[GHC] #10330: Better Template Haskell error message locations

[GHC]

#10330: Better Template Haskell error message locations
-------------------------------------+-------------------------------------
              Reporter:  simonpj     |             Owner:
                  Type:  bug         |            Status:  new
              Priority:  normal      |         Milestone:
             Component:  Template    |           Version:  7.10.1
  Haskell                            |  Operating System:  Unknown/Multiple
              Keywords:              |   Type of failure:  None/Unknown
          Architecture:              |        Blocked By:
  Unknown/Multiple                   |   Related Tickets:
             Test Case:              |
              Blocking:              |
Differential Revisions:              |
-------------------------------------+-------------------------------------
 Consider this TH splice
 {{{
 f x = $( [| let g x = (True && 'x', x) in g True |] )
 }}}
 or more generally
 {{{
 f x = $( g [| x |])
 }}}
 Two problems
  1. If there is a type error in the spliced-in code, there is no record of
 the splice in the error message:
 {{{
 TH.hs:5:10:
     Couldn't match expected type ‘Bool’ with actual type ‘Char’
     In the second argument of ‘(&&)’, namely ‘'x'’
     In the expression: (True && 'x')
     In the expression: ((True && 'x'), x_a2qf)
 }}}
  I'd like to see
 {{{
 TH.hs:5:10:
     Couldn't match expected type ‘Bool’ with actual type ‘Char’
     In the second argument of ‘(&&)’, namely ‘'x'’
     In the expression: (True && 'x')
     In the splice: $( [| let g x = ... in g True |] )   <------- NB
 }}}

  2. (Less important, and harder to fix.)  The error location (column 10)
 is that of the splice (column 10), not where the reported error really is
 (column 31).

  Fixing this would mean round-tripping source location info through TH
 syntax.  Moreover, since the constructed code may be combined from quoted
 snippets spread over some TH libraries, it's far from clear that the
 location info would help.

 Thoughts about fixing (1).  This can only arise for untyped splices.  (For
 typed splices we can't get type errors in spliced code.) So the renamer
 should produce a syntax tree with a variant of `HsPar` saying "here is
 where I spliced in an expression".  Same think for patterns and types; for
 declarations it would be a bit more fiddly.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10330>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler