revlog: use an LRU cache for delta chain bases
Profiling using statprof revealed a hotspot during changegroup
application calculating delta chain bases on generaldelta repos.
Essentially, revlog._addrevision() was performing a lot of redundant
work tracing the delta chain as part of determining when the chain
distance was acceptable. This was most pronounced when adding
revisions to manifests, which can have delta chains thousands of
revisions long.
There was a delta chain base cache on revlogs before, but it only
captured a single revision. This was acceptable before generaldelta,
when _addrevision would build deltas from the previous revision and
thus we'd pretty much guarantee a cache hit when resolving the delta
chain base on a subsequent _addrevision call. However, it isn't
suitable for generaldelta because parent revisions aren't necessarily
the last processed revision.
This patch converts the delta chain base cache to an LRU dict cache.
The cache can hold multiple entries, so generaldelta repos have a
higher chance of getting a cache hit.
The impact of this change when processing changegroup additions is
significant. On a generaldelta conversion of the "mozilla-unified"
repo (which contains heads of the main Firefox repositories in
chronological order - this means there are lots of transitions between
heads in revlog order), this change has the following impact when
performing an `hg unbundle` of an uncompressed bundle of the repo:
before: 5:42 CPU time
after: 4:34 CPU time
Most of this time is saved when applying the changelog and manifest
revlogs:
before: 2:30 CPU time
after: 1:17 CPU time
That nearly a 50% reduction in CPU time applying changesets and
manifests!
Applying a gzipped bundle of the same repo (effectively simulating a
`hg clone` over HTTP) showed a similar speedup:
before: 5:53 CPU time
after: 4:46 CPU time
Wall time improvements were basically the same as CPU time.
I didn't measure explicitly, but it feels like most of the time
is saved when processing manifests. This makes sense, as large
manifests tend to have very long delta chains and thus benefit the
most from this cache.
So, this change effectively makes changegroup application (which is
used by `hg unbundle`, `hg clone`, `hg pull`, `hg unshelve`, and
various other commands) significantly faster when delta chains are
long (which can happen on repos with large numbers of files and thus
large manifests).
In theory, this change can result in more memory utilization. However,
we're caching a dict of ints. At most we have 200 ints + Python object
overhead per revlog. And, the cache is really only populated when
performing read-heavy operations, such as adding changegroups or
scanning an individual revlog. For memory bloat to be an issue, we'd
need to scan/read several revisions from several revlogs all while
having active references to several revlogs. I don't think there are
many operations that do this, so I don't think memory bloat from the
cache will be an issue.
# windows.py - Windows utility function implementations for Mercurial
#
# Copyright 2005-2009 Matt Mackall <mpm@selenic.com> and others
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
from __future__ import absolute_import
import errno
import msvcrt
import os
import re
import stat
import sys
from .i18n import _
from . import (
encoding,
osutil,
win32,
)
try:
import _winreg as winreg
winreg.CloseKey
except ImportError:
import winreg
executablepath = win32.executablepath
getuser = win32.getuser
hidewindow = win32.hidewindow
makedir = win32.makedir
nlinks = win32.nlinks
oslink = win32.oslink
samedevice = win32.samedevice
samefile = win32.samefile
setsignalhandler = win32.setsignalhandler
spawndetached = win32.spawndetached
split = os.path.split
termwidth = win32.termwidth
testpid = win32.testpid
unlink = win32.unlink
umask = 0o022
class mixedfilemodewrapper(object):
"""Wraps a file handle when it is opened in read/write mode.
fopen() and fdopen() on Windows have a specific-to-Windows requirement
that files opened with mode r+, w+, or a+ make a call to a file positioning
function when switching between reads and writes. Without this extra call,
Python will raise a not very intuitive "IOError: [Errno 0] Error."
This class wraps posixfile instances when the file is opened in read/write
mode and automatically adds checks or inserts appropriate file positioning
calls when necessary.
"""
OPNONE = 0
OPREAD = 1
OPWRITE = 2
def __init__(self, fp):
object.__setattr__(self, '_fp', fp)
object.__setattr__(self, '_lastop', 0)
def __getattr__(self, name):
return getattr(self._fp, name)
def __setattr__(self, name, value):
return self._fp.__setattr__(name, value)
def _noopseek(self):
self._fp.seek(0, os.SEEK_CUR)
def seek(self, *args, **kwargs):
object.__setattr__(self, '_lastop', self.OPNONE)
return self._fp.seek(*args, **kwargs)
def write(self, d):
if self._lastop == self.OPREAD:
self._noopseek()
object.__setattr__(self, '_lastop', self.OPWRITE)
return self._fp.write(d)
def writelines(self, *args, **kwargs):
if self._lastop == self.OPREAD:
self._noopeseek()
object.__setattr__(self, '_lastop', self.OPWRITE)
return self._fp.writelines(*args, **kwargs)
def read(self, *args, **kwargs):
if self._lastop == self.OPWRITE:
self._noopseek()
object.__setattr__(self, '_lastop', self.OPREAD)
return self._fp.read(*args, **kwargs)
def readline(self, *args, **kwargs):
if self._lastop == self.OPWRITE:
self._noopseek()
object.__setattr__(self, '_lastop', self.OPREAD)
return self._fp.readline(*args, **kwargs)
def readlines(self, *args, **kwargs):
if self._lastop == self.OPWRITE:
self._noopseek()
object.__setattr__(self, '_lastop', self.OPREAD)
return self._fp.readlines(*args, **kwargs)
def posixfile(name, mode='r', buffering=-1):
'''Open a file with even more POSIX-like semantics'''
try:
fp = osutil.posixfile(name, mode, buffering) # may raise WindowsError
# The position when opening in append mode is implementation defined, so
# make it consistent with other platforms, which position at EOF.
if 'a' in mode:
fp.seek(0, os.SEEK_END)
if '+' in mode:
return mixedfilemodewrapper(fp)
return fp
except WindowsError as err:
# convert to a friendlier exception
raise IOError(err.errno, '%s: %s' % (name, err.strerror))
class winstdout(object):
'''stdout on windows misbehaves if sent through a pipe'''
def __init__(self, fp):
self.fp = fp
def __getattr__(self, key):
return getattr(self.fp, key)
def close(self):
try:
self.fp.close()
except IOError:
pass
def write(self, s):
try:
# This is workaround for "Not enough space" error on
# writing large size of data to console.
limit = 16000
l = len(s)
start = 0
self.softspace = 0
while start < l:
end = start + limit
self.fp.write(s[start:end])
start = end
except IOError as inst:
if inst.errno != 0:
raise
self.close()
raise IOError(errno.EPIPE, 'Broken pipe')
def flush(self):
try:
return self.fp.flush()
except IOError as inst:
if inst.errno != errno.EINVAL:
raise
self.close()
raise IOError(errno.EPIPE, 'Broken pipe')
sys.__stdout__ = sys.stdout = winstdout(sys.stdout)
def _is_win_9x():
'''return true if run on windows 95, 98 or me.'''
try:
return sys.getwindowsversion()[3] == 1
except AttributeError:
return 'command' in os.environ.get('comspec', '')
def openhardlinks():
return not _is_win_9x()
def parsepatchoutput(output_line):
"""parses the output produced by patch and returns the filename"""
pf = output_line[14:]
if pf[0] == '`':
pf = pf[1:-1] # Remove the quotes
return pf
def sshargs(sshcmd, host, user, port):
'''Build argument list for ssh or Plink'''
pflag = 'plink' in sshcmd.lower() and '-P' or '-p'
args = user and ("%s@%s" % (user, host)) or host
return port and ("%s %s %s" % (args, pflag, port)) or args
def setflags(f, l, x):
pass
def copymode(src, dst, mode=None):
pass
def checkexec(path):
return False
def checklink(path):
return False
def setbinary(fd):
# When run without console, pipes may expose invalid
# fileno(), usually set to -1.
fno = getattr(fd, 'fileno', None)
if fno is not None and fno() >= 0:
msvcrt.setmode(fno(), os.O_BINARY)
def pconvert(path):
return path.replace(os.sep, '/')
def localpath(path):
return path.replace('/', '\\')
def normpath(path):
return pconvert(os.path.normpath(path))
def normcase(path):
return encoding.upper(path) # NTFS compares via upper()
# see posix.py for definitions
normcasespec = encoding.normcasespecs.upper
normcasefallback = encoding.upperfallback
def samestat(s1, s2):
return False
# A sequence of backslashes is special iff it precedes a double quote:
# - if there's an even number of backslashes, the double quote is not
# quoted (i.e. it ends the quoted region)
# - if there's an odd number of backslashes, the double quote is quoted
# - in both cases, every pair of backslashes is unquoted into a single
# backslash
# (See http://msdn2.microsoft.com/en-us/library/a1y7w461.aspx )
# So, to quote a string, we must surround it in double quotes, double
# the number of backslashes that precede double quotes and add another
# backslash before every double quote (being careful with the double
# quote we've appended to the end)
_quotere = None
_needsshellquote = None
def shellquote(s):
r"""
>>> shellquote(r'C:\Users\xyz')
'"C:\\Users\\xyz"'
>>> shellquote(r'C:\Users\xyz/mixed')
'"C:\\Users\\xyz/mixed"'
>>> # Would be safe not to quote too, since it is all double backslashes
>>> shellquote(r'C:\\Users\\xyz')
'"C:\\\\Users\\\\xyz"'
>>> # But this must be quoted
>>> shellquote(r'C:\\Users\\xyz/abc')
'"C:\\\\Users\\\\xyz/abc"'
"""
global _quotere
if _quotere is None:
_quotere = re.compile(r'(\\*)("|\\$)')
global _needsshellquote
if _needsshellquote is None:
# ":" is also treated as "safe character", because it is used as a part
# of path name on Windows. "\" is also part of a path name, but isn't
# safe because shlex.split() (kind of) treats it as an escape char and
# drops it. It will leave the next character, even if it is another
# "\".
_needsshellquote = re.compile(r'[^a-zA-Z0-9._:/-]').search
if s and not _needsshellquote(s) and not _quotere.search(s):
# "s" shouldn't have to be quoted
return s
return '"%s"' % _quotere.sub(r'\1\1\\\2', s)
def quotecommand(cmd):
"""Build a command string suitable for os.popen* calls."""
if sys.version_info < (2, 7, 1):
# Python versions since 2.7.1 do this extra quoting themselves
return '"' + cmd + '"'
return cmd
def popen(command, mode='r'):
# Work around "popen spawned process may not write to stdout
# under windows"
# http://bugs.python.org/issue1366
command += " 2> %s" % os.devnull
return os.popen(quotecommand(command), mode)
def explainexit(code):
return _("exited with status %d") % code, code
# if you change this stub into a real check, please try to implement the
# username and groupname functions above, too.
def isowner(st):
return True
def findexe(command):
'''Find executable for command searching like cmd.exe does.
If command is a basename then PATH is searched for command.
PATH isn't searched if command is an absolute or relative path.
An extension from PATHEXT is found and added if not present.
If command isn't found None is returned.'''
pathext = os.environ.get('PATHEXT', '.COM;.EXE;.BAT;.CMD')
pathexts = [ext for ext in pathext.lower().split(os.pathsep)]
if os.path.splitext(command)[1].lower() in pathexts:
pathexts = ['']
def findexisting(pathcommand):
'Will append extension (if needed) and return existing file'
for ext in pathexts:
executable = pathcommand + ext
if os.path.exists(executable):
return executable
return None
if os.sep in command:
return findexisting(command)
for path in os.environ.get('PATH', '').split(os.pathsep):
executable = findexisting(os.path.join(path, command))
if executable is not None:
return executable
return findexisting(os.path.expanduser(os.path.expandvars(command)))
_wantedkinds = set([stat.S_IFREG, stat.S_IFLNK])
def statfiles(files):
'''Stat each file in files. Yield each stat, or None if a file
does not exist or has a type we don't care about.
Cluster and cache stat per directory to minimize number of OS stat calls.'''
dircache = {} # dirname -> filename -> status | None if file does not exist
getkind = stat.S_IFMT
for nf in files:
nf = normcase(nf)
dir, base = os.path.split(nf)
if not dir:
dir = '.'
cache = dircache.get(dir, None)
if cache is None:
try:
dmap = dict([(normcase(n), s)
for n, k, s in osutil.listdir(dir, True)
if getkind(s.st_mode) in _wantedkinds])
except OSError as err:
# Python >= 2.5 returns ENOENT and adds winerror field
# EINVAL is raised if dir is not a directory.
if err.errno not in (errno.ENOENT, errno.EINVAL,
errno.ENOTDIR):
raise
dmap = {}
cache = dircache.setdefault(dir, dmap)
yield cache.get(base, None)
def username(uid=None):
"""Return the name of the user with the given uid.
If uid is None, return the name of the current user."""
return None
def groupname(gid=None):
"""Return the name of the group with the given gid.
If gid is None, return the name of the current group."""
return None
def removedirs(name):
"""special version of os.removedirs that does not remove symlinked
directories or junction points if they actually contain files"""
if osutil.listdir(name):
return
os.rmdir(name)
head, tail = os.path.split(name)
if not tail:
head, tail = os.path.split(head)
while head and tail:
try:
if osutil.listdir(head):
return
os.rmdir(head)
except (ValueError, OSError):
break
head, tail = os.path.split(head)
def unlinkpath(f, ignoremissing=False):
"""unlink and remove the directory if it is empty"""
try:
unlink(f)
except OSError as e:
if not (ignoremissing and e.errno == errno.ENOENT):
raise
# try removing directories that might now be empty
try:
removedirs(os.path.dirname(f))
except OSError:
pass
def rename(src, dst):
'''atomically rename file src to dst, replacing dst if it exists'''
try:
os.rename(src, dst)
except OSError as e:
if e.errno != errno.EEXIST:
raise
unlink(dst)
os.rename(src, dst)
def gethgcmd():
return [sys.executable] + sys.argv[:1]
def groupmembers(name):
# Don't support groups on Windows for now
raise KeyError
def isexec(f):
return False
class cachestat(object):
def __init__(self, path):
pass
def cacheable(self):
return False
def lookupreg(key, valname=None, scope=None):
''' Look up a key/value name in the Windows registry.
valname: value name. If unspecified, the default value for the key
is used.
scope: optionally specify scope for registry lookup, this can be
a sequence of scopes to look up in order. Default (CURRENT_USER,
LOCAL_MACHINE).
'''
if scope is None:
scope = (winreg.HKEY_CURRENT_USER, winreg.HKEY_LOCAL_MACHINE)
elif not isinstance(scope, (list, tuple)):
scope = (scope,)
for s in scope:
try:
val = winreg.QueryValueEx(winreg.OpenKey(s, key), valname)[0]
# never let a Unicode string escape into the wild
return encoding.tolocal(val.encode('UTF-8'))
except EnvironmentError:
pass
expandglobs = True
def statislink(st):
'''check whether a stat result is a symlink'''
return False
def statisexec(st):
'''check whether a stat result is an executable file'''
return False
def poll(fds):
# see posix.py for description
raise NotImplementedError()
def readpipe(pipe):
"""Read all available data from a pipe."""
chunks = []
while True:
size = win32.peekpipe(pipe)
if not size:
break
s = pipe.read(size)
if not s:
break
chunks.append(s)
return ''.join(chunks)
def bindunixsocket(sock, path):
raise NotImplementedError('unsupported platform')