…py#12748)

* BUG: Fix SystemError when pickling datetime64 array with pickle5

Fixes numpygh-12745

* Fix reference and error handling

* Add a test for the fixed reference leak

* Fix for PyPy + simplify #if condition

* Fix comment

…umpy#12722)

* BUG: Fix rounding of denormals in double and float to half casts

Previously the significand was shifted right to align denormals of
different magnitude. This loses some bits that can make a difference
for rounding. This is fixed:
 1. For floats, by inspecting the original last bits when this may
    make a difference (should happen rarely)
 2. For doubles by shifting the bits left to align the denromals and
    thus not lose the lowest orginal bits.

* TST: Test half denormal rounding

The test assumes that half to even is active, if this is ever
changed or allowed to change, the test will fail.

* Fixup: Fixup for halffloat.c

The one code path cannot be used. The other must have been a typo, but
is a valid bug, a new test for it in the next commit.

* TST: half casting lower bits are not lost for denormal results

The first test only tested the off by one, this one specifically tests
that all bits are used to decide if "round to nearest even" should
be used, in the example of rounding towards 0.

* TST: Test not just denormals but all positive finite float16s

* DOC: Add lots of comments and add a short release note.

BUG: fix to check before apply `shlex.split`

BUG: Fix SystemError when pickling datetime64 array with pickle5

BUG: Fix rounding of denormals in double and float to half casts.

TEST: pin mingw version

BUG: ndarrays pickled by 1.16 cannot be loaded by 1.15.4 and lower

BUG: do not Py_DECREF NULL pointer

ENH: add _dtype_ctype to namespace for freeze analysis

BUG: fail if old multiarray module detected

BUG: Do not double-quote arguments passed on to the linker

…s-again

BUG: Do not insert extra double quote into preprocessor macros

- Create 1.16.1-notes.rst
- Create 1.16.1-changelog.rst
- Update mailmap

[ci skip]

DOC: Prepare for 1.16.1 release.

TST: fix vmImage dispatch in Azure

MAINT: remove complicated test of multiarray import failure mode

Previously when doing floor division numpy would sometimes return an incorrect
signed zero. For example:

>>> np.zeros(10)//1
array([-0., -0., -0., -0., -0., -0., -0., -0., -0., -0.])

>>> np.remainder(-1.0,1.0)
-0.0

The reason for this is that whenever div or mod were zero the code was using
the following to pick the sign of zero:

floordiv = (a / b > 0) ? 0.0@c@ : -0.0@c@;

This commit updates these lines to instead use the copysign function which is
how cpython does floor division.

Fixes numpy#12841.

BUG: fix signed zero behavior in npy_divmod

MAINT: Add functions to parse shell-strings in the platform-native way

Fixes numpygh-12979

Among other things, this means setting the `F90` environment variable to `"path to/f90" --coverage` works again.

BUG: Fix regression in parsing of F90 and F77 environment variables

BUG: parse shell escaping in extra_compile_args and extra_link_args

* add an Azure CI test to mimic recent
Anaconda DLL path resolution restrictions
that prevent NumPy imports

* patch numpy.core initialization to use
absolute DLL path loading to overcome
the scenario described above

* Test for DLL pollution in numpy/.libs/ in Azure
CI Windows Python 3.6 builds

* emit a warning with DLL names when > 1 are present
in numpy/.libs/

BLD: Windows absolute path DLL loading

- Create changelog
- Update release notes
- Update mailmap

[ci skip]

DOC: Prepare for 1.16.2 release.

* BUG: Use subprocess.check_output to avoid leaking file handles

This emitted a ResourceWarning as previously written, as the streams
never got closed.

This change also detects errors in `svnversion` and `hg`, rather than
ignoring the error code and hoping stdout contains a revision anyway.

* BUG: Fix distutils.get_version on python 3

Previously, this would throw `TypeError: cannot use a string pattern on
a bytes-like object`, which would then be silenced and never seen.

The previous commit stopped silencing this error, so now everything is
visibly broken rather than silently broken

Changing the strings to raw strings fixes the problem

* BUG: Don't use `close_fds`, which fails on windows

This argument was introduced along with subprocess in
f5afa78 to replace `popen4`, and
probably wasn't necessary in the first place.

BUG: Fixes to numpy.distutils.Configuration.get_version (numpy#13056)

`"Invalid version %r" % (1, 2)` would fail with `TypeError: not all arguments converted during string formatting`

The `Header is not a dictionary` error had a similar problem.

Fixed by changing this entire function to use `.format` in place of `%`, which does not have this gotcha.

Found using LGTM.com

This PR fixes numpy#13074.

Co-Authored-By: rainwoodman <[email protected]>

BUG: Fix errors in string formatting while producing an error

BUG: Convert fortran flags in environment variable

BUG: Remove error-prone borrowed reference handling

…me types

This prevents SystemErrors on overflow, fixing numpygh-13062

BUG: Add error checks when converting integers to datetime types

The patch was a workaround for our broken `exec_command`. It was removed
in numpygh-1217, but then added back in numpygh-1321.  Now that we've fixed
exec_command, I believe we can finally take it out for good.

BUG: Remove our patched version of `distutils.split_quoted`

This fixes TestConversion.test_int_from_huge_longdouble.

This fixes numpy.f2py.tests.test_kind.TestKind

As it was, the difference between stop and start was calculated and
it was assumed that that would be a numpy scalar or array. This is
not true for object input.

Closes numpygh-11927

BUG: Fix testsuite failures on ppc and riscv

BUG: Fix parameter validity checks in ``random.choice``

BUG: Ensure linspace works on object input.

BLD: fix include list for sdist building.

BUG: __array_interface__ offset was always ignored

The warning

[...]/fortranobject.c:138:18: warning: comparison of integers of different signs:
'Py_ssize_t' (aka 'long') and 'unsigned long' [-Wsign-compare]
if (size < sizeof(notalloc)) {
                ~~~~ ^ ~~~~~~~~~~~~~~~~

occurs 17 times when scipy is built.  The change in this pull request
casts `size` to `size_t` before comparing it to `sizeof(...)`.  A
previous check has already eliminated the possibility that `size` is
negative, so this cast is safe.

…numpy#13176)

Fixes numpygh-13165, where a double decref could occur on the error path.

Folded into this commit is some related control flow cleanup requested at review time.

Cast the covariance in the multivariate normal to double
so that the interpretation of tol is cleaner.

closes numpy#10839

MAINT: f2py: Add a cast to avoid a compiler warning.

BUG, MAINT: fix reference count error on invalid input to ndarray.flat

ENH: Cast covariance to double in random mvnormal

BUG: Fix null pointer dereference in PyArray_DTypeFromObjectHelper

Closes numpy#7801.

As a general rule, _every_ use of `.descr` is broken.

Fixes numpy#13297

* minor reviewer adjustments in PR
13301: in test_structured_padded it
is necessary to perform the seek
after writing to the file, and to
specify a fmt option to match
the expected result

The iteration variable has to be intp of course, unfortunately
a test is too slow to be practical (even as a slow test).

This code needs more refactoring, but since it is a minimal fix...

Closes numpygh-13286

This is mentioned in the release notes, so probably should be
discoverable.

- Change those mentions to links.
- Fix the docstring to not cause sphinx to emit warnings.

BUG: ufunc.at iteration variable size fix

BUG: Fix crash when calling savetxt on a padded array

BUG: Use C call to sysctlbyname for AVX detection on MacOS.

DOC: Add as_ctypes_type to the documentation

…xes-fixes

BUG: Return the coefficients array directly

BUG/MAINT: Tidy typeinfo.h and .c

a partial mitigation of numpy#12759.

see also https://nvd.nist.gov/vuln/detail/CVE-2019-6446

BUG: Make allow_pickle=False the default for loading

* Add missing headers that happened to be included before this file
* Fix accidentally exported symbols
* Move registering types in `multiarray.__dict__` to the source file that creates the types
* Add missing error checking to make up for PyStructSequence_InitType2 not being available

DOC: fix some doctest failures

BUG/MAINT: Tidy typeinfo.h and .c