|Version 1 (modified by sweirich@…, 9 years ago) (diff)|
The Class System
This page summarizes the current status of proposals related to the Class System. This discussion has mostly focused on the issues surrounded by MultiParamTypeClassesDilemma, but there are a number of proposals that have been discussed.
Below, the major sections divide the proposals that probably will be included from those that probably won't, leaving the MPTC issue to its own section.
These are the proposals that seem to have some support.
- #32 add FlexibleInstances
- #31 add FlexibleContexts
- #70 allow TypeSynonymInstances
- #17 relax restriction on signatures of class methods
- #81 scoping of type variables in class instances
- #52 Generalised deriving for newtype
Based on current discussion, these proposals will probably not make it into the next Haskell standard. For each one, we summarize the reasons why not.
- #71 Allow Undecidable Instances
Undecidable instances destroy desirable properties of programming languages: decidable type checking and a specification of what programs are an are not allowed in the language. Although some sort of dynamic check could be used to ensure termination, this makes the type system brittle, and it is not clear that programs will be portable from one implementation to another.
- #54 add overlapping or incoherent instances
Overlapping instances add complexity to an already difficult problem. With overlapping instances:
- Haskell would need to be a lot more specific about exactly where context reduction takes place.
- A program that type checks can have its meaning changed by adding an instance declaration
- Similarly adding "import M()" can change the meaning of a program (by changing which instances are visible
- When exactly is overlap permitted? Is this ok?
instance C a Int instance C Bool b
- #18 add closed classes
- #19 control export and import of class instances
- #22 create DerivedInstances A general form of DerivedInstances hasn't really been discussed. But, the special case of NewtypeDeriving is a likely adoption.
- Defaults for superclass methods (no ticket, see List Discussion)
Multiparameter Type Class Dilemma
- #90 solve the MultiParamTypeClassesDilemma
- #49 add multi parameter type classes (MPTCs)
- #36 add FunctionalDependencies (FDs)
- #14 add Associated Types (ATs)
The bottom line is that Haskell' should have multiparameter type classes and functional dependencies. Multiparameter type classes are useful for programming, and something like functional dependencies are necessary to make them work. There has been a lot of discussion about whether Haskell' should include FDs or ATs, but FDs are emerging as the appropriate choice.
Comparison of FDs and ATs
Points in favor of FDs:
- FDs and ATs are the same order of technical difficulty
- We have more experience programming with FDs
- We have a better handle on just what restrictions make type inference tractable for FDs
- ATs are (perhaps) bit weaker in expressiveness, but nicer to program with in some cases. In a few other examples, FDs are nicer.
- ATs are not likely to be implemented in GHC soon
Points against FDs:
- Versions of FDs that are powerful enough for examples (monad transformer library) need complex restrictions to ensure termination and confluence. (It is not clear whether ATs would also need such complex restrictions.)
- ATs may win out in the end, and then we would have standardized too early. But perhaps FDs would not be that prevalent in code and so would be "easy" to update to a newer standard.
What version of FDs to use ?
Current discussion has focused more on the particular restrictions on FDs that should be adopted. See the FunctionalDependencies page for more info.