ooblick.com
It seemed to make some kind of sense at the time

Csh Programming Considered Harmful

By Tom Christiansen <tchrist@convex.COM>.

Resolved: The csh is a tool utterly inadequate for programming, and its use for such purposes should be strictly banned.

I am continually shocked and dismayed to see people write test cases, install scripts, and other random hackery using the csh. Lack of proficiency in the Bourne shell has been known to cause errors in /etc/rc and .cronrc files, which is a problem, because you must write these files in that language.

The csh is seductive because the conditionals are more C-like, so the path of least resistance if chosen and a csh script is written. Sadly, this is a lost cause, and the programmer seldom even realizes it, even when they find that many simple things they wish to do range from cumbersome to impossible in the csh.

File Descriptors

The most common problem encountered in csh programming is that you can't do file-descriptor manipulation. All you are able to do is redirect stdin, or stdout, or dup stderr into stdout. Bourne-compatible shells offer you an abundance of more exotic possibilities.

Writing Files

In the Bourne shell, you can open or dup random file descriptors. For example

    exec 2>errs.out

means that from then on, stderr goes into errs file.

Or what if you just want to throw away stderr and leave stdout alone? Pretty simple operation, eh?

    cmd 2>/dev/null

Works in the Bourne shell. In the csh, you can only make a pitiful attempt like this:

    (cmd > /dev/tty) >& /dev/null

But who said that stdout was my tty? So it's wrong. This simple operation CANNOT BE DONE in the csh.

Along these same lines, you can't direct error messages in csh scripts out stderr as is considered proper. In Bourne shell, you might say:

	echo "$0: cannot find $file" 1>&2

but in the csh, you can't redirect stdout out stderr, so you end up doing something silly like this:

	sh -c 'echo "$0: cannot find $file" 1>&2'

Reading Files

In the csh, all you've got is $<, which reads a line from your tty. What if you've redirected stdin? Tough noogies, you still get your tty, which you really can't redirect. Now, the read statement in the Bourne shell allows you to read from stdin, which catches redirection. It also means that you can do things like this:

    exec 3<file1
    exec 4<file2

Now you can read from fd 3 and get lines from file1, or from file2 through fd 4. In modern, Bourne-like shells, this suffices:

    read some_var 0<&3
    read another_var 0<&4

Although in older ones where read only goes from 0, you trick it:

    exec 5<&0  # save old stdin
    exec 0<&3; read some_var
    exec 0<&4; read another_var
    exec 0<&5  # restore it

Closing FDs

In the Bourne shell, you can close file descriptors you don't want open, like 2>&-, which isn't the same as redirecting it to /dev/null.

More Elaborate Combinations

Maybe you want to pipe stderr to a command and leave stdout alone. Not too hard an idea, right? You can't do this in the csh as I mentioned in 1a. In a Bourne shell, you can do things like this:

    exec 3>&1; grep yyy xxx 2>&1 1>&3 3>&- | sed s/file/foobar/ 1>&2 3>&-
    grep: xxx: No such foobar or directory

Normal output would be unaffected. The closes there were in case something really cared about all its FDs. We send stderr to sed, and then put it back out 2.

Consider the pipeline:

    A | B | C

You want to know the status of C, well, that's easy: it's in $?, or $status in csh. But if you want it from A, you're out of luck -- if you're in the csh, that is. In the Bourne shell, you can get it, although doing so is a bit tricky. Here's something I had to do where I ran dd's stderr into a grep -v pipe to get rid of the records in/out noise, but had to return the dd's exit status, not the grep's:

    device=/dev/rmt8
    dd_noise='^[0-9]+\+[0-9]+ records (in|out)$'
    exec 3>&1
    status=`((dd if=$device ibs=64k 2>&1 1>&3 3>&- 4>&-; echo $? >&4) |
		egrep -v "$dd_noise" 1>&2 3>&- 4>&-) 4>&1`
    exit $status;

Command Orthogonality

Built-ins

The csh is a horrid botch with its built-ins. You can't put them together in many reasonable way. Even simple little things like this:

	    % time | echo

which while nonsensical, shouldn't give me this message:

	    Reset tty pgrp from 9341 to 26678

Others are more fun:

	    % sleep 1 | while
	    while: Too few arguments.
	    [5] 9402
	    % jobs
	    [5]     9402 Done                 sleep |

Some can even hang your shell. Try typing ^Z while you're sourcing something, or redirecting a source command. Just make sure you have another window handy. Or try

    % history | more

on some systems.

Flow control

You can't mix flow-control and commands, like this:

    who | while read line; do
	echo "gotta $line"
    done

You can't combine multiline constructs in a csh using semicolons. There's no easy way to do this

    alias cmd 'if (foo) then bar; else snark; endif'

You can't perform redirections with if statements that are evaluated solely for their exit status:

    if ( { grep vt100 /etc/termcap > /dev/null } ) echo ok

And even pipes don't work:

    if ( { grep vt100 /etc/termcap | sed 's/$/###' } ) echo ok

But these work just fine in the Bourne shell:

    if grep vt100 /etc/termcap > /dev/null ; then echo ok; fi

    if grep vt100 /etc/termcap | sed 's/$/###/' ; then echo ok; fi

Stupid parsing bugs

Certain reasonable things just don't work, like this:

    % kill -1 `cat foo`
    `cat foo`: Ambiguous.

But this is ok:

    % /bin/kill -1 `cat foo`

If you have a stopped job:

    [2]     Stopped              rlogin globhost

You should be able to kill it with

    % kill %?glob
    kill: No match

but

    % fg %?glob

works.

White space can matter:

    if(expr)

may fail on some versions of csh, while

    if (expr)

works!

Signals

In the csh, all you can do with signals is trap SIGINT. In the Bourne shell, you can trap any signal, or the end-of-program exit. For example, to blow away a tempfile on any of a variety of signals:

    $ trap 'rm -f /usr/adm/tmp/i$$ ;
	    echo "ERROR: abnormal exit";
	    exit' 1 2 3 15

    $ trap 'rm tmp.$$' 0   # on program exit

Quoting

You can't quote things reasonably in the csh:

    set foo = "Bill asked, \"How's tricks?\""

doesn't work. This makes it really hard to construct strings with mixed quotes in them. In the Bourne shell, this works just fine. In fact, so does this:

     cd /mnt; /usr/ucb/finger -m -s `ls \`u\``

Dollar signs cannot be escaped in double quotes in the csh. Ug.

    set foo = "this is a \$dollar quoted and this is $HOME not quoted" 
    dollar: Undefined variable.

You have to use backslashes for newlines, and it's just darn hard to get them into strings sometimes.

    set foo = "this \
    and that";
    echo $foo
    this  and that
    echo "$foo"
    Unmatched ".  

Say what? You don't have these problems in the Bourne shell, where it's just fine to write things like this:

    echo 	'This is 
	     some text that contains
	     several newlines.'

As distributed, quoting history references is a challenge. Consider:

    % mail adec23!alberta!pixel.Convex.COM!tchrist
    alberta!pixel.Convex.COM!tchri: Event not found.

Variable Syntax

There's this big difference between global (environment) and local (shell) variables. In csh, you use a totally different syntax to set one from the other.

In Bourne shell, this

    VAR=foo cmds args

is the same as

    (export VAR; VAR=foo; cmd args)

or csh's

    (setenv VAR;  cmd args)

You can't use :t, :h, etc on envariables. Watch:

	echo Try testing with $SHELL:t

It's really nice to be able to say

    ${PAGER-more}

or

    FOO=${BAR:-${BAZ}}

to be able to run the user's PAGER if set, and more otherwise. You can't do this in the csh. It takes more verbiage.

You can't get the process number of the last background command from the csh, something you might like to do if you're starting up several jobs in the background. In the Bourne shell, the pid of the last command put in the background is available in $!.

The csh is also flaky about what it does when it imports an environment variable into a local shell variable, as it does with HOME, USER, PATH, and TERM. Consider this:

    % setenv TERM '`/bin/ls -l / > /dev/tty`'
    % csh -f

And watch the fun!

Expression Evaluation

Consider this statement in the csh:

    if ($?MANPAGER) setenv PAGER $MANPAGER

Despite your attempts to only set PAGER when you want to, the csh aborts:

    MANPAGER: Undefined variable.

That's because it parses the whole line anyway AND EVALUATES IT! You have to write this:

    if ($?MANPAGER) then
	setenv PAGER $MANPAGER
    endif

That's the same problem you have here:

    if ($?X && $X == 'foo') echo ok
    X: Undefined variable

This forces you to write a couple nested if statements. This is highly undesirable because it renders short-circuit booleans useless in situations like these. If the csh were really C-like, you would expect to be able to safely employ this kind of logic. Consider the common C construct:

    if (p && p->member) 

Undefined variables are not fatal errors in the Bourne shell, so this issue does not arise there.

While the csh does have built-in expression handling, it's not what you might think. In fact, it's space sensitive. This is an error

   @ a = 4/2

but this is ok

   @ a = 4 / 2

The ad hoc parsing csh employs fouls you up in other places as well. Consider:

    % alias foo 'echo hi' ; foo
    foo: Command not found.
    % foo
    hi

Error Handling

Wouldn't it be nice to know you had an error in your script before you ran it? That's what the -n flag is for: just check the syntax. This is especially good to make sure seldom taken segments of code code are correct. Alas, the csh implementation of this doesn't work. Consider this statement:

    exit (i)

Of course, they really meant

    exit (1)

or just

    exit 1

Either shell will complain about this. But if you hide this in an if clause, like so:

    #!/bin/csh -fn
    if (1) then
	exit (i)
    endif

The csh tells you there's nothing wrong with this script. The equivalent construct in the Bourne shell, on the other hand, tells you this:

    #!/bin/sh -n
    if (1) then
	exit (i)
    endif

    /tmp/x: syntax error at line 3: `(' unexpected

Random Bugs

Here's one:

    fg %?string
    ^Z
    kill  %?string
    No match.

Huh? Here's another

    !%s%x%s

Coredump, or garbage.

If you have an alias with backquotes, and use that in backquotes in another one, you get a coredump.

Try this:

    % repeat 3 echo "/vmu*"
    /vmu*
    /vmunix
    /vmunix

What???

Here's another one:

    % mkdir tst
    % cd tst
    % touch '[foo]bar'
    % foreach var ( * )
    > echo "File named $var"
    > end
    foreach: No match.

Summary

While some vendors have fixed some of the csh's bugs (the tcsh also does much better here), many have added new ones. Most of its problems can never be solved because they're not actually bugs per se, but rather the direct consequences of braindead design decisions. It's inherently flawed.

Do yourself a favor, and if you have to write a shell script, do it in the Bourne shell. It's on every UNIX system out there. However, behavior can vary.

There are other possibilities.

The Korn shell is the preferred programming shell by many sh addicts, but it still suffers from inherent problems in the Bourne shell's design, such as parsing and evaluation horrors. The Korn shell or its public-domain clones and supersets (like bash) aren't quite so ubiquitous as sh, so it probably wouldn't be wise to write a sharchive in them that you post to the net. When 1003.2 becomes a real standard that companies are forced to adhere to, then we'll be in much better shape. Until then, we'll be stuck with bug-incompatible versions of the sh lying about.

The Plan 9 shell, rc, is much cleaner in its parsing and evaluation; it is not widely available, so you'd be sacrificing portability. No vendor is shipping it yet.

If you don't have to use a shell, but just want an interpreted language, many other free possibilities present themselves, like Perl, REXX, TCL, Scheme, or Python. Of these, Perl is probably the most widely available on UNIX (and many other) systems and certainly comes with the most extensive UNIX interface. Some vendors even ship it standard.

If you have a problem that would ordinarily use sed or awk or sh, but it exceeds their capabilities or must run a little faster, and you don't want to write the silly thing in C, then Perl may be for you. You can get at networking functions, binary data, and most ofthe C library. There are also translators to turn your sed and awk scripts into Perl scripts, as well as a symbolic debugger. Tchrist's personal rule of thumb is that if it's the size that fits in a Makefile, it gets written in the Bourne shell, but anything bigger gets written in Perl.

See the comp.lang.{perl,rexx,tcl} newsgroups for details about these languages (including FAQs), or David Muir Sharnoff's comparison of freely available languages and tools in comp.lang.misc and news.answers.