On Wed, 10 Dec 2025 17:56:22 GMT, Phil Race <[email protected]> wrote:
> This change worries me. How was it tested ? I see MANY places (dozens and > dozens) that Swing internally passes null to these calls. The behaviour of > these is now changed. @prsadhuk please explain how this is a safe change and > if you examined all of the usages of this with null ? In all cases, Swing passes null to these calls in `uninstallKeyboardActions` during `uninstallUI` that is, during L&F change and it doesn't check if the map is removed or not, even though spec mandates it should. It would have seen the issue if uninstall* APIs had tried to use the same map but map is reloaded and replaced by calling these methods in `installUI` so it was never problematic. I dont know why this spec violation is never caught, apart from this bug, even by JCK test. I have tested with all clientlibs test in CI along with SwingSet2 with L&F change and keyboard navigation works ok. ------------- PR Comment: https://git.openjdk.org/jdk/pull/28671#issuecomment-3640036120
