SYNOPSIS

    For English:

     use DateTime::Format::Alami::EN;
     my $parser = DateTime::Format::Alami::EN->new();
     my $dt = $parser->parse_datetime("2 hours 13 minutes from now");

    Or you can also call as class method:

     my $dt = DateTime::Format::Alami::EN->parse_datetime("yesterday");

    To parse duration:

     my $dtdur = DateTime::Format::Alami::EN->parse_datetime_duration("2h"); # 2 hours

    For Indonesian:

     use DateTime::Format::Alami::ID;
     my $parser = DateTime::Format::Alami::ID->new();
     my $dt = $parser->parse_datetime("5 jam lagi");

    Or you can also call as class method:

     my $dt = DateTime::Format::Alami::ID->parse_datetime("hari ini");

    To parse duration:

     my $dtdur = DateTime::Format::Alami::ID->parse_datetime_duration("2h"); # 2 days

DESCRIPTION

    This class parses human/natural date/time/duration string and returns
    DateTime (or DateTime::Duration) object. Currently it supports English
    and Indonesian. The goal of this module is to make it easier to add
    support for other human languages.

    To actually use this class, you must use one of its subclasses for each
    human language that you want to parse.

    There are already some other DateTime human language parsers on CPAN
    and elsewhere, see "SEE ALSO".

HOW IT WORKS

    DateTime::Format::Alami is base class. Each human language is
    implemented in a separate DateTime::Format::Alami::<ISO_CODE> module
    (e.g. DateTime::Format::Alami::EN and DateTime::Format::Alami::EN)
    which is a subclass.

    Parsing is done using a single recursive regex (i.e. containing
    (?&NAME) and (?(DEFINE)) patterns, see perlre). This regex is composed
    from pieces of pattern strings in the p_* and o_* methods, to make it
    easier to override in an OO-fashion.

    A pattern string that is returned by the p_* method is a normal regex
    pattern string that will be compiled using the /x and /i regex
    modifier. The pattern string can also refer to pattern in other o_* or
    p_* method using syntax <o_foo> or <p_foo>. Example, o_today for
    English might be something like:

     sub p_today { "(?: today | this \s+ day )" }

    Other examples:

     sub p_yesterday { "(?: yesterday )" }
    
     sub p_dateymd { join(
         "",
        '(?: <o_dayint> \\s* ?<o_monthname> | <o_monthname> \\s* <o_dayint>\\b|<o_monthint>[ /-]<o_dayint>\\b )',
        '(?: \\s*[,/-]?\\s* <o_yearint>)?'
     )}
    
     sub o_date { "(?: <p_today>|<p_yesterday>|<p_dateymd>)" }
    
     sub p_time { "(?: <o_hour>:<o_minute>(?:<o_second>)? \s* <o_ampm> )" }
    
     sub p_date_time { "(?: <o_date> (?:\s+ at)? <o_time> )" }

    When a pattern from p_* matches, a corresponding action method a_* will
    be invoked. Usually the method will set or modify a DateTime object in
    $self->{_dt}. For example, this is code for a_today:

     sub a_today {
         my $self = shift;
         $self->{_dt} = DateTime->today;
     }

    The patterns from all p_* methods will be combined in an alternation to
    form the final pattern.

    An o_* pattern is just like p_*, but they will not be combined into the
    final pattern and matching it won't execute a corresponding a_* method.

    And there are also w_* methods which return array of strings.

    Parsing duration is similar, except the method names are pdur_*, odur_*
    and adur_*.

ADDING A NEW HUMAN LANGUAGE

    See an example in existing DateTime::Format::Alami::* module. Basically
    you just need to supply the necessary patterns in the p_* methods. If
    you want to introduce new p_* method, don't forget to supply the action
    too in the a_* method.

METHODS

 new => obj

    Constructor. You actually must instantiate subclass instead.

 parse_datetime($str[ , \%opts ]) => obj

    Parse/extract date/time expression in $str. Die if expression cannot be
    parsed. Otherwise return DateTime object (or string/number if format
    option is verbatim/epoch, or hash if format option is combined) or
    array of objects/strings/numbers (if returns option is all/all_cron).

    Known options:

      * time_zone => str

      Will be passed to DateTime constructor.

      * format => str (DateTime|verbatim|epoch|combined)

      The default is DateTime, which will return DateTime object. Other
      choices include verbatim (returns the original text), epoch (returns
      Unix timestamp), combined (returns a hash containing keys like
      DateTime, verbatim, epoch, and other extra information: pos [position
      of pattern in the string], pattern [pattern name], m [raw named
      capture groups], uses_time [whether the date involves time of day]).

      You might think that choosing epoch or verbatim could avoid the
      overhead of DateTime, but actually you can't since DateTime is used
      as the primary format during parsing. The epoch is retrieved from the
      DateTime object using the epoch method.

      * prefers => str (nearest|future|past)

      NOT YET IMPLEMENTED.

      This option decides what happens when an ambiguous date appears in
      the input. For example, "Friday" may refer to any number of Fridays.
      Possible choices are: nearest (prefer the nearest date, the default),
      future (prefer the closest future date), past (prefer the closest
      past date).

      * returns => str (first|last|earliest|latest|all|all_cron)

      If the text has multiple possible dates, then this argument
      determines which date will be returned. Possible choices are: first
      (return the first date found in the string, the default), last
      (return the final date found in the string), earliest (return the
      date found in the string that chronologically precedes any other date
      in the string), latest (return the date found in the string that
      chronologically follows any other date in the string), all (return
      all dates found in the string, in the order they were found in the
      string), all_cron (return all dates found in the string, in
      chronological order).

      When all or all_cron is chosen, function will return array(ref) of
      results instead of a single result, even if there is only a single
      actual result.

 parse_datetime_duration($str[ , \%opts ]) => obj

    Parse/extract duration expression in $str. Die if expression cannot be
    parsed. Otherwise return DateTime::Duration object (or string/number if
    format option is verbatim/seconds, or hash if format option is
    combined) or array of objects/strings/numbers (if returns option is
    all/all_sorted).

    Known options:

      * format => str (Duration|verbatim|seconds|combined)

      The default is Duration, which will return DateTime::Duration object.
      Other choices include verbatim (returns the original text), seconds
      (returns number of seconds, approximated), combined (returns a hash
      containing keys like Duration, verbatim, seconds, and other extra
      information: pos [position of pattern in the string], pattern
      [pattern name], m [raw named capture groups]).

      You might think that choosing seconds or verbatim could avoid the
      overhead of DateTime::Duration, but actually you can't since
      DateTime::Duration is used as the primary format during parsing. The
      number of seconds is calculated from the DateTime::Duration object
      using an approximation (for example, "1 month" does not convert
      exactly to seconds).

      * returns => str (first|last|smallest|largest|all|all_sorted)

      If the text has multiple possible durations, then this argument
      determines which date will be returned. Possible choices are: first
      (return the first duration found in the string, the default), last
      (return the final duration found in the string), smallest (return the
      smallest duration), largest (return the largest duration), all
      (return all durations found in the string, in the order they were
      found in the string), all_sorted (return all durations found in the
      string, in smallest-to-largest order).

      When all or all_sorted is chosen, function will return array(ref) of
      results instead of a single result, even if there is only a single
      actual result.

FAQ

 What does "alami" mean?

    It is an Indonesian word, meaning "natural".

 How does it compare to similar modules?

    DateTime::Format::Natural (DF:Natural) is a more established module
    (first released on 2006) and can understand a bit more English
    expression like 'last day of Sep'. Aside from English, it does not yet
    support other languages.

    DFA:EN's parse_datetime_duration() produces a DateTime::Duration object
    while DF:Natural's parse_datetime_duration() returns two DateTime
    objects instead. In other words, DF:Natural can parse "from 23 Jun to
    29 Jun" in addition to "for 2 weeks".

    DF:Natural in general is slightly more strict about the formats it
    accepts, e.g. it rejects Jun 23st (the error message even gives hints
    that the suffix must be 'rd'). DF:Natural can give a detailed error
    message on why parsing has failed (see its error() method).

    DateTime::Format::Flexible (DF:Flexible) is another established module
    (first released in 2007) that, aside from parsing human expression
    (like 'tomorrow', 'sep 1st') can also parse date/time in several other
    formats like RFC 822, making it a convenient module to use as a
    'one-stop' solution to parse date. Compared to DF:Natural, it has
    better support for timezone but cannot parse some English expressions.
    Aside from English, it currently supports German and Spanish. It does
    not support parsing duration expression.

    This module itself: DateTime::Format::Alami (DF:Alami) is yet another
    implementation. Internally, it uses recursive regex to make parsing
    simpler and adding more languages easier. It requires perl 5.14.0 or
    newer due to the use of (?{ ... }) code blocks inside regular
    expression (while DF:Natural and DF:Flexible can run on perl 5.8+). It
    currently supports English and Indonesian. It supports parsing duration
    expression and returns DateTime::Duration object. It has the smallest
    startup time (see see Bencher::Scenario::DateTimeFormatAlami::Startup).

    Performance-wise, all the modules are within the same order of
    magnitude (see Bencher::Scenario::DateTimeFormatAlami::Parsing).

SEE ALSO

 Similar modules on CPAN

    Date::Extract. DateTime::Format::Alami has some features of
    Date::Extract so it can be used to replace Date::Extract.

    DateTime::Format::Flexible. See "FAQ".

    For Indonesian: DateTime::Format::Indonesian, Date::Extract::ID
    (currently this module uses DateTime::Format::Alami::ID as its
    backend).

    For English: DateTime::Format::Natural. See "FAQ".

 Other modules on CPAN

    DateTime::Format::Human deals with formatting and not parsing.

 Similar non-Perl libraries

    Natt Java library, which the last time I tried sometimes gives weird
    answer, e.g. "32 Oct" becomes 1 Oct in the far future.
    http://natty.joestelmach.com/

    Duckling Clojure library, which can parse date/time as well as numbers
    with some other units like temperature.
    https://github.com/wit-ai/duckling

