As an indirect result of some of conversations in the past few days at Google and Microsoft, I was left thinking that I need a better way to explain to someone why basic research of languages and language constructs is necessary. This crops up every now and then, and it would be useful if I had a simple answer to this. I think I have an approximate answer now, what do you think:

I think basic research about language constructs is necessary because it helps separate concerns. It separates of the quality of the abstraction that the construct provides from the details of the implementation of the construct. In this way one can study and construct and make observation about how valuable it would be as a tool for expressing computation. Once we have a great abstraction, we can then focus our attention on the best possible implementation of the abstraction or some select component of it, now with a clear understanding of the expressive power of the end result.

As an example our present notions of concurrent programming with threads and locks/semaphores is a broken idiom - they are hard to program with, are brittle and don’t compose well. The notion that it would be nicer to write concurrent programs as memory transactions came about and gained favor because the quality of the abstractions that memory transactions provided was far greater than that of locks and semaphores. By quality I mean that memory transactions compose far better, provide guarantees about progress, fairness etc and are not prone to the sorts of pitfalls that lock-based approaches suffer from. This view of transactions matured into a set of key combinators that gave us "composable" memory transactions. Once the idea that "memory transactions are good" gained ground it spurred a cottage industry of people trying to efficiently implement these abstractions.

The underlying point is that the search for abstractions to represent computation is not hindered by details of implementation. What we refer to as an implementation is only the expression of an idea in terms of other abstractions that previous languages have imposed on us. Limiting ourself to the scope of these, baises us against ideas that dont appear natural to them. I like to think of this as a form of "the evil of premature optimisation" at the level of ideas.