Merge pull request #13284 from Carreau/dict-higher

Move dict keys completions higher.

IPython/core/completer.py

@@ -1,2257 +1,2257 b''
 """Completion for IPython.
 
 This module started as fork of the rlcompleter module in the Python standard
 library.  The original enhancements made to rlcompleter have been sent
 upstream and were accepted as of Python 2.3,
 
 This module now support a wide variety of completion mechanism both available
 for normal classic Python code, as well as completer for IPython specific
 Syntax like magics.
 
 Latex and Unicode completion
 ============================
 
 IPython and compatible frontends not only can complete your code, but can help
 you to input a wide range of characters. In particular we allow you to insert
 a unicode character using the tab completion mechanism.
 
 Forward latex/unicode completion
 --------------------------------
 
 Forward completion allows you to easily type a unicode character using its latex
 name, or unicode long description. To do so type a backslash follow by the
 relevant name and press tab:
 
 
 Using latex completion:
 
 .. code::
 
     \\alpha<tab>
     α
 
 or using unicode completion:
 
 
 .. code::
 
     \\GREEK SMALL LETTER ALPHA<tab>
     α
 
 
 Only valid Python identifiers will complete. Combining characters (like arrow or
 dots) are also available, unlike latex they need to be put after the their
 counterpart that is to say, `F\\\\vec<tab>` is correct, not `\\\\vec<tab>F`.
 
 Some browsers are known to display combining characters incorrectly.
 
 Backward latex completion
 -------------------------
 
 It is sometime challenging to know how to type a character, if you are using
 IPython, or any compatible frontend you can prepend backslash to the character
 and press `<tab>` to expand it to its latex form.
 
 .. code::
 
     \\α<tab>
     \\alpha
 
 
 Both forward and backward completions can be deactivated by setting the
 ``Completer.backslash_combining_completions`` option to ``False``.
 
 
 Experimental
 ============
 
 Starting with IPython 6.0, this module can make use of the Jedi library to
 generate completions both using static analysis of the code, and dynamically
 inspecting multiple namespaces. Jedi is an autocompletion and static analysis
 for Python. The APIs attached to this new mechanism is unstable and will
 raise unless use in an :any:`provisionalcompleter` context manager.
 
 You will find that the following are experimental:
 
     - :any:`provisionalcompleter`
     - :any:`IPCompleter.completions`
     - :any:`Completion`
     - :any:`rectify_completions`
 
 .. note::
 
     better name for :any:`rectify_completions` ?
 
 We welcome any feedback on these new API, and we also encourage you to try this
 module in debug mode (start IPython with ``--Completer.debug=True``) in order
 to have extra logging information if :any:`jedi` is crashing, or if current
 IPython completer pending deprecations are returning results not yet handled
 by :any:`jedi`
 
 Using Jedi for tab completion allow snippets like the following to work without
 having to execute any code:
 
    >>> myvar = ['hello', 42]
    ... myvar[1].bi<tab>
 
 Tab completion will be able to infer that ``myvar[1]`` is a real number without
 executing any code unlike the previously available ``IPCompleter.greedy``
 option.
 
 Be sure to update :any:`jedi` to the latest stable version or to try the
 current development version to get better completions.
 """
 
 
 # Copyright (c) IPython Development Team.
 # Distributed under the terms of the Modified BSD License.
 #
 # Some of this code originated from rlcompleter in the Python standard library
 # Copyright (C) 2001 Python Software Foundation, www.python.org
 
 
 import builtins as builtin_mod
 import glob
 import inspect
 import itertools
 import keyword
 import os
 import re
 import string
 import sys
 import time
 import unicodedata
 import uuid
 import warnings
 from contextlib import contextmanager
 from importlib import import_module
 from types import SimpleNamespace
 from typing import Iterable, Iterator, List, Tuple, Union, Any, Sequence, Dict, NamedTuple, Pattern, Optional
 
 from IPython.core.error import TryNext
 from IPython.core.inputtransformer2 import ESC_MAGIC
 from IPython.core.latex_symbols import latex_symbols, reverse_latex_symbol
 from IPython.core.oinspect import InspectColors
 from IPython.utils import generics
 from IPython.utils.dir2 import dir2, get_real_method
 from IPython.utils.path import ensure_dir_exists
 from IPython.utils.process import arg_split
 from traitlets import Bool, Enum, Int, List as ListTrait, Unicode, default, observe
 from traitlets.config.configurable import Configurable
 
 import __main__
 
 # skip module docstests
 __skip_doctest__ = True
 
 try:
     import jedi
     jedi.settings.case_insensitive_completion = False
     import jedi.api.helpers
     import jedi.api.classes
     JEDI_INSTALLED = True
 except ImportError:
     JEDI_INSTALLED = False
 #-----------------------------------------------------------------------------
 # Globals
 #-----------------------------------------------------------------------------
 
 # ranges where we have most of the valid unicode names. We could be more finer
 # grained but is it worth it for performance  While unicode have character in the
 # range 0, 0x110000, we seem to have name for about 10% of those. (131808 as I
 # write this). With below range we cover them all, with a density of ~67%
 # biggest next gap we consider only adds up about 1% density and there are 600
 # gaps that would need hard coding.
 _UNICODE_RANGES = [(32, 0x3134b), (0xe0001, 0xe01f0)]
 
 # Public API
 __all__ = ['Completer','IPCompleter']
 
 if sys.platform == 'win32':
     PROTECTABLES = ' '
 else:
     PROTECTABLES = ' ()[]{}?=\\|;:\'#*"^&'
 
 # Protect against returning an enormous number of completions which the frontend
 # may have trouble processing.
 MATCHES_LIMIT = 500
 
 _deprecation_readline_sentinel = object()
 
 
 class ProvisionalCompleterWarning(FutureWarning):
     """
     Exception raise by an experimental feature in this module.
 
     Wrap code in :any:`provisionalcompleter` context manager if you
     are certain you want to use an unstable feature.
     """
     pass
 
 warnings.filterwarnings('error', category=ProvisionalCompleterWarning)
 
 @contextmanager
 def provisionalcompleter(action='ignore'):
     """
     This context manager has to be used in any place where unstable completer
     behavior and API may be called.
 
     >>> with provisionalcompleter():
     ...     completer.do_experimental_things() # works
 
     >>> completer.do_experimental_things() # raises.
 
     .. note::
 
         Unstable
 
         By using this context manager you agree that the API in use may change
         without warning, and that you won't complain if they do so.
 
         You also understand that, if the API is not to your liking, you should report
         a bug to explain your use case upstream.
 
         We'll be happy to get your feedback, feature requests, and improvements on
         any of the unstable APIs!
     """
     with warnings.catch_warnings():
         warnings.filterwarnings(action, category=ProvisionalCompleterWarning)
         yield
 
 
 def has_open_quotes(s):
     """Return whether a string has open quotes.
 
     This simply counts whether the number of quote characters of either type in
     the string is odd.
 
     Returns
     -------
     If there is an open quote, the quote character is returned.  Else, return
     False.
     """
     # We check " first, then ', so complex cases with nested quotes will get
     # the " to take precedence.
     if s.count('"') % 2:
         return '"'
     elif s.count("'") % 2:
         return "'"
     else:
         return False
 
 
 def protect_filename(s, protectables=PROTECTABLES):
     """Escape a string to protect certain characters."""
     if set(s) & set(protectables):
         if sys.platform == "win32":
             return '"' + s + '"'
         else:
             return "".join(("\\" + c if c in protectables else c) for c in s)
     else:
         return s
 
 
 def expand_user(path:str) -> Tuple[str, bool, str]:
     """Expand ``~``-style usernames in strings.
 
     This is similar to :func:`os.path.expanduser`, but it computes and returns
     extra information that will be useful if the input was being used in
     computing completions, and you wish to return the completions with the
     original '~' instead of its expanded value.
 
     Parameters
     ----------
     path : str
         String to be expanded.  If no ~ is present, the output is the same as the
         input.
 
     Returns
     -------
     newpath : str
         Result of ~ expansion in the input path.
     tilde_expand : bool
         Whether any expansion was performed or not.
     tilde_val : str
         The value that ~ was replaced with.
     """
     # Default values
     tilde_expand = False
     tilde_val = ''
     newpath = path
 
     if path.startswith('~'):
         tilde_expand = True
         rest = len(path)-1
         newpath = os.path.expanduser(path)
         if rest:
             tilde_val = newpath[:-rest]
         else:
             tilde_val = newpath
 
     return newpath, tilde_expand, tilde_val
 
 
 def compress_user(path:str, tilde_expand:bool, tilde_val:str) -> str:
     """Does the opposite of expand_user, with its outputs.
     """
     if tilde_expand:
         return path.replace(tilde_val, '~')
     else:
         return path
 
 
 def completions_sorting_key(word):
     """key for sorting completions
 
     This does several things:
 
     - Demote any completions starting with underscores to the end
     - Insert any %magic and %%cellmagic completions in the alphabetical order
       by their name
     """
     prio1, prio2 = 0, 0
 
     if word.startswith('__'):
         prio1 = 2
     elif word.startswith('_'):
         prio1 = 1
 
     if word.endswith('='):
         prio1 = -1
 
     if word.startswith('%%'):
         # If there's another % in there, this is something else, so leave it alone
         if not "%" in word[2:]:
             word = word[2:]
             prio2 = 2
     elif word.startswith('%'):
         if not "%" in word[1:]:
             word = word[1:]
             prio2 = 1
 
     return prio1, word, prio2
 
 
 class _FakeJediCompletion:
     """
     This is a workaround to communicate to the UI that Jedi has crashed and to
     report a bug. Will be used only id :any:`IPCompleter.debug` is set to true.
 
     Added in IPython 6.0 so should likely be removed for 7.0
 
     """
 
     def __init__(self, name):
 
         self.name = name
         self.complete = name
         self.type = 'crashed'
         self.name_with_symbols = name
         self.signature = ''
         self._origin = 'fake'
 
     def __repr__(self):
         return '<Fake completion object jedi has crashed>'
 
 
 class Completion:
     """
     Completion object used and return by IPython completers.
 
     .. warning::
 
         Unstable
 
         This function is unstable, API may change without warning.
         It will also raise unless use in proper context manager.
 
     This act as a middle ground :any:`Completion` object between the
     :any:`jedi.api.classes.Completion` object and the Prompt Toolkit completion
     object. While Jedi need a lot of information about evaluator and how the
     code should be ran/inspected, PromptToolkit (and other frontend) mostly
     need user facing information.
 
     - Which range should be replaced replaced by what.
     - Some metadata (like completion type), or meta information to displayed to
       the use user.
 
     For debugging purpose we can also store the origin of the completion (``jedi``,
     ``IPython.python_matches``, ``IPython.magics_matches``...).
     """
 
     __slots__ = ['start', 'end', 'text', 'type', 'signature', '_origin']
 
     def __init__(self, start: int, end: int, text: str, *, type: str=None, _origin='', signature='') -> None:
         warnings.warn("``Completion`` is a provisional API (as of IPython 6.0). "
                       "It may change without warnings. "
                       "Use in corresponding context manager.",
                       category=ProvisionalCompleterWarning, stacklevel=2)
 
         self.start = start
         self.end = end
         self.text = text
         self.type = type
         self.signature = signature
         self._origin = _origin
 
     def __repr__(self):
         return '<Completion start=%s end=%s text=%r type=%r, signature=%r,>' % \
                 (self.start, self.end, self.text, self.type or '?', self.signature or '?')
 
     def __eq__(self, other)->Bool:
         """
         Equality and hash do not hash the type (as some completer may not be
         able to infer the type), but are use to (partially) de-duplicate
         completion.
 
         Completely de-duplicating completion is a bit tricker that just
         comparing as it depends on surrounding text, which Completions are not
         aware of.
         """
         return self.start == other.start and \
             self.end == other.end and \
             self.text == other.text
 
     def __hash__(self):
         return hash((self.start, self.end, self.text))
 
 
 _IC = Iterable[Completion]
 
 
 def _deduplicate_completions(text: str, completions: _IC)-> _IC:
     """
     Deduplicate a set of completions.
 
     .. warning::
 
         Unstable
 
         This function is unstable, API may change without warning.
 
     Parameters
     ----------
     text : str
         text that should be completed.
     completions : Iterator[Completion]
         iterator over the completions to deduplicate
 
     Yields
     ------
     `Completions` objects
     Completions coming from multiple sources, may be different but end up having
     the same effect when applied to ``text``. If this is the case, this will
     consider completions as equal and only emit the first encountered.
     Not folded in `completions()` yet for debugging purpose, and to detect when
     the IPython completer does return things that Jedi does not, but should be
     at some point.
     """
     completions = list(completions)
     if not completions:
         return
 
     new_start = min(c.start for c in completions)
     new_end = max(c.end for c in completions)
 
     seen = set()
     for c in completions:
         new_text = text[new_start:c.start] + c.text + text[c.end:new_end]
         if new_text not in seen:
             yield c
             seen.add(new_text)
 
 
 def rectify_completions(text: str, completions: _IC, *, _debug=False)->_IC:
     """
     Rectify a set of completions to all have the same ``start`` and ``end``
 
     .. warning::
 
         Unstable
 
         This function is unstable, API may change without warning.
         It will also raise unless use in proper context manager.
 
     Parameters
     ----------
     text : str
         text that should be completed.
     completions : Iterator[Completion]
         iterator over the completions to rectify
 
     Notes
     -----
     :any:`jedi.api.classes.Completion` s returned by Jedi may not have the same start and end, though
     the Jupyter Protocol requires them to behave like so. This will readjust
     the completion to have the same ``start`` and ``end`` by padding both
     extremities with surrounding text.
 
     During stabilisation should support a ``_debug`` option to log which
     completion are return by the IPython completer and not found in Jedi in
     order to make upstream bug report.
     """
     warnings.warn("`rectify_completions` is a provisional API (as of IPython 6.0). "
                  "It may change without warnings. "
                  "Use in corresponding context manager.",
                   category=ProvisionalCompleterWarning, stacklevel=2)
 
     completions = list(completions)
     if not completions:
         return
     starts = (c.start for c in completions)
     ends = (c.end for c in completions)
 
     new_start = min(starts)
     new_end = max(ends)
 
     seen_jedi = set()
     seen_python_matches = set()
     for c in completions:
         new_text = text[new_start:c.start] + c.text + text[c.end:new_end]
         if c._origin == 'jedi':
             seen_jedi.add(new_text)
         elif c._origin == 'IPCompleter.python_matches':
             seen_python_matches.add(new_text)
         yield Completion(new_start, new_end, new_text, type=c.type, _origin=c._origin, signature=c.signature)
     diff = seen_python_matches.difference(seen_jedi)
     if diff and _debug:
         print('IPython.python matches have extras:', diff)
 
 
 if sys.platform == 'win32':
     DELIMS = ' \t\n`!@#$^&*()=+[{]}|;\'",<>?'
 else:
     DELIMS = ' \t\n`!@#$^&*()=+[{]}\\|;:\'",<>?'
 
 GREEDY_DELIMS = ' =\r\n'
 
 
 class CompletionSplitter(object):
     """An object to split an input line in a manner similar to readline.
 
     By having our own implementation, we can expose readline-like completion in
     a uniform manner to all frontends.  This object only needs to be given the
     line of text to be split and the cursor position on said line, and it
     returns the 'word' to be completed on at the cursor after splitting the
     entire line.
 
     What characters are used as splitting delimiters can be controlled by
     setting the ``delims`` attribute (this is a property that internally
     automatically builds the necessary regular expression)"""
 
     # Private interface
 
     # A string of delimiter characters.  The default value makes sense for
     # IPython's most typical usage patterns.
     _delims = DELIMS
 
     # The expression (a normal string) to be compiled into a regular expression
     # for actual splitting.  We store it as an attribute mostly for ease of
     # debugging, since this type of code can be so tricky to debug.
     _delim_expr = None
 
     # The regular expression that does the actual splitting
     _delim_re = None
 
     def __init__(self, delims=None):
         delims = CompletionSplitter._delims if delims is None else delims
         self.delims = delims
 
     @property
     def delims(self):
         """Return the string of delimiter characters."""
         return self._delims
 
     @delims.setter
     def delims(self, delims):
         """Set the delimiters for line splitting."""
         expr = '[' + ''.join('\\'+ c for c in delims) + ']'
         self._delim_re = re.compile(expr)
         self._delims = delims
         self._delim_expr = expr
 
     def split_line(self, line, cursor_pos=None):
         """Split a line of text with a cursor at the given position.
         """
         l = line if cursor_pos is None else line[:cursor_pos]
         return self._delim_re.split(l)[-1]
 
 
 
 class Completer(Configurable):
 
     greedy = Bool(False,
         help="""Activate greedy completion
         PENDING DEPRECTION. this is now mostly taken care of with Jedi.
 
         This will enable completion on elements of lists, results of function calls, etc.,
         but can be unsafe because the code is actually evaluated on TAB.
         """
     ).tag(config=True)
 
     use_jedi = Bool(default_value=JEDI_INSTALLED,
                     help="Experimental: Use Jedi to generate autocompletions. "
                     "Default to True if jedi is installed.").tag(config=True)
 
     jedi_compute_type_timeout = Int(default_value=400,
         help="""Experimental: restrict time (in milliseconds) during which Jedi can compute types.
         Set to 0 to stop computing types. Non-zero value lower than 100ms may hurt
         performance by preventing jedi to build its cache.
         """).tag(config=True)
 
     debug = Bool(default_value=False,
                  help='Enable debug for the Completer. Mostly print extra '
                       'information for experimental jedi integration.')\
                       .tag(config=True)
 
     backslash_combining_completions = Bool(True,
         help="Enable unicode completions, e.g. \\alpha<tab> . "
              "Includes completion of latex commands, unicode names, and expanding "
              "unicode characters back to latex commands.").tag(config=True)
 
 
 
     def __init__(self, namespace=None, global_namespace=None, **kwargs):
         """Create a new completer for the command line.
 
         Completer(namespace=ns, global_namespace=ns2) -> completer instance.
 
         If unspecified, the default namespace where completions are performed
         is __main__ (technically, __main__.__dict__). Namespaces should be
         given as dictionaries.
 
         An optional second namespace can be given.  This allows the completer
         to handle cases where both the local and global scopes need to be
         distinguished.
         """
 
         # Don't bind to namespace quite yet, but flag whether the user wants a
         # specific namespace or to use __main__.__dict__. This will allow us
         # to bind to __main__.__dict__ at completion time, not now.
         if namespace is None:
             self.use_main_ns = True
         else:
             self.use_main_ns = False
             self.namespace = namespace
 
         # The global namespace, if given, can be bound directly
         if global_namespace is None:
             self.global_namespace = {}
         else:
             self.global_namespace = global_namespace
 
         self.custom_matchers = []
 
         super(Completer, self).__init__(**kwargs)
 
     def complete(self, text, state):
         """Return the next possible completion for 'text'.
 
         This is called successively with state == 0, 1, 2, ... until it
         returns None.  The completion should begin with 'text'.
 
         """
         if self.use_main_ns:
             self.namespace = __main__.__dict__
 
         if state == 0:
             if "." in text:
                 self.matches = self.attr_matches(text)
             else:
                 self.matches = self.global_matches(text)
         try:
             return self.matches[state]
         except IndexError:
             return None
 
     def global_matches(self, text):
         """Compute matches when text is a simple name.
 
         Return a list of all keywords, built-in functions and names currently
         defined in self.namespace or self.global_namespace that match.
 
         """
         matches = []
         match_append = matches.append
         n = len(text)
         for lst in [keyword.kwlist,
                     builtin_mod.__dict__.keys(),
                     self.namespace.keys(),
                     self.global_namespace.keys()]:
             for word in lst:
                 if word[:n] == text and word != "__builtins__":
                     match_append(word)
 
         snake_case_re = re.compile(r"[^_]+(_[^_]+)+?\Z")
         for lst in [self.namespace.keys(),
                     self.global_namespace.keys()]:
             shortened = {"_".join([sub[0] for sub in word.split('_')]) : word
                          for word in lst if snake_case_re.match(word)}
             for word in shortened.keys():
                 if word[:n] == text and word != "__builtins__":
                     match_append(shortened[word])
         return matches
 
     def attr_matches(self, text):
         """Compute matches when text contains a dot.
 
         Assuming the text is of the form NAME.NAME....[NAME], and is
         evaluatable in self.namespace or self.global_namespace, it will be
         evaluated and its attributes (as revealed by dir()) are used as
         possible completions.  (For class instances, class members are
         also considered.)
 
         WARNING: this can still invoke arbitrary C code, if an object
         with a __getattr__ hook is evaluated.
 
         """
 
         # Another option, seems to work great. Catches things like ''.<tab>
         m = re.match(r"(\S+(\.\w+)*)\.(\w*)$", text)
 
         if m:
             expr, attr = m.group(1, 3)
         elif self.greedy:
             m2 = re.match(r"(.+)\.(\w*)$", self.line_buffer)
             if not m2:
                 return []
             expr, attr = m2.group(1,2)
         else:
             return []
 
         try:
             obj = eval(expr, self.namespace)
         except:
             try:
                 obj = eval(expr, self.global_namespace)
             except:
                 return []
 
         if self.limit_to__all__ and hasattr(obj, '__all__'):
             words = get__all__entries(obj)
         else:
             words = dir2(obj)
 
         try:
             words = generics.complete_object(obj, words)
         except TryNext:
             pass
         except AssertionError:
             raise
         except Exception:
             # Silence errors from completion function
             #raise # dbg
             pass
         # Build match list to return
         n = len(attr)
         return [u"%s.%s" % (expr, w) for w in words if w[:n] == attr ]
 
 
 def get__all__entries(obj):
     """returns the strings in the __all__ attribute"""
     try:
         words = getattr(obj, '__all__')
     except:
         return []
 
     return [w for w in words if isinstance(w, str)]
 
 
 def match_dict_keys(keys: List[Union[str, bytes, Tuple[Union[str, bytes]]]], prefix: str, delims: str,
                     extra_prefix: Optional[Tuple[str, bytes]]=None) -> Tuple[str, int, List[str]]:
     """Used by dict_key_matches, matching the prefix to a list of keys
 
     Parameters
     ----------
     keys
         list of keys in dictionary currently being completed.
     prefix
         Part of the text already typed by the user. E.g. `mydict[b'fo`
     delims
         String of delimiters to consider when finding the current key.
     extra_prefix : optional
         Part of the text already typed in multi-key index cases. E.g. for
         `mydict['foo', "bar", 'b`, this would be `('foo', 'bar')`.
 
     Returns
     -------
     A tuple of three elements: ``quote``, ``token_start``, ``matched``, with
     ``quote`` being the quote that need to be used to close current string.
     ``token_start`` the position where the replacement should start occurring,
     ``matches`` a list of replacement/completion
 
     """
     prefix_tuple = extra_prefix if extra_prefix else ()
     Nprefix = len(prefix_tuple)
     def filter_prefix_tuple(key):
         # Reject too short keys
         if len(key) <= Nprefix:
             return False
         # Reject keys with non str/bytes in it
         for k in key:
             if not isinstance(k, (str, bytes)):
                 return False
         # Reject keys that do not match the prefix
         for k, pt in zip(key, prefix_tuple):
             if k != pt:
                 return False
         # All checks passed!
         return True
 
     filtered_keys:List[Union[str,bytes]] = []
     def _add_to_filtered_keys(key):
         if isinstance(key, (str, bytes)):
             filtered_keys.append(key)
 
     for k in keys:
         if isinstance(k, tuple):
             if filter_prefix_tuple(k):
                 _add_to_filtered_keys(k[Nprefix])
         else:
             _add_to_filtered_keys(k)
 
     if not prefix:
         return '', 0, [repr(k) for k in filtered_keys]
     quote_match = re.search('["\']', prefix)
     assert quote_match is not None # silence mypy
     quote = quote_match.group()
     try:
         prefix_str = eval(prefix + quote, {})
     except Exception:
         return '', 0, []
 
     pattern = '[^' + ''.join('\\' + c for c in delims) + ']*$'
     token_match = re.search(pattern, prefix, re.UNICODE)
     assert token_match is not None # silence mypy
     token_start = token_match.start()
     token_prefix = token_match.group()
 
     matched:List[str] = []
     for key in filtered_keys:
         try:
             if not key.startswith(prefix_str):
                 continue
         except (AttributeError, TypeError, UnicodeError):
             # Python 3+ TypeError on b'a'.startswith('a') or vice-versa
             continue
 
         # reformat remainder of key to begin with prefix
         rem = key[len(prefix_str):]
         # force repr wrapped in '
         rem_repr = repr(rem + '"') if isinstance(rem, str) else repr(rem + b'"')
         rem_repr = rem_repr[1 + rem_repr.index("'"):-2]
         if quote == '"':
             # The entered prefix is quoted with ",
             # but the match is quoted with '.
             # A contained " hence needs escaping for comparison:
             rem_repr = rem_repr.replace('"', '\\"')
 
         # then reinsert prefix from start of token
         matched.append('%s%s' % (token_prefix, rem_repr))
     return quote, token_start, matched
 
 
 def cursor_to_position(text:str, line:int, column:int)->int:
     """
     Convert the (line,column) position of the cursor in text to an offset in a
     string.
 
     Parameters
     ----------
     text : str
         The text in which to calculate the cursor offset
     line : int
         Line of the cursor; 0-indexed
     column : int
         Column of the cursor 0-indexed
 
     Returns
     -------
     Position of the cursor in ``text``, 0-indexed.
 
     See Also
     --------
     position_to_cursor : reciprocal of this function
 
     """
     lines = text.split('\n')
     assert line <= len(lines), '{} <= {}'.format(str(line), str(len(lines)))
 
     return sum(len(l) + 1 for l in lines[:line]) + column
 
 def position_to_cursor(text:str, offset:int)->Tuple[int, int]:
     """
     Convert the position of the cursor in text (0 indexed) to a line
     number(0-indexed) and a column number (0-indexed) pair
 
     Position should be a valid position in ``text``.
 
     Parameters
     ----------
     text : str
         The text in which to calculate the cursor offset
     offset : int
         Position of the cursor in ``text``, 0-indexed.
 
     Returns
     -------
     (line, column) : (int, int)
         Line of the cursor; 0-indexed, column of the cursor 0-indexed
 
     See Also
     --------
     cursor_to_position : reciprocal of this function
 
     """
 
     assert 0 <= offset <= len(text) , "0 <= %s <= %s" % (offset , len(text))
 
     before = text[:offset]
     blines = before.split('\n')  # ! splitnes trim trailing \n
     line = before.count('\n')
     col = len(blines[-1])
     return line, col
 
 
 def _safe_isinstance(obj, module, class_name):
     """Checks if obj is an instance of module.class_name if loaded
     """
     return (module in sys.modules and
             isinstance(obj, getattr(import_module(module), class_name)))
 
 def back_unicode_name_matches(text:str) -> Tuple[str, Sequence[str]]:
     """Match Unicode characters back to Unicode name
 
     This does  ``☃`` -> ``\\snowman``
 
     Note that snowman is not a valid python3 combining character but will be expanded.
     Though it will not recombine back to the snowman character by the completion machinery.
 
     This will not either back-complete standard sequences like \\n, \\b ...
 
     Returns
     =======
 
     Return a tuple with two elements:
 
     - The Unicode character that was matched (preceded with a backslash), or
         empty string,
     - a sequence (of 1), name for the match Unicode character, preceded by
         backslash, or empty if no match.
     
     """
     if len(text)<2:
         return '', ()
     maybe_slash = text[-2]
     if maybe_slash != '\\':
         return '', ()
 
     char = text[-1]
     # no expand on quote for completion in strings.
     # nor backcomplete standard ascii keys
     if char in string.ascii_letters or char in ('"',"'"):
         return '', ()
     try :
         unic = unicodedata.name(char)
         return '\\'+char,('\\'+unic,)
     except KeyError:
         pass
     return '', ()
 
 def back_latex_name_matches(text:str) -> Tuple[str, Sequence[str]] :
     """Match latex characters back to unicode name
 
     This does ``\\ℵ`` -> ``\\aleph``
 
     """
     if len(text)<2:
         return '', ()
     maybe_slash = text[-2]
     if maybe_slash != '\\':
         return '', ()
 
 
     char = text[-1]
     # no expand on quote for completion in strings.
     # nor backcomplete standard ascii keys
     if char in string.ascii_letters or char in ('"',"'"):
         return '', ()
     try :
         latex = reverse_latex_symbol[char]
         # '\\' replace the \ as well
         return '\\'+char,[latex]
     except KeyError:
         pass
     return '', ()
 
 
 def _formatparamchildren(parameter) -> str:
     """
     Get parameter name and value from Jedi Private API
 
     Jedi does not expose a simple way to get `param=value` from its API.
 
     Parameters
     ----------
     parameter
         Jedi's function `Param`
 
     Returns
     -------
     A string like 'a', 'b=1', '*args', '**kwargs'
 
     """
     description = parameter.description
     if not description.startswith('param '):
         raise ValueError('Jedi function parameter description have change format.'
                          'Expected "param ...", found %r".' % description)
     return description[6:]
 
 def _make_signature(completion)-> str:
     """
     Make the signature from a jedi completion
 
     Parameters
     ----------
     completion : jedi.Completion
         object does not complete a function type
 
     Returns
     -------
     a string consisting of the function signature, with the parenthesis but
     without the function name. example:
     `(a, *args, b=1, **kwargs)`
 
     """
 
     # it looks like this might work on jedi 0.17
     if hasattr(completion, 'get_signatures'):
         signatures = completion.get_signatures()
         if not signatures:
             return  '(?)'
 
         c0 = completion.get_signatures()[0]
         return '('+c0.to_string().split('(', maxsplit=1)[1]
 
     return '(%s)'% ', '.join([f for f in (_formatparamchildren(p) for signature in completion.get_signatures()
                                           for p in signature.defined_names()) if f])
 
 
 class _CompleteResult(NamedTuple):
     matched_text : str
     matches: Sequence[str]
     matches_origin: Sequence[str]
     jedi_matches: Any
 
 
 class IPCompleter(Completer):
     """Extension of the completer class with IPython-specific features"""
 
     __dict_key_regexps: Optional[Dict[bool,Pattern]] = None
 
     @observe('greedy')
     def _greedy_changed(self, change):
         """update the splitter and readline delims when greedy is changed"""
         if change['new']:
             self.splitter.delims = GREEDY_DELIMS
         else:
             self.splitter.delims = DELIMS
 
     dict_keys_only = Bool(False,
         help="""Whether to show dict key matches only""")
 
     merge_completions = Bool(True,
         help="""Whether to merge completion results into a single list
 
         If False, only the completion results from the first non-empty
         completer will be returned.
         """
     ).tag(config=True)
     omit__names = Enum((0,1,2), default_value=2,
         help="""Instruct the completer to omit private method names
 
         Specifically, when completing on ``object.<tab>``.
 
         When 2 [default]: all names that start with '_' will be excluded.
 
         When 1: all 'magic' names (``__foo__``) will be excluded.
 
         When 0: nothing will be excluded.
         """
     ).tag(config=True)
     limit_to__all__ = Bool(False,
         help="""
         DEPRECATED as of version 5.0.
 
         Instruct the completer to use __all__ for the completion
 
         Specifically, when completing on ``object.<tab>``.
 
         When True: only those names in obj.__all__ will be included.
 
         When False [default]: the __all__ attribute is ignored
         """,
     ).tag(config=True)
 
     profile_completions = Bool(
         default_value=False,
         help="If True, emit profiling data for completion subsystem using cProfile."
     ).tag(config=True)
 
     profiler_output_dir = Unicode(
         default_value=".completion_profiles",
         help="Template for path at which to output profile data for completions."
     ).tag(config=True)
 
     @observe('limit_to__all__')
     def _limit_to_all_changed(self, change):
         warnings.warn('`IPython.core.IPCompleter.limit_to__all__` configuration '
             'value has been deprecated since IPython 5.0, will be made to have '
             'no effects and then removed in future version of IPython.',
             UserWarning)
 
     def __init__(self, shell=None, namespace=None, global_namespace=None,
                  use_readline=_deprecation_readline_sentinel, config=None, **kwargs):
         """IPCompleter() -> completer
 
         Return a completer object.
 
         Parameters
         ----------
         shell
             a pointer to the ipython shell itself.  This is needed
             because this completer knows about magic functions, and those can
             only be accessed via the ipython instance.
         namespace : dict, optional
             an optional dict where completions are performed.
         global_namespace : dict, optional
             secondary optional dict for completions, to
             handle cases (such as IPython embedded inside functions) where
             both Python scopes are visible.
         use_readline : bool, optional
             DEPRECATED, ignored since IPython 6.0, will have no effects
         """
 
         self.magic_escape = ESC_MAGIC
         self.splitter = CompletionSplitter()
 
         if use_readline is not _deprecation_readline_sentinel:
             warnings.warn('The `use_readline` parameter is deprecated and ignored since IPython 6.0.',
                           DeprecationWarning, stacklevel=2)
 
         # _greedy_changed() depends on splitter and readline being defined:
         Completer.__init__(self, namespace=namespace, global_namespace=global_namespace,
                             config=config, **kwargs)
 
         # List where completion matches will be stored
         self.matches = []
         self.shell = shell
         # Regexp to split filenames with spaces in them
         self.space_name_re = re.compile(r'([^\\] )')
         # Hold a local ref. to glob.glob for speed
         self.glob = glob.glob
 
         # Determine if we are running on 'dumb' terminals, like (X)Emacs
         # buffers, to avoid completion problems.
         term = os.environ.get('TERM','xterm')
         self.dumb_terminal = term in ['dumb','emacs']
 
         # Special handling of backslashes needed in win32 platforms
         if sys.platform == "win32":
             self.clean_glob = self._clean_glob_win32
         else:
             self.clean_glob = self._clean_glob
 
         #regexp to parse docstring for function signature
         self.docstring_sig_re = re.compile(r'^[\w|\s.]+\(([^)]*)\).*')
         self.docstring_kwd_re = re.compile(r'[\s|\[]*(\w+)(?:\s*=\s*.*)')
         #use this if positional argument name is also needed
         #= re.compile(r'[\s|\[]*(\w+)(?:\s*=?\s*.*)')
 
         self.magic_arg_matchers = [
             self.magic_config_matches,
             self.magic_color_matches,
         ]
 
         # This is set externally by InteractiveShell
         self.custom_completers = None
 
         # This is a list of names of unicode characters that can be completed
         # into their corresponding unicode value. The list is large, so we
         # laziliy initialize it on first use. Consuming code should access this
         # attribute through the `@unicode_names` property.
         self._unicode_names = None
 
     @property
     def matchers(self) -> List[Any]:
         """All active matcher routines for completion"""
         if self.dict_keys_only:
             return [self.dict_key_matches]
 
         if self.use_jedi:
             return [
                 *self.custom_matchers,
+                self.dict_key_matches,
                 self.file_matches,
                 self.magic_matches,
-                self.dict_key_matches,
             ]
         else:
             return [
                 *self.custom_matchers,
+                self.dict_key_matches,
                 self.python_matches,
                 self.file_matches,
                 self.magic_matches,
                 self.python_func_kw_matches,
-                self.dict_key_matches,
             ]
 
     def all_completions(self, text:str) -> List[str]:
         """
         Wrapper around the completion methods for the benefit of emacs.
         """
         prefix = text.rpartition('.')[0]
         with provisionalcompleter():
             return ['.'.join([prefix, c.text]) if prefix and self.use_jedi else c.text
                     for c in self.completions(text, len(text))]
 
         return self.complete(text)[1]
 
     def _clean_glob(self, text:str):
         return self.glob("%s*" % text)
 
     def _clean_glob_win32(self, text:str):
         return [f.replace("\\","/")
                 for f in self.glob("%s*" % text)]
 
     def file_matches(self, text:str)->List[str]:
         """Match filenames, expanding ~USER type strings.
 
         Most of the seemingly convoluted logic in this completer is an
         attempt to handle filenames with spaces in them.  And yet it's not
         quite perfect, because Python's readline doesn't expose all of the
         GNU readline details needed for this to be done correctly.
 
         For a filename with a space in it, the printed completions will be
         only the parts after what's already been typed (instead of the
         full completions, as is normally done).  I don't think with the
         current (as of Python 2.3) Python readline it's possible to do
         better."""
 
         # chars that require escaping with backslash - i.e. chars
         # that readline treats incorrectly as delimiters, but we
         # don't want to treat as delimiters in filename matching
         # when escaped with backslash
         if text.startswith('!'):
             text = text[1:]
             text_prefix = u'!'
         else:
             text_prefix = u''
 
         text_until_cursor = self.text_until_cursor
         # track strings with open quotes
         open_quotes = has_open_quotes(text_until_cursor)
 
         if '(' in text_until_cursor or '[' in text_until_cursor:
             lsplit = text
         else:
             try:
                 # arg_split ~ shlex.split, but with unicode bugs fixed by us
                 lsplit = arg_split(text_until_cursor)[-1]
             except ValueError:
                 # typically an unmatched ", or backslash without escaped char.
                 if open_quotes:
                     lsplit = text_until_cursor.split(open_quotes)[-1]
                 else:
                     return []
             except IndexError:
                 # tab pressed on empty line
                 lsplit = ""
 
         if not open_quotes and lsplit != protect_filename(lsplit):
             # if protectables are found, do matching on the whole escaped name
             has_protectables = True
             text0,text = text,lsplit
         else:
             has_protectables = False
             text = os.path.expanduser(text)
 
         if text == "":
             return [text_prefix + protect_filename(f) for f in self.glob("*")]
 
         # Compute the matches from the filesystem
         if sys.platform == 'win32':
             m0 = self.clean_glob(text)
         else:
             m0 = self.clean_glob(text.replace('\\', ''))
 
         if has_protectables:
             # If we had protectables, we need to revert our changes to the
             # beginning of filename so that we don't double-write the part
             # of the filename we have so far
             len_lsplit = len(lsplit)
             matches = [text_prefix + text0 +
                        protect_filename(f[len_lsplit:]) for f in m0]
         else:
             if open_quotes:
                 # if we have a string with an open quote, we don't need to
                 # protect the names beyond the quote (and we _shouldn't_, as
                 # it would cause bugs when the filesystem call is made).
                 matches = m0 if sys.platform == "win32" else\
                     [protect_filename(f, open_quotes) for f in m0]
             else:
                 matches = [text_prefix +
                            protect_filename(f) for f in m0]
 
         # Mark directories in input list by appending '/' to their names.
         return [x+'/' if os.path.isdir(x) else x for x in matches]
 
     def magic_matches(self, text:str):
         """Match magics"""
         # Get all shell magics now rather than statically, so magics loaded at
         # runtime show up too.
         lsm = self.shell.magics_manager.lsmagic()
         line_magics = lsm['line']
         cell_magics = lsm['cell']
         pre = self.magic_escape
         pre2 = pre+pre
 
         explicit_magic = text.startswith(pre)
 
         # Completion logic:
         # - user gives %%: only do cell magics
         # - user gives %: do both line and cell magics
         # - no prefix: do both
         # In other words, line magics are skipped if the user gives %% explicitly
         #
         # We also exclude magics that match any currently visible names:
         # https://github.com/ipython/ipython/issues/4877, unless the user has
         # typed a %:
         # https://github.com/ipython/ipython/issues/10754
         bare_text = text.lstrip(pre)
         global_matches = self.global_matches(bare_text)
         if not explicit_magic:
             def matches(magic):
                 """
                 Filter magics, in particular remove magics that match
                 a name present in global namespace.
                 """
                 return ( magic.startswith(bare_text) and
                          magic not in global_matches )
         else:
             def matches(magic):
                 return magic.startswith(bare_text)
 
         comp = [ pre2+m for m in cell_magics if matches(m)]
         if not text.startswith(pre2):
             comp += [ pre+m for m in line_magics if matches(m)]
 
         return comp
 
     def magic_config_matches(self, text:str) -> List[str]:
         """ Match class names and attributes for %config magic """
         texts = text.strip().split()
 
         if len(texts) > 0 and (texts[0] == 'config' or texts[0] == '%config'):
             # get all configuration classes
             classes = sorted(set([ c for c in self.shell.configurables
                                    if c.__class__.class_traits(config=True)
                                    ]), key=lambda x: x.__class__.__name__)
             classnames = [ c.__class__.__name__ for c in classes ]
 
             # return all classnames if config or %config is given
             if len(texts) == 1:
                 return classnames
 
             # match classname
             classname_texts = texts[1].split('.')
             classname = classname_texts[0]
             classname_matches = [ c for c in classnames
                                   if c.startswith(classname) ]
 
             # return matched classes or the matched class with attributes
             if texts[1].find('.') < 0:
                 return classname_matches
             elif len(classname_matches) == 1 and \
                             classname_matches[0] == classname:
                 cls = classes[classnames.index(classname)].__class__
                 help = cls.class_get_help()
                 # strip leading '--' from cl-args:
                 help = re.sub(re.compile(r'^--', re.MULTILINE), '', help)
                 return [ attr.split('=')[0]
                          for attr in help.strip().splitlines()
                          if attr.startswith(texts[1]) ]
         return []
 
     def magic_color_matches(self, text:str) -> List[str] :
         """ Match color schemes for %colors magic"""
         texts = text.split()
         if text.endswith(' '):
             # .split() strips off the trailing whitespace. Add '' back
             # so that: '%colors ' -> ['%colors', '']
             texts.append('')
 
         if len(texts) == 2 and (texts[0] == 'colors' or texts[0] == '%colors'):
             prefix = texts[1]
             return [ color for color in InspectColors.keys()
                      if color.startswith(prefix) ]
         return []
 
     def _jedi_matches(self, cursor_column:int, cursor_line:int, text:str) -> Iterable[Any]:
         """
         Return a list of :any:`jedi.api.Completions` object from a ``text`` and
         cursor position.
 
         Parameters
         ----------
         cursor_column : int
             column position of the cursor in ``text``, 0-indexed.
         cursor_line : int
             line position of the cursor in ``text``, 0-indexed
         text : str
             text to complete
 
         Notes
         -----
         If ``IPCompleter.debug`` is ``True`` may return a :any:`_FakeJediCompletion`
         object containing a string with the Jedi debug information attached.
         """
         namespaces = [self.namespace]
         if self.global_namespace is not None:
             namespaces.append(self.global_namespace)
 
         completion_filter = lambda x:x
         offset = cursor_to_position(text, cursor_line, cursor_column)
         # filter output if we are completing for object members
         if offset:
             pre = text[offset-1]
             if pre == '.':
                 if self.omit__names == 2:
                     completion_filter = lambda c:not c.name.startswith('_')
                 elif self.omit__names == 1:
                     completion_filter = lambda c:not (c.name.startswith('__') and c.name.endswith('__'))
                 elif self.omit__names == 0:
                     completion_filter = lambda x:x
                 else:
                     raise ValueError("Don't understand self.omit__names == {}".format(self.omit__names))
 
         interpreter = jedi.Interpreter(text[:offset], namespaces)
         try_jedi = True
 
         try:
             # find the first token in the current tree -- if it is a ' or " then we are in a string
             completing_string = False
             try:
                 first_child = next(c for c in interpreter._get_module().tree_node.children if hasattr(c, 'value'))
             except StopIteration:
                 pass
             else:
                 # note the value may be ', ", or it may also be ''' or """, or
                 # in some cases, """what/you/typed..., but all of these are
                 # strings.
                 completing_string = len(first_child.value) > 0 and first_child.value[0] in {"'", '"'}
 
             # if we are in a string jedi is likely not the right candidate for
             # now. Skip it.
             try_jedi = not completing_string
         except Exception as e:
             # many of things can go wrong, we are using private API just don't crash.
             if self.debug:
                 print("Error detecting if completing a non-finished string :", e, '|')
 
         if not try_jedi:
             return []
         try:
             return filter(completion_filter, interpreter.complete(column=cursor_column, line=cursor_line + 1))
         except Exception as e:
             if self.debug:
                 return [_FakeJediCompletion('Oops Jedi has crashed, please report a bug with the following:\n"""\n%s\ns"""' % (e))]
             else:
                 return []
 
     def python_matches(self, text:str)->List[str]:
         """Match attributes or global python names"""
         if "." in text:
             try:
                 matches = self.attr_matches(text)
                 if text.endswith('.') and self.omit__names:
                     if self.omit__names == 1:
                         # true if txt is _not_ a __ name, false otherwise:
                         no__name = (lambda txt:
                                     re.match(r'.*\.__.*?__',txt) is None)
                     else:
                         # true if txt is _not_ a _ name, false otherwise:
                         no__name = (lambda txt:
                                     re.match(r'\._.*?',txt[txt.rindex('.'):]) is None)
                     matches = filter(no__name, matches)
             except NameError:
                 # catches <undefined attributes>.<tab>
                 matches = []
         else:
             matches = self.global_matches(text)
         return matches
 
     def _default_arguments_from_docstring(self, doc):
         """Parse the first line of docstring for call signature.
 
         Docstring should be of the form 'min(iterable[, key=func])\n'.
         It can also parse cython docstring of the form
         'Minuit.migrad(self, int ncall=10000, resume=True, int nsplit=1)'.
         """
         if doc is None:
             return []
 
         #care only the firstline
         line = doc.lstrip().splitlines()[0]
 
         #p = re.compile(r'^[\w|\s.]+\(([^)]*)\).*')
         #'min(iterable[, key=func])\n' -> 'iterable[, key=func]'
         sig = self.docstring_sig_re.search(line)
         if sig is None:
             return []
         # iterable[, key=func]' -> ['iterable[' ,' key=func]']
         sig = sig.groups()[0].split(',')
         ret = []
         for s in sig:
             #re.compile(r'[\s|\[]*(\w+)(?:\s*=\s*.*)')
             ret += self.docstring_kwd_re.findall(s)
         return ret
 
     def _default_arguments(self, obj):
         """Return the list of default arguments of obj if it is callable,
         or empty list otherwise."""
         call_obj = obj
         ret = []
         if inspect.isbuiltin(obj):
             pass
         elif not (inspect.isfunction(obj) or inspect.ismethod(obj)):
             if inspect.isclass(obj):
                 #for cython embedsignature=True the constructor docstring
                 #belongs to the object itself not __init__
                 ret += self._default_arguments_from_docstring(
                             getattr(obj, '__doc__', ''))
                 # for classes, check for __init__,__new__
                 call_obj = (getattr(obj, '__init__', None) or
                        getattr(obj, '__new__', None))
             # for all others, check if they are __call__able
             elif hasattr(obj, '__call__'):
                 call_obj = obj.__call__
         ret += self._default_arguments_from_docstring(
                  getattr(call_obj, '__doc__', ''))
 
         _keeps = (inspect.Parameter.KEYWORD_ONLY,
                   inspect.Parameter.POSITIONAL_OR_KEYWORD)
 
         try:
             sig = inspect.signature(obj)
             ret.extend(k for k, v in sig.parameters.items() if
                        v.kind in _keeps)
         except ValueError:
             pass
 
         return list(set(ret))
 
     def python_func_kw_matches(self, text):
         """Match named parameters (kwargs) of the last open function"""
 
         if "." in text: # a parameter cannot be dotted
             return []
         try: regexp = self.__funcParamsRegex
         except AttributeError:
             regexp = self.__funcParamsRegex = re.compile(r'''
                 '.*?(?<!\\)' |    # single quoted strings or
                 ".*?(?<!\\)" |    # double quoted strings or
                 \w+          |    # identifier
                 \S                # other characters
                 ''', re.VERBOSE | re.DOTALL)
         # 1. find the nearest identifier that comes before an unclosed
         # parenthesis before the cursor
         # e.g. for "foo (1+bar(x), pa<cursor>,a=1)", the candidate is "foo"
         tokens = regexp.findall(self.text_until_cursor)
         iterTokens = reversed(tokens); openPar = 0
 
         for token in iterTokens:
             if token == ')':
                 openPar -= 1
             elif token == '(':
                 openPar += 1
                 if openPar > 0:
                     # found the last unclosed parenthesis
                     break
         else:
             return []
         # 2. Concatenate dotted names ("foo.bar" for "foo.bar(x, pa" )
         ids = []
         isId = re.compile(r'\w+$').match
 
         while True:
             try:
                 ids.append(next(iterTokens))
                 if not isId(ids[-1]):
                     ids.pop(); break
                 if not next(iterTokens) == '.':
                     break
             except StopIteration:
                 break
 
         # Find all named arguments already assigned to, as to avoid suggesting
         # them again
         usedNamedArgs = set()
         par_level = -1
         for token, next_token in zip(tokens, tokens[1:]):
             if token == '(':
                 par_level += 1
             elif token == ')':
                 par_level -= 1
 
             if par_level != 0:
                 continue
 
             if next_token != '=':
                 continue
 
             usedNamedArgs.add(token)
 
         argMatches = []
         try:
             callableObj = '.'.join(ids[::-1])
             namedArgs = self._default_arguments(eval(callableObj,
                                                     self.namespace))
 
             # Remove used named arguments from the list, no need to show twice
             for namedArg in set(namedArgs) - usedNamedArgs:
                 if namedArg.startswith(text):
                     argMatches.append("%s=" %namedArg)
         except:
             pass
 
         return argMatches
 
     @staticmethod
     def _get_keys(obj: Any) -> List[Any]:
         # Objects can define their own completions by defining an
         # _ipy_key_completions_() method.
         method = get_real_method(obj, '_ipython_key_completions_')
         if method is not None:
             return method()
 
         # Special case some common in-memory dict-like types
         if isinstance(obj, dict) or\
            _safe_isinstance(obj, 'pandas', 'DataFrame'):
             try:
                 return list(obj.keys())
             except Exception:
                 return []
         elif _safe_isinstance(obj, 'numpy', 'ndarray') or\
              _safe_isinstance(obj, 'numpy', 'void'):
             return obj.dtype.names or []
         return []
 
     def dict_key_matches(self, text:str) -> List[str]:
         "Match string keys in a dictionary, after e.g. 'foo[' "
 
 
         if self.__dict_key_regexps is not None:
             regexps = self.__dict_key_regexps
         else:
             dict_key_re_fmt = r'''(?x)
             (  # match dict-referring expression wrt greedy setting
                 %s
             )
             \[   # open bracket
             \s*  # and optional whitespace
             # Capture any number of str-like objects (e.g. "a", "b", 'c')
             ((?:[uUbB]?  # string prefix (r not handled)
                 (?:
                     '(?:[^']|(?<!\\)\\')*'
                 |
                     "(?:[^"]|(?<!\\)\\")*"
                 )
                 \s*,\s*
             )*)
             ([uUbB]?  # string prefix (r not handled)
                 (?:   # unclosed string
                     '(?:[^']|(?<!\\)\\')*
                 |
                     "(?:[^"]|(?<!\\)\\")*
                 )
             )?
             $
             '''
             regexps = self.__dict_key_regexps = {
                 False: re.compile(dict_key_re_fmt % r'''
                                   # identifiers separated by .
                                   (?!\d)\w+
                                   (?:\.(?!\d)\w+)*
                                   '''),
                 True: re.compile(dict_key_re_fmt % '''
                                  .+
                                  ''')
             }
 
         match = regexps[self.greedy].search(self.text_until_cursor)
 
         if match is None:
             return []
 
         expr, prefix0, prefix = match.groups()
         try:
             obj = eval(expr, self.namespace)
         except Exception:
             try:
                 obj = eval(expr, self.global_namespace)
             except Exception:
                 return []
 
         keys = self._get_keys(obj)
         if not keys:
             return keys
 
         extra_prefix = eval(prefix0) if prefix0 != '' else None
 
         closing_quote, token_offset, matches = match_dict_keys(keys, prefix, self.splitter.delims, extra_prefix=extra_prefix)
         if not matches:
             return matches
 
         # get the cursor position of
         # - the text being completed
         # - the start of the key text
         # - the start of the completion
         text_start = len(self.text_until_cursor) - len(text)
         if prefix:
             key_start = match.start(3)
             completion_start = key_start + token_offset
         else:
             key_start = completion_start = match.end()
 
         # grab the leading prefix, to make sure all completions start with `text`
         if text_start > key_start:
             leading = ''
         else:
             leading = text[text_start:completion_start]
 
         # the index of the `[` character
         bracket_idx = match.end(1)
 
         # append closing quote and bracket as appropriate
         # this is *not* appropriate if the opening quote or bracket is outside
         # the text given to this method
         suf = ''
         continuation = self.line_buffer[len(self.text_until_cursor):]
         if key_start > text_start and closing_quote:
             # quotes were opened inside text, maybe close them
             if continuation.startswith(closing_quote):
                 continuation = continuation[len(closing_quote):]
             else:
                 suf += closing_quote
         if bracket_idx > text_start:
             # brackets were opened inside text, maybe close them
             if not continuation.startswith(']'):
                 suf += ']'
 
         return [leading + k + suf for k in matches]
 
     @staticmethod
     def unicode_name_matches(text:str) -> Tuple[str, List[str]] :
         """Match Latex-like syntax for unicode characters base
         on the name of the character.
 
         This does  ``\\GREEK SMALL LETTER ETA`` -> ``η``
 
         Works only on valid python 3 identifier, or on combining characters that
         will combine to form a valid identifier.
         """
         slashpos = text.rfind('\\')
         if slashpos > -1:
             s = text[slashpos+1:]
             try :
                 unic = unicodedata.lookup(s)
                 # allow combining chars
                 if ('a'+unic).isidentifier():
                     return '\\'+s,[unic]
             except KeyError:
                 pass
         return '', []
 
 
     def latex_matches(self, text:str) -> Tuple[str, Sequence[str]]:
         """Match Latex syntax for unicode characters.
 
         This does both ``\\alp`` -> ``\\alpha`` and ``\\alpha`` -> ``α``
         """
         slashpos = text.rfind('\\')
         if slashpos > -1:
             s = text[slashpos:]
             if s in latex_symbols:
                 # Try to complete a full latex symbol to unicode
                 # \\alpha -> α
                 return s, [latex_symbols[s]]
             else:
                 # If a user has partially typed a latex symbol, give them
                 # a full list of options \al -> [\aleph, \alpha]
                 matches = [k for k in latex_symbols if k.startswith(s)]
                 if matches:
                     return s, matches
         return '', ()
 
     def dispatch_custom_completer(self, text):
         if not self.custom_completers:
             return
 
         line = self.line_buffer
         if not line.strip():
             return None
 
         # Create a little structure to pass all the relevant information about
         # the current completion to any custom completer.
         event = SimpleNamespace()
         event.line = line
         event.symbol = text
         cmd = line.split(None,1)[0]
         event.command = cmd
         event.text_until_cursor = self.text_until_cursor
 
         # for foo etc, try also to find completer for %foo
         if not cmd.startswith(self.magic_escape):
             try_magic = self.custom_completers.s_matches(
                 self.magic_escape + cmd)
         else:
             try_magic = []
 
         for c in itertools.chain(self.custom_completers.s_matches(cmd),
                  try_magic,
                  self.custom_completers.flat_matches(self.text_until_cursor)):
             try:
                 res = c(event)
                 if res:
                     # first, try case sensitive match
                     withcase = [r for r in res if r.startswith(text)]
                     if withcase:
                         return withcase
                     # if none, then case insensitive ones are ok too
                     text_low = text.lower()
                     return [r for r in res if r.lower().startswith(text_low)]
             except TryNext:
                 pass
             except KeyboardInterrupt:
                 """
                 If custom completer take too long,
                 let keyboard interrupt abort and return nothing.
                 """
                 break
 
         return None
 
     def completions(self, text: str, offset: int)->Iterator[Completion]:
         """
         Returns an iterator over the possible completions
 
         .. warning::
 
             Unstable
 
             This function is unstable, API may change without warning.
             It will also raise unless use in proper context manager.
 
         Parameters
         ----------
         text : str
             Full text of the current input, multi line string.
         offset : int
             Integer representing the position of the cursor in ``text``. Offset
             is 0-based indexed.
 
         Yields
         ------
         Completion
 
         Notes
         -----
         The cursor on a text can either be seen as being "in between"
         characters or "On" a character depending on the interface visible to
         the user. For consistency the cursor being on "in between" characters X
         and Y is equivalent to the cursor being "on" character Y, that is to say
         the character the cursor is on is considered as being after the cursor.
 
         Combining characters may span more that one position in the
         text.
 
         .. note::
 
             If ``IPCompleter.debug`` is :any:`True` will yield a ``--jedi/ipython--``
             fake Completion token to distinguish completion returned by Jedi
             and usual IPython completion.
 
         .. note::
 
             Completions are not completely deduplicated yet. If identical
             completions are coming from different sources this function does not
             ensure that each completion object will only be present once.
         """
         warnings.warn("_complete is a provisional API (as of IPython 6.0). "
                       "It may change without warnings. "
                       "Use in corresponding context manager.",
                       category=ProvisionalCompleterWarning, stacklevel=2)
 
         seen = set()
         profiler:Optional[cProfile.Profile]
         try:
             if self.profile_completions:
                 import cProfile
                 profiler = cProfile.Profile()
                 profiler.enable()
             else:
                 profiler = None
 
             for c in self._completions(text, offset, _timeout=self.jedi_compute_type_timeout/1000):
                 if c and (c in seen):
                     continue
                 yield c
                 seen.add(c)
         except KeyboardInterrupt:
             """if completions take too long and users send keyboard interrupt,
             do not crash and return ASAP. """
             pass
         finally:
             if profiler is not None:
                 profiler.disable()
                 ensure_dir_exists(self.profiler_output_dir)
                 output_path = os.path.join(self.profiler_output_dir, str(uuid.uuid4()))
                 print("Writing profiler output to", output_path)
                 profiler.dump_stats(output_path)
 
     def _completions(self, full_text: str, offset: int, *, _timeout) -> Iterator[Completion]:
         """
         Core completion module.Same signature as :any:`completions`, with the
         extra `timeout` parameter (in seconds).
 
         Computing jedi's completion ``.type`` can be quite expensive (it is a
         lazy property) and can require some warm-up, more warm up than just
         computing the ``name`` of a completion. The warm-up can be :
 
             - Long warm-up the first time a module is encountered after
             install/update: actually build parse/inference tree.
 
             - first time the module is encountered in a session: load tree from
             disk.
 
         We don't want to block completions for tens of seconds so we give the
         completer a "budget" of ``_timeout`` seconds per invocation to compute
         completions types, the completions that have not yet been computed will
         be marked as "unknown" an will have a chance to be computed next round
         are things get cached.
 
         Keep in mind that Jedi is not the only thing treating the completion so
         keep the timeout short-ish as if we take more than 0.3 second we still
         have lots of processing to do.
 
         """
         deadline = time.monotonic() + _timeout
 
 
         before = full_text[:offset]
         cursor_line, cursor_column = position_to_cursor(full_text, offset)
 
         matched_text, matches, matches_origin, jedi_matches = self._complete(
             full_text=full_text, cursor_line=cursor_line, cursor_pos=cursor_column)
 
         iter_jm = iter(jedi_matches)
         if _timeout:
             for jm in iter_jm:
                 try:
                     type_ = jm.type
                 except Exception:
                     if self.debug:
                         print("Error in Jedi getting type of ", jm)
                     type_ = None
                 delta = len(jm.name_with_symbols) - len(jm.complete)
                 if type_ == 'function':
                     signature = _make_signature(jm)
                 else:
                     signature = ''
                 yield Completion(start=offset - delta,
                                  end=offset,
                                  text=jm.name_with_symbols,
                                  type=type_,
                                  signature=signature,
                                  _origin='jedi')
 
                 if time.monotonic() > deadline:
                     break
 
         for jm in iter_jm:
             delta = len(jm.name_with_symbols) - len(jm.complete)
             yield Completion(start=offset - delta,
                              end=offset,
                              text=jm.name_with_symbols,
                              type='<unknown>',  # don't compute type for speed
                              _origin='jedi',
                              signature='')
 
 
         start_offset = before.rfind(matched_text)
 
         # TODO:
         # Suppress this, right now just for debug.
         if jedi_matches and matches and self.debug:
             yield Completion(start=start_offset, end=offset, text='--jedi/ipython--',
                              _origin='debug', type='none', signature='')
 
         # I'm unsure if this is always true, so let's assert and see if it
         # crash
         assert before.endswith(matched_text)
         for m, t in zip(matches, matches_origin):
             yield Completion(start=start_offset, end=offset, text=m, _origin=t, signature='', type='<unknown>')
 
 
     def complete(self, text=None, line_buffer=None, cursor_pos=None) -> Tuple[str, Sequence[str]]:
         """Find completions for the given text and line context.
 
         Note that both the text and the line_buffer are optional, but at least
         one of them must be given.
 
         Parameters
         ----------
         text : string, optional
             Text to perform the completion on.  If not given, the line buffer
             is split using the instance's CompletionSplitter object.
         line_buffer : string, optional
             If not given, the completer attempts to obtain the current line
             buffer via readline.  This keyword allows clients which are
             requesting for text completions in non-readline contexts to inform
             the completer of the entire text.
         cursor_pos : int, optional
             Index of the cursor in the full line buffer.  Should be provided by
             remote frontends where kernel has no access to frontend state.
 
         Returns
         -------
         Tuple of two items:
         text : str
             Text that was actually used in the completion.
         matches : list
             A list of completion matches.
 
         Notes
         -----
             This API is likely to be deprecated and replaced by
             :any:`IPCompleter.completions` in the future.
 
         """
         warnings.warn('`Completer.complete` is pending deprecation since '
                 'IPython 6.0 and will be replaced by `Completer.completions`.',
                       PendingDeprecationWarning)
         # potential todo, FOLD the 3rd throw away argument of _complete
         # into the first 2 one.
         return self._complete(line_buffer=line_buffer, cursor_pos=cursor_pos, text=text, cursor_line=0)[:2]
 
     def _complete(self, *, cursor_line, cursor_pos, line_buffer=None, text=None,
                   full_text=None) -> _CompleteResult:
         """
         Like complete but can also returns raw jedi completions as well as the
         origin of the completion text. This could (and should) be made much
         cleaner but that will be simpler once we drop the old (and stateful)
         :any:`complete` API.
 
         With current provisional API, cursor_pos act both (depending on the
         caller) as the offset in the ``text`` or ``line_buffer``, or as the
         ``column`` when passing multiline strings this could/should be renamed
         but would add extra noise.
 
         Parameters
         ----------
         cursor_line :
             Index of the line the cursor is on. 0 indexed.
         cursor_pos :
             Position of the cursor in the current line/line_buffer/text. 0
             indexed.
         line_buffer : optional, str
             The current line the cursor is in, this is mostly due to legacy
             reason that readline coudl only give a us the single current line.
             Prefer `full_text`.
         text : str
             The current "token" the cursor is in, mostly also for historical
             reasons. as the completer would trigger only after the current line
             was parsed.
         full_text : str
             Full text of the current cell.
 
         Returns
         -------
         A tuple of N elements which are (likely):
             matched_text: ? the text that the complete matched
             matches: list of completions ?
             matches_origin: ? list same length as matches, and where each completion came from
             jedi_matches: list of Jedi matches, have it's own structure.
         """
 
 
         # if the cursor position isn't given, the only sane assumption we can
         # make is that it's at the end of the line (the common case)
         if cursor_pos is None:
             cursor_pos = len(line_buffer) if text is None else len(text)
 
         if self.use_main_ns:
             self.namespace = __main__.__dict__
 
         # if text is either None or an empty string, rely on the line buffer
         if (not line_buffer) and full_text:
             line_buffer = full_text.split('\n')[cursor_line]
         if not text: # issue #11508: check line_buffer before calling split_line
             text = self.splitter.split_line(line_buffer, cursor_pos)  if line_buffer else ''
 
         if self.backslash_combining_completions:
             # allow deactivation of these on windows.
             base_text = text if not line_buffer else line_buffer[:cursor_pos]
 
             for meth in (self.latex_matches,
                          self.unicode_name_matches,
                          back_latex_name_matches,
                          back_unicode_name_matches,
                          self.fwd_unicode_match):
                 name_text, name_matches = meth(base_text)
                 if name_text:
                     return _CompleteResult(name_text, name_matches[:MATCHES_LIMIT], \
                            [meth.__qualname__]*min(len(name_matches), MATCHES_LIMIT), ())
 
 
         # If no line buffer is given, assume the input text is all there was
         if line_buffer is None:
             line_buffer = text
 
         self.line_buffer = line_buffer
         self.text_until_cursor = self.line_buffer[:cursor_pos]
 
         # Do magic arg matches
         for matcher in self.magic_arg_matchers:
             matches = list(matcher(line_buffer))[:MATCHES_LIMIT]
             if matches:
                 origins = [matcher.__qualname__] * len(matches)
                 return _CompleteResult(text, matches, origins, ())
 
         # Start with a clean slate of completions
         matches = []
 
         # FIXME: we should extend our api to return a dict with completions for
         # different types of objects.  The rlcomplete() method could then
         # simply collapse the dict into a list for readline, but we'd have
         # richer completion semantics in other environments.
         completions:Iterable[Any] = []
         if self.use_jedi:
             if not full_text:
                 full_text = line_buffer
             completions = self._jedi_matches(
                 cursor_pos, cursor_line, full_text)
 
         if self.merge_completions:
             matches = []
             for matcher in self.matchers:
                 try:
                     matches.extend([(m, matcher.__qualname__)
                                     for m in matcher(text)])
                 except:
                     # Show the ugly traceback if the matcher causes an
                     # exception, but do NOT crash the kernel!
                     sys.excepthook(*sys.exc_info())
         else:
             for matcher in self.matchers:
                 matches = [(m, matcher.__qualname__)
                             for m in matcher(text)]
                 if matches:
                     break
                     
         seen = set()
         filtered_matches = set()
         for m in matches:
             t, c = m
             if t not in seen:
                 filtered_matches.add(m)
                 seen.add(t)
 
         _filtered_matches = sorted(filtered_matches, key=lambda x: completions_sorting_key(x[0]))
 
         custom_res = [(m, 'custom') for m in self.dispatch_custom_completer(text) or []]
         
         _filtered_matches = custom_res or _filtered_matches
         
         _filtered_matches = _filtered_matches[:MATCHES_LIMIT]
         _matches = [m[0] for m in _filtered_matches]
         origins = [m[1] for m in _filtered_matches]
 
         self.matches = _matches
 
         return _CompleteResult(text, _matches, origins, completions)
         
     def fwd_unicode_match(self, text:str) -> Tuple[str, Sequence[str]]:
         """
         Forward match a string starting with a backslash with a list of
         potential Unicode completions.
 
         Will compute list list of Unicode character names on first call and cache it.
 
         Returns
         -------
         At tuple with:
             - matched text (empty if no matches)
             - list of potential completions, empty tuple  otherwise)
         """
         # TODO: self.unicode_names is here a list we traverse each time with ~100k elements.
         # We could do a faster match using a Trie.
 
         # Using pygtrie the following seem to work:
 
         #     s = PrefixSet()
 
         #     for c in range(0,0x10FFFF + 1):
         #         try:
         #             s.add(unicodedata.name(chr(c)))
         #         except ValueError:
         #             pass
         #     [''.join(k) for k in s.iter(prefix)]
 
         # But need to be timed and adds an extra dependency.
 
         slashpos = text.rfind('\\')
         # if text starts with slash
         if slashpos > -1:
             # PERF: It's important that we don't access self._unicode_names
             # until we're inside this if-block. _unicode_names is lazily
             # initialized, and it takes a user-noticeable amount of time to
             # initialize it, so we don't want to initialize it unless we're
             # actually going to use it.
             s = text[slashpos+1:]
             candidates = [x for x in self.unicode_names if x.startswith(s)]
             if candidates:
                 return s, candidates
             else:
                 return '', ()
 
         # if text does not start with slash
         else:
             return '', ()
 
     @property
     def unicode_names(self) -> List[str]:
         """List of names of unicode code points that can be completed.
 
         The list is lazily initialized on first access.
         """
         if self._unicode_names is None:
             names = []
             for c in range(0,0x10FFFF + 1):
                 try:
                     names.append(unicodedata.name(chr(c)))
                 except ValueError:
                     pass
             self._unicode_names = _unicode_name_compute(_UNICODE_RANGES)
 
         return self._unicode_names
 
 def _unicode_name_compute(ranges:List[Tuple[int,int]]) -> List[str]:
     names = []
     for start,stop in ranges:
         for c in range(start, stop) :
             try:
                 names.append(unicodedata.name(chr(c)))
             except ValueError:
                 pass
     return names