[ANN] mpi4py release 4.0.0

[Lisandro Dalcin]

* New features:

  + Add support for the MPI-4.0 standard.

    - Use large count MPI-4 routines.
    - Add persistent collective communication.
    - Add partitioned point-to-point communication.
    - Add new communicator constructors.
    - Add the `Session` class and its methods.

  + Add support for the MPI-4.1 standard.

    - Add non-destructive completion test for multiple requests.
    - Add value-index datatype constructor.
    - Add communicator/session buffer attach/detach/flush.
    - Support for removal of error classes/codes/strings.
    - Support for querying hardware resource information.

  + Add preliminary support for the upcoming MPI-5.0 standard.

    - User-level failure mitigation (ULFM).

  + `mpi4py.util.pool`: New drop-in replacement for `multiprocessing.pool`.

  + `mpi4py.util.sync`: New synchronization utilities.

  + Add runtime check for mismatch between `mpiexec` and MPI library.

  + Support `scikit-build-core`_ as an alternative build backend.

  .. _scikit-build-core: https://scikit-build.readthedocs.io/

  + Support `meson-python`_ as an alternative build backend.

  .. _meson-python: https://meson-python.readthedocs.io/

* Enhancements:

  + `mpi4py.futures`: Support for parallel tasks.

  + `mpi4py.futures`: Report exception tracebacks in workers.

  + `mpi4py.util.pkl5`: Add support for collective communication.

  + Add methods `Datatype.fromcode()`, `Datatype.tocode()` and
    attributes `Datatype.typestr`, `Datatype.typechar` to simplify
    NumPy interoperability for simple cases.

  + Add methods `Comm.Create_errhandler()`, `Win.Create_errhandler()`,
    and `File.Create_errhandler()` to create custom error handlers.

  + Add support for pickle serialization of instances of MPI types.
    All instances of `Datatype`, `Info`, and `Status` can be
    serialized. Instances of `Op` can be serialized only if created
    through `mpi4py` by calling `Op.Create()`. Instances of other MPI
    types can be serialized only if they reference predefined handles.

  + Add `handle` attribute and `fromhandle()` class method to MPI
    classes to ease interoperability with external code. The handle
    value is an unsigned integer guaranteed to fit on the platform's
    ``uintptr_t`` C type.

  + Add lowercase `free()` method to MPI classes to ease MPI object
    deallocation and cleanup. This method eventually attempts to call
    `Free()`, but only if the object's MPI handle is not a null or
    predefined handle, and such call is allowed within the World Model
    init/finalize.

* Backward-incompatible changes:

  + Python 2 is no longer supported, Python 3.6+ is required, but
    typing stubs are supported for Python 3.8+.

  + The `Intracomm.Create_group()` method is no longer defined in the
    base `Comm` class.

  + `Group.Compare()` and `Comm.Compare()` are no longer class methods
    but instance methods. Existing codes using the former class
    methods are expected to continue working.

  + `Group.Translate_ranks()` is no longer a class method but an
    instance method. Existing codes using the former class method are
    expected to continue working.

  + The `LB` and `UB` datatypes are no longer available, use
    `Datatype.Create_resized()` instead.

  + The `HOST` predefined attribute key is no longer available.

  + The `MPI.memory` class has been renamed to `MPI.buffer`. The old
    name is still available as an alias to the new name.

  + The `mpi4py.dl` module is no longer available.

  + The `mpi4py.get_config` function returns an empty dictionary.

* Miscellaneous:

  + The project is now licensed under the BSD-3-Clause license. This
    change is fairly inconsequential for users and distributors. It
    simply adds an additional clause against using contributor names
    for promotional purposes without their consent.

  + Add a new guidelines section to documentation laying out new fair
    play rules. These rules ask companies and outside developers to
    refrain from reusing the ``mpi4py`` name in unaffiliated projects,
    publishing binary mpi4py wheels on the main Python Package Index
    (PyPI), and distributing modified versions with incompatible or
    extended API changes. The primary motivation of these rules is to
    avoid fragmentation and end-user confusion.

--

Lisandro Dalcin
============
Senior Research Scientist

Advanced Algorithm and Numerical Simulations Laboratory (AANSLab)

Computer, Electrical and Mathematical Science and Engineering (CEMSE)

King Abdullah University of Science and Technology (KAUST)

https://cemse.kaust.edu.sa/aanslab