How To Adopt 10 'Good' Unix Habits

An anonymous reader writes to mention an article at the IBM site from earlier this week, which purports to offer good Unix 'habits' to learn. The ten simple suggestions may be common sense to the seasoned admin, but users with less experience may find some helpful hints here. From the article: "Quote variables with caution - Always be careful with shell expansion and variable names. It is generally a good idea to enclose variable calls in double quotation marks, unless you have a good reason not to. Similarly, if you are directly following a variable name with alphanumeric text, be sure also to enclose the variable name in square brackets ([]) to distinguish it from the surrounding text. Otherwise, the shell interprets the trailing text as part of your variable name -- and most likely returns a null value."

FP?

[Shadow-isoHunt]

export POST="first"

Re:

[Anonymous Coward]

Imbecile! POST wasn't defined yet! This won't work on /bin/sh, only with the proprietary bash shell.

Re:

[lahi]

Actually it will work with Korn shell as well, and probably zsh too. Not to mention that many systems have a /bin/sh which is Bourne-compatible but enhanced. Many systems have a /bin/sh implementation that supports this, not just bash-based Linux systems.

-Lasse

welll..

[macadamia_harold]

An anonymous reader writes to mention an article at the IBM site from earlier this week, which purports to offer good Unix 'habits' to learn.

I seriously doubt reading this article is going to get anyone to start showering on a regular basis.

Re:welll..

[AchiIIe]

Some of the points he is making are BS. They are not good `Unix habits` they are simply hacks that marginally reduce the workload but (arguably) increase complexity.

Ie there is NOTHING bad about piping cats. While you might indeed get a ~30% performance increase if you skip the cat, the complexity increases. We often sacrifice performance in order to increase abstraction and understanding.

What makes unix so powerful is its modularity, the fact that you can pipe any output from any application to any applications stdin. This makes it possible to use common tools app1 | app2, app1longoutput | grep thingsIwant. The possibility to mix and match common elements that (arguably) makes unix powerful.

Advice that says "stop piping cats" is akin to "stop using helper functions, they overload the stack, instead do everything in one function"

--
A better articulated article on the programmers intellectual ability vs proper abstraction techniques:
http://www.acm.org/classics/oct95/ [acm.org] - Dijkstra, Edsger - "Go To Statement Considered Harmful"

Re:welll..

[SharpFang]

Ie there is NOTHING bad about piping cats

PETA would disagree.

Re:welll..

[jc42]

Ie there is NOTHING bad about piping cats

PETA would disagree.

Oh? Just imagine the cacophony of a bunch of cats playing bagpipes while "singing" along.

Re:

[t_ban]

Some of the points he is making are BS. They are not good `Unix habits` they are simply hacks that marginally reduce the workload but (arguably) increase complexity. Ie there is NOTHING bad about piping cats. While you might indeed get a ~30% performance increase if you skip the cat, the complexity increases. We often sacrifice performance in order to increase abstraction and understanding. What makes unix so powerful is its modularity, the fact that you can pipe any output from any application to any ap

Re:welll..

[Znork]

I'd tend to agree with the GP. Consider for example if you have excessively badly named files like '-whatever' in a particular directory; cat has very few destructive ways it can go wrong, other commands may be less forgiving, and cause much more surprise.

Further, the assembly line abstraction of cat as 'input the contents of these files into the beginning of my pipeline' is predictable, simple and very clear and readable. Using the filenames in the commands means you have to be certain each command will take filenames, and if you replace the first step (from a grep to an awk, for example), you have to rethink your input method semantics again.

Any typing speed gains and performance improvements you may get will probably get shot the first time some command does something unexpected, or by the extra steps of thought.

And if performance really was a serious concern you probably shouldnt be writing it as a shell script...

Piping cats

[this great guy]

People who argue that piping a single file via cat is the best method are wrong. The following method has all of the advantages you cite, but is also shorter to type, uses less system resources (no cat process, no pipe(2) object), and doesn't require you to "rething the input method" in case you want to change the grep command:

    $ <file.txt grep foobar

Few people know that input-redirection can be established before the command name :)

Re:welll..

[hackstraw]

cat-ing a file and then piping it to grep. surely that is a good point he is making, because grep already takes filenames as an argument?

That list was fairly arbitrary, but the piping cat thing is something that basically only annoys the most anal of anal, and they probably do it sometimes too.

Its common for me to do cat foo and then hit the up arrow and append a pipe to another command instead of editing the whole command line. Computers are pretty fast, and real anal people would use fgrep instead of grep, but again I always use egrep, because I never know when a regular expression will be edited into a more complex one, and to me all of the speeds are the same.

My #1 habit to tell people, although it is not a habit, but just where to start it to learn your shell. No science guys, csh is not a worthy shell in 2006. If you have to suffer with the wacky behavior of a csh variant, at least use tcsh.

My #2 thing to learn is a text editor.

As far as habits go. First and foremost, unalias cp, mv, rm to have the -i flag. In my opinion, that is a BAD habit to start. You WILL lose files sooner or later, and the more painful the better so that you will think so you will stop doing it. the -i flag will NOT stop you from redirecting into a file, and the most dangerous is the -rf flag with rm will override that -i. Remote copies via rcp or scp will not honor the -i flag. Unarchiving an archive will not honor the -i flag. There are tons of ways to lose files, and you will lose them. Its a much better habit to universally save yourself from yourself to not lose them by testing with -i, working off of a copy, and thinking before you hit return, creating new directories to eliminate clobbering a file, NEVER, EVER, do tar cf foo.tar . or tar cf foo.tar *. You will piss yourself and others by doing that.

Actually, this top 10 list is pretty lame, and should be ignored.

Re:

[theonetruekeebler]

The best reason to pipe to grep is to keep filenames out of the output. grep foo ?.txt will produce
a.txt: foo
b.txt: foo
c.txt: foo
whereas cat ?.txt|grep foo produces
foo
foo
foo

I've also seen Unixes where their shells are linked against spectacularly broken libc's. Under Tru64's Bourne and Korn shells, for example, a multithreaded program foo fork bombs when run as foo < z.txt, but works fine as cat z.txt|foo (foo < z.txt under Bash works, though, because Bash is linked against the GNU

Re:welll..

[johnw]

I argue that using> grep file1 file2 file3 regex can be more confusing

Not least to the grep utility, which expects the regex first, not last.

I always use 2 or more args for grep

[benhocking]

E.g.,

grep someFn `find . -name "*.cpp"`

Personally, I was suprised he didn't mention the backtick. Now, that's useful. (Although they can be annoying while camping.)

Re:

[camperdave]

Speaking as a novice, cat|grep is easier to understand. Until I skimmed the fancy article, I was unaware that grep could take a filename like that. I thought that to do filenames you needed fgrep. To an experienced unix guru, it may not be any clearer, but to a novice there's a world of difference.

Re:welll..

[bugg]

I don't think it ever makes sense to use cat with one file - something I have seen far too many people do. To do so, logically, is to tell the commands to run through the file twice.

First you are telling cat to output the entire file, and then you are telling grep to go through the entire output of cat. If you're working with gigabytes of data here, that can quickly be a frustrating exercise! Folks who are in the mentality of using cut | grep and even a visual editor like vi instead of sed are up the creek when they find themselves needing to manipulate and get portions of very large data sets.

Re:

[db32]

I frequently cat bleh | grep whatever just because it was the first way I learned and reflexively do it that way. grep whatever * actually tells me what file it found the match in, rather than just showing me the line. So while piping cats isn't inherently bad, there is frequently better ways to do it that also give more correct output, but it depends on what the desired output is. However "I grepped all the files for 'bleh'" sounds infinitely less disturbing than "I grepped what came out of the cat I pi

Square or Curly brackets?

[Beolach]

enclose the variable name in square brackets ([])

~ $ ls tmp/
a b
~ $ VAR="tmp/*"
~ $ echo $VARa

~ $ echo "$VARa"

~ $ echo "${VAR}a"
tmp/*a
~ $ echo ${VAR}a
tmp/a

Their example correctly uses Curly brackets, {}, but their text says square brackets []. That seems like a typo to me.

Re:

[tigga]

Their example correctly uses Curly brackets, {}, but their text says square brackets []. That seems like a typo to me.

I think somebody needs glasses ;)

Re:

[thc69]

I think somebody needs glasses ;)

Don't you mean:
I think somebody needs glasses ;}

Anyway...Curly brackets? Nyuk nyuk nyuk!

Re:Square or Curly brackets?

[lmfr]

The correct form is {}, not []. There are other things you can use with ${VAR}:

• ${VAR:-text if $VAR is empty}
• ${VAR:=text if $VAR is empty and set VAR to this}
• ${VAR:+text if $VAR is set}
• ${#VAR} -> length of $VAR
• ${VAR#pattern} or ${VAR##pattern} -> remove match of pattern from beginning of $VAR (## -> longest match)
• ${VAR%pattern} or ${VAR%%pattern} -> remove match of pattern from end of $VAR (%% -> longest match)

There are other formats (see the man page), but these are the ones I use the most. Eg:

for i in *.png; do convert "$i" "${i%.*}.jpg"; done

Typo

[seebs]

The quoted paragraph from the article is incorrect -- and it is in the article too -- but the example immediately following it correctly shows the use of braces ("curly brackets"), not square brackets, for variable names in shell.

mkdir

[pfafrich]

His example of good habit with mkdir did not convince me

$ cd tmp/a/b/c || mkdir -p tmp/a/b/c

If the directory exists you end up in the directory, if it does not it creates the directory but leaves you where you first started. Hence you don't know which directory you will be in after the command is executed!

Comment removed

[account_deleted]

Comment removed based on user account deletion

Re:

[plalonde2]

Save a few keystrokes and typos:

( A=tmp/a/b/c cd $A || mkdir -p $A && cd $A )
But this is silly: mkdir -p succeeds even if the directory already exists, so this suffices:
mkdir -p a/b/c && cd a/b/c
I also resent the use of zillions of little command options to "be more efficient". The -C to tar is just plain silly:
( cd a/b/c ; tar xf $ELSEWHERE/foo.tar )
does the same thing, withouth having to look up the fricking -C each time you want to use it. Three guesses which technique has th

Re:

[lahi]

Not only is it silly, but the example given in Listing 4 is WRONG: tar xvf -C tmp/a/b/c newarc.tar.gz will not work, unless there is a version of tar which implies -z given -C. (Someone else already mentioned the misplacement of -C as the filename for -f.) And of course, relying on -z options for tar will not work portably (including versions of AIX, though I'm not sure about AIX 5). This is a case where zcat (given a zcat that knows gzip) should be used: zcat arc.tar.gz| tar xvf -

The latter also works nice

Even that's wrong / Ich bin Grammatiknazi

[The Monster]

The latter also works nicely across machine boundaries:
zcat arc.tar.gz| ssh user@foo 'cd tmp/a/b/c ; tar xvf -'

Only that isn't portable. Some flavors of unix zcat will try to uncompress $1.Z ; clearly there wouldn't be a arc.tar.gz file.

Fortunately, there's an easy fix for that: Use redirected stdin rather than a filename for zcat :

zcat <arc.tar.gz | ssh user@foo 'cd tmp/a/b/c ; tar xvf -'

Some people call me a 'Grammar Nazi' for being so picky about how people use English. I get frustrated

Preview is my friend!

[The Monster]

clearly there wouldn't be a arc.tar.gz.Z file

Re:

[multipartmixed]

You should be using gzcat, not zcat, anyhow. zcat is only portably able to be compress -d.

gzcat will never be broken in the way described, hence the following is fine and portable IME:

gzcat arc.tar.gz | ssh user@foo 'cd tmp/a/b/c && tar -xvf -'

HOWEVER, I find that even vaguely modern CPUs are much faster at gunzipping than typical internet speeds. So, I would use this myself:

cat arc.tar.gz | ssh user@foo 'cd tmp/a/b/c && gzcat | tar -xvf -'

On the otherhand, I would never actually write that,

Re:If I wanted to upload binaries...

[lahi]

Right. So 10 *really* good Unix "habits" would be:
1. Never use csh or any derivative thereof.
2. Know the portable behaviour of your Unix tools.
3. Learn to use ed, one day you'll be glad you did. You can also use ed and ex from scripts or from a command.
4. A shell command is a small program. If you are unsure about a command, test it first, like you would any program.
5. Learn to use the standard shell on your system.
6. Learn useful nonstandard extensions of utilities, but use them with care.
7. Never rely on an extension to the point that you forget how to do it portably. The definition of "portably" is up to you.
8. Learn to use csh enough that you can make do in an emergency, and learn *why* you shouldn't use it.
9. If your standard shell is Bash, learn Korn too. And vice versa. Learn both, how they differ, and how they differ form your standard shell.
10. Sometimes a real C program or a script in a different language is better than using shell.

-Lasse

Re:

[Russ Nelson]

In that case, do this:
mkdir -p tmp/a/b/c; cd tmp/a/b/c

Re:mkdir

[SharpFang]

Especially the habit of using || and && on command line seems ridiculous to me. These have room in two situations:
- scripts
- commands that take long enough that you go have a coffee.

This makes sense:

make install && lilo && reboot

This doesn't:

cd tmp/a/b/c || mkdir -p tmp/a/b/c

If you fail the first part, well, you typed " || " instead of pressing enter.
If you succeed the first part, you typed " || mkdir -p tmp/a/b/c" without a bloody reason.

Type first part. Press enter. Observe result.
If necessary, type the second part, otherwise correct the first without baggage of the second one hanging around.

Re:

[hackstraw]

First, let me be clear that this list is bogus. These 'Good' habits are not that advantagious at all, and I've been using UNIX for quite some time now, and I'm pretty good at it.

Now, being pedantic, the cd a/b/c || mkdir -p /a/b/c is silly, just do the mkdir -p /a/b/c if that is what you want. Don't test for it, just do it!

Re:

[canuck57]

His example of good habit with mkdir did not convince me

$ cd tmp/a/b/c || mkdir -p tmp/a/b/c

If the directory exists you end up in the directory, if it does not it creates the directory but leaves you where you first started. Hence you don't know which directory you will be in after the command is executed!

Actually, the example is a good habit.

If the change directory fails, the script will not make the directory in the wrong place. It is a good practice.

Unix is more than just a shell

[petes_PoV]

This article is really just about good (in the authors opinion) TTY shell usage. There's more to Unix than just its shell
(plus he didn't mention my favourite shortcut: shell history)

How about being more inclusive and expanding this to deal with security features (surely the single biggest benefit?) and the ease of working on remote boxes?

Re:

[SharpFang]

You cannot find anything useful in shell history unless there actually is something

Mod parent 'redundant'.

Shell code in linux based devices

[wertarbyte]

Quote variables with caution

Hey, FON [slashdot.org], are you reading [datenbruch.de] this? Sadly, even shell code in comsumer devices is often deeply flawed. Which is a good thing, if you want to install a new firmware - and a bad thing, if an attacker does it for you.

This article...

[nevali]

...is so littered with basic errors that it really shouldn't be recommended to anybody. How is 'tar xvf -C tmp/a/b/c newarc.tar.gz' expected to work, for example? Quote variables with square brackets? Running subshell commands using ; instead of && ? No mention of 'xargs -0' ? Don't pipe from cat to grep? Does anybody actually care that people do this (primarily so that the syntax is consistent between a munged- and unmunged-grep, and also such that the order of the command-line is logical from a human point of view)? Plus, of course, it's possible that cat | grep could yield better performance than grep alone: if cat uses mmap() to efficiently read the input files, and the kernel's pipe implementation is good, then it could do better than a grep implementation alone that simply read()s the files.

Re:

[delire]

Agreed, the article is incorrect regarding use of tar's -C switch.

This works: tar -C /tmp/foo -xvf /path/to/some.tar.gz

This doesn't: tar xvf -C /tmp/foo /path/to/some.tar.gz

I agree that '&&' is a better practice to encourage than ';'. Chaining commands without heeding their exit status can make a real mess and cause much confusion.

Aside from that I don't think it's a bad top-ten. I think there should've been something in there about command completion. If productivity is important to this

Re:

[shird]

I think whats really missing in the tar command is the '-z' due to it being a .gz

Re:

[delire]

It works anyway, try it. -z is redundant in this case.

Re:

[Haeleth]

I think whats really missing in the tar command is the '-z' due to it being a .gz

This article is about Unix, where there is no such option (see the specification [opengroup.org]).

You may be thinking of GNU tar. I suggest you look up what the acronym GNU stands for, with particular reference to the N.

(This is not to dispute that the example given in the article is broken.)

Re:

[lahi]

Does GNU tar have an embedded EMACS as well? (or will GNU tar users have to wait for the next version?)

Well, at least Michael Stutz compensated for the lacking -z in tar by adding a superfluous -1 option to ls at the beginning of a pipe. I think most ls implementations check whether stdout goes to a terminal to decide whether output should be columnized.

-Lasse

Re:This article...

[treat]

You're the only one who hasn't mentioned xargs -0. I think it's important to elaborate on this. You should never do "find | xargs" or "find | cpio", you should always do "find -print0 | xargs -0" and find -print0 | cpio -0". The former will break if filenames have spaces or newlines in them. You break xargs if filenames have quotes, backslashes, or spaces in them. I never come across a large data set where you can do find | xargs without the -0 options.

If you are encountering data created by untrusted users, don't forget the strange consequences of filenames that contain newlines.

Failing to use -0 is dangerous malpractice.

Very helpful

[EvanED]

I really like this example:

~ $ time grep and tmp/a/longfile.txt | wc -l
2811

real 0m0.097s
user 0m0.006s
sys 0m0.032s
~ $ time grep -c and tmp/a/longfile.txt
2811

real 0m0.013s
user 0m0.006s
sys 0m0.005s

I am so glad that he showed what a difference can make, because I was *really* getting annoyed at having to wait that extra .084 seconds.

Re:

[trip11]

There is no such thing as .084 seconds. Surely you mean .084 hours.

More to the point

[metamatic]

I'll gladly trade 0.1 seconds of CPU time if it means I don't need to remember an extra command line option for the wc command.

That's why I'm of the v7 Unix philosophy. Make a tool that does one thing, and does it well.

If I later need to optimize for speed, then I'll read the man page.

Re:

[martin-boundary]

Especially since that example doesn't account for filesystem caching effects. There's no way of knowing if the bulk of the gain is because of the changed command or because the file is already in RAM, some background process was running, etc.

When timing commands, it's best to repeat the command several times and see if the times change significantly.

Re:

[canuck57]

I am so glad that he showed what a difference can make, because I was *really* getting annoyed at having to wait that extra .084 seconds.

But if in perl, python or java??? Worse yet, for for loop over 1 million files?

Things I had to learn the hard way

[gd23ka]

1. Don't rm with an absolute path because you could easily

#rm -r -f / tmp/dir

when "all" you wanted was

#rm -r -f /tmp/dir

instead do this:

#(cd /tmp ; rm -r -f dir)

or even better use sudo if you have it:

$(cd /tmp ; sudo rm -r -f dir)

2. When logged on as root or when using sudo on a production system think things over
at least twice before hitting enter.

3. Make sure at all times you're on the right machine, logged on as the right user in the right directory.
Set up your shell prompt to look like this user@host /path$

Re:

[SharpFang]

Also DO remember 'su -' changes working directory.

% cd ~
% cd somejunk
% ls
just junk, junk...
% rm -Rf *
somejunk: permission denied
% su -
Password:
# rm -Rf *^H^H^H^H^H^H^H^Hpwd
/

Re:

[sarathmenon]

you could easily
#rm -r -f / tmp/dir
when "all" you wanted was
#rm -r -f /tmp/dir

You are forgetting one thing - there's no solution for stupidity and lack of common sense. While tips like these are generally useful, the person who's going to screw up a system will ignore you, and the zillion other tips that people have taken efforts to write. I've seen people who've run a rm -rf /bin to clean the recycle bin and then wonder what happened.

Its hard to bring in any improvement in the average unix admin,

Re:

[maxwell demon]

When I do an rm which removes more than a single file, I quite often first type it prefixed with echo, look at the output, and after having verified that it is really what I want I go back in the command history, remove the echo and press enter. This is especially good with wildcards, which unintentionally might match something I didn't intend to remove.

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[TeknoHog]

On my keyboard layout I need shift to type a slash

This is exactly why I prefer the US/UK keyboard for unix stuff, even if I live in Finland, Finland, Finland, the country where slash needs a shift.

There are similar annoyances with a few other punctuation characters in the Fi keyboard. Besides, it's easy to setxkbmap when you need to type with your native language.

Re:

[account_deleted]

Comment removed based on user account deletion

absolute drivel

[Anonymous Coward]

This is, without a doubt, the most worthless article I have ever seen, both on Slashdot and on ibm.com, of which I thought better. It is not that the article is boring, but that it is factually incorrect in some places.

"the only excuse to define directories individually was that your mkdir implementation did not support this option, but this is no longer true on most systems. IBM, AIX®, mkdir, GNU mkdir, and others that conform to the Single UNIX Specification now have this option."

This is nonsense. The expansion of the path components in the {braces} is not a function of mkdir(1), but of the shell, and how its argument expansion is configured. I cannot believe that anyone "with 20 years of experience" is brazenly quoting names of standards in an effort to give his ramblings an air of credibility. Actually, wait a minute...

Another bad usage pattern is moving a .tar archive file to a certain directory because it happens to be the directory you want to extract it ...

Better is to check what's in the archive before extracting it in case some inconsiderate fool has failed to put a top-level directory in it.

His research interests include digital publishing and the future of the book.

Let me give you a couple [lotfi.net] of hints [ginpetty.com].

Re:

[portnoy]

He's not talking about the braces construct, he's talking about the -p option to mkdir. Just because you can't read doesn't mean the article is factually incorrect.

Mostly BS

[gweihir]

I don't know whether it is complete BS, as I stopped reading after the 3rd or 4th "good habit". Maybe AIX has an exceedingly slow ''cat'', Linux does not. Maybe the other stuff is valid in some OSes, but then I have not encounterd them. By the same token, really important stuff is missing completely. This looks a bit like a Unix-noveces initial list of things he/she thought about in the first week. Really not worth your time.

Anal Unix Guy

[jellomizer]

The title should be 10 Good Unix Hints. Not Habits.

These things he is stating is like when an other Unix guy is looking over your shoulder and you type in a command that works perfectly fine but has to correct you to do it his way, to save the extra millisecond in CPU while wasting about 5 Seconds to retype the command.

Secondly depending what you are doing on Unix you may or may not need to use these commands often enough to remember all the parameters for every tool. like the mkdir -p, While I know mkd

Re:

[jgrahn]

The title should be 10 Good Unix Hints. Not Habits.

Yes -- and habits is what people desperately need. The people I know primarily need three habits: RTFM when they don't understand something; adjusting their behavior based on the FM; and managing their use of the current directory (i.e. you don't have to cd into a directory to use a file which lives there).

Re:

[TeknoHog]

like the mkdir -p, While I know mkdir could create the entire directory Tree with some parameter, I just never bothered using it, or looking it up.

Interesting... I'm used to using mkdirhier for the same effect. I thought it was simply an alias to mkdir -p, but it turns out to be a shell script (part of X11 imake package) that uses plain mkdir. I better update my scripts to the more portable version now...

Eh? There's more to Unix than shell scripting

[Taagehornet]

"10 good habits that improve your UNIX command line efficiency" would probably have been a better title.

The title did however bring back fond memories of Eric Raymond's The Art of Unix Programming [catb.org]. The book is available online, and if you were hoping for something a bit more substantial as well, then the section Basics of the Unix Philosophy [catb.org] might be worth a read.

Re:

[Russ Nelson]

Yes, some idiots (e.g. Jim Thompson) claim that Eric isn't a good programmer, and yet he somehow has gotten the insight to write this book. Wonder of wonders!

tar comment

[thomasa]

In their example with tar they did

tar xvf

without the dash. (E.g., tar -xvf)

While that does work, I prefer to add
the dash as it makes it more consistent
with the other commands. So I consider
that a bad example. tar is one of the
older commands like dd that have weird
command line syntax.

Actually useful hints

[Artraze]

As has been pointed out, this article is riddled with errors. It's also not very interesting. So in the interest of perhaps actually providing some interesting tips:

In scripts, prefix dangersous commands with an 'echo' for a test run (So you can catch all those rm -rf /).

Single quotes are the best quotes for plain strings. The only reasion to use double quotes is if you need to quote a variable or a single quote.

Completion is fun, but using wildcards is more flexible (though you'll only want to use benign commans like cd, less, etc):
    nano /etc/modules.autoload.d/kernel-2.6
    nano /etc/m*a*d/*6

Note that the use of subpaths reduces the amount of flexibility.
    cd /etc/m* -> /etc/mail
    cd /etc/m*d -> /etc/modules.d
    nano /*/m*/*6 -> /etc/modules.autoload.d/kernel-2.6, and /etc/modules.d/i386 (not quite!)

Finally, as a comment for the article, using:
    test -e $DIR || mkdir -p $DIR
is much better than their suggestion and probaly faster anyway. Though I'd just do "mkdir -p $DIR" and maybe "&>/dev/null" under most circumstances anyway.

That's all I can think of at this point. Anyone else have tips?

No actual habits in the article

[xyloplax]

Here are some more important general IT rules (Unix rules can easily be OS and version dependent and frequently come from usage in YOUR environment)

• Copy before edit
• Tape backup before delete/decommission
• READ YOUR COMMAND before hitting return
• Check where things are symlinked to
• Echo in your scripts instead of destructive commands as a test run
• Test your changes on a lesser-importance box
• Use proper Change Control procedures
• Cover your ass and capture your terminal output
• When taking something out of service, turn it off for a few days/weeks before deleting/purging it

Some good, some bad, some plain pointless.

[Junta]

I'm evaluating the tips based on them being prescriptions for things to do in interactive shell behavior, since that seems to be the theme. Writing scripts changes the situation to make some tips valuable. My number one tip as a response to these is don't try to be too clever (particularly when the biggest benefit of the approach is to say 'look how clever it is!'). Maybe it's because I don't work in a vacuum and all too many times have been called in to clean up where an administrator tried to do something too complicated for their understanding.

mkdir -p is a convenience people should be aware of, but telling people to start getting overly creative with the shell expansion behavior is asking for mistakes/trouble. Also, having a mkdirhier script in case the example isn't supported on all shells is an indication that you shouldn't get overly cozy if you are going to be dealing with a lot of different systems/users with different default shells. The amount of time a lot of people take to figure out the 'clever' way in terms of how to phrase the expansion so the shell will expand it right is often longer than just typing the two lines more that the less thought takes. Not saying this isn't useful, but in my experience too many people mess things up too frequently or take too long to think up the expressions to tell them trying to be clever ends up taking more time than they think they are saving.

Change the path instead of the archive is not that dire to do normally, but if you avoid it, to me it's just easier to be in the target directly and use full path to the archive.

On combining commands, I second that ; can be dangerous and && as a default will make the chain more ready to break, but again I say not trying to be so clever as to put all you want on one line. Some things go wrong that aren't reflected in return codes, doing it one at a time let's you think of those. True, though, that the && never assume the first command works, while your fingers may keep moving and hit enter on next command before your brain realizes the command failed, so && may have merit, but then again taking your time may have more merit.

On the quotation thing, true enough, you must understand how quoting works to do remotely complex things, particularly nested circumstances (i.e. ssh to a system to run a command, where the output will be parsed by two shells.)

On the breaking up long lines thing, in a shell script it may be more necessary, but on an interactive command line it could also indicate you are trying too hard to do things in one chunk. I admit sometimes it does get too wide, but particularly less experienced admins should consider if there were a simpler way to do it in smaller chunks they won't screw up.

Grouping commands is important to know, and harmless (better than repeating the same pipe over and over and more powerful).

I will say xargs is way way over-rated. Too many people, particularly dealing with directory trees containing spaces, get into trouble piping the output of find into anything when IFS causes something like "/tmp/Monthly Report" to be parsed as two different files. find has a competent filtering mechanism (-type, -iname, -name, etc...) and it's own -exec. find is well aware of the state of each file. You could assign IFS to try to avoid it, but using find's built-ins where possible alleviates it.

When you are talking about interactive shell operation, picking the .01s instead of the .09s operation is a bad example. He could have set up a
much larger demonstration that would have been useful, but this just makes people mock the example. In any event, this seems like an okay thing to convey, but I dunno if it would've made my top 10.

Probably a more valid point about using awk, and a common trap I do see people stuck in.

On piping cat, that seems like more an annoyance than anything constructive. Some people use the cat | grep construct because it is so unambiguou

bad ibm no cookie

[illuminatedwax]

Great, IBM, way to ignore the dreaded "xargs" security bug! Seriously, IBM notices some kind of obscure danger about underscores, but completely ignores the fact that xargs separates arguments by newlines??

Let's say I'm a sysadmin and I'm running as root, trying to remove all the files in the /tmp directory by a certain user for some reason:
find /tmp -user 1001 | xargs rm

User 1001 has a directory in /tmp called "haxor\n". Inside there he puts another directory "etc" and inside there he puts a file called "passwd."

Can you guess what happens?
find prints: /tmp/tmp43cc91 /tmp/haxor /tmp/haxor /etc/passwd

xargs sees: ["/tmp/tmp43cc91","/tmp/haxor","","/tmp/haxor","/e tc/passwd"]
Oops!! You just hosed your system!

The correct way to use xargs is to use the -0 switch, which will separate the input by null characters, which cannot appear in filenames. find has a handy -print0 option which will output the correct output:

find /tmp -user 1001 -print0 | xargs -0 rm

And your system is safe.

Incorrect benchmarks!

[Vince]

Does anybody else notice these benchmarks are flawed? For an article discussing the shell, we should know that in this first benchmark, time is only counting the execution time of grep, and not wc, and is thus undercounting how much CPU time is actually used. How about a neat shell trick to correctly run that benchmark?

> ~ $ time grep and tmp/a/longfile.txt | wc -l
> 2811
>
> real 0m0.097s
> user 0m0.006s
> sys 0m0.032s
> ~ $ time grep -c and tmp/a/longfile.txt
> 2811
>
> real 0m0.013s
> user 0m0.006s
> sys 0m0.005s
> ~ $

Re:Don't use shell

[Anonymous Coward]

or even better -- use perl.

Re:

[tgv]

Someone mod up parent and grandparent, PLEASE!

Anyone?

Hello...???

Re:

[Anonymous Coward]

Someone mod up parent and grandparent, PLEASE!

No, don't mod up anybody in this thread. Perl and Python are abominations. Pure, unadulterated Bourne shell is for the true, seasoned *nix user. Just like Java is an answer to a question nobody asked in the GUI world, so too is Perl and Python in the command line world.

Re:

[dheera]

Yuck, I never use bash scripts. I always use Perl scripts. I just do things like

#!/usr/bin/perl
system("blah");
system("blah");
if(perl code perl code) {
system("blah");
}
etc.

why?
1. because i can't remember the awful syntax of the bash if statement. isn't it something like
if[[""$X$$"" == ""$Y""]];; ... fi ?
2. how about accepting command line arguments in bash? in perl it's just $ARGV[0]. nice and simple and like C++ (except for the offset by one) so i don't want to have to bother le

Re:

[wertarbyte]

4. because bash seems to think that sometimes you use x and sometimes you use $x

What's so special about that? If you want the variable, use "x", if you want the value of it, use $x. $ tells the shell to evaluate the next token as an variable name and fetch its value.

Re:

[gmack]

2. how about accepting command line arguments in bash? in perl it's just $ARGV[0]. nice and simple and like C++ (except for the offset by one) so i don't want to have to bother learning another one.

Command line args? $1 $2 etc or $* for all of them.

Re:

[gmack]

It's all about using the right tool for the job. I tend not to use bash for anything complicated enough to require that many command line args. Come to think of it none of my C code ever requires 10 args either.

Re:Don't use shell

[Anonymous Coward]

OK, I agree. Please provide a concise Python script that unpacks a tarball (a .tar.gz or .tar.bz2 file), copies new files in to a said tarball, patches based on the contents of said new files, runs make from various directories in said extracted tarball, and then changes the name of the top-level directory created by the tarball to a new name and repacks the tarball.

Or a concise Python script that opens up a text file of URLs, and extracts the files listed in the URLs:

#!/bin/sh
for a in $( cat file | awk '{print "'\''" $0 "'\''"}' ) ; do
              wget $a
done


Python has it place, and is far better for medium to large projects, and projects where the code needs to be maintainable. Shell, however, works a lot better for automating UNIX tasks than Python does. Not to mention embedded systems: I can compile Busybox to have both a good shell and all of the commands that one would run from shell scripts (including grep, cut, sed, and, yes, awk) in only about 300k. A Python binary is about a megabyte big, and you need about ten megabytes to fit all of the libraries Python 2.4 comes with.

Re:Don't use shell

[Bazman]

If I've got a simple task to do (eg the text-file-of-URLS example) then I knock it up in shell script. By the time that simple task has feature-creeped up to more than 20 lines I start to wish I'd written it in Perl. So I rewrite. By the time that Perl script has crept up to more than 200 lines I start to wish it was written in Python. So I rewrite. By the time that Python script has crept up to 2000 lines I start to wish I'd farmed the job out to a team of programmers, and I give up caring what language its written in and make them do it as a web service. Then I write a small shell script to call their web service. When that shell script has feature-creeped up to more than 20 lines...

Re:Don't use shell

[Anonymous Coward]

This code is not pure shellscript : it uses awk and wget to get the job done...

A Python equivalent might be :

#!/usr/bin/python
import os
for a in file('filename').readlines():
  os.system('wget ' + a)

It's not that much longer, it's much easier to read and less error-prone (especially the awk part), and it uses fewer external utilities.

To me, the *only* advantage of shellscript is that it's the only language that you are sure to find on any Unix system.

Re:Don't use shell

[Haeleth]

This code is not pure shellscript : it uses awk and wget to get the job done...

That's what a pure shellscript is! The whole point of shell scripting is that you use the shell script as glue to tie together simple single-purpose utilities that come as standard with every flavour of Unix ever.*

To me, the *only* advantage of shellscript is that it's the only language that you are sure to find on any Unix system.

No shit, Sherlock! You have clearly never worked in a large organisation, where - believe it or not - you, as a standard user, do not actually get to insist that the already-overworked IT department jump through bureaucratic hoops to install your favourite bloated scripting language, unless you have a damn good business case for it. And probably not even then.

Hint: if the task you want that scripting language to accomplish is trivial to achieve with a simple shell script, you don't have a good business case.

* This doesn't apply to wget, obviously, but if your platform really has no standard alternative, you are more likely to persuade IT to install something small and simple like wget, fetch, curl, etc. than a complete programming environment like Python.

Re:Don't use shell

[JohnFluxx]

Wow, do you think you could be just a little bit more polite next time?

Re:

[multipartmixed]

Right on, brother.

And, for what it's worth, I once had to write an HTTP fetch utility using nothing but SYSVR4 bourne shell, and stuff that shipped stock with a bare-naked install of Solaris 2.5.1.

Why? I needed stuff. NOW. I couldn't wait for next week for an 'admin' to install a web browser, I would have missed my go-live date.

Oh, yes, the box also had an old version of pine on it. I remember, because after HTTP fetching whatever it was, I had to de-base64-it.

I also once wrote an LR(1) parser for arithmeti

Re:Don't use shell

[duguk]

Um, whats wrong with

wget -i filename

Or have I missed something?

Monkeyboi

Re:

[Fred_A]

Um, whats wrong with

wget -i filename

It's not a shellscript, that's what wrong with it. Please pay attention next time.

wget options to read urls from a file

[pikine]

According to man wget, you can use -i or --input-file options to read URLs from a file.

Re:

[swillden]

This can be worked around using the undocumented -e flag. What "-e" does is make it so the script stops running if a program called in the script (except in if and a couple of other places) fails (returns a non-zero exit code).

I think it's really bad that this flag hasn't been documented; it's one of those "inside secrets" that UNIX is nortorious for having. And, yes, Python has really great documentation.

From 'man bash':

-e Exit immediately if a simple command (see SHELL GRAMMAR above) exits with a non-zero status. The shell does not exit if the command that fails is part of the command list immediately following a while or until keyword, part of the test in an if statement, part of a &amp;&amp; or || list, or if the command's return value is being inverted via !. A trap on ERR, if set, is executed before the shell exits.

Seems reasonably well documented to me.

Re:

[martin-boundary]

Nonsense. The bourne shell family are languages with particular strengths, some of which simply don't have a counterpart in perl or python.

Try writing a moderately complex pipeline in perl using native popen() calls. Not only do you have to deal with lots of file descriptors explicitly, but you'll have an unmaintainable mess if you have frequently changing requirements, such as adding or removing extra filters at various places.

Always use the right tool for the job. If you want to do complex logic, by a

Re:

[canuck57]

When possible, avoid shell scripting: the language is flawed by design. Use a modern programming language, such as Python.

How did you get mod'ed up to 5 I don't know.

Python may not be on the system your managing.

Python, while not as heavy as Java, is an elephant in a china shop compared to shell script

Shell scripting is an essential basic admin skill, good admins don't need Python at all.

Java, Python, Perl -- sheshh... they have their places but not for everything. If any one tool could effectively d

Re:

[25albert]

Use a modern programming language, such as Python

I prefer postmodern languages such as Perl [perl.com]

Re:

[Timothy Brownawell]

What is Unix?

*nix is a highly modular component-based software system with a standard interface (flat byte streams) between components, and a basic set of standard components (given in the POSIX standard) that can be relied upon to always be present.

Re:

[netbuzz]

Yup, that's the phrase that jumps out at those of us who don't speak UNIX. Fun with piping cats on this blog post:
http://www.networkworld.com/community/?q=node/9764 [networkworld.com]

Re:

[AndroidCat]

Articles on terminating zombie children are always a treat too.

Re:

[Jesus_666]

Maybe to have heaps of cryptic-looking data scrolling across the terminal window to fool your boss into thinking that you're just playing Tetris because you're waiting for this incredigby important script to finish crunching all that data...

Re:

[Scorpio]

for i in *.JPG ; do mv $i `basename $i .JPG`.jpg ; done

Re:

[value_added]

$ for i in *JPG ; do mv $i ${i/JPG/jpg} ; done

This isn't the first time I've seen this, but it will result in a file MYJPG.JPG being called MYjpg.JPG, ${i//JPG/jpg} would be better as at least the it would end up with the .jpg at the end, but ${i%.JPG}.jpg would be best.

Again, there's lots of ways to do it. To use the trivial JPG -> jpg, example, yes, you're correct in that using the shortest match at the end would be a better approach (excluding other issues). I just wanted to illustrate the redundant

Re:

[Chandon Seldon]

Why is the perl script so hard? The command line would have a bunch of sed going on, whereas the perl script only requires running perl.

I'm guessing... perl -e 'for(`ls`) { chomp; $n = lc $_; system("mv $_ $n"); }'

Re:lowercase uppercase

[multipartmixed]

Oh, for the love of God, stop bitching about how hard this is to do under UNIX without package x.y.z installed and imagine doing it under Mac OS9 or Windows.

Break it down into its constituent parts: Iterate, rename. Whoo! Simple now, eh?

# cd cameraDir
# find . -type f -prune | while read file
> do
> mv "$file" "`echo \"$filename\" | tr '[A-Z]' '[a-z]'`"
> done
#

... should do the trick. If you don't have tr, you can use sed with the y command. But I can't think of the last time I saw a box w/o tr.

Re:

[thomasa]

While it is not the same, it is the best thing on Windows that is close to UNIX.

Re:

[canuck57]

Everyone is talking a different language. No wonder why nobody but nobody uses this operating system.

To biased and pig headed to.

I always stress to new to xNIX types, learn the basics first before getting wrapped up in a perl/java/python or some other cult. Often these other languages are like taking elephants into china shops. And besides, they are not always on a system -- ditto vi versus emacs.