Hi Matthias, I reviewed the v2 patch and tested it on my local build.
I verified that before the patch, mdunlinkfiletag() ignored forknum and segno and always operated on MAIN_FORKNUM segment 0. The new direction with register_unlink_tombstone() makes the intent much clearer, and the added Assert() looks like a good safeguard against accidental misuse. I also had a small comment suggestion near the Assert() to make the tombstone-only behavior slightly more explicit. The patch applied successfully with git apply --3way, and PostgreSQL built and tested successfully on my setup. Overall the patch looks good to me. Regards, Solai
