Mercurial > hg
view mercurial/pycompat.py @ 38732:be4984261611
merge: mark file gets as not thread safe (issue5933)
In default installs, this has the effect of disabling the thread-based
worker on Windows when manifesting files in the working directory. My
measurements have shown that with revlog-based repositories, Mercurial
spends a lot of CPU time in revlog code resolving file data. This ends
up incurring a lot of context switching across threads and slows down
`hg update` operations when going from an empty working directory to
the tip of the repo.
On mozilla-unified (246,351 files) on an i7-6700K (4+4 CPUs):
before: 487s wall
after: 360s wall (equivalent to worker.enabled=false)
cpus=2: 379s wall
Even with only 2 threads, the thread pool is still slower.
The introduction of the thread-based worker (02b36e860e0b) states that
it resulted in a "~50%" speedup for `hg sparse --enable-profile` and
`hg sparse --disable-profile`. This disagrees with my measurement
above. I theorize a few reasons for this:
1) Removal of files from the working directory is I/O - not CPU - bound
and should benefit from a thread pool (unless I/O is insanely fast
and the GIL release is near instantaneous). So tests like `hg sparse
--enable-profile` may exercise deletion throughput and aren't good
benchmarks for worker tasks that are CPU heavy.
2) The patch was authored by someone at Facebook. The results were
likely measured against a repository using remotefilelog. And I
believe that revision retrieval during working directory updates with
remotefilelog will often use a remote store, thus being I/O and not
CPU bound. This probably resulted in an overstated performance gain.
Since there appears to be a need to enable the thread-based worker with
some stores, I've made the flagging of file gets as thread safe
configurable. I've made it experimental because I don't want to formalize
a boolean flag for this option and because this attribute is best
captured against the store implementation. But we don't have a proper
store API for this yet. I'd rather cross this bridge later.
It is possible there are revlog-based repositories that do benefit from
a thread-based worker. I didn't do very comprehensive testing. If there
are, we may want to devise a more proper algorithm for whether to use
the thread-based worker, including possibly config options to limit the
number of threads to use. But until I see evidence that justifies
complexity, simplicity wins.
Differential Revision: https://phab.mercurial-scm.org/D3963
author | Gregory Szorc <gregory.szorc@gmail.com> |
---|---|
date | Wed, 18 Jul 2018 09:49:34 -0700 |
parents | 152f4822d210 |
children | 7eba8f83129b |
line wrap: on
line source
# pycompat.py - portability shim for python 3 # # This software may be used and distributed according to the terms of the # GNU General Public License version 2 or any later version. """Mercurial portability shim for python 3. This contains aliases to hide python version-specific details from the core. """ from __future__ import absolute_import import getopt import inspect import os import shlex import sys import tempfile ispy3 = (sys.version_info[0] >= 3) ispypy = (r'__pypy__' in sys.builtin_module_names) if not ispy3: import cookielib import cPickle as pickle import httplib import Queue as queue import SocketServer as socketserver import xmlrpclib from .thirdparty.concurrent import futures def future_set_exception_info(f, exc_info): f.set_exception_info(*exc_info) else: import concurrent.futures as futures import http.cookiejar as cookielib import http.client as httplib import pickle import queue as queue import socketserver import xmlrpc.client as xmlrpclib def future_set_exception_info(f, exc_info): f.set_exception(exc_info[0]) def identity(a): return a def _rapply(f, xs): if xs is None: # assume None means non-value of optional data return xs if isinstance(xs, (list, set, tuple)): return type(xs)(_rapply(f, x) for x in xs) if isinstance(xs, dict): return type(xs)((_rapply(f, k), _rapply(f, v)) for k, v in xs.items()) return f(xs) def rapply(f, xs): """Apply function recursively to every item preserving the data structure >>> def f(x): ... return 'f(%s)' % x >>> rapply(f, None) is None True >>> rapply(f, 'a') 'f(a)' >>> rapply(f, {'a'}) == {'f(a)'} True >>> rapply(f, ['a', 'b', None, {'c': 'd'}, []]) ['f(a)', 'f(b)', None, {'f(c)': 'f(d)'}, []] >>> xs = [object()] >>> rapply(identity, xs) is xs True """ if f is identity: # fast path mainly for py2 return xs return _rapply(f, xs) if ispy3: import builtins import functools import io import struct fsencode = os.fsencode fsdecode = os.fsdecode oscurdir = os.curdir.encode('ascii') oslinesep = os.linesep.encode('ascii') osname = os.name.encode('ascii') ospathsep = os.pathsep.encode('ascii') ospardir = os.pardir.encode('ascii') ossep = os.sep.encode('ascii') osaltsep = os.altsep if osaltsep: osaltsep = osaltsep.encode('ascii') # os.getcwd() on Python 3 returns string, but it has os.getcwdb() which # returns bytes. getcwd = os.getcwdb sysplatform = sys.platform.encode('ascii') sysexecutable = sys.executable if sysexecutable: sysexecutable = os.fsencode(sysexecutable) bytesio = io.BytesIO # TODO deprecate stringio name, as it is a lie on Python 3. stringio = bytesio def maplist(*args): return list(map(*args)) def rangelist(*args): return list(range(*args)) def ziplist(*args): return list(zip(*args)) rawinput = input getargspec = inspect.getfullargspec # TODO: .buffer might not exist if std streams were replaced; we'll need # a silly wrapper to make a bytes stream backed by a unicode one. stdin = sys.stdin.buffer stdout = sys.stdout.buffer stderr = sys.stderr.buffer # Since Python 3 converts argv to wchar_t type by Py_DecodeLocale() on Unix, # we can use os.fsencode() to get back bytes argv. # # https://hg.python.org/cpython/file/v3.5.1/Programs/python.c#l55 # # TODO: On Windows, the native argv is wchar_t, so we'll need a different # workaround to simulate the Python 2 (i.e. ANSI Win32 API) behavior. if getattr(sys, 'argv', None) is not None: sysargv = list(map(os.fsencode, sys.argv)) bytechr = struct.Struct('>B').pack byterepr = b'%r'.__mod__ class bytestr(bytes): """A bytes which mostly acts as a Python 2 str >>> bytestr(), bytestr(bytearray(b'foo')), bytestr(u'ascii'), bytestr(1) ('', 'foo', 'ascii', '1') >>> s = bytestr(b'foo') >>> assert s is bytestr(s) __bytes__() should be called if provided: >>> class bytesable(object): ... def __bytes__(self): ... return b'bytes' >>> bytestr(bytesable()) 'bytes' There's no implicit conversion from non-ascii str as its encoding is unknown: >>> bytestr(chr(0x80)) # doctest: +ELLIPSIS Traceback (most recent call last): ... UnicodeEncodeError: ... Comparison between bytestr and bytes should work: >>> assert bytestr(b'foo') == b'foo' >>> assert b'foo' == bytestr(b'foo') >>> assert b'f' in bytestr(b'foo') >>> assert bytestr(b'f') in b'foo' Sliced elements should be bytes, not integer: >>> s[1], s[:2] (b'o', b'fo') >>> list(s), list(reversed(s)) ([b'f', b'o', b'o'], [b'o', b'o', b'f']) As bytestr type isn't propagated across operations, you need to cast bytes to bytestr explicitly: >>> s = bytestr(b'foo').upper() >>> t = bytestr(s) >>> s[0], t[0] (70, b'F') Be careful to not pass a bytestr object to a function which expects bytearray-like behavior. >>> t = bytes(t) # cast to bytes >>> assert type(t) is bytes """ def __new__(cls, s=b''): if isinstance(s, bytestr): return s if (not isinstance(s, (bytes, bytearray)) and not hasattr(s, u'__bytes__')): # hasattr-py3-only s = str(s).encode(u'ascii') return bytes.__new__(cls, s) def __getitem__(self, key): s = bytes.__getitem__(self, key) if not isinstance(s, bytes): s = bytechr(s) return s def __iter__(self): return iterbytestr(bytes.__iter__(self)) def __repr__(self): return bytes.__repr__(self)[1:] # drop b'' def iterbytestr(s): """Iterate bytes as if it were a str object of Python 2""" return map(bytechr, s) def maybebytestr(s): """Promote bytes to bytestr""" if isinstance(s, bytes): return bytestr(s) return s def sysbytes(s): """Convert an internal str (e.g. keyword, __doc__) back to bytes This never raises UnicodeEncodeError, but only ASCII characters can be round-trip by sysstr(sysbytes(s)). """ return s.encode(u'utf-8') def sysstr(s): """Return a keyword str to be passed to Python functions such as getattr() and str.encode() This never raises UnicodeDecodeError. Non-ascii characters are considered invalid and mapped to arbitrary but unique code points such that 'sysstr(a) != sysstr(b)' for all 'a != b'. """ if isinstance(s, builtins.str): return s return s.decode(u'latin-1') def strurl(url): """Converts a bytes url back to str""" if isinstance(url, bytes): return url.decode(u'ascii') return url def bytesurl(url): """Converts a str url to bytes by encoding in ascii""" if isinstance(url, str): return url.encode(u'ascii') return url def raisewithtb(exc, tb): """Raise exception with the given traceback""" raise exc.with_traceback(tb) def getdoc(obj): """Get docstring as bytes; may be None so gettext() won't confuse it with _('')""" doc = getattr(obj, u'__doc__', None) if doc is None: return doc return sysbytes(doc) def _wrapattrfunc(f): @functools.wraps(f) def w(object, name, *args): return f(object, sysstr(name), *args) return w # these wrappers are automagically imported by hgloader delattr = _wrapattrfunc(builtins.delattr) getattr = _wrapattrfunc(builtins.getattr) hasattr = _wrapattrfunc(builtins.hasattr) setattr = _wrapattrfunc(builtins.setattr) xrange = builtins.range unicode = str def open(name, mode='r', buffering=-1, encoding=None): return builtins.open(name, sysstr(mode), buffering, encoding) safehasattr = _wrapattrfunc(builtins.hasattr) def _getoptbwrapper(orig, args, shortlist, namelist): """ Takes bytes arguments, converts them to unicode, pass them to getopt.getopt(), convert the returned values back to bytes and then return them for Python 3 compatibility as getopt.getopt() don't accepts bytes on Python 3. """ args = [a.decode('latin-1') for a in args] shortlist = shortlist.decode('latin-1') namelist = [a.decode('latin-1') for a in namelist] opts, args = orig(args, shortlist, namelist) opts = [(a[0].encode('latin-1'), a[1].encode('latin-1')) for a in opts] args = [a.encode('latin-1') for a in args] return opts, args def strkwargs(dic): """ Converts the keys of a python dictonary to str i.e. unicodes so that they can be passed as keyword arguments as dictonaries with bytes keys can't be passed as keyword arguments to functions on Python 3. """ dic = dict((k.decode('latin-1'), v) for k, v in dic.iteritems()) return dic def byteskwargs(dic): """ Converts keys of python dictonaries to bytes as they were converted to str to pass that dictonary as a keyword argument on Python 3. """ dic = dict((k.encode('latin-1'), v) for k, v in dic.iteritems()) return dic # TODO: handle shlex.shlex(). def shlexsplit(s, comments=False, posix=True): """ Takes bytes argument, convert it to str i.e. unicodes, pass that into shlex.split(), convert the returned value to bytes and return that for Python 3 compatibility as shelx.split() don't accept bytes on Python 3. """ ret = shlex.split(s.decode('latin-1'), comments, posix) return [a.encode('latin-1') for a in ret] else: import cStringIO unicode = unicode bytechr = chr byterepr = repr bytestr = str iterbytestr = iter maybebytestr = identity sysbytes = identity sysstr = identity strurl = identity bytesurl = identity # this can't be parsed on Python 3 exec('def raisewithtb(exc, tb):\n' ' raise exc, None, tb\n') def fsencode(filename): """ Partial backport from os.py in Python 3, which only accepts bytes. In Python 2, our paths should only ever be bytes, a unicode path indicates a bug. """ if isinstance(filename, str): return filename else: raise TypeError( "expect str, not %s" % type(filename).__name__) # In Python 2, fsdecode() has a very chance to receive bytes. So it's # better not to touch Python 2 part as it's already working fine. fsdecode = identity def getdoc(obj): return getattr(obj, '__doc__', None) _notset = object() def safehasattr(thing, attr): return getattr(thing, attr, _notset) is not _notset def _getoptbwrapper(orig, args, shortlist, namelist): return orig(args, shortlist, namelist) strkwargs = identity byteskwargs = identity oscurdir = os.curdir oslinesep = os.linesep osname = os.name ospathsep = os.pathsep ospardir = os.pardir ossep = os.sep osaltsep = os.altsep stdin = sys.stdin stdout = sys.stdout stderr = sys.stderr if getattr(sys, 'argv', None) is not None: sysargv = sys.argv sysplatform = sys.platform getcwd = os.getcwd sysexecutable = sys.executable shlexsplit = shlex.split bytesio = cStringIO.StringIO stringio = bytesio maplist = map rangelist = range ziplist = zip rawinput = raw_input getargspec = inspect.getargspec isjython = sysplatform.startswith('java') isdarwin = sysplatform == 'darwin' isposix = osname == 'posix' iswindows = osname == 'nt' def getoptb(args, shortlist, namelist): return _getoptbwrapper(getopt.getopt, args, shortlist, namelist) def gnugetoptb(args, shortlist, namelist): return _getoptbwrapper(getopt.gnu_getopt, args, shortlist, namelist) def mkdtemp(suffix=b'', prefix=b'tmp', dir=None): return tempfile.mkdtemp(suffix, prefix, dir) # text=True is not supported; use util.from/tonativeeol() instead def mkstemp(suffix=b'', prefix=b'tmp', dir=None): return tempfile.mkstemp(suffix, prefix, dir) # mode must include 'b'ytes as encoding= is not supported def namedtempfile(mode=b'w+b', bufsize=-1, suffix=b'', prefix=b'tmp', dir=None, delete=True): mode = sysstr(mode) assert r'b' in mode return tempfile.NamedTemporaryFile(mode, bufsize, suffix=suffix, prefix=prefix, dir=dir, delete=delete)