iglob performance no better than glob

[Kyp]

I have a dir with a large # of files that I need to perform operations
on, but only needing to access a subset of the files, i.e. the first
100 files.

Using glob is very slow, so I ran across iglob, which returns an
iterator, which seemed just like what I wanted. I could iterate over
the files that I wanted, not having to read the entire dir.

So the iglob was faster, but accessing the first file took about the
same time as glob.glob.

Here's some code to compare glob vs. iglob performance, it outputs
the time before/after a glob.iglob('*.*') files.next() sequence and a
glob.glob('*.*') sequence.

#!/usr/bin/env python

import glob,time
print '\nTest of glob.iglob'
print 'before iglob:', time.asctime()
files = glob.iglob('*.*')
print 'after iglob:',time.asctime()
print files.next()
print 'after files.next():', time.asctime()

print '\nTest of glob.glob'
print 'before glob:', time.asctime()
files = glob.glob('*.*')
print 'after glob:',time.asctime()


Here are the results:

Test of glob.iglob
before iglob: Sun Jan 31 11:09:08 2010
after iglob: Sun Jan 31 11:09:08 2010
foo.bar
after files.next(): Sun Jan 31 11:09:59 2010

Test of glob.glob
before glob: Sun Jan 31 11:09:59 2010
after glob: Sun Jan 31 11:10:51 2010

The results are about the same for the 2 approaches, both took about
51 seconds. Am I doing something wrong with iglob?

Is there a way to get the first X # of files from a dir with lots of
files, that does not take a long time to run?

thanx, mark

 

[Skip Montanaro]

So the iglob was faster, but accessing the first file took about the

same time as glob.glob.

I'll wager most of the time required to access the first file is due
to filesystem overhead, not any inherent limitation in Python.

Skip Montanaro

 

[John Bokma]

Is there a way to get the first X # of files from a dir with lots of
files, that does not take a long time to run?

Assuming Linux: what does time

ls thedir | head

give?

with thedir the name of the actual dir

Also how many is many files?

 

[Peter Otten]

I have a dir with a large # of files that I need to perform operations
on, but only needing to access a subset of the files, i.e. the first
100 files.

Using glob is very slow, so I ran across iglob, which returns an
iterator, which seemed just like what I wanted. I could iterate over
the files that I wanted, not having to read the entire dir.

So the iglob was faster, but accessing the first file took about the
same time as glob.glob.

Here's some code to compare glob vs. iglob performance, it outputs
the time before/after a glob.iglob('*.*') files.next() sequence and a
glob.glob('*.*') sequence.

#!/usr/bin/env python

import glob,time
print '\nTest of glob.iglob'
print 'before iglob:', time.asctime()
files = glob.iglob('*.*')
print 'after iglob:',time.asctime()
print files.next()
print 'after files.next():', time.asctime()

print '\nTest of glob.glob'
print 'before glob:', time.asctime()
files = glob.glob('*.*')
print 'after glob:',time.asctime()


Here are the results:

Test of glob.iglob
before iglob: Sun Jan 31 11:09:08 2010
after iglob: Sun Jan 31 11:09:08 2010
foo.bar
after files.next(): Sun Jan 31 11:09:59 2010

Test of glob.glob
before glob: Sun Jan 31 11:09:59 2010
after glob: Sun Jan 31 11:10:51 2010

The results are about the same for the 2 approaches, both took about
51 seconds. Am I doing something wrong with iglob?

No, but iglob() being lazy is pointless in your case because it uses
os.listdir() and fnmatch.filter() underneath which both read the whole
directory before returning anything.

Is there a way to get the first X # of files from a dir with lots of
files, that does not take a long time to run?

Here's my attempt. It turned out to be more work than expected, so I cut a
few corners. It's Linux-only "works on my machine" code, but may give you
some hints on how to proceed.

from ctypes import *
import fnmatch
import glob
import os
import re
from itertools import ifilter, imap

class dirent(Structure):
"works on my machine "
_fields_ = [
("d_ino", c_long),
("d_off", c_long),
("d_reclen", c_ushort),
("d_type", c_ubyte),
("d_name", c_char*256)]


direntp = POINTER(dirent)

LIBC = "libc.so.6"
cdll.LoadLibrary(LIBC)
libc = CDLL(LIBC)
libc.readdir.restype = direntp


def diriter(dir):
"lazy partial replacement for os.listdir()"
# errors? what errors?
dirp = libc.opendir(dir)
if not dirp:
return
try:
while True:
ep = libc.readdir(dirp)
if not ep:
break
yield ep.contents.d_name
finally:
libc.closedir(dirp)


def filter(names, pattern):
"lazy partial replacement for fnmatch.filter()"
import posixpath

pattern = os.path.normcase(pattern)
r = fnmatch.translate(pattern)
r = re.compile(r)

if os.path is not posixpath:
names = imap(os.path.normcase, names)

return ifilter(r.match, names)

def globiter(path):
"lazy partial replacement for glob.glob()"
dir, filename = os.path.split(path)
if glob.has_magic(dir):
raise ValueError("wildcards in directory not supported")
return filter(diriter(dir), filename)


if __name__ == "__main__":
import sys
[pattern] = sys.argv[1:]
for name in globiter(pattern):
print name

Peter

 

[Benjamin Peterson]

So the iglob was faster, but accessing the first file took about the
same time as glob.glob.

That would be because glob is implemented in terms of iglob.

 

[Kyp]

Assuming Linux: what does time

 ls thedir | head

give?

with thedir the name of the actual dir

about 3 seconds.

3.086u 0.201s 0:03.32 98.7% 0+0k 0+0io 0pf+0w

Also how many is many files?

over 100K (I know I should not do that, but it's a temp dir holding
files to be transferred)
thanx, mark

 

[Kyp]

I have a dir with a large # of files that I need to perform operations
on, but only needing to access a subset of the files, i.e. the first
100 files.

Using glob is very slow, so I ran across iglob, which returns an
iterator, which seemed just like what I wanted. I could iterate over
the files that I wanted, not having to read the entire dir.

So the iglob was faster, but accessing the first file took about the
same time as glob.glob.

Here's some code to compare glob vs. iglob performance,  it outputs
the time before/after a glob.iglob('*.*') files.next() sequence and a
glob.glob('*.*') sequence.

#!/usr/bin/env python

import glob,time
print '\nTest of glob.iglob'
print 'before       iglob:', time.asctime()
files = glob.iglob('*.*')
print 'after        iglob:',time.asctime()
print files.next()
print 'after files.next():', time.asctime()

print '\nTest of glob.glob'
print 'before        glob:', time.asctime()
files = glob.glob('*.*')
print 'after         glob:',time.asctime()

Here are the results:

Test of glob.iglob
before       iglob: Sun Jan 31 11:09:08 2010
after        iglob: Sun Jan 31 11:09:08 2010
foo.bar
after files.next(): Sun Jan 31 11:09:59 2010

Test of glob.glob
before        glob: Sun Jan 31 11:09:59 2010
after         glob: Sun Jan 31 11:10:51 2010

The results are about the same for the 2 approaches, both took about
51 seconds. Am I doing something wrong with iglob?

No, but iglob() being lazy is pointless in your case because it uses
os.listdir() and fnmatch.filter() underneath which both read the whole
directory before returning anything.

Is there a way to get the first X # of files from a dir with lots of
files, that does not take a long time to run?

Here's my attempt. It turned out to be more work than expected, so I cut a
few corners. It's Linux-only "works on my machine" code, but may give you
some hints on how to proceed.

from ctypes import *
import fnmatch
import glob
import os
import re
from itertools import ifilter, imap

class dirent(Structure):
    "works on my machine "
    _fields_ = [
        ("d_ino", c_long),
        ("d_off", c_long),
        ("d_reclen", c_ushort),
        ("d_type", c_ubyte),
        ("d_name", c_char*256)]

direntp = POINTER(dirent)

LIBC = "libc.so.6"
cdll.LoadLibrary(LIBC)
libc = CDLL(LIBC)
libc.readdir.restype = direntp

def diriter(dir):
    "lazy partial replacement for os.listdir()"
    # errors? what errors?
    dirp = libc.opendir(dir)
    if not dirp:
        return
    try:
        while True:
            ep = libc.readdir(dirp)
            if not ep:
                break
            yield ep.contents.d_name
    finally:
        libc.closedir(dirp)

def filter(names, pattern):
    "lazy partial replacement for fnmatch.filter()"
    import posixpath

    pattern = os.path.normcase(pattern)
    r = fnmatch.translate(pattern)
    r = re.compile(r)

    if os.path is not posixpath:
        names = imap(os.path.normcase, names)

    return ifilter(r.match, names)

def globiter(path):
    "lazy partial replacement for glob.glob()"
    dir, filename = os.path.split(path)
    if glob.has_magic(dir):
        raise ValueError("wildcards in directory not supported")
    return filter(diriter(dir), filename)

if __name__ == "__main__":
    import sys
    [pattern] = sys.argv[1:]
    for name in globiter(pattern):
        print name

Peter

I'll give it a try, thanx for the reply.
mark

 

[Cameron Simpson]

| > Kyp wrote:
| > > I have a dir with a large # of files that I need to perform operations
| > > on, but only needing to access a subset of the files, i.e. the first
| > > 100 files.
| > > Using glob is very slow, so I ran across iglob, which returns an
| > > iterator, which seemed just like what I wanted. I could iterate over
| > > the files that I wanted, not having to read the entire dir.
[...]
| > > So the iglob was faster, but accessing the first file took about the
| > > same time as glob.glob.
| >
| > > Here's some code to compare glob vs. iglob performance,  it outputs
| > > the time before/after a glob.iglob('*.*') files.next() sequence and a
| > > glob.glob('*.*') sequence.
| >
| > > #!/usr/bin/env python
| >
| > > import glob,time
| > > print '\nTest of glob.iglob'
| > > print 'before       iglob:', time.asctime()
| > > files = glob.iglob('*.*')
| > > print 'after        iglob:',time.asctime()
| > > print files.next()
| > > print 'after files.next():', time.asctime()
| >
| > > print '\nTest of glob.glob'
| > > print 'before        glob:', time.asctime()
| > > files = glob.glob('*.*')
| > > print 'after         glob:',time.asctime()
| >
| > > Here are the results:
| >
| > > Test of glob.iglob
| > > before       iglob: Sun Jan 31 11:09:08 2010
| > > after        iglob: Sun Jan 31 11:09:08 2010
| > > foo.bar
| > > after files.next(): Sun Jan 31 11:09:59 2010
| >
| > > Test of glob.glob
| > > before        glob: Sun Jan 31 11:09:59 2010
| > > after         glob: Sun Jan 31 11:10:51 2010
| >
| > > The results are about the same for the 2 approaches, both took about
| > > 51 seconds. Am I doing something wrong with iglob?
| >
| > No, but iglob() being lazy is pointless in your case because it uses
| > os.listdir() and fnmatch.filter() underneath which both read the whole
| > directory before returning anything.
| >
| > > Is there a way to get the first X # of files from a dir with lots of
| > > files, that does not take a long time to run?
| >
| > Here's my attempt. [...open directory and read native format...]

I'd be inclined first to time os.listdir('.') versus glob.lgo('*.*').

Glob routines tend to lstat() every matching name to ensure the path
exists. That's very slow. If you just do os.listdir() and choose your
100 nmaes, you only need to stat (or just try to open) them.

So time glob.glob("*.*") versus os.listdir(".") first.

Generally, with a large directory, stat time will change performance
immensely.