An Experimental New Type Inference Scheme for Rust

http://smallcultfollowing.com/babysteps/blog/2014/07/09/an-experimental-new-type-inference-scheme-for-rust/

by pohl • 12 years ago

185 11 12 years ago

rayiner• 12 years ago

I really appreciate Niko's efforts into blogging about this sort of thing. It's really neat to watch Rust evolve in a fairly public manner. I'm not sure if there is any other language where these sorts of relatively early design decisions have been played out publicly before.

I also found this article really interesting: http://smallcultfollowing.com/babysteps/blog/2014/05/13/focu....

redline6561• 12 years ago

I always felt Slava Pestov's (now abandoned) blog on Factor's development was a good view into language design: http://factor-language.blogspot.com/

He and Andy Wingo (http://wingolog.org) both write very well on language issues though they may skew towards implementation concerns over design concerns.

pcwalton2• 12 years ago

My colleague zwarich aptly said that "this changes our type inference algorithm from a master's thesis to a blog post". As we approach 1.0, language simplifications like this are really valuable; the fact that it doesn't break existing code is even better.

phkahler• 12 years ago

I would say the fact that is doesn't break existing code is more like the minimum requirement ;-) Even better would be to work in every case the existing system does. But yeah, simplicity is very desirable.

sp332• 12 years ago

The rust language is still unstable. Breaking changes are expected up until the 1.0 release, because they want to make sure they get things right and don't have to live with bad syntax or semantics decisions made in 0.0.1 forever. After 1.0 it will be more stable :)

jmgrosen• 12 years ago

As someone currently working on a type inference scheme himself (although for a completely different purpose) and who is a Rust programmer, this sure was an interesting article! The inner type theorist in me says that losing any amount of expressiveness is a bad thing, but in this case my pragmatism seems to win out and say that this simpler but weaker scheme is probably good enough :)

wyager• 12 years ago

What would this break, theoretically? It's clearly not as rigorous as something based on HM, but I'm curious what the specific disadvantages are.

pcwalton2• 12 years ago

The article gives an example of a program that fails to typecheck, based on function variance.

Note that this algorithm is based on HM. You can't do plain old HM if you have subtyping; you have to do some algorithm based on it. This is just one of many possible algorithms.

freyrs3• 12 years ago

Unidirectional inference in the presence of subtyping is undecidable in the general case. Languages like Haskell and OCaml don't have this problem because they don't include subtyping precisely for this reason.

gsg• 12 years ago

OCaml has subtyping: it addresses the undecidability issue by requiring "cast" annotations for some use cases.

seanmcdirmid• 12 years ago

Semi-unification is always a pain, I really hope this is technically sound and tractable.

An Experimental New Type Inference Scheme for Rust

Related Stories