From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <9ad066c1012da12cc6fe20b64c4aa937@collyer.net>
To: 9fans@cse.psu.edu
From: Geoff Collyer <geoff@collyer.net>
In-Reply-To: <0ca601c3ac7c$063589c0$b9844051@insultant.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Subject: [9fans] mmap (was configure misery)
Date: Sun, 16 Nov 2003 16:30:09 -0800
Topicbox-Message-UUID: 8cfd30ca-eacc-11e9-9e20-41e7f4b1d025

I believe the war story that Boyd is thinking of is that Sun were very
excited about using mmap in SunOS 4.0 (to make the benchmark

	cat bigfile >/dev/null

run in essentially zero time) and even rewrote stdio to use mmap
whenever possible.  I believe their cat still used stdio (or they
rewrote cat to use mmap directly) and they had forgotten in their
excitement that open files have seek offsets and that they are
inherited.  So the shell commands

	(read line; cat; echo $line) <disk-file

would print the entire file, followed by its first line, instead of
all but the first line, followed by the first line.  I can't recall
the real-life example, but it was much more embarrassing.