Allow type families in RuntimeReps (#13105) · Issues · Glasgow Haskell Compiler / GHC

Allow type families in RuntimeReps

Conal Elliott provided me with this puzzle:

type family RepRep a ∷ RuntimeRep

class HasRep a where
  type Rep a ∷ TYPE (RepRep a)
  repr ∷ a → Rep a
  abst ∷ Rep a → a

type instance RepRep Int = IntRep

instance HasRep Int where
  type Rep Int = Int#
  abst n = I# n
  repr (I# n) = n

I think we should accept. However, doing so (even with my solution to #12809 (closed) complete) is hard, because we frequently consult a kind in order to determine a runtime representation. When that kind is TYPE (RepRep Int), the code generator throws up its arms in despair.

The solution here is either to teach the code generator how to normalise types (requiring propagating the FamInstEnvs) or to do a whole-program transformation at some point (zonker? desugarer? maybe we can wait until !CorePrep or even unarisation?) to change ids whose types have kinds like TYPE (RepRep Int) into ids with types with kinds like TYPE IntRep.

But I don't want to let this hold me up at the moment, so I'm posting here as a reminder to revisit this problem.

Trac metadata

Trac field	Value
Version	8.0.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

Conal Elliott provided me with this puzzle:

```
type family RepRep a ∷ RuntimeRep

class HasRep a where
  type Rep a ∷ TYPE (RepRep a)
  repr ∷ a → Rep a
  abst ∷ Rep a → a

type instance RepRep Int = IntRep

instance HasRep Int where
  type Rep Int = Int#
  abst n = I# n
  repr (I# n) = n
```

I think we should accept. However, doing so (even with my solution to #12809 complete) is hard, because we frequently consult a kind in order to determine a runtime representation. When that kind is `TYPE (RepRep Int)`, the code generator throws up its arms in despair.

The solution here is either to teach the code generator how to normalise types (requiring propagating the `FamInstEnvs`) or to do a whole-program transformation at some point (zonker? desugarer? maybe we can wait until !CorePrep or even unarisation?) to change ids whose types have kinds like `TYPE (RepRep Int)` into ids with types with kinds like `TYPE IntRep`.

But I don't want to let this hold me up at the moment, so I'm posting here as a reminder to revisit this problem.

<details><summary>Trac metadata</summary>

| Trac field             | Value        |
| ---------------------- | ------------ |
| Version                | 8.0.1        |
| Type                   | Bug          |
| TypeOfFailure          | OtherFailure |
| Priority               | normal       |
| Resolution             | Unresolved   |
| Component              | Compiler     |
| Test case              |              |
| Differential revisions |              |
| BlockedBy              |              |
| Related                |              |
| Blocking               |              |
| CC                     |              |
| Operating system       |              |
| Architecture           |              |

</details>