yt-dlp/test/test_utils.py
Simon Sawicki f9fb3ce86e
[cleanup] Misc (#8598)
Authored by: bashonly, pukkandan, seproDev, Grub4K

Co-authored-by: bashonly <bashonly@protonmail.com>
Co-authored-by: pukkandan <pukkandan.ytdlp@gmail.com>
Co-authored-by: sepro <4618135+seproDev@users.noreply.github.com>
2023-12-30 22:27:36 +01:00

2405 lines
116 KiB
Python
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#!/usr/bin/env python3
# Allow direct execution
import os
import re
import sys
import unittest
import warnings
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
import contextlib
import io
import itertools
import json
import subprocess
import xml.etree.ElementTree
from yt_dlp.compat import (
compat_etree_fromstring,
compat_HTMLParseError,
compat_os_name,
)
from yt_dlp.utils import (
Config,
DateRange,
ExtractorError,
InAdvancePagedList,
LazyList,
OnDemandPagedList,
Popen,
age_restricted,
args_to_str,
base_url,
caesar,
clean_html,
clean_podcast_url,
cli_bool_option,
cli_option,
cli_valueless_option,
date_from_str,
datetime_from_str,
detect_exe_version,
determine_ext,
determine_file_encoding,
dfxp2srt,
dict_get,
encode_base_n,
encode_compat_str,
encodeFilename,
expand_path,
extract_attributes,
extract_basic_auth,
find_xpath_attr,
fix_xml_ampersands,
float_or_none,
format_bytes,
get_compatible_ext,
get_element_by_attribute,
get_element_by_class,
get_element_html_by_attribute,
get_element_html_by_class,
get_element_text_and_html_by_tag,
get_elements_by_attribute,
get_elements_by_class,
get_elements_html_by_attribute,
get_elements_html_by_class,
get_elements_text_and_html_by_attribute,
int_or_none,
intlist_to_bytes,
iri_to_uri,
is_html,
js_to_json,
limit_length,
locked_file,
lowercase_escape,
match_str,
merge_dicts,
mimetype2ext,
month_by_name,
multipart_encode,
ohdave_rsa_encrypt,
orderedSet,
parse_age_limit,
parse_bitrate,
parse_codecs,
parse_count,
parse_dfxp_time_expr,
parse_duration,
parse_filesize,
parse_iso8601,
parse_qs,
parse_resolution,
pkcs1pad,
prepend_extension,
read_batch_urls,
remove_end,
remove_quotes,
remove_start,
render_table,
replace_extension,
rot47,
sanitize_filename,
sanitize_path,
sanitize_url,
shell_quote,
smuggle_url,
str_or_none,
str_to_int,
strip_jsonp,
strip_or_none,
subtitles_filename,
timeconvert,
traverse_obj,
try_call,
unescapeHTML,
unified_strdate,
unified_timestamp,
unsmuggle_url,
update_url_query,
uppercase_escape,
url_basename,
url_or_none,
urlencode_postdata,
urljoin,
urshift,
variadic,
version_tuple,
xpath_attr,
xpath_element,
xpath_text,
xpath_with_ns,
)
from yt_dlp.utils.networking import (
HTTPHeaderDict,
escape_rfc3986,
normalize_url,
remove_dot_segments,
)
class TestUtil(unittest.TestCase):
def test_timeconvert(self):
self.assertTrue(timeconvert('') is None)
self.assertTrue(timeconvert('bougrg') is None)
def test_sanitize_filename(self):
self.assertEqual(sanitize_filename(''), '')
self.assertEqual(sanitize_filename('abc'), 'abc')
self.assertEqual(sanitize_filename('abc_d-e'), 'abc_d-e')
self.assertEqual(sanitize_filename('123'), '123')
self.assertEqual('abcde', sanitize_filename('abc/de'))
self.assertFalse('/' in sanitize_filename('abc/de///'))
self.assertEqual('abc_de', sanitize_filename('abc/<>\\*|de', is_id=False))
self.assertEqual('xxx', sanitize_filename('xxx/<>\\*|', is_id=False))
self.assertEqual('yes no', sanitize_filename('yes? no', is_id=False))
self.assertEqual('this - that', sanitize_filename('this: that', is_id=False))
self.assertEqual(sanitize_filename('AT&T'), 'AT&T')
aumlaut = 'ä'
self.assertEqual(sanitize_filename(aumlaut), aumlaut)
tests = '\u043a\u0438\u0440\u0438\u043b\u043b\u0438\u0446\u0430'
self.assertEqual(sanitize_filename(tests), tests)
self.assertEqual(
sanitize_filename('New World record at 0:12:34'),
'New World record at 0_12_34')
self.assertEqual(sanitize_filename('--gasdgf'), '--gasdgf')
self.assertEqual(sanitize_filename('--gasdgf', is_id=True), '--gasdgf')
self.assertEqual(sanitize_filename('--gasdgf', is_id=False), '_-gasdgf')
self.assertEqual(sanitize_filename('.gasdgf'), '.gasdgf')
self.assertEqual(sanitize_filename('.gasdgf', is_id=True), '.gasdgf')
self.assertEqual(sanitize_filename('.gasdgf', is_id=False), 'gasdgf')
forbidden = '"\0\\/'
for fc in forbidden:
for fbc in forbidden:
self.assertTrue(fbc not in sanitize_filename(fc))
def test_sanitize_filename_restricted(self):
self.assertEqual(sanitize_filename('abc', restricted=True), 'abc')
self.assertEqual(sanitize_filename('abc_d-e', restricted=True), 'abc_d-e')
self.assertEqual(sanitize_filename('123', restricted=True), '123')
self.assertEqual('abc_de', sanitize_filename('abc/de', restricted=True))
self.assertFalse('/' in sanitize_filename('abc/de///', restricted=True))
self.assertEqual('abc_de', sanitize_filename('abc/<>\\*|de', restricted=True))
self.assertEqual('xxx', sanitize_filename('xxx/<>\\*|', restricted=True))
self.assertEqual('yes_no', sanitize_filename('yes? no', restricted=True))
self.assertEqual('this_-_that', sanitize_filename('this: that', restricted=True))
tests = 'aäb\u4e2d\u56fd\u7684c'
self.assertEqual(sanitize_filename(tests, restricted=True), 'aab_c')
self.assertTrue(sanitize_filename('\xf6', restricted=True) != '') # No empty filename
forbidden = '"\0\\/&!: \'\t\n()[]{}$;`^,#'
for fc in forbidden:
for fbc in forbidden:
self.assertTrue(fbc not in sanitize_filename(fc, restricted=True))
# Handle a common case more neatly
self.assertEqual(sanitize_filename('\u5927\u58f0\u5e26 - Song', restricted=True), 'Song')
self.assertEqual(sanitize_filename('\u603b\u7edf: Speech', restricted=True), 'Speech')
# .. but make sure the file name is never empty
self.assertTrue(sanitize_filename('-', restricted=True) != '')
self.assertTrue(sanitize_filename(':', restricted=True) != '')
self.assertEqual(sanitize_filename(
'ÂÃÄÀÁÅÆÇÈÉÊËÌÍÎÏÐÑÒÓÔÕÖŐØŒÙÚÛÜŰÝÞßàáâãäåæçèéêëìíîïðñòóôõöőøœùúûüűýþÿ', restricted=True),
'AAAAAAAECEEEEIIIIDNOOOOOOOOEUUUUUYTHssaaaaaaaeceeeeiiiionooooooooeuuuuuythy')
def test_sanitize_ids(self):
self.assertEqual(sanitize_filename('_n_cd26wFpw', is_id=True), '_n_cd26wFpw')
self.assertEqual(sanitize_filename('_BD_eEpuzXw', is_id=True), '_BD_eEpuzXw')
self.assertEqual(sanitize_filename('N0Y__7-UOdI', is_id=True), 'N0Y__7-UOdI')
def test_sanitize_path(self):
if sys.platform != 'win32':
return
self.assertEqual(sanitize_path('abc'), 'abc')
self.assertEqual(sanitize_path('abc/def'), 'abc\\def')
self.assertEqual(sanitize_path('abc\\def'), 'abc\\def')
self.assertEqual(sanitize_path('abc|def'), 'abc#def')
self.assertEqual(sanitize_path('<>:"|?*'), '#######')
self.assertEqual(sanitize_path('C:/abc/def'), 'C:\\abc\\def')
self.assertEqual(sanitize_path('C?:/abc/def'), 'C##\\abc\\def')
self.assertEqual(sanitize_path('\\\\?\\UNC\\ComputerName\\abc'), '\\\\?\\UNC\\ComputerName\\abc')
self.assertEqual(sanitize_path('\\\\?\\UNC/ComputerName/abc'), '\\\\?\\UNC\\ComputerName\\abc')
self.assertEqual(sanitize_path('\\\\?\\C:\\abc'), '\\\\?\\C:\\abc')
self.assertEqual(sanitize_path('\\\\?\\C:/abc'), '\\\\?\\C:\\abc')
self.assertEqual(sanitize_path('\\\\?\\C:\\ab?c\\de:f'), '\\\\?\\C:\\ab#c\\de#f')
self.assertEqual(sanitize_path('\\\\?\\C:\\abc'), '\\\\?\\C:\\abc')
self.assertEqual(
sanitize_path('youtube/%(uploader)s/%(autonumber)s-%(title)s-%(upload_date)s.%(ext)s'),
'youtube\\%(uploader)s\\%(autonumber)s-%(title)s-%(upload_date)s.%(ext)s')
self.assertEqual(
sanitize_path('youtube/TheWreckingYard ./00001-Not bad, Especially for Free! (1987 Yamaha 700)-20141116.mp4.part'),
'youtube\\TheWreckingYard #\\00001-Not bad, Especially for Free! (1987 Yamaha 700)-20141116.mp4.part')
self.assertEqual(sanitize_path('abc/def...'), 'abc\\def..#')
self.assertEqual(sanitize_path('abc.../def'), 'abc..#\\def')
self.assertEqual(sanitize_path('abc.../def...'), 'abc..#\\def..#')
self.assertEqual(sanitize_path('../abc'), '..\\abc')
self.assertEqual(sanitize_path('../../abc'), '..\\..\\abc')
self.assertEqual(sanitize_path('./abc'), 'abc')
self.assertEqual(sanitize_path('./../abc'), '..\\abc')
def test_sanitize_url(self):
self.assertEqual(sanitize_url('//foo.bar'), 'http://foo.bar')
self.assertEqual(sanitize_url('httpss://foo.bar'), 'https://foo.bar')
self.assertEqual(sanitize_url('rmtps://foo.bar'), 'rtmps://foo.bar')
self.assertEqual(sanitize_url('https://foo.bar'), 'https://foo.bar')
self.assertEqual(sanitize_url('foo bar'), 'foo bar')
def test_expand_path(self):
def env(var):
return f'%{var}%' if sys.platform == 'win32' else f'${var}'
os.environ['yt_dlp_EXPATH_PATH'] = 'expanded'
self.assertEqual(expand_path(env('yt_dlp_EXPATH_PATH')), 'expanded')
old_home = os.environ.get('HOME')
test_str = R'C:\Documents and Settings\тест\Application Data'
try:
os.environ['HOME'] = test_str
self.assertEqual(expand_path(env('HOME')), os.getenv('HOME'))
self.assertEqual(expand_path('~'), os.getenv('HOME'))
self.assertEqual(
expand_path('~/%s' % env('yt_dlp_EXPATH_PATH')),
'%s/expanded' % os.getenv('HOME'))
finally:
os.environ['HOME'] = old_home or ''
def test_prepend_extension(self):
self.assertEqual(prepend_extension('abc.ext', 'temp'), 'abc.temp.ext')
self.assertEqual(prepend_extension('abc.ext', 'temp', 'ext'), 'abc.temp.ext')
self.assertEqual(prepend_extension('abc.unexpected_ext', 'temp', 'ext'), 'abc.unexpected_ext.temp')
self.assertEqual(prepend_extension('abc', 'temp'), 'abc.temp')
self.assertEqual(prepend_extension('.abc', 'temp'), '.abc.temp')
self.assertEqual(prepend_extension('.abc.ext', 'temp'), '.abc.temp.ext')
def test_replace_extension(self):
self.assertEqual(replace_extension('abc.ext', 'temp'), 'abc.temp')
self.assertEqual(replace_extension('abc.ext', 'temp', 'ext'), 'abc.temp')
self.assertEqual(replace_extension('abc.unexpected_ext', 'temp', 'ext'), 'abc.unexpected_ext.temp')
self.assertEqual(replace_extension('abc', 'temp'), 'abc.temp')
self.assertEqual(replace_extension('.abc', 'temp'), '.abc.temp')
self.assertEqual(replace_extension('.abc.ext', 'temp'), '.abc.temp')
def test_subtitles_filename(self):
self.assertEqual(subtitles_filename('abc.ext', 'en', 'vtt'), 'abc.en.vtt')
self.assertEqual(subtitles_filename('abc.ext', 'en', 'vtt', 'ext'), 'abc.en.vtt')
self.assertEqual(subtitles_filename('abc.unexpected_ext', 'en', 'vtt', 'ext'), 'abc.unexpected_ext.en.vtt')
def test_remove_start(self):
self.assertEqual(remove_start(None, 'A - '), None)
self.assertEqual(remove_start('A - B', 'A - '), 'B')
self.assertEqual(remove_start('B - A', 'A - '), 'B - A')
def test_remove_end(self):
self.assertEqual(remove_end(None, ' - B'), None)
self.assertEqual(remove_end('A - B', ' - B'), 'A')
self.assertEqual(remove_end('B - A', ' - B'), 'B - A')
def test_remove_quotes(self):
self.assertEqual(remove_quotes(None), None)
self.assertEqual(remove_quotes('"'), '"')
self.assertEqual(remove_quotes("'"), "'")
self.assertEqual(remove_quotes(';'), ';')
self.assertEqual(remove_quotes('";'), '";')
self.assertEqual(remove_quotes('""'), '')
self.assertEqual(remove_quotes('";"'), ';')
def test_ordered_set(self):
self.assertEqual(orderedSet([1, 1, 2, 3, 4, 4, 5, 6, 7, 3, 5]), [1, 2, 3, 4, 5, 6, 7])
self.assertEqual(orderedSet([]), [])
self.assertEqual(orderedSet([1]), [1])
# keep the list ordered
self.assertEqual(orderedSet([135, 1, 1, 1]), [135, 1])
def test_unescape_html(self):
self.assertEqual(unescapeHTML('%20;'), '%20;')
self.assertEqual(unescapeHTML('&#x2F;'), '/')
self.assertEqual(unescapeHTML('&#47;'), '/')
self.assertEqual(unescapeHTML('&eacute;'), 'é')
self.assertEqual(unescapeHTML('&#2013266066;'), '&#2013266066;')
self.assertEqual(unescapeHTML('&a&quot;'), '&a"')
# HTML5 entities
self.assertEqual(unescapeHTML('&period;&apos;'), '.\'')
def test_date_from_str(self):
self.assertEqual(date_from_str('yesterday'), date_from_str('now-1day'))
self.assertEqual(date_from_str('now+7day'), date_from_str('now+1week'))
self.assertEqual(date_from_str('now+14day'), date_from_str('now+2week'))
self.assertEqual(date_from_str('20200229+365day'), date_from_str('20200229+1year'))
self.assertEqual(date_from_str('20210131+28day'), date_from_str('20210131+1month'))
def test_datetime_from_str(self):
self.assertEqual(datetime_from_str('yesterday', precision='day'), datetime_from_str('now-1day', precision='auto'))
self.assertEqual(datetime_from_str('now+7day', precision='day'), datetime_from_str('now+1week', precision='auto'))
self.assertEqual(datetime_from_str('now+14day', precision='day'), datetime_from_str('now+2week', precision='auto'))
self.assertEqual(datetime_from_str('20200229+365day', precision='day'), datetime_from_str('20200229+1year', precision='auto'))
self.assertEqual(datetime_from_str('20210131+28day', precision='day'), datetime_from_str('20210131+1month', precision='auto'))
self.assertEqual(datetime_from_str('20210131+59day', precision='day'), datetime_from_str('20210131+2month', precision='auto'))
self.assertEqual(datetime_from_str('now+1day', precision='hour'), datetime_from_str('now+24hours', precision='auto'))
self.assertEqual(datetime_from_str('now+23hours', precision='hour'), datetime_from_str('now+23hours', precision='auto'))
def test_daterange(self):
_20century = DateRange("19000101", "20000101")
self.assertFalse("17890714" in _20century)
_ac = DateRange("00010101")
self.assertTrue("19690721" in _ac)
_firstmilenium = DateRange(end="10000101")
self.assertTrue("07110427" in _firstmilenium)
def test_unified_dates(self):
self.assertEqual(unified_strdate('December 21, 2010'), '20101221')
self.assertEqual(unified_strdate('8/7/2009'), '20090708')
self.assertEqual(unified_strdate('Dec 14, 2012'), '20121214')
self.assertEqual(unified_strdate('2012/10/11 01:56:38 +0000'), '20121011')
self.assertEqual(unified_strdate('1968 12 10'), '19681210')
self.assertEqual(unified_strdate('1968-12-10'), '19681210')
self.assertEqual(unified_strdate('31-07-2022 20:00'), '20220731')
self.assertEqual(unified_strdate('28/01/2014 21:00:00 +0100'), '20140128')
self.assertEqual(
unified_strdate('11/26/2014 11:30:00 AM PST', day_first=False),
'20141126')
self.assertEqual(
unified_strdate('2/2/2015 6:47:40 PM', day_first=False),
'20150202')
self.assertEqual(unified_strdate('Feb 14th 2016 5:45PM'), '20160214')
self.assertEqual(unified_strdate('25-09-2014'), '20140925')
self.assertEqual(unified_strdate('27.02.2016 17:30'), '20160227')
self.assertEqual(unified_strdate('UNKNOWN DATE FORMAT'), None)
self.assertEqual(unified_strdate('Feb 7, 2016 at 6:35 pm'), '20160207')
self.assertEqual(unified_strdate('July 15th, 2013'), '20130715')
self.assertEqual(unified_strdate('September 1st, 2013'), '20130901')
self.assertEqual(unified_strdate('Sep 2nd, 2013'), '20130902')
self.assertEqual(unified_strdate('November 3rd, 2019'), '20191103')
self.assertEqual(unified_strdate('October 23rd, 2005'), '20051023')
def test_unified_timestamps(self):
self.assertEqual(unified_timestamp('December 21, 2010'), 1292889600)
self.assertEqual(unified_timestamp('8/7/2009'), 1247011200)
self.assertEqual(unified_timestamp('Dec 14, 2012'), 1355443200)
self.assertEqual(unified_timestamp('2012/10/11 01:56:38 +0000'), 1349920598)
self.assertEqual(unified_timestamp('1968 12 10'), -33436800)
self.assertEqual(unified_timestamp('1968-12-10'), -33436800)
self.assertEqual(unified_timestamp('28/01/2014 21:00:00 +0100'), 1390939200)
self.assertEqual(
unified_timestamp('11/26/2014 11:30:00 AM PST', day_first=False),
1417001400)
self.assertEqual(
unified_timestamp('2/2/2015 6:47:40 PM', day_first=False),
1422902860)
self.assertEqual(unified_timestamp('Feb 14th 2016 5:45PM'), 1455471900)
self.assertEqual(unified_timestamp('25-09-2014'), 1411603200)
self.assertEqual(unified_timestamp('27.02.2016 17:30'), 1456594200)
self.assertEqual(unified_timestamp('UNKNOWN DATE FORMAT'), None)
self.assertEqual(unified_timestamp('May 16, 2016 11:15 PM'), 1463440500)
self.assertEqual(unified_timestamp('Feb 7, 2016 at 6:35 pm'), 1454870100)
self.assertEqual(unified_timestamp('2017-03-30T17:52:41Q'), 1490896361)
self.assertEqual(unified_timestamp('Sep 11, 2013 | 5:49 AM'), 1378878540)
self.assertEqual(unified_timestamp('December 15, 2017 at 7:49 am'), 1513324140)
self.assertEqual(unified_timestamp('2018-03-14T08:32:43.1493874+00:00'), 1521016363)
self.assertEqual(unified_timestamp('December 31 1969 20:00:01 EDT'), 1)
self.assertEqual(unified_timestamp('Wednesday 31 December 1969 18:01:26 MDT'), 86)
self.assertEqual(unified_timestamp('12/31/1969 20:01:18 EDT', False), 78)
def test_determine_ext(self):
self.assertEqual(determine_ext('http://example.com/foo/bar.mp4/?download'), 'mp4')
self.assertEqual(determine_ext('http://example.com/foo/bar/?download', None), None)
self.assertEqual(determine_ext('http://example.com/foo/bar.nonext/?download', None), None)
self.assertEqual(determine_ext('http://example.com/foo/bar/mp4?download', None), None)
self.assertEqual(determine_ext('http://example.com/foo/bar.m3u8//?download'), 'm3u8')
self.assertEqual(determine_ext('foobar', None), None)
def test_find_xpath_attr(self):
testxml = '''<root>
<node/>
<node x="a"/>
<node x="a" y="c" />
<node x="b" y="d" />
<node x="" />
</root>'''
doc = compat_etree_fromstring(testxml)
self.assertEqual(find_xpath_attr(doc, './/fourohfour', 'n'), None)
self.assertEqual(find_xpath_attr(doc, './/fourohfour', 'n', 'v'), None)
self.assertEqual(find_xpath_attr(doc, './/node', 'n'), None)
self.assertEqual(find_xpath_attr(doc, './/node', 'n', 'v'), None)
self.assertEqual(find_xpath_attr(doc, './/node', 'x'), doc[1])
self.assertEqual(find_xpath_attr(doc, './/node', 'x', 'a'), doc[1])
self.assertEqual(find_xpath_attr(doc, './/node', 'x', 'b'), doc[3])
self.assertEqual(find_xpath_attr(doc, './/node', 'y'), doc[2])
self.assertEqual(find_xpath_attr(doc, './/node', 'y', 'c'), doc[2])
self.assertEqual(find_xpath_attr(doc, './/node', 'y', 'd'), doc[3])
self.assertEqual(find_xpath_attr(doc, './/node', 'x', ''), doc[4])
def test_xpath_with_ns(self):
testxml = '''<root xmlns:media="http://example.com/">
<media:song>
<media:author>The Author</media:author>
<url>http://server.com/download.mp3</url>
</media:song>
</root>'''
doc = compat_etree_fromstring(testxml)
find = lambda p: doc.find(xpath_with_ns(p, {'media': 'http://example.com/'}))
self.assertTrue(find('media:song') is not None)
self.assertEqual(find('media:song/media:author').text, 'The Author')
self.assertEqual(find('media:song/url').text, 'http://server.com/download.mp3')
def test_xpath_element(self):
doc = xml.etree.ElementTree.Element('root')
div = xml.etree.ElementTree.SubElement(doc, 'div')
p = xml.etree.ElementTree.SubElement(div, 'p')
p.text = 'Foo'
self.assertEqual(xpath_element(doc, 'div/p'), p)
self.assertEqual(xpath_element(doc, ['div/p']), p)
self.assertEqual(xpath_element(doc, ['div/bar', 'div/p']), p)
self.assertEqual(xpath_element(doc, 'div/bar', default='default'), 'default')
self.assertEqual(xpath_element(doc, ['div/bar'], default='default'), 'default')
self.assertTrue(xpath_element(doc, 'div/bar') is None)
self.assertTrue(xpath_element(doc, ['div/bar']) is None)
self.assertTrue(xpath_element(doc, ['div/bar'], 'div/baz') is None)
self.assertRaises(ExtractorError, xpath_element, doc, 'div/bar', fatal=True)
self.assertRaises(ExtractorError, xpath_element, doc, ['div/bar'], fatal=True)
self.assertRaises(ExtractorError, xpath_element, doc, ['div/bar', 'div/baz'], fatal=True)
def test_xpath_text(self):
testxml = '''<root>
<div>
<p>Foo</p>
</div>
</root>'''
doc = compat_etree_fromstring(testxml)
self.assertEqual(xpath_text(doc, 'div/p'), 'Foo')
self.assertEqual(xpath_text(doc, 'div/bar', default='default'), 'default')
self.assertTrue(xpath_text(doc, 'div/bar') is None)
self.assertRaises(ExtractorError, xpath_text, doc, 'div/bar', fatal=True)
def test_xpath_attr(self):
testxml = '''<root>
<div>
<p x="a">Foo</p>
</div>
</root>'''
doc = compat_etree_fromstring(testxml)
self.assertEqual(xpath_attr(doc, 'div/p', 'x'), 'a')
self.assertEqual(xpath_attr(doc, 'div/bar', 'x'), None)
self.assertEqual(xpath_attr(doc, 'div/p', 'y'), None)
self.assertEqual(xpath_attr(doc, 'div/bar', 'x', default='default'), 'default')
self.assertEqual(xpath_attr(doc, 'div/p', 'y', default='default'), 'default')
self.assertRaises(ExtractorError, xpath_attr, doc, 'div/bar', 'x', fatal=True)
self.assertRaises(ExtractorError, xpath_attr, doc, 'div/p', 'y', fatal=True)
def test_smuggle_url(self):
data = {"ö": "ö", "abc": [3]}
url = 'https://foo.bar/baz?x=y#a'
smug_url = smuggle_url(url, data)
unsmug_url, unsmug_data = unsmuggle_url(smug_url)
self.assertEqual(url, unsmug_url)
self.assertEqual(data, unsmug_data)
res_url, res_data = unsmuggle_url(url)
self.assertEqual(res_url, url)
self.assertEqual(res_data, None)
smug_url = smuggle_url(url, {'a': 'b'})
smug_smug_url = smuggle_url(smug_url, {'c': 'd'})
res_url, res_data = unsmuggle_url(smug_smug_url)
self.assertEqual(res_url, url)
self.assertEqual(res_data, {'a': 'b', 'c': 'd'})
def test_shell_quote(self):
args = ['ffmpeg', '-i', encodeFilename('ñ€ß\'.mp4')]
self.assertEqual(
shell_quote(args),
"""ffmpeg -i 'ñ€ß'"'"'.mp4'""" if compat_os_name != 'nt' else '''ffmpeg -i "ñ€ß'.mp4"''')
def test_float_or_none(self):
self.assertEqual(float_or_none('42.42'), 42.42)
self.assertEqual(float_or_none('42'), 42.0)
self.assertEqual(float_or_none(''), None)
self.assertEqual(float_or_none(None), None)
self.assertEqual(float_or_none([]), None)
self.assertEqual(float_or_none(set()), None)
def test_int_or_none(self):
self.assertEqual(int_or_none('42'), 42)
self.assertEqual(int_or_none(''), None)
self.assertEqual(int_or_none(None), None)
self.assertEqual(int_or_none([]), None)
self.assertEqual(int_or_none(set()), None)
def test_str_to_int(self):
self.assertEqual(str_to_int('123,456'), 123456)
self.assertEqual(str_to_int('123.456'), 123456)
self.assertEqual(str_to_int(523), 523)
self.assertEqual(str_to_int('noninteger'), None)
self.assertEqual(str_to_int([]), None)
def test_url_basename(self):
self.assertEqual(url_basename('http://foo.de/'), '')
self.assertEqual(url_basename('http://foo.de/bar/baz'), 'baz')
self.assertEqual(url_basename('http://foo.de/bar/baz?x=y'), 'baz')
self.assertEqual(url_basename('http://foo.de/bar/baz#x=y'), 'baz')
self.assertEqual(url_basename('http://foo.de/bar/baz/'), 'baz')
self.assertEqual(
url_basename('http://media.w3.org/2010/05/sintel/trailer.mp4'),
'trailer.mp4')
def test_base_url(self):
self.assertEqual(base_url('http://foo.de/'), 'http://foo.de/')
self.assertEqual(base_url('http://foo.de/bar'), 'http://foo.de/')
self.assertEqual(base_url('http://foo.de/bar/'), 'http://foo.de/bar/')
self.assertEqual(base_url('http://foo.de/bar/baz'), 'http://foo.de/bar/')
self.assertEqual(base_url('http://foo.de/bar/baz?x=z/x/c'), 'http://foo.de/bar/')
self.assertEqual(base_url('http://foo.de/bar/baz&x=z&w=y/x/c'), 'http://foo.de/bar/baz&x=z&w=y/x/')
def test_urljoin(self):
self.assertEqual(urljoin('http://foo.de/', '/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin(b'http://foo.de/', '/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de/', b'/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin(b'http://foo.de/', b'/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('//foo.de/', '/a/b/c.txt'), '//foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de/', 'a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de', '/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de', 'a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de/', 'http://foo.de/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de/', '//foo.de/a/b/c.txt'), '//foo.de/a/b/c.txt')
self.assertEqual(urljoin(None, 'http://foo.de/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin(None, '//foo.de/a/b/c.txt'), '//foo.de/a/b/c.txt')
self.assertEqual(urljoin('', 'http://foo.de/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin(['foobar'], 'http://foo.de/a/b/c.txt'), 'http://foo.de/a/b/c.txt')
self.assertEqual(urljoin('http://foo.de/', None), None)
self.assertEqual(urljoin('http://foo.de/', ''), None)
self.assertEqual(urljoin('http://foo.de/', ['foobar']), None)
self.assertEqual(urljoin('http://foo.de/a/b/c.txt', '.././../d.txt'), 'http://foo.de/d.txt')
self.assertEqual(urljoin('http://foo.de/a/b/c.txt', 'rtmp://foo.de'), 'rtmp://foo.de')
self.assertEqual(urljoin(None, 'rtmp://foo.de'), 'rtmp://foo.de')
def test_url_or_none(self):
self.assertEqual(url_or_none(None), None)
self.assertEqual(url_or_none(''), None)
self.assertEqual(url_or_none('foo'), None)
self.assertEqual(url_or_none('http://foo.de'), 'http://foo.de')
self.assertEqual(url_or_none('https://foo.de'), 'https://foo.de')
self.assertEqual(url_or_none('http$://foo.de'), None)
self.assertEqual(url_or_none('http://foo.de'), 'http://foo.de')
self.assertEqual(url_or_none('//foo.de'), '//foo.de')
self.assertEqual(url_or_none('s3://foo.de'), None)
self.assertEqual(url_or_none('rtmpte://foo.de'), 'rtmpte://foo.de')
self.assertEqual(url_or_none('mms://foo.de'), 'mms://foo.de')
self.assertEqual(url_or_none('rtspu://foo.de'), 'rtspu://foo.de')
self.assertEqual(url_or_none('ftps://foo.de'), 'ftps://foo.de')
def test_parse_age_limit(self):
self.assertEqual(parse_age_limit(None), None)
self.assertEqual(parse_age_limit(False), None)
self.assertEqual(parse_age_limit('invalid'), None)
self.assertEqual(parse_age_limit(0), 0)
self.assertEqual(parse_age_limit(18), 18)
self.assertEqual(parse_age_limit(21), 21)
self.assertEqual(parse_age_limit(22), None)
self.assertEqual(parse_age_limit('18'), 18)
self.assertEqual(parse_age_limit('18+'), 18)
self.assertEqual(parse_age_limit('PG-13'), 13)
self.assertEqual(parse_age_limit('TV-14'), 14)
self.assertEqual(parse_age_limit('TV-MA'), 17)
self.assertEqual(parse_age_limit('TV14'), 14)
self.assertEqual(parse_age_limit('TV_G'), 0)
def test_parse_duration(self):
self.assertEqual(parse_duration(None), None)
self.assertEqual(parse_duration(False), None)
self.assertEqual(parse_duration('invalid'), None)
self.assertEqual(parse_duration('1'), 1)
self.assertEqual(parse_duration('1337:12'), 80232)
self.assertEqual(parse_duration('9:12:43'), 33163)
self.assertEqual(parse_duration('12:00'), 720)
self.assertEqual(parse_duration('00:01:01'), 61)
self.assertEqual(parse_duration('x:y'), None)
self.assertEqual(parse_duration('3h11m53s'), 11513)
self.assertEqual(parse_duration('3h 11m 53s'), 11513)
self.assertEqual(parse_duration('3 hours 11 minutes 53 seconds'), 11513)
self.assertEqual(parse_duration('3 hours 11 mins 53 secs'), 11513)
self.assertEqual(parse_duration('3 hours, 11 minutes, 53 seconds'), 11513)
self.assertEqual(parse_duration('3 hours, 11 mins, 53 secs'), 11513)
self.assertEqual(parse_duration('62m45s'), 3765)
self.assertEqual(parse_duration('6m59s'), 419)
self.assertEqual(parse_duration('49s'), 49)
self.assertEqual(parse_duration('0h0m0s'), 0)
self.assertEqual(parse_duration('0m0s'), 0)
self.assertEqual(parse_duration('0s'), 0)
self.assertEqual(parse_duration('01:02:03.05'), 3723.05)
self.assertEqual(parse_duration('T30M38S'), 1838)
self.assertEqual(parse_duration('5 s'), 5)
self.assertEqual(parse_duration('3 min'), 180)
self.assertEqual(parse_duration('2.5 hours'), 9000)
self.assertEqual(parse_duration('02:03:04'), 7384)
self.assertEqual(parse_duration('01:02:03:04'), 93784)
self.assertEqual(parse_duration('1 hour 3 minutes'), 3780)
self.assertEqual(parse_duration('87 Min.'), 5220)
self.assertEqual(parse_duration('PT1H0.040S'), 3600.04)
self.assertEqual(parse_duration('PT00H03M30SZ'), 210)
self.assertEqual(parse_duration('P0Y0M0DT0H4M20.880S'), 260.88)
self.assertEqual(parse_duration('01:02:03:050'), 3723.05)
self.assertEqual(parse_duration('103:050'), 103.05)
self.assertEqual(parse_duration('1HR 3MIN'), 3780)
self.assertEqual(parse_duration('2hrs 3mins'), 7380)
def test_fix_xml_ampersands(self):
self.assertEqual(
fix_xml_ampersands('"&x=y&z=a'), '"&amp;x=y&amp;z=a')
self.assertEqual(
fix_xml_ampersands('"&amp;x=y&wrong;&z=a'),
'"&amp;x=y&amp;wrong;&amp;z=a')
self.assertEqual(
fix_xml_ampersands('&amp;&apos;&gt;&lt;&quot;'),
'&amp;&apos;&gt;&lt;&quot;')
self.assertEqual(
fix_xml_ampersands('&#1234;&#x1abC;'), '&#1234;&#x1abC;')
self.assertEqual(fix_xml_ampersands('&#&#'), '&amp;#&amp;#')
def test_paged_list(self):
def testPL(size, pagesize, sliceargs, expected):
def get_page(pagenum):
firstid = pagenum * pagesize
upto = min(size, pagenum * pagesize + pagesize)
yield from range(firstid, upto)
pl = OnDemandPagedList(get_page, pagesize)
got = pl.getslice(*sliceargs)
self.assertEqual(got, expected)
iapl = InAdvancePagedList(get_page, size // pagesize + 1, pagesize)
got = iapl.getslice(*sliceargs)
self.assertEqual(got, expected)
testPL(5, 2, (), [0, 1, 2, 3, 4])
testPL(5, 2, (1,), [1, 2, 3, 4])
testPL(5, 2, (2,), [2, 3, 4])
testPL(5, 2, (4,), [4])
testPL(5, 2, (0, 3), [0, 1, 2])
testPL(5, 2, (1, 4), [1, 2, 3])
testPL(5, 2, (2, 99), [2, 3, 4])
testPL(5, 2, (20, 99), [])
def test_read_batch_urls(self):
f = io.StringIO('''\xef\xbb\xbf foo
bar\r
baz
# More after this line\r
; or after this
bam''')
self.assertEqual(read_batch_urls(f), ['foo', 'bar', 'baz', 'bam'])
def test_urlencode_postdata(self):
data = urlencode_postdata({'username': 'foo@bar.com', 'password': '1234'})
self.assertTrue(isinstance(data, bytes))
def test_update_url_query(self):
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'quality': ['HD'], 'format': ['mp4']})),
parse_qs('http://example.com/path?quality=HD&format=mp4'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'system': ['LINUX', 'WINDOWS']})),
parse_qs('http://example.com/path?system=LINUX&system=WINDOWS'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'fields': 'id,formats,subtitles'})),
parse_qs('http://example.com/path?fields=id,formats,subtitles'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'fields': ('id,formats,subtitles', 'thumbnails')})),
parse_qs('http://example.com/path?fields=id,formats,subtitles&fields=thumbnails'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path?manifest=f4m', {'manifest': []})),
parse_qs('http://example.com/path'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path?system=LINUX&system=WINDOWS', {'system': 'LINUX'})),
parse_qs('http://example.com/path?system=LINUX'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'fields': b'id,formats,subtitles'})),
parse_qs('http://example.com/path?fields=id,formats,subtitles'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'width': 1080, 'height': 720})),
parse_qs('http://example.com/path?width=1080&height=720'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'bitrate': 5020.43})),
parse_qs('http://example.com/path?bitrate=5020.43'))
self.assertEqual(parse_qs(update_url_query(
'http://example.com/path', {'test': '第二行тест'})),
parse_qs('http://example.com/path?test=%E7%AC%AC%E4%BA%8C%E8%A1%8C%D1%82%D0%B5%D1%81%D1%82'))
def test_multipart_encode(self):
self.assertEqual(
multipart_encode({b'field': b'value'}, boundary='AAAAAA')[0],
b'--AAAAAA\r\nContent-Disposition: form-data; name="field"\r\n\r\nvalue\r\n--AAAAAA--\r\n')
self.assertEqual(
multipart_encode({'欄位'.encode(): ''.encode()}, boundary='AAAAAA')[0],
b'--AAAAAA\r\nContent-Disposition: form-data; name="\xe6\xac\x84\xe4\xbd\x8d"\r\n\r\n\xe5\x80\xbc\r\n--AAAAAA--\r\n')
self.assertRaises(
ValueError, multipart_encode, {b'field': b'value'}, boundary='value')
def test_dict_get(self):
FALSE_VALUES = {
'none': None,
'false': False,
'zero': 0,
'empty_string': '',
'empty_list': [],
}
d = FALSE_VALUES.copy()
d['a'] = 42
self.assertEqual(dict_get(d, 'a'), 42)
self.assertEqual(dict_get(d, 'b'), None)
self.assertEqual(dict_get(d, 'b', 42), 42)
self.assertEqual(dict_get(d, ('a', )), 42)
self.assertEqual(dict_get(d, ('b', 'a', )), 42)
self.assertEqual(dict_get(d, ('b', 'c', 'a', 'd', )), 42)
self.assertEqual(dict_get(d, ('b', 'c', )), None)
self.assertEqual(dict_get(d, ('b', 'c', ), 42), 42)
for key, false_value in FALSE_VALUES.items():
self.assertEqual(dict_get(d, ('b', 'c', key, )), None)
self.assertEqual(dict_get(d, ('b', 'c', key, ), skip_false_values=False), false_value)
def test_merge_dicts(self):
self.assertEqual(merge_dicts({'a': 1}, {'b': 2}), {'a': 1, 'b': 2})
self.assertEqual(merge_dicts({'a': 1}, {'a': 2}), {'a': 1})
self.assertEqual(merge_dicts({'a': 1}, {'a': None}), {'a': 1})
self.assertEqual(merge_dicts({'a': 1}, {'a': ''}), {'a': 1})
self.assertEqual(merge_dicts({'a': 1}, {}), {'a': 1})
self.assertEqual(merge_dicts({'a': None}, {'a': 1}), {'a': 1})
self.assertEqual(merge_dicts({'a': ''}, {'a': 1}), {'a': ''})
self.assertEqual(merge_dicts({'a': ''}, {'a': 'abc'}), {'a': 'abc'})
self.assertEqual(merge_dicts({'a': None}, {'a': ''}, {'a': 'abc'}), {'a': 'abc'})
def test_encode_compat_str(self):
self.assertEqual(encode_compat_str(b'\xd1\x82\xd0\xb5\xd1\x81\xd1\x82', 'utf-8'), 'тест')
self.assertEqual(encode_compat_str('тест', 'utf-8'), 'тест')
def test_parse_iso8601(self):
self.assertEqual(parse_iso8601('2014-03-23T23:04:26+0100'), 1395612266)
self.assertEqual(parse_iso8601('2014-03-23T22:04:26+0000'), 1395612266)
self.assertEqual(parse_iso8601('2014-03-23T22:04:26Z'), 1395612266)
self.assertEqual(parse_iso8601('2014-03-23T22:04:26.1234Z'), 1395612266)
self.assertEqual(parse_iso8601('2015-09-29T08:27:31.727'), 1443515251)
self.assertEqual(parse_iso8601('2015-09-29T08-27-31.727'), None)
def test_strip_jsonp(self):
stripped = strip_jsonp('cb ([ {"id":"532cb",\n\n\n"x":\n3}\n]\n);')
d = json.loads(stripped)
self.assertEqual(d, [{"id": "532cb", "x": 3}])
stripped = strip_jsonp('parseMetadata({"STATUS":"OK"})\n\n\n//epc')
d = json.loads(stripped)
self.assertEqual(d, {'STATUS': 'OK'})
stripped = strip_jsonp('ps.embedHandler({"status": "success"});')
d = json.loads(stripped)
self.assertEqual(d, {'status': 'success'})
stripped = strip_jsonp('window.cb && window.cb({"status": "success"});')
d = json.loads(stripped)
self.assertEqual(d, {'status': 'success'})
stripped = strip_jsonp('window.cb && cb({"status": "success"});')
d = json.loads(stripped)
self.assertEqual(d, {'status': 'success'})
stripped = strip_jsonp('({"status": "success"});')
d = json.loads(stripped)
self.assertEqual(d, {'status': 'success'})
def test_strip_or_none(self):
self.assertEqual(strip_or_none(' abc'), 'abc')
self.assertEqual(strip_or_none('abc '), 'abc')
self.assertEqual(strip_or_none(' abc '), 'abc')
self.assertEqual(strip_or_none('\tabc\t'), 'abc')
self.assertEqual(strip_or_none('\n\tabc\n\t'), 'abc')
self.assertEqual(strip_or_none('abc'), 'abc')
self.assertEqual(strip_or_none(''), '')
self.assertEqual(strip_or_none(None), None)
self.assertEqual(strip_or_none(42), None)
self.assertEqual(strip_or_none([]), None)
def test_uppercase_escape(self):
self.assertEqual(uppercase_escape(''), '')
self.assertEqual(uppercase_escape('\\U0001d550'), '𝕐')
def test_lowercase_escape(self):
self.assertEqual(lowercase_escape(''), '')
self.assertEqual(lowercase_escape('\\u0026'), '&')
def test_limit_length(self):
self.assertEqual(limit_length(None, 12), None)
self.assertEqual(limit_length('foo', 12), 'foo')
self.assertTrue(
limit_length('foo bar baz asd', 12).startswith('foo bar'))
self.assertTrue('...' in limit_length('foo bar baz asd', 12))
def test_mimetype2ext(self):
self.assertEqual(mimetype2ext(None), None)
self.assertEqual(mimetype2ext('video/x-flv'), 'flv')
self.assertEqual(mimetype2ext('application/x-mpegURL'), 'm3u8')
self.assertEqual(mimetype2ext('text/vtt'), 'vtt')
self.assertEqual(mimetype2ext('text/vtt;charset=utf-8'), 'vtt')
self.assertEqual(mimetype2ext('text/html; charset=utf-8'), 'html')
self.assertEqual(mimetype2ext('audio/x-wav'), 'wav')
self.assertEqual(mimetype2ext('audio/x-wav;codec=pcm'), 'wav')
def test_month_by_name(self):
self.assertEqual(month_by_name(None), None)
self.assertEqual(month_by_name('December', 'en'), 12)
self.assertEqual(month_by_name('décembre', 'fr'), 12)
self.assertEqual(month_by_name('December'), 12)
self.assertEqual(month_by_name('décembre'), None)
self.assertEqual(month_by_name('Unknown', 'unknown'), None)
def test_parse_codecs(self):
self.assertEqual(parse_codecs(''), {})
self.assertEqual(parse_codecs('avc1.77.30, mp4a.40.2'), {
'vcodec': 'avc1.77.30',
'acodec': 'mp4a.40.2',
'dynamic_range': None,
})
self.assertEqual(parse_codecs('mp4a.40.2'), {
'vcodec': 'none',
'acodec': 'mp4a.40.2',
'dynamic_range': None,
})
self.assertEqual(parse_codecs('mp4a.40.5,avc1.42001e'), {
'vcodec': 'avc1.42001e',
'acodec': 'mp4a.40.5',
'dynamic_range': None,
})
self.assertEqual(parse_codecs('avc3.640028'), {
'vcodec': 'avc3.640028',
'acodec': 'none',
'dynamic_range': None,
})
self.assertEqual(parse_codecs(', h264,,newcodec,aac'), {
'vcodec': 'h264',
'acodec': 'aac',
'dynamic_range': None,
})
self.assertEqual(parse_codecs('av01.0.05M.08'), {
'vcodec': 'av01.0.05M.08',
'acodec': 'none',
'dynamic_range': None,
})
self.assertEqual(parse_codecs('vp9.2'), {
'vcodec': 'vp9.2',
'acodec': 'none',
'dynamic_range': 'HDR10',
})
self.assertEqual(parse_codecs('av01.0.12M.10.0.110.09.16.09.0'), {
'vcodec': 'av01.0.12M.10.0.110.09.16.09.0',
'acodec': 'none',
'dynamic_range': 'HDR10',
})
self.assertEqual(parse_codecs('dvhe'), {
'vcodec': 'dvhe',
'acodec': 'none',
'dynamic_range': 'DV',
})
self.assertEqual(parse_codecs('theora, vorbis'), {
'vcodec': 'theora',
'acodec': 'vorbis',
'dynamic_range': None,
})
self.assertEqual(parse_codecs('unknownvcodec, unknownacodec'), {
'vcodec': 'unknownvcodec',
'acodec': 'unknownacodec',
})
self.assertEqual(parse_codecs('unknown'), {})
def test_escape_rfc3986(self):
reserved = "!*'();:@&=+$,/?#[]"
unreserved = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-_.~'
self.assertEqual(escape_rfc3986(reserved), reserved)
self.assertEqual(escape_rfc3986(unreserved), unreserved)
self.assertEqual(escape_rfc3986('тест'), '%D1%82%D0%B5%D1%81%D1%82')
self.assertEqual(escape_rfc3986('%D1%82%D0%B5%D1%81%D1%82'), '%D1%82%D0%B5%D1%81%D1%82')
self.assertEqual(escape_rfc3986('foo bar'), 'foo%20bar')
self.assertEqual(escape_rfc3986('foo%20bar'), 'foo%20bar')
def test_normalize_url(self):
self.assertEqual(
normalize_url('http://wowza.imust.org/srv/vod/telemb/new/UPLOAD/UPLOAD/20224_IncendieHavré_FD.mp4'),
'http://wowza.imust.org/srv/vod/telemb/new/UPLOAD/UPLOAD/20224_IncendieHavre%CC%81_FD.mp4'
)
self.assertEqual(
normalize_url('http://www.ardmediathek.de/tv/Sturm-der-Liebe/Folge-2036-Zu-Mann-und-Frau-erklärt/Das-Erste/Video?documentId=22673108&bcastId=5290'),
'http://www.ardmediathek.de/tv/Sturm-der-Liebe/Folge-2036-Zu-Mann-und-Frau-erkl%C3%A4rt/Das-Erste/Video?documentId=22673108&bcastId=5290'
)
self.assertEqual(
normalize_url('http://тест.рф/фрагмент'),
'http://xn--e1aybc.xn--p1ai/%D1%84%D1%80%D0%B0%D0%B3%D0%BC%D0%B5%D0%BD%D1%82'
)
self.assertEqual(
normalize_url('http://тест.рф/абв?абв=абв#абв'),
'http://xn--e1aybc.xn--p1ai/%D0%B0%D0%B1%D0%B2?%D0%B0%D0%B1%D0%B2=%D0%B0%D0%B1%D0%B2#%D0%B0%D0%B1%D0%B2'
)
self.assertEqual(normalize_url('http://vimeo.com/56015672#at=0'), 'http://vimeo.com/56015672#at=0')
self.assertEqual(normalize_url('http://www.example.com/../a/b/../c/./d.html'), 'http://www.example.com/a/c/d.html')
def test_remove_dot_segments(self):
self.assertEqual(remove_dot_segments('/a/b/c/./../../g'), '/a/g')
self.assertEqual(remove_dot_segments('mid/content=5/../6'), 'mid/6')
self.assertEqual(remove_dot_segments('/ad/../cd'), '/cd')
self.assertEqual(remove_dot_segments('/ad/../cd/'), '/cd/')
self.assertEqual(remove_dot_segments('/..'), '/')
self.assertEqual(remove_dot_segments('/./'), '/')
self.assertEqual(remove_dot_segments('/./a'), '/a')
self.assertEqual(remove_dot_segments('/abc/./.././d/././e/.././f/./../../ghi'), '/ghi')
self.assertEqual(remove_dot_segments('/'), '/')
self.assertEqual(remove_dot_segments('/t'), '/t')
self.assertEqual(remove_dot_segments('t'), 't')
self.assertEqual(remove_dot_segments(''), '')
self.assertEqual(remove_dot_segments('/../a/b/c'), '/a/b/c')
self.assertEqual(remove_dot_segments('../a'), 'a')
self.assertEqual(remove_dot_segments('./a'), 'a')
self.assertEqual(remove_dot_segments('.'), '')
self.assertEqual(remove_dot_segments('////'), '////')
def test_js_to_json_vars_strings(self):
self.assertDictEqual(
json.loads(js_to_json(
'''{
'null': a,
'nullStr': b,
'true': c,
'trueStr': d,
'false': e,
'falseStr': f,
'unresolvedVar': g,
}''',
{
'a': 'null',
'b': '"null"',
'c': 'true',
'd': '"true"',
'e': 'false',
'f': '"false"',
'g': 'var',
}
)),
{
'null': None,
'nullStr': 'null',
'true': True,
'trueStr': 'true',
'false': False,
'falseStr': 'false',
'unresolvedVar': 'var'
}
)
self.assertDictEqual(
json.loads(js_to_json(
'''{
'int': a,
'intStr': b,
'float': c,
'floatStr': d,
}''',
{
'a': '123',
'b': '"123"',
'c': '1.23',
'd': '"1.23"',
}
)),
{
'int': 123,
'intStr': '123',
'float': 1.23,
'floatStr': '1.23',
}
)
self.assertDictEqual(
json.loads(js_to_json(
'''{
'object': a,
'objectStr': b,
'array': c,
'arrayStr': d,
}''',
{
'a': '{}',
'b': '"{}"',
'c': '[]',
'd': '"[]"',
}
)),
{
'object': {},
'objectStr': '{}',
'array': [],
'arrayStr': '[]',
}
)
def test_js_to_json_realworld(self):
inp = '''{
'clip':{'provider':'pseudo'}
}'''
self.assertEqual(js_to_json(inp), '''{
"clip":{"provider":"pseudo"}
}''')
json.loads(js_to_json(inp))
inp = '''{
'playlist':[{'controls':{'all':null}}]
}'''
self.assertEqual(js_to_json(inp), '''{
"playlist":[{"controls":{"all":null}}]
}''')
inp = '''"The CW\\'s \\'Crazy Ex-Girlfriend\\'"'''
self.assertEqual(js_to_json(inp), '''"The CW's 'Crazy Ex-Girlfriend'"''')
inp = '"SAND Number: SAND 2013-7800P\\nPresenter: Tom Russo\\nHabanero Software Training - Xyce Software\\nXyce, Sandia\\u0027s"'
json_code = js_to_json(inp)
self.assertEqual(json.loads(json_code), json.loads(inp))
inp = '''{
0:{src:'skipped', type: 'application/dash+xml'},
1:{src:'skipped', type: 'application/vnd.apple.mpegURL'},
}'''
self.assertEqual(js_to_json(inp), '''{
"0":{"src":"skipped", "type": "application/dash+xml"},
"1":{"src":"skipped", "type": "application/vnd.apple.mpegURL"}
}''')
inp = '''{"foo":101}'''
self.assertEqual(js_to_json(inp), '''{"foo":101}''')
inp = '''{"duration": "00:01:07"}'''
self.assertEqual(js_to_json(inp), '''{"duration": "00:01:07"}''')
inp = '''{segments: [{"offset":-3.885780586188048e-16,"duration":39.75000000000001}]}'''
self.assertEqual(js_to_json(inp), '''{"segments": [{"offset":-3.885780586188048e-16,"duration":39.75000000000001}]}''')
def test_js_to_json_edgecases(self):
on = js_to_json("{abc_def:'1\\'\\\\2\\\\\\'3\"4'}")
self.assertEqual(json.loads(on), {"abc_def": "1'\\2\\'3\"4"})
on = js_to_json('{"abc": true}')
self.assertEqual(json.loads(on), {'abc': True})
# Ignore JavaScript code as well
on = js_to_json('''{
"x": 1,
y: "a",
z: some.code
}''')
d = json.loads(on)
self.assertEqual(d['x'], 1)
self.assertEqual(d['y'], 'a')
# Just drop ! prefix for now though this results in a wrong value
on = js_to_json('''{
a: !0,
b: !1,
c: !!0,
d: !!42.42,
e: !!![],
f: !"abc",
g: !"",
!42: 42
}''')
self.assertEqual(json.loads(on), {
'a': 0,
'b': 1,
'c': 0,
'd': 42.42,
'e': [],
'f': "abc",
'g': "",
'42': 42
})
on = js_to_json('["abc", "def",]')
self.assertEqual(json.loads(on), ['abc', 'def'])
on = js_to_json('[/*comment\n*/"abc"/*comment\n*/,/*comment\n*/"def",/*comment\n*/]')
self.assertEqual(json.loads(on), ['abc', 'def'])
on = js_to_json('[//comment\n"abc" //comment\n,//comment\n"def",//comment\n]')
self.assertEqual(json.loads(on), ['abc', 'def'])
on = js_to_json('{"abc": "def",}')
self.assertEqual(json.loads(on), {'abc': 'def'})
on = js_to_json('{/*comment\n*/"abc"/*comment\n*/:/*comment\n*/"def"/*comment\n*/,/*comment\n*/}')
self.assertEqual(json.loads(on), {'abc': 'def'})
on = js_to_json('{ 0: /* " \n */ ",]" , }')
self.assertEqual(json.loads(on), {'0': ',]'})
on = js_to_json('{ /*comment\n*/0/*comment\n*/: /* " \n */ ",]" , }')
self.assertEqual(json.loads(on), {'0': ',]'})
on = js_to_json('{ 0: // comment\n1 }')
self.assertEqual(json.loads(on), {'0': 1})
on = js_to_json(r'["<p>x<\/p>"]')
self.assertEqual(json.loads(on), ['<p>x</p>'])
on = js_to_json(r'["\xaa"]')
self.assertEqual(json.loads(on), ['\u00aa'])
on = js_to_json("['a\\\nb']")
self.assertEqual(json.loads(on), ['ab'])
on = js_to_json("/*comment\n*/[/*comment\n*/'a\\\nb'/*comment\n*/]/*comment\n*/")
self.assertEqual(json.loads(on), ['ab'])
on = js_to_json('{0xff:0xff}')
self.assertEqual(json.loads(on), {'255': 255})
on = js_to_json('{/*comment\n*/0xff/*comment\n*/:/*comment\n*/0xff/*comment\n*/}')
self.assertEqual(json.loads(on), {'255': 255})
on = js_to_json('{077:077}')
self.assertEqual(json.loads(on), {'63': 63})
on = js_to_json('{/*comment\n*/077/*comment\n*/:/*comment\n*/077/*comment\n*/}')
self.assertEqual(json.loads(on), {'63': 63})
on = js_to_json('{42:42}')
self.assertEqual(json.loads(on), {'42': 42})
on = js_to_json('{/*comment\n*/42/*comment\n*/:/*comment\n*/42/*comment\n*/}')
self.assertEqual(json.loads(on), {'42': 42})
on = js_to_json('{42:4.2e1}')
self.assertEqual(json.loads(on), {'42': 42.0})
on = js_to_json('{ "0x40": "0x40" }')
self.assertEqual(json.loads(on), {'0x40': '0x40'})
on = js_to_json('{ "040": "040" }')
self.assertEqual(json.loads(on), {'040': '040'})
on = js_to_json('[1,//{},\n2]')
self.assertEqual(json.loads(on), [1, 2])
on = js_to_json(R'"\^\$\#"')
self.assertEqual(json.loads(on), R'^$#', msg='Unnecessary escapes should be stripped')
on = js_to_json('\'"\\""\'')
self.assertEqual(json.loads(on), '"""', msg='Unnecessary quote escape should be escaped')
on = js_to_json('[new Date("spam"), \'("eggs")\']')
self.assertEqual(json.loads(on), ['spam', '("eggs")'], msg='Date regex should match a single string')
def test_js_to_json_malformed(self):
self.assertEqual(js_to_json('42a1'), '42"a1"')
self.assertEqual(js_to_json('42a-1'), '42"a"-1')
def test_js_to_json_template_literal(self):
self.assertEqual(js_to_json('`Hello ${name}`', {'name': '"world"'}), '"Hello world"')
self.assertEqual(js_to_json('`${name}${name}`', {'name': '"X"'}), '"XX"')
self.assertEqual(js_to_json('`${name}${name}`', {'name': '5'}), '"55"')
self.assertEqual(js_to_json('`${name}"${name}"`', {'name': '5'}), '"5\\"5\\""')
self.assertEqual(js_to_json('`${name}`', {}), '"name"')
def test_js_to_json_common_constructors(self):
self.assertEqual(json.loads(js_to_json('new Map([["a", 5]])')), {'a': 5})
self.assertEqual(json.loads(js_to_json('Array(5, 10)')), [5, 10])
self.assertEqual(json.loads(js_to_json('new Array(15,5)')), [15, 5])
self.assertEqual(json.loads(js_to_json('new Map([Array(5, 10),new Array(15,5)])')), {'5': 10, '15': 5})
self.assertEqual(json.loads(js_to_json('new Date("123")')), "123")
self.assertEqual(json.loads(js_to_json('new Date(\'2023-10-19\')')), "2023-10-19")
def test_extract_attributes(self):
self.assertEqual(extract_attributes('<e x="y">'), {'x': 'y'})
self.assertEqual(extract_attributes("<e x='y'>"), {'x': 'y'})
self.assertEqual(extract_attributes('<e x=y>'), {'x': 'y'})
self.assertEqual(extract_attributes('<e x="a \'b\' c">'), {'x': "a 'b' c"})
self.assertEqual(extract_attributes('<e x=\'a "b" c\'>'), {'x': 'a "b" c'})
self.assertEqual(extract_attributes('<e x="&#121;">'), {'x': 'y'})
self.assertEqual(extract_attributes('<e x="&#x79;">'), {'x': 'y'})
self.assertEqual(extract_attributes('<e x="&amp;">'), {'x': '&'}) # XML
self.assertEqual(extract_attributes('<e x="&quot;">'), {'x': '"'})
self.assertEqual(extract_attributes('<e x="&pound;">'), {'x': '£'}) # HTML 3.2
self.assertEqual(extract_attributes('<e x="&lambda;">'), {'x': 'λ'}) # HTML 4.0
self.assertEqual(extract_attributes('<e x="&foo">'), {'x': '&foo'})
self.assertEqual(extract_attributes('<e x="\'">'), {'x': "'"})
self.assertEqual(extract_attributes('<e x=\'"\'>'), {'x': '"'})
self.assertEqual(extract_attributes('<e x >'), {'x': None})
self.assertEqual(extract_attributes('<e x=y a>'), {'x': 'y', 'a': None})
self.assertEqual(extract_attributes('<e x= y>'), {'x': 'y'})
self.assertEqual(extract_attributes('<e x=1 y=2 x=3>'), {'y': '2', 'x': '3'})
self.assertEqual(extract_attributes('<e \nx=\ny\n>'), {'x': 'y'})
self.assertEqual(extract_attributes('<e \nx=\n"y"\n>'), {'x': 'y'})
self.assertEqual(extract_attributes("<e \nx=\n'y'\n>"), {'x': 'y'})
self.assertEqual(extract_attributes('<e \nx="\ny\n">'), {'x': '\ny\n'})
self.assertEqual(extract_attributes('<e CAPS=x>'), {'caps': 'x'}) # Names lowercased
self.assertEqual(extract_attributes('<e x=1 X=2>'), {'x': '2'})
self.assertEqual(extract_attributes('<e X=1 x=2>'), {'x': '2'})
self.assertEqual(extract_attributes('<e _:funny-name1=1>'), {'_:funny-name1': '1'})
self.assertEqual(extract_attributes('<e x="Fáilte 世界 \U0001f600">'), {'x': 'Fáilte 世界 \U0001f600'})
self.assertEqual(extract_attributes('<e x="décompose&#769;">'), {'x': 'décompose\u0301'})
# "Narrow" Python builds don't support unicode code points outside BMP.
try:
chr(0x10000)
supports_outside_bmp = True
except ValueError:
supports_outside_bmp = False
if supports_outside_bmp:
self.assertEqual(extract_attributes('<e x="Smile &#128512;!">'), {'x': 'Smile \U0001f600!'})
# Malformed HTML should not break attributes extraction on older Python
self.assertEqual(extract_attributes('<mal"formed/>'), {})
def test_clean_html(self):
self.assertEqual(clean_html('a:\nb'), 'a: b')
self.assertEqual(clean_html('a:\n "b"'), 'a: "b"')
self.assertEqual(clean_html('a<br>\xa0b'), 'a\nb')
def test_intlist_to_bytes(self):
self.assertEqual(
intlist_to_bytes([0, 1, 127, 128, 255]),
b'\x00\x01\x7f\x80\xff')
def test_args_to_str(self):
self.assertEqual(
args_to_str(['foo', 'ba/r', '-baz', '2 be', '']),
'foo ba/r -baz \'2 be\' \'\'' if compat_os_name != 'nt' else 'foo ba/r -baz "2 be" ""'
)
def test_parse_filesize(self):
self.assertEqual(parse_filesize(None), None)
self.assertEqual(parse_filesize(''), None)
self.assertEqual(parse_filesize('91 B'), 91)
self.assertEqual(parse_filesize('foobar'), None)
self.assertEqual(parse_filesize('2 MiB'), 2097152)
self.assertEqual(parse_filesize('5 GB'), 5000000000)
self.assertEqual(parse_filesize('1.2Tb'), 1200000000000)
self.assertEqual(parse_filesize('1.2tb'), 1200000000000)
self.assertEqual(parse_filesize('1,24 KB'), 1240)
self.assertEqual(parse_filesize('1,24 kb'), 1240)
self.assertEqual(parse_filesize('8.5 megabytes'), 8500000)
def test_parse_count(self):
self.assertEqual(parse_count(None), None)
self.assertEqual(parse_count(''), None)
self.assertEqual(parse_count('0'), 0)
self.assertEqual(parse_count('1000'), 1000)
self.assertEqual(parse_count('1.000'), 1000)
self.assertEqual(parse_count('1.1k'), 1100)
self.assertEqual(parse_count('1.1 k'), 1100)
self.assertEqual(parse_count('1,1 k'), 1100)
self.assertEqual(parse_count('1.1kk'), 1100000)
self.assertEqual(parse_count('1.1kk '), 1100000)
self.assertEqual(parse_count('1,1kk'), 1100000)
self.assertEqual(parse_count('100 views'), 100)
self.assertEqual(parse_count('1,100 views'), 1100)
self.assertEqual(parse_count('1.1kk views'), 1100000)
self.assertEqual(parse_count('10M views'), 10000000)
self.assertEqual(parse_count('has 10M views'), 10000000)
def test_parse_resolution(self):
self.assertEqual(parse_resolution(None), {})
self.assertEqual(parse_resolution(''), {})
self.assertEqual(parse_resolution(' 1920x1080'), {'width': 1920, 'height': 1080})
self.assertEqual(parse_resolution('1920×1080 '), {'width': 1920, 'height': 1080})
self.assertEqual(parse_resolution('1920 x 1080'), {'width': 1920, 'height': 1080})
self.assertEqual(parse_resolution('720p'), {'height': 720})
self.assertEqual(parse_resolution('4k'), {'height': 2160})
self.assertEqual(parse_resolution('8K'), {'height': 4320})
self.assertEqual(parse_resolution('pre_1920x1080_post'), {'width': 1920, 'height': 1080})
self.assertEqual(parse_resolution('ep1x2'), {})
self.assertEqual(parse_resolution('1920, 1080'), {'width': 1920, 'height': 1080})
def test_parse_bitrate(self):
self.assertEqual(parse_bitrate(None), None)
self.assertEqual(parse_bitrate(''), None)
self.assertEqual(parse_bitrate('300kbps'), 300)
self.assertEqual(parse_bitrate('1500kbps'), 1500)
self.assertEqual(parse_bitrate('300 kbps'), 300)
def test_version_tuple(self):
self.assertEqual(version_tuple('1'), (1,))
self.assertEqual(version_tuple('10.23.344'), (10, 23, 344))
self.assertEqual(version_tuple('10.1-6'), (10, 1, 6)) # avconv style
def test_detect_exe_version(self):
self.assertEqual(detect_exe_version('''ffmpeg version 1.2.1
built on May 27 2013 08:37:26 with gcc 4.7 (Debian 4.7.3-4)
configuration: --prefix=/usr --extra-'''), '1.2.1')
self.assertEqual(detect_exe_version('''ffmpeg version N-63176-g1fb4685
built on May 15 2014 22:09:06 with gcc 4.8.2 (GCC)'''), 'N-63176-g1fb4685')
self.assertEqual(detect_exe_version('''X server found. dri2 connection failed!
Trying to open render node...
Success at /dev/dri/renderD128.
ffmpeg version 2.4.4 Copyright (c) 2000-2014 the FFmpeg ...'''), '2.4.4')
def test_age_restricted(self):
self.assertFalse(age_restricted(None, 10)) # unrestricted content
self.assertFalse(age_restricted(1, None)) # unrestricted policy
self.assertFalse(age_restricted(8, 10))
self.assertTrue(age_restricted(18, 14))
self.assertFalse(age_restricted(18, 18))
def test_is_html(self):
self.assertFalse(is_html(b'\x49\x44\x43<html'))
self.assertTrue(is_html(b'<!DOCTYPE foo>\xaaa'))
self.assertTrue(is_html( # UTF-8 with BOM
b'\xef\xbb\xbf<!DOCTYPE foo>\xaaa'))
self.assertTrue(is_html( # UTF-16-LE
b'\xff\xfe<\x00h\x00t\x00m\x00l\x00>\x00\xe4\x00'
))
self.assertTrue(is_html( # UTF-16-BE
b'\xfe\xff\x00<\x00h\x00t\x00m\x00l\x00>\x00\xe4'
))
self.assertTrue(is_html( # UTF-32-BE
b'\x00\x00\xFE\xFF\x00\x00\x00<\x00\x00\x00h\x00\x00\x00t\x00\x00\x00m\x00\x00\x00l\x00\x00\x00>\x00\x00\x00\xe4'))
self.assertTrue(is_html( # UTF-32-LE
b'\xFF\xFE\x00\x00<\x00\x00\x00h\x00\x00\x00t\x00\x00\x00m\x00\x00\x00l\x00\x00\x00>\x00\x00\x00\xe4\x00\x00\x00'))
def test_render_table(self):
self.assertEqual(
render_table(
['a', 'empty', 'bcd'],
[[123, '', 4], [9999, '', 51]]),
'a empty bcd\n'
'123 4\n'
'9999 51')
self.assertEqual(
render_table(
['a', 'empty', 'bcd'],
[[123, '', 4], [9999, '', 51]],
hide_empty=True),
'a bcd\n'
'123 4\n'
'9999 51')
self.assertEqual(
render_table(
['\ta', 'bcd'],
[['1\t23', 4], ['\t9999', 51]]),
' a bcd\n'
'1 23 4\n'
'9999 51')
self.assertEqual(
render_table(
['a', 'bcd'],
[[123, 4], [9999, 51]],
delim='-'),
'a bcd\n'
'--------\n'
'123 4\n'
'9999 51')
self.assertEqual(
render_table(
['a', 'bcd'],
[[123, 4], [9999, 51]],
delim='-', extra_gap=2),
'a bcd\n'
'----------\n'
'123 4\n'
'9999 51')
def test_match_str(self):
# Unary
self.assertFalse(match_str('xy', {'x': 1200}))
self.assertTrue(match_str('!xy', {'x': 1200}))
self.assertTrue(match_str('x', {'x': 1200}))
self.assertFalse(match_str('!x', {'x': 1200}))
self.assertTrue(match_str('x', {'x': 0}))
self.assertTrue(match_str('is_live', {'is_live': True}))
self.assertFalse(match_str('is_live', {'is_live': False}))
self.assertFalse(match_str('is_live', {'is_live': None}))
self.assertFalse(match_str('is_live', {}))
self.assertFalse(match_str('!is_live', {'is_live': True}))
self.assertTrue(match_str('!is_live', {'is_live': False}))
self.assertTrue(match_str('!is_live', {'is_live': None}))
self.assertTrue(match_str('!is_live', {}))
self.assertTrue(match_str('title', {'title': 'abc'}))
self.assertTrue(match_str('title', {'title': ''}))
self.assertFalse(match_str('!title', {'title': 'abc'}))
self.assertFalse(match_str('!title', {'title': ''}))
# Numeric
self.assertFalse(match_str('x>0', {'x': 0}))
self.assertFalse(match_str('x>0', {}))
self.assertTrue(match_str('x>?0', {}))
self.assertTrue(match_str('x>1K', {'x': 1200}))
self.assertFalse(match_str('x>2K', {'x': 1200}))
self.assertTrue(match_str('x>=1200 & x < 1300', {'x': 1200}))
self.assertFalse(match_str('x>=1100 & x < 1200', {'x': 1200}))
self.assertTrue(match_str('x > 1:0:0', {'x': 3700}))
# String
self.assertFalse(match_str('y=a212', {'y': 'foobar42'}))
self.assertTrue(match_str('y=foobar42', {'y': 'foobar42'}))
self.assertFalse(match_str('y!=foobar42', {'y': 'foobar42'}))
self.assertTrue(match_str('y!=foobar2', {'y': 'foobar42'}))
self.assertTrue(match_str('y^=foo', {'y': 'foobar42'}))
self.assertFalse(match_str('y!^=foo', {'y': 'foobar42'}))
self.assertFalse(match_str('y^=bar', {'y': 'foobar42'}))
self.assertTrue(match_str('y!^=bar', {'y': 'foobar42'}))
self.assertRaises(ValueError, match_str, 'x^=42', {'x': 42})
self.assertTrue(match_str('y*=bar', {'y': 'foobar42'}))
self.assertFalse(match_str('y!*=bar', {'y': 'foobar42'}))
self.assertFalse(match_str('y*=baz', {'y': 'foobar42'}))
self.assertTrue(match_str('y!*=baz', {'y': 'foobar42'}))
self.assertTrue(match_str('y$=42', {'y': 'foobar42'}))
self.assertFalse(match_str('y$=43', {'y': 'foobar42'}))
# And
self.assertFalse(match_str(
'like_count > 100 & dislike_count <? 50 & description',
{'like_count': 90, 'description': 'foo'}))
self.assertTrue(match_str(
'like_count > 100 & dislike_count <? 50 & description',
{'like_count': 190, 'description': 'foo'}))
self.assertFalse(match_str(
'like_count > 100 & dislike_count <? 50 & description',
{'like_count': 190, 'dislike_count': 60, 'description': 'foo'}))
self.assertFalse(match_str(
'like_count > 100 & dislike_count <? 50 & description',
{'like_count': 190, 'dislike_count': 10}))
# Regex
self.assertTrue(match_str(r'x~=\bbar', {'x': 'foo bar'}))
self.assertFalse(match_str(r'x~=\bbar.+', {'x': 'foo bar'}))
self.assertFalse(match_str(r'x~=^FOO', {'x': 'foo bar'}))
self.assertTrue(match_str(r'x~=(?i)^FOO', {'x': 'foo bar'}))
# Quotes
self.assertTrue(match_str(r'x^="foo"', {'x': 'foo "bar"'}))
self.assertFalse(match_str(r'x^="foo "', {'x': 'foo "bar"'}))
self.assertFalse(match_str(r'x$="bar"', {'x': 'foo "bar"'}))
self.assertTrue(match_str(r'x$=" \"bar\""', {'x': 'foo "bar"'}))
# Escaping &
self.assertFalse(match_str(r'x=foo & bar', {'x': 'foo & bar'}))
self.assertTrue(match_str(r'x=foo \& bar', {'x': 'foo & bar'}))
self.assertTrue(match_str(r'x=foo \& bar & x^=foo', {'x': 'foo & bar'}))
self.assertTrue(match_str(r'x="foo \& bar" & x^=foo', {'x': 'foo & bar'}))
# Example from docs
self.assertTrue(match_str(
r"!is_live & like_count>?100 & description~='(?i)\bcats \& dogs\b'",
{'description': 'Raining Cats & Dogs'}))
# Incomplete
self.assertFalse(match_str('id!=foo', {'id': 'foo'}, True))
self.assertTrue(match_str('x', {'id': 'foo'}, True))
self.assertTrue(match_str('!x', {'id': 'foo'}, True))
self.assertFalse(match_str('x', {'id': 'foo'}, False))
def test_parse_dfxp_time_expr(self):
self.assertEqual(parse_dfxp_time_expr(None), None)
self.assertEqual(parse_dfxp_time_expr(''), None)
self.assertEqual(parse_dfxp_time_expr('0.1'), 0.1)
self.assertEqual(parse_dfxp_time_expr('0.1s'), 0.1)
self.assertEqual(parse_dfxp_time_expr('00:00:01'), 1.0)
self.assertEqual(parse_dfxp_time_expr('00:00:01.100'), 1.1)
self.assertEqual(parse_dfxp_time_expr('00:00:01:100'), 1.1)
def test_dfxp2srt(self):
dfxp_data = '''<?xml version="1.0" encoding="UTF-8"?>
<tt xmlns="http://www.w3.org/ns/ttml" xml:lang="en" xmlns:tts="http://www.w3.org/ns/ttml#parameter">
<body>
<div xml:lang="en">
<p begin="0" end="1">The following line contains Chinese characters and special symbols</p>
<p begin="1" end="2">第二行<br/>♪♪</p>
<p begin="2" dur="1"><span>Third<br/>Line</span></p>
<p begin="3" end="-1">Lines with invalid timestamps are ignored</p>
<p begin="-1" end="-1">Ignore, two</p>
<p begin="3" dur="-1">Ignored, three</p>
</div>
</body>
</tt>'''.encode()
srt_data = '''1
00:00:00,000 --> 00:00:01,000
The following line contains Chinese characters and special symbols
2
00:00:01,000 --> 00:00:02,000
第二行
♪♪
3
00:00:02,000 --> 00:00:03,000
Third
Line
'''
self.assertEqual(dfxp2srt(dfxp_data), srt_data)
dfxp_data_no_default_namespace = b'''<?xml version="1.0" encoding="UTF-8"?>
<tt xml:lang="en" xmlns:tts="http://www.w3.org/ns/ttml#parameter">
<body>
<div xml:lang="en">
<p begin="0" end="1">The first line</p>
</div>
</body>
</tt>'''
srt_data = '''1
00:00:00,000 --> 00:00:01,000
The first line
'''
self.assertEqual(dfxp2srt(dfxp_data_no_default_namespace), srt_data)
dfxp_data_with_style = b'''<?xml version="1.0" encoding="utf-8"?>
<tt xmlns="http://www.w3.org/2006/10/ttaf1" xmlns:ttp="http://www.w3.org/2006/10/ttaf1#parameter" ttp:timeBase="media" xmlns:tts="http://www.w3.org/2006/10/ttaf1#style" xml:lang="en" xmlns:ttm="http://www.w3.org/2006/10/ttaf1#metadata">
<head>
<styling>
<style id="s2" style="s0" tts:color="cyan" tts:fontWeight="bold" />
<style id="s1" style="s0" tts:color="yellow" tts:fontStyle="italic" />
<style id="s3" style="s0" tts:color="lime" tts:textDecoration="underline" />
<style id="s0" tts:backgroundColor="black" tts:fontStyle="normal" tts:fontSize="16" tts:fontFamily="sansSerif" tts:color="white" />
</styling>
</head>
<body tts:textAlign="center" style="s0">
<div>
<p begin="00:00:02.08" id="p0" end="00:00:05.84">default style<span tts:color="red">custom style</span></p>
<p style="s2" begin="00:00:02.08" id="p0" end="00:00:05.84"><span tts:color="lime">part 1<br /></span><span tts:color="cyan">part 2</span></p>
<p style="s3" begin="00:00:05.84" id="p1" end="00:00:09.56">line 3<br />part 3</p>
<p style="s1" tts:textDecoration="underline" begin="00:00:09.56" id="p2" end="00:00:12.36"><span style="s2" tts:color="lime">inner<br /> </span>style</p>
</div>
</body>
</tt>'''
srt_data = '''1
00:00:02,080 --> 00:00:05,840
<font color="white" face="sansSerif" size="16">default style<font color="red">custom style</font></font>
2
00:00:02,080 --> 00:00:05,840
<b><font color="cyan" face="sansSerif" size="16"><font color="lime">part 1
</font>part 2</font></b>
3
00:00:05,840 --> 00:00:09,560
<u><font color="lime">line 3
part 3</font></u>
4
00:00:09,560 --> 00:00:12,360
<i><u><font color="yellow"><font color="lime">inner
</font>style</font></u></i>
'''
self.assertEqual(dfxp2srt(dfxp_data_with_style), srt_data)
dfxp_data_non_utf8 = '''<?xml version="1.0" encoding="UTF-16"?>
<tt xmlns="http://www.w3.org/ns/ttml" xml:lang="en" xmlns:tts="http://www.w3.org/ns/ttml#parameter">
<body>
<div xml:lang="en">
<p begin="0" end="1">Line 1</p>
<p begin="1" end="2">第二行</p>
</div>
</body>
</tt>'''.encode('utf-16')
srt_data = '''1
00:00:00,000 --> 00:00:01,000
Line 1
2
00:00:01,000 --> 00:00:02,000
第二行
'''
self.assertEqual(dfxp2srt(dfxp_data_non_utf8), srt_data)
def test_cli_option(self):
self.assertEqual(cli_option({'proxy': '127.0.0.1:3128'}, '--proxy', 'proxy'), ['--proxy', '127.0.0.1:3128'])
self.assertEqual(cli_option({'proxy': None}, '--proxy', 'proxy'), [])
self.assertEqual(cli_option({}, '--proxy', 'proxy'), [])
self.assertEqual(cli_option({'retries': 10}, '--retries', 'retries'), ['--retries', '10'])
def test_cli_valueless_option(self):
self.assertEqual(cli_valueless_option(
{'downloader': 'external'}, '--external-downloader', 'downloader', 'external'), ['--external-downloader'])
self.assertEqual(cli_valueless_option(
{'downloader': 'internal'}, '--external-downloader', 'downloader', 'external'), [])
self.assertEqual(cli_valueless_option(
{'nocheckcertificate': True}, '--no-check-certificate', 'nocheckcertificate'), ['--no-check-certificate'])
self.assertEqual(cli_valueless_option(
{'nocheckcertificate': False}, '--no-check-certificate', 'nocheckcertificate'), [])
self.assertEqual(cli_valueless_option(
{'checkcertificate': True}, '--no-check-certificate', 'checkcertificate', False), [])
self.assertEqual(cli_valueless_option(
{'checkcertificate': False}, '--no-check-certificate', 'checkcertificate', False), ['--no-check-certificate'])
def test_cli_bool_option(self):
self.assertEqual(
cli_bool_option(
{'nocheckcertificate': True}, '--no-check-certificate', 'nocheckcertificate'),
['--no-check-certificate', 'true'])
self.assertEqual(
cli_bool_option(
{'nocheckcertificate': True}, '--no-check-certificate', 'nocheckcertificate', separator='='),
['--no-check-certificate=true'])
self.assertEqual(
cli_bool_option(
{'nocheckcertificate': True}, '--check-certificate', 'nocheckcertificate', 'false', 'true'),
['--check-certificate', 'false'])
self.assertEqual(
cli_bool_option(
{'nocheckcertificate': True}, '--check-certificate', 'nocheckcertificate', 'false', 'true', '='),
['--check-certificate=false'])
self.assertEqual(
cli_bool_option(
{'nocheckcertificate': False}, '--check-certificate', 'nocheckcertificate', 'false', 'true'),
['--check-certificate', 'true'])
self.assertEqual(
cli_bool_option(
{'nocheckcertificate': False}, '--check-certificate', 'nocheckcertificate', 'false', 'true', '='),
['--check-certificate=true'])
self.assertEqual(
cli_bool_option(
{}, '--check-certificate', 'nocheckcertificate', 'false', 'true', '='),
[])
def test_ohdave_rsa_encrypt(self):
N = 0xab86b6371b5318aaa1d3c9e612a9f1264f372323c8c0f19875b5fc3b3fd3afcc1e5bec527aa94bfa85bffc157e4245aebda05389a5357b75115ac94f074aefcd
e = 65537
self.assertEqual(
ohdave_rsa_encrypt(b'aa111222', e, N),
'726664bd9a23fd0c70f9f1b84aab5e3905ce1e45a584e9cbcf9bcc7510338fc1986d6c599ff990d923aa43c51c0d9013cd572e13bc58f4ae48f2ed8c0b0ba881')
def test_pkcs1pad(self):
data = [1, 2, 3]
padded_data = pkcs1pad(data, 32)
self.assertEqual(padded_data[:2], [0, 2])
self.assertEqual(padded_data[28:], [0, 1, 2, 3])
self.assertRaises(ValueError, pkcs1pad, data, 8)
def test_encode_base_n(self):
self.assertEqual(encode_base_n(0, 30), '0')
self.assertEqual(encode_base_n(80, 30), '2k')
custom_table = '9876543210ZYXWVUTSRQPONMLKJIHGFEDCBA'
self.assertEqual(encode_base_n(0, 30, custom_table), '9')
self.assertEqual(encode_base_n(80, 30, custom_table), '7P')
self.assertRaises(ValueError, encode_base_n, 0, 70)
self.assertRaises(ValueError, encode_base_n, 0, 60, custom_table)
def test_caesar(self):
self.assertEqual(caesar('ace', 'abcdef', 2), 'cea')
self.assertEqual(caesar('cea', 'abcdef', -2), 'ace')
self.assertEqual(caesar('ace', 'abcdef', -2), 'eac')
self.assertEqual(caesar('eac', 'abcdef', 2), 'ace')
self.assertEqual(caesar('ace', 'abcdef', 0), 'ace')
self.assertEqual(caesar('xyz', 'abcdef', 2), 'xyz')
self.assertEqual(caesar('abc', 'acegik', 2), 'ebg')
self.assertEqual(caesar('ebg', 'acegik', -2), 'abc')
def test_rot47(self):
self.assertEqual(rot47('yt-dlp'), r'JE\5=A')
self.assertEqual(rot47('YT-DLP'), r'*%\s{!')
def test_urshift(self):
self.assertEqual(urshift(3, 1), 1)
self.assertEqual(urshift(-3, 1), 2147483646)
GET_ELEMENT_BY_CLASS_TEST_STRING = '''
<span class="foo bar">nice</span>
'''
def test_get_element_by_class(self):
html = self.GET_ELEMENT_BY_CLASS_TEST_STRING
self.assertEqual(get_element_by_class('foo', html), 'nice')
self.assertEqual(get_element_by_class('no-such-class', html), None)
def test_get_element_html_by_class(self):
html = self.GET_ELEMENT_BY_CLASS_TEST_STRING
self.assertEqual(get_element_html_by_class('foo', html), html.strip())
self.assertEqual(get_element_by_class('no-such-class', html), None)
GET_ELEMENT_BY_ATTRIBUTE_TEST_STRING = '''
<div itemprop="author" itemscope>foo</div>
'''
def test_get_element_by_attribute(self):
html = self.GET_ELEMENT_BY_CLASS_TEST_STRING
self.assertEqual(get_element_by_attribute('class', 'foo bar', html), 'nice')
self.assertEqual(get_element_by_attribute('class', 'foo', html), None)
self.assertEqual(get_element_by_attribute('class', 'no-such-foo', html), None)
html = self.GET_ELEMENT_BY_ATTRIBUTE_TEST_STRING
self.assertEqual(get_element_by_attribute('itemprop', 'author', html), 'foo')
def test_get_element_html_by_attribute(self):
html = self.GET_ELEMENT_BY_CLASS_TEST_STRING
self.assertEqual(get_element_html_by_attribute('class', 'foo bar', html), html.strip())
self.assertEqual(get_element_html_by_attribute('class', 'foo', html), None)
self.assertEqual(get_element_html_by_attribute('class', 'no-such-foo', html), None)
html = self.GET_ELEMENT_BY_ATTRIBUTE_TEST_STRING
self.assertEqual(get_element_html_by_attribute('itemprop', 'author', html), html.strip())
GET_ELEMENTS_BY_CLASS_TEST_STRING = '''
<span class="foo bar">nice</span><span class="foo bar">also nice</span>
'''
GET_ELEMENTS_BY_CLASS_RES = ['<span class="foo bar">nice</span>', '<span class="foo bar">also nice</span>']
def test_get_elements_by_class(self):
html = self.GET_ELEMENTS_BY_CLASS_TEST_STRING
self.assertEqual(get_elements_by_class('foo', html), ['nice', 'also nice'])
self.assertEqual(get_elements_by_class('no-such-class', html), [])
def test_get_elements_html_by_class(self):
html = self.GET_ELEMENTS_BY_CLASS_TEST_STRING
self.assertEqual(get_elements_html_by_class('foo', html), self.GET_ELEMENTS_BY_CLASS_RES)
self.assertEqual(get_elements_html_by_class('no-such-class', html), [])
def test_get_elements_by_attribute(self):
html = self.GET_ELEMENTS_BY_CLASS_TEST_STRING
self.assertEqual(get_elements_by_attribute('class', 'foo bar', html), ['nice', 'also nice'])
self.assertEqual(get_elements_by_attribute('class', 'foo', html), [])
self.assertEqual(get_elements_by_attribute('class', 'no-such-foo', html), [])
def test_get_elements_html_by_attribute(self):
html = self.GET_ELEMENTS_BY_CLASS_TEST_STRING
self.assertEqual(get_elements_html_by_attribute('class', 'foo bar', html), self.GET_ELEMENTS_BY_CLASS_RES)
self.assertEqual(get_elements_html_by_attribute('class', 'foo', html), [])
self.assertEqual(get_elements_html_by_attribute('class', 'no-such-foo', html), [])
def test_get_elements_text_and_html_by_attribute(self):
html = self.GET_ELEMENTS_BY_CLASS_TEST_STRING
self.assertEqual(
list(get_elements_text_and_html_by_attribute('class', 'foo bar', html)),
list(zip(['nice', 'also nice'], self.GET_ELEMENTS_BY_CLASS_RES)))
self.assertEqual(list(get_elements_text_and_html_by_attribute('class', 'foo', html)), [])
self.assertEqual(list(get_elements_text_and_html_by_attribute('class', 'no-such-foo', html)), [])
self.assertEqual(list(get_elements_text_and_html_by_attribute(
'class', 'foo', '<a class="foo">nice</a><span class="foo">nice</span>', tag='a')), [('nice', '<a class="foo">nice</a>')])
GET_ELEMENT_BY_TAG_TEST_STRING = '''
random text lorem ipsum</p>
<div>
this should be returned
<span>this should also be returned</span>
<div>
this should also be returned
</div>
closing tag above should not trick, so this should also be returned
</div>
but this text should not be returned
'''
GET_ELEMENT_BY_TAG_RES_OUTERDIV_HTML = GET_ELEMENT_BY_TAG_TEST_STRING.strip()[32:276]
GET_ELEMENT_BY_TAG_RES_OUTERDIV_TEXT = GET_ELEMENT_BY_TAG_RES_OUTERDIV_HTML[5:-6]
GET_ELEMENT_BY_TAG_RES_INNERSPAN_HTML = GET_ELEMENT_BY_TAG_TEST_STRING.strip()[78:119]
GET_ELEMENT_BY_TAG_RES_INNERSPAN_TEXT = GET_ELEMENT_BY_TAG_RES_INNERSPAN_HTML[6:-7]
def test_get_element_text_and_html_by_tag(self):
html = self.GET_ELEMENT_BY_TAG_TEST_STRING
self.assertEqual(
get_element_text_and_html_by_tag('div', html),
(self.GET_ELEMENT_BY_TAG_RES_OUTERDIV_TEXT, self.GET_ELEMENT_BY_TAG_RES_OUTERDIV_HTML))
self.assertEqual(
get_element_text_and_html_by_tag('span', html),
(self.GET_ELEMENT_BY_TAG_RES_INNERSPAN_TEXT, self.GET_ELEMENT_BY_TAG_RES_INNERSPAN_HTML))
self.assertRaises(compat_HTMLParseError, get_element_text_and_html_by_tag, 'article', html)
def test_iri_to_uri(self):
self.assertEqual(
iri_to_uri('https://www.google.com/search?q=foo&ie=utf-8&oe=utf-8&client=firefox-b'),
'https://www.google.com/search?q=foo&ie=utf-8&oe=utf-8&client=firefox-b') # Same
self.assertEqual(
iri_to_uri('https://www.google.com/search?q=Käsesoßenrührlöffel'), # German for cheese sauce stirring spoon
'https://www.google.com/search?q=K%C3%A4seso%C3%9Fenr%C3%BChrl%C3%B6ffel')
self.assertEqual(
iri_to_uri('https://www.google.com/search?q=lt<+gt>+eq%3D+amp%26+percent%25+hash%23+colon%3A+tilde~#trash=?&garbage=#'),
'https://www.google.com/search?q=lt%3C+gt%3E+eq%3D+amp%26+percent%25+hash%23+colon%3A+tilde~#trash=?&garbage=#')
self.assertEqual(
iri_to_uri('http://правозащита38.рф/category/news/'),
'http://xn--38-6kcaak9aj5chl4a3g.xn--p1ai/category/news/')
self.assertEqual(
iri_to_uri('http://www.правозащита38.рф/category/news/'),
'http://www.xn--38-6kcaak9aj5chl4a3g.xn--p1ai/category/news/')
self.assertEqual(
iri_to_uri('https://i❤.ws/emojidomain/👍👏🤝💪'),
'https://xn--i-7iq.ws/emojidomain/%F0%9F%91%8D%F0%9F%91%8F%F0%9F%A4%9D%F0%9F%92%AA')
self.assertEqual(
iri_to_uri('http://日本語.jp/'),
'http://xn--wgv71a119e.jp/')
self.assertEqual(
iri_to_uri('http://导航.中国/'),
'http://xn--fet810g.xn--fiqs8s/')
def test_clean_podcast_url(self):
self.assertEqual(clean_podcast_url('https://www.podtrac.com/pts/redirect.mp3/chtbl.com/track/5899E/traffic.megaphone.fm/HSW7835899191.mp3'), 'https://traffic.megaphone.fm/HSW7835899191.mp3')
self.assertEqual(clean_podcast_url('https://play.podtrac.com/npr-344098539/edge1.pod.npr.org/anon.npr-podcasts/podcast/npr/waitwait/2020/10/20201003_waitwait_wwdtmpodcast201003-015621a5-f035-4eca-a9a1-7c118d90bc3c.mp3'), 'https://edge1.pod.npr.org/anon.npr-podcasts/podcast/npr/waitwait/2020/10/20201003_waitwait_wwdtmpodcast201003-015621a5-f035-4eca-a9a1-7c118d90bc3c.mp3')
self.assertEqual(clean_podcast_url('https://pdst.fm/e/2.gum.fm/chtbl.com/track/chrt.fm/track/34D33/pscrb.fm/rss/p/traffic.megaphone.fm/ITLLC7765286967.mp3?updated=1687282661'), 'https://traffic.megaphone.fm/ITLLC7765286967.mp3?updated=1687282661')
self.assertEqual(clean_podcast_url('https://pdst.fm/e/https://mgln.ai/e/441/www.buzzsprout.com/1121972/13019085-ep-252-the-deep-life-stack.mp3'), 'https://www.buzzsprout.com/1121972/13019085-ep-252-the-deep-life-stack.mp3')
def test_LazyList(self):
it = list(range(10))
self.assertEqual(list(LazyList(it)), it)
self.assertEqual(LazyList(it).exhaust(), it)
self.assertEqual(LazyList(it)[5], it[5])
self.assertEqual(LazyList(it)[5:], it[5:])
self.assertEqual(LazyList(it)[:5], it[:5])
self.assertEqual(LazyList(it)[::2], it[::2])
self.assertEqual(LazyList(it)[1::2], it[1::2])
self.assertEqual(LazyList(it)[5::-1], it[5::-1])
self.assertEqual(LazyList(it)[6:2:-2], it[6:2:-2])
self.assertEqual(LazyList(it)[::-1], it[::-1])
self.assertTrue(LazyList(it))
self.assertFalse(LazyList(range(0)))
self.assertEqual(len(LazyList(it)), len(it))
self.assertEqual(repr(LazyList(it)), repr(it))
self.assertEqual(str(LazyList(it)), str(it))
self.assertEqual(list(LazyList(it, reverse=True)), it[::-1])
self.assertEqual(list(reversed(LazyList(it))[::-1]), it)
self.assertEqual(list(reversed(LazyList(it))[1:3:7]), it[::-1][1:3:7])
def test_LazyList_laziness(self):
def test(ll, idx, val, cache):
self.assertEqual(ll[idx], val)
self.assertEqual(ll._cache, list(cache))
ll = LazyList(range(10))
test(ll, 0, 0, range(1))
test(ll, 5, 5, range(6))
test(ll, -3, 7, range(10))
ll = LazyList(range(10), reverse=True)
test(ll, -1, 0, range(1))
test(ll, 3, 6, range(10))
ll = LazyList(itertools.count())
test(ll, 10, 10, range(11))
ll = reversed(ll)
test(ll, -15, 14, range(15))
def test_format_bytes(self):
self.assertEqual(format_bytes(0), '0.00B')
self.assertEqual(format_bytes(1000), '1000.00B')
self.assertEqual(format_bytes(1024), '1.00KiB')
self.assertEqual(format_bytes(1024**2), '1.00MiB')
self.assertEqual(format_bytes(1024**3), '1.00GiB')
self.assertEqual(format_bytes(1024**4), '1.00TiB')
self.assertEqual(format_bytes(1024**5), '1.00PiB')
self.assertEqual(format_bytes(1024**6), '1.00EiB')
self.assertEqual(format_bytes(1024**7), '1.00ZiB')
self.assertEqual(format_bytes(1024**8), '1.00YiB')
self.assertEqual(format_bytes(1024**9), '1024.00YiB')
def test_hide_login_info(self):
self.assertEqual(Config.hide_login_info(['-u', 'foo', '-p', 'bar']),
['-u', 'PRIVATE', '-p', 'PRIVATE'])
self.assertEqual(Config.hide_login_info(['-u']), ['-u'])
self.assertEqual(Config.hide_login_info(['-u', 'foo', '-u', 'bar']),
['-u', 'PRIVATE', '-u', 'PRIVATE'])
self.assertEqual(Config.hide_login_info(['--username=foo']),
['--username=PRIVATE'])
def test_locked_file(self):
TEXT = 'test_locked_file\n'
FILE = 'test_locked_file.ytdl'
MODES = 'war' # Order is important
try:
for lock_mode in MODES:
with locked_file(FILE, lock_mode, False) as f:
if lock_mode == 'r':
self.assertEqual(f.read(), TEXT * 2, 'Wrong file content')
else:
f.write(TEXT)
for test_mode in MODES:
testing_write = test_mode != 'r'
try:
with locked_file(FILE, test_mode, False):
pass
except (BlockingIOError, PermissionError):
if not testing_write: # FIXME
print(f'Known issue: Exclusive lock ({lock_mode}) blocks read access ({test_mode})')
continue
self.assertTrue(testing_write, f'{test_mode} is blocked by {lock_mode}')
else:
self.assertFalse(testing_write, f'{test_mode} is not blocked by {lock_mode}')
finally:
with contextlib.suppress(OSError):
os.remove(FILE)
def test_determine_file_encoding(self):
self.assertEqual(determine_file_encoding(b''), (None, 0))
self.assertEqual(determine_file_encoding(b'--verbose -x --audio-format mkv\n'), (None, 0))
self.assertEqual(determine_file_encoding(b'\xef\xbb\xbf'), ('utf-8', 3))
self.assertEqual(determine_file_encoding(b'\x00\x00\xfe\xff'), ('utf-32-be', 4))
self.assertEqual(determine_file_encoding(b'\xff\xfe'), ('utf-16-le', 2))
self.assertEqual(determine_file_encoding(b'\xff\xfe# coding: utf-8\n--verbose'), ('utf-16-le', 2))
self.assertEqual(determine_file_encoding(b'# coding: utf-8\n--verbose'), ('utf-8', 0))
self.assertEqual(determine_file_encoding(b'# coding: someencodinghere-12345\n--verbose'), ('someencodinghere-12345', 0))
self.assertEqual(determine_file_encoding(b'#coding:utf-8\n--verbose'), ('utf-8', 0))
self.assertEqual(determine_file_encoding(b'# coding: utf-8 \r\n--verbose'), ('utf-8', 0))
self.assertEqual(determine_file_encoding('# coding: utf-32-be'.encode('utf-32-be')), ('utf-32-be', 0))
self.assertEqual(determine_file_encoding('# coding: utf-16-le'.encode('utf-16-le')), ('utf-16-le', 0))
def test_get_compatible_ext(self):
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None, None], vexts=['mp4'], aexts=['m4a', 'm4a']), 'mkv')
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None], vexts=['flv'], aexts=['flv']), 'flv')
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None], vexts=['mp4'], aexts=['m4a']), 'mp4')
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None], vexts=['mp4'], aexts=['webm']), 'mkv')
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None], vexts=['webm'], aexts=['m4a']), 'mkv')
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None], vexts=['webm'], aexts=['webm']), 'webm')
self.assertEqual(get_compatible_ext(
vcodecs=[None], acodecs=[None], vexts=['webm'], aexts=['weba']), 'webm')
self.assertEqual(get_compatible_ext(
vcodecs=['h264'], acodecs=['mp4a'], vexts=['mov'], aexts=['m4a']), 'mp4')
self.assertEqual(get_compatible_ext(
vcodecs=['av01.0.12M.08'], acodecs=['opus'], vexts=['mp4'], aexts=['webm']), 'webm')
self.assertEqual(get_compatible_ext(
vcodecs=['vp9'], acodecs=['opus'], vexts=['webm'], aexts=['webm'], preferences=['flv', 'mp4']), 'mp4')
self.assertEqual(get_compatible_ext(
vcodecs=['av1'], acodecs=['mp4a'], vexts=['webm'], aexts=['m4a'], preferences=('webm', 'mkv')), 'mkv')
def test_try_call(self):
def total(*x, **kwargs):
return sum(x) + sum(kwargs.values())
self.assertEqual(try_call(None), None,
msg='not a fn should give None')
self.assertEqual(try_call(lambda: 1), 1,
msg='int fn with no expected_type should give int')
self.assertEqual(try_call(lambda: 1, expected_type=int), 1,
msg='int fn with expected_type int should give int')
self.assertEqual(try_call(lambda: 1, expected_type=dict), None,
msg='int fn with wrong expected_type should give None')
self.assertEqual(try_call(total, args=(0, 1, 0, ), expected_type=int), 1,
msg='fn should accept arglist')
self.assertEqual(try_call(total, kwargs={'a': 0, 'b': 1, 'c': 0}, expected_type=int), 1,
msg='fn should accept kwargs')
self.assertEqual(try_call(lambda: 1, expected_type=dict), None,
msg='int fn with no expected_type should give None')
self.assertEqual(try_call(lambda x: {}, total, args=(42, ), expected_type=int), 42,
msg='expect first int result with expected_type int')
def test_variadic(self):
self.assertEqual(variadic(None), (None, ))
self.assertEqual(variadic('spam'), ('spam', ))
self.assertEqual(variadic('spam', allowed_types=dict), 'spam')
with warnings.catch_warnings():
warnings.simplefilter('ignore')
self.assertEqual(variadic('spam', allowed_types=[dict]), 'spam')
def test_traverse_obj(self):
_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': {},
}
# Test base functionality
self.assertEqual(traverse_obj(_TEST_DATA, ('str',)), 'str',
msg='allow tuple path')
self.assertEqual(traverse_obj(_TEST_DATA, ['str']), 'str',
msg='allow list path')
self.assertEqual(traverse_obj(_TEST_DATA, (value for value in ("str",))), 'str',
msg='allow iterable path')
self.assertEqual(traverse_obj(_TEST_DATA, 'str'), 'str',
msg='single items should be treated as a path')
self.assertEqual(traverse_obj(_TEST_DATA, None), _TEST_DATA)
self.assertEqual(traverse_obj(_TEST_DATA, 100), 100)
self.assertEqual(traverse_obj(_TEST_DATA, 1.2), 1.2)
# Test Ellipsis behavior
self.assertCountEqual(traverse_obj(_TEST_DATA, ...),
(item for item in _TEST_DATA.values() if item not in (None, {})),
msg='`...` should give all non discarded values')
self.assertCountEqual(traverse_obj(_TEST_DATA, ('urls', 0, ...)), _TEST_DATA['urls'][0].values(),
msg='`...` selection for dicts should select all values')
self.assertEqual(traverse_obj(_TEST_DATA, (..., ..., 'url')),
['https://www.example.com/0', 'https://www.example.com/1'],
msg='nested `...` queries should work')
self.assertCountEqual(traverse_obj(_TEST_DATA, (..., ..., 'index')), range(4),
msg='`...` query result should be flattened')
self.assertEqual(traverse_obj(iter(range(4)), ...), list(range(4)),
msg='`...` should accept iterables')
# Test function as key
self.assertEqual(traverse_obj(_TEST_DATA, lambda x, y: x == 'urls' and isinstance(y, list)),
[_TEST_DATA['urls']],
msg='function as query key should perform a filter based on (key, value)')
self.assertCountEqual(traverse_obj(_TEST_DATA, lambda _, x: isinstance(x[0], str)), {'str'},
msg='exceptions in the query function should be catched')
self.assertEqual(traverse_obj(iter(range(4)), lambda _, x: x % 2 == 0), [0, 2],
msg='function key should accept iterables')
if __debug__:
with self.assertRaises(Exception, msg='Wrong function signature should raise in debug'):
traverse_obj(_TEST_DATA, lambda a: ...)
with self.assertRaises(Exception, msg='Wrong function signature should raise in debug'):
traverse_obj(_TEST_DATA, lambda a, b, c: ...)
# Test set as key (transformation/type, like `expected_type`)
self.assertEqual(traverse_obj(_TEST_DATA, (..., {str.upper}, )), ['STR'],
msg='Function in set should be a transformation')
self.assertEqual(traverse_obj(_TEST_DATA, (..., {str})), ['str'],
msg='Type in set should be a type filter')
self.assertEqual(traverse_obj(_TEST_DATA, {dict}), _TEST_DATA,
msg='A single set should be wrapped into a path')
self.assertEqual(traverse_obj(_TEST_DATA, (..., {str.upper})), ['STR'],
msg='Transformation function should not raise')
self.assertEqual(traverse_obj(_TEST_DATA, (..., {str_or_none})),
[item for item in map(str_or_none, _TEST_DATA.values()) if item is not None],
msg='Function in set should be a transformation')
self.assertEqual(traverse_obj(_TEST_DATA, ('fail', {lambda _: 'const'})), 'const',
msg='Function in set should always be called')
if __debug__:
with self.assertRaises(Exception, msg='Sets with length != 1 should raise in debug'):
traverse_obj(_TEST_DATA, set())
with self.assertRaises(Exception, msg='Sets with length != 1 should raise in debug'):
traverse_obj(_TEST_DATA, {str.upper, str})
# Test `slice` as a key
_SLICE_DATA = [0, 1, 2, 3, 4]
self.assertEqual(traverse_obj(_TEST_DATA, ('dict', slice(1))), None,
msg='slice on a dictionary should not throw')
self.assertEqual(traverse_obj(_SLICE_DATA, slice(1)), _SLICE_DATA[:1],
msg='slice key should apply slice to sequence')
self.assertEqual(traverse_obj(_SLICE_DATA, slice(1, 2)), _SLICE_DATA[1:2],
msg='slice key should apply slice to sequence')
self.assertEqual(traverse_obj(_SLICE_DATA, slice(1, 4, 2)), _SLICE_DATA[1:4:2],
msg='slice key should apply slice to sequence')
# Test alternative paths
self.assertEqual(traverse_obj(_TEST_DATA, 'fail', 'str'), 'str',
msg='multiple `paths` should be treated as alternative paths')
self.assertEqual(traverse_obj(_TEST_DATA, 'str', 100), 'str',
msg='alternatives should exit early')
self.assertEqual(traverse_obj(_TEST_DATA, 'fail', 'fail'), None,
msg='alternatives should return `default` if exhausted')
self.assertEqual(traverse_obj(_TEST_DATA, (..., 'fail'), 100), 100,
msg='alternatives should track their own branching return')
self.assertEqual(traverse_obj(_TEST_DATA, ('dict', ...), ('data', ...)), list(_TEST_DATA['data']),
msg='alternatives on empty objects should search further')
# Test branch and path nesting
self.assertEqual(traverse_obj(_TEST_DATA, ('urls', (3, 0), 'url')), ['https://www.example.com/0'],
msg='tuple as key should be treated as branches')
self.assertEqual(traverse_obj(_TEST_DATA, ('urls', [3, 0], 'url')), ['https://www.example.com/0'],
msg='list as key should be treated as branches')
self.assertEqual(traverse_obj(_TEST_DATA, ('urls', ((1, 'fail'), (0, 'url')))), ['https://www.example.com/0'],
msg='double nesting in path should be treated as paths')
self.assertEqual(traverse_obj(['0', [1, 2]], [(0, 1), 0]), [1],
msg='do not fail early on branching')
self.assertCountEqual(traverse_obj(_TEST_DATA, ('urls', ((1, ('fail', 'url')), (0, 'url')))),
['https://www.example.com/0', 'https://www.example.com/1'],
msg='tripple nesting in path should be treated as branches')
self.assertEqual(traverse_obj(_TEST_DATA, ('urls', ('fail', (..., 'url')))),
['https://www.example.com/0', 'https://www.example.com/1'],
msg='ellipsis as branch path start gets flattened')
# Test dictionary as key
self.assertEqual(traverse_obj(_TEST_DATA, {0: 100, 1: 1.2}), {0: 100, 1: 1.2},
msg='dict key should result in a dict with the same keys')
self.assertEqual(traverse_obj(_TEST_DATA, {0: ('urls', 0, 'url')}),
{0: 'https://www.example.com/0'},
msg='dict key should allow paths')
self.assertEqual(traverse_obj(_TEST_DATA, {0: ('urls', (3, 0), 'url')}),
{0: ['https://www.example.com/0']},
msg='tuple in dict path should be treated as branches')
self.assertEqual(traverse_obj(_TEST_DATA, {0: ('urls', ((1, 'fail'), (0, 'url')))}),
{0: ['https://www.example.com/0']},
msg='double nesting in dict path should be treated as paths')
self.assertEqual(traverse_obj(_TEST_DATA, {0: ('urls', ((1, ('fail', 'url')), (0, 'url')))}),
{0: ['https://www.example.com/1', 'https://www.example.com/0']},
msg='tripple nesting in dict path should be treated as branches')
self.assertEqual(traverse_obj(_TEST_DATA, {0: 'fail'}), {},
msg='remove `None` values when top level dict key fails')
self.assertEqual(traverse_obj(_TEST_DATA, {0: 'fail'}, default=...), {0: ...},
msg='use `default` if key fails and `default`')
self.assertEqual(traverse_obj(_TEST_DATA, {0: 'dict'}), {},
msg='remove empty values when dict key')
self.assertEqual(traverse_obj(_TEST_DATA, {0: 'dict'}, default=...), {0: ...},
msg='use `default` when dict key and `default`')
self.assertEqual(traverse_obj(_TEST_DATA, {0: {0: 'fail'}}), {},
msg='remove empty values when nested dict key fails')
self.assertEqual(traverse_obj(None, {0: 'fail'}), {},
msg='default to dict if pruned')
self.assertEqual(traverse_obj(None, {0: 'fail'}, default=...), {0: ...},
msg='default to dict if pruned and default is given')
self.assertEqual(traverse_obj(_TEST_DATA, {0: {0: 'fail'}}, default=...), {0: {0: ...}},
msg='use nested `default` when nested dict key fails and `default`')
self.assertEqual(traverse_obj(_TEST_DATA, {0: ('dict', ...)}), {},
msg='remove key if branch in dict key not successful')
# Testing default parameter behavior
_DEFAULT_DATA = {'None': None, 'int': 0, 'list': []}
self.assertEqual(traverse_obj(_DEFAULT_DATA, 'fail'), None,
msg='default value should be `None`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, 'fail', 'fail', default=...), ...,
msg='chained fails should result in default')
self.assertEqual(traverse_obj(_DEFAULT_DATA, 'None', 'int'), 0,
msg='should not short cirquit on `None`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, 'fail', default=1), 1,
msg='invalid dict key should result in `default`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, 'None', default=1), 1,
msg='`None` is a deliberate sentinel and should become `default`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, ('list', 10)), None,
msg='`IndexError` should result in `default`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, (..., 'fail'), default=1), 1,
msg='if branched but not successful return `default` if defined, not `[]`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, (..., 'fail'), default=None), None,
msg='if branched but not successful return `default` even if `default` is `None`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, (..., 'fail')), [],
msg='if branched but not successful return `[]`, not `default`')
self.assertEqual(traverse_obj(_DEFAULT_DATA, ('list', ...)), [],
msg='if branched but object is empty return `[]`, not `default`')
self.assertEqual(traverse_obj(None, ...), [],
msg='if branched but object is `None` return `[]`, not `default`')
self.assertEqual(traverse_obj({0: None}, (0, ...)), [],
msg='if branched but state is `None` return `[]`, not `default`')
branching_paths = [
('fail', ...),
(..., 'fail'),
100 * ('fail',) + (...,),
(...,) + 100 * ('fail',),
]
for branching_path in branching_paths:
self.assertEqual(traverse_obj({}, branching_path), [],
msg='if branched but state is `None`, return `[]` (not `default`)')
self.assertEqual(traverse_obj({}, 'fail', branching_path), [],
msg='if branching in last alternative and previous did not match, return `[]` (not `default`)')
self.assertEqual(traverse_obj({0: 'x'}, 0, branching_path), 'x',
msg='if branching in last alternative and previous did match, return single value')
self.assertEqual(traverse_obj({0: 'x'}, branching_path, 0), 'x',
msg='if branching in first alternative and non-branching path does match, return single value')
self.assertEqual(traverse_obj({}, branching_path, 'fail'), None,
msg='if branching in first alternative and non-branching path does not match, return `default`')
# Testing expected_type behavior
_EXPECTED_TYPE_DATA = {'str': 'str', 'int': 0}
self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, 'str', expected_type=str),
'str', msg='accept matching `expected_type` type')
self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, 'str', expected_type=int),
None, msg='reject non matching `expected_type` type')
self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, 'int', expected_type=lambda x: str(x)),
'0', msg='transform type using type function')
self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, 'str', expected_type=lambda _: 1 / 0),
None, msg='wrap expected_type fuction in try_call')
self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, ..., expected_type=str),
['str'], msg='eliminate items that expected_type fails on')
self.assertEqual(traverse_obj(_TEST_DATA, {0: 100, 1: 1.2}, expected_type=int),
{0: 100}, msg='type as expected_type should filter dict values')
self.assertEqual(traverse_obj(_TEST_DATA, {0: 100, 1: 1.2, 2: 'None'}, expected_type=str_or_none),
{0: '100', 1: '1.2'}, msg='function as expected_type should transform dict values')
self.assertEqual(traverse_obj(_TEST_DATA, ({0: 1.2}, 0, {int_or_none}), expected_type=int),
1, msg='expected_type should not filter non final dict values')
self.assertEqual(traverse_obj(_TEST_DATA, {0: {0: 100, 1: 'str'}}, expected_type=int),
{0: {0: 100}}, msg='expected_type should transform deep dict values')
self.assertEqual(traverse_obj(_TEST_DATA, [({0: '...'}, {0: '...'})], expected_type=type(...)),
[{0: ...}, {0: ...}], msg='expected_type should transform branched dict values')
self.assertEqual(traverse_obj({1: {3: 4}}, [(1, 2), 3], expected_type=int),
[4], msg='expected_type regression for type matching in tuple branching')
self.assertEqual(traverse_obj(_TEST_DATA, ['data', ...], expected_type=int),
[], msg='expected_type regression for type matching in dict result')
# Test get_all behavior
_GET_ALL_DATA = {'key': [0, 1, 2]}
self.assertEqual(traverse_obj(_GET_ALL_DATA, ('key', ...), get_all=False), 0,
msg='if not `get_all`, return only first matching value')
self.assertEqual(traverse_obj(_GET_ALL_DATA, ..., get_all=False), [0, 1, 2],
msg='do not overflatten if not `get_all`')
# Test casesense behavior
_CASESENSE_DATA = {
'KeY': 'value0',
0: {
'KeY': 'value1',
0: {'KeY': 'value2'},
},
}
self.assertEqual(traverse_obj(_CASESENSE_DATA, 'key'), None,
msg='dict keys should be case sensitive unless `casesense`')
self.assertEqual(traverse_obj(_CASESENSE_DATA, 'keY',
casesense=False), 'value0',
msg='allow non matching key case if `casesense`')
self.assertEqual(traverse_obj(_CASESENSE_DATA, (0, ('keY',)),
casesense=False), ['value1'],
msg='allow non matching key case in branch if `casesense`')
self.assertEqual(traverse_obj(_CASESENSE_DATA, (0, ((0, 'keY'),)),
casesense=False), ['value2'],
msg='allow non matching key case in branch path if `casesense`')
# Test traverse_string behavior
_TRAVERSE_STRING_DATA = {'str': 'str', 1.2: 1.2}
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', 0)), None,
msg='do not traverse into string if not `traverse_string`')
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', 0),
traverse_string=True), 's',
msg='traverse into string if `traverse_string`')
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, (1.2, 1),
traverse_string=True), '.',
msg='traverse into converted data if `traverse_string`')
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', ...),
traverse_string=True), 'str',
msg='`...` should result in string (same value) if `traverse_string`')
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', slice(0, None, 2)),
traverse_string=True), 'sr',
msg='`slice` should result in string if `traverse_string`')
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', lambda i, v: i or v == "s"),
traverse_string=True), 'str',
msg='function should result in string if `traverse_string`')
self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', (0, 2)),
traverse_string=True), ['s', 'r'],
msg='branching should result in list if `traverse_string`')
self.assertEqual(traverse_obj({}, (0, ...), traverse_string=True), [],
msg='branching should result in list if `traverse_string`')
self.assertEqual(traverse_obj({}, (0, lambda x, y: True), traverse_string=True), [],
msg='branching should result in list if `traverse_string`')
self.assertEqual(traverse_obj({}, (0, slice(1)), traverse_string=True), [],
msg='branching should result in list if `traverse_string`')
# Test re.Match as input obj
mobj = re.fullmatch(r'0(12)(?P<group>3)(4)?', '0123')
self.assertEqual(traverse_obj(mobj, ...), [x for x in mobj.groups() if x is not None],
msg='`...` on a `re.Match` should give its `groups()`')
self.assertEqual(traverse_obj(mobj, lambda k, _: k in (0, 2)), ['0123', '3'],
msg='function on a `re.Match` should give groupno, value starting at 0')
self.assertEqual(traverse_obj(mobj, 'group'), '3',
msg='str key on a `re.Match` should give group with that name')
self.assertEqual(traverse_obj(mobj, 2), '3',
msg='int key on a `re.Match` should give group with that name')
self.assertEqual(traverse_obj(mobj, 'gRoUp', casesense=False), '3',
msg='str key on a `re.Match` should respect casesense')
self.assertEqual(traverse_obj(mobj, 'fail'), None,
msg='failing str key on a `re.Match` should return `default`')
self.assertEqual(traverse_obj(mobj, 'gRoUpS', casesense=False), None,
msg='failing str key on a `re.Match` should return `default`')
self.assertEqual(traverse_obj(mobj, 8), None,
msg='failing int key on a `re.Match` should return `default`')
self.assertEqual(traverse_obj(mobj, lambda k, _: k in (0, 'group')), ['0123', '3'],
msg='function on a `re.Match` should give group name as well')
def test_http_header_dict(self):
headers = HTTPHeaderDict()
headers['ytdl-test'] = b'0'
self.assertEqual(list(headers.items()), [('Ytdl-Test', '0')])
headers['ytdl-test'] = 1
self.assertEqual(list(headers.items()), [('Ytdl-Test', '1')])
headers['Ytdl-test'] = '2'
self.assertEqual(list(headers.items()), [('Ytdl-Test', '2')])
self.assertTrue('ytDl-Test' in headers)
self.assertEqual(str(headers), str(dict(headers)))
self.assertEqual(repr(headers), str(dict(headers)))
headers.update({'X-dlp': 'data'})
self.assertEqual(set(headers.items()), {('Ytdl-Test', '2'), ('X-Dlp', 'data')})
self.assertEqual(dict(headers), {'Ytdl-Test': '2', 'X-Dlp': 'data'})
self.assertEqual(len(headers), 2)
self.assertEqual(headers.copy(), headers)
headers2 = HTTPHeaderDict({'X-dlp': 'data3'}, **headers, **{'X-dlp': 'data2'})
self.assertEqual(set(headers2.items()), {('Ytdl-Test', '2'), ('X-Dlp', 'data2')})
self.assertEqual(len(headers2), 2)
headers2.clear()
self.assertEqual(len(headers2), 0)
# ensure we prefer latter headers
headers3 = HTTPHeaderDict({'Ytdl-TeSt': 1}, {'Ytdl-test': 2})
self.assertEqual(set(headers3.items()), {('Ytdl-Test', '2')})
del headers3['ytdl-tesT']
self.assertEqual(dict(headers3), {})
headers4 = HTTPHeaderDict({'ytdl-test': 'data;'})
self.assertEqual(set(headers4.items()), {('Ytdl-Test', 'data;')})
# common mistake: strip whitespace from values
# https://github.com/yt-dlp/yt-dlp/issues/8729
headers5 = HTTPHeaderDict({'ytdl-test': ' data; '})
self.assertEqual(set(headers5.items()), {('Ytdl-Test', 'data;')})
def test_extract_basic_auth(self):
assert extract_basic_auth('http://:foo.bar') == ('http://:foo.bar', None)
assert extract_basic_auth('http://foo.bar') == ('http://foo.bar', None)
assert extract_basic_auth('http://@foo.bar') == ('http://foo.bar', 'Basic Og==')
assert extract_basic_auth('http://:pass@foo.bar') == ('http://foo.bar', 'Basic OnBhc3M=')
assert extract_basic_auth('http://user:@foo.bar') == ('http://foo.bar', 'Basic dXNlcjo=')
assert extract_basic_auth('http://user:pass@foo.bar') == ('http://foo.bar', 'Basic dXNlcjpwYXNz')
@unittest.skipUnless(compat_os_name == 'nt', 'Only relevant on Windows')
def test_Popen_windows_escaping(self):
def run_shell(args):
stdout, stderr, error = Popen.run(
args, text=True, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
assert not stderr
assert not error
return stdout
# Test escaping
assert run_shell(['echo', 'test"&']) == '"test""&"\n'
# Test if delayed expansion is disabled
assert run_shell(['echo', '^!']) == '"^!"\n'
assert run_shell('echo "^!"') == '"^!"\n'
if __name__ == '__main__':
unittest.main()