Re: [GHC] #13199: TH-spliced class instances are pretty-printed incorrectly post-#3384

28 Jan 2017

      #13199: TH-spliced class instances are pretty-printed incorrectly post-#3384
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  alanz
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 Ah, so GHC will only print parentheses around types is an explicit
 `HsParTy` is used? In that case, I believe I know what the real culprit
 is. If you dig down deep enough to the code that `runMetaD` is running,
 you'll eventually come to
 [http://git.haskell.org/ghc.git/blob/de78ee6fb77e7505160ab23e6e1b4e66dc87f698...
 the part] in `Convert` that coverts a Template Haskell `InstanceD` to a
 `ClsInstDecl`:

 {{{#!hs
 cvtDec (InstanceD o ctxt ty decs)
   = do  { let doc = text "an instance declaration"
         ; (binds', sigs', fams', ats', adts') <- cvt_ci_decs doc decs
         ; unless (null fams') (failWith (mkBadDecMsg doc fams'))
         ; ctxt' <- cvtContext ctxt
         ; L loc ty' <- cvtType ty
         ; let inst_ty' = mkHsQualTy ctxt loc ctxt' $ L loc ty'
         ; returnJustL $ InstD $ ClsInstD $
           ClsInstDecl { cid_poly_ty = mkLHsSigType inst_ty'
                       , cid_binds = binds'
                       , cid_sigs = Hs.mkClassOpSigs sigs'
                       , cid_tyfam_insts = ats', cid_datafam_insts = adts'
                       , cid_overlap_mode = fmap (L loc . overlap) o } }
 }}}

 In particular, if you trace the value of `ty'` when you run the program
 above, it'll give you `instance C a_a4m5 Maybe b_a4m6`. That's because the
 original Template Haskell AST for this is:

 {{{
 λ> import Language.Haskell.TH
 λ> putStrLn $([d| instance C a (Maybe b) |] >>= stringE . show)
 [InstanceD Nothing [] (AppT (AppT (ConT Bug.C) (VarT
 a_6989586621679027494)) (AppT (ConT GHC.Base.Maybe) (VarT
 b_6989586621679027495))) []]
 }}}

 That is, there are no AST nodes to indicate parentheses. That's because
 when we originally quoted this declaration earlier, `repTy` (located
 [http://git.haskell.org/ghc.git/blob/de78ee6fb77e7505160ab23e6e1b4e66dc87f698...
 here] in `DsMeta`) turns the `HsType` into a Template Haskell `Type`. And
 `repTy` has this as one of its cases:

 {{{#!hs
 repTy (HsParTy t)           = repLTy t
 }}}

 `repTy` strips away all parentheses, and this is by design, according to
 [http://git.haskell.org/ghc.git/blob/de78ee6fb77e7505160ab23e6e1b4e66dc87f698...
 /template-haskell/Language/Haskell/TH/Syntax.hs#l1441 this note]:

 {{{#!hs
 --- * Quoted expressions such as
 ---
 ---   > [| a * b + c |] :: Q Exp
 ---   > [p| a : b : c |] :: Q Pat
 ---   > [t| T + T |] :: Q Type
 ---
 ---   will never contain 'UInfixE', 'UInfixP', 'UInfixT', 'InfixT',
 'ParensE',
 ---   'ParensP', or 'ParensT' constructors.
 }}}

 So I believe the proper fix here is //not// to make `repTy` turn an
 `HsParTy` into a `ParensT`, but rather to change  `cvtType` so that it
 inserts `HsParTy`s appropriately as it converts from a TH AST back to an
 `HsType`, right?

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13199#comment:5
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler