RFR: 8359057: AbstractInterpreter::is_not_reached returns incorrectly with invokedynamic

Matias Saavedra Silva matsaave at openjdk.org
Wed Oct 22 21:09:40 UTC 2025


On Tue, 21 Oct 2025 21:08:01 GMT, Vladimir Ivanov <vlivanov at openjdk.org> wrote:

>>> Just curious, does this have any observable effect that can surface in a regression test?
>> 
>> The initial reporter did not offer a test case that displays this issue and I have not been able to write such a test either. It looks like this code cannot be normally executed so I omitted a test case here.
>
>> Just curious, does this have any observable effect that can surface in a regression test?
> 
> I took a closer look and the bug is benign since, as the code is shaped now, it doesn't affect inlining decisions in any way. Indy instructions are linked to invokers which are force inlined and it overrides `AbstractInterpreter::is_not_reached() == true` case when it comes to deciding whether to inline through the indy call or not.

Thanks for the review @iwanowww!
I agree that this change is trivial so one review is sufficient.

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

PR Comment: https://git.openjdk.org/jdk/pull/27904#issuecomment-3434199028


More information about the hotspot-runtime-dev mailing list