Rev 4294: Fairly significant savings... avoid looking at self._last_recently_used. in lp:///~jameinel/bzr/1.15-lru-gc

[John Arbash Meinel]

At lp:///~jameinel/bzr/1.15-lru-gc

------------------------------------------------------------
revno: 4294
revision-id: john at arbash-meinel.com-20090416204541-gg2qtfggaxwzudvz
parent: john at arbash-meinel.com-20090416203230-ltw57cs1yp0yp5q9
committer: John Arbash Meinel <john at arbash-meinel.com>
branch nick: 1.15-lru-gc
timestamp: Thu 2009-04-16 15:45:41 -0500
message:
  Fairly significant savings... avoid looking at self._last_recently_used.
  We can get the same information from node.next_key, which is a value we need anyway.
  Somewhat surprisingly, this drops us from 7.6s => 7.1s on 2.8M lookups.
-------------- next part --------------
=== modified file 'bzrlib/lru_cache.py'
--- a/bzrlib/lru_cache.py	2009-04-16 20:32:30 +0000
+++ b/bzrlib/lru_cache.py	2009-04-16 20:45:41 +0000
@@ -88,15 +88,15 @@
             return node.value
         # Remove this node from the old location
         node_prev = node.prev
-        if node is self._last_recently_used:
-            self._last_recently_used = node_prev
         next_key = node.next_key
-        node_prev.next_key = next_key
         if next_key is None:
-            node_next = None
+            # 'node' is the _last_recently_used, because it doesn't have a
+            # 'next' item. So move the current lru to the previous node.
+            self._last_recently_used = node_prev
         else:
             node_next = cache[next_key]
             node_next.prev = node_prev
+        node_prev.next_key = next_key
         # Insert this node at the front of the list
         node.next_key = mru.key
         mru.prev = node