On 2/4/13 10:16 AM, Coleen Phillimore wrote:
On 2/4/2013 11:50 AM, Daniel D. Daugherty wrote:
Thanks for the additional comment. Reply below.
On 2/4/13 8:34 AM, Coleen Phillimore wrote:
On 2/4/2013 10:15 AM, Daniel D. Daugherty wrote:
Adding back the missing aliases and people...
Sorry, I meant to send this re-all.
I missed something major in my earlier review.
The metadata->is_valid() flag should only be enabled in debug mode.
It adds a word to all metadata.
Yes, I'm aware that it adds a word to all meta data. My change to
src/share/vm/oops/metadata.hpp was intentional. Personally, I would
prefer that is_valid() remain in the product bits for at least one
release cycle (JDK8). I'm a fan of sanity checks in product bits so
I actually wouldn't mind if it stay there permanently.
However, if you insist, I'll backout the change to
src/share/vm/oops/metadata.hpp and redo the RedefineClasses()
sanity check code to be appropriately #ifdef'ed.
Thanks, Dan. I do. People are actively trying to reduce metadata
right now, so having this additional word will cause a regression that
will be noticed for a special case. is_valid() is a bit of a hack
too. It relies on metadata being returned and given a non-zero bit
pattern which is only enabled in debug (iirc).
I enabled the setting of the non-zero bit pattern also (iirc).
However, I will back out those changes. I don't want to cause
grief for the metadata reduction effort.
There's also a f1_as_method and f2_as_vfinal_method calls that return
Method* which might make the logic of your if statements simpler in
check_no_old_or_obsolete_entries().
Yes, Karen mentioned one of those in her review and asked for
that cleanup to be tracked in a followup bug. I noticed quite
a few "type cleanups" in that area in HSX-24 and I love them.
I'm presuming that you did that work, but I haven't chased the
history to ground.
Dan
Thanks,
Coleen
