Packages in R & Java

-----Original Message-----
From: rossini@blindglobe.net [mailto:rossini@blindglobe.net]
Sent: 27 February 2004 15:32
To: Marsland, John
Cc: r-devel@stat.math.ethz.ch
Subject: Re: [Rd] Packages in R & Java


"Marsland, John" <John.Marsland@CommerzbankIB.com> writes:

Yes, lots of packages do use C/Fortran code ... it would be 

great if there

were more packages that are "pure R" especially since the 

advent of S4

classes and namespaces.

Even with those, quite often we have to move code to C/C++/Fortran for
efficiency.

Is it worth suggesting that we create a designation "pure 

R" for packages

that have no external source code and encourage more of 

these? Indeed, now

the number of packages is so big could they be classified 

to aid potential

users?

Not sure that would be good (encouragement).  It would compromise
solid design principles for aesthetics.  

Going back to the original question, it strikes me that 

C/Fortran code is

most often used to interface to other code - there would be 

a better way of

doing this in a Java environment - or to speed up existing 

R code - where

loops are usually the cause of the speed problem and the 

introduction Java

bytecode would presumably be helpful anyway. Surely there 

could be some sort

of process to generate automated jni bindings?

Argh.  JNI is not a saviour.  It's a royal PITA.

And C/Fortran code is probably just as often used to speed up 
computations.

I don't think a Java implementation of the R parser is as 

bigger task as it

seems ... much of the work to get an interpreted scripting 

language working

could be taken from the Jython project?

See the Omegahat LANGUAGE for a solid start on this.  R-like syntax,
and lots more.  Does anyone still use it?

best,
-tony

-- 
rossini@u.washington.edu