module ActiveRecord module AttributeMethods module Read extend ActiveSupport::Concern ATTRIBUTE_TYPES_CACHED_BY_DEFAULT = [:datetime, :timestamp, :time, :date] included do class_attribute :attribute_types_cached_by_default, instance_writer: false self.attribute_types_cached_by_default = ATTRIBUTE_TYPES_CACHED_BY_DEFAULT end module ClassMethods # +cache_attributes+ allows you to declare which converted attribute # values should be cached. Usually caching only pays off for attributes # with expensive conversion methods, like time related columns (e.g. # +created_at+, +updated_at+). def cache_attributes(*attribute_names) cached_attributes.merge attribute_names.map { |attr| attr.to_s } end # Returns the attributes which are cached. By default time related columns # with datatype :datetime, :timestamp, :time, :date are cached. def cached_attributes @cached_attributes ||= columns.select { |c| cacheable_column?(c) }.map { |col| col.name }.to_set end # Returns +true+ if the provided attribute is being cached. def cache_attribute?(attr_name) cached_attributes.include?(attr_name) end protected # We want to generate the methods via module_eval rather than # define_method, because define_method is slower on dispatch. # Evaluating many similar methods may use more memory as the instruction # sequences are duplicated and cached (in MRI). define_method may # be slower on dispatch, but if you're careful about the closure # created, then define_method will consume much less memory. # # But sometimes the database might return columns with # characters that are not allowed in normal method names (like # 'my_column(omg)'. So to work around this we first define with # the __temp__ identifier, and then use alias method to rename # it to what we want. # # We are also defining a constant to hold the frozen string of # the attribute name. Using a constant means that we do not have # to allocate an object on each call to the attribute method. # Making it frozen means that it doesn't get duped when used to # key the @attributes_cache in read_attribute. def define_method_attribute(name) safe_name = name.unpack('h*').first temp_method = "__temp__#{safe_name}" ActiveRecord::AttributeMethods::AttrNames.set_name_cache safe_name, name generated_attribute_methods.module_eval <<-STR, __FILE__, __LINE__ + 1 def #{temp_method} name = ::ActiveRecord::AttributeMethods::AttrNames::ATTR_#{safe_name} read_attribute(name) { |n| missing_attribute(n, caller) } end STR generated_attribute_methods.module_eval do alias_method name, temp_method undef_method temp_method end end private def cacheable_column?(column) if attribute_types_cached_by_default == ATTRIBUTE_TYPES_CACHED_BY_DEFAULT ! serialized_attributes.include? column.name else attribute_types_cached_by_default.include?(column.type) end end end # Returns the value of the attribute identified by attr_name after # it has been typecast (for example, "2004-12-12" in a data column is cast # to a date object, like Date.new(2004, 12, 12)). def read_attribute(attr_name) # If it's cached, just return it # We use #[] first as a perf optimization for non-nil values. See https://gist.github.com/jonleighton/3552829. name = attr_name.to_s @attributes_cache[name] || @attributes_cache.fetch(name) { column = @columns_hash.fetch(name) { return @attributes.fetch(name) { if name == 'id' && self.class.primary_key != name read_attribute(self.class.primary_key) end } } value = @attributes.fetch(name) { return block_given? ? yield(name) : nil } if self.class.cache_attribute?(name) @attributes_cache[name] = column.type_cast(value) else column.type_cast value end } end private def attribute(attribute_name) read_attribute(attribute_name) end end end end