Draft JEP Announcement: "Computed Constants"

[John Rose]

On 28 Jul 2023, at 13:53, Remi Forax wrote:

> I'm fine with a method <class-field-init> that takes an int (the index from the AutonomousValue attribute) and returns an Object to initialize each autonomous field at the first call. For me, there is no need for a BSM here.
>
> Also I prefer the CC initialization semantics (the field initializer is only executed once) to the one you are using in your prototype, the field initializer can be called several times concurrently but only store the first value that complete.

Yeah, after polishing my prototype it might end up somewhere like that.

The benefit of the CC stuff is that it can clearly provide and document the correct high-level semantics.  And the low-level VM machinery can be simpler (fewer guarantees), right?  Or we could do an all-in-one solution where the JVMS promises to mimic the CC logic example.

But I prefer a layered design, where the VM uses a BSM to route through the CC logic, which then elevates the result to what the JLS or the CC API will guarantee.  So the VM can use the condy-like binding it is accustomed to, which uses CC directly to get the right mutex behavior.