[GHC] #15783: Quoting an internal variable causes an error when splicing

[GHC]

#15783: Quoting an internal variable causes an error when splicing
-------------------------------------+-------------------------------------
        Reporter:  mpickering        |                Owner:  (none)
            Type:  bug               |               Status:  patch
        Priority:  normal            |            Milestone:
       Component:  Template Haskell  |              Version:  8.6.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):  Phab:D5248
       Wiki Page:                    |
-------------------------------------+-------------------------------------
Changes (by RyanGlScott):

 * status:  new => patch
 * differential:   => Phab:D5248


Comment:

 I figured out what's causing this. See Phab:D5248 for a fix.

 It turns out that occurrence analysis was dropping the binding for `d`
 during desugaring (this explains why this bug didn't surface in GHCi, as
 occurrence analysis does not run there). So why was occurrence analysis
 dropping bindings for top-level things referenced from typed TH quotes,
 but not from untyped TH quotes? It turns out that two functions called
 `checkCrossStageLifting` are to blame.

 ...No seriously, there are two completely separate functions named
 `checkCrossStageLifting` in GHC. One exists in `RnSplice`, and only
 handles untyped quotes, and the other exists in `TcExpr`, and only handles
 typed quotes. And wouldn't you know it, `RnSplice.checkCrossStageLifting`
 was doing something that `TcExpr.checkCrossStageLifting` wasn't doing. In
 particular,
 [http://git.haskell.org/ghc.git/blob/879db5595208fb665ff1a0a2b12b9921d3efae0e:/compiler/rename/RnSplice.hs#l800
 these lines] of `RnSplice.checkCrossStageLifting` are crucial:

 {{{#!hs
   | isTopLevel top_lvl
         -- Top-level identifiers in this module,
         -- (which have External Names)
         -- are just like the imported case:
         -- no need for the 'lifting' treatment
         -- E.g.  this is fine:
         --   f x = x
         --   g y = [| f 3 |]
   = when (isExternalName name) (keepAlive name)
     -- See Note [Keeping things alive for Template Haskell]
 }}}

 That call to `keepAlive` ensures that the binding for `f` (from the
 example in the comments) doesn't get discarded during occurrence analysis.
 `TcExpr.checkCrossStageLifting` wasn't doing anything like this, which
 explains why this bug exists.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15783#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler