Webrev updated at https://cr.openjdk.java.net/~weijun/8210476/webrev.01/
The only change is that there is a single Cleaner now for the whole PRNG class. Otherwise, each will maintain its own thread. Thanks Max > On Nov 11, 2018, at 11:30 PM, Weijun Wang <weijun.w...@oracle.com> wrote: > > Please take a review at > > https://cr.openjdk.java.net/~weijun/8210476/webrev.00/ > > Before this fix, every PRNG::nextBytes calls all of CryptAcquireContext, > CryptGenRandom, and CryptReleaseContext. Now, CryptAcquireContext is called > once in PRNG::new, and CryptReleaseContext is called by a Cleaner, and > nextBytes only calls CryptGenRandom. > > I haven't read about thread-safety in any MS document, the current > Windows-PRNG service is marked ThreadSafe=true (in SunMSCAPI.java). If we > cannot be really sure, we can change it to false. > > I've downloaded nearly 1000 Mach5 runs of this test, the enhancement is so > good that I adjusted the test to be stricter. > > Before After > ------ ----- > Count 897 74 > Min 0.38 0.008 > Ave 0.97 0.011 > Max 5.81 0.021 > > Please advise me if the following usage of Cleaner is correct because I > really haven't observed the releaseContext method being called. > > + Cleaner.create().register(this, > + () -> releaseContext(ctxt)); > > Thanks > Max >
