##// END OF EJS Templates
sparse-read: move from a recursive-based approach to a heap-based one...
sparse-read: move from a recursive-based approach to a heap-based one The previous recursive approach was trying to optimise each read slice to have a good density. It had the tendency to over-optimize smaller slices while leaving larger hole in others. The new approach focuses on improving the combined density of all the reads, instead of the individual slices. It slices at the largest gaps first, as they reduce the total amount of read data the most efficiently. Another benefit of this approach is that we iterate over the delta chain only once, reducing the overhead of slicing long delta chains. On the repository we use for tests, the new approach shows similar or faster performance than the current default linear full read. The repository contains about 450,000 revisions with many concurrent topological branches. Tests have been run on two versions of the repository: one built with the current delta constraint, and the other with an unlimited delta span (using 'experimental.maxdeltachainspan=0') Below are timings for building 1% of all the revision in the manifest log using 'hg perfrevlogrevisions -m'. Times are given in seconds. They include the new couple of follow-up changeset in this series. delta-span standard unlimited linear-read 922s 632s sparse-read 814s 566s

File last commit:

r34648:dacfcdd8 default
r34881:9e18ab7f default
Show More
osutil.py
102 lines | 3.5 KiB | text/x-python | PythonLexer
# osutil.py - CFFI version of osutil.c
#
# Copyright 2016 Maciej Fijalkowski <fijall@gmail.com>
#
# 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 os
import stat as statmod
from ..pure.osutil import *
from .. import (
pycompat,
)
if pycompat.isdarwin:
from . import _osutil
ffi = _osutil.ffi
lib = _osutil.lib
listdir_batch_size = 4096
# tweakable number, only affects performance, which chunks
# of bytes do we get back from getattrlistbulk
attrkinds = [None] * 20 # we need the max no for enum VXXX, 20 is plenty
attrkinds[lib.VREG] = statmod.S_IFREG
attrkinds[lib.VDIR] = statmod.S_IFDIR
attrkinds[lib.VLNK] = statmod.S_IFLNK
attrkinds[lib.VBLK] = statmod.S_IFBLK
attrkinds[lib.VCHR] = statmod.S_IFCHR
attrkinds[lib.VFIFO] = statmod.S_IFIFO
attrkinds[lib.VSOCK] = statmod.S_IFSOCK
class stat_res(object):
def __init__(self, st_mode, st_mtime, st_size):
self.st_mode = st_mode
self.st_mtime = st_mtime
self.st_size = st_size
tv_sec_ofs = ffi.offsetof("struct timespec", "tv_sec")
buf = ffi.new("char[]", listdir_batch_size)
def listdirinternal(dfd, req, stat, skip):
ret = []
while True:
r = lib.getattrlistbulk(dfd, req, buf, listdir_batch_size, 0)
if r == 0:
break
if r == -1:
raise OSError(ffi.errno, os.strerror(ffi.errno))
cur = ffi.cast("val_attrs_t*", buf)
for i in range(r):
lgt = cur.length
assert lgt == ffi.cast('uint32_t*', cur)[0]
ofs = cur.name_info.attr_dataoffset
str_lgt = cur.name_info.attr_length
base_ofs = ffi.offsetof('val_attrs_t', 'name_info')
name = str(ffi.buffer(ffi.cast("char*", cur) + base_ofs + ofs,
str_lgt - 1))
tp = attrkinds[cur.obj_type]
if name == "." or name == "..":
continue
if skip == name and tp == statmod.S_ISDIR:
return []
if stat:
mtime = cur.mtime.tv_sec
mode = (cur.accessmask & ~lib.S_IFMT)| tp
ret.append((name, tp, stat_res(st_mode=mode, st_mtime=mtime,
st_size=cur.datalength)))
else:
ret.append((name, tp))
cur = ffi.cast("val_attrs_t*", int(ffi.cast("intptr_t", cur))
+ lgt)
return ret
def listdir(path, stat=False, skip=None):
req = ffi.new("struct attrlist*")
req.bitmapcount = lib.ATTR_BIT_MAP_COUNT
req.commonattr = (lib.ATTR_CMN_RETURNED_ATTRS |
lib.ATTR_CMN_NAME |
lib.ATTR_CMN_OBJTYPE |
lib.ATTR_CMN_ACCESSMASK |
lib.ATTR_CMN_MODTIME)
req.fileattr = lib.ATTR_FILE_DATALENGTH
dfd = lib.open(path, lib.O_RDONLY, 0)
if dfd == -1:
raise OSError(ffi.errno, os.strerror(ffi.errno))
try:
ret = listdirinternal(dfd, req, stat, skip)
finally:
try:
lib.close(dfd)
except BaseException:
pass # we ignore all the errors from closing, not
# much we can do about that
return ret