Token pasting as a vector for variation

[eternaleye]

Token pasting in the preprocessor is sufficient to introduce a switch/case-like construct; this can act as a vector for variation to propagate through the codebase. This example switches on whether LONG_MAX is 2^31-1, 2^63-1, or undefined:

#include <assert.h>
#include <limits.h>
#include <stdio.h>

#define SELECT(cases, selector) _SELECT(cases, selector)
#define _SELECT(cases, selector) cases ## _ ## selector

// switch(val)
//     case 0x7fffffffL: return 4;
#define EVIL_0x7fffffffL 4
//     case 0x7fffffffffffffffL: return 8;
#define EVIL_0x7fffffffffffffffL 8
//     default: return 0;
#define EVIL_LONG_MAX 0
// }

#define LONG_BYTES SELECT(EVIL, LONG_MAX)

int main() {
    assert(LONG_BYTES > 0);
    printf("A long is %d bytes\n", LONG_BYTES);
}

I originally came up with it as an example case of how the preprocessor is hostile to "partial evaluation", as https://github.com/tendra/tendra attempts to perform, as in their approach LONG_MAX would act as undefined in the initial platform-agnostic preprocessing, resulting in an unconditional assert.

[ckaestne]

Conceptually this should be handled in the SuperC lexer that we are using by default (I know that the original TypeChef lexer doesn't handle this correctly).