`unset` requires the variable name, not its expansion

Problem

`unset` takes a variable *name* as its argument, not a value to be expanded. Writing `unset "$var"` first expands `$var` to its current value and then asks `unset` to remove the variable whose *name* matches that value — which is essentially never what you want, and usually unsets nothing (or the wrong variable, if the value happens to coincide with a real variable name). This mistake is subtle because bare `unset` still "runs successfully" with no error even when the argument does not resolve to any meaningful variable name, so the bug is silent unless you specifically check whether the intended variable still exists afterward.

Bad

myvar=hello
unset "$myvar"   # tries to unset a variable literally named "hello"

Good

myvar=hello
unset myvar      # unsets the variable named myvar
# for array elements:
unset 'arr[3]'

Explanation

Pass the bare variable name to `unset` (`unset myvar`), not its expansion. This also applies to array elements and associative array keys, where the index/key should be included literally, e.g. `unset 'arr[index]'`, quoted to prevent glob expansion of the brackets rather than to substitute a value.

Related

SC2178

SC2034

SC2154

When It Matters

unset takes variable names, not values, so unset "$var" removes whatever variable is named by the contents of var — and unset arr[0] with an unquoted index is subject to globbing, so a file named arr0 in the working directory changes what gets unset. The common form unset ${arr[@]} deletes variables named after the array’s contents, which is either a no-op or a very confusing bug.

Second Example

Exceptions

Indirect unset through a name held in a variable is occasionally intentional in generic library code. Make it obvious with a comment, and prefer namerefs (declare -n) in Bash 4.3 and later, which express indirection explicitly.

Faq