…_getlimits. This was causing errors when running the test suite.
(From Ralf Gommers.)
The problem was that if endian.h was not detected, @DEFINE_NPY_HAVE_ENDIAN_H@ was still present in _numpyconfig.h, causing compilation to fail.
…nks to Ryan May.
…ins (#1489) Only Sphinx >= 1.0 is supported now.
Note that the increase in disk image size from 10Mb to 11Mb does not mean numpy is actually that large - more than half that is the included docs. It is the case though that the 2.7 binaries are slightly larger than the 2.5/2.6 ones, by 100kb on Windows and 800kb on OS X.
(cherry picked from commit b4d7338de5338daed4e543921cf3cb82eb2c3675)
… equality with specified tolerances
…types (#1563) The problem was a code path trying to assign x, which returns a scalar -- for actual ndarrays a internal function always returning arrays was used.
…ssible. Closes #1503.
- allow numbers in function signature - print the items in relevance order (not reverse relevance) - include ufunc docstrings
(cherry picked from commit r8511)
…#1429) This makes `argmax` and `argmix` treat NaN as a maximal element. Effectively, this causes propagation of NaNs, which is consistent with the current behavior of amax & amin. (cherry picked from commit r8509)
…element is nan This also fixes NaN propagation in np.maximum/minimum, for example in the case for maximum(1, complex(0, nan)) -> (0, nan) which previously yielded (1, 0). (cherry picked from commit r8508)
… what they are on import (#1130) (cherry picked from commit r8507)