mirror of https://github.com/yt-dlp/yt-dlp
Compare commits
170 Commits
2024.03.10
...
master
Author | SHA1 | Date |
---|---|---|
bashonly | 2e5a47da40 | 1 day ago |
bashonly | 5fdd13006a | 1 day ago |
bashonly | 03334d639d | 2 days ago |
sepro | 8b46ad4d8b | 2 days ago |
Ben Galliart | bef9a9e536 | 3 days ago |
github-actions[bot] | 111b61ddef | 4 days ago |
trueauracoral | 12b248ce60 | 4 days ago |
bashonly | 5e3e19c93c | 4 days ago |
bashonly | c53c2e40fd | 5 days ago |
sepro | ae2194e1dd | 5 days ago |
sepro | 26603d0b34 | 5 days ago |
github-actions[bot] | ed274b60b1 | 5 days ago |
bashonly | ae2af1104f | 5 days ago |
Simon Sawicki | 5c019f6328 | 5 days ago |
ocococococ | 5a2eebc767 | 5 days ago |
imanoreotwe | 119d41f270 | 5 days ago |
bashonly | 347f13dd9b | 5 days ago |
coletdjnz | 96a134dea6 | 5 days ago |
Simon Sawicki | a4da9db87b | 5 days ago |
Simon Sawicki | e897bd8292 | 6 days ago |
HobbyistDev | a2e9031605 | 6 days ago |
Finn R. Gärtner | 3ba8de62d6 | 6 days ago |
bashonly | 0d067e77c3 | 6 days ago |
bashonly | 1463945ae5 | 6 days ago |
bashonly | c92e4e625e | 6 days ago |
bashonly | 90d2da311b | 7 days ago |
sepro | 3779f2a307 | 1 week ago |
c-basalt | 63b569bc5e | 1 week ago |
kclauhk | 82f4f4444e | 1 week ago |
Mozi | eead3bbc01 | 1 week ago |
BohwaZ | 5bbfdb7c99 | 1 week ago |
TuxCoder | 0dd53faeca | 1 week ago |
six | be7db1a5a8 | 1 week ago |
HobbyistDev | 65e709d235 | 1 week ago |
Amir Y. Perehodnik | 06cb063839 | 1 week ago |
panatexxa | 296df0da1d | 1 week ago |
vtexier | 7b5674949f | 1 week ago |
bashonly | f2816634e3 | 1 week ago |
bashonly | beaf832c7a | 1 week ago |
bashonly | eef1e9f44f | 1 week ago |
bashonly | 78c57cc0e0 | 1 week ago |
Simon Sawicki | 3f7999533e | 1 week ago |
bashonly | 4ccd73fea0 | 2 weeks ago |
bashonly | 3584b8390b | 2 weeks ago |
bashonly | 6e36d17f40 | 2 weeks ago |
coletdjnz | c36513f1be | 2 weeks ago |
bashonly | 3e35aa32c7 | 2 weeks ago |
coletdjnz | 53b4d44f55 | 2 weeks ago |
bashonly | c999bac02c | 2 weeks ago |
coletdjnz | 12d8ea8246 | 2 weeks ago |
Justin Keogh | 8e15177b41 | 2 weeks ago |
Roeniss Moon | dd9ad97b1f | 2 weeks ago |
minamotorin | 61b17437dc | 2 weeks ago |
kylegustavo | 7975ddf245 | 2 weeks ago |
Podiumnoche | 6d8a53d870 | 2 weeks ago |
bashonly | 4813173e45 | 2 weeks ago |
bashonly | 41ba4a808b | 2 weeks ago |
Mozi | 351dc0bc33 | 3 weeks ago |
feederbox826 | 518c1afc15 | 3 weeks ago |
WyohKnott | 85ec2a337a | 3 weeks ago |
Jake Finley | b207d26f83 | 3 weeks ago |
sepro | 01395a3434 | 3 weeks ago |
Haxy | cf212d0a33 | 3 weeks ago |
alard | 6db96268c5 | 3 weeks ago |
Eric Lam | 800a43983e | 3 weeks ago |
DaPotato69 | 7e4259dff0 | 3 weeks ago |
Stefan Lobbenmeier | f1f158976e | 3 weeks ago |
llamasblade | 31b417e1d1 | 3 weeks ago |
Hugo Azevedo | fc2879ecb0 | 3 weeks ago |
rrgomes | 0a1a8e3005 | 3 weeks ago |
c-basalt | 4cc99d7b6c | 3 weeks ago |
coletdjnz | 3c7a287e28 | 3 weeks ago |
sepro | 98d71d8c5e | 3 weeks ago |
kclauhk | 00a9f2e1f7 | 3 weeks ago |
Mozi | 73f12119b5 | 3 weeks ago |
Alexandre Huot | 6b54cccdcb | 3 weeks ago |
src-tinkerer | c4b87dd885 | 3 weeks ago |
fireattack | 2338827072 | 3 weeks ago |
fireattack | 06d52c8731 | 3 weeks ago |
sepro | df5c9e733a | 3 weeks ago |
Mozi | b38018b781 | 3 weeks ago |
Rasmus Antons | 145dc6f656 | 3 weeks ago |
bashonly | 5904853ae5 | 4 weeks ago |
Chris Caruso | c8bf48f3a8 | 4 weeks ago |
The-MAGI | 351368cb9a | 4 weeks ago |
sepro | 96da952504 | 4 weeks ago |
bashonly | bec9a59e8e | 4 weeks ago |
bashonly | 036e0d92c6 | 4 weeks ago |
bashonly | cb2fb4a643 | 4 weeks ago |
bashonly | 231c2eacc4 | 4 weeks ago |
bashonly | c4853655cb | 4 weeks ago |
Simon Sawicki | ac817bc83e | 1 month ago |
bashonly | 1a366403d9 | 1 month ago |
Simon Sawicki | 7e26bd53f9 | 1 month ago |
Simon Sawicki | 64766459e3 | 1 month ago |
bashonly | 89f535e265 | 1 month ago |
bashonly | ff38a011d5 | 1 month ago |
bashonly | 8056a3026e | 1 month ago |
Simon Sawicki | 3ee1194288 | 1 month ago |
bashonly | e3b42d8b1b | 1 month ago |
bashonly | c9ce57d9bf | 1 month ago |
bashonly | 02483bea1c | 1 month ago |
bashonly | 315b354429 | 2 months ago |
bashonly | 0c21c53885 | 2 months ago |
github-actions[bot] | 168e72dcd3 | 2 months ago |
Simon Sawicki | ff07792676 | 2 months ago |
bashonly | 216f6a3cb5 | 2 months ago |
bashonly | b19ae095fd | 2 months ago |
Simon Sawicki | 9590cc6b47 | 2 months ago |
luiso1979 | 79a451e576 | 2 months ago |
Leo Heitmann Ruiz | df0e138fc0 | 2 months ago |
bashonly | 2e94602f24 | 2 months ago |
bashonly | 4af9d5c2f6 | 2 months ago |
John Victor | 36b240f9a7 | 2 months ago |
bashonly | fc53ec13ff | 2 months ago |
Dmitry Meyer | 2ab2651a4a | 2 months ago |
bashonly | b15b0c1d21 | 2 months ago |
bashonly | c8a61a9100 | 2 months ago |
Mozi | f2fd449b46 | 2 months ago |
Tomoka1 | 9415f1a5ef | 2 months ago |
bashonly | a48cc86d6f | 2 months ago |
bytedream | 954e57e405 | 2 months ago |
Dong Heon Hee | 9073ae6458 | 2 months ago |
Offert4324 | 4cd9e251b9 | 2 months ago |
bashonly | 0ae16ceb18 | 2 months ago |
bashonly | 443e206ec4 | 2 months ago |
bashonly | 4c3b7a0769 | 2 months ago |
bashonly | 16be117729 | 2 months ago |
trainman261 | b49d5ffc53 | 2 months ago |
HobbyistDev | 36baaa10e0 | 2 months ago |
Kacper Michajłow | 02f93ff51b | 2 months ago |
Mozi | c59de48e2b | 2 months ago |
Mozi | 0284f1fee2 | 2 months ago |
bashonly | e8032503b9 | 2 months ago |
bashonly | 97362712a1 | 2 months ago |
bashonly | 246571ae1d | 2 months ago |
Simon Sawicki | 32abfb00bd | 2 months ago |
pukkandan | c305a25c1b | 2 months ago |
pukkandan | e3a3ed8a98 | 2 months ago |
pukkandan | a25a424323 | 2 months ago |
sepro | 86e3b82261 | 2 months ago |
pukkandan | e7b17fce14 | 2 months ago |
bashonly | a2d0840739 | 2 months ago |
pukkandan | 86a972033e | 2 months ago |
bashonly | 50c2935231 | 2 months ago |
bashonly | 0df63cce69 | 2 months ago |
bashonly | 63f685f341 | 2 months ago |
Simon Sawicki | 3699eeb67c | 2 months ago |
Simon Sawicki | 979ce2e786 | 2 months ago |
bashonly | 58dd0f8d1e | 2 months ago |
bashonly | cb61e20c26 | 2 months ago |
bashonly | 9c42b7eef5 | 2 months ago |
coletdjnz | e5d4f11104 | 2 months ago |
src-tinkerer | bc2b8c0596 | 2 months ago |
sta1us | aa7e9ae4f4 | 2 months ago |
Shreyas Minocha | 07f5b2f757 | 2 months ago |
Daniel Vogt | ff349ff94a | 2 months ago |
Hasan Rüzgar | f859ed3ba1 | 2 months ago |
Aron Buzinkay | 17d248a587 | 2 months ago |
sepro | 388c979ac6 | 2 months ago |
sepro | 22e4dfacb6 | 2 months ago |
Trustin | 86d2f4d248 | 3 months ago |
coletdjnz | 52f5be1f1e | 3 months ago |
coletdjnz | 0b81d4d252 | 3 months ago |
coletdjnz | f849d77ab5 | 3 months ago |
bashonly | f2868b26e9 | 3 months ago |
bashonly | be77923ffe | 3 months ago |
bashonly | 8c05b3ebae | 3 months ago |
jazz1611 | 0da66980d3 | 3 months ago |
bashonly | 17b96974a3 | 3 months ago |
File diff suppressed because one or more lines are too long
Before Width: | Height: | Size: 24 KiB After Width: | Height: | Size: 15 KiB |
@ -0,0 +1,14 @@
|
||||
repos:
|
||||
- repo: local
|
||||
hooks:
|
||||
- id: linter
|
||||
name: Apply linter fixes
|
||||
entry: ruff check --fix .
|
||||
language: system
|
||||
types: [python]
|
||||
require_serial: true
|
||||
- id: format
|
||||
name: Apply formatting fixes
|
||||
entry: autopep8 --in-place .
|
||||
language: system
|
||||
types: [python]
|
@ -0,0 +1,9 @@
|
||||
repos:
|
||||
- repo: local
|
||||
hooks:
|
||||
- id: fix
|
||||
name: Apply code fixes
|
||||
entry: hatch fmt
|
||||
language: system
|
||||
types: [python]
|
||||
require_serial: true
|
@ -0,0 +1,10 @@
|
||||
services:
|
||||
static:
|
||||
build: static
|
||||
environment:
|
||||
channel: ${channel}
|
||||
origin: ${origin}
|
||||
version: ${version}
|
||||
volumes:
|
||||
- ~/build:/build
|
||||
- ../..:/yt-dlp
|
@ -0,0 +1,21 @@
|
||||
FROM alpine:3.19 as base
|
||||
|
||||
RUN apk --update add --no-cache \
|
||||
build-base \
|
||||
python3 \
|
||||
pipx \
|
||||
;
|
||||
|
||||
RUN pipx install pyinstaller
|
||||
# Requires above step to prepare the shared venv
|
||||
RUN ~/.local/share/pipx/shared/bin/python -m pip install -U wheel
|
||||
RUN apk --update add --no-cache \
|
||||
scons \
|
||||
patchelf \
|
||||
binutils \
|
||||
;
|
||||
RUN pipx install staticx
|
||||
|
||||
WORKDIR /yt-dlp
|
||||
COPY entrypoint.sh /entrypoint.sh
|
||||
ENTRYPOINT /entrypoint.sh
|
@ -0,0 +1,13 @@
|
||||
#!/bin/ash
|
||||
set -e
|
||||
|
||||
source ~/.local/share/pipx/venvs/pyinstaller/bin/activate
|
||||
python -m devscripts.install_deps --include secretstorage
|
||||
python -m devscripts.make_lazy_extractors
|
||||
python devscripts/update-version.py -c "${channel}" -r "${origin}" "${version}"
|
||||
python -m bundle.pyinstaller
|
||||
deactivate
|
||||
|
||||
source ~/.local/share/pipx/venvs/staticx/bin/activate
|
||||
staticx /yt-dlp/dist/yt-dlp_linux /build/yt-dlp_linux
|
||||
deactivate
|
@ -1,4 +0,0 @@
|
||||
@echo off
|
||||
|
||||
>&2 echo run_tests.bat is deprecated. Please use `devscripts/run_tests.py` instead
|
||||
python %~dp0run_tests.py %~1
|
@ -1,4 +0,0 @@
|
||||
#!/usr/bin/env sh
|
||||
|
||||
>&2 echo 'run_tests.sh is deprecated. Please use `devscripts/run_tests.py` instead'
|
||||
python3 devscripts/run_tests.py "$1"
|
@ -0,0 +1,26 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
# Allow direct execution
|
||||
import os
|
||||
import sys
|
||||
|
||||
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
||||
|
||||
from pathlib import Path
|
||||
|
||||
from devscripts.make_changelog import create_changelog, create_parser
|
||||
from devscripts.utils import read_file, read_version, write_file
|
||||
|
||||
# Always run after devscripts/update-version.py, and run before `make doc|pypi-files|tar|all`
|
||||
|
||||
if __name__ == '__main__':
|
||||
parser = create_parser()
|
||||
parser.description = 'Update an existing changelog file with an entry for a new release'
|
||||
parser.add_argument(
|
||||
'--changelog-path', type=Path, default=Path(__file__).parent.parent / 'Changelog.md',
|
||||
help='path to the Changelog file')
|
||||
args = parser.parse_args()
|
||||
new_entry = create_changelog(args)
|
||||
|
||||
header, sep, changelog = read_file(args.changelog_path).partition('\n### ')
|
||||
write_file(args.changelog_path, f'{header}{sep}{read_version()}\n{new_entry}\n{sep}{changelog}')
|
@ -1,17 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
# Allow execution from anywhere
|
||||
import os
|
||||
import sys
|
||||
|
||||
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
|
||||
|
||||
import warnings
|
||||
|
||||
from bundle.pyinstaller import main
|
||||
|
||||
warnings.warn(DeprecationWarning('`pyinst.py` is deprecated and will be removed in a future version. '
|
||||
'Use `bundle.pyinstaller` instead'))
|
||||
|
||||
if __name__ == '__main__':
|
||||
main()
|
@ -1,36 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
|
||||
# Allow execution from anywhere
|
||||
import os
|
||||
import sys
|
||||
|
||||
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
|
||||
|
||||
import warnings
|
||||
|
||||
|
||||
if sys.argv[1:2] == ['py2exe']:
|
||||
warnings.warn(DeprecationWarning('`setup.py py2exe` is deprecated and will be removed in a future version. '
|
||||
'Use `bundle.py2exe` instead'))
|
||||
|
||||
import bundle.py2exe
|
||||
|
||||
bundle.py2exe.main()
|
||||
|
||||
elif 'build_lazy_extractors' in sys.argv:
|
||||
warnings.warn(DeprecationWarning('`setup.py build_lazy_extractors` is deprecated and will be removed in a future version. '
|
||||
'Use `devscripts.make_lazy_extractors` instead'))
|
||||
|
||||
import subprocess
|
||||
|
||||
os.chdir(sys.path[0])
|
||||
print('running build_lazy_extractors')
|
||||
subprocess.run([sys.executable, 'devscripts/make_lazy_extractors.py'])
|
||||
|
||||
else:
|
||||
|
||||
print(
|
||||
'ERROR: Building by calling `setup.py` is deprecated. '
|
||||
'Use a build frontend like `build` instead. ',
|
||||
'Refer to https://build.pypa.io for more info', file=sys.stderr)
|
||||
sys.exit(1)
|
@ -0,0 +1,380 @@
|
||||
import abc
|
||||
import base64
|
||||
import contextlib
|
||||
import functools
|
||||
import json
|
||||
import os
|
||||
import random
|
||||
import ssl
|
||||
import threading
|
||||
from http.server import BaseHTTPRequestHandler
|
||||
from socketserver import ThreadingTCPServer
|
||||
|
||||
import pytest
|
||||
|
||||
from test.helper import http_server_port, verify_address_availability
|
||||
from test.test_networking import TEST_DIR
|
||||
from test.test_socks import IPv6ThreadingTCPServer
|
||||
from yt_dlp.dependencies import urllib3
|
||||
from yt_dlp.networking import Request
|
||||
from yt_dlp.networking.exceptions import HTTPError, ProxyError, SSLError
|
||||
|
||||
|
||||
class HTTPProxyAuthMixin:
|
||||
|
||||
def proxy_auth_error(self):
|
||||
self.send_response(407)
|
||||
self.send_header('Proxy-Authenticate', 'Basic realm="test http proxy"')
|
||||
self.end_headers()
|
||||
return False
|
||||
|
||||
def do_proxy_auth(self, username, password):
|
||||
if username is None and password is None:
|
||||
return True
|
||||
|
||||
proxy_auth_header = self.headers.get('Proxy-Authorization', None)
|
||||
if proxy_auth_header is None:
|
||||
return self.proxy_auth_error()
|
||||
|
||||
if not proxy_auth_header.startswith('Basic '):
|
||||
return self.proxy_auth_error()
|
||||
|
||||
auth = proxy_auth_header[6:]
|
||||
|
||||
try:
|
||||
auth_username, auth_password = base64.b64decode(auth).decode().split(':', 1)
|
||||
except Exception:
|
||||
return self.proxy_auth_error()
|
||||
|
||||
if auth_username != (username or '') or auth_password != (password or ''):
|
||||
return self.proxy_auth_error()
|
||||
return True
|
||||
|
||||
|
||||
class HTTPProxyHandler(BaseHTTPRequestHandler, HTTPProxyAuthMixin):
|
||||
def __init__(self, *args, proxy_info=None, username=None, password=None, request_handler=None, **kwargs):
|
||||
self.username = username
|
||||
self.password = password
|
||||
self.proxy_info = proxy_info
|
||||
super().__init__(*args, **kwargs)
|
||||
|
||||
def do_GET(self):
|
||||
if not self.do_proxy_auth(self.username, self.password):
|
||||
self.server.close_request(self.request)
|
||||
return
|
||||
if self.path.endswith('/proxy_info'):
|
||||
payload = json.dumps(self.proxy_info or {
|
||||
'client_address': self.client_address,
|
||||
'connect': False,
|
||||
'connect_host': None,
|
||||
'connect_port': None,
|
||||
'headers': dict(self.headers),
|
||||
'path': self.path,
|
||||
'proxy': ':'.join(str(y) for y in self.connection.getsockname()),
|
||||
})
|
||||
self.send_response(200)
|
||||
self.send_header('Content-Type', 'application/json; charset=utf-8')
|
||||
self.send_header('Content-Length', str(len(payload)))
|
||||
self.end_headers()
|
||||
self.wfile.write(payload.encode())
|
||||
else:
|
||||
self.send_response(404)
|
||||
self.end_headers()
|
||||
|
||||
self.server.close_request(self.request)
|
||||
|
||||
|
||||
if urllib3:
|
||||
import urllib3.util.ssltransport
|
||||
|
||||
class SSLTransport(urllib3.util.ssltransport.SSLTransport):
|
||||
"""
|
||||
Modified version of urllib3 SSLTransport to support server side SSL
|
||||
|
||||
This allows us to chain multiple TLS connections.
|
||||
"""
|
||||
|
||||
def __init__(self, socket, ssl_context, server_hostname=None, suppress_ragged_eofs=True, server_side=False):
|
||||
self.incoming = ssl.MemoryBIO()
|
||||
self.outgoing = ssl.MemoryBIO()
|
||||
|
||||
self.suppress_ragged_eofs = suppress_ragged_eofs
|
||||
self.socket = socket
|
||||
|
||||
self.sslobj = ssl_context.wrap_bio(
|
||||
self.incoming,
|
||||
self.outgoing,
|
||||
server_hostname=server_hostname,
|
||||
server_side=server_side
|
||||
)
|
||||
self._ssl_io_loop(self.sslobj.do_handshake)
|
||||
|
||||
@property
|
||||
def _io_refs(self):
|
||||
return self.socket._io_refs
|
||||
|
||||
@_io_refs.setter
|
||||
def _io_refs(self, value):
|
||||
self.socket._io_refs = value
|
||||
|
||||
def shutdown(self, *args, **kwargs):
|
||||
self.socket.shutdown(*args, **kwargs)
|
||||
else:
|
||||
SSLTransport = None
|
||||
|
||||
|
||||
class HTTPSProxyHandler(HTTPProxyHandler):
|
||||
def __init__(self, request, *args, **kwargs):
|
||||
certfn = os.path.join(TEST_DIR, 'testcert.pem')
|
||||
sslctx = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER)
|
||||
sslctx.load_cert_chain(certfn, None)
|
||||
if isinstance(request, ssl.SSLSocket):
|
||||
request = SSLTransport(request, ssl_context=sslctx, server_side=True)
|
||||
else:
|
||||
request = sslctx.wrap_socket(request, server_side=True)
|
||||
super().__init__(request, *args, **kwargs)
|
||||
|
||||
|
||||
class HTTPConnectProxyHandler(BaseHTTPRequestHandler, HTTPProxyAuthMixin):
|
||||
protocol_version = 'HTTP/1.1'
|
||||
default_request_version = 'HTTP/1.1'
|
||||
|
||||
def __init__(self, *args, username=None, password=None, request_handler=None, **kwargs):
|
||||
self.username = username
|
||||
self.password = password
|
||||
self.request_handler = request_handler
|
||||
super().__init__(*args, **kwargs)
|
||||
|
||||
def do_CONNECT(self):
|
||||
if not self.do_proxy_auth(self.username, self.password):
|
||||
self.server.close_request(self.request)
|
||||
return
|
||||
self.send_response(200)
|
||||
self.end_headers()
|
||||
proxy_info = {
|
||||
'client_address': self.client_address,
|
||||
'connect': True,
|
||||
'connect_host': self.path.split(':')[0],
|
||||
'connect_port': int(self.path.split(':')[1]),
|
||||
'headers': dict(self.headers),
|
||||
'path': self.path,
|
||||
'proxy': ':'.join(str(y) for y in self.connection.getsockname()),
|
||||
}
|
||||
self.request_handler(self.request, self.client_address, self.server, proxy_info=proxy_info)
|
||||
self.server.close_request(self.request)
|
||||
|
||||
|
||||
class HTTPSConnectProxyHandler(HTTPConnectProxyHandler):
|
||||
def __init__(self, request, *args, **kwargs):
|
||||
certfn = os.path.join(TEST_DIR, 'testcert.pem')
|
||||
sslctx = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER)
|
||||
sslctx.load_cert_chain(certfn, None)
|
||||
request = sslctx.wrap_socket(request, server_side=True)
|
||||
self._original_request = request
|
||||
super().__init__(request, *args, **kwargs)
|
||||
|
||||
def do_CONNECT(self):
|
||||
super().do_CONNECT()
|
||||
self.server.close_request(self._original_request)
|
||||
|
||||
|
||||
@contextlib.contextmanager
|
||||
def proxy_server(proxy_server_class, request_handler, bind_ip=None, **proxy_server_kwargs):
|
||||
server = server_thread = None
|
||||
try:
|
||||
bind_address = bind_ip or '127.0.0.1'
|
||||
server_type = ThreadingTCPServer if '.' in bind_address else IPv6ThreadingTCPServer
|
||||
server = server_type(
|
||||
(bind_address, 0), functools.partial(proxy_server_class, request_handler=request_handler, **proxy_server_kwargs))
|
||||
server_port = http_server_port(server)
|
||||
server_thread = threading.Thread(target=server.serve_forever)
|
||||
server_thread.daemon = True
|
||||
server_thread.start()
|
||||
if '.' not in bind_address:
|
||||
yield f'[{bind_address}]:{server_port}'
|
||||
else:
|
||||
yield f'{bind_address}:{server_port}'
|
||||
finally:
|
||||
server.shutdown()
|
||||
server.server_close()
|
||||
server_thread.join(2.0)
|
||||
|
||||
|
||||
class HTTPProxyTestContext(abc.ABC):
|
||||
REQUEST_HANDLER_CLASS = None
|
||||
REQUEST_PROTO = None
|
||||
|
||||
def http_server(self, server_class, *args, **kwargs):
|
||||
return proxy_server(server_class, self.REQUEST_HANDLER_CLASS, *args, **kwargs)
|
||||
|
||||
@abc.abstractmethod
|
||||
def proxy_info_request(self, handler, target_domain=None, target_port=None, **req_kwargs) -> dict:
|
||||
"""return a dict of proxy_info"""
|
||||
|
||||
|
||||
class HTTPProxyHTTPTestContext(HTTPProxyTestContext):
|
||||
# Standard HTTP Proxy for http requests
|
||||
REQUEST_HANDLER_CLASS = HTTPProxyHandler
|
||||
REQUEST_PROTO = 'http'
|
||||
|
||||
def proxy_info_request(self, handler, target_domain=None, target_port=None, **req_kwargs):
|
||||
request = Request(f'http://{target_domain or "127.0.0.1"}:{target_port or "40000"}/proxy_info', **req_kwargs)
|
||||
handler.validate(request)
|
||||
return json.loads(handler.send(request).read().decode())
|
||||
|
||||
|
||||
class HTTPProxyHTTPSTestContext(HTTPProxyTestContext):
|
||||
# HTTP Connect proxy, for https requests
|
||||
REQUEST_HANDLER_CLASS = HTTPSProxyHandler
|
||||
REQUEST_PROTO = 'https'
|
||||
|
||||
def proxy_info_request(self, handler, target_domain=None, target_port=None, **req_kwargs):
|
||||
request = Request(f'https://{target_domain or "127.0.0.1"}:{target_port or "40000"}/proxy_info', **req_kwargs)
|
||||
handler.validate(request)
|
||||
return json.loads(handler.send(request).read().decode())
|
||||
|
||||
|
||||
CTX_MAP = {
|
||||
'http': HTTPProxyHTTPTestContext,
|
||||
'https': HTTPProxyHTTPSTestContext,
|
||||
}
|
||||
|
||||
|
||||
@pytest.fixture(scope='module')
|
||||
def ctx(request):
|
||||
return CTX_MAP[request.param]()
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
'handler', ['Urllib', 'Requests', 'CurlCFFI'], indirect=True)
|
||||
@pytest.mark.parametrize('ctx', ['http'], indirect=True) # pure http proxy can only support http
|
||||
class TestHTTPProxy:
|
||||
def test_http_no_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPProxyHandler) as server_address:
|
||||
with handler(proxies={ctx.REQUEST_PROTO: f'http://{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['connect'] is False
|
||||
assert 'Proxy-Authorization' not in proxy_info['headers']
|
||||
|
||||
def test_http_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPProxyHandler, username='test', password='test') as server_address:
|
||||
with handler(proxies={ctx.REQUEST_PROTO: f'http://test:test@{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert 'Proxy-Authorization' in proxy_info['headers']
|
||||
|
||||
def test_http_bad_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPProxyHandler, username='test', password='test') as server_address:
|
||||
with handler(proxies={ctx.REQUEST_PROTO: f'http://test:bad@{server_address}'}) as rh:
|
||||
with pytest.raises(HTTPError) as exc_info:
|
||||
ctx.proxy_info_request(rh)
|
||||
assert exc_info.value.response.status == 407
|
||||
exc_info.value.response.close()
|
||||
|
||||
def test_http_source_address(self, handler, ctx):
|
||||
with ctx.http_server(HTTPProxyHandler) as server_address:
|
||||
source_address = f'127.0.0.{random.randint(5, 255)}'
|
||||
verify_address_availability(source_address)
|
||||
with handler(proxies={ctx.REQUEST_PROTO: f'http://{server_address}'},
|
||||
source_address=source_address) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['client_address'][0] == source_address
|
||||
|
||||
@pytest.mark.skip_handler('Urllib', 'urllib does not support https proxies')
|
||||
def test_https(self, handler, ctx):
|
||||
with ctx.http_server(HTTPSProxyHandler) as server_address:
|
||||
with handler(verify=False, proxies={ctx.REQUEST_PROTO: f'https://{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['connect'] is False
|
||||
assert 'Proxy-Authorization' not in proxy_info['headers']
|
||||
|
||||
@pytest.mark.skip_handler('Urllib', 'urllib does not support https proxies')
|
||||
def test_https_verify_failed(self, handler, ctx):
|
||||
with ctx.http_server(HTTPSProxyHandler) as server_address:
|
||||
with handler(verify=True, proxies={ctx.REQUEST_PROTO: f'https://{server_address}'}) as rh:
|
||||
# Accept SSLError as may not be feasible to tell if it is proxy or request error.
|
||||
# note: if request proto also does ssl verification, this may also be the error of the request.
|
||||
# Until we can support passing custom cacerts to handlers, we cannot properly test this for all cases.
|
||||
with pytest.raises((ProxyError, SSLError)):
|
||||
ctx.proxy_info_request(rh)
|
||||
|
||||
def test_http_with_idn(self, handler, ctx):
|
||||
with ctx.http_server(HTTPProxyHandler) as server_address:
|
||||
with handler(proxies={ctx.REQUEST_PROTO: f'http://{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh, target_domain='中文.tw')
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['path'].startswith('http://xn--fiq228c.tw')
|
||||
assert proxy_info['headers']['Host'].split(':', 1)[0] == 'xn--fiq228c.tw'
|
||||
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
'handler,ctx', [
|
||||
('Requests', 'https'),
|
||||
('CurlCFFI', 'https'),
|
||||
], indirect=True)
|
||||
class TestHTTPConnectProxy:
|
||||
def test_http_connect_no_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPConnectProxyHandler) as server_address:
|
||||
with handler(verify=False, proxies={ctx.REQUEST_PROTO: f'http://{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['connect'] is True
|
||||
assert 'Proxy-Authorization' not in proxy_info['headers']
|
||||
|
||||
def test_http_connect_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPConnectProxyHandler, username='test', password='test') as server_address:
|
||||
with handler(verify=False, proxies={ctx.REQUEST_PROTO: f'http://test:test@{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert 'Proxy-Authorization' in proxy_info['headers']
|
||||
|
||||
@pytest.mark.skip_handler(
|
||||
'Requests',
|
||||
'bug in urllib3 causes unclosed socket: https://github.com/urllib3/urllib3/issues/3374'
|
||||
)
|
||||
def test_http_connect_bad_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPConnectProxyHandler, username='test', password='test') as server_address:
|
||||
with handler(verify=False, proxies={ctx.REQUEST_PROTO: f'http://test:bad@{server_address}'}) as rh:
|
||||
with pytest.raises(ProxyError):
|
||||
ctx.proxy_info_request(rh)
|
||||
|
||||
def test_http_connect_source_address(self, handler, ctx):
|
||||
with ctx.http_server(HTTPConnectProxyHandler) as server_address:
|
||||
source_address = f'127.0.0.{random.randint(5, 255)}'
|
||||
verify_address_availability(source_address)
|
||||
with handler(proxies={ctx.REQUEST_PROTO: f'http://{server_address}'},
|
||||
source_address=source_address,
|
||||
verify=False) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['client_address'][0] == source_address
|
||||
|
||||
@pytest.mark.skipif(urllib3 is None, reason='requires urllib3 to test')
|
||||
def test_https_connect_proxy(self, handler, ctx):
|
||||
with ctx.http_server(HTTPSConnectProxyHandler) as server_address:
|
||||
with handler(verify=False, proxies={ctx.REQUEST_PROTO: f'https://{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert proxy_info['connect'] is True
|
||||
assert 'Proxy-Authorization' not in proxy_info['headers']
|
||||
|
||||
@pytest.mark.skipif(urllib3 is None, reason='requires urllib3 to test')
|
||||
def test_https_connect_verify_failed(self, handler, ctx):
|
||||
with ctx.http_server(HTTPSConnectProxyHandler) as server_address:
|
||||
with handler(verify=True, proxies={ctx.REQUEST_PROTO: f'https://{server_address}'}) as rh:
|
||||
# Accept SSLError as may not be feasible to tell if it is proxy or request error.
|
||||
# note: if request proto also does ssl verification, this may also be the error of the request.
|
||||
# Until we can support passing custom cacerts to handlers, we cannot properly test this for all cases.
|
||||
with pytest.raises((ProxyError, SSLError)):
|
||||
ctx.proxy_info_request(rh)
|
||||
|
||||
@pytest.mark.skipif(urllib3 is None, reason='requires urllib3 to test')
|
||||
def test_https_connect_proxy_auth(self, handler, ctx):
|
||||
with ctx.http_server(HTTPSConnectProxyHandler, username='test', password='test') as server_address:
|
||||
with handler(verify=False, proxies={ctx.REQUEST_PROTO: f'https://test:test@{server_address}'}) as rh:
|
||||
proxy_info = ctx.proxy_info_request(rh)
|
||||
assert proxy_info['proxy'] == server_address
|
||||
assert 'Proxy-Authorization' in proxy_info['headers']
|
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,444 @@
|
||||
import http.cookies
|
||||
import re
|
||||
import xml.etree.ElementTree
|
||||
|
||||
import pytest
|
||||
|
||||
from yt_dlp.utils import dict_get, int_or_none, str_or_none
|
||||
from yt_dlp.utils.traversal import traverse_obj
|
||||
|
||||
_TEST_DATA = {
|
||||
100: 100,
|
||||
1.2: 1.2,
|
||||
'str': 'str',
|
||||
'None': None,
|
||||
'...': ...,
|
||||
'urls': [
|
||||
{'index': 0, 'url': 'https://www.example.com/0'},
|
||||
{'index': 1, 'url': 'https://www.example.com/1'},
|
||||
],
|
||||
'data': (
|
||||
{'index': 2},
|
||||
{'index': 3},
|
||||
),
|
||||
'dict': {},
|
||||
}
|
||||
|
||||
|
||||
class TestTraversal:
|
||||
def test_traversal_base(self):
|
||||
assert traverse_obj(_TEST_DATA, ('str',)) == 'str', \
|
||||
'allow tuple path'
|
||||
assert traverse_obj(_TEST_DATA, ['str']) == 'str', \
|
||||
'allow list path'
|
||||
assert traverse_obj(_TEST_DATA, (value for value in ("str",))) == 'str', \
|
||||
'allow iterable path'
|
||||
assert traverse_obj(_TEST_DATA, 'str') == 'str', \
|
||||
'single items should be treated as a path'
|
||||
assert traverse_obj(_TEST_DATA, 100) == 100, \
|
||||
'allow int path'
|
||||
assert traverse_obj(_TEST_DATA, 1.2) == 1.2, \
|
||||
'allow float path'
|
||||
assert traverse_obj(_TEST_DATA, None) == _TEST_DATA, \
|
||||
'`None` should not perform any modification'
|
||||
|
||||
def test_traversal_ellipsis(self):
|
||||
assert traverse_obj(_TEST_DATA, ...) == [x for x in _TEST_DATA.values() if x not in (None, {})], \
|
||||
'`...` should give all non discarded values'
|
||||
assert traverse_obj(_TEST_DATA, ('urls', 0, ...)) == list(_TEST_DATA['urls'][0].values()), \
|
||||
'`...` selection for dicts should select all values'
|
||||
assert traverse_obj(_TEST_DATA, (..., ..., 'url')) == ['https://www.example.com/0', 'https://www.example.com/1'], \
|
||||
'nested `...` queries should work'
|
||||
assert traverse_obj(_TEST_DATA, (..., ..., 'index')) == list(range(4)), \
|
||||
'`...` query result should be flattened'
|
||||
assert traverse_obj(iter(range(4)), ...) == list(range(4)), \
|
||||
'`...` should accept iterables'
|
||||
|
||||
def test_traversal_function(self):
|
||||
filter_func = lambda x, y: x == 'urls' and isinstance(y, list)
|
||||
assert traverse_obj(_TEST_DATA, filter_func) == [_TEST_DATA['urls']], \
|
||||
'function as query key should perform a filter based on (key, value)'
|
||||
assert traverse_obj(_TEST_DATA, lambda _, x: isinstance(x[0], str)) == ['str'], \
|
||||
'exceptions in the query function should be catched'
|
||||
assert traverse_obj(iter(range(4)), lambda _, x: x % 2 == 0) == [0, 2], \
|
||||
'function key should accept iterables'
|
||||
# Wrong function signature should raise (debug mode)
|
||||
with pytest.raises(Exception):
|
||||
traverse_obj(_TEST_DATA, lambda a: ...)
|
||||
with pytest.raises(Exception):
|
||||
traverse_obj(_TEST_DATA, lambda a, b, c: ...)
|
||||
|
||||
def test_traversal_set(self):
|
||||
# transformation/type, like `expected_type`
|
||||
assert traverse_obj(_TEST_DATA, (..., {str.upper}, )) == ['STR'], \
|
||||
'Function in set should be a transformation'
|
||||
assert traverse_obj(_TEST_DATA, (..., {str})) == ['str'], \
|
||||
'Type in set should be a type filter'
|
||||
assert traverse_obj(_TEST_DATA, (..., {str, int})) == [100, 'str'], \
|
||||
'Multiple types in set should be a type filter'
|
||||
assert traverse_obj(_TEST_DATA, {dict}) == _TEST_DATA, \
|
||||
'A single set should be wrapped into a path'
|
||||
assert traverse_obj(_TEST_DATA, (..., {str.upper})) == ['STR'], \
|
||||
'Transformation function should not raise'
|
||||
expected = [x for x in map(str_or_none, _TEST_DATA.values()) if x is not None]
|
||||
assert traverse_obj(_TEST_DATA, (..., {str_or_none})) == expected, \
|
||||
'Function in set should be a transformation'
|
||||
assert traverse_obj(_TEST_DATA, ('fail', {lambda _: 'const'})) == 'const', \
|
||||
'Function in set should always be called'
|
||||
# Sets with length < 1 or > 1 not including only types should raise
|
||||
with pytest.raises(Exception):
|
||||
traverse_obj(_TEST_DATA, set())
|
||||
with pytest.raises(Exception):
|
||||
traverse_obj(_TEST_DATA, {str.upper, str})
|
||||
|
||||
def test_traversal_slice(self):
|
||||
_SLICE_DATA = [0, 1, 2, 3, 4]
|
||||
|
||||
assert traverse_obj(_TEST_DATA, ('dict', slice(1))) is None, \
|
||||
'slice on a dictionary should not throw'
|
||||
assert traverse_obj(_SLICE_DATA, slice(1)) == _SLICE_DATA[:1], \
|
||||
'slice key should apply slice to sequence'
|
||||
assert traverse_obj(_SLICE_DATA, slice(1, 2)) == _SLICE_DATA[1:2], \
|
||||
'slice key should apply slice to sequence'
|
||||
assert traverse_obj(_SLICE_DATA, slice(1, 4, 2)) == _SLICE_DATA[1:4:2], \
|
||||
'slice key should apply slice to sequence'
|
||||
|
||||
def test_traversal_alternatives(self):
|
||||
assert traverse_obj(_TEST_DATA, 'fail', 'str') == 'str', \
|
||||
'multiple `paths` should be treated as alternative paths'
|
||||
assert traverse_obj(_TEST_DATA, 'str', 100) == 'str', \
|
||||
'alternatives should exit early'
|
||||
assert traverse_obj(_TEST_DATA, 'fail', 'fail') is None, \
|
||||
'alternatives should return `default` if exhausted'
|
||||
assert traverse_obj(_TEST_DATA, (..., 'fail'), 100) == 100, \
|
||||
'alternatives should track their own branching return'
|
||||
assert traverse_obj(_TEST_DATA, ('dict', ...), ('data', ...)) == list(_TEST_DATA['data']), \
|
||||
'alternatives on empty objects should search further'
|
||||
|
||||
def test_traversal_branching_nesting(self):
|
||||
assert traverse_obj(_TEST_DATA, ('urls', (3, 0), 'url')) == ['https://www.example.com/0'], \
|
||||
'tuple as key should be treated as branches'
|
||||
assert traverse_obj(_TEST_DATA, ('urls', [3, 0], 'url')) == ['https://www.example.com/0'], \
|
||||
'list as key should be treated as branches'
|
||||
assert traverse_obj(_TEST_DATA, ('urls', ((1, 'fail'), (0, 'url')))) == ['https://www.example.com/0'], \
|
||||
'double nesting in path should be treated as paths'
|
||||
assert traverse_obj(['0', [1, 2]], [(0, 1), 0]) == [1], \
|
||||
'do not fail early on branching'
|
||||
expected = ['https://www.example.com/0', 'https://www.example.com/1']
|
||||
assert traverse_obj(_TEST_DATA, ('urls', ((0, ('fail', 'url')), (1, 'url')))) == expected, \
|
||||
'tripple nesting in path should be treated as branches'
|
||||
assert traverse_obj(_TEST_DATA, ('urls', ('fail', (..., 'url')))) == expected, \
|
||||
'ellipsis as branch path start gets flattened'
|
||||
|
||||
def test_traversal_dict(self):
|
||||
assert traverse_obj(_TEST_DATA, {0: 100, 1: 1.2}) == {0: 100, 1: 1.2}, \
|
||||
'dict key should result in a dict with the same keys'
|
||||
expected = {0: 'https://www.example.com/0'}
|
||||
assert traverse_obj(_TEST_DATA, {0: ('urls', 0, 'url')}) == expected, \
|
||||
'dict key should allow paths'
|
||||
expected = {0: ['https://www.example.com/0']}
|
||||
assert traverse_obj(_TEST_DATA, {0: ('urls', (3, 0), 'url')}) == expected, \
|
||||
'tuple in dict path should be treated as branches'
|
||||
assert traverse_obj(_TEST_DATA, {0: ('urls', ((1, 'fail'), (0, 'url')))}) == expected, \
|
||||
'double nesting in dict path should be treated as paths'
|
||||
expected = {0: ['https://www.example.com/1', 'https://www.example.com/0']}
|
||||
assert traverse_obj(_TEST_DATA, {0: ('urls', ((1, ('fail', 'url')), (0, 'url')))}) == expected, \
|
||||
'tripple nesting in dict path should be treated as branches'
|
||||
assert traverse_obj(_TEST_DATA, {0: 'fail'}) == {}, \
|
||||
'remove `None` values when top level dict key fails'
|
||||
assert traverse_obj(_TEST_DATA, {0: 'fail'}, default=...) == {0: ...}, \
|
||||
'use `default` if key fails and `default`'
|
||||
assert traverse_obj(_TEST_DATA, {0: 'dict'}) == {}, \
|
||||
'remove empty values when dict key'
|
||||
assert traverse_obj(_TEST_DATA, {0: 'dict'}, default=...) == {0: ...}, \
|
||||
'use `default` when dict key and `default`'
|
||||
assert traverse_obj(_TEST_DATA, {0: {0: 'fail'}}) == {}, \
|
||||
'remove empty values when nested dict key fails'
|
||||
assert traverse_obj(None, {0: 'fail'}) == {}, \
|
||||
'default to dict if pruned'
|
||||
assert traverse_obj(None, {0: 'fail'}, default=...) == {0: ...}, \
|
||||
'default to dict if pruned and default is given'
|
||||
assert traverse_obj(_TEST_DATA, {0: {0: 'fail'}}, default=...) == {0: {0: ...}}, \
|
||||
'use nested `default` when nested dict key fails and `default`'
|
||||
assert traverse_obj(_TEST_DATA, {0: ('dict', ...)}) == {}, \
|
||||
'remove key if branch in dict key not successful'
|
||||
|
||||
def test_traversal_default(self):
|
||||
_DEFAULT_DATA = {'None': None, 'int': 0, 'list': []}
|
||||
|
||||
assert traverse_obj(_DEFAULT_DATA, 'fail') is None, \
|
||||
'default value should be `None`'
|
||||
assert traverse_obj(_DEFAULT_DATA, 'fail', 'fail', default=...) == ..., \
|
||||
'chained fails should result in default'
|
||||
assert traverse_obj(_DEFAULT_DATA, 'None', 'int') == 0, \
|
||||
'should not short cirquit on `None`'
|
||||
assert traverse_obj(_DEFAULT_DATA, 'fail', default=1) == 1, \
|
||||
'invalid dict key should result in `default`'
|
||||
assert traverse_obj(_DEFAULT_DATA, 'None', default=1) == 1, \
|
||||
'`None` is a deliberate sentinel and should become `default`'
|
||||
assert traverse_obj(_DEFAULT_DATA, ('list', 10)) is None, \
|
||||
'`IndexError` should result in `default`'
|
||||
assert traverse_obj(_DEFAULT_DATA, (..., 'fail'), default=1) == 1, \
|
||||
'if branched but not successful return `default` if defined, not `[]`'
|
||||
assert traverse_obj(_DEFAULT_DATA, (..., 'fail'), default=None) is None, \
|
||||
'if branched but not successful return `default` even if `default` is `None`'
|
||||
assert traverse_obj(_DEFAULT_DATA, (..., 'fail')) == [], \
|
||||
'if branched but not successful return `[]`, not `default`'
|
||||
assert traverse_obj(_DEFAULT_DATA, ('list', ...)) == [], \
|
||||
'if branched but object is empty return `[]`, not `default`'
|
||||
assert traverse_obj(None, ...) == [], \
|
||||
'if branched but object is `None` return `[]`, not `default`'
|
||||
assert traverse_obj({0: None}, (0, ...)) == [], \
|
||||
'if branched but state is `None` return `[]`, not `default`'
|
||||
|
||||
@pytest.mark.parametrize('path', [
|
||||
('fail', ...),
|
||||
(..., 'fail'),
|
||||
100 * ('fail',) + (...,),
|
||||
(...,) + 100 * ('fail',),
|
||||
])
|
||||
def test_traversal_branching(self, path):
|
||||
assert traverse_obj({}, path) == [], \
|
||||
'if branched but state is `None`, return `[]` (not `default`)'
|
||||
assert traverse_obj({}, 'fail', path) == [], \
|
||||
'if branching in last alternative and previous did not match, return `[]` (not `default`)'
|
||||
assert traverse_obj({0: 'x'}, 0, path) == 'x', \
|
||||
'if branching in last alternative and previous did match, return single value'
|
||||
assert traverse_obj({0: 'x'}, path, 0) == 'x', \
|
||||
'if branching in first alternative and non-branching path does match, return single value'
|
||||
assert traverse_obj({}, path, 'fail') is None, \
|
||||
'if branching in first alternative and non-branching path does not match, return `default`'
|
||||
|
||||
def test_traversal_expected_type(self):
|
||||
_EXPECTED_TYPE_DATA = {'str': 'str', 'int': 0}
|
||||
|
||||
assert traverse_obj(_EXPECTED_TYPE_DATA, 'str', expected_type=str) == 'str', \
|
||||
'accept matching `expected_type` type'
|
||||
assert traverse_obj(_EXPECTED_TYPE_DATA, 'str', expected_type=int) is None, \
|
||||
'reject non matching `expected_type` type'
|
||||
assert traverse_obj(_EXPECTED_TYPE_DATA, 'int', expected_type=lambda x: str(x)) == '0', \
|
||||
'transform type using type function'
|
||||
assert traverse_obj(_EXPECTED_TYPE_DATA, 'str', expected_type=lambda _: 1 / 0) is None, \
|
||||
'wrap expected_type fuction in try_call'
|
||||
assert traverse_obj(_EXPECTED_TYPE_DATA, ..., expected_type=str) == ['str'], \
|
||||
'eliminate items that expected_type fails on'
|
||||
assert traverse_obj(_TEST_DATA, {0: 100, 1: 1.2}, expected_type=int) == {0: 100}, \
|
||||
'type as expected_type should filter dict values'
|
||||
assert traverse_obj(_TEST_DATA, {0: 100, 1: 1.2, 2: 'None'}, expected_type=str_or_none) == {0: '100', 1: '1.2'}, \
|
||||
'function as expected_type should transform dict values'
|
||||
assert traverse_obj(_TEST_DATA, ({0: 1.2}, 0, {int_or_none}), expected_type=int) == 1, \
|
||||
'expected_type should not filter non final dict values'
|
||||
assert traverse_obj(_TEST_DATA, {0: {0: 100, 1: 'str'}}, expected_type=int) == {0: {0: 100}}, \
|
||||
'expected_type should transform deep dict values'
|
||||
assert traverse_obj(_TEST_DATA, [({0: '...'}, {0: '...'})], expected_type=type(...)) == [{0: ...}, {0: ...}], \
|
||||
'expected_type should transform branched dict values'
|
||||
assert traverse_obj({1: {3: 4}}, [(1, 2), 3], expected_type=int) == [4], \
|
||||
'expected_type regression for type matching in tuple branching'
|
||||
assert traverse_obj(_TEST_DATA, ['data', ...], expected_type=int) == [], \
|
||||
'expected_type regression for type matching in dict result'
|
||||
|
||||
def test_traversal_get_all(self):
|
||||
_GET_ALL_DATA = {'key': [0, 1, 2]}
|
||||
|
||||
assert traverse_obj(_GET_ALL_DATA, ('key', ...), get_all=False) == 0, \
|
||||
'if not `get_all`, return only first matching value'
|
||||
assert traverse_obj(_GET_ALL_DATA, ..., get_all=False) == [0, 1, 2], \
|
||||
'do not overflatten if not `get_all`'
|
||||
|
||||
def test_traversal_casesense(self):
|
||||
_CASESENSE_DATA = {
|
||||
'KeY': 'value0',
|
||||
0: {
|
||||
'KeY': 'value1',
|
||||
0: {'KeY': 'value2'},
|
||||
},
|
||||
}
|
||||
|
||||
assert traverse_obj(_CASESENSE_DATA, 'key') is None, \
|
||||
'dict keys should be case sensitive unless `casesense`'
|
||||
assert traverse_obj(_CASESENSE_DATA, 'keY', casesense=False) == 'value0', \
|
||||
'allow non matching key case if `casesense`'
|
||||
assert traverse_obj(_CASESENSE_DATA, [0, ('keY',)], casesense=False) == ['value1'], \
|
||||
'allow non matching key case in branch if `casesense`'
|
||||
assert traverse_obj(_CASESENSE_DATA, [0, ([0, 'keY'],)], casesense=False) == ['value2'], \
|
||||
'allow non matching key case in branch path if `casesense`'
|
||||
|
||||
def test_traversal_traverse_string(self):
|
||||
_TRAVERSE_STRING_DATA = {'str': 'str', 1.2: 1.2}
|
||||
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, ('str', 0)) is None, \
|
||||
'do not traverse into string if not `traverse_string`'
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, ('str', 0), traverse_string=True) == 's', \
|
||||
'traverse into string if `traverse_string`'
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, (1.2, 1), traverse_string=True) == '.', \
|
||||
'traverse into converted data if `traverse_string`'
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, ('str', ...), traverse_string=True) == 'str', \
|
||||
'`...` should result in string (same value) if `traverse_string`'
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, ('str', slice(0, None, 2)), traverse_string=True) == 'sr', \
|
||||
'`slice` should result in string if `traverse_string`'
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, ('str', lambda i, v: i or v == "s"), traverse_string=True) == 'str', \
|
||||
'function should result in string if `traverse_string`'
|
||||
assert traverse_obj(_TRAVERSE_STRING_DATA, ('str', (0, 2)), traverse_string=True) == ['s', 'r'], \
|
||||
'branching should result in list if `traverse_string`'
|
||||
assert traverse_obj({}, (0, ...), traverse_string=True) == [], \
|
||||
'branching should result in list if `traverse_string`'
|
||||
assert traverse_obj({}, (0, lambda x, y: True), traverse_string=True) == [], \
|
||||
'branching should result in list if `traverse_string`'
|
||||
assert traverse_obj({}, (0, slice(1)), traverse_string=True) == [], \
|
||||
'branching should result in list if `traverse_string`'
|
||||
|
||||
def test_traversal_re(self):
|
||||
mobj = re.fullmatch(r'0(12)(?P<group>3)(4)?', '0123')
|
||||
assert traverse_obj(mobj, ...) == [x for x in mobj.groups() if x is not None], \
|
||||
'`...` on a `re.Match` should give its `groups()`'
|
||||
assert traverse_obj(mobj, lambda k, _: k in (0, 2)) == ['0123', '3'], \
|
||||
'function on a `re.Match` should give groupno, value starting at 0'
|
||||
assert traverse_obj(mobj, 'group') == '3', \
|
||||
'str key on a `re.Match` should give group with that name'
|
||||
assert traverse_obj(mobj, 2) == '3', \
|
||||
'int key on a `re.Match` should give group with that name'
|
||||
assert traverse_obj(mobj, 'gRoUp', casesense=False) == '3', \
|
||||
'str key on a `re.Match` should respect casesense'
|
||||
assert traverse_obj(mobj, 'fail') is None, \
|
||||
'failing str key on a `re.Match` should return `default`'
|
||||
assert traverse_obj(mobj, 'gRoUpS', casesense=False) is None, \
|
||||
'failing str key on a `re.Match` should return `default`'
|
||||
assert traverse_obj(mobj, 8) is None, \
|
||||
'failing int key on a `re.Match` should return `default`'
|
||||
assert traverse_obj(mobj, lambda k, _: k in (0, 'group')) == ['0123', '3'], \
|
||||
'function on a `re.Match` should give group name as well'
|
||||
|
||||
def test_traversal_xml_etree(self):
|
||||
etree = xml.etree.ElementTree.fromstring('''<?xml version="1.0"?>
|
||||
<data>
|
||||
<country name="Liechtenstein">
|
||||
<rank>1</rank>
|
||||
<year>2008</year>
|
||||
<gdppc>141100</gdppc>
|
||||
<neighbor name="Austria" direction="E"/>
|
||||
<neighbor name="Switzerland" direction="W"/>
|
||||
</country>
|
||||
<country name="Singapore">
|
||||
<rank>4</rank>
|
||||
<year>2011</year>
|
||||
<gdppc>59900</gdppc>
|
||||
<neighbor name="Malaysia" direction="N"/>
|
||||
</country>
|
||||
<country name="Panama">
|
||||
<rank>68</rank>
|
||||
<year>2011</year>
|
||||
<gdppc>13600</gdppc>
|
||||
<neighbor name="Costa Rica" direction="W"/>
|
||||
<neighbor name="Colombia" direction="E"/>
|
||||
</country>
|
||||
</data>''')
|
||||
assert traverse_obj(etree, '') == etree, \
|
||||
'empty str key should return the element itself'
|
||||
assert traverse_obj(etree, 'country') == list(etree), \
|
||||
'str key should lead all children with that tag name'
|
||||
assert traverse_obj(etree, ...) == list(etree), \
|
||||
'`...` as key should return all children'
|
||||
assert traverse_obj(etree, lambda _, x: x[0].text == '4') == [etree[1]], \
|
||||
'function as key should get element as value'
|
||||
assert traverse_obj(etree, lambda i, _: i == 1) == [etree[1]], \
|
||||
'function as key should get index as key'
|
||||
assert traverse_obj(etree, 0) == etree[0], \
|
||||
'int key should return the nth child'
|
||||
expected = ['Austria', 'Switzerland', 'Malaysia', 'Costa Rica', 'Colombia']
|
||||
assert traverse_obj(etree, './/neighbor/@name') == expected, \
|
||||
'`@<attribute>` at end of path should give that attribute'
|
||||
assert traverse_obj(etree, '//neighbor/@fail') == [None, None, None, None, None], \
|
||||
'`@<nonexistant>` at end of path should give `None`'
|
||||
assert traverse_obj(etree, ('//neighbor/@', 2)) == {'name': 'Malaysia', 'direction': 'N'}, \
|
||||
'`@` should give the full attribute dict'
|
||||
assert traverse_obj(etree, '//year/text()') == ['2008', '2011', '2011'], \
|
||||
'`text()` at end of path should give the inner text'
|
||||
assert traverse_obj(etree, '//*[@direction]/@direction') == ['E', 'W', 'N', 'W', 'E'], \
|
||||
'full Python xpath features should be supported'
|
||||
assert traverse_obj(etree, (0, '@name')) == 'Liechtenstein', \
|
||||
'special transformations should act on current element'
|
||||
assert traverse_obj(etree, ('country', 0, ..., 'text()', {int_or_none})) == [1, 2008, 141100], \
|
||||
'special transformations should act on current element'
|
||||
|
||||
def test_traversal_unbranching(self):
|
||||
assert traverse_obj(_TEST_DATA, [(100, 1.2), all]) == [100, 1.2], \
|
||||
'`all` should give all results as list'
|
||||
assert traverse_obj(_TEST_DATA, [(100, 1.2), any]) == 100, \
|
||||
'`any` should give the first result'
|
||||
assert traverse_obj(_TEST_DATA, [100, all]) == [100], \
|
||||
'`all` should give list if non branching'
|
||||
assert traverse_obj(_TEST_DATA, [100, any]) == 100, \
|
||||
'`any` should give single item if non branching'
|
||||
assert traverse_obj(_TEST_DATA, [('dict', 'None', 100), all]) == [100], \
|
||||
'`all` should filter `None` and empty dict'
|
||||
assert traverse_obj(_TEST_DATA, [('dict', 'None', 100), any]) == 100, \
|
||||
'`any` should filter `None` and empty dict'
|
||||
assert traverse_obj(_TEST_DATA, [{
|
||||
'all': [('dict', 'None', 100, 1.2), all],
|
||||
'any': [('dict', 'None', 100, 1.2), any],
|
||||
}]) == {'all': [100, 1.2], 'any': 100}, \
|
||||
'`all`/`any` should apply to each dict path separately'
|
||||
assert traverse_obj(_TEST_DATA, [{
|
||||
'all': [('dict', 'None', 100, 1.2), all],
|
||||
'any': [('dict', 'None', 100, 1.2), any],
|
||||
}], get_all=False) == {'all': [100, 1.2], 'any': 100}, \
|
||||
'`all`/`any` should apply to dict regardless of `get_all`'
|
||||
assert traverse_obj(_TEST_DATA, [('dict', 'None', 100, 1.2), all, {float}]) is None, \
|
||||
'`all` should reset branching status'
|
||||
assert traverse_obj(_TEST_DATA, [('dict', 'None', 100, 1.2), any, {float}]) is None, \
|
||||
'`any` should reset branching status'
|
||||
assert traverse_obj(_TEST_DATA, [('dict', 'None', 100, 1.2), all, ..., {float}]) == [1.2], \
|
||||
'`all` should allow further branching'
|
||||
assert traverse_obj(_TEST_DATA, [('dict', 'None', 'urls', 'data'), any, ..., 'index']) == [0, 1], \
|
||||
'`any` should allow further branching'
|
||||
|
||||
def test_traversal_morsel(self):
|
||||
values = {
|
||||
'expires': 'a',
|
||||
'path': 'b',
|
||||
'comment': 'c',
|
||||
'domain': 'd',
|
||||
'max-age': 'e',
|
||||
'secure': 'f',
|
||||
'httponly': 'g',
|
||||
'version': 'h',
|
||||
'samesite': 'i',
|
||||
}
|
||||
morsel = http.cookies.Morsel()
|
||||
morsel.set('item_key', 'item_value', 'coded_value')
|
||||
morsel.update(values)
|
||||
values['key'] = 'item_key'
|
||||
values['value'] = 'item_value'
|
||||
|
||||
for key, value in values.items():
|
||||
assert traverse_obj(morsel, key) == value, \
|
||||
'Morsel should provide access to all values'
|
||||
assert traverse_obj(morsel, ...) == list(values.values()), \
|
||||
'`...` should yield all values'
|
||||
assert traverse_obj(morsel, lambda k, v: True) == list(values.values()), \
|
||||
'function key should yield all values'
|
||||
assert traverse_obj(morsel, [(None,), any]) == morsel, \
|
||||
'Morsel should not be implicitly changed to dict on usage'
|
||||
|
||||
|
||||
class TestDictGet:
|
||||
def test_dict_get(self):
|
||||
FALSE_VALUES = {
|
||||
'none': None,
|
||||
'false': False,
|
||||
'zero': 0,
|
||||
'empty_string': '',
|
||||
'empty_list': [],
|
||||
}
|
||||
d = {**FALSE_VALUES, 'a': 42}
|
||||
assert dict_get(d, 'a') == 42
|
||||
assert dict_get(d, 'b') is None
|
||||
assert dict_get(d, 'b', 42) == 42
|
||||
assert dict_get(d, ('a',)) == 42
|
||||
assert dict_get(d, ('b', 'a')) == 42
|
||||
assert dict_get(d, ('b', 'c', 'a', 'd')) == 42
|
||||
assert dict_get(d, ('b', 'c')) is None
|
||||
assert dict_get(d, ('b', 'c'), 42) == 42
|
||||
for key, false_value in FALSE_VALUES.items():
|
||||
assert dict_get(d, ('b', 'c', key)) is None
|
||||
assert dict_get(d, ('b', 'c', key), skip_false_values=False) == false_value
|
File diff suppressed because it is too large
Load Diff
@ -1,32 +0,0 @@
|
||||
from .common import InfoExtractor
|
||||
|
||||
|
||||
class CableAVIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://cableav\.tv/(?P<id>[a-zA-Z0-9]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://cableav.tv/lS4iR9lWjN8/',
|
||||
'md5': '7e3fe5e49d61c4233b7f5b0f69b15e18',
|
||||
'info_dict': {
|
||||
'id': 'lS4iR9lWjN8',
|
||||
'ext': 'mp4',
|
||||
'title': '國產麻豆AV 叮叮映畫 DDF001 情欲小說家 - CableAV',
|
||||
'description': '國產AV 480p, 720p 国产麻豆AV 叮叮映画 DDF001 情欲小说家',
|
||||
'thumbnail': r're:^https?://.*\.jpg$',
|
||||
}
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
video_url = self._og_search_video_url(webpage, secure=False)
|
||||
|
||||
formats = self._extract_m3u8_formats(video_url, video_id, 'mp4')
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': self._og_search_title(webpage),
|
||||
'description': self._og_search_description(webpage),
|
||||
'thumbnail': self._og_search_thumbnail(webpage),
|
||||
'formats': formats,
|
||||
}
|
@ -0,0 +1,74 @@
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
determine_ext,
|
||||
int_or_none,
|
||||
parse_iso8601,
|
||||
traverse_obj,
|
||||
urljoin,
|
||||
)
|
||||
|
||||
|
||||
class CaffeineTVIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?caffeine\.tv/[^/?#]+/video/(?P<id>[\da-f-]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://www.caffeine.tv/TsuSurf/video/cffc0a00-e73f-11ec-8080-80017d29f26e',
|
||||
'info_dict': {
|
||||
'id': 'cffc0a00-e73f-11ec-8080-80017d29f26e',
|
||||
'ext': 'mp4',
|
||||
'title': 'GOOOOD MORNINNNNN #highlights',
|
||||
'timestamp': 1654702180,
|
||||
'upload_date': '20220608',
|
||||
'uploader': 'RahJON Wicc',
|
||||
'uploader_id': 'TsuSurf',
|
||||
'duration': 3145,
|
||||
'age_limit': 17,
|
||||
'thumbnail': 'https://www.caffeine.tv/broadcasts/776b6f84-9cd5-42e3-af1d-4a776eeed697/replay/lobby.jpg',
|
||||
'comment_count': int,
|
||||
'view_count': int,
|
||||
'like_count': int,
|
||||
'tags': ['highlights', 'battlerap'],
|
||||
},
|
||||
'params': {
|
||||
'skip_download': 'm3u8',
|
||||
},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
json_data = self._download_json(
|
||||
f'https://api.caffeine.tv/social/public/activity/{video_id}', video_id)
|
||||
broadcast_info = traverse_obj(json_data, ('broadcast_info', {dict})) or {}
|
||||
|
||||
video_url = broadcast_info['video_url']
|
||||
ext = determine_ext(video_url)
|
||||
if ext == 'm3u8':
|
||||
formats = self._extract_m3u8_formats(video_url, video_id, 'mp4')
|
||||
else:
|
||||
formats = [{'url': video_url}]
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'formats': formats,
|
||||
**traverse_obj(json_data, {
|
||||
'like_count': ('like_count', {int_or_none}),
|
||||
'view_count': ('view_count', {int_or_none}),
|
||||
'comment_count': ('comment_count', {int_or_none}),
|
||||
'tags': ('tags', ..., {str}, {lambda x: x or None}),
|
||||
'uploader': ('user', 'name', {str}),
|
||||
'uploader_id': (((None, 'user'), 'username'), {str}, any),
|
||||
'is_live': ('is_live', {bool}),
|
||||
}),
|
||||
**traverse_obj(broadcast_info, {
|
||||
'title': ('broadcast_title', {str}),
|
||||
'duration': ('content_duration', {int_or_none}),
|
||||
'timestamp': ('broadcast_start_time', {parse_iso8601}),
|
||||
'thumbnail': ('preview_image_path', {lambda x: urljoin(url, x)}),
|
||||
}),
|
||||
'age_limit': {
|
||||
# assume Apple Store ratings: https://en.wikipedia.org/wiki/Mobile_software_content_rating_system
|
||||
'FOUR_PLUS': 0,
|
||||
'NINE_PLUS': 9,
|
||||
'TWELVE_PLUS': 12,
|
||||
'SEVENTEEN_PLUS': 17,
|
||||
}.get(broadcast_info.get('content_rating'), 17),
|
||||
}
|
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in New Issue