I think the answer you are after is implied (if not stated) by Vinko's answer, though it is not spelled out simply. To distinguish whether VAR is set but empty or not set, you can use:
if [ -z "${VAR+xxx}" ]; then echo VAR is not set at all; fi
if [ -z "$VAR" ] && [ "${VAR+xxx}" = "xxx" ]; then echo VAR is set but empty; fi
You probably can combine the two tests on the second line into one with:
if [ -z "$VAR" -a "${VAR+xxx}" = "xxx" ]; then echo VAR is set but empty; fi
However, if you read the documentation for Autoconf, you'll find that they do not recommend combining terms with '-a
' and do recommend using separate simple tests combined with &&
. I've not encountered a system where there is a problem; that doesn't mean they didn't used to exist (but they are probably extremely rare these days, even if they weren't as rare in the distant past).
You can find the details of these, and other related shell parameter expansions, the test
or [
command and conditional expressions in the Bash manual.
I was recently asked by email about this answer with the question:
You use two tests, and I understand the second one well, but not the first one. More precisely I don't understand the need for variable expansion
if [ -z "${VAR+xxx}" ]; then echo VAR is not set at all; fi
Wouldn't this accomplish the same?
if [ -z "${VAR}" ]; then echo VAR is not set at all; fi
Fair question - the answer is 'No, your simpler alternative does not do the same thing'.
Suppose I write this before your test:
VAR=
Your test will say "VAR is not set at all", but mine will say (by implication because it echoes nothing) "VAR is set but its value might be empty". Try this script:
(
unset VAR
if [ -z "${VAR+xxx}" ]; then echo JL:1 VAR is not set at all; fi
if [ -z "${VAR}" ]; then echo MP:1 VAR is not set at all; fi
VAR=
if [ -z "${VAR+xxx}" ]; then echo JL:2 VAR is not set at all; fi
if [ -z "${VAR}" ]; then echo MP:2 VAR is not set at all; fi
)
The output is:
JL:1 VAR is not set at all
MP:1 VAR is not set at all
MP:2 VAR is not set at all
In the second pair of tests, the variable is set, but it is set to the empty value. This is the distinction that the ${VAR=value}
and ${VAR:=value}
notations make. Ditto for ${VAR-value}
and ${VAR:-value}
, and ${VAR+value}
and ${VAR:+value}
, and so on.
As Gili points out in his answer, if you run bash
with the set -o nounset
option, then the basic answer above fails with unbound variable
. It is easily remedied:
if [ -z "${VAR+xxx}" ]; then echo VAR is not set at all; fi
if [ -z "${VAR-}" ] && [ "${VAR+xxx}" = "xxx" ]; then echo VAR is set but empty; fi
Or you could cancel the set -o nounset
option with set +u
(set -u
being equivalent to set -o nounset
).