possible fix for 7056328

[John Rose]

On Jun 22, 2011, at 6:51 AM, Ola Bini wrote:

> On 2011-06-21 00.23, John Rose wrote:
>> I have pushed a possible fix for 7056328 to the mlvm queue.  This bug includes Ola's mysterious "class not found" failure.
>> 
>> 292-users (Charlie, Ola):  Please give it a whirl, to find rough edges, etc.
>> 
>> 292-implementors (Christian, Tom):  This is not a formal review request yet, but please make a pre-review.
>> 
> 
> John, good stuff! This runs nicely for me without putting things on the
> BCP. As far as I can tell, the class-not-found error is gone.

I thought so too, except then I tried "java -Xcomp" and the bug came back.  =8-(

After some exploration I found the place where the wires weren't connected properly, and updated the mlvm patch.

Now I think your bug is really dead.

Please try out the patch.  I've also updated the mlvm patches to the latest bsd-port which (quite awesomely) is b146.

Christian has also be running JRuby regression tests, and those look good too.  (I.e., we didn't break anything.)

As a bonus, running the new stress tests found some bugs that *would* have bit y'all eventually.  The worst:  We were compiling asSpreader adapters into code that spread the arguments in reverse order!  (The interpreter doesn't do this, just the compiled code inside an indy instruction binding.  Which nobody noticed yet.)  The root cause was the up-vs-down numbering of stack slots.  The bad code was plausible but wrong; look for "don't do this" in this diff:
  http://cr.openjdk.java.net/~jrose/7056328/webrev.02/src/share/vm/prims/methodHandleWalk.cpp.udiff.html

Here's some relevant discussion on hs-comp-dev:
  http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2011-June/005818.html
  http://cr.openjdk.java.net/~jrose/7056328/webrev.02/

-- John