Lessons from Fortress

Guy Steele says the Fortress project is “wrapping up”, apparently because further implementation work is not expected to be very enlightening. I suppose this is a good time to get round to blogging about it...

Alan Eliasen asks for commentary and lessons learned. Here are some things I've learned from Fortress:

  • If your language includes a flashy experimental syntax, that will dominate perceptions of it, even by knowledgeable and interested people. Fortress includes a lot of potentially useful work about compiling a flexible dynamically-typed language efficiently, but who notices?
  • If you describe generic functions as dynamically resolved overloading, they no longer look weird to O-O programmers. Indeed, they hardly look like a feature at all; they look like the natural way overloading should work. You can even get away with adding them to a non-Lisp language!
  • There's surprisingly little connection between language features and target audience. If the Fortress documentation didn't say it was intended as a replacement for Fortran, I would never have guessed. The mathematical syntax and many numeric types hint at this, but not strongly enough to overcome the dynamic typing and generic functions.
  • Batch compilation is a reasonable thing. Dynamism is traditionally associated with other forms of power and flexibility, but it's OK to do without it (or offer a batch mode that disables it) for the sake of optimization.
  • It's nice to clearly document what features of your language are new or unusual or special, for the convenience of readers who are interested in it as research. Otherwise they may overlook your new ideas among the rest, especially if the language is a complex one like Fortress.
  • Features that express something nonlocally can also have local versions. IIRC, old versions of Fortress had a dispatch form that had the same semantics as method dispatch, but in a typecase-like form without the nonlocality of method definition. (It doesn't seem to exist in the current version, though. Maybe I'm hallucinating and it never existed.)
  • The stated purpose of a language affects how interesting I think it is. Fortress is mostly a compilation research project, but it claims to be a replacement for Fortran, which makes it sound much less interesting than it really is.

If most of these lessons are about documentation and perception rather than about the language itself, that's because I've never used Fortress, only read its documentation and some of the blog posts (which, unfortunately, have vanished, along with most of the website, but may be restored someday).

2 comments:

  1. Actually, dynamic typing is getting to be common in scientific programming languages (Python, Matlab/Octave, R). Julia, which is very new, adds generic functions as well.

    ReplyDelete

It's OK to comment on old posts.