Wandering Thoughts archives

One thing that makes the Bourne shell an odd language

In many ways, the Bourne shell is a relatively conventional programming language. It has a few syntactic abnormalities, a few flourishes created by the fact that it is an engine for running programs (although other languages have featured equivalents of $(...) in the form of various levels of 'eval' functionality), and a different treatment of unquoted words, but the overall control structure is an extremely familiar Algol-style one (which is not surprising, since Steve Bourne really liked Algol).

But the Bourne shell does have one thing that clearly makes it an odd language, namely that it has outsourced what are normally core language functions to external programs. Or rather it started out in its original version by outsourcing those functions; versions of the Bourne shell since then have pulled them back in in various ways. Here I am thinking of both evaluating conditionals via test aka [ and arithmetic via expr (which also does some other things too).

(Bourne shells have had test as a builtin for some time (sometimes with some annoyances) and built in arithmetic is often present these days as $...)

There's no reason why test has to be a separate program and neither test nor expr seems to have existed in Research Unix V6, so they both appeared in V7 along with the Bourne shell itself. They aren't written in BourneGol, so they may not have been written by Steve Bourne himself, but at least test was clearly written as a companion program (the V7 Bourne shell manpage explicitly mentions it, among other things).

I don't know why the original Bourne shell made this decision. It's possible that it was simply forced by the limitations of the PDP-11 environment of V7. Maybe a version of the Bourne shell with test and/or expr built into the main shell code would have either been too big or just considered over-bloated for something that would mostly be used interactively (and thus not be using test et al very often). Or possibly they were just easier to write as separate programs (the V7 expr is just a single yacc file).

Note that there are structural reasons in the Bourne shell to make if et al conditions be the result of commands, instead of restricting them to (only) be actual conditions. But the original Bourne shell could have done this with test or the equivalent as a built-in command, and it certainly has other built in commands. Perhaps test needing to be an actual command was one of the things that pushed it towards not being built in. You can certainly see a spirit of minimalism at work here if you want to (although I have no idea if that's the reason).

(This expands on a tweet of mine.)

Sidebar: It's not clear when test picked up its [ alias

Before I started writing this entry, I expected that test was also known as [ right from the beginning in V7. Now I'm not so sure. On the one hand, the actual V7 shell scripts I can find eg here consistently use test instead of [ and the V7 compile scripts don't seem to create a [ hardlink. On the other hand, the V7 test source already has special magic handling if it's invoked as [.

(There are V7 disk images out there that you can boot up on a PDP-11 emulator, so in theory I could fire one up and see if it has a /bin/[. In practice I'm not that energetic.)