Changes in rpy2¶
Python matrix multiplication (__matmul__ / @) added to R
threading.RLockis added to
rpy2.rinterface_lib.openrliband is used by the context manager
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.
Added callbacks for ptr_R_Busy() and ptr_R_ProcessEvents().
rstart now an objects in
rpy2.rinterface_lib.embedded(set to None until R is initialized).
Unit tests are included in a subpackage
rpy2.testsas was the case before release 3.0.0 (issue #528).
Experimental initialization for Microsoft Windows.
rpy2.situationis now also reporting the rpy2 version.
rpy2.robjecs.package_utils.default_symbol_resolve(). The named parameters default_symbol_check_after present in few methods in
rpy2.robjects.functionswere modified to keep a consistent naming.
The creation of R vectors from Python sequences is now relying on a method
_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.
rpy2.robjects.lib.ggplot2had stopped working with the R package ggplot2 reaching version 3.2.0. (issue #562).
Better handling of recent
pandasarrays 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
No longer allow installation if Python 3 but < 3.5.
Fixed error undefined symbol: DATAPTR if R < 3.5 (issue #565).
Fixed conversion of pandas
Seriesof dtype pandas.Int32Dtype, or pandas.Int64Dtype (issue #544).
Fixed the evaluation of R code using the “R magic” was delaying all output to the end of the execution of that code, independently of whether the attribute cache_display_data was True or False (issue #543).
Fixed conversion of
pandas.Seriesof dtype “object” when all items are either all of the same type or are
Failing to import pandas or numpy when loading the “R magic” extension for jupyter/ipython was hiding the cause of the error in the ImportError exception.
Fallback when an R POSIXct vector does not had an attribute “tzone” (issue #533).
Callback for console reset was not set during R initialization.
Fixed rternalized function returning rpy2 objects (issue #538).
–vanilla is no longer among the default options used to initialize R (issue #534).
Script to install R packages for docker image never made it to version control.
Conversion of R arrays/matrices into numpy object trigged a segfault during garbage collection (issue #524).
rpy2 can be installed without a development environment.
Unit tests are now relying on the Python module pytest.
rpy2.rinterface.NA_Integeris now only defined when the embedded R is initialized.
complete rewrite of
cffiis now used to interface with the R compiled shared library. This allows ABI calls and removes the need to compile binaries. However, if compilation is available (when installing or preparing pre-compiled binaries) faster implementations of performance bottlenecks will be available.
rpy2.rinterface.endr()multiple times is now only ending R the first time it is called (note: an ended R cannot successfully be re-initialized).
The conversion system in the mod:rpy2.robjects.conversion now has only two conversions py2rpy and rpy2py`. py2rpy tries to convert any Python object into an object rpy2 can use with R and rpy2py tries to convert any rpy2 object into a either a non-rpy2 Python object or a mod:rpy2.robjects level object.
The method get for R environments is now called find() to avoid confusion with the method of the same name in Python (
rpy2.robjects.vectors.Arraycan no longer be used to create R arrays of unspecified type. New type-specific classes (for example for vectors
rpy2.robjects.vectors.StrVector) should be used instead.
mod:rpy2.rpy_classic, an implementation of the rpy interface using
rpy2.rinterfaceis no longer available.
Row names in R data frames were lost when converting to pandas data frames (issue #484).
Mismatch between R’s POSIXlt wday and Python time struct_time’s tm_wday (issue #523).
Latest release of
DataFrame.from_items(). (issue #514).
Latest release of
pandasrequires categories to be a list (not an other sequence).
The numpy buffer implemented by R arrays is broken for complex numbers
Missing values in pandas
Categoryseries were creating invalid R factors when converted (issue #493).
Fallback for failure to import numpy or pandas is now dissociated from failure to import
repr()for R POSIX date/time vectors is now showing a string representation of the date/time rather than the timestamp as a float (issue #467).
The HTML representation of R data frame (the default representation in the Jupyter notebook) was displaying an inconsistent number of rows (found while workin on issue #466).
Handle time zones in timezones in Pandas when converting to R data frames (issue #454).
When exiting the Python process, the R cleanup is now explicitly request to happen before Python’s exit. This is preventing possible segfaults the process is terminating (issue #471).
dplyr method ungroup() was missing from
Delegate finding where is local time zone file to either a user-specified module-level variable default_timezone or to the third-party module
The pandas converter is converting
pandas.Seriesof dtype “O” to
rpy2.robjects.vectors.StrVectorobjects, issueing a warning about it (See issue #421).
The conversion of pandas data frame is now working with columns rather than rows (introduce in bug fix for issue #442 below) and this is expected to result in more efficient conversions.
Allow floats in figure sizes for R magic (Pull request #63)
Fixed pickling unpickling of robjects-level instances, regression introduced in fix for issue #432 with release 2.9.1 (issue #443).
Fixed broken unit test for columns of dtype “O” in pandas data frames.
Fixed incorrect conversion of R factors in data frames to columns of integers in pandas data frame (issue #442).
Fixing issue #432 (see Section Bugs fixed below) involved removed the method __reduce__ previously provided for all rpy2 objects representing R objects.
An error when installing with an unsupported R version was fixed (issue #420).
The docstring for rinterface.endr() was improperly stating that the function was not taking any argument (issue #423).
Target version of dplyr and tidyr are now 0.7.4 and 0.7.2 respectively.
Fixed memory leak when pickling objects (issue #432). Fixing the leak caused a slight change in the API (see Section Changes above).
pandasnow handling R ordered factor (issue #398).
jinja2was not listed as a dependency (issue #437).
rpy2.situationto extract and report informations about the environment, such as where is the R HOME, what is the version of R, what is the version of R rpy2 was built with, etc… The module is also designed to be run directly and provide diagnostics: python -m rpy2.situation.
VectorOperationsDelegatornow has a method __matmul__ to implement Python’s matrix multiplication operator (PEP-0645).
A rule to convert R POSIXct vectors to pandas Timestamp vectors was added (issue #418).
_repr_html_()for R vectors to display HTML in jupyter.
Starting several times the singleton
EventProcessorlonger results in a
RuntimeError. This is now only a warning, addressing issue #182.
The target version for the R package dplyr mapped is now 0.7.1, and
rpy2.robjects.lib.dplyr.src_dt()(issue #357) and
rpy2.robjects.lib.dplyr.src_desc()are no longer present.
Environment.keys()is now a iterator to match
dict.keys(), also an interator in Python 3.
Target version of ggplot2 library is 2.2.1.
Option stringsasfactors in the constructor for the class DataFrame. If False, the strings are no longer converted to factors. When converting from pandas data frames the default is to no longer convert columns of strings to factors.
The R “magic” for jupyter is now more consistently using the conversion system, and the use of custom converters through the magic argument -c will work as expected.
Docker-related files moved to directory docker/ (where variants image for rpy2 are available)
numpy.float128()is not available on all platforms. The unit test for it is now skipped on systems where it is not present (issue #347)
R pairlist objects can now be sliced (and issue #380 is resolved).
Passing parameters names that are empty string to R function was causing a segfault (issue #409).
Trying to build an atomic R vector from a Python object that has a length, but it not a sequence nor an iterator was causing a segfault (issue #407).
Trying to build an atomic R vector from a Python object that has a length, but it not a sequence nor an iterator was causing a segfault (issue #407 - backport from rpy2-2.9.0).
The defintion of the method
rpy2.rlike.container.OrdDict.itemswas incorrect, and so was the documentation for rcall (issue #383)
robjects.robject.RSlots.items()is now working (see pull request #57).
The context manager
rpy2.robjects.lib.grdevices.render_to_file()is no longer trying to impose a file name generated by
The symbol LISTSXP (corresponding to R pairlist objects) was not imported from the C module in rpy2.
The functions scale_linetype_discrete and scale_linetype_continuous in ggplot2 were not wrapped by
Fixed the error when the transformation of R “man” pages into Python docstrings was failing when the section “arguments” was missing (issue #368)
Failing to find R in the PATH during the installation of rpy2 is now printing an error message instead of a warning (issue #366)
R’s dplyr::src_dt was moved to dtdplyr::src_dt with dplyr release 0.5.0. To address this, src_dt will become a None if the R package dplyr is discovered to be of version >= 0.5.0 at runtime. (issue #357)
Conversion issue when R symbols were accessed as attribute of the singleton
rpy2.robjects.R. (issue #334)
The rmagic extension for ipython was no longer loading with the latest ipython (version 5.0.0). (issue #359)
The fix to issue #357 (see bugs fixed above) was expanded to cover all R packages wrapped in
rpy2.robjects.liband ensure that the respective Python modules can loaded even if symbols are no longer defined in future versions of the corresponding R packages.
Trying to install rpy2 while R is not in the PATH resulted in an error in setup.py.
rpy2.robjects.SourceCode. The class extends Python’s
strand is meant to represent R source code. An HTML renderer for the ipython notebook (syntax highlighting using
pygmentis also added).
rpy2.robjects.lib.tidyrproviding a custom wrapper for the R library tidyr
The long-deprecated functions
rpy2.rinterface.get_writeconsole()are no longer available. One of
rpy2.rinterface.get_writeconsole_warnerror()respectively should be used instead.
rpy2.robjects.RObject.slotscan now be implictly interated on (the method
__iter__()is now an alias for
The default Python-R conversion is now handling functions. This means that Python function can directly be used as parameters to R functions (when relevant).
Ipython display hook display_png for ggplot2 graphics.
pandas“category” vectors are better handled by the pandas conversion.
rpy2.robjects.lib.grdevicesproviding a custom wrapper for the R library ‘grDevices’, exposing few key functions in the package and providing context managers (render_to_file and render_to_bytesio) designed to simplify the handling of static plots (e.g., webserver producing graphics on the fly or figure embedded in a Jupyter notebook).
Numpy conversion is handling better arrays with dtype equal to “O” when all objects are either all inheriting from
bytes. Such arrays are now producing
R’s own printing of warnings if now transformed to warnings of type rinterface.RRuntimeWarning (it used to be a regular UserWarning)
The family of functions src_* and the function tbl in the R package dplyr have aliases in the module
rpy2.robjects.lib.dplyr, and a class
DataSourcehas been added for convenience.
rpy2.robjects.vectors.DataFramehas a method head corresponding to R’s method of the same name. The method takes the n first row of a data frame.
dplyr’s functions count_ and tally are now exposed as methods for the class
Building/installing rpy2 with a development version of R does not require the use of option –ignore-check-rversion any longer. A warning is simply issue when the R version is “development”.
On MSWindows, the dependency on pywin32 was removed (issue #315)
GroupedDataFramein the dplyr interface module is now inheriting from the definition of DataFrame in that same module (it was previously inheriting from
The default repr() for R objects is now printing the R classes (as suggested in issue #349).
Parameter names to R function that are in UTF-8 are no longer causing a segfault (issue #332)
Looking for a missing key in an R environment (using __getitem__ or [) could raise a LookupError instead of a KeyError.
R environment can now handle unicode keys as UTF-8 (was previously trying Latin1)
rpy2 is interrupting attempts to install with Python < 2.7 with an informative error message (issue #338)
Setting the R class can be done by using a simple Python string (issue #341)
rpy2.robjects.lib.grid.viewport is now returning an instance of class Viewport (defined in the same module) (issue #350)
Python objects exposed to R could lead to segfault when the Python process is exiting (issue #331)
American English spelling was missing for some of the function names to specify colour (color) scales.
Fix for printing R objects on Windows (pull request #47)
Pickling robjects-level objects resulted in rinterface-level objects when unpickled (issue #324).
rpy2.robjects.lib.ggplot2was modified to match the newly released ggplot2-2.0.0. This is introducing API-breaking changes, which breaks the promise to keep the API stable through bugfix releases within series, but without it 2.7.x will not a work with new default installation of the R package ggplot2.
Division and floordivision through the delegator .ro provided with R vectors wrapped by robjects. (issue #320)
Memory leak when unserializing (unpickling) R objects bundled in Python objects (issue #321)
rpy2.robjects.lib.dplyrwas not functioning.
Applied patch by Matthias Klose to fix implict pointer conversions.
pandas2ri.ri2py_dataframeis now propagating the row names in the R data frame into an index in the pandas data frame (issue #285)
methods union, intersect, setdiff, ungroup defined in the R package dplyr were missing from the DataFrame definition in
methods distinct, sample_n, and sample_frac defined in the R package dplyr were missing from the DataFrame definition in
The fix for the inheritance problem with
rpy2.robjects.lib.dplyr.DataFrameintroduced a regression whenever group_by is used.
The methods to perform joins on dplyr DataFrame objects where not working properly.
robjects-level vectors was broken for vectors of length 1 (issue #306)
The ipython notebook-based sections of the documentation were not building
Classes inheriting from
dplyr.DataFramehad dplyr methods returning objects of their parent class.
rpy2.rinterface.RParsingError. Errors occurring when parsing R code through
rpy2.rinterface.parse()raise this exception (previously
rpy2.robjects.conversion.Converterto replace the namedtuple of the same name
rpy2.robjects.converter.ConversionContext. This is a context manager allowing an easy setting of local conversion rules. The constructor has an alias called
rpy2.robjects.lib.dplyrproviding a custom wrapper for the R library dplyr
Environment.items()to iterate through the symbols and associated objects in an R environment.
rpy2.rinterface.ParsingIncompleError, a child class of
rpy2.rinterface.ParsingError, raised when calling
rpy2.rinteface.parse()results in R’s C-level status to be PARSE_INCOMPLETE. This can make the Python implementation of an IDE for R easier.
rpy2.robjects-level objects. The attribute is a
rpy2.robjects.Rslotswhich behaves like a Python mapping to provide access to R-attributes for the object (see issue #275).
The R “magic” for ipython %%R can be passed a local converter (see new features above) by using -c.
Conversion rules were not applied when parsing and evaluating string as R with
Calling the constructor for
rpy2.robjects.vectors.FactorVectorwith an R factor is no longer making a copy, loosing the associated R attributes if any (fixes issue #299).
rpy2 could crash when R was unable to dynamically load the C extension for one of its packages (noticed with issue #303).
rpy2.rinterface.is_initialized()is now a function.
rpy2.robjects.R.__call__()is now calling R’s base::parse() to parse the string rather the parser through R’s C-API. The workaround let’s us retrieve R’s error message in case of failure (see issue #300)
Metaclass RS4Auto_Type facilitating the creation of Python classes from R S4 classes was not handling classes without methods (issue #301)
Check that R >= 3.2 is used at build time (issue #291)
Conversion rules were not applied when parsing and evaluating string as R code with
Because of their long names, the classes
rpy2.robjects.packageshave now the aliases
Typo in function name emitting warnings at build time (issue #283)
The conversion of TaggedList instances is now handling the names of items in the list (issue #286)
Loading the ipython extension in the absence of pandas or numpy is now issuing a warning (issue #279)
Report the existence during build time of a file .Renviron, or the definition of the environment variables R_ENVIRON’ or `R_ENVIRON_USER with a warning. (issue #204)
Moved console writting callback to use ptr_R_WriteConsoleEx rather than ptr_R_WriteConsole. This allows callbacks for warnings and messages. get/set_writeconsole is now replaced by get/set_writeconsole_regular (regular output) and get/set_writeconsole_warnerror (warning and error). In order to conserve backward compatibility an alias for get/set_writeconsole_regular called get/set_writeconsole is provided.
Added callback for ptr_R_ResetConsole.
Categoricalobjects are automatically handled in the pandas converter.
The translation of R symbols into Python symbols used in importr and underlying classes and methods can be customized with a callback. The default translation turning . into _ is default_symbol_r2python.
Translation of named arguments in R function is now sharing code with the translation of R symbols (see point above), providing a consistent way to perform translations.
Utility function sequence_to_vector in robjects to convert Python sequences (e.g., list or tuple) to R vector without having to specify the type (the type is inferred from the list).
robjects.vectorsobject have a property
NAvaluethat contains the NA value for the vector, allowing generic code on R vectors. For example, testing whether any vector contains NA can be written as any(x is myvector.NAvalue for x in myvector). Making numpy /masked/ array is an other application.
The automatic name translation from R to Python used in importr is now slightly more complex. It will not only translate . to _ but should a conflict arise from the existence in R of both the . and _ versions the . version will be appended a _ (in accordance with :pep:0008). The change was discussed in issue #274).
The ipython ‘R magic’ is now starting with a default conversion mode that is pandas2ri if it can find it, then numpy2ri if it can find it, and then the basic conversion.
R vectors are now typed at the C level (IntSexpVector, FloatSexpVector, ListSexpVector, etc…) whenever retrieving them from the embedded R with the low-level rinterface. This is facilitating dispatch on vector type (e.g., with singledispatch now used for the conversion system).
The evaluation of R code through R’s C-level function tryEval caused console output whenever an error occurred. Moving to the seemingly experimental tryEvalSilent makes evaluations less verbose.
Multiple plots in one ipython cell (pull request #44)
simplegeneric was moved of ipython 4.0.0 (pull request #43)
Detection of the R version during setup on Win8 (issues #255 and #258)
Segmentation fault when converting
Serieswith elements of type object (issue #264)
The default converter from Python (non-rpy2) objects to rinterface-level objects was producing robjects-level objects whenever the input was of type
list(discovered while fixing issue #264)
Implemented suggested fix for issue with unlinking files on Windows (issue #191)
Testing rpy2 in the absence of ipython no longer stops with an error (issue #266)
Crash (segfault) when querying an R object in an R environment triggers an error (symbol exists, but associated values resolves to an error - issue #251)
Change in the signature of rcall was not updated in the documentation (issue #259)
Minor update to the documentation (issue #257)
Filter PNG files on size, preventing empty files causing trouble to be ipython notebook rendering of graphics later on (slight modification of the pull request #39)
Fix installation left unresolved with rpy2-2.5.3 (issue #248)
Possible segfault with Python 3.4 (issue #249)
- setup.py has install_requires in addition to requires in the hope to
fix the missing dependency with Python 2 (
singledispatchis required but not installed).
Extracting configuration information from should now work when R is emitting a warning (issue #247)
On OS X the library discovery step can yield nothing (see issue #246). A tentative fix is to issue a warning and keep moving.
String representation of
Check during build_ext if unsupported version of R (pull request #32)
HTMl display of columns of factors in a DataFrame (issue #236)
HTML display of factors (issue #242)
Require singledispatch if Python 3.3 (issue #232)
Fixed bug when R spits out a warning when asked configuration information (issue #233)
Restored printing of compilation information when running setup.py
Fixed installation issue on some systems (issue #234)
Workaround obscure failure message from unittest if Python < 3.4 and
singledispatchcannot be imported (issue #235)
Experimental alternative way to preserve R objects from garbage collection. This can be activated with rinterface.initr(r_preservehash=True) (default is False.
GGPlotobject getting a method
save()mirroring R’s ggplot2::ggsave().
The conversion system is now using generics/single dispatch.
rpy2.ipython.htmlwith HTML display for rpy2 objects
[Experimental] New function
robjects.methods.rs4instance_factory()to type RS4 objects with more specificity.
The script setup.py was rewritten for clarity and ease of maintenance. Now it only uses setuptools.
Use input rather than raw_input in the default console callback with Python 3 (fixes issue #222)
Issues with conversions, pandas, and rmagic (fixes issue #218 and more)
geom_raster was missing from rpy2.robjects.lib.ggplot2 (pull request #30)
Fixed issue with SVG rendering in ipython notebook (issue #217)
Regression with rx2() introduced with new conversion (issue #219)
Fixed documentation (missing import) (issue #213)
Assigning an R DataFrame into an environment was failing if the conversion for Pandas was activated. (Issue #207)
Conversion system slightly changed, with the optional conversions for
pandasmodified accordingly. The changes should only matter if using third-party conversion functions.
The Python 3 version is now a first class citizen. 2to3 is no longer used, and the code base is made directly compatible with Python. This lowers significantly the installation time with Python 3 (which matters when developping rpy2).
The default options to initialize R (rpy2.rinterface.initoptions’) are no longer `(‘rpy2’, ‘–quiet’, ‘–vanilla’, ‘–no-save’) but now (‘rpy2’, ‘–quiet’, ‘–no-save’).
robjects.vectors.ListVectorcan be instanciated from any objects with a method items() with the expectation that the method returns an iterable of (name, value) tuples, or even be an iterable of (name, value) tuples.
For instances of
rpy2.robjects.Function, the __doc__ is now a property fetching information about the parameters in the R signature.
rpy2.robjects.packages.data()to extract the datasets in an R pacakges
ipython’s rmagic is now part of
rpy. To use, %load_ext rpy2.ipython from within IPython.
rpy2.rinterface.SexpEnvironment.keys(), returnings the names in the environment as a tuple of Python strings.
robjects.packages.InstalledPackages, with a companion function
rinterface.SexpSymbolto represent R symbols
Changes in pandas 0.13.0 broke the rpy2 conversion layer (Issue #173)
Crash with R-3.0.2. Changes in R-3.0.2’s C API coupled to a strange behaviour with R promises caused the problem. (PR #150)
ggplot2’s “guides” were missing
ggplot2’s “theme_classic” was missing (PR #143)
ggplot2’s “element_rect” was missing (PR #144)
rpy2.interactive.packages()was broken (PR #142)
Several reports of segfault on OS X (since rpy2-2.3.1 - PR #109)
More fixes in converting DataFrames with dates from pandas
Missing mapping to ggplot2’s scale_shape_discrete function
Better handling of dates in Pandas
Constructor for POSIXct improved (and fixed)
rclassis no longer read-only and can be set (since R allows it)
Importing the module
rpy2.interactiveno longer activates event processing by default (triggering concurrency errors when used with ipython).
rpy2.interactive.ipython(so far plotting automatically a ggplot2 figure in the iPython’s console)
It is now possible to set the
Spurious error when running unit tests with Python 3 and numpy installed
Missing mapping to ggplot2’s geom_dotplot function
Warnings are not longer printed (see Changes below)
Bumped target version of ggplot2 to 0.9.3.1
Warnings are not longer printed. The C-level function in R became hidden in R-3.0, and the cost of an R-level check/print is relatively high if the R code called is very short. This might evolve into printing warnings only if interactive mode in Python (if this can be checked reliably).
rpy2.robjects.lib.ggplot2, a mapping to coord_fixed was missing (PR #120)
Using the parameter lib_loc in a call to
rpy2.robjects.packages.importr()was resulting in an error (PR #119)
Creating a layer through the rpy2.robjects.lib.ggplot2 interface did not accept parameters (PR #122)
Testing the Python version was crashing of a number of unsupported Python versions (<= 2.6) (PR #117)
New module pandas2ri to convert from mod:pandas DataFrame objects
rpy2.robjects.lib.grid.Gparto model their counterparts in R’s grid package as they were previously missing from rpy2.
Building on Win64 (pull request #6)
Fetching data from an R package through importr was masking any R object called data in that package. The data are now under the attribute name __rdata__. This is not completely safe either, although much less likely, a warning will be issued if still masking anything.
More informative error message when failing to build because R CMD config does not return what is expected
default console print callback with Python (issue #112 linked to it)
deprecation warnings with ggplot2 (issue #111 and contributed patch)
C-level API, allowing other C-level modules to make use of utilities without going through the Python level. The exact definition of the API is not yet fixed. For now there is PyRinteractive_IsInitialized() to assess whether R was initialized (through
C-module _rpy_device, allowing one to implement R graphical devices in Python [(very) experimental]
Tracking of R objects kept protected from garbage collection by rpy2 is now possible.
Sexp.rid()to return the identifier of the R object represented by a Python/rpy2 object
Dynamic build of Python docstrings out of the R manual pages
Build dynamic help
Build anonymous R packages from strings
importr(), the datasets are added as an attribute
data, itself an instance of a new class
PackageData. It no longer possible to access datasets are regular objects from a code package (because of changes in R), and the new system is more robust against quirks.
Newest R-2.15 and ggplot2 0.9 broke the ggplot2 interaface in
install process: Library location for some of the R installations
- should compile on win32 (thanks to a patch from Evgeny Cherkashin),
a work to a limited extend
Dynamic construction of S4 classes was looking for R help as ‘class.<class>’ rather than ‘<class>-class’
The cleanup of temporary directories created by R was not happening if the Python process terminated without calline
rpy2.rinterface.endr()(issue #68, and proof-of-principle fix by chrish42)
With the robjects layer, repr() on a list containing non-vector elements was failing
Support for Python 3, and for some of its features ported to Python 2.7
Environment.keys()to list the keys
robjects.vectors.POSIXltto represent vectors of R dates/time
packages.get_packagepath()to get the path to an R package
rpy2.robjects.helpto expose the R help system to Python
Metaclass utilities in
rpy2.robjects.methods, allowing to reflect automatically R S4 classes as Python classes.
rpy2.robjects.vectors.FactorVector.iter_labels()to iterate over the labels
rpy2.robjects.vectors.ListVectorto represent R lists.
rpy2.robjects.vectors.DataFrameaccept any iterable at the condition that the elements iterated through also valid subscripts for it (e.g., given an iterable v, the following is valid:
x[k] for x in v
NAComplexTypefor missing complex values.
SexpExtPtrto represent R objects of type EXTPTR (external pointers).
rpy2.rinterface.parse()to parse a string a R code
rpy2.rinterface.rternalise()to wrap Python function as
SexpClosurethat can be called by R just as it was a function of its own.
rpy2.rinterface.RNULLTypefor R’s C-level NULL value and
rpy2.rinterface.UnboundValueTypefor R’s C-level R_UnboundValue (both singletons).
rinterface.SexpVector.index(), of similar behaviour to
rpy2.rinterface.Sexp.list_attrs()to list the names of all R attributes for a given object.
rpy2.rinterface.ByteSexpVectorto represent R ‘raw’ vectors.
constant R_LEN_T_MAX to store what is the maximum length for a vector in R.
tuple R_VERSION_BUILD to store the version of R rpy2 was built against
Sexp.rclassto return the R class associated with an object
container.OrdDictget proper methods
A new sub-package to provide utilities for interactive work, either for handling R interactive events or use Python for interactive programming (as often done with the R console)
NA_bool, NA_real, NA_integer, NA_character and NA_complex are now deprecated (and removed). NA_Logical, NA_Real, NA_Integer, NA_Character, NA_Complex should be used.
classes representing R vector also inherit their type-specific rinterface-level counterpart.
rpy2.robjects.numpy2riis no longer sufficient to active the conversion. Explicit activation is now needed; the function activate can do that.
ComplexSexpVectorare now defined at the C level, improving performances and memory footprint whenever a lot of instances are created.
Better and more explicit detection system for needed libraries when compiling rpy2 (ported to release 2.1.6)
Long-standing issue with readline fixed (issue #10)
The R class in rpy2.robjects is now truly a singleton
When using numpy 1.5 and Python >= 2.7, the exposed buffer for R numerical (double) vectors or arrays was wrong.
Fixed issue with R arrays with more than 2 dimensions and numpy arrays (issue #47 - backported from the branch 2.2.x).
More fixes for the automated detection of include and libraries at build time.
Further fixes in the automatic detection of includes and libraries needed to compile rpy2 against R. The detection code has been refactored (backport from the 2.2.x branch)
fixes the automatic detection of R_HOME/lib during building/compiling when R_HOME/lib is not in lib/ (issue #54)
rpy2.robjects.lib.ggplot2now has the functions
ylim()exposed (patch contributed anonymously)
Install script when the BLAS library used by R is specified as a library file (patch by Michael Kuhn)
Spurious error message when using DataFrame.from_csvfile() without specifying col_names or row_names
Patch to finally compile with Python < 2.6 (contribDuted by Denis Barbier)
NA_Logical, NA_Real, NA_Integer, NA_Character from
rpy2.rinterfaceare imported by robjects.
NA_bool, NA_real, NA_integer, NA_character and NA_complex are now robjects-level vectors (they were rinterface-level vectors). Consider using the rinterface-defined NAs instead of them.
Missing conditional C definition to compile with Python 2.4 # issue 38
Fixed error when calling robjects.vectors.Vector.iteritems() on an R vector without names
Fixed automatic conversion issues (issue #41)
Issues with NA values # issue 37
Missing manual scale functions in
rpy2.robjects.lib.ggplot2# issue 39
Vector-like objects now in a module
set_accessors()for adding simply accessors to a class inheriting from
RS4_Typefor metaclass-declared accessors
DoubleExtractDelegatorfor extracting the R-way
DataFrame(formerly RDataFrame) can now be created from :rlike.container.OrdDict instances, or any other object inheriting from dict.
FactorVectorto represent R factors
the conversion is now returning subclasses of
robjects.vectors.Vector-formerly RVector- (such as
FloatVector, etc…) rather than only return
StrVectorhas a method
factor()to turn a vector of strings into an R factor
Matrixwas added the methods:
IntVector.tabulate()to count the number of times a value is found in the vector
Vector.sample()to draw a (random) sample of arbitrary size from a vector
NA_Complexas aliases for R’s missing values.
ComplexVectorfor vectors of complex (real + imaginary) elements
packagesto provide utility functions to handle R packages (import of R packages)
functionsto provide classes related to R functions, with the new class
DataFrame.iter_column(), iterating through rows and columns respectively.
DataFrame.rbind()for binding columns or rows to a DataFrame.
Vector.iteritems()to iterate on pairs of names and values.
Robject.__rname__to store the “R name”
New functions for specifying callback functions for R’s front-ends:
MissingArg, exposing R’s special object for representing a “missing” parameter in a function call. (#this was first a patch by Nathaniel Smith with a function getMissingArgSexp)
Initial commit of a callback-based implementation of an R graphical device (this is for the moment very experimental - and not fully working)
SexpClosure.rcall()is now taking 2 parameters, a tuple with the parameters and an
SexpEnvironmentin which the call is to be evaluated.
Sexp.__sexp__now has a setter method. This permits the rebinding of the underlying R SEXP, and allows to expose foo<- type of R methods as Python function/methods with side effects.
Objects of R type RAWSXP are now exposed as instances of class
unserialize()to build Sexp* instances from byte string serialized with R’s own ‘serialize’.
Ability to specify a callback function for R_CleanUp (called upon exiting R) through
ListSexpVectorfor creating R lists easily (complementing
StrSexpVector, and friends)
Array(formerly RArray) are now property-style getters
Pairlists (LISTSXP) now handled
set_interactive()to set whether R is in interactive mode or not (#following an issue reported by Yaroslav Halchenko)
R_NilValue, exposing R’s special object for representing a “NULL”.
ComplexSexpVectorfor vectors of complex (real + imaginary) elements
Scalar Python parameters of type
Nonein a call (using
SexpClosure) are now automatically converted to length-one R vectors (at the exception of None, converted to R_NilValue).
Python slices can now be used on R vector-like objects
Better handling of R’s missing values NA, NA_integer_, NA_real_, and NA_character_.
TaggedList, for creating a TaggedList from any object having a method
The setup.py script is now taking command-line arguments when specifying R library-related paths are wished. python setup.py –help build_ext will list them
RS4 no longer makes R’s slots as Python attributes through
The package is split into modules
The broken variables NA_STRING, NA_INTEGER, NA_LOGICAL, and NA_REAL are removed. The documentation on missing values was revised.
baseNameSpaceEnvwere renamed to
The parameter wantFun in
Environment.get()(formerly REnvironment.get()) is now wantfun
Vector.rdoes not have a __getitem__ method any longer (see in .rx and .rx2 in the new features)
DataFrameare now property-style getters
Arrayare now property-style getters
from_csvfile()and instance method
libto store modules representing R packages
lib.ggplot2for the CRAN package ggplot2.
renaming of few classes, the R prefix:
robjects.vectors.Vectorlost the (now redundant) methods subset and assign. Those operations were just aliases to the
emptyEnvwere renamed to
The parameter wantFun in
SexpEnvironment.get()is now wantfun
The call-back getters and setters are now
Functions also accept named parameters equal to Py_None, and transform them to R NULL (previously only accepted parameters inheriting from Sexp).
TaggedListis now a property (with a getter and a setter)
REnvironment.get()now accepts a named parameter wantFun (like
rinterface.SexpVectorwill now properly raise an exception when trying to create vector-like object of impossible type
Crash when trying to create a SexpVector of a non-vector type
R objects of class matrix are now properly converted into
Robj.as_py()was not working at all (and now it does to some extent)
On win32, printing an object was leaving an open file handle behind each time, leading to an error and the impossibility to print (# bug report and fix by Christopher Gutierrez)
No user-visible change. Win32-specific additions to the C module were made to compile it.
Crash when calling
SexpEnvironment.get()with an empty string #bug report by Walter Moreira
SexpEnvironment.__getitem__()called with an empty string caused unpredictable (and bad) things
Added missing named parameter wantfun to method
REnvironment.get()(making it similar to
Leak in reference counting when creating SexpVector objects fixed (the symptom was a process growing in size when creating R vector from Python list or numpy arrays)
R CMD config LAPACK_LIBS could return an empty string when R was compiled with the veclib framework, causing the setup.py script to raise an exception. setup.py now only print a message about an empty string returned from R CMD config
Numpy arrays with complex elements are no longer causing segfaults
SexpClosure.rcall()with something else that the expected kind of tuple could cause a segfault
process_revents(), a Wrapper for R_ProcessEvents (# suggested by June Kim to help with issues related to interactive display on win32), and for R_RunHandlers on UNIX-like systems (# patch by Nathaniel Smith).
All callbacks are getting a get<callback> to complement the set<callback>. (# Patch by Nathaniel Smith)
Sexp.__deepcopy__()to copy an object (calling Rf_Duplicate)
(# from a patch by Nathaniel Smith)
the default for reading and writing the console are now using sys.stdin and sys.stdout (# patch submitted by Nathaniel Smith)
console IO callbacks (reading and writing) are complemented by one to flush the console
Sexp.do_slot_assign()now creates the slot if missing (design-fix - # patch by Nathaniel Smith)
fixed problem of numpy interface with R boolean vectors. They are now presented as ‘i’ rather than ‘b’ to numpy (# patch submitted by Nathaniel Smith)
The mechanism for setting arbitrary callaback functions for console I/O now ensures that a traceback is printed to stderr whenever an error occurs during the evalutation of the callback (the raised exception used to be silently propagated to the next python call, leading to problems).
Fix installation bug when the include directories contain either ‘-‘ or ‘I’ #spotted by James Yoo
Failing to initialize R now throws a RuntimeError
Copying an R “NA” into Python returns a None (and no longer a True) (#fixes a bug reported by Jeff Gentry)
Property names for the
setnames()(this was likely forgotten for Release 2.0.0).
Property rclass for
Having the environment variable R_HOME specified resulted in an error when importing
rpy2.rinterface# root of the problem spotted by Peter
Setup.py has no longer a (possibly outdated) static hardcoded version number for rpy2
Testing no longer stops with an error in the absence of the third-party module
rpy2.rlike.container.TaggedList.pop()is now returning the element matching the given index
rpy2objects. # adapted from a patch contributed by Nathaniel Smith
Informative message returned as RuntimeError when failing to find R’s HOME
Use the registry to find the R’s HOME on win32 # snatched from Peter’s earlier contribution to rpy-1.x
rpy_classic.RObj.getSexp()moved to a property
py2ri()moved to the new module
conversion. Adding the prefix conversion. to calls to those functions will be enough to update existing code
Informative message returned as RuntimeError when failing to find R’s HOME
Use the registry to find the R’s HOME on win32 # snatched from Peter’s earlier contribution to rpy-1.x
added missing class
does not alias
rpy2.robjects.RDataFramechecks that R lists are data.frames (not all lists are data.frame)
Formerly new attribute
Ris now gone. The documentaion now points to
rpy2.rpy_classicfor this sort of things.
conditional typedef in rinterface.c to compile under win32 # reported and initial proposed fix from Paul Harrington
__pow__ was missing from the delegator object for robjects.RVector (while the documentation was claiming it was there) # bug report by Robert Nuske
Earlier change from Sexp.typeof() to getter Sexp.typeof was not reflected in
rpy2.rpy_classic# bug report by Robert Denham
RFormula, and defined environment as a property
defined names as a property for
Rsingleton. Setting it to True will translate ‘_’ into ‘.’ if the attribute is not found
constructor for RDataFrame now now accepts either
sexpTypeEmbeddedR()is now called
initOptionsis now called
initoptions. Changes of options can only be done through
Sexp.enclos()when R not yet initialized (bug report #2078176)
potential crash of
Sexp.frame()when R not yet initialized
proper reference counting when handling, and deleting,
setup.py: get properly the include directories (no matter where they are) #bug report and fix adapted from Robert Nuske
setup.py: link to external lapack or blas library when relevant
added a MANIFEST.in ensuring that headers get included in the source distribution #missing headers reported by Nicholas Lewin-Koh
rinterface.str_typeint()was causing segfault when called with 99
fixed subsetting for LANGSXP objects
setReadConsole(): specify Python callback for console input
R string vectors can now be built from Python unicode objects
__sexp__to return an opaque C pointer to the underlying R object
rsame()to test if the underlying R objects for two
Sexpare the same.
added emptyEnv (R’s C-level R_EmptyEnv)
R string vectors can now be built from Python unicode objects
functionalwith the functions
indexingwith the function
initEmbeddedR()is only initializing if R is not started (no effect otherwise, and no exception thrown anymore)
Sexp.typeof()was replaced by a Python getter
Sexp.named()was replaced by a Python getter
R objects of type LANGSXP are now one kind of vector (… but this may change again)
R objects of type EXPRSXP are now handled as vectors (… but this may change again)
Rremains a singleton, but does not throw an exception when multiple instances are requested
unable to compile on Python2.4 (definition of aliases to Python2.5-specific were not where they should be).
overflow issues on Python 2.4/64 bits when indexing R vector with very large integers.
handling of negative indexes for
trying to create an instance of
SexpVectorbefore initializing R raises a RuntimeException (used to segfault)
enclos()was not properly exported
setup.py was exiting prematurely when R was compiled against an existing BLAS library
complex vectors should now be handled properly by
package for R-like features in Python
named(), corresponding to R’s C-level NAMED
enclos()for SexpEnvironment corresponding to R’s C-level FRAME and ENCLOS
new experimental class
SexpLangfor R language objects.
R stack checking is disabled (no longer crashes when multithreading)
fixed missing R_PreserveObject for vectors (causing R part of the object to sometimes vanish during garbage collection)
prevents calling an R function when R has been ended (raise
method ‘getnames’ for
new helper class
indexing RVector the “R way” with subset is now possible through a delegating attribute (e.g., myvec.r[True] rather than myvec.subset(True)). #suggested by Michael Sorich
RDataFrame. The constructor
__init__()is still experimental (need for an ordered dictionnary, that will be in before the beta
filled documentation about mapping between objects
many fixes and additions to the documentation
improved GTK console in the demos
changed the major version number to 2 in order to avoid confusion with rpy 1.x # Suggested by Peter and Gregory Warnes
moved test.py to demos/example01.py
changed method name getNames to getnames where available (all lower-case names for methods seems to be the accepted norm in Python).
first public release