[GHC] #11371: Bogus in-scope set in substitutions

[GHC]

#11371: Bogus in-scope set in substitutions
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                Owner:  niteria
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.2.1
       Component:  Compiler          |              Version:  7.10.3
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #11360            |  Differential Rev(s):  phab:D1792,
       Wiki Page:                    |  phab:D1801, phab:D1802
-------------------------------------+-------------------------------------

Comment (by simonpj):

 There is something odd happening here. '''I do not expect the in-scope set
 to be evaluated at all (that is, it should remain as a thunk) when
 substituting into for-all-free types'''.

 That was why I thought it worth writing `piResultTys` (plural). Consider
 `piResultTys (forall a b. blah) [t1, t2]`.   If we substitute one
 `[a->t1]`, and then `[b->t2]` the first will substitute into `forall b.
 blah` and that will force the in-scope set.  But if we do both at once, we
 won't, provided `blah` has no foralls, which is the wildly common case.

 So how can it possibly cost 10% of all of compile time in this single tiny
 function????  That is utterly bizarre.

 Are you sure ASSERTs are off?  They do a lot of forcing!

 Maybe you can just `pprTrace` what is going on, to see if the in-scope
 sets are being unexpectedly forced. If so, fixing that might fix a LOT of
 things.

 Yes, it might be good to give the in-scope set to `piResultTy(s)` when
 it's available, but please, before you do that, can you get more insight
 into what is happening? Thanks!

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11371#comment:31>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler