AnnotationType alternative patch (original) (raw)

Peter Levart peter.levart at gmail.com
Fri Feb 1 22:23:01 UTC 2013

Hello Joel and the list,

While revisiting the patch that I proposed a couple of months ago:

http://mail.openjdk.java.net/pipermail/core-libs-dev/2012-December/012630.html

...and which also includes some other things besides the main fix for scalability and dead-locks, I tried to see what It would take to modify the annotation parsing logic in a way that would make construction of AnnotationType instance infinite-recursion-free without resorting to publishing the reference to the instance in the middle of the constructor. As it turns out the changes are quite simple:

http://freeweb.t-2.net/peter_levart/jdk8-tl/AnnotationType/webrev.01/index.html

I added an overloaded method to AnnotationParser which only parses select annotations (quickly skipping the rest of raw annotation bytes) and implicitly considers those annotations as having RUNTIME retention. This method is used inside the constructor of AnnotationType just for parsing the @Retention and @Inherited meta annotations for annotation class.

This way the infinite recursion is broken and there's no need to publish the half-constructed instance just to prevent it. Everything else is easy: the AnnotationType class is now immutable (all fields final) and can be published to other threads via data race (removed synchronized from AnnotationType.getInstance()). Scalability is infinite and dead-locks are eliminated.

This patch also fixes inconsistent behavior when we have two mutually recursive RUNTIME annotations and one's scope is changed to CLASS and separately compiled - the situation that I described in above-mentioned message.

Regards, Peter

More information about the core-libs-dev mailing list