BUG: Fix numpy.i support for numpy API < 1.7. #7759
Conversation
numpy API < 1.7 does not support NPY_ARRAY_F_CONTIGUOUS.or PyArray_IS_F_CONTIGUOUS. Supporting old APIs is useful as users tend to ship modern numpy.i files to compile against possibly older numpy installations.
charris
changed the title
| %#endif | ||
| %#define array_is_contiguous(a) (PyArray_ISCONTIGUOUS((PyArrayObject*)a)) | ||
| %#define array_is_native(a) (PyArray_ISNOTSWAPPED((PyArrayObject*)a)) | ||
| %#define array_is_fortran(a) (PyArray_IS_F_CONTIGUOUS((PyArrayObject*)a)) |
There was a problem hiding this comment.
PyArray_IS_F_CONTIGUOUS is available in 1.6, it is 1.5 that misses it.
There was a problem hiding this comment.
And nevermind the nevermind ;) PyArray_IS_F_CONTIGUOUS is in 1.6. Note that it is not quite the same as PyArray_ISFORTRAN.
There was a problem hiding this comment.
charris@fc [numpy.git ((v1.6.0))]$ grep -r PyArray_IS_F numpy/core/include/
numpy/core/include/numpy/ndarraytypes.h:#define PyArray_IS_F_CONTIGUOUS(m) PyArray_CHKFLAGS(m, NPY_F_CONTIGUOUS)
|
I'm a bit surprised that folks would still be using such ancient versions of Numpy, but perhaps that is driven by compatibility concerns with old code. |
|
Let's go ahead and put this in as is. I expect users of ancient numpy also expect ancient |
|
Thanks for including the patch! I still have an install target with numpy 1.3 to support, unfortunately. Hopefully they'll actually upgrade somewhere this year, but I'm not counting on it... |
numpy API < 1.7 does not support NPY_ARRAY_F_CONTIGUOUS.or PyArray_IS_F_CONTIGUOUS. Supporting old APIs is useful as users tend to ship modern numpy.i files to compile against possibly older numpy installations.
I tested this fix by searching for the introduction of both constants (numpy 1.7), and compiling against numpy 1.3 and 1.10. This ought to be more than sufficient for this simple change. Note that the code for <=1.7 is equal to what used to be there (actually until 1.9).