mirror of
https://gitlab.torproject.org/tpo/core/tor.git
synced 2024-11-10 21:23:58 +01:00
Make the touch_file unit test work around FS/system time sync issues
Sometimes you can call time() and then touch a file, and have the second come out a little before the first. See #18025 for way more information than you necessarily wanted.
This commit is contained in:
parent
1d6dd288e1
commit
16840e52e5
@ -4612,18 +4612,23 @@ test_util_touch_file(void *arg)
|
||||
struct stat st;
|
||||
write_bytes_to_file(fname, "abc", 3, 1);
|
||||
tt_int_op(0, OP_EQ, stat(fname, &st));
|
||||
tt_i64_op(st.st_mtime, OP_GE, now);
|
||||
/* A subtle point: the filesystem time is not necessarily equal to the
|
||||
* system clock time, since one can be using a monotonic clock, or coarse
|
||||
* monotonic clock, or whatever. So we might wind up with an mtime a few
|
||||
* microseconds ago. Let's just give it a lot of wiggle room. */
|
||||
tt_i64_op(st.st_mtime, OP_GE, now - 1);
|
||||
|
||||
const time_t five_sec_ago = now - 5;
|
||||
struct utimbuf u = { five_sec_ago, five_sec_ago };
|
||||
tt_int_op(0, OP_EQ, utime(fname, &u));
|
||||
tt_int_op(0, OP_EQ, stat(fname, &st));
|
||||
/* Let's hope that utime/stat give the same second as a round-trip? */
|
||||
tt_i64_op(st.st_mtime, OP_EQ, five_sec_ago);
|
||||
|
||||
/* Finally we can touch the file */
|
||||
tt_int_op(0, OP_EQ, touch_file(fname));
|
||||
tt_int_op(0, OP_EQ, stat(fname, &st));
|
||||
tt_i64_op(st.st_mtime, OP_GE, now);
|
||||
tt_i64_op(st.st_mtime, OP_GE, now-1);
|
||||
|
||||
done:
|
||||
;
|
||||
|
Loading…
Reference in New Issue
Block a user