Just about every programming language in existence has the conceptof variables - a symbolic name for a chunk of memory to whichwe can assign values, read and manipulate its contents. The bourneshell is no exception, and this section introduces idea. This is taken further in Variables - Part IIwhich looks into variables which are set for us by the environment.
Let's look back at our first Hello World example. This could be doneusing variables (though it's such a simple example that it doesn'treally warrant it!)
Note that there must be no spaces around the "=" sign: VAR=value
works; VAR = value
doesn't work.In the first case, the shell sees the "=" symbol and treats the command as a variable assignment. In the second case, the shell assumes that VAR must be the name of a command and tries to execute it.
If you think about it, this makes sense - how else could you tell it to run the command VAR with its firstargument being "=" and its second argument being "value"?
Enter the following code into var1.sh:
var.sh
#!/bin/sh MY_MESSAGE="Hello World" echo $MY_MESSAGE
This assigns the string "Hello World" to the variableMY_MESSAGE
then echo
es out the value ofthe variable.
Note that we need the quotes around the string Hello World. Whereaswe could get away with echo Hello World
because echowill take any number of parameters, a variable can only hold one value,so a string with spaces must be quoted to that the shell knows to treatit all as one. Otherwise, the shell will try to execute the command World
after assigning MY_MESSAGE=Hello
The shell does not care about types of variables; they may store strings,integers, real numbers - anything you like.
People used to Perl may be quite happy with this; if you've grown up with C, Pascal,or worse yet Ada, this may seem quite strange.
In truth, these are all stored as strings, but routines which expect a number can treat them as such.
If you assign a string to a variable then try to add 1 to it, you will not get away with it:
$ x="hello" $ y=`expr $x + 1` expr: non-numeric argument $
Since the external program expr
only expects numbers.But there is no syntactic difference between:
MY_MESSAGE="Hello World" MY_SHORT_MESSAGE=hi MY_NUMBER=1 MY_PI=3.142 MY_OTHER_PI="3.142" MY_MIXED=123abc
Note though that special characters must be properly escaped to avoidinterpretation by the shell.
This is discussed further in Escape Characters.
We can interactively set variable names using the read
command; the following script asks you for your name then greets you personally:
var2.sh
#!/bin/sh echo What is your name? read MY_NAME echo "Hello $MY_NAME - hope you're well."
Mario Bacinsky kindly pointed out to me that I had originally missedout the double-quotes in line 3, which meant that the single-quote in theword "you're" was unmatched, causing an error. It is this kind of thingwhich can drive a shell programmer crazy, so watch out for them!
This is using the shell-builtin command read
which reads aline from standard input into the variable supplied.
Note that even if you give it your full name and don't use double quotesaround the echo
command, it still outputs correctly. Howis this done? With the MY_MESSAGE
variable earlier we hadto put double quotes around it to set it.
What happens, is that the read
command automatically placesquotes around its input, so that spaces are treated correctly. (You willneed to quote the output, of course - e.g. echo "$MY_MESSAGE"
).
Variables in the bourne shell do not have to be declared, as they do in languages like C. But if you try to read an undeclared variable,the result is the empty string. You get no warnings or errors. This can causesome subtle bugs - if you assign MY_OBFUSCATED_VARIABLE=Hello
and then echo $MY_OSFUCATED_VARIABLE
, you will get nothing (as the secondOBFUSCATED is mis-spelled).
There is a command called export
which has a fundamentaleffect on the scope of variables. In order to really know what's goingon with your variables, you will need to understand something about howthis is used.
Create a small shell script, myvar2.sh
:
myvar2.sh
#!/bin/sh echo "MYVAR is: $MYVAR" MYVAR="hi there" echo "MYVAR is: $MYVAR"
Now run the script:
$ ./myvar2.sh MYVAR is: MYVAR is: hi there
MYVAR hasn't been set to any value, so it's blank. Then we give it a value, and ithas the expected result.
Now run:
$ MYVAR=hello $ ./myvar2.sh MYVAR is: MYVAR is: hi there
It's still not been set! What's going on?!
When you call myvar2.sh
from your interactive shell, a newshell is spawned to run the script. This is partly because of the #!/bin/sh
line at the start of the script, which we discussed earlier.
We need to export
the variable for it to be inherited by another program - including a shell script. Type:
$ export MYVAR $ ./myvar2.sh MYVAR is: hello MYVAR is: hi there
Now look at line 3 of the script: this is changing the value ofMYVAR
. But there is no way that this will be passed backto your interactive shell. Try reading the value of MYVAR
:
$ echo $MYVAR hello $
Once the shell script exits, its environmentis destroyed. But MYVAR
keeps its value of hello
within your interactive shell.
In order to receive environment changes back from the script, we mustsource the script - this effectively runs the script within ourown interactive shell, instead of spawning another shell to run it.
We can source a script via the "." command:
$ MYVAR=hello $ echo $MYVAR hello $ . ./myvar2.sh MYVAR is: hello MYVAR is: hi there $ echo $MYVAR hi there
The change has now made it out into our shell again! This is how your.profile
or .bash_profile
file works, for example.
Note that in this case, we don't need to export MYVAR
.
Thanks to sway for pointing out that I'd originally said echo MYVAR
above, notecho $MYVAR
as it should be.One other thing worth mentioning at this point about variables, is toconsider the following shell script:
#!/bin/sh echo "What is your name?" read USER_NAME echo "Hello $USER_NAME" echo "I will create you a file called $USER_NAME_file" touch $USER_NAME_file
Think about what result you would expect. For example, if you enter "steve"as your USER_NAME, should the script create steve_file
?
Actually, no. This will cause an error unless there is a variable called USER_NAME_file
. The shell does not know where the variableends and the rest starts. How can we define this?
The answer is, that we enclose the variable itself in curly brackets:
user.sh
#!/bin/sh echo "What is your name?" read USER_NAME echo "Hello $USER_NAME" echo "I will create you a file called ${USER_NAME}_file" touch "${USER_NAME}_file"
The shell now knows that we are referring to the variable USER_NAME
and that we want it suffixed with _file
. This can be thedownfall of many a new shell script programmer, as the source of the problemcan be difficult to track down.
Also note the quotes around "${USER_NAME}_file"
- if the user entered "Steve Parker" (note thespace) then without the quotes, the arguments passed to touch
would be Steve
andParker_file
- that is, we'd effectively be saying touch Steve Parker_file
, whichis two files to be touch
ed, not one. The quotes avoid this. Thanks to Chris for highlighting this.