mjhjmtu

If I had a script which sets variables read-only to some odd values, and sets errexit because of other unsafe operations:

#!/bin/bash
set -e 
declare -r NOTIFY=$(case "$OS" in (macosx) echo macos_notify ;; (linux) echo linux_notify ;; (*) echo : ;; esac)
declare -r SAY=_say
declare -r SUM=_sum
declare -r MV=_mv
set +e 

And I source it to get the definitions, the second time because it's in development:

$ . s.bash 

$ . s.bash 
bash: declare: NOTIFY: readonly variable
Exited

Normally declare -r EXISTING_VAR would neither stop the script nor remove the old, working definition of EXISTING_VAR.

But with errexit, assigning to an existing variable is understandably a failure. The easy options are to remove -r or use set +e for that part of the script.

Barring those, is it possible to write a Bash function to take the place of declare -r but not re-assign if the name already exists?

I tried:

# arg #1: var name, #2: value
set_var_once () 
 # test whether the variable with the 
 # name stored in $1 exists 
 if [[ -z "$!1" ]] 
 then # if it doesn't, set it
 declare -r $1=$2
 fi

I also tried things along the lines of eval "declare -r $1=$(eval $2)", it feels like eval is required somewhere here but I'm not sure where.

All of the versions of set_var_once result in not setting the variable they should.