[utils] Add {expected_type} and Iterable support to traverse_obj()

2 files changed, 266 insertions, 100 deletions

diff --git a/test/test_utils.py b/test/test_utils.py
index 5fab05f7c..1fc16ed05 100644
--- a/test/test_utils.py
+++ b/test/test_utils.py
@@ -79,10 +79,12 @@ from youtube_dl.utils import (
     rot47,
     shell_quote,
     smuggle_url,
+    str_or_none,
     str_to_int,
     strip_jsonp,
     strip_or_none,
     subtitles_filename,
+    T,
     timeconvert,
     traverse_obj,
     try_call,
@@ -1566,6 +1568,7 @@ Line 1
         self.assertEqual(variadic('spam', allowed_types=[dict]), 'spam')
 
     def test_traverse_obj(self):
+        str = compat_str
         _TEST_DATA = {
             100: 100,
             1.2: 1.2,
@@ -1598,8 +1601,8 @@ Line 1
 
         # Test Ellipsis behavior
         self.assertCountEqual(traverse_obj(_TEST_DATA, Ellipsis),
-                              (item for item in _TEST_DATA.values() if item is not None),
-                              msg='`...` should give all values except `None`')
+                              (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, Ellipsis)), _TEST_DATA['urls'][0].values(),
                               msg='`...` selection for dicts should select all values')
         self.assertEqual(traverse_obj(_TEST_DATA, (Ellipsis, Ellipsis, 'url')),
@@ -1607,13 +1610,51 @@ Line 1
                          msg='nested `...` queries should work')
         self.assertCountEqual(traverse_obj(_TEST_DATA, (Ellipsis, Ellipsis, 'index')), range(4),
                               msg='`...` query result should be flattened')
+        self.assertEqual(traverse_obj(iter(range(4)), Ellipsis), 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], compat_str)), ('str',),
-                              msg='exceptions in the query function should be caught')
+        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: Ellipsis)
+            with self.assertRaises(Exception, msg='Wrong function signature should raise in debug'):
+                traverse_obj(_TEST_DATA, lambda a, b, c: Ellipsis)
+
+        # Test set as key (transformation/type, like `expected_type`)
+        self.assertEqual(traverse_obj(_TEST_DATA, (Ellipsis, T(str.upper), )), ['STR'],
+                         msg='Function in set should be a transformation')
+        self.assertEqual(traverse_obj(_TEST_DATA, (Ellipsis, T(str))), ['str'],
+                         msg='Type in set should be a type filter')
+        self.assertEqual(traverse_obj(_TEST_DATA, T(dict)), _TEST_DATA,
+                         msg='A single set should be wrapped into a path')
+        self.assertEqual(traverse_obj(_TEST_DATA, (Ellipsis, T(str.upper))), ['STR'],
+                         msg='Transformation function should not raise')
+        self.assertEqual(traverse_obj(_TEST_DATA, (Ellipsis, T(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')
+        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',
@@ -1659,15 +1700,23 @@ Line 1
                          {0: ['https://www.example.com/1', 'https://www.example.com/0']},
                          msg='triple nesting in dict path should be treated as branches')
         self.assertEqual(traverse_obj(_TEST_DATA, {0: 'fail'}), {},
-                         msg='remove `None` values when dict key')
+                         msg='remove `None` values when top level dict key fails')
         self.assertEqual(traverse_obj(_TEST_DATA, {0: 'fail'}, default=Ellipsis), {0: Ellipsis},
-                         msg='do not remove `None` values if `default`')
-        self.assertEqual(traverse_obj(_TEST_DATA, {0: 'dict'}), {0: {}},
-                         msg='do not remove empty values when dict key')
-        self.assertEqual(traverse_obj(_TEST_DATA, {0: 'dict'}, default=Ellipsis), {0: {}},
-                         msg='do not remove empty values when dict key and a default')
-        self.assertEqual(traverse_obj(_TEST_DATA, {0: ('dict', Ellipsis)}), {0: []},
-                         msg='if branch in dict key not successful, return `[]`')
+                         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=Ellipsis), {0: Ellipsis},
+                         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=Ellipsis), {0: Ellipsis},
+                         msg='default to dict if pruned and default is given')
+        self.assertEqual(traverse_obj(_TEST_DATA, {0: {0: 'fail'}}, default=Ellipsis), {0: {0: Ellipsis}},
+                         msg='use nested `default` when nested dict key fails and `default`')
+        self.assertEqual(traverse_obj(_TEST_DATA, {0: ('dict', Ellipsis)}), {},
+                         msg='remove key if branch in dict key not successful')
 
         # Testing default parameter behavior
         _DEFAULT_DATA = {'None': None, 'int': 0, 'list': []}
@@ -1691,20 +1740,55 @@ Line 1
                          msg='if branched but not successful return `[]`, not `default`')
         self.assertEqual(traverse_obj(_DEFAULT_DATA, ('list', Ellipsis)), [],
                          msg='if branched but object is empty return `[]`, not `default`')
+        self.assertEqual(traverse_obj(None, Ellipsis), [],
+                         msg='if branched but object is `None` return `[]`, not `default`')
+        self.assertEqual(traverse_obj({0: None}, (0, Ellipsis)), [],
+                         msg='if branched but state is `None` return `[]`, not `default`')
+
+        branching_paths = [
+            ('fail', Ellipsis),
+            (Ellipsis, 'fail'),
+            100 * ('fail',) + (Ellipsis,),
+            (Ellipsis,) + 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=compat_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: compat_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 function in try_call')
-        self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, Ellipsis, expected_type=compat_str), ['str'],
-                         msg='eliminate items that expected_type fails on')
+        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 function in try_call')
+        self.assertEqual(traverse_obj(_EXPECTED_TYPE_DATA, Ellipsis, 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(Ellipsis)),
+                         [{0: Ellipsis}, {0: Ellipsis}], 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', Ellipsis], expected_type=int),
+                         [], msg='expected_type regression for type matching in dict result')
 
         # Test get_all behavior
         _GET_ALL_DATA = {'key': [0, 1, 2]}
@@ -1749,14 +1833,23 @@ Line 1
                                       _traverse_string=True), '.',
                          msg='traverse into converted data if `traverse_string`')
         self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', Ellipsis),
-                                      _traverse_string=True), list('str'),
-                         msg='`...` branching into string should result in list')
+                                      _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 into string should result in list')
-        self.assertEqual(traverse_obj(_TRAVERSE_STRING_DATA, ('str', lambda _, x: x),
-                                      _traverse_string=True), list('str'),
-                         msg='function branching into string should result in list')
+                         msg='branching should result in list if `traverse_string`')
+        self.assertEqual(traverse_obj({}, (0, Ellipsis), _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 is_user_input behavior
         _IS_USER_INPUT_DATA = {'range8': list(range(8))}
@@ -1793,6 +1886,8 @@ Line 1
                          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_get_first(self):
         self.assertEqual(get_first([{'a': None}, {'a': 'spam'}], 'a'), 'spam')
diff --git a/youtube_dl/utils.py b/youtube_dl/utils.py
index 83f67bd95..dbdbe5f59 100644
--- a/youtube_dl/utils.py
+++ b/youtube_dl/utils.py
@@ -16,6 +16,7 @@ import email.header
 import errno
 import functools
 import gzip
+import inspect
 import io
 import itertools
 import json
@@ -3881,7 +3882,7 @@ def detect_exe_version(output, version_re=None, unrecognized='present'):
         return unrecognized
 
 
-class LazyList(compat_collections_abc.Sequence):
+class LazyList(compat_collections_abc.Iterable):
     """Lazy immutable list from an iterable
     Note that slices of a LazyList are lists and not LazyList"""
 
@@ -4223,10 +4224,16 @@ def multipart_encode(data, boundary=None):
     return out, content_type
 
 
-def variadic(x, allowed_types=(compat_str, bytes, dict)):
-    if not isinstance(allowed_types, tuple) and isinstance(allowed_types, compat_collections_abc.Iterable):
+def is_iterable_like(x, allowed_types=compat_collections_abc.Iterable, blocked_types=NO_DEFAULT):
+    if blocked_types is NO_DEFAULT:
+        blocked_types = (compat_str, bytes, compat_collections_abc.Mapping)
+    return isinstance(x, allowed_types) and not isinstance(x, blocked_types)
+
+
+def variadic(x, allowed_types=NO_DEFAULT):
+    if isinstance(allowed_types, compat_collections_abc.Iterable):
         allowed_types = tuple(allowed_types)
-    return x if isinstance(x, compat_collections_abc.Iterable) and not isinstance(x, allowed_types) else (x,)
+    return x if is_iterable_like(x, blocked_types=allowed_types) else (x,)
 
 
 def dict_get(d, key_or_keys, default=None, skip_false_values=True):
@@ -5993,7 +6000,7 @@ def clean_podcast_url(url):
 
 def traverse_obj(obj, *paths, **kwargs):
     """
-    Safely traverse nested `dict`s and `Sequence`s
+    Safely traverse nested `dict`s and `Iterable`s
 
     >>> obj = [{}, {"key": "value"}]
     >>> traverse_obj(obj, (1, "key"))
@@ -6001,14 +6008,17 @@ def traverse_obj(obj, *paths, **kwargs):
 
     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`, `Sequence` and `re.Match`.
-    A value of None is treated as the absence of a value.
+    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.
-        - `str`/`int`:      Return `obj[key]`. For `re.Match, return `obj.group(key)`.
+        - `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.
@@ -6016,6 +6026,9 @@ def traverse_obj(obj, *paths, **kwargs):
         - `function`:       Branch out and return values filtered by the function.
                             Read as: `[value for key, value in obj if function(key, value)]`.
                             For `Sequence`s, `key` is the index of the value.
+                            For `Iterable`s, `key` is the enumeration count 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()}`.
 
@@ -6024,8 +6037,12 @@ def traverse_obj(obj, *paths, **kwargs):
     @params paths           Paths which to traverse by.
     Keyword arguments:
     @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.
 
@@ -6036,12 +6053,15 @@ def traverse_obj(obj, *paths, **kwargs):
     @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.
                             A list is always returned if the last path branches and no `default` is given.
+                            If a path ends on a `dict` that result will always be a `dict`.
     """
 
     # parameter defaults
@@ -6055,7 +6075,6 @@ def traverse_obj(obj, *paths, **kwargs):
     # instant compat
     str = compat_str
 
-    is_sequence = lambda x: isinstance(x, compat_collections_abc.Sequence) and not isinstance(x, (str, bytes))
     casefold = lambda k: compat_casefold(k) if isinstance(k, str) else k
 
     if isinstance(expected_type, type):
@@ -6063,128 +6082,180 @@ def traverse_obj(obj, *paths, **kwargs):
     else:
         type_test = lambda val: try_call(expected_type or IDENTITY, args=(val,))
 
+    def lookup_or_none(v, k, getter=None):
+        try:
+            return getter(v, k) if getter else v[k]
+        except IndexError:
+            return None
+
     def from_iterable(iterables):
         # chain.from_iterable(['ABC', 'DEF']) --> A B C D E F
         for it in iterables:
             for item in it:
                 yield item
 
-    def apply_key(key, obj):
-        if obj is None:
-            return
+    def apply_key(key, obj, is_last):
+        branching = False
+
+        if obj is None and _traverse_string:
+            if key is Ellipsis or callable(key) or isinstance(key, slice):
+                branching = True
+                result = ()
+            else:
+                result = None
 
         elif key is None:
-            yield obj
+            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):
+                result = obj if isinstance(obj, item) else None
+            else:
+                result = try_call(item, args=(obj,))
 
         elif isinstance(key, (list, tuple)):
-            for branch in key:
-                _, result = apply_path(obj, branch)
-                for item in result:
-                    yield item
+            branching = True
+            result = from_iterable(
+                apply_path(obj, branch, is_last)[0] for branch in key)
 
         elif key is Ellipsis:
-            result = []
+            branching = True
             if isinstance(obj, compat_collections_abc.Mapping):
                 result = obj.values()
-            elif is_sequence(obj):
+            elif is_iterable_like(obj):
                 result = obj
             elif isinstance(obj, compat_re_Match):
                 result = obj.groups()
             elif _traverse_string:
+                branching = False
                 result = str(obj)
-            for item in result:
-                yield item
+            else:
+                result = ()
 
         elif callable(key):
-            if is_sequence(obj):
-                iter_obj = enumerate(obj)
-            elif isinstance(obj, compat_collections_abc.Mapping):
+            branching = True
+            if isinstance(obj, compat_collections_abc.Mapping):
                 iter_obj = obj.items()
+            elif is_iterable_like(obj):
+                iter_obj = enumerate(obj)
             elif isinstance(obj, compat_re_Match):
-                iter_obj = enumerate(itertools.chain([obj.group()], obj.groups()))
+                iter_obj = itertools.chain(
+                    enumerate(itertools.chain((obj.group(),), obj.groups())),
+                    obj.groupdict().items())
             elif _traverse_string:
+                branching = False
                 iter_obj = enumerate(str(obj))
             else:
-                return
-            for item in (v for k, v in iter_obj if try_call(key, args=(k, v))):
-                yield item
+                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)) for k, v in key.items())
-            yield dict((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)
+            iter_obj = ((k, _traverse_obj(obj, v, False, is_last)) for k, v in key.items())
+            result = dict((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, compat_collections_abc.Mapping):
-            yield (obj.get(key) if casesense or (key in obj)
-                   else next((v for k, v in obj.items() if casefold(k) == key), None))
+            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, compat_re_Match):
+            result = None
             if isinstance(key, int) or casesense:
-                try:
-                    yield obj.group(key)
-                    return
-                except IndexError:
-                    pass
-            if not isinstance(key, str):
-                return
+                result = lookup_or_none(obj, key, getter=compat_re_Match.group)
 
-            yield next((v for k, v in obj.groupdict().items() if casefold(k) == key), None)
+            elif isinstance(key, str):
+                result = next((v for k, v in obj.groupdict().items()
+                              if casefold(k) == key), None)
 
         else:
-            if _is_user_input:
-                key = (int_or_none(key) if ':' not in key
-                       else slice(*map(int_or_none, key.split(':'))))
+            result = None
+            if isinstance(key, (int, slice)):
+                if is_iterable_like(obj, compat_collections_abc.Sequence):
+                    branching = isinstance(key, slice)
+                    result = lookup_or_none(obj, key)
+                elif _traverse_string:
+                    result = lookup_or_none(str(obj), key)
+
+        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
 
-            if not isinstance(key, (int, slice)):
-                return
+        for item in iterator:
+            yield False, prev
+            prev = item
 
-            if not is_sequence(obj):
-                if not _traverse_string:
-                    return
-                obj = str(obj)
+        yield True, prev
 
-            try:
-                yield obj[key]
-            except IndexError:
-                pass
-
-    def apply_path(start_obj, path):
+    def apply_path(start_obj, path, test_type):
         objs = (start_obj,)
         has_branched = False
 
-        for key in variadic(path):
-            if _is_user_input and key == ':':
-                key = Ellipsis
+        key = None
+        for last, key in lazy_last(variadic(path, (str, bytes, dict, set))):
+            if _is_user_input and isinstance(key, str):
+                if key == ':':
+                    key = Ellipsis
+                elif ':' in key:
+                    key = slice(*map(int_or_none, key.split(':')))
+                elif int_or_none(key) is not None:
+                    key = int(key)
 
             if not casesense and isinstance(key, str):
                 key = compat_casefold(key)
 
-            if key is Ellipsis or isinstance(key, (list, tuple)) or callable(key):
-                has_branched = True
+            if __debug__ and callable(key):
+                # Verify function signature
+                inspect.getcallargs(key, None, None)
+
+            new_objs = []
+            for obj in objs:
+                branching, results = apply_key(key, obj, last)
+                has_branched |= branching
+                new_objs.append(results)
+
+            objs = from_iterable(new_objs)
 
-            key_func = functools.partial(apply_key, key)
-            objs = from_iterable(map(key_func, objs))
+        if test_type and not isinstance(key, (dict, list, tuple)):
+            objs = map(type_test, objs)
 
-        return has_branched, objs
+        return objs, has_branched, isinstance(key, dict)
 
-    def _traverse_obj(obj, path, use_list=True):
-        has_branched, results = apply_path(obj, path)
-        results = LazyList(x for x in map(type_test, results) if x is not None)
+    def _traverse_obj(obj, path, allow_empty, test_type):
+        results, has_branched, is_dict = apply_path(obj, path, test_type)
+        results = LazyList(x for x in results if x not in (None, {}))
 
         if get_all and has_branched:
-            return results.exhaust() if results or use_list else None
+            if results:
+                return results.exhaust()
+            if allow_empty:
+                return [] if default is NO_DEFAULT else default
+            return None
 
-        return results[0] if results else None
+        return results[0] if results else {} if allow_empty and is_dict else None
 
     for index, path in enumerate(paths, 1):
-        use_list = default is NO_DEFAULT and index == len(paths)
-        result = _traverse_obj(obj, path, use_list)
+        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 T(x):
+    """ For use in yt-dl instead of {type} or set((type,)) """
+    return set((x,))
+
+
 def get_first(obj, keys, **kwargs):
     return traverse_obj(obj, (Ellipsis,) + tuple(variadic(keys)), get_all=False, **kwargs)