== The Kind field in DTypes ==

The `PyArray_Descr` structure (defined in [source:/trunk/numpy/core/include/numpy/ndarrayobject.h#1172 ndarrayobject.h]) has a character field "kind", which roughly corresponds to the base type of the scalar. This field is exported as the "kind" field of a dtype. Valid values are described in [source:/trunk/numpy/core/numerictypes.py numerictypes.py], and are

|| '''Character''' || '''Basic numeric type''' || '''Base dtype''' ||
|| b || Boolean || `bool_` ||
|| i || Signed integer || `number` ||
|| u || Unsigned integer || `unsignedinteger` ||
|| f || Floating point || `floating` ||
|| c || Complex number || `complexfloating` ||
|| S || String || `str_` ||
|| U || Unicode string || `unicode_` ||
|| V || Void type || `void` ||
|| O || Python object || `object_` ||

The kind is used when determining a common dtype for two dtypes. It also can used along with the byte-width to uniquely identify the dtype (`i8`, for instance, is a 64-bit signed integer). It is also used in Numpy and Scipy in a handful of places for checking integerness (or similiar), although those places could be replaced with `issubsctype`.

Kind should not be confused with the typecode (another holdover from Numeric).

(cookedm: IMO, we don't need kind, and it should be deprecated.)