Records, Intersection type and lambda
B. Blaser
bsrbnd at gmail.com
Wed Jan 6 13:02:14 UTC 2021
Hi Vicente,
I saw you've integrated the initial fix with no additional change
which seems reasonable.
However, did you also try the example I provided below as I'm not
absolutely sure this works?
Otherwise, you may look at the experimental fix I suggested.
Thanks,
Bernard
On Sun, 27 Dec 2020 at 22:31, B. Blaser <bsrbnd at gmail.com> wrote:
>
> Hi Vicente,
>
> On Wed, 23 Dec 2020 at 04:29, Vicente Romero <vicente.romero at oracle.com> wrote:
> >
> > Hi Bernard,
> >
> > what side effects do you see?
>
> By tweaking Remi's example:
>
> public class RecLCE {
> interface Foo { }
> interface F<T extends Foo> {
> void call(T t);
> }
>
> record Bar1() implements Foo { }
> record Bar2() implements Foo { }
> record Pair<P extends Foo>(P p1, P p2) {
>
> static <P extends Foo> Pair<P> of(P p1, P p2) {
> return new Pair<P>(p1, p2);
> }
>
> void forEach(F<P> f) {
> f.call(p1);
> f.call(p2);
> }
> }
>
> static class Hello {
> void m(Foo foo) {
> System.out.println(foo.getClass());
> }
> }
>
> public static void main(String[] args) {
> var pair = Pair.of(new Bar1(), new Bar2());
> pair.forEach(new Hello()::m);
> }
> }
>
> The line you've added is causing the lambda to use the instantiated
> type 'Record' which isn't a subtype of 'Foo', see javap's output:
> Method arguments:
> #57 (LRecLCE$Foo;)V
> #58 REF_invokeStatic
> RecLCE.lambda$main$0:(LRecLCE$Hello;Ljava/lang/Record;)V
> #61 (Ljava/lang/Record;)V
>
> So, it should use the un-instantiated SAM parameter type 'Foo' instead
> which my initial fix was suggesting:
> Method arguments:
> #57 (LRecLCE$Foo;)V
> #58 REF_invokeStatic RecLCE.lambda$main$0:(LRecLCE$Hello;LRecLCE$Foo;)V
> #61 (Ljava/lang/Record;)V
>
> Unfortunately, this very example reveals another issue as the
> instantiated type 'Record' is still not convertible to 'Foo' although
> all was working fine with Remis's initial example:
> Method arguments:
> #55 (Ljava/lang/Object;)V
> #57 REF_invokeStatic
> RecordIntersectionTypeAndLambda.lambda$main$0:(LRecordIntersectionTypeAndLambda$Hello;Ljava/lang/Object;)V
> #60 (Ljava/lang/Record;)V
>
> Referring to 'LambdaMetafactory::metafactory', we see that
> 'instantiatedMethodType' may be the same or a specialization of
> 'samMethodType' suggesting to use the SAM type too when the
> instantiated type is compound like in the experimental fix below
> (langtools:tier1 is OK on jdk14u):
> Method arguments:
> #57 (LRecLCE$Foo;)V
> #58 REF_invokeStatic RecLCE.lambda$main$0:(LRecLCE$Hello;LRecLCE$Foo;)V
> #57 (LRecLCE$Foo;)V
>
> What do you think?
> Bernard
>
> diff --git a/src/jdk.compiler/share/classes/com/sun/tools/javac/comp/LambdaToMethod.java
> b/src/jdk.compiler/share/classes/com/sun/tools/javac/comp/LambdaToMethod.java
> --- a/src/jdk.compiler/share/classes/com/sun/tools/javac/comp/LambdaToMethod.java
> +++ b/src/jdk.compiler/share/classes/com/sun/tools/javac/comp/LambdaToMethod.java
> @@ -438,6 +438,7 @@
> List<JCExpression> indy_args =
> translate(syntheticInits.toList(), localContext.prev);
>
> //convert to an invokedynamic call
> + localContext.useSAM = tree.useSAM;
> result = makeMetafactoryIndyCall(context, sym.asHandle(), indy_args);
> }
>
> @@ -888,6 +889,7 @@
> private final ListBuffer<JCVariableDecl> params = new ListBuffer<>();
>
> private JCExpression receiverExpression = null;
> + private boolean useSAM = false;
>
> MemberReferenceToLambda(JCMemberReference tree,
> ReferenceTranslationContext localContext, Symbol owner) {
> this.tree = tree;
> @@ -911,6 +913,7 @@
> slam.target = tree.target;
> slam.type = tree.type;
> slam.pos = tree.pos;
> + slam.useSAM = useSAM;
> return slam;
> } finally {
> make.at(prevPos);
> @@ -954,6 +957,7 @@
>
> // Failsafe -- assure match-up
> boolean checkForIntersection = tree.varargsElement !=
> null || implSize == descPTypes.size();
> + useSAM = checkForIntersection &&
> localContext.interfaceParameterIsIntersectionOrUnionType();
>
> // Use parameter types of the implementation method
> unless the unerased
> // SAM parameter type is an intersection type, in that case use the
> @@ -963,18 +967,7 @@
> // are used as pointers to the current parameter type information
> // and are thus not usable afterwards.
> for (int i = 0; implPTypes.nonEmpty() && i < last; ++i) {
> - // By default use the implementation method parmeter type
> - Type parmType = implPTypes.head;
> - // If the unerased parameter type is a type variable whose
> - // bound is an intersection (eg. <T extends A & B>) then
> - // use the SAM parameter type
> - if (checkForIntersection && descPTypes.head.getKind()
> == TypeKind.TYPEVAR) {
> - TypeVar tv = (TypeVar) descPTypes.head;
> - if (tv.getUpperBound().getKind() ==
> TypeKind.INTERSECTION) {
> - parmType = samPTypes.head;
> - }
> - }
> - addParameter("x$" + i, parmType, true);
> + addParameter("x$" + i, useSAM ? samPTypes.head :
> implPTypes.head, true);
>
> // Advance to the next parameter
> implPTypes = implPTypes.tail;
> @@ -1094,7 +1087,7 @@
> List<LoadableConstant> staticArgs = List.of(
> typeToMethodType(samSym.type),
> refSym.asHandle(),
> - typeToMethodType(tree.getDescriptorType(types)));
> + typeToMethodType(context.useSAM ? samSym.type :
> tree.getDescriptorType(types)));
>
> //computed indy arg types
> ListBuffer<Type> indy_args_types = new ListBuffer<>();
> @@ -1826,6 +1819,8 @@
> /** list of methods to be bridged by the meta-factory */
> final List<Symbol> bridges;
>
> + boolean useSAM = false;
> +
> TranslationContext(T tree) {
> this.tree = tree;
> this.owner = owner(true);
> diff --git a/src/jdk.compiler/share/classes/com/sun/tools/javac/tree/JCTree.java
> b/src/jdk.compiler/share/classes/com/sun/tools/javac/tree/JCTree.java
> --- a/src/jdk.compiler/share/classes/com/sun/tools/javac/tree/JCTree.java
> +++ b/src/jdk.compiler/share/classes/com/sun/tools/javac/tree/JCTree.java
> @@ -1892,6 +1892,7 @@
> public JCTree body;
> public boolean canCompleteNormally = true;
> public ParameterKind paramKind;
> + public boolean useSAM = false;
>
> public JCLambda(List<JCVariableDecl> params,
> JCTree body) {
More information about the compiler-dev
mailing list