Implementation looks good to me. Looks like the read-only fsync tests are failing though.

       EROFS, EINVAL
              fd  is  bound to a special file (e.g., a pipe, FIFO, or socket) which does not sup‐
              port synchronization.

Maybe one of those would serve as a better test for the failure case.

Implementation looks good to me. Looks like the read-only fsync tests are failing though.

Passes on Windows but a quick Linux test indicates you can fsync a RO file descriptor and get a success (0) return, I guess as an effective no-op. So it's not a good test, will remove it and add a failure case on e.g. a socket FD.
@nikic

If the resource indicated by $stream is not a file pointer, an E_WARNING is emitted.

idk if this is severe enough, only thing i am really stuck on atm.

Thank you for this PR, I'm happy it's voted accepted :)

I was testing this out for a bit, and from what I see, I cannot get it to work with custom stream wrappers. Is that intended?
For example:

stream_wrapper_register("test", "TestStream");

class TestStream {
     function stream_open($path, $mode, $options, &$opened_path)
     {
         return true;
     }
    function stream_fsync() {
        return true;
    }
    function stream_set_option() {
         return true;
    }
}

var_dump($r = fopen('test://dsa', 'w'));
fsync($r);

This results in the warning Warning: fsync(): Can't fsync this stream!

Thank you.

I was testing this out for a bit, and from what I see, I cannot get it to work with custom stream wrappers. Is that intended?

@Ayesh yes, this is intentional; fsync only works with streams which can be internally cast to plain file streams (specifically, streams which have a file descriptor). It simply doesn't make sense to try and fsync() anything else, since it is by definition a file system operation.

@nikic comments addressed, I think this is good to merge now.