Closes ticket 993. Review: numpy#229 There's some concern about future maintainability (by David C.), but overall the benefits seem to outweigh the possible drawbacks.
…ously, see http://projects.scipy.org/numpy/ticket/993
Also make npy_PyFile_OpenFile mode argument const char*.
This may be more a workaround for a Python 3K inconsistency or bug, but it is probably better to use unicode for these sorts of calls. The fix itself is a bit of a hack, but so is the *_to_* function. It really needs a rewrite.
The python3 unicode strings don't have the __buffer__ interface.
Since type_num is not considered in PyArray_EquivTypes (for dtype equality comparison), seems reasonable to ignore it for hashing as well.
This fixes a problem on 64 bit Windows where a 64 bit index is needed but long is only 32 bits.
This is caused by the inconsistent floating point handling of Python itself. On Windows with 2.5: >>> "%s" % 1e-6 '1e-006' With 2.6: >>> "%s" % 1e-6 '1e-06' Reviewed as PR-225.
Add parameters in NoseTester constructor as well as in NoseTester.test() to control which warnings raise an error. An explicit difference is made between the master branch and release branches. This was mentioned on the ML: http://thread.gmane.org/gmane.comp.python.numeric.general/48210
Thanks to Matthew Brett for tracking this down on a 64-bit SPARC running Debian.
Don't do from numpy.testing import *, put spaces after commas, etc.
* add-isclose-function: ENH: Made "isclose()" NA-aware, added release note and versionadded. TST: Test that isclose().all() == allclose() for all test cases. TST: Added masked array and scalar tests for "isclose()". ENH: Allow "isclose()" to work with subclasses of ndarray (such as masked arrays). ENH: Added "isclose()".