Some thoughts on the Next Generation Statistical Computing Environment
Duncan Temple Lang

R has been terifically successful for a variety of reasons.  However,
some of the design is based on work from almost 40 years ago.  As the
information technology landscape has changed so dramatically and
continues to evolve, it is important to ask whether we are still on
the right path.  We need to plan not just for the immediate needs of
the statistical computing community, but also build infrastructure for
the future.  We need this to enable us and others to
experiment with new paradigms and innovate rather than simply program.

R is being used for different purposes than the original interactive
EDA environment. Developing software with these tools is probably
suboptimal.  While one can extend R through the R language, it is very
difficult to extend the system itself. This makes it difficult to
introduce new data structures at the system level and leaves them as
second class objects.  We need extensible data types to be able to
take advantage of application-specific information to do complex,
efficient computations. Additionally, we continue to need to interface
to other languages, extending the notion of interface and relying on
meta-data, be it dynamic/run-time or static and treat external objects
natively.

Given the limited resources we have in our community, we need to be
intelligent in how we leverage the work of other communities in shared
infrastructure.  I'll discuss some of the possible approaches we might
consider for evolving or building the next generation system and
discuss some of the tradeoffs and sociological aspects of such
development.  The key notions are the traditional staples -
extensibility, components, meta-data. But importantly, they need to be
applied at the right level and different audiences need to be
identified and characterized.