Ah, I see what's up with the man page. It's documenting the Glibc getdents64() wrapper interface, so that's why it uses the Glibc types. But gpgme isn't using that wrapper, it's doing the syscall directly, so it should use the types the kernel uses, which as you've noticed are just generic unsigned and signed 64-bit integers, matching what my patch does.

In D600#6448, @ikloecker wrote:

$ man getdents64

getdents64()
    The  original  Linux getdents() system call did not handle large filesystems and large file offsets.  Consequently, Linux 2.4
    added getdents64(), with wider types for the d_ino and d_off fields.  In addition, getdents64() supports an  explicit  d_type
    field.

    The  getdents64()  system call is like getdents(), except that its second argument is a pointer to a buffer containing struc‐
    tures of the following type:

        struct linux_dirent64 {
            ino64_t        d_ino;    /* 64-bit inode number */
            off64_t        d_off;    /* 64-bit offset to next structure */
            unsigned short d_reclen; /* Size of this dirent */
            unsigned char  d_type;   /* File type */
            char           d_name[]; /* Filename (null-terminated) */
        };

In D600#6445, @gniibe wrote:

If it is intentional change by musl (requiring some changes by an application), we can use __ino64_t_defined and __off64_t_defined macro to see if those types are defined or not.

off64_t mat not the same as int64_t

In D600#6438, @werner wrote:

This breaks existing 32 bit systems with the 64 bit types.