Personal website of Martin Tournoij (“arp242”); writing about programming (CV) and various other things.

Working on GoatCounter and moreGitHub Sponsors.

Contact at or GitHub.

This page's author

Discussions: Lobsters.

I like to configure programs with commandline flags. Why include thousands of lines of code to parse confusing file formats when you can just run prog -foo? Want to support loading from environment variables? prog -foo "$ARG", or if you want to set a default use prog -foo "${ARG:-default}". Want to inspect the flags the application was launched with? ps. Want the documentation? prog -help. Want a config file? Create a wrapper script to call the program with the desired flags. Want to read the documen­ta­tion? That’s what the usage text is for.

Is this suitable for everything? No, of course not. In particular config files will work better for apps you launch frequently and have a lot of settings such as, say, git. But for servers or other programs that aren’t invoked very often I find this works very well.

For large and complex applications it’s probably not a good idea to configure everything with flags. Configuring something like Postfix with just flags would clearly be silly. Most applications are not complex; and if your application does grow it’ll be easy to replace the flag parsing code – which is usually just a few lines – with something more advanced.

I see plenty of fairly simple programs that have complex configuration schemes just to load a few settings. The project I worked on this week has 2.8k lines of code and 8 settings and uses a config library with 2.2k lines of code plus 9.5k from the YAML library. I don’t think that really makes things easier.

I’ve never really liked the idea of directly using environment variables for configuration. It’s not very explicit, can be surprising (with subprocesses inheriting parent), there are no warnings on typos, and it’s hard to inspect. Using flags allows you to use environment variables while fixing most of those issues: it’s explicit, clear which variables are used, and you get errors on typos.


# Error out on undefined variables to catch typos or other errors such as env
# vars not being defined when you expected it (not enough systems do this!)
set -u

# exec replaces the process; prevents needless /bin/sh from dangling around.
exec prog \
    -a "$A"          `# Always include A from environment, error out if unset`  \
    -b "${B:-d}"     `# Include B from environment, use "d" if unset` \
    "${C:+-c "$C"}"  `# Only pass -c $C if $C is set, pass nothing at all if unset` \
    "$@"             `# Include all argument passed to the shell script`

You can probably do something similar with systemd unit files. Personally I’d just call that shell script from systemd; it’s just so much easier to understand, test, and debug.

The biggest issue is that secrets may appear in ps output. Mounting procfs with hidepid=1 fixes the worst of it. That said, passing secrets securely is complex and most methods are not completely secure (a full guide is beyond the scope of this post).

Another approach that I rather like is the suckless one: configuration is done by editing config.h and recompiling the program. These tools are aimed at advanced users and are fairly small and quick to recompile (you don’t really want to recompile Vim every time you change a setting). Still, I think it can work well for many problems. Not everything needs to be a setting.