mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-18 05:19:59 +00:00
parent
85b87c991a
commit
d710a6ca7c
4 changed files with 261 additions and 11 deletions
|
@ -4,8 +4,18 @@
|
||||||
|
|
||||||
import pytest
|
import pytest
|
||||||
|
|
||||||
from yt_dlp.utils import dict_get, int_or_none, str_or_none
|
from yt_dlp.utils import (
|
||||||
from yt_dlp.utils.traversal import traverse_obj
|
ExtractorError,
|
||||||
|
determine_ext,
|
||||||
|
dict_get,
|
||||||
|
int_or_none,
|
||||||
|
str_or_none,
|
||||||
|
)
|
||||||
|
from yt_dlp.utils.traversal import (
|
||||||
|
traverse_obj,
|
||||||
|
require,
|
||||||
|
subs_list_to_dict,
|
||||||
|
)
|
||||||
|
|
||||||
_TEST_DATA = {
|
_TEST_DATA = {
|
||||||
100: 100,
|
100: 100,
|
||||||
|
@ -420,6 +430,71 @@ def test_traversal_morsel(self):
|
||||||
assert traverse_obj(morsel, [(None,), any]) == morsel, \
|
assert traverse_obj(morsel, [(None,), any]) == morsel, \
|
||||||
'Morsel should not be implicitly changed to dict on usage'
|
'Morsel should not be implicitly changed to dict on usage'
|
||||||
|
|
||||||
|
def test_traversal_filter(self):
|
||||||
|
data = [None, False, True, 0, 1, 0.0, 1.1, '', 'str', {}, {0: 0}, [], [1]]
|
||||||
|
|
||||||
|
assert traverse_obj(data, [..., filter]) == [True, 1, 1.1, 'str', {0: 0}, [1]], \
|
||||||
|
'`filter` should filter falsy values'
|
||||||
|
|
||||||
|
|
||||||
|
class TestTraversalHelpers:
|
||||||
|
def test_traversal_require(self):
|
||||||
|
with pytest.raises(ExtractorError):
|
||||||
|
traverse_obj(_TEST_DATA, ['None', {require('value')}])
|
||||||
|
assert traverse_obj(_TEST_DATA, ['str', {require('value')}]) == 'str', \
|
||||||
|
'`require` should pass through non `None` values'
|
||||||
|
|
||||||
|
def test_subs_list_to_dict(self):
|
||||||
|
assert traverse_obj([
|
||||||
|
{'name': 'de', 'url': 'https://example.com/subs/de.vtt'},
|
||||||
|
{'name': 'en', 'url': 'https://example.com/subs/en1.ass'},
|
||||||
|
{'name': 'en', 'url': 'https://example.com/subs/en2.ass'},
|
||||||
|
], [..., {
|
||||||
|
'id': 'name',
|
||||||
|
'url': 'url',
|
||||||
|
}, all, {subs_list_to_dict}]) == {
|
||||||
|
'de': [{'url': 'https://example.com/subs/de.vtt'}],
|
||||||
|
'en': [
|
||||||
|
{'url': 'https://example.com/subs/en1.ass'},
|
||||||
|
{'url': 'https://example.com/subs/en2.ass'},
|
||||||
|
],
|
||||||
|
}, 'function should build subtitle dict from list of subtitles'
|
||||||
|
assert traverse_obj([
|
||||||
|
{'name': 'de', 'url': 'https://example.com/subs/de.ass'},
|
||||||
|
{'name': 'de'},
|
||||||
|
{'name': 'en', 'content': 'content'},
|
||||||
|
{'url': 'https://example.com/subs/en'},
|
||||||
|
], [..., {
|
||||||
|
'id': 'name',
|
||||||
|
'data': 'content',
|
||||||
|
'url': 'url',
|
||||||
|
}, all, {subs_list_to_dict}]) == {
|
||||||
|
'de': [{'url': 'https://example.com/subs/de.ass'}],
|
||||||
|
'en': [{'data': 'content'}],
|
||||||
|
}, 'subs with mandatory items missing should be filtered'
|
||||||
|
assert traverse_obj([
|
||||||
|
{'url': 'https://example.com/subs/de.ass', 'name': 'de'},
|
||||||
|
{'url': 'https://example.com/subs/en', 'name': 'en'},
|
||||||
|
], [..., {
|
||||||
|
'id': 'name',
|
||||||
|
'ext': ['url', {lambda x: determine_ext(x, default_ext=None)}],
|
||||||
|
'url': 'url',
|
||||||
|
}, all, {subs_list_to_dict(ext='ext')}]) == {
|
||||||
|
'de': [{'url': 'https://example.com/subs/de.ass', 'ext': 'ass'}],
|
||||||
|
'en': [{'url': 'https://example.com/subs/en', 'ext': 'ext'}],
|
||||||
|
}, '`ext` should set default ext but leave existing value untouched'
|
||||||
|
assert traverse_obj([
|
||||||
|
{'name': 'en', 'url': 'https://example.com/subs/en2', 'prio': True},
|
||||||
|
{'name': 'en', 'url': 'https://example.com/subs/en1', 'prio': False},
|
||||||
|
], [..., {
|
||||||
|
'id': 'name',
|
||||||
|
'quality': ['prio', {int}],
|
||||||
|
'url': 'url',
|
||||||
|
}, all, {subs_list_to_dict(ext='ext')}]) == {'en': [
|
||||||
|
{'url': 'https://example.com/subs/en1', 'ext': 'ext'},
|
||||||
|
{'url': 'https://example.com/subs/en2', 'ext': 'ext'},
|
||||||
|
]}, '`quality` key should sort subtitle list accordingly'
|
||||||
|
|
||||||
|
|
||||||
class TestDictGet:
|
class TestDictGet:
|
||||||
def test_dict_get(self):
|
def test_dict_get(self):
|
||||||
|
|
|
@ -573,13 +573,13 @@ class InfoExtractor:
|
||||||
|
|
||||||
def _login_hint(self, method=NO_DEFAULT, netrc=None):
|
def _login_hint(self, method=NO_DEFAULT, netrc=None):
|
||||||
password_hint = f'--username and --password, --netrc-cmd, or --netrc ({netrc or self._NETRC_MACHINE}) to provide account credentials'
|
password_hint = f'--username and --password, --netrc-cmd, or --netrc ({netrc or self._NETRC_MACHINE}) to provide account credentials'
|
||||||
|
cookies_hint = 'See https://github.com/yt-dlp/yt-dlp/wiki/FAQ#how-do-i-pass-cookies-to-yt-dlp for how to manually pass cookies'
|
||||||
return {
|
return {
|
||||||
None: '',
|
None: '',
|
||||||
'any': f'Use --cookies, --cookies-from-browser, {password_hint}',
|
'any': f'Use --cookies, --cookies-from-browser, {password_hint}. {cookies_hint}',
|
||||||
'password': f'Use {password_hint}',
|
'password': f'Use {password_hint}',
|
||||||
'cookies': (
|
'cookies': f'Use --cookies-from-browser or --cookies for the authentication. {cookies_hint}',
|
||||||
'Use --cookies-from-browser or --cookies for the authentication. '
|
'session_cookies': f'Use --cookies for the authentication (--cookies-from-browser might not work). {cookies_hint}',
|
||||||
'See https://github.com/yt-dlp/yt-dlp/wiki/FAQ#how-do-i-pass-cookies-to-yt-dlp for how to manually pass cookies'),
|
|
||||||
}[method if method is not NO_DEFAULT else 'any' if self.supports_login() else 'cookies']
|
}[method if method is not NO_DEFAULT else 'any' if self.supports_login() else 'cookies']
|
||||||
|
|
||||||
def __init__(self, downloader=None):
|
def __init__(self, downloader=None):
|
||||||
|
|
|
@ -1984,11 +1984,30 @@ def urljoin(base, path):
|
||||||
return urllib.parse.urljoin(base, path)
|
return urllib.parse.urljoin(base, path)
|
||||||
|
|
||||||
|
|
||||||
def int_or_none(v, scale=1, default=None, get_attr=None, invscale=1):
|
def partial_application(func):
|
||||||
|
sig = inspect.signature(func)
|
||||||
|
|
||||||
|
@functools.wraps(func)
|
||||||
|
def wrapped(*args, **kwargs):
|
||||||
|
try:
|
||||||
|
sig.bind(*args, **kwargs)
|
||||||
|
except TypeError:
|
||||||
|
return functools.partial(func, *args, **kwargs)
|
||||||
|
else:
|
||||||
|
return func(*args, **kwargs)
|
||||||
|
|
||||||
|
return wrapped
|
||||||
|
|
||||||
|
|
||||||
|
@partial_application
|
||||||
|
def int_or_none(v, scale=1, default=None, get_attr=None, invscale=1, base=None):
|
||||||
if get_attr and v is not None:
|
if get_attr and v is not None:
|
||||||
v = getattr(v, get_attr, None)
|
v = getattr(v, get_attr, None)
|
||||||
|
if invscale == 1 and scale < 1:
|
||||||
|
invscale = int(1 / scale)
|
||||||
|
scale = 1
|
||||||
try:
|
try:
|
||||||
return int(v) * invscale // scale
|
return (int(v) if base is None else int(v, base=base)) * invscale // scale
|
||||||
except (ValueError, TypeError, OverflowError):
|
except (ValueError, TypeError, OverflowError):
|
||||||
return default
|
return default
|
||||||
|
|
||||||
|
@ -2006,9 +2025,13 @@ def str_to_int(int_str):
|
||||||
return int_or_none(int_str)
|
return int_or_none(int_str)
|
||||||
|
|
||||||
|
|
||||||
|
@partial_application
|
||||||
def float_or_none(v, scale=1, invscale=1, default=None):
|
def float_or_none(v, scale=1, invscale=1, default=None):
|
||||||
if v is None:
|
if v is None:
|
||||||
return default
|
return default
|
||||||
|
if invscale == 1 and scale < 1:
|
||||||
|
invscale = int(1 / scale)
|
||||||
|
scale = 1
|
||||||
try:
|
try:
|
||||||
return float(v) * invscale / scale
|
return float(v) * invscale / scale
|
||||||
except (ValueError, TypeError):
|
except (ValueError, TypeError):
|
||||||
|
|
|
@ -1,18 +1,35 @@
|
||||||
|
from __future__ import annotations
|
||||||
|
|
||||||
|
import collections
|
||||||
import collections.abc
|
import collections.abc
|
||||||
import contextlib
|
import contextlib
|
||||||
|
import functools
|
||||||
import http.cookies
|
import http.cookies
|
||||||
import inspect
|
import inspect
|
||||||
import itertools
|
import itertools
|
||||||
import re
|
import re
|
||||||
|
import typing
|
||||||
import xml.etree.ElementTree
|
import xml.etree.ElementTree
|
||||||
|
|
||||||
from ._utils import (
|
from ._utils import (
|
||||||
IDENTITY,
|
IDENTITY,
|
||||||
NO_DEFAULT,
|
NO_DEFAULT,
|
||||||
|
ExtractorError,
|
||||||
LazyList,
|
LazyList,
|
||||||
deprecation_warning,
|
deprecation_warning,
|
||||||
|
get_elements_html_by_class,
|
||||||
|
get_elements_html_by_attribute,
|
||||||
|
get_elements_by_attribute,
|
||||||
|
get_element_html_by_attribute,
|
||||||
|
get_element_by_attribute,
|
||||||
|
get_element_html_by_id,
|
||||||
|
get_element_by_id,
|
||||||
|
get_element_html_by_class,
|
||||||
|
get_elements_by_class,
|
||||||
|
get_element_text_and_html_by_tag,
|
||||||
is_iterable_like,
|
is_iterable_like,
|
||||||
try_call,
|
try_call,
|
||||||
|
url_or_none,
|
||||||
variadic,
|
variadic,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
@ -54,6 +71,7 @@ def traverse_obj(
|
||||||
Read as: `{key: traverse_obj(obj, path) for key, path in dct.items()}`.
|
Read as: `{key: traverse_obj(obj, path) for key, path in dct.items()}`.
|
||||||
- `any`-builtin: Take the first matching object and return it, resetting branching.
|
- `any`-builtin: Take the first matching object and return it, resetting branching.
|
||||||
- `all`-builtin: Take all matching objects and return them as a list, resetting branching.
|
- `all`-builtin: Take all matching objects and return them as a list, resetting branching.
|
||||||
|
- `filter`-builtin: Return the value if it is truthy, `None` otherwise.
|
||||||
|
|
||||||
`tuple`, `list`, and `dict` all support nested paths and branches.
|
`tuple`, `list`, and `dict` all support nested paths and branches.
|
||||||
|
|
||||||
|
@ -247,6 +265,10 @@ def apply_path(start_obj, path, test_type):
|
||||||
objs = (list(filtered_objs),)
|
objs = (list(filtered_objs),)
|
||||||
continue
|
continue
|
||||||
|
|
||||||
|
if key is filter:
|
||||||
|
objs = filter(None, objs)
|
||||||
|
continue
|
||||||
|
|
||||||
if __debug__ and callable(key):
|
if __debug__ and callable(key):
|
||||||
# Verify function signature
|
# Verify function signature
|
||||||
inspect.signature(key).bind(None, None)
|
inspect.signature(key).bind(None, None)
|
||||||
|
@ -277,13 +299,143 @@ def _traverse_obj(obj, path, allow_empty, test_type):
|
||||||
return results[0] if results else {} if allow_empty and is_dict else None
|
return results[0] if results else {} if allow_empty and is_dict else None
|
||||||
|
|
||||||
for index, path in enumerate(paths, 1):
|
for index, path in enumerate(paths, 1):
|
||||||
result = _traverse_obj(obj, path, index == len(paths), True)
|
is_last = index == len(paths)
|
||||||
|
try:
|
||||||
|
result = _traverse_obj(obj, path, is_last, True)
|
||||||
if result is not None:
|
if result is not None:
|
||||||
return result
|
return result
|
||||||
|
except _RequiredError as e:
|
||||||
|
if is_last:
|
||||||
|
# Reraise to get cleaner stack trace
|
||||||
|
raise ExtractorError(e.orig_msg, expected=e.expected) from None
|
||||||
|
|
||||||
return None if default is NO_DEFAULT else default
|
return None if default is NO_DEFAULT else default
|
||||||
|
|
||||||
|
|
||||||
|
def value(value, /):
|
||||||
|
return lambda _: value
|
||||||
|
|
||||||
|
|
||||||
|
def require(name, /, *, expected=False):
|
||||||
|
def func(value):
|
||||||
|
if value is None:
|
||||||
|
raise _RequiredError(f'Unable to extract {name}', expected=expected)
|
||||||
|
|
||||||
|
return value
|
||||||
|
|
||||||
|
return func
|
||||||
|
|
||||||
|
|
||||||
|
class _RequiredError(ExtractorError):
|
||||||
|
pass
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def subs_list_to_dict(*, ext: str | None = None) -> collections.abc.Callable[[list[dict]], dict[str, list[dict]]]: ...
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def subs_list_to_dict(subs: list[dict] | None, /, *, ext: str | None = None) -> dict[str, list[dict]]: ...
|
||||||
|
|
||||||
|
|
||||||
|
def subs_list_to_dict(subs: list[dict] | None = None, /, *, ext=None):
|
||||||
|
"""
|
||||||
|
Convert subtitles from a traversal into a subtitle dict.
|
||||||
|
The path should have an `all` immediately before this function.
|
||||||
|
|
||||||
|
Arguments:
|
||||||
|
`ext` The default value for `ext` in the subtitle dict
|
||||||
|
|
||||||
|
In the dict you can set the following additional items:
|
||||||
|
`id` The subtitle id to sort the dict into
|
||||||
|
`quality` The sort order for each subtitle
|
||||||
|
"""
|
||||||
|
if subs is None:
|
||||||
|
return functools.partial(subs_list_to_dict, ext=ext)
|
||||||
|
|
||||||
|
result = collections.defaultdict(list)
|
||||||
|
|
||||||
|
for sub in subs:
|
||||||
|
if not url_or_none(sub.get('url')) and not sub.get('data'):
|
||||||
|
continue
|
||||||
|
sub_id = sub.pop('id', None)
|
||||||
|
if sub_id is None:
|
||||||
|
continue
|
||||||
|
if ext is not None and not sub.get('ext'):
|
||||||
|
sub['ext'] = ext
|
||||||
|
result[sub_id].append(sub)
|
||||||
|
result = dict(result)
|
||||||
|
|
||||||
|
for subs in result.values():
|
||||||
|
subs.sort(key=lambda x: x.pop('quality', 0) or 0)
|
||||||
|
|
||||||
|
return result
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def find_element(*, attr: str, value: str, tag: str | None = None, html=False): ...
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def find_element(*, cls: str, html=False): ...
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def find_element(*, id: str, tag: str | None = None, html=False): ...
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def find_element(*, tag: str, html=False): ...
|
||||||
|
|
||||||
|
|
||||||
|
def find_element(*, tag=None, id=None, cls=None, attr=None, value=None, html=False):
|
||||||
|
# deliberately using `id=` and `cls=` for ease of readability
|
||||||
|
assert tag or id or cls or (attr and value), 'One of tag, id, cls or (attr AND value) is required'
|
||||||
|
if not tag:
|
||||||
|
tag = r'[\w:.-]+'
|
||||||
|
|
||||||
|
if attr and value:
|
||||||
|
assert not cls, 'Cannot match both attr and cls'
|
||||||
|
assert not id, 'Cannot match both attr and id'
|
||||||
|
func = get_element_html_by_attribute if html else get_element_by_attribute
|
||||||
|
return functools.partial(func, attr, value, tag=tag)
|
||||||
|
|
||||||
|
elif cls:
|
||||||
|
assert not id, 'Cannot match both cls and id'
|
||||||
|
assert tag is None, 'Cannot match both cls and tag'
|
||||||
|
func = get_element_html_by_class if html else get_elements_by_class
|
||||||
|
return functools.partial(func, cls)
|
||||||
|
|
||||||
|
elif id:
|
||||||
|
func = get_element_html_by_id if html else get_element_by_id
|
||||||
|
return functools.partial(func, id, tag=tag)
|
||||||
|
|
||||||
|
index = int(bool(html))
|
||||||
|
return lambda html: get_element_text_and_html_by_tag(tag, html)[index]
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def find_elements(*, cls: str, html=False): ...
|
||||||
|
|
||||||
|
|
||||||
|
@typing.overload
|
||||||
|
def find_elements(*, attr: str, value: str, tag: str | None = None, html=False): ...
|
||||||
|
|
||||||
|
|
||||||
|
def find_elements(*, tag=None, cls=None, attr=None, value=None, html=False):
|
||||||
|
# deliberately using `cls=` for ease of readability
|
||||||
|
assert cls or (attr and value), 'One of cls or (attr AND value) is required'
|
||||||
|
|
||||||
|
if attr and value:
|
||||||
|
assert not cls, 'Cannot match both attr and cls'
|
||||||
|
func = get_elements_html_by_attribute if html else get_elements_by_attribute
|
||||||
|
return functools.partial(func, attr, value, tag=tag or r'[\w:.-]+')
|
||||||
|
|
||||||
|
assert not tag, 'Cannot match both cls and tag'
|
||||||
|
func = get_elements_html_by_class if html else get_elements_by_class
|
||||||
|
return functools.partial(func, cls)
|
||||||
|
|
||||||
|
|
||||||
def get_first(obj, *paths, **kwargs):
|
def get_first(obj, *paths, **kwargs):
|
||||||
return traverse_obj(obj, *((..., *variadic(keys)) for keys in paths), **kwargs, get_all=False)
|
return traverse_obj(obj, *((..., *variadic(keys)) for keys in paths), **kwargs, get_all=False)
|
||||||
|
|
||||||
|
|
Loading…
Reference in a new issue