On Fri, 30 May 2025 04:11:36 GMT, Sergey Bylokhov <s...@openjdk.org> wrote:
> The patch for [JDK-8357299](https://bugs.openjdk.org/browse/JDK-8357299) does > not completely fix the regression it was intended to address. Instead of > bailing out on overflow, it introduces additional logic around src and clip, > which might lead to attempts to draw invalid source pixels. > > A test case demonstrating the issue is attached to > [JDK-8358103](https://bugs.openjdk.org/browse/JDK-8358103)(passed on jdk22 > failed on latest openjdk/jdk). > > There are some unresolved discussions related to that patch, see > [PR](https://github.com/openjdk/jdk/pull/25340). I believe it would be better > to bail out early in case of overflow than risk incorrect calculations. > > Therefore, I propose reverting > [JDK-8357299](https://bugs.openjdk.org/browse/JDK-8357299) until a proper fix > is available. This approach would also simplify backports, since only the > patch for [JDK-8358103](https://bugs.openjdk.org/browse/JDK-8358103) would > need to be backported > > @prsadhuk, @prrace please take a look Marked as reviewed by psadhukhan (Reviewer). > A test case demonstrating the issue is attached to > [JDK-8358103](https://bugs.openjdk.org/browse/JDK-8358103)(passed on jdk22 > failed on latest openjdk/jdk). I believe this new testcase is also failing with jdk24.0.1 too where UNSAFE_TO_ADD macro was added for overflow check and caused [JDK-8357299](https://bugs.openjdk.org/browse/JDK-8357299) issue so both testcase is failing with jdk24.0.1 but if you feel the earlier bailout is better than this, then it's ok. ------------- PR Review: https://git.openjdk.org/jdk/pull/25537#pullrequestreview-2880172293 PR Comment: https://git.openjdk.org/jdk/pull/25537#issuecomment-2921224425
