[lworld] RFR: 8377791: [lworld] Assert in java_lang_Class::restore_archived_mirror

[Matias Saavedra Silva]

On Thu, 12 Feb 2026 21:21:54 GMT, Frederic Parain <fparain at openjdk.org> wrote:

>> Different array layouts result in distinct array klass metadata which are stored in a linked list with the base ObjArrayKlass acting as the head. Multiple layouts are currently uncommon, so this assert was sufficient:
>> 
>> `assert(objarray_k == k || objarray_k->next_refined_array_klass() == k) failed: must be`
>> 
>> This assert is not correct since `k` could refer to another refined array klass further down the linked list. This patch corrects this assert by adding a find method to iterate through the linked list in case there are multiple layouts for the same type. A new test was added to catch this failure. Verified with tier 1-5 tests.
>
> src/hotspot/share/classfile/javaClasses.cpp line 1273:
> 
>> 1271:     }
>> 1272:   } else {
>> 1273:     ObjArrayKlass* objarray_k = (ObjArrayKlass*)as_Klass(m);
> 
> This is the else clause of the `if (!k->is_array_klass())` , which means that k is either and objArrayKlass (or sub-type) or a typeArrayKlass. How is it safe to unconditionally cast to objArrayKlass?

You're right, this seems to work by accident. In this case, `k` could be a TypeArrayKlass and casting it to ObjArrayKlass doesn't seem to break anything so it passes the assert. I'll update this case to an `else if`

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

PR Review Comment: https://git.openjdk.org/valhalla/pull/2095#discussion_r2801172960