Interesting interview with Xavier Leroy

Interesting interview, covering things like how CompCert came about, the early history of OCaml, etc.


JY: What do you think are the most interesting or important problems that the community could or should be focusing on in the next few years?

XL: Maybe it’s time to go back to programming language design and try to have a little more imagination. In the 90’s, we thought a new PL would solve all our problems, and nowadays, we tend to think that we are here to develop verification tools for old languages. Maybe we should go back a little bit to domain-specific languages. A pet peeve of mine is how hard it is to program massively parallel computations. With GPUs in particular, highly parallel hardware has become very affordable, yet we still program it with dialects of C and shoddy tools.

Quite a lot packed in this forward-looking bit.

Here’s a paradigm without a language: static & strongly-typed APL.


If we ignore syntax, a prototype of that in a modern ML derivative wouldn’t be too bad. If you want to make it feel like APL, that would be a challenge. Either way, an interesting thought experiment.


Take a look at Remora.


If I understand this vision of parallel computing correctly, I think Paul Chiusano’s Unison language aims (among other things) to make distributed computing seamless in this sense:

A very cool project (ML heritage, type inference, rich-text editing, the works).