On Apr 5, 2009, at 1:10 AM, Charles Oliver Nutter wrote: > - follow the instructions in the MLVM repo, with the following > modifications: > -- clone bsd-port instead of openjdk directly I changed the README to reflect this reality. > -- remove "testable" from your guards to get invokedynamic support > along > with method handles (though John says there's still some bugs to fix) The indy stuff is now marked testable. (Not stable, just testable.) > -- add "projects" to your guards to also apply John's demo projects > for > anonymous classes and method handles (and I believe the invokedynamic > code in the method handle project is function too, right John?) Partially. Some tests pass, some fail. I'm working in it... > -- use the *second* set of instructions for applying the patches, > with > all the raw hg queue commands > -- once everything has been applied, follow Stephen Bannasch's > instructions for getting it to build as normal > -- when running with opcode 186 (new-style invokedynamic) you'll > need to > turn off verification with -Xverify:none I committed a patch to the old plugin-based verifier in the jdk repo, which should fix that.
> With this process I managed to get JRuby to the point of producing an > invokedynamic error (missing bootstrap), which to me is a great > success. > I hope to have things re-wired in the next day or two and will > report back. That's great! I think the road is clear for your next step. Upcoming problems are likely to be either something low-level in the new JVM code, or else a missing piece in the java.dyn.MethodHandles API. Thanks for working on this! -- John _______________________________________________ mlvm-dev mailing list mlvm-dev@openjdk.java.net http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
