GHC/SuperClass

[in draft May 2017]

There's an idiom that's briefly mentioned in the User Guide under Equality constraints [1], but is more general and more powerful.

You get the power from a combination of extensions:

• -XMultiParamTypeClasses [2]
• Superclass constraints, which can also be MultiParam [3]
• -XFunctionalDependencies [4]
• and/or superclass Equality constraints with Type families, having the effect of FunDeps [5]
• The overall effect can satisfy the FunDep Coverage conditions, which means you might not need -XUndecidableInstances [6]

The User Guide [7] says:

Equality constraints can also appear in class and instance contexts. The former enable a simple translation of programs using functional dependencies into programs using family synonyms instead.

That's not wrong; but by talking about "class and instance contexts" in the same breath does fail to emphasise that:

• whereas Instance constraints don't apply until after the instance has been selected
  (And it's a common newbie mistake to think they restrict the selection.)
• superclass constraints apply during type improvement before selecting instances
  as described here [8] "superclass context is critical ..." [thanks @DFeuer]
• This behaviour means that even if you don't explicitly put a FunDep | ... -> ... on a class:
  • If you put a superclass equality constraint on a class, that induces a FunDep as explained in [9];
    but just as much
  • If you put a superclass class constraint on a class, and that superclass has a FunDep, that also induces a FunDep.
  • and so on for a superclass constraint of a superclass constraint of a superclass constraint ....

Functional dependency-like behaviour through superclass constraints

Constraints in the class context can achieve the effect of Functional dependencies even without the explicit vertical bar and dependency arrow syntax in the class declaration. Either:

• Using an Equality constraint to a Type family [10]; or
• A superclass constraint that does have an explicit Functional dependency.
• Or a superclass constraint which is itself constrained in one of those ways.

A class declaration of the form

class C a b | a -> b

can equivalently be given as

class (F a ~ b) => C a b where
  type F a

That is, we represent every functional dependency (FD) a1 .. an -> b by an FD type family F a1 .. an and a superclass context equality F a1 .. an ~ b, essentially giving a name to the functional dependency. [See [11]; using (~) needs -XTypeFamilies or -XGADTs.] In class instances, we define the type instances of FD families in accordance with the class head.

Or class C can equivalently be given as

class (D a b) => C a b ...

class D a b | a -> b
-- or
class (G a ~ b) => D a b where
  type G a

The class instances for D will closely follow those for C. So there is not much gain in expressivity. Consider also:

class (D a c) => E a b c ...

-- class D as above

for which the D instances will be more compact than E.

Note that D might not declare any methods, but merely be used for type improvement.

Method signatures for class C are not affected by either of these ways for achieving dependencies.

Reference this thread: [12], with some fancier examples of pseudo-FunDeps.

Is this power made clear in the theoretical literature? It's kinda implicit in the TypeFamilies2008 paper; but note that was written concurrently with Type Families development; and published before the work was completed. Specifically, supporting Type Families and (~) in superclass constraints was delivered a long time after the rest of the work.