[GHC] #7989: "No constructor has all these fields" message can be improved

[GHC]

#7989: "No constructor has all these fields" message can be improved
---------------------------------+------------------------------------------
    Reporter:  akio              |       Owner:                  
        Type:  feature request   |      Status:  patch           
    Priority:  normal            |   Milestone:                  
   Component:  Compiler          |     Version:  7.7             
    Keywords:                    |          Os:  Unknown/Multiple
Architecture:  Unknown/Multiple  |     Failure:  Other           
  Difficulty:  Unknown           |    Testcase:                  
   Blockedby:                    |    Blocking:                  
     Related:                    |  
---------------------------------+------------------------------------------

Comment(by akio):

 I updated the patch.

 Replying to [comment:7 simonpj]:
 > OK.... It's not easy to understand that code!  Maybe it would be worth a
 few more named intermediate values, with types, and a comment?

 I did this.

 >
 > You could produce a very big set when a very small set exists, couldn't
 you?

 Yes, the size of the set is only bounded by the number of constructors.

 >
 > Could you ameliorate that as follows?  Use `T` to mean the data type of
 the first field in `rbinds`.
 >  * Sort the `membership` list in order of increasing number of `True`
 fields.  So the earlier fields `f` in the list have few `T` constructors
 that have field `f`.
 >  * Now a field with a different type will be right at the front of the
 list.
 >  * And the `(&&)` process will terminate more quickly because of all the
 `False` items.

 I implemented this. However I'm afraid that the worst-case behavior of the
 algorithm hasn't changed much. Given a datatype with 2N constructors and
 2N fields, it can find a set of N fields even though there is one with
 just 2 fields. (an example is when the constructor ```i``` has the field
 ```j``` iff ```mod (i + j) (2*N) < N```)

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7989#comment:8>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler