Does CRT really support well-known timezones, like TZ=UTC? I have a feeling that it accepts value of TZ variable as long as it can parse it according to format documented for _tzset[1].
Format for TZ supported by CRT is subset of what POSIX[2] describes. In particular, CRT does not support specifying exact time when the switch happens, but only offset from UTC. Microsoft docs say that CRT always uses US rules for this, which is very unfortunate. Another thing that I think worth mentioning - it seems that CRT functions _ftime and localtime will always try to reparse value of TZ and update timezone information (variables _daylight, _dstbias, _timezone and _tzname). If two functions (any of _tzset, _ftime or localtime) are called from different threads at the same time, and value of TZ was different upon entry, this may result in inconsistent result. I don't think there's much we can do about. [1] https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/tzset [2] https://pubs.opengroup.org/onlinepubs/9799919799/basedefs/V1_chap08.html ________________________________ From: Martin Storsjö <[email protected]> Sent: Monday, November 17, 2025 11:18 PM To: Pali Rohár <[email protected]> Cc: [email protected] <[email protected]> Subject: Re: [Mingw-w64-public] [PATCH 7/7] crt: Extend t_time test for _localtime64 and _mktime64 On Sat, 25 Oct 2025, Pali Rohár wrote: > --- > mingw-w64-crt/testcases/t_time.c | 96 ++++++++++++++++++++++++++++++-- > 1 file changed, 92 insertions(+), 4 deletions(-) > > diff --git a/mingw-w64-crt/testcases/t_time.c > b/mingw-w64-crt/testcases/t_time.c > index 50b07d0083a7..75a15d816d31 100644 > --- a/mingw-w64-crt/testcases/t_time.c > +++ b/mingw-w64-crt/testcases/t_time.c > @@ -10,14 +10,17 @@ int __cdecl ftime64(struct __timeb64 *tb64); > > int main() > { > - time_t t; > - __time32_t t32; > - __time64_t t64; > + time_t t, t_; > + __time32_t t32, t32_; > + __time64_t t64, t64_; > struct timeb tb; > struct _timeb tb_; > struct __timeb32 tb32; > struct __timeb64 tb64; > struct tm *htm; > + struct tm tm1; > + struct tm tm2; > + struct tm tm3; > int ret1, ret2, ret3; > const char *str; > const wchar_t *wstr; > @@ -95,7 +98,49 @@ int main() > assert (htm->tm_yday == 74); > assert (htm->tm_isdst == 0); > > - /* ctime returns time string in local timezone, so set local timezone to > UTC to have test timezone independent */ > + time_t times[] = { > + 1700000000 /* Tue Nov 14 22:13:20 UTC 2023 */, > + 1600000000 /* Sun Sep 13 12:26:40 UTC 2020 */, > + }; > + for (size_t i = 0; i < sizeof(times)/sizeof(*times); i++) { > + t = times[i]; > + htm = localtime (&t); > + tm1 = *htm; > + printf ("localtime(%lld): sec=%d min=%d hour=%d mday=%d mon=%d > year=%d wday=%d yday=%d isdst=%d\n", > + (long long)t, htm->tm_sec, htm->tm_min, htm->tm_hour, htm->tm_mday, > htm->tm_mon, htm->tm_year, htm->tm_wday, htm->tm_yday, htm->tm_isdst); > + t_ = mktime (htm); > + printf ("mktime(): %lld sec=%d min=%d hour=%d mday=%d mon=%d > year=%d wday=%d yday=%d isdst=%d\n", > + (long long)t_, htm->tm_sec, htm->tm_min, htm->tm_hour, htm->tm_mday, > htm->tm_mon, htm->tm_year, htm->tm_wday, htm->tm_yday, htm->tm_isdst); > + assert (t_ == t); > + assert (memcmp (htm, &tm1, sizeof(tm1)) == 0); > + > + t32 = t; > + htm = _localtime32 (&t32); > + tm2 = *htm; > + printf ("_localtime32(%d): sec=%d min=%d hour=%d mday=%d mon=%d year=%d > wday=%d yday=%d isdst=%d\n", > + t32, htm->tm_sec, htm->tm_min, htm->tm_hour, htm->tm_mday, > htm->tm_mon, htm->tm_year, htm->tm_wday, htm->tm_yday, htm->tm_isdst); > + t32_ = _mktime32 (htm); > + printf ("_mktime32(): %d sec=%d min=%d hour=%d mday=%d mon=%d year=%d > wday=%d yday=%d isdst=%d\n", > + t32_, htm->tm_sec, htm->tm_min, htm->tm_hour, htm->tm_mday, > htm->tm_mon, htm->tm_year, htm->tm_wday, htm->tm_yday, htm->tm_isdst); > + assert (t32_ == t32); > + assert (memcmp (htm, &tm2, sizeof(tm2)) == 0); > + > + t64 = t; > + htm = _localtime64 (&t64); > + tm3 = *htm; > + printf ("_localtime64(%lld): sec=%d min=%d hour=%d mday=%d mon=%d > year=%d wday=%d yday=%d isdst=%d\n", > + t64, htm->tm_sec, htm->tm_min, htm->tm_hour, htm->tm_mday, > htm->tm_mon, htm->tm_year, htm->tm_wday, htm->tm_yday, htm->tm_isdst); > + t64_ = _mktime64 (htm); > + printf ("_mktime64(): %lld sec=%d min=%d hour=%d mday=%d mon=%d > year=%d wday=%d yday=%d isdst=%d\n", > + t64_, htm->tm_sec, htm->tm_min, htm->tm_hour, htm->tm_mday, > htm->tm_mon, htm->tm_year, htm->tm_wday, htm->tm_yday, htm->tm_isdst); > + assert (t64_ == t64); > + assert (memcmp (htm, &tm3, sizeof(tm3)) == 0); > + > + assert (memcmp (&tm1, &tm2, sizeof(tm1)) == 0); > + assert (memcmp (&tm2, &tm3, sizeof(tm2)) == 0); > + } > + > + /* ctime, localtime and mktime returns time string in local timezone, so > set local timezone to UTC to have test timezone independent */ > putenv ("TZ=UTC"); In addition to setting TZ=UTC here, is it possible to set some other well-known timezone that would be available everywhere, which we could use for some testcases for cases closer to the DST switchover time? We could do that for both mktime and localtime IMO. In itself, these test additions probably are fine though - the above would be nice extras. But given that corner case handling close to DST switchovers, I think it would be good to actually check such a case. So this patchset in itself looks ok, except for the one issue I found in 2/7. // Martin _______________________________________________ Mingw-w64-public mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mingw-w64-public _______________________________________________ Mingw-w64-public mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
