Compile-time type hierarchy information in pattern switch

[Remi Forax]

----- Mail original -----
> De: "Brian Goetz" <brian.goetz at oracle.com>
> À: "Tagir Valeev" <amaembo at gmail.com>, "amber-spec-experts" <amber-spec-experts at openjdk.java.net>
> Envoyé: Jeudi 5 Avril 2018 21:44:30
> Objet: Re: Compile-time type hierarchy information in pattern switch

>> Is it too harsh to reject the whole class if the assumptions on class
>> hierarchy which were necessary to compile the switch statements used
>> in the class are not valid at runtime?
> 
> That is one of the questions!  And the other question is: is this too
> expensive to do this check at runtime, given that it will fail so
> infrequently.
> 
> If we can detect it cheaply enough, though, we can also repair the
> situation and fall back to linear testing of patterns.  This seems
> better (we can execute the statement the user wrote) than failing. My
> real question is can I punt on trying to detect it, and still optimize
> the common cases  down to O(1) dispatch....

the way to detect it is to use the DAG of the supertypes (lazily constructed*), from the last to the first case, the idea is to propagate the index of down to the super types, if during the propagation, you find a supertype which is also a case and with an index lower that the currently propagated, then it's a failure. 

Rémi

* you do not have to actually create the DAG, just be able to traverse it from the subtype to the supertypes.