[lworld] RFR: 8375441: [lworld] C2: assert(is_instance()) failed: bad cast [v2]

[Marc Chevalier]

On Mon, 2 Feb 2026 16:33:05 GMT, Quan Anh Mai <qamai at openjdk.org> wrote:

>> Marc Chevalier has updated the pull request incrementally with 10 additional commits since the last revision:
>> 
>>  - Details
>>  - exception handling
>>  - oops
>>  - Also cache correctly whole stable fields
>>  - fix comment
>>  - Remove debug
>>  - ShouldNotReachHere
>>  - explicit casts
>>  - Clean up
>>  - expand_constant
>
> src/hotspot/share/ci/ciInstance.cpp line 154:
> 
>> 152:     return ciConstant();
>> 153:   }
>> 154:   if (field->original_holder() == nullptr) {
> 
> I think simply comparing `field` and `containing_field` is enough.

I thought too, but then, I was surprised.

For instance from `GraphBuilder::access_field` (that does `ciConstant field_value = field->constant_value_of(const_oop);`), we get the field from `ciField* field = stream()->get_field(will_link);`. Deep inside, the field is created in `ciField* ciEnv::get_field_by_index_impl(ciInstanceKlass* accessor, int index, Bytecodes::Code bc)` and put in the `field_cache` (whose role is not clear to me). The interesting part is that it's a field created here.

And later, when the fields are computed by `void ciInstanceKlass::compute_nonstatic_fields()`, new `ciField` objects are created, but the field_cache ones aren't reused. In such a case, I get a `field` and `containing_field` that have identical content but with different pointers. If I simply compare the pointers, the `field` is wrongly interpreted as a sub-field of `containing_field`. I guess I could compare their symbols (since offset and signatures might not be enough), but that doesn't seem much better to me.

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

PR Review Comment: https://git.openjdk.org/valhalla/pull/1923#discussion_r2758606244