Changes between Version 27 and Version 28 of GhcKinds/KindInference


Ignore:
Timestamp:
Jul 25, 2014 12:24:45 PM (5 years ago)
Author:
simonpj
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GhcKinds/KindInference

    v27 v28  
    144144We need two rules, depending on whether or not a CUSK is detected.
    145145
    146 The first rule requires the equations to be fully parametric in its kinds, whereas the second allows non-parametric equations and polymorphic recursion. '''Simon:''' I don't know what this means. '''Richard:''' "Non-parametric" here refers to using a kind-indexed closed type family, where the equations match on the kinds, not just the types. It has a flavor of polymorphic recursion in that the instantiations for the kind variables are not always the same at every usage.
     146The first rule requires the equations to be fully parametric in its kinds, whereas the second allows non-parametric equations and polymorphic recursion. '''Simon:''' I don't know what this means. '''Richard:''' "Non-parametric" here refers to using a kind-indexed closed type family, where the equations match on the kinds, not just the types. It has a flavor of polymorphic recursion in that the instantiations for the kind variables are not always the same at every usage.  '''Simon'' does it really matter? Could we simply nuke this paragraph? Introducing an entirely new term, that then has to be explained, and is not necessary to the exposition, seems to add nothing.
    147147
    148148For closed type families, these rules are ''different'' than the implementation today, because kind inference for closed type families today is ill-specified. See comment:18:ticket:9200.