| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |\
| |
| | |
BUG: maskna: PEP3118 code wasn't raising an error on NA-masked arrays
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Numpy uses its 'base' attribute in two different ways, both
to manage the memory of an external data buffer, and to track
an array for 'UPDATEIFCOPY' purposes. The nditer was creating
a temporary view with negative strides to match the strides of
input arrays, but then also wanted to use UPDATEIFCOPY. Unfortunately
this choice to overload 'base' for both purposes prevents that
from working, and the result was a memory leak.
This bug has been fixed by choosing instead to retain positive
strides and not fully match the order of input memory when
doing the UPDATEIFCOPY.
|
| | | |
|
| |\ \
| |/
|/| |
BUG: Fix format/variable incompatibility.
|
| | |
| |
| |
| |
| | |
This shows up in python 2.4 where Py_ssize_t is an int.
Also remove unused variable.
|
| |/
|
|
|
|
|
| |
The --cover-inclusive argument means that coverage.py tries to include every
single .py file in the source tree in the coverage report. This leads to test
errors, because it tries to import files like setupscons.py (which will of
course directly fail for anyone not having numscons installed).
|
| |
|
|
|
| |
In Python 3 the pickle.loads function expects a bytes argument, so
convert the test strings.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
| |
The purpose of this is to allow dtypes with metadata to
have faster C-level performance. With this, datetime64
metadata lookup can become a single C pointer dereference,
compared to a C-string to Python-string conversion +
a Python dict lookup as it is presently.
|
| | |
|
| |
|
|
|
|
| |
with datetime64
This bug was found by running the pandas master against numpy master.
|
| |
|
|
|
| |
Fix incorrect python version checks in test_print.py.
Fix missing build_err_msg import and wrong variable in test_io.py.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
| |
Two templates were broken up and specialized. The corresponding
substitution loops could also be specialized, but this wasn't
done this time through. Breaking up the templates clarifies the
code at the expense of extra text, but I think the improved
clarity is a good tradeoff.
|
| |
|
|
|
| |
Also break templated @from@_to_@to@ functions into two templates
to simplify the code a bit.
|
| |
|
|
| |
MyPyUnicode_Resize and MyPyUnicode_New are no longer used, so remove them.
|
| |
|
|
|
|
|
|
| |
PyArray_Scalar needed to use PyObject_* memory management since the numpy
unicode object inherits tp_alloc from PyUnicode_Type.
Observation: realloc returns null on error and that isn't always checked in
Numpy.
|
| |
|
|
|
| |
Use PyUnicode_FromUCS4 in UNICODE_getitime so that the same memory
management is used for all parts of the returned PyUnicodeObject.
|
| | |
|
| |
|
|
|
| |
This removes the include dependency on PY_UNICODE_WIDE in preparation
for defining more general unicode functions in the ucsnarray.c file.
|
| |
|
|
| |
This fix is kept separate to facilitate cherry picking for a backport.
|
| |\
| |
| |
| |
| | |
Fix filter importwarnings
This PR fixes the issue left over from merging PR-251.
|
| | | |
|
| | | |
|
| |/ |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |\
| |
| | |
TST: filter ImportWarnings in NoseTester.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Warnings show up when a directory with the same name as a Python file or
compiled extension is seen which doesn't have an __init__.py file in it. This
situation is very common, for example in SciPy where many extensions are
created from source files located under a directory with the same name.
This filter is located within a context manager, so only filters when running
tests.
|
| |\ \
| | |
| | | |
Fix some test warnings
|
| | | | |
|
| | |/ |
|
| |/
|
|
|
|
|
|
|
|
|
| |
This reverts commit 0a3a954c3750419552c076aa444563ac20def14a. The fix
caused segfaults for python built using the default ucs2 unicode size. It
looks like the cause is memory corruption in the PyObject pooled memory. In
particular, the routines in ucsnarrow.c lack the checks and fallbacks
present in the original Python source code in Objects/unicodeobject.c. I
think the best fix is to rewrite the relevant bits of numpy code so that
they don't use the MyPyUnicode_* routines at all, but rather the routines
present in Python.
|
| | |
|
| |
|
|
|
|
| |
I don't know that the crash is python2.7 specific, it may also occur
in earlier versions of python. The cause was mismatched memory
allocation/deallocation of maskna data.
|
| |
|
|
|
|
|
| |
The usefullness of this test is questionable, as it only shows up
as a core dump when run under python-debug for python versions >= 2.7,
and then it will only fail if run before other failing tests. It is
included as it may be useful as a record.
|
| | |
|
| |
|
|
|
|
|
|
| |
In Python >= 2.7 the memory management of unicode objects changed from
PyMem_* to PyObject_*. Numpy had its own modified copy of some of the
Python code for handling unicode objects and it needed to be updated
to deal with the change. Thanks to Joseph Miessner for tracking this down
and finding the fix.
|
| |
|
|
| |
Equal and nearly-equal size requirement is not true when passing a 1-D array of indices.
|
| | |
|
| | |
|
| | |
|
