Robert Bradshaw wrote:
> On Oct 8, 2009, at 6:04 AM, Dag Sverre Seljebotn wrote:
> 
>> I experience this:
>>
>> <Py_ssize_t>np.sqrt(some_float64)
>>  => TypeError: 'numpy.float64' object cannot be interpreted as an  
>> index
>>
>> However
>>  a) <int>np.sqrt(...) works fine, as does <long> and <long long>.
>>  b) In Python, int(np.sqrt(...)) and long(np.sqrt(...)) both works  
>> fine
>>
>> Is Py_ssize_t "special" when it comes to converting from Python  
>> objects?
>> If so I find it mildly confusing and something which only increases
>> learning curve...
> 
> Py_ssize_t is meant for indexing, and fails for the same reason that
> 
>  >>> L = range(10)
>  >>> L[2]
> 2
>  >>> L[2.5]
> Traceback (most recent call last):
>    File "<stdin>", line 1, in <module>
> TypeError: list indices must be integers, not float
> 
> fails. This is a feature, not a bug.

OK, if this is a feature, help me out with this then:

I need to iterate from 0 to "lmax". When I allocate an array, it is 
always allocated of size (lmax + 1)**2, and it is convenient to be able 
to deduce lmax from the array shape.

What type should I use for lmax to conveniently convert from a float 
(because sqrt is float, but I know I can safely truncate) and guarantee 
that I support the range of NumPy arrays (which essentially use 
Py_ssize_t to store shape info)? The type must be signed, otherwise I 
fall into the range(-lmax, lmax) trap all the time.

(For the record, I think the type system is complicated enough without 
this feature -- does it have usecases where an explicit call to e.g. a 
cython.asindex() function to invoke __index__ isn't clearer?)

-- 
Dag Sverre
_______________________________________________
Cython-dev mailing list
[email protected]
http://codespeak.net/mailman/listinfo/cython-dev

Reply via email to