Trimming Whitespace in bash

You need to trim leading and/or trailing whitespace from lines for fields of data.

These solutions rely on a bash-specific treatment of read and $REPLY.

See the end of the discussion for an alternate solution.
First, we’ll show a file with some leading and trailing whitespace. Note we add ~~ to show the whitespace.

Note the ➝ denotes a literal tab character in the output:

To trim both leading and trailing whitespace use $IFS add the built-in REPLY variable (see the discussion for why this works):

To trim only leading or only trailing spaces, use a simple pattern match:

Trimming only leading or only trailing whitespace (including tab) is a bit more complicated:

OK, at this point you are probably looking at these lines and wondering how we’re going to make this comprehensible.

It turns out there’s a simple, if subtle explanation. Here we go. The first example used the default $REPLY variable that read uses when you do not supply your own variable name(s).

Chet Ramey (maintainer of bash) made a design decision that, “[if] there are no variables, save the text of the line read to the variable $REPLY [unchanged, else parse using $IFS].”

But when we supply one or more variable names to read, it does parse the input, using the values in $IFS (which are space, tab, and newline by default).

One step of that parsing process is to trim leading and trailing whitespace—just what we want:

To trim leading or trailing (but not both) spaces is easy using the ${##} or ${%%} operators.

But covering tabs is a little harder. If we had only tabs, we could use the ${##} or ${%%} operators and insert literal tabs using the Ctrl-V Ctrl-I key sequence.

But that’s risky since it’s probable there’s a mix of spaces and tabs, and some text editors or unwary users may strip out the tabs.

So we turn on extended globbing and use a character class to make our intent clear.

The [:space:] character class would work without extglob, but we need to say “one or more occurrences” using +( ) or else it will trim a single space or tabs, but not multiples or both on the same line.

Here’s a different take, exploiting the same $IFS parsing, but to parse out fields (or words) instead of records (or lines):

Finally, although the original solutions rely on Chet’s design decision about read and $REPLY, this solution does not: