Uploaded Test files

This commit is contained in:
Batuhan Berk Başoğlu 2020-11-12 11:05:57 -05:00
parent f584ad9d97
commit 2e81cb7d99
16627 changed files with 2065359 additions and 102444 deletions

View file

@ -0,0 +1,43 @@
"""
Jedi is a static analysis tool for Python that is typically used in
IDEs/editors plugins. Jedi has a focus on autocompletion and goto
functionality. Other features include refactoring, code search and finding
references.
Jedi has a simple API to work with. There is a reference implementation as a
`VIM-Plugin <https://github.com/davidhalter/jedi-vim>`_. Autocompletion in your
REPL is also possible, IPython uses it natively and for the CPython REPL you
can install it. Jedi is well tested and bugs should be rare.
Here's a simple example of the autocompletion feature:
>>> import jedi
>>> source = '''
... import json
... json.lo'''
>>> script = jedi.Script(source, path='example.py')
>>> script
<Script: 'example.py' ...>
>>> completions = script.complete(3, len('json.lo'))
>>> completions
[<Completion: load>, <Completion: loads>]
>>> print(completions[0].complete)
ad
>>> print(completions[0].name)
load
"""
__version__ = '0.17.2'
from jedi.api import Script, Interpreter, set_debug_function, \
preload_module, names
from jedi import settings
from jedi.api.environment import find_virtualenvs, find_system_environments, \
get_default_environment, InvalidPythonEnvironment, create_environment, \
get_system_environment, InterpreterEnvironment
from jedi.api.project import Project, get_default_project
from jedi.api.exceptions import InternalError, RefactoringError
# Finally load the internal plugins. This is only internal.
from jedi.plugins import registry
del registry

View file

@ -0,0 +1,72 @@
import sys
from os.path import join, dirname, abspath, isdir
def _start_linter():
"""
This is a pre-alpha API. You're not supposed to use it at all, except for
testing. It will very likely change.
"""
import jedi
if '--debug' in sys.argv:
jedi.set_debug_function()
for path in sys.argv[2:]:
if path.startswith('--'):
continue
if isdir(path):
import fnmatch
import os
paths = []
for root, dirnames, filenames in os.walk(path):
for filename in fnmatch.filter(filenames, '*.py'):
paths.append(os.path.join(root, filename))
else:
paths = [path]
try:
for p in paths:
for error in jedi.Script(path=p)._analysis():
print(error)
except Exception:
if '--pdb' in sys.argv:
import traceback
traceback.print_exc()
import pdb
pdb.post_mortem()
else:
raise
def _complete():
import jedi
import pdb
if '-d' in sys.argv:
sys.argv.remove('-d')
jedi.set_debug_function()
try:
completions = jedi.Script(sys.argv[2]).complete()
for c in completions:
c.docstring()
c.type
except Exception as e:
print(repr(e))
pdb.post_mortem()
else:
print(completions)
if len(sys.argv) == 2 and sys.argv[1] == 'repl':
# don't want to use __main__ only for repl yet, maybe we want to use it for
# something else. So just use the keyword ``repl`` for now.
print(join(dirname(abspath(__file__)), 'api', 'replstartup.py'))
elif len(sys.argv) > 1 and sys.argv[1] == '_linter':
_start_linter()
elif len(sys.argv) > 1 and sys.argv[1] == '_complete':
_complete()
else:
print('Command not implemented: %s' % sys.argv[1])

View file

@ -0,0 +1,631 @@
"""
To ensure compatibility from Python ``2.7`` - ``3.x``, a module has been
created. Clearly there is huge need to use conforming syntax.
"""
from __future__ import print_function
import atexit
import errno
import functools
import sys
import os
import re
import pkgutil
import warnings
import subprocess
import weakref
try:
import importlib
except ImportError:
pass
from zipimport import zipimporter
from jedi.file_io import KnownContentFileIO, ZipFileIO
is_py3 = sys.version_info[0] >= 3
is_py35 = is_py3 and sys.version_info[1] >= 5
py_version = int(str(sys.version_info[0]) + str(sys.version_info[1]))
if sys.version_info[:2] < (3, 5):
"""
A super-minimal shim around listdir that behave like
scandir for the information we need.
"""
class _DirEntry:
def __init__(self, name, basepath):
self.name = name
self.basepath = basepath
def is_dir(self):
path_for_name = os.path.join(self.basepath, self.name)
return os.path.isdir(path_for_name)
def scandir(dir):
return [_DirEntry(name, dir) for name in os.listdir(dir)]
else:
from os import scandir
class DummyFile(object):
def __init__(self, loader, string):
self.loader = loader
self.string = string
def read(self):
return self.loader.get_source(self.string)
def close(self):
del self.loader
def find_module_py34(string, path=None, full_name=None, is_global_search=True):
spec = None
loader = None
for finder in sys.meta_path:
if is_global_search and finder != importlib.machinery.PathFinder:
p = None
else:
p = path
try:
find_spec = finder.find_spec
except AttributeError:
# These are old-school clases that still have a different API, just
# ignore those.
continue
spec = find_spec(string, p)
if spec is not None:
loader = spec.loader
if loader is None and not spec.has_location:
# This is a namespace package.
full_name = string if not path else full_name
implicit_ns_info = ImplicitNSInfo(full_name, spec.submodule_search_locations._path)
return implicit_ns_info, True
break
return find_module_py33(string, path, loader)
def find_module_py33(string, path=None, loader=None, full_name=None, is_global_search=True):
loader = loader or importlib.machinery.PathFinder.find_module(string, path)
if loader is None and path is None: # Fallback to find builtins
try:
with warnings.catch_warnings(record=True):
# Mute "DeprecationWarning: Use importlib.util.find_spec()
# instead." While we should replace that in the future, it's
# probably good to wait until we deprecate Python 3.3, since
# it was added in Python 3.4 and find_loader hasn't been
# removed in 3.6.
loader = importlib.find_loader(string)
except ValueError as e:
# See #491. Importlib might raise a ValueError, to avoid this, we
# just raise an ImportError to fix the issue.
raise ImportError("Originally " + repr(e))
if loader is None:
raise ImportError("Couldn't find a loader for {}".format(string))
return _from_loader(loader, string)
def _from_loader(loader, string):
try:
is_package_method = loader.is_package
except AttributeError:
is_package = False
else:
is_package = is_package_method(string)
try:
get_filename = loader.get_filename
except AttributeError:
return None, is_package
else:
module_path = cast_path(get_filename(string))
# To avoid unicode and read bytes, "overwrite" loader.get_source if
# possible.
try:
f = type(loader).get_source
except AttributeError:
raise ImportError("get_source was not defined on loader")
if is_py3 and f is not importlib.machinery.SourceFileLoader.get_source:
# Unfortunately we are reading unicode here, not bytes.
# It seems hard to get bytes, because the zip importer
# logic just unpacks the zip file and returns a file descriptor
# that we cannot as easily access. Therefore we just read it as
# a string in the cases where get_source was overwritten.
code = loader.get_source(string)
else:
code = _get_source(loader, string)
if code is None:
return None, is_package
if isinstance(loader, zipimporter):
return ZipFileIO(module_path, code, cast_path(loader.archive)), is_package
return KnownContentFileIO(module_path, code), is_package
def _get_source(loader, fullname):
"""
This method is here as a replacement for SourceLoader.get_source. That
method returns unicode, but we prefer bytes.
"""
path = loader.get_filename(fullname)
try:
return loader.get_data(path)
except OSError:
raise ImportError('source not available through get_data()',
name=fullname)
def find_module_pre_py3(string, path=None, full_name=None, is_global_search=True):
# This import is here, because in other places it will raise a
# DeprecationWarning.
import imp
try:
module_file, module_path, description = imp.find_module(string, path)
module_type = description[2]
is_package = module_type is imp.PKG_DIRECTORY
if is_package:
# In Python 2 directory package imports are returned as folder
# paths, not __init__.py paths.
p = os.path.join(module_path, '__init__.py')
try:
module_file = open(p)
module_path = p
except FileNotFoundError:
pass
elif module_type != imp.PY_SOURCE:
if module_file is not None:
module_file.close()
module_file = None
if module_file is None:
return None, is_package
with module_file:
code = module_file.read()
return KnownContentFileIO(cast_path(module_path), code), is_package
except ImportError:
pass
if path is None:
path = sys.path
for item in path:
loader = pkgutil.get_importer(item)
if loader:
loader = loader.find_module(string)
if loader is not None:
return _from_loader(loader, string)
raise ImportError("No module named {}".format(string))
find_module = find_module_py34 if is_py3 else find_module_pre_py3
find_module.__doc__ = """
Provides information about a module.
This function isolates the differences in importing libraries introduced with
python 3.3 on; it gets a module name and optionally a path. It will return a
tuple containin an open file for the module (if not builtin), the filename
or the name of the module if it is a builtin one and a boolean indicating
if the module is contained in a package.
"""
class ImplicitNSInfo(object):
"""Stores information returned from an implicit namespace spec"""
def __init__(self, name, paths):
self.name = name
self.paths = paths
if is_py3:
all_suffixes = importlib.machinery.all_suffixes
else:
def all_suffixes():
# Is deprecated and raises a warning in Python 3.6.
import imp
return [suffix for suffix, _, _ in imp.get_suffixes()]
# unicode function
try:
unicode = unicode
except NameError:
unicode = str
# re-raise function
if is_py3:
def reraise(exception, traceback):
raise exception.with_traceback(traceback)
else:
eval(compile("""
def reraise(exception, traceback):
raise exception, None, traceback
""", 'blub', 'exec'))
reraise.__doc__ = """
Re-raise `exception` with a `traceback` object.
Usage::
reraise(Exception, sys.exc_info()[2])
"""
def use_metaclass(meta, *bases):
""" Create a class with a metaclass. """
if not bases:
bases = (object,)
return meta("Py2CompatibilityMetaClass", bases, {})
try:
encoding = sys.stdout.encoding
if encoding is None:
encoding = 'utf-8'
except AttributeError:
encoding = 'ascii'
def u(string, errors='strict'):
"""Cast to unicode DAMMIT!
Written because Python2 repr always implicitly casts to a string, so we
have to cast back to a unicode (and we now that we always deal with valid
unicode, because we check that in the beginning).
"""
if isinstance(string, bytes):
return unicode(string, encoding='UTF-8', errors=errors)
return string
def cast_path(obj):
"""
Take a bytes or str path and cast it to unicode.
Apparently it is perfectly fine to pass both byte and unicode objects into
the sys.path. This probably means that byte paths are normal at other
places as well.
Since this just really complicates everything and Python 2.7 will be EOL
soon anyway, just go with always strings.
"""
return u(obj, errors='replace')
def force_unicode(obj):
# Intentionally don't mix those two up, because those two code paths might
# be different in the future (maybe windows?).
return cast_path(obj)
try:
import builtins # module name in python 3
except ImportError:
import __builtin__ as builtins # noqa: F401
import ast # noqa: F401
def literal_eval(string):
return ast.literal_eval(string)
try:
from itertools import zip_longest
except ImportError:
from itertools import izip_longest as zip_longest # Python 2 # noqa: F401
try:
FileNotFoundError = FileNotFoundError
except NameError:
FileNotFoundError = IOError
try:
IsADirectoryError = IsADirectoryError
except NameError:
IsADirectoryError = IOError
try:
PermissionError = PermissionError
except NameError:
PermissionError = IOError
try:
NotADirectoryError = NotADirectoryError
except NameError:
class NotADirectoryError(Exception):
# Don't implement this for Python 2 anymore.
pass
def no_unicode_pprint(dct):
"""
Python 2/3 dict __repr__ may be different, because of unicode differens
(with or without a `u` prefix). Normally in doctests we could use `pprint`
to sort dicts and check for equality, but here we have to write a separate
function to do that.
"""
import pprint
s = pprint.pformat(dct)
print(re.sub("u'", "'", s))
def utf8_repr(func):
"""
``__repr__`` methods in Python 2 don't allow unicode objects to be
returned. Therefore cast them to utf-8 bytes in this decorator.
"""
def wrapper(self):
result = func(self)
if isinstance(result, unicode):
return result.encode('utf-8')
else:
return result
if is_py3:
return func
else:
return wrapper
if is_py3:
import queue
else:
import Queue as queue # noqa: F401
try:
# Attempt to load the C implementation of pickle on Python 2 as it is way
# faster.
import cPickle as pickle
except ImportError:
import pickle
def pickle_load(file):
try:
if is_py3:
return pickle.load(file, encoding='bytes')
return pickle.load(file)
# Python on Windows don't throw EOF errors for pipes. So reraise them with
# the correct type, which is caught upwards.
except OSError:
if sys.platform == 'win32':
raise EOFError()
raise
def _python2_dct_keys_to_unicode(data):
"""
Python 2 stores object __dict__ entries as bytes, not unicode, correct it
here. Python 2 can deal with both, Python 3 expects unicode.
"""
if isinstance(data, tuple):
return tuple(_python2_dct_keys_to_unicode(x) for x in data)
elif isinstance(data, list):
return list(_python2_dct_keys_to_unicode(x) for x in data)
elif hasattr(data, '__dict__') and type(data.__dict__) == dict:
data.__dict__ = {unicode(k): v for k, v in data.__dict__.items()}
return data
def pickle_dump(data, file, protocol):
try:
if not is_py3:
data = _python2_dct_keys_to_unicode(data)
pickle.dump(data, file, protocol)
# On Python 3.3 flush throws sometimes an error even though the writing
# operation should be completed.
file.flush()
# Python on Windows don't throw EPIPE errors for pipes. So reraise them with
# the correct type and error number.
except OSError:
if sys.platform == 'win32':
raise IOError(errno.EPIPE, "Broken pipe")
raise
# Determine the highest protocol version compatible for a given list of Python
# versions.
def highest_pickle_protocol(python_versions):
protocol = 4
for version in python_versions:
if version[0] == 2:
# The minimum protocol version for the versions of Python that we
# support (2.7 and 3.3+) is 2.
return 2
if version[1] < 4:
protocol = 3
return protocol
try:
from inspect import Parameter
except ImportError:
class Parameter(object):
POSITIONAL_ONLY = object()
POSITIONAL_OR_KEYWORD = object()
VAR_POSITIONAL = object()
KEYWORD_ONLY = object()
VAR_KEYWORD = object()
class GeneralizedPopen(subprocess.Popen):
def __init__(self, *args, **kwargs):
if os.name == 'nt':
try:
# Was introduced in Python 3.7.
CREATE_NO_WINDOW = subprocess.CREATE_NO_WINDOW
except AttributeError:
CREATE_NO_WINDOW = 0x08000000
kwargs['creationflags'] = CREATE_NO_WINDOW
# The child process doesn't need file descriptors except 0, 1, 2.
# This is unix only.
kwargs['close_fds'] = 'posix' in sys.builtin_module_names
super(GeneralizedPopen, self).__init__(*args, **kwargs)
# shutil.which is not available on Python 2.7.
def which(cmd, mode=os.F_OK | os.X_OK, path=None):
"""Given a command, mode, and a PATH string, return the path which
conforms to the given mode on the PATH, or None if there is no such
file.
`mode` defaults to os.F_OK | os.X_OK. `path` defaults to the result
of os.environ.get("PATH"), or can be overridden with a custom search
path.
"""
# Check that a given file can be accessed with the correct mode.
# Additionally check that `file` is not a directory, as on Windows
# directories pass the os.access check.
def _access_check(fn, mode):
return (os.path.exists(fn) and os.access(fn, mode)
and not os.path.isdir(fn))
# If we're given a path with a directory part, look it up directly rather
# than referring to PATH directories. This includes checking relative to the
# current directory, e.g. ./script
if os.path.dirname(cmd):
if _access_check(cmd, mode):
return cmd
return None
if path is None:
path = os.environ.get("PATH", os.defpath)
if not path:
return None
path = path.split(os.pathsep)
if sys.platform == "win32":
# The current directory takes precedence on Windows.
if os.curdir not in path:
path.insert(0, os.curdir)
# PATHEXT is necessary to check on Windows.
pathext = os.environ.get("PATHEXT", "").split(os.pathsep)
# See if the given file matches any of the expected path extensions.
# This will allow us to short circuit when given "python.exe".
# If it does match, only test that one, otherwise we have to try
# others.
if any(cmd.lower().endswith(ext.lower()) for ext in pathext):
files = [cmd]
else:
files = [cmd + ext for ext in pathext]
else:
# On other platforms you don't have things like PATHEXT to tell you
# what file suffixes are executable, so just pass on cmd as-is.
files = [cmd]
seen = set()
for dir in path:
normdir = os.path.normcase(dir)
if normdir not in seen:
seen.add(normdir)
for thefile in files:
name = os.path.join(dir, thefile)
if _access_check(name, mode):
return name
return None
if not is_py3:
# Simplified backport of Python 3 weakref.finalize:
# https://github.com/python/cpython/blob/ded4737989316653469763230036b04513cb62b3/Lib/weakref.py#L502-L662
class finalize(object):
"""Class for finalization of weakrefable objects.
finalize(obj, func, *args, **kwargs) returns a callable finalizer
object which will be called when obj is garbage collected. The
first time the finalizer is called it evaluates func(*arg, **kwargs)
and returns the result. After this the finalizer is dead, and
calling it just returns None.
When the program exits any remaining finalizers will be run.
"""
# Finalizer objects don't have any state of their own.
# This ensures that they cannot be part of a ref-cycle.
__slots__ = ()
_registry = {}
def __init__(self, obj, func, *args, **kwargs):
info = functools.partial(func, *args, **kwargs)
info.weakref = weakref.ref(obj, self)
self._registry[self] = info
# To me it's an absolute mystery why in Python 2 we need _=None. It
# makes really no sense since it's never really called. Then again it
# might be called by Python 2.7 itself, but weakref.finalize is not
# documented in Python 2 and therefore shouldn't be randomly called.
# We never call this stuff with a parameter and therefore this
# parameter should not be needed. But it is. ~dave
def __call__(self, _=None):
"""Return func(*args, **kwargs) if alive."""
info = self._registry.pop(self, None)
if info:
return info()
@classmethod
def _exitfunc(cls):
if not cls._registry:
return
for finalizer in list(cls._registry):
try:
finalizer()
except Exception:
sys.excepthook(*sys.exc_info())
assert finalizer not in cls._registry
atexit.register(finalize._exitfunc)
weakref.finalize = finalize
if is_py3 and sys.version_info[1] > 5:
from inspect import unwrap
else:
# Only Python >=3.6 does properly limit the amount of unwraps. This is very
# relevant in the case of unittest.mock.patch.
# Below is the implementation of Python 3.7.
def unwrap(func, stop=None):
"""Get the object wrapped by *func*.
Follows the chain of :attr:`__wrapped__` attributes returning the last
object in the chain.
*stop* is an optional callback accepting an object in the wrapper chain
as its sole argument that allows the unwrapping to be terminated early if
the callback returns a true value. If the callback never returns a true
value, the last object in the chain is returned as usual. For example,
:func:`signature` uses this to stop unwrapping if any object in the
chain has a ``__signature__`` attribute defined.
:exc:`ValueError` is raised if a cycle is encountered.
"""
if stop is None:
def _is_wrapper(f):
return hasattr(f, '__wrapped__')
else:
def _is_wrapper(f):
return hasattr(f, '__wrapped__') and not stop(f)
f = func # remember the original func for error reporting
# Memoise by id to tolerate non-hashable objects, but store objects to
# ensure they aren't destroyed, which would allow their IDs to be reused.
memo = {id(f): f}
recursion_limit = sys.getrecursionlimit()
while _is_wrapper(func):
func = func.__wrapped__
id_func = id(func)
if (id_func in memo) or (len(memo) >= recursion_limit):
raise ValueError('wrapper loop when unwrapping {!r}'.format(f))
memo[id_func] = func
return func

View file

@ -0,0 +1,917 @@
"""
The API basically only provides one class. You can create a :class:`Script` and
use its methods.
Additionally you can add a debug function with :func:`set_debug_function`.
Alternatively, if you don't need a custom function and are happy with printing
debug messages to stdout, simply call :func:`set_debug_function` without
arguments.
"""
import os
import sys
import warnings
from functools import wraps
import parso
from parso.python import tree
from jedi._compatibility import force_unicode, cast_path, is_py3
from jedi.parser_utils import get_executable_nodes
from jedi import debug
from jedi import settings
from jedi import cache
from jedi.file_io import KnownContentFileIO
from jedi.api import classes
from jedi.api import interpreter
from jedi.api import helpers
from jedi.api.helpers import validate_line_column
from jedi.api.completion import Completion, search_in_module
from jedi.api.keywords import KeywordName
from jedi.api.environment import InterpreterEnvironment
from jedi.api.project import get_default_project, Project
from jedi.api.errors import parso_to_jedi_errors
from jedi.api import refactoring
from jedi.api.refactoring.extract import extract_function, extract_variable
from jedi.inference import InferenceState
from jedi.inference import imports
from jedi.inference.references import find_references
from jedi.inference.arguments import try_iter_content
from jedi.inference.helpers import infer_call_of_leaf
from jedi.inference.sys_path import transform_path_to_dotted
from jedi.inference.syntax_tree import tree_name_to_values
from jedi.inference.value import ModuleValue
from jedi.inference.base_value import ValueSet
from jedi.inference.value.iterable import unpack_tuple_to_dict
from jedi.inference.gradual.conversion import convert_names, convert_values
from jedi.inference.gradual.utils import load_proper_stub_module
from jedi.inference.utils import to_list
# Jedi uses lots and lots of recursion. By setting this a little bit higher, we
# can remove some "maximum recursion depth" errors.
sys.setrecursionlimit(3000)
def _no_python2_support(func):
# TODO remove when removing Python 2/3.5
@wraps(func)
def wrapper(self, *args, **kwargs):
if self._inference_state.grammar.version_info < (3, 6) or sys.version_info < (3, 6):
raise NotImplementedError(
"No support for refactorings/search on Python 2/3.5"
)
return func(self, *args, **kwargs)
return wrapper
class Script(object):
"""
A Script is the base for completions, goto or whatever you want to do with
Jedi. The counter part of this class is :class:`Interpreter`, which works
with actual dictionaries and can work with a REPL. This class
should be used when a user edits code in an editor.
You can either use the ``code`` parameter or ``path`` to read a file.
Usually you're going to want to use both of them (in an editor).
The Script's ``sys.path`` is very customizable:
- If `project` is provided with a ``sys_path``, that is going to be used.
- If `environment` is provided, its ``sys.path`` will be used
(see :func:`Environment.get_sys_path <jedi.api.environment.Environment.get_sys_path>`);
- Otherwise ``sys.path`` will match that of the default environment of
Jedi, which typically matches the sys path that was used at the time
when Jedi was imported.
Most methods have a ``line`` and a ``column`` parameter. Lines in Jedi are
always 1-based and columns are always zero based. To avoid repetition they
are not always documented. You can omit both line and column. Jedi will
then just do whatever action you are calling at the end of the file. If you
provide only the line, just will complete at the end of that line.
.. warning:: By default :attr:`jedi.settings.fast_parser` is enabled, which means
that parso reuses modules (i.e. they are not immutable). With this setting
Jedi is **not thread safe** and it is also not safe to use multiple
:class:`.Script` instances and its definitions at the same time.
If you are a normal plugin developer this should not be an issue. It is
an issue for people that do more complex stuff with Jedi.
This is purely a performance optimization and works pretty well for all
typical usages, however consider to turn the setting off if it causes
you problems. See also
`this discussion <https://github.com/davidhalter/jedi/issues/1240>`_.
:param code: The source code of the current file, separated by newlines.
:type code: str
:param line: Deprecated, please use it directly on e.g. ``.complete``
:type line: int
:param column: Deprecated, please use it directly on e.g. ``.complete``
:type column: int
:param path: The path of the file in the file system, or ``''`` if
it hasn't been saved yet.
:type path: str or None
:param encoding: Deprecated, cast to unicode yourself. The encoding of
``code``, if it is not a ``unicode`` object (default ``'utf-8'``).
:type encoding: str
:param sys_path: Deprecated, use the project parameter.
:type sys_path: typing.List[str]
:param Environment environment: Provide a predefined :ref:`Environment <environments>`
to work with a specific Python version or virtualenv.
:param Project project: Provide a :class:`.Project` to make sure finding
references works well, because the right folder is searched. There are
also ways to modify the sys path and other things.
"""
def __init__(self, code=None, line=None, column=None, path=None,
encoding=None, sys_path=None, environment=None,
project=None, source=None):
self._orig_path = path
# An empty path (also empty string) should always result in no path.
self.path = os.path.abspath(path) if path else None
if encoding is None:
encoding = 'utf-8'
else:
warnings.warn(
"Deprecated since version 0.17.0. You should cast to valid "
"unicode yourself, especially if you are not using utf-8.",
DeprecationWarning,
stacklevel=2
)
if line is not None:
warnings.warn(
"Providing the line is now done in the functions themselves "
"like `Script(...).complete(line, column)`",
DeprecationWarning,
stacklevel=2
)
if column is not None:
warnings.warn(
"Providing the column is now done in the functions themselves "
"like `Script(...).complete(line, column)`",
DeprecationWarning,
stacklevel=2
)
if source is not None:
code = source
warnings.warn(
"Use the code keyword argument instead.",
DeprecationWarning,
stacklevel=2
)
if code is None:
# TODO add a better warning than the traceback!
with open(path, 'rb') as f:
code = f.read()
if sys_path is not None and not is_py3:
sys_path = list(map(force_unicode, sys_path))
if project is None:
# Load the Python grammar of the current interpreter.
project = get_default_project(
os.path.dirname(self.path) if path else None
)
# TODO deprecate and remove sys_path from the Script API.
if sys_path is not None:
project._sys_path = sys_path
warnings.warn(
"Deprecated since version 0.17.0. Use the project API instead, "
"which means Script(project=Project(dir, sys_path=sys_path)) instead.",
DeprecationWarning,
stacklevel=2
)
self._inference_state = InferenceState(
project, environment=environment, script_path=self.path
)
debug.speed('init')
self._module_node, code = self._inference_state.parse_and_get_code(
code=code,
path=self.path,
encoding=encoding,
use_latest_grammar=path and path.endswith('.pyi'),
cache=False, # No disk cache, because the current script often changes.
diff_cache=settings.fast_parser,
cache_path=settings.cache_directory,
)
debug.speed('parsed')
self._code_lines = parso.split_lines(code, keepends=True)
self._code = code
self._pos = line, column
cache.clear_time_caches()
debug.reset_time()
# Cache the module, this is mostly useful for testing, since this shouldn't
# be called multiple times.
@cache.memoize_method
def _get_module(self):
names = None
is_package = False
if self.path is not None:
import_names, is_p = transform_path_to_dotted(
self._inference_state.get_sys_path(add_parent_paths=False),
self.path
)
if import_names is not None:
names = import_names
is_package = is_p
if self.path is None:
file_io = None
else:
file_io = KnownContentFileIO(cast_path(self.path), self._code)
if self.path is not None and self.path.endswith('.pyi'):
# We are in a stub file. Try to load the stub properly.
stub_module = load_proper_stub_module(
self._inference_state,
file_io,
names,
self._module_node
)
if stub_module is not None:
return stub_module
if names is None:
names = ('__main__',)
module = ModuleValue(
self._inference_state, self._module_node,
file_io=file_io,
string_names=names,
code_lines=self._code_lines,
is_package=is_package,
)
if names[0] not in ('builtins', '__builtin__', 'typing'):
# These modules are essential for Jedi, so don't overwrite them.
self._inference_state.module_cache.add(names, ValueSet([module]))
return module
def _get_module_context(self):
return self._get_module().as_context()
def __repr__(self):
return '<%s: %s %r>' % (
self.__class__.__name__,
repr(self._orig_path),
self._inference_state.environment,
)
@validate_line_column
def complete(self, line=None, column=None, **kwargs):
"""
Completes objects under the cursor.
Those objects contain information about the completions, more than just
names.
:param fuzzy: Default False. Will return fuzzy completions, which means
that e.g. ``ooa`` will match ``foobar``.
:return: Completion objects, sorted by name. Normal names appear
before "private" names that start with ``_`` and those appear
before magic methods and name mangled names that start with ``__``.
:rtype: list of :class:`.Completion`
"""
return self._complete(line, column, **kwargs)
def _complete(self, line, column, fuzzy=False): # Python 2...
with debug.increase_indent_cm('complete'):
completion = Completion(
self._inference_state, self._get_module_context(), self._code_lines,
(line, column), self.get_signatures, fuzzy=fuzzy,
)
return completion.complete()
def completions(self, fuzzy=False):
warnings.warn(
"Deprecated since version 0.16.0. Use Script(...).complete instead.",
DeprecationWarning,
stacklevel=2
)
return self.complete(*self._pos, fuzzy=fuzzy)
@validate_line_column
def infer(self, line=None, column=None, **kwargs):
"""
Return the definitions of under the cursor. It is basically a wrapper
around Jedi's type inference.
This method follows complicated paths and returns the end, not the
first definition. The big difference between :meth:`goto` and
:meth:`infer` is that :meth:`goto` doesn't
follow imports and statements. Multiple objects may be returned,
because depending on an option you can have two different versions of a
function.
:param only_stubs: Only return stubs for this method.
:param prefer_stubs: Prefer stubs to Python objects for this method.
:rtype: list of :class:`.Name`
"""
with debug.increase_indent_cm('infer'):
return self._infer(line, column, **kwargs)
def goto_definitions(self, **kwargs):
warnings.warn(
"Deprecated since version 0.16.0. Use Script(...).infer instead.",
DeprecationWarning,
stacklevel=2
)
return self.infer(*self._pos, **kwargs)
def _infer(self, line, column, only_stubs=False, prefer_stubs=False):
pos = line, column
leaf = self._module_node.get_name_of_position(pos)
if leaf is None:
leaf = self._module_node.get_leaf_for_position(pos)
if leaf is None or leaf.type == 'string':
return []
context = self._get_module_context().create_context(leaf)
values = helpers.infer(self._inference_state, context, leaf)
values = convert_values(
values,
only_stubs=only_stubs,
prefer_stubs=prefer_stubs,
)
defs = [classes.Name(self._inference_state, c.name) for c in values]
# The additional set here allows the definitions to become unique in an
# API sense. In the internals we want to separate more things than in
# the API.
return helpers.sorted_definitions(set(defs))
def goto_assignments(self, follow_imports=False, follow_builtin_imports=False, **kwargs):
warnings.warn(
"Deprecated since version 0.16.0. Use Script(...).goto instead.",
DeprecationWarning,
stacklevel=2
)
return self.goto(*self._pos,
follow_imports=follow_imports,
follow_builtin_imports=follow_builtin_imports,
**kwargs)
@validate_line_column
def goto(self, line=None, column=None, **kwargs):
"""
Goes to the name that defined the object under the cursor. Optionally
you can follow imports.
Multiple objects may be returned, depending on an if you can have two
different versions of a function.
:param follow_imports: The method will follow imports.
:param follow_builtin_imports: If ``follow_imports`` is True will try
to look up names in builtins (i.e. compiled or extension modules).
:param only_stubs: Only return stubs for this method.
:param prefer_stubs: Prefer stubs to Python objects for this method.
:rtype: list of :class:`.Name`
"""
with debug.increase_indent_cm('goto'):
return self._goto(line, column, **kwargs)
def _goto(self, line, column, follow_imports=False, follow_builtin_imports=False,
only_stubs=False, prefer_stubs=False):
tree_name = self._module_node.get_name_of_position((line, column))
if tree_name is None:
# Without a name we really just want to jump to the result e.g.
# executed by `foo()`, if we the cursor is after `)`.
return self.infer(line, column, only_stubs=only_stubs, prefer_stubs=prefer_stubs)
name = self._get_module_context().create_name(tree_name)
# Make it possible to goto the super class function/attribute
# definitions, when they are overwritten.
names = []
if name.tree_name.is_definition() and name.parent_context.is_class():
class_node = name.parent_context.tree_node
class_value = self._get_module_context().create_value(class_node)
mro = class_value.py__mro__()
next(mro) # Ignore the first entry, because it's the class itself.
for cls in mro:
names = cls.goto(tree_name.value)
if names:
break
if not names:
names = list(name.goto())
if follow_imports:
names = helpers.filter_follow_imports(names, follow_builtin_imports)
names = convert_names(
names,
only_stubs=only_stubs,
prefer_stubs=prefer_stubs,
)
defs = [classes.Name(self._inference_state, d) for d in set(names)]
# Avoid duplicates
return list(set(helpers.sorted_definitions(defs)))
@_no_python2_support
def search(self, string, **kwargs):
"""
Searches a name in the current file. For a description of how the
search string should look like, please have a look at
:meth:`.Project.search`.
:param bool all_scopes: Default False; searches not only for
definitions on the top level of a module level, but also in
functions and classes.
:yields: :class:`.Name`
"""
return self._search(string, **kwargs) # Python 2 ...
def _search(self, string, all_scopes=False):
return self._search_func(string, all_scopes=all_scopes)
@to_list
def _search_func(self, string, all_scopes=False, complete=False, fuzzy=False):
names = self._names(all_scopes=all_scopes)
wanted_type, wanted_names = helpers.split_search_string(string)
return search_in_module(
self._inference_state,
self._get_module_context(),
names=names,
wanted_type=wanted_type,
wanted_names=wanted_names,
complete=complete,
fuzzy=fuzzy,
)
def complete_search(self, string, **kwargs):
"""
Like :meth:`.Script.search`, but completes that string. If you want to
have all possible definitions in a file you can also provide an empty
string.
:param bool all_scopes: Default False; searches not only for
definitions on the top level of a module level, but also in
functions and classes.
:param fuzzy: Default False. Will return fuzzy completions, which means
that e.g. ``ooa`` will match ``foobar``.
:yields: :class:`.Completion`
"""
return self._search_func(string, complete=True, **kwargs)
@validate_line_column
def help(self, line=None, column=None):
"""
Used to display a help window to users. Uses :meth:`.Script.goto` and
returns additional definitions for keywords and operators.
Typically you will want to display :meth:`.BaseName.docstring` to the
user for all the returned definitions.
The additional definitions are ``Name(...).type == 'keyword'``.
These definitions do not have a lot of value apart from their docstring
attribute, which contains the output of Python's :func:`help` function.
:rtype: list of :class:`.Name`
"""
definitions = self.goto(line, column, follow_imports=True)
if definitions:
return definitions
leaf = self._module_node.get_leaf_for_position((line, column))
if leaf is not None and leaf.type in ('keyword', 'operator', 'error_leaf'):
def need_pydoc():
if leaf.value in ('(', ')', '[', ']'):
if leaf.parent.type == 'trailer':
return False
if leaf.parent.type == 'atom':
return False
grammar = self._inference_state.grammar
# This parso stuff is not public, but since I control it, this
# is fine :-) ~dave
reserved = grammar._pgen_grammar.reserved_syntax_strings.keys()
return leaf.value in reserved
if need_pydoc():
name = KeywordName(self._inference_state, leaf.value)
return [classes.Name(self._inference_state, name)]
return []
def usages(self, **kwargs):
warnings.warn(
"Deprecated since version 0.16.0. Use Script(...).get_references instead.",
DeprecationWarning,
stacklevel=2
)
return self.get_references(*self._pos, **kwargs)
@validate_line_column
def get_references(self, line=None, column=None, **kwargs):
"""
Lists all references of a variable in a project. Since this can be
quite hard to do for Jedi, if it is too complicated, Jedi will stop
searching.
:param include_builtins: Default ``True``. If ``False``, checks if a reference
is a builtin (e.g. ``sys``) and in that case does not return it.
:param scope: Default ``'project'``. If ``'file'``, include references in
the current module only.
:rtype: list of :class:`.Name`
"""
def _references(include_builtins=True, scope='project'):
if scope not in ('project', 'file'):
raise ValueError('Only the scopes "file" and "project" are allowed')
tree_name = self._module_node.get_name_of_position((line, column))
if tree_name is None:
# Must be syntax
return []
names = find_references(self._get_module_context(), tree_name, scope == 'file')
definitions = [classes.Name(self._inference_state, n) for n in names]
if not include_builtins or scope == 'file':
definitions = [d for d in definitions if not d.in_builtin_module()]
return helpers.sorted_definitions(definitions)
return _references(**kwargs)
def call_signatures(self):
warnings.warn(
"Deprecated since version 0.16.0. Use Script(...).get_signatures instead.",
DeprecationWarning,
stacklevel=2
)
return self.get_signatures(*self._pos)
@validate_line_column
def get_signatures(self, line=None, column=None):
"""
Return the function object of the call under the cursor.
E.g. if the cursor is here::
abs(# <-- cursor is here
This would return the ``abs`` function. On the other hand::
abs()# <-- cursor is here
This would return an empty list..
:rtype: list of :class:`.Signature`
"""
pos = line, column
call_details = helpers.get_signature_details(self._module_node, pos)
if call_details is None:
return []
context = self._get_module_context().create_context(call_details.bracket_leaf)
definitions = helpers.cache_signatures(
self._inference_state,
context,
call_details.bracket_leaf,
self._code_lines,
pos
)
debug.speed('func_call followed')
# TODO here we use stubs instead of the actual values. We should use
# the signatures from stubs, but the actual values, probably?!
return [classes.Signature(self._inference_state, signature, call_details)
for signature in definitions.get_signatures()]
@validate_line_column
def get_context(self, line=None, column=None):
"""
Returns the scope context under the cursor. This basically means the
function, class or module where the cursor is at.
:rtype: :class:`.Name`
"""
pos = (line, column)
leaf = self._module_node.get_leaf_for_position(pos, include_prefixes=True)
if leaf.start_pos > pos or leaf.type == 'endmarker':
previous_leaf = leaf.get_previous_leaf()
if previous_leaf is not None:
leaf = previous_leaf
module_context = self._get_module_context()
n = tree.search_ancestor(leaf, 'funcdef', 'classdef')
if n is not None and n.start_pos < pos <= n.children[-1].start_pos:
# This is a bit of a special case. The context of a function/class
# name/param/keyword is always it's parent context, not the
# function itself. Catch all the cases here where we are before the
# suite object, but still in the function.
context = module_context.create_value(n).as_context()
else:
context = module_context.create_context(leaf)
while context.name is None:
context = context.parent_context # comprehensions
definition = classes.Name(self._inference_state, context.name)
while definition.type != 'module':
name = definition._name # TODO private access
tree_name = name.tree_name
if tree_name is not None: # Happens with lambdas.
scope = tree_name.get_definition()
if scope.start_pos[1] < column:
break
definition = definition.parent()
return definition
def _analysis(self):
self._inference_state.is_analysis = True
self._inference_state.analysis_modules = [self._module_node]
module = self._get_module_context()
try:
for node in get_executable_nodes(self._module_node):
context = module.create_context(node)
if node.type in ('funcdef', 'classdef'):
# Resolve the decorators.
tree_name_to_values(self._inference_state, context, node.children[1])
elif isinstance(node, tree.Import):
import_names = set(node.get_defined_names())
if node.is_nested():
import_names |= set(path[-1] for path in node.get_paths())
for n in import_names:
imports.infer_import(context, n)
elif node.type == 'expr_stmt':
types = context.infer_node(node)
for testlist in node.children[:-1:2]:
# Iterate tuples.
unpack_tuple_to_dict(context, types, testlist)
else:
if node.type == 'name':
defs = self._inference_state.infer(context, node)
else:
defs = infer_call_of_leaf(context, node)
try_iter_content(defs)
self._inference_state.reset_recursion_limitations()
ana = [a for a in self._inference_state.analysis if self.path == a.path]
return sorted(set(ana), key=lambda x: x.line)
finally:
self._inference_state.is_analysis = False
def get_names(self, **kwargs):
"""
Returns names defined in the current file.
:param all_scopes: If True lists the names of all scopes instead of
only the module namespace.
:param definitions: If True lists the names that have been defined by a
class, function or a statement (``a = b`` returns ``a``).
:param references: If True lists all the names that are not listed by
``definitions=True``. E.g. ``a = b`` returns ``b``.
:rtype: list of :class:`.Name`
"""
names = self._names(**kwargs)
return [classes.Name(self._inference_state, n) for n in names]
def get_syntax_errors(self):
"""
Lists all syntax errors in the current file.
:rtype: list of :class:`.SyntaxError`
"""
return parso_to_jedi_errors(self._inference_state.grammar, self._module_node)
def _names(self, all_scopes=False, definitions=True, references=False):
# Set line/column to a random position, because they don't matter.
module_context = self._get_module_context()
defs = [
module_context.create_name(name)
for name in helpers.get_module_names(
self._module_node,
all_scopes=all_scopes,
definitions=definitions,
references=references,
)
]
return sorted(defs, key=lambda x: x.start_pos)
@_no_python2_support
def rename(self, line=None, column=None, **kwargs):
"""
Renames all references of the variable under the cursor.
:param new_name: The variable under the cursor will be renamed to this
string.
:raises: :exc:`.RefactoringError`
:rtype: :class:`.Refactoring`
"""
return self._rename(line, column, **kwargs)
def _rename(self, line, column, new_name): # Python 2...
definitions = self.get_references(line, column, include_builtins=False)
return refactoring.rename(self._inference_state, definitions, new_name)
@_no_python2_support
def extract_variable(self, line, column, **kwargs):
"""
Moves an expression to a new statemenet.
For example if you have the cursor on ``foo`` and provide a
``new_name`` called ``bar``::
foo = 3.1
x = int(foo + 1)
the code above will become::
foo = 3.1
bar = foo + 1
x = int(bar)
:param new_name: The expression under the cursor will be renamed to
this string.
:param int until_line: The the selection range ends at this line, when
omitted, Jedi will be clever and try to define the range itself.
:param int until_column: The the selection range ends at this column, when
omitted, Jedi will be clever and try to define the range itself.
:raises: :exc:`.RefactoringError`
:rtype: :class:`.Refactoring`
"""
return self._extract_variable(line, column, **kwargs) # Python 2...
@validate_line_column
def _extract_variable(self, line, column, new_name, until_line=None, until_column=None):
if until_line is None and until_column is None:
until_pos = None
else:
if until_line is None:
until_line = line
if until_column is None:
until_column = len(self._code_lines[until_line - 1])
until_pos = until_line, until_column
return extract_variable(
self._inference_state, self.path, self._module_node,
new_name, (line, column), until_pos
)
@_no_python2_support
def extract_function(self, line, column, **kwargs):
"""
Moves an expression to a new function.
For example if you have the cursor on ``foo`` and provide a
``new_name`` called ``bar``::
global_var = 3
def x():
foo = 3.1
x = int(foo + 1 + global_var)
the code above will become::
global_var = 3
def bar(foo):
return int(foo + 1 + global_var)
def x():
foo = 3.1
x = bar(foo)
:param new_name: The expression under the cursor will be replaced with
a function with this name.
:param int until_line: The the selection range ends at this line, when
omitted, Jedi will be clever and try to define the range itself.
:param int until_column: The the selection range ends at this column, when
omitted, Jedi will be clever and try to define the range itself.
:raises: :exc:`.RefactoringError`
:rtype: :class:`.Refactoring`
"""
return self._extract_function(line, column, **kwargs) # Python 2...
@validate_line_column
def _extract_function(self, line, column, new_name, until_line=None, until_column=None):
if until_line is None and until_column is None:
until_pos = None
else:
if until_line is None:
until_line = line
if until_column is None:
until_column = len(self._code_lines[until_line - 1])
until_pos = until_line, until_column
return extract_function(
self._inference_state, self.path, self._get_module_context(),
new_name, (line, column), until_pos
)
@_no_python2_support
def inline(self, line=None, column=None):
"""
Inlines a variable under the cursor. This is basically the opposite of
extracting a variable. For example with the cursor on bar::
foo = 3.1
bar = foo + 1
x = int(bar)
the code above will become::
foo = 3.1
x = int(foo + 1)
:raises: :exc:`.RefactoringError`
:rtype: :class:`.Refactoring`
"""
names = [d._name for d in self.get_references(line, column, include_builtins=True)]
return refactoring.inline(self._inference_state, names)
class Interpreter(Script):
"""
Jedi's API for Python REPLs.
Implements all of the methods that are present in :class:`.Script` as well.
In addition to completions that normal REPL completion does like
``str.upper``, Jedi also supports code completion based on static code
analysis. For example Jedi will complete ``str().upper``.
>>> from os.path import join
>>> namespace = locals()
>>> script = Interpreter('join("").up', [namespace])
>>> print(script.complete()[0].name)
upper
All keyword arguments are same as the arguments for :class:`.Script`.
:param str code: Code to parse.
:type namespaces: typing.List[dict]
:param namespaces: A list of namespace dictionaries such as the one
returned by :func:`globals` and :func:`locals`.
"""
_allow_descriptor_getattr_default = True
def __init__(self, code, namespaces, **kwds):
try:
namespaces = [dict(n) for n in namespaces]
except Exception:
raise TypeError("namespaces must be a non-empty list of dicts.")
environment = kwds.get('environment', None)
if environment is None:
environment = InterpreterEnvironment()
else:
if not isinstance(environment, InterpreterEnvironment):
raise TypeError("The environment needs to be an InterpreterEnvironment subclass.")
super(Interpreter, self).__init__(code, environment=environment,
project=Project(os.getcwd()), **kwds)
self.namespaces = namespaces
self._inference_state.allow_descriptor_getattr = self._allow_descriptor_getattr_default
@cache.memoize_method
def _get_module_context(self):
tree_module_value = ModuleValue(
self._inference_state, self._module_node,
file_io=KnownContentFileIO(self.path, self._code),
string_names=('__main__',),
code_lines=self._code_lines,
)
return interpreter.MixedModuleContext(
tree_module_value,
self.namespaces,
)
def names(source=None, path=None, encoding='utf-8', all_scopes=False,
definitions=True, references=False, environment=None):
warnings.warn(
"Deprecated since version 0.16.0. Use Script(...).get_names instead.",
DeprecationWarning,
stacklevel=2
)
return Script(source, path=path, encoding=encoding).get_names(
all_scopes=all_scopes,
definitions=definitions,
references=references,
)
def preload_module(*modules):
"""
Preloading modules tells Jedi to load a module now, instead of lazy parsing
of modules. This can be useful for IDEs, to control which modules to load
on startup.
:param modules: different module names, list of string.
"""
for m in modules:
s = "import %s as x; x." % m
Script(s, path=None).complete(1, len(s))
def set_debug_function(func_cb=debug.print_to_stdout, warnings=True,
notices=True, speed=True):
"""
Define a callback debug function to get all the debug messages.
If you don't specify any arguments, debug messages will be printed to stdout.
:param func_cb: The callback function for debug messages.
"""
debug.debug_function = func_cb
debug.enable_warning = warnings
debug.enable_notice = notices
debug.enable_speed = speed

View file

@ -0,0 +1,925 @@
"""
There are a couple of classes documented in here:
- :class:`.BaseName` as an abstact base class for almost everything.
- :class:`.Name` used in a lot of places
- :class:`.Completion` for completions
- :class:`.BaseSignature` as a base class for signatures
- :class:`.Signature` for :meth:`.Script.get_signatures` only
- :class:`.ParamName` used for parameters of signatures
- :class:`.Refactoring` for refactorings
- :class:`.SyntaxError` for :meth:`.Script.get_syntax_errors` only
These classes are the much biggest part of the API, because they contain
the interesting information about all operations.
"""
import re
import sys
import warnings
from parso.python.tree import search_ancestor
from jedi import settings
from jedi import debug
from jedi.inference.utils import unite
from jedi.cache import memoize_method
from jedi.inference import imports
from jedi.inference.imports import ImportName
from jedi.inference.compiled.mixed import MixedName
from jedi.inference.gradual.typeshed import StubModuleValue
from jedi.inference.gradual.conversion import convert_names, convert_values
from jedi.inference.base_value import ValueSet
from jedi.api.keywords import KeywordName
from jedi.api import completion_cache
from jedi.api.helpers import filter_follow_imports
def _sort_names_by_start_pos(names):
return sorted(names, key=lambda s: s.start_pos or (0, 0))
def defined_names(inference_state, context):
"""
List sub-definitions (e.g., methods in class).
:type scope: Scope
:rtype: list of Name
"""
filter = next(context.get_filters())
names = [name for name in filter.values()]
return [Name(inference_state, n) for n in _sort_names_by_start_pos(names)]
def _values_to_definitions(values):
return [Name(c.inference_state, c.name) for c in values]
class BaseName(object):
"""
The base class for all definitions, completions and signatures.
"""
_mapping = {
'posixpath': 'os.path',
'riscospath': 'os.path',
'ntpath': 'os.path',
'os2emxpath': 'os.path',
'macpath': 'os.path',
'genericpath': 'os.path',
'posix': 'os',
'_io': 'io',
'_functools': 'functools',
'_collections': 'collections',
'_socket': 'socket',
'_sqlite3': 'sqlite3',
'__builtin__': 'builtins',
}
_tuple_mapping = dict((tuple(k.split('.')), v) for (k, v) in {
'argparse._ActionsContainer': 'argparse.ArgumentParser',
}.items())
def __init__(self, inference_state, name):
self._inference_state = inference_state
self._name = name
"""
An instance of :class:`parso.python.tree.Name` subclass.
"""
self.is_keyword = isinstance(self._name, KeywordName)
@memoize_method
def _get_module_context(self):
# This can take a while to complete, because in the worst case of
# imports (consider `import a` completions), we need to load all
# modules starting with a first.
return self._name.get_root_context()
@property
def module_path(self):
"""
Shows the file path of a module. e.g. ``/usr/lib/python2.7/os.py``
:rtype: str or None
"""
module = self._get_module_context()
if module.is_stub() or not module.is_compiled():
# Compiled modules should not return a module path even if they
# have one.
return self._get_module_context().py__file__()
return None
@property
def name(self):
"""
Name of variable/function/class/module.
For example, for ``x = None`` it returns ``'x'``.
:rtype: str or None
"""
return self._name.get_public_name()
@property
def type(self):
"""
The type of the definition.
Here is an example of the value of this attribute. Let's consider
the following source. As what is in ``variable`` is unambiguous
to Jedi, :meth:`jedi.Script.infer` should return a list of
definition for ``sys``, ``f``, ``C`` and ``x``.
>>> from jedi._compatibility import no_unicode_pprint
>>> from jedi import Script
>>> source = '''
... import keyword
...
... class C:
... pass
...
... class D:
... pass
...
... x = D()
...
... def f():
... pass
...
... for variable in [keyword, f, C, x]:
... variable'''
>>> script = Script(source)
>>> defs = script.infer()
Before showing what is in ``defs``, let's sort it by :attr:`line`
so that it is easy to relate the result to the source code.
>>> defs = sorted(defs, key=lambda d: d.line)
>>> no_unicode_pprint(defs) # doctest: +NORMALIZE_WHITESPACE
[<Name full_name='keyword', description='module keyword'>,
<Name full_name='__main__.C', description='class C'>,
<Name full_name='__main__.D', description='instance D'>,
<Name full_name='__main__.f', description='def f'>]
Finally, here is what you can get from :attr:`type`:
>>> defs = [str(d.type) for d in defs] # It's unicode and in Py2 has u before it.
>>> defs[0]
'module'
>>> defs[1]
'class'
>>> defs[2]
'instance'
>>> defs[3]
'function'
Valid values for type are ``module``, ``class``, ``instance``, ``function``,
``param``, ``path``, ``keyword`` and ``statement``.
"""
tree_name = self._name.tree_name
resolve = False
if tree_name is not None:
# TODO move this to their respective names.
definition = tree_name.get_definition()
if definition is not None and definition.type == 'import_from' and \
tree_name.is_definition():
resolve = True
if isinstance(self._name, imports.SubModuleName) or resolve:
for value in self._name.infer():
return value.api_type
return self._name.api_type
@property
def module_name(self):
"""
The module name, a bit similar to what ``__name__`` is in a random
Python module.
>>> from jedi import Script
>>> source = 'import json'
>>> script = Script(source, path='example.py')
>>> d = script.infer()[0]
>>> print(d.module_name) # doctest: +ELLIPSIS
json
"""
return self._get_module_context().py__name__()
def in_builtin_module(self):
"""
Returns True, if this is a builtin module.
"""
value = self._get_module_context().get_value()
if isinstance(value, StubModuleValue):
return any(v.is_compiled() for v in value.non_stub_value_set)
return value.is_compiled()
@property
def line(self):
"""The line where the definition occurs (starting with 1)."""
start_pos = self._name.start_pos
if start_pos is None:
return None
return start_pos[0]
@property
def column(self):
"""The column where the definition occurs (starting with 0)."""
start_pos = self._name.start_pos
if start_pos is None:
return None
return start_pos[1]
def get_definition_start_position(self):
"""
The (row, column) of the start of the definition range. Rows start with
1, columns start with 0.
:rtype: Optional[Tuple[int, int]]
"""
if self._name.tree_name is None:
return None
definition = self._name.tree_name.get_definition()
if definition is None:
return self._name.start_pos
return definition.start_pos
def get_definition_end_position(self):
"""
The (row, column) of the end of the definition range. Rows start with
1, columns start with 0.
:rtype: Optional[Tuple[int, int]]
"""
if self._name.tree_name is None:
return None
definition = self._name.tree_name.get_definition()
if definition is None:
return self._name.tree_name.end_pos
if self.type in ("function", "class"):
last_leaf = definition.get_last_leaf()
if last_leaf.type == "newline":
return last_leaf.get_previous_leaf().end_pos
return last_leaf.end_pos
return definition.end_pos
def docstring(self, raw=False, fast=True):
r"""
Return a document string for this completion object.
Example:
>>> from jedi import Script
>>> source = '''\
... def f(a, b=1):
... "Document for function f."
... '''
>>> script = Script(source, path='example.py')
>>> doc = script.infer(1, len('def f'))[0].docstring()
>>> print(doc)
f(a, b=1)
<BLANKLINE>
Document for function f.
Notice that useful extra information is added to the actual
docstring, e.g. function signatures are prepended to their docstrings.
If you need the actual docstring, use ``raw=True`` instead.
>>> print(script.infer(1, len('def f'))[0].docstring(raw=True))
Document for function f.
:param fast: Don't follow imports that are only one level deep like
``import foo``, but follow ``from foo import bar``. This makes
sense for speed reasons. Completing `import a` is slow if you use
the ``foo.docstring(fast=False)`` on every object, because it
parses all libraries starting with ``a``.
"""
if isinstance(self._name, ImportName) and fast:
return ''
doc = self._get_docstring()
if raw:
return doc
signature_text = self._get_docstring_signature()
if signature_text and doc:
return signature_text + '\n\n' + doc
else:
return signature_text + doc
def _get_docstring(self):
return self._name.py__doc__()
def _get_docstring_signature(self):
return '\n'.join(
signature.to_string()
for signature in self._get_signatures(for_docstring=True)
)
@property
def description(self):
"""
A description of the :class:`.Name` object, which is heavily used
in testing. e.g. for ``isinstance`` it returns ``def isinstance``.
Example:
>>> from jedi._compatibility import no_unicode_pprint
>>> from jedi import Script
>>> source = '''
... def f():
... pass
...
... class C:
... pass
...
... variable = f if random.choice([0,1]) else C'''
>>> script = Script(source) # line is maximum by default
>>> defs = script.infer(column=3)
>>> defs = sorted(defs, key=lambda d: d.line)
>>> no_unicode_pprint(defs) # doctest: +NORMALIZE_WHITESPACE
[<Name full_name='__main__.f', description='def f'>,
<Name full_name='__main__.C', description='class C'>]
>>> str(defs[0].description) # strip literals in python2
'def f'
>>> str(defs[1].description)
'class C'
"""
typ = self.type
tree_name = self._name.tree_name
if typ == 'param':
return typ + ' ' + self._name.to_string()
if typ in ('function', 'class', 'module', 'instance') or tree_name is None:
if typ == 'function':
# For the description we want a short and a pythonic way.
typ = 'def'
return typ + ' ' + self._name.get_public_name()
definition = tree_name.get_definition(include_setitem=True) or tree_name
# Remove the prefix, because that's not what we want for get_code
# here.
txt = definition.get_code(include_prefix=False)
# Delete comments:
txt = re.sub(r'#[^\n]+\n', ' ', txt)
# Delete multi spaces/newlines
txt = re.sub(r'\s+', ' ', txt).strip()
return txt
@property
def full_name(self):
"""
Dot-separated path of this object.
It is in the form of ``<module>[.<submodule>[...]][.<object>]``.
It is useful when you want to look up Python manual of the
object at hand.
Example:
>>> from jedi import Script
>>> source = '''
... import os
... os.path.join'''
>>> script = Script(source, path='example.py')
>>> print(script.infer(3, len('os.path.join'))[0].full_name)
os.path.join
Notice that it returns ``'os.path.join'`` instead of (for example)
``'posixpath.join'``. This is not correct, since the modules name would
be ``<module 'posixpath' ...>```. However most users find the latter
more practical.
"""
if not self._name.is_value_name:
return None
names = self._name.get_qualified_names(include_module_names=True)
if names is None:
return None
names = list(names)
try:
names[0] = self._mapping[names[0]]
except KeyError:
pass
return '.'.join(names)
def is_stub(self):
"""
Returns True if the current name is defined in a stub file.
"""
if not self._name.is_value_name:
return False
return self._name.get_root_context().is_stub()
def is_side_effect(self):
"""
Checks if a name is defined as ``self.foo = 3``. In case of self, this
function would return False, for foo it would return True.
"""
tree_name = self._name.tree_name
if tree_name is None:
return False
return tree_name.is_definition() and tree_name.parent.type == 'trailer'
def goto(self, **kwargs):
"""
Like :meth:`.Script.goto` (also supports the same params), but does it
for the current name. This is typically useful if you are using
something like :meth:`.Script.get_names()`.
:param follow_imports: The goto call will follow imports.
:param follow_builtin_imports: If follow_imports is True will try to
look up names in builtins (i.e. compiled or extension modules).
:param only_stubs: Only return stubs for this goto call.
:param prefer_stubs: Prefer stubs to Python objects for this goto call.
:rtype: list of :class:`Name`
"""
with debug.increase_indent_cm('goto for %s' % self._name):
return self._goto(**kwargs)
def goto_assignments(self, **kwargs): # Python 2...
warnings.warn(
"Deprecated since version 0.16.0. Use .goto.",
DeprecationWarning,
stacklevel=2
)
return self.goto(**kwargs)
def _goto(self, follow_imports=False, follow_builtin_imports=False,
only_stubs=False, prefer_stubs=False):
if not self._name.is_value_name:
return []
names = self._name.goto()
if follow_imports:
names = filter_follow_imports(names, follow_builtin_imports)
names = convert_names(
names,
only_stubs=only_stubs,
prefer_stubs=prefer_stubs,
)
return [self if n == self._name else Name(self._inference_state, n)
for n in names]
def infer(self, **kwargs): # Python 2...
"""
Like :meth:`.Script.infer`, it can be useful to understand which type
the current name has.
Return the actual definitions. I strongly recommend not using it for
your completions, because it might slow down |jedi|. If you want to
read only a few objects (<=20), it might be useful, especially to get
the original docstrings. The basic problem of this function is that it
follows all results. This means with 1000 completions (e.g. numpy),
it's just very, very slow.
:param only_stubs: Only return stubs for this goto call.
:param prefer_stubs: Prefer stubs to Python objects for this type
inference call.
:rtype: list of :class:`Name`
"""
with debug.increase_indent_cm('infer for %s' % self._name):
return self._infer(**kwargs)
def _infer(self, only_stubs=False, prefer_stubs=False):
assert not (only_stubs and prefer_stubs)
if not self._name.is_value_name:
return []
# First we need to make sure that we have stub names (if possible) that
# we can follow. If we don't do that, we can end up with the inferred
# results of Python objects instead of stubs.
names = convert_names([self._name], prefer_stubs=True)
values = convert_values(
ValueSet.from_sets(n.infer() for n in names),
only_stubs=only_stubs,
prefer_stubs=prefer_stubs,
)
resulting_names = [c.name for c in values]
return [self if n == self._name else Name(self._inference_state, n)
for n in resulting_names]
@property
@memoize_method
def params(self):
warnings.warn(
"Deprecated since version 0.16.0. Use get_signatures()[...].params",
DeprecationWarning,
stacklevel=2
)
# Only return the first one. There might be multiple one, especially
# with overloading.
for signature in self._get_signatures():
return [
Name(self._inference_state, n)
for n in signature.get_param_names(resolve_stars=True)
]
if self.type == 'function' or self.type == 'class':
# Fallback, if no signatures were defined (which is probably by
# itself a bug).
return []
raise AttributeError('There are no params defined on this.')
def parent(self):
"""
Returns the parent scope of this identifier.
:rtype: Name
"""
if not self._name.is_value_name:
return None
if self.type in ('function', 'class', 'param') and self._name.tree_name is not None:
# Since the parent_context doesn't really match what the user
# thinks of that the parent is here, we do these cases separately.
# The reason for this is the following:
# - class: Nested classes parent_context is always the
# parent_context of the most outer one.
# - function: Functions in classes have the module as
# parent_context.
# - param: The parent_context of a param is not its function but
# e.g. the outer class or module.
cls_or_func_node = self._name.tree_name.get_definition()
parent = search_ancestor(cls_or_func_node, 'funcdef', 'classdef', 'file_input')
context = self._get_module_context().create_value(parent).as_context()
else:
context = self._name.parent_context
if context is None:
return None
while context.name is None:
# Happens for comprehension contexts
context = context.parent_context
return Name(self._inference_state, context.name)
def __repr__(self):
return "<%s %sname=%r, description=%r>" % (
self.__class__.__name__,
'full_' if self.full_name else '',
self.full_name or self.name,
self.description,
)
def get_line_code(self, before=0, after=0):
"""
Returns the line of code where this object was defined.
:param before: Add n lines before the current line to the output.
:param after: Add n lines after the current line to the output.
:return str: Returns the line(s) of code or an empty string if it's a
builtin.
"""
if not self._name.is_value_name:
return ''
lines = self._name.get_root_context().code_lines
if lines is None:
# Probably a builtin module, just ignore in that case.
return ''
index = self._name.start_pos[0] - 1
start_index = max(index - before, 0)
return ''.join(lines[start_index:index + after + 1])
def _get_signatures(self, for_docstring=False):
if for_docstring and self._name.api_type == 'statement' and not self.is_stub():
# For docstrings we don't resolve signatures if they are simple
# statements and not stubs. This is a speed optimization.
return []
if isinstance(self._name, MixedName):
# While this would eventually happen anyway, it's basically just a
# shortcut to not infer anything tree related, because it's really
# not necessary.
return self._name.infer_compiled_value().get_signatures()
names = convert_names([self._name], prefer_stubs=True)
return [sig for name in names for sig in name.infer().get_signatures()]
def get_signatures(self):
"""
Returns all potential signatures for a function or a class. Multiple
signatures are typical if you use Python stubs with ``@overload``.
:rtype: list of :class:`BaseSignature`
"""
return [
BaseSignature(self._inference_state, s)
for s in self._get_signatures()
]
def execute(self):
"""
Uses type inference to "execute" this identifier and returns the
executed objects.
:rtype: list of :class:`Name`
"""
return _values_to_definitions(self._name.infer().execute_with_values())
def get_type_hint(self):
"""
Returns type hints like ``Iterable[int]`` or ``Union[int, str]``.
This method might be quite slow, especially for functions. The problem
is finding executions for those functions to return something like
``Callable[[int, str], str]``.
:rtype: str
"""
return self._name.infer().get_type_hint()
class Completion(BaseName):
"""
``Completion`` objects are returned from :meth:`.Script.complete`. They
provide additional information about a completion.
"""
def __init__(self, inference_state, name, stack, like_name_length,
is_fuzzy, cached_name=None):
super(Completion, self).__init__(inference_state, name)
self._like_name_length = like_name_length
self._stack = stack
self._is_fuzzy = is_fuzzy
self._cached_name = cached_name
# Completion objects with the same Completion name (which means
# duplicate items in the completion)
self._same_name_completions = []
def _complete(self, like_name):
append = ''
if settings.add_bracket_after_function \
and self.type == 'function':
append = '('
name = self._name.get_public_name()
if like_name:
name = name[self._like_name_length:]
return name + append
@property
def complete(self):
"""
Only works with non-fuzzy completions. Returns None if fuzzy
completions are used.
Return the rest of the word, e.g. completing ``isinstance``::
isinstan# <-- Cursor is here
would return the string 'ce'. It also adds additional stuff, depending
on your ``settings.py``.
Assuming the following function definition::
def foo(param=0):
pass
completing ``foo(par`` would give a ``Completion`` which ``complete``
would be ``am=``.
"""
if self._is_fuzzy:
return None
return self._complete(True)
@property
def name_with_symbols(self):
"""
Similar to :attr:`.name`, but like :attr:`.name` returns also the
symbols, for example assuming the following function definition::
def foo(param=0):
pass
completing ``foo(`` would give a ``Completion`` which
``name_with_symbols`` would be "param=".
"""
return self._complete(False)
def docstring(self, raw=False, fast=True):
"""
Documented under :meth:`BaseName.docstring`.
"""
if self._like_name_length >= 3:
# In this case we can just resolve the like name, because we
# wouldn't load like > 100 Python modules anymore.
fast = False
return super(Completion, self).docstring(raw=raw, fast=fast)
def _get_docstring(self):
if self._cached_name is not None:
return completion_cache.get_docstring(
self._cached_name,
self._name.get_public_name(),
lambda: self._get_cache()
)
return super(Completion, self)._get_docstring()
def _get_docstring_signature(self):
if self._cached_name is not None:
return completion_cache.get_docstring_signature(
self._cached_name,
self._name.get_public_name(),
lambda: self._get_cache()
)
return super(Completion, self)._get_docstring_signature()
def _get_cache(self):
return (
super(Completion, self).type,
super(Completion, self)._get_docstring_signature(),
super(Completion, self)._get_docstring(),
)
@property
def type(self):
"""
Documented under :meth:`BaseName.type`.
"""
# Purely a speed optimization.
if self._cached_name is not None:
return completion_cache.get_type(
self._cached_name,
self._name.get_public_name(),
lambda: self._get_cache()
)
return super(Completion, self).type
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, self._name.get_public_name())
class Name(BaseName):
"""
*Name* objects are returned from many different APIs including
:meth:`.Script.goto` or :meth:`.Script.infer`.
"""
def __init__(self, inference_state, definition):
super(Name, self).__init__(inference_state, definition)
@property
def desc_with_module(self):
warnings.warn(
"Deprecated since version 0.17.0. No replacement for now, maybe .full_name helps",
DeprecationWarning,
stacklevel=2
)
return "%s:%s" % (self.module_name, self.description)
@memoize_method
def defined_names(self):
"""
List sub-definitions (e.g., methods in class).
:rtype: list of :class:`Name`
"""
defs = self._name.infer()
return sorted(
unite(defined_names(self._inference_state, d.as_context()) for d in defs),
key=lambda s: s._name.start_pos or (0, 0)
)
def is_definition(self):
"""
Returns True, if defined as a name in a statement, function or class.
Returns False, if it's a reference to such a definition.
"""
if self._name.tree_name is None:
return True
else:
return self._name.tree_name.is_definition()
def __eq__(self, other):
return self._name.start_pos == other._name.start_pos \
and self.module_path == other.module_path \
and self.name == other.name \
and self._inference_state == other._inference_state
def __ne__(self, other):
return not self.__eq__(other)
def __hash__(self):
return hash((self._name.start_pos, self.module_path, self.name, self._inference_state))
class BaseSignature(Name):
"""
These signatures are returned by :meth:`BaseName.get_signatures`
calls.
"""
def __init__(self, inference_state, signature):
super(BaseSignature, self).__init__(inference_state, signature.name)
self._signature = signature
@property
def params(self):
"""
Returns definitions for all parameters that a signature defines.
This includes stuff like ``*args`` and ``**kwargs``.
:rtype: list of :class:`.ParamName`
"""
return [ParamName(self._inference_state, n)
for n in self._signature.get_param_names(resolve_stars=True)]
def to_string(self):
"""
Returns a text representation of the signature. This could for example
look like ``foo(bar, baz: int, **kwargs)``.
:rtype: str
"""
return self._signature.to_string()
class Signature(BaseSignature):
"""
A full signature object is the return value of
:meth:`.Script.get_signatures`.
"""
def __init__(self, inference_state, signature, call_details):
super(Signature, self).__init__(inference_state, signature)
self._call_details = call_details
self._signature = signature
@property
def index(self):
"""
Returns the param index of the current cursor position.
Returns None if the index cannot be found in the curent call.
:rtype: int
"""
return self._call_details.calculate_index(
self._signature.get_param_names(resolve_stars=True)
)
@property
def bracket_start(self):
"""
Returns a line/column tuple of the bracket that is responsible for the
last function call. The first line is 1 and the first column 0.
:rtype: int, int
"""
return self._call_details.bracket_leaf.start_pos
def __repr__(self):
return '<%s: index=%r %s>' % (
type(self).__name__,
self.index,
self._signature.to_string(),
)
class ParamName(Name):
def infer_default(self):
"""
Returns default values like the ``1`` of ``def foo(x=1):``.
:rtype: list of :class:`.Name`
"""
return _values_to_definitions(self._name.infer_default())
def infer_annotation(self, **kwargs):
"""
:param execute_annotation: Default True; If False, values are not
executed and classes are returned instead of instances.
:rtype: list of :class:`.Name`
"""
return _values_to_definitions(self._name.infer_annotation(ignore_stars=True, **kwargs))
def to_string(self):
"""
Returns a simple representation of a param, like
``f: Callable[..., Any]``.
:rtype: str
"""
return self._name.to_string()
@property
def kind(self):
"""
Returns an enum instance of :mod:`inspect`'s ``Parameter`` enum.
:rtype: :py:attr:`inspect.Parameter.kind`
"""
if sys.version_info < (3, 5):
raise NotImplementedError(
'Python 2 is end-of-life, the new feature is not available for it'
)
return self._name.get_kind()

View file

@ -0,0 +1,665 @@
import re
from textwrap import dedent
from parso.python.token import PythonTokenTypes
from parso.python import tree
from parso.tree import search_ancestor, Leaf
from parso import split_lines
from jedi._compatibility import Parameter
from jedi import debug
from jedi import settings
from jedi.api import classes
from jedi.api import helpers
from jedi.api import keywords
from jedi.api.strings import complete_dict
from jedi.api.file_name import complete_file_name
from jedi.inference import imports
from jedi.inference.base_value import ValueSet
from jedi.inference.helpers import infer_call_of_leaf, parse_dotted_names
from jedi.inference.context import get_global_filters
from jedi.inference.value import TreeInstance, ModuleValue
from jedi.inference.names import ParamNameWrapper, SubModuleName
from jedi.inference.gradual.conversion import convert_values, convert_names
from jedi.parser_utils import cut_value_at_position
from jedi.plugins import plugin_manager
class ParamNameWithEquals(ParamNameWrapper):
def get_public_name(self):
return self.string_name + '='
def _get_signature_param_names(signatures, positional_count, used_kwargs):
# Add named params
for call_sig in signatures:
for i, p in enumerate(call_sig.params):
# Allow protected access, because it's a public API.
# TODO reconsider with Python 2 drop
kind = p._name.get_kind()
if i < positional_count and kind == Parameter.POSITIONAL_OR_KEYWORD:
continue
if kind in (Parameter.POSITIONAL_OR_KEYWORD, Parameter.KEYWORD_ONLY) \
and p.name not in used_kwargs:
yield ParamNameWithEquals(p._name)
def _must_be_kwarg(signatures, positional_count, used_kwargs):
if used_kwargs:
return True
must_be_kwarg = True
for signature in signatures:
for i, p in enumerate(signature.params):
# TODO reconsider with Python 2 drop
kind = p._name.get_kind()
if kind is Parameter.VAR_POSITIONAL:
# In case there were not already kwargs, the next param can
# always be a normal argument.
return False
if i >= positional_count and kind in (Parameter.POSITIONAL_OR_KEYWORD,
Parameter.POSITIONAL_ONLY):
must_be_kwarg = False
break
if not must_be_kwarg:
break
return must_be_kwarg
def filter_names(inference_state, completion_names, stack, like_name, fuzzy, cached_name):
comp_dct = set()
if settings.case_insensitive_completion:
like_name = like_name.lower()
for name in completion_names:
string = name.string_name
if settings.case_insensitive_completion:
string = string.lower()
if helpers.match(string, like_name, fuzzy=fuzzy):
new = classes.Completion(
inference_state,
name,
stack,
len(like_name),
is_fuzzy=fuzzy,
cached_name=cached_name,
)
k = (new.name, new.complete) # key
if k not in comp_dct:
comp_dct.add(k)
tree_name = name.tree_name
if tree_name is not None:
definition = tree_name.get_definition()
if definition is not None and definition.type == 'del_stmt':
continue
yield new
def _remove_duplicates(completions, other_completions):
names = {d.name for d in other_completions}
return [c for c in completions if c.name not in names]
def get_user_context(module_context, position):
"""
Returns the scope in which the user resides. This includes flows.
"""
leaf = module_context.tree_node.get_leaf_for_position(position, include_prefixes=True)
return module_context.create_context(leaf)
def get_flow_scope_node(module_node, position):
node = module_node.get_leaf_for_position(position, include_prefixes=True)
while not isinstance(node, (tree.Scope, tree.Flow)):
node = node.parent
return node
@plugin_manager.decorate()
def complete_param_names(context, function_name, decorator_nodes):
# Basically there's no way to do param completion. The plugins are
# responsible for this.
return []
class Completion:
def __init__(self, inference_state, module_context, code_lines, position,
signatures_callback, fuzzy=False):
self._inference_state = inference_state
self._module_context = module_context
self._module_node = module_context.tree_node
self._code_lines = code_lines
# The first step of completions is to get the name
self._like_name = helpers.get_on_completion_name(self._module_node, code_lines, position)
# The actual cursor position is not what we need to calculate
# everything. We want the start of the name we're on.
self._original_position = position
self._signatures_callback = signatures_callback
self._fuzzy = fuzzy
def complete(self):
leaf = self._module_node.get_leaf_for_position(
self._original_position,
include_prefixes=True
)
string, start_leaf, quote = _extract_string_while_in_string(leaf, self._original_position)
prefixed_completions = complete_dict(
self._module_context,
self._code_lines,
start_leaf or leaf,
self._original_position,
None if string is None else quote + string,
fuzzy=self._fuzzy,
)
if string is not None and not prefixed_completions:
prefixed_completions = list(complete_file_name(
self._inference_state, self._module_context, start_leaf, quote, string,
self._like_name, self._signatures_callback,
self._code_lines, self._original_position,
self._fuzzy
))
if string is not None:
if not prefixed_completions and '\n' in string:
# Complete only multi line strings
prefixed_completions = self._complete_in_string(start_leaf, string)
return prefixed_completions
cached_name, completion_names = self._complete_python(leaf)
completions = list(filter_names(self._inference_state, completion_names,
self.stack, self._like_name,
self._fuzzy, cached_name=cached_name))
return (
# Removing duplicates mostly to remove False/True/None duplicates.
_remove_duplicates(prefixed_completions, completions)
+ sorted(completions, key=lambda x: (x.name.startswith('__'),
x.name.startswith('_'),
x.name.lower()))
)
def _complete_python(self, leaf):
"""
Analyzes the current context of a completion and decides what to
return.
Technically this works by generating a parser stack and analysing the
current stack for possible grammar nodes.
Possible enhancements:
- global/nonlocal search global
- yield from / raise from <- could be only exceptions/generators
- In args: */**: no completion
- In params (also lambda): no completion before =
"""
grammar = self._inference_state.grammar
self.stack = stack = None
self._position = (
self._original_position[0],
self._original_position[1] - len(self._like_name)
)
cached_name = None
try:
self.stack = stack = helpers.get_stack_at_position(
grammar, self._code_lines, leaf, self._position
)
except helpers.OnErrorLeaf as e:
value = e.error_leaf.value
if value == '.':
# After ErrorLeaf's that are dots, we will not do any
# completions since this probably just confuses the user.
return cached_name, []
# If we don't have a value, just use global completion.
return cached_name, self._complete_global_scope()
allowed_transitions = \
list(stack._allowed_transition_names_and_token_types())
if 'if' in allowed_transitions:
leaf = self._module_node.get_leaf_for_position(self._position, include_prefixes=True)
previous_leaf = leaf.get_previous_leaf()
indent = self._position[1]
if not (leaf.start_pos <= self._position <= leaf.end_pos):
indent = leaf.start_pos[1]
if previous_leaf is not None:
stmt = previous_leaf
while True:
stmt = search_ancestor(
stmt, 'if_stmt', 'for_stmt', 'while_stmt', 'try_stmt',
'error_node',
)
if stmt is None:
break
type_ = stmt.type
if type_ == 'error_node':
first = stmt.children[0]
if isinstance(first, Leaf):
type_ = first.value + '_stmt'
# Compare indents
if stmt.start_pos[1] == indent:
if type_ == 'if_stmt':
allowed_transitions += ['elif', 'else']
elif type_ == 'try_stmt':
allowed_transitions += ['except', 'finally', 'else']
elif type_ == 'for_stmt':
allowed_transitions.append('else')
completion_names = []
kwargs_only = False
if any(t in allowed_transitions for t in (PythonTokenTypes.NAME,
PythonTokenTypes.INDENT)):
# This means that we actually have to do type inference.
nonterminals = [stack_node.nonterminal for stack_node in stack]
nodes = _gather_nodes(stack)
if nodes and nodes[-1] in ('as', 'def', 'class'):
# No completions for ``with x as foo`` and ``import x as foo``.
# Also true for defining names as a class or function.
return cached_name, list(self._complete_inherited(is_function=True))
elif "import_stmt" in nonterminals:
level, names = parse_dotted_names(nodes, "import_from" in nonterminals)
only_modules = not ("import_from" in nonterminals and 'import' in nodes)
completion_names += self._get_importer_names(
names,
level,
only_modules=only_modules,
)
elif nonterminals[-1] in ('trailer', 'dotted_name') and nodes[-1] == '.':
dot = self._module_node.get_leaf_for_position(self._position)
cached_name, n = self._complete_trailer(dot.get_previous_leaf())
completion_names += n
elif self._is_parameter_completion():
completion_names += self._complete_params(leaf)
else:
# Apparently this looks like it's good enough to filter most cases
# so that signature completions don't randomly appear.
# To understand why this works, three things are important:
# 1. trailer with a `,` in it is either a subscript or an arglist.
# 2. If there's no `,`, it's at the start and only signatures start
# with `(`. Other trailers could start with `.` or `[`.
# 3. Decorators are very primitive and have an optional `(` with
# optional arglist in them.
if nodes[-1] in ['(', ','] \
and nonterminals[-1] in ('trailer', 'arglist', 'decorator'):
signatures = self._signatures_callback(*self._position)
if signatures:
call_details = signatures[0]._call_details
used_kwargs = list(call_details.iter_used_keyword_arguments())
positional_count = call_details.count_positional_arguments()
completion_names += _get_signature_param_names(
signatures,
positional_count,
used_kwargs,
)
kwargs_only = _must_be_kwarg(signatures, positional_count, used_kwargs)
if not kwargs_only:
completion_names += self._complete_global_scope()
completion_names += self._complete_inherited(is_function=False)
if not kwargs_only:
current_line = self._code_lines[self._position[0] - 1][:self._position[1]]
completion_names += self._complete_keywords(
allowed_transitions,
only_values=not (not current_line or current_line[-1] in ' \t.;'
and current_line[-3:] != '...')
)
return cached_name, completion_names
def _is_parameter_completion(self):
tos = self.stack[-1]
if tos.nonterminal == 'lambdef' and len(tos.nodes) == 1:
# We are at the position `lambda `, where basically the next node
# is a param.
return True
if tos.nonterminal in 'parameters':
# Basically we are at the position `foo(`, there's nothing there
# yet, so we have no `typedargslist`.
return True
# var args is for lambdas and typed args for normal functions
return tos.nonterminal in ('typedargslist', 'varargslist') and tos.nodes[-1] == ','
def _complete_params(self, leaf):
stack_node = self.stack[-2]
if stack_node.nonterminal == 'parameters':
stack_node = self.stack[-3]
if stack_node.nonterminal == 'funcdef':
context = get_user_context(self._module_context, self._position)
node = search_ancestor(leaf, 'error_node', 'funcdef')
if node is not None:
if node.type == 'error_node':
n = node.children[0]
if n.type == 'decorators':
decorators = n.children
elif n.type == 'decorator':
decorators = [n]
else:
decorators = []
else:
decorators = node.get_decorators()
function_name = stack_node.nodes[1]
return complete_param_names(context, function_name.value, decorators)
return []
def _complete_keywords(self, allowed_transitions, only_values):
for k in allowed_transitions:
if isinstance(k, str) and k.isalpha():
if not only_values or k in ('True', 'False', 'None'):
yield keywords.KeywordName(self._inference_state, k)
def _complete_global_scope(self):
context = get_user_context(self._module_context, self._position)
debug.dbg('global completion scope: %s', context)
flow_scope_node = get_flow_scope_node(self._module_node, self._position)
filters = get_global_filters(
context,
self._position,
flow_scope_node
)
completion_names = []
for filter in filters:
completion_names += filter.values()
return completion_names
def _complete_trailer(self, previous_leaf):
inferred_context = self._module_context.create_context(previous_leaf)
values = infer_call_of_leaf(inferred_context, previous_leaf)
debug.dbg('trailer completion values: %s', values, color='MAGENTA')
# The cached name simply exists to make speed optimizations for certain
# modules.
cached_name = None
if len(values) == 1:
v, = values
if v.is_module():
if len(v.string_names) == 1:
module_name = v.string_names[0]
if module_name in ('numpy', 'tensorflow', 'matplotlib', 'pandas'):
cached_name = module_name
return cached_name, self._complete_trailer_for_values(values)
def _complete_trailer_for_values(self, values):
user_context = get_user_context(self._module_context, self._position)
return complete_trailer(user_context, values)
def _get_importer_names(self, names, level=0, only_modules=True):
names = [n.value for n in names]
i = imports.Importer(self._inference_state, names, self._module_context, level)
return i.completion_names(self._inference_state, only_modules=only_modules)
def _complete_inherited(self, is_function=True):
"""
Autocomplete inherited methods when overriding in child class.
"""
leaf = self._module_node.get_leaf_for_position(self._position, include_prefixes=True)
cls = tree.search_ancestor(leaf, 'classdef')
if cls is None:
return
# Complete the methods that are defined in the super classes.
class_value = self._module_context.create_value(cls)
if cls.start_pos[1] >= leaf.start_pos[1]:
return
filters = class_value.get_filters(is_instance=True)
# The first dict is the dictionary of class itself.
next(filters)
for filter in filters:
for name in filter.values():
# TODO we should probably check here for properties
if (name.api_type == 'function') == is_function:
yield name
def _complete_in_string(self, start_leaf, string):
"""
To make it possible for people to have completions in doctests or
generally in "Python" code in docstrings, we use the following
heuristic:
- Having an indented block of code
- Having some doctest code that starts with `>>>`
- Having backticks that doesn't have whitespace inside it
"""
def iter_relevant_lines(lines):
include_next_line = False
for l in code_lines:
if include_next_line or l.startswith('>>>') or l.startswith(' '):
yield re.sub(r'^( *>>> ?| +)', '', l)
else:
yield None
include_next_line = bool(re.match(' *>>>', l))
string = dedent(string)
code_lines = split_lines(string, keepends=True)
relevant_code_lines = list(iter_relevant_lines(code_lines))
if relevant_code_lines[-1] is not None:
# Some code lines might be None, therefore get rid of that.
relevant_code_lines = ['\n' if c is None else c for c in relevant_code_lines]
return self._complete_code_lines(relevant_code_lines)
match = re.search(r'`([^`\s]+)', code_lines[-1])
if match:
return self._complete_code_lines([match.group(1)])
return []
def _complete_code_lines(self, code_lines):
module_node = self._inference_state.grammar.parse(''.join(code_lines))
module_value = ModuleValue(
self._inference_state,
module_node,
code_lines=code_lines,
)
module_value.parent_context = self._module_context
return Completion(
self._inference_state,
module_value.as_context(),
code_lines=code_lines,
position=module_node.end_pos,
signatures_callback=lambda *args, **kwargs: [],
fuzzy=self._fuzzy
).complete()
def _gather_nodes(stack):
nodes = []
for stack_node in stack:
if stack_node.dfa.from_rule == 'small_stmt':
nodes = []
else:
nodes += stack_node.nodes
return nodes
_string_start = re.compile(r'^\w*(\'{3}|"{3}|\'|")')
def _extract_string_while_in_string(leaf, position):
def return_part_of_leaf(leaf):
kwargs = {}
if leaf.line == position[0]:
kwargs['endpos'] = position[1] - leaf.column
match = _string_start.match(leaf.value, **kwargs)
if not match:
return None, None, None
start = match.group(0)
if leaf.line == position[0] and position[1] < leaf.column + match.end():
return None, None, None
return cut_value_at_position(leaf, position)[match.end():], leaf, start
if position < leaf.start_pos:
return None, None, None
if leaf.type == 'string':
return return_part_of_leaf(leaf)
leaves = []
while leaf is not None:
if leaf.type == 'error_leaf' and ('"' in leaf.value or "'" in leaf.value):
if len(leaf.value) > 1:
return return_part_of_leaf(leaf)
prefix_leaf = None
if not leaf.prefix:
prefix_leaf = leaf.get_previous_leaf()
if prefix_leaf is None or prefix_leaf.type != 'name' \
or not all(c in 'rubf' for c in prefix_leaf.value.lower()):
prefix_leaf = None
return (
''.join(cut_value_at_position(l, position) for l in leaves),
prefix_leaf or leaf,
('' if prefix_leaf is None else prefix_leaf.value)
+ cut_value_at_position(leaf, position),
)
if leaf.line != position[0]:
# Multi line strings are always simple error leaves and contain the
# whole string, single line error leaves are atherefore important
# now and since the line is different, it's not really a single
# line string anymore.
break
leaves.insert(0, leaf)
leaf = leaf.get_previous_leaf()
return None, None, None
def complete_trailer(user_context, values):
completion_names = []
for value in values:
for filter in value.get_filters(origin_scope=user_context.tree_node):
completion_names += filter.values()
if not value.is_stub() and isinstance(value, TreeInstance):
completion_names += _complete_getattr(user_context, value)
python_values = convert_values(values)
for c in python_values:
if c not in values:
for filter in c.get_filters(origin_scope=user_context.tree_node):
completion_names += filter.values()
return completion_names
def _complete_getattr(user_context, instance):
"""
A heuristic to make completion for proxy objects work. This is not
intended to work in all cases. It works exactly in this case:
def __getattr__(self, name):
...
return getattr(any_object, name)
It is important that the return contains getattr directly, otherwise it
won't work anymore. It's really just a stupid heuristic. It will not
work if you write e.g. `return (getatr(o, name))`, because of the
additional parentheses. It will also not work if you move the getattr
to some other place that is not the return statement itself.
It is intentional that it doesn't work in all cases. Generally it's
really hard to do even this case (as you can see below). Most people
will write it like this anyway and the other ones, well they are just
out of luck I guess :) ~dave.
"""
names = (instance.get_function_slot_names(u'__getattr__')
or instance.get_function_slot_names(u'__getattribute__'))
functions = ValueSet.from_sets(
name.infer()
for name in names
)
for func in functions:
tree_node = func.tree_node
if tree_node is None or tree_node.type != 'funcdef':
continue
for return_stmt in tree_node.iter_return_stmts():
# Basically until the next comment we just try to find out if a
# return statement looks exactly like `return getattr(x, name)`.
if return_stmt.type != 'return_stmt':
continue
atom_expr = return_stmt.children[1]
if atom_expr.type != 'atom_expr':
continue
atom = atom_expr.children[0]
trailer = atom_expr.children[1]
if len(atom_expr.children) != 2 or atom.type != 'name' \
or atom.value != 'getattr':
continue
arglist = trailer.children[1]
if arglist.type != 'arglist' or len(arglist.children) < 3:
continue
context = func.as_context()
object_node = arglist.children[0]
# Make sure it's a param: foo in __getattr__(self, foo)
name_node = arglist.children[2]
name_list = context.goto(name_node, name_node.start_pos)
if not any(n.api_type == 'param' for n in name_list):
continue
# Now that we know that these are most probably completion
# objects, we just infer the object and return them as
# completions.
objects = context.infer_node(object_node)
return complete_trailer(user_context, objects)
return []
def search_in_module(inference_state, module_context, names, wanted_names,
wanted_type, complete=False, fuzzy=False,
ignore_imports=False, convert=False):
for s in wanted_names[:-1]:
new_names = []
for n in names:
if s == n.string_name:
if n.tree_name is not None and n.api_type == 'module' \
and ignore_imports:
continue
new_names += complete_trailer(
module_context,
n.infer()
)
debug.dbg('dot lookup on search %s from %s', new_names, names[:10])
names = new_names
last_name = wanted_names[-1].lower()
for n in names:
string = n.string_name.lower()
if complete and helpers.match(string, last_name, fuzzy=fuzzy) \
or not complete and string == last_name:
if isinstance(n, SubModuleName):
names = [v.name for v in n.infer()]
else:
names = [n]
if convert:
names = convert_names(names)
for n2 in names:
if complete:
def_ = classes.Completion(
inference_state, n2,
stack=None,
like_name_length=len(last_name),
is_fuzzy=fuzzy,
)
else:
def_ = classes.Name(inference_state, n2)
if not wanted_type or wanted_type == def_.type:
yield def_

View file

@ -0,0 +1,25 @@
_cache = {}
def save_entry(module_name, name, cache):
try:
module_cache = _cache[module_name]
except KeyError:
module_cache = _cache[module_name] = {}
module_cache[name] = cache
def _create_get_from_cache(number):
def _get_from_cache(module_name, name, get_cache_values):
try:
return _cache[module_name][name][number]
except KeyError:
v = get_cache_values()
save_entry(module_name, name, v)
return v[number]
return _get_from_cache
get_type = _create_get_from_cache(0)
get_docstring_signature = _create_get_from_cache(1)
get_docstring = _create_get_from_cache(2)

View file

@ -0,0 +1,489 @@
"""
Environments are a way to activate different Python versions or Virtualenvs for
static analysis. The Python binary in that environment is going to be executed.
"""
import os
import sys
import hashlib
import filecmp
from collections import namedtuple
from jedi._compatibility import highest_pickle_protocol, which
from jedi.cache import memoize_method, time_cache
from jedi.inference.compiled.subprocess import CompiledSubprocess, \
InferenceStateSameProcess, InferenceStateSubprocess
import parso
_VersionInfo = namedtuple('VersionInfo', 'major minor micro')
_SUPPORTED_PYTHONS = ['3.8', '3.7', '3.6', '3.5', '2.7']
_SAFE_PATHS = ['/usr/bin', '/usr/local/bin']
_CONDA_VAR = 'CONDA_PREFIX'
_CURRENT_VERSION = '%s.%s' % (sys.version_info.major, sys.version_info.minor)
class InvalidPythonEnvironment(Exception):
"""
If you see this exception, the Python executable or Virtualenv you have
been trying to use is probably not a correct Python version.
"""
class _BaseEnvironment(object):
@memoize_method
def get_grammar(self):
version_string = '%s.%s' % (self.version_info.major, self.version_info.minor)
return parso.load_grammar(version=version_string)
@property
def _sha256(self):
try:
return self._hash
except AttributeError:
self._hash = _calculate_sha256_for_file(self.executable)
return self._hash
def _get_info():
return (
sys.executable,
sys.prefix,
sys.version_info[:3],
)
class Environment(_BaseEnvironment):
"""
This class is supposed to be created by internal Jedi architecture. You
should not create it directly. Please use create_environment or the other
functions instead. It is then returned by that function.
"""
_subprocess = None
def __init__(self, executable, env_vars=None):
self._start_executable = executable
self._env_vars = env_vars
# Initialize the environment
self._get_subprocess()
def _get_subprocess(self):
if self._subprocess is not None and not self._subprocess.is_crashed:
return self._subprocess
try:
self._subprocess = CompiledSubprocess(self._start_executable,
env_vars=self._env_vars)
info = self._subprocess._send(None, _get_info)
except Exception as exc:
raise InvalidPythonEnvironment(
"Could not get version information for %r: %r" % (
self._start_executable,
exc))
# Since it could change and might not be the same(?) as the one given,
# set it here.
self.executable = info[0]
"""
The Python executable, matches ``sys.executable``.
"""
self.path = info[1]
"""
The path to an environment, matches ``sys.prefix``.
"""
self.version_info = _VersionInfo(*info[2])
"""
Like :data:`sys.version_info`: a tuple to show the current
Environment's Python version.
"""
# py2 sends bytes via pickle apparently?!
if self.version_info.major == 2:
self.executable = self.executable.decode()
self.path = self.path.decode()
# Adjust pickle protocol according to host and client version.
self._subprocess._pickle_protocol = highest_pickle_protocol([
sys.version_info, self.version_info])
return self._subprocess
def __repr__(self):
version = '.'.join(str(i) for i in self.version_info)
return '<%s: %s in %s>' % (self.__class__.__name__, version, self.path)
def get_inference_state_subprocess(self, inference_state):
return InferenceStateSubprocess(inference_state, self._get_subprocess())
@memoize_method
def get_sys_path(self):
"""
The sys path for this environment. Does not include potential
modifications from e.g. appending to :data:`sys.path`.
:returns: list of str
"""
# It's pretty much impossible to generate the sys path without actually
# executing Python. The sys path (when starting with -S) itself depends
# on how the Python version was compiled (ENV variables).
# If you omit -S when starting Python (normal case), additionally
# site.py gets executed.
return self._get_subprocess().get_sys_path()
class _SameEnvironmentMixin(object):
def __init__(self):
self._start_executable = self.executable = sys.executable
self.path = sys.prefix
self.version_info = _VersionInfo(*sys.version_info[:3])
self._env_vars = None
class SameEnvironment(_SameEnvironmentMixin, Environment):
pass
class InterpreterEnvironment(_SameEnvironmentMixin, _BaseEnvironment):
def get_inference_state_subprocess(self, inference_state):
return InferenceStateSameProcess(inference_state)
def get_sys_path(self):
return sys.path
def _get_virtual_env_from_var(env_var='VIRTUAL_ENV'):
"""Get virtualenv environment from VIRTUAL_ENV environment variable.
It uses `safe=False` with ``create_environment``, because the environment
variable is considered to be safe / controlled by the user solely.
"""
var = os.environ.get(env_var)
if var:
# Under macOS in some cases - notably when using Pipenv - the
# sys.prefix of the virtualenv is /path/to/env/bin/.. instead of
# /path/to/env so we need to fully resolve the paths in order to
# compare them.
if os.path.realpath(var) == os.path.realpath(sys.prefix):
return _try_get_same_env()
try:
return create_environment(var, safe=False)
except InvalidPythonEnvironment:
pass
def _calculate_sha256_for_file(path):
sha256 = hashlib.sha256()
with open(path, 'rb') as f:
for block in iter(lambda: f.read(filecmp.BUFSIZE), b''):
sha256.update(block)
return sha256.hexdigest()
def get_default_environment():
"""
Tries to return an active Virtualenv or conda environment.
If there is no VIRTUAL_ENV variable or no CONDA_PREFIX variable set
set it will return the latest Python version installed on the system. This
makes it possible to use as many new Python features as possible when using
autocompletion and other functionality.
:returns: :class:`.Environment`
"""
virtual_env = _get_virtual_env_from_var()
if virtual_env is not None:
return virtual_env
conda_env = _get_virtual_env_from_var(_CONDA_VAR)
if conda_env is not None:
return conda_env
return _try_get_same_env()
def _try_get_same_env():
env = SameEnvironment()
if not os.path.basename(env.executable).lower().startswith('python'):
# This tries to counter issues with embedding. In some cases (e.g.
# VIM's Python Mac/Windows, sys.executable is /foo/bar/vim. This
# happens, because for Mac a function called `_NSGetExecutablePath` is
# used and for Windows `GetModuleFileNameW`. These are both platform
# specific functions. For all other systems sys.executable should be
# alright. However here we try to generalize:
#
# 1. Check if the executable looks like python (heuristic)
# 2. In case it's not try to find the executable
# 3. In case we don't find it use an interpreter environment.
#
# The last option will always work, but leads to potential crashes of
# Jedi - which is ok, because it happens very rarely and even less,
# because the code below should work for most cases.
if os.name == 'nt':
# The first case would be a virtualenv and the second a normal
# Python installation.
checks = (r'Scripts\python.exe', 'python.exe')
else:
# For unix it looks like Python is always in a bin folder.
checks = (
'bin/python%s.%s' % (sys.version_info[0], sys.version[1]),
'bin/python%s' % (sys.version_info[0]),
'bin/python',
)
for check in checks:
guess = os.path.join(sys.exec_prefix, check)
if os.path.isfile(guess):
# Bingo - We think we have our Python.
return Environment(guess)
# It looks like there is no reasonable Python to be found.
return InterpreterEnvironment()
# If no virtualenv is found, use the environment we're already
# using.
return env
def get_cached_default_environment():
var = os.environ.get('VIRTUAL_ENV') or os.environ.get(_CONDA_VAR)
environment = _get_cached_default_environment()
# Under macOS in some cases - notably when using Pipenv - the
# sys.prefix of the virtualenv is /path/to/env/bin/.. instead of
# /path/to/env so we need to fully resolve the paths in order to
# compare them.
if var and os.path.realpath(var) != os.path.realpath(environment.path):
_get_cached_default_environment.clear_cache()
return _get_cached_default_environment()
return environment
@time_cache(seconds=10 * 60) # 10 Minutes
def _get_cached_default_environment():
try:
return get_default_environment()
except InvalidPythonEnvironment:
# It's possible that `sys.executable` is wrong. Typically happens
# when Jedi is used in an executable that embeds Python. For further
# information, have a look at:
# https://github.com/davidhalter/jedi/issues/1531
return InterpreterEnvironment()
def find_virtualenvs(paths=None, **kwargs):
"""
:param paths: A list of paths in your file system to be scanned for
Virtualenvs. It will search in these paths and potentially execute the
Python binaries.
:param safe: Default True. In case this is False, it will allow this
function to execute potential `python` environments. An attacker might
be able to drop an executable in a path this function is searching by
default. If the executable has not been installed by root, it will not
be executed.
:param use_environment_vars: Default True. If True, the VIRTUAL_ENV
variable will be checked if it contains a valid VirtualEnv.
CONDA_PREFIX will be checked to see if it contains a valid conda
environment.
:yields: :class:`.Environment`
"""
def py27_comp(paths=None, safe=True, use_environment_vars=True):
if paths is None:
paths = []
_used_paths = set()
if use_environment_vars:
# Using this variable should be safe, because attackers might be
# able to drop files (via git) but not environment variables.
virtual_env = _get_virtual_env_from_var()
if virtual_env is not None:
yield virtual_env
_used_paths.add(virtual_env.path)
conda_env = _get_virtual_env_from_var(_CONDA_VAR)
if conda_env is not None:
yield conda_env
_used_paths.add(conda_env.path)
for directory in paths:
if not os.path.isdir(directory):
continue
directory = os.path.abspath(directory)
for path in os.listdir(directory):
path = os.path.join(directory, path)
if path in _used_paths:
# A path shouldn't be inferred twice.
continue
_used_paths.add(path)
try:
executable = _get_executable_path(path, safe=safe)
yield Environment(executable)
except InvalidPythonEnvironment:
pass
return py27_comp(paths, **kwargs)
def find_system_environments(**kwargs):
"""
Ignores virtualenvs and returns the Python versions that were installed on
your system. This might return nothing, if you're running Python e.g. from
a portable version.
The environments are sorted from latest to oldest Python version.
:yields: :class:`.Environment`
"""
for version_string in _SUPPORTED_PYTHONS:
try:
yield get_system_environment(version_string, **kwargs)
except InvalidPythonEnvironment:
pass
# TODO: this function should probably return a list of environments since
# multiple Python installations can be found on a system for the same version.
def get_system_environment(version, **kwargs):
"""
Return the first Python environment found for a string of the form 'X.Y'
where X and Y are the major and minor versions of Python.
:raises: :exc:`.InvalidPythonEnvironment`
:returns: :class:`.Environment`
"""
exe = which('python' + version)
if exe:
if exe == sys.executable:
return SameEnvironment()
return Environment(exe)
if os.name == 'nt':
for exe in _get_executables_from_windows_registry(version):
try:
return Environment(exe, **kwargs)
except InvalidPythonEnvironment:
pass
raise InvalidPythonEnvironment("Cannot find executable python%s." % version)
def create_environment(path, safe=True, **kwargs):
"""
Make it possible to manually create an Environment object by specifying a
Virtualenv path or an executable path and optional environment variables.
:raises: :exc:`.InvalidPythonEnvironment`
:returns: :class:`.Environment`
TODO: make env_vars a kwarg when Python 2 is dropped. For now, preserve API
"""
return _create_environment(path, safe, **kwargs)
def _create_environment(path, safe=True, env_vars=None):
if os.path.isfile(path):
_assert_safe(path, safe)
return Environment(path, env_vars=env_vars)
return Environment(_get_executable_path(path, safe=safe), env_vars=env_vars)
def _get_executable_path(path, safe=True):
"""
Returns None if it's not actually a virtual env.
"""
if os.name == 'nt':
python = os.path.join(path, 'Scripts', 'python.exe')
else:
python = os.path.join(path, 'bin', 'python')
if not os.path.exists(python):
raise InvalidPythonEnvironment("%s seems to be missing." % python)
_assert_safe(python, safe)
return python
def _get_executables_from_windows_registry(version):
# The winreg module is named _winreg on Python 2.
try:
import winreg
except ImportError:
import _winreg as winreg
# TODO: support Python Anaconda.
sub_keys = [
r'SOFTWARE\Python\PythonCore\{version}\InstallPath',
r'SOFTWARE\Wow6432Node\Python\PythonCore\{version}\InstallPath',
r'SOFTWARE\Python\PythonCore\{version}-32\InstallPath',
r'SOFTWARE\Wow6432Node\Python\PythonCore\{version}-32\InstallPath'
]
for root_key in [winreg.HKEY_CURRENT_USER, winreg.HKEY_LOCAL_MACHINE]:
for sub_key in sub_keys:
sub_key = sub_key.format(version=version)
try:
with winreg.OpenKey(root_key, sub_key) as key:
prefix = winreg.QueryValueEx(key, '')[0]
exe = os.path.join(prefix, 'python.exe')
if os.path.isfile(exe):
yield exe
except WindowsError:
pass
def _assert_safe(executable_path, safe):
if safe and not _is_safe(executable_path):
raise InvalidPythonEnvironment(
"The python binary is potentially unsafe.")
def _is_safe(executable_path):
# Resolve sym links. A venv typically is a symlink to a known Python
# binary. Only virtualenvs copy symlinks around.
real_path = os.path.realpath(executable_path)
if _is_unix_safe_simple(real_path):
return True
# Just check the list of known Python versions. If it's not in there,
# it's likely an attacker or some Python that was not properly
# installed in the system.
for environment in find_system_environments():
if environment.executable == real_path:
return True
# If the versions don't match, just compare the binary files. If we
# don't do that, only venvs will be working and not virtualenvs.
# venvs are symlinks while virtualenvs are actual copies of the
# Python files.
# This still means that if the system Python is updated and the
# virtualenv's Python is not (which is probably never going to get
# upgraded), it will not work with Jedi. IMO that's fine, because
# people should just be using venv. ~ dave
if environment._sha256 == _calculate_sha256_for_file(real_path):
return True
return False
def _is_unix_safe_simple(real_path):
if _is_unix_admin():
# In case we are root, just be conservative and
# only execute known paths.
return any(real_path.startswith(p) for p in _SAFE_PATHS)
uid = os.stat(real_path).st_uid
# The interpreter needs to be owned by root. This means that it wasn't
# written by a user and therefore attacking Jedi is not as simple.
# The attack could look like the following:
# 1. A user clones a repository.
# 2. The repository has an innocent looking folder called foobar. jedi
# searches for the folder and executes foobar/bin/python --version if
# there's also a foobar/bin/activate.
# 3. The attacker has gained code execution, since he controls
# foobar/bin/python.
return uid == 0
def _is_unix_admin():
try:
return os.getuid() == 0
except AttributeError:
return False # Windows

View file

@ -0,0 +1,46 @@
"""
This file is about errors in Python files and not about exception handling in
Jedi.
"""
def parso_to_jedi_errors(grammar, module_node):
return [SyntaxError(e) for e in grammar.iter_errors(module_node)]
class SyntaxError(object):
"""
Syntax errors are generated by :meth:`.Script.get_syntax_errors`.
"""
def __init__(self, parso_error):
self._parso_error = parso_error
@property
def line(self):
"""The line where the error starts (starting with 1)."""
return self._parso_error.start_pos[0]
@property
def column(self):
"""The column where the error starts (starting with 0)."""
return self._parso_error.start_pos[1]
@property
def until_line(self):
"""The line where the error ends (starting with 1)."""
return self._parso_error.end_pos[0]
@property
def until_column(self):
"""The column where the error ends (starting with 0)."""
return self._parso_error.end_pos[1]
def get_message(self):
return self._parso_error.message
def __repr__(self):
return '<%s from=%s to=%s>' % (
self.__class__.__name__,
self._parso_error.start_pos,
self._parso_error.end_pos,
)

View file

@ -0,0 +1,31 @@
class _JediError(Exception):
pass
class InternalError(_JediError):
"""
This error might happen a subprocess is crashing. The reason for this is
usually broken C code in third party libraries. This is not a very common
thing and it is safe to use Jedi again. However using the same calls might
result in the same error again.
"""
class WrongVersion(_JediError):
"""
This error is reserved for the future, shouldn't really be happening at the
moment.
"""
class RefactoringError(_JediError):
"""
Refactorings can fail for various reasons. So if you work with refactorings
like :meth:`.Script.rename`, :meth:`.Script.inline`,
:meth:`.Script.extract_variable` and :meth:`.Script.extract_function`, make
sure to catch these. The descriptions in the errors are ususally valuable
for end users.
A typical ``RefactoringError`` would tell the user that inlining is not
possible if no name is under the cursor.
"""

View file

@ -0,0 +1,156 @@
import os
from jedi._compatibility import FileNotFoundError, force_unicode, scandir
from jedi.api import classes
from jedi.api.strings import StringName, get_quote_ending
from jedi.api.helpers import match
from jedi.inference.helpers import get_str_or_none
class PathName(StringName):
api_type = u'path'
def complete_file_name(inference_state, module_context, start_leaf, quote, string,
like_name, signatures_callback, code_lines, position, fuzzy):
# First we want to find out what can actually be changed as a name.
like_name_length = len(os.path.basename(string))
addition = _get_string_additions(module_context, start_leaf)
if string.startswith('~'):
string = os.path.expanduser(string)
if addition is None:
return
string = addition + string
# Here we use basename again, because if strings are added like
# `'foo' + 'bar`, it should complete to `foobar/`.
must_start_with = os.path.basename(string)
string = os.path.dirname(string)
sigs = signatures_callback(*position)
is_in_os_path_join = sigs and all(s.full_name == 'os.path.join' for s in sigs)
if is_in_os_path_join:
to_be_added = _add_os_path_join(module_context, start_leaf, sigs[0].bracket_start)
if to_be_added is None:
is_in_os_path_join = False
else:
string = to_be_added + string
base_path = os.path.join(inference_state.project.path, string)
try:
listed = sorted(scandir(base_path), key=lambda e: e.name)
# OSError: [Errno 36] File name too long: '...'
except (FileNotFoundError, OSError):
return
quote_ending = get_quote_ending(quote, code_lines, position)
for entry in listed:
name = entry.name
if match(name, must_start_with, fuzzy=fuzzy):
if is_in_os_path_join or not entry.is_dir():
name += quote_ending
else:
name += os.path.sep
yield classes.Completion(
inference_state,
PathName(inference_state, name[len(must_start_with) - like_name_length:]),
stack=None,
like_name_length=like_name_length,
is_fuzzy=fuzzy,
)
def _get_string_additions(module_context, start_leaf):
def iterate_nodes():
node = addition.parent
was_addition = True
for child_node in reversed(node.children[:node.children.index(addition)]):
if was_addition:
was_addition = False
yield child_node
continue
if child_node != '+':
break
was_addition = True
addition = start_leaf.get_previous_leaf()
if addition != '+':
return ''
context = module_context.create_context(start_leaf)
return _add_strings(context, reversed(list(iterate_nodes())))
def _add_strings(context, nodes, add_slash=False):
string = ''
first = True
for child_node in nodes:
values = context.infer_node(child_node)
if len(values) != 1:
return None
c, = values
s = get_str_or_none(c)
if s is None:
return None
if not first and add_slash:
string += os.path.sep
string += force_unicode(s)
first = False
return string
def _add_os_path_join(module_context, start_leaf, bracket_start):
def check(maybe_bracket, nodes):
if maybe_bracket.start_pos != bracket_start:
return None
if not nodes:
return ''
context = module_context.create_context(nodes[0])
return _add_strings(context, nodes, add_slash=True) or ''
if start_leaf.type == 'error_leaf':
# Unfinished string literal, like `join('`
value_node = start_leaf.parent
index = value_node.children.index(start_leaf)
if index > 0:
error_node = value_node.children[index - 1]
if error_node.type == 'error_node' and len(error_node.children) >= 2:
index = -2
if error_node.children[-1].type == 'arglist':
arglist_nodes = error_node.children[-1].children
index -= 1
else:
arglist_nodes = []
return check(error_node.children[index + 1], arglist_nodes[::2])
return None
# Maybe an arglist or some weird error case. Therefore checked below.
searched_node_child = start_leaf
while searched_node_child.parent is not None \
and searched_node_child.parent.type not in ('arglist', 'trailer', 'error_node'):
searched_node_child = searched_node_child.parent
if searched_node_child.get_first_leaf() is not start_leaf:
return None
searched_node = searched_node_child.parent
if searched_node is None:
return None
index = searched_node.children.index(searched_node_child)
arglist_nodes = searched_node.children[:index]
if searched_node.type == 'arglist':
trailer = searched_node.parent
if trailer.type == 'error_node':
trailer_index = trailer.children.index(searched_node)
assert trailer_index >= 2
assert trailer.children[trailer_index - 1] == '('
return check(trailer.children[trailer_index - 1], arglist_nodes[::2])
elif trailer.type == 'trailer':
return check(trailer.children[0], arglist_nodes[::2])
elif searched_node.type == 'trailer':
return check(searched_node.children[0], [])
elif searched_node.type == 'error_node':
# Stuff like `join(""`
return check(arglist_nodes[-1], [])

View file

@ -0,0 +1,522 @@
"""
Helpers for the API
"""
import re
from collections import namedtuple
from textwrap import dedent
from itertools import chain
from functools import wraps
from parso.python.parser import Parser
from parso.python import tree
from jedi._compatibility import u, Parameter
from jedi.inference.base_value import NO_VALUES
from jedi.inference.syntax_tree import infer_atom
from jedi.inference.helpers import infer_call_of_leaf
from jedi.inference.compiled import get_string_value_set
from jedi.cache import signature_time_cache, memoize_method
from jedi.parser_utils import get_parent_scope
CompletionParts = namedtuple('CompletionParts', ['path', 'has_dot', 'name'])
def _start_match(string, like_name):
return string.startswith(like_name)
def _fuzzy_match(string, like_name):
if len(like_name) <= 1:
return like_name in string
pos = string.find(like_name[0])
if pos >= 0:
return _fuzzy_match(string[pos + 1:], like_name[1:])
return False
def match(string, like_name, fuzzy=False):
if fuzzy:
return _fuzzy_match(string, like_name)
else:
return _start_match(string, like_name)
def sorted_definitions(defs):
# Note: `or ''` below is required because `module_path` could be
return sorted(defs, key=lambda x: (x.module_path or '', x.line or 0, x.column or 0, x.name))
def get_on_completion_name(module_node, lines, position):
leaf = module_node.get_leaf_for_position(position)
if leaf is None or leaf.type in ('string', 'error_leaf'):
# Completions inside strings are a bit special, we need to parse the
# string. The same is true for comments and error_leafs.
line = lines[position[0] - 1]
# The first step of completions is to get the name
return re.search(r'(?!\d)\w+$|$', line[:position[1]]).group(0)
elif leaf.type not in ('name', 'keyword'):
return ''
return leaf.value[:position[1] - leaf.start_pos[1]]
def _get_code(code_lines, start_pos, end_pos):
# Get relevant lines.
lines = code_lines[start_pos[0] - 1:end_pos[0]]
# Remove the parts at the end of the line.
lines[-1] = lines[-1][:end_pos[1]]
# Remove first line indentation.
lines[0] = lines[0][start_pos[1]:]
return ''.join(lines)
class OnErrorLeaf(Exception):
@property
def error_leaf(self):
return self.args[0]
def _get_code_for_stack(code_lines, leaf, position):
# It might happen that we're on whitespace or on a comment. This means
# that we would not get the right leaf.
if leaf.start_pos >= position:
# If we're not on a comment simply get the previous leaf and proceed.
leaf = leaf.get_previous_leaf()
if leaf is None:
return u('') # At the beginning of the file.
is_after_newline = leaf.type == 'newline'
while leaf.type == 'newline':
leaf = leaf.get_previous_leaf()
if leaf is None:
return u('')
if leaf.type == 'error_leaf' or leaf.type == 'string':
if leaf.start_pos[0] < position[0]:
# On a different line, we just begin anew.
return u('')
# Error leafs cannot be parsed, completion in strings is also
# impossible.
raise OnErrorLeaf(leaf)
else:
user_stmt = leaf
while True:
if user_stmt.parent.type in ('file_input', 'suite', 'simple_stmt'):
break
user_stmt = user_stmt.parent
if is_after_newline:
if user_stmt.start_pos[1] > position[1]:
# This means that it's actually a dedent and that means that we
# start without value (part of a suite).
return u('')
# This is basically getting the relevant lines.
return _get_code(code_lines, user_stmt.get_start_pos_of_prefix(), position)
def get_stack_at_position(grammar, code_lines, leaf, pos):
"""
Returns the possible node names (e.g. import_from, xor_test or yield_stmt).
"""
class EndMarkerReached(Exception):
pass
def tokenize_without_endmarker(code):
# TODO This is for now not an official parso API that exists purely
# for Jedi.
tokens = grammar._tokenize(code)
for token in tokens:
if token.string == safeword:
raise EndMarkerReached()
elif token.prefix.endswith(safeword):
# This happens with comments.
raise EndMarkerReached()
elif token.string.endswith(safeword):
yield token # Probably an f-string literal that was not finished.
raise EndMarkerReached()
else:
yield token
# The code might be indedented, just remove it.
code = dedent(_get_code_for_stack(code_lines, leaf, pos))
# We use a word to tell Jedi when we have reached the start of the
# completion.
# Use Z as a prefix because it's not part of a number suffix.
safeword = 'ZZZ_USER_WANTS_TO_COMPLETE_HERE_WITH_JEDI'
code = code + ' ' + safeword
p = Parser(grammar._pgen_grammar, error_recovery=True)
try:
p.parse(tokens=tokenize_without_endmarker(code))
except EndMarkerReached:
return p.stack
raise SystemError(
"This really shouldn't happen. There's a bug in Jedi:\n%s"
% list(tokenize_without_endmarker(code))
)
def infer(inference_state, context, leaf):
if leaf.type == 'name':
return inference_state.infer(context, leaf)
parent = leaf.parent
definitions = NO_VALUES
if parent.type == 'atom':
# e.g. `(a + b)`
definitions = context.infer_node(leaf.parent)
elif parent.type == 'trailer':
# e.g. `a()`
definitions = infer_call_of_leaf(context, leaf)
elif isinstance(leaf, tree.Literal):
# e.g. `"foo"` or `1.0`
return infer_atom(context, leaf)
elif leaf.type in ('fstring_string', 'fstring_start', 'fstring_end'):
return get_string_value_set(inference_state)
return definitions
def filter_follow_imports(names, follow_builtin_imports=False):
for name in names:
if name.is_import():
new_names = list(filter_follow_imports(
name.goto(),
follow_builtin_imports=follow_builtin_imports,
))
found_builtin = False
if follow_builtin_imports:
for new_name in new_names:
if new_name.start_pos is None:
found_builtin = True
if found_builtin:
yield name
else:
for new_name in new_names:
yield new_name
else:
yield name
class CallDetails(object):
def __init__(self, bracket_leaf, children, position):
['bracket_leaf', 'call_index', 'keyword_name_str']
self.bracket_leaf = bracket_leaf
self._children = children
self._position = position
@property
def index(self):
return _get_index_and_key(self._children, self._position)[0]
@property
def keyword_name_str(self):
return _get_index_and_key(self._children, self._position)[1]
@memoize_method
def _list_arguments(self):
return list(_iter_arguments(self._children, self._position))
def calculate_index(self, param_names):
positional_count = 0
used_names = set()
star_count = -1
args = self._list_arguments()
if not args:
if param_names:
return 0
else:
return None
is_kwarg = False
for i, (star_count, key_start, had_equal) in enumerate(args):
is_kwarg |= had_equal | (star_count == 2)
if star_count:
pass # For now do nothing, we don't know what's in there here.
else:
if i + 1 != len(args): # Not last
if had_equal:
used_names.add(key_start)
else:
positional_count += 1
for i, param_name in enumerate(param_names):
kind = param_name.get_kind()
if not is_kwarg:
if kind == Parameter.VAR_POSITIONAL:
return i
if kind in (Parameter.POSITIONAL_OR_KEYWORD, Parameter.POSITIONAL_ONLY):
if i == positional_count:
return i
if key_start is not None and not star_count == 1 or star_count == 2:
if param_name.string_name not in used_names \
and (kind == Parameter.KEYWORD_ONLY
or kind == Parameter.POSITIONAL_OR_KEYWORD
and positional_count <= i):
if star_count:
return i
if had_equal:
if param_name.string_name == key_start:
return i
else:
if param_name.string_name.startswith(key_start):
return i
if kind == Parameter.VAR_KEYWORD:
return i
return None
def iter_used_keyword_arguments(self):
for star_count, key_start, had_equal in list(self._list_arguments()):
if had_equal and key_start:
yield key_start
def count_positional_arguments(self):
count = 0
for star_count, key_start, had_equal in self._list_arguments()[:-1]:
if star_count:
break
count += 1
return count
def _iter_arguments(nodes, position):
def remove_after_pos(name):
if name.type != 'name':
return None
return name.value[:position[1] - name.start_pos[1]]
# Returns Generator[Tuple[star_count, Optional[key_start: str], had_equal]]
nodes_before = [c for c in nodes if c.start_pos < position]
if nodes_before[-1].type == 'arglist':
for x in _iter_arguments(nodes_before[-1].children, position):
yield x # Python 2 :(
return
previous_node_yielded = False
stars_seen = 0
for i, node in enumerate(nodes_before):
if node.type == 'argument':
previous_node_yielded = True
first = node.children[0]
second = node.children[1]
if second == '=':
if second.start_pos < position:
yield 0, first.value, True
else:
yield 0, remove_after_pos(first), False
elif first in ('*', '**'):
yield len(first.value), remove_after_pos(second), False
else:
# Must be a Comprehension
first_leaf = node.get_first_leaf()
if first_leaf.type == 'name' and first_leaf.start_pos >= position:
yield 0, remove_after_pos(first_leaf), False
else:
yield 0, None, False
stars_seen = 0
elif node.type in ('testlist', 'testlist_star_expr'): # testlist is Python 2
for n in node.children[::2]:
if n.type == 'star_expr':
stars_seen = 1
n = n.children[1]
yield stars_seen, remove_after_pos(n), False
stars_seen = 0
# The count of children is even if there's a comma at the end.
previous_node_yielded = bool(len(node.children) % 2)
elif isinstance(node, tree.PythonLeaf) and node.value == ',':
if not previous_node_yielded:
yield stars_seen, '', False
stars_seen = 0
previous_node_yielded = False
elif isinstance(node, tree.PythonLeaf) and node.value in ('*', '**'):
stars_seen = len(node.value)
elif node == '=' and nodes_before[-1]:
previous_node_yielded = True
before = nodes_before[i - 1]
if before.type == 'name':
yield 0, before.value, True
else:
yield 0, None, False
# Just ignore the star that is probably a syntax error.
stars_seen = 0
if not previous_node_yielded:
if nodes_before[-1].type == 'name':
yield stars_seen, remove_after_pos(nodes_before[-1]), False
else:
yield stars_seen, '', False
def _get_index_and_key(nodes, position):
"""
Returns the amount of commas and the keyword argument string.
"""
nodes_before = [c for c in nodes if c.start_pos < position]
if nodes_before[-1].type == 'arglist':
return _get_index_and_key(nodes_before[-1].children, position)
key_str = None
last = nodes_before[-1]
if last.type == 'argument' and last.children[1] == '=' \
and last.children[1].end_pos <= position:
# Checked if the argument
key_str = last.children[0].value
elif last == '=':
key_str = nodes_before[-2].value
return nodes_before.count(','), key_str
def _get_signature_details_from_error_node(node, additional_children, position):
for index, element in reversed(list(enumerate(node.children))):
# `index > 0` means that it's a trailer and not an atom.
if element == '(' and element.end_pos <= position and index > 0:
# It's an error node, we don't want to match too much, just
# until the parentheses is enough.
children = node.children[index:]
name = element.get_previous_leaf()
if name is None:
continue
if name.type == 'name' or name.parent.type in ('trailer', 'atom'):
return CallDetails(element, children + additional_children, position)
def get_signature_details(module, position):
leaf = module.get_leaf_for_position(position, include_prefixes=True)
# It's easier to deal with the previous token than the next one in this
# case.
if leaf.start_pos >= position:
# Whitespace / comments after the leaf count towards the previous leaf.
leaf = leaf.get_previous_leaf()
if leaf is None:
return None
# Now that we know where we are in the syntax tree, we start to look at
# parents for possible function definitions.
node = leaf.parent
while node is not None:
if node.type in ('funcdef', 'classdef', 'decorated', 'async_stmt'):
# Don't show signatures if there's stuff before it that just
# makes it feel strange to have a signature.
return None
additional_children = []
for n in reversed(node.children):
if n.start_pos < position:
if n.type == 'error_node':
result = _get_signature_details_from_error_node(
n, additional_children, position
)
if result is not None:
return result
additional_children[0:0] = n.children
continue
additional_children.insert(0, n)
# Find a valid trailer
if node.type == 'trailer' and node.children[0] == '(' \
or node.type == 'decorator' and node.children[2] == '(':
# Additionally we have to check that an ending parenthesis isn't
# interpreted wrong. There are two cases:
# 1. Cursor before paren -> The current signature is good
# 2. Cursor after paren -> We need to skip the current signature
if not (leaf is node.children[-1] and position >= leaf.end_pos):
leaf = node.get_previous_leaf()
if leaf is None:
return None
return CallDetails(
node.children[0] if node.type == 'trailer' else node.children[2],
node.children,
position
)
node = node.parent
return None
@signature_time_cache("call_signatures_validity")
def cache_signatures(inference_state, context, bracket_leaf, code_lines, user_pos):
"""This function calculates the cache key."""
line_index = user_pos[0] - 1
before_cursor = code_lines[line_index][:user_pos[1]]
other_lines = code_lines[bracket_leaf.start_pos[0]:line_index]
whole = ''.join(other_lines + [before_cursor])
before_bracket = re.match(r'.*\(', whole, re.DOTALL)
module_path = context.get_root_context().py__file__()
if module_path is None:
yield None # Don't cache!
else:
yield (module_path, before_bracket, bracket_leaf.start_pos)
yield infer(
inference_state,
context,
bracket_leaf.get_previous_leaf(),
)
def validate_line_column(func):
@wraps(func)
def wrapper(self, line=None, column=None, *args, **kwargs):
line = max(len(self._code_lines), 1) if line is None else line
if not (0 < line <= len(self._code_lines)):
raise ValueError('`line` parameter is not in a valid range.')
line_string = self._code_lines[line - 1]
line_len = len(line_string)
if line_string.endswith('\r\n'):
line_len -= 2
elif line_string.endswith('\n'):
line_len -= 1
column = line_len if column is None else column
if not (0 <= column <= line_len):
raise ValueError('`column` parameter (%d) is not in a valid range '
'(0-%d) for line %d (%r).' % (
column, line_len, line, line_string))
return func(self, line, column, *args, **kwargs)
return wrapper
def get_module_names(module, all_scopes, definitions=True, references=False):
"""
Returns a dictionary with name parts as keys and their call paths as
values.
"""
def def_ref_filter(name):
is_def = name.is_definition()
return definitions and is_def or references and not is_def
names = list(chain.from_iterable(module.get_used_names().values()))
if not all_scopes:
# We have to filter all the names that don't have the module as a
# parent_scope. There's None as a parent, because nodes in the module
# node have the parent module and not suite as all the others.
# Therefore it's important to catch that case.
def is_module_scope_name(name):
parent_scope = get_parent_scope(name)
# async functions have an extra wrapper. Strip it.
if parent_scope and parent_scope.type == 'async_stmt':
parent_scope = parent_scope.parent
return parent_scope in (module, None)
names = [n for n in names if is_module_scope_name(n)]
return filter(def_ref_filter, names)
def split_search_string(name):
type, _, dotted_names = name.rpartition(' ')
if type == 'def':
type = 'function'
return type, dotted_names.split('.')

View file

@ -0,0 +1,75 @@
"""
TODO Some parts of this module are still not well documented.
"""
from jedi.inference import compiled
from jedi.inference.base_value import ValueSet
from jedi.inference.filters import ParserTreeFilter, MergedFilter
from jedi.inference.names import TreeNameDefinition
from jedi.inference.compiled import mixed
from jedi.inference.compiled.access import create_access_path
from jedi.inference.context import ModuleContext
def _create(inference_state, obj):
return compiled.create_from_access_path(
inference_state, create_access_path(inference_state, obj)
)
class NamespaceObject(object):
def __init__(self, dct):
self.__dict__ = dct
class MixedTreeName(TreeNameDefinition):
def infer(self):
"""
In IPython notebook it is typical that some parts of the code that is
provided was already executed. In that case if something is not properly
inferred, it should still infer from the variables it already knows.
"""
inferred = super(MixedTreeName, self).infer()
if not inferred:
for compiled_value in self.parent_context.mixed_values:
for f in compiled_value.get_filters():
values = ValueSet.from_sets(
n.infer() for n in f.get(self.string_name)
)
if values:
return values
return inferred
class MixedParserTreeFilter(ParserTreeFilter):
name_class = MixedTreeName
class MixedModuleContext(ModuleContext):
def __init__(self, tree_module_value, namespaces):
super(MixedModuleContext, self).__init__(tree_module_value)
self.mixed_values = [
self._get_mixed_object(
_create(self.inference_state, NamespaceObject(n))
) for n in namespaces
]
def _get_mixed_object(self, compiled_value):
return mixed.MixedObject(
compiled_value=compiled_value,
tree_value=self._value
)
def get_filters(self, until_position=None, origin_scope=None):
yield MergedFilter(
MixedParserTreeFilter(
parent_context=self,
until_position=until_position,
origin_scope=origin_scope
),
self.get_global_filter(),
)
for mixed_object in self.mixed_values:
for filter in mixed_object.get_filters(until_position, origin_scope):
yield filter

View file

@ -0,0 +1,57 @@
import pydoc
from jedi.inference.utils import ignored
from jedi.inference.names import AbstractArbitraryName
try:
from pydoc_data import topics as pydoc_topics
except ImportError:
# Python 2
try:
import pydoc_topics
except ImportError:
# This is for Python 3 embeddable version, which dont have
# pydoc_data module in its file python3x.zip.
pydoc_topics = None
class KeywordName(AbstractArbitraryName):
api_type = u'keyword'
def py__doc__(self):
return imitate_pydoc(self.string_name)
def imitate_pydoc(string):
"""
It's not possible to get the pydoc's without starting the annoying pager
stuff.
"""
if pydoc_topics is None:
return ''
# str needed because of possible unicode stuff in py2k (pydoc doesn't work
# with unicode strings)
string = str(string)
h = pydoc.help
with ignored(KeyError):
# try to access symbols
string = h.symbols[string]
string, _, related = string.partition(' ')
def get_target(s):
return h.topics.get(s, h.keywords.get(s))
while isinstance(string, str):
string = get_target(string)
try:
# is a tuple now
label, related = string
except TypeError:
return ''
try:
return pydoc_topics.topics[label].strip() if pydoc_topics else ''
except KeyError:
return ''

View file

@ -0,0 +1,427 @@
"""
Projects are a way to handle Python projects within Jedi. For simpler plugins
you might not want to deal with projects, but if you want to give the user more
flexibility to define sys paths and Python interpreters for a project,
:class:`.Project` is the perfect way to allow for that.
Projects can be saved to disk and loaded again, to allow project definitions to
be used across repositories.
"""
import os
import errno
import json
import sys
from jedi._compatibility import FileNotFoundError, PermissionError, \
IsADirectoryError, NotADirectoryError
from jedi import debug
from jedi.api.environment import get_cached_default_environment, create_environment
from jedi.api.exceptions import WrongVersion
from jedi.api.completion import search_in_module
from jedi.api.helpers import split_search_string, get_module_names
from jedi._compatibility import force_unicode
from jedi.inference.imports import load_module_from_path, \
load_namespace_from_path, iter_module_names
from jedi.inference.sys_path import discover_buildout_paths
from jedi.inference.cache import inference_state_as_method_param_cache
from jedi.inference.references import recurse_find_python_folders_and_files, search_in_file_ios
from jedi.file_io import FolderIO
from jedi.common import traverse_parents
_CONFIG_FOLDER = '.jedi'
_CONTAINS_POTENTIAL_PROJECT = \
'setup.py', '.git', '.hg', 'requirements.txt', 'MANIFEST.in', 'pyproject.toml'
_SERIALIZER_VERSION = 1
def _try_to_skip_duplicates(func):
def wrapper(*args, **kwargs):
found_tree_nodes = []
found_modules = []
for definition in func(*args, **kwargs):
tree_node = definition._name.tree_name
if tree_node is not None and tree_node in found_tree_nodes:
continue
if definition.type == 'module' and definition.module_path is not None:
if definition.module_path in found_modules:
continue
found_modules.append(definition.module_path)
yield definition
found_tree_nodes.append(tree_node)
return wrapper
def _remove_duplicates_from_path(path):
used = set()
for p in path:
if p in used:
continue
used.add(p)
yield p
def _force_unicode_list(lst):
return list(map(force_unicode, lst))
class Project(object):
"""
Projects are a simple way to manage Python folders and define how Jedi does
import resolution. It is mostly used as a parameter to :class:`.Script`.
Additionally there are functions to search a whole project.
"""
_environment = None
@staticmethod
def _get_config_folder_path(base_path):
return os.path.join(base_path, _CONFIG_FOLDER)
@staticmethod
def _get_json_path(base_path):
return os.path.join(Project._get_config_folder_path(base_path), 'project.json')
@classmethod
def load(cls, path):
"""
Loads a project from a specific path. You should not provide the path
to ``.jedi/project.json``, but rather the path to the project folder.
:param path: The path of the directory you want to use as a project.
"""
with open(cls._get_json_path(path)) as f:
version, data = json.load(f)
if version == 1:
return cls(**data)
else:
raise WrongVersion(
"The Jedi version of this project seems newer than what we can handle."
)
def save(self):
"""
Saves the project configuration in the project in ``.jedi/project.json``.
"""
data = dict(self.__dict__)
data.pop('_environment', None)
data.pop('_django', None) # TODO make django setting public?
data = {k.lstrip('_'): v for k, v in data.items()}
# TODO when dropping Python 2 use pathlib.Path.mkdir(parents=True, exist_ok=True)
try:
os.makedirs(self._get_config_folder_path(self._path))
except OSError as e:
if e.errno != errno.EEXIST:
raise
with open(self._get_json_path(self._path), 'w') as f:
return json.dump((_SERIALIZER_VERSION, data), f)
def __init__(self, path, **kwargs):
"""
:param path: The base path for this project.
:param environment_path: The Python executable path, typically the path
of a virtual environment.
:param load_unsafe_extensions: Default False, Loads extensions that are not in the
sys path and in the local directories. With this option enabled,
this is potentially unsafe if you clone a git repository and
analyze it's code, because those compiled extensions will be
important and therefore have execution privileges.
:param sys_path: list of str. You can override the sys path if you
want. By default the ``sys.path.`` is generated by the
environment (virtualenvs, etc).
:param added_sys_path: list of str. Adds these paths at the end of the
sys path.
:param smart_sys_path: If this is enabled (default), adds paths from
local directories. Otherwise you will have to rely on your packages
being properly configured on the ``sys.path``.
"""
def py2_comp(path, environment_path=None, load_unsafe_extensions=False,
sys_path=None, added_sys_path=(), smart_sys_path=True):
self._path = os.path.abspath(path)
self._environment_path = environment_path
self._sys_path = sys_path
self._smart_sys_path = smart_sys_path
self._load_unsafe_extensions = load_unsafe_extensions
self._django = False
self.added_sys_path = list(added_sys_path)
"""The sys path that is going to be added at the end of the """
py2_comp(path, **kwargs)
@property
def path(self):
"""
The base path for this project.
"""
return self._path
@inference_state_as_method_param_cache()
def _get_base_sys_path(self, inference_state):
# The sys path has not been set explicitly.
sys_path = list(inference_state.environment.get_sys_path())
try:
sys_path.remove('')
except ValueError:
pass
return sys_path
@inference_state_as_method_param_cache()
def _get_sys_path(self, inference_state, add_parent_paths=True, add_init_paths=False):
"""
Keep this method private for all users of jedi. However internally this
one is used like a public method.
"""
suffixed = list(self.added_sys_path)
prefixed = []
if self._sys_path is None:
sys_path = list(self._get_base_sys_path(inference_state))
else:
sys_path = list(self._sys_path)
if self._smart_sys_path:
prefixed.append(self._path)
if inference_state.script_path is not None:
suffixed += discover_buildout_paths(inference_state, inference_state.script_path)
if add_parent_paths:
# Collect directories in upward search by:
# 1. Skipping directories with __init__.py
# 2. Stopping immediately when above self._path
traversed = []
for parent_path in traverse_parents(inference_state.script_path):
if parent_path == self._path or not parent_path.startswith(self._path):
break
if not add_init_paths \
and os.path.isfile(os.path.join(parent_path, "__init__.py")):
continue
traversed.append(parent_path)
# AFAIK some libraries have imports like `foo.foo.bar`, which
# leads to the conclusion to by default prefer longer paths
# rather than shorter ones by default.
suffixed += reversed(traversed)
if self._django:
prefixed.append(self._path)
path = prefixed + sys_path + suffixed
return list(_force_unicode_list(_remove_duplicates_from_path(path)))
def get_environment(self):
if self._environment is None:
if self._environment_path is not None:
self._environment = create_environment(self._environment_path, safe=False)
else:
self._environment = get_cached_default_environment()
return self._environment
def search(self, string, **kwargs):
"""
Searches a name in the whole project. If the project is very big,
at some point Jedi will stop searching. However it's also very much
recommended to not exhaust the generator. Just display the first ten
results to the user.
There are currently three different search patterns:
- ``foo`` to search for a definition foo in any file or a file called
``foo.py`` or ``foo.pyi``.
- ``foo.bar`` to search for the ``foo`` and then an attribute ``bar``
in it.
- ``class foo.bar.Bar`` or ``def foo.bar.baz`` to search for a specific
API type.
:param bool all_scopes: Default False; searches not only for
definitions on the top level of a module level, but also in
functions and classes.
:yields: :class:`.Name`
"""
return self._search(string, **kwargs)
def complete_search(self, string, **kwargs):
"""
Like :meth:`.Script.search`, but completes that string. An empty string
lists all definitions in a project, so be careful with that.
:param bool all_scopes: Default False; searches not only for
definitions on the top level of a module level, but also in
functions and classes.
:yields: :class:`.Completion`
"""
return self._search_func(string, complete=True, **kwargs)
def _search(self, string, all_scopes=False): # Python 2..
return self._search_func(string, all_scopes=all_scopes)
@_try_to_skip_duplicates
def _search_func(self, string, complete=False, all_scopes=False):
# Using a Script is they easiest way to get an empty module context.
from jedi import Script
s = Script('', project=self)
inference_state = s._inference_state
empty_module_context = s._get_module_context()
if inference_state.grammar.version_info < (3, 6) or sys.version_info < (3, 6):
raise NotImplementedError(
"No support for refactorings/search on Python 2/3.5"
)
debug.dbg('Search for string %s, complete=%s', string, complete)
wanted_type, wanted_names = split_search_string(string)
name = wanted_names[0]
stub_folder_name = name + '-stubs'
ios = recurse_find_python_folders_and_files(FolderIO(self._path))
file_ios = []
# 1. Search for modules in the current project
for folder_io, file_io in ios:
if file_io is None:
file_name = folder_io.get_base_name()
if file_name == name or file_name == stub_folder_name:
f = folder_io.get_file_io('__init__.py')
try:
m = load_module_from_path(inference_state, f).as_context()
except FileNotFoundError:
f = folder_io.get_file_io('__init__.pyi')
try:
m = load_module_from_path(inference_state, f).as_context()
except FileNotFoundError:
m = load_namespace_from_path(inference_state, folder_io).as_context()
else:
continue
else:
file_ios.append(file_io)
file_name = os.path.basename(file_io.path)
if file_name in (name + '.py', name + '.pyi'):
m = load_module_from_path(inference_state, file_io).as_context()
else:
continue
debug.dbg('Search of a specific module %s', m)
for x in search_in_module(
inference_state,
m,
names=[m.name],
wanted_type=wanted_type,
wanted_names=wanted_names,
complete=complete,
convert=True,
ignore_imports=True,
):
yield x # Python 2...
# 2. Search for identifiers in the project.
for module_context in search_in_file_ios(inference_state, file_ios, name):
names = get_module_names(module_context.tree_node, all_scopes=all_scopes)
names = [module_context.create_name(n) for n in names]
names = _remove_imports(names)
for x in search_in_module(
inference_state,
module_context,
names=names,
wanted_type=wanted_type,
wanted_names=wanted_names,
complete=complete,
ignore_imports=True,
):
yield x # Python 2...
# 3. Search for modules on sys.path
sys_path = [
p for p in self._get_sys_path(inference_state)
# Exclude folders that are handled by recursing of the Python
# folders.
if not p.startswith(self._path)
]
names = list(iter_module_names(inference_state, empty_module_context, sys_path))
for x in search_in_module(
inference_state,
empty_module_context,
names=names,
wanted_type=wanted_type,
wanted_names=wanted_names,
complete=complete,
convert=True,
):
yield x # Python 2...
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self._path)
def _is_potential_project(path):
for name in _CONTAINS_POTENTIAL_PROJECT:
if os.path.exists(os.path.join(path, name)):
return True
return False
def _is_django_path(directory):
""" Detects the path of the very well known Django library (if used) """
try:
with open(os.path.join(directory, 'manage.py'), 'rb') as f:
return b"DJANGO_SETTINGS_MODULE" in f.read()
except (FileNotFoundError, IsADirectoryError, PermissionError):
return False
def get_default_project(path=None):
"""
If a project is not defined by the user, Jedi tries to define a project by
itself as well as possible. Jedi traverses folders until it finds one of
the following:
1. A ``.jedi/config.json``
2. One of the following files: ``setup.py``, ``.git``, ``.hg``,
``requirements.txt`` and ``MANIFEST.in``.
"""
if path is None:
path = os.getcwd()
check = os.path.realpath(path)
probable_path = None
first_no_init_file = None
for dir in traverse_parents(check, include_current=True):
try:
return Project.load(dir)
except (FileNotFoundError, IsADirectoryError, PermissionError):
pass
except NotADirectoryError:
continue
if first_no_init_file is None:
if os.path.exists(os.path.join(dir, '__init__.py')):
# In the case that a __init__.py exists, it's in 99% just a
# Python package and the project sits at least one level above.
continue
else:
first_no_init_file = dir
if _is_django_path(dir):
project = Project(dir)
project._django = True
return project
if probable_path is None and _is_potential_project(dir):
probable_path = dir
if probable_path is not None:
# TODO search for setup.py etc
return Project(probable_path)
if first_no_init_file is not None:
return Project(first_no_init_file)
curdir = path if os.path.isdir(path) else os.path.dirname(path)
return Project(curdir)
def _remove_imports(names):
return [
n for n in names
if n.tree_name is None or n.api_type != 'module'
]

View file

@ -0,0 +1,249 @@
from os.path import dirname, basename, join, relpath
import os
import re
import difflib
from parso import split_lines
from jedi.api.exceptions import RefactoringError
EXPRESSION_PARTS = (
'or_test and_test not_test comparison '
'expr xor_expr and_expr shift_expr arith_expr term factor power atom_expr'
).split()
class ChangedFile(object):
def __init__(self, inference_state, from_path, to_path,
module_node, node_to_str_map):
self._inference_state = inference_state
self._from_path = from_path
self._to_path = to_path
self._module_node = module_node
self._node_to_str_map = node_to_str_map
def get_diff(self):
old_lines = split_lines(self._module_node.get_code(), keepends=True)
new_lines = split_lines(self.get_new_code(), keepends=True)
# Add a newline at the end if it's missing. Otherwise the diff will be
# very weird. A `diff -u file1 file2` would show the string:
#
# \ No newline at end of file
#
# This is not necessary IMO, because Jedi does not really play with
# newlines and the ending newline does not really matter in Python
# files. ~dave
if old_lines[-1] != '':
old_lines[-1] += '\n'
if new_lines[-1] != '':
new_lines[-1] += '\n'
project_path = self._inference_state.project.path
if self._from_path is None:
from_p = ''
else:
from_p = relpath(self._from_path, project_path)
if self._to_path is None:
to_p = ''
else:
to_p = relpath(self._to_path, project_path)
diff = difflib.unified_diff(
old_lines, new_lines,
fromfile=from_p,
tofile=to_p,
)
# Apparently there's a space at the end of the diff - for whatever
# reason.
return ''.join(diff).rstrip(' ')
def get_new_code(self):
return self._inference_state.grammar.refactor(self._module_node, self._node_to_str_map)
def apply(self):
if self._from_path is None:
raise RefactoringError(
'Cannot apply a refactoring on a Script with path=None'
)
with open(self._from_path, 'w', newline='') as f:
f.write(self.get_new_code())
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self._from_path)
class Refactoring(object):
def __init__(self, inference_state, file_to_node_changes, renames=()):
self._inference_state = inference_state
self._renames = renames
self._file_to_node_changes = file_to_node_changes
def get_changed_files(self):
"""
Returns a path to ``ChangedFile`` map.
"""
def calculate_to_path(p):
if p is None:
return p
for from_, to in renames:
if p.startswith(from_):
p = to + p[len(from_):]
return p
renames = self.get_renames()
return {
path: ChangedFile(
self._inference_state,
from_path=path,
to_path=calculate_to_path(path),
module_node=next(iter(map_)).get_root_node(),
node_to_str_map=map_
) for path, map_ in sorted(self._file_to_node_changes.items())
}
def get_renames(self):
"""
Files can be renamed in a refactoring.
Returns ``Iterable[Tuple[str, str]]``.
"""
return sorted(self._renames)
def get_diff(self):
text = ''
project_path = self._inference_state.project.path
for from_, to in self.get_renames():
text += 'rename from %s\nrename to %s\n' \
% (relpath(from_, project_path), relpath(to, project_path))
return text + ''.join(f.get_diff() for f in self.get_changed_files().values())
def apply(self):
"""
Applies the whole refactoring to the files, which includes renames.
"""
for f in self.get_changed_files().values():
f.apply()
for old, new in self.get_renames():
os.rename(old, new)
def _calculate_rename(path, new_name):
name = basename(path)
dir_ = dirname(path)
if name in ('__init__.py', '__init__.pyi'):
parent_dir = dirname(dir_)
return dir_, join(parent_dir, new_name)
ending = re.search(r'\.pyi?$', name).group(0)
return path, join(dir_, new_name + ending)
def rename(inference_state, definitions, new_name):
file_renames = set()
file_tree_name_map = {}
if not definitions:
raise RefactoringError("There is no name under the cursor")
for d in definitions:
tree_name = d._name.tree_name
if d.type == 'module' and tree_name is None:
file_renames.add(_calculate_rename(d.module_path, new_name))
else:
# This private access is ok in a way. It's not public to
# protect Jedi users from seeing it.
if tree_name is not None:
fmap = file_tree_name_map.setdefault(d.module_path, {})
fmap[tree_name] = tree_name.prefix + new_name
return Refactoring(inference_state, file_tree_name_map, file_renames)
def inline(inference_state, names):
if not names:
raise RefactoringError("There is no name under the cursor")
if any(n.api_type == 'module' for n in names):
raise RefactoringError("Cannot inline imports or modules")
if any(n.tree_name is None for n in names):
raise RefactoringError("Cannot inline builtins/extensions")
definitions = [n for n in names if n.tree_name.is_definition()]
if len(definitions) == 0:
raise RefactoringError("No definition found to inline")
if len(definitions) > 1:
raise RefactoringError("Cannot inline a name with multiple definitions")
if len(names) == 1:
raise RefactoringError("There are no references to this name")
tree_name = definitions[0].tree_name
expr_stmt = tree_name.get_definition()
if expr_stmt.type != 'expr_stmt':
type_ = dict(
funcdef='function',
classdef='class',
).get(expr_stmt.type, expr_stmt.type)
raise RefactoringError("Cannot inline a %s" % type_)
if len(expr_stmt.get_defined_names(include_setitem=True)) > 1:
raise RefactoringError("Cannot inline a statement with multiple definitions")
first_child = expr_stmt.children[1]
if first_child.type == 'annassign' and len(first_child.children) == 4:
first_child = first_child.children[2]
if first_child != '=':
if first_child.type == 'annassign':
raise RefactoringError(
'Cannot inline a statement that is defined by an annotation'
)
else:
raise RefactoringError(
'Cannot inline a statement with "%s"'
% first_child.get_code(include_prefix=False)
)
rhs = expr_stmt.get_rhs()
replace_code = rhs.get_code(include_prefix=False)
references = [n for n in names if not n.tree_name.is_definition()]
file_to_node_changes = {}
for name in references:
tree_name = name.tree_name
path = name.get_root_context().py__file__()
s = replace_code
if rhs.type == 'testlist_star_expr' \
or tree_name.parent.type in EXPRESSION_PARTS \
or tree_name.parent.type == 'trailer' \
and tree_name.parent.get_next_sibling() is not None:
s = '(' + replace_code + ')'
of_path = file_to_node_changes.setdefault(path, {})
n = tree_name
prefix = n.prefix
par = n.parent
if par.type == 'trailer' and par.children[0] == '.':
prefix = par.parent.children[0].prefix
n = par
for some_node in par.parent.children[:par.parent.children.index(par)]:
of_path[some_node] = ''
of_path[n] = prefix + s
path = definitions[0].get_root_context().py__file__()
changes = file_to_node_changes.setdefault(path, {})
changes[expr_stmt] = _remove_indent_of_prefix(expr_stmt.get_first_leaf().prefix)
next_leaf = expr_stmt.get_next_leaf()
# Most of the time we have to remove the newline at the end of the
# statement, but if there's a comment we might not need to.
if next_leaf.prefix.strip(' \t') == '' \
and (next_leaf.type == 'newline' or next_leaf == ';'):
changes[next_leaf] = ''
return Refactoring(inference_state, file_to_node_changes)
def _remove_indent_of_prefix(prefix):
r"""
Removes the last indentation of a prefix, e.g. " \n \n " becomes " \n \n".
"""
return ''.join(split_lines(prefix, keepends=True)[:-1])

View file

@ -0,0 +1,387 @@
from textwrap import dedent
from parso import split_lines
from jedi import debug
from jedi.api.exceptions import RefactoringError
from jedi.api.refactoring import Refactoring, EXPRESSION_PARTS
from jedi.common import indent_block
from jedi.parser_utils import function_is_classmethod, function_is_staticmethod
_DEFINITION_SCOPES = ('suite', 'file_input')
_VARIABLE_EXCTRACTABLE = EXPRESSION_PARTS + \
('atom testlist_star_expr testlist test lambdef lambdef_nocond '
'keyword name number string fstring').split()
def extract_variable(inference_state, path, module_node, name, pos, until_pos):
nodes = _find_nodes(module_node, pos, until_pos)
debug.dbg('Extracting nodes: %s', nodes)
is_expression, message = _is_expression_with_error(nodes)
if not is_expression:
raise RefactoringError(message)
generated_code = name + ' = ' + _expression_nodes_to_string(nodes)
file_to_node_changes = {path: _replace(nodes, name, generated_code, pos)}
return Refactoring(inference_state, file_to_node_changes)
def _is_expression_with_error(nodes):
"""
Returns a tuple (is_expression, error_string).
"""
if any(node.type == 'name' and node.is_definition() for node in nodes):
return False, 'Cannot extract a name that defines something'
if nodes[0].type not in _VARIABLE_EXCTRACTABLE:
return False, 'Cannot extract a "%s"' % nodes[0].type
return True, ''
def _find_nodes(module_node, pos, until_pos):
"""
Looks up a module and tries to find the appropriate amount of nodes that
are in there.
"""
start_node = module_node.get_leaf_for_position(pos, include_prefixes=True)
if until_pos is None:
if start_node.type == 'operator':
next_leaf = start_node.get_next_leaf()
if next_leaf is not None and next_leaf.start_pos == pos:
start_node = next_leaf
if _is_not_extractable_syntax(start_node):
start_node = start_node.parent
if start_node.parent.type == 'trailer':
start_node = start_node.parent.parent
while start_node.parent.type in EXPRESSION_PARTS:
start_node = start_node.parent
nodes = [start_node]
else:
# Get the next leaf if we are at the end of a leaf
if start_node.end_pos == pos:
next_leaf = start_node.get_next_leaf()
if next_leaf is not None:
start_node = next_leaf
# Some syntax is not exactable, just use its parent
if _is_not_extractable_syntax(start_node):
start_node = start_node.parent
# Find the end
end_leaf = module_node.get_leaf_for_position(until_pos, include_prefixes=True)
if end_leaf.start_pos > until_pos:
end_leaf = end_leaf.get_previous_leaf()
if end_leaf is None:
raise RefactoringError('Cannot extract anything from that')
parent_node = start_node
while parent_node.end_pos < end_leaf.end_pos:
parent_node = parent_node.parent
nodes = _remove_unwanted_expression_nodes(parent_node, pos, until_pos)
# If the user marks just a return statement, we return the expression
# instead of the whole statement, because the user obviously wants to
# extract that part.
if len(nodes) == 1 and start_node.type in ('return_stmt', 'yield_expr'):
return [nodes[0].children[1]]
return nodes
def _replace(nodes, expression_replacement, extracted, pos,
insert_before_leaf=None, remaining_prefix=None):
# Now try to replace the nodes found with a variable and move the code
# before the current statement.
definition = _get_parent_definition(nodes[0])
if insert_before_leaf is None:
insert_before_leaf = definition.get_first_leaf()
first_node_leaf = nodes[0].get_first_leaf()
lines = split_lines(insert_before_leaf.prefix, keepends=True)
if first_node_leaf is insert_before_leaf:
if remaining_prefix is not None:
# The remaining prefix has already been calculated.
lines[:-1] = remaining_prefix
lines[-1:-1] = [indent_block(extracted, lines[-1]) + '\n']
extracted_prefix = ''.join(lines)
replacement_dct = {}
if first_node_leaf is insert_before_leaf:
replacement_dct[nodes[0]] = extracted_prefix + expression_replacement
else:
if remaining_prefix is None:
p = first_node_leaf.prefix
else:
p = remaining_prefix + _get_indentation(nodes[0])
replacement_dct[nodes[0]] = p + expression_replacement
replacement_dct[insert_before_leaf] = extracted_prefix + insert_before_leaf.value
for node in nodes[1:]:
replacement_dct[node] = ''
return replacement_dct
def _expression_nodes_to_string(nodes):
return ''.join(n.get_code(include_prefix=i != 0) for i, n in enumerate(nodes))
def _suite_nodes_to_string(nodes, pos):
n = nodes[0]
prefix, part_of_code = _split_prefix_at(n.get_first_leaf(), pos[0] - 1)
code = part_of_code + n.get_code(include_prefix=False) \
+ ''.join(n.get_code() for n in nodes[1:])
return prefix, code
def _split_prefix_at(leaf, until_line):
"""
Returns a tuple of the leaf's prefix, split at the until_line
position.
"""
# second means the second returned part
second_line_count = leaf.start_pos[0] - until_line
lines = split_lines(leaf.prefix, keepends=True)
return ''.join(lines[:-second_line_count]), ''.join(lines[-second_line_count:])
def _get_indentation(node):
return split_lines(node.get_first_leaf().prefix)[-1]
def _get_parent_definition(node):
"""
Returns the statement where a node is defined.
"""
while node is not None:
if node.parent.type in _DEFINITION_SCOPES:
return node
node = node.parent
raise NotImplementedError('We should never even get here')
def _remove_unwanted_expression_nodes(parent_node, pos, until_pos):
"""
This function makes it so for `1 * 2 + 3` you can extract `2 + 3`, even
though it is not part of the expression.
"""
typ = parent_node.type
is_suite_part = typ in ('suite', 'file_input')
if typ in EXPRESSION_PARTS or is_suite_part:
nodes = parent_node.children
for i, n in enumerate(nodes):
if n.end_pos > pos:
start_index = i
if n.type == 'operator':
start_index -= 1
break
for i, n in reversed(list(enumerate(nodes))):
if n.start_pos < until_pos:
end_index = i
if n.type == 'operator':
end_index += 1
# Something like `not foo or bar` should not be cut after not
for n2 in nodes[i:]:
if _is_not_extractable_syntax(n2):
end_index += 1
else:
break
break
nodes = nodes[start_index:end_index + 1]
if not is_suite_part:
nodes[0:1] = _remove_unwanted_expression_nodes(nodes[0], pos, until_pos)
nodes[-1:] = _remove_unwanted_expression_nodes(nodes[-1], pos, until_pos)
return nodes
return [parent_node]
def _is_not_extractable_syntax(node):
return node.type == 'operator' \
or node.type == 'keyword' and node.value not in ('None', 'True', 'False')
def extract_function(inference_state, path, module_context, name, pos, until_pos):
nodes = _find_nodes(module_context.tree_node, pos, until_pos)
assert len(nodes)
is_expression, _ = _is_expression_with_error(nodes)
context = module_context.create_context(nodes[0])
is_bound_method = context.is_bound_method()
params, return_variables = list(_find_inputs_and_outputs(module_context, context, nodes))
# Find variables
# Is a class method / method
if context.is_module():
insert_before_leaf = None # Leaf will be determined later
else:
node = _get_code_insertion_node(context.tree_node, is_bound_method)
insert_before_leaf = node.get_first_leaf()
if is_expression:
code_block = 'return ' + _expression_nodes_to_string(nodes) + '\n'
remaining_prefix = None
has_ending_return_stmt = False
else:
has_ending_return_stmt = _is_node_ending_return_stmt(nodes[-1])
if not has_ending_return_stmt:
# Find the actually used variables (of the defined ones). If none are
# used (e.g. if the range covers the whole function), return the last
# defined variable.
return_variables = list(_find_needed_output_variables(
context,
nodes[0].parent,
nodes[-1].end_pos,
return_variables
)) or [return_variables[-1]] if return_variables else []
remaining_prefix, code_block = _suite_nodes_to_string(nodes, pos)
after_leaf = nodes[-1].get_next_leaf()
first, second = _split_prefix_at(after_leaf, until_pos[0])
code_block += first
code_block = dedent(code_block)
if not has_ending_return_stmt:
output_var_str = ', '.join(return_variables)
code_block += 'return ' + output_var_str + '\n'
# Check if we have to raise RefactoringError
_check_for_non_extractables(nodes[:-1] if has_ending_return_stmt else nodes)
decorator = ''
self_param = None
if is_bound_method:
if not function_is_staticmethod(context.tree_node):
function_param_names = context.get_value().get_param_names()
if len(function_param_names):
self_param = function_param_names[0].string_name
params = [p for p in params if p != self_param]
if function_is_classmethod(context.tree_node):
decorator = '@classmethod\n'
else:
code_block += '\n'
function_code = '%sdef %s(%s):\n%s' % (
decorator,
name,
', '.join(params if self_param is None else [self_param] + params),
indent_block(code_block)
)
function_call = '%s(%s)' % (
('' if self_param is None else self_param + '.') + name,
', '.join(params)
)
if is_expression:
replacement = function_call
else:
if has_ending_return_stmt:
replacement = 'return ' + function_call + '\n'
else:
replacement = output_var_str + ' = ' + function_call + '\n'
replacement_dct = _replace(nodes, replacement, function_code, pos,
insert_before_leaf, remaining_prefix)
if not is_expression:
replacement_dct[after_leaf] = second + after_leaf.value
file_to_node_changes = {path: replacement_dct}
return Refactoring(inference_state, file_to_node_changes)
def _check_for_non_extractables(nodes):
for n in nodes:
try:
children = n.children
except AttributeError:
if n.value == 'return':
raise RefactoringError(
'Can only extract return statements if they are at the end.')
if n.value == 'yield':
raise RefactoringError('Cannot extract yield statements.')
else:
_check_for_non_extractables(children)
def _is_name_input(module_context, names, first, last):
for name in names:
if name.api_type == 'param' or not name.parent_context.is_module():
if name.get_root_context() is not module_context:
return True
if name.start_pos is None or not (first <= name.start_pos < last):
return True
return False
def _find_inputs_and_outputs(module_context, context, nodes):
first = nodes[0].start_pos
last = nodes[-1].end_pos
inputs = []
outputs = []
for name in _find_non_global_names(nodes):
if name.is_definition():
if name not in outputs:
outputs.append(name.value)
else:
if name.value not in inputs:
name_definitions = context.goto(name, name.start_pos)
if not name_definitions \
or _is_name_input(module_context, name_definitions, first, last):
inputs.append(name.value)
# Check if outputs are really needed:
return inputs, outputs
def _find_non_global_names(nodes):
for node in nodes:
try:
children = node.children
except AttributeError:
if node.type == 'name':
yield node
else:
# We only want to check foo in foo.bar
if node.type == 'trailer' and node.children[0] == '.':
continue
for x in _find_non_global_names(children): # Python 2...
yield x
def _get_code_insertion_node(node, is_bound_method):
if not is_bound_method or function_is_staticmethod(node):
while node.parent.type != 'file_input':
node = node.parent
while node.parent.type in ('async_funcdef', 'decorated', 'async_stmt'):
node = node.parent
return node
def _find_needed_output_variables(context, search_node, at_least_pos, return_variables):
"""
Searches everything after at_least_pos in a node and checks if any of the
return_variables are used in there and returns those.
"""
for node in search_node.children:
if node.start_pos < at_least_pos:
continue
return_variables = set(return_variables)
for name in _find_non_global_names([node]):
if not name.is_definition() and name.value in return_variables:
return_variables.remove(name.value)
yield name.value
def _is_node_ending_return_stmt(node):
t = node.type
if t == 'simple_stmt':
return _is_node_ending_return_stmt(node.children[0])
return t == 'return_stmt'

View file

@ -0,0 +1,29 @@
"""
To use Jedi completion in Python interpreter, add the following in your shell
setup (e.g., ``.bashrc``). This works only on Linux/Mac, because readline is
not available on Windows. If you still want Jedi autocompletion in your REPL,
just use IPython instead::
export PYTHONSTARTUP="$(python -m jedi repl)"
Then you will be able to use Jedi completer in your Python interpreter::
$ python
Python 2.7.2+ (default, Jul 20 2012, 22:15:08)
[GCC 4.6.1] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import os
>>> os.path.join('a', 'b').split().in<TAB> # doctest: +SKIP
..dex ..sert
"""
import jedi.utils
from jedi import __version__ as __jedi_version__
print('REPL completion using Jedi %s' % __jedi_version__)
jedi.utils.setup_readline(fuzzy=False)
del jedi
# Note: try not to do many things here, as it will contaminate global
# namespace of the interpreter.

View file

@ -0,0 +1,109 @@
"""
This module is here for string completions. This means mostly stuff where
strings are returned, like `foo = dict(bar=3); foo["ba` would complete to
`"bar"]`.
It however does the same for numbers. The difference between string completions
and other completions is mostly that this module doesn't return defined
names in a module, but pretty much an arbitrary string.
"""
import re
from jedi._compatibility import unicode
from jedi.inference.names import AbstractArbitraryName
from jedi.inference.helpers import infer_call_of_leaf
from jedi.api.classes import Completion
from jedi.parser_utils import cut_value_at_position
_sentinel = object()
class StringName(AbstractArbitraryName):
api_type = u'string'
is_value_name = False
def complete_dict(module_context, code_lines, leaf, position, string, fuzzy):
bracket_leaf = leaf
if bracket_leaf != '[':
bracket_leaf = leaf.get_previous_leaf()
cut_end_quote = ''
if string:
cut_end_quote = get_quote_ending(string, code_lines, position, invert_result=True)
if bracket_leaf == '[':
if string is None and leaf is not bracket_leaf:
string = cut_value_at_position(leaf, position)
context = module_context.create_context(bracket_leaf)
before_bracket_leaf = bracket_leaf.get_previous_leaf()
if before_bracket_leaf.type in ('atom', 'trailer', 'name'):
values = infer_call_of_leaf(context, before_bracket_leaf)
return list(_completions_for_dicts(
module_context.inference_state,
values,
'' if string is None else string,
cut_end_quote,
fuzzy=fuzzy,
))
return []
def _completions_for_dicts(inference_state, dicts, literal_string, cut_end_quote, fuzzy):
for dict_key in sorted(_get_python_keys(dicts), key=lambda x: repr(x)):
dict_key_str = _create_repr_string(literal_string, dict_key)
if dict_key_str.startswith(literal_string):
name = StringName(inference_state, dict_key_str[:-len(cut_end_quote) or None])
yield Completion(
inference_state,
name,
stack=None,
like_name_length=len(literal_string),
is_fuzzy=fuzzy
)
def _create_repr_string(literal_string, dict_key):
if not isinstance(dict_key, (unicode, bytes)) or not literal_string:
return repr(dict_key)
r = repr(dict_key)
prefix, quote = _get_string_prefix_and_quote(literal_string)
if quote is None:
return r
if quote == r[0]:
return prefix + r
return prefix + quote + r[1:-1] + quote
def _get_python_keys(dicts):
for dct in dicts:
if dct.array_type == 'dict':
for key in dct.get_key_values():
dict_key = key.get_safe_value(default=_sentinel)
if dict_key is not _sentinel:
yield dict_key
def _get_string_prefix_and_quote(string):
match = re.match(r'(\w*)("""|\'{3}|"|\')', string)
if match is None:
return None, None
return match.group(1), match.group(2)
def _matches_quote_at_position(code_lines, quote, position):
string = code_lines[position[0] - 1][position[1]:position[1] + len(quote)]
return string == quote
def get_quote_ending(string, code_lines, position, invert_result=False):
_, quote = _get_string_prefix_and_quote(string)
if quote is None:
return ''
# Add a quote only if it's not already there.
if _matches_quote_at_position(code_lines, quote, position) != invert_result:
return ''
return quote

View file

@ -0,0 +1,114 @@
"""
This caching is very important for speed and memory optimizations. There's
nothing really spectacular, just some decorators. The following cache types are
available:
- ``time_cache`` can be used to cache something for just a limited time span,
which can be useful if there's user interaction and the user cannot react
faster than a certain time.
This module is one of the reasons why |jedi| is not thread-safe. As you can see
there are global variables, which are holding the cache information. Some of
these variables are being cleaned after every API usage.
"""
import time
from functools import wraps
from jedi import settings
from parso.cache import parser_cache
_time_caches = {}
def clear_time_caches(delete_all=False):
""" Jedi caches many things, that should be completed after each completion
finishes.
:param delete_all: Deletes also the cache that is normally not deleted,
like parser cache, which is important for faster parsing.
"""
global _time_caches
if delete_all:
for cache in _time_caches.values():
cache.clear()
parser_cache.clear()
else:
# normally just kill the expired entries, not all
for tc in _time_caches.values():
# check time_cache for expired entries
for key, (t, value) in list(tc.items()):
if t < time.time():
# delete expired entries
del tc[key]
def signature_time_cache(time_add_setting):
"""
This decorator works as follows: Call it with a setting and after that
use the function with a callable that returns the key.
But: This function is only called if the key is not available. After a
certain amount of time (`time_add_setting`) the cache is invalid.
If the given key is None, the function will not be cached.
"""
def _temp(key_func):
dct = {}
_time_caches[time_add_setting] = dct
def wrapper(*args, **kwargs):
generator = key_func(*args, **kwargs)
key = next(generator)
try:
expiry, value = dct[key]
if expiry > time.time():
return value
except KeyError:
pass
value = next(generator)
time_add = getattr(settings, time_add_setting)
if key is not None:
dct[key] = time.time() + time_add, value
return value
return wrapper
return _temp
def time_cache(seconds):
def decorator(func):
cache = {}
@wraps(func)
def wrapper(*args, **kwargs):
key = (args, frozenset(kwargs.items()))
try:
created, result = cache[key]
if time.time() < created + seconds:
return result
except KeyError:
pass
result = func(*args, **kwargs)
cache[key] = time.time(), result
return result
wrapper.clear_cache = lambda: cache.clear()
return wrapper
return decorator
def memoize_method(method):
"""A normal memoize function."""
@wraps(method)
def wrapper(self, *args, **kwargs):
cache_dict = self.__dict__.setdefault('_memoize_method_dct', {})
dct = cache_dict.setdefault(method, {})
key = (args, frozenset(kwargs.items()))
try:
return dct[key]
except KeyError:
result = method(self, *args, **kwargs)
dct[key] = result
return result
return wrapper

View file

@ -0,0 +1,36 @@
import os
from contextlib import contextmanager
def traverse_parents(path, include_current=False):
if not include_current:
path = os.path.dirname(path)
previous = None
while previous != path:
yield path
previous = path
path = os.path.dirname(path)
@contextmanager
def monkeypatch(obj, attribute_name, new_value):
"""
Like pytest's monkeypatch, but as a value manager.
"""
old_value = getattr(obj, attribute_name)
try:
setattr(obj, attribute_name, new_value)
yield
finally:
setattr(obj, attribute_name, old_value)
def indent_block(text, indention=' '):
"""This function indents a text block with a default of four spaces."""
temp = ''
while text and text[-1] == '\n':
temp += text[-1]
text = text[:-1]
lines = text.split('\n')
return '\n'.join(map(lambda s: indention + s, lines)) + temp

View file

@ -0,0 +1,143 @@
import os
import time
from contextlib import contextmanager
from jedi._compatibility import encoding, is_py3, u
_inited = False
def _lazy_colorama_init():
"""
Lazily init colorama if necessary, not to screw up stdout if debugging is
not enabled.
This version of the function does nothing.
"""
try:
if os.name == 'nt':
# Does not work on Windows, as pyreadline and colorama interfere
raise ImportError
else:
# Use colorama for nicer console output.
from colorama import Fore, init
from colorama import initialise
def _lazy_colorama_init(): # noqa: F811
"""
Lazily init colorama if necessary, not to screw up stdout is
debug not enabled.
This version of the function does init colorama.
"""
global _inited
if not _inited:
# pytest resets the stream at the end - causes troubles. Since
# after every output the stream is reset automatically we don't
# need this.
initialise.atexit_done = True
try:
init(strip=False)
except Exception:
# Colorama fails with initializing under vim and is buggy in
# version 0.3.6.
pass
_inited = True
except ImportError:
class Fore(object):
RED = ''
GREEN = ''
YELLOW = ''
MAGENTA = ''
RESET = ''
BLUE = ''
NOTICE = object()
WARNING = object()
SPEED = object()
enable_speed = False
enable_warning = False
enable_notice = False
# callback, interface: level, str
debug_function = None
_debug_indent = 0
_start_time = time.time()
def reset_time():
global _start_time, _debug_indent
_start_time = time.time()
_debug_indent = 0
def increase_indent(func):
"""Decorator for makin """
def wrapper(*args, **kwargs):
with increase_indent_cm():
return func(*args, **kwargs)
return wrapper
@contextmanager
def increase_indent_cm(title=None, color='MAGENTA'):
global _debug_indent
if title:
dbg('Start: ' + title, color=color)
_debug_indent += 1
try:
yield
finally:
_debug_indent -= 1
if title:
dbg('End: ' + title, color=color)
def dbg(message, *args, **kwargs):
""" Looks at the stack, to see if a debug message should be printed. """
# Python 2 compatibility, because it doesn't understand default args
color = kwargs.pop('color', 'GREEN')
assert color
if debug_function and enable_notice:
i = ' ' * _debug_indent
_lazy_colorama_init()
debug_function(color, i + 'dbg: ' + message % tuple(u(repr(a)) for a in args))
def warning(message, *args, **kwargs):
format = kwargs.pop('format', True)
assert not kwargs
if debug_function and enable_warning:
i = ' ' * _debug_indent
if format:
message = message % tuple(u(repr(a)) for a in args)
debug_function('RED', i + 'warning: ' + message)
def speed(name):
if debug_function and enable_speed:
now = time.time()
i = ' ' * _debug_indent
debug_function('YELLOW', i + 'speed: ' + '%s %s' % (name, now - _start_time))
def print_to_stdout(color, str_out):
"""
The default debug function that prints to standard out.
:param str color: A string that is an attribute of ``colorama.Fore``.
"""
col = getattr(Fore, color)
_lazy_colorama_init()
if not is_py3:
str_out = str_out.encode(encoding, 'replace')
print(col + str_out + Fore.RESET)
# debug_function = print_to_stdout

View file

@ -0,0 +1,83 @@
import os
from parso import file_io
class AbstractFolderIO(object):
def __init__(self, path):
self.path = path
def get_base_name(self):
raise NotImplementedError
def list(self):
raise NotImplementedError
def get_file_io(self, name):
raise NotImplementedError
def get_parent_folder(self):
raise NotImplementedError
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self.path)
class FolderIO(AbstractFolderIO):
def get_base_name(self):
return os.path.basename(self.path)
def list(self):
return os.listdir(self.path)
def get_file_io(self, name):
return FileIO(os.path.join(self.path, name))
def get_parent_folder(self):
return FolderIO(os.path.dirname(self.path))
def walk(self):
for root, dirs, files in os.walk(self.path):
root_folder_io = FolderIO(root)
original_folder_ios = [FolderIO(os.path.join(root, d)) for d in dirs]
modified_folder_ios = list(original_folder_ios)
yield (
root_folder_io,
modified_folder_ios,
[FileIO(os.path.join(root, f)) for f in files],
)
modified_iterator = iter(reversed(modified_folder_ios))
current = next(modified_iterator, None)
i = len(original_folder_ios)
for folder_io in reversed(original_folder_ios):
i -= 1 # Basically enumerate but reversed
if current is folder_io:
current = next(modified_iterator, None)
else:
del dirs[i]
class FileIOFolderMixin(object):
def get_parent_folder(self):
return FolderIO(os.path.dirname(self.path))
class ZipFileIO(file_io.KnownContentFileIO, FileIOFolderMixin):
"""For .zip and .egg archives"""
def __init__(self, path, code, zip_path):
super(ZipFileIO, self).__init__(path, code)
self._zip_path = zip_path
def get_last_modified(self):
try:
return os.path.getmtime(self._zip_path)
except OSError: # Python 3 would probably only need FileNotFoundError
return None
class FileIO(file_io.FileIO, FileIOFolderMixin):
pass
class KnownContentFileIO(file_io.KnownContentFileIO, FileIOFolderMixin):
pass

View file

@ -0,0 +1,197 @@
"""
Type inference of Python code in |jedi| is based on three assumptions:
* The code uses as least side effects as possible. Jedi understands certain
list/tuple/set modifications, but there's no guarantee that Jedi detects
everything (list.append in different modules for example).
* No magic is being used:
- metaclasses
- ``setattr()`` / ``__import__()``
- writing to ``globals()``, ``locals()``, ``object.__dict__``
* The programmer is not a total dick, e.g. like `this
<https://github.com/davidhalter/jedi/issues/24>`_ :-)
The actual algorithm is based on a principle I call lazy type inference. That
said, the typical entry point for static analysis is calling
``infer_expr_stmt``. There's separate logic for autocompletion in the API, the
inference_state is all about inferring an expression.
TODO this paragraph is not what jedi does anymore, it's similar, but not the
same.
Now you need to understand what follows after ``infer_expr_stmt``. Let's
make an example::
import datetime
datetime.date.toda# <-- cursor here
First of all, this module doesn't care about completion. It really just cares
about ``datetime.date``. At the end of the procedure ``infer_expr_stmt`` will
return the ``date`` class.
To *visualize* this (simplified):
- ``InferenceState.infer_expr_stmt`` doesn't do much, because there's no assignment.
- ``Context.infer_node`` cares for resolving the dotted path
- ``InferenceState.find_types`` searches for global definitions of datetime, which
it finds in the definition of an import, by scanning the syntax tree.
- Using the import logic, the datetime module is found.
- Now ``find_types`` is called again by ``infer_node`` to find ``date``
inside the datetime module.
Now what would happen if we wanted ``datetime.date.foo.bar``? Two more
calls to ``find_types``. However the second call would be ignored, because the
first one would return nothing (there's no foo attribute in ``date``).
What if the import would contain another ``ExprStmt`` like this::
from foo import bar
Date = bar.baz
Well... You get it. Just another ``infer_expr_stmt`` recursion. It's really
easy. Python can obviously get way more complicated then this. To understand
tuple assignments, list comprehensions and everything else, a lot more code had
to be written.
Jedi has been tested very well, so you can just start modifying code. It's best
to write your own test first for your "new" feature. Don't be scared of
breaking stuff. As long as the tests pass, you're most likely to be fine.
I need to mention now that lazy type inference is really good because it
only *inferes* what needs to be *inferred*. All the statements and modules
that are not used are just being ignored.
"""
import parso
from jedi.file_io import FileIO
from jedi import debug
from jedi import settings
from jedi.inference import imports
from jedi.inference import recursion
from jedi.inference.cache import inference_state_function_cache
from jedi.inference import helpers
from jedi.inference.names import TreeNameDefinition
from jedi.inference.base_value import ContextualizedNode, \
ValueSet, iterate_values
from jedi.inference.value import ClassValue, FunctionValue
from jedi.inference.syntax_tree import infer_expr_stmt, \
check_tuple_assignments, tree_name_to_values
from jedi.inference.imports import follow_error_node_imports_if_possible
from jedi.plugins import plugin_manager
class InferenceState(object):
def __init__(self, project, environment=None, script_path=None):
if environment is None:
environment = project.get_environment()
self.environment = environment
self.script_path = script_path
self.compiled_subprocess = environment.get_inference_state_subprocess(self)
self.grammar = environment.get_grammar()
self.latest_grammar = parso.load_grammar(version='3.7')
self.memoize_cache = {} # for memoize decorators
self.module_cache = imports.ModuleCache() # does the job of `sys.modules`.
self.stub_module_cache = {} # Dict[Tuple[str, ...], Optional[ModuleValue]]
self.compiled_cache = {} # see `inference.compiled.create()`
self.inferred_element_counts = {}
self.mixed_cache = {} # see `inference.compiled.mixed._create()`
self.analysis = []
self.dynamic_params_depth = 0
self.is_analysis = False
self.project = project
self.access_cache = {}
self.allow_descriptor_getattr = False
self.flow_analysis_enabled = True
self.reset_recursion_limitations()
def import_module(self, import_names, sys_path=None, prefer_stubs=True):
return imports.import_module_by_names(
self, import_names, sys_path, prefer_stubs=prefer_stubs)
@staticmethod
@plugin_manager.decorate()
def execute(value, arguments):
debug.dbg('execute: %s %s', value, arguments)
with debug.increase_indent_cm():
value_set = value.py__call__(arguments=arguments)
debug.dbg('execute result: %s in %s', value_set, value)
return value_set
@property
@inference_state_function_cache()
def builtins_module(self):
module_name = u'builtins'
if self.environment.version_info.major == 2:
module_name = u'__builtin__'
builtins_module, = self.import_module((module_name,), sys_path=())
return builtins_module
@property
@inference_state_function_cache()
def typing_module(self):
typing_module, = self.import_module((u'typing',))
return typing_module
def reset_recursion_limitations(self):
self.recursion_detector = recursion.RecursionDetector()
self.execution_recursion_detector = recursion.ExecutionRecursionDetector(self)
def get_sys_path(self, **kwargs):
"""Convenience function"""
return self.project._get_sys_path(self, **kwargs)
def infer(self, context, name):
def_ = name.get_definition(import_name_always=True)
if def_ is not None:
type_ = def_.type
is_classdef = type_ == 'classdef'
if is_classdef or type_ == 'funcdef':
if is_classdef:
c = ClassValue(self, context, name.parent)
else:
c = FunctionValue.from_context(context, name.parent)
return ValueSet([c])
if type_ == 'expr_stmt':
is_simple_name = name.parent.type not in ('power', 'trailer')
if is_simple_name:
return infer_expr_stmt(context, def_, name)
if type_ == 'for_stmt':
container_types = context.infer_node(def_.children[3])
cn = ContextualizedNode(context, def_.children[3])
for_types = iterate_values(container_types, cn)
n = TreeNameDefinition(context, name)
return check_tuple_assignments(n, for_types)
if type_ in ('import_from', 'import_name'):
return imports.infer_import(context, name)
if type_ == 'with_stmt':
return tree_name_to_values(self, context, name)
elif type_ == 'param':
return context.py__getattribute__(name.value, position=name.end_pos)
else:
result = follow_error_node_imports_if_possible(context, name)
if result is not None:
return result
return helpers.infer_call_of_leaf(context, name)
def parse_and_get_code(self, code=None, path=None, encoding='utf-8',
use_latest_grammar=False, file_io=None, **kwargs):
if code is None:
if file_io is None:
file_io = FileIO(path)
code = file_io.read()
# We cannot just use parso, because it doesn't use errors='replace'.
code = parso.python_bytes_to_unicode(code, encoding=encoding, errors='replace')
if len(code) > settings._cropped_file_size:
code = code[:settings._cropped_file_size]
grammar = self.latest_grammar if use_latest_grammar else self.grammar
return grammar.parse(code=code, path=path, file_io=file_io, **kwargs), code
def parse(self, *args, **kwargs):
return self.parse_and_get_code(*args, **kwargs)[0]

View file

@ -0,0 +1,217 @@
"""
Module for statical analysis.
"""
from parso.python import tree
from jedi._compatibility import force_unicode
from jedi import debug
from jedi.inference.helpers import is_string
CODES = {
'attribute-error': (1, AttributeError, 'Potential AttributeError.'),
'name-error': (2, NameError, 'Potential NameError.'),
'import-error': (3, ImportError, 'Potential ImportError.'),
'type-error-too-many-arguments': (4, TypeError, None),
'type-error-too-few-arguments': (5, TypeError, None),
'type-error-keyword-argument': (6, TypeError, None),
'type-error-multiple-values': (7, TypeError, None),
'type-error-star-star': (8, TypeError, None),
'type-error-star': (9, TypeError, None),
'type-error-operation': (10, TypeError, None),
'type-error-not-iterable': (11, TypeError, None),
'type-error-isinstance': (12, TypeError, None),
'type-error-not-subscriptable': (13, TypeError, None),
'value-error-too-many-values': (14, ValueError, None),
'value-error-too-few-values': (15, ValueError, None),
}
class Error(object):
def __init__(self, name, module_path, start_pos, message=None):
self.path = module_path
self._start_pos = start_pos
self.name = name
if message is None:
message = CODES[self.name][2]
self.message = message
@property
def line(self):
return self._start_pos[0]
@property
def column(self):
return self._start_pos[1]
@property
def code(self):
# The class name start
first = self.__class__.__name__[0]
return first + str(CODES[self.name][0])
def __unicode__(self):
return '%s:%s:%s: %s %s' % (self.path, self.line, self.column,
self.code, self.message)
def __str__(self):
return self.__unicode__()
def __eq__(self, other):
return (self.path == other.path and self.name == other.name
and self._start_pos == other._start_pos)
def __ne__(self, other):
return not self.__eq__(other)
def __hash__(self):
return hash((self.path, self._start_pos, self.name))
def __repr__(self):
return '<%s %s: %s@%s,%s>' % (self.__class__.__name__,
self.name, self.path,
self._start_pos[0], self._start_pos[1])
class Warning(Error):
pass
def add(node_context, error_name, node, message=None, typ=Error, payload=None):
exception = CODES[error_name][1]
if _check_for_exception_catch(node_context, node, exception, payload):
return
# TODO this path is probably not right
module_context = node_context.get_root_context()
module_path = module_context.py__file__()
issue_instance = typ(error_name, module_path, node.start_pos, message)
debug.warning(str(issue_instance), format=False)
node_context.inference_state.analysis.append(issue_instance)
return issue_instance
def _check_for_setattr(instance):
"""
Check if there's any setattr method inside an instance. If so, return True.
"""
module = instance.get_root_context()
node = module.tree_node
if node is None:
# If it's a compiled module or doesn't have a tree_node
return False
try:
stmt_names = node.get_used_names()['setattr']
except KeyError:
return False
return any(node.start_pos < n.start_pos < node.end_pos
# Check if it's a function called setattr.
and not (n.parent.type == 'funcdef' and n.parent.name == n)
for n in stmt_names)
def add_attribute_error(name_context, lookup_value, name):
message = ('AttributeError: %s has no attribute %s.' % (lookup_value, name))
# Check for __getattr__/__getattribute__ existance and issue a warning
# instead of an error, if that happens.
typ = Error
if lookup_value.is_instance() and not lookup_value.is_compiled():
# TODO maybe make a warning for __getattr__/__getattribute__
if _check_for_setattr(lookup_value):
typ = Warning
payload = lookup_value, name
add(name_context, 'attribute-error', name, message, typ, payload)
def _check_for_exception_catch(node_context, jedi_name, exception, payload=None):
"""
Checks if a jedi object (e.g. `Statement`) sits inside a try/catch and
doesn't count as an error (if equal to `exception`).
Also checks `hasattr` for AttributeErrors and uses the `payload` to compare
it.
Returns True if the exception was catched.
"""
def check_match(cls, exception):
if not cls.is_class():
return False
for python_cls in exception.mro():
if cls.py__name__() == python_cls.__name__ \
and cls.parent_context.is_builtins_module():
return True
return False
def check_try_for_except(obj, exception):
# Only nodes in try
iterator = iter(obj.children)
for branch_type in iterator:
next(iterator) # The colon
suite = next(iterator)
if branch_type == 'try' \
and not (branch_type.start_pos < jedi_name.start_pos <= suite.end_pos):
return False
for node in obj.get_except_clause_tests():
if node is None:
return True # An exception block that catches everything.
else:
except_classes = node_context.infer_node(node)
for cls in except_classes:
from jedi.inference.value import iterable
if isinstance(cls, iterable.Sequence) and \
cls.array_type == 'tuple':
# multiple exceptions
for lazy_value in cls.py__iter__():
for typ in lazy_value.infer():
if check_match(typ, exception):
return True
else:
if check_match(cls, exception):
return True
def check_hasattr(node, suite):
try:
assert suite.start_pos <= jedi_name.start_pos < suite.end_pos
assert node.type in ('power', 'atom_expr')
base = node.children[0]
assert base.type == 'name' and base.value == 'hasattr'
trailer = node.children[1]
assert trailer.type == 'trailer'
arglist = trailer.children[1]
assert arglist.type == 'arglist'
from jedi.inference.arguments import TreeArguments
args = TreeArguments(node_context.inference_state, node_context, arglist)
unpacked_args = list(args.unpack())
# Arguments should be very simple
assert len(unpacked_args) == 2
# Check name
key, lazy_value = unpacked_args[1]
names = list(lazy_value.infer())
assert len(names) == 1 and is_string(names[0])
assert force_unicode(names[0].get_safe_value()) == payload[1].value
# Check objects
key, lazy_value = unpacked_args[0]
objects = lazy_value.infer()
return payload[0] in objects
except AssertionError:
return False
obj = jedi_name
while obj is not None and not isinstance(obj, (tree.Function, tree.Class)):
if isinstance(obj, tree.Flow):
# try/except catch check
if obj.type == 'try_stmt' and check_try_for_except(obj, exception):
return True
# hasattr check
if exception == AttributeError and obj.type in ('if_stmt', 'while_stmt'):
if check_hasattr(obj.children[1], obj.children[3]):
return True
obj = obj.parent
return False

View file

@ -0,0 +1,343 @@
import re
from parso.python import tree
from jedi._compatibility import zip_longest
from jedi import debug
from jedi.inference.utils import PushBackIterator
from jedi.inference import analysis
from jedi.inference.lazy_value import LazyKnownValue, LazyKnownValues, \
LazyTreeValue, get_merged_lazy_value
from jedi.inference.names import ParamName, TreeNameDefinition, AnonymousParamName
from jedi.inference.base_value import NO_VALUES, ValueSet, ContextualizedNode
from jedi.inference.value import iterable
from jedi.inference.cache import inference_state_as_method_param_cache
def try_iter_content(types, depth=0):
"""Helper method for static analysis."""
if depth > 10:
# It's possible that a loop has references on itself (especially with
# CompiledValue). Therefore don't loop infinitely.
return
for typ in types:
try:
f = typ.py__iter__
except AttributeError:
pass
else:
for lazy_value in f():
try_iter_content(lazy_value.infer(), depth + 1)
class ParamIssue(Exception):
pass
def repack_with_argument_clinic(clinic_string):
"""
Transforms a function or method with arguments to the signature that is
given as an argument clinic notation.
Argument clinic is part of CPython and used for all the functions that are
implemented in C (Python 3.7):
str.split.__text_signature__
# Results in: '($self, /, sep=None, maxsplit=-1)'
"""
def decorator(func):
def wrapper(value, arguments):
try:
args = tuple(iterate_argument_clinic(
value.inference_state,
arguments,
clinic_string,
))
except ParamIssue:
return NO_VALUES
else:
return func(value, *args)
return wrapper
return decorator
def iterate_argument_clinic(inference_state, arguments, clinic_string):
"""Uses a list with argument clinic information (see PEP 436)."""
clinic_args = list(_parse_argument_clinic(clinic_string))
iterator = PushBackIterator(arguments.unpack())
for i, (name, optional, allow_kwargs, stars) in enumerate(clinic_args):
if stars == 1:
lazy_values = []
for key, argument in iterator:
if key is not None:
iterator.push_back((key, argument))
break
lazy_values.append(argument)
yield ValueSet([iterable.FakeTuple(inference_state, lazy_values)])
lazy_values
continue
elif stars == 2:
raise NotImplementedError()
key, argument = next(iterator, (None, None))
if key is not None:
debug.warning('Keyword arguments in argument clinic are currently not supported.')
raise ParamIssue
if argument is None and not optional:
debug.warning('TypeError: %s expected at least %s arguments, got %s',
name, len(clinic_args), i)
raise ParamIssue
value_set = NO_VALUES if argument is None else argument.infer()
if not value_set and not optional:
# For the stdlib we always want values. If we don't get them,
# that's ok, maybe something is too hard to resolve, however,
# we will not proceed with the type inference of that function.
debug.warning('argument_clinic "%s" not resolvable.', name)
raise ParamIssue
yield value_set
def _parse_argument_clinic(string):
allow_kwargs = False
optional = False
while string:
# Optional arguments have to begin with a bracket. And should always be
# at the end of the arguments. This is therefore not a proper argument
# clinic implementation. `range()` for exmple allows an optional start
# value at the beginning.
match = re.match(r'(?:(?:(\[),? ?|, ?|)(\**\w+)|, ?/)\]*', string)
string = string[len(match.group(0)):]
if not match.group(2): # A slash -> allow named arguments
allow_kwargs = True
continue
optional = optional or bool(match.group(1))
word = match.group(2)
stars = word.count('*')
word = word[stars:]
yield (word, optional, allow_kwargs, stars)
if stars:
allow_kwargs = True
class _AbstractArgumentsMixin(object):
def unpack(self, funcdef=None):
raise NotImplementedError
def get_calling_nodes(self):
return []
class AbstractArguments(_AbstractArgumentsMixin):
context = None
argument_node = None
trailer = None
def unpack_arglist(arglist):
if arglist is None:
return
# Allow testlist here as well for Python2's class inheritance
# definitions.
if not (arglist.type in ('arglist', 'testlist') or (
# in python 3.5 **arg is an argument, not arglist
arglist.type == 'argument' and arglist.children[0] in ('*', '**'))):
yield 0, arglist
return
iterator = iter(arglist.children)
for child in iterator:
if child == ',':
continue
elif child in ('*', '**'):
c = next(iterator, None)
assert c is not None
yield len(child.value), c
elif child.type == 'argument' and \
child.children[0] in ('*', '**'):
assert len(child.children) == 2
yield len(child.children[0].value), child.children[1]
else:
yield 0, child
class TreeArguments(AbstractArguments):
def __init__(self, inference_state, context, argument_node, trailer=None):
"""
:param argument_node: May be an argument_node or a list of nodes.
"""
self.argument_node = argument_node
self.context = context
self._inference_state = inference_state
self.trailer = trailer # Can be None, e.g. in a class definition.
@classmethod
@inference_state_as_method_param_cache()
def create_cached(cls, *args, **kwargs):
return cls(*args, **kwargs)
def unpack(self, funcdef=None):
named_args = []
for star_count, el in unpack_arglist(self.argument_node):
if star_count == 1:
arrays = self.context.infer_node(el)
iterators = [_iterate_star_args(self.context, a, el, funcdef)
for a in arrays]
for values in list(zip_longest(*iterators)):
# TODO zip_longest yields None, that means this would raise
# an exception?
yield None, get_merged_lazy_value(
[v for v in values if v is not None]
)
elif star_count == 2:
arrays = self.context.infer_node(el)
for dct in arrays:
for key, values in _star_star_dict(self.context, dct, el, funcdef):
yield key, values
else:
if el.type == 'argument':
c = el.children
if len(c) == 3: # Keyword argument.
named_args.append((c[0].value, LazyTreeValue(self.context, c[2]),))
else: # Generator comprehension.
# Include the brackets with the parent.
sync_comp_for = el.children[1]
if sync_comp_for.type == 'comp_for':
sync_comp_for = sync_comp_for.children[1]
comp = iterable.GeneratorComprehension(
self._inference_state,
defining_context=self.context,
sync_comp_for_node=sync_comp_for,
entry_node=el.children[0],
)
yield None, LazyKnownValue(comp)
else:
yield None, LazyTreeValue(self.context, el)
# Reordering arguments is necessary, because star args sometimes appear
# after named argument, but in the actual order it's prepended.
for named_arg in named_args:
yield named_arg
def _as_tree_tuple_objects(self):
for star_count, argument in unpack_arglist(self.argument_node):
default = None
if argument.type == 'argument':
if len(argument.children) == 3: # Keyword argument.
argument, default = argument.children[::2]
yield argument, default, star_count
def iter_calling_names_with_star(self):
for name, default, star_count in self._as_tree_tuple_objects():
# TODO this function is a bit strange. probably refactor?
if not star_count or not isinstance(name, tree.Name):
continue
yield TreeNameDefinition(self.context, name)
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self.argument_node)
def get_calling_nodes(self):
old_arguments_list = []
arguments = self
while arguments not in old_arguments_list:
if not isinstance(arguments, TreeArguments):
break
old_arguments_list.append(arguments)
for calling_name in reversed(list(arguments.iter_calling_names_with_star())):
names = calling_name.goto()
if len(names) != 1:
break
if isinstance(names[0], AnonymousParamName):
# Dynamic parameters should not have calling nodes, because
# they are dynamic and extremely random.
return []
if not isinstance(names[0], ParamName):
break
executed_param_name = names[0].get_executed_param_name()
arguments = executed_param_name.arguments
break
if arguments.argument_node is not None:
return [ContextualizedNode(arguments.context, arguments.argument_node)]
if arguments.trailer is not None:
return [ContextualizedNode(arguments.context, arguments.trailer)]
return []
class ValuesArguments(AbstractArguments):
def __init__(self, values_list):
self._values_list = values_list
def unpack(self, funcdef=None):
for values in self._values_list:
yield None, LazyKnownValues(values)
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self._values_list)
class TreeArgumentsWrapper(_AbstractArgumentsMixin):
def __init__(self, arguments):
self._wrapped_arguments = arguments
@property
def context(self):
return self._wrapped_arguments.context
@property
def argument_node(self):
return self._wrapped_arguments.argument_node
@property
def trailer(self):
return self._wrapped_arguments.trailer
def unpack(self, func=None):
raise NotImplementedError
def get_calling_nodes(self):
return self._wrapped_arguments.get_calling_nodes()
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self._wrapped_arguments)
def _iterate_star_args(context, array, input_node, funcdef=None):
if not array.py__getattribute__('__iter__'):
if funcdef is not None:
# TODO this funcdef should not be needed.
m = "TypeError: %s() argument after * must be a sequence, not %s" \
% (funcdef.name.value, array)
analysis.add(context, 'type-error-star', input_node, message=m)
try:
iter_ = array.py__iter__
except AttributeError:
pass
else:
for lazy_value in iter_():
yield lazy_value
def _star_star_dict(context, array, input_node, funcdef):
from jedi.inference.value.instance import CompiledInstance
if isinstance(array, CompiledInstance) and array.name.string_name == 'dict':
# For now ignore this case. In the future add proper iterators and just
# make one call without crazy isinstance checks.
return {}
elif isinstance(array, iterable.Sequence) and array.array_type == 'dict':
return array.exact_key_items()
else:
if funcdef is not None:
m = "TypeError: %s argument after ** must be a mapping, not %s" \
% (funcdef.name.value, array)
analysis.add(context, 'type-error-star-star', input_node, message=m)
return {}

View file

@ -0,0 +1,556 @@
"""
Values are the "values" that Python would return. However Values are at the
same time also the "values" that a user is currently sitting in.
A ValueSet is typically used to specify the return of a function or any other
static analysis operation. In jedi there are always multiple returns and not
just one.
"""
from functools import reduce
from operator import add
from parso.python.tree import Name
from jedi import debug
from jedi._compatibility import zip_longest, unicode
from jedi.parser_utils import clean_scope_docstring
from jedi.inference.helpers import SimpleGetItemNotFound
from jedi.inference.utils import safe_property
from jedi.inference.cache import inference_state_as_method_param_cache
from jedi.cache import memoize_method
sentinel = object()
class HelperValueMixin(object):
def get_root_context(self):
value = self
if value.parent_context is None:
return value.as_context()
while True:
if value.parent_context is None:
return value
value = value.parent_context
def execute(self, arguments):
return self.inference_state.execute(self, arguments=arguments)
def execute_with_values(self, *value_list):
from jedi.inference.arguments import ValuesArguments
arguments = ValuesArguments([ValueSet([value]) for value in value_list])
return self.inference_state.execute(self, arguments)
def execute_annotation(self):
return self.execute_with_values()
def gather_annotation_classes(self):
return ValueSet([self])
def merge_types_of_iterate(self, contextualized_node=None, is_async=False):
return ValueSet.from_sets(
lazy_value.infer()
for lazy_value in self.iterate(contextualized_node, is_async)
)
def _get_value_filters(self, name_or_str):
origin_scope = name_or_str if isinstance(name_or_str, Name) else None
for f in self.get_filters(origin_scope=origin_scope):
yield f
# This covers the case where a stub files are incomplete.
if self.is_stub():
from jedi.inference.gradual.conversion import convert_values
for c in convert_values(ValueSet({self})):
for f in c.get_filters():
yield f
def goto(self, name_or_str, name_context=None, analysis_errors=True):
from jedi.inference import finder
filters = self._get_value_filters(name_or_str)
names = finder.filter_name(filters, name_or_str)
debug.dbg('context.goto %s in (%s): %s', name_or_str, self, names)
return names
def py__getattribute__(self, name_or_str, name_context=None, position=None,
analysis_errors=True):
"""
:param position: Position of the last statement -> tuple of line, column
"""
if name_context is None:
name_context = self
names = self.goto(name_or_str, name_context, analysis_errors)
values = ValueSet.from_sets(name.infer() for name in names)
if not values:
n = name_or_str.value if isinstance(name_or_str, Name) else name_or_str
values = self.py__getattribute__alternatives(n)
if not names and not values and analysis_errors:
if isinstance(name_or_str, Name):
from jedi.inference import analysis
analysis.add_attribute_error(
name_context, self, name_or_str)
debug.dbg('context.names_to_types: %s -> %s', names, values)
return values
def py__await__(self):
await_value_set = self.py__getattribute__(u"__await__")
if not await_value_set:
debug.warning('Tried to run __await__ on value %s', self)
return await_value_set.execute_with_values()
def py__name__(self):
return self.name.string_name
def iterate(self, contextualized_node=None, is_async=False):
debug.dbg('iterate %s', self)
if is_async:
from jedi.inference.lazy_value import LazyKnownValues
# TODO if no __aiter__ values are there, error should be:
# TypeError: 'async for' requires an object with __aiter__ method, got int
return iter([
LazyKnownValues(
self.py__getattribute__('__aiter__').execute_with_values()
.py__getattribute__('__anext__').execute_with_values()
.py__getattribute__('__await__').execute_with_values()
.py__stop_iteration_returns()
) # noqa
])
return self.py__iter__(contextualized_node)
def is_sub_class_of(self, class_value):
with debug.increase_indent_cm('subclass matching of %s <=> %s' % (self, class_value),
color='BLUE'):
for cls in self.py__mro__():
if cls.is_same_class(class_value):
debug.dbg('matched subclass True', color='BLUE')
return True
debug.dbg('matched subclass False', color='BLUE')
return False
def is_same_class(self, class2):
# Class matching should prefer comparisons that are not this function.
if type(class2).is_same_class != HelperValueMixin.is_same_class:
return class2.is_same_class(self)
return self == class2
@memoize_method
def as_context(self, *args, **kwargs):
return self._as_context(*args, **kwargs)
class Value(HelperValueMixin):
"""
To be implemented by subclasses.
"""
tree_node = None
# Possible values: None, tuple, list, dict and set. Here to deal with these
# very important containers.
array_type = None
api_type = 'not_defined_please_report_bug'
def __init__(self, inference_state, parent_context=None):
self.inference_state = inference_state
self.parent_context = parent_context
def py__getitem__(self, index_value_set, contextualized_node):
from jedi.inference import analysis
# TODO this value is probably not right.
analysis.add(
contextualized_node.context,
'type-error-not-subscriptable',
contextualized_node.node,
message="TypeError: '%s' object is not subscriptable" % self
)
return NO_VALUES
def py__simple_getitem__(self, index):
raise SimpleGetItemNotFound
def py__iter__(self, contextualized_node=None):
if contextualized_node is not None:
from jedi.inference import analysis
analysis.add(
contextualized_node.context,
'type-error-not-iterable',
contextualized_node.node,
message="TypeError: '%s' object is not iterable" % self)
return iter([])
def py__next__(self, contextualized_node=None):
return self.py__iter__(contextualized_node)
def get_signatures(self):
return []
def is_class(self):
return False
def is_class_mixin(self):
return False
def is_instance(self):
return False
def is_function(self):
return False
def is_module(self):
return False
def is_namespace(self):
return False
def is_compiled(self):
return False
def is_bound_method(self):
return False
def is_builtins_module(self):
return False
def py__bool__(self):
"""
Since Wrapper is a super class for classes, functions and modules,
the return value will always be true.
"""
return True
def py__doc__(self):
try:
self.tree_node.get_doc_node
except AttributeError:
return ''
else:
return clean_scope_docstring(self.tree_node)
def get_safe_value(self, default=sentinel):
if default is sentinel:
raise ValueError("There exists no safe value for value %s" % self)
return default
def execute_operation(self, other, operator):
debug.warning("%s not possible between %s and %s", operator, self, other)
return NO_VALUES
def py__call__(self, arguments):
debug.warning("no execution possible %s", self)
return NO_VALUES
def py__stop_iteration_returns(self):
debug.warning("Not possible to return the stop iterations of %s", self)
return NO_VALUES
def py__getattribute__alternatives(self, name_or_str):
"""
For now a way to add values in cases like __getattr__.
"""
return NO_VALUES
def py__get__(self, instance, class_value):
debug.warning("No __get__ defined on %s", self)
return ValueSet([self])
def py__get__on_class(self, calling_instance, instance, class_value):
return NotImplemented
def get_qualified_names(self):
# Returns Optional[Tuple[str, ...]]
return None
def is_stub(self):
# The root value knows if it's a stub or not.
return self.parent_context.is_stub()
def _as_context(self):
raise NotImplementedError('Not all values need to be converted to contexts: %s', self)
@property
def name(self):
raise NotImplementedError
def get_type_hint(self, add_class_info=True):
return None
def infer_type_vars(self, value_set):
"""
When the current instance represents a type annotation, this method
tries to find information about undefined type vars and returns a dict
from type var name to value set.
This is for example important to understand what `iter([1])` returns.
According to typeshed, `iter` returns an `Iterator[_T]`:
def iter(iterable: Iterable[_T]) -> Iterator[_T]: ...
This functions would generate `int` for `_T` in this case, because it
unpacks the `Iterable`.
Parameters
----------
`self`: represents the annotation of the current parameter to infer the
value for. In the above example, this would initially be the
`Iterable[_T]` of the `iterable` parameter and then, when recursing,
just the `_T` generic parameter.
`value_set`: represents the actual argument passed to the parameter
we're inferrined for, or (for recursive calls) their types. In the
above example this would first be the representation of the list
`[1]` and then, when recursing, just of `1`.
"""
return {}
def iterate_values(values, contextualized_node=None, is_async=False):
"""
Calls `iterate`, on all values but ignores the ordering and just returns
all values that the iterate functions yield.
"""
return ValueSet.from_sets(
lazy_value.infer()
for lazy_value in values.iterate(contextualized_node, is_async=is_async)
)
class _ValueWrapperBase(HelperValueMixin):
@safe_property
def name(self):
from jedi.inference.names import ValueName
wrapped_name = self._wrapped_value.name
if wrapped_name.tree_name is not None:
return ValueName(self, wrapped_name.tree_name)
else:
from jedi.inference.compiled import CompiledValueName
return CompiledValueName(self, wrapped_name.string_name)
@classmethod
@inference_state_as_method_param_cache()
def create_cached(cls, inference_state, *args, **kwargs):
return cls(*args, **kwargs)
def __getattr__(self, name):
assert name != '_wrapped_value', 'Problem with _get_wrapped_value'
return getattr(self._wrapped_value, name)
class LazyValueWrapper(_ValueWrapperBase):
@safe_property
@memoize_method
def _wrapped_value(self):
with debug.increase_indent_cm('Resolve lazy value wrapper'):
return self._get_wrapped_value()
def __repr__(self):
return '<%s>' % (self.__class__.__name__)
def _get_wrapped_value(self):
raise NotImplementedError
class ValueWrapper(_ValueWrapperBase):
def __init__(self, wrapped_value):
self._wrapped_value = wrapped_value
def __repr__(self):
return '%s(%s)' % (self.__class__.__name__, self._wrapped_value)
class TreeValue(Value):
def __init__(self, inference_state, parent_context, tree_node):
super(TreeValue, self).__init__(inference_state, parent_context)
self.tree_node = tree_node
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self.tree_node)
class ContextualizedNode(object):
def __init__(self, context, node):
self.context = context
self.node = node
def get_root_context(self):
return self.context.get_root_context()
def infer(self):
return self.context.infer_node(self.node)
def __repr__(self):
return '<%s: %s in %s>' % (self.__class__.__name__, self.node, self.context)
def _getitem(value, index_values, contextualized_node):
# The actual getitem call.
result = NO_VALUES
unused_values = set()
for index_value in index_values:
index = index_value.get_safe_value(default=None)
if type(index) in (float, int, str, unicode, slice, bytes):
try:
result |= value.py__simple_getitem__(index)
continue
except SimpleGetItemNotFound:
pass
unused_values.add(index_value)
# The index was somehow not good enough or simply a wrong type.
# Therefore we now iterate through all the values and just take
# all results.
if unused_values or not index_values:
result |= value.py__getitem__(
ValueSet(unused_values),
contextualized_node
)
debug.dbg('py__getitem__ result: %s', result)
return result
class ValueSet(object):
def __init__(self, iterable):
self._set = frozenset(iterable)
for value in iterable:
assert not isinstance(value, ValueSet)
@classmethod
def _from_frozen_set(cls, frozenset_):
self = cls.__new__(cls)
self._set = frozenset_
return self
@classmethod
def from_sets(cls, sets):
"""
Used to work with an iterable of set.
"""
aggregated = set()
for set_ in sets:
if isinstance(set_, ValueSet):
aggregated |= set_._set
else:
aggregated |= frozenset(set_)
return cls._from_frozen_set(frozenset(aggregated))
def __or__(self, other):
return self._from_frozen_set(self._set | other._set)
def __and__(self, other):
return self._from_frozen_set(self._set & other._set)
def __iter__(self):
for element in self._set:
yield element
def __bool__(self):
return bool(self._set)
def __len__(self):
return len(self._set)
def __repr__(self):
return 'S{%s}' % (', '.join(str(s) for s in self._set))
def filter(self, filter_func):
return self.__class__(filter(filter_func, self._set))
def __getattr__(self, name):
def mapper(*args, **kwargs):
return self.from_sets(
getattr(value, name)(*args, **kwargs)
for value in self._set
)
return mapper
def __eq__(self, other):
return self._set == other._set
def __ne__(self, other):
return not self.__eq__(other)
def __hash__(self):
return hash(self._set)
def py__class__(self):
return ValueSet(c.py__class__() for c in self._set)
def iterate(self, contextualized_node=None, is_async=False):
from jedi.inference.lazy_value import get_merged_lazy_value
type_iters = [c.iterate(contextualized_node, is_async=is_async) for c in self._set]
for lazy_values in zip_longest(*type_iters):
yield get_merged_lazy_value(
[l for l in lazy_values if l is not None]
)
def execute(self, arguments):
return ValueSet.from_sets(c.inference_state.execute(c, arguments) for c in self._set)
def execute_with_values(self, *args, **kwargs):
return ValueSet.from_sets(c.execute_with_values(*args, **kwargs) for c in self._set)
def goto(self, *args, **kwargs):
return reduce(add, [c.goto(*args, **kwargs) for c in self._set], [])
def py__getattribute__(self, *args, **kwargs):
return ValueSet.from_sets(c.py__getattribute__(*args, **kwargs) for c in self._set)
def get_item(self, *args, **kwargs):
return ValueSet.from_sets(_getitem(c, *args, **kwargs) for c in self._set)
def try_merge(self, function_name):
value_set = self.__class__([])
for c in self._set:
try:
method = getattr(c, function_name)
except AttributeError:
pass
else:
value_set |= method()
return value_set
def gather_annotation_classes(self):
return ValueSet.from_sets([c.gather_annotation_classes() for c in self._set])
def get_signatures(self):
return [sig for c in self._set for sig in c.get_signatures()]
def get_type_hint(self, add_class_info=True):
t = [v.get_type_hint(add_class_info=add_class_info) for v in self._set]
type_hints = sorted(filter(None, t))
if len(type_hints) == 1:
return type_hints[0]
optional = 'None' in type_hints
if optional:
type_hints.remove('None')
if len(type_hints) == 0:
return None
elif len(type_hints) == 1:
s = type_hints[0]
else:
s = 'Union[%s]' % ', '.join(type_hints)
if optional:
s = 'Optional[%s]' % s
return s
def infer_type_vars(self, value_set):
# Circular
from jedi.inference.gradual.annotation import merge_type_var_dicts
type_var_dict = {}
for value in self._set:
merge_type_var_dicts(
type_var_dict,
value.infer_type_vars(value_set),
)
return type_var_dict
NO_VALUES = ValueSet([])
def iterator_to_value_set(func):
def wrapper(*args, **kwargs):
return ValueSet(func(*args, **kwargs))
return wrapper

View file

@ -0,0 +1,126 @@
"""
- the popular ``_memoize_default`` works like a typical memoize and returns the
default otherwise.
- ``CachedMetaClass`` uses ``_memoize_default`` to do the same with classes.
"""
from functools import wraps
from jedi import debug
_NO_DEFAULT = object()
_RECURSION_SENTINEL = object()
def _memoize_default(default=_NO_DEFAULT, inference_state_is_first_arg=False,
second_arg_is_inference_state=False):
""" This is a typical memoization decorator, BUT there is one difference:
To prevent recursion it sets defaults.
Preventing recursion is in this case the much bigger use than speed. I
don't think, that there is a big speed difference, but there are many cases
where recursion could happen (think about a = b; b = a).
"""
def func(function):
def wrapper(obj, *args, **kwargs):
# TODO These checks are kind of ugly and slow.
if inference_state_is_first_arg:
cache = obj.memoize_cache
elif second_arg_is_inference_state:
cache = args[0].memoize_cache # needed for meta classes
else:
cache = obj.inference_state.memoize_cache
try:
memo = cache[function]
except KeyError:
cache[function] = memo = {}
key = (obj, args, frozenset(kwargs.items()))
if key in memo:
return memo[key]
else:
if default is not _NO_DEFAULT:
memo[key] = default
rv = function(obj, *args, **kwargs)
memo[key] = rv
return rv
return wrapper
return func
def inference_state_function_cache(default=_NO_DEFAULT):
def decorator(func):
return _memoize_default(default=default, inference_state_is_first_arg=True)(func)
return decorator
def inference_state_method_cache(default=_NO_DEFAULT):
def decorator(func):
return _memoize_default(default=default)(func)
return decorator
def inference_state_as_method_param_cache():
def decorator(call):
return _memoize_default(second_arg_is_inference_state=True)(call)
return decorator
class CachedMetaClass(type):
"""
This is basically almost the same than the decorator above, it just caches
class initializations. Either you do it this way or with decorators, but
with decorators you lose class access (isinstance, etc).
"""
@inference_state_as_method_param_cache()
def __call__(self, *args, **kwargs):
return super(CachedMetaClass, self).__call__(*args, **kwargs)
def inference_state_method_generator_cache():
"""
This is a special memoizer. It memoizes generators and also checks for
recursion errors and returns no further iterator elemends in that case.
"""
def func(function):
@wraps(function)
def wrapper(obj, *args, **kwargs):
cache = obj.inference_state.memoize_cache
try:
memo = cache[function]
except KeyError:
cache[function] = memo = {}
key = (obj, args, frozenset(kwargs.items()))
if key in memo:
actual_generator, cached_lst = memo[key]
else:
actual_generator = function(obj, *args, **kwargs)
cached_lst = []
memo[key] = actual_generator, cached_lst
i = 0
while True:
try:
next_element = cached_lst[i]
if next_element is _RECURSION_SENTINEL:
debug.warning('Found a generator recursion for %s' % obj)
# This means we have hit a recursion.
return
except IndexError:
cached_lst.append(_RECURSION_SENTINEL)
next_element = next(actual_generator, None)
if next_element is None:
cached_lst.pop()
return
cached_lst[-1] = next_element
yield next_element
i += 1
return wrapper
return func

View file

@ -0,0 +1,68 @@
from jedi._compatibility import unicode
from jedi.inference.compiled.value import CompiledValue, CompiledName, \
CompiledValueFilter, CompiledValueName, create_from_access_path
from jedi.inference.base_value import LazyValueWrapper
def builtin_from_name(inference_state, string):
typing_builtins_module = inference_state.builtins_module
if string in ('None', 'True', 'False'):
builtins, = typing_builtins_module.non_stub_value_set
filter_ = next(builtins.get_filters())
else:
filter_ = next(typing_builtins_module.get_filters())
name, = filter_.get(string)
value, = name.infer()
return value
class ExactValue(LazyValueWrapper):
"""
This class represents exact values, that makes operations like additions
and exact boolean values possible, while still being a "normal" stub.
"""
def __init__(self, compiled_value):
self.inference_state = compiled_value.inference_state
self._compiled_value = compiled_value
def __getattribute__(self, name):
if name in ('get_safe_value', 'execute_operation', 'access_handle',
'negate', 'py__bool__', 'is_compiled'):
return getattr(self._compiled_value, name)
return super(ExactValue, self).__getattribute__(name)
def _get_wrapped_value(self):
instance, = builtin_from_name(
self.inference_state, self._compiled_value.name.string_name).execute_with_values()
return instance
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self._compiled_value)
def create_simple_object(inference_state, obj):
"""
Only allows creations of objects that are easily picklable across Python
versions.
"""
assert type(obj) in (int, float, str, bytes, unicode, slice, complex, bool), obj
compiled_value = create_from_access_path(
inference_state,
inference_state.compiled_subprocess.create_simple_object(obj)
)
return ExactValue(compiled_value)
def get_string_value_set(inference_state):
return builtin_from_name(inference_state, u'str').execute_with_values()
def load_module(inference_state, dotted_name, **kwargs):
# Temporary, some tensorflow builtins cannot be loaded, so it's tried again
# and again and it's really slow.
if dotted_name.startswith('tensorflow.'):
return None
access_path = inference_state.compiled_subprocess.load_module(dotted_name=dotted_name, **kwargs)
if access_path is None:
return None
return create_from_access_path(inference_state, access_path)

View file

@ -0,0 +1,564 @@
from __future__ import print_function
import inspect
import types
import sys
import operator as op
from collections import namedtuple
import warnings
import re
from jedi._compatibility import unicode, is_py3, builtins, \
py_version, force_unicode
from jedi.inference.compiled.getattr_static import getattr_static
ALLOWED_GETITEM_TYPES = (str, list, tuple, unicode, bytes, bytearray, dict)
MethodDescriptorType = type(str.replace)
# These are not considered classes and access is granted even though they have
# a __class__ attribute.
NOT_CLASS_TYPES = (
types.BuiltinFunctionType,
types.CodeType,
types.FrameType,
types.FunctionType,
types.GeneratorType,
types.GetSetDescriptorType,
types.LambdaType,
types.MemberDescriptorType,
types.MethodType,
types.ModuleType,
types.TracebackType,
MethodDescriptorType
)
if is_py3:
NOT_CLASS_TYPES += (
types.MappingProxyType,
types.SimpleNamespace,
types.DynamicClassAttribute,
)
# Those types don't exist in typing.
MethodDescriptorType = type(str.replace)
WrapperDescriptorType = type(set.__iter__)
# `object.__subclasshook__` is an already executed descriptor.
object_class_dict = type.__dict__["__dict__"].__get__(object)
ClassMethodDescriptorType = type(object_class_dict['__subclasshook__'])
_sentinel = object()
# Maps Python syntax to the operator module.
COMPARISON_OPERATORS = {
'==': op.eq,
'!=': op.ne,
'is': op.is_,
'is not': op.is_not,
'<': op.lt,
'<=': op.le,
'>': op.gt,
'>=': op.ge,
}
_OPERATORS = {
'+': op.add,
'-': op.sub,
}
_OPERATORS.update(COMPARISON_OPERATORS)
ALLOWED_DESCRIPTOR_ACCESS = (
types.FunctionType,
types.GetSetDescriptorType,
types.MemberDescriptorType,
MethodDescriptorType,
WrapperDescriptorType,
ClassMethodDescriptorType,
staticmethod,
classmethod,
)
def safe_getattr(obj, name, default=_sentinel):
try:
attr, is_get_descriptor = getattr_static(obj, name)
except AttributeError:
if default is _sentinel:
raise
return default
else:
if isinstance(attr, ALLOWED_DESCRIPTOR_ACCESS):
# In case of descriptors that have get methods we cannot return
# it's value, because that would mean code execution.
# Since it's an isinstance call, code execution is still possible,
# but this is not really a security feature, but much more of a
# safety feature. Code execution is basically always possible when
# a module is imported. This is here so people don't shoot
# themselves in the foot.
return getattr(obj, name)
return attr
SignatureParam = namedtuple(
'SignatureParam',
'name has_default default default_string has_annotation annotation annotation_string kind_name'
)
def shorten_repr(func):
def wrapper(self):
r = func(self)
if len(r) > 50:
r = r[:50] + '..'
return r
return wrapper
def create_access(inference_state, obj):
return inference_state.compiled_subprocess.get_or_create_access_handle(obj)
def load_module(inference_state, dotted_name, sys_path):
temp, sys.path = sys.path, sys_path
try:
__import__(dotted_name)
except ImportError:
# If a module is "corrupt" or not really a Python module or whatever.
print('Module %s not importable in path %s.' % (dotted_name, sys_path), file=sys.stderr)
return None
except Exception:
# Since __import__ pretty much makes code execution possible, just
# catch any error here and print it.
import traceback
print("Cannot import:\n%s" % traceback.format_exc(), file=sys.stderr)
return None
finally:
sys.path = temp
# Just access the cache after import, because of #59 as well as the very
# complicated import structure of Python.
module = sys.modules[dotted_name]
return create_access_path(inference_state, module)
class AccessPath(object):
def __init__(self, accesses):
self.accesses = accesses
# Writing both of these methods here looks a bit ridiculous. However with
# the differences of Python 2/3 it's actually necessary, because we will
# otherwise have a accesses attribute that is bytes instead of unicode.
def __getstate__(self):
return self.accesses
def __setstate__(self, value):
self.accesses = value
def create_access_path(inference_state, obj):
access = create_access(inference_state, obj)
return AccessPath(access.get_access_path_tuples())
def _force_unicode_decorator(func):
return lambda *args, **kwargs: force_unicode(func(*args, **kwargs))
def get_api_type(obj):
if inspect.isclass(obj):
return u'class'
elif inspect.ismodule(obj):
return u'module'
elif inspect.isbuiltin(obj) or inspect.ismethod(obj) \
or inspect.ismethoddescriptor(obj) or inspect.isfunction(obj):
return u'function'
# Everything else...
return u'instance'
class DirectObjectAccess(object):
def __init__(self, inference_state, obj):
self._inference_state = inference_state
self._obj = obj
def __repr__(self):
return '%s(%s)' % (self.__class__.__name__, self.get_repr())
def _create_access(self, obj):
return create_access(self._inference_state, obj)
def _create_access_path(self, obj):
return create_access_path(self._inference_state, obj)
def py__bool__(self):
return bool(self._obj)
def py__file__(self):
try:
return self._obj.__file__
except AttributeError:
return None
def py__doc__(self):
return force_unicode(inspect.getdoc(self._obj)) or u''
def py__name__(self):
if not _is_class_instance(self._obj) or \
inspect.ismethoddescriptor(self._obj): # slots
cls = self._obj
else:
try:
cls = self._obj.__class__
except AttributeError:
# happens with numpy.core.umath._UFUNC_API (you get it
# automatically by doing `import numpy`.
return None
try:
return force_unicode(cls.__name__)
except AttributeError:
return None
def py__mro__accesses(self):
return tuple(self._create_access_path(cls) for cls in self._obj.__mro__[1:])
def py__getitem__all_values(self):
if isinstance(self._obj, dict):
return [self._create_access_path(v) for v in self._obj.values()]
return self.py__iter__list()
def py__simple_getitem__(self, index):
if type(self._obj) not in ALLOWED_GETITEM_TYPES:
# Get rid of side effects, we won't call custom `__getitem__`s.
return None
return self._create_access_path(self._obj[index])
def py__iter__list(self):
if not hasattr(self._obj, '__getitem__'):
return None
if type(self._obj) not in ALLOWED_GETITEM_TYPES:
# Get rid of side effects, we won't call custom `__getitem__`s.
return []
lst = []
for i, part in enumerate(self._obj):
if i > 20:
# Should not go crazy with large iterators
break
lst.append(self._create_access_path(part))
return lst
def py__class__(self):
return self._create_access_path(self._obj.__class__)
def py__bases__(self):
return [self._create_access_path(base) for base in self._obj.__bases__]
def py__path__(self):
paths = getattr(self._obj, '__path__', None)
# Avoid some weird hacks that would just fail, because they cannot be
# used by pickle.
if not isinstance(paths, list) \
or not all(isinstance(p, (bytes, unicode)) for p in paths):
return None
return paths
@_force_unicode_decorator
@shorten_repr
def get_repr(self):
builtins = 'builtins', '__builtin__'
if inspect.ismodule(self._obj):
return repr(self._obj)
# Try to avoid execution of the property.
if safe_getattr(self._obj, '__module__', default='') in builtins:
return repr(self._obj)
type_ = type(self._obj)
if type_ == type:
return type.__repr__(self._obj)
if safe_getattr(type_, '__module__', default='') in builtins:
# Allow direct execution of repr for builtins.
return repr(self._obj)
return object.__repr__(self._obj)
def is_class(self):
return inspect.isclass(self._obj)
def is_function(self):
return inspect.isfunction(self._obj) or inspect.ismethod(self._obj)
def is_module(self):
return inspect.ismodule(self._obj)
def is_instance(self):
return _is_class_instance(self._obj)
def ismethoddescriptor(self):
return inspect.ismethoddescriptor(self._obj)
def get_qualified_names(self):
def try_to_get_name(obj):
return getattr(obj, '__qualname__', getattr(obj, '__name__', None))
if self.is_module():
return ()
name = try_to_get_name(self._obj)
if name is None:
name = try_to_get_name(type(self._obj))
if name is None:
return ()
return tuple(force_unicode(n) for n in name.split('.'))
def dir(self):
return list(map(force_unicode, dir(self._obj)))
def has_iter(self):
try:
iter(self._obj)
return True
except TypeError:
return False
def is_allowed_getattr(self, name, unsafe=False):
# TODO this API is ugly.
if unsafe:
# Unsafe is mostly used to check for __getattr__/__getattribute__.
# getattr_static works for properties, but the underscore methods
# are just ignored (because it's safer and avoids more code
# execution). See also GH #1378.
# Avoid warnings, see comment in the next function.
with warnings.catch_warnings(record=True):
warnings.simplefilter("always")
try:
return hasattr(self._obj, name), False
except Exception:
# Obviously has an attribute (propably a property) that
# gets executed, so just avoid all exceptions here.
return False, False
try:
attr, is_get_descriptor = getattr_static(self._obj, name)
except AttributeError:
return False, False
else:
if is_get_descriptor and type(attr) not in ALLOWED_DESCRIPTOR_ACCESS:
# In case of descriptors that have get methods we cannot return
# it's value, because that would mean code execution.
return True, True
return True, False
def getattr_paths(self, name, default=_sentinel):
try:
# Make sure no warnings are printed here, this is autocompletion,
# warnings should not be shown. See also GH #1383.
with warnings.catch_warnings(record=True):
warnings.simplefilter("always")
return_obj = getattr(self._obj, name)
except Exception as e:
if default is _sentinel:
if isinstance(e, AttributeError):
# Happens e.g. in properties of
# PyQt4.QtGui.QStyleOptionComboBox.currentText
# -> just set it to None
raise
# Just in case anything happens, return an AttributeError. It
# should not crash.
raise AttributeError
return_obj = default
access = self._create_access(return_obj)
if inspect.ismodule(return_obj):
return [access]
try:
module = return_obj.__module__
except AttributeError:
pass
else:
if module is not None:
try:
__import__(module)
# For some modules like _sqlite3, the __module__ for classes is
# different, in this case it's sqlite3. So we have to try to
# load that "original" module, because it's not loaded yet. If
# we don't do that, we don't really have a "parent" module and
# we would fall back to builtins.
except ImportError:
pass
module = inspect.getmodule(return_obj)
if module is None:
module = inspect.getmodule(type(return_obj))
if module is None:
module = builtins
return [self._create_access(module), access]
def get_safe_value(self):
if type(self._obj) in (bool, bytes, float, int, str, unicode, slice) or self._obj is None:
return self._obj
raise ValueError("Object is type %s and not simple" % type(self._obj))
def get_api_type(self):
return get_api_type(self._obj)
def get_array_type(self):
if isinstance(self._obj, dict):
return 'dict'
return None
def get_key_paths(self):
def iter_partial_keys():
# We could use list(keys()), but that might take a lot more memory.
for (i, k) in enumerate(self._obj.keys()):
# Limit key listing at some point. This is artificial, but this
# way we don't get stalled because of slow completions
if i > 50:
break
yield k
return [self._create_access_path(k) for k in iter_partial_keys()]
def get_access_path_tuples(self):
accesses = [create_access(self._inference_state, o) for o in self._get_objects_path()]
return [(access.py__name__(), access) for access in accesses]
def _get_objects_path(self):
def get():
obj = self._obj
yield obj
try:
obj = obj.__objclass__
except AttributeError:
pass
else:
yield obj
try:
# Returns a dotted string path.
imp_plz = obj.__module__
except AttributeError:
# Unfortunately in some cases like `int` there's no __module__
if not inspect.ismodule(obj):
yield builtins
else:
if imp_plz is None:
# Happens for example in `(_ for _ in []).send.__module__`.
yield builtins
else:
try:
yield sys.modules[imp_plz]
except KeyError:
# __module__ can be something arbitrary that doesn't exist.
yield builtins
return list(reversed(list(get())))
def execute_operation(self, other_access_handle, operator):
other_access = other_access_handle.access
op = _OPERATORS[operator]
return self._create_access_path(op(self._obj, other_access._obj))
def get_annotation_name_and_args(self):
"""
Returns Tuple[Optional[str], Tuple[AccessPath, ...]]
"""
if sys.version_info < (3, 5):
return None, ()
name = None
args = ()
if safe_getattr(self._obj, '__module__', default='') == 'typing':
m = re.match(r'typing.(\w+)\[', repr(self._obj))
if m is not None:
name = m.group(1)
import typing
if sys.version_info >= (3, 8):
args = typing.get_args(self._obj)
else:
args = safe_getattr(self._obj, '__args__', default=None)
return name, tuple(self._create_access_path(arg) for arg in args)
def needs_type_completions(self):
return inspect.isclass(self._obj) and self._obj != type
def _annotation_to_str(self, annotation):
if py_version < 30:
return ''
return inspect.formatannotation(annotation)
def get_signature_params(self):
return [
SignatureParam(
name=p.name,
has_default=p.default is not p.empty,
default=self._create_access_path(p.default),
default_string=repr(p.default),
has_annotation=p.annotation is not p.empty,
annotation=self._create_access_path(p.annotation),
annotation_string=self._annotation_to_str(p.annotation),
kind_name=str(p.kind)
) for p in self._get_signature().parameters.values()
]
def _get_signature(self):
obj = self._obj
if py_version < 33:
raise ValueError("inspect.signature was introduced in 3.3")
try:
return inspect.signature(obj)
except (RuntimeError, TypeError):
# Reading the code of the function in Python 3.6 implies there are
# at least these errors that might occur if something is wrong with
# the signature. In that case we just want a simple escape for now.
raise ValueError
def get_return_annotation(self):
try:
o = self._obj.__annotations__.get('return')
except AttributeError:
return None
if o is None:
return None
try:
# Python 2 doesn't have typing.
import typing
except ImportError:
pass
else:
try:
o = typing.get_type_hints(self._obj).get('return')
except Exception:
pass
return self._create_access_path(o)
def negate(self):
return self._create_access_path(-self._obj)
def get_dir_infos(self):
"""
Used to return a couple of infos that are needed when accessing the sub
objects of an objects
"""
tuples = dict(
(force_unicode(name), self.is_allowed_getattr(name))
for name in self.dir()
)
return self.needs_type_completions(), tuples
def _is_class_instance(obj):
"""Like inspect.* methods."""
try:
cls = obj.__class__
except AttributeError:
return False
else:
# The isinstance check for cls is just there so issubclass doesn't
# raise an exception.
return cls != type and isinstance(cls, type) and not issubclass(cls, NOT_CLASS_TYPES)

View file

@ -0,0 +1,184 @@
"""
A static version of getattr.
This is a backport of the Python 3 code with a little bit of additional
information returned to enable Jedi to make decisions.
"""
import types
from jedi import debug
from jedi._compatibility import py_version
_sentinel = object()
def _check_instance(obj, attr):
instance_dict = {}
try:
instance_dict = object.__getattribute__(obj, "__dict__")
except AttributeError:
pass
return dict.get(instance_dict, attr, _sentinel)
def _check_class(klass, attr):
for entry in _static_getmro(klass):
if _shadowed_dict(type(entry)) is _sentinel:
try:
return entry.__dict__[attr]
except KeyError:
pass
return _sentinel
def _is_type(obj):
try:
_static_getmro(obj)
except TypeError:
return False
return True
def _shadowed_dict_newstyle(klass):
dict_attr = type.__dict__["__dict__"]
for entry in _static_getmro(klass):
try:
class_dict = dict_attr.__get__(entry)["__dict__"]
except KeyError:
pass
else:
if not (type(class_dict) is types.GetSetDescriptorType
and class_dict.__name__ == "__dict__"
and class_dict.__objclass__ is entry):
return class_dict
return _sentinel
def _static_getmro_newstyle(klass):
mro = type.__dict__['__mro__'].__get__(klass)
if not isinstance(mro, (tuple, list)):
# There are unfortunately no tests for this, I was not able to
# reproduce this in pure Python. However should still solve the issue
# raised in GH #1517.
debug.warning('mro of %s returned %s, should be a tuple' % (klass, mro))
return ()
return mro
if py_version >= 30:
_shadowed_dict = _shadowed_dict_newstyle
_get_type = type
_static_getmro = _static_getmro_newstyle
else:
def _shadowed_dict(klass):
"""
In Python 2 __dict__ is not overwritable:
class Foo(object): pass
setattr(Foo, '__dict__', 4)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: __dict__ must be a dictionary object
It applies to both newstyle and oldstyle classes:
class Foo(object): pass
setattr(Foo, '__dict__', 4)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
AttributeError: attribute '__dict__' of 'type' objects is not writable
It also applies to instances of those objects. However to keep things
straight forward, newstyle classes always use the complicated way of
accessing it while oldstyle classes just use getattr.
"""
if type(klass) is _oldstyle_class_type:
return getattr(klass, '__dict__', _sentinel)
return _shadowed_dict_newstyle(klass)
class _OldStyleClass:
pass
_oldstyle_instance_type = type(_OldStyleClass())
_oldstyle_class_type = type(_OldStyleClass)
def _get_type(obj):
type_ = object.__getattribute__(obj, '__class__')
if type_ is _oldstyle_instance_type:
# Somehow for old style classes we need to access it directly.
return obj.__class__
return type_
def _static_getmro(klass):
if type(klass) is _oldstyle_class_type:
def oldstyle_mro(klass):
"""
Oldstyle mro is a really simplistic way of look up mro:
https://stackoverflow.com/questions/54867/what-is-the-difference-between-old-style-and-new-style-classes-in-python
"""
yield klass
for base in klass.__bases__:
for yield_from in oldstyle_mro(base):
yield yield_from
return oldstyle_mro(klass)
return _static_getmro_newstyle(klass)
def _safe_hasattr(obj, name):
return _check_class(_get_type(obj), name) is not _sentinel
def _safe_is_data_descriptor(obj):
return _safe_hasattr(obj, '__set__') or _safe_hasattr(obj, '__delete__')
def getattr_static(obj, attr, default=_sentinel):
"""Retrieve attributes without triggering dynamic lookup via the
descriptor protocol, __getattr__ or __getattribute__.
Note: this function may not be able to retrieve all attributes
that getattr can fetch (like dynamically created attributes)
and may find attributes that getattr can't (like descriptors
that raise AttributeError). It can also return descriptor objects
instead of instance members in some cases. See the
documentation for details.
Returns a tuple `(attr, is_get_descriptor)`. is_get_descripter means that
the attribute is a descriptor that has a `__get__` attribute.
"""
instance_result = _sentinel
if not _is_type(obj):
klass = _get_type(obj)
dict_attr = _shadowed_dict(klass)
if (dict_attr is _sentinel or type(dict_attr) is types.MemberDescriptorType):
instance_result = _check_instance(obj, attr)
else:
klass = obj
klass_result = _check_class(klass, attr)
if instance_result is not _sentinel and klass_result is not _sentinel:
if _safe_hasattr(klass_result, '__get__') \
and _safe_is_data_descriptor(klass_result):
# A get/set descriptor has priority over everything.
return klass_result, True
if instance_result is not _sentinel:
return instance_result, False
if klass_result is not _sentinel:
return klass_result, _safe_hasattr(klass_result, '__get__')
if obj is klass:
# for types we check the metaclass too
for entry in _static_getmro(type(klass)):
if _shadowed_dict(type(entry)) is _sentinel:
try:
return entry.__dict__[attr], False
except KeyError:
pass
if default is not _sentinel:
return default, False
raise AttributeError(attr)

View file

@ -0,0 +1,304 @@
"""
Used only for REPL Completion.
"""
import inspect
import os
import sys
from jedi.parser_utils import get_cached_code_lines
from jedi._compatibility import unwrap
from jedi import settings
from jedi.cache import memoize_method
from jedi.inference import compiled
from jedi.file_io import FileIO
from jedi.inference.names import NameWrapper
from jedi.inference.base_value import ValueSet, ValueWrapper, NO_VALUES
from jedi.inference.value import ModuleValue
from jedi.inference.cache import inference_state_function_cache, \
inference_state_method_cache
from jedi.inference.compiled.access import ALLOWED_GETITEM_TYPES, get_api_type
from jedi.inference.gradual.conversion import to_stub
from jedi.inference.context import CompiledContext, CompiledModuleContext, \
TreeContextMixin
_sentinel = object()
class MixedObject(ValueWrapper):
"""
A ``MixedObject`` is used in two ways:
1. It uses the default logic of ``parser.python.tree`` objects,
2. except for getattr calls and signatures. The names dicts are generated
in a fashion like ``CompiledValue``.
This combined logic makes it possible to provide more powerful REPL
completion. It allows side effects that are not noticable with the default
parser structure to still be completeable.
The biggest difference from CompiledValue to MixedObject is that we are
generally dealing with Python code and not with C code. This will generate
fewer special cases, because we in Python you don't have the same freedoms
to modify the runtime.
"""
def __init__(self, compiled_value, tree_value):
super(MixedObject, self).__init__(tree_value)
self.compiled_value = compiled_value
self.access_handle = compiled_value.access_handle
def get_filters(self, *args, **kwargs):
yield MixedObjectFilter(
self.inference_state, self.compiled_value, self._wrapped_value)
def get_signatures(self):
# Prefer `inspect.signature` over somehow analyzing Python code. It
# should be very precise, especially for stuff like `partial`.
return self.compiled_value.get_signatures()
@inference_state_method_cache(default=NO_VALUES)
def py__call__(self, arguments):
# Fallback to the wrapped value if to stub returns no values.
values = to_stub(self._wrapped_value)
if not values:
values = self._wrapped_value
return values.py__call__(arguments)
def get_safe_value(self, default=_sentinel):
if default is _sentinel:
return self.compiled_value.get_safe_value()
else:
return self.compiled_value.get_safe_value(default)
@property
def array_type(self):
return self.compiled_value.array_type
def get_key_values(self):
return self.compiled_value.get_key_values()
def py__simple_getitem__(self, index):
python_object = self.compiled_value.access_handle.access._obj
if type(python_object) in ALLOWED_GETITEM_TYPES:
return self.compiled_value.py__simple_getitem__(index)
return self._wrapped_value.py__simple_getitem__(index)
def negate(self):
return self.compiled_value.negate()
def _as_context(self):
if self.parent_context is None:
return MixedModuleContext(self)
return MixedContext(self)
def __repr__(self):
return '<%s: %s; %s>' % (
type(self).__name__,
self.access_handle.get_repr(),
self._wrapped_value,
)
class MixedContext(CompiledContext, TreeContextMixin):
@property
def compiled_value(self):
return self._value.compiled_value
class MixedModuleContext(CompiledModuleContext, MixedContext):
pass
class MixedName(NameWrapper):
"""
The ``CompiledName._compiled_value`` is our MixedObject.
"""
def __init__(self, wrapped_name, parent_tree_value):
super(MixedName, self).__init__(wrapped_name)
self._parent_tree_value = parent_tree_value
@property
def start_pos(self):
values = list(self.infer())
if not values:
# This means a start_pos that doesn't exist (compiled objects).
return 0, 0
return values[0].name.start_pos
@memoize_method
def infer(self):
compiled_value = self._wrapped_name.infer_compiled_value()
tree_value = self._parent_tree_value
if tree_value.is_instance() or tree_value.is_class():
tree_values = tree_value.py__getattribute__(self.string_name)
if compiled_value.is_function():
return ValueSet({MixedObject(compiled_value, v) for v in tree_values})
module_context = tree_value.get_root_context()
return _create(self._inference_state, compiled_value, module_context)
class MixedObjectFilter(compiled.CompiledValueFilter):
def __init__(self, inference_state, compiled_value, tree_value):
super(MixedObjectFilter, self).__init__(inference_state, compiled_value)
self._tree_value = tree_value
def _create_name(self, name):
return MixedName(
super(MixedObjectFilter, self)._create_name(name),
self._tree_value,
)
@inference_state_function_cache()
def _load_module(inference_state, path):
return inference_state.parse(
path=path,
cache=True,
diff_cache=settings.fast_parser,
cache_path=settings.cache_directory
).get_root_node()
def _get_object_to_check(python_object):
"""Check if inspect.getfile has a chance to find the source."""
if sys.version_info[0] > 2:
try:
python_object = unwrap(python_object)
except ValueError:
# Can return a ValueError when it wraps around
pass
if (inspect.ismodule(python_object)
or inspect.isclass(python_object)
or inspect.ismethod(python_object)
or inspect.isfunction(python_object)
or inspect.istraceback(python_object)
or inspect.isframe(python_object)
or inspect.iscode(python_object)):
return python_object
try:
return python_object.__class__
except AttributeError:
raise TypeError # Prevents computation of `repr` within inspect.
def _find_syntax_node_name(inference_state, python_object):
original_object = python_object
try:
python_object = _get_object_to_check(python_object)
path = inspect.getsourcefile(python_object)
except TypeError:
# The type might not be known (e.g. class_with_dict.__weakref__)
return None
if path is None or not os.path.exists(path):
# The path might not exist or be e.g. <stdin>.
return None
file_io = FileIO(path)
module_node = _load_module(inference_state, path)
if inspect.ismodule(python_object):
# We don't need to check names for modules, because there's not really
# a way to write a module in a module in Python (and also __name__ can
# be something like ``email.utils``).
code_lines = get_cached_code_lines(inference_state.grammar, path)
return module_node, module_node, file_io, code_lines
try:
name_str = python_object.__name__
except AttributeError:
# Stuff like python_function.__code__.
return None
if name_str == '<lambda>':
return None # It's too hard to find lambdas.
# Doesn't always work (e.g. os.stat_result)
names = module_node.get_used_names().get(name_str, [])
# Only functions and classes are relevant. If a name e.g. points to an
# import, it's probably a builtin (like collections.deque) and needs to be
# ignored.
names = [
n for n in names
if n.parent.type in ('funcdef', 'classdef') and n.parent.name == n
]
if not names:
return None
try:
code = python_object.__code__
# By using the line number of a code object we make the lookup in a
# file pretty easy. There's still a possibility of people defining
# stuff like ``a = 3; foo(a); a = 4`` on the same line, but if people
# do so we just don't care.
line_nr = code.co_firstlineno
except AttributeError:
pass
else:
line_names = [name for name in names if name.start_pos[0] == line_nr]
# There's a chance that the object is not available anymore, because
# the code has changed in the background.
if line_names:
names = line_names
code_lines = get_cached_code_lines(inference_state.grammar, path)
# It's really hard to actually get the right definition, here as a last
# resort we just return the last one. This chance might lead to odd
# completions at some points but will lead to mostly correct type
# inference, because people tend to define a public name in a module only
# once.
tree_node = names[-1].parent
if tree_node.type == 'funcdef' and get_api_type(original_object) == 'instance':
# If an instance is given and we're landing on a function (e.g.
# partial in 3.5), something is completely wrong and we should not
# return that.
return None
return module_node, tree_node, file_io, code_lines
@inference_state_function_cache()
def _create(inference_state, compiled_value, module_context):
# TODO accessing this is bad, but it probably doesn't matter that much,
# because we're working with interpreteters only here.
python_object = compiled_value.access_handle.access._obj
result = _find_syntax_node_name(inference_state, python_object)
if result is None:
# TODO Care about generics from stuff like `[1]` and don't return like this.
if type(python_object) in (dict, list, tuple):
return ValueSet({compiled_value})
tree_values = to_stub(compiled_value)
if not tree_values:
return ValueSet({compiled_value})
else:
module_node, tree_node, file_io, code_lines = result
if module_context is None or module_context.tree_node != module_node:
root_compiled_value = compiled_value.get_root_context().get_value()
# TODO this __name__ might be wrong.
name = root_compiled_value.py__name__()
string_names = tuple(name.split('.'))
module_value = ModuleValue(
inference_state, module_node,
file_io=file_io,
string_names=string_names,
code_lines=code_lines,
is_package=root_compiled_value.is_package(),
)
if name is not None:
inference_state.module_cache.add(string_names, ValueSet([module_value]))
module_context = module_value.as_context()
tree_values = ValueSet({module_context.create_value(tree_node)})
if tree_node.type == 'classdef':
if not compiled_value.is_class():
# Is an instance, not a class.
tree_values = tree_values.execute_with_values()
return ValueSet(
MixedObject(compiled_value, tree_value=tree_value)
for tree_value in tree_values
)

View file

@ -0,0 +1,404 @@
"""
Makes it possible to do the compiled analysis in a subprocess. This has two
goals:
1. Making it safer - Segfaults and RuntimeErrors as well as stdout/stderr can
be ignored and dealt with.
2. Make it possible to handle different Python versions as well as virtualenvs.
"""
import os
import sys
import subprocess
import socket
import errno
import traceback
from functools import partial
from threading import Thread
try:
from queue import Queue, Empty
except ImportError:
from Queue import Queue, Empty # python 2.7
from jedi._compatibility import queue, is_py3, force_unicode, \
pickle_dump, pickle_load, GeneralizedPopen, weakref
from jedi import debug
from jedi.cache import memoize_method
from jedi.inference.compiled.subprocess import functions
from jedi.inference.compiled.access import DirectObjectAccess, AccessPath, \
SignatureParam
from jedi.api.exceptions import InternalError
_MAIN_PATH = os.path.join(os.path.dirname(__file__), '__main__.py')
def _enqueue_output(out, queue):
for line in iter(out.readline, b''):
queue.put(line)
def _add_stderr_to_debug(stderr_queue):
while True:
# Try to do some error reporting from the subprocess and print its
# stderr contents.
try:
line = stderr_queue.get_nowait()
line = line.decode('utf-8', 'replace')
debug.warning('stderr output: %s' % line.rstrip('\n'))
except Empty:
break
def _get_function(name):
return getattr(functions, name)
def _cleanup_process(process, thread):
try:
process.kill()
process.wait()
except OSError:
# Raised if the process is already killed.
pass
thread.join()
for stream in [process.stdin, process.stdout, process.stderr]:
try:
stream.close()
except OSError:
# Raised if the stream is broken.
pass
class _InferenceStateProcess(object):
def __init__(self, inference_state):
self._inference_state_weakref = weakref.ref(inference_state)
self._inference_state_id = id(inference_state)
self._handles = {}
def get_or_create_access_handle(self, obj):
id_ = id(obj)
try:
return self.get_access_handle(id_)
except KeyError:
access = DirectObjectAccess(self._inference_state_weakref(), obj)
handle = AccessHandle(self, access, id_)
self.set_access_handle(handle)
return handle
def get_access_handle(self, id_):
return self._handles[id_]
def set_access_handle(self, handle):
self._handles[handle.id] = handle
class InferenceStateSameProcess(_InferenceStateProcess):
"""
Basically just an easy access to functions.py. It has the same API
as InferenceStateSubprocess and does the same thing without using a subprocess.
This is necessary for the Interpreter process.
"""
def __getattr__(self, name):
return partial(_get_function(name), self._inference_state_weakref())
class InferenceStateSubprocess(_InferenceStateProcess):
def __init__(self, inference_state, compiled_subprocess):
super(InferenceStateSubprocess, self).__init__(inference_state)
self._used = False
self._compiled_subprocess = compiled_subprocess
def __getattr__(self, name):
func = _get_function(name)
def wrapper(*args, **kwargs):
self._used = True
result = self._compiled_subprocess.run(
self._inference_state_weakref(),
func,
args=args,
kwargs=kwargs,
)
# IMO it should be possible to create a hook in pickle.load to
# mess with the loaded objects. However it's extremely complicated
# to work around this so just do it with this call. ~ dave
return self._convert_access_handles(result)
return wrapper
def _convert_access_handles(self, obj):
if isinstance(obj, SignatureParam):
return SignatureParam(*self._convert_access_handles(tuple(obj)))
elif isinstance(obj, tuple):
return tuple(self._convert_access_handles(o) for o in obj)
elif isinstance(obj, list):
return [self._convert_access_handles(o) for o in obj]
elif isinstance(obj, AccessHandle):
try:
# Rewrite the access handle to one we're already having.
obj = self.get_access_handle(obj.id)
except KeyError:
obj.add_subprocess(self)
self.set_access_handle(obj)
elif isinstance(obj, AccessPath):
return AccessPath(self._convert_access_handles(obj.accesses))
return obj
def __del__(self):
if self._used and not self._compiled_subprocess.is_crashed:
self._compiled_subprocess.delete_inference_state(self._inference_state_id)
class CompiledSubprocess(object):
is_crashed = False
# Start with 2, gets set after _get_info.
_pickle_protocol = 2
def __init__(self, executable, env_vars=None):
self._executable = executable
self._env_vars = env_vars
self._inference_state_deletion_queue = queue.deque()
self._cleanup_callable = lambda: None
def __repr__(self):
pid = os.getpid()
return '<%s _executable=%r, _pickle_protocol=%r, is_crashed=%r, pid=%r>' % (
self.__class__.__name__,
self._executable,
self._pickle_protocol,
self.is_crashed,
pid,
)
@memoize_method
def _get_process(self):
debug.dbg('Start environment subprocess %s', self._executable)
parso_path = sys.modules['parso'].__file__
args = (
self._executable,
_MAIN_PATH,
os.path.dirname(os.path.dirname(parso_path)),
'.'.join(str(x) for x in sys.version_info[:3]),
)
process = GeneralizedPopen(
args,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
# Use system default buffering on Python 2 to improve performance
# (this is already the case on Python 3).
bufsize=-1,
env=self._env_vars
)
self._stderr_queue = Queue()
self._stderr_thread = t = Thread(
target=_enqueue_output,
args=(process.stderr, self._stderr_queue)
)
t.daemon = True
t.start()
# Ensure the subprocess is properly cleaned up when the object
# is garbage collected.
self._cleanup_callable = weakref.finalize(self,
_cleanup_process,
process,
t)
return process
def run(self, inference_state, function, args=(), kwargs={}):
# Delete old inference_states.
while True:
try:
inference_state_id = self._inference_state_deletion_queue.pop()
except IndexError:
break
else:
self._send(inference_state_id, None)
assert callable(function)
return self._send(id(inference_state), function, args, kwargs)
def get_sys_path(self):
return self._send(None, functions.get_sys_path, (), {})
def _kill(self):
self.is_crashed = True
self._cleanup_callable()
def _send(self, inference_state_id, function, args=(), kwargs={}):
if self.is_crashed:
raise InternalError("The subprocess %s has crashed." % self._executable)
if not is_py3:
# Python 2 compatibility
kwargs = {force_unicode(key): value for key, value in kwargs.items()}
data = inference_state_id, function, args, kwargs
try:
pickle_dump(data, self._get_process().stdin, self._pickle_protocol)
except (socket.error, IOError) as e:
# Once Python2 will be removed we can just use `BrokenPipeError`.
# Also, somehow in windows it returns EINVAL instead of EPIPE if
# the subprocess dies.
if e.errno not in (errno.EPIPE, errno.EINVAL):
# Not a broken pipe
raise
self._kill()
raise InternalError("The subprocess %s was killed. Maybe out of memory?"
% self._executable)
try:
is_exception, traceback, result = pickle_load(self._get_process().stdout)
except EOFError as eof_error:
try:
stderr = self._get_process().stderr.read().decode('utf-8', 'replace')
except Exception as exc:
stderr = '<empty/not available (%r)>' % exc
self._kill()
_add_stderr_to_debug(self._stderr_queue)
raise InternalError(
"The subprocess %s has crashed (%r, stderr=%s)." % (
self._executable,
eof_error,
stderr,
))
_add_stderr_to_debug(self._stderr_queue)
if is_exception:
# Replace the attribute error message with a the traceback. It's
# way more informative.
result.args = (traceback,)
raise result
return result
def delete_inference_state(self, inference_state_id):
"""
Currently we are not deleting inference_state instantly. They only get
deleted once the subprocess is used again. It would probably a better
solution to move all of this into a thread. However, the memory usage
of a single inference_state shouldn't be that high.
"""
# With an argument - the inference_state gets deleted.
self._inference_state_deletion_queue.append(inference_state_id)
class Listener(object):
def __init__(self, pickle_protocol):
self._inference_states = {}
# TODO refactor so we don't need to process anymore just handle
# controlling.
self._process = _InferenceStateProcess(Listener)
self._pickle_protocol = pickle_protocol
def _get_inference_state(self, function, inference_state_id):
from jedi.inference import InferenceState
try:
inference_state = self._inference_states[inference_state_id]
except KeyError:
from jedi import InterpreterEnvironment
inference_state = InferenceState(
# The project is not actually needed. Nothing should need to
# access it.
project=None,
environment=InterpreterEnvironment()
)
self._inference_states[inference_state_id] = inference_state
return inference_state
def _run(self, inference_state_id, function, args, kwargs):
if inference_state_id is None:
return function(*args, **kwargs)
elif function is None:
del self._inference_states[inference_state_id]
else:
inference_state = self._get_inference_state(function, inference_state_id)
# Exchange all handles
args = list(args)
for i, arg in enumerate(args):
if isinstance(arg, AccessHandle):
args[i] = inference_state.compiled_subprocess.get_access_handle(arg.id)
for key, value in kwargs.items():
if isinstance(value, AccessHandle):
kwargs[key] = inference_state.compiled_subprocess.get_access_handle(value.id)
return function(inference_state, *args, **kwargs)
def listen(self):
stdout = sys.stdout
# Mute stdout. Nobody should actually be able to write to it,
# because stdout is used for IPC.
sys.stdout = open(os.devnull, 'w')
stdin = sys.stdin
if sys.version_info[0] > 2:
stdout = stdout.buffer
stdin = stdin.buffer
# Python 2 opens streams in text mode on Windows. Set stdout and stdin
# to binary mode.
elif sys.platform == 'win32':
import msvcrt
msvcrt.setmode(stdout.fileno(), os.O_BINARY)
msvcrt.setmode(stdin.fileno(), os.O_BINARY)
while True:
try:
payload = pickle_load(stdin)
except EOFError:
# It looks like the parent process closed.
# Don't make a big fuss here and just exit.
exit(0)
try:
result = False, None, self._run(*payload)
except Exception as e:
result = True, traceback.format_exc(), e
pickle_dump(result, stdout, self._pickle_protocol)
class AccessHandle(object):
def __init__(self, subprocess, access, id_):
self.access = access
self._subprocess = subprocess
self.id = id_
def add_subprocess(self, subprocess):
self._subprocess = subprocess
def __repr__(self):
try:
detail = self.access
except AttributeError:
detail = '#' + str(self.id)
return '<%s of %s>' % (self.__class__.__name__, detail)
def __getstate__(self):
return self.id
def __setstate__(self, state):
self.id = state
def __getattr__(self, name):
if name in ('id', 'access') or name.startswith('_'):
raise AttributeError("Something went wrong with unpickling")
# if not is_py3: print >> sys.stderr, name
# print('getattr', name, file=sys.stderr)
return partial(self._workaround, force_unicode(name))
def _workaround(self, name, *args, **kwargs):
"""
TODO Currently we're passing slice objects around. This should not
happen. They are also the only unhashable objects that we're passing
around.
"""
if args and isinstance(args[0], slice):
return self._subprocess.get_compiled_method_return(self.id, name, *args, **kwargs)
return self._cached_results(name, *args, **kwargs)
@memoize_method
def _cached_results(self, name, *args, **kwargs):
return self._subprocess.get_compiled_method_return(self.id, name, *args, **kwargs)

View file

@ -0,0 +1,55 @@
import os
import sys
def _get_paths():
# Get the path to jedi.
_d = os.path.dirname
_jedi_path = _d(_d(_d(_d(_d(__file__)))))
_parso_path = sys.argv[1]
# The paths are the directory that jedi and parso lie in.
return {'jedi': _jedi_path, 'parso': _parso_path}
# Remove the first entry, because it's simply a directory entry that equals
# this directory.
del sys.path[0]
if sys.version_info > (3, 4):
from importlib.machinery import PathFinder
class _ExactImporter(object):
def __init__(self, path_dct):
self._path_dct = path_dct
def find_module(self, fullname, path=None):
if path is None and fullname in self._path_dct:
p = self._path_dct[fullname]
loader = PathFinder.find_module(fullname, path=[p])
return loader
return None
# Try to import jedi/parso.
sys.meta_path.insert(0, _ExactImporter(_get_paths()))
from jedi.inference.compiled import subprocess # NOQA
sys.meta_path.pop(0)
else:
import imp
def load(name):
paths = list(_get_paths().values())
fp, pathname, description = imp.find_module(name, paths)
return imp.load_module(name, fp, pathname, description)
load('parso')
load('jedi')
from jedi.inference.compiled import subprocess # NOQA
from jedi._compatibility import highest_pickle_protocol # noqa: E402
# Retrieve the pickle protocol.
host_sys_version = [int(x) for x in sys.argv[2].split('.')]
pickle_protocol = highest_pickle_protocol([sys.version_info, host_sys_version])
# And finally start the client.
subprocess.Listener(pickle_protocol=pickle_protocol).listen()

View file

@ -0,0 +1,115 @@
from __future__ import print_function
import sys
import os
import re
import inspect
from jedi._compatibility import find_module, cast_path, force_unicode, \
all_suffixes, scandir
from jedi.inference.compiled import access
from jedi import debug
from jedi import parser_utils
def get_sys_path():
return list(map(cast_path, sys.path))
def load_module(inference_state, **kwargs):
return access.load_module(inference_state, **kwargs)
def get_compiled_method_return(inference_state, id, attribute, *args, **kwargs):
handle = inference_state.compiled_subprocess.get_access_handle(id)
return getattr(handle.access, attribute)(*args, **kwargs)
def create_simple_object(inference_state, obj):
return access.create_access_path(inference_state, obj)
def get_module_info(inference_state, sys_path=None, full_name=None, **kwargs):
"""
Returns Tuple[Union[NamespaceInfo, FileIO, None], Optional[bool]]
"""
if sys_path is not None:
sys.path, temp = sys_path, sys.path
try:
return find_module(full_name=full_name, **kwargs)
except ImportError:
return None, None
finally:
if sys_path is not None:
sys.path = temp
def get_builtin_module_names(inference_state):
return list(map(force_unicode, sys.builtin_module_names))
def _test_raise_error(inference_state, exception_type):
"""
Raise an error to simulate certain problems for unit tests.
"""
raise exception_type
def _test_print(inference_state, stderr=None, stdout=None):
"""
Force some prints in the subprocesses. This exists for unit tests.
"""
if stderr is not None:
print(stderr, file=sys.stderr)
sys.stderr.flush()
if stdout is not None:
print(stdout)
sys.stdout.flush()
def _get_init_path(directory_path):
"""
The __init__ file can be searched in a directory. If found return it, else
None.
"""
for suffix in all_suffixes():
path = os.path.join(directory_path, '__init__' + suffix)
if os.path.exists(path):
return path
return None
def safe_literal_eval(inference_state, value):
return parser_utils.safe_literal_eval(value)
def iter_module_names(*args, **kwargs):
return list(_iter_module_names(*args, **kwargs))
def _iter_module_names(inference_state, paths):
# Python modules/packages
for path in paths:
try:
dirs = scandir(path)
except OSError:
# The file might not exist or reading it might lead to an error.
debug.warning("Not possible to list directory: %s", path)
continue
for dir_entry in dirs:
name = dir_entry.name
# First Namespaces then modules/stubs
if dir_entry.is_dir():
# pycache is obviously not an interestin namespace. Also the
# name must be a valid identifier.
# TODO use str.isidentifier, once Python 2 is removed
if name != '__pycache__' and not re.search(r'\W|^\d', name):
yield name
else:
if name.endswith('.pyi'): # Stub files
modname = name[:-4]
else:
modname = inspect.getmodulename(name)
if modname and '.' not in modname:
if modname != '__init__':
yield modname

View file

@ -0,0 +1,629 @@
"""
Imitate the parser representation.
"""
import re
from functools import partial
from jedi import debug
from jedi.inference.utils import to_list
from jedi._compatibility import force_unicode, Parameter, cast_path
from jedi.cache import memoize_method
from jedi.inference.filters import AbstractFilter
from jedi.inference.names import AbstractNameDefinition, ValueNameMixin, \
ParamNameInterface
from jedi.inference.base_value import Value, ValueSet, NO_VALUES
from jedi.inference.lazy_value import LazyKnownValue
from jedi.inference.compiled.access import _sentinel
from jedi.inference.cache import inference_state_function_cache
from jedi.inference.helpers import reraise_getitem_errors
from jedi.inference.signature import BuiltinSignature
from jedi.inference.context import CompiledContext, CompiledModuleContext
class CheckAttribute(object):
"""Raises :exc:`AttributeError` if the attribute X is not available."""
def __init__(self, check_name=None):
# Remove the py in front of e.g. py__call__.
self.check_name = check_name
def __call__(self, func):
self.func = func
if self.check_name is None:
self.check_name = force_unicode(func.__name__[2:])
return self
def __get__(self, instance, owner):
if instance is None:
return self
# This might raise an AttributeError. That's wanted.
instance.access_handle.getattr_paths(self.check_name)
return partial(self.func, instance)
class CompiledValue(Value):
def __init__(self, inference_state, access_handle, parent_context=None):
super(CompiledValue, self).__init__(inference_state, parent_context)
self.access_handle = access_handle
def py__call__(self, arguments):
return_annotation = self.access_handle.get_return_annotation()
if return_annotation is not None:
# TODO the return annotation may also be a string.
return create_from_access_path(
self.inference_state,
return_annotation
).execute_annotation()
try:
self.access_handle.getattr_paths(u'__call__')
except AttributeError:
return super(CompiledValue, self).py__call__(arguments)
else:
if self.access_handle.is_class():
from jedi.inference.value import CompiledInstance
return ValueSet([
CompiledInstance(self.inference_state, self.parent_context, self, arguments)
])
else:
return ValueSet(self._execute_function(arguments))
@CheckAttribute()
def py__class__(self):
return create_from_access_path(self.inference_state, self.access_handle.py__class__())
@CheckAttribute()
def py__mro__(self):
return (self,) + tuple(
create_from_access_path(self.inference_state, access)
for access in self.access_handle.py__mro__accesses()
)
@CheckAttribute()
def py__bases__(self):
return tuple(
create_from_access_path(self.inference_state, access)
for access in self.access_handle.py__bases__()
)
def get_qualified_names(self):
return self.access_handle.get_qualified_names()
def py__bool__(self):
return self.access_handle.py__bool__()
def is_class(self):
return self.access_handle.is_class()
def is_function(self):
return self.access_handle.is_function()
def is_module(self):
return self.access_handle.is_module()
def is_compiled(self):
return True
def is_stub(self):
return False
def is_instance(self):
return self.access_handle.is_instance()
def py__doc__(self):
return self.access_handle.py__doc__()
@to_list
def get_param_names(self):
try:
signature_params = self.access_handle.get_signature_params()
except ValueError: # Has no signature
params_str, ret = self._parse_function_doc()
if not params_str:
tokens = []
else:
tokens = params_str.split(',')
if self.access_handle.ismethoddescriptor():
tokens.insert(0, 'self')
for p in tokens:
name, _, default = p.strip().partition('=')
yield UnresolvableParamName(self, name, default)
else:
for signature_param in signature_params:
yield SignatureParamName(self, signature_param)
def get_signatures(self):
_, return_string = self._parse_function_doc()
return [BuiltinSignature(self, return_string)]
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self.access_handle.get_repr())
@memoize_method
def _parse_function_doc(self):
doc = self.py__doc__()
if doc is None:
return '', ''
return _parse_function_doc(doc)
@property
def api_type(self):
return self.access_handle.get_api_type()
def get_filters(self, is_instance=False, origin_scope=None):
yield self._ensure_one_filter(is_instance)
@memoize_method
def _ensure_one_filter(self, is_instance):
return CompiledValueFilter(self.inference_state, self, is_instance)
def py__simple_getitem__(self, index):
with reraise_getitem_errors(IndexError, KeyError, TypeError):
try:
access = self.access_handle.py__simple_getitem__(index)
except AttributeError:
return super(CompiledValue, self).py__simple_getitem__(index)
if access is None:
return NO_VALUES
return ValueSet([create_from_access_path(self.inference_state, access)])
def py__getitem__(self, index_value_set, contextualized_node):
all_access_paths = self.access_handle.py__getitem__all_values()
if all_access_paths is None:
# This means basically that no __getitem__ has been defined on this
# object.
return super(CompiledValue, self).py__getitem__(index_value_set, contextualized_node)
return ValueSet(
create_from_access_path(self.inference_state, access)
for access in all_access_paths
)
def py__iter__(self, contextualized_node=None):
# Python iterators are a bit strange, because there's no need for
# the __iter__ function as long as __getitem__ is defined (it will
# just start with __getitem__(0). This is especially true for
# Python 2 strings, where `str.__iter__` is not even defined.
if not self.access_handle.has_iter():
for x in super(CompiledValue, self).py__iter__(contextualized_node):
yield x
access_path_list = self.access_handle.py__iter__list()
if access_path_list is None:
# There is no __iter__ method on this object.
return
for access in access_path_list:
yield LazyKnownValue(create_from_access_path(self.inference_state, access))
def py__name__(self):
return self.access_handle.py__name__()
@property
def name(self):
name = self.py__name__()
if name is None:
name = self.access_handle.get_repr()
return CompiledValueName(self, name)
def _execute_function(self, params):
from jedi.inference import docstrings
from jedi.inference.compiled import builtin_from_name
if self.api_type != 'function':
return
for name in self._parse_function_doc()[1].split():
try:
# TODO wtf is this? this is exactly the same as the thing
# below. It uses getattr as well.
self.inference_state.builtins_module.access_handle.getattr_paths(name)
except AttributeError:
continue
else:
bltn_obj = builtin_from_name(self.inference_state, name)
for result in self.inference_state.execute(bltn_obj, params):
yield result
for type_ in docstrings.infer_return_types(self):
yield type_
def get_safe_value(self, default=_sentinel):
try:
return self.access_handle.get_safe_value()
except ValueError:
if default == _sentinel:
raise
return default
def execute_operation(self, other, operator):
try:
return ValueSet([create_from_access_path(
self.inference_state,
self.access_handle.execute_operation(other.access_handle, operator)
)])
except TypeError:
return NO_VALUES
def execute_annotation(self):
if self.access_handle.get_repr() == 'None':
# None as an annotation doesn't need to be executed.
return ValueSet([self])
name, args = self.access_handle.get_annotation_name_and_args()
arguments = [
ValueSet([create_from_access_path(self.inference_state, path)])
for path in args
]
if name == 'Union':
return ValueSet.from_sets(arg.execute_annotation() for arg in arguments)
elif name:
# While with_generics only exists on very specific objects, we
# should probably be fine, because we control all the typing
# objects.
return ValueSet([
v.with_generics(arguments)
for v in self.inference_state.typing_module.py__getattribute__(name)
]).execute_annotation()
return super(CompiledValue, self).execute_annotation()
def negate(self):
return create_from_access_path(self.inference_state, self.access_handle.negate())
def get_metaclasses(self):
return NO_VALUES
def _as_context(self):
return CompiledContext(self)
@property
def array_type(self):
return self.access_handle.get_array_type()
def get_key_values(self):
return [
create_from_access_path(self.inference_state, k)
for k in self.access_handle.get_key_paths()
]
def get_type_hint(self, add_class_info=True):
if self.access_handle.get_repr() in ('None', "<class 'NoneType'>"):
return 'None'
return None
class CompiledModule(CompiledValue):
file_io = None # For modules
def _as_context(self):
return CompiledModuleContext(self)
def py__path__(self):
paths = self.access_handle.py__path__()
if paths is None:
return None
return map(cast_path, paths)
def is_package(self):
return self.py__path__() is not None
@property
def string_names(self):
# For modules
name = self.py__name__()
if name is None:
return ()
return tuple(name.split('.'))
def py__file__(self):
return cast_path(self.access_handle.py__file__())
class CompiledName(AbstractNameDefinition):
def __init__(self, inference_state, parent_value, name):
self._inference_state = inference_state
self.parent_context = parent_value.as_context()
self._parent_value = parent_value
self.string_name = name
def py__doc__(self):
value, = self.infer()
return value.py__doc__()
def _get_qualified_names(self):
parent_qualified_names = self.parent_context.get_qualified_names()
if parent_qualified_names is None:
return None
return parent_qualified_names + (self.string_name,)
def get_defining_qualified_value(self):
context = self.parent_context
if context.is_module() or context.is_class():
return self.parent_context.get_value() # Might be None
return None
def __repr__(self):
try:
name = self.parent_context.name # __name__ is not defined all the time
except AttributeError:
name = None
return '<%s: (%s).%s>' % (self.__class__.__name__, name, self.string_name)
@property
def api_type(self):
api = self.infer()
# If we can't find the type, assume it is an instance variable
if not api:
return "instance"
return next(iter(api)).api_type
@memoize_method
def infer(self):
return ValueSet([self.infer_compiled_value()])
def infer_compiled_value(self):
return create_from_name(self._inference_state, self._parent_value, self.string_name)
class SignatureParamName(ParamNameInterface, AbstractNameDefinition):
def __init__(self, compiled_value, signature_param):
self.parent_context = compiled_value.parent_context
self._signature_param = signature_param
@property
def string_name(self):
return self._signature_param.name
def to_string(self):
s = self._kind_string() + self.string_name
if self._signature_param.has_annotation:
s += ': ' + self._signature_param.annotation_string
if self._signature_param.has_default:
s += '=' + self._signature_param.default_string
return s
def get_kind(self):
return getattr(Parameter, self._signature_param.kind_name)
def infer(self):
p = self._signature_param
inference_state = self.parent_context.inference_state
values = NO_VALUES
if p.has_default:
values = ValueSet([create_from_access_path(inference_state, p.default)])
if p.has_annotation:
annotation = create_from_access_path(inference_state, p.annotation)
values |= annotation.execute_with_values()
return values
class UnresolvableParamName(ParamNameInterface, AbstractNameDefinition):
def __init__(self, compiled_value, name, default):
self.parent_context = compiled_value.parent_context
self.string_name = name
self._default = default
def get_kind(self):
return Parameter.POSITIONAL_ONLY
def to_string(self):
string = self.string_name
if self._default:
string += '=' + self._default
return string
def infer(self):
return NO_VALUES
class CompiledValueName(ValueNameMixin, AbstractNameDefinition):
def __init__(self, value, name):
self.string_name = name
self._value = value
self.parent_context = value.parent_context
class EmptyCompiledName(AbstractNameDefinition):
"""
Accessing some names will raise an exception. To avoid not having any
completions, just give Jedi the option to return this object. It infers to
nothing.
"""
def __init__(self, inference_state, name):
self.parent_context = inference_state.builtins_module
self.string_name = name
def infer(self):
return NO_VALUES
class CompiledValueFilter(AbstractFilter):
def __init__(self, inference_state, compiled_value, is_instance=False):
self._inference_state = inference_state
self.compiled_value = compiled_value
self.is_instance = is_instance
def get(self, name):
access_handle = self.compiled_value.access_handle
return self._get(
name,
lambda name, unsafe: access_handle.is_allowed_getattr(name, unsafe),
lambda name: name in access_handle.dir(),
check_has_attribute=True
)
def _get(self, name, allowed_getattr_callback, in_dir_callback, check_has_attribute=False):
"""
To remove quite a few access calls we introduced the callback here.
"""
# Always use unicode objects in Python 2 from here.
name = force_unicode(name)
if self._inference_state.allow_descriptor_getattr:
pass
has_attribute, is_descriptor = allowed_getattr_callback(
name,
unsafe=self._inference_state.allow_descriptor_getattr
)
if check_has_attribute and not has_attribute:
return []
if (is_descriptor or not has_attribute) \
and not self._inference_state.allow_descriptor_getattr:
return [self._get_cached_name(name, is_empty=True)]
if self.is_instance and not in_dir_callback(name):
return []
return [self._get_cached_name(name)]
@memoize_method
def _get_cached_name(self, name, is_empty=False):
if is_empty:
return EmptyCompiledName(self._inference_state, name)
else:
return self._create_name(name)
def values(self):
from jedi.inference.compiled import builtin_from_name
names = []
needs_type_completions, dir_infos = self.compiled_value.access_handle.get_dir_infos()
# We could use `unsafe` here as well, especially as a parameter to
# get_dir_infos. But this would lead to a lot of property executions
# that are probably not wanted. The drawback for this is that we
# have a different name for `get` and `values`. For `get` we always
# execute.
for name in dir_infos:
names += self._get(
name,
lambda name, unsafe: dir_infos[name],
lambda name: name in dir_infos,
)
# ``dir`` doesn't include the type names.
if not self.is_instance and needs_type_completions:
for filter in builtin_from_name(self._inference_state, u'type').get_filters():
names += filter.values()
return names
def _create_name(self, name):
return CompiledName(
self._inference_state,
self.compiled_value,
name
)
def __repr__(self):
return "<%s: %s>" % (self.__class__.__name__, self.compiled_value)
docstr_defaults = {
'floating point number': u'float',
'character': u'str',
'integer': u'int',
'dictionary': u'dict',
'string': u'str',
}
def _parse_function_doc(doc):
"""
Takes a function and returns the params and return value as a tuple.
This is nothing more than a docstring parser.
TODO docstrings like utime(path, (atime, mtime)) and a(b [, b]) -> None
TODO docstrings like 'tuple of integers'
"""
doc = force_unicode(doc)
# parse round parentheses: def func(a, (b,c))
try:
count = 0
start = doc.index('(')
for i, s in enumerate(doc[start:]):
if s == '(':
count += 1
elif s == ')':
count -= 1
if count == 0:
end = start + i
break
param_str = doc[start + 1:end]
except (ValueError, UnboundLocalError):
# ValueError for doc.index
# UnboundLocalError for undefined end in last line
debug.dbg('no brackets found - no param')
end = 0
param_str = u''
else:
# remove square brackets, that show an optional param ( = None)
def change_options(m):
args = m.group(1).split(',')
for i, a in enumerate(args):
if a and '=' not in a:
args[i] += '=None'
return ','.join(args)
while True:
param_str, changes = re.subn(r' ?\[([^\[\]]+)\]',
change_options, param_str)
if changes == 0:
break
param_str = param_str.replace('-', '_') # see: isinstance.__doc__
# parse return value
r = re.search(u'-[>-]* ', doc[end:end + 7])
if r is None:
ret = u''
else:
index = end + r.end()
# get result type, which can contain newlines
pattern = re.compile(r'(,\n|[^\n-])+')
ret_str = pattern.match(doc, index).group(0).strip()
# New object -> object()
ret_str = re.sub(r'[nN]ew (.*)', r'\1()', ret_str)
ret = docstr_defaults.get(ret_str, ret_str)
return param_str, ret
def create_from_name(inference_state, compiled_value, name):
access_paths = compiled_value.access_handle.getattr_paths(name, default=None)
value = None
for access_path in access_paths:
value = create_cached_compiled_value(
inference_state,
access_path,
parent_context=None if value is None else value.as_context(),
)
return value
def _normalize_create_args(func):
"""The cache doesn't care about keyword vs. normal args."""
def wrapper(inference_state, obj, parent_context=None):
return func(inference_state, obj, parent_context)
return wrapper
def create_from_access_path(inference_state, access_path):
value = None
for name, access in access_path.accesses:
value = create_cached_compiled_value(
inference_state,
access,
parent_context=None if value is None else value.as_context()
)
return value
@_normalize_create_args
@inference_state_function_cache()
def create_cached_compiled_value(inference_state, access_handle, parent_context):
assert not isinstance(parent_context, CompiledValue)
if parent_context is None:
cls = CompiledModule
else:
cls = CompiledValue
return cls(inference_state, access_handle, parent_context)

View file

@ -0,0 +1,500 @@
from abc import abstractmethod
from contextlib import contextmanager
from parso.tree import search_ancestor
from parso.python.tree import Name
from jedi.inference.filters import ParserTreeFilter, MergedFilter, \
GlobalNameFilter
from jedi.inference.names import AnonymousParamName, TreeNameDefinition
from jedi.inference.base_value import NO_VALUES, ValueSet
from jedi.parser_utils import get_parent_scope
from jedi import debug
from jedi import parser_utils
class AbstractContext(object):
# Must be defined: inference_state and tree_node and parent_context as an attribute/property
def __init__(self, inference_state):
self.inference_state = inference_state
self.predefined_names = {}
@abstractmethod
def get_filters(self, until_position=None, origin_scope=None):
raise NotImplementedError
def goto(self, name_or_str, position):
from jedi.inference import finder
filters = _get_global_filters_for_name(
self, name_or_str if isinstance(name_or_str, Name) else None, position,
)
names = finder.filter_name(filters, name_or_str)
debug.dbg('context.goto %s in (%s): %s', name_or_str, self, names)
return names
def py__getattribute__(self, name_or_str, name_context=None, position=None,
analysis_errors=True):
"""
:param position: Position of the last statement -> tuple of line, column
"""
if name_context is None:
name_context = self
names = self.goto(name_or_str, position)
string_name = name_or_str.value if isinstance(name_or_str, Name) else name_or_str
# This paragraph is currently needed for proper branch type inference
# (static analysis).
found_predefined_types = None
if self.predefined_names and isinstance(name_or_str, Name):
node = name_or_str
while node is not None and not parser_utils.is_scope(node):
node = node.parent
if node.type in ("if_stmt", "for_stmt", "comp_for", 'sync_comp_for'):
try:
name_dict = self.predefined_names[node]
types = name_dict[string_name]
except KeyError:
continue
else:
found_predefined_types = types
break
if found_predefined_types is not None and names:
from jedi.inference import flow_analysis
check = flow_analysis.reachability_check(
context=self,
value_scope=self.tree_node,
node=name_or_str,
)
if check is flow_analysis.UNREACHABLE:
values = NO_VALUES
else:
values = found_predefined_types
else:
values = ValueSet.from_sets(name.infer() for name in names)
if not names and not values and analysis_errors:
if isinstance(name_or_str, Name):
from jedi.inference import analysis
message = ("NameError: name '%s' is not defined." % string_name)
analysis.add(name_context, 'name-error', name_or_str, message)
debug.dbg('context.names_to_types: %s -> %s', names, values)
if values:
return values
return self._check_for_additional_knowledge(name_or_str, name_context, position)
def _check_for_additional_knowledge(self, name_or_str, name_context, position):
name_context = name_context or self
# Add isinstance and other if/assert knowledge.
if isinstance(name_or_str, Name) and not name_context.is_instance():
flow_scope = name_or_str
base_nodes = [name_context.tree_node]
if any(b.type in ('comp_for', 'sync_comp_for') for b in base_nodes):
return NO_VALUES
from jedi.inference.finder import check_flow_information
while True:
flow_scope = get_parent_scope(flow_scope, include_flows=True)
n = check_flow_information(name_context, flow_scope,
name_or_str, position)
if n is not None:
return n
if flow_scope in base_nodes:
break
return NO_VALUES
def get_root_context(self):
parent_context = self.parent_context
if parent_context is None:
return self
return parent_context.get_root_context()
def is_module(self):
return False
def is_builtins_module(self):
return False
def is_class(self):
return False
def is_stub(self):
return False
def is_instance(self):
return False
def is_compiled(self):
return False
def is_bound_method(self):
return False
@abstractmethod
def py__name__(self):
raise NotImplementedError
def get_value(self):
raise NotImplementedError
@property
def name(self):
return None
def get_qualified_names(self):
return ()
def py__doc__(self):
return ''
@contextmanager
def predefine_names(self, flow_scope, dct):
predefined = self.predefined_names
predefined[flow_scope] = dct
try:
yield
finally:
del predefined[flow_scope]
class ValueContext(AbstractContext):
"""
Should be defined, otherwise the API returns empty types.
"""
def __init__(self, value):
super(ValueContext, self).__init__(value.inference_state)
self._value = value
@property
def tree_node(self):
return self._value.tree_node
@property
def parent_context(self):
return self._value.parent_context
def is_module(self):
return self._value.is_module()
def is_builtins_module(self):
return self._value == self.inference_state.builtins_module
def is_class(self):
return self._value.is_class()
def is_stub(self):
return self._value.is_stub()
def is_instance(self):
return self._value.is_instance()
def is_compiled(self):
return self._value.is_compiled()
def is_bound_method(self):
return self._value.is_bound_method()
def py__name__(self):
return self._value.py__name__()
@property
def name(self):
return self._value.name
def get_qualified_names(self):
return self._value.get_qualified_names()
def py__doc__(self):
return self._value.py__doc__()
def get_value(self):
return self._value
def __repr__(self):
return '%s(%s)' % (self.__class__.__name__, self._value)
class TreeContextMixin(object):
def infer_node(self, node):
from jedi.inference.syntax_tree import infer_node
return infer_node(self, node)
def create_value(self, node):
from jedi.inference import value
if node == self.tree_node:
assert self.is_module()
return self.get_value()
parent_context = self.create_context(node)
if node.type in ('funcdef', 'lambdef'):
func = value.FunctionValue.from_context(parent_context, node)
if parent_context.is_class():
class_value = parent_context.parent_context.create_value(parent_context.tree_node)
instance = value.AnonymousInstance(
self.inference_state, parent_context.parent_context, class_value)
func = value.BoundMethod(
instance=instance,
class_context=class_value.as_context(),
function=func
)
return func
elif node.type == 'classdef':
return value.ClassValue(self.inference_state, parent_context, node)
else:
raise NotImplementedError("Probably shouldn't happen: %s" % node)
def create_context(self, node):
def from_scope_node(scope_node, is_nested=True):
if scope_node == self.tree_node:
return self
if scope_node.type in ('funcdef', 'lambdef', 'classdef'):
return self.create_value(scope_node).as_context()
elif scope_node.type in ('comp_for', 'sync_comp_for'):
parent_scope = parser_utils.get_parent_scope(scope_node)
parent_context = from_scope_node(parent_scope)
if node.start_pos >= scope_node.children[-1].start_pos:
return parent_context
return CompForContext(parent_context, scope_node)
raise Exception("There's a scope that was not managed: %s" % scope_node)
def parent_scope(node):
while True:
node = node.parent
if parser_utils.is_scope(node):
return node
elif node.type in ('argument', 'testlist_comp'):
if node.children[1].type in ('comp_for', 'sync_comp_for'):
return node.children[1]
elif node.type == 'dictorsetmaker':
for n in node.children[1:4]:
# In dictionaries it can be pretty much anything.
if n.type in ('comp_for', 'sync_comp_for'):
return n
scope_node = parent_scope(node)
if scope_node.type in ('funcdef', 'classdef'):
colon = scope_node.children[scope_node.children.index(':')]
if node.start_pos < colon.start_pos:
parent = node.parent
if not (parent.type == 'param' and parent.name == node):
scope_node = parent_scope(scope_node)
return from_scope_node(scope_node, is_nested=True)
def create_name(self, tree_name):
definition = tree_name.get_definition()
if definition and definition.type == 'param' and definition.name == tree_name:
funcdef = search_ancestor(definition, 'funcdef', 'lambdef')
func = self.create_value(funcdef)
return AnonymousParamName(func, tree_name)
else:
context = self.create_context(tree_name)
return TreeNameDefinition(context, tree_name)
class FunctionContext(TreeContextMixin, ValueContext):
def get_filters(self, until_position=None, origin_scope=None):
yield ParserTreeFilter(
self.inference_state,
parent_context=self,
until_position=until_position,
origin_scope=origin_scope
)
class ModuleContext(TreeContextMixin, ValueContext):
def py__file__(self):
return self._value.py__file__()
def get_filters(self, until_position=None, origin_scope=None):
filters = self._value.get_filters(origin_scope)
# Skip the first filter and replace it.
next(filters, None)
yield MergedFilter(
ParserTreeFilter(
parent_context=self,
until_position=until_position,
origin_scope=origin_scope
),
self.get_global_filter(),
)
for f in filters: # Python 2...
yield f
def get_global_filter(self):
return GlobalNameFilter(self, self.tree_node)
@property
def string_names(self):
return self._value.string_names
@property
def code_lines(self):
return self._value.code_lines
def get_value(self):
"""
This is the only function that converts a context back to a value.
This is necessary for stub -> python conversion and vice versa. However
this method shouldn't be moved to AbstractContext.
"""
return self._value
class NamespaceContext(TreeContextMixin, ValueContext):
def get_filters(self, until_position=None, origin_scope=None):
return self._value.get_filters()
def get_value(self):
return self._value
@property
def string_names(self):
return self._value.string_names
def py__file__(self):
return self._value.py__file__()
class ClassContext(TreeContextMixin, ValueContext):
def get_filters(self, until_position=None, origin_scope=None):
yield self.get_global_filter(until_position, origin_scope)
def get_global_filter(self, until_position=None, origin_scope=None):
return ParserTreeFilter(
parent_context=self,
until_position=until_position,
origin_scope=origin_scope
)
class CompForContext(TreeContextMixin, AbstractContext):
def __init__(self, parent_context, comp_for):
super(CompForContext, self).__init__(parent_context.inference_state)
self.tree_node = comp_for
self.parent_context = parent_context
def get_filters(self, until_position=None, origin_scope=None):
yield ParserTreeFilter(self)
def get_value(self):
return None
def py__name__(self):
return '<comprehension context>'
def __repr__(self):
return '%s(%s)' % (self.__class__.__name__, self.tree_node)
class CompiledContext(ValueContext):
def get_filters(self, until_position=None, origin_scope=None):
return self._value.get_filters()
class CompiledModuleContext(CompiledContext):
code_lines = None
def get_value(self):
return self._value
@property
def string_names(self):
return self._value.string_names
def py__file__(self):
return self._value.py__file__()
def _get_global_filters_for_name(context, name_or_none, position):
# For functions and classes the defaults don't belong to the
# function and get inferred in the value before the function. So
# make sure to exclude the function/class name.
if name_or_none is not None:
ancestor = search_ancestor(name_or_none, 'funcdef', 'classdef', 'lambdef')
lambdef = None
if ancestor == 'lambdef':
# For lambdas it's even more complicated since parts will
# be inferred later.
lambdef = ancestor
ancestor = search_ancestor(name_or_none, 'funcdef', 'classdef')
if ancestor is not None:
colon = ancestor.children[-2]
if position is not None and position < colon.start_pos:
if lambdef is None or position < lambdef.children[-2].start_pos:
position = ancestor.start_pos
return get_global_filters(context, position, name_or_none)
def get_global_filters(context, until_position, origin_scope):
"""
Returns all filters in order of priority for name resolution.
For global name lookups. The filters will handle name resolution
themselves, but here we gather possible filters downwards.
>>> from jedi._compatibility import u, no_unicode_pprint
>>> from jedi import Script
>>> script = Script(u('''
... x = ['a', 'b', 'c']
... def func():
... y = None
... '''))
>>> module_node = script._module_node
>>> scope = next(module_node.iter_funcdefs())
>>> scope
<Function: func@3-5>
>>> context = script._get_module_context().create_context(scope)
>>> filters = list(get_global_filters(context, (4, 0), None))
First we get the names from the function scope.
>>> no_unicode_pprint(filters[0]) # doctest: +ELLIPSIS
MergedFilter(<ParserTreeFilter: ...>, <GlobalNameFilter: ...>)
>>> sorted(str(n) for n in filters[0].values()) # doctest: +NORMALIZE_WHITESPACE
['<TreeNameDefinition: string_name=func start_pos=(3, 4)>',
'<TreeNameDefinition: string_name=x start_pos=(2, 0)>']
>>> filters[0]._filters[0]._until_position
(4, 0)
>>> filters[0]._filters[1]._until_position
Then it yields the names from one level "lower". In this example, this is
the module scope (including globals).
As a side note, you can see, that the position in the filter is None on the
globals filter, because there the whole module is searched.
>>> list(filters[1].values()) # package modules -> Also empty.
[]
>>> sorted(name.string_name for name in filters[2].values()) # Module attributes
['__doc__', '__name__', '__package__']
Finally, it yields the builtin filter, if `include_builtin` is
true (default).
>>> list(filters[3].values()) # doctest: +ELLIPSIS
[...]
"""
base_context = context
from jedi.inference.value.function import BaseFunctionExecutionContext
while context is not None:
# Names in methods cannot be resolved within the class.
for filter in context.get_filters(
until_position=until_position,
origin_scope=origin_scope):
yield filter
if isinstance(context, (BaseFunctionExecutionContext, ModuleContext)):
# The position should be reset if the current scope is a function.
until_position = None
context = context.parent_context
b = next(base_context.inference_state.builtins_module.get_filters(), None)
assert b is not None
# Add builtins to the global scope.
yield b

View file

@ -0,0 +1,307 @@
"""
Docstrings are another source of information for functions and classes.
:mod:`jedi.inference.dynamic_params` tries to find all executions of functions,
while the docstring parsing is much easier. There are three different types of
docstrings that |jedi| understands:
- `Sphinx <http://sphinx-doc.org/markup/desc.html#info-field-lists>`_
- `Epydoc <http://epydoc.sourceforge.net/manual-fields.html>`_
- `Numpydoc <https://github.com/numpy/numpy/blob/master/doc/HOWTO_DOCUMENT.rst.txt>`_
For example, the sphinx annotation ``:type foo: str`` clearly states that the
type of ``foo`` is ``str``.
As an addition to parameter searching, this module also provides return
annotations.
"""
import re
import warnings
from textwrap import dedent
from parso import parse, ParserSyntaxError
from jedi._compatibility import u
from jedi import debug
from jedi.common import indent_block
from jedi.inference.cache import inference_state_method_cache
from jedi.inference.base_value import iterator_to_value_set, ValueSet, \
NO_VALUES
from jedi.inference.lazy_value import LazyKnownValues
DOCSTRING_PARAM_PATTERNS = [
r'\s*:type\s+%s:\s*([^\n]+)', # Sphinx
r'\s*:param\s+(\w+)\s+%s:[^\n]*', # Sphinx param with type
r'\s*@type\s+%s:\s*([^\n]+)', # Epydoc
]
DOCSTRING_RETURN_PATTERNS = [
re.compile(r'\s*:rtype:\s*([^\n]+)', re.M), # Sphinx
re.compile(r'\s*@rtype:\s*([^\n]+)', re.M), # Epydoc
]
REST_ROLE_PATTERN = re.compile(r':[^`]+:`([^`]+)`')
_numpy_doc_string_cache = None
def _get_numpy_doc_string_cls():
global _numpy_doc_string_cache
if isinstance(_numpy_doc_string_cache, (ImportError, SyntaxError)):
raise _numpy_doc_string_cache
from numpydoc.docscrape import NumpyDocString
_numpy_doc_string_cache = NumpyDocString
return _numpy_doc_string_cache
def _search_param_in_numpydocstr(docstr, param_str):
"""Search `docstr` (in numpydoc format) for type(-s) of `param_str`."""
with warnings.catch_warnings():
warnings.simplefilter("ignore")
try:
# This is a non-public API. If it ever changes we should be
# prepared and return gracefully.
params = _get_numpy_doc_string_cls()(docstr)._parsed_data['Parameters']
except Exception:
return []
for p_name, p_type, p_descr in params:
if p_name == param_str:
m = re.match(r'([^,]+(,[^,]+)*?)(,[ ]*optional)?$', p_type)
if m:
p_type = m.group(1)
return list(_expand_typestr(p_type))
return []
def _search_return_in_numpydocstr(docstr):
"""
Search `docstr` (in numpydoc format) for type(-s) of function returns.
"""
with warnings.catch_warnings():
warnings.simplefilter("ignore")
try:
doc = _get_numpy_doc_string_cls()(docstr)
except Exception:
return
try:
# This is a non-public API. If it ever changes we should be
# prepared and return gracefully.
returns = doc._parsed_data['Returns']
returns += doc._parsed_data['Yields']
except Exception:
return
for r_name, r_type, r_descr in returns:
# Return names are optional and if so the type is in the name
if not r_type:
r_type = r_name
for type_ in _expand_typestr(r_type):
yield type_
def _expand_typestr(type_str):
"""
Attempts to interpret the possible types in `type_str`
"""
# Check if alternative types are specified with 'or'
if re.search(r'\bor\b', type_str):
for t in type_str.split('or'):
yield t.split('of')[0].strip()
# Check if like "list of `type`" and set type to list
elif re.search(r'\bof\b', type_str):
yield type_str.split('of')[0]
# Check if type has is a set of valid literal values eg: {'C', 'F', 'A'}
elif type_str.startswith('{'):
node = parse(type_str, version='3.7').children[0]
if node.type == 'atom':
for leaf in node.children[1].children:
if leaf.type == 'number':
if '.' in leaf.value:
yield 'float'
else:
yield 'int'
elif leaf.type == 'string':
if 'b' in leaf.string_prefix.lower():
yield 'bytes'
else:
yield 'str'
# Ignore everything else.
# Otherwise just work with what we have.
else:
yield type_str
def _search_param_in_docstr(docstr, param_str):
"""
Search `docstr` for type(-s) of `param_str`.
>>> _search_param_in_docstr(':type param: int', 'param')
['int']
>>> _search_param_in_docstr('@type param: int', 'param')
['int']
>>> _search_param_in_docstr(
... ':type param: :class:`threading.Thread`', 'param')
['threading.Thread']
>>> bool(_search_param_in_docstr('no document', 'param'))
False
>>> _search_param_in_docstr(':param int param: some description', 'param')
['int']
"""
# look at #40 to see definitions of those params
patterns = [re.compile(p % re.escape(param_str))
for p in DOCSTRING_PARAM_PATTERNS]
for pattern in patterns:
match = pattern.search(docstr)
if match:
return [_strip_rst_role(match.group(1))]
return _search_param_in_numpydocstr(docstr, param_str)
def _strip_rst_role(type_str):
"""
Strip off the part looks like a ReST role in `type_str`.
>>> _strip_rst_role(':class:`ClassName`') # strip off :class:
'ClassName'
>>> _strip_rst_role(':py:obj:`module.Object`') # works with domain
'module.Object'
>>> _strip_rst_role('ClassName') # do nothing when not ReST role
'ClassName'
See also:
http://sphinx-doc.org/domains.html#cross-referencing-python-objects
"""
match = REST_ROLE_PATTERN.match(type_str)
if match:
return match.group(1)
else:
return type_str
def _infer_for_statement_string(module_context, string):
code = dedent(u("""
def pseudo_docstring_stuff():
'''
Create a pseudo function for docstring statements.
Need this docstring so that if the below part is not valid Python this
is still a function.
'''
{}
"""))
if string is None:
return []
for element in re.findall(r'((?:\w+\.)*\w+)\.', string):
# Try to import module part in dotted name.
# (e.g., 'threading' in 'threading.Thread').
string = 'import %s\n' % element + string
# Take the default grammar here, if we load the Python 2.7 grammar here, it
# will be impossible to use `...` (Ellipsis) as a token. Docstring types
# don't need to conform with the current grammar.
debug.dbg('Parse docstring code %s', string, color='BLUE')
grammar = module_context.inference_state.latest_grammar
try:
module = grammar.parse(code.format(indent_block(string)), error_recovery=False)
except ParserSyntaxError:
return []
try:
funcdef = next(module.iter_funcdefs())
# First pick suite, then simple_stmt and then the node,
# which is also not the last item, because there's a newline.
stmt = funcdef.children[-1].children[-1].children[-2]
except (AttributeError, IndexError):
return []
if stmt.type not in ('name', 'atom', 'atom_expr'):
return []
from jedi.inference.value import FunctionValue
function_value = FunctionValue(
module_context.inference_state,
module_context,
funcdef
)
func_execution_context = function_value.as_context()
# Use the module of the param.
# TODO this module is not the module of the param in case of a function
# call. In that case it's the module of the function call.
# stuffed with content from a function call.
return list(_execute_types_in_stmt(func_execution_context, stmt))
def _execute_types_in_stmt(module_context, stmt):
"""
Executing all types or general elements that we find in a statement. This
doesn't include tuple, list and dict literals, because the stuff they
contain is executed. (Used as type information).
"""
definitions = module_context.infer_node(stmt)
return ValueSet.from_sets(
_execute_array_values(module_context.inference_state, d)
for d in definitions
)
def _execute_array_values(inference_state, array):
"""
Tuples indicate that there's not just one return value, but the listed
ones. `(str, int)` means that it returns a tuple with both types.
"""
from jedi.inference.value.iterable import SequenceLiteralValue, FakeTuple, FakeList
if isinstance(array, SequenceLiteralValue) and array.array_type in ('tuple', 'list'):
values = []
for lazy_value in array.py__iter__():
objects = ValueSet.from_sets(
_execute_array_values(inference_state, typ)
for typ in lazy_value.infer()
)
values.append(LazyKnownValues(objects))
cls = FakeTuple if array.array_type == 'tuple' else FakeList
return {cls(inference_state, values)}
else:
return array.execute_annotation()
@inference_state_method_cache()
def infer_param(function_value, param):
def infer_docstring(docstring):
return ValueSet(
p
for param_str in _search_param_in_docstr(docstring, param.name.value)
for p in _infer_for_statement_string(module_context, param_str)
)
module_context = function_value.get_root_context()
func = param.get_parent_function()
if func.type == 'lambdef':
return NO_VALUES
types = infer_docstring(function_value.py__doc__())
if function_value.is_bound_method() \
and function_value.py__name__() == '__init__':
types |= infer_docstring(function_value.class_context.py__doc__())
debug.dbg('Found param types for docstring: %s', types, color='BLUE')
return types
@inference_state_method_cache()
@iterator_to_value_set
def infer_return_types(function_value):
def search_return_in_docstr(code):
for p in DOCSTRING_RETURN_PATTERNS:
match = p.search(code)
if match:
yield _strip_rst_role(match.group(1))
# Check for numpy style return hint
for type_ in _search_return_in_numpydocstr(code):
yield type_
for type_str in search_return_in_docstr(function_value.py__doc__()):
for value in _infer_for_statement_string(function_value.get_root_context(), type_str):
yield value

View file

@ -0,0 +1,226 @@
"""
One of the really important features of |jedi| is to have an option to
understand code like this::
def foo(bar):
bar. # completion here
foo(1)
There's no doubt wheter bar is an ``int`` or not, but if there's also a call
like ``foo('str')``, what would happen? Well, we'll just show both. Because
that's what a human would expect.
It works as follows:
- |Jedi| sees a param
- search for function calls named ``foo``
- execute these calls and check the input.
"""
from jedi import settings
from jedi import debug
from jedi.parser_utils import get_parent_scope
from jedi.inference.cache import inference_state_method_cache
from jedi.inference.arguments import TreeArguments
from jedi.inference.param import get_executed_param_names
from jedi.inference.helpers import is_stdlib_path
from jedi.inference.utils import to_list
from jedi.inference.value import instance
from jedi.inference.base_value import ValueSet, NO_VALUES
from jedi.inference.references import get_module_contexts_containing_name
from jedi.inference import recursion
MAX_PARAM_SEARCHES = 20
def _avoid_recursions(func):
def wrapper(function_value, param_index):
inf = function_value.inference_state
with recursion.execution_allowed(inf, function_value.tree_node) as allowed:
# We need to catch recursions that may occur, because an
# anonymous functions can create an anonymous parameter that is
# more or less self referencing.
if allowed:
inf.dynamic_params_depth += 1
try:
return func(function_value, param_index)
finally:
inf.dynamic_params_depth -= 1
return NO_VALUES
return wrapper
@debug.increase_indent
@_avoid_recursions
def dynamic_param_lookup(function_value, param_index):
"""
A dynamic search for param values. If you try to complete a type:
>>> def func(foo):
... foo
>>> func(1)
>>> func("")
It is not known what the type ``foo`` without analysing the whole code. You
have to look for all calls to ``func`` to find out what ``foo`` possibly
is.
"""
funcdef = function_value.tree_node
if not settings.dynamic_params:
return NO_VALUES
path = function_value.get_root_context().py__file__()
if path is not None and is_stdlib_path(path):
# We don't want to search for references in the stdlib. Usually people
# don't work with it (except if you are a core maintainer, sorry).
# This makes everything slower. Just disable it and run the tests,
# you will see the slowdown, especially in 3.6.
return NO_VALUES
if funcdef.type == 'lambdef':
string_name = _get_lambda_name(funcdef)
if string_name is None:
return NO_VALUES
else:
string_name = funcdef.name.value
debug.dbg('Dynamic param search in %s.', string_name, color='MAGENTA')
module_context = function_value.get_root_context()
arguments_list = _search_function_arguments(module_context, funcdef, string_name)
values = ValueSet.from_sets(
get_executed_param_names(
function_value, arguments
)[param_index].infer()
for arguments in arguments_list
)
debug.dbg('Dynamic param result finished', color='MAGENTA')
return values
@inference_state_method_cache(default=None)
@to_list
def _search_function_arguments(module_context, funcdef, string_name):
"""
Returns a list of param names.
"""
compare_node = funcdef
if string_name == '__init__':
cls = get_parent_scope(funcdef)
if cls.type == 'classdef':
string_name = cls.name.value
compare_node = cls
found_arguments = False
i = 0
inference_state = module_context.inference_state
if settings.dynamic_params_for_other_modules:
module_contexts = get_module_contexts_containing_name(
inference_state, [module_context], string_name,
# Limit the amounts of files to be opened massively.
limit_reduction=5,
)
else:
module_contexts = [module_context]
for for_mod_context in module_contexts:
for name, trailer in _get_potential_nodes(for_mod_context, string_name):
i += 1
# This is a simple way to stop Jedi's dynamic param recursion
# from going wild: The deeper Jedi's in the recursion, the less
# code should be inferred.
if i * inference_state.dynamic_params_depth > MAX_PARAM_SEARCHES:
return
random_context = for_mod_context.create_context(name)
for arguments in _check_name_for_execution(
inference_state, random_context, compare_node, name, trailer):
found_arguments = True
yield arguments
# If there are results after processing a module, we're probably
# good to process. This is a speed optimization.
if found_arguments:
return
def _get_lambda_name(node):
stmt = node.parent
if stmt.type == 'expr_stmt':
first_operator = next(stmt.yield_operators(), None)
if first_operator == '=':
first = stmt.children[0]
if first.type == 'name':
return first.value
return None
def _get_potential_nodes(module_value, func_string_name):
try:
names = module_value.tree_node.get_used_names()[func_string_name]
except KeyError:
return
for name in names:
bracket = name.get_next_leaf()
trailer = bracket.parent
if trailer.type == 'trailer' and bracket == '(':
yield name, trailer
def _check_name_for_execution(inference_state, context, compare_node, name, trailer):
from jedi.inference.value.function import BaseFunctionExecutionContext
def create_args(value):
arglist = trailer.children[1]
if arglist == ')':
arglist = None
args = TreeArguments(inference_state, context, arglist, trailer)
from jedi.inference.value.instance import InstanceArguments
if value.tree_node.type == 'classdef':
created_instance = instance.TreeInstance(
inference_state,
value.parent_context,
value,
args
)
return InstanceArguments(created_instance, args)
else:
if value.is_bound_method():
args = InstanceArguments(value.instance, args)
return args
for value in inference_state.infer(context, name):
value_node = value.tree_node
if compare_node == value_node:
yield create_args(value)
elif isinstance(value.parent_context, BaseFunctionExecutionContext) \
and compare_node.type == 'funcdef':
# Here we're trying to find decorators by checking the first
# parameter. It's not very generic though. Should find a better
# solution that also applies to nested decorators.
param_names = value.parent_context.get_param_names()
if len(param_names) != 1:
continue
values = param_names[0].infer()
if [v.tree_node for v in values] == [compare_node]:
# Found a decorator.
module_context = context.get_root_context()
execution_context = value.as_context(create_args(value))
potential_nodes = _get_potential_nodes(module_context, param_names[0].string_name)
for name, trailer in potential_nodes:
if value_node.start_pos < name.start_pos < value_node.end_pos:
random_context = execution_context.create_context(name)
iterator = _check_name_for_execution(
inference_state,
random_context,
compare_node,
name,
trailer
)
for arguments in iterator:
yield arguments

View file

@ -0,0 +1,353 @@
"""
Filters are objects that you can use to filter names in different scopes. They
are needed for name resolution.
"""
from abc import abstractmethod
import weakref
from parso.tree import search_ancestor
from jedi._compatibility import use_metaclass
from jedi.inference import flow_analysis
from jedi.inference.base_value import ValueSet, ValueWrapper, \
LazyValueWrapper
from jedi.parser_utils import get_cached_parent_scope
from jedi.inference.utils import to_list
from jedi.inference.names import TreeNameDefinition, ParamName, \
AnonymousParamName, AbstractNameDefinition
_definition_name_cache = weakref.WeakKeyDictionary()
class AbstractFilter(object):
_until_position = None
def _filter(self, names):
if self._until_position is not None:
return [n for n in names if n.start_pos < self._until_position]
return names
@abstractmethod
def get(self, name):
raise NotImplementedError
@abstractmethod
def values(self):
raise NotImplementedError
class FilterWrapper(object):
name_wrapper_class = None
def __init__(self, wrapped_filter):
self._wrapped_filter = wrapped_filter
def wrap_names(self, names):
return [self.name_wrapper_class(name) for name in names]
def get(self, name):
return self.wrap_names(self._wrapped_filter.get(name))
def values(self):
return self.wrap_names(self._wrapped_filter.values())
def _get_definition_names(used_names, name_key):
try:
for_module = _definition_name_cache[used_names]
except KeyError:
for_module = _definition_name_cache[used_names] = {}
try:
return for_module[name_key]
except KeyError:
names = used_names.get(name_key, ())
result = for_module[name_key] = tuple(
name for name in names if name.is_definition(include_setitem=True)
)
return result
class AbstractUsedNamesFilter(AbstractFilter):
name_class = TreeNameDefinition
def __init__(self, parent_context, parser_scope):
self._parser_scope = parser_scope
self._module_node = self._parser_scope.get_root_node()
self._used_names = self._module_node.get_used_names()
self.parent_context = parent_context
def get(self, name):
return self._convert_names(self._filter(
_get_definition_names(self._used_names, name),
))
def _convert_names(self, names):
return [self.name_class(self.parent_context, name) for name in names]
def values(self):
return self._convert_names(
name
for name_key in self._used_names
for name in self._filter(
_get_definition_names(self._used_names, name_key),
)
)
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, self.parent_context)
class ParserTreeFilter(AbstractUsedNamesFilter):
def __init__(self, parent_context, node_context=None, until_position=None,
origin_scope=None):
"""
node_context is an option to specify a second value for use cases
like the class mro where the parent class of a new name would be the
value, but for some type inference it's important to have a local
value of the other classes.
"""
if node_context is None:
node_context = parent_context
super(ParserTreeFilter, self).__init__(parent_context, node_context.tree_node)
self._node_context = node_context
self._origin_scope = origin_scope
self._until_position = until_position
def _filter(self, names):
names = super(ParserTreeFilter, self)._filter(names)
names = [n for n in names if self._is_name_reachable(n)]
return list(self._check_flows(names))
def _is_name_reachable(self, name):
parent = name.parent
if parent.type == 'trailer':
return False
base_node = parent if parent.type in ('classdef', 'funcdef') else name
return get_cached_parent_scope(self._used_names, base_node) == self._parser_scope
def _check_flows(self, names):
for name in sorted(names, key=lambda name: name.start_pos, reverse=True):
check = flow_analysis.reachability_check(
context=self._node_context,
value_scope=self._parser_scope,
node=name,
origin_scope=self._origin_scope
)
if check is not flow_analysis.UNREACHABLE:
yield name
if check is flow_analysis.REACHABLE:
break
class _FunctionExecutionFilter(ParserTreeFilter):
def __init__(self, parent_context, function_value, until_position, origin_scope):
super(_FunctionExecutionFilter, self).__init__(
parent_context,
until_position=until_position,
origin_scope=origin_scope,
)
self._function_value = function_value
def _convert_param(self, param, name):
raise NotImplementedError
@to_list
def _convert_names(self, names):
for name in names:
param = search_ancestor(name, 'param')
# Here we don't need to check if the param is a default/annotation,
# because those are not definitions and never make it to this
# point.
if param:
yield self._convert_param(param, name)
else:
yield TreeNameDefinition(self.parent_context, name)
class FunctionExecutionFilter(_FunctionExecutionFilter):
def __init__(self, *args, **kwargs):
self._arguments = kwargs.pop('arguments') # Python 2
super(FunctionExecutionFilter, self).__init__(*args, **kwargs)
def _convert_param(self, param, name):
return ParamName(self._function_value, name, self._arguments)
class AnonymousFunctionExecutionFilter(_FunctionExecutionFilter):
def _convert_param(self, param, name):
return AnonymousParamName(self._function_value, name)
class GlobalNameFilter(AbstractUsedNamesFilter):
def get(self, name):
try:
names = self._used_names[name]
except KeyError:
return []
return self._convert_names(self._filter(names))
@to_list
def _filter(self, names):
for name in names:
if name.parent.type == 'global_stmt':
yield name
def values(self):
return self._convert_names(
name for name_list in self._used_names.values()
for name in self._filter(name_list)
)
class DictFilter(AbstractFilter):
def __init__(self, dct):
self._dct = dct
def get(self, name):
try:
value = self._convert(name, self._dct[name])
except KeyError:
return []
else:
return list(self._filter([value]))
def values(self):
def yielder():
for item in self._dct.items():
try:
yield self._convert(*item)
except KeyError:
pass
return self._filter(yielder())
def _convert(self, name, value):
return value
def __repr__(self):
keys = ', '.join(self._dct.keys())
return '<%s: for {%s}>' % (self.__class__.__name__, keys)
class MergedFilter(object):
def __init__(self, *filters):
self._filters = filters
def get(self, name):
return [n for filter in self._filters for n in filter.get(name)]
def values(self):
return [n for filter in self._filters for n in filter.values()]
def __repr__(self):
return '%s(%s)' % (self.__class__.__name__, ', '.join(str(f) for f in self._filters))
class _BuiltinMappedMethod(ValueWrapper):
"""``Generator.__next__`` ``dict.values`` methods and so on."""
api_type = u'function'
def __init__(self, value, method, builtin_func):
super(_BuiltinMappedMethod, self).__init__(builtin_func)
self._value = value
self._method = method
def py__call__(self, arguments):
# TODO add TypeError if params are given/or not correct.
return self._method(self._value, arguments)
class SpecialMethodFilter(DictFilter):
"""
A filter for methods that are defined in this module on the corresponding
classes like Generator (for __next__, etc).
"""
class SpecialMethodName(AbstractNameDefinition):
api_type = u'function'
def __init__(self, parent_context, string_name, value, builtin_value):
callable_, python_version = value
if python_version is not None and \
python_version != parent_context.inference_state.environment.version_info.major:
raise KeyError
self.parent_context = parent_context
self.string_name = string_name
self._callable = callable_
self._builtin_value = builtin_value
def infer(self):
for filter in self._builtin_value.get_filters():
# We can take the first index, because on builtin methods there's
# always only going to be one name. The same is true for the
# inferred values.
for name in filter.get(self.string_name):
builtin_func = next(iter(name.infer()))
break
else:
continue
break
return ValueSet([
_BuiltinMappedMethod(self.parent_context, self._callable, builtin_func)
])
def __init__(self, value, dct, builtin_value):
super(SpecialMethodFilter, self).__init__(dct)
self.value = value
self._builtin_value = builtin_value
"""
This value is what will be used to introspect the name, where as the
other value will be used to execute the function.
We distinguish, because we have to.
"""
def _convert(self, name, value):
return self.SpecialMethodName(self.value, name, value, self._builtin_value)
class _OverwriteMeta(type):
def __init__(cls, name, bases, dct):
super(_OverwriteMeta, cls).__init__(name, bases, dct)
base_dct = {}
for base_cls in reversed(cls.__bases__):
try:
base_dct.update(base_cls.overwritten_methods)
except AttributeError:
pass
for func in cls.__dict__.values():
try:
base_dct.update(func.registered_overwritten_methods)
except AttributeError:
pass
cls.overwritten_methods = base_dct
class _AttributeOverwriteMixin(object):
def get_filters(self, *args, **kwargs):
yield SpecialMethodFilter(self, self.overwritten_methods, self._wrapped_value)
for filter in self._wrapped_value.get_filters(*args, **kwargs):
yield filter
class LazyAttributeOverwrite(use_metaclass(_OverwriteMeta, _AttributeOverwriteMixin,
LazyValueWrapper)):
def __init__(self, inference_state):
self.inference_state = inference_state
class AttributeOverwrite(use_metaclass(_OverwriteMeta, _AttributeOverwriteMixin,
ValueWrapper)):
pass
def publish_method(method_name, python_version_match=None):
def decorator(func):
dct = func.__dict__.setdefault('registered_overwritten_methods', {})
dct[method_name] = func, python_version_match
return func
return decorator

View file

@ -0,0 +1,146 @@
"""
Searching for names with given scope and name. This is very central in Jedi and
Python. The name resolution is quite complicated with descripter,
``__getattribute__``, ``__getattr__``, ``global``, etc.
If you want to understand name resolution, please read the first few chapters
in http://blog.ionelmc.ro/2015/02/09/understanding-python-metaclasses/.
Flow checks
+++++++++++
Flow checks are not really mature. There's only a check for ``isinstance``. It
would check whether a flow has the form of ``if isinstance(a, type_or_tuple)``.
Unfortunately every other thing is being ignored (e.g. a == '' would be easy to
check for -> a is a string). There's big potential in these checks.
"""
from parso.tree import search_ancestor
from parso.python.tree import Name
from jedi import settings
from jedi.inference.arguments import TreeArguments
from jedi.inference.value import iterable
from jedi.inference.base_value import NO_VALUES
from jedi.parser_utils import is_scope
def filter_name(filters, name_or_str):
"""
Searches names that are defined in a scope (the different
``filters``), until a name fits.
"""
string_name = name_or_str.value if isinstance(name_or_str, Name) else name_or_str
names = []
for filter in filters:
names = filter.get(string_name)
if names:
break
return list(_remove_del_stmt(names))
def _remove_del_stmt(names):
# Catch del statements and remove them from results.
for name in names:
if name.tree_name is not None:
definition = name.tree_name.get_definition()
if definition is not None and definition.type == 'del_stmt':
continue
yield name
def check_flow_information(value, flow, search_name, pos):
""" Try to find out the type of a variable just with the information that
is given by the flows: e.g. It is also responsible for assert checks.::
if isinstance(k, str):
k. # <- completion here
ensures that `k` is a string.
"""
if not settings.dynamic_flow_information:
return None
result = None
if is_scope(flow):
# Check for asserts.
module_node = flow.get_root_node()
try:
names = module_node.get_used_names()[search_name.value]
except KeyError:
return None
names = reversed([
n for n in names
if flow.start_pos <= n.start_pos < (pos or flow.end_pos)
])
for name in names:
ass = search_ancestor(name, 'assert_stmt')
if ass is not None:
result = _check_isinstance_type(value, ass.assertion, search_name)
if result is not None:
return result
if flow.type in ('if_stmt', 'while_stmt'):
potential_ifs = [c for c in flow.children[1::4] if c != ':']
for if_test in reversed(potential_ifs):
if search_name.start_pos > if_test.end_pos:
return _check_isinstance_type(value, if_test, search_name)
return result
def _get_isinstance_trailer_arglist(node):
if node.type in ('power', 'atom_expr') and len(node.children) == 2:
# This might be removed if we analyze and, etc
first, trailer = node.children
if first.type == 'name' and first.value == 'isinstance' \
and trailer.type == 'trailer' and trailer.children[0] == '(':
return trailer
return None
def _check_isinstance_type(value, node, search_name):
lazy_cls = None
trailer = _get_isinstance_trailer_arglist(node)
if trailer is not None and len(trailer.children) == 3:
arglist = trailer.children[1]
args = TreeArguments(value.inference_state, value, arglist, trailer)
param_list = list(args.unpack())
# Disallow keyword arguments
if len(param_list) == 2 and len(arglist.children) == 3:
(key1, _), (key2, lazy_value_cls) = param_list
if key1 is None and key2 is None:
call = _get_call_string(search_name)
is_instance_call = _get_call_string(arglist.children[0])
# Do a simple get_code comparison of the strings . They should
# just have the same code, and everything will be all right.
# There are ways that this is not correct, if some stuff is
# redefined in between. However here we don't care, because
# it's a heuristic that works pretty well.
if call == is_instance_call:
lazy_cls = lazy_value_cls
if lazy_cls is None:
return None
value_set = NO_VALUES
for cls_or_tup in lazy_cls.infer():
if isinstance(cls_or_tup, iterable.Sequence) and cls_or_tup.array_type == 'tuple':
for lazy_value in cls_or_tup.py__iter__():
value_set |= lazy_value.infer().execute_with_values()
else:
value_set |= cls_or_tup.execute_with_values()
return value_set
def _get_call_string(node):
if node.parent.type == 'atom_expr':
return _get_call_string(node.parent)
code = ''
leaf = node.get_first_leaf()
end = node.get_last_leaf().end_pos
while leaf.start_pos < end:
code += leaf.value
leaf = leaf.get_next_leaf()
return code

View file

@ -0,0 +1,123 @@
from jedi.parser_utils import get_flow_branch_keyword, is_scope, get_parent_scope
from jedi.inference.recursion import execution_allowed
from jedi.inference.helpers import is_big_annoying_library
class Status(object):
lookup_table = {}
def __init__(self, value, name):
self._value = value
self._name = name
Status.lookup_table[value] = self
def invert(self):
if self is REACHABLE:
return UNREACHABLE
elif self is UNREACHABLE:
return REACHABLE
else:
return UNSURE
def __and__(self, other):
if UNSURE in (self, other):
return UNSURE
else:
return REACHABLE if self._value and other._value else UNREACHABLE
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, self._name)
REACHABLE = Status(True, 'reachable')
UNREACHABLE = Status(False, 'unreachable')
UNSURE = Status(None, 'unsure')
def _get_flow_scopes(node):
while True:
node = get_parent_scope(node, include_flows=True)
if node is None or is_scope(node):
return
yield node
def reachability_check(context, value_scope, node, origin_scope=None):
if is_big_annoying_library(context) \
or not context.inference_state.flow_analysis_enabled:
return UNSURE
first_flow_scope = get_parent_scope(node, include_flows=True)
if origin_scope is not None:
origin_flow_scopes = list(_get_flow_scopes(origin_scope))
node_flow_scopes = list(_get_flow_scopes(node))
branch_matches = True
for flow_scope in origin_flow_scopes:
if flow_scope in node_flow_scopes:
node_keyword = get_flow_branch_keyword(flow_scope, node)
origin_keyword = get_flow_branch_keyword(flow_scope, origin_scope)
branch_matches = node_keyword == origin_keyword
if flow_scope.type == 'if_stmt':
if not branch_matches:
return UNREACHABLE
elif flow_scope.type == 'try_stmt':
if not branch_matches and origin_keyword == 'else' \
and node_keyword == 'except':
return UNREACHABLE
if branch_matches:
break
# Direct parents get resolved, we filter scopes that are separate
# branches. This makes sense for autocompletion and static analysis.
# For actual Python it doesn't matter, because we're talking about
# potentially unreachable code.
# e.g. `if 0:` would cause all name lookup within the flow make
# unaccessible. This is not a "problem" in Python, because the code is
# never called. In Jedi though, we still want to infer types.
while origin_scope is not None:
if first_flow_scope == origin_scope and branch_matches:
return REACHABLE
origin_scope = origin_scope.parent
return _break_check(context, value_scope, first_flow_scope, node)
def _break_check(context, value_scope, flow_scope, node):
reachable = REACHABLE
if flow_scope.type == 'if_stmt':
if flow_scope.is_node_after_else(node):
for check_node in flow_scope.get_test_nodes():
reachable = _check_if(context, check_node)
if reachable in (REACHABLE, UNSURE):
break
reachable = reachable.invert()
else:
flow_node = flow_scope.get_corresponding_test_node(node)
if flow_node is not None:
reachable = _check_if(context, flow_node)
elif flow_scope.type in ('try_stmt', 'while_stmt'):
return UNSURE
# Only reachable branches need to be examined further.
if reachable in (UNREACHABLE, UNSURE):
return reachable
if value_scope != flow_scope and value_scope != flow_scope.parent:
flow_scope = get_parent_scope(flow_scope, include_flows=True)
return reachable & _break_check(context, value_scope, flow_scope, node)
else:
return reachable
def _check_if(context, node):
with execution_allowed(context.inference_state, node) as allowed:
if not allowed:
return UNSURE
types = context.infer_node(node)
values = set(x.py__bool__() for x in types)
if len(values) == 1:
return Status.lookup_table[values.pop()]
else:
return UNSURE

Some files were not shown because too many files have changed in this diff Show more