README.rdoc

= Geocoder

Geocoder is a complete geocoding solution for Ruby. With Rails it adds geocoding (by street or IP address), reverse geocoding (find street address based on given coordinates), and distance queries. It's as simple as calling +geocode+ on your objects, and then using a scope like <tt>Venue.near("Billings, MT")</tt>.


== Compatibility

* Supports multiple Ruby versions: Ruby 1.8.7, 1.9.2, and JRuby.
* Supports multiple databases: MySQL, PostgreSQL, SQLite, and MongoDB (1.7.0 and higher).
* Supports Rails 3. If you need to use it with Rails 2 please see the <tt>rails2</tt> branch (no longer maintained, limited feature set).
* Works very well outside of Rails but you'll need to install either the +json+ (for MRI) or +json_pure+ (for JRuby) gem.


== Install

=== As a Gem

Add to your Gemfile:

  gem "geocoder"

and run at the command prompt:

  bundle install

=== Or As a Plugin

At the command prompt:

  rails plugin install git://github.com/alexreisner/geocoder.git


== Configure Object Geocoding

In the below, note that addresses may be street or IP addresses.

=== ActiveRecord

Your model must have two attributes (database columns) for storing latitude and longitude coordinates. By default they should be called +latitude+ and +longitude+ but this can be changed (see "More on Configuration" below):

  rails generate migration AddLatitudeAndLongitudeToModel latitude:float longitude:float
  rake db:migrate

For reverse geocoding your model must provide a method that returns an address. This can be a single attribute, but it can also be a method that returns a string assembled from different attributes (eg: +city+, +state+, and +country+).

Next, your model must tell Geocoder which method returns your object's geocodable address:

  geocoded_by :full_street_address   # can also be an IP address
  after_validation :geocode          # auto-fetch coordinates

For reverse geocoding, tell Geocoder which attributes store latitude and longitude:

  reverse_geocoded_by :lat, :lon
  after_validation :reverse_geocode  # auto-fetch address

=== Mongoid

First, your model must have an array field for storing coordinates:

  field :coordinates, :type => Array

You may also want an address field, like this:

  field :address

but if you store address components (city, state, country, etc) in separate fields you can instead define a method called +address+ that combines them into a single string which will be used to query the geocoding service.

Once your fields are defined, include the <tt>Geocoder::Model::Mongoid</tt> module and then call <tt>geocoded_by</tt>:

  include Geocoder::Model::Mongoid
  geocoded_by :address               # can also be an IP address
  after_validation :geocode          # auto-fetch coordinates

Reverse geocoding is similar:

  include Geocoder::Model::Mongoid
  reverse_geocoded_by :coordinates
  after_validation :reverse_geocode  # auto-fetch address

=== Bulk Geocoding

If you have just added geocoding to an existing application with a lot of objects you can use this Rake task to geocode them all:

  rake geocode:all CLASS=YourModel

Geocoder will print warnings if you exceed the rate limit for your geocoding service.


== Request Geocoding by IP Address

Geocoder adds a +location+ method to the standard <tt>Rack::Request</tt> object so you can easily look up the location of any HTTP request by IP address. For example, in a Rails controller or a Sinatra app:

  # returns Geocoder::Result object
  result = request.location

See "Advanced Geocoding" below for more information about Geocoder::Result objects.


== Location-Aware Database Queries

To find objects by location, use the following scopes:

  Venue.near('Omaha, NE, US', 20)    # venues within 20 miles of Omaha
  Venue.near([40.71, 100.23], 20)    # venues within 20 miles of a point
  Venue.geocoded                     # venues with coordinates
  Venue.not_geocoded                 # venues without coordinates

With geocoded objects you can do things like this:

  obj.nearbys(30)                      # other objects within 30 miles
  obj.distance_from([40.714,-100.234]) # distance from arbitrary point to object
  obj.bearing_to("Paris, France")      # direction from object to arbitrary point

Some utility methods are also available:

  # look up coordinates of some location (like searching Google Maps)
  Geocoder.coordinates("25 Main St, Cooperstown, NY")
   => [42.700149, -74.922767]

  # distance (in miles) between Eiffel Tower and Empire State Building
  Geocoder::Calculations.distance_between([47.858205,2.294359], [40.748433,-73.985655])
   => 3619.77359999382

  # find the geographic center (aka center of gravity) of objects or points
  Geocoder::Calculations.geographic_center([city1, city2, [40.22,-73.99], city4])
   => [35.14968, -90.048929]

Please see the code for more methods and detailed information about arguments (eg, working with kilometers).


== Distance and Bearing

When you run a location-aware query the returned objects have two attributes added to them (only w/ ActiveRecord):

* <tt>obj.distance</tt> - number of miles from the search point to this object
* <tt>obj.bearing</tt> - direction from the search point to this object

Results are automatically sorted by distance from the search point, closest to farthest. Bearing is given as a number of clockwise degrees from due north, for example:

* <tt>0</tt> - due north
* <tt>180</tt> - due south
* <tt>90</tt> - due east
* <tt>270</tt> - due west
* <tt>230.1</tt> - southwest
* <tt>359.9</tt> - almost due north

You can convert these numbers to compass point names by using the utility method provided:

  Geocoder::Calculations.compass_point(355) # => "N"
  Geocoder::Calculations.compass_point(45)  # => "NE"
  Geocoder::Calculations.compass_point(208) # => "SW"

<i>Note: when using SQLite +distance+ and +bearing+ values are provided for interface consistency only. They are not very accurate.</i>

To calculate accurate distance and bearing with SQLite or Mongoid:

  obj.distance_to([43.9,-98.6])  # distance from obj to point
  obj.bearing_to([43.9,-98.6])   # bearing from obj to point
  obj.bearing_from(obj2)         # bearing from obj2 to obj

The <tt>bearing_from/to</tt> methods take a single argument which can be: a <tt>[lat,lon]</tt> array, a geocoded object, or a geocodable address (string). The <tt>distance_from/to</tt> methods also take a units argument (<tt>:mi</tt> or <tt>:km</tt>).


== More on Configuration

You are not stuck with using the +latitude+ and +longitude+ database column names (with ActiveRecord) or the +coordinates+ array (Mongoid) for storing coordinates. For example:

  geocoded_by :address, :latitude  => :lat, :longitude => :lon # ActiveRecord
  geocoded_by :address, :coordinates => :coords                # Mongoid

The +address+ method can return any string you'd use to search Google Maps. For example, any of the following are acceptable:

* "714 Green St, Big Town, MO"
* "Eiffel Tower, Paris, FR"
* "Paris, TX, US"

If your model has +street+, +city+, +state+, and +country+ attributes you might do something like this:

  geocoded_by :address

  def address
    [street, city, state, country].compact.join(', ')
  end

For reverse geocoding you can also specify an alternate name attribute where the address will be stored, for example:

  reverse_geocoded_by :lat, :lon, :address => :location  # ActiveRecord
  reverse_geocoded_by :coordinates, :address => :loc     # Mongoid


== Advanced Geocoding

So far we have looked at shortcuts for assigning geocoding results to object attributes. However, if you need to do something fancy you can skip the auto-assignment by providing a block (takes the object to be geocoded and an array of <tt>Geocoder::Result</tt> objects) in which you handle the parsed geocoding result any way you like, for example:

  reverse_geocoded_by :lat, :lon do |obj,results|
    if geo = results.first
      obj.city    = geo.city
      obj.zipcode = geo.postal_code
      obj.country = geo.country_code
    end
  end
  after_validation :reverse_geocode

Every <tt>Geocoder::Result</tt> object, +result+, provides the following data:

* <tt>result.latitude</tt> - float
* <tt>result.longitude</tt> - float
* <tt>result.coordinates</tt> - array of the above two
* <tt>result.address</tt> - string
* <tt>result.city</tt> - string
* <tt>result.postal_code</tt> - string
* <tt>result.country_name</tt> - string
* <tt>result.country_code</tt> - string

and if you're familiar with the results returned by the geocoding service you're using, you can access even more (see code comments for details: <tt>lib/geocoder/results/*</tt>).


== Geocoding Services

By default Geocoder uses Google's geocoding API to fetch coordinates and addresses. However if you wish to use Yahoo's geocoding API you can simply add this to an initializer:

  # config/initializers/geocoder.rb
  Geocoder::Configuration.lookup = :yahoo

Street address geocoding services currently supported (valid settings for the above):

* Google: <tt>:google</tt>
* Yahoo: <tt>:yahoo</tt>
* Geocoder.ca: <tt>:geocoder_ca</tt> (US and Canada only)

Note that the result objects returned by different geocoding services all implement the methods listed above. Beyond that, however, you must be familiar with your particular subclass of <tt>Geocoder::Result</tt> and the geocoding service's result structure:

* Google: http://code.google.com/apis/maps/documentation/geocoding/#JSON
* Yahoo: http://developer.yahoo.com/geo/placefinder/guide/responses.html
* Geocoder.ca: (???)
* FreeGeoIP: http://github.com/fiorix/freegeoip/blob/master/README.rst

=== API Keys

To use your Google API key or Yahoo app ID:

  Geocoder::Configuration.api_key = "..."

To obtain an API key (not required):

* Yahoo: https://developer.apps.yahoo.com/wsregapp
* Google: http://code.google.com/apis/maps/signup.html

=== Timeout

You can set the timeout used for connections to the geocoding service. The default is 3 seconds but if you want to set it to 5, for example, put the following in an initializer:

  Geocoder::Configuration.timeout = 5

=== Language

You can set the language used for reverse geocoding results to German, for example, by setting the following:

  Geocoder::Configuration.language = :de

For a list of supported languages see the documentation for the geocoding service you're using.

=== HTTPS

If you want to use HTTPS for geocoding service connections:

  Geocoder::Configuration.use_https = true

Note that currently the only service that supports HTTPS is Google.


== Caching Results

It's a good idea, when relying on any external service, to cache retrieved data. When implemented correctly it improves your app's response time and stability. It's easy to cache geocoding results with Geocoder, just configure a cache store:

  Geocoder::Configuration.cache = Redis.new

This example uses Redis, but the cache store can be any object that supports these methods:

* <tt>store#[](key)</tt>         - retrieves a value
* <tt>store#[]=(key, value)</tt> - stores a value
* <tt>store#keys</tt>            - lists all keys

Even a plain Ruby hash will work, though it's not a great choice (cleared out when app is restarted, not shared between app instances, etc).

You can also set a custom prefix to be used for cache keys:

  Geocoder::Configuration.cache_prefix = "..."

By default the prefix is <tt>geocoder:</tt>

If you need to expire cached content:

  Geocoder.cache.expire("http://...") # expire cached result for a URL
  Geocoder.cache.expire(:all)         # expire all cached results

Do *not* include the prefix when passing a URL to be expired. Expiring <tt>:all</tt> will only expire keys with the configured prefix (won't kill every entry in your key/value store).


== Forward and Reverse Geocoding in the Same Model

If you apply both forward and reverse geocoding functionality to the same model (say users can supply an address or coordinates and you want to fill in whatever's missing), you will provide two address methods:

* one for storing the fetched address (reverse geocoding)
* one for providing an address to use when fetching coordinates (forward geocoding)

For example:

  class Venue

    # build an address from street, city, and state attributes
    geocoded_by :address_from_components

    # store the fetched address in the full_address attribute
    reverse_geocoded_by :latitude, :longitude, :address => :full_address
  end

However, there can be only one set of latitude/longitude attributes, and whichever you specify last will be used. For example:

  class Venue

    geocoded_by :address,
      :latitude  => :fetched_latitude,  # this will be overridden by the below
      :longitude => :fetched_longitude  # same here

    reverse_geocoded_by :latitude, :longitude
  end

The reason for this is that we don't want ambiguity when doing distance calculations. We need a single, authoritative source for coordinates!


== Use Outside of Rails

You can use Geocoder outside of Rails by calling the <tt>Geocoder.search</tt> method:

  results = Geocoder.search("McCarren Park, Brooklyn, NY")

This returns an array of <tt>Geocoder::Result</tt> objects with all information provided by the geocoding service. Please see above and in the code for details.


== Notes on Mongoid

=== The Near Method

Mongoid document classes have a built-in +near+ scope, but since it only works two-dimensions Geocoder overrides it with its own spherical +near+ method in geocoded classes.

=== Latitude/Longitude Order

Coordinates are generally printed and spoken as latitude, then logitude ([lat,lon]). Geocoder respects this convention and always expects method arguments to be given in [lat,lon] order. However, MongoDB requires that coordinates be stored in [lon,lat] order as per the GeoJSON spec (http://geojson.org/geojson-spec.html#positions), so internally they are stored "backwards." However, this does not affect order of arguments to methods when using Mongoid. I mention this only in case you notice it and freak out. Don't worry. Everything is going to be OK.


== Distance Queries in SQLite

SQLite's lack of trigonometric functions requires an alternate implementation of the +near+ scope. When using SQLite, Geocoder will automatically use a less accurate algorithm for finding objects near a given point. Results of this algorithm should not be trusted too much as it will return objects that are outside the given radius, along with inaccurate distance and bearing calculations.


=== Discussion

There are few options for finding objects near a given point in SQLite without installing extensions:

1. Use a square instead of a circle for finding nearby points. For example, if you want to find points near 40.71, 100.23, search for objects with latitude between 39.71 and 41.71 and longitude between 99.23 and 101.23. One degree of latitude or longitude is at most 69 miles so divide your radius (in miles) by 69.0 to get the amount to add and subtract from your center coordinates to get the upper and lower bounds. The results will not be very accurate (you'll get points outside the desired radius), but you will get all the points within the required radius.

2. Load all objects into memory and compute distances between them using the <tt>Geocoder::Calculations.distance_between</tt> method. This will produce accurate results but will be very slow (and use a lot of memory) if you have a lot of objects in your database.

3. If you have a large number of objects (so you can't use approach #2) and you need accurate results (better than approach #1 will give), you can use a combination of the two. Get all the objects within a square around your center point, and then eliminate the ones that are too far away using <tt>Geocoder::Calculations.distance_between</tt>.

Because Geocoder needs to provide this functionality as a scope, we must go with option #1, but feel free to implement #2 or #3 if you need more accuracy.


== Tests

Geocoder comes with a test suite (just run <tt>rake test</tt>) that mocks ActiveRecord and is focused on testing the aspects of Geocoder that do not involve executing database queries. Geocoder uses many database engine-specific queries which must be tested against all supported databases (SQLite, MySQL, etc). Ideally this involves creating a full, working Rails application, and that seems beyond the scope of the included test suite. As such, I have created a separate repository which includes a full-blown Rails application and some utilities for easily running tests against multiple environments:

http://github.com/alexreisner/geocoder_test


== Known Issue

You cannot use the +near+ scope with another scope that provides an +includes+ option because the +SELECT+ clause generated by +near+ will overwrite it (or vice versa). Instead, try using +joins+ and pass a <tt>:select</tt> option to the +near+ scope to get the columns you want. For example, in Rails 2 syntax:

  # instead of :includes => :venues:
  City.near("Omaha, NE", 20, :select => "venues.*").all(:joins => :venues)

If anyone has a more elegant solution to this problem I am very interested in seeing it.


== Roadmap

* add support for more ORMs (Mongoid, DataMapper)
* add support for more geocoding services
* maintain the same simple interface


Copyright (c) 2009-11 Alex Reisner, released under the MIT license