From d65595c8dd994b47b7fdfdf9427bb546c577467b Mon Sep 17 00:00:00 2001 From: Gennadiy Rozental Date: Tue, 27 Jun 2023 17:22:39 -0700 Subject: Import of CCTZ from GitHub. PiperOrigin-RevId: 543896343 Change-Id: Ia91b3e082b764b750bbbe9a3ce63192263d51438 --- absl/time/internal/cctz/include/cctz/time_zone.h | 1 + absl/time/internal/cctz/src/time_zone_format.cc | 6 +- absl/time/internal/cctz/src/time_zone_if.cc | 12 +- absl/time/internal/cctz/src/time_zone_if.h | 9 +- absl/time/internal/cctz/src/time_zone_impl.cc | 6 +- absl/time/internal/cctz/src/time_zone_impl.h | 4 + absl/time/internal/cctz/src/time_zone_info.cc | 602 +++++++++++---------- absl/time/internal/cctz/src/time_zone_info.h | 25 +- absl/time/internal/cctz/src/time_zone_libc.cc | 10 +- absl/time/internal/cctz/src/time_zone_libc.h | 8 +- .../internal/cctz/src/time_zone_lookup_test.cc | 4 + 11 files changed, 361 insertions(+), 326 deletions(-) diff --git a/absl/time/internal/cctz/include/cctz/time_zone.h b/absl/time/internal/cctz/include/cctz/time_zone.h index 6e382dc6..b2b0cf6f 100644 --- a/absl/time/internal/cctz/include/cctz/time_zone.h +++ b/absl/time/internal/cctz/include/cctz/time_zone.h @@ -23,6 +23,7 @@ #include #include #include +#include // NOLINT: We use std::ratio in this header #include #include diff --git a/absl/time/internal/cctz/src/time_zone_format.cc b/absl/time/internal/cctz/src/time_zone_format.cc index 0bea75a4..96268a83 100644 --- a/absl/time/internal/cctz/src/time_zone_format.cc +++ b/absl/time/internal/cctz/src/time_zone_format.cc @@ -14,15 +14,13 @@ #if !defined(HAS_STRPTIME) #if !defined(_MSC_VER) && !defined(__MINGW32__) && !defined(__VXWORKS__) -#define HAS_STRPTIME \ - 1 // assume everyone has strptime() except windows - // and VxWorks +#define HAS_STRPTIME 1 // Assume everyone else has strptime(). #endif #endif #if defined(HAS_STRPTIME) && HAS_STRPTIME #if !defined(_XOPEN_SOURCE) && !defined(__OpenBSD__) -#define _XOPEN_SOURCE // Definedness suffices for strptime. +#define _XOPEN_SOURCE // Definedness suffices for strptime(). #endif #endif diff --git a/absl/time/internal/cctz/src/time_zone_if.cc b/absl/time/internal/cctz/src/time_zone_if.cc index 1d44dde7..0e65cd9e 100644 --- a/absl/time/internal/cctz/src/time_zone_if.cc +++ b/absl/time/internal/cctz/src/time_zone_if.cc @@ -23,19 +23,19 @@ ABSL_NAMESPACE_BEGIN namespace time_internal { namespace cctz { -std::unique_ptr TimeZoneIf::Load(const std::string& name) { +std::unique_ptr TimeZoneIf::UTC() { return TimeZoneInfo::UTC(); } + +std::unique_ptr TimeZoneIf::Make(const std::string& name) { // Support "libc:localtime" and "libc:*" to access the legacy // localtime and UTC support respectively from the C library. // NOTE: The "libc:*" zones are internal, test-only interfaces, and // are subject to change/removal without notice. Do not use them. if (name.compare(0, 5, "libc:") == 0) { - return std::unique_ptr(new TimeZoneLibC(name.substr(5))); + return TimeZoneLibC::Make(name.substr(5)); } - // Otherwise use the "zoneinfo" implementation by default. - std::unique_ptr tz(new TimeZoneInfo); - if (!tz->Load(name)) tz.reset(); - return std::unique_ptr(tz.release()); + // Otherwise use the "zoneinfo" implementation. + return TimeZoneInfo::Make(name); } // Defined out-of-line to avoid emitting a weak vtable in all TUs. diff --git a/absl/time/internal/cctz/src/time_zone_if.h b/absl/time/internal/cctz/src/time_zone_if.h index 7d3e42d3..bec9beb5 100644 --- a/absl/time/internal/cctz/src/time_zone_if.h +++ b/absl/time/internal/cctz/src/time_zone_if.h @@ -33,8 +33,9 @@ namespace cctz { // Subclasses implement the functions for civil-time conversions in the zone. class TimeZoneIf { public: - // A factory function for TimeZoneIf implementations. - static std::unique_ptr Load(const std::string& name); + // Factory functions for TimeZoneIf implementations. + static std::unique_ptr UTC(); // never fails + static std::unique_ptr Make(const std::string& name); virtual ~TimeZoneIf(); @@ -51,7 +52,9 @@ class TimeZoneIf { virtual std::string Description() const = 0; protected: - TimeZoneIf() {} + TimeZoneIf() = default; + TimeZoneIf(const TimeZoneIf&) = delete; + TimeZoneIf& operator=(const TimeZoneIf&) = delete; }; // Convert between time_point and a count of seconds since the diff --git a/absl/time/internal/cctz/src/time_zone_impl.cc b/absl/time/internal/cctz/src/time_zone_impl.cc index f34e3aec..aadbb77d 100644 --- a/absl/time/internal/cctz/src/time_zone_impl.cc +++ b/absl/time/internal/cctz/src/time_zone_impl.cc @@ -99,11 +99,13 @@ void time_zone::Impl::ClearTimeZoneMapTestOnly() { } } +time_zone::Impl::Impl() : name_("UTC"), zone_(TimeZoneIf::UTC()) {} + time_zone::Impl::Impl(const std::string& name) - : name_(name), zone_(TimeZoneIf::Load(name_)) {} + : name_(name), zone_(TimeZoneIf::Make(name_)) {} const time_zone::Impl* time_zone::Impl::UTCImpl() { - static const Impl* utc_impl = new Impl("UTC"); // never fails + static const Impl* utc_impl = new Impl; return utc_impl; } diff --git a/absl/time/internal/cctz/src/time_zone_impl.h b/absl/time/internal/cctz/src/time_zone_impl.h index 7d747ba9..8308a3b4 100644 --- a/absl/time/internal/cctz/src/time_zone_impl.h +++ b/absl/time/internal/cctz/src/time_zone_impl.h @@ -78,7 +78,11 @@ class time_zone::Impl { std::string Description() const { return zone_->Description(); } private: + Impl(); explicit Impl(const std::string& name); + Impl(const Impl&) = delete; + Impl& operator=(const Impl&) = delete; + static const Impl* UTCImpl(); const std::string name_; diff --git a/absl/time/internal/cctz/src/time_zone_info.cc b/absl/time/internal/cctz/src/time_zone_info.cc index 787426f7..4bc7b50b 100644 --- a/absl/time/internal/cctz/src/time_zone_info.cc +++ b/absl/time/internal/cctz/src/time_zone_info.cc @@ -134,6 +134,49 @@ std::int_fast64_t Decode64(const char* cp) { return static_cast(v - s64maxU - 1) - s64max - 1; } +struct Header { // counts of: + std::size_t timecnt; // transition times + std::size_t typecnt; // transition types + std::size_t charcnt; // zone abbreviation characters + std::size_t leapcnt; // leap seconds (we expect none) + std::size_t ttisstdcnt; // UTC/local indicators (unused) + std::size_t ttisutcnt; // standard/wall indicators (unused) + + bool Build(const tzhead& tzh); + std::size_t DataLength(std::size_t time_len) const; +}; + +// Builds the in-memory header using the raw bytes from the file. +bool Header::Build(const tzhead& tzh) { + std::int_fast32_t v; + if ((v = Decode32(tzh.tzh_timecnt)) < 0) return false; + timecnt = static_cast(v); + if ((v = Decode32(tzh.tzh_typecnt)) < 0) return false; + typecnt = static_cast(v); + if ((v = Decode32(tzh.tzh_charcnt)) < 0) return false; + charcnt = static_cast(v); + if ((v = Decode32(tzh.tzh_leapcnt)) < 0) return false; + leapcnt = static_cast(v); + if ((v = Decode32(tzh.tzh_ttisstdcnt)) < 0) return false; + ttisstdcnt = static_cast(v); + if ((v = Decode32(tzh.tzh_ttisutcnt)) < 0) return false; + ttisutcnt = static_cast(v); + return true; +} + +// How many bytes of data are associated with this header. The result +// depends upon whether this is a section with 4-byte or 8-byte times. +std::size_t Header::DataLength(std::size_t time_len) const { + std::size_t len = 0; + len += (time_len + 1) * timecnt; // unix_time + type_index + len += (4 + 1 + 1) * typecnt; // utc_offset + is_dst + abbr_index + len += 1 * charcnt; // abbreviations + len += (time_len + 4) * leapcnt; // leap-time + TAI-UTC + len += 1 * ttisstdcnt; // UTC/local indicators + len += 1 * ttisutcnt; // standard/wall indicators + return len; +} + // Does the rule for future transitions call for year-round daylight time? // See tz/zic.c:stringzone() for the details on how such rules are encoded. bool AllYearDST(const PosixTimeZone& posix) { @@ -217,98 +260,6 @@ inline civil_second YearShift(const civil_second& cs, year_t shift) { } // namespace -// What (no leap-seconds) UTC+seconds zoneinfo would look like. -bool TimeZoneInfo::ResetToBuiltinUTC(const seconds& offset) { - transition_types_.resize(1); - TransitionType& tt(transition_types_.back()); - tt.utc_offset = static_cast(offset.count()); - tt.is_dst = false; - tt.abbr_index = 0; - - // We temporarily add some redundant, contemporary (2015 through 2025) - // transitions for performance reasons. See TimeZoneInfo::LocalTime(). - // TODO: Fix the performance issue and remove the extra transitions. - transitions_.clear(); - transitions_.reserve(12); - for (const std::int_fast64_t unix_time : { - -(1LL << 59), // a "first half" transition - 1420070400LL, // 2015-01-01T00:00:00+00:00 - 1451606400LL, // 2016-01-01T00:00:00+00:00 - 1483228800LL, // 2017-01-01T00:00:00+00:00 - 1514764800LL, // 2018-01-01T00:00:00+00:00 - 1546300800LL, // 2019-01-01T00:00:00+00:00 - 1577836800LL, // 2020-01-01T00:00:00+00:00 - 1609459200LL, // 2021-01-01T00:00:00+00:00 - 1640995200LL, // 2022-01-01T00:00:00+00:00 - 1672531200LL, // 2023-01-01T00:00:00+00:00 - 1704067200LL, // 2024-01-01T00:00:00+00:00 - 1735689600LL, // 2025-01-01T00:00:00+00:00 - }) { - Transition& tr(*transitions_.emplace(transitions_.end())); - tr.unix_time = unix_time; - tr.type_index = 0; - tr.civil_sec = LocalTime(tr.unix_time, tt).cs; - tr.prev_civil_sec = tr.civil_sec - 1; - } - - default_transition_type_ = 0; - abbreviations_ = FixedOffsetToAbbr(offset); - abbreviations_.append(1, '\0'); - future_spec_.clear(); // never needed for a fixed-offset zone - extended_ = false; - - tt.civil_max = LocalTime(seconds::max().count(), tt).cs; - tt.civil_min = LocalTime(seconds::min().count(), tt).cs; - - transitions_.shrink_to_fit(); - return true; -} - -// Builds the in-memory header using the raw bytes from the file. -bool TimeZoneInfo::Header::Build(const tzhead& tzh) { - std::int_fast32_t v; - if ((v = Decode32(tzh.tzh_timecnt)) < 0) return false; - timecnt = static_cast(v); - if ((v = Decode32(tzh.tzh_typecnt)) < 0) return false; - typecnt = static_cast(v); - if ((v = Decode32(tzh.tzh_charcnt)) < 0) return false; - charcnt = static_cast(v); - if ((v = Decode32(tzh.tzh_leapcnt)) < 0) return false; - leapcnt = static_cast(v); - if ((v = Decode32(tzh.tzh_ttisstdcnt)) < 0) return false; - ttisstdcnt = static_cast(v); - if ((v = Decode32(tzh.tzh_ttisutcnt)) < 0) return false; - ttisutcnt = static_cast(v); - return true; -} - -// How many bytes of data are associated with this header. The result -// depends upon whether this is a section with 4-byte or 8-byte times. -std::size_t TimeZoneInfo::Header::DataLength(std::size_t time_len) const { - std::size_t len = 0; - len += (time_len + 1) * timecnt; // unix_time + type_index - len += (4 + 1 + 1) * typecnt; // utc_offset + is_dst + abbr_index - len += 1 * charcnt; // abbreviations - len += (time_len + 4) * leapcnt; // leap-time + TAI-UTC - len += 1 * ttisstdcnt; // UTC/local indicators - len += 1 * ttisutcnt; // standard/wall indicators - return len; -} - -// zic(8) can generate no-op transitions when a zone changes rules at an -// instant when there is actually no discontinuity. So we check whether -// two transitions have equivalent types (same offset/is_dst/abbr). -bool TimeZoneInfo::EquivTransitions(std::uint_fast8_t tt1_index, - std::uint_fast8_t tt2_index) const { - if (tt1_index == tt2_index) return true; - const TransitionType& tt1(transition_types_[tt1_index]); - const TransitionType& tt2(transition_types_[tt2_index]); - if (tt1.utc_offset != tt2.utc_offset) return false; - if (tt1.is_dst != tt2.is_dst) return false; - if (tt1.abbr_index != tt2.abbr_index) return false; - return true; -} - // Find/make a transition type with these attributes. bool TimeZoneInfo::GetTransitionType(std::int_fast32_t utc_offset, bool is_dst, const std::string& abbr, @@ -341,6 +292,20 @@ bool TimeZoneInfo::GetTransitionType(std::int_fast32_t utc_offset, bool is_dst, return true; } +// zic(8) can generate no-op transitions when a zone changes rules at an +// instant when there is actually no discontinuity. So we check whether +// two transitions have equivalent types (same offset/is_dst/abbr). +bool TimeZoneInfo::EquivTransitions(std::uint_fast8_t tt1_index, + std::uint_fast8_t tt2_index) const { + if (tt1_index == tt2_index) return true; + const TransitionType& tt1(transition_types_[tt1_index]); + const TransitionType& tt2(transition_types_[tt2_index]); + if (tt1.utc_offset != tt2.utc_offset) return false; + if (tt1.is_dst != tt2.is_dst) return false; + if (tt1.abbr_index != tt2.abbr_index) return false; + return true; +} + // Use the POSIX-TZ-environment-variable-style string to handle times // in years after the last transition stored in the zoneinfo data. bool TimeZoneInfo::ExtendTransitions() { @@ -394,206 +359,19 @@ bool TimeZoneInfo::ExtendTransitions() { auto dst_trans_off = TransOffset(leap_year, jan1_weekday, posix.dst_start); auto std_trans_off = TransOffset(leap_year, jan1_weekday, posix.dst_end); dst.unix_time = jan1_time + dst_trans_off - posix.std_offset; - std.unix_time = jan1_time + std_trans_off - posix.dst_offset; - const auto* ta = dst.unix_time < std.unix_time ? &dst : &std; - const auto* tb = dst.unix_time < std.unix_time ? &std : &dst; - if (last_time < tb->unix_time) { - if (last_time < ta->unix_time) transitions_.push_back(*ta); - transitions_.push_back(*tb); - } - if (last_year_ == limit) break; - jan1_time += kSecsPerYear[leap_year]; - jan1_weekday = (jan1_weekday + kDaysPerYear[leap_year]) % 7; - leap_year = !leap_year && IsLeap(last_year_ + 1); - } - - return true; -} - -bool TimeZoneInfo::Load(ZoneInfoSource* zip) { - // Read and validate the header. - tzhead tzh; - if (zip->Read(&tzh, sizeof(tzh)) != sizeof(tzh)) return false; - if (strncmp(tzh.tzh_magic, TZ_MAGIC, sizeof(tzh.tzh_magic)) != 0) - return false; - Header hdr; - if (!hdr.Build(tzh)) return false; - std::size_t time_len = 4; - if (tzh.tzh_version[0] != '\0') { - // Skip the 4-byte data. - if (zip->Skip(hdr.DataLength(time_len)) != 0) return false; - // Read and validate the header for the 8-byte data. - if (zip->Read(&tzh, sizeof(tzh)) != sizeof(tzh)) return false; - if (strncmp(tzh.tzh_magic, TZ_MAGIC, sizeof(tzh.tzh_magic)) != 0) - return false; - if (tzh.tzh_version[0] == '\0') return false; - if (!hdr.Build(tzh)) return false; - time_len = 8; - } - if (hdr.typecnt == 0) return false; - if (hdr.leapcnt != 0) { - // This code assumes 60-second minutes so we do not want - // the leap-second encoded zoneinfo. We could reverse the - // compensation, but the "right" encoding is rarely used - // so currently we simply reject such data. - return false; - } - if (hdr.ttisstdcnt != 0 && hdr.ttisstdcnt != hdr.typecnt) return false; - if (hdr.ttisutcnt != 0 && hdr.ttisutcnt != hdr.typecnt) return false; - - // Read the data into a local buffer. - std::size_t len = hdr.DataLength(time_len); - std::vector tbuf(len); - if (zip->Read(tbuf.data(), len) != len) return false; - const char* bp = tbuf.data(); - - // Decode and validate the transitions. - transitions_.reserve(hdr.timecnt + 2); - transitions_.resize(hdr.timecnt); - for (std::size_t i = 0; i != hdr.timecnt; ++i) { - transitions_[i].unix_time = (time_len == 4) ? Decode32(bp) : Decode64(bp); - bp += time_len; - if (i != 0) { - // Check that the transitions are ordered by time (as zic guarantees). - if (!Transition::ByUnixTime()(transitions_[i - 1], transitions_[i])) - return false; // out of order - } - } - bool seen_type_0 = false; - for (std::size_t i = 0; i != hdr.timecnt; ++i) { - transitions_[i].type_index = Decode8(bp++); - if (transitions_[i].type_index >= hdr.typecnt) return false; - if (transitions_[i].type_index == 0) seen_type_0 = true; - } - - // Decode and validate the transition types. - transition_types_.reserve(hdr.typecnt + 2); - transition_types_.resize(hdr.typecnt); - for (std::size_t i = 0; i != hdr.typecnt; ++i) { - transition_types_[i].utc_offset = - static_cast(Decode32(bp)); - if (transition_types_[i].utc_offset >= kSecsPerDay || - transition_types_[i].utc_offset <= -kSecsPerDay) - return false; - bp += 4; - transition_types_[i].is_dst = (Decode8(bp++) != 0); - transition_types_[i].abbr_index = Decode8(bp++); - if (transition_types_[i].abbr_index >= hdr.charcnt) return false; - } - - // Determine the before-first-transition type. - default_transition_type_ = 0; - if (seen_type_0 && hdr.timecnt != 0) { - std::uint_fast8_t index = 0; - if (transition_types_[0].is_dst) { - index = transitions_[0].type_index; - while (index != 0 && transition_types_[index].is_dst) --index; - } - while (index != hdr.typecnt && transition_types_[index].is_dst) ++index; - if (index != hdr.typecnt) default_transition_type_ = index; - } - - // Copy all the abbreviations. - abbreviations_.reserve(hdr.charcnt + 10); - abbreviations_.assign(bp, hdr.charcnt); - bp += hdr.charcnt; - - // Skip the unused portions. We've already dispensed with leap-second - // encoded zoneinfo. The ttisstd/ttisgmt indicators only apply when - // interpreting a POSIX spec that does not include start/end rules, and - // that isn't the case here (see "zic -p"). - bp += (time_len + 4) * hdr.leapcnt; // leap-time + TAI-UTC - bp += 1 * hdr.ttisstdcnt; // UTC/local indicators - bp += 1 * hdr.ttisutcnt; // standard/wall indicators - assert(bp == tbuf.data() + tbuf.size()); - - future_spec_.clear(); - if (tzh.tzh_version[0] != '\0') { - // Snarf up the NL-enclosed future POSIX spec. Note - // that version '3' files utilize an extended format. - auto get_char = [](ZoneInfoSource* azip) -> int { - unsigned char ch; // all non-EOF results are positive - return (azip->Read(&ch, 1) == 1) ? ch : EOF; - }; - if (get_char(zip) != '\n') return false; - for (int c = get_char(zip); c != '\n'; c = get_char(zip)) { - if (c == EOF) return false; - future_spec_.push_back(static_cast(c)); - } - } - - // We don't check for EOF so that we're forwards compatible. - - // If we did not find version information during the standard loading - // process (as of tzh_version '3' that is unsupported), then ask the - // ZoneInfoSource for any out-of-bound version string it may be privy to. - if (version_.empty()) { - version_ = zip->Version(); - } - - // Trim redundant transitions. zic may have added these to work around - // differences between the glibc and reference implementations (see - // zic.c:dontmerge) or to avoid bugs in old readers. For us, they just - // get in the way when we do future_spec_ extension. - while (hdr.timecnt > 1) { - if (!EquivTransitions(transitions_[hdr.timecnt - 1].type_index, - transitions_[hdr.timecnt - 2].type_index)) { - break; - } - hdr.timecnt -= 1; - } - transitions_.resize(hdr.timecnt); - - // Ensure that there is always a transition in the first half of the - // time line (the second half is handled below) so that the signed - // difference between a civil_second and the civil_second of its - // previous transition is always representable, without overflow. - if (transitions_.empty() || transitions_.front().unix_time >= 0) { - Transition& tr(*transitions_.emplace(transitions_.begin())); - tr.unix_time = -(1LL << 59); // -18267312070-10-26T17:01:52+00:00 - tr.type_index = default_transition_type_; - } - - // Extend the transitions using the future specification. - if (!ExtendTransitions()) return false; - - // Ensure that there is always a transition in the second half of the - // time line (the first half is handled above) so that the signed - // difference between a civil_second and the civil_second of its - // previous transition is always representable, without overflow. - const Transition& last(transitions_.back()); - if (last.unix_time < 0) { - const std::uint_fast8_t type_index = last.type_index; - Transition& tr(*transitions_.emplace(transitions_.end())); - tr.unix_time = 2147483647; // 2038-01-19T03:14:07+00:00 - tr.type_index = type_index; - } - - // Compute the local civil time for each transition and the preceding - // second. These will be used for reverse conversions in MakeTime(). - const TransitionType* ttp = &transition_types_[default_transition_type_]; - for (std::size_t i = 0; i != transitions_.size(); ++i) { - Transition& tr(transitions_[i]); - tr.prev_civil_sec = LocalTime(tr.unix_time, *ttp).cs - 1; - ttp = &transition_types_[tr.type_index]; - tr.civil_sec = LocalTime(tr.unix_time, *ttp).cs; - if (i != 0) { - // Check that the transitions are ordered by civil time. Essentially - // this means that an offset change cannot cross another such change. - // No one does this in practice, and we depend on it in MakeTime(). - if (!Transition::ByCivilTime()(transitions_[i - 1], tr)) - return false; // out of order + std.unix_time = jan1_time + std_trans_off - posix.dst_offset; + const auto* ta = dst.unix_time < std.unix_time ? &dst : &std; + const auto* tb = dst.unix_time < std.unix_time ? &std : &dst; + if (last_time < tb->unix_time) { + if (last_time < ta->unix_time) transitions_.push_back(*ta); + transitions_.push_back(*tb); } + if (last_year_ == limit) break; + jan1_time += kSecsPerYear[leap_year]; + jan1_weekday = (jan1_weekday + kDaysPerYear[leap_year]) % 7; + leap_year = !leap_year && IsLeap(last_year_ + 1); } - // Compute the maximum/minimum civil times that can be converted to a - // time_point for each of the zone's transition types. - for (auto& tt : transition_types_) { - tt.civil_max = LocalTime(seconds::max().count(), tt).cs; - tt.civil_min = LocalTime(seconds::min().count(), tt).cs; - } - - transitions_.shrink_to_fit(); return true; } @@ -795,6 +573,240 @@ std::unique_ptr FuchsiaZoneInfoSource::Open( } // namespace +// What (no leap-seconds) UTC+seconds zoneinfo would look like. +bool TimeZoneInfo::ResetToBuiltinUTC(const seconds& offset) { + transition_types_.resize(1); + TransitionType& tt(transition_types_.back()); + tt.utc_offset = static_cast(offset.count()); + tt.is_dst = false; + tt.abbr_index = 0; + + // We temporarily add some redundant, contemporary (2015 through 2025) + // transitions for performance reasons. See TimeZoneInfo::LocalTime(). + // TODO: Fix the performance issue and remove the extra transitions. + transitions_.clear(); + transitions_.reserve(12); + for (const std::int_fast64_t unix_time : { + -(1LL << 59), // a "first half" transition + 1420070400LL, // 2015-01-01T00:00:00+00:00 + 1451606400LL, // 2016-01-01T00:00:00+00:00 + 1483228800LL, // 2017-01-01T00:00:00+00:00 + 1514764800LL, // 2018-01-01T00:00:00+00:00 + 1546300800LL, // 2019-01-01T00:00:00+00:00 + 1577836800LL, // 2020-01-01T00:00:00+00:00 + 1609459200LL, // 2021-01-01T00:00:00+00:00 + 1640995200LL, // 2022-01-01T00:00:00+00:00 + 1672531200LL, // 2023-01-01T00:00:00+00:00 + 1704067200LL, // 2024-01-01T00:00:00+00:00 + 1735689600LL, // 2025-01-01T00:00:00+00:00 + }) { + Transition& tr(*transitions_.emplace(transitions_.end())); + tr.unix_time = unix_time; + tr.type_index = 0; + tr.civil_sec = LocalTime(tr.unix_time, tt).cs; + tr.prev_civil_sec = tr.civil_sec - 1; + } + + default_transition_type_ = 0; + abbreviations_ = FixedOffsetToAbbr(offset); + abbreviations_.append(1, '\0'); + future_spec_.clear(); // never needed for a fixed-offset zone + extended_ = false; + + tt.civil_max = LocalTime(seconds::max().count(), tt).cs; + tt.civil_min = LocalTime(seconds::min().count(), tt).cs; + + transitions_.shrink_to_fit(); + return true; +} + +bool TimeZoneInfo::Load(ZoneInfoSource* zip) { + // Read and validate the header. + tzhead tzh; + if (zip->Read(&tzh, sizeof(tzh)) != sizeof(tzh)) return false; + if (strncmp(tzh.tzh_magic, TZ_MAGIC, sizeof(tzh.tzh_magic)) != 0) + return false; + Header hdr; + if (!hdr.Build(tzh)) return false; + std::size_t time_len = 4; + if (tzh.tzh_version[0] != '\0') { + // Skip the 4-byte data. + if (zip->Skip(hdr.DataLength(time_len)) != 0) return false; + // Read and validate the header for the 8-byte data. + if (zip->Read(&tzh, sizeof(tzh)) != sizeof(tzh)) return false; + if (strncmp(tzh.tzh_magic, TZ_MAGIC, sizeof(tzh.tzh_magic)) != 0) + return false; + if (tzh.tzh_version[0] == '\0') return false; + if (!hdr.Build(tzh)) return false; + time_len = 8; + } + if (hdr.typecnt == 0) return false; + if (hdr.leapcnt != 0) { + // This code assumes 60-second minutes so we do not want + // the leap-second encoded zoneinfo. We could reverse the + // compensation, but the "right" encoding is rarely used + // so currently we simply reject such data. + return false; + } + if (hdr.ttisstdcnt != 0 && hdr.ttisstdcnt != hdr.typecnt) return false; + if (hdr.ttisutcnt != 0 && hdr.ttisutcnt != hdr.typecnt) return false; + + // Read the data into a local buffer. + std::size_t len = hdr.DataLength(time_len); + std::vector tbuf(len); + if (zip->Read(tbuf.data(), len) != len) return false; + const char* bp = tbuf.data(); + + // Decode and validate the transitions. + transitions_.reserve(hdr.timecnt + 2); + transitions_.resize(hdr.timecnt); + for (std::size_t i = 0; i != hdr.timecnt; ++i) { + transitions_[i].unix_time = (time_len == 4) ? Decode32(bp) : Decode64(bp); + bp += time_len; + if (i != 0) { + // Check that the transitions are ordered by time (as zic guarantees). + if (!Transition::ByUnixTime()(transitions_[i - 1], transitions_[i])) + return false; // out of order + } + } + bool seen_type_0 = false; + for (std::size_t i = 0; i != hdr.timecnt; ++i) { + transitions_[i].type_index = Decode8(bp++); + if (transitions_[i].type_index >= hdr.typecnt) return false; + if (transitions_[i].type_index == 0) seen_type_0 = true; + } + + // Decode and validate the transition types. + transition_types_.reserve(hdr.typecnt + 2); + transition_types_.resize(hdr.typecnt); + for (std::size_t i = 0; i != hdr.typecnt; ++i) { + transition_types_[i].utc_offset = + static_cast(Decode32(bp)); + if (transition_types_[i].utc_offset >= kSecsPerDay || + transition_types_[i].utc_offset <= -kSecsPerDay) + return false; + bp += 4; + transition_types_[i].is_dst = (Decode8(bp++) != 0); + transition_types_[i].abbr_index = Decode8(bp++); + if (transition_types_[i].abbr_index >= hdr.charcnt) return false; + } + + // Determine the before-first-transition type. + default_transition_type_ = 0; + if (seen_type_0 && hdr.timecnt != 0) { + std::uint_fast8_t index = 0; + if (transition_types_[0].is_dst) { + index = transitions_[0].type_index; + while (index != 0 && transition_types_[index].is_dst) --index; + } + while (index != hdr.typecnt && transition_types_[index].is_dst) ++index; + if (index != hdr.typecnt) default_transition_type_ = index; + } + + // Copy all the abbreviations. + abbreviations_.reserve(hdr.charcnt + 10); + abbreviations_.assign(bp, hdr.charcnt); + bp += hdr.charcnt; + + // Skip the unused portions. We've already dispensed with leap-second + // encoded zoneinfo. The ttisstd/ttisgmt indicators only apply when + // interpreting a POSIX spec that does not include start/end rules, and + // that isn't the case here (see "zic -p"). + bp += (time_len + 4) * hdr.leapcnt; // leap-time + TAI-UTC + bp += 1 * hdr.ttisstdcnt; // UTC/local indicators + bp += 1 * hdr.ttisutcnt; // standard/wall indicators + assert(bp == tbuf.data() + tbuf.size()); + + future_spec_.clear(); + if (tzh.tzh_version[0] != '\0') { + // Snarf up the NL-enclosed future POSIX spec. Note + // that version '3' files utilize an extended format. + auto get_char = [](ZoneInfoSource* azip) -> int { + unsigned char ch; // all non-EOF results are positive + return (azip->Read(&ch, 1) == 1) ? ch : EOF; + }; + if (get_char(zip) != '\n') return false; + for (int c = get_char(zip); c != '\n'; c = get_char(zip)) { + if (c == EOF) return false; + future_spec_.push_back(static_cast(c)); + } + } + + // We don't check for EOF so that we're forwards compatible. + + // If we did not find version information during the standard loading + // process (as of tzh_version '3' that is unsupported), then ask the + // ZoneInfoSource for any out-of-bound version string it may be privy to. + if (version_.empty()) { + version_ = zip->Version(); + } + + // Trim redundant transitions. zic may have added these to work around + // differences between the glibc and reference implementations (see + // zic.c:dontmerge) or to avoid bugs in old readers. For us, they just + // get in the way when we do future_spec_ extension. + while (hdr.timecnt > 1) { + if (!EquivTransitions(transitions_[hdr.timecnt - 1].type_index, + transitions_[hdr.timecnt - 2].type_index)) { + break; + } + hdr.timecnt -= 1; + } + transitions_.resize(hdr.timecnt); + + // Ensure that there is always a transition in the first half of the + // time line (the second half is handled below) so that the signed + // difference between a civil_second and the civil_second of its + // previous transition is always representable, without overflow. + if (transitions_.empty() || transitions_.front().unix_time >= 0) { + Transition& tr(*transitions_.emplace(transitions_.begin())); + tr.unix_time = -(1LL << 59); // -18267312070-10-26T17:01:52+00:00 + tr.type_index = default_transition_type_; + } + + // Extend the transitions using the future specification. + if (!ExtendTransitions()) return false; + + // Ensure that there is always a transition in the second half of the + // time line (the first half is handled above) so that the signed + // difference between a civil_second and the civil_second of its + // previous transition is always representable, without overflow. + const Transition& last(transitions_.back()); + if (last.unix_time < 0) { + const std::uint_fast8_t type_index = last.type_index; + Transition& tr(*transitions_.emplace(transitions_.end())); + tr.unix_time = 2147483647; // 2038-01-19T03:14:07+00:00 + tr.type_index = type_index; + } + + // Compute the local civil time for each transition and the preceding + // second. These will be used for reverse conversions in MakeTime(). + const TransitionType* ttp = &transition_types_[default_transition_type_]; + for (std::size_t i = 0; i != transitions_.size(); ++i) { + Transition& tr(transitions_[i]); + tr.prev_civil_sec = LocalTime(tr.unix_time, *ttp).cs - 1; + ttp = &transition_types_[tr.type_index]; + tr.civil_sec = LocalTime(tr.unix_time, *ttp).cs; + if (i != 0) { + // Check that the transitions are ordered by civil time. Essentially + // this means that an offset change cannot cross another such change. + // No one does this in practice, and we depend on it in MakeTime(). + if (!Transition::ByCivilTime()(transitions_[i - 1], tr)) + return false; // out of order + } + } + + // Compute the maximum/minimum civil times that can be converted to a + // time_point for each of the zone's transition types. + for (auto& tt : transition_types_) { + tt.civil_max = LocalTime(seconds::max().count(), tt).cs; + tt.civil_min = LocalTime(seconds::min().count(), tt).cs; + } + + transitions_.shrink_to_fit(); + return true; +} + bool TimeZoneInfo::Load(const std::string& name) { // We can ensure that the loading of UTC or any other fixed-offset // zone never fails because the simple, fixed-offset state can be @@ -816,6 +828,18 @@ bool TimeZoneInfo::Load(const std::string& name) { return zip != nullptr && Load(zip.get()); } +std::unique_ptr TimeZoneInfo::UTC() { + auto tz = std::unique_ptr(new TimeZoneInfo); + tz->ResetToBuiltinUTC(seconds::zero()); + return tz; +} + +std::unique_ptr TimeZoneInfo::Make(const std::string& name) { + auto tz = std::unique_ptr(new TimeZoneInfo); + if (!tz->Load(name)) tz.reset(); // fallback to UTC + return tz; +} + // BreakTime() translation for a particular transition type. time_zone::absolute_lookup TimeZoneInfo::LocalTime( std::int_fast64_t unix_time, const TransitionType& tt) const { diff --git a/absl/time/internal/cctz/src/time_zone_info.h b/absl/time/internal/cctz/src/time_zone_info.h index 2467ff55..689df6f9 100644 --- a/absl/time/internal/cctz/src/time_zone_info.h +++ b/absl/time/internal/cctz/src/time_zone_info.h @@ -18,6 +18,7 @@ #include #include #include +#include #include #include @@ -64,12 +65,9 @@ struct TransitionType { // A time zone backed by the IANA Time Zone Database (zoneinfo). class TimeZoneInfo : public TimeZoneIf { public: - TimeZoneInfo() = default; - TimeZoneInfo(const TimeZoneInfo&) = delete; - TimeZoneInfo& operator=(const TimeZoneInfo&) = delete; - - // Loads the zoneinfo for the given name, returning true if successful. - bool Load(const std::string& name); + // Factories. + static std::unique_ptr UTC(); // never fails + static std::unique_ptr Make(const std::string& name); // TimeZoneIf implementations. time_zone::absolute_lookup BreakTime( @@ -83,17 +81,9 @@ class TimeZoneInfo : public TimeZoneIf { std::string Description() const override; private: - struct Header { // counts of: - std::size_t timecnt; // transition times - std::size_t typecnt; // transition types - std::size_t charcnt; // zone abbreviation characters - std::size_t leapcnt; // leap seconds (we expect none) - std::size_t ttisstdcnt; // UTC/local indicators (unused) - std::size_t ttisutcnt; // standard/wall indicators (unused) - - bool Build(const tzhead& tzh); - std::size_t DataLength(std::size_t time_len) const; - }; + TimeZoneInfo() = default; + TimeZoneInfo(const TimeZoneInfo&) = delete; + TimeZoneInfo& operator=(const TimeZoneInfo&) = delete; bool GetTransitionType(std::int_fast32_t utc_offset, bool is_dst, const std::string& abbr, std::uint_least8_t* index); @@ -102,6 +92,7 @@ class TimeZoneInfo : public TimeZoneIf { bool ExtendTransitions(); bool ResetToBuiltinUTC(const seconds& offset); + bool Load(const std::string& name); bool Load(ZoneInfoSource* zip); // Helpers for BreakTime() and MakeTime(). diff --git a/absl/time/internal/cctz/src/time_zone_libc.cc b/absl/time/internal/cctz/src/time_zone_libc.cc index e503a858..d0146122 100644 --- a/absl/time/internal/cctz/src/time_zone_libc.cc +++ b/absl/time/internal/cctz/src/time_zone_libc.cc @@ -62,7 +62,7 @@ auto tm_zone(const std::tm& tm) -> decltype(tzname[0]) { } #elif defined(__native_client__) || defined(__myriad2__) || \ defined(__EMSCRIPTEN__) -// Uses the globals: 'timezone' and 'tzname'. +// Uses the globals: '_timezone' and 'tzname'. auto tm_gmtoff(const std::tm& tm) -> decltype(_timezone + 0) { const bool is_dst = tm.tm_isdst > 0; return _timezone + (is_dst ? 60 * 60 : 0); @@ -193,8 +193,9 @@ std::time_t find_trans(std::time_t lo, std::time_t hi, tm_gmtoff_t offset) { } // namespace -TimeZoneLibC::TimeZoneLibC(const std::string& name) - : local_(name == "localtime") {} +std::unique_ptr TimeZoneLibC::Make(const std::string& name) { + return std::unique_ptr(new TimeZoneLibC(name)); +} time_zone::absolute_lookup TimeZoneLibC::BreakTime( const time_point& tp) const { @@ -323,6 +324,9 @@ std::string TimeZoneLibC::Description() const { return local_ ? "localtime" : "UTC"; } +TimeZoneLibC::TimeZoneLibC(const std::string& name) + : local_(name == "localtime") {} + } // namespace cctz } // namespace time_internal ABSL_NAMESPACE_END diff --git a/absl/time/internal/cctz/src/time_zone_libc.h b/absl/time/internal/cctz/src/time_zone_libc.h index 1da9039a..2d6fa686 100644 --- a/absl/time/internal/cctz/src/time_zone_libc.h +++ b/absl/time/internal/cctz/src/time_zone_libc.h @@ -27,10 +27,10 @@ namespace cctz { // A time zone backed by gmtime_r(3), localtime_r(3), and mktime(3), // and which therefore only supports UTC and the local time zone. -// TODO: Add support for fixed offsets from UTC. class TimeZoneLibC : public TimeZoneIf { public: - explicit TimeZoneLibC(const std::string& name); + // Factory. + static std::unique_ptr Make(const std::string& name); // TimeZoneIf implementations. time_zone::absolute_lookup BreakTime( @@ -44,6 +44,10 @@ class TimeZoneLibC : public TimeZoneIf { std::string Description() const override; private: + explicit TimeZoneLibC(const std::string& name); + TimeZoneLibC(const TimeZoneLibC&) = delete; + TimeZoneLibC& operator=(const TimeZoneLibC&) = delete; + const bool local_; // localtime or UTC }; diff --git a/absl/time/internal/cctz/src/time_zone_lookup_test.cc b/absl/time/internal/cctz/src/time_zone_lookup_test.cc index 38f10f48..fc659266 100644 --- a/absl/time/internal/cctz/src/time_zone_lookup_test.cc +++ b/absl/time/internal/cctz/src/time_zone_lookup_test.cc @@ -734,6 +734,10 @@ TEST(TimeZone, UTC) { time_zone loaded_utc0; EXPECT_TRUE(load_time_zone("UTC0", &loaded_utc0)); EXPECT_EQ(loaded_utc0, utc); + + time_zone loaded_bad; + EXPECT_FALSE(load_time_zone("Invalid/TimeZone", &loaded_bad)); + EXPECT_EQ(loaded_bad, utc); } TEST(TimeZone, NamedTimeZones) { -- cgit v1.2.3