=============
New features
------------
- :obj:`rpy2.rinterface.NULL` can be refererenced in code before the embedded
R is initialized.
- `LD_LIBRARY_PATH` is obtained from the executable `RScript` and set prior
to initializing R at the :mod:`rpy2.rinterface` level (issue 833).
- The package is now fully typed with type hints.
- The :func:`rpy2.rinterface.evalr` accepts the named parameters `envir`
and `enclos`, mirroring better the signature of the R function `eval`.
- New function :func:`rpy2.rinterface.evalr_expr` to evaluate R expressions.
- :mod:`rpy2.situation` is now using a module-level (called `logger`) to
facilitate troubleshooting. When the module is run as an executable
the flag `--verbose` can take the possible values DEBUG, INFO, WARNING, ERROR.
Changes
-------
- Better customizablity of conversion rules from R objects with `typeof`
`INTSXP` (array of integers), `VECSEXP` (R list), and `LGLSXP` (array of
booleans) :mod:`pandas`.
- The representation of R integer is no longer using a comma separator
for thousands (issue 797).
- Requirement is now Python>=3.7. :mod:`rpy2` is no longer tested with Python 3.6.
- :mod:`rpy2.robject.lib.grid` was refactored to fix the conversion. This
fixes issue 804. A classes now have a `classmethod` `r` for the R
constructor (the `__init__` is the Python constructor).
- :obj:`rpy2.robjects.vectors.__all__` is removed.
- Cleaner class design by having :class:`rpy2.rinterface.NumpyArrayMixin`
removed and replaced by
:class:`rpy2.rinterface.SexpVectorWithNumpyInterface`.
- With Python < 3.8 the module :mod:`typing_extensions` is now a requirement.
- Initialization and update steps for
:class:`rpy2.robjects.conversion.NameClassMap` were updated to ensure
type hints are correct. This probably solved cryptic bugs with conversion
system.
- :meth:`rpy2.rinterface.LangSexpVector.from_string` was at the robject level (class method of
:class:`rpy2.robjects.language.LangVector`) but is now at the interface level.
- :meth:`rpy2.rinterface.get_evaluation_context` is deprecated. The module
level context variable `evaluation_context` should be used instead. The
use of the context variable should address issues when using nested
:meth:`rpy2.rinterface.local_context`.
- Added recommendation to try :meth:`pandas.DataFrame.infer_objects` when
the converion to R data frames fails because of mixed types (issue 794).
- No longer implicit dependency on :mod:`numpy` for memory views. The Python
API has an unresolved issue around handling FORTRAN-ordered arrays without
going to C-level (https://bugs.python.org/issue34778), and C-extension
as added to rpy2 to cut the dependency on numpy for just this (PR 856).
Bugs fixed
----------
- R warning about partial argument name matching is now cleared (issue 821).
- Setting a slice for a StrSexpVector where some of the elements are not strings
no longer fails.
- Assigning items to an R array of strings using a slice of more than one
element was likely to fail.
- The :mod:`rpy2.rinterface`-level conversion of Python integers to R integers is now
checking that integers are not larger than what R can handle.
- Trying to access items in an R environment before R is initialized is no
longer ending with a segfault (issue 842).