Release 3.5.13
==============

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

- R is setting a number of environment variables in a wrapper script. Those
  variables were not set by default by rpy2 and this could lead to issues (see issue #1033).
  The variables are now added to `os.environ`.

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

- `python -m rpy2.situation LD_LIBRARY_PATH` could incorrectly truncate the value
  for `LD_LIBRARY_PATH` when the variable `LD_LIBRARY_PATH` is already defined.  

Changes
-------

- The dependency on :mod:`pytz`, a deprecated package, was
  removed and replaced by a dependency to :mod:`zoneinfo`.
  If Python < 3.9 the package :mod:`backports.zoneinfo`
  is an added dependency.

- The depency on :mod:`tzlocal` does no longer limit it
  to `<5.0` (issue #1035).

- Remove dependency on :mod:`distutils` in Python's standard library. That package is
  deprecated and removed in Python 3.12 (see https://peps.python.org/pep-0632/ and
  issue #1040).

Release 3.5.12

New features

• :func:rpy2.rinterface.evalr_expr_with_visible to evaluate
  an expression and return the visibility status of the
  value returned.

• More events in the intialization and termination of an
  embedded R logged by :mod:rpy2.rinterface_lib.embedded.logger
  at INFO and DEBUG levels.

• :meth:rpy2.robjects.R.__call__ has 2 named arguments "visible" and
  "print_r_warnings" to handle R "invisible" results and print R warnings
  at the end of the evaluation. Invisible results happen when doing
  rpy2.robjects.r("x <- 1"). R will return the value of x "invisibly".
  The default are invisible is True and print_r_warnings is True.

Bugs fixed

• "R magic" cells are now show R warnings at the end of the
  evaluation of an R cell (#issue 226).

Changes

• Evaluating a string as R code using :meth:rpy2.robjects.R.__call__
  (e.g., rpy2.robjects.r("1+2")) now shows R warnings at the end
  of the evaluation by default.

• :meth:rpy2.robjects.R.__call__ returns the results invisibly by default
  (see section "New features" for this release).

Release 3.5.11

New features

• :class:pandas.Categorical objects are now converted to R
  factor objects.

Bugs fixed

• :mod:pandas 2.0 makes API-breaking changes that made
  conversion fail (issue #1013).

• The default converter for R magic was dropping R factors
  to :mod:numpy arrays of integers rather than convert them
  to pandas :class:Category (issue #1010).

Changes

• The fix for issue #1010 (see above) changes the numpy conversion
  for R factors. It returns an arrays with the level factors rather
  than the integer codes for the levels.

• :func:rpy2.rojects.pandas2ri.py2rpy_categoryseries raises a
  :class:DeprecationWarning.

Release 3.5.10

New features

• :mod:rpy2.situation reports cffi interface type information.

• The ipython/jupyter R magic can now have jpeg as a default graphics format
  for static figures in addition to png and svg.

Bugs fixed

• Building the path to the R shared library was incorrectly
  using the output of R CMD config LIBnn (issue #982).

• R started raising warnings when calling formals on BUILTINSXP R objects.
  Internal functions in :mod:rpy.robjects.functions calling formals no
  longer propagate the warnings.

New features

• Type hints in rpy2 are now checked with mypy.

Bugs fixed

• Building the path to the R shared library was incorrectly
  using the output of R CMD config LIBnn (issue #982).

• R started raising warnings when calling formals on SPECIALSXP R objects.
  Internal functions rpy.robjects.functions calling formals no longer
  propagate the warnings.

• The numpy converter was not turning NA_character_ into
  None (issue #979).

• rpy2.situation included an eager import rpy2 that could cause a version
  mismatch error with some build/install toolchain (issue #984). The import is now
  lazy/delayed.

• Installation targets pandas, all, and test now specify pandas>=1.2.0
  (which should limit frequencies of issues like #998).

Changes

• rpy2.robjects.conversion.Converter objects are no longer functioning context
  managers. An exception is now raised when trying to use it that way.
  The feature was introduced with rpy2-3.5.7 but it does not guarantee the locality
  of a context manager and can result in permanently changed conversion rules.
  The method rpy2.robjects.conversion.Converter.context should be used instead.

Release 3.5.8

New features

• Default value for the interactivity status of the embedded R
  as initialization is now a private variable
  _DEFAULT_R_INTERACTIVE in :mod:rpy2.rinterface_lib.embedded.

Bugs fixed

• :func:rpy2.rinterface.initr_checkenv (aliased to
  :func:rpy2.rinterface.initr) did not include
  initialization parameters for interactive, _want_setcallbacks,
  and _c_stack_limit (issue #976).

• The numpy converter was not listing R objects of type CHARSXP as
  vectors. This make R NA_character_ values wrapped in rpy2 Python
  objects to be passed as-is (issue #983).

Changelog here:
https://github.com/rpy2/rpy2/blob/RELEASE_3_5_7/NEWS