파이썬 3.4의 새로운 기능

저자

R. David Murray <rdmurray@bitdance.com> (편집자)

이 기사에서는 3.3과 비교하여 파이썬 3.4의 새로운 기능에 관해 설명합니다. 파이썬 3.4는 2014년 3월 16일에 출시되었습니다. 자세한 내용은, changelog를 참조하십시오.

더 보기

PEP 429 -- 파이썬 3.4 출시 일정

요약 -- 배포 주요 사항

새로운 문법 기능:

  • 파이썬 3.4에는 새로운 문법 기능이 추가되지 않았습니다.

기타 새로운 기능:

새로운 라이브러리 모듈:

크게 개선된 라이브러리 모듈:

보안 향상:

CPython 구현 개선:

다른 많은 소규모 개선 사항, CPython 최적화, 폐지 및 잠재적 이식 문제를 포함하여 사용자가 직면하는 변경 사항의 포괄적인 목록을 계속 읽으십시오.

새로운 기능

PEP 453: 파이썬 설치 시 PIP의 명시적 부트 스트래핑

기본적으로 pip 부트 스트랩하기

새로운 (PEP 453에 정의된) ensurepip 모듈은 pip 설치 프로그램을 파이썬 설치와 가상 환경으로 부트 스트랩하는 표준 교차 플랫폼 메커니즘을 제공합니다. 파이썬 3.4.0에 포함된 pip 버전은 pip 1.5.4이며, 향후 3.4.x 유지 보수 릴리스는 포함된 버전을 릴리스 후보를 만드는 시점에 사용 가능한 최신 버전의 pip로 갱신할 것입니다.

기본적으로, pipX와 pipX.Y 명령은 pip 파이썬 패키지와 해당 종속성과 함께 모든 플랫폼에 설치됩니다 (여기서 X.Y는 파이썬 설치 버전을 나타냅니다). 윈도우와 모든 플랫폼의 가상 환경에서 버전이 지정되지 않은 pip 명령도 설치됩니다. 다른 플랫폼에서는, 시스템 전체에 적용되는 버전이 지정되지 않은 pip 명령은 일반적으로 별도로 설치된 파이썬 2 버전을 가리킵니다.

pyvenv 명령 줄 유틸리티와 venv 모듈은 ensurepip 모듈을 사용하여 pip를 가상 환경에서 쉽게 사용할 수 있도록 합니다. 명령 줄 유틸리티를 사용할 때, pip가 기본적으로 설치되지만, venv 모듈을 사용할 때는 pip의 API 설치를 명시적으로 요청해야 합니다.

CPython의 POSIX 시스템에서의 소스 빌드의 경우, make install과 make altinstall 명령은 기본적으로 pip를 부트 스트랩합니다. 이 동작은 구성 옵션을 통해 제어할 수 있으며, Makefile 옵션을 통해 재정의 할 수 있습니다.

윈도우와 Mac OS X에서, CPython 설치 프로그램은 이제 CPython 자체와 함께 pip를 기본적으로 설치합니다 (사용자는 설치 프로세스 중에 설치를 거부할 수 있습니다). 윈도우 사용자는 기본적으로 명령 줄에서 pip를 사용할 수 있도록 자동 PATH 수정을 선택해야합니다, 그렇지 않으면 윈도우 용 파이썬 런처를 통해 py -m pip로 계속 액세스 할 수 있습니다.

PEP 에서 논의되었듯이, 이 명령을 호출할 때 (일반적으로 시스템 패키지 관리자를 사용해서) 해당 플랫폼에 설치하는 방법에 대한 명확하고 간단한 지침을 제공하는 한, 플랫폼 패키저는 기본적으로 이러한 명령을 설치하지 않도록 선택할 수 있습니다.

참고

파이썬 2와 파이썬 3의 병행 설치 간의 충돌을 방지하기 위해, ensurepip가 직접 호출될 때 기본적으로 버전이 지정된 pip3와 pip3.4 명령만 부트 스트랩됩니다 - 버전이 지정되지 않은 pip 명령도 요청하려면 --default-pip 옵션이 필요합니다. pyvenv와 윈도우 설치 프로그램은 해당 환경에서 버전이 지정되지 않은 pip 명령을 사용할 수 있도록 보장하며, 여러 파이썬 설치가 있는 시스템에서 모호함을 피하기 위해 pip는 직접 호출되는 대신 항상 -m 스위치를 통해 호출될 수 있습니다.

문서 변경

이 변경의 일환으로, 문서의 파이썬 모듈 설치하기와 파이썬 모듈 배포하기 섹션이 짧은 시작 및 FAQ 문서로 완전히 재 설계되었습니다. 대부분의 패키징 문서는 이제 파이썬 패키징 사용자 지침서와 개별 프로젝트의 설명서를 유지하는 파이썬 패키징 위원회(Python Packaging Authority)로 옮겨졌습니다.

그러나, 이 마이그레이션은 현재 아직 불완전하므로, 해당 지침서의 레거시 버전은 파이썬 모듈 설치 (레거시 버전)와 파이썬 모듈 배포 (레거시 버전)에서 계속 제공됩니다.

더 보기

PEP 453 -- 파이썬 설치시 pip의 명시적 부트 스트랩

Donald Stufft와 Nick Coghlan이 작성한 PEP, Donald Stufft, Nick Coghlan, Martin von Löwis 및 Ned Deily가 구현했습니다.

PEP 446: 새로 만든어진 파일 기술자는 상속할 수 없습니다

PEP 446은 새로 만들어진 파일 기술자를 상속 불가능으로 만듭니다. 일반적으로, 이것은 응용 프로그램이 원하는 동작입니다: 새 프로세스를 시작할 때, 현재 열려있는 파일이 새 프로세스에서도 열리면 찾기 어려운 모든 종류의 버그로 이어지고, 잠재적으로 보안 문제가 발생할 수 있습니다.

그러나, 상속이 필요한 경우가 있습니다. 이러한 경우를 지원하기 위해, 다음과 같은 새로운 함수와 메서드를 제공합니다:

더 보기

PEP 446 -- 새로 만들어진 파일 기술자를 상속 불가능하게 만듭니다

Victor Stinner가 작성하고 구현 한 PEP.

코덱 처리 개선

처음 도입 된 이후, codecs 모듈은 항상 형 중립 동적 인코딩과 디코딩 시스템으로 작동하도록 설계되었습니다. 그러나, 파이썬 텍스트 모델, 특히 내장 str, bytes 및 bytearray 형에 대한 형 제한 편의 메서드와의 밀접한 결합은, 역사적으로 그 사실을 모호하게 만들었습니다.

상황을 명확히 하기 위한 핵심 단계로, codecs.encode()와 codecs.decode() 편의 함수가 이제 파이썬 2.7, 3.3 및 3.4에서 올바르게 설명되었습니다. 이 함수들은 파이썬 2.4부터 codecs 모듈에 존재했지만 (그리고 회귀 테스트 스위트에서 다루었습니다), 이전에는 실행시간 인트로스펙션을 통해서만 발견할 수 있었습니다.

str, bytes 및 bytearray의 편의 메서드와 달리, codecs 편의 함수는 유니코드 텍스트 인코딩(파이썬 3에서)이나 basestring <-> basestring 변환(파이썬 2에서)으로 제한되지 않고, 파이썬 2와 파이썬 3 모두에서 임의의 코덱을 지원합니다.

파이썬 3.4에서, 인터프리터는 표준 라이브러리에서 제공되는 알려진 비 텍스트 인코딩을 식별하고 적절한 경우 다음과 같은 범용 편의 함수로 사용자를 안내할 수 있습니다:

>>> b"abcdef".decode("hex")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
LookupError: 'hex' is not a text encoding; use codecs.decode() to handle arbitrary codecs

>>> "hello".encode("rot13")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
LookupError: 'rot13' is not a text encoding; use codecs.encode() to handle arbitrary codecs

>>> open("foo.txt", encoding="hex")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
LookupError: 'hex' is not a text encoding; use codecs.open() to handle arbitrary codecs

관련 변경에서, 이전 버전과의 호환성을 깨지 않고 실행 가능할 때마다, 인코딩과 디코딩 연산 중에 발생한 예외는 에러를 생성하는 코덱의 이름을 언급하는 같은 형의 연결된 예외(chained exception)로 래핑됩니다:

>>> import codecs

>>> codecs.decode(b"abcdefgh", "hex")
Traceback (most recent call last):
  File "/usr/lib/python3.4/encodings/hex_codec.py", line 20, in hex_decode
    return (binascii.a2b_hex(input), len(input))
binascii.Error: Non-hexadecimal digit found

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
binascii.Error: decoding with 'hex' codec failed (Error: Non-hexadecimal digit found)

>>> codecs.encode("hello", "bz2")
Traceback (most recent call last):
  File "/usr/lib/python3.4/encodings/bz2_codec.py", line 17, in bz2_encode
    return (bz2.compress(input), len(input))
  File "/usr/lib/python3.4/bz2.py", line 498, in compress
    return comp.compress(data) + comp.flush()
TypeError: 'str' does not support the buffer interface

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: encoding with 'bz2' codec failed (TypeError: 'str' does not support the buffer interface)

마지막으로, 위의 예에서 볼 수 있듯이, 이러한 개선으로 인해 파이썬 3.2에서 자체적으로 복원된 비 유니코드 코덱의 편의 별칭을 복원할 수 있었습니다. 즉, 바이너리 데이터를 16진수 표현 으로부터/으로 인코딩하는 것은 (예를들어) 이제 다음과 같이 작성할 수 있습니다:

>>> from codecs import encode, decode
>>> encode(b"hello", "hex")
b'68656c6c6f'
>>> decode(b"68656c6c6f", "hex")
b'hello'

표준 라이브러리에서 제공되는 바이너리와 텍스트 변환은 바이너리 변환과 텍스트 변환에 자세히 설명되어 있습니다.

(Contributed by Nick Coghlan in bpo-7475, bpo-17827, bpo-17828 and bpo-19619.)

PEP 451: 임포트 시스템을 위한 ModuleSpec 형

PEP 451은 임포트 절차가 모듈을 로드하는 데 사용할 모듈 (즉, 모듈 명세)에 대한 정보의 캡슐화를 제공합니다. 이는 임포트 구현과 여러 임포트 관련 API를 단순화하는 데 도움이됩니다. 변경은 여러 미래의 임포트 관련 구현의 디딤돌이기도 합니다.

PEP의 공개 변경 사항은 완전히 이전 버전과 호환됩니다. 또한 임포터 저자를 제외한 모든 사람에게 투명해야합니다. 키 파인더와 로더 메서드는 폐지되었지만, 계속 작동합니다. 새로운 임포터는 PEP에 설명된 새로운 메서드를 사용해야합니다. 새 메서드를 구현하려면 기존 임포터를 업데이트해야 합니다. 교체해야 하는 메서드와 이들이 교체될 메서드 목록은 폐지 섹션을 참조하십시오.

기타 언어 변경

핵심 파이썬 언어에 대한 몇 가지 작은 변경 사항은 다음과 같습니다:

  • 유니코드 데이터베이스가 UCD 버전 6.3으로 갱신되었습니다.

  • min()과 max()는 이제 평가중인 이터러블에 요소가 없는 경우 반환하는 값을 지정하는 데 사용할 수 있는 default 키워드 전용 인자를 받아들입니다. (Contributed by Julian Berman in bpo-18111.)

  • 모듈 객체는 이제 weakref 가능합니다.

  • 모듈 __file__ 어트리뷰트(및 관련 값)는 이제 스크립트가 상대 경로를 사용하여 직접 실행될 때의 __main__.__file__을 제외하고, 기본적으로 항상 절대 경로를 포함해야합니다. (Contributed by Brett Cannon in bpo-18416.)

  • 모든 UTF-* 코덱(UTF-7 제외)은 이제 surrogatepass 에러 처리기를 사용하지 않는 한 인코딩과 디코딩 중에 서로게이트를 거부합니다. 단, UTF-16 디코더(이것은 유효한 서로게이트 쌍을 허용합니다)와 UTF-16 인코더(비 BMP 문자를 인코딩하는 동안 생성합니다)는 예외입니다. (Contributed by Victor Stinner, Kang-Hao (Kenny) Lu and Serhiy Storchaka in bpo-12892.)

  • 새 독일어 EBCDIC 코덱 cp273. (Contributed by Michael Bierenfeld and Andrew Kuchling in bpo-1097797.)

  • 새 우크라이나어 코덱 cp1125. (Contributed by Serhiy Storchaka in bpo-19668.)

  • bytes.join()과 bytearray.join()은 이제 임의의 버퍼 객체를 인자로 받아들입니다. (Contributed by Antoine Pitrou in bpo-15958.)

  • int 생성자는 이제 __index__ 메서드가 있는 모든 객체를 base 인자로 받아들입니다. (Contributed by Mark Dickinson in bpo-16772.)

  • 이제 프레임 객체에는 프레임에서 지역 변수에 대한 모든 참조를 지우는 clear() 메서드가 있습니다. (Contributed by Antoine Pitrou in bpo-17934.)

  • memoryview는 이제 Sequence로 등록되었으며, reversed() 내장을 지원합니다. (Contributed by Nick Coghlan and Claudiu Popa in bpo-18690 and bpo-19078.)

  • 인자 클리닉의 도입과 inspect와 pydoc 모듈에 대한 다른 변경의 결과로 help()가 보고하는 서명이 여러 경우에서 수정되고 개선되었습니다.

  • __length_hint__()는 이제 공식 언어 명세의 일부입니다 (PEP 424를 참조하십시오). (Contributed by Armin Ronacher in bpo-16148.)

새로운 모듈

asyncio

새로운 (PEP 3156에 정의된) asyncio 모듈은 파이썬 용 표준 플러그 형 이벤트 루프 모델을 제공하여, 표준 라이브러리에서 견고한 비동기 IO 지원을 제공하고, 다른 이벤트 루프 구현이 표준 라이브러리와 서로 상호 운용 할 수 있도록합니다.

파이썬 3.4의 경우, 이 모듈은 잠정 API로 간주됩니다.

더 보기

PEP 3156 -- 비동기 IO 지원 재부팅: "asyncio" 모듈

Guido van Rossum이 주도하여 작성하고 구현한 PEP.

ensurepip

새로운 ensurepip 모듈은 PEP 453 구현을위한 기본 인프라입니다. 정상적인 이벤트 과정에서 최종 사용자는 이 모듈과 상호 작용할 필요가 없지만, 설치나 가상 환경으로의 자동화된 부트스트래핑이 거부 된 경우 pip를 수동으로 부트스트랩하는 데 사용할 수 있습니다.

ensurepip에는 함께 제공되는 CPython 릴리스의 첫 번째 릴리스 후보를 기준으로 최신 pip 번들 사본이 포함되어 있습니다 (유지 보수 릴리스와 기능 릴리스 모두에 적용됩니다). ensurepip는 인터넷에 접속하지 않습니다. 설치시 인터넷 액세스가 가능하면 ensurepip가 실행된 후 번들로 제공되는 pip를 사용하여 pip를 번들된 버전보다 최신 릴리스로 업그레이드 할 수 있습니다. (이러한 업그레이드 된 pip 버전은 별도로 설치된 패키지로 간주되며 파이썬을 제거해도 제거되지 않음에 유의하십시오.)

pip가 이미 설치되어있을 때 호출하면 아무 작업도 수행하지 않기 때문에 모듈 이름은 ensurepip입니다. 또한 기존에 설치된 pip 버전이 번들 사본보다 오래된 경우 pip 번들 사본을 설치하도록하는 --upgrade 옵션도 있습니다.

enum

새로운 (PEP 435에 정의된) enum 모듈은 열거 형의 표준 구현을 제공하여, 다른 모듈(가령 socket)이 더 많은 정보를 제공하는 에러 메시지를 제공하고 불투명 한 정수 상수를 이전 버전과 호환되는 열거 값으로 대체함으로써 더 나은 디버깅 지원을 제공하도록 합니다.

더 보기

PEP 435 -- 파이썬 표준 라이브러리에 Enum 형 추가

Barry Warsaw, Eli Bendersky 및 Ethan Furman이 작성하고, Ethan Furman이 구현한 PEP.

pathlib

새로운 pathlib 모듈은 다양한 운영 체제에 적합한 의미 체계를 가진 파일 시스템 경로를 나타내는 클래스를 제공합니다. 경로 클래스는 I/O 없이 순전히 계산 연산을 제공하는 순수 경로와 순수 경로에서 상속되지만 I/O 연산도 제공하는 구상 경로로 나뉩니다.

파이썬 3.4의 경우, 이 모듈은 잠정 API로 간주됩니다.

더 보기

PEP 428 -- pathlib 모듈 -- 객체 지향 파일 시스템 경로

Antoine Pitrou가 작성하고 구현한 PEP.

selectors

새로운 (PEP 3156 구현의 일부로 만들어진) selectors 모듈은 select 모듈 프리미티브를 기반으로하는, 고수준의 효율적인 I/O 멀티플렉싱을 허용합니다.

statistics

새로운 (PEP 450에서 정의된) statistics 모듈은 표준 라이브러리에서 직접 몇 가지 핵심 통계 기능을 제공합니다. 이 모듈은 데이터 계열의 평균, 중앙값, 모드, 분산 및 표준 편차 계산을 지원합니다.

더 보기

PEP 450 -- 표준 라이브러리에 통계 모듈 추가

Steven D'Aprano가 작성하고 구현한 PEP

tracemalloc

새로운 (PEP 454에 정의된) tracemalloc 모듈은 파이썬에서 할당한 메모리 블록을 추적하는 디버그 도구입니다. 다음 정보를 제공합니다:

  • 객체가 할당된 위치 추적

  • 파일명과 줄 번호 당 할당 된 메모리 블록에 대한 통계: 할당 된 메모리 블록의 총 크기, 수 및 평균 크기

  • 두 스냅 샷 간의 차이를 계산하여 메모리 누수 감지

더 보기

PEP 454 -- 파이썬 메모리 할당을 추적하기 위해 새 tracemalloc 모듈 추가

Victor Stinner가 작성하고 구현한 PEP

개선된 모듈

abc

새 함수 abc.get_cache_token()을 사용하여 객체 그래프의 변경에 영향을 받는 캐시를 무효화 할 시기를 알 수 있습니다. (Contributed by Łukasz Langa in bpo-16832.)

새 클래스 ABC는 메타 클래스로 ABCMeta를 갖습니다. ABC를 베이스 클래스로 사용하는 것은 본질적으로 metaclass=abc.ABCMeta를 지정하는 것과 같은 효과가 있지만, 입력하기가 더 간단하고 읽기 쉽습니다. (Contributed by Bruno Dupuis in bpo-16049.)

aifc

getparams() 메서드는 이제 일반 튜플이 아닌 네임드 튜플을 반환합니다. (Contributed by Claudiu Popa in bpo-17818.)

aifc.open()은 이제 컨텍스트 관리 프로토콜을 지원합니다: with 블록에서 사용하면, 반환된 객체의 close() 메서드가 블록 끝에서 자동으로 호출됩니다. (Contributed by Serhiy Storchacha in bpo-16486.)

writeframesraw()와 writeframes() 메서드는 이제 모든 바이트열류 객체를 받아들입니다. (Contributed by Serhiy Storchaka in bpo-8311.)

argparse

FileType 클래스는 이제 open()으로 전달되는 encoding과 errors 인자를 받아들입니다. (Contributed by Lucas Maystre in bpo-11175.)

audioop

audioop는 이제 24-비트 샘플을 지원합니다. (Contributed by Serhiy Storchaka in bpo-12866.)

새로운 byteswap() 함수는 빅 엔디안 샘플을 리틀 엔디안으로 또는 그 반대로 변환합니다. (Contributed by Serhiy Storchaka in bpo-19641.)

모든 audioop 함수는 이제 모든 바이트열류 객체를 받아들입니다. 문자열은 허용되지 않습니다: 이전에는 작동하지 않았으나, 이제 즉시 에러가 발생합니다. (Contributed by Serhiy Storchaka in bpo-16685.)

base64

base64의 인코딩과 디코딩 함수는 이전에 bytes나 bytearray 인스턴스를 요구했던 곳에서 모든 바이트열류 객체를 받아들입니다. (Contributed by Nick Coghlan in bpo-17839.)

새로운 함수 a85encode(), a85decode(), b85encode() 및 b85decode()는 각각 Ascii85와 git/mercurial Base85 형식으로 바이너리 데이터를 인코딩하고 디코딩하는 기능을 제공합니다. a85 함수에는 Adobe 변형을 포함하여 Ascii85 인코딩 변형과 호환되도록 하는 데 사용할 수 있는 옵션이 있습니다. (Contributed by Martin Morrison, the Mercurial project, Serhiy Storchaka, and Antoine Pitrou in bpo-17618.)

collections

ChainMap.new_child() 메서드는 이제 체인에 추가할 자식 맵을 지정하는 m 인자를 받아들입니다. 이를 통해 기존 매핑 및/또는 사용자 정의 매핑형을 자식으로 사용할 수 있습니다. (Contributed by Vinay Sajip in bpo-16613.)

colorsys

RGB --- YIQ 변환에 대한 계수의 자릿수가 FCC NTSC 버전과 일치하도록 확장되었습니다. 결과의 변화는 1% 미만이어야 하며 다른 곳에서 찾은 결과와 더 잘 일치 할 수 있습니다. (Contributed by Brian Landers and Serhiy Storchaka in bpo-14323.)

contextlib

새로운 contextlib.suppress 컨텍스트 관리자는 단일 문에서 예외를 의도적으로 억제하는 코드의 의도를 명확히 하는 데 도움이 됩니다. (Contributed by Raymond Hettinger in bpo-15806 and Zero Piraeus in bpo-19266.)

새로운 contextlib.redirect_stdout() 컨텍스트 관리자를 사용하면 유틸리티 스크립트가 sys.stdout에 출력을 기록하고 리디렉션 옵션을 제공하지 않는 유연하지 않은 API를 더 쉽게 처리 할 수 있습니다. 컨텍스트 관리자를 사용하여, sys.stdout 출력을 다른 스트림으로 리디렉션 하거나, io.StringIO와 함께 사용하여 문자열로 리디렉션 할 수 있습니다. 후자는, 예를 들어, 명령 줄 인터페이스를 구현하기 위해 작성된 함수에서 출력을 캡처하는 데 특히 유용할 수 있습니다. sys.stdout의 전역 상태에 영향을 미치기 때문에 유틸리티 스크립트에만 권장됩니다. (Contributed by Raymond Hettinger in bpo-15805.)

contextlib 설명서는 또한 일회용, 재사용 가능 및 재진입 가능 컨텍스트 관리자 간의 차이점에 대한 논의를 포함하도록 갱신되었습니다.

dbm

dbm.open() 객체는 이제 컨텍스트 관리 프로토콜을 지원합니다. with 문에서 사용하면, 데이터베이스 객체의 close 메서드가 블록 끝에서 자동으로 호출됩니다. (Contributed by Claudiu Popa and Nick Coghlan in bpo-19282.)

dis

show_code(), dis(), distb() 및 disassemble() 함수는 이제 출력을 기록할 곳을 제어하는 키워드 전용 file 인자를 받아들입니다.

The dis module is now built around an Instruction class that provides object oriented access to the details of each individual bytecode operation.

A new method, get_instructions(), provides an iterator that emits the Instruction stream for a given piece of Python code. Thus it is now possible to write a program that inspects and manipulates a bytecode object in ways different from those provided by the dis module itself. For example:

>>> import dis
>>> for instr in dis.get_instructions(lambda x: x + 1):
...     print(instr.opname)
LOAD_FAST
LOAD_CONST
BINARY_ADD
RETURN_VALUE

The various display tools in the dis module have been rewritten to use these new components.

In addition, a new application-friendly class Bytecode provides an object-oriented API for inspecting bytecode in both in human-readable form and for iterating over instructions. The Bytecode constructor takes the same arguments that get_instruction() does (plus an optional current_offset), and the resulting object can be iterated to produce Instruction objects. But it also has a dis method, equivalent to calling dis on the constructor argument, but returned as a multi-line string:

>>> bytecode = dis.Bytecode(lambda x: x + 1, current_offset=3)
>>> for instr in bytecode:
...     print('{} ({})'.format(instr.opname, instr.opcode))
LOAD_FAST (124)
LOAD_CONST (100)
BINARY_ADD (23)
RETURN_VALUE (83)
>>> bytecode.dis().splitlines()       
['  1           0 LOAD_FAST                0 (x)',
 '      -->     3 LOAD_CONST               1 (1)',
 '              6 BINARY_ADD',
 '              7 RETURN_VALUE']

Bytecode also has a class method, from_traceback(), that provides the ability to manipulate a traceback (that is, print(Bytecode.from_traceback(tb).dis()) is equivalent to distb(tb)).

(Contributed by Nick Coghlan, Ryan Kelly and Thomas Kluyver in bpo-11816 and Claudiu Popa in bpo-17916.)

New function stack_effect() computes the effect on the Python stack of a given opcode and argument, information that is not otherwise available. (Contributed by Larry Hastings in bpo-19722.)

doctest

A new option flag, FAIL_FAST, halts test running as soon as the first failure is detected. (Contributed by R. David Murray and Daniel Urban in bpo-16522.)

The doctest command line interface now uses argparse, and has two new options, -o and -f. -o allows doctest options to be specified on the command line, and -f is a shorthand for -o FAIL_FAST (to parallel the similar option supported by the unittest CLI). (Contributed by R. David Murray in bpo-11390.)

doctest will now find doctests in extension module __doc__ strings. (Contributed by Zachary Ware in bpo-3158.)

email

as_string() now accepts a policy argument to override the default policy of the message when generating a string representation of it. This means that as_string can now be used in more circumstances, instead of having to create and use a generator in order to pass formatting parameters to its flatten method. (Contributed by R. David Murray in bpo-18600.)

New method as_bytes() added to produce a bytes representation of the message in a fashion similar to how as_string produces a string representation. It does not accept the maxheaderlen argument, but does accept the unixfrom and policy arguments. The Message __bytes__() method calls it, meaning that bytes(mymsg) will now produce the intuitive result: a bytes object containing the fully formatted message. (Contributed by R. David Murray in bpo-18600.)

The Message.set_param() message now accepts a replace keyword argument. When specified, the associated header will be updated without changing its location in the list of headers. For backward compatibility, the default is False. (Contributed by R. David Murray in bpo-18891.)

A pair of new subclasses of Message have been added (EmailMessage and MIMEPart), along with a new sub-module, contentmanager and a new policy attribute content_manager. All documentation is currently in the new module, which is being added as part of email's new provisional API. These classes provide a number of new methods that make extracting content from and inserting content into email messages much easier. For details, see the contentmanager documentation and the email: 예제. These API additions complete the bulk of the work that was planned as part of the email6 project. The currently provisional API is scheduled to become final in Python 3.5 (possibly with a few minor additions in the area of error handling). (Contributed by R. David Murray in bpo-18891.)

filecmp

A new clear_cache() function provides the ability to clear the filecmp comparison cache, which uses os.stat() information to determine if the file has changed since the last compare. This can be used, for example, if the file might have been changed and re-checked in less time than the resolution of a particular filesystem's file modification time field. (Contributed by Mark Levitt in bpo-18149.)

New module attribute DEFAULT_IGNORES provides the list of directories that are used as the default value for the ignore parameter of the dircmp() function. (Contributed by Eli Bendersky in bpo-15442.)

functools

The new partialmethod() descriptor brings partial argument application to descriptors, just as partial() provides for normal callables. The new descriptor also makes it easier to get arbitrary callables (including partial() instances) to behave like normal instance methods when included in a class definition. (Contributed by Alon Horev and Nick Coghlan in bpo-4331.)

The new singledispatch() decorator brings support for single-dispatch generic functions to the Python standard library. Where object oriented programming focuses on grouping multiple operations on a common set of data into a class, a generic function focuses on grouping multiple implementations of an operation that allows it to work with different kinds of data.

더 보기

PEP 443 -- Single-dispatch generic functions

Łukasz Langa가 작성하고 구현한 PEP.

total_ordering() now supports a return value of NotImplemented from the underlying comparison function. (Contributed by Katie Miller in bpo-10042.)

A pure-python version of the partial() function is now in the stdlib; in CPython it is overridden by the C accelerated version, but it is available for other implementations to use. (Contributed by Brian Thorne in bpo-12428.)

gc

New function get_stats() returns a list of three per-generation dictionaries containing the collections statistics since interpreter startup. (Contributed by Antoine Pitrou in bpo-16351.)

glob

A new function escape() provides a way to escape special characters in a filename so that they do not become part of the globbing expansion but are instead matched literally. (Contributed by Serhiy Storchaka in bpo-8402.)

hashlib

A new hashlib.pbkdf2_hmac() function provides the PKCS#5 password-based key derivation function 2. (Contributed by Christian Heimes in bpo-18582.)

The name attribute of hashlib hash objects is now a formally supported interface. It has always existed in CPython's hashlib (although it did not return lower case names for all supported hashes), but it was not a public interface and so some other Python implementations have not previously supported it. (Contributed by Jason R. Coombs in bpo-18532.)

hmac

hmac now accepts bytearray as well as bytes for the key argument to the new() function, and the msg parameter to both the new() function and the update() method now accepts any type supported by the hashlib module. (Contributed by Jonas Borgström in bpo-18240.)

The digestmod argument to the hmac.new() function may now be any hash digest name recognized by hashlib. In addition, the current behavior in which the value of digestmod defaults to MD5 is deprecated: in a future version of Python there will be no default value. (Contributed by Christian Heimes in bpo-17276.)

With the addition of block_size and name attributes (and the formal documentation of the digest_size attribute), the hmac module now conforms fully to the PEP 247 API. (Contributed by Christian Heimes in bpo-18775.)

html

New function unescape() function converts HTML5 character references to the corresponding Unicode characters. (Contributed by Ezio Melotti in bpo-2927.)

HTMLParser accepts a new keyword argument convert_charrefs that, when True, automatically converts all character references. For backward-compatibility, its value defaults to False, but it will change to True in a future version of Python, so you are invited to set it explicitly and update your code to use this new feature. (Contributed by Ezio Melotti in bpo-13633.)

The strict argument of HTMLParser is now deprecated. (Contributed by Ezio Melotti in bpo-15114.)

http

send_error() now accepts an optional additional explain parameter which can be used to provide an extended error description, overriding the hardcoded default if there is one. This extended error description will be formatted using the error_message_format attribute and sent as the body of the error response. (Contributed by Karl Cow in bpo-12921.)

The http.server command line interface now has a -b/--bind option that causes the server to listen on a specific address. (Contributed by Malte Swart in bpo-17764.)

idlelib와 IDLE

Since idlelib implements the IDLE shell and editor and is not intended for import by other programs, it gets improvements with every release. See Lib/idlelib/NEWS.txt for a cumulative list of changes since 3.3.0, as well as changes made in future 3.4.x releases. This file is also available from the IDLE Help ‣ About IDLE dialog.

importlib

The InspectLoader ABC defines a new method, source_to_code() that accepts source data and a path and returns a code object. The default implementation is equivalent to compile(data, path, 'exec', dont_inherit=True). (Contributed by Eric Snow and Brett Cannon in bpo-15627.)

InspectLoader also now has a default implementation for the get_code() method. However, it will normally be desirable to override the default implementation for performance reasons. (Contributed by Brett Cannon in bpo-18072.)

The reload() function has been moved from imp to importlib as part of the imp module deprecation. (Contributed by Berker Peksag in bpo-18193.)

importlib.util now has a MAGIC_NUMBER attribute providing access to the bytecode version number. This replaces the get_magic() function in the deprecated imp module. (Contributed by Brett Cannon in bpo-18192.)

New importlib.util functions cache_from_source() and source_from_cache() replace the same-named functions in the deprecated imp module. (Contributed by Brett Cannon in bpo-18194.)

The importlib bootstrap NamespaceLoader now conforms to the InspectLoader ABC, which means that runpy and python -m can now be used with namespace packages. (Contributed by Brett Cannon in bpo-18058.)

importlib.util has a new function decode_source() that decodes source from bytes using universal newline processing. This is useful for implementing InspectLoader.get_source() methods.

importlib.machinery.ExtensionFileLoader now has a get_filename() method. This was inadvertently omitted in the original implementation. (Contributed by Eric Snow in bpo-19152.)

inspect

The inspect module now offers a basic command line interface to quickly display source code and other information for modules, classes and functions. (Contributed by Claudiu Popa and Nick Coghlan in bpo-18626.)

unwrap() makes it easy to unravel wrapper function chains created by functools.wraps() (and any other API that sets the __wrapped__ attribute on a wrapper function). (Contributed by Daniel Urban, Aaron Iles and Nick Coghlan in bpo-13266.)

As part of the implementation of the new enum module, the inspect module now has substantially better support for custom __dir__ methods and dynamic class attributes provided through metaclasses. (Contributed by Ethan Furman in bpo-18929 and bpo-19030.)

getfullargspec() and getargspec() now use the signature() API. This allows them to support a much broader range of callables, including those with __signature__ attributes, those with metadata provided by argument clinic, functools.partial() objects and more. Note that, unlike signature(), these functions still ignore __wrapped__ attributes, and report the already bound first argument for bound methods, so it is still necessary to update your code to use signature() directly if those features are desired. (Contributed by Yury Selivanov in bpo-17481.)

signature() now supports duck types of CPython functions, which adds support for functions compiled with Cython. (Contributed by Stefan Behnel and Yury Selivanov in bpo-17159.)

ipaddress

ipaddress was added to the standard library in Python 3.3 as a provisional API. With the release of Python 3.4, this qualification has been removed: ipaddress is now considered a stable API, covered by the normal standard library requirements to maintain backwards compatibility.

A new is_global property is True if an address is globally routeable. (Contributed by Peter Moody in bpo-17400.)

logging

The TimedRotatingFileHandler has a new atTime parameter that can be used to specify the time of day when rollover should happen. (Contributed by Ronald Oussoren in bpo-9556.)

SocketHandler and DatagramHandler now support Unix domain sockets (by setting port to None). (Contributed by Vinay Sajip in commit ce46195b56a9.)

fileConfig() now accepts a configparser.RawConfigParser subclass instance for the fname parameter. This facilitates using a configuration file when logging configuration is just a part of the overall application configuration, or where the application modifies the configuration before passing it to fileConfig(). (Contributed by Vinay Sajip in bpo-16110.)

Logging configuration data received from a socket via the logging.config.listen() function can now be validated before being processed by supplying a verification function as the argument to the new verify keyword argument. (Contributed by Vinay Sajip in bpo-15452.)

marshal

The default marshal version has been bumped to 3. The code implementing the new version restores the Python2 behavior of recording only one copy of interned strings and preserving the interning on deserialization, and extends this "one copy" ability to any object type (including handling recursive references). This reduces both the size of .pyc files and the amount of memory a module occupies in memory when it is loaded from a .pyc (or .pyo) file. (Contributed by Kristján Valur Jónsson in bpo-16475, with additional speedups by Antoine Pitrou in bpo-19219.)

mmap

mmap objects can now be weakrefed. (Contributed by Valerie Lambert in bpo-4885.)

multiprocessing

On Unix two new start methods, spawn and forkserver, have been added for starting processes using multiprocessing. These make the mixing of processes with threads more robust, and the spawn method matches the semantics that multiprocessing has always used on Windows. New function get_all_start_methods() reports all start methods available on the platform, get_start_method() reports the current start method, and set_start_method() sets the start method. (Contributed by Richard Oudkerk in bpo-8713.)

multiprocessing also now has the concept of a context, which determines how child processes are created. New function get_context() returns a context that uses a specified start method. It has the same API as the multiprocessing module itself, so you can use it to create Pools and other objects that will operate within that context. This allows a framework and an application or different parts of the same application to use multiprocessing without interfering with each other. (Contributed by Richard Oudkerk in bpo-18999.)

Except when using the old fork start method, child processes no longer inherit unneeded handles/file descriptors from their parents (part of bpo-8713).

multiprocessing now relies on runpy (which implements the -m switch) to initialise __main__ appropriately in child processes when using the spawn or forkserver start methods. This resolves some edge cases where combining multiprocessing, the -m command line switch, and explicit relative imports could cause obscure failures in child processes. (Contributed by Nick Coghlan in bpo-19946.)

operator

New function length_hint() provides an implementation of the specification for how the __length_hint__() special method should be used, as part of the PEP 424 formal specification of this language feature. (Contributed by Armin Ronacher in bpo-16148.)

There is now a pure-python version of the operator module available for reference and for use by alternate implementations of Python. (Contributed by Zachary Ware in bpo-16694.)

os

There are new functions to get and set the inheritable flag of a file descriptor (os.get_inheritable(), os.set_inheritable()) or a Windows handle (os.get_handle_inheritable(), os.set_handle_inheritable()).

New function cpu_count() reports the number of CPUs available on the platform on which Python is running (or None if the count can't be determined). The multiprocessing.cpu_count() function is now implemented in terms of this function). (Contributed by Trent Nelson, Yogesh Chaudhari, Victor Stinner, and Charles-François Natali in bpo-17914.)

os.path.samestat() is now available on the Windows platform (and the os.path.samefile() implementation is now shared between Unix and Windows). (Contributed by Brian Curtin in bpo-11939.)

os.path.ismount() now recognizes volumes mounted below a drive root on Windows. (Contributed by Tim Golden in bpo-9035.)

os.open() supports two new flags on platforms that provide them, O_PATH (un-opened file descriptor), and O_TMPFILE (unnamed temporary file; as of 3.4.0 release available only on Linux systems with a kernel version of 3.11 or newer that have uapi headers). (Contributed by Christian Heimes in bpo-18673 and Benjamin Peterson, respectively.)

pdb

pdb has been enhanced to handle generators, yield, and yield from in a more useful fashion. This is especially helpful when debugging asyncio based programs. (Contributed by Andrew Svetlov and Xavier de Gaye in bpo-16596.)

The print command has been removed from pdb, restoring access to the Python print() function from the pdb command line. Python2's pdb did not have a print command; instead, entering print executed the print statement. In Python3 print was mistakenly made an alias for the pdb p command. p, however, prints the repr of its argument, not the str like the Python2 print command did. Worse, the Python3 pdb print command shadowed the Python3 print function, making it inaccessible at the pdb prompt. (Contributed by Connor Osborn in bpo-18764.)

pickle

pickle now supports (but does not use by default) a new pickle protocol, protocol 4. This new protocol addresses a number of issues that were present in previous protocols, such as the serialization of nested classes, very large strings and containers, and classes whose __new__() method takes keyword-only arguments. It also provides some efficiency improvements.

더 보기

PEP 3154 -- Pickle protocol 4

Antoine Pitrou가 작성하고 Alexandre Vassalotti가 구현한 PEP.

plistlib

plistlib now has an API that is similar to the standard pattern for stdlib serialization protocols, with new load(), dump(), loads(), and dumps() functions. (The older API is now deprecated.) In addition to the already supported XML plist format (FMT_XML), it also now supports the binary plist format (FMT_BINARY). (Contributed by Ronald Oussoren and others in bpo-14455.)

poplib

Two new methods have been added to poplib: capa(), which returns the list of capabilities advertised by the POP server, and stls(), which switches a clear-text POP3 session into an encrypted POP3 session if the POP server supports it. (Contributed by Lorenzo Catucci in bpo-4473.)

pprint

The pprint module's PrettyPrinter class and its pformat(), and pprint() functions have a new option, compact, that controls how the output is formatted. Currently setting compact to True means that sequences will be printed with as many sequence elements as will fit within width on each (indented) line. (Contributed by Serhiy Storchaka in bpo-19132.)

Long strings are now wrapped using Python's normal line continuation syntax. (Contributed by Antoine Pitrou in bpo-17150.)

pty

pty.spawn() now returns the status value from os.waitpid() on the child process, instead of None. (Contributed by Gregory P. Smith.)

pydoc

The pydoc module is now based directly on the inspect.signature() introspection API, allowing it to provide signature information for a wider variety of callable objects. This change also means that __wrapped__ attributes are now taken into account when displaying help information. (Contributed by Larry Hastings in bpo-19674.)

The pydoc module no longer displays the self parameter for already bound methods. Instead, it aims to always display the exact current signature of the supplied callable. (Contributed by Larry Hastings in bpo-20710.)

In addition to the changes that have been made to pydoc directly, its handling of custom __dir__ methods and various descriptor behaviours has also been improved substantially by the underlying changes in the inspect module.

As the help() builtin is based on pydoc, the above changes also affect the behaviour of help().

re

New fullmatch() function and regex.fullmatch() method anchor the pattern at both ends of the string to match. This provides a way to be explicit about the goal of the match, which avoids a class of subtle bugs where $ characters get lost during code changes or the addition of alternatives to an existing regular expression. (Contributed by Matthew Barnett in bpo-16203.)

The repr of regex objects now includes the pattern and the flags; the repr of match objects now includes the start, end, and the part of the string that matched. (Contributed by Hugo Lopes Tavares and Serhiy Storchaka in bpo-13592 and bpo-17087.)

resource

New prlimit() function, available on Linux platforms with a kernel version of 2.6.36 or later and glibc of 2.13 or later, provides the ability to query or set the resource limits for processes other than the one making the call. (Contributed by Christian Heimes in bpo-16595.)

On Linux kernel version 2.6.36 or later, there are also some new Linux specific constants: RLIMIT_MSGQUEUE, RLIMIT_NICE, RLIMIT_RTPRIO, RLIMIT_RTTIME, and RLIMIT_SIGPENDING. (Contributed by Christian Heimes in bpo-19324.)

On FreeBSD version 9 and later, there some new FreeBSD specific constants: RLIMIT_SBSIZE, RLIMIT_SWAP, and RLIMIT_NPTS. (Contributed by Claudiu Popa in bpo-19343.)

select

epoll objects now support the context management protocol. When used in a with statement, the close() method will be called automatically at the end of the block. (Contributed by Serhiy Storchaka in bpo-16488.)

devpoll objects now have fileno() and close() methods, as well as a new attribute closed. (Contributed by Victor Stinner in bpo-18794.)

shelve

Shelf instances may now be used in with statements, and will be automatically closed at the end of the with block. (Contributed by Filip Gruszczyński in bpo-13896.)

shutil

copyfile() now raises a specific Error subclass, SameFileError, when the source and destination are the same file, which allows an application to take appropriate action on this specific error. (Contributed by Atsuo Ishimoto and Hynek Schlawack in bpo-1492704.)

smtpd

The SMTPServer and SMTPChannel classes now accept a map keyword argument which, if specified, is passed in to asynchat.async_chat as its map argument. This allows an application to avoid affecting the global socket map. (Contributed by Vinay Sajip in bpo-11959.)

smtplib

SMTPException is now a subclass of OSError, which allows both socket level errors and SMTP protocol level errors to be caught in one try/except statement by code that only cares whether or not an error occurred. (Contributed by Ned Jackson Lovely in bpo-2118.)

socket

The socket module now supports the CAN_BCM protocol on platforms that support it. (Contributed by Brian Thorne in bpo-15359.)

Socket objects have new methods to get or set their inheritable flag, get_inheritable() and set_inheritable().

The socket.AF_* and socket.SOCK_* constants are now enumeration values using the new enum module. This allows meaningful names to be printed during debugging, instead of integer "magic numbers".

The AF_LINK constant is now available on BSD and OSX.

inet_pton() and inet_ntop() are now supported on Windows. (Contributed by Atsuo Ishimoto in bpo-7171.)

sqlite3

A new boolean parameter to the connect() function, uri, can be used to indicate that the database parameter is a uri (see the SQLite URI documentation). (Contributed by poq in bpo-13773.)

ssl

PROTOCOL_TLSv1_1 and PROTOCOL_TLSv1_2 (TLSv1.1 and TLSv1.2 support) have been added; support for these protocols is only available if Python is linked with OpenSSL 1.0.1 or later. (Contributed by Michele Orrù and Antoine Pitrou in bpo-16692.)

New function create_default_context() provides a standard way to obtain an SSLContext whose settings are intended to be a reasonable balance between compatibility and security. These settings are more stringent than the defaults provided by the SSLContext constructor, and may be adjusted in the future, without prior deprecation, if best-practice security requirements change. The new recommended best practice for using stdlib libraries that support SSL is to use create_default_context() to obtain an SSLContext object, modify it if needed, and then pass it as the context argument of the appropriate stdlib API. (Contributed by Christian Heimes in bpo-19689.)

SSLContext method load_verify_locations() accepts a new optional argument cadata, which can be used to provide PEM or DER encoded certificates directly via strings or bytes, respectively. (Contributed by Christian Heimes in bpo-18138.)

New function get_default_verify_paths() returns a named tuple of the paths and environment variables that the set_default_verify_paths() method uses to set OpenSSL's default cafile and capath. This can be an aid in debugging default verification issues. (Contributed by Christian Heimes in bpo-18143.)

SSLContext has a new method, cert_store_stats(), that reports the number of loaded X.509 certs, X.509 CA certs, and certificate revocation lists (crls), as well as a get_ca_certs() method that returns a list of the loaded CA certificates. (Contributed by Christian Heimes in bpo-18147.)

If OpenSSL 0.9.8 or later is available, SSLContext has a new attribute verify_flags that can be used to control the certificate verification process by setting it to some combination of the new constants VERIFY_DEFAULT, VERIFY_CRL_CHECK_LEAF, VERIFY_CRL_CHECK_CHAIN, or VERIFY_X509_STRICT. OpenSSL does not do any CRL verification by default. (Contributed by Christien Heimes in bpo-8813.)

New SSLContext method load_default_certs() loads a set of default "certificate authority" (CA) certificates from default locations, which vary according to the platform. It can be used to load both TLS web server authentication certificates (purpose=SERVER_AUTH) for a client to use to verify a server, and certificates for a server to use in verifying client certificates (purpose=CLIENT_AUTH). (Contributed by Christian Heimes in bpo-19292.)

Two new windows-only functions, enum_certificates() and enum_crls() provide the ability to retrieve certificates, certificate information, and CRLs from the Windows cert store. (Contributed by Christian Heimes in bpo-17134.)

Support for server-side SNI (Server Name Indication) using the new ssl.SSLContext.set_servername_callback() method. (Contributed by Daniel Black in bpo-8109.)

The dictionary returned by SSLSocket.getpeercert() contains additional X509v3 extension items: crlDistributionPoints, calIssuers, and OCSP URIs. (Contributed by Christian Heimes in bpo-18379.)

stat

The stat module is now backed by a C implementation in _stat. A C implementation is required as most of the values aren't standardized and are platform-dependent. (Contributed by Christian Heimes in bpo-11016.)

The module supports new ST_MODE flags, S_IFDOOR, S_IFPORT, and S_IFWHT. (Contributed by Christian Hiemes in bpo-11016.)

struct

New function iter_unpack and a new struct.Struct.iter_unpack() method on compiled formats provide streamed unpacking of a buffer containing repeated instances of a given format of data. (Contributed by Antoine Pitrou in bpo-17804.)

subprocess

check_output() now accepts an input argument that can be used to provide the contents of stdin for the command that is run. (Contributed by Zack Weinberg in bpo-16624.)

getstatus() and getstatusoutput() now work on Windows. This change was actually inadvertently made in 3.3.4. (Contributed by Tim Golden in bpo-10197.)

sunau

The getparams() method now returns a namedtuple rather than a plain tuple. (Contributed by Claudiu Popa in bpo-18901.)

sunau.open() now supports the context management protocol: when used in a with block, the close method of the returned object will be called automatically at the end of the block. (Contributed by Serhiy Storchaka in bpo-18878.)

AU_write.setsampwidth() now supports 24 bit samples, thus adding support for writing 24 sample using the module. (Contributed by Serhiy Storchaka in bpo-19261.)

The writeframesraw() and writeframes() methods now accept any bytes-like object. (Contributed by Serhiy Storchaka in bpo-8311.)

sys

New function sys.getallocatedblocks() returns the current number of blocks allocated by the interpreter. (In CPython with the default --with-pymalloc setting, this is allocations made through the PyObject_Malloc() API.) This can be useful for tracking memory leaks, especially if automated via a test suite. (Contributed by Antoine Pitrou in bpo-13390.)

When the Python interpreter starts in interactive mode, it checks for an __interactivehook__ attribute on the sys module. If the attribute exists, its value is called with no arguments just before interactive mode is started. The check is made after the PYTHONSTARTUP file is read, so it can be set there. The site module sets it to a function that enables tab completion and history saving (in ~/.python-history) if the platform supports readline. If you do not want this (new) behavior, you can override it in PYTHONSTARTUP, sitecustomize, or usercustomize by deleting this attribute from sys (or setting it to some other callable). (Contributed by Éric Araujo and Antoine Pitrou in bpo-5845.)

tarfile

The tarfile module now supports a simple 명령 줄 인터페이스 when called as a script directly or via -m. This can be used to create and extract tarfile archives. (Contributed by Berker Peksag in bpo-13477.)

textwrap

The TextWrapper class has two new attributes/constructor arguments: max_lines, which limits the number of lines in the output, and placeholder, which is a string that will appear at the end of the output if it has been truncated because of max_lines. Building on these capabilities, a new convenience function shorten() collapses all of the whitespace in the input to single spaces and produces a single line of a given width that ends with the placeholder (by default, [...]). (Contributed by Antoine Pitrou and Serhiy Storchaka in bpo-18585 and bpo-18725.)

threading

The Thread object representing the main thread can be obtained from the new main_thread() function. In normal conditions this will be the thread from which the Python interpreter was started. (Contributed by Andrew Svetlov in bpo-18882.)

traceback

A new traceback.clear_frames() function takes a traceback object and clears the local variables in all of the frames it references, reducing the amount of memory consumed. (Contributed by Andrew Kuchling in bpo-1565525.)

types

A new DynamicClassAttribute() descriptor provides a way to define an attribute that acts normally when looked up through an instance object, but which is routed to the class __getattr__ when looked up through the class. This allows one to have properties active on a class, and have virtual attributes on the class with the same name (see Enum for an example). (Contributed by Ethan Furman in bpo-19030.)

urllib

urllib.request now supports data: URLs via the DataHandler class. (Contributed by Mathias Panzenböck in bpo-16423.)

The http method that will be used by a Request class can now be specified by setting a method class attribute on the subclass. (Contributed by Jason R Coombs in bpo-18978.)

Request objects are now reusable: if the full_url or data attributes are modified, all relevant internal properties are updated. This means, for example, that it is now possible to use the same Request object in more than one OpenerDirector.open() call with different data arguments, or to modify a Request's url rather than recomputing it from scratch. There is also a new remove_header() method that can be used to remove headers from a Request. (Contributed by Alexey Kachayev in bpo-16464, Daniel Wozniak in bpo-17485, and Damien Brecht and Senthil Kumaran in bpo-17272.)

HTTPError objects now have a headers attribute that provides access to the HTTP response headers associated with the error. (Contributed by Berker Peksag in bpo-15701.)

unittest

The TestCase class has a new method, subTest(), that produces a context manager whose with block becomes a "sub-test". This context manager allows a test method to dynamically generate subtests by, say, calling the subTest context manager inside a loop. A single test method can thereby produce an indefinite number of separately-identified and separately-counted tests, all of which will run even if one or more of them fail. For example:

class NumbersTest(unittest.TestCase):
    def test_even(self):
        for i in range(6):
            with self.subTest(i=i):
                self.assertEqual(i % 2, 0)

will result in six subtests, each identified in the unittest verbose output with a label consisting of the variable name i and a particular value for that variable (i=0, i=1, etc). See 부분 테스트(subtest)를 사용하여 테스트 반복 구별 짓기 for the full version of this example. (Contributed by Antoine Pitrou in bpo-16997.)

unittest.main() now accepts an iterable of test names for defaultTest, where previously it only accepted a single test name as a string. (Contributed by Jyrki Pulliainen in bpo-15132.)

If SkipTest is raised during test discovery (that is, at the module level in the test file), it is now reported as a skip instead of an error. (Contributed by Zach Ware in bpo-16935.)

discover() now sorts the discovered files to provide consistent test ordering. (Contributed by Martin Melin and Jeff Ramnani in bpo-16709.)

TestSuite now drops references to tests as soon as the test has been run, if the test is successful. On Python interpreters that do garbage collection, this allows the tests to be garbage collected if nothing else is holding a reference to the test. It is possible to override this behavior by creating a TestSuite subclass that defines a custom _removeTestAtIndex method. (Contributed by Tom Wardill, Matt McClure, and Andrew Svetlov in bpo-11798.)

A new test assertion context-manager, assertLogs(), will ensure that a given block of code emits a log message using the logging module. By default the message can come from any logger and have a priority of INFO or higher, but both the logger name and an alternative minimum logging level may be specified. The object returned by the context manager can be queried for the LogRecords and/or formatted messages that were logged. (Contributed by Antoine Pitrou in bpo-18937.)

Test discovery now works with namespace packages (Contributed by Claudiu Popa in bpo-17457.)

unittest.mock objects now inspect their specification signatures when matching calls, which means an argument can now be matched by either position or name, instead of only by position. (Contributed by Antoine Pitrou in bpo-17015.)

mock_open() objects now have readline and readlines methods. (Contributed by Toshio Kuratomi in bpo-17467.)

venv

venv now includes activation scripts for the csh and fish shells. (Contributed by Andrew Svetlov in bpo-15417.)

EnvBuilder and the create() convenience function take a new keyword argument with_pip, which defaults to False, that controls whether or not EnvBuilder ensures that pip is installed in the virtual environment. (Contributed by Nick Coghlan in bpo-19552 as part of the PEP 453 implementation.)

wave

The getparams() method now returns a namedtuple rather than a plain tuple. (Contributed by Claudiu Popa in bpo-17487.)

wave.open() now supports the context management protocol. (Contributed by Claudiu Popa in bpo-17616.)

wave can now write output to unseekable files. (Contributed by David Jones, Guilherme Polo, and Serhiy Storchaka in bpo-5202.)

The writeframesraw() and writeframes() methods now accept any bytes-like object. (Contributed by Serhiy Storchaka in bpo-8311.)

weakref

New WeakMethod class simulates weak references to bound methods. (Contributed by Antoine Pitrou in bpo-14631.)

New finalize class makes it possible to register a callback to be invoked when an object is garbage collected, without needing to carefully manage the lifecycle of the weak reference itself. (Contributed by Richard Oudkerk in bpo-15528.)

The callback, if any, associated with a ref is now exposed via the __callback__ attribute. (Contributed by Mark Dickinson in bpo-17643.)

xml.etree

A new parser, XMLPullParser, allows a non-blocking applications to parse XML documents. An example can be seen at 비 블로킹 구문 분석을 위한 풀(pull) API. (Contributed by Antoine Pitrou in bpo-17741.)

The xml.etree.ElementTree tostring() and tostringlist() functions, and the ElementTree write() method, now have a short_empty_elements keyword-only parameter providing control over whether elements with no content are written in abbreviated (<tag />) or expanded (<tag></tag>) form. (Contributed by Ariel Poliak and Serhiy Storchaka in bpo-14377.)

zipfile

The writepy() method of the PyZipFile class has a new filterfunc option that can be used to control which directories and files are added to the archive. For example, this could be used to exclude test files from the archive. (Contributed by Christian Tismer in bpo-19274.)

The allowZip64 parameter to ZipFile and PyZipfile is now True by default. (Contributed by William Mallard in bpo-17201.)

CPython Implementation Changes

PEP 445: Customization of CPython Memory Allocators

PEP 445 adds new C level interfaces to customize memory allocation in the CPython interpreter.

더 보기

PEP 445 -- Add new APIs to customize Python memory allocators

Victor Stinner가 작성하고 구현 한 PEP.

PEP 442: Safe Object Finalization

PEP 442 removes the current limitations and quirks of object finalization in CPython. With it, objects with __del__() methods, as well as generators with finally clauses, can be finalized when they are part of a reference cycle.

As part of this change, module globals are no longer forcibly set to None during interpreter shutdown in most cases, instead relying on the normal operation of the cyclic garbage collector. This avoids a whole class of interpreter-shutdown-time errors, usually involving __del__ methods, that have plagued Python since the cyclic GC was first introduced.

더 보기

PEP 442 -- Safe object finalization

Antoine Pitrou가 작성하고 구현한 PEP.

PEP 456: Secure and Interchangeable Hash Algorithm

PEP 456 follows up on earlier security fix work done on Python's hash algorithm to address certain DOS attacks to which public facing APIs backed by dictionary lookups may be subject. (See bpo-14621 for the start of the current round of improvements.) The PEP unifies CPython's hash code to make it easier for a packager to substitute a different hash algorithm, and switches Python's default implementation to a SipHash implementation on platforms that have a 64 bit data type. Any performance differences in comparison with the older FNV algorithm are trivial.

The PEP adds additional fields to the sys.hash_info named tuple to describe the hash algorithm in use by the currently executing binary. Otherwise, the PEP does not alter any existing CPython APIs.

PEP 436: Argument Clinic

"Argument Clinic" (PEP 436) is now part of the CPython build process and can be used to simplify the process of defining and maintaining accurate signatures for builtins and standard library extension modules implemented in C.

Some standard library extension modules have been converted to use Argument Clinic in Python 3.4, and pydoc and inspect have been updated accordingly.

It is expected that signature metadata for programmatic introspection will be added to additional callables implemented in C as part of Python 3.4 maintenance releases.

참고

The Argument Clinic PEP is not fully up to date with the state of the implementation. This has been deemed acceptable by the release manager and core development team in this case, as Argument Clinic will not be made available as a public API for third party use in Python 3.4.

더 보기

PEP 436 -- The Argument Clinic DSL

Larry Hastings가 작성하고 구현한 PEP.

Other Build and C API Changes

  • The new PyType_GetSlot() function has been added to the stable ABI, allowing retrieval of function pointers from named type slots when using the limited API. (Contributed by Martin von Löwis in bpo-17162.)

  • The new Py_SetStandardStreamEncoding() pre-initialization API allows applications embedding the CPython interpreter to reliably force a particular encoding and error handler for the standard streams. (Contributed by Bastien Montagne and Nick Coghlan in bpo-16129.)

  • Most Python C APIs that don't mutate string arguments are now correctly marked as accepting const char * rather than char *. (Contributed by Serhiy Storchaka in bpo-1772673.)

  • A new shell version of python-config can be used even when a python interpreter is not available (for example, in cross compilation scenarios).

  • PyUnicode_FromFormat() now supports width and precision specifications for %s, %A, %U, %V, %S, and %R. (Contributed by Ysj Ray and Victor Stinner in bpo-7330.)

  • New function PyStructSequence_InitType2() supplements the existing PyStructSequence_InitType() function. The difference is that it returns 0 on success and -1 on failure.

  • The CPython source can now be compiled using the address sanity checking features of recent versions of GCC and clang: the false alarms in the small object allocator have been silenced. (Contributed by Dhiru Kholia in bpo-18596.)

  • The Windows build now uses Address Space Layout Randomization and Data Execution Prevention. (Contributed by Christian Heimes in bpo-16632.)

  • New function PyObject_LengthHint() is the C API equivalent of operator.length_hint(). (Contributed by Armin Ronacher in bpo-16148.)

Other Improvements

  • The python command has a new option, -I, which causes it to run in "isolated mode", which means that sys.path contains neither the script's directory nor the user's site-packages directory, and all PYTHON* environment variables are ignored (it implies both -s and -E). Other restrictions may also be applied in the future, with the goal being to isolate the execution of a script from the user's environment. This is appropriate, for example, when Python is used to run a system script. On most POSIX systems it can and should be used in the #! line of system scripts. (Contributed by Christian Heimes in bpo-16499.)

  • Tab-completion is now enabled by default in the interactive interpreter on systems that support readline. History is also enabled by default, and is written to (and read from) the file ~/.python-history. (Contributed by Antoine Pitrou and Éric Araujo in bpo-5845.)

  • Invoking the Python interpreter with --version now outputs the version to standard output instead of standard error (bpo-18338). Similar changes were made to argparse (bpo-18920) and other modules that have script-like invocation capabilities (bpo-18922).

  • The CPython Windows installer now adds .py to the PATHEXT variable when extensions are registered, allowing users to run a python script at the windows command prompt by just typing its name without the .py extension. (Contributed by Paul Moore in bpo-18569.)

  • A new make target coverage-report will build python, run the test suite, and generate an HTML coverage report for the C codebase using gcov and lcov.

  • The -R option to the python regression test suite now also checks for memory allocation leaks, using sys.getallocatedblocks(). (Contributed by Antoine Pitrou in bpo-13390.)

  • python -m now works with namespace packages.

  • The stat module is now implemented in C, which means it gets the values for its constants from the C header files, instead of having the values hard-coded in the python module as was previously the case.

  • Loading multiple python modules from a single OS module (.so, .dll) now works correctly (previously it silently returned the first python module in the file). (Contributed by Václav Šmilauer in bpo-16421.)

  • A new opcode, LOAD_CLASSDEREF, has been added to fix a bug in the loading of free variables in class bodies that could be triggered by certain uses of __prepare__. (Contributed by Benjamin Peterson in bpo-17853.)

  • A number of MemoryError-related crashes were identified and fixed by Victor Stinner using his PEP 445-based pyfailmalloc tool (bpo-18408, bpo-18520).

  • The pyvenv command now accepts a --copies option to use copies rather than symlinks even on systems where symlinks are the default. (Contributed by Vinay Sajip in bpo-18807.)

  • The pyvenv command also accepts a --without-pip option to suppress the otherwise-automatic bootstrapping of pip into the virtual environment. (Contributed by Nick Coghlan in bpo-19552 as part of the PEP 453 implementation.)

  • The encoding name is now optional in the value set for the PYTHONIOENCODING environment variable. This makes it possible to set just the error handler, without changing the default encoding. (Contributed by Serhiy Storchaka in bpo-18818.)

  • The bz2, lzma, and gzip module open functions now support x (exclusive creation) mode. (Contributed by Tim Heaney and Vajrasky Kok in bpo-19201, bpo-19222, and bpo-19223.)

Significant Optimizations

  • The UTF-32 decoder is now 3x to 4x faster. (Contributed by Serhiy Storchaka in bpo-14625.)

  • The cost of hash collisions for sets is now reduced. Each hash table probe now checks a series of consecutive, adjacent key/hash pairs before continuing to make random probes through the hash table. This exploits cache locality to make collision resolution less expensive. The collision resolution scheme can be described as a hybrid of linear probing and open addressing. The number of additional linear probes defaults to nine. This can be changed at compile-time by defining LINEAR_PROBES to be any value. Set LINEAR_PROBES=0 to turn-off linear probing entirely. (Contributed by Raymond Hettinger in bpo-18771.)

  • The interpreter starts about 30% faster. A couple of measures lead to the speedup. The interpreter loads fewer modules on startup, e.g. the re, collections and locale modules and their dependencies are no longer imported by default. The marshal module has been improved to load compiled Python code faster. (Contributed by Antoine Pitrou, Christian Heimes and Victor Stinner in bpo-19219, bpo-19218, bpo-19209, bpo-19205 and bpo-9548.)

  • bz2.BZ2File is now as fast or faster than the Python2 version for most cases. lzma.LZMAFile has also been optimized. (Contributed by Serhiy Storchaka and Nadeem Vawda in bpo-16034.)

  • random.getrandbits() is 20%-40% faster for small integers (the most common use case). (Contributed by Serhiy Storchaka in bpo-16674.)

  • By taking advantage of the new storage format for strings, pickling of strings is now significantly faster. (Contributed by Victor Stinner and Antoine Pitrou in bpo-15596.)

  • A performance issue in io.FileIO.readall() has been solved. This particularly affects Windows, and significantly speeds up the case of piping significant amounts of data through subprocess. (Contributed by Richard Oudkerk in bpo-15758.)

  • html.escape() is now 10x faster. (Contributed by Matt Bryant in bpo-18020.)

  • On Windows, the native VirtualAlloc is now used instead of the CRT malloc in obmalloc. Artificial benchmarks show about a 3% memory savings.

  • os.urandom() now uses a lazily-opened persistent file descriptor so as to avoid using many file descriptors when run in parallel from multiple threads. (Contributed by Antoine Pitrou in bpo-18756.)

폐지

This section covers various APIs and other features that have been deprecated in Python 3.4, and will be removed in Python 3.5 or later. In most (but not all) cases, using the deprecated APIs will produce a DeprecationWarning when the interpreter is run with deprecation warnings enabled (for example, by using -Wd).

Deprecations in the Python API

Deprecated Features

  • Running IDLE with the -n flag (no subprocess) is deprecated. However, the feature will not be removed until bpo-18823 is resolved.

  • The site module adding a "site-python" directory to sys.path, if it exists, is deprecated (bpo-19375).

제거

Operating Systems No Longer Supported

Support for the following operating systems has been removed from the source and build tools:

  • OS/2 (bpo-16135).

  • Windows 2000 (changeset e52df05b496a).

  • Windows systems where COMSPEC points to command.com (bpo-14470).

  • VMS (bpo-16136).

API and Feature Removals

The following obsolete and previously deprecated APIs and features have been removed:

  • The unmaintained Misc/TextMate and Misc/vim directories have been removed (see the devguide for suggestions on what to use instead).

  • The SO makefile macro is removed (it was replaced by the SHLIB_SUFFIX and EXT_SUFFIX macros) (bpo-16754).

  • The PyThreadState.tick_counter field has been removed; its value has been meaningless since Python 3.2, when the "new GIL" was introduced (bpo-19199).

  • PyLoader and PyPycLoader have been removed from importlib. (Contributed by Taras Lyapun in bpo-15641.)

  • The strict argument to HTTPConnection and HTTPSConnection has been removed. HTTP 0.9-style "Simple Responses" are no longer supported.

  • The deprecated urllib.request.Request getter and setter methods add_data, has_data, get_data, get_type, get_host, get_selector, set_proxy, get_origin_req_host, and is_unverifiable have been removed (use direct attribute access instead).

  • Support for loading the deprecated TYPE_INT64 has been removed from marshal. (Contributed by Dan Riti in bpo-15480.)

  • inspect.Signature: positional-only parameters are now required to have a valid name.

  • object.__format__() no longer accepts non-empty format strings, it now raises a TypeError instead. Using a non-empty string has been deprecated since Python 3.2. This change has been made to prevent a situation where previously working (but incorrect) code would start failing if an object gained a __format__ method, which means that your code may now raise a TypeError if you are using an 's' format code with objects that do not have a __format__ method that handles it. See bpo-7994 for background.

  • difflib.SequenceMatcher.isbjunk() and difflib.SequenceMatcher.isbpopular() were deprecated in 3.2, and have now been removed: use x in sm.bjunk and x in sm.bpopular, where sm is a SequenceMatcher object (bpo-13248).

Code Cleanups

  • The unused and undocumented internal Scanner class has been removed from the pydoc module.

  • The private and effectively unused _gestalt module has been removed, along with the private platform functions _mac_ver_lookup, _mac_ver_gstalt, and _bcd2str, which would only have ever been called on badly broken OSX systems (see bpo-18393).

  • The hardcoded copies of certain stat constants that were included in the tarfile module namespace have been removed.

파이썬 3.4로 이식하기

This section lists previously described changes and other bugfixes that may require changes to your code.

Changes in 'python' Command Behavior

  • In a posix shell, setting the PATH environment variable to an empty value is equivalent to not setting it at all. However, setting PYTHONPATH to an empty value was not equivalent to not setting it at all: setting PYTHONPATH to an empty value was equivalent to setting it to ., which leads to confusion when reasoning by analogy to how PATH works. The behavior now conforms to the posix convention for PATH.

  • The [X refs, Y blocks] output of a debug (--with-pydebug) build of the CPython interpreter is now off by default. It can be re-enabled using the -X showrefcount option. (Contributed by Ezio Melotti in bpo-17323.)

  • The python command and most stdlib scripts (as well as argparse) now output --version information to stdout instead of stderr (for issue list see Other Improvements above).

파이썬 API의 변경

  • The ABCs defined in importlib.abc now either raise the appropriate exception or return a default value instead of raising NotImplementedError blindly. This will only affect code calling super() and falling through all the way to the ABCs. For compatibility, catch both NotImplementedError or the appropriate exception as needed.

  • The module type now initializes the __package__ and __loader__ attributes to None by default. To determine if these attributes were set in a backwards-compatible fashion, use e.g. getattr(module, '__loader__', None) is not None. (bpo-17115.)

  • importlib.util.module_for_loader() now sets __loader__ and __package__ unconditionally to properly support reloading. If this is not desired then you will need to set these attributes manually. You can use importlib.util.module_to_load() for module management.

  • Import now resets relevant attributes (e.g. __name__, __loader__, __package__, __file__, __cached__) unconditionally when reloading. Note that this restores a pre-3.3 behavior in that it means a module is re-found when re-loaded (bpo-19413).

  • Frozen packages no longer set __path__ to a list containing the package name, they now set it to an empty list. The previous behavior could cause the import system to do the wrong thing on submodule imports if there was also a directory with the same name as the frozen package. The correct way to determine if a module is a package or not is to use hasattr(module, '__path__') (bpo-18065).

  • Frozen modules no longer define a __file__ attribute. It's semantically incorrect for frozen modules to set the attribute as they are not loaded from any explicit location. If you must know that a module comes from frozen code then you can see if the module's __spec__.location is set to 'frozen', check if the loader is a subclass of importlib.machinery.FrozenImporter, or if Python 2 compatibility is necessary you can use imp.is_frozen().

  • py_compile.compile() now raises FileExistsError if the file path it would write to is a symlink or a non-regular file. This is to act as a warning that import will overwrite those files with a regular file regardless of what type of file path they were originally.

  • importlib.abc.SourceLoader.get_source() no longer raises ImportError when the source code being loaded triggers a SyntaxError or UnicodeDecodeError. As ImportError is meant to be raised only when source code cannot be found but it should, it was felt to be over-reaching/overloading of that meaning when the source code is found but improperly structured. If you were catching ImportError before and wish to continue to ignore syntax or decoding issues, catch all three exceptions now.

  • functools.update_wrapper() and functools.wraps() now correctly set the __wrapped__ attribute to the function being wrapped, even if that function also had its __wrapped__ attribute set. This means __wrapped__ attributes now correctly link a stack of decorated functions rather than every __wrapped__ attribute in the chain referring to the innermost function. Introspection libraries that assumed the previous behaviour was intentional can use inspect.unwrap() to access the first function in the chain that has no __wrapped__ attribute.

  • inspect.getfullargspec() has been reimplemented on top of inspect.signature() and hence handles a much wider variety of callable objects than it did in the past. It is expected that additional builtin and extension module callables will gain signature metadata over the course of the Python 3.4 series. Code that assumes that inspect.getfullargspec() will fail on non-Python callables may need to be adjusted accordingly.

  • importlib.machinery.PathFinder now passes on the current working directory to objects in sys.path_hooks for the empty string. This results in sys.path_importer_cache never containing '', thus iterating through sys.path_importer_cache based on sys.path will not find all keys. A module's __file__ when imported in the current working directory will also now have an absolute path, including when using -m with the interpreter (except for __main__.__file__ when a script has been executed directly using a relative path) (Contributed by Brett Cannon in bpo-18416). is specified on the command-line) (bpo-18416).

  • The removal of the strict argument to HTTPConnection and HTTPSConnection changes the meaning of the remaining arguments if you are specifying them positionally rather than by keyword. If you've been paying attention to deprecation warnings your code should already be specifying any additional arguments via keywords.

  • Strings between from __future__ import ... statements now always raise a SyntaxError. Previously if there was no leading docstring, an interstitial string would sometimes be ignored. This brings CPython into compliance with the language spec; Jython and PyPy already were. (bpo-17434).

  • ssl.SSLSocket.getpeercert() and ssl.SSLSocket.do_handshake() now raise an OSError with ENOTCONN when the SSLSocket is not connected, instead of the previous behavior of raising an AttributeError. In addition, getpeercert() will raise a ValueError if the handshake has not yet been done.

  • base64.b32decode() now raises a binascii.Error when the input string contains non-b32-alphabet characters, instead of a TypeError. This particular TypeError was missed when the other TypeErrors were converted. (Contributed by Serhiy Storchaka in bpo-18011.) Note: this change was also inadvertently applied in Python 3.3.3.

  • The file attribute is now automatically closed when the creating cgi.FieldStorage instance is garbage collected. If you were pulling the file object out separately from the cgi.FieldStorage instance and not keeping the instance alive, then you should either store the entire cgi.FieldStorage instance or read the contents of the file before the cgi.FieldStorage instance is garbage collected.

  • Calling read or write on a closed SSL socket now raises an informative ValueError rather than the previous more mysterious AttributeError (bpo-9177).

  • slice.indices() no longer produces an OverflowError for huge values. As a consequence of this fix, slice.indices() now raises a ValueError if given a negative length; previously it returned nonsense values (bpo-14794).

  • The complex constructor, unlike the cmath functions, was incorrectly accepting float values if an object's __complex__ special method returned one. This now raises a TypeError. (bpo-16290.)

  • The int constructor in 3.2 and 3.3 erroneously accepts float values for the base parameter. It is unlikely anyone was doing this, but if so, it will now raise a TypeError (bpo-16772).

  • Defaults for keyword-only arguments are now evaluated after defaults for regular keyword arguments, instead of before. Hopefully no one wrote any code that depends on the previous buggy behavior (bpo-16967).

  • Stale thread states are now cleared after fork(). This may cause some system resources to be released that previously were incorrectly kept perpetually alive (for example, database connections kept in thread-local storage). (bpo-17094.)

  • Parameter names in __annotations__ dicts are now mangled properly, similarly to __kwdefaults__. (Contributed by Yury Selivanov in bpo-20625.)

  • hashlib.hash.name now always returns the identifier in lower case. Previously some builtin hashes had uppercase names, but now that it is a formal public interface the naming has been made consistent (bpo-18532).

  • Because unittest.TestSuite now drops references to tests after they are run, test harnesses that re-use a TestSuite to re-run a set of tests may fail. Test suites should not be re-used in this fashion since it means state is retained between test runs, breaking the test isolation that unittest is designed to provide. However, if the lack of isolation is considered acceptable, the old behavior can be restored by creating a TestSuite subclass that defines a _removeTestAtIndex method that does nothing (see TestSuite.__iter__()) (bpo-11798).

  • unittest now uses argparse for command line parsing. There are certain invalid command forms that used to work that are no longer allowed; in theory this should not cause backward compatibility issues since the disallowed command forms didn't make any sense and are unlikely to be in use.

  • The re.split(), re.findall(), and re.sub() functions, and the group() and groups() methods of match objects now always return a bytes object when the string to be matched is a bytes-like object. Previously the return type matched the input type, so if your code was depending on the return value being, say, a bytearray, you will need to change your code.

  • audioop functions now raise an error immediately if passed string input, instead of failing randomly later on (bpo-16685).

  • The new convert_charrefs argument to HTMLParser currently defaults to False for backward compatibility, but will eventually be changed to default to True. It is recommended that you add this keyword, with the appropriate value, to any HTMLParser calls in your code (bpo-13633).

  • Since the digestmod argument to the hmac.new() function will in the future have no default, all calls to hmac.new() should be changed to explicitly specify a digestmod (bpo-17276).

  • Calling sysconfig.get_config_var() with the SO key, or looking SO up in the results of a call to sysconfig.get_config_vars() is deprecated. This key should be replaced by EXT_SUFFIX or SHLIB_SUFFIX, depending on the context (bpo-19555).

  • Any calls to open functions that specify U should be modified. U is ineffective in Python3 and will eventually raise an error if used. Depending on the function, the equivalent of its old Python2 behavior can be achieved using either a newline argument, or if necessary by wrapping the stream in TextIOWrapper to use its newline argument (bpo-15204).

  • If you use pyvenv in a script and desire that pip not be installed, you must add --without-pip to your command invocation.

  • The default behavior of json.dump() and json.dumps() when an indent is specified has changed: it no longer produces trailing spaces after the item separating commas at the ends of lines. This will matter only if you have tests that are doing white-space-sensitive comparisons of such output (bpo-16333).

  • doctest now looks for doctests in extension module __doc__ strings, so if your doctest test discovery includes extension modules that have things that look like doctests in them you may see test failures you've never seen before when running your tests (bpo-3158).

  • The collections.abc module has been slightly refactored as part of the Python startup improvements. As a consequence of this, it is no longer the case that importing collections automatically imports collections.abc. If your program depended on the (undocumented) implicit import, you will need to add an explicit import collections.abc (bpo-20784).

C API의 변경

  • PyEval_EvalFrameEx(), PyObject_Repr(), and PyObject_Str(), along with some other internal C APIs, now include a debugging assertion that ensures they are not used in situations where they may silently discard a currently active exception. In cases where discarding the active exception is expected and desired (for example, because it has already been saved locally with PyErr_Fetch() or is being deliberately replaced with a different exception), an explicit PyErr_Clear() call will be needed to avoid triggering the assertion when invoking these operations (directly or indirectly) and running against a version of Python that is compiled with assertions enabled.

  • PyErr_SetImportError() now sets TypeError when its msg argument is not set. Previously only NULL was returned with no exception set.

  • The result of the PyOS_ReadlineFunctionPointer callback must now be a string allocated by PyMem_RawMalloc() or PyMem_RawRealloc(), or NULL if an error occurred, instead of a string allocated by PyMem_Malloc() or PyMem_Realloc() (bpo-16742)

  • PyThread_set_key_value() now always set the value. In Python 3.3, the function did nothing if the key already exists (if the current value is a non-NULL pointer).

  • The f_tstate (thread state) field of the PyFrameObject structure has been removed to fix a bug: see bpo-14432 for the rationale.

Changed in 3.4.3

PEP 476: Enabling certificate verification by default for stdlib http clients

http.client and modules which use it, such as urllib.request and xmlrpc.client, will now verify that the server presents a certificate which is signed by a CA in the platform trust store and whose hostname matches the hostname being requested by default, significantly improving security for many applications.

For applications which require the old previous behavior, they can pass an alternate context:

import urllib.request
import ssl

# This disables all verification
context = ssl._create_unverified_context()

# This allows using a specific certificate for the host, which doesn't need
# to be in the trust store
context = ssl.create_default_context(cafile="/path/to/file.crt")

urllib.request.urlopen("https://invalid-cert", context=context)