Strongly typed languages (like C, Java, etc.) provide very useful about possible errors in code. Similarly, high-level, un-typed languages like R, Matlab, Perl, Python avoid the need for declarations, allow reuse of variable names, and generally make interactive and prototyping use more productive and rapid. Ousterhout wrote about the difference between the two. The Omegahat language and several other languages have explored a potentially much richer framework which represents a compromise between the two extremes - optional typing. Essentially, we can add information about parameter and return types for functions if we know it. Otherwise, we can omit it. This can apply to individual parameters or individual combinations of inputs. Developers that provide type information are rewarded by having the interpreter perform run-time checking. The information also can be used potentially in compilation (e.g. generating byte code). And, importantly, it makes it significantly easier or even feasible to export the functions to other languages and computational interfaces. By annotating R functions, we can export methodology to other languages in an automated manner.

The OptionalTyping package provides a prototype mechanism of specifying optional type information about parameters and return values of R functions. In addition to merely specifying the types to aid relectance and the generation of interfaces to these functions, the package also provides an implementation entirely in R of a type checking mechanism. This allows us to experiment with and express different semantics for the type validation.

Software

R Package - OptionalTyping_0.1-0.tar.gz This can be installed as a regular R package:

  R CMD INSTALL OptionalTyping_0.1-0.tar.gz

Documentation

Some thoughts on aspects of the optional typing approach

Duncan Temple Lang <duncan@wald.ucdavis.edu>

Last modified: Fri Sep 30 21:44:35 PDT 2005