seamusabshere-date-performance

Date::Performance

This package adds some semblance of performance to Ruby's core Date class using a combination of different techniques:

1. Implements various core Date methods in C. This is nowhere near a complete rewrite of all Date features but many of the hot spots have been replaced with machine code.

2. Provide alternate implementations of strftime and strptime in C. The stock date formatting and parsing methods are extremely slow compared to their libc counterparts. Date#sys_strftime and Date::sys_strptime are light facades on top of the system's strftime(2) and strptime(2). The system methods run 10x and 50x (yes, fifty-ecks) faster than their Ruby based counterparts, respectively. Unfortunately, strftime(2) and strptime(2) implementations vary from system to system and have various limitations not found in the core Date implementation so can not safely be used as replacements for the core methods.

3. Memoization. The Date::Memoize module can be used to speed certain types of repetitive date processing significantly. This file must be required separately.

Synopsis

This package is mostly transparent after an initial require:

require 'date/performance'
Date.new 1912, 6, 23
# Wow! That was fast!

Date::Performance is not used directly but automatically replaces core Date methods when required.

In addition to the C extension, the Date::Memoization module can be used to speed things up even further in some cases by making a trade off between space and time:

require 'date/memoize'
Date.new 1912, 6, 23
Date.parse '1912-06-23'

Requiring the file automatically replaces Date::new / Date::civil, Date::parse, and Date::jd methods with memoized versions.

Installation / Hacking

This package has been tested on the following platforms:

• FreeBSD 5.4 (x86) and 6.1 (AMD64)
• Linux / Fedora Core 6 (x86)
• MacOS X (Intel)

The easiest way to install the package is to use RubyGems:

$ gem install date-performance --source=http://tomayko.com

Old versions and other dist formats are available at:

http://tomayko.com/dist/date-performance/

A git repository is also available:

$ git clone git://github.com/rtomayko/date-performance.git

Background

The Date class is often the cause of poor performance in Ruby programs. A frequent suggestion is to use the Time class, which is much faster, but that solution has correctness problems in a wide range of data typing cases. It is often the case that you want separate Date, Time, and DateTime types.

There are a couple of reasons why Date runs slowly when compared with Time. The common assumption is that this is mostly due to Time being written in C and Date being written in Ruby. While that clearly has an impact, I would argue that the reason Date is slow is because it's not designed to be fast. The code opts for readability over performance in almost every case. This is a feature.

Have you read the date.rb documentation [1]? The implementation is pretty hard core; it can handle a lot of weird cases that Time [2] does not and would appear to be a correct implementation of date handling, which has the usual side-effect of being slow.

The Date implementation uses a single Astronomical Julian Day (AJD) number to represent dates internally. In fact, Date#initialize takes a single ajd argument, which means that all date forms that are commonly used (UNIX timestamp, Civil, etc.) must be converted to an AJD before we can even instantiate the thing.

The real performance hit seems to come from the various rational number operations performed on the way from a civil, ordinal, and julian date to an AJD.

When I began writing Date::Performance, I was getting pretty big (3x - 4x) performance boosts in many places simply by optimizing the Ruby code a bit. These boosts came at the expense of readability, however, and so the decision was made to go for maximum unreadability and implement the boosts in C.

There's a nice balance here: the Ruby implementation reads like a spec, while the C version implements it for quickness.

Memoization

In addition to the C extension, this package includes a separate Date::Memoize module that can further speed up date processing in situations where the range of dates being manipulated is fairly dense and the same dates are being created repeatedly. Working with databases and flat files are two examples where memoization may help significantly.

The Date::Memoize module replaces various Date constructor methods (new, civil, and parse) with memoized[http://en.wikipedia.org/wiki/Memoization] versions (see Date::Memoization for details). The best way to determine whether memoization is right for you is to add it to your project and see what happens.

License

MIT. See the COPYING file included in the distribution for more information.

See Also

• [1] Ruby Date Implementation Notes and Documentation: http://www.ruby-doc.org/docs/rdoc/1.9/files/_/lib/date_rb.html

• [2] Ruby Time documentation http://www.ruby-doc.org/docs/rdoc/1.9/classes/Time.html