From mboxrd@z Thu Jan  1 00:00:00 1970
From: norman@oclsc.org (Norman Wilson)
Date: Sun, 11 Sep 2016 21:43:34 -0400 (EDT)
Subject: [TUHS] Shell control through external commands
Message-ID: <20160912014334.434304422E@lignose.oclsc.org>

I remember reading about #! in the early 1980s, and
having mixed feelings about it, as I still do.  The
basic idea is fine, if annoyingly limited; but that
the kernel has to decide, in effect, whether to treat
a header as binary or text bothers me.  Were I designing
a new system from scratch today, I'd just make the
header all text; the small extra space and time for
the kernel to parse that for binaries doesn't matter
any more.  It certainly did when #! was invented,
though.

I had the impression at the time that it came from
Berkeley, but I think I later heard from the horse's
mouth that it was originally Dennis's idea.

I don't think anyone has yet laid out the complete
story of what came before:

1.  Originally, the shell would exec(file), and if
exec returned ENOEXEC, would open the file and treat
it as shell commands.

2.  Then came the C shell, and a problem: did file
contain commands for csh or sh?  A hack emerged:
if csh encountered a script file, it would read
the first character; if that was '#' it was a
csh script, otherwise it handed off to /bin/sh.

None of this helped when some program other than
the shell called exec on a shell script.  That's one
reason execlp and execvp appeared.  (The other is that
they observe $PATH if the command pathname has a
single element.)

I don't know offhand whether there was ever an execlp/vp
that implemented the #-means-csh convention.  Anyone
else remember?

Norman Wilson
Toronto ON