[GHC] #8543: `Coercible` ought to work for recursive newtypes

[GHC]

#8543: `Coercible` ought to work for recursive newtypes
--------------------------------------------+------------------------------
        Reporter:  nomeata                  |            Owner:  nomeata
            Type:  task                     |           Status:  closed
        Priority:  normal                   |        Milestone:
       Component:  Compiler (Type checker)  |          Version:  7.6.3
      Resolution:  duplicate                |         Keywords:
Operating System:  Unknown/Multiple         |     Architecture:
 Type of failure:  None/Unknown             |  Unknown/Multiple
       Test Case:                           |       Difficulty:  Unknown
        Blocking:                           |       Blocked By:
                                            |  Related Tickets:  #8503
--------------------------------------------+------------------------------
Changes (by simonpj):

 * cc: dimitris@…, sweirich@…, diatchki (added)
 * status:  new => closed
 * resolution:   => duplicate


Old description:

> At least on a best-effort basis.
>
> The approach (developed by SPJ and me) is to
>
> A) Add a feature to the constraint solver to prevent recursive
> dictionaries for specially marked instances (for now only used for
> `Coercible`). Rationale: Such dictionaries (which are fine for most
> classes like `Show`) would make `coerce` diverge. Implementation: Use the
> depth counter and do not use lower depths to solver constraints with a
> higher depth. This is of course a very conservative approximation, but
> should be sufficient.
>
> B) Use the regular constraint depth bound to prevent looping at compile
> time. In order for that to be more useful, count constraint solving and
> type function resolving separately.

New description:

 The approach (developed by SPJ and me) is to

 A) Add a feature to the constraint solver to prevent recursive
 dictionaries for specially marked instances (for now only used for
 `Coercible`). Rationale: Such dictionaries (which are fine for most
 classes like `Show`) would make `coerce` diverge. Implementation: Use the
 depth counter and do not use lower depths to solver constraints with a
 higher depth. This is of course a very conservative approximation, but
 should be sufficient.

 B) Use the regular constraint depth bound to prevent looping at compile
 time. In order for that to be more useful, count constraint solving and
 type function resolving separately.

--

Comment:

 I think it's better to keep this ticket with #8503, which has all the
 context. I've moved the comment there.

 Simon

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8543#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler