On Tue, 17 Mar 2026 09:51:16 GMT, Fredrik Bredberg <[email protected]> wrote:
>> On x86, last_sp and sender_sp are essentially the same. Which means the >> method handle linker adjustment can make last_sp point beyond SP, yet >> safe_for_sender only cares about sender_sp/unextended_sp, not last_sp. The >> more I look at this, the more I feel like the proposed change is not the way >> to go. I do not agree with the comments in the JDK-8302320 PR discussion >> that we should change unextended_sp rather than fix safe_for_sender. I am >> in the opposite camp: fix only safe_for_sender. > > Thank you @dean-long for all the info you added to this PR. > I don't have any strong opinion where to fix this issue, or even if it should > be fixed. > But I feel that we should somehow collect the info you added for the future. > All the stack pointer variants like `expression`, `sender`, `unextended`, > `last` etc. are utterly confusing when starting to read the VM source. Adding > that some platforms have padding needs doesn't make it easier. Hi Dean, thanks for your comments. I now tend to think that the original issue isn't really a problem. Given possible side effects of the proposed changes and small but important platform differences, enforcing some relation just for the sake of it is not worth it. Me and @fbredber will try to make a comprehensive collection of information about different flavors of frame-related pointers one can find in the VM. ------------- PR Comment: https://git.openjdk.org/jdk/pull/29744#issuecomment-4080842933
