switch statement and lambda

[Brian Goetz]

I see why this is tempting, but I am going to suggest we wait.  As part of Valhalla, we would like for `void` to become a real type some day; that will require evaluating all the places in the JLS where we treat statement and expressions differently, or make exceptions like this.  Until we’ve completed this analysis, I’m reluctant to add more special cases. 


> On Apr 9, 2019, at 3:10 PM, forax at univ-mlv.fr wrote:
> 
> 
> 
> De: "Gavin Bierman" <gavin.bierman at oracle.com>
> À: "Remi Forax" <forax at univ-mlv.fr>
> Cc: "amber-spec-experts" <amber-spec-experts at openjdk.java.net>
> Envoyé: Mardi 9 Avril 2019 19:28:57
> Objet: Re: switch statement and lambda
> 
> On 6 Apr 2019, at 21:17, Remi Forax <forax at univ-mlv.fr <mailto:forax at univ-mlv.fr>> wrote:
> 
> Currently this code doesn't compile
>  IntConsumer c = x -> switch(x) { default -> System.out.println(x); };
> 
> I believe it should because this is the basic pattern for supporting the actor model,
> you consume a message and do a side effect* depending on the type of the message,
> translated in Java, you want a lambda that takes a message as parameter, calls a switch to do the pattern matching and return void.
> 
> I understand, although this is actually to do with the way lambda expressions are typed, rather than the switch expression. In JLS 15.27.3 "Type of a Lambda Expression”, there is a special case:
> 
> • If the function type's result is void, the lambda body is either a statement expression (§14.8) or a void-compatible block.
> 
> Which means that the following code typechecks:
> 
> IntConsumer ic = x -> System.out.println(x);
> 
> but it breaks as soon as we nest the statement expression, e.g.
> 
> IntConsumer ic2 = x -> true ? System.out.println(x) : System.out.println(x); // Compilation error: target-type for conditional expression cannot be void
> This is what is happening in your example. So to deal with this we’d either have to make typechecking of lambdas smarter - either by replacing the typing rule for lambdas above with something more compositional, or by making void a first-class type, or we could perhaps add a pattern-matching form of lambda, which has a void-aware typing rule. I’m not sure about any of these options for now.
> 
> yes,
> i'm proposing to create a special case for a switch inside a lambda expression for the same reason we have a special treatment for methods.
> 
> By example, this does not compile
> https://github.com/fora <https://github.com/fora>x/loom-fiber/blob/cea7b86c26c2e86b00fb72e5098a37983e8b6441/src/main/java/fr.umlv.loom/fr/umlv/loom/actor/CounterStringActorExprSwitchDemo.java#L17
> while this compiles
> https://github.com/forax/loom-fiber/blob/6d8a0a6ba870580d43988cd0d507d19d57653d62/src/main/java/fr.umlv.loom/fr/umlv/loom/actor/CounterStringActorExprSwitchDemo.java#L17 <https://github.com/forax/loom-fiber/blob/6d8a0a6ba870580d43988cd0d507d19d57653d62/src/main/java/fr.umlv.loom/fr/umlv/loom/actor/CounterStringActorExprSwitchDemo.java#L17>
> 
> 
> 
> Cheers,
> Gavin
> 
> cheers,
> Rémi

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.java.net/pipermail/amber-spec-experts/attachments/20190409/30a9f746/attachment-0001.html>