changeset 46872:8bca353b1ebc

match: convert O(n) to O(log n) in exactmatcher.visitchildrenset When using narrow, during rebase this is called (at least) once per directory in the set of files in the commit being rebased. Every time it's called, we did the set arithmetic (now extracted and cached), which was probably pretty cheap but not necessary to repeat each time, looped over every item in the matcher and kept things that started with the directory we were querying. With very large narrowspecs, and a commit that touched a file in a large number of directories, this was slow. In a pathological repo, the rebase of a single commit (that touched over 17k files, I believe in approximately as many directories) with a narrowspec that had >32k entries took 8,246s of profiled time, with 5,007s of that spent in visitchildrenset (transitively). With this change, the time spent in visitchildrenset is less than 34s (which is where my profile cut off). Most of the remaining time was network access due to our custom remotefilelog-based setup not properly prefetching. Differential Revision: https://phab.mercurial-scm.org/D10294
author Kyle Lippincott <spectral@google.com>
date Wed, 31 Mar 2021 12:46:54 -0700
parents 887f89b100ac
children 0abf5eba0042
files mercurial/match.py
diffstat 1 files changed, 28 insertions(+), 3 deletions(-) [+]
line wrap: on
line diff
--- a/mercurial/match.py	Tue Mar 30 13:05:22 2021 -0700
+++ b/mercurial/match.py	Wed Mar 31 12:46:54 2021 -0700
@@ -7,6 +7,7 @@
 
 from __future__ import absolute_import, print_function
 
+import bisect
 import copy
 import itertools
 import os
@@ -798,14 +799,38 @@
     def visitdir(self, dir):
         return dir in self._dirs
 
+    @propertycache
+    def _visitchildrenset_candidates(self):
+        """A memoized set of candidates for visitchildrenset."""
+        return self._fileset | self._dirs - {b''}
+
+    @propertycache
+    def _sorted_visitchildrenset_candidates(self):
+        """A memoized sorted list of candidates for visitchildrenset."""
+        return sorted(self._visitchildrenset_candidates)
+
     def visitchildrenset(self, dir):
         if not self._fileset or dir not in self._dirs:
             return set()
 
-        candidates = self._fileset | self._dirs - {b''}
-        if dir != b'':
+        if dir == b'':
+            candidates = self._visitchildrenset_candidates
+        else:
+            candidates = self._sorted_visitchildrenset_candidates
             d = dir + b'/'
-            candidates = {c[len(d) :] for c in candidates if c.startswith(d)}
+            # Use bisect to find the first element potentially starting with d
+            # (i.e. >= d). This should always find at least one element (we'll
+            # assert later if this is not the case).
+            first = bisect.bisect_left(candidates, d)
+            # We need a representation of the first element that is > d that
+            # does not start with d, so since we added a `/` on the end of dir,
+            # we'll add whatever comes after slash (we could probably assume
+            # that `0` is after `/`, but let's not) to the end of dir instead.
+            dnext = dir + encoding.strtolocal(chr(ord(b'/') + 1))
+            # Use bisect to find the first element >= d_next
+            last = bisect.bisect_left(candidates, dnext, lo=first)
+            dlen = len(d)
+            candidates = {c[dlen:] for c in candidates[first:last]}
         # self._dirs includes all of the directories, recursively, so if
         # we're attempting to match foo/bar/baz.txt, it'll have '', 'foo',
         # 'foo/bar' in it. Thus we can safely ignore a candidate that has a