Path: utzoo!attcan!uunet!zephyr.ens.tek.com!uw-beaver!milton!dali.cs.montana.edu!uakari.primate.wisc.edu!zaphod.mps.ohio-state.edu!rpi!uupsi!cmcl2!kramden.acf.nyu.edu!brnstnd
From: brnstnd@kramden.acf.nyu.edu (Dan Bernstein)
Newsgroups: comp.unix.wizards
Subject: Re: Should find traverse symbolic links?
Message-ID: <6528:Mar201:43:4391@kramden.acf.nyu.edu>
Date: 2 Mar 91 01:43:43 GMT
References: <15319@smoke.brl.mil> <124226@uunet.UU.NET> <1888@mitisft.Convergent.COM>
Organization: IR
Lines: 23

Side issue: Traversing the entire file system makes sense, but it really
doesn't work for many applications. Any program that returns a list of
all objects satisfying property P, one at a time, had better make two
things true: 1. If an object is returned in the list, it must have had
property P sometime while the program was running. 2. There must be some
period of time so that (a) the program is not running except during that
period (and maybe less); (b) if any object has property P throughout
that period, then it is returned in the list.

find guarantees #1 but not #2. If someone moves a directory while you're
removing all core files or recording all setuid scripts or whatever, you
may miss some files.

(Note that most versions of readdir() don't guarantee #2, and some don't
even guarantee #1. This unreliability is a royal pain.)

It would be much better to have openinodes() and readinodes() calls to
traverse the set of inodes in a given filesystem, one by one. Then both
#1 and #2 would be true, and a lot of file-tree walk code could be done
away with. This could even be made available to normal users who want to
walk through all the files they own.

---Dan