After pushing the changes for the 'size_t' stuff, now I believe that
things were a bit broken before the patch, and they still are.
+ Go Cython.Compiler.PyrexTypes
+ Look at the module-level tuple variable 'rank_to_type_name'. You can
see that 'Py_ssize_t' and 'size_t' are AFTER the type 'PY_LONG_LONG'.
+ Now look at the implementation of the function 'widest_numeric_type()'.
Then, this bit of Cython::
cdef f():
cdef unsigned long long a = 5
cdef size_t b = 7
return a + b
generates C code like this::
/* "/u/dalcinl/Devel/cython-devel/tmp.pyx":5
* cdef unsigned long long a = 5
* cdef size_t b = 7
* return a + b # <<<<<<<<<<<<<<
*/
/* __pyx_t_1 allocated */
__pyx_t_1 = __pyx_PyInt_FromSize_t((__pyx_v_a + __pyx_v_b)); if
(unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 5;
__pyx_clineno = __LINE__; goto __pyx_L1_error;}
This is IMHO totally wrong !!
Could we assume that PY_LONG_LONG is the widest integral?? Should we
assume that 'Py_ssize_t' and 'size_t' are between 'long' and
'PY_LONG_LONG'? In such case, I'll have to fix the 'rank_to_type_name'
tuple, and perhaps the 'sign_and_rank_to_type' dictionary (used as a
fallback in function 'widest_numeric_type()')
However, I'm not sure how to we will handle code like the one below
with the current implementation,
cdef f():
cdef Py_ssize_t a = 5
cdef size_t b = 7
return a + b
I believe the value of 'a+b' should be promoted to 'signed long long',
but it cannot see the way to implent such behavior in the current
Cython codebase.
--
Lisandro Dalcín
---------------
Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
PTLC - Güemes 3450, (3000) Santa Fe, Argentina
Tel/Fax: +54-(0)342-451.1594
_______________________________________________
Cython-dev mailing list
[email protected]
http://codespeak.net/mailman/listinfo/cython-dev
