Python standard library changes in recent years

With each major Python release, all the attention goes to the new language features: the walrus operator, dictionary merging, pattern matching. There is also a lot of writing about asyncio and typing modules — they are developing rapidly and are obviously important for the core team.

The rest of the standard library modules receive undeservedly little attention. I want to fix this and tell you about the novelties introduced in versions 3.8–3.10.

Note. This is not an exhaustive list, of course. I write only about those changes that interested me personally. But since I am not too much different from the "average" Python backend developer, it is likely that you will also be interested. Let me know if I missed something.

The modules are in alphabetical order, so if you get bored with the first (little-known) ones, do not be discouraged — it gets more exciting further.


All new features and improvements in the article are accompanied by examples. You can try them in the playground or run locally. If you have an older local Python, run it using Docker:

$ docker run -it --rm python:3.10-alpine


array module provides compact typed numeric arrays. It is used much less frequently than the famous list counterpart.

array.index() methods finds the value in the array and returns the index of the found element. Now it supports optional start and stop parameters, which define the search interval (3.10+):

from array import array
arr = array("i", [7, 11, 19, 42])

idx = arr.index(11)
# idx == 1

idx = arr.index(11, 2)
# ValueError: array.index(x): x not in array


Contributed by: Anders LorentsenZackery Spytz


base64 module encodes binary data into ASCII strings using Base16, Base32, and Base64 algorithms.

It received a couple of new functions: b32hexencode() and b32hexdecode(), which use an extended 32-character alphabet according to RFC 4648 (3.10+):

import base64
bytes = b"python is awesome"




Contributed by: Filipe Laíns


bisect module works with sorted lists using binary search method. Main functions are:

  • bisect() finds an item in the list;
  • insort() adds an item while retaining the order.
import bisect

lst = [7, 11, 19, 42]
idx = bisect.bisect(lst, 12)
# idx == 2

bisect.insort(lst, 12)
# [7, 11, 12, 19, 42]

Since version 3.10, all module functions support the optional key parameter. It is a function that returns the value of a list item. It is convenient to use if the elements cannot be compared directly:

import bisect
import operator

p1 = {"id": 11, "name": "Diane"}
p2 = {"id": 12, "name": "Bob"}
p3 = {"id": 13, "name": "Emma"}

key = operator.itemgetter("name")
people = sorted([p1, p2, p3], key=key)
# Bob, Diane, Emma

idx = bisect.bisect(people, "Dan")
# TypeError: '<' not supported between instances of 'str' and 'dict'

idx = bisect.bisect(people, "Dan", key=key)
# idx == 1


Contributed by: Raymond Hettinger


builtins module contains all the built-in functions and classes that programmers use without imports: int, list, len(), open() and the like.

import builtins

list is builtins.list
# True

len is builtins.len
# True

The string received str.removeprefix() and str.removesuffix() methods, which cut off the string head and tail respectively (3.9+):

s = "Python is awesome"

s.removeprefix("Python is ")
# 'awesome'

s.removesuffix(" is awesome")
# 'Python'

The integer received the int.bit_count() method, which returns the number of ones in the binary representation of the integer (3.10+):

n = 42

# '0b101010'

# 3

The dictionary methods dict.key(), dict.values() and dict.items() return view objects that reference dictionary data. Previously, it was impossible to get a reverse link to the dictionary from these objects, but now it can be done — through the .mapping attribute (3.10+):

people = {
    "Diane": 70,
    "Bob": 78,
    "Emma": 84

keys = people.keys()
# dict_keys(['Diane', 'Bob', 'Emma'])

# 78

Collection merging zip() function received the strict parameter. It ensures that the sequences are of the same length (3.10+):

keys = ["Diane", "Bob", "Emma"]
vals = [70, 78, 84, 42]

pairs = zip(keys, vals)
# [('Diane', 70), ('Bob', 78), ('Emma', 84)]

pairs = zip(keys, vals, strict=True)
# ValueError: zip() argument 2 is longer than argument 1


Contributed by: Dennis SweeneyNiklas FiekasBrandt Bucher


dataclasses module generates classes according to the specification.

Dataclasses can now use slots, which are great for creating compact data objects with a fixed set of properties (3.10+).

Regular dataclass:

from dataclasses import dataclass

class Person:
    id: int
    name: str

diane = Person(id=11, name="Diane")
# {'id': 11, 'name': 'Diane'}
diane.salary = 70
# ok

Dataclass with slots:

from dataclasses import dataclass

class SlotPerson:
    id: int
    name: str

bob = SlotPerson(id=12, name="Bob")
# AttributeError: 'SlotPerson' object has no attribute '__dict__'
# ('id', 'name')
bob.salary = 78
# AttributeError: 'SlotPerson' object has no attribute 'salary'

Besides, the dataclass can now be forced to accept keyword-only parameters when creating an object (3.10+):

from dataclasses import dataclass

class KeywordPerson:
    id: int
    name: str

diane = KeywordPerson(id=11, name="Diane")
# ok
diane = KeywordPerson(11, "Diane")
# TypeError: KeywordPerson.__init__() takes 1 positional argument but 3 were given


Contributed by: Yurii KarabasEric V. Smith


datetime module (unsurprisingly) deals with date and time.

It received new date.fromisocalendar() and datetime.fromisocalendar() constructors, which create a date from the (year, week, week_day) trio (3.8+):

import datetime as dt

day =, 9, 13)
# datetime.IsoCalendarDate(year=2022, week=37, weekday=2)

year, week, day = day.isocalendar()
next_day =, week, day+1)
#, 9, 14)

Besides, the .isocalendar() method now returns a named IsoCalendarDate instead of the regular tuple (3.9+). You can see it in the example above.


Contributed by: Paul GanssleDong-hee Na


fractions module works with rational numbers.

It received the Fraction.as_integer_ratio() method to return a fraction as a (numerator, denominator) pair, thereby fixing the age-old shame of the usual float (3.8+):

# (1, 4)

# (1, 2)

# (3602879701896397, 18014398509481984)
# oopsie
from fractions import Fraction

# (1, 5)
# so much better

To be fair, decimal.Decimal learned to do this back in 3.6. But it's still nice.


Contributed by: Lisa RoachRaymond Hettinger


functools module is a collection of higher-order auxiliary functions. One of them is lru_cache(), which caches expensive calculations:

import functools
import time

def find_user(name):
    # imitating slow search
    user = {"id": 11, "name": "Diane"}
    return user

# kinda slow

# blazingly fast

Previously, it required to explicitly set the cache size. And now you can specify @lru_cache without arguments, using the default size of 128 (3.8+).

Besides, you can get the cache parameters (3.9+):

# {'maxsize': 256, 'typed': False}

If you don't mind the memory usage, you can use the unlimited @cache instead of @lru_cache (3.9+).

New @cached_property decorator caches the calculated object property (3.8+):

import functools
import statistics

class Dataset:
    def __init__(self, seq):
        self._data = tuple(seq)

    def stdev(self):
        return statistics.stdev(self._data)

dataset = Dataset(range(1_000_000))

# kinda slow

# blazingly fast

And @singledispatchmethod overloads the method depending on the parameter type (3.8+):

import functools

class Divider:
    def divide(self, dividend, divisor):
        raise NotImplementedError("Do not know how to divide those")

    def _(self, dividend: int, divisor: int):
        return dividend // divisor

    def _(self, dividend: str, divisor: int):
        # this is really stupid, I know
        newlen = len(dividend) // divisor
        return dividend[:newlen]

divider = Divider()
divider.divide(10, 2)
# 5

divider.divide("hello world", 2)
# 'hello'

Smells like Java to me.


Contributed by: Raymond HettingerCarl MeyerEthan Smith


glob module searches for files and directories that match the template.

Now thanks to the root_dir parameter in glob() and iglob() functions you can specify the root directory of the search (3.10+):

import glob
import os

# '/'

glob.glob("*", root_dir="/usr")
# ['local', 'share', 'bin', 'lib', 'sbin', 'src']

It's a small thing, but it's nice.


Contributed by: Serhiy Storchaka


graphlib module works with graphs. And you know what? This is a brand-new module! (3.9+)

So far, it has only one feature — topological graph sorting (an ordering of vertices such that for any u → v, the vertex u comes before v):

from graphlib import TopologicalSorter

graph = {"Diane": {"Bob", "Cindy"}, "Cindy": {"Alice"}, "Bob": {"Alice"}}
# Alice → Bob → Diane
#     ↳ Cindy ↗

sorter = TopologicalSorter(graph)
# ['Alice', 'Cindy', 'Bob', 'Diane']


Contributed by: Pablo GalindoTim PetersLarry Hastings


itertools module provides a variety of iterators for memory-efficient collection processing.

One of them is the accumulate() function, which calculates the rolling aggregate. Now it allows the initial parameter, which sets the initial value (3.8+):

import itertools

seq = [7, 11, 19, 42]

accumulator = itertools.accumulate(seq)
# [7, 18, 37, 79]

accumulator = itertools.accumulate(seq, initial=100)
# [100, 107, 118, 137, 179]

And the shiny new pairwise() function traverses the collection and yields pairs of consecutive elements (3.10+):

import itertools

seq = [7, 11, 19, 42]
pairer = itertools.pairwise(seq)

# [(7, 11), (11, 19), (19, 42)]


Contributed by: Lisa RoachRaymond Hettinger


math module includes an abundance of mathematical functions.

There are a lot of news here:

  • dist() calculates the Euclidean distance between points (3.8+);
  • perm() and comb() count the number of permutations and combinations (3.8+);
  • lcm() computes the least common multiple (3.9+);
  • gcd() now computes the greatest common divisor for an arbitrary number of arguments (3.9+).
import math

math.dist((1,1), (4, 5))
# 5.0

math.perm(5, 2)
# 20

math.comb(5, 2)
# 10

math.lcm(9, 27, 60)
# 540

math.gcd(9, 27, 60)
# 3

And prod() multiplies the sequence elements (3.8+):

import math

seq = range(3, 9)
# 20160


Contributed by: Raymond HettingerYash AggarwalKeller FuchsSerhiy StorchakaMark DickinsonAnanthakrishnanPablo Galindo


random module handles random numbers.

New randbytes() method generates a random byte string (3.9+):

import random

# b'\x8b\xd4\x8f\xc9'


Contributed by: Victor Stinner


shlex module splits the string into tokens according to the Unix command line rules.

And now it also joins the tokens back into the string — thanks to the join() function (3.8+):

import shlex

tokens = ["echo", "-n", "Python is awesome"]
# "echo -n 'Python is awesome'"


Contributed by: Bo Bayles


shutil module works with files and directories: copies, moves and deletes them.

Copying directories has now become a little more convenient — kudos to the dirs_exist_ok parameter in the copytree() function (3.8+). If it is on, the function allows the target directory to exist:

from pathlib import Path
import shutil

tmp = Path("/tmp")

src = tmp.joinpath("src")
# /tmp/src
# /tmp/src/src.txt

dst = tmp.joinpath("dst")
# /tmp/dst

shutil.copytree(src, dst)
# FileExistsError: [Errno 17] File exists: '/tmp/dst'
shutil.copytree(src, dst, dirs_exist_ok=True)
# PosixPath('/tmp/dst')


Contributed by: Josh Bronson


statistics module handles mathematical statistics. Like math, it has greatly improved in recent releases. Not scipy yet, but it's not the kindergarten version Python had in 3.4.

See for yourself:

  • fmean() computes the arithmetic mean (like mean(), only faster) (3.8+);
  • geometric_mean() computes the geometric mean (3.8+);
  • multimode() returns the modes (the most frequent values in the dataset), even if there are multiple ones (in contrast to mode()) (3.8+);
  • quantiles() splits the dataset into quantiles and returns the cut points (3.8+).
import statistics

seq = list(range(1, 10))

# 5.0

# 4.147166274396913

# [1, 2, 3, 4, 5, 6, 7, 8, 9]
statistics.multimode("python is awesome")
# ['o', ' ', 's', 'e']

# [2.5, 5.0, 7.5]

NormalDist describes the normal distribution of a random variable (3.8+):

from statistics import NormalDist

birth_weights = NormalDist.from_samples([2.5, 3.1, 2.1, 2.4, 2.7, 3.5])
drug_effects = NormalDist(0.4, 0.15)
combined = birth_weights + drug_effects

round(combined.mean, 1)
# 3.1

round(combined.stdev, 1)
# 0.5

The module received Pearson correlation() and covariance() functions (3.10+):

import statistics

x = [1, 2, 3, 4, 5, 6, 7, 8, 9]
y = [9, 8, 7, 6, 5, 4, 3, 2, 1]

statistics.correlation(x, x)
# 1.0

statistics.correlation(x, y)
# -1.0

statistics.covariance(x, x)
# 7.5

statistics.covariance(x, y)
# -7.5

And even the linear_regression() calculator (3.10+):

import statistics

movies_by_year = {
    2000: 371,
    2003: 507,
    2006: 608,
    2009: 520,
    2012: 669,
    2015: 708,
    2018: 873,
    2021: 403,

x = movies_by_year.keys()
y = movies_by_year.values()
slope, intercept = statistics.linear_regression(x, y)

year_2022 = round(slope * 2022 + intercept)
# 697

By the way, the statistics module is also famous for its excellent documentation. Check it out.


Contributed by: Raymond HettingerSteven D’ApranoTimothy Wolodzko


zoneinfo module provides information about time zones around the world. Another new module! (3.9+)

Before the zoneinfo appearance, Python had a single ascetic timezone.utc time zone. Well, not anymore:

import datetime as dt
from zoneinfo import ZoneInfo

utc = dt.datetime(2022, 9, 13, hour=21, tzinfo=dt.timezone.utc)
# 2022-09-13 21:00:00+00:00

paris = utc.astimezone(ZoneInfo("Europe/Paris"))
# 2022-09-13 23:00:00+02:00

tokyo = utc.astimezone(ZoneInfo("Asia/Tokyo"))
# 2022-09-14 06:00:00+09:00

sydney = utc.astimezone(ZoneInfo("Australia/Sydney"))
# 2022-09-14 07:00:00+10:00


Contributed by: Paul Ganssle


We have reviewed as many as 17 modules contributed by 27 devs — and this is without taking into account asyncio, typing and many other lower-level ones. As you can see, the standard library is actively developing. And the new features are quite reasonable. I hope you will find the described novelties useful!

I would also like to specifically thank the contributors for their amazing work:

  • Carl Meyer for the functools.cached_property() decorator;
  • Dennis Sweeney for the str.removeprefix() and str.removesuffix() methods;
  • Ethan Smith for the functools.singledispatchmethod() decorator;
  • Filipe Laíns for the base64.b32hexencode() and base64.b32hexdecode() functions;
  • Lisa Roach for the Fraction.as_integer_ratio() method and itertools.accumulate() improvements;
  • Niklas Fiekas for the int.bit_count() method;
  • Pablo Galindo for the whole graphlib module and function;
  • Paul Ganssle for the whole zoneinfo module;
  • Raymond Hettinger for lots of functions in the statistics module, itertools.pairwise() function, key parameter in the bisect module and his community work;
  • Serhiy Storchaka and Yash Aggarwal for the combinatorics in the math module;
  • Timothy Wolodzko for the covariance(), correlation(), and linear_regression() functions in the statistics module;
  • Victor Stinner for the random.randbytes() method.

★ Subscribe to keep up with new posts.