<chapter xmlns="http://docbook.org/ns/docbook" version="5.0"
	 xml:id="std.time" xreflabel="Time">
<?dbhtml filename="time.html"?>

<info><title>
  Time
  <indexterm><primary>Time</primary></indexterm>
</title>
  <keywordset>
    <keyword>ISO C++</keyword>
    <keyword>library</keyword>
  </keywordset>
</info>

<!-- Sect1 01 : Time zone database -->
<section xml:id="std.time.zone.db" xreflabel=""><info><title>Time zone database</title></info>
<?dbhtml filename="tzdb.html"?>

   <para>
     Since C++20 the <code>&lt;chrono&gt;</code> header provides time zone
     support via <code>std::chrono::tzdb</code>, a complete interface to the
     <link xmlns:xlink="http://www.w3.org/1999/xlink"
     xlink:href="https://www.iana.org/time-zones">IANA Time Zone Database</link>.
     The library supports converting times between UTC and the local time
     in a specific time zone, querying UTC offsets and DST information for a
     given time zone on a given date, and finding the system's current time zone.
     Key entry points are <code>std::chrono::get_tzdb()</code> to access
     the database, <code>std::chrono::locate_zone()</code> to look up
     a zone by name, and <code>std::chrono::current_zone()</code> to obtain
     the system's local time zone. The <code>std::chrono::zoned_time</code>
     class template represents a time point in a given time zone.
     The database can be reloaded at runtime
     via <code>std::chrono::reload_tzdb()</code> to pick up updated zone
     data without restarting the program.
   </para>

   <para>
     The <link xmlns:xlink="http://www.w3.org/1999/xlink"
     xlink:href="https://howardhinnant.github.io/date/tz.html">Time Zone Database
     Parser</link> by Howard E. Hinnant is a valuable source of information and
     examples of usage for the <code>date</code> library, on which the standard
     API is based.
   </para>

   <section xml:id="zone.db.zic" xreflabel="tzdb uses zic format"><info><title>tzdb uses zic format</title></info>

   <para>
     When support is enabled (which is default) the time zone database class,
     <code>tzdb</code>, uses the <link xmlns:xlink="http://www.w3.org/1999/xlink"
     xlink:href="https://man7.org/linux/man-pages/man8/zic.8.html#FILES">
     <code>zic</code> file format</link>, both as the format of the input
     files, and for in-memory representation. This results in a reduced
     memory footprint for many programs, as time zone data is stored in
     a compact form until needed by the application.
   </para>
   <para>
     The transitions to/from DST (Daylight Savings Time) are generated and
     cached on the fly, when information
     for a given zone is requested. Consequently, the initial request for
     UTC conversion for a given <code>time_zone</code> object is slower.
     To mitigate this, a <code>time_zone</code> can be expanded for a
     particular time range at the start
     of the program (and after each reload), by iterating over the
     <code>sys_info</code> sequence, as in the example below.
   </para>
   <programlisting>
     void
     expand_zone(const std::chrono::time_zone* tz,
                 const std::chrono::sys_seconds start,
                 const std::chrono::sys_seconds end)
     {
       std::chrono::sys_seconds time = start;
       while (time &lt; end) {
         const std::chrono::sys_info&amp; info = tz->get_info(time);
         time = info.end + std::chrono::seconds(1);
       }
     }
   </programlisting>
  </section>

  <section xml:id="zone.current" xreflabel="current zone"><info><title>current zone</title></info>

   <para>
     The <code>time_zone</code> pointer returned from
     <code>std::chrono::current_zone</code>, is determined by looking up
     (by <code>std::chrono::locate_zone</code>) IANA zone name
     determined as follows (for platforms other than AIX and Windows):
   </para>
   <itemizedlist>
     <listitem>
       On systems supporting <code>readlink</code>, suffix components of the
       path of the file linked by <code>/etc/localtime</code>. The suffixes
       are matched in the order of increasing length, starting from the final
       component, until the match is found in the time zone database.
       Any repeated slashes (<code>//</code>) are eliminated.
     </listitem>
     <listitem>
       Names stored in the files <code>/etc/timezone</code> (Debian
       derivatives) or <code>/var/db/zoneinfo</code> (FreeBSD) in that order.
     </listitem>
     <listitem>
       Values for <code>TIMEZONE</code> and <code>ZONE</code> keys in
       <code>/etc/sysconfig/clock</code>.
     </listitem>
     <listitem>
       <code>UTC</code>.
     </listitem>
   </itemizedlist>

   <para>
     For AIX, the value of <code>TZ</code> environment value is used, with
     fallback to <code>UTC</code>.
   </para>

   <para>
     On Windows <code>TimeZoneKeyName</code> is mapped to the IANA zone, using
     internal mapping hard-coded in library, with fallback to <code>UTC</code>:
   </para>
   <itemizedlist>
     <listitem>
       If the returned <code>TimeZoneKeyName</code> is empty or DST is disabled,
       then a value based on <code>Bias</code> is returned: <code>Etc/UTC</code>
       for zero, <code>Etc/GMT+/-N</code> for value that is multiply of 60,
       and empty string otherwise.
     </listitem>
     <listitem>
       If the mapping contains a single entry for <code>TimeZoneKeyName</code>,
       that name is used.
     </listitem>
     <listitem>
       If <code>TimeZoneKeyName</code> corresponds to multiple IANA zones,
       a 2-letter country code is used, retrieved by applying <code>GetGeoInfoW</code>
       on the result of <code>GetUserGeoID(GEOCLASS_NATION)</code>.
       If determining the country code fails, or no entry exists for a given code,
       the first mapping entry (<code>001</code>) is returned.
     </listitem>
   </itemizedlist>
  </section>

  <section xml:id="zone.db.sources" xreflabel="data sources"><info><title>data sources</title></info>
    <para>
      Depending on the build configuration (see
      <code>--with-libstdcxx-zoneinfo=</code> documentation in
      <xref linkend="manual.intro.setup.configure">Configure section</xref>).
      the content of the time zone database is sourced from
      <filename>tzdata.zi</filename> and <filename>leapseconds</filename> files
      located in <emphasis>zoneinfo_dir</emphasis>, or from static information
      embedded in the library.
    </para>

    <para>
      By default, <emphasis>zoneinfo_dir</emphasis> is set to the
      system-specific default directory (if a suitable dir is known for target),
      usually <code>/usr/share/zoneinfo</code>. If no such directory
      exists, or it does not contain the required files in the correct format,
      embedded static data is used as fallback.
    </para>

    <para>
      In addition to <code>--with-libstdcxx-zoneinfo=</code> configure option
      (used during GCC build), the path of <emphasis>zoneinfo_dir</emphasis>
      can be overridden by the application by providing a definition of the
      <code>__gnu_cxx::zoneinfo_dir_override()</code> function. The
      returned path should be a directory that contains <code>tzdata.zi</code>
      and <code>leapseconds</code> files in the
      <link xmlns:xlink="http://www.w3.org/1999/xlink"
      xlink:href="https://man7.org/linux/man-pages/man8/zic.8.html#FILES">
      <code>zic</code> format</link>.
    </para>

    <para>
      The embedded timezone information corresponds to the copy of the
      IANA database at the time of the release, and its version can be queried
      using <code>std::chrono::get_tzdb().version</code>.
    </para>

    <para>
      The full time zone database can be disabled when GCC is configured,
      in which case a minimal time zone database is provided. This minimal
      database can be identified by <code>chrono::get_tzdb().version</code>
      being set to <code>"ersatz"</code>. This database contains leapseconds
      data, <code>Etc/UTC</code>, <code>Etc/GMT</code> zones, and their aliases
      (<code>Etc/Zulu</code>, <code>Etc/UCT</code>, <code>Etc/Universal</code>,
      <code>Etc/Greenwich</code>, <code>Etc/GMT0</code>, <code>Etc/GMT+0</code>,
      <code>Etc/GMT-0</code>). This information is sufficient to support
      conversion between <code>utc_clock</code> and <code>sys_clock</code>,
      as well as a <code>UTC</code> fallback for <code>current_zone</code>
      (<xref linkend="zone.current">link</xref>).
    </para>
  </section>

  <section xml:id="zone.db.iterator" xreflabel="tzdb_list::const_iterator extends lifetime">
    <info><title>tzdb_list::const_iterator extends lifetime</title></info>

    <para>
      The <code>std::chrono::reload_tzdb()</code> function may be used to load
      updated content of the <code>tzdata.zi</code> and <code>leapseconds</code>
      files from <emphasis>zoneinfo_dir</emphasis> (if
      <xref linkend="zone.db.sources">enabled</xref>). If the version is different
      (<code>std::chrono::remote_version() != std::chrono::get_tzdb().version</code>),
      a new element is added at the front of the <code>tzdb_list</code>.
      This new database is used for subsequent calls to <code>current_zone</code>
      and <code>locate_zone</code>.
    </para>

    <para>
      The above process is thread-safe, and does not invalidate nor change any
      pre-existing pointers to <code>time_zone</code> objects. However, for a
      long-running application it may lead to accumulation of time zone data,
      and thus increased memory usage. In most cases, this is acceptable even
      for long-running applications, due to infrequent updates to the IANA database,
      and the reduced footprint of the libstdc++ implementation (see
      <xref linkend="zone.db.zic">link</xref>).
    </para>

    <para>
      If accumulating old <code>tzdb</code> databases is not acceptable
      (e.g. due to memory constraints), entries
      may be removed from <code>tzdb_list</code> using <code>tzdb_list::erase_after</code>.
      When using this function, you are responsible for ensuring that the application
      is no longer using any <code>time_zone*</code> to a removed database.
      If you erase a <code>tzdb</code> while some part of the application is still
      using it (or one of its <code>time_zone</code> objects) you will create
      a dangling pointer, leading to undefined behaviour.
      To make <code>tzdb_list::erase_after</code> safer, libstdc++ provides an
      <emphasis>extension</emphasis> that avoids creating dangling pointers.
      The <code>tzdb_list</code> uses <code>shared_ptr&lt;tzdb&gt;</code> to
      refer to each entry in the list, and <code>tzdb_list::const_iterator</code>
      also uses a <code>shared_ptr&lt;tzdb&gt;</code> to refer to its target.
      This means that iterators into the list share ownership of the list elements,
      so that erasing an element from the list does not destroy it if there are
      any iterators which share ownership of the element. The application can
      use this to ensure that <code>tzdb</code> and <code>time_zone</code>
      objects are not destroyed while they're still being used. The lifetime
      of a <code>tzdb</code> that is still in use can be extended by holding
      onto a <code>tzdb_list::const_iterator</code> that refers to it.
    </para>
  </section>

</section>

</chapter>