RCL uses CFFI to communicate with the R environment, RCLmath links with libRmath and makes some functions accessible in a much simpler way. The code is available under the MIT license. Note that there is another project working on communication between Common Lisp and R: RCLG.

rcl

This package give access to the full R environment.

	AllegroCL 8.0	CLISP 2.41	CMUCL 19d	LispWorks 5.0	OpenMCL 1.1	SBCL 1.0
Windows/i386	OK	OK	NA	OK	NA	OK
Linux/i386	OK	OK	fails	OK	NA	fails
MacOSX/i386	OK	NO FFI	NA	fails	NA	OK
MacOSX/PPC	OK	OK	OK	fails	fails	OK

The results above correspond to FedoraCore6 running on a PIV, Tiger running on a G4 and a CoreDuo, and WindowsXP running in VMWare Fusion (using R 2.5.0/2.5.1RC). Note that a success means only that some basic operations work, no extensive testing is performed in any platform.

The failures seem to have the same origin. Somehow we can't find anything in the R environment: (r-header *r-globalenv*) returns in all this failing combinations (4 0 0 32768 0 0 0 0 0 0) (in working systems we get (4 0 2 32768 1 0 0 0 0 0), where the numbers correspond to the fields in the header: (TYPE OBJ NAMED GP MARK DEBUG TRACE FIN GCGEN GCCLS)).

Download: rcl.tgz (older snapshots rcl-070316.tgz, rcl-060716.tgz,rcl-060703.tgz,rcl-060623.tgz).

When decoding an object with attributes, the list of attributes is returned as a secondary value (it's also possible to print attributes to standard-output whenever encountered decoding an object, to be aware of their existence).

CL-USER> (in-package :rcl)
#<PACKAGE "RCL">
R> (r-init)
T
R> (r "R.Version")
("i386-apple-darwin8.10.1" "i386" "darwin8.10.1" "i386, darwin8.10.1" "Patched"
 "2" "6.0" "2007" "10" "03" "43075" "R"
 "R version 2.6.0 Patched (2007-10-03 r43075)")
((:NAMES "platform" "arch" "os" "system" "status" "major" "minor" "year"
  "month" "day" "svn rev" "language" "version.string"))

In its current state it might be possible to use RCL successfully for simple tasks, but there are a number of issues:

there is no error handling ("R_tryEval" is used, it always returns with a flag that indicates if there was an error), but at least the error messages and warnings can be redirected to a lisp stream and are not lost
there is no event loop (so the "quartz" device doesn't work properly, for example), CarbonEL might help with this (untested)
there might be issues related to garbage collection or other effects due to the ugly interection between the lisp and the R worlds; I made no attemp to handle all these problems (see RCLG fore a more robust approach)
the conversion R<->lisp has to be improved: better encapsulation of R objects is desirable, and complex objects (in particular data frames) are handled very poorly

RCL

A Common Lisp Interface to R

rcl