Skip to content

Instantly share code, notes, and snippets.

@benhoyt
Created November 9, 2012 09:59
Show Gist options
  • Save benhoyt/4044946 to your computer and use it in GitHub Desktop.
Save benhoyt/4044946 to your computer and use it in GitHub Desktop.
Speed up os.walk() significantly by using file attributes from FindFirst/Next or readdir
"""Speed up os.walk() significantly by using file attributes that
FindFirst/Next give us instead of doing an extra stat(). Can also do the same
thing with opendir/readdir on Linux.
This is doubly useful when the user (caller of os.walk) is doing *another*
stat() to get say the file sizes.
On my tests (Windows 64-bit) our walk() is about 5x as fast as os.walk() for
large directory trees, and 9x as fast if you're doing the file size thing.
Note that these timings are "once it's in the cache", not first-time timings.
Other advantages to using FindFirst/Next:
* You can write a version of listdir() which yield results rather than getting
all filenames at once -- better for "streaming" large directories.
* You can use its built-in wildcard matching (globbing), which is presumably
faster than getting all filenames and calling fnmatch on the result. (This one
you couldn't do with opendir/readdir.)
This isn't just me who noticed these were issues. See also:
http://stackoverflow.com/questions/2485719/very-quickly-getting-total-size-of-folder
http://stackoverflow.com/questions/4403598/list-files-in-a-folder-as-a-stream-to-begin-process-immediately
http://bugs.python.org/issue15200 -- Titled "Faster os.walk", but actually unrelated
"""
import ctypes
import os
import stat
if os.name == 'nt':
from ctypes import wintypes
INVALID_HANDLE_VALUE = ctypes.c_void_p(-1).value
ERROR_FILE_NOT_FOUND = 2
ERROR_NO_MORE_FILES = 18
FILE_ATTRIBUTE_DIRECTORY = 16
FILE_ATTRIBUTE_READONLY = 1
SECONDS_BETWEEN_EPOCHS = 11644473600 # seconds between 1601-01-01 and 1970-01-01
kernel32 = ctypes.windll.kernel32
_FindFirstFile = kernel32.FindFirstFileW
_FindFirstFile.argtypes = [wintypes.LPCWSTR, ctypes.POINTER(wintypes.WIN32_FIND_DATAW)]
_FindFirstFile.restype = wintypes.HANDLE
_FindNextFile = kernel32.FindNextFileW
_FindNextFile.argtypes = [wintypes.HANDLE, ctypes.POINTER(wintypes.WIN32_FIND_DATAW)]
_FindNextFile.restype = wintypes.BOOL
_FindClose = kernel32.FindClose
_FindClose.argtypes = [wintypes.HANDLE]
_FindClose.restype = wintypes.BOOL
def _attributes_to_mode(attributes):
"""Convert Win32 dwFileAttributes to st_mode. Taken from CPython's
Modules/posixmodule.c.
"""
mode = 0
if attributes & FILE_ATTRIBUTE_DIRECTORY:
mode |= stat.S_IFDIR | 0111
else:
mode |= stat.S_IFREG
if attributes & FILE_ATTRIBUTE_READONLY:
mode |= 0444
else:
mode |= 0666
return mode
def _filetime_to_time(filetime):
total = filetime.dwHighDateTime << 32 | filetime.dwLowDateTime
return total / 10000000.0 - SECONDS_BETWEEN_EPOCHS
def _find_data_to_stat(data):
st_mode = _attributes_to_mode(data.dwFileAttributes)
st_ino = 0
st_dev = 0
st_nlink = 0
st_uid = 0
st_gid = 0
st_size = data.nFileSizeHigh << 32 | data.nFileSizeLow
st_atime = _filetime_to_time(data.ftLastAccessTime)
st_mtime = _filetime_to_time(data.ftLastWriteTime)
st_ctime = _filetime_to_time(data.ftCreationTime)
st = os.stat_result((st_mode, st_ino, st_dev, st_nlink, st_uid, st_gid,
st_size, st_atime, st_mtime, st_ctime))
return st
def listdir_stat(dirname='.', glob=None):
dirname = os.path.abspath(dirname)
filename = os.path.join(dirname, glob or '*')
data = wintypes.WIN32_FIND_DATAW()
data_p = ctypes.byref(data)
handle = _FindFirstFile(filename, data_p)
if handle == INVALID_HANDLE_VALUE:
error = ctypes.GetLastError()
if error == ERROR_FILE_NOT_FOUND:
return
raise ctypes.WinError()
try:
while True:
if data.cFileName not in ('.', '..'):
st = _find_data_to_stat(data)
yield (data.cFileName, st)
success = _FindNextFile(handle, data_p)
if not success:
error = ctypes.GetLastError()
if error == ERROR_NO_MORE_FILES:
break
raise ctypes.WinError()
finally:
if not _FindClose(handle):
raise ctypes.WinError()
# Linux/posix -- this is only half-tested and doesn't work at the moment, but
# leaving here for future use
else:
import ctypes
import ctypes.util
class DIR(ctypes.Structure):
pass
DIR_p = ctypes.POINTER(DIR)
class dirent(ctypes.Structure):
_fields_ = (
('d_ino', ctypes.c_long),
('d_off', ctypes.c_long),
('d_reclen', ctypes.c_ushort),
('d_type', ctypes.c_byte),
('d_name', ctypes.c_char * 256)
)
dirent_p = ctypes.POINTER(dirent)
_libc = ctypes.CDLL(ctypes.util.find_library('c'))
_opendir = _libc.opendir
_opendir.argtypes = [ctypes.c_char_p]
_opendir.restype = DIR_p
_readdir = _libc.readdir
_readdir.argtypes = [DIR_p]
_readdir.restype = dirent_p
_closedir = _libc.closedir
_closedir.argtypes = [DIR_p]
_closedir.restype = ctypes.c_int
DT_DIR = 4
def _type_to_stat(d_type):
if d_type == DT_DIR:
st_mode = stat.S_IFDIR | 0111
else:
st_mode = stat.S_IFREG
st = os.stat_result((st_mode, None, None, None, None, None,
None, None, None, None))
return st
def listdir_stat(dirname='.', glob=None):
dir_p = _opendir(dirname)
try:
while True:
p = _readdir(dir_p)
if not p:
break
name = p.contents.d_name
if name not in ('.', '..'):
st = _type_to_stat(p.contents.d_type)
yield (name, st)
finally:
_closedir(dir_p)
def walk(top):
try:
names_stats = list(listdir_stat(top))
except OSError:
return
dirs, nondirs = [], []
for name, st in names_stats:
if stat.S_ISDIR(st.st_mode):
dirs.append((name, st))
else:
nondirs.append((name, st))
yield top, dirs, nondirs
for name, st in dirs:
new_path = os.path.join(top, name)
for x in walk(new_path):
yield x
if __name__ == '__main__':
import datetime
import sys
import time
size = 0
time0 = time.clock()
for root, dirs, files in walk('.'):
for file, st in files:
size += st.st_size
pass
elapsed1 = time.clock() - time0
print 'our walk', elapsed1, size
time0 = time.clock()
size = 0
for root, dirs, files in os.walk('.'):
for file in files:
size += os.path.getsize(os.path.join(root, file))
pass
elapsed2 = time.clock() - time0
print 'os.walk', elapsed2, size
print 'ours was', elapsed2 / elapsed1, 'times faster'
@benhoyt
Copy link
Author

benhoyt commented Nov 9, 2012

See also my post to python-ideas on this.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment