VAR=
VAR=”" (Same as above)
VAR=” “, and
VAR=”something” (which is same as #3, VAR=” “)

The solution has to defend against all these options.

Assuming (Big assumption) that VAR’s value is always greater than a single alpha-numeric, one can test for the length of VAR.

$ export SOMEVAR=
$ defined “${SOMEVAR} || echo “notset”
notset

$ defined “${SOMEVAR}” || echo “notset”
notset
$ export SOMEVAR=”sometext”
$ defined “${SOMEVAR}” || echo “notset”
$ export SOMEVAR=
$ defined “${SOMEVAR}” || echo “notset”

It’ll show “notset” when a variable is defined but empty.

-z only tells you whether a variable has content, not whether it’s set or not.

defined() {
[[ ! -z "${1}" ]]
}

$ defined “${SOMEVAR}” || echo “notset”
notset
$ export SOMEVAR=”sometext”
$ defined “${SOMEVAR}” || echo “notset”

defined()
{
[ "${!1-one} == "${!1-two}" ]
}

By way of explanation, it uses three bash expansions. First, you get the arguments to a function via ${1}. Second, you get indirection by adding the bang. ${!a} returns the value of the variable named by a. Third, it uses default values, just like your original code. I usually use “one” and “two” instead of “X” and “Y”. It doesn’t mater, so long as the values are different.

(I wasn’t the previous commiter. I’m very indebted to both the original post and the previous comment. It was very helpful to me!)

# 10:56:58 garrod@sioprism 2.15 $ cd ~ && set | grep BASH
BASH=/bin/bash
BASH_VERSINFO=([0]=”2″ [1]=”05b” [2]=”0″ [3]=”1″ [4]=”release” [5]=”ia64-redhat-linux-gnu”)
BASH_VERSION=’2.05b.0(1)-release’

Example:
$ defined FOO || echo notset
notset
$ FOO=X
$ defined FOO || echo notset
$ FOO=
$ defined FOO || echo notset
$ unset FOO
$ defined FOO || echo notset
notset
$ defined BASH_VERSION[2] || echo notset
$ defined BASH_VERSION[20] || echo notset
notset
$