HPR4607: UNIX Curio #3 - basename and dirname

This show has been flagged as Clean by the host.

This series is dedicated to exploring little-known—and occasionally useful—trinkets lurking in the dusty corners of UNIX-like operating systems.

Hopefully it doesn't seem like I'm picking on
Linux Journal
, but like UNIX Curio #1 (HPR4587), this column has been inspired by
an article of theirs


¹
. The author was demonstrating a clever
bash
script that would take a filename and send the file to standard output or, if the filename ended in .gz, decompress it and send the result to standard output. Slightly rearranged, he had:

F=`echo $1 | perl -pe 's/\.gz$//'`

if [[ -f $F ]] ; then

  cat $F

elif [[ -f $F.gz ]] ; then

  gunzip -c $F

fi

He took some heat on the web site and in letters to the magazine for cranking up a whole Perl interpreter just to chop the .gz off the end of a filename. Our curio for today is a standard UNIX utility made for just this purpose called


basename




²
. Along with its brother


dirname




³
, it is used to pull apart pathnames to get the part you want. What
basename
does is remove any leading path on the name given to it, and if a suffix is specified as well, removes that also. If a directory path with a trailing slash is given, it returns the last part with no slashes. Here are some examples:

$ basename /bin/gzip

gzip

$ basename /bin/gzip .so

gzip

$ basename /usr/lib/libz.so .so

libz

$ basename /usr/lib/

lib

The counterpart,
dirname
, does essentially the opposite. It removes the last part of the pathname and returns a directory name (with no trailing slash):

$ dirname /usr/lib/libz.so

/usr/lib

$ dirname /usr/lib/

/usr

$ dirname file_in_this_dir

.

So we can replace the first line of the script up top with
F=`dirname $1`/`basename $1 .gz`
, get the same result, and be sure it will work on any UNIX-like system, no Perl necessary. The more observant among you may be thinking "
sed
could do that, too!" and you're right;
F=`echo $1 | sed 's/\.gz$//'`
also would work anywhere.

One might suspect that as a general-purpose text processor,
sed
would be slower than
basename
and
dirname
. To see how they compared, we ran each method against a randomly-generated list of 5,000 filenames. Turns out the critics were right, as Perl ran the longest at 59 seconds. Using
basename
/
dirname
took 44 seconds—a nice improvement, but
sed
blew past it at 34 seconds. Probably the fact that only one call to
sed
was needed versus two for
basename
and
dirname
made the difference.

Helpful suggestions in response to the article revealed a shell curio. You may have seen the brace syntax for parameters. For example, to show a filename $F with an "X" appended, you can't use
echo $FX
because that means a parameter named
FX
. Instead, you'd use
echo ${F}X
and the shell only interprets what's inside the braces as the parameter name.

Modifiers can also go inside the braces


⁴
and one of these, %, is just what we need to chop off that extension. This works in
bash
,
zsh
, and any shell conforming to the current POSIX standard, but not
csh
and friends or older implementations of the Bourne shell. We can rewrite the first line of the original script as simply
F=${1%.gz}
and forgo any outside utilities. Performance? Under half a second to process those 5,000 filenames. Not bad at all.

References:

1. 
   
   
   Treating Compressed and Uncompressed Data Sources the Same
   https://www.linuxjournal.com/content/treating-compressed-and-uncompressed-data-sources-same



2. 
   
   
   Basename specification
   https://pubs.opengroup.org/onlinepubs/009695399/utilities/basename.html



3. 
   
   
   Dirname specification
   https://pubs.opengroup.org/onlinepubs/009695399/utilities/dirname.html



4. 
   
   
   Shell Command Language: Parameter Expansion
   https://pubs.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_06_02

This article was originally written in July 2010. The podcast episode was recorded in March 2026.

Provide feedback on this episode.