Looks good. +1 Attila. /M
> On 04 Nov 2015, at 14:59, Attila Szegedi <attila.szeg...@oracle.com> wrote: > > Please review JDK-8141425 "Improve caching in NashornCallSiteDescriptor" at > <http://cr.openjdk.java.net/~attila/8141425/webrev.jdk9> for > <https://bugs.openjdk.java.net/browse/JDK-8141425> > > Few words on the implementation: > - I went with one String->NamedOperation map per Nashorn operation, as this > makes it unnecessary to create a NamedOperation to use as a key (and then > have the map use its equals/hashCode) > - I went with global caching, which necessitates using weak/weak synchronized > maps. Alternatively, we could’ve used a per-class caching as for the call > site descriptors themselves, but as one class is one function in Nashorn, > that’d result in reuse of NamedOperation objects only within a single > function. (But we’d get rid of both synchronization and use of weak > references.) > > Thanks, > Attila.
