= ThirdBase: A Fast and Easy Date/DateTime Class for Ruby

ThirdBase differs from Ruby's standard Date/DateTime class in the following
ways:

- ThirdBase is roughly 2-10 times faster depending on usage
- ThirdBase has a lower memory footprint
- ThirdBase supports pluggable parsers
- ThirdBase doesn't depend on Ruby's Rational class
- ThirdBase always uses the gregorian calendar

== Background

The Ruby standard Date class tries to be all things to all people.  While
it does a decent job, it's slow enough to be the bottleneck in some
applications.  If we decide not to care about the Date of Calendar Reform
and the fact that the Astronomical Julian Date differs from the Julian
Date, much of the complexity of Ruby's standard Date/DateTime class can
be removed, and there can be significant improvements in speed.

== Resources

* {RDoc}[http://third-base.rubyforge.org]
* {Source code}[http://github.com/jeremyevans/third_base]
* {Bug tracking}[http://rubyforge.org/projects/third-base/]

To check out the source code:
  
    git clone git://github.com/jeremyevans/third_base.git

== Installation

  sudo gem install third_base

== Usage and Compatibility

There are three ways that ThirdBase can be used:

=== Alongside the standard Date/DateTime class

Usage:

  require 'third_base'

If you just require it, you can use ThirdBase::Date and ThirdBase::DateTime
alongside the standard Date and DateTime classes. This ensures compatibility
with all existing software, but doesn't provide any performance increase to any
class not explicitly using ThirdBase.

=== Replace Date and DateTime with ThirdBase's

Usage:

  require 'third_base'
  include ThirdBase

This is the least compatible method.  It may work for some applications but
will break most, because if they use "require 'date'", they will get a
superclass mismatch.  Also ThirdBase::Date is not completely API compatible
with the standard Date class, so it could break depending on how the
application used Date.

If you aren't using any libraries that use ruby's standard Date class, this is
an easy way to be able to use Date and DateTime to refer to ThirdBase's
versions instead of Ruby's standard versions.

Note that rubygems indirectly uses the standard Date class, so if you want to
do this, you'll have to unpack the gem and put it in the $LOAD_PATH manually.

One case in which this pattern is useful is if you want to use ThirdBase within
your libraries as the date class, but with other libaries that use the standard
version as the date class.  To do this:

  require 'third_base'
  class YourLibrary
    include ThirdBase
    def today
      Date.today
    end
  end
  
This makes it so that references to Date within YourLibrary use
ThirdBase::Date, while references to Date outside YourLibrary use the standard
Date class.

=== Use ThirdBase's compatibility mode via the third_base executable

Usage:

  $ third_base irb 
  $ third_base mongrel_rails
  $ third_base ruby -rdate -e "p Date.ancestors"

This should be used if you want to make all libraries use ThirdBase's Date
class.  Doing this means that even if they "require 'date'", they will use
ThirdBases's versions.  More explicity, it will define Date and DateTime
as subclasses of ThirdBase::Date and ThirdBase::DateTime, and make them as
API compatible as possible.

You could get this by using "require 'third_base/compat'".  Unfortunately,
that doesn't work if you are using rubygems (and ThirdBase is mainly
distributed as a gem), because rubygems indirectly requires date.

The third_base executable modifies the RUBYLIB and RUBYOPT environment
variables and should ensure that even if a ruby library requires 'date', they
will get the ThirdBase version with the compatibility API.  To use the
third_base executable, you just prepend it to any command that you want to run.

This is the middle ground.  It should work for most applications, but as
ThirdBase's compatibility API is not 100% compatible with the standard Date
class, things can still break. See the next section for some differences.

If you have good unit tests/specs, you can try using this in your application
then running your specs (e.g. third_base rake spec).  Assuming good coverage,
if you have no errors, it should be OK to use, and you'll get a nice speedup.

== Incompatibilities with the standard Date class when using third_base/compat

* The marshalling format is different
* The new! class methods take different arguments
* Methods which returned rationals now return integers or floats
* ajd and amjd are now considered the same as jd and mjd, respectively
* The gregorian calendar is now the only calendar used
* All parsed two digit years are mapped to a year between 1969 and 2068
* Default parsing may be different, but the user can modify the parsers used
* Potentially others, but hopefully anything else can be fixed

== Pluggable Parsers

The standard Date class has a hard coded parsing routine that cannot be easily
modified by the user.  ThirdBase uses a different approach, by allowing the
user to add parsers and change the order of parsers.  There are some default
parsers built into ThirdBase's Date and DateTime, and they should work well for
the majority of American users.  However, there is no guarantee that it
includes a parser for the format you want to parse (though you can add a parser
that will do so).

The user should note that ThirdBases's Date and DateTime classes have
completely separate parsers, and modifying one does not affect the other.

=== Adding Parser Types

ThirdBase's parsers are separated into parser types.  The Date class has
four parser types built in: :iso, :us, :num, and :eu, of which only :iso,
:us, and :num are used by default.   DateTime has all of the parser types
that Date has, and an additional one called :time.

To add a parser type:

  Date.add_parser_type(:mine)
  DateTime.add_parser_type(:mine)

=== Adding Regexp Parsers to Parser Types

A ThirdBase Date/Datetime regexp parser consists of two parts, a regular
expression, and a block that takes a MatchData object and returns a 
Date/DateTime instance or a hash to be passed to Date/DateTime.new!.  The
block is only called if the regular expression matches the string to be
parsed, and it can return nil if it is not able to successfully parse the
string (even if the string matches the regular expression).  To add a
parser, you use the add_parser class method, which takes an argument
specifying which parser family to use, the regular expression, and a block
that is used as a proc for the parser:

To add a parser to a parser type:

  Date.add_parser(:mine, /\Atoday\z/i) do |m|
    t = Time.now
    {:civil=>[t.year, t.mon, t.day]}
  end

  DateTime.add_parser(:mine, /\Anow\z/i) do |m|
    t = Time.now
    {:civil=>[t.year, t.mon, t.day], :parts=>[t.hour, t.min, t.sec, t.usec] \
      :offset=>t.utc_offset}
  end

If you add a DateTime parser that may guess at certain values instead of
parsing them out of the string, you should include a :not_parsed entry which
is an array of symbols indicating items that were not parsed directly out
of the string.  This is only necessary if you are using the compatibility
mode and want better compatibility for Date._parse (which ruby's Time.parse
method uses internally):

  DateTime.add_parser(:mine, /\A(\d\d)_(\d\d)_(\d{4})\z/i) do |m|
    {:civil=>[m[3].to_i, m[2].to_i, m[1].to_i], :parts=>[0,0,0,0], :offset=>0, \
     :not_parsed=>[:hour, :min, :sec, :sec_fraction, :offset, :zone]}
  end

The entries in :not_parsed should match keys that can be returned by
Date._parse, so in addition to the ones listed above, you can also use
:year, :mon, and :mday.

Adding a parser to a parser type adds it to the front of the array of parsers 
for that type, so it will be tried before other parsers for that type.  It is
an error to add a parser to a parser type that doesn't exist.

=== Adding strptime Parsers to Parser Types (New in 1.2.0)

ThirdBase 1.2.0 added the ability to more easily create common parsers
using the strptime format string syntax.  These are created similar to regexp
parsers, but use a format string, and the block is then optional (and should
be omitted unless you know what you are doing):

  DateTime.add_parser(:mine, '%Z %m~%Y~%d %S`%M`%H')

=== Modifying the Order of Parsers Types

You can change the order in which parsers types are tried by using the
use_parsers class method, which takes multiple arguments specifying the order
of parser types:

To modify the order of parser types:

  Date.use_parsers(:mine, :num, :iso, :us)
  DateTime.use_parsers(:time, :iso, :mine, :eu, :num)

== Performance

=== Synthetic Benchmark

  Date vs. ThirdBase::Date: 20000 Iterations
        user     system      total        real
  Date.new                  1.210000   0.000000   1.210000 (  1.209048)
  ThirdBase::Date.new       0.240000   0.000000   0.240000 (  0.237548)
  Date.new >>               4.100000   0.010000   4.110000 (  4.107972)
  ThirdBase::Date.new >>    0.580000   0.010000   0.590000 (  0.585797)
  Date.new +                1.580000   0.030000   1.610000 (  1.613447)
  ThirdBase::Date.new +     0.810000   0.000000   0.810000 (  0.803092)
  Date.parse                6.180000   0.180000   6.360000 (  6.364501)
  ThirdBase::Date.parse     0.540000   0.000000   0.540000 (  0.532560)
  Date.strptime             6.680000   0.030000   6.710000 (  6.707893)
  ThirdBase::Date.strptime  2.200000   0.040000   2.240000 (  2.241585)
  DateTime vs. ThirdBase::DateTime: 20000 Iterations
        user     system      total        real
  DateTime.new                  3.490000   0.270000   3.760000 (  3.760513)
  ThirdBase::DateTime.new       0.350000   0.000000   0.350000 (  0.357525)
  DateTime.new >>               6.720000   0.230000   6.950000 (  6.953825)
  ThirdBase::DateTime.new >>    0.840000   0.020000   0.860000 (  0.854347)
  DateTime.new +                3.730000   0.170000   3.900000 (  3.894309)
  ThirdBase::DateTime.new +     0.780000   0.060000   0.840000 (  0.834865)
  DateTime.parse                8.450000   0.400000   8.850000 (  8.854514)
  ThirdBase::DateTime.parse     0.980000   0.040000   1.020000 (  1.015109)
  DateTime.strptime            10.860000   0.380000  11.240000 ( 11.243913)
  ThirdBase::DateTime.strptime  3.410000   0.160000   3.570000 (  3.574491)

=== Real World Example

ThirdBase was written to solve a real world problem, slow retrieval of records
from a database because they contained many date fields. The table in
question (employees), has 23 fields, 5 of which are date fields.  Here are
the results of selecting all records for the database via Sequel, both with
and without third_base:

  $ script/benchmarker 100 Employee.all
              user     system      total        real
  #1     25.990000   0.040000  26.030000 ( 27.587781)
  $ third_base script/benchmarker 100 Employee.all
              user     system      total        real
  #1     13.640000   0.100000  13.740000 ( 15.018741)

Note that the times above include the time to query the database and
instantiate all of the Model objects.  In this instance you can see that
ThirdBase doubles performance with no change to the existing code.  This is
do to the fact that previously, date-related code took about 3/4 of the
processing time:

  ruby-prof graph profile without ThirdBase for Employee.all 100 times:

  75.87%   1.05%    101.51      1.40      0.00    100.12            85500     <Class::Date>#new (/usr/local/lib/ruby/1.8/date.rb:725}

  ruby-prof graph profile with ThirdBase for Employee.all 100 times:

  36.43%   1.29%     18.01      0.64      0.00     17.37            85500     <Class::ThirdBase::Date>#new 

ThirdBase still takes up over a third of the processing time, but the total
time it takes has been reduced by a factor of 5.  There may be opportunities
to further speed up ThirdBase--while it was designed to be faster than the
default Date class, there have been no attempts to optimize its performance.

== License

ThirdBase is released under the MIT License.  See the LICENSE file for details.

== Author

Jeremy Evans <code@jeremyevans.net>