[ci skip]

Implement `__dir__()` on group nodes to include in the list of
autocompletions also those children that are named as valid python
identifiers.

This tests autocompletion of group (node) names.

+ Add regular Group-attributes in its dir() test-case.

feat(group): autocomplete valid python-identifiers

+ Docs updated.
+ Simple test-case.
+ Not suitable for PY-2 (no `__dir__()` in datamodel).
+ Symetrical to #624(Groups).

+ See http://bugs.python.org/issue12166 &
http://stackoverflow.com/questions/15507848/the-correct-way-to-override-the-dir-method-in-python
why not fiddling with __dir__()
on PY2.  
Let's conform to this:
http://www.python3statement.org/#sections20-statement

+ Test dir() results for attributes.
# See discussion in #265.

feat(attributes): __dir__() autocompletes also valid py-indentifiers

The release notes for 3.3.0 -> 3.4.0 went missing.

[ci skip]

+ Add `Group.__getitem__()` change by ankostis.

Define `Group.__getitem__()` method

__item__ is not a standard dunder. Removed.

While trying to build PyTables using icc/icpc, the build would fail with the following error:

error: identifier "_mm256_castsi128_si256" is undefined
  _mm256_loadu2_m128i(const __m128i* const hiaddr, const __m128i* const loaddr)
  ^

test.c(5): error: expected a type specifier
  _mm256_loadu2_m128i(const __m128i* const hiaddr, const __m128i* const loaddr)
  ^

test.c(5): error: expected a type specifier
  _mm256_loadu2_m128i(const __m128i* const hiaddr, const __m128i* const loaddr)
  ^

test.c(5): error #141: unnamed prototyped parameters not allowed when body is present
  _mm256_loadu2_m128i(const __m128i* const hiaddr, const __m128i* const loaddr)
  ^

test.c(8): error: identifier "loaddr" is undefined
      _mm256_castsi128_si256(_mm_loadu_si128(loaddr)), _mm_loadu_si128(hiaddr), 1);
                                             ^

test.c(8): error: identifier "hiaddr" is undefined
      _mm256_castsi128_si256(_mm_loadu_si128(loaddr)), _mm_loadu_si128(hiaddr), 1);

It was diagnosed that immintrin.h from include directory of icc already defines split load/store intrinsics.
Ensuring that section of shuffle-avx2.c is not declared while building with icc/icpc, allows builds to go through.

enable building PyTables with Intel compiler (icc/icpc)

PEP 519 defines a filesystem path protocol, and enhances Python builtin I/O
functions to support "path-like" objects. As part of this PEP, the `os` module
also gained a `fspath` function, which takes an arbitrary path-like object and
returns a string.

If the os.fspath function is available, use it to convert `filename` to a
string.

Method Node._get_obj_timestamps() returns an object with attributes
atime, mtime, ctime and btime from the object's H5O_info_t struct.
These times as integer seconds since the UNIX epoch.

HDF5 stores timestamps associated with an object by default: access
time (atime), modification time (mtime), change time (ctime), and
birth time (btime).  Of these, only ctime has been implemented as of
HDF version 1.8.15.

This property retrieves the "track times" creation property of a
dataset using H5Pget_obj_track_times.

Retrieval of the property works in a newly created dataset but the
property does not seem to survive closing and reopening the file
(bug?).  So, the property has dubious value.  It might be better to
just advise users to check whether the ctime for the dataset is 0.

Allow disabling object time tracking when creating Tables, Arrays,
CArrays, EArrays, or VLArrays using H5Pset_obj_track_times, via an
optional keyword argument "track_times" in constructors, default True
for backwards compatibility.

Disabling object time tracking makes it at least possible to get
bitwise reproducible output files from different runs of the same
application.

From the HDF5 FAQ:
"If you turn off the create/modify/access time tracking for
  objects created (with the H5Pset_obj_track_times() routine),
  everything should be bit-for-bit reproducible. Coincidentally, it
  makes accessing those objects faster and the size of their metadata
  smaller also. You do lose the ability to know when the object was
  created/modified/accessed."

The user intent to track times / not track times is passed in to the
corresponding functions in the C source via the attribute
._want_track_times of the Leaf class, the way filters are.  These C
functions (H5ARRAYmake, H5TBOmake_table, and H5VLARRAYmake) have a new
boolean argument, track_times, before the data pointer argument.

DOC small documentation fixes

Thanks to @molpopgen gh-649

Avoids warnings in comparisons in Python 3

Fix slicing outside of data samples

New c-blosc 1.14.3 sources