hg: lambda/lambda/jdk: 3 new changesets (original) (raw)

Brian Goetz brian.goetz at oracle.com
Sun Apr 22 10:16:30 PDT 2012

Also, there's room for a special case where there is exactly one captured parameter, and the implementation method handle is an instance method. This means the sole parameter is the receiver, and we can used bindTo instead of insertArgs, which is a shorter capture path.

The "translate by method handle proxies" is largely a proof-of-concept to demonstrate that we can change implementation strategies on a run-to-run basis while holding the bytecode stable. As a translation strategy, it is not great. It will be slower (uses dynamic proxy in the implementation), doesn't yet have a path to supporting serialization, and the MHProxy implementation has bugs (filed CR 7163223 to track these.)

The good news is that, modulo the above bug report, the approach works. We have two metafactory implementations, choosable by setting a system property, and I have tests that generate thousands of combinations of argument and return types, generate source code, compile it, assert that the compiler yields the right diagnostics, for the classes that pass the compiler loads them, runs them, and asserts that we get the right answer. And both metafactories pass these tests, modulo bug 7163223.

(To run these tests: go to the lambda repo, cd to jdk/combo-tests, drop the appropriate version of the testng JAR into the lib/ subdirectory, and do "ant test." We're in the process of integrating testng support into jtreg, at which point these tests will move to the main test area.)

What this experiment with MH proxies does point us towards is implementing the second realistic metafactory implementation, the one that generates a single wrapper class per SAM type, whose constructor takes a MethodHandle, and whose SAM methods invoke that handle. That's coming soon, but first I am working on some more tests regarding bridge methods.

(Lots more details in the document I just posted to this list.)

On 4/22/2012 6:23 AM, Rémi Forax wrote:

On 04/21/2012 11:58 PM, brian.goetz at oracle.com wrote:

Changeset: 59aa44ba1555 Author: briangoetz Date: 2012-04-21 17:56 -0400 URL: http://hg.openjdk.java.net/lambda/lambda/jdk/rev/59aa44ba1555

Minor improvements in combo-test framework and lambda tests ! combo-tests/build.xml ! combo-tests/tests/tools/javac/combo/JavacTemplateTestBase.java ! combo-tests/tests/tools/javac/combo/Template.java ! combo-tests/tests/tools/javac/lambda/LambdaConversionTest.java Changeset: b166fa7adaea Author: briangoetz Date: 2012-04-21 17:56 -0400 URL: http://hg.openjdk.java.net/lambda/lambda/jdk/rev/b166fa7adaea Minor improvements in combo-test framework and lambda tests ! test-ng/build.xml ! test-ng/tests/org/openjdk/tests/java/util/functions/MappersTest.java Changeset: d0e63cae6a1c Author: briangoetz Date: 2012-04-21 17:57 -0400 URL: http://hg.openjdk.java.net/lambda/lambda/jdk/rev/d0e63cae6a1c Allow metafactory option to be choosable at runtime via lambda.metafactory system property; first cut at second translation strategy (method handle proxies) ! .hgignore ! src/share/classes/java/lang/invoke/InnerClassGenerator.java ! src/share/classes/java/lang/invoke/LambdaMetafactory.java + src/share/classes/java/lang/invoke/MagicLambdaImpl.java ! src/share/classes/java/lang/invoke/MethodHandleProxies.java + src/share/classes/java/lang/invoke/MethodHandleProxyLambdaMetafactory.java

Hi Brian, I think you don't need the MhMetafactoryCallSite because the arguments of MethodHandle.asInterfaceInstance are known during the bootstrap. The idea is that instead of the method factory,() you can bind the method bindTo itself. I will try to come with a code, it will be more clear :) Rémi

More information about the lambda-dev mailing list