module ReactiveRecord # the point is to collect up a all records needed, with whatever attributes were required + primary key, and inheritance column # or get all scope arrays, with the record ids # the incoming vector includes the terminal method # output is a hash tree of the form # tree ::= {method => tree | [value]} | method's value is either a nested tree or a single value which is wrapped in array # {:id => primary_key_id_value} | if its the id method we leave the array off because we know it must be an int # {integer => tree} for collections, each item retrieved will be represented by its id # # example # { # "User" => { # ["find", 12] => { # :id => 12 # "email" => ["mitch@catprint.com"] # "todos" => { # "active" => { # 123 => # { # id: 123, # title: ["get fetch_records_from_db done"] # }, # 119 => # { # id: 119 # title: ["go for a swim"] # } # ] # } # } # } # } # } # } # To build this tree we first fill values for each individual vector, saving all the intermediate data # when all these are built we build the above hash structure # basic # [Todo, [find, 123], title] # -> [[Todo, [find, 123], title], "get fetch_records_from_db done", 123] # [User, [find_by_email, "mitch@catprint.com"], first_name] # -> [[User, [find_by_email, "mitch@catprint.com"], first_name], "Mitch", 12] # misses # [User, [find_by_email, "foobar@catprint.com"], first_name] # nothing is found so nothing is downloaded # prerendering may do this # [User, [find_by_email, "foobar@catprint.com"]] # which will return a cache object whose id is nil, and value is nil # scoped collection # [User, [find, 12], todos, active, *, title] # -> [[User, [find, 12], todos, active, *, title], "get fetch_records_from_db done", 12, 123] # -> [[User, [find, 12], todos, active, *, title], "go for a swim", 12, 119] # collection with nested belongs_to # [User, [find, 12], todos, *, team] # -> [[User, [find, 12], todos, *, team, name], "developers", 12, 123, 252] # [[User, [find, 12], todos, *, team, name], nil, 12, 119] <- no team defined for todo<119> so list ends early # collections that are empty will deliver nothing # [User, [find, 13], todos, *, team, name] # no todos for user 13 # evaluation will get this far: [[User, [find, 13], todos], nil, 13] # nothing will match [User, [find, 13], todos, team, name] so nothing will be downloaded # aggregate # [User, [find, 12], address, zip_code] # -> [[User, [find, 12], address, zip_code]], "14622", 12] <- note parent id is returned # aggregate with a belongs_to # [User, [find, 12], address, country, country_code] # -> [[User, [find, 12], address, country, country_code], "US", 12, 342] # collection * (for iterators etc) # [User, [find, 12], todos, overdue, *all] # -> [[User, [find, 12], todos, active, *all], [119, 123], 12] # [Todo, [find, 119], owner, todos, active, *all] # -> [[Todo, [find, 119], owner, todos, active, *all], [119, 123], 119, 12] class ServerDataCache # SECURITY - SAFE def initialize(acting_user, preloaded_records) @acting_user = acting_user @cache = [] @requested_cache_items = [] @preloaded_records = preloaded_records end if RUBY_ENGINE != 'opal' def self.get_model(str) # We don't want to open a security hole by allowing some client side string to # autoload a class, which would happen if we did a simple str.constantize. # # Because all AR models are loaded at boot time on the server to define the # ActiveRecord::Base.public_columns_hash method any model which the client has # access to should already be loaded. # # If str is not already loaded then we have an access violation. unless const_defined? str Hyperloop::InternalPolicy.raise_operation_access_violation end str.constantize end # SECURITY - NOW SAFE def [](*vector) root = CacheItem.new(@cache, @acting_user, vector[0], @preloaded_records) vector[1..-1].inject(root) { |cache_item, method| cache_item.apply_method method if cache_item } vector[0] = ServerDataCache.get_model(vector[0]) last_value = nil @cache.each do |cache_item| next if cache_item.root != root || @requested_cache_items.include?(cache_item) @requested_cache_items << cache_item last_value = cache_item end last_value end # SECURITY - SAFE def self.[](models, associations, vectors, acting_user) ActiveRecord::Base.public_columns_hash result = nil ActiveRecord::Base.transaction do cache = new(acting_user, ReactiveRecord::Base.save_records(models, associations, acting_user, false, false)) vectors.each { |vector| cache[*vector] } result = cache.as_json raise ActiveRecord::Rollback, "This Rollback is intentional!" end result end # SECURITY - SAFE def clear_requests @requested_cache_items = [] end # SECURITY - SAFE def as_json @requested_cache_items.inject({}) do |hash, cache_item| hash.deep_merge! cache_item.as_hash end end # SECURITY - SAFE def select(&block); @cache.select(&block); end # SECURITY - SAFE def detect(&block); @cache.detect(&block); end # SECURITY - SAFE def inject(initial, &block); @cache.inject(initial) █ end class CacheItem attr_reader :vector attr_reader :absolute_vector attr_reader :root attr_reader :acting_user # SECURITY - SAFE def value @value # which is a ActiveRecord object end # SECURITY - SAFE def method @vector.last end # SECURITY - NOW SAFE def self.new(db_cache, acting_user, klass, preloaded_records) klass_constant = ServerDataCache.get_model(klass) if existing = db_cache.detect { |cached_item| cached_item.vector == [klass_constant] } return existing end super end # SECURITY - NOW SAFE def initialize(db_cache, acting_user, klass, preloaded_records) klass = ServerDataCache.get_model(klass) @db_cache = db_cache @acting_user = acting_user @vector = @absolute_vector = [klass] @value = klass @parent = nil @root = self @preloaded_records = preloaded_records db_cache << self end # SECURITY - UNSAFE def apply_method_to_cache(method) @db_cache.inject(nil) do |representative, cache_item| if cache_item.vector == vector # TODO: Security - this is the wrong check in the wrong place... if @value.class < ActiveRecord::Base and @value.attributes.has_key?(method) # TODO: second check is not needed, its built into check_permmissions, check should be does class respond to check_permissions... @value.check_permission_with_acting_user(@acting_user, :view_permitted?, method) end if method == "*" cache_item.apply_star || representative elsif method == "*all" cache_item.build_new_cache_item(cache_item.value.collect { |record| record.id }, method, method) elsif method == "*count" cache_item.build_new_cache_item(cache_item.value.count, method, method) elsif preloaded_value = @preloaded_records[cache_item.absolute_vector + [method]] cache_item.build_new_cache_item(preloaded_value, method, method) elsif aggregation = cache_item.aggregation?(method) cache_item.build_new_cache_item(aggregation.mapping.collect { |attribute, accessor| cache_item.value[attribute] }, method, method) else if !cache_item.value || cache_item.value.class == Array representative else # TODO: Security. Protect the send(*method). But its complicated.. method can be an attribute, scope, relationship or actual method. # Each needs some protection logic. begin cache_item.build_new_cache_item(cache_item.value.send(*method), method, method) rescue Exception => e # ReactiveRecord::Pry::rescued(e) ::Rails.logger.debug "\033[0;31;1mERROR: HyperModel exception caught when applying #{method} to db object #{cache_item.value}: #{e}\033[0;30;21m" raise e, "HyperModel fetching records failed, exception caught when applying #{method} to db object #{cache_item.value}: #{e}", e.backtrace end end end else representative end end end # SECURITY - SAFE def aggregation?(method) if method.is_a?(String) && @value.class.respond_to?(:reflect_on_aggregation) aggregation = @value.class.reflect_on_aggregation(method.to_sym) if aggregation && !(aggregation.klass < ActiveRecord::Base) && @value.send(method) aggregation end end end # SECURITY - SAFE def apply_star if @value && @value.length > 0 i = -1 @value.inject(nil) do |representative, current_value| i += 1 if preloaded_value = @preloaded_records[@absolute_vector + ["*#{i}"]] build_new_cache_item(preloaded_value, "*", "*#{i}") else build_new_cache_item(current_value, "*", "*#{i}") end end else build_new_cache_item([], "*", "*") end end # SECURITY - SAFE # TODO replace instance_eval with a method like clone_new_child(....) def build_new_cache_item(new_value, method, absolute_method) new_parent = self self.clone.instance_eval do @vector = @vector + [method] # don't push it on since you need a new vector! @absolute_vector = @absolute_vector + [absolute_method] @value = new_value @db_cache << self @parent = new_parent @root = new_parent.root self end end # SECURITY - SAFE def apply_method(method) if method.is_a? Array and method.first == "find_by_id" method[0] = "find" elsif method.is_a? String and method =~ /^\*[0-9]+$/ method = "*" end new_vector = vector + [method] @db_cache.detect { |cached_item| cached_item.vector == new_vector} || apply_method_to_cache(method) end # SECURITY - SAFE def jsonize(method) # sadly standard json converts {[:foo, nil] => 123} to {"['foo', nil]": 123} # luckily [:foo, nil] does convert correctly # so we check the methods and force proper conversion method.is_a?(Array) ? method.to_json : method end def as_hash(children = nil) unless children return {} if @value.is_a?(Class) and (@value < ActiveRecord::Base) children = [@value.is_a?(BigDecimal) ? @value.to_f : @value] end if @parent if method == "*" if @value.is_a? Array # this happens when a scope is empty there is test case, but @parent.as_hash({}) # does it work for all edge cases? else @parent.as_hash({@value.id => children}) end elsif @value.class < ActiveRecord::Base and children.is_a? Hash @parent.as_hash({jsonize(method) => children.merge({ :id => (method.is_a?(Array) && method.first == "new") ? [nil] : [@value.id], @value.class.inheritance_column => [@value[@value.class.inheritance_column]], })}) elsif method == "*all" @parent.as_hash({jsonize(method) => children.first}) else @parent.as_hash({jsonize(method) => children}) end else {method.name => children} end end def to_json @value.to_json end end end =begin tree is a hash, target is the object that will be filled in with the data hanging off the key. first time around target == nil, so for each key, value pair we do this: load_from_json(value, Object.const_get(JSON.parse(key))) keys: ':*all': target.replace tree["*all"].collect { |id| target.proxy_association.klass.find(id) } Example: {'*all': [1, 7, 19, 23]} target is a collection and will now have 4 records: 1, 7, 19, 23 'id': if value is an array then target.id = value.first Example: {'id': [17]} Example the target is a record, and its id is now set to 17 '*count': target.set_count_state(value.first) note: set_count_state sets the count of a collection and updates the associated state variable integer-like-string-or-number: target.push_and_update_belongs_to(key) note: collection will be a has_many association, so we are doing a target << find(key), and updating both ends of the relationship [:new, nnn] do a ReactiveRecord::Base.find_by_object_id(target.base_class, method[1]) and that becomes the new target, with val being passed allow_change [...] and current target is NOT an ActiveRecord Model (??? a collection ???) then send key to target, and that becomes new target but note if value is an array then the scope returned nil, so we destroy the bogus record, and set new target back to nil new_target.destroy and new_target = nil if value.is_a? Array [...] and current target IS AN ActiveRecord Model (not a collection) then target.backing_record.update_attribute([method], target.backing_record.convert(method, value.first)) aggregation: target.class.respond_to?(:reflect_on_aggregation) and aggregation = target.class.reflect_on_aggregation(method) and !(aggregation.klass < ActiveRecord::Base) target.send "#{method}=", aggregation.deserialize(value.first) other-string-method-name: if value is a an array then value.first is the new value and we do target.send "{key}=", value.first if value is a hash =end def self.load_from_json(tree, target = nil) ignore_all = nil # have to process *all before any other items # we leave the "*all" key in just for debugging purposes, and then skip it below if sorted_collection = tree["*all"] target.replace sorted_collection.collect { |id| target.proxy_association.klass.find(id) } end if id_value = tree["id"] and id_value.is_a? Array target.id = id_value.first end tree.each do |method, value| method = JSON.parse(method) rescue method new_target = nil if method == "*all" next # its already been processed above elsif !target load_from_json(value, Object.const_get(method)) elsif method == "*count" target.set_count_state(value.first) elsif method.is_a? Integer or method =~ /^[0-9]+$/ new_target = target.push_and_update_belongs_to(method) #target << (new_target = target.proxy_association.klass.find(method)) elsif method.is_a? Array if method[0] == "new" new_target = ReactiveRecord::Base.find_by_object_id(target.base_class, method[1]) elsif !(target.class < ActiveRecord::Base) new_target = target.send(*method) # value is an array if scope returns nil, so we destroy the bogus record new_target.destroy and new_target = nil if value.is_a? Array else target.backing_record.update_attribute([method], target.backing_record.convert(method, value.first)) end elsif target.class.respond_to?(:reflect_on_aggregation) and aggregation = target.class.reflect_on_aggregation(method) and !(aggregation.klass < ActiveRecord::Base) target.send "#{method}=", aggregation.deserialize(value.first) elsif value.is_a? Array target.send "#{method}=", value.first unless method == "id" # we handle ids first so things sync nicely elsif value.is_a? Hash and value[:id] and value[:id].first and association = target.class.reflect_on_association(method) # not sure if its necessary to check the id above... is it possible to for the method to be an association but not have an id? new_target = association.klass.find(value[:id].first) target.send "#{method}=", new_target elsif !(target.class < ActiveRecord::Base) new_target = target.send(*method) # value is an array if scope returns nil, so we destroy the bogus record new_target.destroy and new_target = nil if value.is_a? Array else new_target = target.send("#{method}=", target.send(method)) end load_from_json(value, new_target) if new_target end end end end