What does this expr do in a shell script?

Question

What does the following do exactly?

newProg=`expr "${newProg}" : ".* -> \(.*\)$"` 
if expr "x${newProg}" : 'x/' >/dev/null; then 
 prog="${newProg}" 
else 
 progdir=`dirname "${prog}"` 
 prog="${progdir}/${newProg}" 
fi 

Answer 1

The first line uses expr to:

• Match the value of $newProg against the basic regular expression .* -> \(.*\)$ (anchored on both ends). That's anything with -> in it.
• Set newProg to the part of the original value matched by the capturing subexpression written in parentheses (the part after ->).

The greedy matching semantics are such that this always captures and outputs the last such match, so there is never a -> in the output.

This would, for example, parse the output of ls -l on a symlink to find the target (given well-behaved filenames, etc, usual caveats).

The second line uses expr again to check whether the new value of newProg begins with a /, i.e., it represents an absolute path.

In the case where it does, prog is just set to that value. Otherwise, prog is set to a computed relative path using dirname on $prog, a variable not defined in this snippet. The effect will presumably be to construct a path relative to that original location.

---

If you are only targeting a more capable shell or environment than pure POSIX, there is likely to be a more reliable way of doing this than expr. readlink is a very commonly-available command also, which would save on parsing ls.

• cas: +1. the only thing i'd add to that is that parsing the output of ls is always a bad idea, and using expr to parse ls is also inefficient and leads to ugly, difficult to read shell code....especially when every version of test aka [ in existence has -h and/or -L options to test whether a file is a symlink or not.
• cas: A GNU readlink-based alternative to this could look like [ -L "$newProg" ] && prog="$(readlink -m "$newprog")" (1 line rather than 7). And if $progdir is required (i.e. not just a temporary throwaway variable), append && progdir="$(dirname "$prog")" to that line. BTW, the [ -L ... ] && test isn't required because running readlink -m on a non-symlink filename just returns that filename (with the full path).