Using @Stable breaks down some of the complexity. And the CHM (with the winning computeIfAbsent) breaks down lots more. Nice job!
I'm a little sad that it needs two CHMs (CACHE and CLASS_CACHE). This must be because there are three or four states per key, instead of the usual two. These states should be documented for posterity. — John On Nov 9, 2015, at 4:16 AM, Peter Levart <peter.lev...@gmail.com> wrote: > > Hi all, > > Thanks for analysis, reviews and discussion that hopefully beheaded this > hydra. > > I added an assert to method setSpeciesDataToConcreteBMHClass() that verifies > the presence of @Stable annotation on the SPECIES_DATA field of the generated > class. It caught a bug I made when I specified a binary class name > "java/lang/invoke/Stable" in FieldVisitor.visitAnnotation() instead of type > signature "Ljava/lang/invoke/Stable;" > > The consequence was not drastic - the field was just not being interpreted by > VM as @Stable. > > With that adjustment and successful re-run of jtreg tests, I ask for > re-confirmation to push the following: > > http://cr.openjdk.java.net/~plevart/jdk9-dev/BMH.race/webrev.07/ > > Regards, Peter > > On 11/06/2015 01:31 PM, Vladimir Ivanov wrote: >> Peter, >> >>> >>> http://cr.openjdk.java.net/~plevart/jdk9-dev/BMH.race/webrev.06/ >>> >> >> Looks really good! Reviewed. >> >>> - the SPECIES_DATA field in generated class can not be static final when >>> it is initialized out of <clinit>. Just static. It was just static >>> before when it was initialized in <clinit> and could be static final. We >>> can make it @Stable at least. >>> >> Not sure it is on hot path, but I'm all for marking the field as @Stable >> >>> - there's no need for empty <clinit> method now. Removed. >>> >>> Basic java/lang/invoke jtreg tests pass. >> >> Best regards, >> Vladimir Ivanov >
