As a matter of style we usually prefer to use %ju and a cast to uintmax_t instead of PRIu64.

I don't quite understand the change. off_t is indeed signed, so why do we want to hide negative values? They should result in errors anyway.

ggate defines its offset in unsigned int 64 as below.
So, printing negative numbers from printf indicates using a wrong formatter.

The PR formatter is to be architecture independent.
Is it better to use C99 provided formatter?
If not, we can use %u for unsigned 32bit int and %llu for unsigned 64bit, cannot we?

https://svnweb.freebsd.org/base/head/sbin/ggate/shared/ggate.h?revision=351936&view=markup#l90

90 struct g_gate_hdr {
91 uint8_t gh_cmd; /* command */
92 uint64_t gh_offset; /* device offset */
93 uint32_t gh_length; /* size of block */
94 uint64_t gh_seq; /* request number */
95 uint16_t gh_error; /* error value (0 if ok) */
96 } __packed;

In D21388#585695, @ota_j.email.ne.jp wrote:

ggate defines its offset in unsigned int 64 as below.
So, printing negative numbers from printf indicates using a wrong formatter.

I'm sorry, I see now. For some reason I thought that r_offset is off_t, not uint64_t.

The PR formatter is to be architecture independent.
Is it better to use C99 provided formatter?
If not, we can use %u for unsigned 32bit int and %llu for unsigned 64bit, cannot we?

For printing off_t (i.e., a typedef), it is better to cast to intmax_t. For printing a raw uint64_t it is ok to use PR formatters. They are generally discouraged in FreeBSD, but I don't object to it. I do not think it is right to use %llu though.

So, either of the suggested approaches are ok with me. Sorry again for the confusion.