statx struct's stx_size pointer compatibility with uint64_t/size_t

[Dominique Martinet <asmadeus at codewreck dot org>]

Hi,

I appologize if this is not the right place to ask, I'm using the glibc
header so going here but it is just as much a linux uapi question...
Cc'd a few linux kernel folks who worked on this as well for their
opinion, any input appreciated.


A coworker ran into an incompatible-pointer-type compiler warning when
trying to pass &statxbuf.stx_size to a function expecting a size_t,
which boils down to this:
-------------
#define _GNU_SOURCE
#include <stdint.h>
#include <sys/stat.h>

static void test(uint64_t *size) {   
    (void)size;
}

int main() {   
    struct statx statxbuf;

    test(&statxbuf.stx_size);
    return 0;
}
-------------
Giving this warning:
-------------
t.c: In function ‘main’:
t.c:12:10: warning: passing argument 1 of ‘test’ from incompatible
pointer type [-Wincompatible-pointer-types]
   12 |     test(&statxbuf.stx_size);
      |          ^~~~~~~~~~~~~~~~~~
      |          |
      |          __u64 * {aka long long unsigned int *}
t.c:5:28: note: expected ‘uint64_t *’ {aka ‘long unsigned int *’} but
argument is of type ‘__u64 *’ {aka ‘long long unsigned int *’}
    5 | static void test(uint64_t *size) {
      |                  ~~~~~~~~~~^~~~
-------------
(same happens with size_t)

The final warning is probably some standard defining long and long long
cannot be treated as compatible even on archs where it is, but it's a
bit of a shame that manipulating __u64 and uint64_t yield such errors
when passing pointers around -- the types pretty much guarantee they
have to be compatible and it is just an arbitrary choice that made one
be long and the other long long on x86_64 unless I misunderstood
something?


I'm a bit at loss of what to advise here.
We need to pass the value as a pointer because it can be updated, our
use case is that the symlink size can be wrong in /proc/x/fd/ and we
will want the correct value back in a statx struct here[1].

What would be the "recommended" way of doing this?
Any chance the field could change to be uint64_t-compatible in the
future? Not sure what that implies regarding e.g. backwards
compatibility though...


[1] https://github.com/cea-hpc/robinhood/blob/1ed74893c088d78783acd2e25e8009a483510ff7/src/backends/posix.c#L248

Thanks,
-- 
Dominique