a paper on invokedynamic

[John Rose]

On Nov 16, 2009, at 7:21 AM, Emmanuel Castro wrote:

> I am not sure of my English skills, however I noticed a strange sentence :
> "The four method invocation instructions in the JVM correspond directly to to different kinds of Java method calls".
> I think there is a "to" in surpus in "to to".

In this case your English skills are clearly superior to mine.  Thanks for pointing out the typo!

Item (b) in 2.1 also has a missing article "a" in "a symbolic entity".

More fundamentally, Dean Long has pointed out that my account in 1. of pre-instruction linkage is oversimplified; the fix is in bold:

Member lookup and matching of names and types is performed once only as a symbolic reference is linked, and specifically when it is resolved ([Lindholm99] 5.4.3).  This happens before the first time a particular member reference instruction executes.

And also in 3.:

Like all invoke instructions, invokedynamic can be linked lazily, as late as the first execution of each dynamic call site.  

And 3.1:

Before an unlinked dynamic call site is executed, the JVM calls the site’s bootstrap method.

Perhaps I can get these fixes into the final paper.

-- John
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/mlvm-dev/attachments/20091116/af8b517c/attachment.html