On Mon, 16 Feb 2026 14:51:52 GMT, Joel Sikström <[email protected]> wrote:
> Hello,
>
> We should consider moving the enum ArrayProperties from ArrayKlass to its own
> class, ArrayProperties. In addition to making the code easier to read and
> understand, this allows us to have explicit setters/getters, replacing the
> bit-fiddling expressions that are used in many places. The
> ArrayProperties-specific methods in ArrayKlass have been moved to be methods
> in the new ArrayProperties class instead.
>
> Perhaps the most controversial change in this PR is the removal of
> `ArrayKlass::ArrayProperties::DEFAULT` in favor of using a default
> constructor for ArrayProperties. The semantics are still the same, i.e.,
> asking `.is_null_restricted()` or `.is_non_atomic()` will be false for the
> default constructed property. With this I've also removed the unused fields
> from ArrayProperties (DUMMY and comments).
>
> I did consider using define macros to generate enum+getters+setters, but I
> opted for the stamped-out version instead.
>
> Testing:
> * Running through tier1-2
src/hotspot/share/oops/objArrayKlass.cpp line 425:
> 423: if (next_refined_array_klass() == nullptr) {
> 424: ObjArrayKlass* first = this;
> 425: if (!is_refArray_klass() && !is_flatArray_klass() &&
> (props.is_null_restricted() || props.is_non_atomic())) {
The check for this was `props != ArrayKlass::ArrayProperties::DEFAULT` before
this, which included the INVALID state. I ran this through tier1-2 with the
assert `assert(!props.is_invalid(), "surely");`, which did not hit.
I'd appreciate feedback on if we still want to check for the invalid property
here!
-------------
PR Review Comment:
https://git.openjdk.org/valhalla/pull/2114#discussion_r2812812833
