Joblib

---------------------------

Due to maintenance issues, 1.4.1 was not valid and we bumped the version to 1.4.2


- Fix a backward incompatible change in ``MemorizedFunc.call`` which needs to
return the metadata. Also make sure that ``NotMemorizedFunc.call`` return
an empty dict for metadata for consistency.
https://github.com/joblib/joblib/pull/1576

---------------------------

- Allow caching co-routines with `Memory.cache`.
https://github.com/joblib/joblib/pull/894

- Try to cast ``n_jobs`` to int in parallel and raise an error if
it fails. This means that ``n_jobs=2.3`` will now result in
``effective_n_jobs=2`` instead of failing.
https://github.com/joblib/joblib/pull/1539

- Ensure that errors in the task generator given to Parallel's call
are raised in the results consumming thread.
https://github.com/joblib/joblib/pull/1491

- Adjust codebase to NumPy 2.0 by changing ``np.NaN`` to ``np.nan``
and importing ``byte_bounds`` from ``np.lib.array_utils``.
https://github.com/joblib/joblib/pull/1501

- The parameter ``return_as`` in ``joblib.Parallel`` can now be set to
``generator_unordered``. In this case the results will be returned in the
order of task completion rather than the order of submission.
https://github.com/joblib/joblib/pull/1463

- dask backend now supports ``return_as=generator`` and
``return_as=generator_unordered``.
https://github.com/joblib/joblib/pull/1520

- Vendor cloudpickle 3.0.0 and end support for Python 3.7 which has
reached end of life.
https://github.com/joblib/joblib/pull/1487
https://github.com/joblib/joblib/pull/1515

---------------------------

- Fix a regression in ``joblib.Parallel`` introduced in 1.3.0 where
explicitly setting ``n_jobs=None`` was not interpreted as "unset".
https://github.com/joblib/joblib/pull/1475

- Fix a regression in ``joblib.Parallel`` introduced in 1.3.0 where
``joblib.Parallel`` logging methods exposed from inheritance to
``joblib.Logger`` didn't work because of missing logger
initialization.
https://github.com/joblib/joblib/pull/1494

- Various maintenance updates to the doc, the ci and the test.
https://github.com/joblib/joblib/pull/1480,
https://github.com/joblib/joblib/pull/1481,
https://github.com/joblib/joblib/pull/1476,
https://github.com/joblib/joblib/pull/1492

---------------------------

- Fix compatibility with python 3.7 by vendor loky 3.4.1
which is compatible with this version.
https://github.com/joblib/joblib/pull/1472

---------------------------

- Ensure native byte order for memmap arrays in ``joblib.load``.
https://github.com/joblib/joblib/issues/1353

- Add ability to change default Parallel backend in tests by setting the
``JOBLIB_TESTS_DEFAULT_PARALLEL_BACKEND`` environment variable.
https://github.com/joblib/joblib/pull/1356

- Fix temporary folder creation in `joblib.Parallel` on Linux subsystems on Windows
which do have `/dev/shm` but don't have the `os.statvfs` function
https://github.com/joblib/joblib/issues/1353

- Drop runtime dependency on ``distutils``. ``distutils`` is going away
in Python 3.12 and is deprecated from Python 3.10 onwards. This import
was kept around to avoid breaking scikit-learn, however it's now been
long enough since scikit-learn deployed a fixed (version 1.1 was released
in May 2022) that it should be safe to remove this.
https://github.com/joblib/joblib/pull/1361

- A warning is raised when a pickling error occurs during caching operations.
In version 1.5, this warning will be turned into an error. For all other
errors, a new warning has been introduced: ``joblib.memory.CacheWarning``.
https://github.com/joblib/joblib/pull/1359

- Avoid (module, name) collisions when caching nested functions. This fix
changes the module name of nested functions, invalidating caches from
previous versions of Joblib.
https://github.com/joblib/joblib/pull/1374

- Add ``cache_validation_callback`` in :meth:`joblib.Memory.cache`, to allow
custom cache invalidation based on the metadata of the function call.
https://github.com/joblib/joblib/pull/1149

- Add a ``return_as`` parameter for ``Parallel``, that enables consuming
results asynchronously.
https://github.com/joblib/joblib/pull/1393,
https://github.com/joblib/joblib/pull/1458

- Improve the behavior of ``joblib`` for ``n_jobs=1``, with simplified
tracebacks and more efficient running time.
https://github.com/joblib/joblib/pull/1393

- Add the ``parallel_config`` context manager to allow for more fine-grained
control over the backend configuration. It should be used in place of the
``parallel_backend`` context manager. In particular, it has the advantage
of not requiring to set a specific backend in the context manager.
https://github.com/joblib/joblib/pull/1392,
https://github.com/joblib/joblib/pull/1457

- Add ``items_limit`` and ``age_limit`` in :meth:`joblib.Memory.reduce_size`
to make it easy to limit the number of items and remove items that have
not been accessed for a long time in the cache.
https://github.com/joblib/joblib/pull/1200

- Deprecate ``bytes_limit`` in ``Memory`` as this is not automatically enforced,
the limit can be directly passed to :meth:`joblib.Memory.reduce_size` which
needs to be called to actually enforce the limit.
https://github.com/joblib/joblib/pull/1447

- Vendor ``loky`` 3.4.0 which includes various fixes.
https://github.com/joblib/joblib/pull/1422

- Various updates to the documentation and to benchmarking tools.
https://github.com/joblib/joblib/pull/1343,
https://github.com/joblib/joblib/pull/1348,
https://github.com/joblib/joblib/pull/1411,
https://github.com/joblib/joblib/pull/1451,
https://github.com/joblib/joblib/pull/1427,
https://github.com/joblib/joblib/pull/1400

- Move project metadata to ``pyproject.toml``.
https://github.com/joblib/joblib/pull/1382,
https://github.com/joblib/joblib/pull/1433

- Add more tests to improve python ``nogil`` support.
https://github.com/joblib/joblib/pull/1394,
https://github.com/joblib/joblib/pull/1395

-------------

- Fix a security issue where ``eval(pre_dispatch)`` could potentially run
arbitrary code. Now only basic numerics are supported.
https://github.com/joblib/joblib/pull/1327

- Make sure that joblib works even when multiprocessing is not available,
for instance with Pyodide
https://github.com/joblib/joblib/pull/1256

- Avoid unnecessary warnings when workers and main process delete
the temporary memmap folder contents concurrently.
https://github.com/joblib/joblib/pull/1263

- Fix memory alignment bug for pickles containing numpy arrays.
This is especially important when loading the pickle with
``mmap_mode != None`` as the resulting ``numpy.memmap`` object
would not be able to correct the misalignment without performing
a memory copy.
This bug would cause invalid computation and segmentation faults
with native code that would directly access the underlying data
buffer of a numpy array, for instance C/C++/Cython code compiled
with older GCC versions or some old OpenBLAS written in platform
specific assembly.
https://github.com/joblib/joblib/pull/1254

- Vendor cloudpickle 2.2.0 which adds support for PyPy 3.8+.

- Vendor loky 3.3.0 which fixes several bugs including:

- robustly forcibly terminating worker processes in case of a crash
(https://github.com/joblib/joblib/pull/1269);

- avoiding leaking worker processes in case of nested loky parallel
calls;

- reliability spawn the correct number of reusable workers.