RFR: 8343493: Perform module checks during MetaspaceShared::map_archives()

[Calvin Cheung]

On Tue, 5 Nov 2024 01:00:00 GMT, Ioi Lam <iklam at openjdk.org> wrote:

> Currently, `Modules::serialize()` and `Modules::serialize_addmods_names()` store/retrieve the main module name, as well as the set of modules specified by `--add-modules`. These functions are called from `MetaspaceShared::serialize()`, *after* the JVM has decided to use the CDS archive.
> 
> However, with JEP 483 ([JDK-8315737](https://bugs.openjdk.org/browse/JDK-8315737)), we need the module information *before* deciding whether to use the CDS archive. For example, if `-XX:+AOTClassLinking` is specified during the CDS archive creation, the archive cannot be used if we specified an incompatible main module at runtime.
> 
> This RFE added a new function, `MetaspaceShared::early_serialize()`, which is called from `MetaspaceShared::map_archives()`. `Modules::serialize()` and `Modules::serialize_addmods_names()` are called within this context so that we can use the module information to decide whether to use the CDS archive or not.
> 
> I also added some comments about what "serialize" does, since it's rather non-obvious.

Looks good overall. One comment:

src/hotspot/share/cds/filemap.hpp line 197:

> 195:   bool    _use_secondary_supers_table;            // save the flag UseSecondarySupersTable
> 196:   size_t  _cloned_vtables_offset;                 // The address of the first cloned vtable
> 197:   size_t  _early_serialized_data_offset;          // Data accessed using {ReadClosure,WriteClosure}::serialize()

Should we print this field in `FileMapHeader::print()`?

-------------

PR Review: https://git.openjdk.org/jdk/pull/21890#pullrequestreview-2414704238
PR Review Comment: https://git.openjdk.org/jdk/pull/21890#discussion_r1828781008