[jdk17u-dev] RFR: 8364660: ClassVerifier::ends_in_athrow() should be removed

[Dean Long]

On Wed, 29 Oct 2025 15:07:23 GMT, Goetz Lindenmaier <goetz at openjdk.org> wrote:

>> I backport this for parity with 17.0.18-oracle from 21.
>> 
>> Resolved Copyright in verifier.hpp.
>> 
>> Skipped VerifierImpl.java, that was added by https://bugs.openjdk.org/browse/JDK-8294982: Implementation of Classfile API
>
> HI @matias9927, @dean-long, @liach
> I backported this pretty much as-is to 21 and 17. 
> @schmelter-sap noted in his review for 17 that the spec change is only for 22+. 
> So is it ok to backport it as I did?  Or do I need to modify the change?

@GoeLin , I can't seem to find the comment by @schmelter-sap, but I think going all the way back to jdk 8, the HotSpot implementation and the specs agree that the same exception handler can't handle both the before and after state, so if we get to the point where ends_in_athrow() is called, verification is going to fail no matter what, because of a stackmap mismatch.  We have had some internal discussions here at Oracle, and agreed that ends_in_athrow() can be removed going back to jdk 8, so I think this PR is correct.

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

PR Comment: https://git.openjdk.org/jdk17u-dev/pull/4128#issuecomment-3464618304