Hi Martin,
On 11/19/2014 01:42 AM, Martin Buchholz wrote:
Hi Joe, Peter, Paul
This is the followup on thread safety I promised Peter.
Looks good.
I made the WildcardTypeImpl.[upperBoundASTs, lowerBoundASTs] and
TypeVariableImpl.boundASTs fields volatile in my version of patch
(instead of final):
http://cr.openjdk.java.net/~plevart/jdk9-dev/GenericsReflectionRaces/webrev.01/
...so that after the structure they point to has been parsed into bound
types, they can be thrown away. The comments indicate that possibility
already, but the implementor was afraid to do it because of possible
races. I think I got it right here.
As I said before, I'm willing to take the risk to use fields written
via a race if we have a reflective test proving it correct, and the
bug contains the 0.1 version of such a test.
https://bugs.openjdk.java.net/browse/JDK-8065172
http://cr.openjdk.java.net/~martin/webrevs/openjdk9/core-reflection-more-safety/
Inspired by your 0.1 version of a test, I created version 0.2:
http://cr.openjdk.java.net/~plevart/misc/SunReflectDataRaces/
...which uses some of your code, but rewritten with heavy use of
Stream(s). When run with your proposed patch applied (and all previous
patches that are already commited), it finds 20 possible data races. I
haven't yet studied any of them.
Regards, Peter
