On Mon, 21 Jul 2025 21:08:13 GMT, Chen Liang <li...@openjdk.org> wrote:
> > heapShared.cpp contains a list of hard coded classes & fields for archiving > > the initialized states in CDS archive. With the support for using > > annotation for pre-init being added to the mainline, the support can be > > applied to those classes as well. > > I have looked into that; there is some difficulty, in particular, there are a > lot of special handling for LMF classes, such as setting the protection > domain, which distinguishes them from the LF/String concat classes. Since > this is not as pure a migration, I am planning to do as a separate task; most > likely, the string concat classes would be taken as the general hidden class > template (which works for hidden classes like pattern matching, not sure > about MethodHandleProxies, which also have publicly accessible instances like > LMF) SG. > > > There could be cases where AOTC is not enabled but CDS or heap archiving > > with pre-initialization is enabled, it's good to consider more general > > naming for the annotations rather than using AOT-only-context based > > annotations, like `AOTSafeClassInitializer` and `AOTRuntimeSetup`. > > Now we are using the term AOT not only to describe code compilation (like > JIT), but for general preprocessing, like jlink. So the term AOT is suitable. Ok, that seems reasonable then. > > > As discussed during Leyden/premain meetings, there are classes/fields that > > are not suitable for pre-init and caching, such as Thread, Random, System, > > etc. We want to annotate or mark them so AOTC or CDS archiving ensures > > those states are not cached. > > Also considered for a future patch. This relates to your safeguard proposal; > we might add the safeguard implementation together with the warning > annotation. SG. ------------- PR Comment: https://git.openjdk.org/jdk/pull/25922#issuecomment-3099037148
