NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/unix/question/14775

Internet Message Format | 1992-12-17 | 1.8 KB

Path: sparky!uunet!think.com!barmar
From: [e-mail] (Barry Margolin)
Newsgroups: comp.unix.questions
Subject: Re: cat a > a , getting it straight
Date: 17 Dec 1992 20:18:23 GMT
Organization: Thinking Machines Corporation, Cambridge MA, USA
Lines: 34
Message-ID: <[e-mail]>
References: <[e-mail]> <[e-mail]> <[e-mail]>
NNTP-Posting-Host: telecaster.think.com

In article <[e-mail]> [e-mail] (John Navarra) writes:
>[riemann:62] ~/tmp -> cat a | cat > a
>[riemann:63] ~/tmp -> cat a
>file a
>
>But, this output is not predictable. Sometimes, a was clobbered several
>times in a row. Sometimes, a was preserved several times in a row. Once,
>I even got this:

The shell runs both parts of a pipeline concurrently.  Sometimes the "cat
a" manages to run to completion before "cat > a" gets started, so it can
read the file before it gets clobbered.  Other times the "cat > a" starts
first, so "cat a" sees an empty input file.

>[casbah:58] ~/tmp -> cat a | cat > a
>cat: write error: Bad address

In this case, "cat > a" started in the middle of "cat a".  On Suns, the cat
command works by mmap()ing the input file if it's an ordinary file, and
then using that memory as the output buffer in a write() call.  What
happened here is that the file was truncated between the mmap() and the
write(), so the write() tried to access memory that was no longer
allocated.

>Somebody is not doing their job right. I couldn't reproduce this on non-Sun
>machines though.

Depends on what you consider "right".  You're not supposed to be able to do
this, so anything that happens is right.
--
Barry Margolin
System Manager, Thinking Machines Corp.

[e-mail]          {uunet,harvard}!think!barmar