Accidentally applying the Strategy Pattern
March 11, 2014
Accidental strategy pattern
A recent project had me working on task extracting lists from a number of websites. My initial solution was to have a single class. It didn’t take long before it become verbose, with a number of branching if/else statements depending on the site being extracted. While my solution worked OK for one or two sites. I wasn’t going to be maintainable for 100’s of sites.
There two types of code:
In examining the code, the code could be separated into:
- Re-useable code
- Detailed/ Unique code; specific to the website being scrapped.
The re-useable code captures the patterns that needs to be followed in order to extract the listings from the site. These steps or pattern include: interacting with the HTML response/ request cycle, iterating through the listings and saving the results into the database.
The second type of code is particular to the website being scrapped and would be different for each website. For example, the code used to identify each record in the HTML and to parse the resulting text.
The solution that I had stumbled upon was to separate out the re-usable code from the website specific code. This is done using procs to store the website specific code (or rules). I had accidentally adopted the strategy pattern.
An example of the Strategy Pattern
class Person
attr_accessor :name
def initialize(&rules)
instance_exec(&rules)
end
end
Person.new do
# the rules go here
end
The strategy pattern separates the code that changes "rules" from the part that is re-used. In the above case, the re-used portion of code is contained in the Person class. The strength of this pattern is that you can change the rules at run-time and it enhances clarity as the intent of the code is not distorted by the particular implementation. The implementation of this pattern uses instance_exec to execute the rules - but more on that later.
Using lambdas and instance_exec
I didn’t have a good handle on lambdas. It seemed strange to me that the context was assigned when it was created rather than on execution. Thankfully, time spend with JavaScript, has cured me of this, as its very similar to JavaScript's concept of closures. The big difference to lambdas is how this and self is assigned. In JavaScript, this is determined by the calling object. However, in Ruby's lambdas, self is assigned when the lambda was created.
Take a look at the following example:
@some_variable ='some variable'
HTML_FORMATER = lambda do
puts @some_variable
end
class Report
def initialize(&format)
@some_variable ='this is variable in Report'
format.call()
end
end
Report.new(&HTML_FORMATER)
=> some variable
The @some_variable references the value "some variable" as this was the context when the block was created. However, if we want to change the content to refer to the object calling the method, this can achieved using instance_exec.
To see this in action
class Report
def initialize(&format)
@some_variable ='this is some variable in Report'
instance_exec(&format)
end
end
Report.new(&HTML_FORMATER)
=>this is some variable in Report
The @some_variable references the value ”this is some variable in Report” . The context used is the Report instance, which is what I am after.
My code as a final example:
I have implemented this as a rack task so that I can run it daily using CRON. I have simplified it to show the overall strategy.
# encoding=UTF-8
require 'rubygems'
require 'faraday'
require 'nokogiri'
require 'open-uri'
require 'yaml'
require 'ostruct'
# rake task to
namespace :api do
desc 'Search Engine Ranking'
task :search_engine_ranking => :environment do
def env_die
abort 'Fatal error: You need to specify a valid ENV.'
end
# Ensure that ENV is set to something
if ENV['RAILS_ENV'].nil?
env_die
end
class SearchEngine
attr_accessor :params, :conn, :response, :results,
:search_engine_request,:count
def initialize(keywords, base_url)
@params = OpenStruct.new
@count =0
@params.base_url = base_url
@params.path ='/search?q=' + keywords.join('+')
@conn = Faraday.new(:url => @params.base_url) do |faraday|
faraday.response :logger # log requests to STDOUT
faraday.request :url_encoded # form-encode POST params
faraday.adapter Faraday.default_adapter # Net::HTTP request
end
yield(@params)
self.instance_exec(params, ¶ms.rule_for_setup)
unless params.rule_for_setup.nil?
@search_engine_request =
SearchEngineRequest.create(keywords: keywords.join('+'),
site: base_url)
end
def get_action(path=params.path)
puts 'request on #{path}'
@response = conn.get path do |response|
response.headers['Cookie'] = params.cookie
unless params.cookie.nil?
end
end
def extract_records
doc = Nokogiri::HTML response.body
doc.css(self.instance_exec(¶ms.rule_to_indentify_records))
end
def process_results
@results =[]
extract_records.each do |record|
@count +=1
match, url, path =
self.instance_exec(record, ¶ms.rule_to_extract_record)
@results << {order_id: @count, url: url,
path: path, match?: match } unless match.nil?
end
save_results
end
def save_results
results.each do | record|
puts record
search_engine_request.search_engine_results.create(
{ rank_order: record[:order_id],
match: record[:match?],
path: record[:path][0..100],
url: record[:url][0..100] })
end
end
def loop_through_the_pages
0.step(100,10).each do |start|
get_action(params.url_first_part + start.to_s + params.url_last_part)
end
process_results
end
end
url = 'https://www.xxxx.com'
keywords =['ecommerce', 'spree']
search_engine = SearchEngine.new(keywords,url) do |params|
params.rule_to_indentify_records = lambda{ '#results li'}
params.rule_to_extract_record = lambda{ |record|
search_url = begin
record.css('h3 > a')[0].attribute_nodes.
find{|i| i.name='href'}.value
rescue
''
end
match = begin
record.text.match(%r{xx -your expr xxxxxx}).nil? ? false : true
rescue
false
end
url_path_match = search_url.
match(/.*http:\/\/(?<url>.*?)\/(?<path>.*)/)
url_path_match.nil? ? [] :
[match,url_path_match['url'], url_path_match['path']]
}
params.rule_for_setup = lambda{ |p|
response = get_action(p.path)
p.cookie = response.headers['set-cookie']
doc = Nokogiri::HTML response.body
raw_url =doc.css('.sb_pag li a')[2].attribute_nodes.
find{|i| i.name =='href'}.value
match_url=raw_url.match(/(?<first_part>.+?first=)\d+(?<last_part>.*)/)
p.url_first_part, p.url_last_part =
[match_url['first_part'], match_url['last_part']]
}
end
search_engine.loop_through_the_pages
end
end
