[Cython] template types

[Robert Bradshaw]

On May 6, 2009, at 12:47 AM, Dag Sverre Seljebotn wrote:

> Stefan Behnel wrote:
>> I wouldn't care so much about that for now. Whatever declaration we
>> choose, it's easily added when the feature itself is there, and I  
>> think
>> parametrised types are such an advanced feature that users can  
>> well wait
>> for it to become available in pure Python mode.
>>
>> The priorities should be to make the feature simple and well  
>> readable in
>> Cython code, and then 'somehow' available in Python code.
>
> I disagree with this principle. (I guess it might be a matter of taste
> in the end, so I'll just state my opinion and try not to drag this on
> for too long.)
>
> In perhaps about a year, with Python 3 argument decorators (def foo(x:
> int) -> float) and a bit of type inference, "pure Python mode" could
> actually become as convenient as Cython syntax and the preferred  
> syntax
> we'd teach new users in my opinion.

I think "cdef int i,j,k" is in always going to be easier than

i = cython.declare(int)
j = cython.declare(int)
k = cython.declare(int)

but in general I agree with this philosophy.

> And anyway, I'm in favor of avoiding to have to invent two kinds of
> syntax for new features. Especially when the benefits are so small!  
> (One
> uses a template a lot more than writing them.)

I'm guessing you mean you're *not* in favor?

> It's just a mess to have to tell users "this one is the Cython syntax
> but you can also use this one for Python compatability" etc.; that
> problem is there to stay but I don't see the sense of keeping  
> adding to it.

Only if there's a non-pure Python syntax that's a lot cleaner (which  
I think the class Foo[T] is).

- Robert