The War Against Shell File Extensions

I see what you did there, you created a file called script.sh. Why do you have to be like this? How can we wash off this shame?

Pedantic Explanation

Extensions don’t matter. Executable files are recognized by the executable file permission bit (x) and, for compiled binaries, by the ELF header. Scripts use an interpreter defined in the shebang line (#!). There are a lot of terms here, and you have no way to know how much I truly understand them, so you might say: “It’s just for providing context in the file name.”

But I have a more pedantic answer. Let’s say you name your script script.sh, but the shebang is:

#!/usr/bin/env bash

The interpreter is not sh, and with this shebang, we can’t even be sure that it’s really bash.

Moreover, the interpreter may not even be what it seems:

$ ls -la /usr/bin/sh
lrwxrwxrwx 1 root root 4 Feb  4  2025 /usr/bin/sh -> dash

For example, in the Debian 13 installation that I use, sh is symlinked to dash, which is a POSIX-compliant shell just like sh. So should the script be called script.dash? On another system, it might use a different interpreter.

Silver Lining

So a lot of pedanticims happened here, but I have a guide on how to use shell script extensions if you really want to be a baby.

If it’s a POSIX-compliant script, you can use .sh. If it’s for a specific shell, use that shell’s extension, for example, .bash. But the important thing is: never use extensions if it’s a command that will be available system-wide.

Less Pedantic Explanation

So let’s say I have a rare condition that causes a love for creating sha1 hashes from text, so I create a script called hshr:

#!/usr/bin/env bash

main() {
  if [ -z "$1" ]; then
    echo "No argument provided."
    exit 1
  fi
  echo -n "$1" | sha1sum | awk '{print $1}'
}

main "$1"

Maybe I want it to be accessible throughout the whole system, so I symlink it to the appropriate path. Then I can run it just like any other command.

$ sudo ln -sf "$PWD"/hshr /usr/local/bin/hshr
$ hshr "example text"
d9e989f651cdd269d7f9bb8a215d024d8d283688

You’re still not convinced.

One day I realize that I want to rewrite this in C, so I whip up method like this, which will be called with the first argument which is the text, and will print out the sha1 hash:

void sha1(char *s)
{
    unsigned char hash[SHA_DIGEST_LENGTH];
    SHA1((unsigned char *)s, strlen(s), hash);

    for (int i = 0; i < SHA_DIGEST_LENGTH; i++) {
        printf("%02x", hash[i]);
    }

    printf("\n");
}

I build it and create a symlink to the same place, and try it out again:

$ gcc -o hshr hshr.c -lcrypto
$ sudo ln -sf "$PWD"/hshr /usr/local/bin/hshr
$ hshr "example text"
d9e989f651cdd269d7f9bb8a215d024d8d283688

The command implementation has been completely switched, but the command output and usage remain the same. Voilà, this is my point.

I rest my case, dear jury. Please protect your ears while the mic hits the floor.

Then, while I’m enjoying my well-earned victory, you step up to me and whisper in my ear: “Technically, it’s enough to not use the extension in the symlink target.”