commit 16840e52e594e221069efc9bf7b65ebfb5e2d639 Author: Nick Mathewson <nickm@torproject.org> Date: Mon Jan 11 10:03:00 2016 -0500 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. --- src/test/test_util.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/src/test/test_util.c b/src/test/test_util.c index 863105f..37f7d93 100644 --- a/src/test/test_util.c +++ b/src/test/test_util.c @@ -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: ;