unboxedTupleNameDegree_maybe doesn't work with GHC 9.10

[RyanGlScott]

If you run this program with GHC 9.8 or earlier:

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

import Language.Haskell.TH.Desugar

main :: IO ()
main = print $ unboxedTupleNameDegree_maybe ''(#,#)

You will get this output, as expected:

$ runghc-9.8 Bug.hs
Just 2

With GHC 9.10, however, you will instead get:

$ runghc-9.10 Bug.hs
Nothing

Ultimately, this happens due to the fact that the names of unboxed tuple type constructors have changed in GHC 9.10. In GHC 9.8, we have:

λ> :set -XUnboxedTuples -XTemplateHaskell
λ> ''(#,#)
GHC.Prim.(#,#)

But in GHC 9.10, we have:

λ> :set -XUnboxedTuples -XTemplateHaskell
λ> ''(#,#)
GHC.Types.Tuple2#

The code for parsing unboxed tuple type constructors, however, assumes that only the former name will be used. We need to update this code to account for the possibility of names like Tuple2#.

[RyanGlScott]

We should also check for things like Sum2# in unboxedSumNameDegree_maybe.

[RyanGlScott]

Sigh... my testing in #215 wasn't thorough enough. While the example in #213 (comment) is now fixed, making a small tweak to it causes it to fail again:

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

import Language.Haskell.TH
import Language.Haskell.TH.Desugar

main :: IO ()
main = print $ unboxedTupleNameDegree_maybe $ unboxedTupleTypeName 2

$ runghc-9.8 Bug.hs
Just 2
$ runghc-9.10 Bug.hs
Nothing

This one is really weird, as it turns out that unboxedTupleTypeName 2 returns a completely different Name than ''(#,#) does:

λ> unboxedTupleTypeName 2
GHC.Prim.Tuple2#
λ> ''(#,#)
GHC.Types.Tuple2#

See GHC#24750. This is unfortunate from a th-desugar perspective because we desugar uses of UnboxedTupleT <N> to ConT (unboxedTupleTypeName <N>), so it is particularly important to get the unboxedTupleTypeName case correct.

My hope is that we can fix GHC#24750 upstream so that this example Just Works™. If not, we'll need to add a special case for the (non-existent) GHC.Prim.Tuple<N># family of Names.

[RyanGlScott]

The fix for GHC#24750 is present in GHC 9.10.1, which means that the program in #213 (comment) now works as expected. I've added that program (among others) as regression tests in 8179363.