# Copyright 2014-2018 The ODL contributors
#
# This file is part of ODL.
#
# This Source Code Form is subject to the terms of the Mozilla Public License,
# v. 2.0. If a copy of the MPL was not distributed with this file, You can
# obtain one at https://mozilla.org/MPL/2.0/.
"""Entry points for adding more spaces to ODL using external packages.
External packages can add an implementation of `TensorSpace` by hooking
into the setuptools entry point ``'odl.space'`` and exposing the methods
``tensor_space_impl`` and ``tensor_space_impl_names``.
This is used with functions such as `rn`, `cn`, `tensor_space` or
`uniform_discr` in order to allow arbitrary implementations.
See Also
--------
NumpyTensorSpace : Numpy-based implementation of `TensorSpace`
"""
from __future__ import print_function, division, absolute_import
from odl.space.npy_tensors import NumpyTensorSpace
# We don't expose anything to odl.space
__all__ = ()
IS_INITIALIZED = False
TENSOR_SPACE_IMPLS = {'numpy': NumpyTensorSpace}
def _initialize_if_needed():
"""Initialize ``TENSOR_SPACE_IMPLS`` if not already done."""
global IS_INITIALIZED, TENSOR_SPACE_IMPLS
if not IS_INITIALIZED:
# pkg_resources has long import time
from pkg_resources import iter_entry_points
for entry_point in iter_entry_points(group='odl.space', name=None):
try:
module = entry_point.load()
except ImportError:
pass
else:
TENSOR_SPACE_IMPLS.update(module.tensor_space_impls())
IS_INITIALIZED = True
[docs]def tensor_space_impl_names():
"""A tuple of strings with valid tensor space implementation names."""
_initialize_if_needed()
return tuple(TENSOR_SPACE_IMPLS.keys())
[docs]def tensor_space_impl(impl):
"""Tensor space class corresponding to the given impl name.
Parameters
----------
impl : str
Name of the implementation, see `tensor_space_impl_names` for
the full list.
Returns
-------
tensor_space_impl : type
Class inheriting from `TensorSpace`.
Raises
------
ValueError
If ``impl`` is not a valid name of a tensor space imlementation.
"""
if impl != 'numpy':
# Shortcut to improve "import odl" times since most users do not use
# non-numpy backends
_initialize_if_needed()
try:
return TENSOR_SPACE_IMPLS[impl]
except KeyError:
raise ValueError("`impl` {!r} does not correspond to a valid tensor "
"space implmentation".format(impl))