JEP 405 Record Patterns and Backwards Compatible Changes

[Brian Goetz]

> I agree, but i think this duality is restricted to the canonical 
> constructor and the (canonical) deconstructor, as we can see with "with".

No, there's a story for how overload selection will be done on ctor/dtor 
pairs from `with`.  Until then, records can prefer the canonical pair; 
it so happens to turn out that the story will be compatible with this 
selection.  We'll get there in steps.

> Constructors get overloading for free from being methods even if 
> usually what the user want is just default values.
> Deconstructors can not re-use the same overloading algorithm as 
> methods (cf refining overloading of 
> reconstruction-records-and-classes.md), so the duality breaks here.

They will have a different set of overload rules, which basically come 
from "reversing the arrows" in the constructor overload restrictions / 
selection.  So they're different in exactly the way you'd expect a dual 
to be different.

> Also i think that named pattern methods are a better mechanism than 
> deconstructors overloading because a pattern methods is not 
> unconditional while a deconstructor is,

Remember, this is not the list for language design discussions.

But the two are not inconsistent; just as users can choose between 
constructors or factories, they can choose between deconstruction and 
named patterns.  Hopefully they will choose consistently; ctors+dtors, 
or factories+named patterns.  But again, not the place to have the "I 
think it should work differently" discussion.