RFR: 5043030 (reflect) unnecessary object creation in reflection (original) (raw)

Joel Borggrén-Franck joel.franck at oracle.com
Fri Jun 13 08:46:50 UTC 2014

Hi Andrej,

On 2014-06-12, Andrej Golovnin wrote:

> I don’t think it is that desirable given that we inflate after 15 calls, > and it adds code to the VM. >

As I already wrote it is just needed for the consistent behavior.

I don't want consistency here. I want the uninflated invocations/field gets to return new boxes for now.

> Lets start with your current tests and the reflection tests in jdk/test > and try to get a coverage metric. With the new build system, supply > “—with-jtreg=some path” to configure and you can run a test suite called > jdklang which includes reflection with > > make TEST=jdklang {CONCURRENCY=[num physical cores/4]} test >

The failed tests:

jdk/lambda/LambdaTranslationTest1.java jdk/lambda/LambdaTranslationTest2.java From my standpoint, this tests have bugs. They make use of Locale sensitive APIs and assume that the current Locale has the same formatting rules as Locale.US. I use a locale with a different formatting rules. This tests fail even without my patch.

If you start a separate thread posting a reproducer I'll file bugs for this.

Before I start to change something I think we should make the decision whether we are going to drop the VM part or not.

I am dropping the VM part.

Again, I have very limited time to work on this. We are looking into other changes in reflection that may or may not change and/or supersede this work. I think this would be a fine contribution to get in to 9, a good baseline to build further work upon, but the scope has to be limited. Given the risk-reward here my bar for accepting this is quite high.

I'm looking forward to a rewritten test and some coverage information.

cheers /Joel

More information about the core-libs-dev mailing list