We have seen that referencing a variable, $var, fetches its value. But, what about the value of a value? What about $$var?
The actual notation is \$$var, usually preceded by an eval (and sometimes an echo). This is called an indirect reference.
Example 28-1. Indirect Variable References
1 #!/bin/bash 2 # ind-ref.sh: Indirect variable referencing. 3 # Accessing the contents of the contents of a variable. 4 5 # First, let's fool around a little. 6 7 var=23 8 9 echo "\$var = $var" # $var = 23 10 # So far, everything as expected. But ... 11 12 echo "\$\$var = $$var" # $$var = 4570var 13 # Not useful ... 14 # \$\$ expanded to PID of the script 15 # -- refer to the entry on the $$ variable -- 16 #+ and "var" is echoed as plain text. 17 # (Thank you, Jakob Bohm, for pointing this out.) 18 19 echo "\\\$\$var = \$$var" # \$$var = $23 20 # As expected. The first $ is escaped and pasted on to 21 #+ the value of var ($var = 23 ). 22 # Meaningful, but still not useful. 23 24 # Now, let's start over and do it the right way. 25 26 # ============================================== # 27 28 29 a=letter_of_alphabet # Variable "a" holds the name of another variable. 30 letter_of_alphabet=z 31 32 echo 33 34 # Direct reference. 35 echo "a = $a" # a = letter_of_alphabet 36 37 # Indirect reference. 38 eval a=\$$a 39 # ^^^ Forcing an eval(uation), and ... 40 # ^ Escaping the first $ ... 41 # ------------------------------------------------------------------------ 42 # The 'eval' forces an update of $a, sets it to the updated value of \$$a. 43 # So, we see why 'eval' so often shows up in indirect reference notation. 44 # ------------------------------------------------------------------------ 45 echo "Now a = $a" # Now a = z 46 47 echo 48 49 50 # Now, let's try changing the second-order reference. 51 52 t=table_cell_3 53 table_cell_3=24 54 echo "\"table_cell_3\" = $table_cell_3" # "table_cell_3" = 24 55 echo -n "dereferenced \"t\" = "; eval echo \$$t # dereferenced "t" = 24 56 # In this simple case, the following also works (why?). 57 # eval t=\$$t; echo "\"t\" = $t" 58 59 echo 60 61 t=table_cell_3 62 NEW_VAL=387 63 table_cell_3=$NEW_VAL 64 echo "Changing value of \"table_cell_3\" to $NEW_VAL." 65 echo "\"table_cell_3\" now $table_cell_3" 66 echo -n "dereferenced \"t\" now "; eval echo \$$t 67 # "eval" takes the two arguments "echo" and "\$$t" (set equal to $table_cell_3) 68 69 70 echo 71 72 # (Thanks, Stephane Chazelas, for clearing up the above behavior.) 73 74 75 # A more straightforward method is the ${!t} notation, discussed in the 76 #+ "Bash, version 2" section. 77 # See also ex78.sh. 78 79 exit 0 |
Of what practical use is indirect referencing of variables? It gives Bash a little of the functionality of pointers in C, for instance, in table lookup. And, it also has some other very interesting applications. . . .
Nils Radtke shows how to build "dynamic" variable names and evaluate their contents. This can be useful when sourcing configuration files.
1 #!/bin/bash 2 3 4 # --------------------------------------------- 5 # This could be "sourced" from a separate file. 6 isdnMyProviderRemoteNet=172.16.0.100 7 isdnYourProviderRemoteNet=10.0.0.10 8 isdnOnlineService="MyProvider" 9 # --------------------------------------------- 10 11 12 remoteNet=$(eval "echo \$$(echo isdn${isdnOnlineService}RemoteNet)") 13 remoteNet=$(eval "echo \$$(echo isdnMyProviderRemoteNet)") 14 remoteNet=$(eval "echo \$isdnMyProviderRemoteNet") 15 remoteNet=$(eval "echo $isdnMyProviderRemoteNet") 16 17 echo "$remoteNet" # 172.16.0.100 18 19 # ================================================================ 20 21 # And, it gets even better. 22 23 # Consider the following snippet given a variable named getSparc, 24 #+ but no such variable getIa64: 25 26 chkMirrorArchs () { 27 arch="$1"; 28 if [ "$(eval "echo \${$(echo get$(echo -ne $arch | 29 sed 's/^\(.\).*/\1/g' | tr 'a-z' 'A-Z'; echo $arch | 30 sed 's/^.\(.*\)/\1/g')):-false}")" = true ] 31 then 32 return 0; 33 else 34 return 1; 35 fi; 36 } 37 38 getSparc="true" 39 unset getIa64 40 chkMirrorArchs sparc 41 echo $? # 0 42 # True 43 44 chkMirrorArchs Ia64 45 echo $? # 1 46 # False 47 48 # Notes: 49 # ----- 50 # Even the to-be-substituted variable name part is built explicitly. 51 # The parameters to the chkMirrorArchs calls are all lower case. 52 # The variable name is composed of two parts: "get" and "Sparc" . . . |
Example 28-2. Passing an indirect reference to awk
1 #!/bin/bash 2 3 # Another version of the "column totaler" script 4 #+ that adds up a specified column (of numbers) in the target file. 5 # This one uses indirect references. 6 7 ARGS=2 8 E_WRONGARGS=85 9 10 if [ $# -ne "$ARGS" ] # Check for proper number of command-line args. 11 then 12 echo "Usage: `basename $0` filename column-number" 13 exit $E_WRONGARGS 14 fi 15 16 filename=$1 # Name of file to operate on. 17 column_number=$2 # Which column to total up. 18 19 #===== Same as original script, up to this point =====# 20 21 22 # A multi-line awk script is invoked by 23 # awk " 24 # ... 25 # ... 26 # ... 27 # " 28 29 30 # Begin awk script. 31 # ------------------------------------------------- 32 awk " 33 34 { total += \$${column_number} # Indirect reference 35 } 36 END { 37 print total 38 } 39 40 " "$filename" 41 # Note that awk doesn't need an eval preceding \$$. 42 # ------------------------------------------------- 43 # End awk script. 44 45 # Indirect variable reference avoids the hassles 46 #+ of referencing a shell variable within the embedded awk script. 47 # Thanks, Stephane Chazelas. 48 49 50 exit $? |
This method of indirect referencing is a bit tricky. If the second order variable changes its value, then the first order variable must be properly dereferenced (as in the above example). Fortunately, the ${!variable} notation introduced with version 2 of Bash (see Example 37-2 and Example A-22) makes indirect referencing more intuitive. |