On 11/10/2013 15:30, Aleksey Shipilev wrote:
:
While I agree with the fix, profiling is the no-go way to confirm the
performance regression and the fix, especially with the broken test you
have in [3]. You need to build targeted benchmarks, see e.g. jmh:
http://openjdk.java.net/projects/code-tools/jmh/
[1] http://cr.openjdk.java.net/~aefimov/8008733/webrev.00/
Looks good. What a mess is that code anyway!
I don't know this code but I see that m_dtmManager is initialized in the
init method (invoked by the constructor). Also there doesn't seem to be
serializable involved. So the change looks okay.
Do you have a sponsor to help you get this in? Also will the benchmarks
or tests to verify the performance be going anywhere (this might be a
question for Aleksey as ideally we should be building up a suite a good
micro benchmarks as we fix or improve things).
-Alan.
