RFR: 8253555: Make ByteSize and WordSize typed scoped enums

[Stefan Karlsson]

On Thu, 24 Sep 2020 06:23:34 GMT, Stefan Karlsson <stefank at openjdk.org> wrote:

> It's a common bug in HotSpot that byte values are used when the code asks for words, and vice versa. One attempt to
> prevent that has been to use the classes ByteSize and WordSize, to make the compiler detect these problems. The current
> implementation is a bit problematic, because both types are int typedefs in release builds, and wrapper classes in
> debug builds. This means that we can't use these types in overload resolution. I propose that we fix that by changing
> the type to scoped enums with a fixed int type. The compiler will then be apple to completely separate ByteSize,
> WordSize, and ints. We also don't need different code in release vs debug builds.  FWIW, I once created a size_t
> versions of these classes and annotated all the metaspace code with these types. This experiment flushed out a handful
> of bugs (some of them were known and the reason for trying that experiment)  There are some controversy about having
> these classes around. See for example: JDK-8041956: remove ByteSize and WordSize classes  I hope we can stay away from
> that discussion in this PR.

Hmm. All these comments came through the mailing list bot, and I can't address them individually. :( I implemented the
minimally necessary for this. The templateInterpreterGenerator_arm.cpp change was needed to get this to compile. The
surrounding code also performs an in_bytes call for the add.

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

PR: https://git.openjdk.java.net/jdk/pull/328