From mboxrd@z Thu Jan  1 00:00:00 1970
MIME-Version: 1.0
Date: Thu, 23 Jan 2014 22:27:42 -0500
Message-ID: <CACzOosDi7uK-JPvTcdN3So9VzhkjmfvBto0YwUTkdWka41r65Q@mail.gmail.com>
From: cherry <lunaria21@gmail.com>
To: Fans of the OS Plan 9 from Bell Labs <9fans@9fans.net>
Content-Type: multipart/alternative; boundary=001a1134890ad9ace704f0aef01e
Subject: [9fans] echo -n and zero-length write(2)
Topicbox-Message-UUID: b45fb15e-ead8-11e9-9d60-3106f5b1d025

--001a1134890ad9ace704f0aef01e
Content-Type: text/plain; charset=ISO-8859-1

Hello Fans,

When echo(1) is invoked with -n and no or empty arguments, it calls
write(2) with zero length. When writing to a pipe, the other end of the
pipe read()'s zero byte, which causes it thinks it's got to the end and
close the pipe, and commands after the "echo -n" would fail to write. I
start thinking whether this is the desired behavior. On one hand, it is an
easy way to do a zero-length write() through echo -n, also it is consistent
for echo(1): to do a write(2). On the other hand, it makes rc script
behaves differently if sending to a pipe, and probably some people would
expect empty echo -n is just a no-op. Which should be better?

The context is, as posted on the werc list, it's template.awk generates rc
script with echo -n, and the output of the script is sent to a pipe. When
the argument is empty it fails. My last post was wrapping echo as

fn echo { builtin echo $* | cat }

But invoking cat for every echo hurts the performance. Perhaps the most
efficient way is to write an ad hoc echo that does not do zero-length
write(2)? Any suggestion?

Thanks,
- cherry

--001a1134890ad9ace704f0aef01e
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr">Hello Fans,<br><br>When echo(1) is invoked with -n and no =
or empty arguments, it calls write(2) with zero length. When writing to a p=
ipe, the other end of the pipe read()&#39;s zero byte, which causes it thin=
ks it&#39;s got to the end and close the pipe, and commands after the &quot=
;echo -n&quot; would fail to write. I start thinking whether this is the de=
sired=20
behavior. On one hand, it is an easy way to do a zero-length write() throug=
h echo -n, also it is consistent for echo(1): to do a write(2). On
 the other hand, it makes rc script behaves differently if sending to a=20
pipe, and probably some people would expect empty echo -n is just a no-op. =
Which should be better?<br><br>The context is, as posted on the werc list, =
it&#39;s template.awk generates rc script with echo -n, and the output of t=
he script is sent to a pipe. When the argument is empty it fails. My last p=
ost was wrapping echo as<br>
<br>fn echo { builtin echo $* | cat }<br><br>But invoking cat for every ech=
o hurts the performance. Perhaps the most efficient way is to write an ad h=
oc echo that does not do zero-length write(2)? Any suggestion?<br><br>Thank=
s,<br>
- cherry<br><br>

</div>

--001a1134890ad9ace704f0aef01e--