Patchwork revset: prefetch all attributes before loop in _revsbetween

login
register
mail settings
Submitter Pierre-Yves David
Date June 13, 2015, 1:37 a.m.
Message ID <89b2644c6933cd28a83b.1434159443@marginatus.alto.octopoid.net>
Download mbox | patch
Permalink /patch/9624/
State Accepted
Headers show

Comments

Pierre-Yves David - June 13, 2015, 1:37 a.m.
# HG changeset patch
# User Pierre-Yves David <pierre-yves.david@fb.com>
# Date 1434048166 25200
#      Thu Jun 11 11:42:46 2015 -0700
# Node ID 89b2644c6933cd28a83baf800c2fd3ab3530f468
# Parent  4388714b846e21cb2fefb686220c7c0d2df1b85a
revset: prefetch all attributes before loop in _revsbetween

Python is slow at attributes lookup. No, really, I mean -slow-. prefetching
these three methods give use a measurable performance boost.

revset #0: 0::tip
   plain
0) 0.037655
1) 0.034290  91%
Matt Mackall - June 15, 2015, 5:36 p.m.
On Fri, 2015-06-12 at 18:37 -0700, Pierre-Yves David wrote:
> # HG changeset patch
> # User Pierre-Yves David <pierre-yves.david@fb.com>
> # Date 1434048166 25200
> #      Thu Jun 11 11:42:46 2015 -0700
> # Node ID 89b2644c6933cd28a83baf800c2fd3ab3530f468
> # Parent  4388714b846e21cb2fefb686220c7c0d2df1b85a
> revset: prefetch all attributes before loop in _revsbetween

Queued for default, thanks.

Patch

diff --git a/mercurial/revset.py b/mercurial/revset.py
--- a/mercurial/revset.py
+++ b/mercurial/revset.py
@@ -89,27 +89,31 @@  def _revsbetween(repo, roots, heads):
     seen = {}
     # XXX this should be 'parentset.min()' assuming 'parentset' is a smartset
     # (and if it is not, it should.)
     minroot = min(roots)
     roots = set(roots)
+    # prefetch all the things! (because python is slow)
+    reached = reachable.add
+    dovisit = visit.append
+    nextvisit = visit.pop
     # open-code the post-order traversal due to the tiny size of
     # sys.getrecursionlimit()
     while visit:
-        rev = visit.pop()
+        rev = nextvisit()
         if rev in roots:
-            reachable.add(rev)
+            reached(rev)
         parents = parentrevs(rev)
         seen[rev] = parents
         for parent in parents:
             if parent >= minroot and parent not in seen:
-                visit.append(parent)
+                dovisit(parent)
     if not reachable:
         return baseset()
     for rev in sorted(seen):
         for parent in seen[rev]:
             if parent in reachable:
-                reachable.add(rev)
+                reached(rev)
     return baseset(sorted(reachable))
 
 elements = {
     "(": (21, ("group", 1, ")"), ("func", 1, ")")),
     "##": (20, None, ("_concat", 20)),