How to Use the Plan 9 C Compiler [pdf]

Or it could be a testament to how little real-world use Plan 9 saw. This is the same reason Latin doesn't change.

I'm drawing a distinction between adding extensions (which can still be problematic if they can't be disabled) and ignoring or rejecting parts of the language that are required for correctness. The simplest and most obvious example mentioned fairly early in the paper is const but there are plenty of others.

I can compile nearly any C program using gcc and get correct behavior. Obviously I can also compile a lot of other things using gcc, including things that are not found in C, and if I do so then I cannot expect any definition of C to specify the resulting behavior. But most C programs will not compile using the Plan 9 compilers at all, because they do not accept the C language by any meaningful definition. In the case of const, it's also true that many dangerously incorrect programs will compile when using Plan 9's compiler, even though any definition of C prohibits that explicitly. These are qualitative differences.

Let's be even more concrete. In case 1, we consider a language exactly like C (pick whatever C you want here), but with the following additional preprocessor symbols always defined:

    #define _FBEGIN {
    #define _FEND }

This language allows, but does not require, the use of alternate tokens instead of { and } around function bodies. Because the tokens are taken from a namespace reserved by the standards, every correct C program is also a correct program in this language, and will have the same behavior. I submit that one may reasonably claim that a compiler for this language is a C compiler that supports extensions, even though a very limited subset of incorrect programs will fail to compile in a different way than they would when submitted to a C compiler.

In case 2, we consider the same language except that in this case the use of { and } around function bodies is not accepted. Instead, one must use _FBEGIN and _FEND. I submit that a compiler for this language is not a C compiler, as no (useful) correct C program is accepted by it.

What the Plan 9 compiler authors did, repeatedly, reduces to case 2. There is nothing unethical about creating a new language. There is something unethical about creating a new language and not only deliberately giving it the same name as a similar (very popular) language, but disingenuously acting as if it were that language. At best, the implication that all other (again, very popular) standards and implementations of C are wrong simply for not being Plan 9's is obnoxiously smug.

> I guess the designers of C realized there are better alternatives when implementing their own compiler.

Or maybe they went off the rails and lost their minds, like Bill Shockley.

volatile is mostly useful for telling the compiler not to optimize out a variable. For example, if you use memory-mapped I/O, you can mark a pointer to it as volatile to tell the compiler that the hardware will be changing it and it shouldn't make any assumptions about that while optimizing. That's lower level than most people go, though, so I understand that it's not useful for the average C programmer.