Building Python Document 30% faster.

Index: search.py
===================================================================
--- search.py	(リビジョン 71045)
+++ search.py	(作業コピー)
@@ -14,8 +14,14 @@
 
 from docutils.nodes import Text, NodeVisitor
 
-from sphinx.util.stemmer import PorterStemmer
 from sphinx.util import jsdump, rpartition
+try:
+    # PyStemmer is wrapper for stemmer in c
+    import Stemmer as PyStemmer
+    PYSTEMMER = True
+except ImportError:
+    from sphinx.util.stemmer import PorterStemmer
+    PYSTEMMER = False
 
 
 word_re = re.compile(r'\w+(?u)')
@@ -62,17 +68,31 @@
 js_index = _JavaScriptIndex()
 
 
-class Stemmer(PorterStemmer):
-    """
-    All those porter stemmer implementations look hideous.
-    make at least the stem method nicer.
-    """
+if PYSTEMMER:
+    class Stemmer(object):
 
-    def stem(self, word):
-        word = word.lower()
-        return PorterStemmer.stem(self, word, 0, len(word) - 1)
+        def __init__(self):
+            self._stemmer = PyStemmer.Stemmer('english')
 
+        def stem(self, word):
+            return self._stemmer.stemWord(word.lower())
 
+        def stemWords(self, iter):
+            import itertools
+            return self._stemmer.stemWords(itertools.imap(lambda x: x.lower(), iter))
+else:
+    class Stemmer(PorterStemmer):
+        """
+        All those porter stemmer implementations look hideous.
+        make at least the stem method nicer.
+        """
+
+        def stem(self, word):
+            word = word.lower()
+            return PorterStemmer.stem(self, word, 0, len(word) - 1)
+
+
+
 class WordCollector(NodeVisitor):
     """
     A special visitor that collects words for the `IndexBuilder`.
@@ -196,11 +216,11 @@
         visitor = WordCollector(doctree)
         doctree.walk(visitor)
 
-        def add_term(word, prefix='', stem=self._stemmer.stem):
+        def add_term(word, stem=self._stemmer.stem):
             word = stem(word)
             if len(word) < 3 or word in stopwords or word.isdigit():
                 return
-            self._mapping.setdefault(prefix + word, set()).add(filename)
+            self._mapping.setdefault(word, set()).add(filename)
 
         for word in word_re.findall(title):
             add_term(word)
Index: builders/html.py
===================================================================
--- builders/html.py	(リビジョン 71045)
+++ builders/html.py	(作業コピー)
@@ -23,7 +23,7 @@
 
 from docutils import nodes
 from docutils.io import DocTreeInput, StringOutput
-from docutils.core import publish_parts
+from docutils.core import Publisher, publish_parts
 from docutils.utils import new_document
 from docutils.frontend import OptionParser
 from docutils.readers.doctree import Reader as DoctreeReader
@@ -181,14 +181,26 @@
         """Utility: Render a lone doctree node."""
         doc = new_document('<partial node>')
         doc.append(node)
-        return publish_parts(
-            doc,
-            source_class=DocTreeInput,
-            reader=DoctreeReader(),
-            writer=HTMLWriter(self),
-            settings_overrides={'output_encoding': 'unicode'}
-        )
 
+        # cache publisher object.
+        if 'publisher' not in self.__dict__:
+            self.publisher = Publisher(
+                    source_class = DocTreeInput,
+                    destination_class=StringOutput)
+            self.publisher.set_components('standalone',
+                    'restructuredtext', 'pseudoxml')
+
+        pub = self.publisher
+
+        pub.reader = DoctreeReader()
+        pub.writer = HTMLWriter(self)
+        pub.process_programmatic_settings(
+                None, {'output_encoding': 'unicode'}, None)
+        pub.set_source(doc, None)
+        pub.set_destination(None, None)
+        pub.publish()
+        return pub.writer.parts
+        
     def prepare_writing(self, docnames):
         from sphinx.search import IndexBuilder

Index: parsers/rst/states.py
===================================================================
--- parsers/rst/states.py	(リビジョン 71045)
+++ parsers/rst/states.py	(作業コピー)
@@ -207,6 +207,7 @@
     """
 
     nested_sm = NestedStateMachine
+    nested_sm_cache = []
 
     def __init__(self, state_machine, debug=0):
         self.nested_sm_kwargs = {'state_classes': state_classes,
@@ -255,21 +256,33 @@
         Create a new StateMachine rooted at `node` and run it over the input
         `block`.
         """
+        use_default = 0
         if state_machine_class is None:
             state_machine_class = self.nested_sm
+            use_default += 1
         if state_machine_kwargs is None:
             state_machine_kwargs = self.nested_sm_kwargs
+            use_default += 1
         block_length = len(block)
-        state_machine = state_machine_class(debug=self.debug,
+
+        if use_default == 2 and self.nested_sm_cache: #use default sm cache
+            state_machine = self.nested_sm_cache.pop()
+        else:
+            state_machine = state_machine_class(debug=self.debug,
                                             **state_machine_kwargs)
         state_machine.run(block, input_offset, memo=self.memo,
                           node=node, match_titles=match_titles)
-        state_machine.unlink()
+        if use_default == 2:
+            self.nested_sm_cache.append(state_machine)
+        else:
+            state_machine.unlink()
+
         new_offset = state_machine.abs_line_offset()
         # No `block.parent` implies disconnected -- lines aren't in sync:
         if block.parent and (len(block) - block_length) != 0:
             # Adjustment for block if modified in nested parse:
             self.state_machine.next_line(len(block) - block_length)
+
         return new_offset
 
     def nested_list_parse(self, block, input_offset, node, initial_state,