Euphonic

- New Features:

- A system has been added for reference data in JSON files. These
are accessed via ``euphonic.utils.get_reference_data`` and some
data has been added for coherent scattering lengths and cross-sections.
This system has been made available to the
``calculate_structure_factor()`` method; it is no longer necessary to
craft a data dict every time a program uses this function.

- Changes:

- Fixed structure factor formula in docs (``|F(Q, nu)|`` -> ``|F(Q, \\nu)|^2``
and ``e^(Q.r)`` -> ``e^(iQ.r)``)

- Bug fixes:

- Fix ``'born':null`` in ``ForceConstants`` .json files when Born is not
present in the calculation (see
`c20679c <https://github.com/pace-neutrons/Euphonic/commit/c20679c>`_)
- Fix incorrect calculation of LO-TO splitting when ``reduce_qpts=True``,
as the 'reduced' q rather than the actual q was used as the q-direction
(see `3958072 <https://github.com/pace-neutrons/Euphonic/commit/3958072>`_)
- Fix interpolation for materials with non-symmetric supcercell matrices,
see `81 <https://github.com/pace-neutrons/Euphonic/issues/81>`_
- Fix interpolation for force constants read from Phonopy for materials that
have a primitive matrix and more than 1 species, see
`77 <https://github.com/pace-neutrons/Euphonic/issues/77>`_

- Breaking Changes:

- There has been a major refactor, for see the latest
`docs <https://euphonic.readthedocs.io/en/latest>`_ for how to use, or
`here <https://euphonic.readthedocs.io/en/latest/refactor.html>`_ for
refactor details
- Python 2 is no longer supported. Supported Python versions are ``3.6``,
``3.7`` and ``3.8``

- New Features:

- Euphonic can now read Phonopy input! See
`the docs <https://euphonic.readthedocs.io/en/latest>`_
for details.

- Improvements:

- Added ``fall_back_on_python`` boolean keyword argument to
``ForceConstants.calculate_qpoint_phonon_modes`` to control
whether the Python implementation is used as a fallback to the C
extension or not, see
`35 <https://github.com/pace-neutrons/Euphonic/issues/35>`_
- Added ``--python-only`` option to ``setup.py`` to enable install
without the C extension

- Bug fixes:

- On reading CASTEP phonon file header information, switch from a fixed
number of lines skipped to a search for a specific line, fixing issue
`23 <https://github.com/pace-neutrons/Euphonic/issues/23>`_
- Fix NaN frequencies/eigenvectors for consecutive gamma points, see
`25 <https://github.com/pace-neutrons/Euphonic/issues/25>`_
- Fix issue saving plots to file with dispersion.py, see
`27 <https://github.com/pace-neutrons/Euphonic/issues/27>`_
- Fix incorrect frequencies at gamma point when using dipole correction
in C, `45 <https://github.com/pace-neutrons/Euphonic/issues/45>`_

- Bug fixes:

- Add MANIFEST.in for PyPI distribution

- Bug fixes:

- Cannot easily upload C header files to PyPI without an accompanying source
file, so refactor C files to avoid this

- There are several breaking changes:

- Changes to the object instantiation API. The former interface
``InterpolationData(seedname)`` has been changed to
``InterpolationData.from_castep(seedname)`,` in anticipation of more codes
being added which require more varied arguments.
- Changes to the Debye-Waller calculation API when calculating the structure
factor. The previous ``dw_arg`` kwarg accepted either a seedname or length
3 list describing a grid. The new kwarg is now ``dw_data`` and accepts a
``PhononData`` or ``InterpolationData`` object with the frequencies
calculated on a grid. This is to make it clearer to the user exactly what
arguments are being used when calculating phonons on the grid.
- Changes to parallel functionality. The previous parallel implementation
based on Python's multiprocessing has been removed and replaced by a
C/OpenMP version. This has both better performance and is more robust. As
a result the ``n_procs`` kwarg to ``calculate_fine_phonons`` has been
replaced by ``use_c`` and ``n_threads`` kwargs.

- Improvements:

- The parallel implementation based on Python's multiprocessing has been
removed and now uses C/OpenMP which both has better performance and is more
robust
- Documentation has been moved to readthedocs and is more detailed
- Clearer interface for calculating the Debye-Waller factor
- Better error handling (e.g. empty ``InterpolationData`` objects, Matplotlib
is not installed...)

- Bug fixes:

- Fix gwidth for DOS not being converted to correct units
- Fix qwidth for S(Q,w) broadening being incorrectly calculated

- Undo removal of SIGINT handler before launching multiprocessing Pool, as this causes an error when calling from Matlab, and doesn't significantly improve usability