Lambda the Ultimate

I think that R is an example of a powerful DSL. As I understand, that the higher level language the better it describes final intention of its used the easier it is to perform all sorts of optimization tricks under the hood.

I think tweaking R runtime to perform better under multicore environment is easier than doing same for C or Java or Python.

It may have changed a lot since in the 4 years since I used it at university but using it standardly, it is bad for large data sets. A lot of it is written in fortran. It does most processing in memory. However, I still use it, as most statistics can be done one smallish aggregated data sets anyway and memory is large enough now. It has everything statistical you could ever want unless you are professor in stats and most of them use it anyway. If its not there the profs normally create a module for it anyway for others to use.

It is funny that R and Perl are the two most widely used programming languages in Bioinformatics. The only reason I can imagine is that they provided something (text processing and statistic computing) needed in that field. I say funny, it is because according to me, both of them have quite obscure syntax and semantics, yet many bioinformaticians (especially those as biologist origin) have to struggle to learn them. Poor them!

At first glance, I read it as "R stand for Regression".
Good joke ;-)

There are a bunch of
statistics textbooks that use R, and the article itself is apparently controversial within the R/S community.

But by far the best thing is the picture.

Whoever thought they'd see an illustration of static scoping in the Times!

For a complete treatment of the state of parallel processing in R see this as yet unpublished JSS article.

The authors are far more optimistic than I on R's prospects. R's semantics guarantee virtually unlimited access to the executing environment, which would be hard to replicate across boxes. For example, R allows functions to call parent.frame() which returns a copy of the environment in which the function was called -- dynamic scope on demand. Functions can then change the environment, adding variables, changing state, you name it. Mutability stands in opposition to parallelization and R is crazy mutable.

Even on a single box there is tons of non-reentrant C and Fortran code below the hood. Not that it couldn't be rewritten, but rather that it won't.

R has a great community, but the community's collective strength is statistics, not algorithms and building concurrent systems. As someone who expects to be using R for years (against my will largely), I'm hoping for the best, but not expecting big improvements. At the same time, smart people are smart people. Perhaps I'm underestimating my colleagues.