What's the difference between set, export and env and when should I use each?

Question

Every so often I'll bash out a bash script and it strikes me there are a few ways of setting a variable:

key=value
env key=value
export key=value

When you're inside a script or a single command (for instance, I'll often chain a variable with a Wine launcher to set the right Wine prefix) these seem to be completely interchangeable but surely that can't be the case.

What's the difference between these three methods and can you give me an example of when I would specifically want to use each one?

_{Definitely related to What is the difference between `VAR=...` and `export VAR=...`? but I want to know how env fits into this too, and some examples showing the benefits of each would be nice too :)}

Answer 1

Let us consider a specific example. The grep command uses an environment variable called GREP_OPTIONS to set default options.

Now. Given that the file test.txt contains the following lines:

line one
line two

running the command grep one test.txt will return

line one

If you run grep with the -v option, it will return the non-matching lines, so the output will be

line two

We will now try to set the option with an environmental variable.

1. Environment variables set without export will not be inherited in the environment of the commands you are calling.

   GREP_OPTIONS='-v'
   grep one test.txt
   

   The result:

   line one
   

   Obviously, the option -v did not get passed to grep.

   You want to use this form when you are setting a variable only for the shell to use, for example in for i in * ; do you do not want to export $i.

2. However, the variable is passed on to the environment of that particular command line, so you can do

   GREP_OPTIONS='-v' grep one test.txt
   

   which will return the expected

   line two
   

   You use this form to temporarily change the environment of this particular instance of the program launched.

3. Exporting a variable causes the variable to be inherited:

   export GREP_OPTIONS='-v'
   grep one test.txt
   

   returns now

   line two
   

   This is the most common way of setting variables for use of subsequently started processes in a shell

4. This was all done in bash. export is a bash builtin; VAR=whatever is bash syntax. env, on another hand, is a program in itself. When env is called, following things happen:

   1. The command env gets executed as a new process
   2. env modifies the environment, and
   3. calls the command that was provided as an argument. The env process is replaced by the command process.

   Example:

   env GREP_OPTIONS='-v' grep one test.txt
   

   This command will launch two new processes: (i) env and (ii) grep (actually, the second process will replace the first one). From the point of view of the grep process, the result is exactly the same as running

   GREP_OPTIONS='-v' grep one test.txt
   

   However, you can use this idiom if you are outside of bash or don't want to launch another shell (for example, when you are using the exec() family of functions rather than the system() call).

Additional note on #!/usr/bin/env

This is also why the idiom #!/usr/bin/env interpreter is used rather than #!/usr/bin/interpreter. env does not require a full path to a program, because it uses the execvp() function which searches through the PATH variable just like a shell does, and then replaces itself by the command run. Thus, it can be used to find out where an interpreter (like perl or python) "sits" on the path.

It also means that by modifying the current path you can influence which python variant will be called. This makes the following possible:

echo -e '#!/usr/bin/bash\n\necho I am an evil interpreter!' > python
chmod a+x ./python
export PATH=.
python

instead of running Python, will result in

I am an evil interpreter!