Mandy Chung wrote:
Alan, David,
I revise the fix including the following:
1. Use the lazy initialization holder class idiom to initialize the bcp
variable to simplify the synchronization.
2. Group the initialization of system properties in a new private method
System.initializeSystemProperties so that it can save a copy for internal
implementation use to address the deadlock issue and remove the
properties that are intended for private internal use from public access
before initializing the System.props. This removes the 2 calls to
System.setProperties.
3. Reorder sun.net.* in FILES_java.gmk
The new webrev is at:
http://cr.openjdk.java.net/~mchung/6977738/webrev.01/
This looks much better (and thanks for doing the various
drive-by-clean-ups).
I assume BootClassPathHolder should be private, and that the bcp field
can be final.
Minor nit but but the IllegalStateExceptions thrown by
VM.getSavedProperty capitalizes the first letter of the message whereas
VM.saveAndRemoveProperties doesn't.
I wonder if it would be worth seeing if saveAndRemoveProperties could
grab the value of the java.lang.Integer.IntegerCache.high property. That
would allow us to be handle it more consistently with these other
properties that we save during initialization. The cache initialization
could then use VM.getSavedProperty to get the value. Might be worth
thinking about but you would need to be careful to ensure that
Integer.valueOf isn't called as otherwise isn't not going to work.
-Alan.
