gwyn@brl-smoke.ARPA (Doug Gwyn ) (07/05/87)
It has been pointed out to me off-line that current UNIX implementations maintain the CWD as an inode pointer, so actually all the stat info is kept resident, not just the dev+inumber. (Actually I knew this, but I've gotten into the habit of thinking of the sys+dev+inumber as the unique handle for an inode.) The main advantage to this approach is that it is more efficient to access the inode when opening a relative pathname than to find one from a prefix, even if a clever implementation (hashing, caching, etc.) is used for prefixes. The main disadvantage as I see it is that special filesystem types have to pretend to support a directory structure like the disk file system, including manufacturing ".." entries (for /bin/pwd).