Numpy.distutils was imported together with numpy due to the change in commit 025c8c7; this is unwanted.
… (Ticket #1895/1896) This interacted with the NPY_ITER_GROWINNER flag which would incorrectly grow the length because it thought nothing was being buffered.
…d_library. This is related to PEP 3149, tags in shared library extensions. Only applies to Linux (for now). See also #1749.
This bug was in the double reduction loop optimization used during buffering, and was triggered by the bug report example because it reduced everything to a single value.
…or of --include-paths. Updated docs.
This general aspect of object arrays is fraught with errors, for example, eval(repr(x)) produces a 2 instead of 1 dimensional array when given the x in the first example of the bug. This fix at least makes the result requested in the bug report work.
Also: 1. Note that the polynomial package is preferred over poly1d. 2. Remove deprecation of mapparms in __init__.py as it interferes with the documention of the method of the same name. This is probably safe as it is unlikely to be used yet. 3. Make some improvements to the documentation in polytemplate.
… NULL (fixes #1863 This appears to have been a longstanding bug, but has come to the surface because PyArray_NewLikeArray uses the function with a non-NULL strides parameter and is used more frequently then other such uses.
…be verbose on unspecified use modules. (backport of 026bc92)
…han the buffer size (ticket #1837) The symptoms of this bug were showing up only for a greater number of operands because einsum runs specialized loops for fewer operands and dimensions.
The call to the stransfer function had incorrect arguments. This didn't show up because all the tests of einsum, the only current user of the function, produced contiguous arrays which are special cased.
The ticket's bug is visible in einsum, but is caused by an error in how the reduction double loop interacts with buffering in the nditer. This test case directly tests the nditer triggering the bad code.
… bug was noticable with ifort but not with gfortran.
…ted size-to-shape mapping patch and implemented two argument size function in C.
On my system (Python 2.7, RHEL5), the call to func.__name__ returns a temporary Python string object with only a single reference. Dereferencing it before copying out its contents results in reading freed memory.