TH treats non-functions in function position inconsistently

[heisenbug]

I start GHCi like this:

~/bin/ghci -XTemplateHaskell

My aim is to use TH to rewrite certain application-looking (syntactically) constructs to semantically valid haskell code, that passes the type-checker (c.f. Conor McBride's "idiom brackets").

But I already fail at very simple things, because GHC seems to type-check the TH quotation's innards...

This works:

Prelude> :t [| id id |]
[| id id |]
  :: Language.Haskell.TH.Syntax.Q Language.Haskell.TH.Syntax.Exp

This is also accepted, though it is clearly not typeable:

Prelude> :t [| 1 1 |]
[| 1 1 |]
  :: Language.Haskell.TH.Syntax.Q Language.Haskell.TH.Syntax.Exp

Encouraged by this I try:

Prelude> :t [| False True |]
<interactive>:1:4:
    Couldn't match expected type `Bool -> t0' with actual type `Bool'
    The function `False' is applied to one argument,
    but its type `Bool' has none
    In the Template Haskell quotation [| False True |]
    In the expression: [| False True |]

Bummer! Somehow the type-checker does get into business with "expressions" inside quotations.

I believe this is a bug, and a quotation should be built for [| False True |] in spite of False True not being typeable.

I know that using quasi-quotations I can build my own parser and rewrite engine around this, but that is a tad heavyweight for me and I'd like to understand the root of the above inconsistency and why the type checker gets active in some cases above, but not in others.

[simonpj]

Read: ​http://hackage.haskell.org/trac/ghc/blog/Template%20Haskell%20Proposal which tackles precisely the question you raise.

I guess you are another customer for this work, which is (still) pending.

Simon

[Krzysztof Gogolewski <krz.gogolewski@…>]

In e595c81ee0951f17d34b918123dfa7711710fe89/testsuite:

Add tests for new TH #4124, #4128, #4170, #4364, #6062

[monoidal]

This is now fixed thanks to new TH. Woo hoo!