Rpy2

=============

Bugs fixed
----------

- An incomplete backport of the bug fixed in 3.2.3 broke the ABI mode.

=============

Bugs fixed
-----------

- Error when parsing strings as R codes could result in a segfault.

=============

Bugs fixed
----------

- Python format error when trying to report that the system is not reported
on Windows (issue 597).

- The setup script would error on build if R is not installed. It is now
printing an error message.

=============

Bugs fixed
----------

- The wrapper for the R package `dbplyr` could not import the underlying
package (refactoring elsewhere was not propagated there).

- Creating R objects called `names` `globalenv` caused the method
:meth:`Sexp.names` to fail (issue 587).

- Whenever the pandas conversion was activated :class:`FloatSexpVector` instances
with the R class `POSIXct` attached where not corrected mapped back to pandas
datetime arrays. (issue 594).

- Fix installation when an installation when a prefix without write access is used
(issue 588).

=============

New features
------------

- rpy2 can built and used with :mod:`cffi`'s ABI or API modes (releases 3.0.x and
3.1.x were using the ABI mode exclusively). At the time of writing the default
is still the ABI mode but the choice can be controlled through the environment variable
`RPY2_CFFI_MODE`. If set, possible values are `ABI` (default if the environment
variable is not set), `API`, or `BOTH`. When the latter, both `API` and `ABI`
modes are built, and the choice of which one to use can be made at run time.

Changes
-------

- The "consoleread" callback (reading input to the R console) is now assuming UTF-8
(was previously assuming ASCII) and is no longer trying to add a "new line" character
at the end of the input.

- Querying an R environment with an invalid key will generate a :class:`TypeError`
or a :class:`ValueError` depending on the issue (rather than always :class:`ValueError`
before.

Bugs fixed
----------

- `setup.py` is now again compatible with Python2 (issue 580).

- Unit tests were failing if numpy is not installed.

- :mod:`rpy2.situation` is no longer breaking when R is not the in path and
there is no environment variable `R_HOME`.

- Build script for the cffi interface is now using the environment
variable `R_HOME` whenever defined (rather that always infer it from the
R in the PATH).

- Converting R strings back to Python was incorrectly using `Latin1` while `UTF-8` was
intended (issue 537).

=============

New features
------------

- Python matrix multiplication (`__matmul__` / ``) added to
R :class:`Matrix` objects.

- An :class:`threading.RLock` is added to :mod:`rpy2.rinterface_lib.openrlib` and is
used by the context manager :func:`rpy2.rinterface_lib.memorymanagement.rmemory`
to ensure that protect/unprotect cycles cannot be broken by thread switching, at least
as long as the context manager is used to handle such cycles (see issue 571).

- The documentation covers the use of notebooks (mainly Jupyter/Jupyterlab).

- The PNG output in Jupyter notebooks R cells can now specify an argument `--type`
(passed as the named argument `type` in the R function `png`).
For example on some Linux systems and R installations, the type `cairo`
can fix issues when alpha transparency is used.

Changes
-------

- Added callbacks for `ptr_R_Busy()` and `ptr_R_ProcessEvents()`.

- `rstart` now an objects in :mod:`rpy2.rinterface_lib.embedded`
(set to `None` until R is initialized).

- Unit tests are included in a subpackage :mod:`rpy2.tests` as was the
case before release 3.0.0 (issue 528).

- Experimental initialization for Microsoft Windows.

- :mod:`rpy2.situation` is now also reporting the rpy2 version.

- :func:`rpy2.robjecs.package_utils.default_symbol_check_after` was
renamed :func:`rpy2.robjecs.package_utils.default_symbol_resolve`.
The named parameters `default_symbol_check_after` present in few methods
in :mod:`rpy2.robjects.packages` and :mod:`rpy2.robjects.functions` were
modified to keep a consistent naming.

- Trying to instantiate an :class:`rpy2.rlike.container.OrdDict` with a
a :class:`dict` will result in a :class:`TypeError` rather than a
:class:`ValueError`.

- Methods of :class:`rpy2.rlike.container.OrdDict` now raises a
:class:`NotImplementedError` when not implemented.

- The creation of R vectors from Python sequences is now relying on a method
:meth:`_populate_r_vector` that allows vectorized implementation to
to improve speed.

- Continuous integration tests run against Python 3.6, 3.7, and 3.8. It is
no longer checked against Python 3.5.

Bugs fixed
----------

- `aes` in :mod:`rpy2.robjects.lib.ggplot2` had stopped working with the
R package ggplot2 reaching version 3.2.0. (issue 562).

- Better handling of recent :mod:`pandas` arrays with missing values
(related to issue 544).

- The mapping of the R operator `%in%` reachable through the attribute `ro`
of R vectors was always returning `True`. It is now working properly.

- R POSIXct vectors with `NA` dates were triggering an error when converted
in a data frame converted to :mod:`pandas` (issue 561).