Override the default :eager_loader option for many_*_many associations to work with an identity_map. If the :eager_graph association option is used, you'll probably have to use :uniq=>true on the current association and :cartesian_product_number=>2 on the association mentioned by :eager_graph, otherwise you'll end up with duplicates because the row proc will be getting called multiple times for the same object. If you do have duplicates and you use :eager_graph, they'll probably be lost. Making that work correctly would require changing a lot of the core architecture, such as how graphing and eager graphing work.
# File lib/sequel/plugins/identity_map.rb, line 48 def associate(type, name, opts = {}, &block) if opts[:eager_loader] super elsif type == :many_to_many opts = super el = opts[:eager_loader] model = self left_pk = opts[:left_primary_key] uses_lcks = opts[:uses_left_composite_keys] uses_rcks = opts[:uses_right_composite_keys] right = opts[:right_key] rcks = opts[:right_keys] join_table = opts[:join_table] left = opts[:left_key] lcks = opts[:left_keys] left_key_alias = opts[:left_key_alias] ||= opts.default_associated_key_alias opts[:eager_loader] = lambda do |eo| return el.call(eo) unless model.identity_map h = eo[:id_map] eo[:rows].each{|object| object.associations[name] = []} r = uses_rcks ? rcks.zip(opts.right_primary_keys) : [[right, opts.right_primary_key]] l = uses_lcks ? [[lcks.map{|k| SQL::QualifiedIdentifier.new(join_table, k)}, h.keys]] : [[left, h.keys]] # Replace the row proc to remove the left key alias before calling the previous row proc. # Associate the value of the left key alias with the associated object (through its object_id). # When loading the associated objects, lookup the left key alias value and associate the # associated objects to the main objects if the left key alias value matches the left primary key # value of the main object. # # The deleting of the left key alias from the hash before calling the previous row proc # is necessary when an identity map is used, otherwise if the same associated object is returned more than # once for the association, it won't know which of current objects to associate it to. ds = opts.associated_class.inner_join(join_table, r + l) pr = ds.row_proc h2 = {} ds.row_proc = proc do |hash| hash_key = if uses_lcks left_key_alias.map{|k| hash.delete(k)} else hash.delete(left_key_alias) end obj = pr.call(hash) (h2[obj.object_id] ||= []) << hash_key obj end model.eager_loading_dataset(opts, ds, Array(opts.select), eo[:associations], eo) .all do |assoc_record| if hash_keys = h2.delete(assoc_record.object_id) hash_keys.each do |hash_key| if objects = h[hash_key] objects.each{|object| object.associations[name].push(assoc_record)} end end end end end opts elsif type == :many_through_many opts = super el = opts[:eager_loader] model = self left_pk = opts[:left_primary_key] left_key = opts[:left_key] uses_lcks = opts[:uses_left_composite_keys] left_keys = Array(left_key) left_key_alias = opts[:left_key_alias] opts[:eager_loader] = lambda do |eo| return el.call(eo) unless model.identity_map h = eo[:id_map] eo[:rows].each{|object| object.associations[name] = []} ds = opts.associated_class opts.reverse_edges.each{|t| ds = ds.join(t[:table], Array(t[:left]).zip(Array(t[:right])), :table_alias=>t[:alias])} ft = opts.final_reverse_edge conds = uses_lcks ? [[left_keys.map{|k| SQL::QualifiedIdentifier.new(ft[:table], k)}, h.keys]] : [[left_key, h.keys]] # See above comment in many_to_many eager_loader ds = ds.join(ft[:table], Array(ft[:left]).zip(Array(ft[:right])) + conds, :table_alias=>ft[:alias]) pr = ds.row_proc h2 = {} ds.row_proc = proc do |hash| hash_key = if uses_lcks left_key_alias.map{|k| hash.delete(k)} else hash.delete(left_key_alias) end obj = pr.call(hash) (h2[obj.object_id] ||= []) << hash_key obj end model.eager_loading_dataset(opts, ds, Array(opts.select), eo[:associations], eo).all do |assoc_record| if hash_keys = h2.delete(assoc_record.object_id) hash_keys.each do |hash_key| if objects = h[hash_key] objects.each{|object| object.associations[name].push(assoc_record)} end end end end end opts else super end end
If the identity map is in use, check it for a current copy of the object. If a copy does not exist, create a new object and add it to the identity map. If a copy exists, add any values in the given row that aren't currently in the object to the object's values. This allows you to only request certain fields in an initial query, make modifications to some of those fields and request other, potentially overlapping fields in a new query, and not have the second query override fields you modified.
# File lib/sequel/plugins/identity_map.rb, line 172 def call(row) return super unless (idm = identity_map) && (pk = primary_key) if (k = identity_map_key(Array(pk).map{|x| row[x]})) && (o = idm[k]) o.merge_db_update(row) else o = super if (k = identity_map_key(o.pk)) idm[k] = o end end o end
Returns the current thread-local identity map. Should be a hash if there is an active identity map, and nil otherwise.
# File lib/sequel/plugins/identity_map.rb, line 154 def identity_map Thread.current[:sequel_identity_map] end
The identity map key for an object of the current class with the given pk. May not always be correct for a class which uses STI.
# File lib/sequel/plugins/identity_map.rb, line 160 def identity_map_key(pk) pk = Array(pk) "#{self}:#{pk.join(',')}" unless pk.compact.empty? end
Take a block and inside that block use an identity map to ensure a 1-1 correspondence of objects to the database row they represent.
# File lib/sequel/plugins/identity_map.rb, line 187 def with_identity_map return yield if identity_map begin self.identity_map = {} yield ensure self.identity_map = nil end end
Generated with the Darkfish Rdoc Generator 2.