twice before you speak.
This is why we have two ears, but only one mouth."