From: Robert Klemme <shortcutter@...>
Date: 2007-06-01T20:25:20+09:00
Subject: Re: Set intersection with a specific criteria: a better idiom?

On 31.05.2007 19:21, Joel VanderWerf wrote:
> Eric Torreborre wrote:
>> Hi,
>>
>> I have been trying to intersect 2 sets of objects that where I want 
>> the "intersected" objects to share a common property.
>>
>> This is what I came up with:
>>
>> Here, I want to get all the Errors having the same file name:
>>
>> def intersection_set
>>   Set.new(event_errors).&(Set.new(socket_channel_errors)) {|x| 
>> file_name(x)}
>> end
>>
>> To allow that I extended the Set class with:
>>
>> def &(enum, &property)
>>     n = self.class.new
>>     if block_given?
>>       mapped = self.map{|e| property.call(e)}
>>       enum.each { |o| n.add(o) if mapped.include?(property.call(o)) }
>>     else
>>       enum.each { |o| n.add(o) if include?(o) }
>>     end
>>     n
>> end
>>  
>> I have met this requirement quite some times while scripting and I may 
>> miss something here.
>> Do you know a more idiomatic, Ruby-way of doing that?

Basically you want to do the intersection based on a specific criterion. 
  My first choice in such a scenario would be a Hash based solution, 
i.e. you create an index based on what you consider the key field.  See 
attachment for a sample implementation.

Regarding your question whether to include your code in std lib or 
facets lib: I vote "no".  Here's why: a set intersection will leave you 
with a single uniform set of items, namely all items present in all sets 
that take part in the intersection.  Your solution is not directly an 
intersection but it creates a copy of one set with all elements from the 
copy removed that do not fit a certain criterion.  More specifically, 
you kind of reversed the logic because the resulting set has only 
elements from the /parameter/ (i.e. the right hand side of the operator) 
and not from the left hand side.  So while for intersection order of 
arguments does not matter, it does for substraction:

irb(main):001:0> require 'set'
=> true
irb(main):002:0> s1 = (1..10).to_set
=> #<Set: {5, 6, 1, 7, 2, 8, 3, 9, 4, 10}>
irb(main):003:0> s2 = (5..15).to_set
=> #<Set: {5, 11, 6, 12, 7, 13, 8, 14, 9, 15, 10}>
irb(main):004:0> s1 - s2
=> #<Set: {1, 2, 3, 4}>
irb(main):005:0> s2 - s1
=> #<Set: {11, 12, 13, 14, 15}>

I would rewrite your original code like this:

require 'set'
...
# use a set for efficiency reasons
keys = event_errors.map {|x| file_name x}.to_set
# or, more efficient:
keys = event_errors.inject(Set.new) {|s,x| s << file_name(x)}
# create the result
result = socket_channel_errors.select {|x| keys.include? file_name(x)}

Since this is basically a two liner I don't see any necessity to change 
std libs.

Kind regards

	robert