From owner-freebsd-current  Tue Mar 26 22:30:45 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id WAA28809
          for current-outgoing; Tue, 26 Mar 1996 22:30:45 -0800 (PST)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id WAA28794
          Tue, 26 Mar 1996 22:30:41 -0800 (PST)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.12/8.6.9) id RAA19092; Wed, 27 Mar 1996 17:29:02 +1100
Date: Wed, 27 Mar 1996 17:29:02 +1100
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199603270629.RAA19092@godzilla.zeta.org.au>
To: loodvrij@gridpoint.com, phk@critter.tfs.com
Subject: Re: Patch to talkd
Cc: asami@cs.berkeley.edu, freebsd-current@freebsd.org,
        gibbs@freefall.freebsd.org, jkh@time.cdrom.com,
        joerg_wunsch@uriah.heep.sax.de, paul@netcraft.co.uk
Sender: owner-current@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

>> All of this would be much better if the timestamp on the tty device-node
>> was only updated on input...
>> 
>These problems seem to be caused by programs like systat and elm which sit
>there in read() and respond to signals.  When the process handles a
>signal, the atime is set.  Doesn't seem right to me.

For all special files, the atime is (bogusly) marked for update when a
read() is started and it isn't marked for update (as is required) upon
successful completion of the read().

For ufs regular files, the atime is marked for update (and immediately
updated) upon both successful and unsuccessful completion.

Bruce