Experimental parser: more inlining

tools/lexer_generator/code_generator.py

 # Copyright 2013 the V8 project authors. All rights reserved.
 # Redistribution and use in source and binary forms, with or without
 # modification, are permitted provided that the following conditions are
 # met:
 #
 #     * Redistributions of source code must retain the above copyright
 #       notice, this list of conditions and the following disclaimer.
 #     * Redistributions in binary form must reproduce the above
 #       copyright notice, this list of conditions and the following
 #       disclaimer in the documentation and/or other materials provided
@@ skipping 96 matching lines @@
     transitions = sorted(transitions, cmp)
     # map transition keys to disjoint ranges
     disjoint_keys = {'value' : []}
     def f((key, state)):
       ranges = list(key.range_iter())
       disjoint_keys['value'] += ranges
       return (ranges, state)
     transitions = map(f, transitions)
     disjoint_keys = sorted(disjoint_keys['value'], CodeGenerator.__range_cmp)
     # dictionary object representing state
+    (class_keys, distinct_keys, ranges) = (0, 0, 0)
+    for (t, r) in disjoint_keys:
+      if t == 'CLASS':
+        class_keys += 1
+      elif t == 'LATIN_1':
+        distinct_keys += r[1] - r[0] + 1
+        ranges += 1
+      else:
+        raise Exception()
     return {
       'node_number' : state.node_number(),
       'original_node_number' : state.node_number(),
       'transitions' : transitions,
       'switch_transitions' : [],
       'disjoint_keys' : disjoint_keys,
       'inline' : None,
       'depth' : None,
       'action' : action,
       'entry_action' : entry_action,
       'match_action' : match_action,
+      'class_keys' : class_keys,
+      'distinct_keys' : distinct_keys,
+      'ranges' : ranges
     }
 
   @staticmethod
   def __compute_depths(node_number, depth, id_map):
     state = id_map[node_number]
     if state['depth'] != None:
       return
     state['depth'] = depth
     for (k, transition_node) in state['transitions']:
       CodeGenerator.__compute_depths(transition_node, depth + 1, id_map)
 
-  @staticmethod
-  def __set_inline(count, state):
+  def __set_inline(self, count, state):
     assert state['inline'] == None
     inline = False
+    # inline terminal states
     if not state['transitions']:
       inline = True
-    # TODO add in 1 or 2 element if blocks
+    # inline next to terminal states with 1 or 2 transitions
+    elif state['distinct_keys'] < 3 and state['class_keys'] == 0:
+      inline = True
+      # ensure state terminates in 1 step
+      for key, state_id in state['transitions']:
+        if self.__dfa_states[state_id]['transitions']:
+          inline = False
+          break
     state['inline'] = inline
     return count + 1 if inline else count
 
   @staticmethod
   def __split_transitions(split_count, state):
     '''Goes through the transitions for 'state' and decides which of them should
     use the if statement and which should use the switch statement.'''
     assert not state['switch_transitions']
-    (class_keys, distinct_keys, ranges) = (0, 0, 0)
-    for (t, r) in state['disjoint_keys']:
-      if t == 'CLASS':
-        class_keys += 1
-      elif t == 'LATIN_1':
-        distinct_keys += r[1] - r[0] + 1
-        ranges += 1
-      else:
-        raise Exception()
+    (distinct_keys, ranges) = (state['distinct_keys'], state['ranges'])
     if distinct_keys <= 7 or float(distinct_keys)/float(ranges) >= 7.0:
       return split_count
     switch_transitions = []
     if_transitions = []
     for (ranges, node_id) in state['transitions']:
       i = []
       s = []
       for r in ranges:
         if r[0] == 'CLASS':
           i.append(r)
@@ skipping 16 matching lines @@
     dfa_states = sorted(dfa_states, cmp=self.__state_cmp)
     # remap all node numbers
     for i, state in enumerate(dfa_states):
       state['node_number'] = i
     def f((key, original_node_number)):
       return (key, id_map[original_node_number]['node_number'])
     for state in dfa_states:
       state['transitions'] = map(f, state['transitions'])
     assert id_map[self.__start_node_number]['node_number'] == 0
     assert len(dfa_states) == self.__dfa.node_count()
+    # store states
+    self.__dfa_states = dfa_states
     # set nodes to inline
     if self.__inline:
-      inlined = reduce(CodeGenerator.__set_inline, dfa_states, 0)
+      inlined = reduce(self.__set_inline, dfa_states, 0)
       if self.__log:
         print "%s states inlined" % inlined
     elif self.__log:
       print "no inlining"
     # split transitions
     if self.__switching:
       switched = reduce(CodeGenerator.__split_transitions, dfa_states, 0)
       if self.__log:
         print "%s states use switch (instead of if)" % switched
     elif self.__log:
       print "no switching"
-    # store states
-    self.__dfa_states = dfa_states
 
   def process(self):
 
     self.__canonicalize_traversal()
 
     dfa_states = self.__dfa_states
 
     default_action = self.__default_action
     assert(default_action and default_action.match_action())
     default_action = default_action.match_action()
 
     sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
     template_env = jinja2.Environment(
       loader = jinja2.PackageLoader('lexer_generator', '.'),
       undefined = jinja2.StrictUndefined)
     template = template_env.get_template('code_generator.jinja')
 
     return template.render(
       start_node_number = 0,
       debug_print = self.__debug_print,
       default_action = default_action,
       dfa_states = dfa_states)