Ruby Hash whose key is a function of the object? - ruby

Question

For instance,
s1 = Student.new(1, "Bob", "Podunk High")
hash[1] = s1
puts hash[1].name #produces "Bob"
s1.id = 15
puts hash[15].name #produces "Bob"
puts hash[1].name #fails
This is not exactly Hash-like behavior and insertions with the wrong key still needs to be defined.
While I can certainly roll my own container that behaves this way but it will be hard to make it fast, ie not search through the whole container every time [] is called. Just wondering if someone smarter has already made something I can steal.
EDIT: Some good ideas below helped me focus my requirements:
avoid the O(n) lookup time
allow multiple containers to the same object (association not composition)
have different data types (eg. that might use name instead of id) without too much reimplementation

Answer 1

You can implement it yourself.
Look at the draft solution:
class Campus
attr_reader :students
def initialize
#students = []
end
def [](ind)
students.detect{|s| s.id == ind}
end
def <<(st)
raise "Yarrr, not a student" if st.class != Student
raise "We already have got one with id #{st.id}" if self[st.id]
students << st
end
end
class Student
attr_accessor :id, :name, :prop
def initialize(id, name, prop)
#id, #name, #prop = id, name, prop
end
end
campus = Campus.new
st1 = Student.new(1, "Pedro", "Math")
st2 = Student.new(2, "Maria", "Opera")
campus << st1
campus << st2
campus[1]
#=> Student...id:1,name:pedro...
campus[2].name
#=> Maria
campus[2].id = 10
campus[2]
#=> error
campus[10].name
#=> Maria
Or you can play around Array class (or Hash, if you really need it):
class StrangeArray < Array
def [](ind)
self.detect{|v| v.id == ind} || raise "nothing found" # if you really need to raise an error
end
def <<(st)
raise "Looks like a duplicate" if self[st.id]
self.push(st)
end
end
campus = StrangeArray.new
campus << Student.new(15, 'Michael', 'Music')
campus << Student.new(40, 'Lisa', 'Medicine')
campus[1]
#=> error 'not found'
campus[15].prop
#=> Music
campus[15].id = 20
campus[20].prop
#=> Music
etc
And after #tadman's correct comment you can use reference to your hash right into your Student class:
class Student
attr_accessor :name, :prop
attr_reader :id, :campus
def initialize(id, name, prop, camp=nil)
#id, #name, #prop = id, name, prop
self.campus = camp if camp
end
def id=(new_id)
if campus
rase "this id is already taken in campus" if campus[new_id]
campus.delete id
campus[new_id] = self
end
#id = new_id
end
def campus=(camp)
rase "this id is already taken in campus" if camp[id]
#campus = camp
camp[#id] = self
end
end
campus = {}
st1 = Student.new(1, "John", "Math")
st2 = Student.new(2, "Lisa", "Math", campus)
# so now in campus is only Lisa
st1.campus = campus
# we've just pushed John in campus
campus[1].name
#=> John
campus[1].id = 10
campus[10].name
#=> John

Answer 2

While the Hash object might not behave the way you want it to, you can always customize the objects being inserted to be hashed a particular way.
You can do this by adding two new methods to your existing class:
class Student
def hash
self.id
end
def eql?(student)
self.id == student.id
end
end
By defining hash to return a value based on id, Hash will consider these two candidates for the same spot in the hash. The second definition declares "hash equivalence" between any two objects that have the same hash value.
This will work well provided your id values fit into a conventional 32-bit Fixnum and aren't 64-bit BIGINT database values.
As fl00r points out, this will only work if your id is immutable. For most databases this tends to be the case. Changing id on the fly is probably a really bad idea, though, as it can lead to total chaos and mind-blowing bugs.

Answer 3

This is a hard problem. Database vendors can make money because it is a hard problem. You are basically looking to implement traditional RDBMS indices: search through derived data, to provide fast lookup to the data it was derived from, while allowing that data to change. If you want to access the data from multiple threads, you'll quickly run into all the issues that make it hard to make a database ACID compliant.
I suggest putting the data into a database, adding the necessary indices and letting the database -- an application optimized for this exact purpose -- do the work.

Answer 4

The container must be notified when your key has been changed, otherwise you must search the key on the fly in lg(n).
If you rarely change key and lookup a lot, just rebuild the hash:
def build_hash_on_attribute(objects, attribute)
Hash[objects.collect { |e| [e.send(method), e] }]
end
s1 = OpenStruct.new id: 1, name: 's1'
h = build_hash_on_attribute([s1], :id)
h[1].name # => 's1'
h[1].id = 15
# rebuild the whole index after any key attribute has been changed
h = build_hash_on_attribute(h.values, :id)
h[1] # => nil
h[15].name # => 's1'
Update 02/12: Add a solution using observer pattern
Or you do need such automatically index building, you can use observer pattern like below or decorator pattern. But you need to use the wrapped objects in decorator pattern.
gist: https://gist.github.com/1807324
module AttrChangeEmitter
def self.included(base)
base.extend ClassMethods
base.send :include, InstanceMethods
end
module ClassMethods
def attr_change_emitter(*attrs)
attrs.each do |attr|
class_eval do
alias_method "#{attr}_without_emitter=", "#{attr}="
define_method "#{attr}_with_emitter=" do |v|
previous_value = send("#{attr}")
send "#{attr}_without_emitter=", v
attr_change_listeners_on(attr).each do |listener|
listener.call self, previous_value, v
end
end
alias_method "#{attr}=", "#{attr}_with_emitter="
end
end
end
end
module InstanceMethods
def attr_change_listeners_on(attr)
#attr_change_listeners_on ||= {}
#attr_change_listeners_on[attr.to_sym] ||= []
end
def add_attr_change_listener_on(attr, block)
listeners = attr_change_listeners_on(attr)
listeners << block unless listeners.include?(block)
end
def remove_attr_change_listener_on(attr, block)
attr_change_listeners_on(attr).delete block
end
end
end
class AttrChangeAwareHash
include Enumerable
def initialize(attr = :id)
#attr = attr.to_sym
#hash = {}
end
def each(&block)
#hash.values.each(&block)
end
def on_entity_attr_change(e, previous_value, new_value)
if #hash[previous_value].equal? e
#hash.delete(previous_value)
# remove the original one in slot new_value
delete_by_key(new_value)
#hash[new_value] = e
end
end
def add(v)
delete(v)
v.add_attr_change_listener_on(#attr, self.method(:on_entity_attr_change))
k = v.send(#attr)
#hash[k] = v
end
alias_method :<<, :add
def delete(v)
k = v.send(#attr)
delete_by_key(k) if #hash[k].equal?(v)
end
def delete_by_key(k)
v = #hash.delete(k)
v.remove_attr_change_listener_on(#attr, self.method(:on_entity_attr_change)) if v
v
end
def [](k)
#hash[k]
end
end
class Student
include AttrChangeEmitter
attr_accessor :id, :name
attr_change_emitter :id, :name
def initialize(id, name)
self.id = id
self.name = name
end
end
indexByIDA = AttrChangeAwareHash.new(:id)
indexByIDB = AttrChangeAwareHash.new(:id)
indexByName = AttrChangeAwareHash.new(:name)
s1 = Student.new(1, 'John')
s2 = Student.new(2, 'Bill')
s3 = Student.new(3, 'Kate')
indexByIDA << s1
indexByIDA << s3
indexByIDB << s1
indexByIDB << s2
indexByName << s1
indexByName << s2
indexByName << s3
puts indexByIDA[1].name # => John
puts indexByIDB[2].name # => Bill
puts indexByName['John'].id # => 1
s2.id = 15
s2.name = 'Batman'
p indexByIDB[2] # => nil
puts indexByIDB[15].name # => Batman
indexByName.each do |v|
v.name = v.name.downcase
end
p indexByName['John'] # => nil
puts indexByName['john'].id # => 1
p indexByName.collect { |v| [v.id, v.name] }
# => [[1, "john"], [3, "kate"], [15, "batman"]]
indexByName.delete_by_key 'john'
indexByName.delete(s2)
s2.id = 1 # set batman id to 1 to overwrite john
p indexByIDB.collect { |v| [v.id, v.name] }
# => [[1, "batman"]]
p indexByName.collect { |v| [v.id, v.name] }
# => [[3, "kate"]]