Hacker News Reader

Comment by fosap | Hacker News Reader

IMO a types stack languages does not make much sense. If the stack side effect is typed a function does alway have to return the same number of arguments and consume the same number. That IMO takes the by far biggest advantage of postfix languages away.

View on HN

Dynamic stack effects are pretty hard to reason about in nontrivial programs. Even Factor, which is largely dynamic wrt typing and dispatch, has a very limited number of words with dynamic stack effect.

Also, static types don’t necessarily preclude dynamic stack effects, as long as those effects can be characterised somehow by the type system. In my statically typed concatenative language, Kitten, I originally considered a notion of “regular” types that would let you do something like this:

    { :: r -> r End
    } :: r End a* -> r [a]

    { 1 2 3 } :: [Int]

In practice that proved to be too much of an implementation headache for not much benefit. Fixed arities also let you avoid sentinel values (as above) and parentheses (as in Lisps).

Factor's typing discipline is largely dynamic. However, they enforce the usage of annotations for higher order operations (recursive and inline) and strongly encourage the use of stack effect annotations. A stack checker, which is effectively a simple type system, proved pretty useful for debugging & optimizing.

See: http://docs.factorcode.org/content/article-inference.html

I do not have many experience with postfix languages. I mostly use RPL for simple ad-hoc programs, and I dynamic stack effects all the time. Mostly for logging and for error messages. In the "real world" one would not do that, so maybe I'm overestimating it.

Factor supports both dynamic and statically-unbalanced stack effects, but the former risks destabilizing the runtime and the later requires that defunctionalization (read: inlining) ultimately resolve to a static, balanced stack effect. It's all documented quite well in subpages of that article I linked you to.

In practice, all of the unbalanced stack-effects will become trivially balanced at word boundaries & dynamic stack-effects only occur when metaprogramming.

Reply on news.ycombinator.com