尚未翻译
本页面尚未翻译成中文，因此显示英文原文。 帮助翻译
Pathlib Module Complexity¶
The pathlib module provides an object-oriented approach to filesystem path handling.
Complexity Reference¶
Unless a row says otherwise, n is the number of components in the path and the bound covers one call on an already-parsed path. Nothing in the first table touches the filesystem.
Classes¶
| Name | Time | Space | Notes |
|---|---|---|---|
PurePath, PurePosixPath, PureWindowsPath |
O(n) | O(n) | Construction only; these never touch the filesystem |
Path, PosixPath, WindowsPath |
O(n) | O(n) | Path() returns the flavour matching the running platform |
UnsupportedOperation |
O(1) | O(1) | Python 3.13+; raised where a flavour has no such operation |
Pure Path Operations¶
| Operation | Time | Space | Notes |
|---|---|---|---|
PurePath(str) |
O(n) | O(n) | n = length of the string; from 3.12 both are O(1), the string being stored and parsed on first use |
path / segment, PurePath.joinpath() |
O(n) | O(n) | n = combined length; from 3.12 the segments are stored unparsed and the cost is deferred with the rest |
PurePath.parts |
O(n) | O(n) | Rebuilt on every access from 3.12; cached after the first before that |
PurePath.parent |
O(n) | O(n) | Builds a new path from the tail |
PurePath.parents |
O(1) | O(1) | A lazy sequence; parents[i] costs O(n - i) and list(parents) is O(n²) |
PurePath.name/stem/suffix/anchor/drive/root |
O(1) | O(1) | Read from the parsed path |
PurePath.suffixes |
O(m) | O(m) | m = length of the final component |
PurePath.with_name(), .with_stem(), .with_suffix(), .with_segments() |
O(n) | O(n) | with_segments() is 3.12+ |
PurePath.relative_to(), .is_relative_to() |
O(n²) | O(n) | O(n) before 3.12 |
PurePath.match(), .full_match() |
O(p + n) | O(1) | p = components in the pattern; full_match() is 3.13+ |
PurePath.as_posix(), str(path) |
O(n) | O(n) | The string is cached after the first build |
PurePath.as_uri() |
O(n) | O(n) | Return file:// URI; deprecated on PurePath in 3.14, so call it on a Path |
PurePath.is_absolute() |
O(1) | O(1) | Reads the anchor |
PurePath.is_reserved() |
O(1) | O(1) | Windows names only; deprecated in 3.13 |
PurePath.parser |
O(1) | O(1) | Python 3.13+; the posixpath or ntpath module behind this flavour |
Filesystem Operations¶
| Operation | Time | Space | Notes |
|---|---|---|---|
Path.cwd() |
O(n) | O(n) | n = length of current directory |
Path.home() |
O(n) | O(n) | n = length of home directory |
Path.exists() |
O(1) | O(1) | One stat call |
Path.is_file() |
O(1) | O(1) | One stat call |
Path.is_dir() |
O(1) | O(1) | One stat call |
Path.is_symlink() |
O(1) | O(1) | One lstat call |
Path.is_junction() |
O(1) | O(1) | Python 3.12+; always False on POSIX |
Path.is_mount() |
O(1), O(n) on 3.12 | O(1) | n = path components; stats the path and its parent; on 3.12 it resolves the parent instead, at one lstat per component |
Path.is_socket() |
O(1) | O(1) | One stat call |
Path.is_fifo() |
O(1) | O(1) | One stat call |
Path.is_block_device() |
O(1) | O(1) | One stat call |
Path.is_char_device() |
O(1) | O(1) | One stat call |
Path.stat() |
O(1) | O(1) | Get file stats |
Path.lstat() |
O(1) | O(1) | Like stat but don't follow symlinks |
Path.info |
O(1) | O(1) | Python 3.14+; caches what it stats, so repeated queries cost one syscall between them |
Path.resolve() |
O(n) | O(n) | n = path components; one lstat each |
Path.absolute() |
O(n) | O(n) | Make absolute without resolving symlinks |
Path.expanduser() |
O(n) | O(n) | Expand ~ to home directory |
Path.iterdir() |
O(d) | O(d) | d = directory entries; the whole listing is read before the first item |
Path.walk() |
O(n) | O(w + d) | Python 3.12+; n = total entries, d = max depth, w = entries queued but not yet walked |
Path.glob(pattern) |
O(n) | O(w); O(w + d) with ** |
n = entries scanned, not entries matched; w = entries of the largest directory scanned, d = depth reached |
Path.rglob(pattern) |
O(n) | O(w + d) | glob('**/' + pattern); n = entries in the tree, d = depth reached |
Path.mkdir() |
O(1) | O(1) | O(d) for d missing components with parents=True |
Path.touch() |
O(1) | O(1) | Create file or update timestamp |
Path.rename(target) |
O(1) | O(1) | Rename path |
Path.replace(target) |
O(1) | O(1) | Rename, overwriting target if exists |
Path.unlink() |
O(1) | O(1) | Delete file |
Path.rmdir() |
O(1) | O(1) | Delete empty directory |
Path.symlink_to(target) |
O(1) | O(1) | Create symlink |
Path.hardlink_to(target) |
O(1) | O(1) | Create hard link |
Path.link_to(target) |
O(1) | O(1) | Removed in 3.12; use hardlink_to(), whose arguments are the other way round |
Path.readlink() |
O(n) | O(n) | n = length of symlink target |
Path.chmod(mode) |
O(1) | O(1) | Change file mode |
Path.lchmod(mode) |
O(1) | O(1) | chmod without following symlinks |
Path.owner() |
O(1) | O(1) | One stat call plus a user-database lookup |
Path.group() |
O(1) | O(1) | One stat call plus a group-database lookup |
Path.samefile(other) |
O(1) | O(1) | Two stat calls |
Path.open(mode) |
O(1) | O(1) | Open file (returns file object) |
Path.read_text() |
O(n) | O(n) | Read file as text |
Path.read_bytes() |
O(n) | O(n) | Read file as bytes |
Path.write_text() |
O(n) | O(n) | Write text to file |
Path.write_bytes() |
O(n) | O(n) | Write bytes to file |
Path.copy(target), .copy_into(dir) |
O(b) | O(1) | Python 3.14+; b = bytes copied, streamed rather than held |
Path.move(target), .move_into(dir) |
O(1) | O(1) | Python 3.14+; O(b) across filesystems, where it copies and deletes |
Path.from_uri(uri) |
O(n) | O(n) | Create Path from URI (Python 3.13+) |
Path Construction¶
Time Complexity: O(n)¶
Where n = length of the path string.
from pathlib import Path
# Simple path: O(n) in the length of the string
path = Path('/home/user/documents/file.txt')
# Relative path: O(n)
path = Path('docs/README.md')
# Current directory: O(n) in path length
cwd = Path.cwd()
# Home directory: O(n) in path length
home = Path.home()
Space Complexity: O(n)¶
from pathlib import Path
# Stores the path string
path = Path('/' + 'a' * 10000) # O(n) space
Construction Is Deferred From Python 3.12¶
Before 3.12 the constructor splits the string into components immediately. From 3.12 it only keeps the arguments; the split happens on the first operation that needs it, and the result is cached on the object. Building a path costs O(1) there, and building paths you never use costs nothing beyond the reference.
from pathlib import Path
# 3.12+: O(1) - nothing is parsed yet
path = Path('/' + 'a' * 100000)
# The deferred parse is paid here, once
text = str(path) # O(n)
# Cached: subsequent uses do not pay it again
text = str(path) # O(1)
The same applies to joining, which is why / is cheap on recent versions and
proportional to the path on older ones.
Pure Paths Never Touch the Filesystem¶
PurePath and its two flavours are string manipulation with a path-shaped
API. Every operation in the pure table above costs what parsing and rebuilding
a string costs, and none of them can fail because a file is missing or a
directory is unreadable.
from pathlib import PurePosixPath, PureWindowsPath
# No such directory anywhere; all of this still works
p = PurePosixPath('/nowhere/at/all/report.tar.gz')
p.name # 'report.tar.gz' O(1)
p.stem # 'report.tar' O(1)
p.suffix # '.gz' O(1)
p.suffixes # ['.tar', '.gz'] O(m) in the final component
p.parts # ('/', 'nowhere', 'at', 'all', 'report.tar.gz') O(n)
# Windows semantics on any platform
PureWindowsPath('C:/Users/x').drive # 'C:' O(1)
Path inherits all of it and adds the operations that do make system calls.
Two Things That Cost More Than They Look¶
relative_to() Is Quadratic From Python 3.12¶
relative_to() and is_relative_to() walk the parents of one path looking for
the other, and testing each candidate scans the parents again. Before 3.12 the
comparison was a slice; from 3.12 the search is quadratic in the number of
components, which turns a deep path into real work.
from pathlib import PurePosixPath
deep = PurePosixPath('/' + '/'.join(f'd{i}' for i in range(200)) + '/f.txt')
# O(n²) from 3.12, O(n) before it
deep.relative_to('/d0')
For a path a handful of components long this is invisible. For one two hundred deep, called in a loop over many paths, it is the loop's dominant cost.
parts Is Rebuilt On Every Access From Python 3.12¶
parts returns a fresh tuple each time it is read. Up to 3.11 the tuple was
cached after the first access; from 3.12 it is not, so reading it inside a loop
pays O(n) each time round.
from pathlib import PurePosixPath
p = PurePosixPath('/a/b/c/d/e/f.txt')
# Read once, use many times
parts = p.parts
first, last = parts[0], parts[-1]
parents is the opposite: it is a lazy sequence, O(1) to obtain, and
parents[i] builds only the path it returns. Materialising the whole of it
with list() is O(n²), because each of the n parents holds up to n components.
Path Operations¶
exists(), is_file(), is_dir()¶
Time Complexity: O(1)¶
Each makes exactly one stat call, whatever the path's length or the size of the directory holding it.
from pathlib import Path
path = Path('file.txt')
# Single stat system call: O(1)
if path.exists(): # O(1) - stat call
print('exists')
if path.is_file(): # O(1) - stat call
print('is file')
if path.is_dir(): # O(1) - stat call
print('is directory')
Space Complexity: O(1)¶
from pathlib import Path
path = Path('large_name' * 20)
# No additional memory needed
exists = path.exists() # O(1) space
Path Resolution¶
resolve()¶
Time Complexity: O(n)¶
Where n = number of path components; each one costs an lstat call, and a symlink adds the components of its target.
from pathlib import Path
# Simple resolution: O(n) where n = components
path = Path('docs/../files/./file.txt')
absolute = path.resolve() # O(n) - normalize and resolve
# With symlinks: O(n) where n = symlinks + components
# Each symlink read is a system call
path = Path('/path/with/symlinks/file.txt')
absolute = path.resolve() # O(n)
Space Complexity: O(n)¶
from pathlib import Path
# Stores resolved path
path = Path('relative/path').resolve() # O(n) space
Directory Operations¶
iterdir()¶
Time Complexity: O(d)¶
Where d = number of directory entries.
from pathlib import Path
path = Path.home()
# List all entries: O(d) where d = entries
for item in path.iterdir(): # O(d)
print(item)
# Count entries: O(d)
count = sum(1 for _ in path.iterdir()) # O(d)
Space Complexity: O(d)¶
iterdir() returns an iterator, but it is not a streaming one: the directory
is read in full and every name is materialised before the first item is
yielded. Iterating instead of calling list() avoids holding the Path
objects, not the listing they are built from.
from pathlib import Path
path = Path.home()
# The listing is already in memory by the time this yields anything
for item in path.iterdir(): # O(d) space, not O(1)
print(item)
# Building a list on top of it costs the Path objects as well
entries = list(path.iterdir()) # O(d) space
A big directory costs memory even if you stop early
next(iter(path.iterdir())) reads the whole directory. Where that
matters, os.scandir() is the streaming alternative - see
OS Module.
glob()¶
Time Complexity: O(n)¶
Where n = number of entries scanned, which is every entry of every directory the pattern opens, not the number of entries that match.
from pathlib import Path
path = Path('.')
# Simple glob: O(n) where n = entries in this directory
py_files = list(path.glob('*.py')) # O(n)
# Nested glob: O(n) over every entry in the subtree
txt_files = list(path.glob('**/*.txt')) # O(n)
# A non-matching leading component still costs the scan that rejects it
results = list(path.glob('[a-z]*/data/*.json')) # O(n)
Space Complexity: O(w)¶
Where w = entries of the largest directory scanned. Like iterdir(), glob()
reads a directory in one go, so the peak follows the widest directory it opens
rather than the number of matches it yields. A ** in the pattern walks the
tree as rglob() does, and pays its O(w + d) space instead.
from pathlib import Path
# Iterating does not make the scan incremental within a directory
for match in Path('.').glob('*'): # O(w) space
print(match)
# Collecting the results adds the matches on top
matches = list(Path('.').glob('*')) # O(w + matches) space
# A ** pattern pays rglob's depth term instead of the flat O(w)
txt_files = list(Path('.').glob('**/*.txt')) # O(w + d) space
rglob()¶
Time Complexity: O(n)¶
Where n = total entries in the directory tree.
from pathlib import Path
path = Path('.')
# Recursive glob: O(n) where n = all entries in the tree
all_py = list(path.rglob('*.py')) # O(n)
# rglob(pattern) is glob('**/' + pattern)
all_txt = list(path.rglob('*.txt')) # O(n)
Space Complexity: O(w + d)¶
Where w = entries of the largest directory scanned and d = depth reached. Depth does not fold into width: a deeper tree costs more than a shallow one holding the same entries.
from pathlib import Path
# All results in memory, on top of the walk's own O(w + d)
matches = list(Path('.').rglob('*.py')) # O(matches) space
File I/O¶
read_text() / read_bytes()¶
Time Complexity: O(n)¶
Where n = file size.
from pathlib import Path
path = Path('data.txt')
path.write_text('example')
# Read entire file: O(n) where n = file size
content = path.read_text() # O(n)
# Read binary: O(n)
data = path.read_bytes() # O(n)
# Read with encoding: O(n)
content = path.read_text(encoding='utf-8') # O(n)
Space Complexity: O(n)¶
from pathlib import Path
Path('large_file.txt').write_text('x' * 1000)
# Stores entire file content
content = Path('large_file.txt').read_text() # O(n) space
write_text() / write_bytes()¶
Time Complexity: O(n)¶
Where n = content size.
from pathlib import Path
path = Path('output.txt')
# Write text: O(n) in the length of the string
path.write_text('Hello, World!')
# Write bytes: O(n)
path.write_bytes(b'Binary data')
# Overwrites file: O(n) regardless of original size
path.write_text('x' * 1000000)
Space Complexity: O(n)¶
from pathlib import Path
# Encodes to bytes before writing
content = 'x' * 1000000
Path('file.txt').write_text(content) # O(n) extra space
File System Modifications¶
mkdir(), rmdir(), unlink(), rename()¶
Time Complexity: O(1)¶
from pathlib import Path
# Create directory: O(1) - single system call
Path('new_dir').mkdir() # O(1)
# Create with parents: O(d) where d = missing components
Path('a/b/c/d').mkdir(parents=True, exist_ok=True) # O(d)
# Remove empty directory: O(1)
Path('new_dir').rmdir() # O(1)
# Rename: O(1) - same filesystem
Path('a/b/c/d').rename('a/b/c/e') # O(1)
# Remove file: O(1)
Path('file.txt').touch() # O(1)
Path('file.txt').unlink() # O(1)
Space Complexity: O(1)¶
from pathlib import Path
# No significant memory allocation
Path('dir').mkdir() # O(1) space
Path('dir').rmdir() # O(1) space
Common Patterns¶
Safe File Operations¶
from pathlib import Path
# Check before operating
path = Path('file.txt')
path.write_text('contents')
if path.exists() and path.is_file(): # O(1) + O(1)
content = path.read_text() # O(n)
Directory Tree Processing¶
from pathlib import Path
def process(text):
return len(text)
# Process all files: O(n) to walk, plus the size of each file read
for file in Path('.').rglob('*.py'): # O(n) to find files
content = file.read_text() # O(size) per file
process(content)
Path Construction¶
from pathlib import Path
names = ['a.txt', 'b.txt']
# Build paths: O(n) in total path length, O(1) from Python 3.12
base = Path('data')
file = base / 'subdir' / 'file.txt'
# Multiple files
for name in names:
path = base / name
Safe Deletion¶
from pathlib import Path
import shutil
path = Path('directory')
# Delete directory and contents: O(n) where n = total entries
if path.is_dir(): # O(1)
shutil.rmtree(path) # O(n)
# Or using pathlib
# Path.rmdir() only works on empty directories
Comparison with os.path¶
from pathlib import Path
import os
# pathlib (object-oriented)
p = Path('data') / 'file.txt' # O(n), O(1) from Python 3.12
exists = p.exists() # O(1)
# os.path (functional)
path = os.path.join('data', 'file.txt') # O(n)
exists = os.path.exists(path) # O(1)
# The filesystem calls cost the same either way. The string handling does not:
# from 3.12 pathlib defers it, so a path that is built and never used is free,
# while os.path.join always produces the joined string.
Performance Characteristics¶
Best Practices¶
from pathlib import Path
def process(path):
return path.name
# Good: scan once
base = Path('.')
files = list(base.glob('*.py')) # O(n) once
for file in files: # Iterate cached results
process(file) # O(1) per file
# Bad: repeated glob calls rescan the directory every time
for i in range(100):
files = list(Path('.').glob('*.py')) # O(n) * 100
Python 3.14: info, copy and move¶
Path.info is a cached view of what one stat call already told the
interpreter. Each query on the same object is free; each Path.exists() is
another syscall.
from pathlib import Path
path = Path('report.txt')
path.write_text('contents')
# One stat call answers all three
info = path.info
info.exists(), info.is_file(), info.is_dir()
# Three separate stat calls
path.exists(), path.is_file(), path.is_dir()
The cache never expires, so a long-lived Path whose file changes underneath
it keeps answering from the old stat. Build a fresh Path where that matters.
copy() streams, so it is O(b) in the bytes moved and O(1) in space however
large the file is. move() on one filesystem is a rename and costs neither.
from pathlib import Path
source = Path('report.txt')
source.write_text('contents')
source.copy(Path('backup.txt')) # O(b) time, O(1) space
source.move(Path('archive.txt')) # O(1) on the same filesystem
Version Notes¶
- Python 3.4+: pathlib introduced
- Python 3.10+:
Path.hardlink_to()added, deprecatingPath.link_to() - Python 3.12+:
Path.walk(),Path.is_junction()andPurePath.with_segments()added; construction and joining defer parsing to first use;partsis rebuilt per access andrelative_to()becomes quadratic;Path.link_to()removed - Python 3.13+:
Path.from_uri(),PurePath.full_match(),PurePath.parserandUnsupportedOperationadded;iterdir()reads the directory when it is called rather than at the firstnext();PurePath.is_reserved()deprecated - Python 3.14+:
Path.copy(),Path.copy_into(),Path.move(),Path.move_into()andPath.infoadded;PurePath.as_uri()deprecated in favour ofPath.as_uri()
Related Documentation¶
- OS Module - Low-level filesystem operations
- Glob Module - Unix-style pathname expansion