libc::execv incorrect definition

[carllerche]

It should be:

pub unsafe fn execv(prog: *const c_char, argv: *const *const c_char) -> c_int

[mahkoh]

char *const argv[] == char *const *argv in function signatures.

#include <stdio.h>

#define typename(x) _Generic((x), \
    char *const *:      "char *const *", \
    default:            "unreachable!()")

void f(char *const x[]) {
    puts(typename(x));
}

int main(void) {
    f(0);
}

(only works in clang)

[mahkoh]

But the definition is still wrong. char *const * is a pointer to constant pointers to mutable characters. I think the following definition is correct.

pub unsafe fn execv(prog: *const c_char, argv: *const *mut c_char) -> c_int

[mahkoh]

Interestingly enough, the example in POSIX 2008 disregards the inner mutability in its execv example:

#include <unistd.h>

int main(void) {
    char *const cmd[] = { "ls", "-l", (char *)0 };
    execv("/bin/ls", cmd);
}

test.c:4:24: warning: initializing 'char *const' with an expression of type
      'const char [3]' discards qualifiers
      [-Wincompatible-pointer-types-discards-qualifiers]
        char *const cmd[] = { "ls", "-l", (char *)0 };
                              ^~~~
test.c:4:30: warning: initializing 'char *const' with an expression of type
      'const char [3]' discards qualifiers
      [-Wincompatible-pointer-types-discards-qualifiers]
        char *const cmd[] = { "ls", "-l", (char *)0 };
                                    ^~~~

I'm not quite sure what to believe.

[mahkoh]

A thorough reading of POSIX 2008 reveals that execv will never modify any level of the argv parameter. See the following screenshot for a detailed explanation (compatibility with old code).

So @carllerche's definition is correct after all.

[steveklabnik]

Today, this is

                pub fn execv(prog: *const c_char,
                             argv: *mut *const c_char) -> intptr_t;

so it appears that it's still wrong.

[geofft]

PR #25641 should fix this. I didn't notice this issue at the time, but I reached the same conclusions about the actual const guarantees (and C's legacy problems therewith).