Split out immutable GadtConstraint #16602
Conversation
This comment was marked as resolved.
This comment was marked as resolved.
| sealed trait GadtConstraintHandling(private var myGadt: GadtConstraint) { | ||
| this: ConstraintHandling => | ||
|
|
||
| def gadt: GadtConstraint = myGadt | ||
| private def gadt_=(g: GadtConstraint) = myGadt = g |
There was a problem hiding this comment.
Maybe we should not make GadtConstraint a private variable in GadtConstraintHandling, but declare two methods for retrieving and updating GADT constraints in this trait, and require the class that extends this trait to provide this two methods, just like what we do in ConstraintHandling (here). Then GadtConstraintHandling will be a stateless mixin for GADT constraint manipulation, and Context.fresh would do the job of GadtConstraint.fresh as long as we put the GadtConstraint in the context.
So that doesn't work, because we want code that's running in an inner context to have changes apply to an outer context. |
This change eagerly allocates a GadtConstraint whenever its state is changed, rather than simply changing the reference(s) in, what's now called,
GadtConstraintHandling, such asconstraint. Previous to a recent change, in order to restore the GADT constraints, a fresh copy was eagerly created in a number of cases, just so it can be used to restore, for the cases in which it must be restored. The recent change exposed the underlying pieces of the GADT constraint and allowed those to be restored as components. Now we still do that, but as a packaged up GadtConstraint that we created once, eagerly. That also makes sure that invariants in components are upheld, like the mappings.