0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-18 01:35:12 +00:00
yt-dlp/yt_dlp/utils/traversal.py
2024-03-30 19:54:43 +01:00

285 lines
12 KiB
Python

import collections.abc
import contextlib
import inspect
import itertools
import re
import xml.etree.ElementTree
from ._utils import (
IDENTITY,
NO_DEFAULT,
LazyList,
deprecation_warning,
is_iterable_like,
try_call,
variadic,
)
def traverse_obj(
obj, *paths, default=NO_DEFAULT, expected_type=None, get_all=True,
casesense=True, is_user_input=NO_DEFAULT, traverse_string=False):
"""
Safely traverse nested `dict`s and `Iterable`s
>>> obj = [{}, {"key": "value"}]
>>> traverse_obj(obj, (1, "key"))
'value'
Each of the provided `paths` is tested and the first producing a valid result will be returned.
The next path will also be tested if the path branched but no results could be found.
Supported values for traversal are `Mapping`, `Iterable` and `re.Match`.
Unhelpful values (`{}`, `None`) are treated as the absence of a value and discarded.
The paths will be wrapped in `variadic`, so that `'key'` is conveniently the same as `('key', )`.
The keys in the path can be one of:
- `None`: Return the current object.
- `set`: Requires the only item in the set to be a type or function,
like `{type}`/`{func}`. If a `type`, returns only values
of this type. If a function, returns `func(obj)`.
- `str`/`int`: Return `obj[key]`. For `re.Match`, return `obj.group(key)`.
- `slice`: Branch out and return all values in `obj[key]`.
- `Ellipsis`: Branch out and return a list of all values.
- `tuple`/`list`: Branch out and return a list of all matching values.
Read as: `[traverse_obj(obj, branch) for branch in branches]`.
- `function`: Branch out and return values filtered by the function.
Read as: `[value for key, value in obj if function(key, value)]`.
For `Iterable`s, `key` is the index of the value.
For `re.Match`es, `key` is the group number (0 = full match)
as well as additionally any group names, if given.
- `dict` Transform the current object and return a matching dict.
Read as: `{key: traverse_obj(obj, path) for key, path in dct.items()}`.
`tuple`, `list`, and `dict` all support nested paths and branches.
@params paths Paths which to traverse by.
@param default Value to return if the paths do not match.
If the last key in the path is a `dict`, it will apply to each value inside
the dict instead, depth first. Try to avoid if using nested `dict` keys.
@param expected_type If a `type`, only accept final values of this type.
If any other callable, try to call the function on each result.
If the last key in the path is a `dict`, it will apply to each value inside
the dict instead, recursively. This does respect branching paths.
@param get_all If `False`, return the first matching result, otherwise all matching ones.
@param casesense If `False`, consider string dictionary keys as case insensitive.
`traverse_string` is only meant to be used by YoutubeDL.prepare_outtmpl and is not part of the API
@param traverse_string Whether to traverse into objects as strings.
If `True`, any non-compatible object will first be
converted into a string and then traversed into.
The return value of that path will be a string instead,
not respecting any further branching.
@returns The result of the object traversal.
If successful, `get_all=True`, and the path branches at least once,
then a list of results is returned instead.
If no `default` is given and the last path branches, a `list` of results
is always returned. If a path ends on a `dict` that result will always be a `dict`.
"""
if is_user_input is not NO_DEFAULT:
deprecation_warning('The is_user_input parameter is deprecated and no longer works')
casefold = lambda k: k.casefold() if isinstance(k, str) else k
if isinstance(expected_type, type):
type_test = lambda val: val if isinstance(val, expected_type) else None
else:
type_test = lambda val: try_call(expected_type or IDENTITY, args=(val,))
def apply_key(key, obj, is_last):
branching = False
result = None
if obj is None and traverse_string:
if key is ... or callable(key) or isinstance(key, slice):
branching = True
result = ()
elif key is None:
result = obj
elif isinstance(key, set):
assert len(key) == 1, 'Set should only be used to wrap a single item'
item = next(iter(key))
if isinstance(item, type):
if isinstance(obj, item):
result = obj
else:
result = try_call(item, args=(obj,))
elif isinstance(key, (list, tuple)):
branching = True
result = itertools.chain.from_iterable(
apply_path(obj, branch, is_last)[0] for branch in key)
elif key is ...:
branching = True
if isinstance(obj, collections.abc.Mapping):
result = obj.values()
elif is_iterable_like(obj) or isinstance(obj, xml.etree.ElementTree.Element):
result = obj
elif isinstance(obj, re.Match):
result = obj.groups()
elif traverse_string:
branching = False
result = str(obj)
else:
result = ()
elif callable(key):
branching = True
if isinstance(obj, collections.abc.Mapping):
iter_obj = obj.items()
elif is_iterable_like(obj) or isinstance(obj, xml.etree.ElementTree.Element):
iter_obj = enumerate(obj)
elif isinstance(obj, re.Match):
iter_obj = itertools.chain(
enumerate((obj.group(), *obj.groups())),
obj.groupdict().items())
elif traverse_string:
branching = False
iter_obj = enumerate(str(obj))
else:
iter_obj = ()
result = (v for k, v in iter_obj if try_call(key, args=(k, v)))
if not branching: # string traversal
result = ''.join(result)
elif isinstance(key, dict):
iter_obj = ((k, _traverse_obj(obj, v, False, is_last)) for k, v in key.items())
result = {
k: v if v is not None else default for k, v in iter_obj
if v is not None or default is not NO_DEFAULT
} or None
elif isinstance(obj, collections.abc.Mapping):
result = (try_call(obj.get, args=(key,)) if casesense or try_call(obj.__contains__, args=(key,)) else
next((v for k, v in obj.items() if casefold(k) == key), None))
elif isinstance(obj, re.Match):
if isinstance(key, int) or casesense:
with contextlib.suppress(IndexError):
result = obj.group(key)
elif isinstance(key, str):
result = next((v for k, v in obj.groupdict().items() if casefold(k) == key), None)
elif isinstance(key, (int, slice)):
if is_iterable_like(obj, (collections.abc.Sequence, xml.etree.ElementTree.Element)):
branching = isinstance(key, slice)
with contextlib.suppress(IndexError):
result = obj[key]
elif traverse_string:
with contextlib.suppress(IndexError):
result = str(obj)[key]
elif isinstance(obj, xml.etree.ElementTree.Element) and isinstance(key, str):
xpath, _, special = key.rpartition('/')
if not special.startswith('@') and special != 'text()':
xpath = key
special = None
# Allow abbreviations of relative paths, absolute paths error
if xpath.startswith('/'):
xpath = f'.{xpath}'
elif xpath and not xpath.startswith('./'):
xpath = f'./{xpath}'
def apply_specials(element):
if special is None:
return element
if special == '@':
return element.attrib
if special.startswith('@'):
return try_call(element.attrib.get, args=(special[1:],))
if special == 'text()':
return element.text
assert False, f'apply_specials is missing case for {special!r}'
if xpath:
result = list(map(apply_specials, obj.iterfind(xpath)))
else:
result = apply_specials(obj)
return branching, result if branching else (result,)
def lazy_last(iterable):
iterator = iter(iterable)
prev = next(iterator, NO_DEFAULT)
if prev is NO_DEFAULT:
return
for item in iterator:
yield False, prev
prev = item
yield True, prev
def apply_path(start_obj, path, test_type):
objs = (start_obj,)
has_branched = False
key = None
for last, key in lazy_last(variadic(path, (str, bytes, dict, set))):
if not casesense and isinstance(key, str):
key = key.casefold()
if key in (any, all):
has_branched = False
filtered_objs = (obj for obj in objs if obj not in (None, {}))
if key is any:
objs = (next(filtered_objs, None),)
else:
objs = (list(filtered_objs),)
continue
if __debug__ and callable(key):
# Verify function signature
inspect.signature(key).bind(None, None)
new_objs = []
for obj in objs:
branching, results = apply_key(key, obj, last)
has_branched |= branching
new_objs.append(results)
objs = itertools.chain.from_iterable(new_objs)
if test_type and not isinstance(key, (dict, list, tuple)):
objs = map(type_test, objs)
return objs, has_branched, isinstance(key, dict)
def _traverse_obj(obj, path, allow_empty, test_type):
results, has_branched, is_dict = apply_path(obj, path, test_type)
results = LazyList(item for item in results if item not in (None, {}))
if get_all and has_branched:
if results:
return results.exhaust()
if allow_empty:
return [] if default is NO_DEFAULT else default
return None
return results[0] if results else {} if allow_empty and is_dict else None
for index, path in enumerate(paths, 1):
result = _traverse_obj(obj, path, index == len(paths), True)
if result is not None:
return result
return None if default is NO_DEFAULT else default
def get_first(obj, *paths, **kwargs):
return traverse_obj(obj, *((..., *variadic(keys)) for keys in paths), **kwargs, get_all=False)
def dict_get(d, key_or_keys, default=None, skip_false_values=True):
for val in map(d.get, variadic(key_or_keys)):
if val is not None and (val or not skip_false_values):
return val
return default