Thomas Touhey

On Mastodon, Koz Ross (quitter.se, his instance, disappeared since this post was published) asked:

UNIX historians, why does main return a *signed* int? As far as I’m aware, exit statuses can only be positive or 0.

I first made a quick answer, but then I decided to check out for some sources, and it eventually got too big to fit into a Mastodon post, so here’s the reason of this article.

One of the reasons is that the unsigned keyword did not exist in C until the mid 1970s, as hardware did not support unsigned operations until then, and Unix had been up for several years already. But the “when should I use an unsigned type” question is still asked frequently today.

First of all, you shouldn’t use unsigned for everything that can’t be under zero. Here’s what Peter Van Der Linden says in Expert C Programming:

Avoid unnecessary complexity by minimizing your use of unsigned types. Specifically, don’t use an unsigned type to represent a quantity just because it will never be negative (e.g. “age” or “national_debt”).

Use a signed type like int and you won’t have to worry about boundary cases in the detailed rules for promoting mixed types.

Only use unsigned types for bitfields or binary masks. Use casts in expressions, to make all the operands signed or unsigned, so the compiler does not have to choose the result type.

Mixed types promotion is when you have to adapt two integers of different types to apply an operation on them, such as when you are multiplying signed int and unsigned int. Also, keep in mind that the unsigned version of a type only allows you to use one more bit: for example, as int is at least 16-bits wide, you can store up to 32767 into it (use long/unsigned long if you want to store up to 32-bits in a portable fashion).

This is used in ctype.h as well, as transferring an int to a function is usually quicker than transferring an unsigned char (usually because your function has to transmit it as an int, then the other function has to apply the & 255 mask on it).

Now, you may ask: yeah, but what if I want to return 32768 as a return code? Well you can’t. Here’s the information about it that you can find in the Single Unix Specification (basically POSIX), 2.13. Status Information section:

Status information is data associated with a process detailing a change in the state of the process. It shall consist of: […]

• If the new state is terminated:

  • The low-order 8 bits of the status argument that the process passed to _Exit(), _exit() or exit(), or the low-order 8 bits of the value the process returned from main().

So you are only allowed to return up to 255 as the exit code, which the int can contain in a portable fashion. And because of the definition of the exit code, you can just take return_code & 255.