On Wed, Dec 17, 2025 at 2:29 PM Kirill Reshke <[email protected]> wrote: > > I did run a test and this indeed triggers assertion if somebody writes > something like [0]. Code at [0] works (although never testes) only > because is passed > InvalidXLogRecPtr as recptr to visibilitymap_set. Maybe it is worth to > add comment nearby this > > /* > * Avoid relying on all_visible_according_to_vm as a proxy for the > * page-level PD_ALL_VISIBLE bit being set, since it might have become > * stale -- even when all_visible is set > */ > > To explain why is it OK to make conditional MarkBufferDirty?
I actually propose we never do that (because the buffer should always be dirty anyway, and it isn't too expensive to mark an already dirty buffer dirty again). I've proposed a refactoring of this code, which includes comments about this expectation in 0001 on this thread [1] (which you are also participating in). I proposed the test separately since it seems independently valuable but 0001 in [1] contains the same test as the one in this thread actually. - Melanie [1] https://www.postgresql.org/message-id/CAAKRu_Yt6EH5aFSJBm-k7PrNM4bTt56fTRbyU7gqYXe4cW%2BF9g%40mail.gmail.com
