Lambda vs anon class optimization effect on CHA

[John Rose]

On Aug 10, 2016, at 12:02 PM, Aleksey Shipilev <aleksey.shipilev at gmail.com> wrote:

> I would venture to guess that we can wire up CHA to look at what classes
> have been _initialized_ rather than what classes were _resolved_, but
> there is no telling how much bugs would that expose.

I agree with both of your points.

Even more sharply:  A class which is never _instantiated_ does not contribute
to the concrete types encountered at virtual call sites.  It can be treated as if it
had been marked ACC_ABSTRACT, until the first instantiation.  (CHA makes
a distinction between abstract and non-abstract.)

I agree that changing this code might cause bugs.  Here's where I held back, FTR:

http://hg.openjdk.java.net/jdk9/jdk9/hotspot/file/tip/src/share/vm/code/dependencies.cpp#l1440

In any case, since loaded classes can transition to initialized, and
then can be instantiated (if not abstract), there will have to be new dependency
tracking to guard against these transitions, just as we now guard against
new subclasses getting loaded.

— John
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20160812/05bd8949/attachment.html>