Why does local -n work on array variables when the manual explicitly says it does not? Is the manual wrong? Is this relying on some sort of bash "undefined behavior?" Is the manual out-of-date? Am I missing something?
Here is the information I'm looking at from the bash manual:
Run man bash and search for local [ using the regular expression search pattern local \[. It says (emphasis added):
local [option] [name[=value] ... | - ]
For each argument, a local variable named name is created, and assigned value. The
option can be any of the options accepted by declare.
(See also help local).
So, the options that can be passed to the bash local built-in command are the same as those of declare. Let's find what options can be passed to declare:
Run man bash and search for declare [ using the regular expression search pattern declare \[. Under the -n entry there for declare [-aAfFgilnrtux] [-p] [name[=value] ...] you'll see (emphasis added):
-n Give each name the nameref attribute, making it a name reference to another
variable. That other variable is defined by the value of name. All references, assignments, and attribute modifications to name, except those using or
changing the -n attribute itself, are performed on the variable referenced by
name's value. The nameref attribute cannot be applied to array variables.
(See also help declare).
So, despite it saying "The nameref attribute cannot be applied to array variables,", it does work just fine on array variables!
To prove that, here is a demo of it working fine for regular bash arrays:
function foo {
# declare a local **reference variable** (hence `-n`) named `data_ref`
# which is a reference to the value stored in the first parameter
# passed in
local -n data_ref="$1"
echo "${data_ref[0]}"
echo "${data_ref[1]}"
}
# declare a regular bash "indexed" array
declare -a data
data+=("Fred Flintstone")
data+=("Barney Rubble")
foo "data"
Sample output:
Fred Flintstone
Barney Rubble
...and here is it working just fine on associative bash arrays (ie: bash hash tables, "dictionaries", or "unordered maps"):
function foo {
# declare a local **reference variable** (hence `-n`) named `data_ref`
# which is a reference to the value stored in the first parameter
# passed in
local -n data_ref="$1"
echo "${data_ref["a"]}"
echo "${data_ref["b"]}"
}
# declare a bash associative array
declare -A data
data["a"]="Fred Flintstone"
data["b"]="Barney Rubble"
foo "data"
Sample output:
Fred Flintstone
Barney Rubble
My bash --version is 4.4.20(1)-release, on Linux Ubuntu 18.04:
GNU bash, version 4.4.20(1)-release (x86_64-pc-linux-gnu)
Copyright (C) 2016 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software; you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.
References:
I modified the above code samples from #Todd Lehman's answer here: How to pass an associative array as argument to a function in Bash?
Note: in case anyone is searching for it, this question itself also answers the questions: "How do I pass a bash array as a parameter to a function?" and "How do I pass a bash associative array as a parameter to a function?"
You are applying -n to a scalar variable, and then have that nameref point to an array. This is fine.
What you can't do is apply -n to an array variable to create an array of namerefs:
# Declare two variables
declare foo=42 bar=1337
# Make an array of namerefs
declare -n array=(foo bar) # ERROR
# ${array[0]} is foo so this should print $foo, i.e. 42?
echo "${array[0]}"
If you try, Bash stops you:
bash: declare: array: reference variable cannot be an array
There are multiple ways to get a result from a function in a bash script, one is to use reference variables like local -n out_ref="$1, which is also my preferred way.
My bash version is:
GNU bash, Version 5.0.3(1)-release
Recently, one of my bash functions needed to produce a associative array as a result, like in this example code:
#!/bin/bash
testFunction() {
local -n out_ref="$1"
out_ref[name]="Fry"
out_ref[company]="Planet Express"
}
declare -A employee
testFunction employee
echo -e "employee[name]: ${employee[name]}"
echo -e "employee[company]: ${employee[company]}"
I declare the variable employee as an associative array with declare -A.
The output is:
employee[name]: Fry
employee[company]: Planet Express
If I remove the line declare -A employee, the output is:
employee[name]: Planet Express
employee[company]: Planet Express
Is there a way to move the declaration of the associative array into the function, so the user of that function does not need to do it beforehand?
Use declare -Ag "$1" inside the function, so that you declare employee as a global variable.
Without declare -A employee, employee is basically treated like an ordinary indexed array. In that context, name and employee are both undefined variables evaluated in an arithmetic context, so both evaluate to 0.
I'm not entirely sure how local -n is implemented (local -nA is not allowed to mark the reference as an associative array), but it appears that lacking a "real" associate array to assign to, the value of employee is just whatever the last value assigned to any key of the reference out_ref. (Probably the same as above: both name and company evaluate to 0 in the arithmetic context, so you are assigning to out_ref[0] twice (and because out_ref does not refer to an array-marked name, you are just assigning to out_ref itself.)
Expanding on what these better minds have provided, here's a functioning example that shows it in actual use.
A() {
declare -Ag "$1"
local -n ref="$1"
ref[A]='in A'
}
B() {
declare -Ag "$1"
local -n ref="$1"
ref[B]='in B'
}
A foo
B foo
declare -p foo
Saved and executed as a file named tst, here's the run and the output -
$: ./tst
declare -A foo=([A]="in A" [B]="in B" )
Note the "redeclaration" of the global variable in multiple functions does not impugn the data integrity.
I am confused about a bash script.
I have the following code:
function grep_search() {
magic_way_to_define_magic_variable_$1=`ls | tail -1`
echo $magic_variable_$1
}
I want to be able to create a variable name containing the first argument of the command and bearing the value of e.g. the last line of ls.
So to illustrate what I want:
$ ls | tail -1
stack-overflow.txt
$ grep_search() open_box
stack-overflow.txt
So, how should I define/declare $magic_way_to_define_magic_variable_$1 and how should I call it within the script?
I have tried eval, ${...}, \$${...}, but I am still confused.
I've been looking for better way of doing it recently. Associative array sounded like overkill for me. Look what I found:
suffix=bzz
declare prefix_$suffix=mystr
...and then...
varname=prefix_$suffix
echo ${!varname}
From the docs:
The ‘$’ character introduces parameter expansion, command substitution, or arithmetic expansion. ...
The basic form of parameter expansion is ${parameter}. The value of parameter is substituted. ...
If the first character of parameter is an exclamation point (!), and parameter is not a nameref, it introduces a level of indirection. Bash uses the value formed by expanding the rest of parameter as the new parameter; this is then expanded and that value is used in the rest of the expansion, rather than the expansion of the original parameter. This is known as indirect expansion. The value is subject to tilde expansion, parameter expansion, command substitution, and arithmetic expansion. ...
Use an associative array, with command names as keys.
# Requires bash 4, though
declare -A magic_variable=()
function grep_search() {
magic_variable[$1]=$( ls | tail -1 )
echo ${magic_variable[$1]}
}
If you can't use associative arrays (e.g., you must support bash 3), you can use declare to create dynamic variable names:
declare "magic_variable_$1=$(ls | tail -1)"
and use indirect parameter expansion to access the value.
var="magic_variable_$1"
echo "${!var}"
See BashFAQ: Indirection - Evaluating indirect/reference variables.
Beyond associative arrays, there are several ways of achieving dynamic variables in Bash. Note that all these techniques present risks, which are discussed at the end of this answer.
In the following examples I will assume that i=37 and that you want to alias the variable named var_37 whose initial value is lolilol.
Method 1. Using a “pointer” variable
You can simply store the name of the variable in an indirection variable, not unlike a C pointer. Bash then has a syntax for reading the aliased variable: ${!name} expands to the value of the variable whose name is the value of the variable name. You can think of it as a two-stage expansion: ${!name} expands to $var_37, which expands to lolilol.
name="var_$i"
echo "$name" # outputs “var_37”
echo "${!name}" # outputs “lolilol”
echo "${!name%lol}" # outputs “loli”
# etc.
Unfortunately, there is no counterpart syntax for modifying the aliased variable. Instead, you can achieve assignment with one of the following tricks.
1a. Assigning with eval
eval is evil, but is also the simplest and most portable way of achieving our goal. You have to carefully escape the right-hand side of the assignment, as it will be evaluated twice. An easy and systematic way of doing this is to evaluate the right-hand side beforehand (or to use printf %q).
And you should check manually that the left-hand side is a valid variable name, or a name with index (what if it was evil_code # ?). By contrast, all other methods below enforce it automatically.
# check that name is a valid variable name:
# note: this code does not support variable_name[index]
shopt -s globasciiranges
[[ "$name" == [a-zA-Z_]*([a-zA-Z_0-9]) ]] || exit
value='babibab'
eval "$name"='$value' # carefully escape the right-hand side!
echo "$var_37" # outputs “babibab”
Downsides:
does not check the validity of the variable name.
eval is evil.
eval is evil.
eval is evil.
1b. Assigning with read
The read builtin lets you assign values to a variable of which you give the name, a fact which can be exploited in conjunction with here-strings:
IFS= read -r -d '' "$name" <<< 'babibab'
echo "$var_37" # outputs “babibab\n”
The IFS part and the option -r make sure that the value is assigned as-is, while the option -d '' allows to assign multi-line values. Because of this last option, the command returns with an non-zero exit code.
Note that, since we are using a here-string, a newline character is appended to the value.
Downsides:
somewhat obscure;
returns with a non-zero exit code;
appends a newline to the value.
1c. Assigning with printf
Since Bash 3.1 (released 2005), the printf builtin can also assign its result to a variable whose name is given. By contrast with the previous solutions, it just works, no extra effort is needed to escape things, to prevent splitting and so on.
printf -v "$name" '%s' 'babibab'
echo "$var_37" # outputs “babibab”
Downsides:
Less portable (but, well).
Method 2. Using a “reference” variable
Since Bash 4.3 (released 2014), the declare builtin has an option -n for creating a variable which is a “name reference” to another variable, much like C++ references. Just as in Method 1, the reference stores the name of the aliased variable, but each time the reference is accessed (either for reading or assigning), Bash automatically resolves the indirection.
In addition, Bash has a special and very confusing syntax for getting the value of the reference itself, judge by yourself: ${!ref}.
declare -n ref="var_$i"
echo "${!ref}" # outputs “var_37”
echo "$ref" # outputs “lolilol”
ref='babibab'
echo "$var_37" # outputs “babibab”
This does not avoid the pitfalls explained below, but at least it makes the syntax straightforward.
Downsides:
Not portable.
Risks
All these aliasing techniques present several risks. The first one is executing arbitrary code each time you resolve the indirection (either for reading or for assigning). Indeed, instead of a scalar variable name, like var_37, you may as well alias an array subscript, like arr[42]. But Bash evaluates the contents of the square brackets each time it is needed, so aliasing arr[$(do_evil)] will have unexpected effects… As a consequence, only use these techniques when you control the provenance of the alias.
function guillemots {
declare -n var="$1"
var="«${var}»"
}
arr=( aaa bbb ccc )
guillemots 'arr[1]' # modifies the second cell of the array, as expected
guillemots 'arr[$(date>>date.out)1]' # writes twice into date.out
# (once when expanding var, once when assigning to it)
The second risk is creating a cyclic alias. As Bash variables are identified by their name and not by their scope, you may inadvertently create an alias to itself (while thinking it would alias a variable from an enclosing scope). This may happen in particular when using common variable names (like var). As a consequence, only use these techniques when you control the name of the aliased variable.
function guillemots {
# var is intended to be local to the function,
# aliasing a variable which comes from outside
declare -n var="$1"
var="«${var}»"
}
var='lolilol'
guillemots var # Bash warnings: “var: circular name reference”
echo "$var" # outputs anything!
Source:
BashFaq/006: How can I use variable variables (indirect variables, pointers, references) or associative arrays?
BashFAQ/048: eval command and security issues
Example below returns value of $name_of_var
var=name_of_var
echo $(eval echo "\$$var")
Use declare
There is no need on using prefixes like on other answers, neither arrays. Use just declare, double quotes, and parameter expansion.
I often use the following trick to parse argument lists contanining one to n arguments formatted as key=value otherkey=othervalue etc=etc, Like:
# brace expansion just to exemplify
for variable in {one=foo,two=bar,ninja=tip}
do
declare "${variable%=*}=${variable#*=}"
done
echo $one $two $ninja
# foo bar tip
But expanding the argv list like
for v in "$#"; do declare "${v%=*}=${v#*=}"; done
Extra tips
# parse argv's leading key=value parameters
for v in "$#"; do
case "$v" in ?*=?*) declare "${v%=*}=${v#*=}";; *) break;; esac
done
# consume argv's leading key=value parameters
while test $# -gt 0; do
case "$1" in ?*=?*) declare "${1%=*}=${1#*=}";; *) break;; esac
shift
done
Combining two highly rated answers here into a complete example that is hopefully useful and self-explanatory:
#!/bin/bash
intro="You know what,"
pet1="cat"
pet2="chicken"
pet3="cow"
pet4="dog"
pet5="pig"
# Setting and reading dynamic variables
for i in {1..5}; do
pet="pet$i"
declare "sentence$i=$intro I have a pet ${!pet} at home"
done
# Just reading dynamic variables
for i in {1..5}; do
sentence="sentence$i"
echo "${!sentence}"
done
echo
echo "Again, but reading regular variables:"
echo $sentence1
echo $sentence2
echo $sentence3
echo $sentence4
echo $sentence5
Output:
You know what, I have a pet cat at home
You know what, I have a pet chicken at home
You know what, I have a pet cow at home
You know what, I have a pet dog at home
You know what, I have a pet pig at home
Again, but reading regular variables:
You know what, I have a pet cat at home
You know what, I have a pet chicken at home
You know what, I have a pet cow at home
You know what, I have a pet dog at home
You know what, I have a pet pig at home
This will work too
my_country_code="green"
x="country"
eval z='$'my_"$x"_code
echo $z ## o/p: green
In your case
eval final_val='$'magic_way_to_define_magic_variable_"$1"
echo $final_val
This should work:
function grep_search() {
declare magic_variable_$1="$(ls | tail -1)"
echo "$(tmpvar=magic_variable_$1 && echo ${!tmpvar})"
}
grep_search var # calling grep_search with argument "var"
An extra method that doesn't rely on which shell/bash version you have is by using envsubst. For example:
newvar=$(echo '$magic_variable_'"${dynamic_part}" | envsubst)
For zsh (newers mac os versions), you should use
real_var="holaaaa"
aux_var="real_var"
echo ${(P)aux_var}
holaaaa
Instead of "!"
As per BashFAQ/006, you can use read with here string syntax for assigning indirect variables:
function grep_search() {
read "$1" <<<$(ls | tail -1);
}
Usage:
$ grep_search open_box
$ echo $open_box
stack-overflow.txt
Even though it's an old question, I still had some hard time with fetching dynamic variables names, while avoiding the eval (evil) command.
Solved it with declare -n which creates a reference to a dynamic value, this is especially useful in CI/CD processes, where the required secret names of the CI/CD service are not known until runtime. Here's how:
# Bash v4.3+
# -----------------------------------------------------------
# Secerts in CI/CD service, injected as environment variables
# AWS_ACCESS_KEY_ID_DEV, AWS_SECRET_ACCESS_KEY_DEV
# AWS_ACCESS_KEY_ID_STG, AWS_SECRET_ACCESS_KEY_STG
# -----------------------------------------------------------
# Environment variables injected by CI/CD service
# BRANCH_NAME="DEV"
# -----------------------------------------------------------
declare -n _AWS_ACCESS_KEY_ID_REF=AWS_ACCESS_KEY_ID_${BRANCH_NAME}
declare -n _AWS_SECRET_ACCESS_KEY_REF=AWS_SECRET_ACCESS_KEY_${BRANCH_NAME}
export AWS_ACCESS_KEY_ID=${_AWS_ACCESS_KEY_ID_REF}
export AWS_SECRET_ACCESS_KEY=${_AWS_SECRET_ACCESS_KEY_REF}
echo $AWS_ACCESS_KEY_ID $AWS_SECRET_ACCESS_KEY
aws s3 ls
Wow, most of the syntax is horrible! Here is one solution with some simpler syntax if you need to indirectly reference arrays:
#!/bin/bash
foo_1=(fff ddd) ;
foo_2=(ggg ccc) ;
for i in 1 2 ;
do
eval mine=( \${foo_$i[#]} ) ;
echo ${mine[#]}" " ;
done ;
For simpler use cases I recommend the syntax described in the Advanced Bash-Scripting Guide.
KISS approach:
a=1
c="bam"
let "$c$a"=4
echo $bam1
results in 4
I want to be able to create a variable name containing the first argument of the command
script.sh file:
#!/usr/bin/env bash
function grep_search() {
eval $1=$(ls | tail -1)
}
Test:
$ source script.sh
$ grep_search open_box
$ echo $open_box
script.sh
As per help eval:
Execute arguments as a shell command.
You may also use Bash ${!var} indirect expansion, as already mentioned, however it doesn't support retrieving of array indices.
For further read or examples, check BashFAQ/006 about Indirection.
We are not aware of any trick that can duplicate that functionality in POSIX or Bourne shells without eval, which can be difficult to do securely. So, consider this a use at your own risk hack.
However, you should re-consider using indirection as per the following notes.
Normally, in bash scripting, you won't need indirect references at all. Generally, people look at this for a solution when they don't understand or know about Bash Arrays or haven't fully considered other Bash features such as functions.
Putting variable names or any other bash syntax inside parameters is frequently done incorrectly and in inappropriate situations to solve problems that have better solutions. It violates the separation between code and data, and as such puts you on a slippery slope toward bugs and security issues. Indirection can make your code less transparent and harder to follow.
For indexed arrays, you can reference them like so:
foo=(a b c)
bar=(d e f)
for arr_var in 'foo' 'bar'; do
declare -a 'arr=("${'"$arr_var"'[#]}")'
# do something with $arr
echo "\$$arr_var contains:"
for char in "${arr[#]}"; do
echo "$char"
done
done
Associative arrays can be referenced similarly but need the -A switch on declare instead of -a.
POSIX compliant answer
For this solution you'll need to have r/w permissions to the /tmp folder.
We create a temporary file holding our variables and leverage the -a flag of the set built-in:
$ man set
...
-a Each variable or function that is created or modified is given the export attribute and marked for export to the environment of subsequent commands.
Therefore, if we create a file holding our dynamic variables, we can use set to bring them to life inside our script.
The implementation
#!/bin/sh
# Give the temp file a unique name so you don't mess with any other files in there
ENV_FILE="/tmp/$(date +%s)"
MY_KEY=foo
MY_VALUE=bar
echo "$MY_KEY=$MY_VALUE" >> "$ENV_FILE"
# Now that our env file is created and populated, we can use "set"
set -a; . "$ENV_FILE"; set +a
rm "$ENV_FILE"
echo "$foo"
# Output is "bar" (without quotes)
Explaining the steps above:
# Enables the -a behavior
set -a
# Sources the env file
. "$ENV_FILE"
# Disables the -a behavior
set +a
While I think declare -n is still the best way to do it there is another way nobody mentioned it, very useful in CI/CD
function dynamic(){
export a_$1="bla"
}
dynamic 2
echo $a_2
This function will not support spaces so dynamic "2 3" will return an error.
for varname=$prefix_suffix format, just use:
varname=${prefix}_suffix
I know that there are arrays in bash-scripting. For example:
JOBS=("JOB1", "JOB2", "JOB3")
Then one can refer to, say, JOB2 as follows:
${JOBS[1]}
In a bash script I recently encountered a normal (non-array) variable:
JOB="XYZ"
Later in the script, this variable was referred to as follows:
${JOB[0]}
and:
${JOB[1]}
Since JOB is not an array, I do not understand what ${JOB[number]} is expanded to.
Is this just a programming mistake? Or is there some construct regarding normal variables I am not aware of?
bash doesn't have array values at all; it provides array syntax to use with names that have an array attribute set:
$ a=()
$ declare -p a
declare -a a=()
The -a in the output indicates that the array attribute is set on a, having been set implicitly by the preceding array assignment.
For such variables, the name-plus-index acts as a sort of "virtual" name, and like any other name, expands to the empty string if the name doesn't actually exist, as is the case with ${JOB[1]}. A side effect of the implementation is the $foo and ${foo[0]} are generally equivalent, whether or not foo has its array attribute set.
I believe newer versions of Bash support one-dimensional arrays and the syntax can be seen in the following link as follows :
Advanced Bash-Scripting Guide
If Variable is not an array, ${Variable[Number]} will work only for Number=0 but not for any other number as ${Variable[0]} and $Variable - both are same.
I'm writing a bash script to analyse some files. In a first iteration I create associative arrays with word counts for each category that is analysed. These categories are not known in advance so the names of these associative arrays are variable but all with the same prefix count_$category. The associative arrays have a word as key and its occurrence count in that category as value.
After all files are analysed, I have to summarise the results for each category. I can iterate over the variable names using ${count_*} but how can I access the associative arrays behind those variable names? For each associative array (each count_* variable) I should iterate over the words and their counts.
I have already tried with indirect access like this but it doesn't work:
for categorycount in ${count_*} # categorycount now holds the name of the associative array variable for each category
do
array=${!categorycount}
for word in ${!array[#]}
do
echo "$word occurred ${array[$word]} times"
done
done
The modern (bash 4.3+) approach uses "namevars", a facility borrowed from ksh:
for _count_var in "${!count_#}"; do
declare -n count=$_count_var # make count an alias for $_count_var
for key in "${!count[#]}"; do # iterate over keys, via same
echo "$key occurred ${count[$key]} times" # extract value, likewise
done
unset -n count # clear that alias
done
declare -n count=$count_var allows "${count[foo]}" to be used to look up item foo in the associative array named count_var; similarly, count[foo]=bar will assign to that same item. unset -n count then removes this mapping.
Prior to bash 4.3:
for _count_var in "${!count_#}"; do
printf -v cmd '_count_keys=( "${!%q[#]}" )' "$_count_var" && eval "$cmd"
for key in "${_count_keys[#]}"; do
var="$_count_var[$key]"
echo "$key occurred ${!var} times"
done
done
Note the use of %q, rather than substituting a variable name directly into a string, to generate the command to eval. Even though in this case we're probably safe (because the set of possible variable names is restricted), following this practice reduces the amount of context that needs to be considered to determine whether an indirect expansion is secure.
In both cases, note that internal variables (_count_var, _count_keys, etc) use names that don't match the count_* pattern.