Template Haskell pretty-printer doesn't parenthesize infix datatype names in data declarations

If you run this program:

{-# LANGUAGE ExistentialQuantification #-}
{-# LANGUAGE PolyKinds #-}
{-# LANGUAGE TemplateHaskell #-}
{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE TypeOperators #-}
module Foo where

import Language.Haskell.TH

main :: IO ()
main = do
  putStrLn $([d| data a :~: b where Refl1 :: a :~: a |] >>= stringE . pprint)
  putStrLn $([d| data a :~~: b  = a ~ b => Refl2     |] >>= stringE . pprint)

$ /opt/ghc/8.2.1/bin/runghc Foo.hs
data :~:_0 a_1 b_2 where Refl1_3 :: :~:_0 a_4 a_4
data :~~:_0 a_1 b_2 = a_1 ~ b_2 => Refl2_3

It'll print the output incorrectly. Those infix names :~: and :~~: ought to be surrounded by parentheses, since they're used in prefix position.

Trac metadata

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

changed weight to 5

added Tbug TemplateHaskell Trac import newcomer labels

I am not sure, but I think that it's enough to change pprName' Alone to pprName' Applied here.

There are 8 failed tests in testsuite/tests/th because of this change:

• GHC.Types.: becomes (GHC.Types.:)
• Data.Type.Equality.~ becomes (Data.Type.Equality.~)
• and something else

There is another way to fix it.

Change ppr t to pprName' Applied t here, which outputs:

data (:~:_0) a_1 b_2 where Refl1_3 :: :~:_0 a_4 a_4
data (:~~:_0) a_1 b_2 = a_1 ~ b_2 => Refl2_3

This line is responsible for the kind of Refl1_3, but I haven't figured out how to fix it yet.

Thanks for looking into this, ak3n!

I don't believe changing the implementation of pprName is what you want here, since that will affect far more than just infix datatypes names, which is what you really care about. I think your second suggestion (changing uses of ppr to pprName' Applied) is on the right track.

As far as why the type of Refl1_3 is pretty-printed incorrectly, I //think// it's due to the pretty-printer for Type. It has a case for ConT here:

pprParendType (ConT c)            = ppr c

I think this should be pprName' Applied c instead. This is consistent with the pretty-printer's treatment for Exp as well:

pprExp _ (ConE c)     = pprName' Applied c

So that feels like the right solution.

Yep, I tried to run the whole test suite and it finished with 417 unexpected failures.

Thanks for pointing out pprParendType! I didn't find it yesterday.

Also, should we change pprParendType (VarT v) = ppr v too? Because of pprExp _ (VarE v) = pprName' Applied v.

assigned to @trac-ak3n

Replying to [ticket:13887#comment:139005 ak3n]:

Yep, I tried to run the whole test suite and it finished with 417 unexpected failures.

Well, chances are that most of the tests were pretty-printing information incorrectly, so this perhaps isn't so surprising. Make sure to eyeball the new test results to confirm that they look correct, and if so, you can run make accept TEST="Test1 Test2 ... TestN" to have GHC accept the new results. (If you're not sure, feel free to consult.)

Also, should we change pprParendType (VarT v) = ppr v too? Because of pprExp _ (VarE v) = pprName' Applied v.

Good question. For consistency, I think we probably should apply this change to VarE as well. That being said, I'm not sure if you can actually have a type-level variable that's syntactically infix, so I don't think you can actually observe this behavior in practice. For instance, I tried this:

λ> $([d| type F ((**) :: * -> * -> *) x y = x ** y |] >>= stringE . pprint)

<interactive>:30:14: error:
    Unexpected type ‘((**) :: * -> * -> *)’
    In the type declaration for ‘F’
    A type declaration should have form
      type F a b c = ...

But in case this ever changes, it would be nice to have Template Haskell Just Work™, so I'd be fine with this change.

Oh, sorry, I was unclear, there were 417 unexpected failures with bad pprName version.

The variant with proper pprName' Applied is fine.

Trac metadata

Trac field	Value
Differential revisions	- → D3717

mentioned in commit ef7fd0ae

closed

changed milestone to %8.2.1

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

I found yet another example of this sort of thing:

{-# LANGUAGE DataKinds #-}
{-# LANGUAGE TemplateHaskell #-}
module Main where

import Data.Proxy
import GHC.Generics
import Language.Haskell.TH

main :: IO ()
main = do
  putStrLn $([t| Proxy  (:*:) |] >>= stringE . pprint)
  putStrLn $([t| Proxy '(:*:) |] >>= stringE . pprint)
  putStrLn $([t| Proxy '(:)   |] >>= stringE . pprint)

Data.Proxy.Proxy (GHC.Generics.:*:)
Data.Proxy.Proxy 'GHC.Generics.:*:
Data.Proxy.Proxy 'GHC.Types.:

Unfortunately, we must have overlooked applying this fix to //promoted// infix constructors.

I've got a patch at D3802 which addresses this.

Trac metadata

Trac field	Value
Differential revisions	D3717 → D3717, D3802

mentioned in commit 7089dc2f

added incorrect runtime result label

added Pnormal label