[rust-dev] "Instance Chains: Type Class Programming Without Overlapping Instances" and type classes in Rust

[Patrick Walton]

On 1/9/12 5:36 PM, Tim Chevalier wrote:
> The problem the paper addresses is in Haskell, where having multiple
> instances in scope for the same class and type can cause unpredictable
> behavior. (The paper explains the basic problem in more detail pretty
> well.) It seems like there's an analogous issue in Rust when you
> import multiple sets of methods for the same type class, which I
> understand is handled with a compile failure. A difference is that
> with instance chains, an instance is not a "first-class" thing that
> you can import (like an item in Rust terminology) -- rather, you use
> type-level if-then expressions to disambiguate which instances apply.

Skimming the paper, I don't see where the modularity problem is 
addressed (e.g. one module defines an instance of Show int, another 
module defines an instance of Show int, now the two cannot be linked). 
It only solves the problem in the overlapping case, in which both 
typeclass instances apply, but one is more specific than the other (e.g. 
Show [int] and Show ['a]). My initial reaction is that Rust-style named 
instances are the simpler solution. We make the programmer decide 
explicitly which implementation to use; we never try to guess.

Patrick