module Vool # Sending in a dynamic language is off course not as simple as just calling. # The function that needs to be called depends after all on the receiver, # and no guarantees can be made on what that is. # # It helps to know that usually (>99%) the class of the receiver does not change. # Our stategy then is to cache the functions and only dynamically determine it in # case of a miss (the 1%, and first invocation) # # As cache key we must use the type of the object (which is the first word of _every_ object) # as that is constant, and function implementations depend on the type (not class) class SendStatement < CallStatement attr_reader :block def block return nil if arguments.empty? bl = arguments.last bl.is_a?(BlockStatement) ? bl : nil end def add_block( block ) @arguments << block end def each(&block) super self.block.each(&block) if self.block end # lazy init this, to keep the dependency (which goes to parfait and booting) at bay def dynamic_call @dynamic ||= Mom::DynamicCall.new() end # A Send breaks down to 2 steps: # - Setting up the next message, with receiver, arguments, and (importantly) return address # - a CachedCall , or a SimpleCall, depending on wether the receiver type can be determined def to_mom( compiler ) @receiver = SelfExpression.new(compiler.receiver_type) if @receiver.is_a?(SelfExpression) if(@receiver.ct_type) method = @receiver.ct_type.resolve_method(self.name) return simple_call(compiler, method) if method end cached_call(compiler) end def message_setup(compiler,called_method) setup = Mom::MessageSetup.new( called_method ) mom_receive = @receiver.slot_definition(compiler) arg_target = [:message , :next_message , :arguments] args = [] @arguments.each_with_index do |arg , index| # +1 because of type args << Mom::SlotLoad.new(self, arg_target + [index + 1] , arg.slot_definition(compiler)) end setup << Mom::ArgumentTransfer.new(self, mom_receive , args ) end def simple_call(compiler, called_method) message_setup(compiler,called_method) << Mom::SimpleCall.new(called_method) end # this breaks cleanly into two parts: # - check the cached type and if neccessary update # - call the cached method def cached_call(compiler) cache_check(compiler) << call_cached_method(compiler) end # check that current type is the cached type # if not, change and find method for the type (simple_call to resolve_method) # conceptually easy in ruby, but we have to compile that "easy" ruby def cache_check(compiler) ok = Mom::Label.new(self,"cache_ok_#{self.object_id}") check = build_condition(ok, compiler) # if cached_type != current_type check << Mom::SlotLoad.new(self,[dynamic_call.cache_entry, :cached_type] , receiver_type_definition(compiler)) check << Mom::ResolveMethod.new(self, @name , dynamic_call.cache_entry ) check << ok end # to call the method (that we know now to be in the cache), we move the method # to reg1, do the setup (very similar to static) and call def call_cached_method(compiler) message_setup(compiler,dynamic_call.cache_entry) << dynamic_call end def to_s(depth = 0) sen = "#{receiver}.#{name}(#{@arguments.collect{|a| a.to_s}.join(', ')})" at_depth(depth , sen) end private def receiver_type_definition(compiler) defi = @receiver.slot_definition(compiler) defi.slots << :type defi end def build_condition(ok_label, compiler) cached_type = Mom::SlotDefinition.new(dynamic_call.cache_entry , [:cached_type]) current_type = receiver_type_definition(compiler) Mom::NotSameCheck.new(cached_type , current_type, ok_label) end end end