On Mon, 16 Feb 2026 15:21:46 GMT, Tobias Hartmann <[email protected]> wrote:
>> ## Issue
>>
>> The test `compiler/ciReplay/TestInliningProtectionDomain.java` still fails
>> with `--enable-preview` even when using `java.lang.Record` as return class.
>>
>> ## Causes
>>
>> `java.lang.Record` is part of the migrated value classes that are added to
>> the class loader if _enable preview_ is on and consequently it gets resolved
>> early, making `ProtectionDomainTestNoOtherCompilationPrivate` fail.
>>
>> ## Fix
>> To avoid further potential issues (with another lib class that could
>> potentially be loaded early in the future) I replaced the `Record` return
>> type with a new test-local `class SigType {}` in the three `bar()` methods.
>> Additionally the private-case replay assertion had to be updated to expect
>> `unloaded signature classes` instead of `disallowed by ciReplay` (`Record`
>> was resolved early and inlining was explicitly denied by replay. `SigType`
>> is simply not resolved). The purpose of the test (that replay compilation
>> produces the same inlining outcome as the normal run) should still be met.
>>
>> ## Testing
>> Tier 1-3+
>
> Third time's the charm - looks good to me! 😄
I hope it's the definitive one! 😉 Thanks for your review @TobiHartmann.
-------------
PR Comment: https://git.openjdk.org/valhalla/pull/2113#issuecomment-3913654493
