- tz - lists.iana.org

Etc/GMT timezone offsets are inverted
by Pedro A. López-Valencia July 11, 2017

July 11, 2017

I just noticed that all GMT+- timezones are inverted so that GMT+5 is really GMT-5. This applies to all GMT-offset timezone definitions. -- https://about.me/palopezv

2 1

IERS Bulletin C54
by Michael Deckers July 11, 2017

July 11, 2017

The IERS has issued Bulletin C54 which assures that no leap second will occur near 2018-01-01. Interestingly, the issue date 2017-07-06 in the txt version at [https://datacenter.iers.org/eop/-/somos/5Rgv/getTX/16/bulletinc-054.txt] differs from the issue date 2017-07-09 in the xml version at [https://datacenter.iers.org/eop/-/somos/5Rgv/latestXL/16/bulletinc-054.txt/…] Michael Deckers. --- This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus

5 7

backzone/Europe/Guernsey
by Michael Deckers July 10, 2017

July 10, 2017

For the connoisseurs of olde time stamps: The "LMT location" for backzone/Europe/Guernsey is given as "49.27N -2.33E" -- seemingly in degrees and decimal fractions of degrees. But the point ( +49.27°, -2.33° ) = ( +49°16′12″, -002°19′48″ ) is in open waters about 15 km south of Guernsey and east of Sark in the Channel. The point ( +49°27′, -002°33′ ), however, is in St Peter Port on the island, so that arc minutes and not hundredths of degrees must be meant. This also agrees with zone.tab. Hence: -# LMT Location - 49.27N -2.33E - St.Peter Port +# LMT Location - 49°27′N -2°33′E - St.Peter Port -Zone Europe/Guernsey -0:09:19 - LMT 1913 Jun 18 +Zone Europe/Guernsey -0:10:12 - LMT 1913 Jun 18 Michael Deckers. --- This email has been checked for viruses by Avast antivirus software. https://www.avast.com/antivirus

3 3

Re: [tz] TZ & AIX
by Paul Eggert July 8, 2017

July 8, 2017

Jorge Fábregas wrote: > https://www.ibm.com/developerworks/aix/library/au-aix-olson-time-zone/index… Thanks for the heads-up. I installed the attached.

2 1

Orillia's 1912 DST pops up again in Toronto Star
by Paul Eggert July 8, 2017

July 8, 2017

Today's Toronto Star reports on the two-week daylight saving time interlude in Orillia, Ontario. Today's article doesn't mention the same newspaper's 1997-10-14 article on the same topic, and it's not clear what's new this time around. Still, the story is amusingly told. Proposed patch attached.

1 0

DST in the (odd) news
by Arthur David Olson July 7, 2017

July 7, 2017

"Alarm clock removed from family's home after 14 years of daily beeping" "Over the years, it beeped at 6.50pm or 7.50pm each day, depending on whether Daylight [Saving] Time was in place." https://www.theguardian.com/us-news/2017/jul/07/alarm-clock-wall-pennsylvan… @dashdashado

1 0

[PROPOSED] Avoid zic problems with errno == EOVERFLOW
by Paul Eggert June 26, 2017

June 26, 2017

Although this patch does not fix any bugs, it makes zic’s ‘stat’ calls less likely to fail unnecessarily with errno == EOVERFLOW, and this should be a bit faster in some cases. Problem reported by Nick Alcock in: https://sourceware.org/ml/libc-alpha/2017-06/msg01264.html * private.h (_FILE_OFFSET_BITS, _LARGE_FILES) (_DARWIN_USE_64_BIT_INODE): New macros. --- private.h | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/private.h b/private.h index f5e39b5..2e8415e 100644 --- a/private.h +++ b/private.h @@ -102,6 +102,16 @@ /* Enable strtoimax on pre-C99 Solaris 11. */ #define __EXTENSIONS__ 1 +/* To avoid having 'stat' fail unnecessarily with errno == EOVERFLOW, + enable large files on GNUish systems ... */ +#ifndef _FILE_OFFSET_BITS +# define _FILE_OFFSET_BITS 64 +#endif +/* ... and on AIX ... */ +#define _LARGE_FILES 1 +/* ... and enable large inode numbers on Mac OS X 10.5 and later. */ +#define _DARWIN_USE_64_BIT_INODE 1 + /* ** Nested includes */ -- 2.9.4

1 0

[PROPOSED] Port to GCC 4.8.4
by Paul Eggert June 16, 2017

June 16, 2017

Problem reported by Stan Shebs and Joseph Myers in: https://sourceware.org/ml/libc-alpha/2017-06/msg00741.html * NEWS, Makefile: Document this. * private.h (HAVE_GENERIC): New macro, so that we do not use _Generic with GCC 4.8. (TIME_T_MIN_NO_PADDING, TIME_T_MAX_NO_PADDING): New macros. (TIME_T_MIN, TIME_T_MAX): Rename from time_t_min and time_t_max, and turn them into macros so that problems are less likely to occur for subset uses like glibc where these macros are never expanded. All uses changed. Use new macros to work around GCC 4.8 bug. --- Makefile | 1 + NEWS | 3 +++ date.c | 2 +- localtime.c | 26 +++++++++++++------------- private.h | 60 ++++++++++++++++++++++++++++++++++++++++-------------------- 5 files changed, 58 insertions(+), 34 deletions(-) diff --git a/Makefile b/Makefile index 2b31f3f..347ece7 100644 --- a/Makefile +++ b/Makefile @@ -142,6 +142,7 @@ LDLIBS= # -DHAVE_DECL_ASCTIME_R=0 if <time.h> does not declare asctime_r # -DHAVE_DECL_ENVIRON if <unistd.h> declares 'environ' # -DHAVE_DIRECT_H if mkdir needs <direct.h> (MS-Windows) +# -DHAVE_GENERIC=0 if _Generic does not work # -DHAVE_GETTEXT if 'gettext' works (e.g., GNU/Linux, FreeBSD, Solaris) # -DHAVE_INCOMPATIBLE_CTIME_R if your system's time.h declares # ctime_r and asctime_r incompatibly with the POSIX standard diff --git a/NEWS b/NEWS index ca4fd8e..cfdd620 100644 --- a/NEWS +++ b/NEWS @@ -94,6 +94,9 @@ Unreleased, experimental changes zdump.c no longer assumes snprintf. (Reported by Jonathan Leffler.) + Calculation of time_t extrema works around a bug in GCC 4.8.4 + (Reported by Stan Shebs and Joseph Myers.) + Several minor changes have been made to the code to make it a bit easier to port to MS-Windows and Solaris. (Thanks to Kees Dekker for reporting the problems.) diff --git a/date.c b/date.c index b29e162..2cc533f 100644 --- a/date.c +++ b/date.c @@ -97,7 +97,7 @@ main(const int argc, char *argv[]) secs = strtoimax (optarg, &endarg, 0); if (*endarg || optarg == endarg) errno = EINVAL; - else if (! (time_t_min <= secs && secs <= time_t_max)) + else if (! (TIME_T_MIN <= secs && secs <= TIME_T_MAX)) errno = ERANGE; if (errno) { perror(optarg); diff --git a/localtime.c b/localtime.c index 8c944f3..2e1d621 100644 --- a/localtime.c +++ b/localtime.c @@ -464,17 +464,17 @@ tzloadbody(char const *name, struct state *sp, bool doextend, sp->charcnt = charcnt; /* Read transitions, discarding those out of time_t range. - But pretend the last transition before time_t_min - occurred at time_t_min. */ + But pretend the last transition before TIME_T_MIN + occurred at TIME_T_MIN. */ timecnt = 0; for (i = 0; i < sp->timecnt; ++i) { int_fast64_t at = stored == 4 ? detzcode(p) : detzcode64(p); - sp->types[i] = at <= time_t_max; + sp->types[i] = at <= TIME_T_MAX; if (sp->types[i]) { time_t attime - = ((TYPE_SIGNED(time_t) ? at < time_t_min : at < 0) - ? time_t_min : at); + = ((TYPE_SIGNED(time_t) ? at < TIME_T_MIN : at < 0) + ? TIME_T_MIN : at); if (timecnt && attime <= sp->ats[timecnt - 1]) { if (attime < sp->ats[timecnt - 1]) return EINVAL; @@ -524,7 +524,7 @@ tzloadbody(char const *name, struct state *sp, bool doextend, /* Leap seconds cannot occur before the Epoch. */ if (tr < 0) return EINVAL; - if (tr <= time_t_max) { + if (tr <= TIME_T_MAX) { /* Leap seconds cannot occur more than once per UTC month, and UTC months are at least 28 days long (minus 1 second for a negative leap second). Each leap second's @@ -1795,12 +1795,12 @@ increment_overflow_time(time_t *tp, int_fast32_t j) { /* ** This is like - ** 'if (! (time_t_min <= *tp + j && *tp + j <= time_t_max)) ...', + ** 'if (! (TIME_T_MIN <= *tp + j && *tp + j <= TIME_T_MAX)) ...', ** except that it does the right thing even if *tp + j would overflow. */ if (! (j < 0 - ? (TYPE_SIGNED(time_t) ? time_t_min - j <= *tp : -1 - j < *tp) - : *tp <= time_t_max - j)) + ? (TYPE_SIGNED(time_t) ? TIME_T_MIN - j <= *tp : -1 - j < *tp) + : *tp <= TIME_T_MAX - j)) return true; *tp += j; return false; @@ -1936,8 +1936,8 @@ time2sub(struct tm *const tmp, /* ** Do a binary search (this works whatever time_t's type is). */ - lo = time_t_min; - hi = time_t_max; + lo = TIME_T_MIN; + hi = TIME_T_MAX; for ( ; ; ) { t = lo / 2 + hi / 2; if (t < lo) @@ -1954,12 +1954,12 @@ time2sub(struct tm *const tmp, } else dir = tmcomp(&mytm, &yourtm); if (dir != 0) { if (t == lo) { - if (t == time_t_max) + if (t == TIME_T_MAX) return WRONG; ++t; ++lo; } else if (t == hi) { - if (t == time_t_min) + if (t == TIME_T_MIN) return WRONG; --t; --hi; diff --git a/private.h b/private.h index 2b07251..f5e39b5 100644 --- a/private.h +++ b/private.h @@ -27,6 +27,21 @@ #define HAVE_DECL_ASCTIME_R 1 #endif +#if !defined HAVE_GENERIC && defined __has_extension +# if __has_extension(c_generic_selections) +# define HAVE_GENERIC 1 +# else +# define HAVE_GENERIC 0 +# endif +#endif +/* _Generic is buggy in pre-4.9 GCC. */ +#if !defined HAVE_GENERIC && defined __GNUC__ +# define HAVE_GENERIC (4 < __GNUC__ + (9 <= __GNUC_MINOR__)) +#endif +#ifndef HAVE_GENERIC +# define HAVE_GENERIC (201112 <= __STDC_VERSION__) +#endif + #ifndef HAVE_GETTEXT #define HAVE_GETTEXT 0 #endif /* !defined HAVE_GETTEXT */ @@ -561,27 +576,32 @@ time_t time2posix_z(timezone_t, time_t) ATTRIBUTE_PURE; #define MINVAL(t, b) \ ((t) (TYPE_SIGNED(t) ? - TWOS_COMPLEMENT(t) - MAXVAL(t, b) : 0)) -/* The minimum and maximum finite time values. This implementation - assumes no padding if time_t is signed and either the compiler is - pre-C11 or time_t is not one of the standard signed integer types. */ -#if 201112 <= __STDC_VERSION__ -static time_t const time_t_min - = (TYPE_SIGNED(time_t) - ? _Generic((time_t) 0, - signed char: SCHAR_MIN, short: SHRT_MIN, - int: INT_MIN, long: LONG_MIN, long long: LLONG_MIN, - default: MINVAL(time_t, TYPE_BIT(time_t))) - : 0); -static time_t const time_t_max - = (TYPE_SIGNED(time_t) - ? _Generic((time_t) 0, - signed char: SCHAR_MAX, short: SHRT_MAX, - int: INT_MAX, long: LONG_MAX, long long: LLONG_MAX, - default: MAXVAL(time_t, TYPE_BIT(time_t))) - : -1); +/* The extreme time values, assuming no padding. */ +#define TIME_T_MIN_NO_PADDING MINVAL(time_t, TYPE_BIT(time_t)) +#define TIME_T_MAX_NO_PADDING MAXVAL(time_t, TYPE_BIT(time_t)) + +/* The extreme time values. These are macros, not constants, so that + any portability problem occur only when compiling .c files that use + the macros, which is safer for applications that need only zdump and zic. + This implementation assumes no padding if time_t is signed and + either the compiler lacks support for _Generic or time_t is not one + of the standard signed integer types. */ +#if HAVE_GENERIC +# define TIME_T_MIN \ + _Generic((time_t) 0, \ + signed char: SCHAR_MIN, short: SHRT_MIN, \ + int: INT_MIN, long: LONG_MIN, long long: LLONG_MIN, \ + default: TIME_T_MIN_NO_PADDING) +# define TIME_T_MAX \ + (TYPE_SIGNED(time_t) \ + ? _Generic((time_t) 0, \ + signed char: SCHAR_MAX, short: SHRT_MAX, \ + int: INT_MAX, long: LONG_MAX, long long: LLONG_MAX, \ + default: TIME_T_MAX_NO_PADDING) \ + : (time_t) -1) #else -static time_t const time_t_min = MINVAL(time_t, TYPE_BIT(time_t)); -static time_t const time_t_max = MAXVAL(time_t, TYPE_BIT(time_t)); +# define TIME_T_MIN TIME_T_MIN_NO_PADDING +# define TIME_T_MAX TIME_T_MAX_NO_PADDING #endif /* -- 2.9.4

1 0

[PROPOSED] Port check_time_t_alternatives to June 8 AIX fix
by Paul Eggert June 16, 2017

June 16, 2017

* private.h (TZ_TIME_T): Declare strftime. This is helpful now that we use our own strftime. Declare strftime_l too, if present. --- private.h | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/private.h b/private.h index 982ddf3..2b07251 100644 --- a/private.h +++ b/private.h @@ -424,6 +424,12 @@ typedef time_tz tz_time_t; char *ctime(time_t const *); char *ctime_r(time_t const *, char *); double difftime(time_t, time_t) ATTRIBUTE_CONST; +size_t strftime(char *restrict, size_t, char const *restrict, + struct tm const *restrict); +# if HAVE_STRFTIME_L +size_t strftime_l(char *restrict, size_t, char const *restrict, + struct tm const *restrict, locale_t); +# endif struct tm *gmtime(time_t const *); struct tm *gmtime_r(time_t const *restrict, struct tm *restrict); struct tm *localtime(time_t const *); -- 2.9.4

1 0

Re: [tz] 1 s error in America/Adak and America/Nome
by Paul Eggert June 15, 2017

June 15, 2017

On 06/10/2017 07:42 AM, Michael H Deckers wrote: > I do not understand the evidence for the proposed changes. A change of > time scale by 24 h is only observable though the day of the week and > the day number in a calendar (which may be combined with a switch of > calendars), but not through the time of day. > > Where two successive days are observed as Sundays, this is evidence > for the change of time scale having occurred at the instant between > the two. Yes. However, the the question is whether America/Sitka should record the time kept by those in formal control of Sitka, or the time kept by Toomas Ahllund and his fellow workers who made the transition on Sunday in order to have a longer weekend. If the former, the transition should be at 15:30 Friday (Gregorian); if the latter, it should presumably be at 00:00 Sunday (Gregorian). I chose the former, partly because it reflects formal transfer of control, and partly because we have a specific time-of-day recorded for it. > And if a precise instant for the acquisition of Alaska is > to be taken as the instant of the time zone change, then this > assumption > should be applied to all locations (where nothing else is known). > But the proposed changes to tzdb amount to the change happening when > UT = 1867-10-19 + 00:41:37 h for Yakutat > UT = 1867-10-19 + 00:31:13 h for Sitka, Juneau, Metlakatla > UT = 1867-10-19 + 00:31:03 h for Adak, Nome > UT = 1867-10-19 + 00:30:59 h for Anchorage > I do not see support for such variations. Thanks for reporting that; those were due to bugs in my calculations. I installed the attached patch to fix the bugs, and to try to clarify the point you mentioned.

5 12