[lworld] RFR: 8378273: [lworld] Use acmp map to detect oops for VarHandle [v2]

[Dan Heidinga]

On Fri, 20 Feb 2026 16:42:30 GMT, Chen Liang <liach at openjdk.org> wrote:

>> src/java.base/share/classes/jdk/internal/value/ValueClass.java line 66:
>> 
>>> 64:         int[] map = Unsafe.getUnsafe().getFieldMap(c);
>>> 65:         int nbNonRef = map[0];
>>> 66:         return nbNonRef * 2 + 1 < map.length;
>> 
>> This method encodes the acmp map layout in a second place in Java.  The first being in ValueObjectMethods.  Rather than propagating this encoding, can we move this method to ValueObjectMethods so the encoding is kept in one place?
>
> Unfortunately, `ValueObjectMethods` is package-private in `java.lang.runtime` and must not be exported to the public. To consolidate uses of the acmp map, we might have to move `ValueObjectMethods` elsewhere so `jdk.internal.value.ValueClass` can access its without VOM being exported to the public.

I think it makes more sense to move this chunk of the method:

int[] map = Unsafe.getUnsafe().getFieldMap(c);
int nbNonRef = map[0];
return nbNonRef * 2 + 1 < map.length;

to Unsafe.java as that keeps the decoding code close to the fetching code.

As much as I'd like all the acmp_map decoding in one place, it doesn't make sense to move ValueObjectMethods at this point

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

PR Review Comment: https://git.openjdk.org/valhalla/pull/2143#discussion_r2834576359