Add stpncpy to legacy support

[ryandesign (Ryan Carsten Schmidt)]

dash, for example, currently fails to build on Mac OS X 10.6 and earlier because of the lack of stpncpy; see #69572. And this prevents the installation of other ports, like littleutils, that depend on dash.

[mascguy (Christopher Nielsen)]

Given the relatively simple logic involved, I was expecting implementations of stpncpy to be reasonably similar. But the first three variations I reviewed, are quite different indeed.

GCC - via ​opensource.apple.com/source/gcc/gcc-5465/libiberty/stpncpy.c.auto.html:

char *stpncpy(char *dst, const char *src, size_t len) {
  size_t n = strlen(src);
  if (n > len)
    n = len;
  return strncpy(dst, src, len) + n;
}

LIBC - via ​opensource.apple.com/source/Libc/Libc-825.26/string/stpncpy.c.auto.html:

char *stpncpy(char * restrict dst, const char * restrict src, size_t maxlen) {
    const size_t srclen = strnlen(src, maxlen);
    if (srclen < maxlen) {
        //  The stpncpy() and strncpy() functions copy at most maxlen
        //  characters from src into dst.
        memcpy(dst, src, srclen);
        //  If src is less than maxlen characters long, the remainder
        //  of dst is filled with '\0' characters.
        memset(dst+srclen, 0, maxlen-srclen);
        //  The stpcpy() and stpncpy() functions return a pointer to the
        //  terminating '\0' character of dst.
        return dst+srclen;
    } else {
        //  The stpncpy() and strncpy() functions copy at most maxlen
        //  characters from src into dst.
        memcpy(dst, src, maxlen);
        //  If stpncpy() does not terminate dst with a NUL character, it
        //  instead returns a pointer to src[maxlen] (which does not
        //  necessarily refer to a valid memory location.)
        return dst+maxlen;
    }
}

And a 3rd, from ​https://lore.kernel.org/dash/6cacf608-326e-211a-7d37-164041710980@knaff.lu/:

char *stpncpy(char *dst, const char *src, int len) {
    int i,nullSeen=0;
    char *ret=dst+len;
    for(i=0; i<len; i++) {
        if(nullSeen)
            dst[i] = '\0';
        else {
            dst[i] = src[i];
            if(dst[i] == '\0') {
                nullSeen = 1;
                ret = dst+i;
            }
        }
    }
    return ret;
}

So we apparently have an embarrassment of riches...

[fhgwright (Fred Wright)]

I was about to attack this, so you can feel free to assign it to me.

The obvious thing to do is to copy Apple's implementation from the later OS versions where they provide them. It looks like this happened in Libc-825.24, and never changed since. It's a dumb two-pass implementation (as is the GCC version shown above) that fails to avoid the extra overhead that motivated stpcpy()/stpncpy() in the first place, but in keeping with the philosophy of never doing anything better than Apple in legacy-support, it makes sense. Thus, the only real benefit over strcpy()/strncpy() is in compatibility and convenience.

The larger part of the work is in creating tests, since Apple doesn't have any. One could argue that adding tests for code copied verbatim from Apple shouldn't really be necessary, but in the interests of completeness, it makes sense to add them.

[fhgwright (Fred Wright)]

In 12a4dca86937f7d072f8ed64fd7df91b60619a41/macports-legacy-support (master):

Add stpncpy() support

Imports Apple's stpncpy.c verbatim from Libc-825.24.

Removes 'restrict' qualifiers for pre-C99 compatibility.

Fixes whitespace; adds modification notice and proper build
conditional.

Updates the legacy-support headers.

Updates the README.

Closes: #69571

TESTED:
Tested on 10.4-10.5 ppc, 10.4-10.6 i386, 10.5-10.6 ppc (i386 Rosetta),
10.5-12.x x86_64, 11.x-14.x arm64.
Passes all tests, including newly added stpncpy() tests.