New 'number' type and -number:type option

[Charles]

Cobra defaults to using 'decimal' instead of 'float64' or 'float32' because decimal can accurately represent the numbers that people often input such as 0.1. As an added bonus, decimal has more significant digits and a wider range of values. By "Cobra defaults to 1" I mean that:

class Sample
    def main is shared
        rate = 0.1
        # rate's type is decimal, not float
        # or consider:
        .util(4.5)  # woops. compile-time error. forgot to say 4.5f
    def util(x as float)
        pass

The benefits of decimal are covered in more detail under "Accurate Math" in the Comparison to Python.

But decimal is not perfect. There is a performance penalty (which I've never gotten around to measuring) and there is the fact that many libraries oriented towards graphics, mathematics, games, neural nets, etc. will be based on float64. This is made more painful by the fact that .NET generic classes cannot be parameterized by numeric type in any meaningful way.

I certainly want developers in those areas to feel welcome and productive with Cobra. Its compiled performance, slick syntax, unit tests and (optional) static typing make it an otherwise great choice for those areas.

So in development, I have added a new -number:TYPE command line argument. The choices for TYPE include the fractional numeric types in the language: float32, float64, float and decimal. ('float' is an alias for 'float64') The default is 'decimal' but you can override this now like so:

Code: Select all

cobra -number:float myprog.cobra

Now the literal "4.5" will be typed as 'float', not 'decimal'. No "f" suffix required.

Furthermore, you can use the keyword 'number' as a type now:

def sum(a as number, b as number) as number
    return a + b

If you use 'number' throughout your code and you are not passing your values to a library, then you can change the definition of number to see how that affects the accuracy and speed of your program. But the primary motivation for this feature is to make life easier for programmers that normally work with float.

If you have any questions, please ask. If you test this out of the repository and have any problems, let me know here in the forums.

-Chuck

[Charles]

By the way, you can always dictate the type of a fractional literal by using a suffix. This works independent of the -number: option.

d  = 1.0d    # decimal
f3 = 1.0f32  # float32
fa = 1.0f    # float / float64
fb = 1.0f64  # float / float64

[hopscc]

If you have multiple source files that require different types for 'number' (e.g 3 use default decimal and one needs float64 say) having to specify them on the commandline
a) forces you to use at least two command line invocations (which)
b) demands you use a build system or script to keep the typing and file info together.
It would be better if this default number' typing was also settable in the cobra source file ..... like in a compiler directive ...

Heres a patch that adds a compiler directive doing just that - sets the default type of 'number' to the specified type from the directive on.

Code: Select all

%% number float32
    def y is shared
        m as number = 48
        print '[m.getType] m=[m]'
        assert m.getType is Single

%% number decimal
    def x is shared
        m as number = 46
        print '[m.getType] m=[m]'
        assert m.getType is Decimal


[hopscc]

And unit test file