REL: 2.9.0

This is the first release of the h5py 2.9 series

This release support Python 2.7 and Python 3.4-3.7

Features:
 - Convenient high-level API for creating virtual datasets
 - Construct an h5py.File from a python file-like object
 - File now takes parameters for controlling the chunk cache
 - Option to preserve insertion order of attribute and group entries
 - Can now create datasets backed by external non-HDF5 files
 - List and tuples of rriunicode strings can be stored as HDF5 attributes

Bug fixes:
 - Fix reading/writing of float128 data
 - Fix converting data to float16 when creating a dataset

Support for old Python
 - Drops support for Python 3.3
 - Drops support for Python 2.6

REL: 2.9.0rc1

The first RC for 2.9.0

REL: 2.8.0

This is the first release of the h5py 2.8 series

Features
 - allow registration of new drivers
 - add option to track object creation order

Bug fixes
 - allow slices with stop < start as empty
 - improve building with custom hdf5 installation
 - ensure libver defaults to LIBVER_EARLIEST
 - improve handling of float types
 - support non-ascii external paths on py2
 - avoid crashing IPython auto-completer

API changes
 - deprecate ``h5t.available_ftypes``.  This is no longer
   used internally and will be removed in the future.  There is
   no replacement public API.
 - Do not sort fields in compound types.  This is account for changes
   in numpy 1.14.
 - Minimum version of cython is now 0.23

Wheels hdf5 version
 - Wheels upload to pypi will be built against hdf5 1.10.2

REL: 2.8.0rc1

This is the first RC for h5py 2.8.0

Features
 - allow registration of new drivers
 - add option to track object creation order

Bug fixes
 - allow slices with stop < start as empty
 - improve building with custom hdf5 installation
 - ensure libver defaults to LIBVER_EARLIEST
 - improve handling of float types
 - support non-ascii external paths on py2
 - avoid crashing IPython auto-completer

API changes
 - deprecate ``h5t.available_ftypes``.  This is no longer
   used internally and will be removed in the future.  There is
   no replacement public API.
 - Do not sort fields in compound types.  This is account for changes
   in numpy 1.14.
 - Minimum version of cython is now 0.23

REL: 2.7.1

First bugfix release in 2.7 series.

This has several critical fixes for possible segfault and deadlocks.

- Fixed critical issue with cyclic gc which resulted in segfaults
- Avoid unaligned access fixing h5py on sparc64
- Fixed compilation issues for some library locations
- Fix deadlock between phil and the import lock in py2
- Improve windows handling if filenames
- Allow close to be called on file multiple times
- Warn on loaded vs complied hdf5 version issues
- Fix overflow computing size of dataset on windows
- Do not mangle capitalization of filenames in error messages
- Fix longdouble on ppc64le
- Fix compounds structs with variable-size members

REL: 2.7.0

h5py version 2.7.0

 - Dropped support for python 3.2.
 - Added support for `pathlib`.
 - Improved auto-complete in IPython/jupyter for py3k.
 - Added low-level support for Direct Chunk Write and
   File Image Operations.
 - Support for reading non-standard float types.
 - Improved support for compound data types.
 - Improved support for ppc.
 - Many bug fixes.

REL: 2.7.0rc3

Third release candidate for 2.7.0.

Bug fix for ppc64el architecture.

REL: v2.7rc2

This is the second release candidate for h5py 2.7

 - Dropped support for python 3.2.
 - Added support for `pathlib`.
 - Improved auto-complete in IPython/jupyter for py3k.
 - Added low-level support for Direct Chunk Write and
   File Image Operations.
 - Support for reading non-standard float types.
 - Improved support for compound data types.
 - Many bug fixes.
 - Version correctly bumped.

REL: v2.7rc1

This is the first release candidate for h5py 2.7

 - Dropped support for python 3.2.
 - Added support for `pathlib`.
 - Improved auto-complete in IPython/jupyter for py3k.
 - Added low-level support for Direct Chunk Write and
   File Image Operations.
 - Support for reading non-standard float types.
 - Improved support for compound data types.
 - Many bug fixes.

Tag 2.6.0