October 2025 - tz - lists.iana.org

[PROPOSED] New compile-time option THREAD_TM_MULTI
by Paul Eggert Oct. 14, 2025

Oct. 14, 2025

* Makefile, NEWS: Mention it. * localtime.c (THREAD_TM_MULTI): Default to 0. The remaining changes affect only the THREAD_TM_MULTI case. (enum tm_multi): New enum. (N_TM_MULTI): New constant. (tm_multi_key, tm_multi_key_err): New static vars. (tm_multi): New static function. (localtime, gmtime, offtime): Use it. --- Makefile | 8 ++++++++ NEWS | 9 +++++++++ localtime.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++++--- 3 files changed, 70 insertions(+), 3 deletions(-) diff --git a/Makefile b/Makefile index 8939869a..b172542a 100644 --- a/Makefile +++ b/Makefile @@ -316,6 +316,14 @@ LDLIBS= # This can improve paralellism and thus save real time # if many threads call tzcode functions simultaneously. # It also costs CPU time and thus energy. +# -DTHREAD_TM_MULTI to have gmtime, localtime, and offtime +# return different struct tm * addresses in different threads. +# This supports unportable programs that call +# gmtime/localtime/offtime when they should call +# gmtime_r/localtime_r/offtime_r to avoid races. +# Because the corresponding storage is freed on thread exit, +# this option is incompatible with POSIX.1-2024 and earlier. +# It also costs CPU time and memory. # -Dtime_tz=\"T\" to use T as the time_t type, rather than the system time_t # This is intended for internal use only; it mangles external names. # -DTZ_CHANGE_INTERVAL=N if functions depending on TZ should check diff --git a/NEWS b/NEWS index 5bfb7bb4..46bf48ed 100644 --- a/NEWS +++ b/NEWS @@ -80,6 +80,15 @@ Unreleased, experimental changes rarely changing and many threads call tzcode simultaneously. It costs more CPU time and energy. + The new CFLAGS option -TTHREAD_TM_MULTI causes localtime to return + a pointer to thread-specific memory, as FreeBSD does, instead of + to the same memory in all threads. This supports unportable + programs that incorrectly use localtime instead of localtime_r. + This option affects gmtime and offtime similarly to localtime. + Because the corresponding storage is freed on thread exit, this + option is incompatible with POSIX.1-2024 and earlier. It also + costs CPU time and memory. + tzcode now uses mempcpy if available, guessing its availability. Compile with -DHAVE_MEMPCPY=1 or 0 to override the guess. diff --git a/localtime.c b/localtime.c index 1191f810..709b954c 100644 --- a/localtime.c +++ b/localtime.c @@ -52,6 +52,10 @@ struct stat { char st_ctime, st_dev, st_ino; } # define THREAD_RWLOCK 0 #endif +#ifndef THREAD_TM_MULTI +# define THREAD_TM_MULTI 0 +#endif + #if THREAD_SAFE # include <pthread.h> @@ -171,6 +175,52 @@ once(once_t *once_control, void (*init_routine)(void)) #endif } +enum tm_multi { LOCALTIME_TM_MULTI, GMTIME_TM_MULTI, OFFTIME_TM_MULTI }; + +#if THREAD_SAFE && THREAD_TM_MULTI + +enum { N_TM_MULTI = OFFTIME_TM_MULTI + 1 }; +static pthread_key_t tm_multi_key; +static int tm_multi_key_err; + +static void +tm_multi_key_init(void) +{ + tm_multi_key_err = pthread_key_create(&tm_multi_key, free); +} + +#endif + +/* Return TMP, or a thread-specific struct tm * selected by WHICH. */ +static struct tm * +tm_multi(struct tm *tmp, ATTRIBUTE_MAYBE_UNUSED enum tm_multi which) +{ +#if THREAD_SAFE && THREAD_TM_MULTI + /* It is OK to check is_threaded() separately here; even if it + returns a different value in other places in the caller, + this function's behavior is still valid. */ + if (is_threaded()) { + /* Try to get a thread-specific struct tm *. + Fall back on TMP if this fails. */ + static pthread_once_t tm_multi_once = PTHREAD_ONCE_INIT; + pthread_once(&tm_multi_once, tm_multi_key_init); + if (!tm_multi_key_err) { + struct tm *p = pthread_getspecific(tm_multi_key); + if (!p) { + p = malloc(N_TM_MULTI * sizeof *p); + if (p && pthread_setspecific(tm_multi_key, p) != 0) { + free(p); + p = NULL; + } + } + if (p) + return &p[which]; + } + } +#endif + return tmp; +} + /* Unless intptr_t is missing, pacify gcc -Wcast-qual on char const * exprs. Use this carefully, as the casts disable type checking. This is a macro so that it can be used in static initializers. */ @@ -2156,7 +2206,7 @@ localtime(const time_t *timep) # if !SUPPORT_C89 static struct tm tm; # endif - return localtime_tzset(timep, &tm, true); + return localtime_tzset(timep, tm_multi(&tm, LOCALTIME_TM_MULTI), true); } struct tm * @@ -2208,7 +2258,7 @@ gmtime(const time_t *timep) # if !SUPPORT_C89 static struct tm tm; # endif - return gmtime_r(timep, &tm); + return gmtime_r(timep, tm_multi(&tm, GMTIME_TM_MULTI)); } # if STD_INSPIRED @@ -2229,7 +2279,7 @@ offtime(time_t const *timep, long offset) # if !SUPPORT_C89 static struct tm tm; # endif - return offtime_r(timep, offset, &tm); + return offtime_r(timep, offset, tm_multi(&tm, OFFTIME_TM_MULTI)); } # endif -- 2.48.1

1 0

[PROPOSED 1/3] Define fullname only if needed
by Paul Eggert Oct. 14, 2025

Oct. 14, 2025

* localtime.c (union local_storage): Define fullname member only if needed. This saves space on unlikely platforms where sizeof (struct file_analysis) < PATH_MAX, and is closer to FreeBSD. (tzloadbody): Mention fullname only if declared. --- localtime.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/localtime.c b/localtime.c index a78962a6..c9836b2d 100644 --- a/localtime.c +++ b/localtime.c @@ -771,7 +771,7 @@ union local_storage { struct state st; } u; -#ifdef PATH_MAX +#if defined PATH_MAX && !OPENAT_TZDIR && !SUPPRESS_TZDIR /* The name of the file to be opened. */ char fullname[PATH_MAX]; #endif @@ -890,7 +890,7 @@ tzloadbody(char const *name, struct state *sp, char tzloadflags, cp = mempcpy(lsp, tzdirslash, tzdirslashlen); cp = mempcpy(cp, name, namelen); *cp = '\0'; -#ifdef PATH_MAX +#if defined PATH_MAX && !OPENAT_TZDIR && !SUPPRESS_TZDIR name = lsp->fullname; #else name = (char *) lsp; -- 2.48.1

1 0

[PROPOSED 1/2] Simplify function parameter decl style
by Paul Eggert Oct. 14, 2025

Oct. 14, 2025

* localtime.c (time2sub, time2, time1): Use simpler style for declaring function parameters. Although this style has worked since C89, I didn’t know about it until reading Jens Gustedt’s book Modern C. Switching to this style now can simplify future future changes that also use a functional programming style. --- localtime.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/localtime.c b/localtime.c index 63b8ea0a..24972b20 100644 --- a/localtime.c +++ b/localtime.c @@ -2442,8 +2442,8 @@ mktmcpy(struct tm *dest, struct tm const *src) static time_t time2sub(struct tm *const tmp, - struct tm *(*funcp)(struct state const *, time_t const *, - int_fast32_t, struct tm *), + struct tm *funcp(struct state const *, time_t const *, + int_fast32_t, struct tm *), struct state const *sp, const int_fast32_t offset, bool *okayp, @@ -2661,8 +2661,8 @@ label: static time_t time2(struct tm * const tmp, - struct tm *(*funcp)(struct state const *, time_t const *, - int_fast32_t, struct tm *), + struct tm *funcp(struct state const *, time_t const *, + int_fast32_t, struct tm *), struct state const *sp, const int_fast32_t offset, bool *okayp) @@ -2680,8 +2680,8 @@ time2(struct tm * const tmp, static time_t time1(struct tm *const tmp, - struct tm *(*funcp)(struct state const *, time_t const *, - int_fast32_t, struct tm *), + struct tm *funcp(struct state const *, time_t const *, + int_fast32_t, struct tm *), struct state const *sp, const int_fast32_t offset) { -- 2.48.1

1 1

[PROPOSED] Port recent PATH_MAX change to strict C23
by Paul Eggert Oct. 14, 2025

Oct. 14, 2025

* localtime.c (tzdirslashlen): Now an enum, not a const size_t, as strict C23 requires this for the static_assert that uses tzdirslashlen when PATH_MAX is defined. --- localtime.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/localtime.c b/localtime.c index 3f3bdce5..ec13ed71 100644 --- a/localtime.c +++ b/localtime.c @@ -676,7 +676,7 @@ union input_buffer { ATTRIBUTE_NONSTRING #endif static char const tzdirslash[sizeof TZDIR + OPENAT_TZDIR] = TZDIR "/"; -static size_t const tzdirslashlen = sizeof TZDIR; +enum { tzdirslashlen = sizeof TZDIR }; /* Local storage needed for 'tzloadbody'. */ union local_storage { -- 2.48.1

3 5

[PROPOSED 1/2] Optionally use read-write locks instead of mutexes
by Paul Eggert Oct. 13, 2025

Oct. 13, 2025

This avoids bottlenecks when many threads access local time but no thread is changing TZ. It costs CPU time and energy. * Makefile, NEWS: Mention this. * localtime.c (THREAD_RWLOCK): New macro, defaulting to 0. (locallock) [THREAD_RWLOCK]: Now of type pthread_rwlock_t, not pthread_mutex_t. All uses changed. (rd2wrlock): New function. (tzset_unlocked): If THREAD_RWLOCK, upgrade the read lock to a write lock if needed. (tzset_unlocked, gmtcheck): Upgrade the read to a write lock in the unlikely case where writing is needed. --- Makefile | 5 +++ NEWS | 5 +++ localtime.c | 100 +++++++++++++++++++++++++++++++++++++--------------- 3 files changed, 82 insertions(+), 28 deletions(-) diff --git a/Makefile b/Makefile index 85c2c50b..c2c82c3a 100644 --- a/Makefile +++ b/Makefile @@ -304,6 +304,11 @@ LDLIBS= # -DTHREAD_SAFE to make localtime.c thread-safe, as POSIX requires; # not needed by the main-program tz code, which is single-threaded. # Append other compiler flags as needed, e.g., -pthread on GNU/Linux. +# The following option can also be used: +# -DTHREAD_RWLOCK to use read-write locks intead of mutexes. +# This can improve paralellism and thus save real time +# if many threads call tzcode functions simultaneously. +# It also costs CPU time and thus energy. # -Dtime_tz=\"T\" to use T as the time_t type, rather than the system time_t # This is intended for internal use only; it mangles external names. # -DTZ_CHANGE_INTERVAL=N if functions depending on TZ should check diff --git a/NEWS b/NEWS index 64d9336d..8ba97ab4 100644 --- a/NEWS +++ b/NEWS @@ -68,6 +68,11 @@ Unreleased, experimental changes transition begun in release 96k, which removed spaces in tzdata because the spaces break time string parsers. + The new CFLAGS option -DTHREAD_RWLOCK uses read-write locks, as + macOS does, instead of mutexes. This saves real time when TZ is + rarely changing and many threads call tzcode simultaneously. + It costs more CPU time and energy. + tzcode now uses mempcpy if available, guessing its availability. Compile with -DHAVE_MEMPCPY=1 or 0 to override the guess. diff --git a/localtime.c b/localtime.c index ec13ed71..3a38ef04 100644 --- a/localtime.c +++ b/localtime.c @@ -44,16 +44,39 @@ struct stat { char st_ctime, st_dev, st_ino; } # define st_ctim st_ctimespec #endif +#ifndef THREAD_RWLOCK +# define THREAD_RWLOCK 0 +#endif + #if defined THREAD_SAFE && THREAD_SAFE # include <pthread.h> +# if THREAD_RWLOCK +static pthread_rwlock_t locallock = PTHREAD_RWLOCK_INITIALIZER; +static int lock(void) { return pthread_rwlock_rdlock(&locallock); } +static void unlock(void) { pthread_rwlock_unlock(&locallock); } +# else static pthread_mutex_t locallock = PTHREAD_MUTEX_INITIALIZER; static int lock(void) { return pthread_mutex_lock(&locallock); } static void unlock(void) { pthread_mutex_unlock(&locallock); } +# endif #else static int lock(void) { return 0; } static void unlock(void) { } #endif +/* Upgrade a read lock to a write lock. + Return 0 on success, an errno value otherwise. */ +static int +rd2wrlock(void) +{ +# if THREAD_RWLOCK + unlock(); + return pthread_rwlock_wrlock(&locallock); +# else + return 0; +# endif +} + /* Unless intptr_t is missing, pacify gcc -Wcast-qual on char const * exprs. Use this carefully, as the casts disable type checking. This is a macro so that it can be used in static initializers. */ @@ -1733,38 +1756,55 @@ zoneinit(struct state *sp, char const *name, char tzloadflags) } /* Like tzset(), but in a critical section. + If THREAD_RWLOCK the caller has a read lock, + and this function might ugrade it to a write lock. If tz_change_interval is positive the time is NOW; otherwise ignore NOW. */ static void tzset_unlocked(monotime_t now) { - char const *name = getenv("TZ"); - struct state *sp = lclptr; - char tzloadflags = TZLOAD_FROMENV | TZLOAD_TZSTRING; - size_t namelen = sizeof lcl_TZname + 1; /* placeholder for no name */ - - if (name) { - namelen = strnlen(name, sizeof lcl_TZname); - - /* Abbreviate a string like "/usr/share/zoneinfo/America/Los_Angeles" - to its shorter equivalent "America/Los_Angeles". */ - if (!SUPPRESS_TZDIR && tzdirslashlen < namelen - && memcmp(name, tzdirslash, tzdirslashlen) == 0) { - char const *p = name + tzdirslashlen; - while (*p == '/') - p++; - if (*p && *p != ':') { - name = p; - namelen = strnlen(name, sizeof lcl_TZname); - tzloadflags |= TZLOAD_TZDIR_SUB; + char const *name; + struct state *sp; + char tzloadflags; + size_t namelen; + bool writing = false; + + for (;;) { + name = getenv("TZ"); + sp = lclptr; + tzloadflags = TZLOAD_FROMENV | TZLOAD_TZSTRING; + namelen = sizeof lcl_TZname + 1; /* placeholder for no name */ + + if (name) { + namelen = strnlen(name, sizeof lcl_TZname); + + /* Abbreviate a string like "/usr/share/zoneinfo/America/Los_Angeles" + to its shorter equivalent "America/Los_Angeles". */ + if (!SUPPRESS_TZDIR && tzdirslashlen < namelen + && memcmp(name, tzdirslash, tzdirslashlen) == 0) { + char const *p = name + tzdirslashlen; + while (*p == '/') + p++; + if (*p && *p != ':') { + name = p; + namelen = strnlen(name, sizeof lcl_TZname); + tzloadflags |= TZLOAD_TZDIR_SUB; + } } } + + if ((tz_change_interval <= 0 ? tz_change_interval < 0 : fresh_tzdata(now)) + && (name + ? 0 < lcl_is_set && strcmp(lcl_TZname, name) == 0 + : lcl_is_set < 0)) + return; + + if (!THREAD_RWLOCK || writing) + break; + if (rd2wrlock() != 0) + return; + writing = true; } - if ((tz_change_interval <= 0 ? tz_change_interval < 0 : fresh_tzdata(now)) - && (name - ? 0 < lcl_is_set && strcmp(lcl_TZname, name) == 0 - : lcl_is_set < 0)) - return; # if ALL_STATE if (! sp) lclptr = sp = malloc(sizeof *lclptr); @@ -1829,12 +1869,16 @@ gmtcheck(void) if (lock() != 0) return; if (! gmt_is_set) { + if (rd2wrlock() != 0) + return; + if (!THREAD_RWLOCK || !gmt_is_set) { #if ALL_STATE - gmtptr = malloc(sizeof *gmtptr); + gmtptr = malloc(sizeof *gmtptr); #endif - if (gmtptr) - gmtload(gmtptr); - gmt_is_set = true; + if (gmtptr) + gmtload(gmtptr); + gmt_is_set = true; + } } unlock(); } -- 2.51.0

1 0

[PROPOSED] Port better to platforms lacking mempcpy
by Paul Eggert Oct. 13, 2025

Oct. 13, 2025

* private.h (mempcpy) [!HAVE_MEMPCPY]: Args are pointer to void, not pointer to char, for compatibility with glibc. --- private.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/private.h b/private.h index f6f12aa2..befa6db1 100644 --- a/private.h +++ b/private.h @@ -844,7 +844,7 @@ extern char **environ; #endif #if !HAVE_MEMPCPY static void * -mempcpy(char *restrict s1, char const *restrict s2, size_t n) +mempcpy(void *restrict s1, void const *restrict s2, size_t n) { char *p = memcpy(s1, s2, n); return p + n; -- 2.51.0

1 0

[PROPOSED 1/3] tzset sets errno if lock fails
by Paul Eggert Oct. 10, 2025

Oct. 10, 2025

* localtime.c (tzset): Set errno if the lock fails. This should never happen, but if it does the least we can do is set errno, as other functions do. --- localtime.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/localtime.c b/localtime.c index 54db0ab7..ef094560 100644 --- a/localtime.c +++ b/localtime.c @@ -1788,8 +1788,11 @@ void tzset(void) { monotime_t now = get_monotonic_time(); - if (lock() != 0) + int err = lock(); + if (err != 0) { + errno = err; return; + } tzset_unlocked(now); unlock(); } -- 2.51.0

1 3

[Feature Request] Add validation in zic for invalid DST intervals
by 任春辉 Oct. 10, 2025

Oct. 10, 2025

Hello tzdata maintainers, I would like to propose a feature enhancement for the zic compiler. Currently, the NOS time zone rules are defined as: Rule NOS 1970 2037 - Jan 27 23:43:01 2:0 SDST Rule NOS 1970 2037 - Jan 28 0:51:01 0 S Zone OS/NOS +7:0:0 NOS NO%s Issue: - The first rule enters DST at 23:43:01 with a +2 hour offset, moving local time to 01:43:01. - The second rule is intended to revert to standard time at 00:51:01. - Due to the offset, 00:51:01 local time never occurs, so the second rule will never trigger. - This creates unreachable local times and may cause incorrect results in time calculations. Proposal: - Add validation in the zic command to detect DST rules where the offset causes the subsequent rule's local time to be unreachable. - The validation could issue a warning or error to alert the author of potentially invalid rules. - This would improve the robustness of time zone definitions and prevent subtle bugs in downstream applications. Thank you for your consideration. Best regards

2 1

[PROPOSED 1/2] Simplify port to NetBSD struct __state
by Paul Eggert Oct. 8, 2025

Oct. 8, 2025

NetBSD uses ‘struct __state’ where tzcode uses ‘struct state’. tzcode 2014g worked around this by a tricky #define and #undef of the relevant names. However, as I recall, this was motivated by an old compatibility issue as to whether the functions would take pointer-to-const or plain pointer. That issue has been resolved in favor of the latter (which is what NetBSD did all along, despite its attempt to do the former), so use a less-tricky workaround similar to what NetBSD libc is already using. * private.h (localtime_rz, mktime_z, posix2time_z, time2posix_z) (timezone_t, tzalloc, tzfree): Do not #define and #undef. (state) [NETBSD_INSPIRED && _NETBSD_SOURCE]: New macro. --- private.h | 28 +++++++--------------------- 1 file changed, 7 insertions(+), 21 deletions(-) diff --git a/private.h b/private.h index 23f1d9fb..4599d0b9 100644 --- a/private.h +++ b/private.h @@ -194,30 +194,15 @@ ** Nested includes */ -/* Avoid clashes with NetBSD by renaming NetBSD's declarations. - If defining the 'timezone' variable, avoid a clash with FreeBSD's +/* If defining the 'timezone' variable, avoid a clash with FreeBSD's 'timezone' function by renaming its declaration. */ -#define localtime_rz sys_localtime_rz -#define mktime_z sys_mktime_z -#define posix2time_z sys_posix2time_z -#define time2posix_z sys_time2posix_z #if defined USG_COMPAT && USG_COMPAT == 2 # define timezone sys_timezone #endif -#define timezone_t sys_timezone_t -#define tzalloc sys_tzalloc -#define tzfree sys_tzfree #include <time.h> -#undef localtime_rz -#undef mktime_z -#undef posix2time_z -#undef time2posix_z #if defined USG_COMPAT && USG_COMPAT == 2 # undef timezone #endif -#undef timezone_t -#undef tzalloc -#undef tzfree #include <stddef.h> @@ -920,15 +905,16 @@ time_t posix2time(time_t); #endif /* -** Define functions that are ABI compatible with NetBSD but have -** better prototypes. NetBSD 6.1.4 defines a pointer type timezone_t -** and labors under the misconception that 'const timezone_t' is a -** pointer to a constant. This use of 'const' is ineffective, so it -** is not done here. What we call 'struct state' NetBSD calls +** Define functions that are ABI compatible with NetBSD. +** What we call 'struct state' NetBSD calls ** 'struct __state', but this is a private name so it doesn't matter. */ #if NETBSD_INSPIRED +# ifdef _NETBSD_SOURCE +# define state __state +# else typedef struct state *timezone_t; +# endif struct tm *localtime_rz(timezone_t restrict, time_t const *restrict, struct tm *restrict); time_t mktime_z(timezone_t restrict, struct tm *restrict); -- 2.51.0

1 1

[PROPOSED] Support FreeBSD-style TZif polling
by Paul Eggert Oct. 7, 2025

Oct. 7, 2025

I installed the attached proposed patch, which adds a tzcode option to periodically poll the TZif file to see whether it changed. You can enable this option by compiling with -DTZ_CHANGE_INTERVAL=61 (or choose your own favorite polling interval instead of 61 seconds). The option affects localtime and other functions whose behavior depends on the TZ environment variable (or its absence); it does not affect tzalloc-related functions. This patch affects tzcode's behavior only if you compile with the new option. The remaining major difference between the functionalities of tzcode and FreeBSD is FreeBSD's support for calling localtime (not localtime_r) from multiple threads. I plan to take a look at that some time soon, with the idea of adding that as another tzcode option. While preparing this patch I noticed some issues in FreeBSD localtime.c, noted below. The attached patch addresses these issues. * A bug can occur in the first minute after reboot, where FreeBSD’s tzdata_is_fresh assumes that since the monotonic clock is less than 61, the cache must be fresh. * Prefer CLOCK_MONOTONIC_COARSE aka CLOCK_MONOTONIC_FAST if available, as its precision suffices here and it should be more efficient than CLOCK_MONOTONIC. * When checking the file timestamp, the code should not fail merely because fstat fails due to EOVERFLOW. Instead, it should assume the file has changed, as this results in better behavior for the user. * When checking the file timestamp the code should not use strlcpy to copy its name. The strlcpy could silently truncate the name in the unlikely case where a relative file name is shorter than PATH_MAX but the absolute file name is not. More importantly, the file contents are what matter here, and since the file name does not matter there is no need to copy it. * It’s a bit better to check st_ctim, not just st_ctime, as st_ctim has subsecond resolution and is more likely to avoid false matches on platforms with dicey st_dev or st_ino. * For efficiency do not bother checking st_mtim/st_mtime, as st_ctim is updated whenever st_mtim is. * For efficiency do not call fstat twice in the unlikely case where TZ is an absolute pathname. * tzdata_is_fresh is poorly named, as it returns 0 if the cache is fresh, 1 if not. And as a nit, in traditional English, “data” is a plural word. The attached patch uses a function fresh_tzdata instead, with the more common interpretation of true and false. * The attached patch ports to platforms with 32-bit time_t and without tm_gmtoff. On these platforms, strftime.c includes localtime.c and localtime.h uses ‘#define time_t timex_t’. For this reason, fresh_tzdata cannot use time_t (which might be timex_t); it needs to use the underlying system time_t. * The attached patch ports to platforms lacking <sys/stat.h>, struct timespec, clock_gettime, or CLOCK_MONOTONIC/CLOCK_MONOTONIC_COARSE.

4 13