https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114260
--- Comment #2 from Howard Hinnant <howard.hinnant at gmail dot com> --- This: 2024-03-05 00:00:00 2024-03-04 23:59:33 looks like correct output to me. sys_time and utc_time map to the same civil calendar date/time (except during a leap second). That is 2024-03-05 00:00:00 sys_time, converted to utc_time, also should print out as 2024-03-05 00:00:00. The only difference between sys_time and utc_time is that utc_time counts the leap seconds since 1970 (27 at this point). This means if you look at the difference in .time_since_epoch(), utc_time will be 27 seconds longer, even though it prints out as the same date and time. Consequently, 2024-03-05 00:00:00 in utc_time is *not* a multiple of 86400s, but rather 27s greater than a multiple of 86400s. And all round<days>(udays) does is round the .time_since_epoch() to the nearest multiple of 86400s. Which in utc_time is 27s earlier, or 2024-03-04 23:59:33. So in summary, if you make a change, and *don't* get this output, then you've introduced a bug.
