Adding new stuff

This commit is contained in:
ViktorBarzin 2017-07-09 00:23:01 +03:00
parent 9ef8a96f9a
commit 0b3d063cb3
1580 changed files with 0 additions and 0 deletions

View file

@ -0,0 +1,43 @@
"""
Jedi is a static analysis tool for Python that can be used in IDEs/editors. Its
historic focus is autocompletion, but does static analysis for now as well.
Jedi is fast and is very well tested. It understands Python on a deeper level
than all other static analysis frameworks for Python.
Jedi has support for two different goto functions. It's possible to search for
related names and to list all names in a Python file and infer them. Jedi
understands docstrings and you can use Jedi autocompletion in your REPL as
well.
Jedi uses a very simple API to connect with IDE's. There's a reference
implementation as a `VIM-Plugin <https://github.com/davidhalter/jedi-vim>`_,
which uses Jedi's autocompletion. We encourage you to use Jedi in your IDEs.
It's really easy.
To give you a simple example how you can use the Jedi library, here is an
example for the autocompletion feature:
>>> import jedi
>>> source = '''
... import datetime
... datetime.da'''
>>> script = jedi.Script(source, 3, len('datetime.da'), 'example.py')
>>> script
<Script: 'example.py'>
>>> completions = script.completions()
>>> completions #doctest: +ELLIPSIS
[<Completion: date>, <Completion: datetime>, ...]
>>> print(completions[0].complete)
te
>>> print(completions[0].name)
date
As you see Jedi is pretty simple and allows you to concentrate on writing a
good text editor, while still having very good IDE features for Python.
"""
__version__ = '0.10.0'
from jedi.api import Script, Interpreter, NotFoundError, set_debug_function
from jedi.api import preload_module, defined_names, names
from jedi import settings

View file

@ -0,0 +1,48 @@
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 path in paths:
for error in jedi.Script(path=path)._analysis():
print(error)
except Exception:
if '--pdb' in sys.argv:
import traceback
traceback.print_exc()
import pdb
pdb.post_mortem()
else:
raise
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()

View file

@ -0,0 +1,263 @@
"""
To ensure compatibility from Python ``2.6`` - ``3.3``, a module has been
created. Clearly there is huge need to use conforming syntax.
"""
import sys
import imp
import os
import re
import pkgutil
try:
import importlib
except ImportError:
pass
is_py3 = sys.version_info[0] >= 3
is_py33 = is_py3 and sys.version_info.minor >= 3
is_py34 = is_py3 and sys.version_info.minor >= 4
is_py35 = is_py3 and sys.version_info.minor >= 5
is_py26 = not is_py3 and sys.version_info[1] < 7
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_py33(string, path=None):
loader = importlib.machinery.PathFinder.find_module(string, path)
if loader is None and path is None: # Fallback to find builtins
try:
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 {0}".format(string))
try:
is_package = loader.is_package(string)
if is_package:
if hasattr(loader, 'path'):
module_path = os.path.dirname(loader.path)
else:
# At least zipimporter does not have path attribute
module_path = os.path.dirname(loader.get_filename(string))
if hasattr(loader, 'archive'):
module_file = DummyFile(loader, string)
else:
module_file = None
else:
module_path = loader.get_filename(string)
module_file = DummyFile(loader, string)
except AttributeError:
# ExtensionLoader has not attribute get_filename, instead it has a
# path attribute that we can use to retrieve the module path
try:
module_path = loader.path
module_file = DummyFile(loader, string)
except AttributeError:
module_path = string
module_file = None
finally:
is_package = False
if hasattr(loader, 'archive'):
module_path = loader.archive
return module_file, module_path, is_package
def find_module_pre_py33(string, path=None):
try:
module_file, module_path, description = imp.find_module(string, path)
module_type = description[2]
return module_file, module_path, module_type is imp.PKG_DIRECTORY
except ImportError:
pass
if path is None:
path = sys.path
for item in path:
loader = pkgutil.get_importer(item)
if loader:
try:
loader = loader.find_module(string)
if loader:
is_package = loader.is_package(string)
is_archive = hasattr(loader, 'archive')
try:
module_path = loader.get_filename(string)
except AttributeError:
# fallback for py26
try:
module_path = loader._get_filename(string)
except AttributeError:
continue
if is_package:
module_path = os.path.dirname(module_path)
if is_archive:
module_path = loader.archive
file = None
if not is_package or is_archive:
file = DummyFile(loader, string)
return (file, module_path, is_package)
except ImportError:
pass
raise ImportError("No module named {0}".format(string))
find_module = find_module_py33 if is_py33 else find_module_pre_py33
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.
"""
# unicode function
try:
unicode = unicode
except NameError:
unicode = str
if is_py3:
u = lambda s: s
else:
u = lambda s: s.decode('utf-8')
u.__doc__ = """
Decode a raw string into unicode object. Do nothing in Python 3.
"""
# exec function
if is_py3:
def exec_function(source, global_map):
exec(source, global_map)
else:
eval(compile("""def exec_function(source, global_map):
exec source in global_map """, 'blub', 'exec'))
# 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])
"""
class Python3Method(object):
def __init__(self, func):
self.func = func
def __get__(self, obj, objtype):
if obj is None:
return lambda *args, **kwargs: self.func(*args, **kwargs)
else:
return lambda *args, **kwargs: self.func(obj, *args, **kwargs)
def use_metaclass(meta, *bases):
""" Create a class with a metaclass. """
if not bases:
bases = (object,)
return meta("HackClass", bases, {})
try:
encoding = sys.stdout.encoding
if encoding is None:
encoding = 'utf-8'
except AttributeError:
encoding = 'ascii'
def u(string):
"""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 is_py3:
return str(string)
elif not isinstance(string, unicode):
return unicode(str(string), 'UTF-8')
return string
try:
import builtins # module name in python 3
except ImportError:
import __builtin__ as builtins
import ast
def literal_eval(string):
# py3.0, py3.1 and py32 don't support unicode literals. Support those, I
# don't want to write two versions of the tokenizer.
if is_py3 and sys.version_info.minor < 3:
if re.match('[uU][\'"]', string):
string = string[1:]
return ast.literal_eval(string)
try:
from itertools import zip_longest
except ImportError:
from itertools import izip_longest as zip_longest # Python 2
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

View file

@ -0,0 +1,463 @@
"""
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.
.. warning:: Please, note that Jedi is **not thread safe**.
"""
import os
import warnings
import sys
from jedi._compatibility import unicode
from jedi.parser import load_grammar
from jedi.parser import tree
from jedi.parser.fast import FastParser
from jedi.parser.utils import save_parser
from jedi import debug
from jedi import settings
from jedi import common
from jedi import cache
from jedi.api import classes
from jedi.api import interpreter
from jedi.api import usages
from jedi.api import helpers
from jedi.api.completion import Completion
from jedi.evaluate import Evaluator
from jedi.evaluate import representation as er
from jedi.evaluate import imports
from jedi.evaluate.param import try_iter_content
from jedi.evaluate.helpers import get_module_names
from jedi.evaluate.sys_path import get_venv_path
from jedi.evaluate.iterable import unpack_tuple_to_dict
# Jedi uses lots and lots of recursion. By setting this a little bit higher, we
# can remove some "maximum recursion depth" errors.
sys.setrecursionlimit(2000)
class NotFoundError(Exception):
"""A custom error to avoid catching the wrong exceptions.
.. deprecated:: 0.9.0
Not in use anymore, Jedi just returns no goto result if you're not on a
valid name.
.. todo:: Remove!
"""
class Script(object):
"""
A Script is the base for completions, goto or whatever you want to do with
|jedi|.
You can either use the ``source`` parameter or ``path`` to read a file.
Usually you're going to want to use both of them (in an editor).
The script might be analyzed in a different ``sys.path`` than |jedi|:
- if `sys_path` parameter is not ``None``, it will be used as ``sys.path``
for the script;
- if `sys_path` parameter is ``None`` and ``VIRTUAL_ENV`` environment
variable is defined, ``sys.path`` for the specified environment will be
guessed (see :func:`jedi.evaluate.sys_path.get_venv_path`) and used for
the script;
- otherwise ``sys.path`` will match that of |jedi|.
:param source: The source code of the current file, separated by newlines.
:type source: str
:param line: The line to perform actions on (starting with 1).
:type line: int
:param column: The column of the cursor (starting with 0).
: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: The encoding of ``source``, if it is not a
``unicode`` object (default ``'utf-8'``).
:type encoding: str
:param source_encoding: The encoding of ``source``, if it is not a
``unicode`` object (default ``'utf-8'``).
:type encoding: str
:param sys_path: ``sys.path`` to use during analysis of the script
:type sys_path: list
"""
def __init__(self, source=None, line=None, column=None, path=None,
encoding='utf-8', source_path=None, source_encoding=None,
sys_path=None):
if source_path is not None:
warnings.warn("Use path instead of source_path.", DeprecationWarning)
path = source_path
if source_encoding is not None:
warnings.warn("Use encoding instead of source_encoding.", DeprecationWarning)
encoding = source_encoding
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 source is None:
# TODO add a better warning than the traceback!
with open(path, 'rb') as f:
source = f.read()
self._source = common.source_to_unicode(source, encoding)
self._code_lines = common.splitlines(self._source)
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_len = len(self._code_lines[line - 1])
column = line_len if column is None else column
if not (0 <= column <= line_len):
raise ValueError('`column` parameter is not in a valid range.')
self._pos = line, column
self._path = path
cache.clear_time_caches()
debug.reset_time()
self._grammar = load_grammar(version='%s.%s' % sys.version_info[:2])
if sys_path is None:
venv = os.getenv('VIRTUAL_ENV')
if venv:
sys_path = list(get_venv_path(venv))
self._evaluator = Evaluator(self._grammar, sys_path=sys_path)
debug.speed('init')
def _get_module(self):
cache.invalidate_star_import_cache(self._path)
parser = FastParser(self._grammar, self._source, self.path)
save_parser(self.path, parser, pickling=False)
module = self._evaluator.wrap(parser.module)
imports.add_module(self._evaluator, unicode(module.name), module)
return parser.module
@property
def source_path(self):
"""
.. deprecated:: 0.7.0
Use :attr:`.path` instead.
.. todo:: Remove!
"""
warnings.warn("Use path instead of source_path.", DeprecationWarning)
return self.path
def __repr__(self):
return '<%s: %s>' % (self.__class__.__name__, repr(self._orig_path))
def completions(self):
"""
Return :class:`classes.Completion` objects. Those objects contain
information about the completions, more than just names.
:return: Completion objects, sorted by name and __ comes last.
:rtype: list of :class:`classes.Completion`
"""
debug.speed('completions start')
completion = Completion(
self._evaluator, self._get_module(), self._code_lines,
self._pos, self.call_signatures
)
completions = completion.completions()
debug.speed('completions end')
return completions
def goto_definitions(self):
"""
Return the definitions of a the path under the cursor. goto function!
This follows complicated paths and returns the end, not the first
definition. The big difference between :meth:`goto_assignments` and
:meth:`goto_definitions` is that :meth:`goto_assignments` doesn't
follow imports and statements. Multiple objects may be returned,
because Python itself is a dynamic language, which means depending on
an option you can have two different versions of a function.
:rtype: list of :class:`classes.Definition`
"""
leaf = self._get_module().name_for_position(self._pos)
if leaf is None:
leaf = self._get_module().get_leaf_for_position(self._pos)
if leaf is None:
return []
definitions = helpers.evaluate_goto_definition(self._evaluator, leaf)
names = [s.name for s in definitions]
defs = [classes.Definition(self._evaluator, name) for name in names]
# 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):
"""
Return the first definition found, while optionally following imports.
Multiple objects may be returned, because Python itself is a
dynamic language, which means depending on an option you can have two
different versions of a function.
:rtype: list of :class:`classes.Definition`
"""
def filter_follow_imports(names):
for name in names:
definition = name.get_definition()
if definition.type in ('import_name', 'import_from'):
imp = imports.ImportWrapper(self._evaluator, name)
for name in filter_follow_imports(imp.follow(is_goto=True)):
yield name
else:
yield name
names = self._goto()
if follow_imports:
names = filter_follow_imports(names)
defs = [classes.Definition(self._evaluator, d) for d in set(names)]
return helpers.sorted_definitions(defs)
def _goto(self):
"""
Used for goto_assignments and usages.
"""
name = self._get_module().name_for_position(self._pos)
if name is None:
return []
return list(self._evaluator.goto(name))
def usages(self, additional_module_paths=()):
"""
Return :class:`classes.Definition` objects, which contain all
names that point to the definition of the name under the cursor. This
is very useful for refactoring (renaming), or to show all usages of a
variable.
.. todo:: Implement additional_module_paths
:rtype: list of :class:`classes.Definition`
"""
temp, settings.dynamic_flow_information = \
settings.dynamic_flow_information, False
try:
user_stmt = self._get_module().get_statement_for_position(self._pos)
definitions = self._goto()
if not definitions and isinstance(user_stmt, tree.Import):
# For not defined imports (goto doesn't find something, we take
# the name as a definition. This is enough, because every name
# points to it.
name = user_stmt.name_for_position(self._pos)
if name is None:
# Must be syntax
return []
definitions = [name]
if not definitions:
# Without a definition for a name we cannot find references.
return []
if not isinstance(user_stmt, tree.Import):
# import case is looked at with add_import_name option
definitions = usages.usages_add_import_modules(self._evaluator,
definitions)
module = set([d.get_parent_until() for d in definitions])
module.add(self._get_module())
names = usages.usages(self._evaluator, definitions, module)
for d in set(definitions):
names.append(classes.Definition(self._evaluator, d))
finally:
settings.dynamic_flow_information = temp
return helpers.sorted_definitions(set(names))
def call_signatures(self):
"""
Return the function object of the call you're currently in.
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:`classes.CallSignature`
"""
call_signature_details = \
helpers.get_call_signature_details(self._get_module(), self._pos)
if call_signature_details is None:
return []
with common.scale_speed_settings(settings.scale_call_signatures):
definitions = helpers.cache_call_signatures(
self._evaluator,
call_signature_details.bracket_leaf,
self._code_lines,
self._pos
)
debug.speed('func_call followed')
return [classes.CallSignature(self._evaluator, d.name,
call_signature_details.bracket_leaf.start_pos,
call_signature_details.call_index,
call_signature_details.keyword_name_str)
for d in definitions if hasattr(d, 'py__call__')]
def _analysis(self):
self._evaluator.is_analysis = True
self._evaluator.analysis_modules = [self._get_module()]
try:
for node in self._get_module().nodes_to_execute():
if node.type in ('funcdef', 'classdef'):
if node.type == 'classdef':
continue
raise NotImplementedError
er.Function(self._evaluator, node).get_decorated_func()
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.paths())
for n in import_names:
imports.ImportWrapper(self._evaluator, n).follow()
elif node.type == 'expr_stmt':
types = self._evaluator.eval_element(node)
for testlist in node.children[:-1:2]:
# Iterate tuples.
unpack_tuple_to_dict(self._evaluator, types, testlist)
else:
try_iter_content(self._evaluator.goto_definitions(node))
self._evaluator.reset_recursion_limitations()
ana = [a for a in self._evaluator.analysis if self.path == a.path]
return sorted(set(ana), key=lambda x: x.line)
finally:
self._evaluator.is_analysis = False
class Interpreter(Script):
"""
Jedi API for Python REPLs.
In addition to completion of simple attribute access, Jedi
supports code completion based on static code analysis.
Jedi can complete attributes of object which is not initialized
yet.
>>> from os.path import join
>>> namespace = locals()
>>> script = Interpreter('join("").up', [namespace])
>>> print(script.completions()[0].name)
upper
"""
def __init__(self, source, namespaces, **kwds):
"""
Parse `source` and mixin interpreted Python objects from `namespaces`.
:type source: str
:arg source: Code to parse.
:type namespaces: list of dict
:arg namespaces: a list of namespace dictionaries such as the one
returned by :func:`locals`.
Other optional arguments are same as the ones for :class:`Script`.
If `line` and `column` are None, they are assumed be at the end of
`source`.
"""
try:
namespaces = [dict(n) for n in namespaces]
except Exception:
raise TypeError("namespaces must be a non-empty list of dicts.")
super(Interpreter, self).__init__(source, **kwds)
self.namespaces = namespaces
parser_module = super(Interpreter, self)._get_module()
self._module = interpreter.MixedModule(self._evaluator, parser_module, self.namespaces)
def _get_module(self):
return self._module
def defined_names(source, path=None, encoding='utf-8'):
"""
Get all definitions in `source` sorted by its position.
This functions can be used for listing functions, classes and
data defined in a file. This can be useful if you want to list
them in "sidebar". Each element in the returned list also has
`defined_names` method which can be used to get sub-definitions
(e.g., methods in class).
:rtype: list of classes.Definition
.. deprecated:: 0.9.0
Use :func:`names` instead.
.. todo:: Remove!
"""
warnings.warn("Use call_signatures instead.", DeprecationWarning)
return names(source, path, encoding)
def names(source=None, path=None, encoding='utf-8', all_scopes=False,
definitions=True, references=False):
"""
Returns a list of `Definition` objects, containing name parts.
This means you can call ``Definition.goto_assignments()`` and get the
reference of a name.
The parameters are the same as in :py:class:`Script`, except or the
following ones:
: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``.
"""
def def_ref_filter(_def):
is_def = _def.is_definition()
return definitions and is_def or references and not is_def
# Set line/column to a random position, because they don't matter.
script = Script(source, line=1, column=0, path=path, encoding=encoding)
defs = [classes.Definition(script._evaluator, name_part)
for name_part in get_module_names(script._get_module(), all_scopes)]
return sorted(filter(def_ref_filter, defs), key=lambda x: (x.line, x.column))
def preload_module(*modules):
"""
Preloading modules tells Jedi to load a module now, instead of lazy parsing
of modules. Usful 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, 1, len(s), None).completions()
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, with n params.
"""
debug.debug_function = func_cb
debug.enable_warning = warnings
debug.enable_notice = notices
debug.enable_speed = speed

View file

@ -0,0 +1,764 @@
"""
The :mod:`jedi.api.classes` module contains the return classes of the API.
These classes are the much bigger part of the whole API, because they contain
the interesting information about completion and goto operations.
"""
import warnings
from itertools import chain
import re
from jedi._compatibility import unicode, use_metaclass
from jedi import settings
from jedi import common
from jedi.parser import tree
from jedi.parser.utils import load_parser
from jedi.evaluate.cache import memoize_default, CachedMetaClass
from jedi.evaluate import representation as er
from jedi.evaluate import iterable
from jedi.evaluate import imports
from jedi.evaluate import compiled
from jedi.evaluate.compiled import mixed
from jedi.api import keywords
from jedi.evaluate.finder import filter_definition_names
def defined_names(evaluator, scope):
"""
List sub-definitions (e.g., methods in class).
:type scope: Scope
:rtype: list of Definition
"""
dct = scope.names_dict
names = list(chain.from_iterable(dct.values()))
names = filter_definition_names(names, scope)
return [Definition(evaluator, d) for d in sorted(names, key=lambda s: s.start_pos)]
class BaseDefinition(object):
_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',
'_sqlite3': 'sqlite3',
'__builtin__': '',
'builtins': '',
}
_tuple_mapping = dict((tuple(k.split('.')), v) for (k, v) in {
'argparse._ActionsContainer': 'argparse.ArgumentParser',
'_sre.SRE_Match': 're.MatchObject',
'_sre.SRE_Pattern': 're.RegexObject',
}.items())
def __init__(self, evaluator, name):
self._evaluator = evaluator
self._name = name
"""
An instance of :class:`jedi.parser.reprsentation.Name` subclass.
"""
self._definition = evaluator.wrap(self._name.get_definition())
self.is_keyword = isinstance(self._definition, keywords.Keyword)
# generate a path to the definition
self._module = name.get_parent_until()
if self.in_builtin_module():
self.module_path = None
else:
self.module_path = self._module.path
"""Shows the file path of a module. e.g. ``/usr/lib/python2.7/os.py``"""
@property
def name(self):
"""
Name of variable/function/class/module.
For example, for ``x = None`` it returns ``'x'``.
:rtype: str or None
"""
return unicode(self._name)
@property
def start_pos(self):
"""
.. deprecated:: 0.7.0
Use :attr:`.line` and :attr:`.column` instead.
.. todo:: Remove!
"""
warnings.warn("Use line/column instead.", DeprecationWarning)
return self._name.start_pos
@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.goto_definitions` should return a list of
definition for ``sys``, ``f``, ``C`` and ``x``.
>>> 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.goto_definitions()
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)
>>> defs # doctest: +NORMALIZE_WHITESPACE
[<Definition module keyword>, <Definition class C>,
<Definition class D>, <Definition def f>]
Finally, here is what you can get from :attr:`type`:
>>> defs[0].type
'module'
>>> defs[1].type
'class'
>>> defs[2].type
'instance'
>>> defs[3].type
'function'
"""
stripped = self._definition
if isinstance(stripped, er.InstanceElement):
stripped = stripped.var
if isinstance(stripped, (compiled.CompiledObject, mixed.MixedObject)):
return stripped.api_type()
elif isinstance(stripped, iterable.Array):
return 'instance'
elif isinstance(stripped, tree.Import):
return 'import'
string = type(stripped).__name__.lower().replace('wrapper', '')
if string == 'exprstmt':
return 'statement'
else:
return string
def _path(self):
"""The path to a module/class/function definition."""
path = []
par = self._definition
while par is not None:
if isinstance(par, tree.Import):
path += imports.ImportWrapper(self._evaluator, self._name).import_path
break
try:
name = par.name
except AttributeError:
pass
else:
if isinstance(par, er.ModuleWrapper):
# TODO just make the path dotted from the beginning, we
# shouldn't really split here.
path[0:0] = par.py__name__().split('.')
break
else:
path.insert(0, unicode(name))
par = par.parent
return path
@property
def module_name(self):
"""
The module name.
>>> from jedi import Script
>>> source = 'import json'
>>> script = Script(source, path='example.py')
>>> d = script.goto_definitions()[0]
>>> print(d.module_name) # doctest: +ELLIPSIS
json
"""
return str(self._module.name)
def in_builtin_module(self):
"""Whether this is a builtin module."""
return isinstance(self._module, compiled.CompiledObject)
@property
def line(self):
"""The line where the definition occurs (starting with 1)."""
if self.in_builtin_module():
return None
return self._name.start_pos[0]
@property
def column(self):
"""The column where the definition occurs (starting with 0)."""
if self.in_builtin_module():
return None
return self._name.start_pos[1]
def docstring(self, raw=False):
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, 1, len('def f'), 'example.py')
>>> doc = script.goto_definitions()[0].docstring()
>>> print(doc)
f(a, b=1)
<BLANKLINE>
Document for function f.
Notice that useful extra information is added to the actual
docstring. For function, it is call signature. If you need
actual docstring, use ``raw=True`` instead.
>>> print(script.goto_definitions()[0].docstring(raw=True))
Document for function f.
"""
if raw:
return _Help(self._definition).raw()
else:
return _Help(self._definition).full()
@property
def doc(self):
"""
.. deprecated:: 0.8.0
Use :meth:`.docstring` instead.
.. todo:: Remove!
"""
warnings.warn("Use docstring() instead.", DeprecationWarning)
return self.docstring()
@property
def raw_doc(self):
"""
.. deprecated:: 0.8.0
Use :meth:`.docstring` instead.
.. todo:: Remove!
"""
warnings.warn("Use docstring() instead.", DeprecationWarning)
return self.docstring(raw=True)
@property
def description(self):
"""A textual description of the object."""
return unicode(self._name)
@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, 3, len('os.path.join'), 'example.py')
>>> print(script.goto_definitions()[0].full_name)
os.path.join
Notice that it correctly returns ``'os.path.join'`` instead of
(for example) ``'posixpath.join'``.
"""
path = [unicode(p) for p in self._path()]
# TODO add further checks, the mapping should only occur on stdlib.
if not path:
return None # for keywords the path is empty
with common.ignored(KeyError):
path[0] = self._mapping[path[0]]
for key, repl in self._tuple_mapping.items():
if tuple(path[:len(key)]) == key:
path = [repl] + path[len(key):]
return '.'.join(path if path[0] else path[1:])
def goto_assignments(self):
defs = self._evaluator.goto(self._name)
return [Definition(self._evaluator, d) for d in defs]
@memoize_default()
def _follow_statements_imports(self):
"""
Follow both statements and imports, as far as possible.
"""
if self._definition.isinstance(tree.ExprStmt):
return self._evaluator.eval_statement(self._definition)
elif self._definition.isinstance(tree.Import):
return imports.ImportWrapper(self._evaluator, self._name).follow()
else:
return set([self._definition])
@property
@memoize_default()
def params(self):
"""
Raises an ``AttributeError``if the definition is not callable.
Otherwise returns a list of `Definition` that represents the params.
"""
followed = list(self._follow_statements_imports())
if not followed or not hasattr(followed[0], 'py__call__'):
raise AttributeError()
followed = followed[0] # only check the first one.
if followed.type in ('funcdef', 'lambda'):
if isinstance(followed, er.InstanceElement):
params = followed.params[1:]
else:
params = followed.params
elif followed.isinstance(er.compiled.CompiledObject):
params = followed.params
elif isinstance(followed, er.Class):
try:
sub = followed.get_subscope_by_name('__init__')
params = sub.params[1:] # ignore self
except KeyError:
return []
elif isinstance(followed, er.Instance):
try:
sub = followed.get_subscope_by_name('__call__')
params = sub.params[1:] # ignore self
except KeyError:
return []
else:
return []
return [_Param(self._evaluator, p.name) for p in params]
def parent(self):
scope = self._definition.get_parent_scope()
scope = self._evaluator.wrap(scope)
return Definition(self._evaluator, scope.name)
def __repr__(self):
return "<%s %s>" % (type(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 self.in_builtin_module():
return ''
path = self._definition.get_parent_until().path
parser = load_parser(path)
lines = common.splitlines(parser.source)
line_nr = self._name.start_pos[0]
start_line_nr = line_nr - before
return '\n'.join(lines[start_line_nr:line_nr + after + 1])
class Completion(BaseDefinition):
"""
`Completion` objects are returned from :meth:`api.Script.completions`. They
provide additional information about a completion.
"""
def __init__(self, evaluator, name, stack, like_name_length):
super(Completion, self).__init__(evaluator, name)
self._like_name_length = like_name_length
self._stack = stack
# 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 = '('
if isinstance(self._definition, tree.Param) and self._stack is not None:
node_names = list(self._stack.get_node_names(self._evaluator.grammar))
if 'trailer' in node_names and 'argument' not in node_names:
append += '='
name = str(self._name)
if like_name:
name = name[self._like_name_length:]
return name + append
@property
def complete(self):
"""
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`.
"""
return self._complete(True)
@property
def name_with_symbols(self):
"""
Similar to :attr:`name`, but like :attr:`name`
returns also the symbols, for example::
list()
would return ``.append`` and others (which means it adds a dot).
"""
return self._complete(False)
@property
def description(self):
"""Provide a description of the completion object."""
if self._definition is None:
return ''
t = self.type
if t == 'statement' or t == 'import':
desc = self._definition.get_code()
else:
desc = '.'.join(unicode(p) for p in self._path())
line = '' if self.in_builtin_module else '@%s' % self.line
return '%s: %s%s' % (t, desc, line)
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, self._name)
def docstring(self, raw=False, fast=True):
"""
: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``.
"""
definition = self._definition
if isinstance(definition, tree.Import):
i = imports.ImportWrapper(self._evaluator, self._name)
if len(i.import_path) > 1 or not fast:
followed = self._follow_statements_imports()
if followed:
# TODO: Use all of the followed objects as input to Documentation.
definition = list(followed)[0]
if raw:
return _Help(definition).raw()
else:
return _Help(definition).full()
@property
def type(self):
"""
The type of the completion objects. Follows imports. For a further
description, look at :attr:`jedi.api.classes.BaseDefinition.type`.
"""
if isinstance(self._definition, tree.Import):
i = imports.ImportWrapper(self._evaluator, self._name)
if len(i.import_path) <= 1:
return 'module'
followed = self.follow_definition()
if followed:
# Caveat: Only follows the first one, ignore the other ones.
# This is ok, since people are almost never interested in
# variations.
return followed[0].type
return super(Completion, self).type
@memoize_default()
def _follow_statements_imports(self):
# imports completion is very complicated and needs to be treated
# separately in Completion.
definition = self._definition
if definition.isinstance(tree.Import):
i = imports.ImportWrapper(self._evaluator, self._name)
return i.follow()
return super(Completion, self)._follow_statements_imports()
@memoize_default()
def follow_definition(self):
"""
Return the original 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 PITA-slow.
"""
defs = self._follow_statements_imports()
return [Definition(self._evaluator, d.name) for d in defs]
class Definition(use_metaclass(CachedMetaClass, BaseDefinition)):
"""
*Definition* objects are returned from :meth:`api.Script.goto_assignments`
or :meth:`api.Script.goto_definitions`.
"""
def __init__(self, evaluator, definition):
super(Definition, self).__init__(evaluator, definition)
@property
def description(self):
"""
A description of the :class:`.Definition` object, which is heavily used
in testing. e.g. for ``isinstance`` it returns ``def isinstance``.
Example:
>>> from jedi import Script
>>> source = '''
... def f():
... pass
...
... class C:
... pass
...
... variable = f if random.choice([0,1]) else C'''
>>> script = Script(source, column=3) # line is maximum by default
>>> defs = script.goto_definitions()
>>> defs = sorted(defs, key=lambda d: d.line)
>>> defs
[<Definition def f>, <Definition class C>]
>>> str(defs[0].description) # strip literals in python2
'def f'
>>> str(defs[1].description)
'class C'
"""
d = self._definition
if isinstance(d, er.InstanceElement):
d = d.var
if isinstance(d, compiled.CompiledObject):
typ = d.api_type()
if typ == 'instance':
typ = 'class' # The description should be similar to Py objects.
d = typ + ' ' + d.name.get_code()
elif isinstance(d, iterable.Array):
d = 'class ' + d.type
elif isinstance(d, (tree.Class, er.Class, er.Instance)):
d = 'class ' + unicode(d.name)
elif isinstance(d, (er.Function, tree.Function)):
d = 'def ' + unicode(d.name)
elif isinstance(d, tree.Module):
# only show module name
d = 'module %s' % self.module_name
elif isinstance(d, tree.Param):
d = d.get_code().strip()
if d.endswith(','):
d = d[:-1] # Remove the comma.
else: # ExprStmt
try:
first_leaf = d.first_leaf()
except AttributeError:
# `d` is already a Leaf (Name).
first_leaf = d
# Remove the prefix, because that's not what we want for get_code
# here.
old, first_leaf.prefix = first_leaf.prefix, ''
try:
d = d.get_code()
finally:
first_leaf.prefix = old
# Delete comments:
d = re.sub('#[^\n]+\n', ' ', d)
# Delete multi spaces/newlines
return re.sub('\s+', ' ', d).strip()
@property
def desc_with_module(self):
"""
In addition to the definition, also return the module.
.. warning:: Don't use this function yet, its behaviour may change. If
you really need it, talk to me.
.. todo:: Add full path. This function is should return a
`module.class.function` path.
"""
position = '' if self.in_builtin_module else '@%s' % (self.line)
return "%s:%s%s" % (self.module_name, self.description, position)
@memoize_default()
def defined_names(self):
"""
List sub-definitions (e.g., methods in class).
:rtype: list of Definition
"""
defs = self._follow_statements_imports()
# For now we don't want base classes or evaluate decorators.
defs = [d.base if isinstance(d, (er.Class, er.Function)) else d for d in defs]
iterable = (defined_names(self._evaluator, d) for d in defs)
iterable = list(iterable)
return list(chain.from_iterable(iterable))
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.
"""
return self._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._evaluator == other._evaluator
def __ne__(self, other):
return not self.__eq__(other)
def __hash__(self):
return hash((self._name.start_pos, self.module_path, self.name, self._evaluator))
class CallSignature(Definition):
"""
`CallSignature` objects is the return value of `Script.function_definition`.
It knows what functions you are currently in. e.g. `isinstance(` would
return the `isinstance` function. without `(` it would return nothing.
"""
def __init__(self, evaluator, executable_name, bracket_start_pos, index, key_name_str):
super(CallSignature, self).__init__(evaluator, executable_name)
self._index = index
self._key_name_str = key_name_str
self._bracket_start_pos = bracket_start_pos
@property
def index(self):
"""
The Param index of the current call.
Returns None if the index cannot be found in the curent call.
"""
if self._key_name_str is not None:
for i, param in enumerate(self.params):
if self._key_name_str == param.name:
return i
if self.params and self.params[-1]._name.get_definition().stars == 2:
return i
else:
return None
if self._index >= len(self.params):
for i, param in enumerate(self.params):
# *args case
if param._name.get_definition().stars == 1:
return i
return None
return self._index
@property
def bracket_start(self):
"""
The indent of the bracket that is responsible for the last function
call.
"""
return self._bracket_start_pos
@property
def call_name(self):
"""
.. deprecated:: 0.8.0
Use :attr:`.name` instead.
.. todo:: Remove!
The name (e.g. 'isinstance') as a string.
"""
warnings.warn("Use name instead.", DeprecationWarning)
return unicode(self.name)
@property
def module(self):
"""
.. deprecated:: 0.8.0
Use :attr:`.module_name` for the module name.
.. todo:: Remove!
"""
return self._executable.get_parent_until()
def __repr__(self):
return '<%s: %s index %s>' % (type(self).__name__, self._name,
self.index)
class _Param(Definition):
"""
Just here for backwards compatibility.
"""
def get_code(self):
"""
.. deprecated:: 0.8.0
Use :attr:`.description` and :attr:`.name` instead.
.. todo:: Remove!
A function to get the whole code of the param.
"""
warnings.warn("Use description instead.", DeprecationWarning)
return self.description
class _Help(object):
"""
Temporary implementation, will be used as `Script.help() or something in
the future.
"""
def __init__(self, definition):
self._name = definition
def full(self):
try:
return self._name.doc
except AttributeError:
return self.raw()
def raw(self):
"""
The raw docstring ``__doc__`` for any object.
See :attr:`doc` for example.
"""
try:
return self._name.raw_doc
except AttributeError:
return ''

View file

@ -0,0 +1,256 @@
from itertools import chain
from jedi.parser import token
from jedi.parser import tree
from jedi import debug
from jedi import settings
from jedi.api import classes
from jedi.api import helpers
from jedi.evaluate import imports
from jedi.api import keywords
from jedi.evaluate import compiled
from jedi.evaluate.helpers import call_of_leaf
from jedi.evaluate.finder import global_names_dict_generator, filter_definition_names
def get_call_signature_param_names(call_signatures):
# add named params
for call_sig in call_signatures:
# Allow protected access, because it's a public API.
module = call_sig._name.get_parent_until()
# Compiled modules typically don't allow keyword arguments.
if not isinstance(module, compiled.CompiledObject):
for p in call_sig.params:
# Allow access on _definition here, because it's a
# public API and we don't want to make the internal
# Name object public.
if p._definition.stars == 0: # no *args/**kwargs
yield p._name
def filter_names(evaluator, completion_names, stack, like_name):
comp_dct = {}
for name in set(completion_names):
if settings.case_insensitive_completion \
and str(name).lower().startswith(like_name.lower()) \
or str(name).startswith(like_name):
if isinstance(name.parent, (tree.Function, tree.Class)):
# TODO I think this is a hack. It should be an
# er.Function/er.Class before that.
name = evaluator.wrap(name.parent).name
new = classes.Completion(
evaluator,
name,
stack,
len(like_name)
)
k = (new.name, new.complete) # key
if k in comp_dct and settings.no_completion_duplicates:
comp_dct[k]._same_name_completions.append(new)
else:
comp_dct[k] = new
yield new
def get_user_scope(module, position):
"""
Returns the scope in which the user resides. This includes flows.
"""
user_stmt = module.get_statement_for_position(position)
if user_stmt is None:
def scan(scope):
for s in scope.children:
if s.start_pos <= position <= s.end_pos:
if isinstance(s, (tree.Scope, tree.Flow)):
return scan(s) or s
elif s.type in ('suite', 'decorated'):
return scan(s)
return None
return scan(module) or module
else:
return user_stmt.get_parent_scope(include_flows=True)
class Completion:
def __init__(self, evaluator, module, code_lines, position, call_signatures_method):
self._evaluator = evaluator
self._module = evaluator.wrap(module)
self._code_lines = code_lines
# The first step of completions is to get the name
self._like_name = helpers.get_on_completion_name(module, 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._position = position[0], position[1] - len(self._like_name)
self._call_signatures_method = call_signatures_method
def completions(self):
completion_names = self._get_context_completions()
completions = filter_names(self._evaluator, completion_names,
self.stack, self._like_name)
return sorted(completions, key=lambda x: (x.name.startswith('__'),
x.name.startswith('_'),
x.name.lower()))
def _get_context_completions(self):
"""
Analyzes the context that a completion is made in 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._evaluator.grammar
try:
self.stack = helpers.get_stack_at_position(
grammar, self._code_lines, self._module, self._position
)
except helpers.OnErrorLeaf as e:
self.stack = None
if e.error_leaf.value == '.':
# After ErrorLeaf's that are dots, we will not do any
# completions since this probably just confuses the user.
return []
# If we don't have a context, just use global completion.
return self._global_completions()
allowed_keywords, allowed_tokens = \
helpers.get_possible_completion_types(grammar, self.stack)
completion_names = list(self._get_keyword_completion_names(allowed_keywords))
if token.NAME in allowed_tokens:
# This means that we actually have to do type inference.
symbol_names = list(self.stack.get_node_names(grammar))
nodes = list(self.stack.get_nodes())
if "import_stmt" in symbol_names:
level = 0
only_modules = True
level, names = self._parse_dotted_names(nodes)
if "import_from" in symbol_names:
if 'import' in nodes:
only_modules = False
else:
assert "import_name" in symbol_names
completion_names += self._get_importer_names(
names,
level,
only_modules
)
elif 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 list(self._get_class_context_completions(is_function=True))
elif symbol_names[-1] == 'trailer' and nodes[-1] == '.':
dot = self._module.get_leaf_for_position(self._position)
atom_expr = call_of_leaf(dot.get_previous_leaf())
completion_names += self._trailer_completions(atom_expr)
else:
completion_names += self._global_completions()
completion_names += self._get_class_context_completions(is_function=False)
if 'trailer' in symbol_names:
call_signatures = self._call_signatures_method()
completion_names += get_call_signature_param_names(call_signatures)
return completion_names
def _get_keyword_completion_names(self, keywords_):
for k in keywords_:
yield keywords.keyword(self._evaluator, k).name
def _global_completions(self):
scope = get_user_scope(self._module, self._position)
if not scope.is_scope(): # Might be a flow (if/while/etc).
scope = scope.get_parent_scope()
scope = self._evaluator.wrap(scope)
debug.dbg('global completion scope: %s', scope)
names_dicts = global_names_dict_generator(
self._evaluator,
scope,
self._position
)
completion_names = []
for names_dict, pos in names_dicts:
names = list(chain.from_iterable(names_dict.values()))
if not names:
continue
completion_names += filter_definition_names(
names, self._module.get_statement_for_position(self._position), pos
)
return completion_names
def _trailer_completions(self, atom_expr):
scopes = self._evaluator.eval_element(atom_expr)
completion_names = []
debug.dbg('trailer completion scopes: %s', scopes)
for s in scopes:
names = []
for names_dict in s.names_dicts(search_global=False):
names += chain.from_iterable(names_dict.values())
completion_names += filter_definition_names(
names, self._module.get_statement_for_position(self._position)
)
return completion_names
def _parse_dotted_names(self, nodes):
level = 0
names = []
for node in nodes[1:]:
if node in ('.', '...'):
if not names:
level += len(node.value)
elif node.type == 'dotted_name':
names += node.children[::2]
elif node.type == 'name':
names.append(node)
else:
break
return level, names
def _get_importer_names(self, names, level=0, only_modules=True):
names = [str(n) for n in names]
i = imports.Importer(self._evaluator, names, self._module, level)
return i.completion_names(self._evaluator, only_modules=only_modules)
def _get_class_context_completions(self, is_function=True):
"""
Autocomplete inherited methods when overriding in child class.
"""
leaf = self._module.get_leaf_for_position(self._position, include_prefixes=True)
cls = leaf.get_parent_until(tree.Class)
if isinstance(cls, (tree.Class, tree.Function)):
# Complete the methods that are defined in the super classes.
cls = self._evaluator.wrap(cls)
else:
return
if cls.start_pos[1] >= leaf.start_pos[1]:
return
names_dicts = cls.names_dicts(search_global=False, is_instance=True)
# The first dict is the dictionary of class itself.
next(names_dicts)
for names_dict in names_dicts:
for values in names_dict.values():
for value in values:
if (value.parent.type == 'funcdef') == is_function:
yield value

View file

@ -0,0 +1,315 @@
"""
Helpers for the API
"""
import re
from collections import namedtuple
from jedi._compatibility import u
from jedi.evaluate.helpers import call_of_leaf
from jedi import parser
from jedi.parser import tokenize
from jedi.cache import time_cache
from jedi import common
CompletionParts = namedtuple('CompletionParts', ['path', 'has_dot', '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))
def get_on_completion_name(module, lines, position):
leaf = module.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 '\n'.join(lines)
class OnErrorLeaf(Exception):
@property
def error_leaf(self):
return self.args[0]
def _is_on_comment(leaf, position):
# We might be on a comment.
if leaf.type == 'endmarker':
try:
dedent = leaf.get_previous_leaf()
if dedent.type == 'dedent' and dedent.prefix:
# TODO This is needed because the fast parser uses multiple
# endmarker tokens within a file which is obviously ugly.
# This is so ugly that I'm not even commenting how it exactly
# happens, but let me tell you that I want to get rid of it.
leaf = dedent
except IndexError:
pass
comment_lines = common.splitlines(leaf.prefix)
difference = leaf.start_pos[0] - position[0]
prefix_start_pos = leaf.get_start_pos_of_prefix()
if difference == 0:
indent = leaf.start_pos[1]
elif position[0] == prefix_start_pos[0]:
indent = prefix_start_pos[1]
else:
indent = 0
line = comment_lines[-difference - 1][:position[1] - indent]
return '#' in line
def _get_code_for_stack(code_lines, module, position):
leaf = module.get_leaf_for_position(position, include_prefixes=True)
# 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 _is_on_comment(leaf, position):
return u('')
# If we're not on a comment simply get the previous leaf and proceed.
try:
leaf = leaf.get_previous_leaf()
except IndexError:
return u('') # At the beginning of the file.
is_after_newline = leaf.type == 'newline'
while leaf.type == 'newline':
try:
leaf = leaf.get_previous_leaf()
except IndexError:
return u('')
if leaf.type in ('indent', 'dedent'):
return u('')
elif leaf.type == 'error_leaf' or leaf.type == 'string':
# Error leafs cannot be parsed, completion in strings is also
# impossible.
raise OnErrorLeaf(leaf)
else:
if leaf == ';':
user_stmt = leaf.parent
else:
user_stmt = leaf.get_definition()
if user_stmt.parent.type == 'simple_stmt':
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 context (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, module, pos):
"""
Returns the possible node names (e.g. import_from, xor_test or yield_stmt).
"""
class EndMarkerReached(Exception):
pass
def tokenize_without_endmarker(code):
tokens = tokenize.source_tokens(code, use_exact_op_types=True)
for token_ in tokens:
if token_.string == safeword:
raise EndMarkerReached()
else:
yield token_
code = _get_code_for_stack(code_lines, module, 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'
# Remove as many indents from **all** code lines as possible.
code = code + safeword
p = parser.ParserWithRecovery(grammar, code, start_parsing=False)
try:
p.parse(tokenizer=tokenize_without_endmarker(code))
except EndMarkerReached:
return Stack(p.stack)
raise SystemError("This really shouldn't happen. There's a bug in Jedi.")
class Stack(list):
def get_node_names(self, grammar):
for dfa, state, (node_number, nodes) in self:
yield grammar.number2symbol[node_number]
def get_nodes(self):
for dfa, state, (node_number, nodes) in self:
for node in nodes:
yield node
def get_possible_completion_types(grammar, stack):
def add_results(label_index):
try:
grammar_labels.append(inversed_tokens[label_index])
except KeyError:
try:
keywords.append(inversed_keywords[label_index])
except KeyError:
t, v = grammar.labels[label_index]
assert t >= 256
# See if it's a symbol and if we're in its first set
inversed_keywords
itsdfa = grammar.dfas[t]
itsstates, itsfirst = itsdfa
for first_label_index in itsfirst.keys():
add_results(first_label_index)
inversed_keywords = dict((v, k) for k, v in grammar.keywords.items())
inversed_tokens = dict((v, k) for k, v in grammar.tokens.items())
keywords = []
grammar_labels = []
def scan_stack(index):
dfa, state, node = stack[index]
states, first = dfa
arcs = states[state]
for label_index, new_state in arcs:
if label_index == 0:
# An accepting state, check the stack below.
scan_stack(index - 1)
else:
add_results(label_index)
scan_stack(-1)
return keywords, grammar_labels
def evaluate_goto_definition(evaluator, leaf):
if leaf.type == 'name':
# In case of a name we can just use goto_definition which does all the
# magic itself.
return evaluator.goto_definitions(leaf)
node = None
parent = leaf.parent
if parent.type == 'atom':
node = leaf.parent
elif parent.type == 'trailer':
node = call_of_leaf(leaf)
if node is None:
return []
return evaluator.eval_element(node)
CallSignatureDetails = namedtuple(
'CallSignatureDetails',
['bracket_leaf', 'call_index', 'keyword_name_str']
)
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':
nodes_before = [c for c in nodes_before[-1].children if c.start_pos < position]
key_str = None
if nodes_before:
last = nodes_before[-1]
if last.type == 'argument' 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_call_signature_details_from_error_node(node, 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.type == 'name' or name.parent.type in ('trailer', 'atom'):
return CallSignatureDetails(
element,
*_get_index_and_key(children, position)
)
def get_call_signature_details(module, position):
leaf = module.get_leaf_for_position(position, include_prefixes=True)
if leaf == ')':
if leaf.end_pos == position:
leaf = leaf.get_next_leaf()
# 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'):
# Don't show call signatures if there's stuff before it that just
# makes it feel strange to have a call signature.
return None
for n in node.children[::-1]:
if n.start_pos < position and n.type == 'error_node':
result = _get_call_signature_details_from_error_node(n, position)
if result is not None:
return result
if node.type == 'trailer' and node.children[0] == '(':
leaf = node.get_previous_leaf()
return CallSignatureDetails(
node.children[0], *_get_index_and_key(node.children, position))
node = node.parent
return None
@time_cache("call_signatures_validity")
def cache_call_signatures(evaluator, bracket_leaf, code_lines, user_pos):
"""This function calculates the cache key."""
index = user_pos[0] - 1
before_cursor = code_lines[index][:user_pos[1]]
other_lines = code_lines[bracket_leaf.start_pos[0]:index]
whole = '\n'.join(other_lines + [before_cursor])
before_bracket = re.match(r'.*\(', whole, re.DOTALL)
module_path = bracket_leaf.get_parent_until().path
if module_path is None:
yield None # Don't cache!
else:
yield (module_path, before_bracket, bracket_leaf.start_pos)
yield evaluate_goto_definition(
evaluator,
bracket_leaf.get_previous_leaf()
)

View file

@ -0,0 +1,67 @@
"""
TODO Some parts of this module are still not well documented.
"""
import copy
from jedi.cache import underscore_memoization
from jedi.evaluate import helpers
from jedi.evaluate.representation import ModuleWrapper
from jedi.evaluate.compiled import mixed
class MixedModule(object):
resets_positions = True
type = 'mixed_module'
def __init__(self, evaluator, parser_module, namespaces):
self._evaluator = evaluator
self._namespaces = namespaces
self._namespace_objects = [type('jedi_namespace', (), n) for n in namespaces]
self._wrapped_module = ModuleWrapper(evaluator, parser_module)
# Usually we are dealing with very small code sizes when it comes to
# interpreter modules. In this case we just copy the whole syntax tree
# to be able to modify it.
self._parser_module = copy.deepcopy(parser_module)
for child in self._parser_module.children:
child.parent = self
def names_dicts(self, search_global):
for names_dict in self._wrapped_module.names_dicts(search_global):
yield names_dict
for namespace_obj in self._namespace_objects:
m = mixed.MixedObject(self._evaluator, namespace_obj, self._parser_module.name)
for names_dict in m.names_dicts(False):
yield names_dict
def __getattr__(self, name):
return getattr(self._parser_module, name)
class LazyName(helpers.FakeName):
def __init__(self, evaluator, module, name, value):
super(LazyName, self).__init__(name)
self._module = module
self._evaluator = evaluator
self._value = value
self._name = name
def is_definition(self):
return True
@property
@underscore_memoization
def parent(self):
"""
Creating fake statements for the interpreter.
Here we are trying to link back to Python code, if possible. This means
we try to find the python module for a name (not the builtin).
"""
return mixed.create(self._evaluator, self._value)
@parent.setter
def parent(self, value):
"""Needed because the super class tries to set parent."""

View file

@ -0,0 +1,132 @@
import pydoc
import keyword
from jedi._compatibility import is_py3, is_py35
from jedi import common
from jedi.evaluate.helpers import FakeName
from jedi.parser.tree import Leaf
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
if is_py3:
if is_py35:
# in python 3.5 async and await are not proper keywords, but for
# completion pursposes should as as though they are
keys = keyword.kwlist + ["async", "await"]
else:
keys = keyword.kwlist
else:
keys = keyword.kwlist + ['None', 'False', 'True']
def has_inappropriate_leaf_keyword(pos, module):
relevant_errors = filter(
lambda error: error.first_pos[0] == pos[0],
module.error_statement_stacks)
for error in relevant_errors:
if error.next_token in keys:
return True
return False
def completion_names(evaluator, stmt, pos, module):
keyword_list = all_keywords(evaluator)
if not isinstance(stmt, Leaf) or has_inappropriate_leaf_keyword(pos, module):
keyword_list = filter(
lambda keyword: not keyword.only_valid_as_leaf,
keyword_list
)
return [keyword.name for keyword in keyword_list]
def all_keywords(evaluator, pos=(0, 0)):
return set([Keyword(evaluator, k, pos) for k in keys])
def keyword(evaluator, string, pos=(0, 0)):
if string in keys:
return Keyword(evaluator, string, pos)
else:
return None
def get_operator(evaluator, string, pos):
return Keyword(evaluator, string, pos)
keywords_only_valid_as_leaf = (
'continue',
'break',
)
class Keyword(object):
type = 'completion_keyword'
def __init__(self, evaluator, name, pos):
self.name = FakeName(name, self, pos)
self.start_pos = pos
self.parent = evaluator.BUILTINS
def get_parent_until(self):
return self.parent
@property
def only_valid_as_leaf(self):
return self.name.value in keywords_only_valid_as_leaf
@property
def names(self):
""" For a `parsing.Name` like comparision """
return [self.name]
@property
def docstr(self):
return imitate_pydoc(self.name)
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, self.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 common.ignored(KeyError):
# try to access symbols
string = h.symbols[string]
string, _, related = string.partition(' ')
get_target = lambda s: 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] if pydoc_topics else ''
except KeyError:
return ''

View file

@ -0,0 +1,27 @@
"""
To use Jedi completion in Python interpreter, add the following in your shell
setup (e.g., ``.bashrc``)::
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().split().in<TAB> # doctest: +SKIP
os.path.join().split().index os.path.join().split().insert
"""
import jedi.utils
from jedi import __version__ as __jedi_version__
print('REPL completion using Jedi %s' % __jedi_version__)
jedi.utils.setup_readline()
del jedi
# Note: try not to do many things here, as it will contaminate global
# namespace of the interpreter.

View file

@ -0,0 +1,49 @@
from jedi._compatibility import unicode
from jedi.api import classes
from jedi.parser import tree
from jedi.evaluate import imports
def usages(evaluator, definition_names, mods):
"""
:param definitions: list of Name
"""
def compare_array(definitions):
""" `definitions` are being compared by module/start_pos, because
sometimes the id's of the objects change (e.g. executions).
"""
result = []
for d in definitions:
module = d.get_parent_until()
result.append((module, d.start_pos))
return result
search_name = unicode(list(definition_names)[0])
compare_definitions = compare_array(definition_names)
mods |= set([d.get_parent_until() for d in definition_names])
definitions = []
for m in imports.get_modules_containing_name(evaluator, mods, search_name):
try:
check_names = m.used_names[search_name]
except KeyError:
continue
for name in check_names:
result = evaluator.goto(name)
if [c for c in compare_array(result) if c in compare_definitions]:
definitions.append(classes.Definition(evaluator, name))
# Previous definitions might be imports, so include them
# (because goto might return that import name).
compare_definitions += compare_array([name])
return definitions
def usages_add_import_modules(evaluator, definitions):
""" Adds the modules of the imports """
new = set()
for d in definitions:
imp_or_stmt = d.get_definition()
if isinstance(imp_or_stmt, tree.Import):
s = imports.ImportWrapper(evaluator, d)
new |= set(s.follow(is_goto=True))
return set(definitions) | new

View file

@ -0,0 +1,127 @@
"""
This caching is very important for speed and memory optimizations. There's
nothing really spectacular, just some decorators. The following cache types are
available:
- module caching (`load_parser` and `save_parser`), which uses pickle and is
really important to assure low load times of modules like ``numpy``.
- ``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 jedi import settings
from jedi.parser.utils import parser_cache
from jedi.parser.utils import underscore_memoization
_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 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 memoize_method(method):
"""A normal memoize function."""
def wrapper(self, *args, **kwargs):
dct = self.__dict__.setdefault('_memoize_method_dct', {})
key = (args, frozenset(kwargs.items()))
try:
return dct[key]
except KeyError:
result = method(self, *args, **kwargs)
dct[key] = result
return result
return wrapper
def cache_star_import(func):
@time_cache("star_import_cache_validity")
def wrapper(self):
yield self.base # The cache key
yield func(self)
return wrapper
def _invalidate_star_import_cache_module(module, only_main=False):
""" Important if some new modules are being reparsed """
try:
t, modules = _time_caches['star_import_cache_validity'][module]
except KeyError:
pass
else:
del _time_caches['star_import_cache_validity'][module]
# This stuff was part of load_parser. However since we're most likely
# not going to use star import caching anymore, just ignore it.
#else:
# In case there is already a module cached and this module
# has to be reparsed, we also need to invalidate the import
# caches.
# _invalidate_star_import_cache_module(parser_cache_item.parser.module)
def invalidate_star_import_cache(path):
"""On success returns True."""
try:
parser_cache_item = parser_cache[path]
except KeyError:
pass
else:
_invalidate_star_import_cache_module(parser_cache_item.parser.module)

View file

@ -0,0 +1,169 @@
""" A universal module with functions / classes without dependencies. """
import sys
import contextlib
import functools
import re
from itertools import chain
from ast import literal_eval
from jedi._compatibility import unicode, reraise
from jedi import settings
class UncaughtAttributeError(Exception):
"""
Important, because `__getattr__` and `hasattr` catch AttributeErrors
implicitly. This is really evil (mainly because of `__getattr__`).
`hasattr` in Python 2 is even more evil, because it catches ALL exceptions.
Therefore this class originally had to be derived from `BaseException`
instead of `Exception`. But because I removed relevant `hasattr` from
the code base, we can now switch back to `Exception`.
:param base: return values of sys.exc_info().
"""
def safe_property(func):
return property(reraise_uncaught(func))
def reraise_uncaught(func):
"""
Re-throw uncaught `AttributeError`.
Usage: Put ``@rethrow_uncaught`` in front of the function
which does **not** suppose to raise `AttributeError`.
AttributeError is easily get caught by `hasattr` and another
``except AttributeError`` clause. This becomes problem when you use
a lot of "dynamic" attributes (e.g., using ``@property``) because you
can't distinguish if the property does not exist for real or some code
inside of the "dynamic" attribute through that error. In a well
written code, such error should not exist but getting there is very
difficult. This decorator is to help us getting there by changing
`AttributeError` to `UncaughtAttributeError` to avoid unexpected catch.
This helps us noticing bugs earlier and facilitates debugging.
.. note:: Treating StopIteration here is easy.
Add that feature when needed.
"""
@functools.wraps(func)
def wrapper(*args, **kwds):
try:
return func(*args, **kwds)
except AttributeError:
exc_info = sys.exc_info()
reraise(UncaughtAttributeError(exc_info[1]), exc_info[2])
return wrapper
class PushBackIterator(object):
def __init__(self, iterator):
self.pushes = []
self.iterator = iterator
self.current = None
def push_back(self, value):
self.pushes.append(value)
def __iter__(self):
return self
def next(self):
""" Python 2 Compatibility """
return self.__next__()
def __next__(self):
if self.pushes:
self.current = self.pushes.pop()
else:
self.current = next(self.iterator)
return self.current
@contextlib.contextmanager
def scale_speed_settings(factor):
a = settings.max_executions
b = settings.max_until_execution_unique
settings.max_executions *= factor
settings.max_until_execution_unique *= factor
try:
yield
finally:
settings.max_executions = a
settings.max_until_execution_unique = b
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
@contextlib.contextmanager
def ignored(*exceptions):
"""
Context manager that ignores all of the specified exceptions. This will
be in the standard library starting with Python 3.4.
"""
try:
yield
except exceptions:
pass
def source_to_unicode(source, encoding=None):
def detect_encoding():
"""
For the implementation of encoding definitions in Python, look at:
- http://www.python.org/dev/peps/pep-0263/
- http://docs.python.org/2/reference/lexical_analysis.html#encoding-declarations
"""
byte_mark = literal_eval(r"b'\xef\xbb\xbf'")
if source.startswith(byte_mark):
# UTF-8 byte-order mark
return 'utf-8'
first_two_lines = re.match(br'(?:[^\n]*\n){0,2}', source).group(0)
possible_encoding = re.search(br"coding[=:]\s*([-\w.]+)",
first_two_lines)
if possible_encoding:
return possible_encoding.group(1)
else:
# the default if nothing else has been set -> PEP 263
return encoding if encoding is not None else 'utf-8'
if isinstance(source, unicode):
# only cast str/bytes
return source
encoding = detect_encoding()
if not isinstance(encoding, unicode):
encoding = unicode(encoding, 'utf-8', 'replace')
# cast to unicode by default
return unicode(source, encoding, 'replace')
def splitlines(string):
"""
A splitlines for Python code. In contrast to Python's ``str.splitlines``,
looks at form feeds and other special characters as normal text. Just
splits ``\n`` and ``\r\n``.
Also different: Returns ``['']`` for an empty string input.
"""
return re.split('\n|\r\n', string)
def unite(iterable):
"""Turns a two dimensional array into a one dimensional."""
return set(chain.from_iterable(iterable))
def to_list(func):
def wrapper(*args, **kwargs):
return list(func(*args, **kwargs))
return wrapper

View file

@ -0,0 +1,133 @@
from jedi._compatibility import encoding, is_py3, u
import inspect
import os
import time
def _lazy_colorama_init():
"""
Lazily init colorama if necessary, not to screw up stdout is debug not
enabled.
This version of the function does nothing.
"""
pass
_inited=False
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():
"""
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()
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 = ''
NOTICE = object()
WARNING = object()
SPEED = object()
enable_speed = False
enable_warning = False
enable_notice = False
# callback, interface: level, str
debug_function = None
ignored_modules = ['jedi.parser']
_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):
global _debug_indent
_debug_indent += 1
try:
return func(*args, **kwargs)
finally:
_debug_indent -= 1
return wrapper
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:
frm = inspect.stack()[1]
mod = inspect.getmodule(frm[0])
if not (mod.__name__ in ignored_modules):
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 +1,534 @@
"""
Evaluation 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 called lazy evaluation. If you
don't know about it, google it. That said, the typical entry point for static
analysis is calling ``eval_statement``. There's separate logic for
autocompletion in the API, the evaluator is all about evaluating an expression.
Now you need to understand what follows after ``eval_statement``. 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 ``eval_statement`` will
return the ``date`` class.
To *visualize* this (simplified):
- ``Evaluator.eval_statement`` doesn't do much, because there's no assignment.
- ``Evaluator.eval_element`` cares for resolving the dotted path
- ``Evaluator.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 ``eval_element`` 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 ``eval_statement`` 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 evaluation is really good because it
only *evaluates* what needs to be *evaluated*. All the statements and modules
that are not used are just being ignored.
"""
import copy
import sys
from itertools import chain
from jedi.parser import tree
from jedi import debug
from jedi.evaluate import representation as er
from jedi.evaluate import imports
from jedi.evaluate import recursion
from jedi.evaluate import iterable
from jedi.evaluate.cache import memoize_default
from jedi.evaluate import stdlib
from jedi.evaluate import finder
from jedi.evaluate import compiled
from jedi.evaluate import precedence
from jedi.evaluate import param
from jedi.evaluate import helpers
class Evaluator(object):
def __init__(self, grammar, sys_path=None):
self.grammar = grammar
self.memoize_cache = {} # for memoize decorators
# To memorize modules -> equals `sys.modules`.
self.modules = {} # like `sys.modules`.
self.compiled_cache = {} # see `evaluate.compiled.create()`
self.mixed_cache = {} # see `evaluate.compiled.mixed.create()`
self.analysis = []
self.predefined_if_name_dict_dict = {}
self.dynamic_params_depth = 0
self.is_analysis = False
if sys_path is None:
sys_path = sys.path
self.sys_path = copy.copy(sys_path)
try:
self.sys_path.remove('')
except ValueError:
pass
self.reset_recursion_limitations()
# Constants
self.BUILTINS = compiled.get_special_object(self, 'BUILTINS')
def reset_recursion_limitations(self):
self.recursion_detector = recursion.RecursionDetector(self)
self.execution_recursion_detector = recursion.ExecutionRecursionDetector(self)
def wrap(self, element):
if isinstance(element, (er.Wrapper, er.InstanceElement,
er.ModuleWrapper, er.FunctionExecution, er.Instance, compiled.CompiledObject)) or element is None:
# TODO this is so ugly, please refactor.
return element
if element.type == 'classdef':
return er.Class(self, element)
elif element.type == 'funcdef':
return er.Function(self, element)
elif element.type == 'lambda':
return er.LambdaWrapper(self, element)
elif element.type == 'file_input':
return er.ModuleWrapper(self, element)
else:
return element
def find_types(self, scope, name_str, position=None, search_global=False,
is_goto=False):
"""
This is the search function. The most important part to debug.
`remove_statements` and `filter_statements` really are the core part of
this completion.
:param position: Position of the last statement -> tuple of line, column
:return: List of Names. Their parents are the types.
"""
f = finder.NameFinder(self, scope, name_str, position)
scopes = f.scopes(search_global)
if is_goto:
return f.filter_name(scopes)
return f.find(scopes, attribute_lookup=not search_global)
#@memoize_default(default=[], evaluator_is_first_arg=True)
#@recursion.recursion_decorator
@debug.increase_indent
def eval_statement(self, stmt, seek_name=None):
"""
The starting point of the completion. A statement always owns a call
list, which are the calls, that a statement does. In case multiple
names are defined in the statement, `seek_name` returns the result for
this name.
:param stmt: A `tree.ExprStmt`.
"""
debug.dbg('eval_statement %s (%s)', stmt, seek_name)
rhs = stmt.get_rhs()
types = self.eval_element(rhs)
if seek_name:
types = finder.check_tuple_assignments(self, types, seek_name)
first_operation = stmt.first_operation()
if first_operation not in ('=', None) and not isinstance(stmt, er.InstanceElement): # TODO don't check for this.
# `=` is always the last character in aug assignments -> -1
operator = copy.copy(first_operation)
operator.value = operator.value[:-1]
name = str(stmt.get_defined_names()[0])
parent = self.wrap(stmt.get_parent_scope())
left = self.find_types(parent, name, stmt.start_pos, search_global=True)
for_stmt = stmt.get_parent_until(tree.ForStmt)
if isinstance(for_stmt, tree.ForStmt) and types \
and for_stmt.defines_one_name():
# Iterate through result and add the values, that's possible
# only in for loops without clutter, because they are
# predictable. Also only do it, if the variable is not a tuple.
node = for_stmt.get_input_node()
for_iterables = self.eval_element(node)
ordered = list(iterable.py__iter__(self, for_iterables, node))
for index_types in ordered:
dct = {str(for_stmt.children[1]): index_types}
self.predefined_if_name_dict_dict[for_stmt] = dct
t = self.eval_element(rhs)
left = precedence.calculate(self, left, operator, t)
types = left
if ordered:
# If there are no for entries, we cannot iterate and the
# types are defined by += entries. Therefore the for loop
# is never called.
del self.predefined_if_name_dict_dict[for_stmt]
else:
types = precedence.calculate(self, left, operator, types)
debug.dbg('eval_statement result %s', types)
return types
def eval_element(self, element):
if isinstance(element, iterable.AlreadyEvaluated):
return set(element)
elif isinstance(element, iterable.MergedNodes):
return iterable.unite(self.eval_element(e) for e in element)
if_stmt = element.get_parent_until((tree.IfStmt, tree.ForStmt, tree.IsScope))
predefined_if_name_dict = self.predefined_if_name_dict_dict.get(if_stmt)
if predefined_if_name_dict is None and isinstance(if_stmt, tree.IfStmt):
if_stmt_test = if_stmt.children[1]
name_dicts = [{}]
# If we already did a check, we don't want to do it again -> If
# predefined_if_name_dict_dict is filled, we stop.
# We don't want to check the if stmt itself, it's just about
# the content.
if element.start_pos > if_stmt_test.end_pos:
# Now we need to check if the names in the if_stmt match the
# names in the suite.
if_names = helpers.get_names_of_node(if_stmt_test)
element_names = helpers.get_names_of_node(element)
str_element_names = [str(e) for e in element_names]
if any(str(i) in str_element_names for i in if_names):
for if_name in if_names:
definitions = self.goto_definitions(if_name)
# Every name that has multiple different definitions
# causes the complexity to rise. The complexity should
# never fall below 1.
if len(definitions) > 1:
if len(name_dicts) * len(definitions) > 16:
debug.dbg('Too many options for if branch evaluation %s.', if_stmt)
# There's only a certain amount of branches
# Jedi can evaluate, otherwise it will take to
# long.
name_dicts = [{}]
break
original_name_dicts = list(name_dicts)
name_dicts = []
for definition in definitions:
new_name_dicts = list(original_name_dicts)
for i, name_dict in enumerate(new_name_dicts):
new_name_dicts[i] = name_dict.copy()
new_name_dicts[i][str(if_name)] = [definition]
name_dicts += new_name_dicts
else:
for name_dict in name_dicts:
name_dict[str(if_name)] = definitions
if len(name_dicts) > 1:
result = set()
for name_dict in name_dicts:
self.predefined_if_name_dict_dict[if_stmt] = name_dict
try:
result |= self._eval_element_not_cached(element)
finally:
del self.predefined_if_name_dict_dict[if_stmt]
return result
else:
return self._eval_element_if_evaluated(element)
return self._eval_element_cached(element)
else:
if predefined_if_name_dict:
return self._eval_element_not_cached(element)
else:
return self._eval_element_if_evaluated(element)
return self._eval_element_cached(element)
def _eval_element_if_evaluated(self, element):
"""
TODO This function is temporary: Merge with eval_element.
"""
parent = element
while parent is not None:
parent = parent.parent
predefined_if_name_dict = self.predefined_if_name_dict_dict.get(parent)
if predefined_if_name_dict is not None:
return self._eval_element_not_cached(element)
return self._eval_element_cached(element)
@memoize_default(default=set(), evaluator_is_first_arg=True)
def _eval_element_cached(self, element):
return self._eval_element_not_cached(element)
@debug.increase_indent
def _eval_element_not_cached(self, element):
debug.dbg('eval_element %s@%s', element, element.start_pos)
types = set()
if isinstance(element, (tree.Name, tree.Literal)) or tree.is_node(element, 'atom'):
types = self._eval_atom(element)
elif isinstance(element, tree.Keyword):
# For False/True/None
if element.value in ('False', 'True', 'None'):
types.add(compiled.builtin_from_name(self, element.value))
# else: print e.g. could be evaluated like this in Python 2.7
elif element.isinstance(tree.Lambda):
types = set([er.LambdaWrapper(self, element)])
elif element.isinstance(er.LambdaWrapper):
types = set([element]) # TODO this is no real evaluation.
elif element.type == 'expr_stmt':
types = self.eval_statement(element)
elif element.type in ('power', 'atom_expr'):
types = self._eval_atom(element.children[0])
for trailer in element.children[1:]:
if trailer == '**': # has a power operation.
right = self.eval_element(element.children[2])
types = set(precedence.calculate(self, types, trailer, right))
break
types = self.eval_trailer(types, trailer)
elif element.type in ('testlist_star_expr', 'testlist',):
# The implicit tuple in statements.
types = set([iterable.ImplicitTuple(self, element)])
elif element.type in ('not_test', 'factor'):
types = self.eval_element(element.children[-1])
for operator in element.children[:-1]:
types = set(precedence.factor_calculate(self, types, operator))
elif element.type == 'test':
# `x if foo else y` case.
types = (self.eval_element(element.children[0]) |
self.eval_element(element.children[-1]))
elif element.type == 'operator':
# Must be an ellipsis, other operators are not evaluated.
assert element.value == '...'
types = set([compiled.create(self, Ellipsis)])
elif element.type == 'dotted_name':
types = self._eval_atom(element.children[0])
for next_name in element.children[2::2]:
types = set(chain.from_iterable(self.find_types(typ, next_name)
for typ in types))
types = types
elif element.type == 'eval_input':
types = self._eval_element_not_cached(element.children[0])
else:
types = precedence.calculate_children(self, element.children)
debug.dbg('eval_element result %s', types)
return types
def _eval_atom(self, atom):
"""
Basically to process ``atom`` nodes. The parser sometimes doesn't
generate the node (because it has just one child). In that case an atom
might be a name or a literal as well.
"""
if isinstance(atom, tree.Name):
# This is the first global lookup.
stmt = atom.get_definition()
scope = stmt.get_parent_until(tree.IsScope, include_current=True)
if isinstance(scope, (tree.Function, er.FunctionExecution)):
# Adjust scope: If the name is not in the suite, it's a param
# default or annotation and will be resolved as part of the
# parent scope.
colon = scope.children.index(':')
if atom.start_pos < scope.children[colon + 1].start_pos:
scope = scope.get_parent_scope()
if isinstance(stmt, tree.CompFor):
stmt = stmt.get_parent_until((tree.ClassOrFunc, tree.ExprStmt))
if stmt.type != 'expr_stmt':
# We only need to adjust the start_pos for statements, because
# there the name cannot be used.
stmt = atom
return self.find_types(scope, atom, stmt.start_pos, search_global=True)
elif isinstance(atom, tree.Literal):
return set([compiled.create(self, atom.eval())])
else:
c = atom.children
if c[0].type == 'string':
# Will be one string.
types = self._eval_atom(c[0])
for string in c[1:]:
right = self._eval_atom(string)
types = precedence.calculate(self, types, '+', right)
return types
# Parentheses without commas are not tuples.
elif c[0] == '(' and not len(c) == 2 \
and not(tree.is_node(c[1], 'testlist_comp')
and len(c[1].children) > 1):
return self.eval_element(c[1])
try:
comp_for = c[1].children[1]
except (IndexError, AttributeError):
pass
else:
if comp_for == ':':
# Dict comprehensions have a colon at the 3rd index.
try:
comp_for = c[1].children[3]
except IndexError:
pass
if comp_for.type == 'comp_for':
return set([iterable.Comprehension.from_atom(self, atom)])
return set([iterable.Array(self, atom)])
def eval_trailer(self, types, trailer):
trailer_op, node = trailer.children[:2]
if node == ')': # `arglist` is optional.
node = ()
new_types = set()
if trailer_op == '[':
new_types |= iterable.py__getitem__(self, types, trailer)
else:
for typ in types:
debug.dbg('eval_trailer: %s in scope %s', trailer, typ)
if trailer_op == '.':
new_types |= self.find_types(typ, node)
elif trailer_op == '(':
new_types |= self.execute(typ, node, trailer)
return new_types
def execute_evaluated(self, obj, *args):
"""
Execute a function with already executed arguments.
"""
args = [iterable.AlreadyEvaluated([arg]) for arg in args]
return self.execute(obj, args)
@debug.increase_indent
def execute(self, obj, arguments=(), trailer=None):
if not isinstance(arguments, param.Arguments):
arguments = param.Arguments(self, arguments, trailer)
if self.is_analysis:
arguments.eval_all()
if obj.isinstance(er.Function):
obj = obj.get_decorated_func()
debug.dbg('execute: %s %s', obj, arguments)
try:
# Some stdlib functions like super(), namedtuple(), etc. have been
# hard-coded in Jedi to support them.
return stdlib.execute(self, obj, arguments)
except stdlib.NotInStdLib:
pass
try:
func = obj.py__call__
except AttributeError:
debug.warning("no execution possible %s", obj)
return set()
else:
types = func(arguments)
debug.dbg('execute result: %s in %s', types, obj)
return types
def goto_definitions(self, name):
def_ = name.get_definition()
is_simple_name = name.parent.type not in ('power', 'trailer')
if is_simple_name:
if name.parent.type == 'classdef' and name.parent.name == name:
return [self.wrap(name.parent)]
if name.parent.type in ('file_input', 'funcdef'):
return [self.wrap(name.parent)]
if def_.type == 'expr_stmt' and name in def_.get_defined_names():
return self.eval_statement(def_, name)
elif def_.type == 'for_stmt':
container_types = self.eval_element(def_.children[3])
for_types = iterable.py__iter__types(self, container_types, def_.children[3])
return finder.check_tuple_assignments(self, for_types, name)
elif def_.type in ('import_from', 'import_name'):
return imports.ImportWrapper(self, name).follow()
call = helpers.call_of_leaf(name)
return self.eval_element(call)
def goto(self, name):
def resolve_implicit_imports(names):
for name in names:
if isinstance(name.parent, helpers.FakeImport):
# Those are implicit imports.
s = imports.ImportWrapper(self, name)
for n in s.follow(is_goto=True):
yield n
else:
yield name
stmt = name.get_definition()
par = name.parent
if par.type == 'argument' and par.children[1] == '=' and par.children[0] == name:
# Named param goto.
trailer = par.parent
if trailer.type == 'arglist':
trailer = trailer.parent
if trailer.type != 'classdef':
if trailer.type == 'decorator':
types = self.eval_element(trailer.children[1])
else:
i = trailer.parent.children.index(trailer)
to_evaluate = trailer.parent.children[:i]
types = self.eval_element(to_evaluate[0])
for trailer in to_evaluate[1:]:
types = self.eval_trailer(types, trailer)
param_names = []
for typ in types:
try:
params = typ.params
except AttributeError:
pass
else:
param_names += [param.name for param in params
if param.name.value == name.value]
return param_names
elif isinstance(par, tree.ExprStmt) and name in par.get_defined_names():
# Only take the parent, because if it's more complicated than just
# a name it's something you can "goto" again.
return [name]
elif isinstance(par, (tree.Param, tree.Function, tree.Class)) and par.name is name:
return [name]
elif isinstance(stmt, tree.Import):
modules = imports.ImportWrapper(self, name).follow(is_goto=True)
return list(resolve_implicit_imports(modules))
elif par.type == 'dotted_name': # Is a decorator.
index = par.children.index(name)
if index > 0:
new_dotted = helpers.deep_ast_copy(par)
new_dotted.children[index - 1:] = []
types = self.eval_element(new_dotted)
return resolve_implicit_imports(iterable.unite(
self.find_types(typ, name, is_goto=True) for typ in types
))
scope = name.get_parent_scope()
if tree.is_node(par, 'trailer') and par.children[0] == '.':
call = helpers.call_of_leaf(name, cut_own_trailer=True)
types = self.eval_element(call)
return resolve_implicit_imports(iterable.unite(
self.find_types(typ, name, is_goto=True) for typ in types
))
else:
if stmt.type != 'expr_stmt':
# We only need to adjust the start_pos for statements, because
# there the name cannot be used.
stmt = name
return self.find_types(scope, name, stmt.start_pos,
search_global=True, is_goto=True)

View file

@ -0,0 +1,216 @@
"""
Module for statical analysis.
"""
from jedi import debug
from jedi.parser import tree
from jedi.evaluate.compiled import CompiledObject
from jedi.common import unite
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(evaluator, name, jedi_obj, message=None, typ=Error, payload=None):
from jedi.evaluate.iterable import MergedNodes
while isinstance(jedi_obj, MergedNodes):
if len(jedi_obj) != 1:
# TODO is this kosher?
return
jedi_obj = list(jedi_obj)[0]
exception = CODES[name][1]
if _check_for_exception_catch(evaluator, jedi_obj, exception, payload):
return
module_path = jedi_obj.get_parent_until().path
instance = typ(name, module_path, jedi_obj.start_pos, message)
debug.warning(str(instance), format=False)
evaluator.analysis.append(instance)
def _check_for_setattr(instance):
"""
Check if there's any setattr method inside an instance. If so, return True.
"""
module = instance.get_parent_until()
try:
stmts = module.used_names['setattr']
except KeyError:
return False
return any(instance.start_pos < stmt.start_pos < instance.end_pos
for stmt in stmts)
def add_attribute_error(evaluator, scope, name):
message = ('AttributeError: %s has no attribute %s.' % (scope, name))
from jedi.evaluate.representation import Instance
# Check for __getattr__/__getattribute__ existance and issue a warning
# instead of an error, if that happens.
if isinstance(scope, Instance):
typ = Warning
try:
scope.get_subscope_by_name('__getattr__')
except KeyError:
try:
scope.get_subscope_by_name('__getattribute__')
except KeyError:
if not _check_for_setattr(scope):
typ = Error
else:
typ = Error
payload = scope, name
add(evaluator, 'attribute-error', name, message, typ, payload)
def _check_for_exception_catch(evaluator, jedi_obj, 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):
try:
return isinstance(cls, CompiledObject) and issubclass(exception, cls.obj)
except TypeError:
return False
def check_try_for_except(obj, exception):
# Only nodes in try
iterator = iter(obj.children)
for branch_type in iterator:
colon = next(iterator)
suite = next(iterator)
if branch_type == 'try' \
and not (branch_type.start_pos < jedi_obj.start_pos <= suite.end_pos):
return False
for node in obj.except_clauses():
if node is None:
return True # An exception block that catches everything.
else:
except_classes = evaluator.eval_element(node)
for cls in except_classes:
from jedi.evaluate import iterable
if isinstance(cls, iterable.Array) and cls.type == 'tuple':
# multiple exceptions
for typ in unite(cls.py__iter__()):
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_obj.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.evaluate.param import Arguments
args = list(Arguments(evaluator, arglist).unpack())
# Arguments should be very simple
assert len(args) == 2
# Check name
key, values = args[1]
assert len(values) == 1
names = list(evaluator.eval_element(values[0]))
assert len(names) == 1 and isinstance(names[0], CompiledObject)
assert names[0].obj == str(payload[1])
# Check objects
key, values = args[0]
assert len(values) == 1
objects = evaluator.eval_element(values[0])
return payload[0] in objects
except AssertionError:
return False
obj = jedi_obj
while obj is not None and not obj.isinstance(tree.Function, tree.Class):
if obj.isinstance(tree.Flow):
# try/except catch check
if obj.isinstance(tree.TryStmt) and check_try_for_except(obj, exception):
return True
# hasattr check
if exception == AttributeError and obj.isinstance(tree.IfStmt, tree.WhileStmt):
if check_hasattr(obj.children[1], obj.children[3]):
return True
obj = obj.parent
return False

View file

@ -0,0 +1,58 @@
"""
- the popular ``memoize_default`` works like a typical memoize and returns the
default otherwise.
- ``CachedMetaClass`` uses ``memoize_default`` to do the same with classes.
"""
import inspect
NO_DEFAULT = object()
def memoize_default(default=NO_DEFAULT, evaluator_is_first_arg=False, second_arg_is_evaluator=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):
if evaluator_is_first_arg:
cache = obj.memoize_cache
elif second_arg_is_evaluator: # needed for meta classes
cache = args[0].memoize_cache
else:
cache = obj._evaluator.memoize_cache
try:
memo = cache[function]
except KeyError:
memo = {}
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)
if inspect.isgenerator(rv):
rv = list(rv)
memo[key] = rv
return rv
return wrapper
return func
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).
"""
@memoize_default(None, second_arg_is_evaluator=True)
def __call__(self, *args, **kwargs):
return super(CachedMetaClass, self).__call__(*args, **kwargs)

View file

@ -0,0 +1,544 @@
"""
Imitate the parser representation.
"""
import inspect
import re
import sys
import os
from functools import partial
from jedi._compatibility import builtins as _builtins, unicode
from jedi import debug
from jedi.cache import underscore_memoization, memoize_method
from jedi.parser.tree import Param, Base, Operator, zero_position_modifier
from jedi.evaluate.helpers import FakeName
from . import fake
_sep = os.path.sep
if os.path.altsep is not None:
_sep += os.path.altsep
_path_re = re.compile('(?:\.[^{0}]+|[{0}]__init__\.py)$'.format(re.escape(_sep)))
del _sep
class CheckAttribute(object):
"""Raises an AttributeError if the attribute X isn't available."""
def __init__(self, func):
self.func = func
# Remove the py in front of e.g. py__call__.
self.check_name = func.__name__[2:]
def __get__(self, instance, owner):
# This might raise an AttributeError. That's wanted.
getattr(instance.obj, self.check_name)
return partial(self.func, instance)
class CompiledObject(Base):
# comply with the parser
start_pos = 0, 0
path = None # modules have this attribute - set it to None.
used_names = {} # To be consistent with modules.
def __init__(self, evaluator, obj, parent=None):
self._evaluator = evaluator
self.obj = obj
self.parent = parent
@CheckAttribute
def py__call__(self, params):
if inspect.isclass(self.obj):
from jedi.evaluate.representation import Instance
return set([Instance(self._evaluator, self, params)])
else:
return set(self._execute_function(params))
@CheckAttribute
def py__class__(self):
return create(self._evaluator, self.obj.__class__)
@CheckAttribute
def py__mro__(self):
return tuple(create(self._evaluator, cls) for cls in self.obj.__mro__)
@CheckAttribute
def py__bases__(self):
return tuple(create(self._evaluator, cls) for cls in self.obj.__bases__)
def py__bool__(self):
return bool(self.obj)
def py__file__(self):
return self.obj.__file__
def is_class(self):
return inspect.isclass(self.obj)
@property
def doc(self):
return inspect.getdoc(self.obj) or ''
@property
def params(self):
params_str, ret = self._parse_function_doc()
tokens = params_str.split(',')
if inspect.ismethoddescriptor(self.obj):
tokens.insert(0, 'self')
params = []
for p in tokens:
parts = [FakeName(part) for part in p.strip().split('=')]
if len(parts) > 1:
parts.insert(1, Operator(zero_position_modifier, '=', (0, 0)))
params.append(Param(parts, self))
return params
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, repr(self.obj))
@underscore_memoization
def _parse_function_doc(self):
if self.doc is None:
return '', ''
return _parse_function_doc(self.doc)
def api_type(self):
obj = self.obj
if inspect.isclass(obj):
return 'class'
elif inspect.ismodule(obj):
return 'module'
elif inspect.isbuiltin(obj) or inspect.ismethod(obj) \
or inspect.ismethoddescriptor(obj) or inspect.isfunction(obj):
return 'function'
# Everything else...
return 'instance'
@property
def type(self):
"""Imitate the tree.Node.type values."""
cls = self._get_class()
if inspect.isclass(cls):
return 'classdef'
elif inspect.ismodule(cls):
return 'file_input'
elif inspect.isbuiltin(cls) or inspect.ismethod(cls) or \
inspect.ismethoddescriptor(cls):
return 'funcdef'
@underscore_memoization
def _cls(self):
"""
We used to limit the lookups for instantiated objects like list(), but
this is not the case anymore. Python itself
"""
# Ensures that a CompiledObject is returned that is not an instance (like list)
return self
def _get_class(self):
if not fake.is_class_instance(self.obj) or \
inspect.ismethoddescriptor(self.obj): # slots
return self.obj
try:
return self.obj.__class__
except AttributeError:
# happens with numpy.core.umath._UFUNC_API (you get it
# automatically by doing `import numpy`.
return type
@property
def names_dict(self):
# For compatibility with `representation.Class`.
return self.names_dicts(False)[0]
def names_dicts(self, search_global, is_instance=False):
return self._names_dict_ensure_one_dict(is_instance)
@memoize_method
def _names_dict_ensure_one_dict(self, is_instance):
"""
search_global shouldn't change the fact that there's one dict, this way
there's only one `object`.
"""
return [LazyNamesDict(self._evaluator, self, is_instance)]
def get_subscope_by_name(self, name):
if name in dir(self.obj):
return CompiledName(self._evaluator, self, name).parent
else:
raise KeyError("CompiledObject doesn't have an attribute '%s'." % name)
@CheckAttribute
def py__getitem__(self, index):
if type(self.obj) not in (str, list, tuple, unicode, bytes, bytearray, dict):
# Get rid of side effects, we won't call custom `__getitem__`s.
return set()
return set([create(self._evaluator, self.obj[index])])
@CheckAttribute
def py__iter__(self):
if type(self.obj) not in (str, list, tuple, unicode, bytes, bytearray, dict):
# Get rid of side effects, we won't call custom `__getitem__`s.
return
for part in self.obj:
yield set([create(self._evaluator, part)])
@property
def name(self):
try:
name = self._get_class().__name__
except AttributeError:
name = repr(self.obj)
return FakeName(name, self)
def _execute_function(self, params):
if self.type != 'funcdef':
return
for name in self._parse_function_doc()[1].split():
try:
bltn_obj = getattr(_builtins, name)
except AttributeError:
continue
else:
if bltn_obj is None:
# We want to evaluate everything except None.
# TODO do we?
continue
bltn_obj = create(self._evaluator, bltn_obj)
for result in self._evaluator.execute(bltn_obj, params):
yield result
@property
@underscore_memoization
def subscopes(self):
"""
Returns only the faked scopes - the other ones are not important for
internal analysis.
"""
module = self.get_parent_until()
faked_subscopes = []
for name in dir(self.obj):
try:
faked_subscopes.append(
fake.get_faked(module.obj, self.obj, parent=self, name=name)
)
except fake.FakeDoesNotExist:
pass
return faked_subscopes
def is_scope(self):
return True
def get_self_attributes(self):
return [] # Instance compatibility
def get_imports(self):
return [] # Builtins don't have imports
class CompiledName(FakeName):
def __init__(self, evaluator, compiled_obj, name):
super(CompiledName, self).__init__(name)
self._evaluator = evaluator
self._compiled_obj = compiled_obj
self.name = name
def __repr__(self):
try:
name = self._compiled_obj.name # __name__ is not defined all the time
except AttributeError:
name = None
return '<%s: (%s).%s>' % (type(self).__name__, name, self.name)
def is_definition(self):
return True
@property
@underscore_memoization
def parent(self):
module = self._compiled_obj.get_parent_until()
return _create_from_name(self._evaluator, module, self._compiled_obj, self.name)
@parent.setter
def parent(self, value):
pass # Just ignore this, FakeName tries to overwrite the parent attribute.
class LazyNamesDict(object):
"""
A names_dict instance for compiled objects, resembles the parser.tree.
"""
name_class = CompiledName
def __init__(self, evaluator, compiled_obj, is_instance=False):
self._evaluator = evaluator
self._compiled_obj = compiled_obj
self._is_instance = is_instance
def __iter__(self):
return (v[0].value for v in self.values())
@memoize_method
def __getitem__(self, name):
try:
getattr(self._compiled_obj.obj, name)
except AttributeError:
raise KeyError('%s in %s not found.' % (name, self._compiled_obj))
except Exception:
# This is a bit ugly. We're basically returning this to make
# lookups possible without having the actual attribute. However
# this makes proper completion possible.
return [FakeName(name, create(self._evaluator, None), is_definition=True)]
return [self.name_class(self._evaluator, self._compiled_obj, name)]
def values(self):
obj = self._compiled_obj.obj
values = []
for name in dir(obj):
try:
values.append(self[name])
except KeyError:
# The dir function can be wrong.
pass
is_instance = self._is_instance or fake.is_class_instance(obj)
# ``dir`` doesn't include the type names.
if not inspect.ismodule(obj) and obj != type and not is_instance:
values += create(self._evaluator, type).names_dict.values()
return values
def dotted_from_fs_path(fs_path, sys_path):
"""
Changes `/usr/lib/python3.4/email/utils.py` to `email.utils`. I.e.
compares the path with sys.path and then returns the dotted_path. If the
path is not in the sys.path, just returns None.
"""
if os.path.basename(fs_path).startswith('__init__.'):
# We are calculating the path. __init__ files are not interesting.
fs_path = os.path.dirname(fs_path)
# prefer
# - UNIX
# /path/to/pythonX.Y/lib-dynload
# /path/to/pythonX.Y/site-packages
# - Windows
# C:\path\to\DLLs
# C:\path\to\Lib\site-packages
# over
# - UNIX
# /path/to/pythonX.Y
# - Windows
# C:\path\to\Lib
path = ''
for s in sys_path:
if (fs_path.startswith(s) and len(path) < len(s)):
path = s
# - Window
# X:\path\to\lib-dynload/datetime.pyd => datetime
module_path = fs_path[len(path):].lstrip(os.path.sep).lstrip('/')
# - Window
# Replace like X:\path\to\something/foo/bar.py
return _path_re.sub('', module_path).replace(os.path.sep, '.').replace('/', '.')
def load_module(evaluator, path=None, name=None):
sys_path = evaluator.sys_path
if path is not None:
dotted_path = dotted_from_fs_path(path, sys_path=sys_path)
else:
dotted_path = name
if dotted_path is None:
p, _, dotted_path = path.partition(os.path.sep)
sys_path.insert(0, p)
temp, sys.path = sys.path, sys_path
try:
__import__(dotted_path)
except RuntimeError:
if 'PySide' in dotted_path or 'PyQt' in dotted_path:
# RuntimeError: the PyQt4.QtCore and PyQt5.QtCore modules both wrap
# the QObject class.
# See https://github.com/davidhalter/jedi/pull/483
return None
raise
except ImportError:
# If a module is "corrupt" or not really a Python module or whatever.
debug.warning('Module %s not importable.', path)
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_path]
return create(evaluator, module)
docstr_defaults = {
'floating point number': 'float',
'character': 'str',
'integer': 'int',
'dictionary': 'dict',
'string': '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'
"""
# 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 = ''
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('-[>-]* ', doc[end:end + 7])
if r is None:
ret = ''
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(evaluator, module, parent, name):
try:
return fake.get_faked(module.obj, parent.obj, parent=parent, name=name)
except fake.FakeDoesNotExist:
pass
try:
obj = getattr(parent.obj, name)
except AttributeError:
# Happens e.g. in properties of
# PyQt4.QtGui.QStyleOptionComboBox.currentText
# -> just set it to None
obj = None
return create(evaluator, obj, parent)
def builtin_from_name(evaluator, string):
bltn_obj = getattr(_builtins, string)
return create(evaluator, bltn_obj)
def _a_generator(foo):
"""Used to have an object to return for generators."""
yield 42
yield foo
_SPECIAL_OBJECTS = {
'FUNCTION_CLASS': type(load_module),
'METHOD_CLASS': type(CompiledObject.is_class),
'MODULE_CLASS': type(os),
'GENERATOR_OBJECT': _a_generator(1.0),
'BUILTINS': _builtins,
}
def get_special_object(evaluator, identifier):
obj = _SPECIAL_OBJECTS[identifier]
return create(evaluator, obj, parent=create(evaluator, _builtins))
def compiled_objects_cache(attribute_name):
def decorator(func):
"""
This decorator caches just the ids, oopposed to caching the object itself.
Caching the id has the advantage that an object doesn't need to be
hashable.
"""
def wrapper(evaluator, obj, parent=None, module=None):
cache = getattr(evaluator, attribute_name)
# Do a very cheap form of caching here.
key = id(obj), id(parent)
try:
return cache[key][0]
except KeyError:
# TODO this whole decorator looks way too ugly and this if
# doesn't make it better. Find a more generic solution.
if parent or module:
result = func(evaluator, obj, parent, module)
else:
result = func(evaluator, obj)
# Need to cache all of them, otherwise the id could be overwritten.
cache[key] = result, obj, parent, module
return result
return wrapper
return decorator
@compiled_objects_cache('compiled_cache')
def create(evaluator, obj, parent=None, module=None):
"""
A very weird interface class to this module. The more options provided the
more acurate loading compiled objects is.
"""
if inspect.ismodule(obj):
if parent is not None:
# Modules don't have parents, be careful with caching: recurse.
return create(evaluator, obj)
else:
if parent is None and obj != _builtins:
return create(evaluator, obj, create(evaluator, _builtins))
try:
return fake.get_faked(module and module.obj, obj, parent=parent)
except fake.FakeDoesNotExist:
pass
return CompiledObject(evaluator, obj, parent)

View file

@ -0,0 +1,203 @@
"""
Loads functions that are mixed in to the standard library. E.g. builtins are
written in C (binaries), but my autocompletion only understands Python code. By
mixing in Python code, the autocompletion should work much better for builtins.
"""
import os
import inspect
import types
from jedi._compatibility import is_py3, builtins, unicode, is_py34
from jedi.parser import ParserWithRecovery, load_grammar
from jedi.parser import tree as pt
from jedi.evaluate.helpers import FakeName
modules = {}
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
)
if is_py34:
NOT_CLASS_TYPES += (types.DynamicClassAttribute,)
class FakeDoesNotExist(Exception):
pass
def _load_faked_module(module):
module_name = module.__name__
if module_name == '__builtin__' and not is_py3:
module_name = 'builtins'
try:
return modules[module_name]
except KeyError:
path = os.path.dirname(os.path.abspath(__file__))
try:
with open(os.path.join(path, 'fake', module_name) + '.pym') as f:
source = f.read()
except IOError:
modules[module_name] = None
return
grammar = load_grammar(version='3.4')
module = ParserWithRecovery(grammar, unicode(source), module_name).module
modules[module_name] = module
if module_name == 'builtins' and not is_py3:
# There are two implementations of `open` for either python 2/3.
# -> Rename the python2 version (`look at fake/builtins.pym`).
open_func = search_scope(module, 'open')
open_func.children[1] = FakeName('open_python3')
open_func = search_scope(module, 'open_python2')
open_func.children[1] = FakeName('open')
return module
def search_scope(scope, obj_name):
for s in scope.subscopes:
if str(s.name) == obj_name:
return s
def get_module(obj):
if inspect.ismodule(obj):
return obj
try:
obj = obj.__objclass__
except AttributeError:
pass
try:
imp_plz = obj.__module__
except AttributeError:
# Unfortunately in some cases like `int` there's no __module__
return builtins
else:
if imp_plz is None:
# Happens for example in `(_ for _ in []).send.__module__`.
return builtins
else:
try:
return __import__(imp_plz)
except ImportError:
# __module__ can be something arbitrary that doesn't exist.
return builtins
def _faked(module, obj, name):
# Crazy underscore actions to try to escape all the internal madness.
if module is None:
module = get_module(obj)
faked_mod = _load_faked_module(module)
if faked_mod is None:
return None
# Having the module as a `parser.representation.module`, we need to scan
# for methods.
if name is None:
if inspect.isbuiltin(obj):
return search_scope(faked_mod, obj.__name__)
elif not inspect.isclass(obj):
# object is a method or descriptor
try:
objclass = obj.__objclass__
except AttributeError:
return None
else:
cls = search_scope(faked_mod, objclass.__name__)
if cls is None:
return None
return search_scope(cls, obj.__name__)
else:
if obj == module:
return search_scope(faked_mod, name)
else:
try:
cls_name = obj.__name__
except AttributeError:
return None
cls = search_scope(faked_mod, cls_name)
if cls is None:
return None
return search_scope(cls, name)
def memoize_faked(obj):
"""
A typical memoize function that ignores issues with non hashable results.
"""
cache = obj.cache = {}
def memoizer(*args, **kwargs):
key = (obj, args, frozenset(kwargs.items()))
try:
result = cache[key]
except TypeError:
return obj(*args, **kwargs)
except KeyError:
result = obj(*args, **kwargs)
if result is not None:
cache[key] = obj(*args, **kwargs)
return result
else:
return result
return memoizer
@memoize_faked
def _get_faked(module, obj, name=None):
obj = type(obj) if is_class_instance(obj) else obj
result = _faked(module, obj, name)
if result is None or isinstance(result, pt.Class):
# We're not interested in classes. What we want is functions.
raise FakeDoesNotExist
else:
# Set the docstr which was previously not set (faked modules don't
# contain it).
doc = '"""%s"""' % obj.__doc__ # TODO need escapes.
suite = result.children[-1]
string = pt.String(pt.zero_position_modifier, doc, (0, 0), '')
new_line = pt.Newline('\n', (0, 0), '')
docstr_node = pt.Node('simple_stmt', [string, new_line])
suite.children.insert(2, docstr_node)
return result
def get_faked(module, obj, name=None, parent=None):
faked = _get_faked(module, obj, name)
faked.parent = parent
return faked
def is_class_instance(obj):
"""Like inspect.* methods."""
try:
cls = obj.__class__
except AttributeError:
return False
else:
return cls != type and not issubclass(cls, NOT_CLASS_TYPES)

View file

@ -0,0 +1,9 @@
class partial():
def __init__(self, func, *args, **keywords):
self.__func = func
self.__args = args
self.__keywords = keywords
def __call__(self, *args, **kwargs):
# TODO should be **dict(self.__keywords, **kwargs)
return self.__func(*(self.__args + args), **self.__keywords)

View file

@ -0,0 +1,26 @@
def connect(database, timeout=None, isolation_level=None, detect_types=None, factory=None):
return Connection()
class Connection():
def cursor(self):
return Cursor()
class Cursor():
def cursor(self):
return Cursor()
def fetchone(self):
return Row()
def fetchmany(self, size=cursor.arraysize):
return [self.fetchone()]
def fetchall(self):
return [self.fetchone()]
class Row():
def keys(self):
return ['']

View file

@ -0,0 +1,99 @@
def compile():
class SRE_Match():
endpos = int()
lastgroup = int()
lastindex = int()
pos = int()
string = str()
regs = ((int(), int()),)
def __init__(self, pattern):
self.re = pattern
def start(self):
return int()
def end(self):
return int()
def span(self):
return int(), int()
def expand(self):
return str()
def group(self, nr):
return str()
def groupdict(self):
return {str(): str()}
def groups(self):
return (str(),)
class SRE_Pattern():
flags = int()
groupindex = {}
groups = int()
pattern = str()
def findall(self, string, pos=None, endpos=None):
"""
findall(string[, pos[, endpos]]) --> list.
Return a list of all non-overlapping matches of pattern in string.
"""
return [str()]
def finditer(self, string, pos=None, endpos=None):
"""
finditer(string[, pos[, endpos]]) --> iterator.
Return an iterator over all non-overlapping matches for the
RE pattern in string. For each match, the iterator returns a
match object.
"""
yield SRE_Match(self)
def match(self, string, pos=None, endpos=None):
"""
match(string[, pos[, endpos]]) --> match object or None.
Matches zero or more characters at the beginning of the string
pattern
"""
return SRE_Match(self)
def scanner(self, string, pos=None, endpos=None):
pass
def search(self, string, pos=None, endpos=None):
"""
search(string[, pos[, endpos]]) --> match object or None.
Scan through string looking for a match, and return a corresponding
MatchObject instance. Return None if no position in the string matches.
"""
return SRE_Match(self)
def split(self, string, maxsplit=0]):
"""
split(string[, maxsplit = 0]) --> list.
Split string by the occurrences of pattern.
"""
return [str()]
def sub(self, repl, string, count=0):
"""
sub(repl, string[, count = 0]) --> newstring
Return the string obtained by replacing the leftmost non-overlapping
occurrences of pattern in string by the replacement repl.
"""
return str()
def subn(self, repl, string, count=0):
"""
subn(repl, string[, count = 0]) --> (newstring, number of subs)
Return the tuple (new_string, number_of_subs_made) found by replacing
the leftmost non-overlapping occurrences of pattern with the
replacement repl.
"""
return (str(), int())
return SRE_Pattern()

View file

@ -0,0 +1,8 @@
def proxy(object, callback=None):
return object
class weakref():
def __init__(self, object, callback=None):
self.__object = object
def __call__(self):
return self.__object

View file

@ -0,0 +1,262 @@
"""
Pure Python implementation of some builtins.
This code is not going to be executed anywhere.
These implementations are not always correct, but should work as good as
possible for the auto completion.
"""
def next(iterator, default=None):
if random.choice([0, 1]):
if hasattr("next"):
return iterator.next()
else:
return iterator.__next__()
else:
if default is not None:
return default
def iter(collection, sentinel=None):
if sentinel:
yield collection()
else:
for c in collection:
yield c
def range(start, stop=None, step=1):
return [0]
class file():
def __iter__(self):
yield ''
def next(self):
return ''
class xrange():
# Attention: this function doesn't exist in Py3k (there it is range).
def __iter__(self):
yield 1
def count(self):
return 1
def index(self):
return 1
def open(file, mode='r', buffering=-1, encoding=None, errors=None, newline=None, closefd=True):
import io
return io.TextIOWrapper(file, mode, buffering, encoding, errors, newline, closefd)
def open_python2(name, mode=None, buffering=None):
return file(name, mode, buffering)
#--------------------------------------------------------
# descriptors
#--------------------------------------------------------
class property():
def __init__(self, fget, fset=None, fdel=None, doc=None):
self.fget = fget
self.fset = fset
self.fdel = fdel
self.__doc__ = doc
def __get__(self, obj, cls):
return self.fget(obj)
def __set__(self, obj, value):
self.fset(obj, value)
def __delete__(self, obj):
self.fdel(obj)
def setter(self, func):
self.fset = func
return self
def getter(self, func):
self.fget = func
return self
def deleter(self, func):
self.fdel = func
return self
class staticmethod():
def __init__(self, func):
self.__func = func
def __get__(self, obj, cls):
return self.__func
class classmethod():
def __init__(self, func):
self.__func = func
def __get__(self, obj, cls):
def _method(*args, **kwargs):
return self.__func(cls, *args, **kwargs)
return _method
#--------------------------------------------------------
# array stuff
#--------------------------------------------------------
class list():
def __init__(self, iterable=[]):
self.__iterable = []
for i in iterable:
self.__iterable += [i]
def __iter__(self):
for i in self.__iterable:
yield i
def __getitem__(self, y):
return self.__iterable[y]
def pop(self):
return self.__iterable[int()]
class tuple():
def __init__(self, iterable=[]):
self.__iterable = []
for i in iterable:
self.__iterable += [i]
def __iter__(self):
for i in self.__iterable:
yield i
def __getitem__(self, y):
return self.__iterable[y]
def index(self):
return 1
def count(self):
return 1
class set():
def __init__(self, iterable=[]):
self.__iterable = iterable
def __iter__(self):
for i in self.__iterable:
yield i
def pop(self):
return list(self.__iterable)[-1]
def copy(self):
return self
def difference(self, other):
return self - other
def intersection(self, other):
return self & other
def symmetric_difference(self, other):
return self ^ other
def union(self, other):
return self | other
class frozenset():
def __init__(self, iterable=[]):
self.__iterable = iterable
def __iter__(self):
for i in self.__iterable:
yield i
def copy(self):
return self
class dict():
def __init__(self, **elements):
self.__elements = elements
def clear(self):
# has a strange docstr
pass
def get(self, k, d=None):
# TODO implement
try:
#return self.__elements[k]
pass
except KeyError:
return d
def values(self):
return self.__elements.values()
def setdefault(self, k, d):
# TODO maybe also return the content
return d
class enumerate():
def __init__(self, sequence, start=0):
self.__sequence = sequence
def __iter__(self):
for i in self.__sequence:
yield 1, i
def __next__(self):
return next(self.__iter__())
def next(self):
return next(self.__iter__())
class reversed():
def __init__(self, sequence):
self.__sequence = sequence
def __iter__(self):
for i in self.__sequence:
yield i
def __next__(self):
return next(self.__iter__())
def next(self):
return next(self.__iter__())
def sorted(iterable, cmp=None, key=None, reverse=False):
return iterable
#--------------------------------------------------------
# basic types
#--------------------------------------------------------
class int():
def __init__(self, x, base=None):
pass
class str():
def __init__(self, obj):
pass
class type():
def mro():
return [object]

View file

@ -0,0 +1,4 @@
class datetime():
@staticmethod
def now():
return datetime()

View file

@ -0,0 +1,6 @@
class TextIOWrapper():
def __next__(self):
return str()
def __iter__(self):
yield str()

View file

@ -0,0 +1,5 @@
def getcwd():
return ''
def getcwdu():
return ''

View file

@ -0,0 +1,158 @@
"""
Used only for REPL Completion.
"""
import inspect
import os
from jedi import common
from jedi.parser.fast import FastParser
from jedi.evaluate import compiled
from jedi.cache import underscore_memoization
class MixedObject(object):
"""
A ``MixedObject`` is used in two ways:
1. It uses the default logic of ``parser.tree`` objects,
2. except for getattr calls. The names dicts are generated in a fashion
like ``CompiledObject``.
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 CompiledObject 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, evaluator, obj, node_name):
self._evaluator = evaluator
self.obj = obj
self.node_name = node_name
self.definition = node_name.get_definition()
@property
def names_dict(self):
return LazyMixedNamesDict(self._evaluator, self)
def names_dicts(self, search_global):
# TODO is this needed?
assert search_global is False
return [self.names_dict]
def api_type(self):
mappings = {
'expr_stmt': 'statement',
'classdef': 'class',
'funcdef': 'function',
'file_input': 'module',
}
return mappings[self.definition.type]
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, repr(self.obj))
def __getattr__(self, name):
return getattr(self.definition, name)
class MixedName(compiled.CompiledName):
"""
The ``CompiledName._compiled_object`` is our MixedObject.
"""
@property
@underscore_memoization
def parent(self):
return create(self._evaluator, getattr(self._compiled_obj.obj, self.name))
@parent.setter
def parent(self, value):
pass # Just ignore this, Name tries to overwrite the parent attribute.
@property
def start_pos(self):
if isinstance(self.parent, MixedObject):
return self.parent.node_name.start_pos
# This means a start_pos that doesn't exist (compiled objects).
return (0, 0)
class LazyMixedNamesDict(compiled.LazyNamesDict):
name_class = MixedName
def parse(grammar, path):
with open(path) as f:
source = f.read()
source = common.source_to_unicode(source)
return FastParser(grammar, source, path)
def _load_module(evaluator, path, python_object):
module = parse(evaluator.grammar, path).module
python_module = inspect.getmodule(python_object)
evaluator.modules[python_module.__name__] = module
return module
def find_syntax_node_name(evaluator, python_object):
try:
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
module = _load_module(evaluator, path, python_object)
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``).
return module
name_str = python_object.__name__
if name_str == '<lambda>':
return None # It's too hard to find lambdas.
names = module.used_names[name_str]
names = [n for n in names if n.is_definition()]
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:
return line_names[-1]
# 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.
return names[-1]
@compiled.compiled_objects_cache('mixed_cache')
def create(evaluator, obj):
name = find_syntax_node_name(evaluator, obj)
if name is None:
return compiled.create(evaluator, obj)
else:
return MixedObject(evaluator, obj, name)

View file

@ -0,0 +1,204 @@
"""
Docstrings are another source of information for functions and classes.
:mod:`jedi.evaluate.dynamic` tries to find all executions of functions, while
the docstring parsing is much easier. There are two 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>`_
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.
"""
from ast import literal_eval
import re
from itertools import chain
from textwrap import dedent
from jedi.evaluate.cache import memoize_default
from jedi.parser import ParserWithRecovery, load_grammar
from jedi.parser.tree import Class
from jedi.common import indent_block
from jedi.evaluate.iterable import Array, FakeSequence, AlreadyEvaluated
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':[^`]+:`([^`]+)`')
try:
from numpydoc.docscrape import NumpyDocString
except ImportError:
def _search_param_in_numpydocstr(docstr, param_str):
return []
else:
def _search_param_in_numpydocstr(docstr, param_str):
"""Search `docstr` (in numpydoc format) for type(-s) of `param_str`."""
params = NumpyDocString(docstr)._parsed_data['Parameters']
for p_name, p_type, p_descr in params:
if p_name == param_str:
m = re.match('([^,]+(,[^,]+)*?)(,[ ]*optional)?$', p_type)
if m:
p_type = m.group(1)
if p_type.startswith('{'):
types = set(type(x).__name__ for x in literal_eval(p_type))
return list(types)
else:
return [p_type]
return []
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) or
[])
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 _evaluate_for_statement_string(evaluator, string, module):
code = dedent("""
def pseudo_docstring_stuff():
# Create a pseudo function for docstring statements.
%s
""")
if string is None:
return []
for element in re.findall('((?:\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.
p = ParserWithRecovery(load_grammar(), code % indent_block(string))
try:
pseudo_cls = p.module.subscopes[0]
# First pick suite, then simple_stmt (-2 for DEDENT) and then the node,
# which is also not the last item, because there's a newline.
stmt = pseudo_cls.children[-1].children[-2].children[-2]
except (AttributeError, IndexError):
return []
# 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.
pseudo_cls.parent = module
return list(_execute_types_in_stmt(evaluator, stmt))
def _execute_types_in_stmt(evaluator, 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 = evaluator.eval_element(stmt)
return chain.from_iterable(_execute_array_values(evaluator, d) for d in definitions)
def _execute_array_values(evaluator, 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.
"""
if isinstance(array, Array):
values = []
for types in array.py__iter__():
objects = set(chain.from_iterable(_execute_array_values(evaluator, typ) for typ in types))
values.append(AlreadyEvaluated(objects))
return [FakeSequence(evaluator, values, array.type)]
else:
return evaluator.execute(array)
@memoize_default(None, evaluator_is_first_arg=True)
def follow_param(evaluator, param):
def eval_docstring(docstring):
return set(
[p for param_str in _search_param_in_docstr(docstring, str(param.name))
for p in _evaluate_for_statement_string(evaluator, param_str, module)]
)
func = param.parent_function
module = param.get_parent_until()
types = eval_docstring(func.raw_doc)
if func.name.value == '__init__':
cls = func.get_parent_until(Class)
if cls.type == 'classdef':
types |= eval_docstring(cls.raw_doc)
return types
@memoize_default(None, evaluator_is_first_arg=True)
def find_return_types(evaluator, func):
def search_return_in_docstr(code):
for p in DOCSTRING_RETURN_PATTERNS:
match = p.search(code)
if match:
return _strip_rst_role(match.group(1))
type_str = search_return_in_docstr(func.raw_doc)
return _evaluate_for_statement_string(evaluator, type_str, func.get_parent_until())

View file

@ -0,0 +1,149 @@
"""
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. This work with a ``ParamListener``.
"""
from itertools import chain
from jedi._compatibility import unicode
from jedi.parser import tree
from jedi import settings
from jedi import debug
from jedi.evaluate.cache import memoize_default
from jedi.evaluate import imports
MAX_PARAM_SEARCHES = 20
class ParamListener(object):
"""
This listener is used to get the params for a function.
"""
def __init__(self):
self.param_possibilities = []
def execute(self, params):
self.param_possibilities += params
@debug.increase_indent
def search_params(evaluator, param):
"""
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.
"""
if not settings.dynamic_params:
return set()
evaluator.dynamic_params_depth += 1
try:
func = param.get_parent_until(tree.Function)
debug.dbg('Dynamic param search for %s in %s.', param, str(func.name), color='MAGENTA')
# Compare the param names.
names = [n for n in search_function_call(evaluator, func)
if n.value == param.name.value]
# Evaluate the ExecutedParams to types.
result = set(chain.from_iterable(n.parent.eval(evaluator) for n in names))
debug.dbg('Dynamic param result %s', result, color='MAGENTA')
return result
finally:
evaluator.dynamic_params_depth -= 1
@memoize_default([], evaluator_is_first_arg=True)
def search_function_call(evaluator, func):
"""
Returns a list of param names.
"""
from jedi.evaluate import representation as er
def get_possible_nodes(module, func_name):
try:
names = module.used_names[func_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 undecorate(typ):
# We have to remove decorators, because they are not the
# "original" functions, this way we can easily compare.
# At the same time we also have to remove InstanceElements.
if typ.isinstance(er.Function, er.Instance) \
and typ.decorates is not None:
return typ.decorates
elif isinstance(typ, er.InstanceElement):
return typ.var
else:
return typ
current_module = func.get_parent_until()
func_name = unicode(func.name)
compare = func
if func_name == '__init__':
cls = func.get_parent_scope()
if isinstance(cls, tree.Class):
func_name = unicode(cls.name)
compare = cls
# add the listener
listener = ParamListener()
func.listeners.add(listener)
try:
result = []
i = 0
for mod in imports.get_modules_containing_name(evaluator, [current_module], func_name):
for name, trailer in get_possible_nodes(mod, func_name):
i += 1
# This is a simple way to stop Jedi's dynamic param recursion
# from going wild: The deeper Jedi's in the recursin, the less
# code should be evaluated.
if i * evaluator.dynamic_params_depth > MAX_PARAM_SEARCHES:
return listener.param_possibilities
for typ in evaluator.goto_definitions(name):
undecorated = undecorate(typ)
if evaluator.wrap(compare) == undecorated:
# Only if we have the correct function we execute
# it, otherwise just ignore it.
evaluator.eval_trailer([typ], trailer)
result = listener.param_possibilities
# If there are results after processing a module, we're probably
# good to process.
if result:
return result
finally:
# cleanup: remove the listener; important: should not stick.
func.listeners.remove(listener)
return set()

View file

@ -0,0 +1,632 @@
"""
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 itertools import chain
from jedi._compatibility import unicode
from jedi.parser import tree
from jedi import debug
from jedi import common
from jedi.common import unite
from jedi import settings
from jedi.evaluate import representation as er
from jedi.evaluate import dynamic
from jedi.evaluate import compiled
from jedi.evaluate import docstrings
from jedi.evaluate import pep0484
from jedi.evaluate import iterable
from jedi.evaluate import imports
from jedi.evaluate import analysis
from jedi.evaluate import flow_analysis
from jedi.evaluate import param
from jedi.evaluate import helpers
from jedi.evaluate.cache import memoize_default
def filter_after_position(names, position):
"""
Removes all names after a certain position. If position is None, just
returns the names list.
"""
if position is None:
return names
names_new = []
for n in names:
# Filter positions and also allow list comprehensions and lambdas.
if n.start_pos[0] is not None and n.start_pos < position \
or isinstance(n.get_definition(), (tree.CompFor, tree.Lambda)):
names_new.append(n)
return names_new
def filter_definition_names(names, origin, position=None):
"""
Filter names that are actual definitions in a scope. Names that are just
used will be ignored.
"""
if not names:
return []
# Just calculate the scope from the first
stmt = names[0].get_definition()
scope = stmt.get_parent_scope()
if not (isinstance(scope, er.FunctionExecution) and
isinstance(scope.base, er.LambdaWrapper)):
names = filter_after_position(names, position)
names = [name for name in names if name.is_definition()]
# Private name mangling (compile.c) disallows access on names
# preceeded by two underscores `__` if used outside of the class. Names
# that also end with two underscores (e.g. __id__) are not affected.
for name in list(names):
if name.value.startswith('__') and not name.value.endswith('__'):
if filter_private_variable(scope, origin):
names.remove(name)
return names
class NameFinder(object):
def __init__(self, evaluator, scope, name_str, position=None):
self._evaluator = evaluator
# Make sure that it's not just a syntax tree node.
self.scope = evaluator.wrap(scope)
self.name_str = name_str
self.position = position
self._found_predefined_if_name = None
@debug.increase_indent
def find(self, scopes, attribute_lookup):
"""
:params bool attribute_lookup: Tell to logic if we're accessing the
attribute or the contents of e.g. a function.
"""
# TODO rename scopes to names_dicts
names = self.filter_name(scopes)
if self._found_predefined_if_name is not None:
return self._found_predefined_if_name
types = self._names_to_types(names, attribute_lookup)
if not names and not types \
and not (isinstance(self.name_str, tree.Name) and
isinstance(self.name_str.parent.parent, tree.Param)):
if not isinstance(self.name_str, (str, unicode)): # TODO Remove?
if attribute_lookup:
analysis.add_attribute_error(self._evaluator,
self.scope, self.name_str)
else:
message = ("NameError: name '%s' is not defined."
% self.name_str)
analysis.add(self._evaluator, 'name-error', self.name_str,
message)
debug.dbg('finder._names_to_types: %s -> %s', names, types)
return types
def scopes(self, search_global=False):
if search_global:
return global_names_dict_generator(self._evaluator, self.scope, self.position)
else:
return ((n, None) for n in self.scope.names_dicts(search_global))
def names_dict_lookup(self, names_dict, position):
def get_param(scope, el):
if isinstance(el.get_parent_until(tree.Param), tree.Param):
return scope.param_by_name(str(el))
return el
search_str = str(self.name_str)
try:
names = names_dict[search_str]
if not names: # We want names, otherwise stop.
return []
except KeyError:
return []
names = filter_definition_names(names, self.name_str, position)
name_scope = None
# Only the names defined in the last position are valid definitions.
last_names = []
for name in reversed(sorted(names, key=lambda name: name.start_pos)):
stmt = name.get_definition()
name_scope = self._evaluator.wrap(stmt.get_parent_scope())
if isinstance(self.scope, er.Instance) and not isinstance(name_scope, er.Instance):
# Instances should not be checked for positioning, because we
# don't know in which order the functions are called.
last_names.append(name)
continue
if isinstance(name_scope, compiled.CompiledObject):
# Let's test this. TODO need comment. shouldn't this be
# filtered before?
last_names.append(name)
continue
if isinstance(stmt, er.ModuleWrapper):
# In case of REPL completion, we can infer modules names that
# don't really have a definition (because they are really just
# namespaces). In this case we can just add it.
last_names.append(name)
continue
if isinstance(name, compiled.CompiledName) \
or isinstance(name, er.InstanceName) and isinstance(name._origin_name, compiled.CompiledName):
last_names.append(name)
continue
if isinstance(self.name_str, tree.Name):
origin_scope = self.name_str.get_parent_until(tree.Scope, reverse=True)
scope = self.name_str
check = None
while True:
scope = scope.parent
if scope.type in ("if_stmt", "for_stmt", "comp_for"):
try:
name_dict = self._evaluator.predefined_if_name_dict_dict[scope]
types = set(name_dict[str(self.name_str)])
except KeyError:
continue
else:
if self.name_str.start_pos < scope.children[1].end_pos:
# It doesn't make any sense to check if
# statements in the if statement itself, just
# deliver types.
self._found_predefined_if_name = types
else:
check = flow_analysis.break_check(self._evaluator, self.scope,
origin_scope)
if check is flow_analysis.UNREACHABLE:
self._found_predefined_if_name = set()
else:
self._found_predefined_if_name = types
break
if isinstance(scope, tree.IsScope) or scope is None:
break
else:
origin_scope = None
if isinstance(stmt.parent, compiled.CompiledObject):
# TODO seriously? this is stupid.
continue
check = flow_analysis.break_check(self._evaluator, name_scope,
stmt, origin_scope)
if check is not flow_analysis.UNREACHABLE:
last_names.append(name)
if check is flow_analysis.REACHABLE:
break
if isinstance(name_scope, er.FunctionExecution):
# Replace params
return [get_param(name_scope, n) for n in last_names]
return last_names
def filter_name(self, names_dicts):
"""
Searches names that are defined in a scope (the different
`names_dicts`), until a name fits.
"""
names = []
for names_dict, position in names_dicts:
names = self.names_dict_lookup(names_dict, position)
if names:
break
debug.dbg('finder.filter_name "%s" in (%s): %s@%s', self.name_str,
self.scope, names, self.position)
return list(self._clean_names(names))
def _clean_names(self, names):
"""
``NameFinder.filter_name`` should only output names with correct
wrapper parents. We don't want to see AST classes out in the
evaluation, so remove them already here!
"""
for n in names:
definition = n.parent
if isinstance(definition, (compiled.CompiledObject,
iterable.BuiltinMethod)):
# TODO this if should really be removed by changing the type of
# those classes.
yield n
elif definition.type in ('funcdef', 'classdef', 'file_input'):
yield self._evaluator.wrap(definition).name
else:
yield n
def _check_getattr(self, inst):
"""Checks for both __getattr__ and __getattribute__ methods"""
result = set()
# str is important, because it shouldn't be `Name`!
name = compiled.create(self._evaluator, str(self.name_str))
with common.ignored(KeyError):
result = inst.execute_subscope_by_name('__getattr__', name)
if not result:
# This is a little bit special. `__getattribute__` is in Python
# executed before `__getattr__`. But: I know no use case, where
# this could be practical and where jedi would return wrong types.
# If you ever find something, let me know!
# We are inversing this, because a hand-crafted `__getattribute__`
# could still call another hand-crafted `__getattr__`, but not the
# other way around.
with common.ignored(KeyError):
result = inst.execute_subscope_by_name('__getattribute__', name)
return result
def _names_to_types(self, names, attribute_lookup):
types = set()
# Add isinstance and other if/assert knowledge.
if isinstance(self.name_str, tree.Name):
# Ignore FunctionExecution parents for now.
flow_scope = self.name_str
until = flow_scope.get_parent_until(er.FunctionExecution)
while not isinstance(until, er.FunctionExecution):
flow_scope = flow_scope.get_parent_scope(include_flows=True)
if flow_scope is None:
break
# TODO check if result is in scope -> no evaluation necessary
n = check_flow_information(self._evaluator, flow_scope,
self.name_str, self.position)
if n:
return n
for name in names:
new_types = _name_to_types(self._evaluator, name, self.scope)
if isinstance(self.scope, (er.Class, er.Instance)) and attribute_lookup:
types |= set(self._resolve_descriptors(name, new_types))
else:
types |= set(new_types)
if not names and isinstance(self.scope, er.Instance):
# handling __getattr__ / __getattribute__
return self._check_getattr(self.scope)
return types
def _resolve_descriptors(self, name, types):
# The name must not be in the dictionary, but part of the class
# definition. __get__ is only called if the descriptor is defined in
# the class dictionary.
name_scope = name.get_definition().get_parent_scope()
if not isinstance(name_scope, (er.Instance, tree.Class)):
return types
result = set()
for r in types:
try:
desc_return = r.get_descriptor_returns
except AttributeError:
result.add(r)
else:
result |= desc_return(self.scope)
return result
def _get_global_stmt_scopes(evaluator, global_stmt, name):
global_stmt_scope = global_stmt.get_parent_scope()
module = global_stmt_scope.get_parent_until()
for used_name in module.used_names[str(name)]:
if used_name.parent.type == 'global_stmt':
yield evaluator.wrap(used_name.get_parent_scope())
@memoize_default(set(), evaluator_is_first_arg=True)
def _name_to_types(evaluator, name, scope):
types = []
typ = name.get_definition()
if typ.isinstance(tree.ForStmt):
types = pep0484.find_type_from_comment_hint_for(evaluator, typ, name)
if types:
return types
if typ.isinstance(tree.WithStmt):
types = pep0484.find_type_from_comment_hint_with(evaluator, typ, name)
if types:
return types
if typ.isinstance(tree.ForStmt, tree.CompFor):
container_types = evaluator.eval_element(typ.children[3])
for_types = iterable.py__iter__types(evaluator, container_types, typ.children[3])
types = check_tuple_assignments(evaluator, for_types, name)
elif isinstance(typ, tree.Param):
types = _eval_param(evaluator, typ, scope)
elif typ.isinstance(tree.ExprStmt):
types = _remove_statements(evaluator, typ, name)
elif typ.isinstance(tree.WithStmt):
types = evaluator.eval_element(typ.node_from_name(name))
elif isinstance(typ, tree.Import):
types = imports.ImportWrapper(evaluator, name).follow()
elif typ.type == 'global_stmt':
for s in _get_global_stmt_scopes(evaluator, typ, name):
finder = NameFinder(evaluator, s, str(name))
names_dicts = finder.scopes(search_global=True)
# For global_stmt lookups, we only need the first possible scope,
# which means the function itself.
names_dicts = [next(names_dicts)]
types += finder.find(names_dicts, attribute_lookup=False)
elif isinstance(typ, tree.TryStmt):
# TODO an exception can also be a tuple. Check for those.
# TODO check for types that are not classes and add it to
# the static analysis report.
exceptions = evaluator.eval_element(name.get_previous_sibling().get_previous_sibling())
types = set(chain.from_iterable(evaluator.execute(t) for t in exceptions))
else:
if typ.isinstance(er.Function):
typ = typ.get_decorated_func()
types = set([typ])
return types
def _remove_statements(evaluator, stmt, name):
"""
This is the part where statements are being stripped.
Due to lazy evaluation, statements like a = func; b = a; b() have to be
evaluated.
"""
types = set()
# Remove the statement docstr stuff for now, that has to be
# implemented with the evaluator class.
#if stmt.docstr:
#res_new.append(stmt)
check_instance = None
if isinstance(stmt, er.InstanceElement) and stmt.is_class_var:
check_instance = stmt.instance
stmt = stmt.var
pep0484types = \
pep0484.find_type_from_comment_hint_assign(evaluator, stmt, name)
if pep0484types:
return pep0484types
types |= evaluator.eval_statement(stmt, seek_name=name)
if check_instance is not None:
# class renames
types = set([er.get_instance_el(evaluator, check_instance, a, True)
if isinstance(a, (er.Function, tree.Function))
else a for a in types])
return types
def _eval_param(evaluator, param, scope):
res_new = set()
func = param.get_parent_scope()
cls = func.parent.get_parent_until((tree.Class, tree.Function))
from jedi.evaluate.param import ExecutedParam, Arguments
if isinstance(cls, tree.Class) and param.position_nr == 0 \
and not isinstance(param, ExecutedParam):
# This is where we add self - if it has never been
# instantiated.
if isinstance(scope, er.InstanceElement):
res_new.add(scope.instance)
else:
inst = er.Instance(evaluator, evaluator.wrap(cls),
Arguments(evaluator, ()), is_generated=True)
res_new.add(inst)
return res_new
# Instances are typically faked, if the instance is not called from
# outside. Here we check it for __init__ functions and return.
if isinstance(func, er.InstanceElement) \
and func.instance.is_generated and str(func.name) == '__init__':
param = func.var.params[param.position_nr]
# Add pep0484 and docstring knowledge.
pep0484_hints = pep0484.follow_param(evaluator, param)
doc_params = docstrings.follow_param(evaluator, param)
if pep0484_hints or doc_params:
return list(set(pep0484_hints) | set(doc_params))
if isinstance(param, ExecutedParam):
return res_new | param.eval(evaluator)
else:
# Param owns no information itself.
res_new |= dynamic.search_params(evaluator, param)
if not res_new:
if param.stars:
t = 'tuple' if param.stars == 1 else 'dict'
typ = list(evaluator.find_types(evaluator.BUILTINS, t))[0]
res_new = evaluator.execute(typ)
if param.default:
res_new |= evaluator.eval_element(param.default)
return res_new
def check_flow_information(evaluator, 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 = set()
if flow.is_scope():
# Check for asserts.
try:
names = reversed(flow.names_dict[search_name.value])
except (KeyError, AttributeError):
names = []
for name in names:
ass = name.get_parent_until(tree.AssertStmt)
if isinstance(ass, tree.AssertStmt) and pos is not None and ass.start_pos < pos:
result = _check_isinstance_type(evaluator, ass.assertion(), search_name)
if result:
break
if isinstance(flow, (tree.IfStmt, tree.WhileStmt)):
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(evaluator, if_test, search_name)
return result
def _check_isinstance_type(evaluator, element, search_name):
try:
assert element.type in ('power', 'atom_expr')
# this might be removed if we analyze and, etc
assert len(element.children) == 2
first, trailer = element.children
assert isinstance(first, tree.Name) and first.value == 'isinstance'
assert trailer.type == 'trailer' and trailer.children[0] == '('
assert len(trailer.children) == 3
# arglist stuff
arglist = trailer.children[1]
args = param.Arguments(evaluator, arglist, trailer)
lst = list(args.unpack())
# Disallow keyword arguments
assert len(lst) == 2 and lst[0][0] is None and lst[1][0] is None
name = lst[0][1][0] # first argument, values, first value
# Do a simple get_code comparison. They should just have the same code,
# and everything will be all right.
classes = lst[1][1][0]
call = helpers.call_of_leaf(search_name)
assert name.get_code(normalized=True) == call.get_code(normalized=True)
except AssertionError:
return set()
result = set()
for cls_or_tup in evaluator.eval_element(classes):
if isinstance(cls_or_tup, iterable.Array) and cls_or_tup.type == 'tuple':
for typ in unite(cls_or_tup.py__iter__()):
result |= evaluator.execute(typ)
else:
result |= evaluator.execute(cls_or_tup)
return result
def global_names_dict_generator(evaluator, scope, position):
"""
For global name lookups. Yields tuples of (names_dict, position). If the
position is None, the position does not matter anymore in that scope.
This function is used to include names from outer scopes. For example, when
the current scope is function:
>>> from jedi._compatibility import u, no_unicode_pprint
>>> from jedi.parser import ParserWithRecovery, load_grammar
>>> parser = ParserWithRecovery(load_grammar(), u('''
... x = ['a', 'b', 'c']
... def func():
... y = None
... '''))
>>> scope = parser.module.subscopes[0]
>>> scope
<Function: func@3-5>
`global_names_dict_generator` is a generator. First it yields names from
most inner scope.
>>> from jedi.evaluate import Evaluator
>>> evaluator = Evaluator(load_grammar())
>>> scope = evaluator.wrap(scope)
>>> pairs = list(global_names_dict_generator(evaluator, scope, (4, 0)))
>>> no_unicode_pprint(pairs[0])
({'func': [], 'y': [<Name: y@4,4>]}, (4, 0))
Then it yields the names from one level "lower". In this example, this
is the most outer scope. As you can see, the position in the tuple is now
None, because typically the whole module is loaded before the function is
called.
>>> no_unicode_pprint(pairs[1])
({'func': [<Name: func@3,4>], 'x': [<Name: x@2,0>]}, None)
After that we have a few underscore names that are part of the module.
>>> sorted(pairs[2][0].keys())
['__doc__', '__file__', '__name__', '__package__']
>>> pairs[3] # global names -> there are none in our example.
({}, None)
>>> pairs[4] # package modules -> Also none.
({}, None)
Finally, it yields names from builtin, if `include_builtin` is
true (default).
>>> pairs[5][0].values() #doctest: +ELLIPSIS
[[<CompiledName: ...>], ...]
"""
in_func = False
while scope is not None:
if not (scope.type == 'classdef' and in_func):
# Names in methods cannot be resolved within the class.
for names_dict in scope.names_dicts(True):
yield names_dict, position
if hasattr(scope, 'resets_positions'):
# TODO This is so ugly, seriously. However there's
# currently no good way of influencing
# global_names_dict_generator when it comes to certain
# objects.
position = None
if scope.type == 'funcdef':
# The position should be reset if the current scope is a function.
in_func = True
position = None
scope = evaluator.wrap(scope.get_parent_scope())
# Add builtins to the global scope.
for names_dict in evaluator.BUILTINS.names_dicts(True):
yield names_dict, None
def check_tuple_assignments(evaluator, types, name):
"""
Checks if tuples are assigned.
"""
for index, node in name.assignment_indexes():
iterated = iterable.py__iter__(evaluator, types, node)
for _ in range(index + 1):
try:
types = next(iterated)
except StopIteration:
# We could do this with the default param in next. But this
# would allow this loop to run for a very long time if the
# index number is high. Therefore break if the loop is
# finished.
types = set()
break
return types
def filter_private_variable(scope, origin_node):
"""Check if a variable is defined inside the same class or outside."""
instance = scope.get_parent_scope()
coming_from = origin_node
while coming_from is not None \
and not isinstance(coming_from, (tree.Class, compiled.CompiledObject)):
coming_from = coming_from.get_parent_scope()
# CompiledObjects don't have double underscore attributes, but Jedi abuses
# those for fakes (builtins.pym -> list).
if isinstance(instance, compiled.CompiledObject):
return instance != coming_from
else:
return isinstance(instance, er.Instance) and instance.base.base != coming_from

View file

@ -0,0 +1,91 @@
from jedi.parser import tree
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 break_check(evaluator, base_scope, stmt, origin_scope=None):
element_scope = evaluator.wrap(stmt.get_parent_scope(include_flows=True))
# 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 element_scope == origin_scope:
return REACHABLE
origin_scope = origin_scope.parent
x = _break_check(evaluator, stmt, base_scope, element_scope)
return x
def _break_check(evaluator, stmt, base_scope, element_scope):
element_scope = evaluator.wrap(element_scope)
base_scope = evaluator.wrap(base_scope)
reachable = REACHABLE
if isinstance(element_scope, tree.IfStmt):
if element_scope.node_after_else(stmt):
for check_node in element_scope.check_nodes():
reachable = _check_if(evaluator, check_node)
if reachable in (REACHABLE, UNSURE):
break
reachable = reachable.invert()
else:
node = element_scope.node_in_which_check_node(stmt)
if node is not None:
reachable = _check_if(evaluator, node)
elif isinstance(element_scope, (tree.TryStmt, tree.WhileStmt)):
return UNSURE
# Only reachable branches need to be examined further.
if reachable in (UNREACHABLE, UNSURE):
return reachable
if element_scope.type == 'file_input':
# The definition is in another module and therefore just return what we
# have generated.
return reachable
if base_scope != element_scope and base_scope != element_scope.parent:
return reachable & _break_check(evaluator, stmt, base_scope, element_scope.parent)
else:
return reachable
def _check_if(evaluator, node):
types = evaluator.eval_element(node)
values = set(x.py__bool__() for x in types)
if len(values) == 1:
return Status.lookup_table[values.pop()]
else:
return UNSURE

View file

@ -0,0 +1,200 @@
import copy
from itertools import chain
from jedi.parser import tree
def deep_ast_copy(obj, parent=None, new_elements=None):
"""
Much, much faster than copy.deepcopy, but just for Parser elements (Doesn't
copy parents).
"""
if new_elements is None:
new_elements = {}
def copy_node(obj):
# If it's already in the cache, just return it.
try:
return new_elements[obj]
except KeyError:
# Actually copy and set attributes.
new_obj = copy.copy(obj)
new_elements[obj] = new_obj
# Copy children
new_children = []
for child in obj.children:
typ = child.type
if typ in ('newline', 'operator', 'keyword', 'number', 'string',
'indent', 'dedent', 'endmarker', 'error_leaf'):
# At the moment we're not actually copying those primitive
# elements, because there's really no need to. The parents are
# obviously wrong, but that's not an issue.
new_child = child
elif typ == 'name':
new_elements[child] = new_child = copy.copy(child)
new_child.parent = new_obj
else: # Is a BaseNode.
new_child = copy_node(child)
new_child.parent = new_obj
new_children.append(new_child)
new_obj.children = new_children
# Copy the names_dict (if there is one).
try:
names_dict = obj.names_dict
except AttributeError:
pass
else:
try:
new_obj.names_dict = new_names_dict = {}
except AttributeError: # Impossible to set CompFor.names_dict
pass
else:
for string, names in names_dict.items():
new_names_dict[string] = [new_elements[n] for n in names]
return new_obj
if isinstance(obj, tree.BaseNode):
new_obj = copy_node(obj)
else:
# Special case of a Name object.
new_elements[obj] = new_obj = copy.copy(obj)
if parent is not None:
new_obj.parent = parent
return new_obj
def call_of_leaf(leaf, cut_own_trailer=False):
"""
Creates a "call" node that consist of all ``trailer`` and ``power``
objects. E.g. if you call it with ``append``::
list([]).append(3) or None
You would get a node with the content ``list([]).append`` back.
This generates a copy of the original ast node.
If you're using the leaf, e.g. the bracket `)` it will return ``list([])``.
# TODO remove cut_own_trailer option, since its always used with it. Just
# ignore it, It's not what we want anyway. Or document it better?
"""
trailer = leaf.parent
# The leaf may not be the last or first child, because there exist three
# different trailers: `( x )`, `[ x ]` and `.x`. In the first two examples
# we should not match anything more than x.
if trailer.type != 'trailer' or leaf not in (trailer.children[0], trailer.children[-1]):
if trailer.type == 'atom':
return trailer
return leaf
power = trailer.parent
index = power.children.index(trailer)
power = deep_ast_copy(power)
if cut_own_trailer:
cut = index
else:
cut = index + 1
power.children[cut:] = []
if power.type == 'error_node':
start = index
while True:
start -= 1
if power.children[start].type != 'trailer':
break
transformed = tree.Node('power', power.children[start:])
transformed.parent = power.parent
return transformed
return power
def get_names_of_node(node):
try:
children = node.children
except AttributeError:
if node.type == 'name':
return [node]
else:
return []
else:
return list(chain.from_iterable(get_names_of_node(c) for c in children))
def get_module_names(module, all_scopes):
"""
Returns a dictionary with name parts as keys and their call paths as
values.
"""
if all_scopes:
dct = module.used_names
else:
dct = module.names_dict
return chain.from_iterable(dct.values())
class FakeImport(tree.ImportName):
def __init__(self, name, parent, level=0):
super(FakeImport, self).__init__([])
self.parent = parent
self._level = level
self.name = name
def get_defined_names(self):
return [self.name]
def aliases(self):
return {}
@property
def level(self):
return self._level
@property
def start_pos(self):
return 0, 0
def paths(self):
return [[self.name]]
def is_definition(self):
return True
class FakeName(tree.Name):
def __init__(self, name_str, parent=None, start_pos=(0, 0), is_definition=None):
"""
In case is_definition is defined (not None), that bool value will be
returned.
"""
super(FakeName, self).__init__(tree.zero_position_modifier, name_str, start_pos)
self.parent = parent
self._is_definition = is_definition
def get_definition(self):
return self.parent
def is_definition(self):
if self._is_definition is None:
return super(FakeName, self).is_definition()
else:
return self._is_definition
class LazyName(FakeName):
def __init__(self, name, parent_callback, is_definition=None):
super(LazyName, self).__init__(name, is_definition=is_definition)
self._parent_callback = parent_callback
@property
def parent(self):
return self._parent_callback()
@parent.setter
def parent(self, value):
pass # Do nothing, super classes can try to set the parent.

View file

@ -0,0 +1,517 @@
"""
:mod:`jedi.evaluate.imports` is here to resolve import statements and return
the modules/classes/functions/whatever, which they stand for. However there's
not any actual importing done. This module is about finding modules in the
filesystem. This can be quite tricky sometimes, because Python imports are not
always that simple.
This module uses imp for python up to 3.2 and importlib for python 3.3 on; the
correct implementation is delegated to _compatibility.
This module also supports import autocompletion, which means to complete
statements like ``from datetim`` (curser at the end would return ``datetime``).
"""
import imp
import os
import pkgutil
import sys
from itertools import chain
from jedi._compatibility import find_module, unicode
from jedi import common
from jedi import debug
from jedi.parser import fast
from jedi.parser import tree
from jedi.parser.utils import save_parser, load_parser, parser_cache
from jedi.evaluate import sys_path
from jedi.evaluate import helpers
from jedi import settings
from jedi.common import source_to_unicode
from jedi.evaluate import compiled
from jedi.evaluate import analysis
from jedi.evaluate.cache import memoize_default, NO_DEFAULT
def completion_names(evaluator, imp, pos):
name = imp.name_for_position(pos)
module = evaluator.wrap(imp.get_parent_until())
if name is None:
level = 0
for node in imp.children:
if node.end_pos <= pos:
if node in ('.', '...'):
level += len(node.value)
import_path = []
else:
# Completion on an existing name.
# The import path needs to be reduced by one, because we're completing.
import_path = imp.path_for_name(name)[:-1]
level = imp.level
importer = Importer(evaluator, tuple(import_path), module, level)
if isinstance(imp, tree.ImportFrom):
c = imp.children
only_modules = c[c.index('import')].start_pos >= pos
else:
only_modules = True
return importer.completion_names(evaluator, only_modules)
class ImportWrapper(tree.Base):
def __init__(self, evaluator, name):
self._evaluator = evaluator
self._name = name
self._import = name.get_parent_until(tree.Import)
self.import_path = self._import.path_for_name(name)
@memoize_default()
def follow(self, is_goto=False):
module = self._evaluator.wrap(self._import.get_parent_until())
import_path = self._import.path_for_name(self._name)
from_import_name = None
try:
from_names = self._import.get_from_names()
except AttributeError:
# Is an import_name
pass
else:
if len(from_names) + 1 == len(import_path):
# We have to fetch the from_names part first and then check
# if from_names exists in the modules.
from_import_name = import_path[-1]
import_path = from_names
importer = Importer(self._evaluator, tuple(import_path),
module, self._import.level)
types = importer.follow()
#if self._import.is_nested() and not self.nested_resolve:
# scopes = [NestedImportModule(module, self._import)]
if from_import_name is not None:
types = set(chain.from_iterable(
self._evaluator.find_types(t, unicode(from_import_name),
is_goto=is_goto)
for t in types))
if not types:
path = import_path + [from_import_name]
importer = Importer(self._evaluator, tuple(path),
module, self._import.level)
types = importer.follow()
# goto only accepts `Name`
if is_goto:
types = set(s.name for s in types)
else:
# goto only accepts `Name`
if is_goto:
types = set(s.name for s in types)
debug.dbg('after import: %s', types)
return types
class NestedImportModule(tree.Module):
"""
TODO while there's no use case for nested import module right now, we might
be able to use them for static analysis checks later on.
"""
def __init__(self, module, nested_import):
self._module = module
self._nested_import = nested_import
def _get_nested_import_name(self):
"""
Generates an Import statement, that can be used to fake nested imports.
"""
i = self._nested_import
# This is not an existing Import statement. Therefore, set position to
# 0 (0 is not a valid line number).
zero = (0, 0)
names = [unicode(name) for name in i.namespace_names[1:]]
name = helpers.FakeName(names, self._nested_import)
new = tree.Import(i._sub_module, zero, zero, name)
new.parent = self._module
debug.dbg('Generated a nested import: %s', new)
return helpers.FakeName(str(i.namespace_names[1]), new)
def __getattr__(self, name):
return getattr(self._module, name)
def __repr__(self):
return "<%s: %s of %s>" % (self.__class__.__name__, self._module,
self._nested_import)
def _add_error(evaluator, name, message=None):
if hasattr(name, 'parent'):
# Should be a name, not a string!
analysis.add(evaluator, 'import-error', name, message)
def get_init_path(directory_path):
"""
The __init__ file can be searched in a directory. If found return it, else
None.
"""
for suffix, _, _ in imp.get_suffixes():
path = os.path.join(directory_path, '__init__' + suffix)
if os.path.exists(path):
return path
return None
class Importer(object):
def __init__(self, evaluator, import_path, module, level=0):
"""
An implementation similar to ``__import__``. Use `follow`
to actually follow the imports.
*level* specifies whether to use absolute or relative imports. 0 (the
default) means only perform absolute imports. Positive values for level
indicate the number of parent directories to search relative to the
directory of the module calling ``__import__()`` (see PEP 328 for the
details).
:param import_path: List of namespaces (strings or Names).
"""
debug.speed('import %s' % (import_path,))
self._evaluator = evaluator
self.level = level
self.module = module
try:
self.file_path = module.py__file__()
except AttributeError:
# Can be None for certain compiled modules like 'builtins'.
self.file_path = None
if level:
base = module.py__package__().split('.')
if base == ['']:
base = []
if level > len(base):
path = module.py__file__()
if path is not None:
import_path = list(import_path)
for i in range(level):
path = os.path.dirname(path)
dir_name = os.path.basename(path)
# This is not the proper way to do relative imports. However, since
# Jedi cannot be sure about the entry point, we just calculate an
# absolute path here.
if dir_name:
import_path.insert(0, dir_name)
else:
_add_error(self._evaluator, import_path[-1])
import_path = []
# TODO add import error.
debug.warning('Attempted relative import beyond top-level package.')
else:
# Here we basically rewrite the level to 0.
import_path = tuple(base) + tuple(import_path)
self.import_path = import_path
@property
def str_import_path(self):
"""Returns the import path as pure strings instead of `Name`."""
return tuple(str(name) for name in self.import_path)
@memoize_default()
def sys_path_with_modifications(self):
in_path = []
sys_path_mod = list(sys_path.sys_path_with_modifications(self._evaluator, self.module))
if self.file_path is not None:
# If you edit e.g. gunicorn, there will be imports like this:
# `from gunicorn import something`. But gunicorn is not in the
# sys.path. Therefore look if gunicorn is a parent directory, #56.
if self.import_path: # TODO is this check really needed?
for path in sys_path.traverse_parents(self.file_path):
if os.path.basename(path) == self.str_import_path[0]:
in_path.append(os.path.dirname(path))
# Since we know nothing about the call location of the sys.path,
# it's a possibility that the current directory is the origin of
# the Python execution.
sys_path_mod.insert(0, os.path.dirname(self.file_path))
return in_path + sys_path_mod
@memoize_default(NO_DEFAULT)
def follow(self):
if not self.import_path:
return set()
return self._do_import(self.import_path, self.sys_path_with_modifications())
def _do_import(self, import_path, sys_path):
"""
This method is very similar to importlib's `_gcd_import`.
"""
import_parts = [str(i) for i in import_path]
# Handle "magic" Flask extension imports:
# ``flask.ext.foo`` is really ``flask_foo`` or ``flaskext.foo``.
if len(import_path) > 2 and import_parts[:2] == ['flask', 'ext']:
# New style.
ipath = ('flask_' + str(import_parts[2]),) + import_path[3:]
modules = self._do_import(ipath, sys_path)
if modules:
return modules
else:
# Old style
return self._do_import(('flaskext',) + import_path[2:], sys_path)
module_name = '.'.join(import_parts)
try:
return set([self._evaluator.modules[module_name]])
except KeyError:
pass
if len(import_path) > 1:
# This is a recursive way of importing that works great with
# the module cache.
bases = self._do_import(import_path[:-1], sys_path)
if not bases:
return set()
# We can take the first element, because only the os special
# case yields multiple modules, which is not important for
# further imports.
parent_module = list(bases)[0]
# This is a huge exception, we follow a nested import
# ``os.path``, because it's a very important one in Python
# that is being achieved by messing with ``sys.modules`` in
# ``os``.
if [str(i) for i in import_path] == ['os', 'path']:
return self._evaluator.find_types(parent_module, 'path')
try:
paths = parent_module.py__path__()
except AttributeError:
# The module is not a package.
_add_error(self._evaluator, import_path[-1])
return set()
else:
debug.dbg('search_module %s in paths %s', module_name, paths)
for path in paths:
# At the moment we are only using one path. So this is
# not important to be correct.
try:
module_file, module_path, is_pkg = \
find_module(import_parts[-1], [path])
break
except ImportError:
module_path = None
if module_path is None:
_add_error(self._evaluator, import_path[-1])
return set()
else:
parent_module = None
try:
debug.dbg('search_module %s in %s', import_parts[-1], self.file_path)
# Override the sys.path. It works only good that way.
# Injecting the path directly into `find_module` did not work.
sys.path, temp = sys_path, sys.path
try:
module_file, module_path, is_pkg = \
find_module(import_parts[-1])
finally:
sys.path = temp
except ImportError:
# The module is not a package.
_add_error(self._evaluator, import_path[-1])
return set()
source = None
if is_pkg:
# In this case, we don't have a file yet. Search for the
# __init__ file.
if module_path.endswith(('.zip', '.egg')):
source = module_file.loader.get_source(module_name)
else:
module_path = get_init_path(module_path)
elif module_file:
source = module_file.read()
module_file.close()
if module_file is None and not module_path.endswith(('.py', '.zip', '.egg')):
module = compiled.load_module(self._evaluator, module_path)
else:
module = _load_module(self._evaluator, module_path, source, sys_path, parent_module)
if module is None:
# The file might raise an ImportError e.g. and therefore not be
# importable.
return set()
self._evaluator.modules[module_name] = module
return set([module])
def _generate_name(self, name):
# Create a pseudo import to be able to follow them.
name = helpers.FakeName(name)
imp = helpers.FakeImport(name, parent=self.module)
name.parent = imp
return name
def _get_module_names(self, search_path=None):
"""
Get the names of all modules in the search_path. This means file names
and not names defined in the files.
"""
names = []
# add builtin module names
if search_path is None:
names += [self._generate_name(name) for name in sys.builtin_module_names]
if search_path is None:
search_path = self.sys_path_with_modifications()
for module_loader, name, is_pkg in pkgutil.iter_modules(search_path):
names.append(self._generate_name(name))
return names
def completion_names(self, evaluator, only_modules=False):
"""
:param only_modules: Indicates wheter it's possible to import a
definition that is not defined in a module.
"""
from jedi.evaluate import finder
names = []
if self.import_path:
# flask
if self.str_import_path == ('flask', 'ext'):
# List Flask extensions like ``flask_foo``
for mod in self._get_module_names():
modname = str(mod)
if modname.startswith('flask_'):
extname = modname[len('flask_'):]
names.append(self._generate_name(extname))
# Now the old style: ``flaskext.foo``
for dir in self.sys_path_with_modifications():
flaskext = os.path.join(dir, 'flaskext')
if os.path.isdir(flaskext):
names += self._get_module_names([flaskext])
for scope in self.follow():
# Non-modules are not completable.
if not scope.type == 'file_input': # not a module
continue
# namespace packages
if isinstance(scope, tree.Module) and scope.path.endswith('__init__.py'):
paths = scope.py__path__()
names += self._get_module_names(paths)
if only_modules:
# In the case of an import like `from x.` we don't need to
# add all the variables.
if ('os',) == self.str_import_path and not self.level:
# os.path is a hardcoded exception, because it's a
# ``sys.modules`` modification.
names.append(self._generate_name('path'))
continue
for names_dict in scope.names_dicts(search_global=False):
_names = list(chain.from_iterable(names_dict.values()))
if not _names:
continue
_names = finder.filter_definition_names(_names, scope)
names += _names
else:
# Empty import path=completion after import
if not self.level:
names += self._get_module_names()
if self.file_path is not None:
path = os.path.abspath(self.file_path)
for i in range(self.level - 1):
path = os.path.dirname(path)
names += self._get_module_names([path])
return names
def _load_module(evaluator, path=None, source=None, sys_path=None, parent_module=None):
def load(source):
dotted_path = path and compiled.dotted_from_fs_path(path, sys_path)
if path is not None and path.endswith(('.py', '.zip', '.egg')) \
and dotted_path not in settings.auto_import_modules:
if source is None:
with open(path, 'rb') as f:
source = f.read()
else:
return compiled.load_module(evaluator, path)
p = path
p = fast.FastParser(evaluator.grammar, common.source_to_unicode(source), p)
save_parser(path, p)
from jedi.evaluate.representation import ModuleWrapper
return ModuleWrapper(evaluator, p.module, parent_module)
if sys_path is None:
sys_path = evaluator.sys_path
cached = load_parser(path)
module = load(source) if cached is None else cached.module
module = evaluator.wrap(module)
return module
def add_module(evaluator, module_name, module):
if '.' not in module_name:
# We cannot add paths with dots, because that would collide with
# the sepatator dots for nested packages. Therefore we return
# `__main__` in ModuleWrapper.py__name__(), which is similar to
# Python behavior.
evaluator.modules[module_name] = module
def get_modules_containing_name(evaluator, mods, name):
"""
Search a name in the directories of modules.
"""
def check_python_file(path):
try:
return parser_cache[path].parser.module
except KeyError:
try:
return check_fs(path)
except IOError:
return None
def check_fs(path):
with open(path, 'rb') as f:
source = source_to_unicode(f.read())
if name in source:
module_name = os.path.basename(path)[:-3] # Remove `.py`.
module = _load_module(evaluator, path, source)
add_module(evaluator, module_name, module)
return module
# skip non python modules
mods = set(m for m in mods if not isinstance(m, compiled.CompiledObject))
mod_paths = set()
for m in mods:
mod_paths.add(m.path)
yield m
if settings.dynamic_params_for_other_modules:
paths = set(settings.additional_dynamic_modules)
for p in mod_paths:
if p is not None:
# We need abspath, because the seetings paths might not already
# have been converted to absolute paths.
d = os.path.dirname(os.path.abspath(p))
for entry in os.listdir(d):
if entry not in mod_paths:
if entry.endswith('.py'):
paths.add(d + os.path.sep + entry)
for p in sorted(paths):
# make testing easier, sort it - same results on every interpreter
c = check_python_file(p)
if c is not None and c not in mods and not isinstance(c, compiled.CompiledObject):
yield c

View file

@ -0,0 +1,863 @@
"""
Contains all classes and functions to deal with lists, dicts, generators and
iterators in general.
Array modifications
*******************
If the content of an array (``set``/``list``) is requested somewhere, the
current module will be checked for appearances of ``arr.append``,
``arr.insert``, etc. If the ``arr`` name points to an actual array, the
content will be added
This can be really cpu intensive, as you can imagine. Because |jedi| has to
follow **every** ``append`` and check wheter it's the right array. However this
works pretty good, because in *slow* cases, the recursion detector and other
settings will stop this process.
It is important to note that:
1. Array modfications work only in the current module.
2. Jedi only checks Array additions; ``list.pop``, etc are ignored.
"""
from jedi.common import unite, safe_property
from jedi import debug
from jedi import settings
from jedi._compatibility import use_metaclass, unicode, zip_longest
from jedi.parser import tree
from jedi.evaluate import compiled
from jedi.evaluate import helpers
from jedi.evaluate.cache import CachedMetaClass, memoize_default
from jedi.evaluate import analysis
from jedi.evaluate import pep0484
from jedi import common
class IterableWrapper(tree.Base):
def is_class(self):
return False
@memoize_default()
def _get_names_dict(self, names_dict):
builtin_methods = {}
for cls in reversed(type(self).mro()):
try:
builtin_methods.update(cls.builtin_methods)
except AttributeError:
pass
if not builtin_methods:
return names_dict
dct = {}
for names in names_dict.values():
for name in names:
name_str = name.value
try:
method = builtin_methods[name_str, self.type]
except KeyError:
dct[name_str] = [name]
else:
parent = BuiltinMethod(self, method, name.parent)
dct[name_str] = [helpers.FakeName(name_str, parent, is_definition=True)]
return dct
class BuiltinMethod(IterableWrapper):
"""``Generator.__next__`` ``dict.values`` methods and so on."""
def __init__(self, builtin, method, builtin_func):
self._builtin = builtin
self._method = method
self._builtin_func = builtin_func
def py__call__(self, params):
return self._method(self._builtin)
def __getattr__(self, name):
return getattr(self._builtin_func, name)
def has_builtin_methods(cls):
cls.builtin_methods = {}
for func in cls.__dict__.values():
try:
cls.builtin_methods.update(func.registered_builtin_methods)
except AttributeError:
pass
return cls
def register_builtin_method(method_name, type=None):
def wrapper(func):
dct = func.__dict__.setdefault('registered_builtin_methods', {})
dct[method_name, type] = func
return func
return wrapper
@has_builtin_methods
class GeneratorMixin(object):
type = None
@register_builtin_method('send')
@register_builtin_method('next')
@register_builtin_method('__next__')
def py__next__(self):
# TODO add TypeError if params are given.
return unite(self.py__iter__())
@memoize_default()
def names_dicts(self, search_global=False): # is always False
gen_obj = compiled.get_special_object(self._evaluator, 'GENERATOR_OBJECT')
yield self._get_names_dict(gen_obj.names_dict)
def py__bool__(self):
return True
def py__class__(self):
gen_obj = compiled.get_special_object(self._evaluator, 'GENERATOR_OBJECT')
return gen_obj.py__class__()
class Generator(use_metaclass(CachedMetaClass, IterableWrapper, GeneratorMixin)):
"""Handling of `yield` functions."""
def __init__(self, evaluator, func, var_args):
super(Generator, self).__init__()
self._evaluator = evaluator
self.func = func
self.var_args = var_args
def py__iter__(self):
from jedi.evaluate.representation import FunctionExecution
f = FunctionExecution(self._evaluator, self.func, self.var_args)
return f.get_yield_types()
def __getattr__(self, name):
if name not in ['start_pos', 'end_pos', 'parent', 'get_imports',
'doc', 'docstr', 'get_parent_until',
'get_code', 'subscopes']:
raise AttributeError("Accessing %s of %s is not allowed."
% (self, name))
return getattr(self.func, name)
def __repr__(self):
return "<%s of %s>" % (type(self).__name__, self.func)
class Comprehension(IterableWrapper):
@staticmethod
def from_atom(evaluator, atom):
bracket = atom.children[0]
if bracket == '{':
if atom.children[1].children[1] == ':':
cls = DictComprehension
else:
cls = SetComprehension
elif bracket == '(':
cls = GeneratorComprehension
elif bracket == '[':
cls = ListComprehension
return cls(evaluator, atom)
def __init__(self, evaluator, atom):
self._evaluator = evaluator
self._atom = atom
def _get_comprehension(self):
# The atom contains a testlist_comp
return self._atom.children[1]
def _get_comp_for(self):
# The atom contains a testlist_comp
return self._get_comprehension().children[1]
@memoize_default()
def _eval_node(self, index=0):
"""
The first part `x + 1` of the list comprehension:
[x + 1 for x in foo]
"""
comp_for = self._get_comp_for()
# For nested comprehensions we need to search the last one.
from jedi.evaluate.representation import InstanceElement
node = self._get_comprehension().children[index]
if isinstance(node, InstanceElement):
# This seems to be a strange case that I haven't found a way to
# write tests against. However since it's my new goal to get rid of
# InstanceElement anyway, I don't care.
node = node.var
last_comp = list(comp_for.get_comp_fors())[-1]
return helpers.deep_ast_copy(node, parent=last_comp)
def _nested(self, comp_fors):
evaluator = self._evaluator
comp_for = comp_fors[0]
input_node = comp_for.children[3]
input_types = evaluator.eval_element(input_node)
iterated = py__iter__(evaluator, input_types, input_node)
exprlist = comp_for.children[1]
for i, types in enumerate(iterated):
evaluator.predefined_if_name_dict_dict[comp_for] = \
unpack_tuple_to_dict(evaluator, types, exprlist)
try:
for result in self._nested(comp_fors[1:]):
yield result
except IndexError:
iterated = evaluator.eval_element(self._eval_node())
if self.type == 'dict':
yield iterated, evaluator.eval_element(self._eval_node(2))
else:
yield iterated
finally:
del evaluator.predefined_if_name_dict_dict[comp_for]
@memoize_default(default=[])
@common.to_list
def _iterate(self):
comp_fors = tuple(self._get_comp_for().get_comp_fors())
for result in self._nested(comp_fors):
yield result
def py__iter__(self):
return self._iterate()
def __repr__(self):
return "<%s of %s>" % (type(self).__name__, self._atom)
@has_builtin_methods
class ArrayMixin(object):
@memoize_default()
def names_dicts(self, search_global=False): # Always False.
# `array.type` is a string with the type, e.g. 'list'.
scope = compiled.builtin_from_name(self._evaluator, self.type)
# builtins only have one class -> [0]
scopes = self._evaluator.execute_evaluated(scope, self)
names_dicts = list(scopes)[0].names_dicts(search_global)
#yield names_dicts[0]
yield self._get_names_dict(names_dicts[1])
def py__bool__(self):
return None # We don't know the length, because of appends.
def py__class__(self):
return compiled.builtin_from_name(self._evaluator, self.type)
@safe_property
def parent(self):
return self._evaluator.BUILTINS
@property
def name(self):
return FakeSequence(self._evaluator, [], self.type).name
@memoize_default()
def dict_values(self):
return unite(self._evaluator.eval_element(v) for k, v in self._items())
@register_builtin_method('values', type='dict')
def _imitate_values(self):
items = self.dict_values()
return create_evaluated_sequence_set(self._evaluator, items, sequence_type='list')
#return set([FakeSequence(self._evaluator, [AlreadyEvaluated(items)], 'tuple')])
@register_builtin_method('items', type='dict')
def _imitate_items(self):
items = [set([FakeSequence(self._evaluator, (k, v), 'tuple')])
for k, v in self._items()]
return create_evaluated_sequence_set(self._evaluator, *items, sequence_type='list')
class ListComprehension(Comprehension, ArrayMixin):
type = 'list'
def py__getitem__(self, index):
all_types = list(self.py__iter__())
result = all_types[index]
if isinstance(index, slice):
return create_evaluated_sequence_set(
self._evaluator,
unite(result),
sequence_type='list'
)
return result
class SetComprehension(Comprehension, ArrayMixin):
type = 'set'
@has_builtin_methods
class DictComprehension(Comprehension, ArrayMixin):
type = 'dict'
def _get_comp_for(self):
return self._get_comprehension().children[3]
def py__iter__(self):
for keys, values in self._iterate():
yield keys
def py__getitem__(self, index):
for keys, values in self._iterate():
for k in keys:
if isinstance(k, compiled.CompiledObject):
if k.obj == index:
return values
return self.dict_values()
def dict_values(self):
return unite(values for keys, values in self._iterate())
@register_builtin_method('items', type='dict')
def _imitate_items(self):
items = set(FakeSequence(self._evaluator,
(AlreadyEvaluated(keys), AlreadyEvaluated(values)), 'tuple')
for keys, values in self._iterate())
return create_evaluated_sequence_set(self._evaluator, items, sequence_type='list')
class GeneratorComprehension(Comprehension, GeneratorMixin):
pass
class Array(IterableWrapper, ArrayMixin):
mapping = {'(': 'tuple',
'[': 'list',
'{': 'dict'}
def __init__(self, evaluator, atom):
self._evaluator = evaluator
self.atom = atom
self.type = Array.mapping[atom.children[0]]
"""The builtin name of the array (list, set, tuple or dict)."""
c = self.atom.children
array_node = c[1]
if self.type == 'dict' and array_node != '}' \
and (not hasattr(array_node, 'children')
or ':' not in array_node.children):
self.type = 'set'
@property
def name(self):
return helpers.FakeName(self.type, parent=self)
def py__getitem__(self, index):
"""Here the index is an int/str. Raises IndexError/KeyError."""
if self.type == 'dict':
for key, value in self._items():
for k in self._evaluator.eval_element(key):
if isinstance(k, compiled.CompiledObject) \
and index == k.obj:
return self._evaluator.eval_element(value)
raise KeyError('No key found in dictionary %s.' % self)
# Can raise an IndexError
if isinstance(index, slice):
return set([self])
else:
return self._evaluator.eval_element(self._items()[index])
def __getattr__(self, name):
if name not in ['start_pos', 'get_only_subelement', 'parent',
'get_parent_until', 'items']:
raise AttributeError('Strange access on %s: %s.' % (self, name))
return getattr(self.atom, name)
# @memoize_default()
def py__iter__(self):
"""
While values returns the possible values for any array field, this
function returns the value for a certain index.
"""
if self.type == 'dict':
# Get keys.
types = set()
for k, _ in self._items():
types |= self._evaluator.eval_element(k)
# We don't know which dict index comes first, therefore always
# yield all the types.
for _ in types:
yield types
else:
for value in self._items():
yield self._evaluator.eval_element(value)
additions = check_array_additions(self._evaluator, self)
if additions:
yield additions
def _values(self):
"""Returns a list of a list of node."""
if self.type == 'dict':
return unite(v for k, v in self._items())
else:
return self._items()
def _items(self):
c = self.atom.children
array_node = c[1]
if array_node in (']', '}', ')'):
return [] # Direct closing bracket, doesn't contain items.
if tree.is_node(array_node, 'testlist_comp'):
return array_node.children[::2]
elif tree.is_node(array_node, 'dictorsetmaker'):
kv = []
iterator = iter(array_node.children)
for key in iterator:
op = next(iterator, None)
if op is None or op == ',':
kv.append(key) # A set.
else:
assert op == ':' # A dict.
kv.append((key, next(iterator)))
next(iterator, None) # Possible comma.
return kv
else:
return [array_node]
def __repr__(self):
return "<%s of %s>" % (type(self).__name__, self.atom)
class _FakeArray(Array):
def __init__(self, evaluator, container, type):
self.type = type
self._evaluator = evaluator
self.atom = container
class ImplicitTuple(_FakeArray):
def __init__(self, evaluator, testlist):
super(ImplicitTuple, self).__init__(evaluator, testlist, 'tuple')
self._testlist = testlist
def _items(self):
return self._testlist.children[::2]
class FakeSequence(_FakeArray):
def __init__(self, evaluator, sequence_values, type):
"""
type should be one of "tuple", "list"
"""
super(FakeSequence, self).__init__(evaluator, sequence_values, type)
self._sequence_values = sequence_values
def _items(self):
return self._sequence_values
def create_evaluated_sequence_set(evaluator, *types_order, **kwargs):
"""
``sequence_type`` is a named argument, that doesn't work in Python2. For backwards
compatibility reasons, we're now using kwargs.
"""
sequence_type = kwargs.pop('sequence_type')
assert not kwargs
sets = tuple(AlreadyEvaluated(types) for types in types_order)
return set([FakeSequence(evaluator, sets, sequence_type)])
class AlreadyEvaluated(frozenset):
"""A simple container to add already evaluated objects to an array."""
def get_code(self, normalized=False):
# For debugging purposes.
return str(self)
class MergedNodes(frozenset):
pass
class FakeDict(_FakeArray):
def __init__(self, evaluator, dct):
super(FakeDict, self).__init__(evaluator, dct, 'dict')
self._dct = dct
def py__iter__(self):
yield set(compiled.create(self._evaluator, key) for key in self._dct)
def py__getitem__(self, index):
return unite(self._evaluator.eval_element(v) for v in self._dct[index])
def _items(self):
for key, values in self._dct.items():
# TODO this is not proper. The values could be multiple values?!
yield key, values[0]
class MergedArray(_FakeArray):
def __init__(self, evaluator, arrays):
super(MergedArray, self).__init__(evaluator, arrays, arrays[-1].type)
self._arrays = arrays
def py__iter__(self):
for array in self._arrays:
for types in array.py__iter__():
yield types
def py__getitem__(self, index):
return unite(self.py__iter__())
def _items(self):
for array in self._arrays:
for a in array._items():
yield a
def __len__(self):
return sum(len(a) for a in self._arrays)
def unpack_tuple_to_dict(evaluator, types, exprlist):
"""
Unpacking tuple assignments in for statements and expr_stmts.
"""
if exprlist.type == 'name':
return {exprlist.value: types}
elif exprlist.type == 'atom' and exprlist.children[0] in '([':
return unpack_tuple_to_dict(evaluator, types, exprlist.children[1])
elif exprlist.type in ('testlist', 'testlist_comp', 'exprlist',
'testlist_star_expr'):
dct = {}
parts = iter(exprlist.children[::2])
n = 0
for iter_types in py__iter__(evaluator, types, exprlist):
n += 1
try:
part = next(parts)
except StopIteration:
analysis.add(evaluator, 'value-error-too-many-values', part,
message="ValueError: too many values to unpack (expected %s)" % n)
else:
dct.update(unpack_tuple_to_dict(evaluator, iter_types, part))
has_parts = next(parts, None)
if types and has_parts is not None:
analysis.add(evaluator, 'value-error-too-few-values', has_parts,
message="ValueError: need more than %s values to unpack" % n)
return dct
elif exprlist.type == 'power' or exprlist.type == 'atom_expr':
# Something like ``arr[x], var = ...``.
# This is something that is not yet supported, would also be difficult
# to write into a dict.
return {}
elif exprlist.type == 'star_expr': # `a, *b, c = x` type unpackings
# Currently we're not supporting them.
return {}
raise NotImplementedError
def py__iter__(evaluator, types, node=None):
debug.dbg('py__iter__')
type_iters = []
for typ in types:
try:
iter_method = typ.py__iter__
except AttributeError:
if node is not None:
analysis.add(evaluator, 'type-error-not-iterable', node,
message="TypeError: '%s' object is not iterable" % typ)
else:
type_iters.append(iter_method())
#for result in iter_method():
#yield result
for t in zip_longest(*type_iters, fillvalue=set()):
yield unite(t)
def py__iter__types(evaluator, types, node=None):
"""
Calls `py__iter__`, but ignores the ordering in the end and just returns
all types that it contains.
"""
return unite(py__iter__(evaluator, types, node))
def py__getitem__(evaluator, types, trailer):
from jedi.evaluate.representation import Class
result = set()
trailer_op, node, trailer_cl = trailer.children
assert trailer_op == "["
assert trailer_cl == "]"
# special case: PEP0484 typing module, see
# https://github.com/davidhalter/jedi/issues/663
for typ in list(types):
if isinstance(typ, Class):
typing_module_types = \
pep0484.get_types_for_typing_module(evaluator, typ, node)
if typing_module_types is not None:
types.remove(typ)
result |= typing_module_types
if not types:
# all consumed by special cases
return result
for index in create_index_types(evaluator, node):
if isinstance(index, (compiled.CompiledObject, Slice)):
index = index.obj
if type(index) not in (float, int, str, unicode, slice):
# If the index is not clearly defined, we have to get all the
# possiblities.
for typ in list(types):
if isinstance(typ, Array) and typ.type == 'dict':
types.remove(typ)
result |= typ.dict_values()
return result | py__iter__types(evaluator, types)
for typ in types:
# The actual getitem call.
try:
getitem = typ.py__getitem__
except AttributeError:
analysis.add(evaluator, 'type-error-not-subscriptable', trailer_op,
message="TypeError: '%s' object is not subscriptable" % typ)
else:
try:
result |= getitem(index)
except IndexError:
result |= py__iter__types(evaluator, set([typ]))
except KeyError:
# Must be a dict. Lists don't raise KeyErrors.
result |= typ.dict_values()
return result
def check_array_additions(evaluator, array):
""" Just a mapper function for the internal _check_array_additions """
if array.type not in ('list', 'set'):
# TODO also check for dict updates
return set()
is_list = array.type == 'list'
try:
current_module = array.atom.get_parent_until()
except AttributeError:
# If there's no get_parent_until, it's a FakeSequence or another Fake
# type. Those fake types are used inside Jedi's engine. No values may
# be added to those after their creation.
return set()
return _check_array_additions(evaluator, array, current_module, is_list)
@memoize_default(default=set(), evaluator_is_first_arg=True)
@debug.increase_indent
def _check_array_additions(evaluator, compare_array, module, is_list):
"""
Checks if a `Array` has "add" (append, insert, extend) statements:
>>> a = [""]
>>> a.append(1)
"""
debug.dbg('Dynamic array search for %s' % compare_array, color='MAGENTA')
if not settings.dynamic_array_additions or isinstance(module, compiled.CompiledObject):
debug.dbg('Dynamic array search aborted.', color='MAGENTA')
return set()
def check_additions(arglist, add_name):
params = list(param.Arguments(evaluator, arglist).unpack())
result = set()
if add_name in ['insert']:
params = params[1:]
if add_name in ['append', 'add', 'insert']:
for key, nodes in params:
result |= unite(evaluator.eval_element(node) for node in nodes)
elif add_name in ['extend', 'update']:
for key, nodes in params:
for node in nodes:
types = evaluator.eval_element(node)
result |= py__iter__types(evaluator, types, node)
return result
from jedi.evaluate import representation as er, param
def get_execution_parent(element):
""" Used to get an Instance/FunctionExecution parent """
if isinstance(element, Array):
node = element.atom
else:
# Is an Instance with an
# Arguments([AlreadyEvaluated([_ArrayInstance])]) inside
# Yeah... I know... It's complicated ;-)
node = list(element.var_args.argument_node[0])[0].var_args.trailer
if isinstance(node, er.InstanceElement) or node is None:
return node
return node.get_parent_until(er.FunctionExecution)
temp_param_add, settings.dynamic_params_for_other_modules = \
settings.dynamic_params_for_other_modules, False
search_names = ['append', 'extend', 'insert'] if is_list else ['add', 'update']
comp_arr_parent = get_execution_parent(compare_array)
added_types = set()
for add_name in search_names:
try:
possible_names = module.used_names[add_name]
except KeyError:
continue
else:
for name in possible_names:
# Check if the original scope is an execution. If it is, one
# can search for the same statement, that is in the module
# dict. Executions are somewhat special in jedi, since they
# literally copy the contents of a function.
if isinstance(comp_arr_parent, er.FunctionExecution):
if comp_arr_parent.start_pos < name.start_pos < comp_arr_parent.end_pos:
name = comp_arr_parent.name_for_position(name.start_pos)
else:
# Don't check definitions that are not defined in the
# same function. This is not "proper" anyway. It also
# improves Jedi's speed for array lookups, since we
# don't have to check the whole source tree anymore.
continue
trailer = name.parent
power = trailer.parent
trailer_pos = power.children.index(trailer)
try:
execution_trailer = power.children[trailer_pos + 1]
except IndexError:
continue
else:
if execution_trailer.type != 'trailer' \
or execution_trailer.children[0] != '(' \
or execution_trailer.children[1] == ')':
continue
power = helpers.call_of_leaf(name, cut_own_trailer=True)
# InstanceElements are special, because they don't get copied,
# but have this wrapper around them.
if isinstance(comp_arr_parent, er.InstanceElement):
power = er.get_instance_el(evaluator, comp_arr_parent.instance, power)
if evaluator.recursion_detector.push_stmt(power):
# Check for recursion. Possible by using 'extend' in
# combination with function calls.
continue
try:
if compare_array in evaluator.eval_element(power):
# The arrays match. Now add the results
added_types |= check_additions(execution_trailer.children[1], add_name)
finally:
evaluator.recursion_detector.pop_stmt()
# reset settings
settings.dynamic_params_for_other_modules = temp_param_add
debug.dbg('Dynamic array result %s' % added_types, color='MAGENTA')
return added_types
def check_array_instances(evaluator, instance):
"""Used for set() and list() instances."""
if not settings.dynamic_array_additions:
return instance.var_args
ai = _ArrayInstance(evaluator, instance)
from jedi.evaluate import param
return param.Arguments(evaluator, [AlreadyEvaluated([ai])])
class _ArrayInstance(IterableWrapper):
"""
Used for the usage of set() and list().
This is definitely a hack, but a good one :-)
It makes it possible to use set/list conversions.
In contrast to Array, ListComprehension and all other iterable types, this
is something that is only used inside `evaluate/compiled/fake/builtins.py`
and therefore doesn't need `names_dicts`, `py__bool__` and so on, because
we don't use these operations in `builtins.py`.
"""
def __init__(self, evaluator, instance):
self._evaluator = evaluator
self.instance = instance
self.var_args = instance.var_args
def py__iter__(self):
try:
_, first_nodes = next(self.var_args.unpack())
except StopIteration:
types = set()
else:
types = unite(self._evaluator.eval_element(node) for node in first_nodes)
for types in py__iter__(self._evaluator, types, first_nodes[0]):
yield types
module = self.var_args.get_parent_until()
if module is None:
return
is_list = str(self.instance.name) == 'list'
additions = _check_array_additions(self._evaluator, self.instance, module, is_list)
if additions:
yield additions
class Slice(object):
def __init__(self, evaluator, start, stop, step):
self._evaluator = evaluator
# all of them are either a Precedence or None.
self._start = start
self._stop = stop
self._step = step
@property
def obj(self):
"""
Imitate CompiledObject.obj behavior and return a ``builtin.slice()``
object.
"""
def get(element):
if element is None:
return None
result = self._evaluator.eval_element(element)
if len(result) != 1:
# For simplicity, we want slices to be clear defined with just
# one type. Otherwise we will return an empty slice object.
raise IndexError
try:
return list(result)[0].obj
except AttributeError:
return None
try:
return slice(get(self._start), get(self._stop), get(self._step))
except IndexError:
return slice(None, None, None)
def create_index_types(evaluator, index):
"""
Handles slices in subscript nodes.
"""
if index == ':':
# Like array[:]
return set([Slice(evaluator, None, None, None)])
elif tree.is_node(index, 'subscript'): # subscript is a slice operation.
# Like array[:3]
result = []
for el in index.children:
if el == ':':
if not result:
result.append(None)
elif tree.is_node(el, 'sliceop'):
if len(el.children) == 2:
result.append(el.children[1])
else:
result.append(el)
result += [None] * (3 - len(result))
return set([Slice(evaluator, *result)])
# No slices
return evaluator.eval_element(index)

View file

@ -0,0 +1,100 @@
"""
This module is not intended to be used in jedi, rather it will be fed to the
jedi-parser to replace classes in the typing module
"""
try:
from collections import abc
except ImportError:
# python 2
import collections as abc
def factory(typing_name, indextypes):
class Iterable(abc.Iterable):
def __iter__(self):
while True:
yield indextypes[0]()
class Iterator(Iterable, abc.Iterator):
def next(self):
""" needed for python 2 """
return self.__next__()
def __next__(self):
return indextypes[0]()
class Sequence(abc.Sequence):
def __getitem__(self, index):
return indextypes[0]()
class MutableSequence(Sequence, abc.MutableSequence):
pass
class List(MutableSequence, list):
pass
class Tuple(Sequence, tuple):
def __getitem__(self, index):
if indextypes[1] == Ellipsis:
# https://www.python.org/dev/peps/pep-0484/#the-typing-module
# Tuple[int, ...] means a tuple of ints of indetermined length
return indextypes[0]()
else:
return indextypes[index]()
class AbstractSet(Iterable, abc.Set):
pass
class MutableSet(AbstractSet, abc.MutableSet):
pass
class KeysView(Iterable, abc.KeysView):
pass
class ValuesView(abc.ValuesView):
def __iter__(self):
while True:
yield indextypes[1]()
class ItemsView(abc.ItemsView):
def __iter__(self):
while True:
yield (indextypes[0](), indextypes[1]())
class Mapping(Iterable, abc.Mapping):
def __getitem__(self, item):
return indextypes[1]()
def keys(self):
return KeysView()
def values(self):
return ValuesView()
def items(self):
return ItemsView()
class MutableMapping(Mapping, abc.MutableMapping):
pass
class Dict(MutableMapping, dict):
pass
dct = {
"Sequence": Sequence,
"MutableSequence": MutableSequence,
"List": List,
"Iterable": Iterable,
"Iterator": Iterator,
"AbstractSet": AbstractSet,
"MutableSet": MutableSet,
"Mapping": Mapping,
"MutableMapping": MutableMapping,
"Tuple": Tuple,
"KeysView": KeysView,
"ItemsView": ItemsView,
"ValuesView": ValuesView,
"Dict": Dict,
}
return dct[typing_name]

View file

@ -0,0 +1,438 @@
from collections import defaultdict
from itertools import chain
from jedi._compatibility import unicode, zip_longest
from jedi import debug
from jedi import common
from jedi.parser import tree
from jedi.evaluate import iterable
from jedi.evaluate import analysis
from jedi.evaluate import precedence
from jedi.evaluate.helpers import FakeName
from jedi.cache import underscore_memoization
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
# CompiledObject). Therefore don't loop infinitely.
return
for typ in types:
try:
f = typ.py__iter__
except AttributeError:
pass
else:
for iter_types in f():
try_iter_content(iter_types, depth + 1)
class Arguments(tree.Base):
def __init__(self, evaluator, argument_node, trailer=None):
"""
The argument_node is either a parser node or a list of evaluated
objects. Those evaluated objects may be lists of evaluated objects
themselves (one list for the first argument, one for the second, etc).
:param argument_node: May be an argument_node or a list of nodes.
"""
self.argument_node = argument_node
self._evaluator = evaluator
self.trailer = trailer # Can be None, e.g. in a class definition.
def _split(self):
if isinstance(self.argument_node, (tuple, list)):
for el in self.argument_node:
yield 0, el
else:
if not (tree.is_node(self.argument_node, 'arglist') or (
# in python 3.5 **arg is an argument, not arglist
(tree.is_node(self.argument_node, 'argument') and
self.argument_node.children[0] in ('*', '**')))):
yield 0, self.argument_node
return
iterator = iter(self.argument_node.children)
for child in iterator:
if child == ',':
continue
elif child in ('*', '**'):
yield len(child.value), next(iterator)
elif tree.is_node(child, 'argument') and \
child.children[0] in ('*', '**'):
assert len(child.children) == 2
yield len(child.children[0].value), child.children[1]
else:
yield 0, child
def get_parent_until(self, *args, **kwargs):
if self.trailer is None:
try:
element = self.argument_node[0]
from jedi.evaluate.iterable import AlreadyEvaluated
if isinstance(element, AlreadyEvaluated):
element = list(self._evaluator.eval_element(element))[0]
except IndexError:
return None
else:
return element.get_parent_until(*args, **kwargs)
else:
return self.trailer.get_parent_until(*args, **kwargs)
def as_tuple(self):
for stars, argument in self._split():
if tree.is_node(argument, 'argument'):
argument, default = argument.children[::2]
else:
default = None
yield argument, default, stars
def unpack(self, func=None):
named_args = []
for stars, el in self._split():
if stars == 1:
arrays = self._evaluator.eval_element(el)
iterators = [_iterate_star_args(self._evaluator, a, el, func)
for a in arrays]
iterators = list(iterators)
for values in list(zip_longest(*iterators)):
yield None, [v for v in values if v is not None]
elif stars == 2:
arrays = self._evaluator.eval_element(el)
dicts = [_star_star_dict(self._evaluator, a, el, func)
for a in arrays]
for dct in dicts:
for key, values in dct.items():
yield key, values
else:
if tree.is_node(el, 'argument'):
c = el.children
if len(c) == 3: # Keyword argument.
named_args.append((c[0].value, (c[2],)))
else: # Generator comprehension.
# Include the brackets with the parent.
comp = iterable.GeneratorComprehension(
self._evaluator, self.argument_node.parent)
yield None, (iterable.AlreadyEvaluated([comp]),)
elif isinstance(el, (list, tuple)):
yield None, el
else:
yield None, (el,)
# Reordering var_args is necessary, because star args sometimes appear
# after named argument, but in the actual order it's prepended.
for key_arg in named_args:
yield key_arg
def _reorder_var_args(var_args):
named_index = None
new_args = []
for i, stmt in enumerate(var_args):
if isinstance(stmt, tree.ExprStmt):
if named_index is None and stmt.assignment_details:
named_index = i
if named_index is not None:
expression_list = stmt.expression_list()
if expression_list and expression_list[0] == '*':
new_args.insert(named_index, stmt)
named_index += 1
continue
new_args.append(stmt)
return new_args
def eval_argument_clinic(self, arguments):
"""Uses a list with argument clinic information (see PEP 436)."""
iterator = self.unpack()
for i, (name, optional, allow_kwargs) in enumerate(arguments):
key, va_values = next(iterator, (None, []))
if key is not None:
raise NotImplementedError
if not va_values and not optional:
debug.warning('TypeError: %s expected at least %s arguments, got %s',
name, len(arguments), i)
raise ValueError
values = set(chain.from_iterable(self._evaluator.eval_element(el)
for el in va_values))
if not values 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 evaluation of that function.
debug.warning('argument_clinic "%s" not resolvable.', name)
raise ValueError
yield values
def scope(self):
# Returns the scope in which the arguments are used.
return (self.trailer or self.argument_node).get_parent_until(tree.IsScope)
def eval_args(self):
# TODO this method doesn't work with named args and a lot of other
# things. Use unpack.
return [self._evaluator.eval_element(el) for stars, el in self._split()]
def __repr__(self):
return '<%s: %s>' % (type(self).__name__, self.argument_node)
def get_calling_var_args(self):
if tree.is_node(self.argument_node, 'arglist', 'argument') \
or self.argument_node == () and self.trailer is not None:
return _get_calling_var_args(self._evaluator, self)
else:
return None
def eval_all(self, func=None):
"""
Evaluates all arguments as a support for static analysis
(normally Jedi).
"""
for key, element_values in self.unpack():
for element in element_values:
types = self._evaluator.eval_element(element)
try_iter_content(types)
class ExecutedParam(tree.Param):
"""Fake a param and give it values."""
def __init__(self, original_param, var_args, values):
self._original_param = original_param
self.var_args = var_args
self._values = values
def eval(self, evaluator):
types = set()
for v in self._values:
types |= evaluator.eval_element(v)
return types
@property
def position_nr(self):
# Need to use the original logic here, because it uses the parent.
return self._original_param.position_nr
@property
@underscore_memoization
def name(self):
return FakeName(str(self._original_param.name), self, self.start_pos)
def __getattr__(self, name):
return getattr(self._original_param, name)
def _get_calling_var_args(evaluator, var_args):
old_var_args = None
while var_args != old_var_args:
old_var_args = var_args
for name, default, stars in reversed(list(var_args.as_tuple())):
if not stars or not isinstance(name, tree.Name):
continue
names = evaluator.goto(name)
if len(names) != 1:
break
param = names[0].get_definition()
if not isinstance(param, ExecutedParam):
if isinstance(param, tree.Param):
# There is no calling var_args in this case - there's just
# a param without any input.
return None
break
# We never want var_args to be a tuple. This should be enough for
# now, we can change it later, if we need to.
if isinstance(param.var_args, Arguments):
var_args = param.var_args
return var_args.argument_node or var_args.trailer
def get_params(evaluator, func, var_args):
param_names = []
param_dict = {}
for param in func.params:
param_dict[str(param.name)] = param
unpacked_va = list(var_args.unpack(func))
from jedi.evaluate.representation import InstanceElement
if isinstance(func, InstanceElement):
# Include self at this place.
unpacked_va.insert(0, (None, [iterable.AlreadyEvaluated([func.instance])]))
var_arg_iterator = common.PushBackIterator(iter(unpacked_va))
non_matching_keys = defaultdict(lambda: [])
keys_used = {}
keys_only = False
had_multiple_value_error = False
for param in func.params:
# The value and key can both be null. There, the defaults apply.
# args / kwargs will just be empty arrays / dicts, respectively.
# Wrong value count is just ignored. If you try to test cases that are
# not allowed in Python, Jedi will maybe not show any completions.
default = [] if param.default is None else [param.default]
key, va_values = next(var_arg_iterator, (None, default))
while key is not None:
keys_only = True
k = unicode(key)
try:
key_param = param_dict[unicode(key)]
except KeyError:
non_matching_keys[key] = va_values
else:
param_names.append(ExecutedParam(key_param, var_args, va_values).name)
if k in keys_used:
had_multiple_value_error = True
m = ("TypeError: %s() got multiple values for keyword argument '%s'."
% (func.name, k))
calling_va = _get_calling_var_args(evaluator, var_args)
if calling_va is not None:
analysis.add(evaluator, 'type-error-multiple-values',
calling_va, message=m)
else:
try:
keys_used[k] = param_names[-1]
except IndexError:
# TODO this is wrong stupid and whatever.
pass
key, va_values = next(var_arg_iterator, (None, ()))
values = []
if param.stars == 1:
# *args param
lst_values = [iterable.MergedNodes(va_values)] if va_values else []
for key, va_values in var_arg_iterator:
# Iterate until a key argument is found.
if key:
var_arg_iterator.push_back((key, va_values))
break
if va_values:
lst_values.append(iterable.MergedNodes(va_values))
seq = iterable.FakeSequence(evaluator, lst_values, 'tuple')
values = [iterable.AlreadyEvaluated([seq])]
elif param.stars == 2:
# **kwargs param
dct = iterable.FakeDict(evaluator, dict(non_matching_keys))
values = [iterable.AlreadyEvaluated([dct])]
non_matching_keys = {}
else:
# normal param
if va_values:
values = va_values
else:
# No value: Return an empty container
values = []
if not keys_only:
calling_va = var_args.get_calling_var_args()
if calling_va is not None:
m = _error_argument_count(func, len(unpacked_va))
analysis.add(evaluator, 'type-error-too-few-arguments',
calling_va, message=m)
# Now add to result if it's not one of the previously covered cases.
if (not keys_only or param.stars == 2):
param_names.append(ExecutedParam(param, var_args, values).name)
keys_used[unicode(param.name)] = param_names[-1]
if keys_only:
# All arguments should be handed over to the next function. It's not
# about the values inside, it's about the names. Jedi needs to now that
# there's nothing to find for certain names.
for k in set(param_dict) - set(keys_used):
param = param_dict[k]
values = [] if param.default is None else [param.default]
param_names.append(ExecutedParam(param, var_args, values).name)
if not (non_matching_keys or had_multiple_value_error
or param.stars or param.default):
# add a warning only if there's not another one.
calling_va = _get_calling_var_args(evaluator, var_args)
if calling_va is not None:
m = _error_argument_count(func, len(unpacked_va))
analysis.add(evaluator, 'type-error-too-few-arguments',
calling_va, message=m)
for key, va_values in non_matching_keys.items():
m = "TypeError: %s() got an unexpected keyword argument '%s'." \
% (func.name, key)
for value in va_values:
analysis.add(evaluator, 'type-error-keyword-argument', value.parent, message=m)
remaining_params = list(var_arg_iterator)
if remaining_params:
m = _error_argument_count(func, len(unpacked_va))
# Just report an error for the first param that is not needed (like
# cPython).
first_key, first_values = remaining_params[0]
for v in first_values:
if first_key is not None:
# Is a keyword argument, return the whole thing instead of just
# the value node.
v = v.parent
try:
non_kw_param = keys_used[first_key]
except KeyError:
pass
else:
origin_args = non_kw_param.parent.var_args.argument_node
# TODO calculate the var_args tree and check if it's in
# the tree (if not continue).
# print('\t\tnonkw', non_kw_param.parent.var_args.argument_node, )
if origin_args not in [f.parent.parent for f in first_values]:
continue
analysis.add(evaluator, 'type-error-too-many-arguments',
v, message=m)
return param_names
def _iterate_star_args(evaluator, array, input_node, func=None):
from jedi.evaluate.representation import Instance
if isinstance(array, iterable.Array):
# TODO ._items is not the call we want here. Replace in the future.
for node in array._items():
yield node
elif isinstance(array, iterable.Generator):
for types in array.py__iter__():
yield iterable.AlreadyEvaluated(types)
elif isinstance(array, Instance) and array.name.get_code() == 'tuple':
debug.warning('Ignored a tuple *args input %s' % array)
else:
if func is not None:
m = "TypeError: %s() argument after * must be a sequence, not %s" \
% (func.name.value, array)
analysis.add(evaluator, 'type-error-star', input_node, message=m)
def _star_star_dict(evaluator, array, input_node, func):
dct = defaultdict(lambda: [])
from jedi.evaluate.representation import Instance
if isinstance(array, Instance) and array.name.get_code() == 'dict':
# For now ignore this case. In the future add proper iterators and just
# make one call without crazy isinstance checks.
return {}
if isinstance(array, iterable.FakeDict):
return array._dct
elif isinstance(array, iterable.Array) and array.type == 'dict':
# TODO bad call to non-public API
for key_node, value in array._items():
for key in evaluator.eval_element(key_node):
if precedence.is_string(key):
dct[key.obj].append(value)
else:
if func is not None:
m = "TypeError: %s argument after ** must be a mapping, not %s" \
% (func.name.value, array)
analysis.add(evaluator, 'type-error-star-star', input_node, message=m)
return dict(dct)
def _error_argument_count(func, actual_count):
default_arguments = sum(1 for p in func.params if p.default or p.stars)
if default_arguments == 0:
before = 'exactly '
else:
before = 'from %s to ' % (len(func.params) - default_arguments)
return ('TypeError: %s() takes %s%s arguments (%s given).'
% (func.name, before, len(func.params), actual_count))

View file

@ -0,0 +1,195 @@
"""
PEP 0484 ( https://www.python.org/dev/peps/pep-0484/ ) describes type hints
through function annotations. There is a strong suggestion in this document
that only the type of type hinting defined in PEP0484 should be allowed
as annotations in future python versions.
The (initial / probably incomplete) implementation todo list for pep-0484:
v Function parameter annotations with builtin/custom type classes
v Function returntype annotations with builtin/custom type classes
v Function parameter annotations with strings (forward reference)
v Function return type annotations with strings (forward reference)
v Local variable type hints
v Assigned types: `Url = str\ndef get(url:Url) -> str:`
v Type hints in `with` statements
x Stub files support
x support `@no_type_check` and `@no_type_check_decorator`
x support for typing.cast() operator
x support for type hint comments for functions, `# type: (int, str) -> int`.
See comment from Guido https://github.com/davidhalter/jedi/issues/662
"""
import itertools
import os
from jedi.parser import \
Parser, load_grammar, ParseError, ParserWithRecovery, tree
from jedi.evaluate.cache import memoize_default
from jedi.common import unite
from jedi.evaluate import compiled
from jedi import debug
from jedi import _compatibility
import re
def _evaluate_for_annotation(evaluator, annotation, index=None):
"""
Evaluates a string-node, looking for an annotation
If index is not None, the annotation is expected to be a tuple
and we're interested in that index
"""
if annotation is not None:
definitions = evaluator.eval_element(
_fix_forward_reference(evaluator, annotation))
if index is not None:
definitions = list(itertools.chain.from_iterable(
definition.py__getitem__(index) for definition in definitions
if definition.type == 'tuple' and
len(list(definition.py__iter__())) >= index))
return list(itertools.chain.from_iterable(
evaluator.execute(d) for d in definitions))
else:
return []
def _fix_forward_reference(evaluator, node):
evaled_nodes = evaluator.eval_element(node)
if len(evaled_nodes) != 1:
debug.warning("Eval'ed typing index %s should lead to 1 object, "
" not %s" % (node, evaled_nodes))
return node
evaled_node = list(evaled_nodes)[0]
if isinstance(evaled_node, compiled.CompiledObject) and \
isinstance(evaled_node.obj, str):
try:
p = Parser(load_grammar(), _compatibility.unicode(evaled_node.obj),
start_symbol='eval_input')
newnode = p.get_parsed_node()
except ParseError:
debug.warning('Annotation not parsed: %s' % evaled_node.obj)
return node
else:
module = node.get_parent_until()
p.position_modifier.line = module.end_pos[0]
newnode.parent = module
return newnode
else:
return node
@memoize_default(None, evaluator_is_first_arg=True)
def follow_param(evaluator, param):
annotation = param.annotation()
return _evaluate_for_annotation(evaluator, annotation)
@memoize_default(None, evaluator_is_first_arg=True)
def find_return_types(evaluator, func):
annotation = func.py__annotations__().get("return", None)
return _evaluate_for_annotation(evaluator, annotation)
_typing_module = None
def _get_typing_replacement_module():
"""
The idea is to return our jedi replacement for the PEP-0484 typing module
as discussed at https://github.com/davidhalter/jedi/issues/663
"""
global _typing_module
if _typing_module is None:
typing_path = \
os.path.abspath(os.path.join(__file__, "../jedi_typing.py"))
with open(typing_path) as f:
code = _compatibility.unicode(f.read())
p = ParserWithRecovery(load_grammar(), code)
_typing_module = p.module
return _typing_module
def get_types_for_typing_module(evaluator, typ, node):
from jedi.evaluate.iterable import FakeSequence
if not typ.base.get_parent_until().name.value == "typing":
return None
# we assume that any class using [] in a module called
# "typing" with a name for which we have a replacement
# should be replaced by that class. This is not 100%
# airtight but I don't have a better idea to check that it's
# actually the PEP-0484 typing module and not some other
if tree.is_node(node, "subscriptlist"):
nodes = node.children[::2] # skip the commas
else:
nodes = [node]
del node
nodes = [_fix_forward_reference(evaluator, node) for node in nodes]
# hacked in Union and Optional, since it's hard to do nicely in parsed code
if typ.name.value == "Union":
return unite(evaluator.eval_element(node) for node in nodes)
if typ.name.value == "Optional":
return evaluator.eval_element(nodes[0])
typing = _get_typing_replacement_module()
factories = evaluator.find_types(typing, "factory")
assert len(factories) == 1
factory = list(factories)[0]
assert factory
function_body_nodes = factory.children[4].children
valid_classnames = set(child.name.value
for child in function_body_nodes
if isinstance(child, tree.Class))
if typ.name.value not in valid_classnames:
return None
compiled_classname = compiled.create(evaluator, typ.name.value)
args = FakeSequence(evaluator, nodes, "tuple")
result = evaluator.execute_evaluated(factory, compiled_classname, args)
return result
def find_type_from_comment_hint_for(evaluator, node, name):
return \
_find_type_from_comment_hint(evaluator, node, node.children[1], name)
def find_type_from_comment_hint_with(evaluator, node, name):
assert len(node.children[1].children) == 3, \
"Can only be here when children[1] is 'foo() as f'"
return _find_type_from_comment_hint(
evaluator, node, node.children[1].children[2], name)
def find_type_from_comment_hint_assign(evaluator, node, name):
return \
_find_type_from_comment_hint(evaluator, node, node.children[0], name)
def _find_type_from_comment_hint(evaluator, node, varlist, name):
index = None
if varlist.type in ("testlist_star_expr", "exprlist"):
# something like "a, b = 1, 2"
index = 0
for child in varlist.children:
if child == name:
break
if child.type == "operator":
continue
index += 1
else:
return []
comment = node.get_following_comment_same_line()
if comment is None:
return []
match = re.match(r"^#\s*type:\s*([^#]*)", comment)
if not match:
return []
annotation = tree.String(
tree.zero_position_modifier,
repr(str(match.group(1).strip())),
node.start_pos)
annotation.parent = node.parent
return _evaluate_for_annotation(evaluator, annotation, index)

View file

@ -0,0 +1,178 @@
"""
Handles operator precedence.
"""
import operator
from jedi._compatibility import unicode
from jedi.parser import tree
from jedi import debug
from jedi.evaluate.compiled import CompiledObject, create, builtin_from_name
from jedi.evaluate import analysis
# Maps Python syntax to the operator module.
COMPARISON_OPERATORS = {
'==': operator.eq,
'!=': operator.ne,
'is': operator.is_,
'is not': operator.is_not,
'<': operator.lt,
'<=': operator.le,
'>': operator.gt,
'>=': operator.ge,
}
def literals_to_types(evaluator, result):
# Changes literals ('a', 1, 1.0, etc) to its type instances (str(),
# int(), float(), etc).
new_result = set()
for typ in result:
if is_literal(typ):
# Literals are only valid as long as the operations are
# correct. Otherwise add a value-free instance.
cls = builtin_from_name(evaluator, typ.name.value)
new_result |= evaluator.execute(cls)
else:
new_result.add(typ)
return new_result
def calculate_children(evaluator, children):
"""
Calculate a list of children with operators.
"""
iterator = iter(children)
types = evaluator.eval_element(next(iterator))
for operator in iterator:
right = next(iterator)
if tree.is_node(operator, 'comp_op'): # not in / is not
operator = ' '.join(str(c.value) for c in operator.children)
# handle lazy evaluation of and/or here.
if operator in ('and', 'or'):
left_bools = set([left.py__bool__() for left in types])
if left_bools == set([True]):
if operator == 'and':
types = evaluator.eval_element(right)
elif left_bools == set([False]):
if operator != 'and':
types = evaluator.eval_element(right)
# Otherwise continue, because of uncertainty.
else:
types = calculate(evaluator, types, operator,
evaluator.eval_element(right))
debug.dbg('calculate_children types %s', types)
return types
def calculate(evaluator, left_result, operator, right_result):
result = set()
if not left_result or not right_result:
# illegal slices e.g. cause left/right_result to be None
result = (left_result or set()) | (right_result or set())
result = literals_to_types(evaluator, result)
else:
# I don't think there's a reasonable chance that a string
# operation is still correct, once we pass something like six
# objects.
if len(left_result) * len(right_result) > 6:
result = literals_to_types(evaluator, left_result | right_result)
else:
for left in left_result:
for right in right_result:
result |= _element_calculate(evaluator, left, operator, right)
return result
def factor_calculate(evaluator, types, operator):
"""
Calculates `+`, `-`, `~` and `not` prefixes.
"""
for typ in types:
if operator == '-':
if _is_number(typ):
yield create(evaluator, -typ.obj)
elif operator == 'not':
value = typ.py__bool__()
if value is None: # Uncertainty.
return
yield create(evaluator, not value)
else:
yield typ
def _is_number(obj):
return isinstance(obj, CompiledObject) \
and isinstance(obj.obj, (int, float))
def is_string(obj):
return isinstance(obj, CompiledObject) \
and isinstance(obj.obj, (str, unicode))
def is_literal(obj):
return _is_number(obj) or is_string(obj)
def _is_tuple(obj):
from jedi.evaluate import iterable
return isinstance(obj, iterable.Array) and obj.type == 'tuple'
def _is_list(obj):
from jedi.evaluate import iterable
return isinstance(obj, iterable.Array) and obj.type == 'list'
def _element_calculate(evaluator, left, operator, right):
from jedi.evaluate import iterable, representation as er
l_is_num = _is_number(left)
r_is_num = _is_number(right)
if operator == '*':
# for iterables, ignore * operations
if isinstance(left, iterable.Array) or is_string(left):
return set([left])
elif isinstance(right, iterable.Array) or is_string(right):
return set([right])
elif operator == '+':
if l_is_num and r_is_num or is_string(left) and is_string(right):
return set([create(evaluator, left.obj + right.obj)])
elif _is_tuple(left) and _is_tuple(right) or _is_list(left) and _is_list(right):
return set([iterable.MergedArray(evaluator, (left, right))])
elif operator == '-':
if l_is_num and r_is_num:
return set([create(evaluator, left.obj - right.obj)])
elif operator == '%':
# With strings and numbers the left type typically remains. Except for
# `int() % float()`.
return set([left])
elif operator in COMPARISON_OPERATORS:
operation = COMPARISON_OPERATORS[operator]
if isinstance(left, CompiledObject) and isinstance(right, CompiledObject):
# Possible, because the return is not an option. Just compare.
left = left.obj
right = right.obj
try:
result = operation(left, right)
except TypeError:
# Could be True or False.
return set([create(evaluator, True), create(evaluator, False)])
else:
return set([create(evaluator, result)])
elif operator == 'in':
return set()
def check(obj):
"""Checks if a Jedi object is either a float or an int."""
return isinstance(obj, er.Instance) and obj.name.get_code() in ('int', 'float')
# Static analysis, one is a number, the other one is not.
if operator in ('+', '-') and l_is_num != r_is_num \
and not (check(left) or check(right)):
message = "TypeError: unsupported operand type(s) for +: %s and %s"
analysis.add(evaluator, 'type-error-operation', operator,
message % (left, right))
return set([left, right])

View file

@ -0,0 +1,157 @@
"""
Recursions are the recipe of |jedi| to conquer Python code. However, someone
must stop recursions going mad. Some settings are here to make |jedi| stop at
the right time. You can read more about them :ref:`here <settings-recursion>`.
Next to :mod:`jedi.evaluate.cache` this module also makes |jedi| not
thread-safe. Why? ``execution_recursion_decorator`` uses class variables to
count the function calls.
"""
from jedi import debug
from jedi import settings
from jedi.evaluate import iterable
def recursion_decorator(func):
def run(evaluator, stmt, *args, **kwargs):
rec_detect = evaluator.recursion_detector
if rec_detect.push_stmt(stmt):
return set()
else:
result = func(evaluator, stmt, *args, **kwargs)
rec_detect.pop_stmt()
return result
return run
class RecursionDetector(object):
"""
A decorator to detect recursions in statements. In a recursion a statement
at the same place, in the same module may not be executed two times.
"""
def __init__(self, evaluator):
self.top = None
self.current = None
self._evaluator = evaluator
def push_stmt(self, stmt):
self.current = _RecursionNode(self._evaluator, stmt, self.current)
check = self._check_recursion()
if check:
debug.warning('catched stmt recursion: %s against %s @%s', stmt,
check.stmt, stmt.start_pos)
self.pop_stmt()
return True
return False
def pop_stmt(self):
if self.current is not None:
# I don't know how current can be None, but sometimes it happens
# with Python3.
self.current = self.current.parent
def _check_recursion(self):
test = self.current
while True:
test = test.parent
if self.current == test:
return test
if not test:
return False
def node_statements(self):
result = []
n = self.current
while n:
result.insert(0, n.stmt)
n = n.parent
return result
class _RecursionNode(object):
""" A node of the RecursionDecorator. """
def __init__(self, evaluator, stmt, parent):
self._evaluator = evaluator
self.script = stmt.get_parent_until()
self.position = stmt.start_pos
self.parent = parent
self.stmt = stmt
# Don't check param instances, they are not causing recursions
# The same's true for the builtins, because the builtins are really
# simple.
self.is_ignored = self.script == self._evaluator.BUILTINS
def __eq__(self, other):
if not other:
return None
return self.script == other.script \
and self.position == other.position \
and not self.is_ignored and not other.is_ignored
def execution_recursion_decorator(func):
def run(execution, **kwargs):
detector = execution._evaluator.execution_recursion_detector
if detector.push_execution(execution):
result = set()
else:
result = func(execution, **kwargs)
detector.pop_execution()
return result
return run
class ExecutionRecursionDetector(object):
"""
Catches recursions of executions.
"""
def __init__(self, evaluator):
self.recursion_level = 0
self.parent_execution_funcs = []
self.execution_funcs = set()
self.execution_count = 0
self._evaluator = evaluator
def __call__(self, execution):
debug.dbg('Execution recursions: %s', execution, self.recursion_level,
self.execution_count, len(self.execution_funcs))
if self.check_recursion(execution):
result = set()
else:
result = self.func(execution)
self.pop_execution()
return result
def pop_execution(self):
self.parent_execution_funcs.pop()
self.recursion_level -= 1
def push_execution(self, execution):
in_par_execution_funcs = execution.base in self.parent_execution_funcs
in_execution_funcs = execution.base in self.execution_funcs
self.recursion_level += 1
self.execution_count += 1
self.execution_funcs.add(execution.base)
self.parent_execution_funcs.append(execution.base)
if self.execution_count > settings.max_executions:
return True
if isinstance(execution.base, (iterable.Array, iterable.Generator)):
return False
module = execution.get_parent_until()
if module == self._evaluator.BUILTINS:
return False
if in_par_execution_funcs:
if self.recursion_level > settings.max_function_recursion_level:
return True
if in_execution_funcs and \
len(self.execution_funcs) > settings.max_until_execution_unique:
return True
if self.execution_count > settings.max_executions_without_builtins:
return True
return False

View file

@ -0,0 +1,974 @@
"""
Like described in the :mod:`jedi.parser.tree` module,
there's a need for an ast like module to represent the states of parsed
modules.
But now there are also structures in Python that need a little bit more than
that. An ``Instance`` for example is only a ``Class`` before it is
instantiated. This class represents these cases.
So, why is there also a ``Class`` class here? Well, there are decorators and
they change classes in Python 3.
Representation modules also define "magic methods". Those methods look like
``py__foo__`` and are typically mappable to the Python equivalents ``__call__``
and others. Here's a list:
====================================== ========================================
**Method** **Description**
-------------------------------------- ----------------------------------------
py__call__(params: Array) On callable objects, returns types.
py__bool__() Returns True/False/None; None means that
there's no certainty.
py__bases__() Returns a list of base classes.
py__mro__() Returns a list of classes (the mro).
py__iter__() Returns a generator of a set of types.
py__class__() Returns the class of an instance.
py__getitem__(index: int/str) Returns a a set of types of the index.
Can raise an IndexError/KeyError.
py__file__() Only on modules.
py__package__() Only on modules. For the import system.
py__path__() Only on modules. For the import system.
====================================== ========================================
__
"""
import os
import pkgutil
import imp
import re
from itertools import chain
from jedi._compatibility import use_metaclass, unicode, Python3Method, is_py3
from jedi.parser import tree
from jedi import debug
from jedi import common
from jedi.cache import underscore_memoization, cache_star_import
from jedi.evaluate.cache import memoize_default, CachedMetaClass, NO_DEFAULT
from jedi.evaluate import compiled
from jedi.evaluate.compiled import mixed
from jedi.evaluate import recursion
from jedi.evaluate import iterable
from jedi.evaluate import docstrings
from jedi.evaluate import pep0484
from jedi.evaluate import helpers
from jedi.evaluate import param
from jedi.evaluate import flow_analysis
from jedi.evaluate import imports
class Executed(tree.Base):
"""
An instance is also an executable - because __init__ is called
:param var_args: The param input array, consist of a parser node or a list.
"""
def __init__(self, evaluator, base, var_args=()):
self._evaluator = evaluator
self.base = base
self.var_args = var_args
def is_scope(self):
return True
def get_parent_until(self, *args, **kwargs):
return tree.Base.get_parent_until(self, *args, **kwargs)
@common.safe_property
def parent(self):
return self.base.parent
class Instance(use_metaclass(CachedMetaClass, Executed)):
"""
This class is used to evaluate instances.
"""
def __init__(self, evaluator, base, var_args, is_generated=False):
super(Instance, self).__init__(evaluator, base, var_args)
self.decorates = None
# Generated instances are classes that are just generated by self
# (No var_args) used.
self.is_generated = is_generated
if base.name.get_code() in ['list', 'set'] \
and evaluator.BUILTINS == base.get_parent_until():
# compare the module path with the builtin name.
self.var_args = iterable.check_array_instances(evaluator, self)
elif not is_generated:
# Need to execute the __init__ function, because the dynamic param
# searching needs it.
try:
method = self.get_subscope_by_name('__init__')
except KeyError:
pass
else:
evaluator.execute(method, self.var_args)
def is_class(self):
return False
@property
def py__call__(self):
def actual(params):
return self._evaluator.execute(method, params)
try:
method = self.get_subscope_by_name('__call__')
except KeyError:
# Means the Instance is not callable.
raise AttributeError
return actual
def py__class__(self):
return self.base
def py__bool__(self):
# Signalize that we don't know about the bool type.
return None
@memoize_default()
def _get_method_execution(self, func):
func = get_instance_el(self._evaluator, self, func, True)
return FunctionExecution(self._evaluator, func, self.var_args)
def _get_func_self_name(self, func):
"""
Returns the name of the first param in a class method (which is
normally self.
"""
try:
return str(func.params[0].name)
except IndexError:
return None
def _self_names_dict(self, add_mro=True):
names = {}
# This loop adds the names of the self object, copies them and removes
# the self.
for sub in self.base.subscopes:
if isinstance(sub, tree.Class):
continue
# Get the self name, if there's one.
self_name = self._get_func_self_name(sub)
if self_name is None:
continue
if sub.name.value == '__init__' and not self.is_generated:
# ``__init__`` is special because the params need are injected
# this way. Therefore an execution is necessary.
if not sub.get_decorators():
# __init__ decorators should generally just be ignored,
# because to follow them and their self variables is too
# complicated.
sub = self._get_method_execution(sub)
for name_list in sub.names_dict.values():
for name in name_list:
if name.value == self_name and name.get_previous_sibling() is None:
trailer = name.get_next_sibling()
if tree.is_node(trailer, 'trailer') \
and len(trailer.children) == 2 \
and trailer.children[0] == '.':
name = trailer.children[1] # After dot.
if name.is_definition():
arr = names.setdefault(name.value, [])
arr.append(get_instance_el(self._evaluator, self, name))
return names
def get_subscope_by_name(self, name):
sub = self.base.get_subscope_by_name(name)
return get_instance_el(self._evaluator, self, sub, True)
def execute_subscope_by_name(self, name, *args):
method = self.get_subscope_by_name(name)
return self._evaluator.execute_evaluated(method, *args)
def get_descriptor_returns(self, obj):
""" Throws a KeyError if there's no method. """
# Arguments in __get__ descriptors are obj, class.
# `method` is the new parent of the array, don't know if that's good.
none_obj = compiled.create(self._evaluator, None)
args = [obj, obj.base] if isinstance(obj, Instance) else [none_obj, obj]
try:
return self.execute_subscope_by_name('__get__', *args)
except KeyError:
return set([self])
@memoize_default()
def names_dicts(self, search_global):
yield self._self_names_dict()
for s in self.base.py__mro__()[1:]:
if not isinstance(s, compiled.CompiledObject):
# Compiled objects don't have `self.` names.
for inst in self._evaluator.execute(s):
yield inst._self_names_dict(add_mro=False)
for names_dict in self.base.names_dicts(search_global=False, is_instance=True):
yield LazyInstanceDict(self._evaluator, self, names_dict)
def py__getitem__(self, index):
try:
method = self.get_subscope_by_name('__getitem__')
except KeyError:
debug.warning('No __getitem__, cannot access the array.')
return set()
else:
index_obj = compiled.create(self._evaluator, index)
return self._evaluator.execute_evaluated(method, index_obj)
def py__iter__(self):
try:
method = self.get_subscope_by_name('__iter__')
except KeyError:
debug.warning('No __iter__ on %s.' % self)
return
else:
iters = self._evaluator.execute(method)
for generator in iters:
if isinstance(generator, Instance):
# `__next__` logic.
name = '__next__' if is_py3 else 'next'
try:
yield generator.execute_subscope_by_name(name)
except KeyError:
debug.warning('Instance has no __next__ function in %s.', generator)
else:
for typ in generator.py__iter__():
yield typ
@property
@underscore_memoization
def name(self):
name = self.base.name
return helpers.FakeName(unicode(name), self, name.start_pos)
def __getattr__(self, name):
if name not in ['start_pos', 'end_pos', 'get_imports', 'type',
'doc', 'raw_doc']:
raise AttributeError("Instance %s: Don't touch this (%s)!"
% (self, name))
return getattr(self.base, name)
def __repr__(self):
dec = ''
if self.decorates is not None:
dec = " decorates " + repr(self.decorates)
return "<%s of %s(%s)%s>" % (type(self).__name__, self.base,
self.var_args, dec)
class LazyInstanceDict(object):
def __init__(self, evaluator, instance, dct):
self._evaluator = evaluator
self._instance = instance
self._dct = dct
def __getitem__(self, name):
return [get_instance_el(self._evaluator, self._instance, var, True)
for var in self._dct[name]]
def values(self):
return [self[key] for key in self._dct]
class InstanceName(tree.Name):
def __init__(self, origin_name, parent):
super(InstanceName, self).__init__(tree.zero_position_modifier,
origin_name.value,
origin_name.start_pos)
self._origin_name = origin_name
self.parent = parent
def is_definition(self):
return self._origin_name.is_definition()
def get_instance_el(evaluator, instance, var, is_class_var=False):
"""
Returns an InstanceElement if it makes sense, otherwise leaves the object
untouched.
Basically having an InstanceElement is context information. That is needed
in quite a lot of cases, which includes Nodes like ``power``, that need to
know where a self name comes from for example.
"""
if isinstance(var, tree.Name):
parent = get_instance_el(evaluator, instance, var.parent, is_class_var)
return InstanceName(var, parent)
elif var.type != 'funcdef' \
and isinstance(var, (Instance, compiled.CompiledObject, tree.Leaf,
tree.Module, FunctionExecution)):
return var
var = evaluator.wrap(var)
return InstanceElement(evaluator, instance, var, is_class_var)
class InstanceElement(use_metaclass(CachedMetaClass, tree.Base)):
"""
InstanceElement is a wrapper for any object, that is used as an instance
variable (e.g. self.variable or class methods).
"""
def __init__(self, evaluator, instance, var, is_class_var):
self._evaluator = evaluator
self.instance = instance
self.var = var
self.is_class_var = is_class_var
@common.safe_property
@memoize_default()
def parent(self):
par = self.var.parent
if isinstance(par, Class) and par == self.instance.base \
or isinstance(par, tree.Class) \
and par == self.instance.base.base:
par = self.instance
else:
par = get_instance_el(self._evaluator, self.instance, par,
self.is_class_var)
return par
def get_parent_until(self, *args, **kwargs):
return tree.BaseNode.get_parent_until(self, *args, **kwargs)
def get_definition(self):
return self.get_parent_until((tree.ExprStmt, tree.IsScope, tree.Import))
def get_decorated_func(self):
""" Needed because the InstanceElement should not be stripped """
func = self.var.get_decorated_func()
func = get_instance_el(self._evaluator, self.instance, func)
return func
def get_rhs(self):
return get_instance_el(self._evaluator, self.instance,
self.var.get_rhs(), self.is_class_var)
def is_definition(self):
return self.var.is_definition()
@property
def children(self):
# Copy and modify the array.
return [get_instance_el(self._evaluator, self.instance, command, self.is_class_var)
for command in self.var.children]
@property
@memoize_default()
def name(self):
name = self.var.name
return helpers.FakeName(unicode(name), self, name.start_pos)
def __iter__(self):
for el in self.var.__iter__():
yield get_instance_el(self._evaluator, self.instance, el,
self.is_class_var)
def __getitem__(self, index):
return get_instance_el(self._evaluator, self.instance, self.var[index],
self.is_class_var)
def __getattr__(self, name):
return getattr(self.var, name)
def isinstance(self, *cls):
return isinstance(self.var, cls)
def is_scope(self):
"""
Since we inherit from Base, it would overwrite the action we want here.
"""
return self.var.is_scope()
def py__call__(self, params):
if isinstance(self.var, compiled.CompiledObject):
# This check is a bit strange, but CompiledObject itself is a bit
# more complicated than we would it actually like to be.
return self.var.py__call__(params)
else:
return Function.py__call__(self, params)
def __repr__(self):
return "<%s of %s>" % (type(self).__name__, self.var)
class Wrapper(tree.Base):
def is_scope(self):
return True
def is_class(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
@property
@underscore_memoization
def name(self):
name = self.base.name
return helpers.FakeName(unicode(name), self, name.start_pos)
class Class(use_metaclass(CachedMetaClass, Wrapper)):
"""
This class is not only important to extend `tree.Class`, it is also a
important for descriptors (if the descriptor methods are evaluated or not).
"""
def __init__(self, evaluator, base):
self._evaluator = evaluator
self.base = base
@memoize_default(default=())
def py__mro__(self):
def add(cls):
if cls not in mro:
mro.append(cls)
mro = [self]
# TODO Do a proper mro resolution. Currently we are just listing
# classes. However, it's a complicated algorithm.
for cls in self.py__bases__():
# TODO detect for TypeError: duplicate base class str,
# e.g. `class X(str, str): pass`
try:
mro_method = cls.py__mro__
except AttributeError:
# TODO add a TypeError like:
"""
>>> class Y(lambda: test): pass
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: function() argument 1 must be code, not str
>>> class Y(1): pass
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: int() takes at most 2 arguments (3 given)
"""
pass
else:
add(cls)
for cls_new in mro_method():
add(cls_new)
return tuple(mro)
@memoize_default(default=())
def py__bases__(self):
arglist = self.base.get_super_arglist()
if arglist:
args = param.Arguments(self._evaluator, arglist)
return list(chain.from_iterable(args.eval_args()))
else:
return [compiled.create(self._evaluator, object)]
def py__call__(self, params):
return set([Instance(self._evaluator, self, params)])
def py__class__(self):
return compiled.create(self._evaluator, type)
@property
def params(self):
try:
return self.get_subscope_by_name('__init__').params
except KeyError:
return [] # object.__init__
def names_dicts(self, search_global, is_instance=False):
if search_global:
yield self.names_dict
else:
for scope in self.py__mro__():
if isinstance(scope, compiled.CompiledObject):
yield scope.names_dicts(False, is_instance)[0]
else:
yield scope.names_dict
def is_class(self):
return True
def get_subscope_by_name(self, name):
for s in self.py__mro__():
for sub in reversed(s.subscopes):
if sub.name.value == name:
return sub
raise KeyError("Couldn't find subscope.")
def __getattr__(self, name):
if name not in ['start_pos', 'end_pos', 'parent', 'raw_doc',
'doc', 'get_imports', 'get_parent_until', 'get_code',
'subscopes', 'names_dict', 'type']:
raise AttributeError("Don't touch this: %s of %s !" % (name, self))
return getattr(self.base, name)
def __repr__(self):
return "<e%s of %s>" % (type(self).__name__, self.base)
class Function(use_metaclass(CachedMetaClass, Wrapper)):
"""
Needed because of decorators. Decorators are evaluated here.
"""
def __init__(self, evaluator, func, is_decorated=False):
""" This should not be called directly """
self._evaluator = evaluator
self.base = self.base_func = func
self.is_decorated = is_decorated
# A property that is set by the decorator resolution.
self.decorates = None
@memoize_default()
def get_decorated_func(self):
"""
Returns the function, that should to be executed in the end.
This is also the places where the decorators are processed.
"""
f = self.base_func
decorators = self.base_func.get_decorators()
if not decorators or self.is_decorated:
return self
# Only enter it, if has not already been processed.
if not self.is_decorated:
for dec in reversed(decorators):
debug.dbg('decorator: %s %s', dec, f)
dec_results = self._evaluator.eval_element(dec.children[1])
trailer = dec.children[2:-1]
if trailer:
# Create a trailer and evaluate it.
trailer = tree.Node('trailer', trailer)
trailer.parent = dec
dec_results = self._evaluator.eval_trailer(dec_results, trailer)
if not len(dec_results):
debug.warning('decorator not found: %s on %s', dec, self.base_func)
return self
decorator = dec_results.pop()
if dec_results:
debug.warning('multiple decorators found %s %s',
self.base_func, dec_results)
# Create param array.
if isinstance(f, Function):
old_func = f # TODO this is just hacky. change.
elif f.type == 'funcdef':
old_func = Function(self._evaluator, f, is_decorated=True)
else:
old_func = f
wrappers = self._evaluator.execute_evaluated(decorator, old_func)
if not len(wrappers):
debug.warning('no wrappers found %s', self.base_func)
return self
if len(wrappers) > 1:
# TODO resolve issue with multiple wrappers -> multiple types
debug.warning('multiple wrappers found %s %s',
self.base_func, wrappers)
f = list(wrappers)[0]
if isinstance(f, (Instance, Function)):
f.decorates = self
debug.dbg('decorator end %s', f)
return f
def names_dicts(self, search_global):
if search_global:
yield self.names_dict
else:
scope = self.py__class__()
for names_dict in scope.names_dicts(False):
yield names_dict
@Python3Method
def py__call__(self, params):
if self.base.is_generator():
return set([iterable.Generator(self._evaluator, self, params)])
else:
return FunctionExecution(self._evaluator, self, params).get_return_types()
@memoize_default()
def py__annotations__(self):
parser_func = self.base
return_annotation = parser_func.annotation()
if return_annotation:
dct = {'return': return_annotation}
else:
dct = {}
for function_param in parser_func.params:
param_annotation = function_param.annotation()
if param_annotation is not None:
dct[function_param.name.value] = param_annotation
return dct
def py__class__(self):
# This differentiation is only necessary for Python2. Python3 does not
# use a different method class.
if isinstance(self.base.get_parent_scope(), tree.Class):
name = 'METHOD_CLASS'
else:
name = 'FUNCTION_CLASS'
return compiled.get_special_object(self._evaluator, name)
def __getattr__(self, name):
return getattr(self.base_func, name)
def __repr__(self):
dec = ''
if self.decorates is not None:
dec = " decorates " + repr(self.decorates)
return "<e%s of %s%s>" % (type(self).__name__, self.base_func, dec)
class LambdaWrapper(Function):
def get_decorated_func(self):
return self
class FunctionExecution(Executed):
"""
This class is used to evaluate functions and their returns.
This is the most complicated class, because it contains the logic to
transfer parameters. It is even more complicated, because there may be
multiple calls to functions and recursion has to be avoided. But this is
responsibility of the decorators.
"""
type = 'funcdef'
def __init__(self, evaluator, base, *args, **kwargs):
super(FunctionExecution, self).__init__(evaluator, base, *args, **kwargs)
self._copy_dict = {}
funcdef = base.base_func
if isinstance(funcdef, mixed.MixedObject):
# The extra information in mixed is not needed anymore. We can just
# unpack it and give it the tree object.
funcdef = funcdef.definition
# Just overwrite the old version. We don't need it anymore.
funcdef = helpers.deep_ast_copy(funcdef, new_elements=self._copy_dict)
for child in funcdef.children:
if child.type not in ('operator', 'keyword'):
# Not all nodes are properly copied by deep_ast_copy.
child.parent = self
self.children = funcdef.children
self.names_dict = funcdef.names_dict
@memoize_default(default=set())
@recursion.execution_recursion_decorator
def get_return_types(self, check_yields=False):
func = self.base
if func.isinstance(LambdaWrapper):
return self._evaluator.eval_element(self.children[-1])
if func.listeners:
# Feed the listeners, with the params.
for listener in func.listeners:
listener.execute(self._get_params())
# If we do have listeners, that means that there's not a regular
# execution ongoing. In this case Jedi is interested in the
# inserted params, not in the actual execution of the function.
return set()
if check_yields:
types = set()
returns = self.yields
else:
returns = self.returns
types = set(docstrings.find_return_types(self._evaluator, func))
types |= set(pep0484.find_return_types(self._evaluator, func))
for r in returns:
check = flow_analysis.break_check(self._evaluator, self, r)
if check is flow_analysis.UNREACHABLE:
debug.dbg('Return unreachable: %s', r)
else:
if check_yields:
types |= iterable.unite(self._eval_yield(r))
else:
types |= self._evaluator.eval_element(r.children[1])
if check is flow_analysis.REACHABLE:
debug.dbg('Return reachable: %s', r)
break
return types
def _eval_yield(self, yield_expr):
element = yield_expr.children[1]
if element.type == 'yield_arg':
# It must be a yield from.
yield_from_types = self._evaluator.eval_element(element.children[1])
for result in iterable.py__iter__(self._evaluator, yield_from_types, element):
yield result
else:
yield self._evaluator.eval_element(element)
@recursion.execution_recursion_decorator
def get_yield_types(self):
yields = self.yields
stopAt = tree.ForStmt, tree.WhileStmt, FunctionExecution, tree.IfStmt
for_parents = [(x, x.get_parent_until((stopAt))) for x in yields]
# Calculate if the yields are placed within the same for loop.
yields_order = []
last_for_stmt = None
for yield_, for_stmt in for_parents:
# For really simple for loops we can predict the order. Otherwise
# we just ignore it.
parent = for_stmt.parent
if parent.type == 'suite':
parent = parent.parent
if for_stmt.type == 'for_stmt' and parent == self \
and for_stmt.defines_one_name(): # Simplicity for now.
if for_stmt == last_for_stmt:
yields_order[-1][1].append(yield_)
else:
yields_order.append((for_stmt, [yield_]))
elif for_stmt == self:
yields_order.append((None, [yield_]))
else:
yield self.get_return_types(check_yields=True)
return
last_for_stmt = for_stmt
evaluator = self._evaluator
for for_stmt, yields in yields_order:
if for_stmt is None:
# No for_stmt, just normal yields.
for yield_ in yields:
for result in self._eval_yield(yield_):
yield result
else:
input_node = for_stmt.get_input_node()
for_types = evaluator.eval_element(input_node)
ordered = iterable.py__iter__(evaluator, for_types, input_node)
for index_types in ordered:
dct = {str(for_stmt.children[1]): index_types}
evaluator.predefined_if_name_dict_dict[for_stmt] = dct
for yield_in_same_for_stmt in yields:
for result in self._eval_yield(yield_in_same_for_stmt):
yield result
del evaluator.predefined_if_name_dict_dict[for_stmt]
def names_dicts(self, search_global):
yield self.names_dict
@memoize_default(default=NO_DEFAULT)
def _get_params(self):
"""
This returns the params for an TODO and is injected as a
'hack' into the tree.Function class.
This needs to be here, because Instance can have __init__ functions,
which act the same way as normal functions.
"""
return param.get_params(self._evaluator, self.base, self.var_args)
def param_by_name(self, name):
return [n for n in self._get_params() if str(n) == name][0]
def name_for_position(self, position):
return tree.Function.name_for_position(self, position)
def __getattr__(self, name):
if name not in ['start_pos', 'end_pos', 'imports', 'name', 'type']:
raise AttributeError('Tried to access %s: %s. Why?' % (name, self))
return getattr(self.base, name)
@common.safe_property
@memoize_default()
def returns(self):
return tree.Scope._search_in_scope(self, tree.ReturnStmt)
@common.safe_property
@memoize_default()
def yields(self):
return tree.Scope._search_in_scope(self, tree.YieldExpr)
@common.safe_property
@memoize_default()
def statements(self):
return tree.Scope._search_in_scope(self, tree.ExprStmt)
@common.safe_property
@memoize_default()
def subscopes(self):
return tree.Scope._search_in_scope(self, tree.Scope)
def __repr__(self):
return "<%s of %s>" % (type(self).__name__, self.base)
class GlobalName(helpers.FakeName):
def __init__(self, name):
"""
We need to mark global names somehow. Otherwise they are just normal
names that are not definitions.
"""
super(GlobalName, self).__init__(name.value, name.parent,
name.start_pos, is_definition=True)
class ModuleWrapper(use_metaclass(CachedMetaClass, tree.Module, Wrapper)):
def __init__(self, evaluator, module, parent_module=None):
self._evaluator = evaluator
self.base = self._module = module
self._parent_module = parent_module
def names_dicts(self, search_global):
yield self.base.names_dict
yield self._module_attributes_dict()
for star_module in self.star_imports():
yield star_module.names_dict
yield dict((str(n), [GlobalName(n)]) for n in self.base.global_names)
yield self._sub_modules_dict()
# I'm not sure if the star import cache is really that effective anymore
# with all the other really fast import caches. Recheck. Also we would need
# to push the star imports into Evaluator.modules, if we reenable this.
#@cache_star_import
@memoize_default([])
def star_imports(self):
modules = []
for i in self.base.imports:
if i.is_star_import():
name = i.star_import_name()
new = imports.ImportWrapper(self._evaluator, name).follow()
for module in new:
if isinstance(module, tree.Module):
modules += module.star_imports()
modules += new
return modules
@memoize_default()
def _module_attributes_dict(self):
def parent_callback():
# Create a string type object (without a defined string in it):
return list(self._evaluator.execute(compiled.create(self._evaluator, str)))[0]
names = ['__file__', '__package__', '__doc__', '__name__']
# All the additional module attributes are strings.
return dict((n, [helpers.LazyName(n, parent_callback, is_definition=True)])
for n in names)
@property
@memoize_default()
def name(self):
return helpers.FakeName(unicode(self.base.name), self, (1, 0))
def _get_init_directory(self):
"""
:return: The path to the directory of a package. None in case it's not
a package.
"""
for suffix, _, _ in imp.get_suffixes():
ending = '__init__' + suffix
py__file__ = self.py__file__()
if py__file__ is not None and py__file__.endswith(ending):
# Remove the ending, including the separator.
return self.py__file__()[:-len(ending) - 1]
return None
def py__name__(self):
for name, module in self._evaluator.modules.items():
if module == self:
return name
return '__main__'
def py__file__(self):
"""
In contrast to Python's __file__ can be None.
"""
if self._module.path is None:
return None
return os.path.abspath(self._module.path)
def py__package__(self):
if self._get_init_directory() is None:
return re.sub(r'\.?[^\.]+$', '', self.py__name__())
else:
return self.py__name__()
def _py__path__(self):
if self._parent_module is None:
search_path = self._evaluator.sys_path
else:
search_path = self._parent_module.py__path__()
init_path = self.py__file__()
if os.path.basename(init_path) == '__init__.py':
with open(init_path, 'rb') as f:
content = common.source_to_unicode(f.read())
# these are strings that need to be used for namespace packages,
# the first one is ``pkgutil``, the second ``pkg_resources``.
options = ('declare_namespace(__name__)', 'extend_path(__path__')
if options[0] in content or options[1] in content:
# It is a namespace, now try to find the rest of the
# modules on sys_path or whatever the search_path is.
paths = set()
for s in search_path:
other = os.path.join(s, unicode(self.name))
if os.path.isdir(other):
paths.add(other)
return list(paths)
# Default to this.
return [self._get_init_directory()]
@property
def py__path__(self):
"""
Not seen here, since it's a property. The callback actually uses a
variable, so use it like::
foo.py__path__(sys_path)
In case of a package, this returns Python's __path__ attribute, which
is a list of paths (strings).
Raises an AttributeError if the module is not a package.
"""
path = self._get_init_directory()
if path is None:
raise AttributeError('Only packages have __path__ attributes.')
else:
return self._py__path__
@memoize_default()
def _sub_modules_dict(self):
"""
Lists modules in the directory of this module (if this module is a
package).
"""
path = self._module.path
names = {}
if path is not None and path.endswith(os.path.sep + '__init__.py'):
mods = pkgutil.iter_modules([os.path.dirname(path)])
for module_loader, name, is_pkg in mods:
fake_n = helpers.FakeName(name)
# It's obviously a relative import to the current module.
imp = helpers.FakeImport(fake_n, self, level=1)
fake_n.parent = imp
names[name] = [fake_n]
# TODO add something like this in the future, its cleaner than the
# import hacks.
# ``os.path`` is a hardcoded exception, because it's a
# ``sys.modules`` modification.
#if str(self.name) == 'os':
# names.append(helpers.FakeName('path', parent=self))
return names
def py__class__(self):
return compiled.get_special_object(self._evaluator, 'MODULE_CLASS')
def __getattr__(self, name):
return getattr(self._module, name)
def __repr__(self):
return "<%s: %s>" % (type(self).__name__, self._module)

View file

@ -0,0 +1,110 @@
"""An adapted copy of relevant site-packages functionality from Python stdlib.
This file contains some functions related to handling site-packages in Python
with jedi-specific modifications:
- the functions operate on sys_path argument rather than global sys.path
- in .pth files "import ..." lines that allow execution of arbitrary code are
skipped to prevent code injection into jedi interpreter
"""
# Copyright (c) 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010,
# 2011, 2012, 2013, 2014, 2015 Python Software Foundation; All Rights Reserved
from __future__ import print_function
import sys
import os
def makepath(*paths):
dir = os.path.join(*paths)
try:
dir = os.path.abspath(dir)
except OSError:
pass
return dir, os.path.normcase(dir)
def _init_pathinfo(sys_path):
"""Return a set containing all existing directory entries from sys_path"""
d = set()
for dir in sys_path:
try:
if os.path.isdir(dir):
dir, dircase = makepath(dir)
d.add(dircase)
except TypeError:
continue
return d
def addpackage(sys_path, sitedir, name, known_paths):
"""Process a .pth file within the site-packages directory:
For each line in the file, either combine it with sitedir to a path
and add that to known_paths, or execute it if it starts with 'import '.
"""
if known_paths is None:
known_paths = _init_pathinfo(sys_path)
reset = 1
else:
reset = 0
fullname = os.path.join(sitedir, name)
try:
f = open(fullname, "r")
except OSError:
return
with f:
for n, line in enumerate(f):
if line.startswith("#"):
continue
try:
if line.startswith(("import ", "import\t")):
# Change by immerrr: don't evaluate import lines to prevent
# code injection into jedi through pth files.
#
# exec(line)
continue
line = line.rstrip()
dir, dircase = makepath(sitedir, line)
if not dircase in known_paths and os.path.exists(dir):
sys_path.append(dir)
known_paths.add(dircase)
except Exception:
print("Error processing line {:d} of {}:\n".format(n+1, fullname),
file=sys.stderr)
import traceback
for record in traceback.format_exception(*sys.exc_info()):
for line in record.splitlines():
print(' '+line, file=sys.stderr)
print("\nRemainder of file ignored", file=sys.stderr)
break
if reset:
known_paths = None
return known_paths
def addsitedir(sys_path, sitedir, known_paths=None):
"""Add 'sitedir' argument to sys_path if missing and handle .pth files in
'sitedir'"""
if known_paths is None:
known_paths = _init_pathinfo(sys_path)
reset = 1
else:
reset = 0
sitedir, sitedircase = makepath(sitedir)
if not sitedircase in known_paths:
sys_path.append(sitedir) # Add path component
known_paths.add(sitedircase)
try:
names = os.listdir(sitedir)
except OSError:
return
names = [name for name in names if name.endswith(".pth")]
for name in sorted(names):
addpackage(sys_path, sitedir, name, known_paths)
if reset:
known_paths = None
return known_paths

View file

@ -0,0 +1,280 @@
"""
Implementations of standard library functions, because it's not possible to
understand them with Jedi.
To add a new implementation, create a function and add it to the
``_implemented`` dict at the bottom of this module.
Note that this module exists only to implement very specific functionality in
the standard library. The usual way to understand the standard library is the
compiled module that returns the types for C-builtins.
"""
import collections
import re
from jedi._compatibility import unicode
from jedi.common import unite
from jedi.evaluate import compiled
from jedi.evaluate import representation as er
from jedi.evaluate import iterable
from jedi.parser import ParserWithRecovery
from jedi.parser import tree
from jedi import debug
from jedi.evaluate import precedence
from jedi.evaluate import param
from jedi.evaluate import analysis
class NotInStdLib(LookupError):
pass
def execute(evaluator, obj, arguments):
try:
obj_name = str(obj.name)
except AttributeError:
pass
else:
if obj.parent == evaluator.BUILTINS:
module_name = 'builtins'
elif isinstance(obj.parent, tree.Module):
module_name = str(obj.parent.name)
else:
module_name = ''
# for now we just support builtin functions.
try:
func = _implemented[module_name][obj_name]
except KeyError:
pass
else:
return func(evaluator, obj, arguments)
raise NotInStdLib()
def _follow_param(evaluator, arguments, index):
try:
key, values = list(arguments.unpack())[index]
except IndexError:
return set()
else:
return unite(evaluator.eval_element(v) for v in values)
def argument_clinic(string, want_obj=False, want_scope=False, want_arguments=False):
"""
Works like Argument Clinic (PEP 436), to validate function params.
"""
clinic_args = []
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('(?:(?:(\[),? ?|, ?|)(\w+)|, ?/)\]*', 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)
clinic_args.append((word, optional, allow_kwargs))
def f(func):
def wrapper(evaluator, obj, arguments):
debug.dbg('builtin start %s' % obj, color='MAGENTA')
try:
lst = list(arguments.eval_argument_clinic(clinic_args))
except ValueError:
return set()
else:
kwargs = {}
if want_scope:
kwargs['scope'] = arguments.scope()
if want_obj:
kwargs['obj'] = obj
if want_arguments:
kwargs['arguments'] = arguments
return func(evaluator, *lst, **kwargs)
finally:
debug.dbg('builtin end', color='MAGENTA')
return wrapper
return f
@argument_clinic('object, name[, default], /')
def builtins_getattr(evaluator, objects, names, defaults=None):
# follow the first param
for obj in objects:
if not isinstance(obj, (er.Instance, er.Class, tree.Module, compiled.CompiledObject)):
debug.warning('getattr called without instance')
continue
for name in names:
if precedence.is_string(name):
return evaluator.find_types(obj, name.obj)
else:
debug.warning('getattr called without str')
continue
return set()
@argument_clinic('object[, bases, dict], /')
def builtins_type(evaluator, objects, bases, dicts):
if bases or dicts:
# It's a type creation... maybe someday...
return set()
else:
return set([o.py__class__() for o in objects])
class SuperInstance(er.Instance):
"""To be used like the object ``super`` returns."""
def __init__(self, evaluator, cls):
su = cls.py_mro()[1]
super().__init__(evaluator, su and su[0] or self)
@argument_clinic('[type[, obj]], /', want_scope=True)
def builtins_super(evaluator, types, objects, scope):
# TODO make this able to detect multiple inheritance super
accept = (tree.Function, er.FunctionExecution)
if scope.isinstance(*accept):
wanted = (tree.Class, er.Instance)
cls = scope.get_parent_until(accept + wanted,
include_current=False)
if isinstance(cls, wanted):
if isinstance(cls, tree.Class):
cls = er.Class(evaluator, cls)
elif isinstance(cls, er.Instance):
cls = cls.base
su = cls.py__bases__()
if su:
return evaluator.execute(su[0])
return set()
@argument_clinic('sequence, /', want_obj=True, want_arguments=True)
def builtins_reversed(evaluator, sequences, obj, arguments):
# While we could do without this variable (just by using sequences), we
# want static analysis to work well. Therefore we need to generated the
# values again.
first_arg = next(arguments.as_tuple())[0]
ordered = list(iterable.py__iter__(evaluator, sequences, first_arg))
rev = [iterable.AlreadyEvaluated(o) for o in reversed(ordered)]
# Repack iterator values and then run it the normal way. This is
# necessary, because `reversed` is a function and autocompletion
# would fail in certain cases like `reversed(x).__iter__` if we
# just returned the result directly.
rev = iterable.AlreadyEvaluated(
[iterable.FakeSequence(evaluator, rev, 'list')]
)
return set([er.Instance(evaluator, obj, param.Arguments(evaluator, [rev]))])
@argument_clinic('obj, type, /', want_arguments=True)
def builtins_isinstance(evaluator, objects, types, arguments):
bool_results = set([])
for o in objects:
try:
mro_func = o.py__class__().py__mro__
except AttributeError:
# This is temporary. Everything should have a class attribute in
# Python?! Maybe we'll leave it here, because some numpy objects or
# whatever might not.
return set([compiled.create(True), compiled.create(False)])
mro = mro_func()
for cls_or_tup in types:
if cls_or_tup.is_class():
bool_results.add(cls_or_tup in mro)
elif str(cls_or_tup.name) == 'tuple' \
and cls_or_tup.get_parent_scope() == evaluator.BUILTINS:
# Check for tuples.
classes = unite(cls_or_tup.py__iter__())
bool_results.add(any(cls in mro for cls in classes))
else:
_, nodes = list(arguments.unpack())[1]
for node in nodes:
message = 'TypeError: isinstance() arg 2 must be a ' \
'class, type, or tuple of classes and types, ' \
'not %s.' % cls_or_tup
analysis.add(evaluator, 'type-error-isinstance', node, message)
return set(compiled.create(evaluator, x) for x in bool_results)
def collections_namedtuple(evaluator, obj, arguments):
"""
Implementation of the namedtuple function.
This has to be done by processing the namedtuple class template and
evaluating the result.
.. note:: |jedi| only supports namedtuples on Python >2.6.
"""
# Namedtuples are not supported on Python 2.6
if not hasattr(collections, '_class_template'):
return set()
# Process arguments
# TODO here we only use one of the types, we should use all.
name = list(_follow_param(evaluator, arguments, 0))[0].obj
_fields = list(_follow_param(evaluator, arguments, 1))[0]
if isinstance(_fields, compiled.CompiledObject):
fields = _fields.obj.replace(',', ' ').split()
elif isinstance(_fields, iterable.Array):
try:
fields = [v.obj for v in unite(_fields.py__iter__())]
except AttributeError:
return set()
else:
return set()
# Build source
source = collections._class_template.format(
typename=name,
field_names=fields,
num_fields=len(fields),
arg_list=', '.join(fields),
repr_fmt=', '.join(collections._repr_template.format(name=name) for name in fields),
field_defs='\n'.join(collections._field_template.format(index=index, name=name)
for index, name in enumerate(fields))
)
# Parse source
generated_class = ParserWithRecovery(evaluator.grammar, unicode(source)).module.subscopes[0]
return set([er.Class(evaluator, generated_class)])
@argument_clinic('first, /')
def _return_first_param(evaluator, firsts):
return firsts
_implemented = {
'builtins': {
'getattr': builtins_getattr,
'type': builtins_type,
'super': builtins_super,
'reversed': builtins_reversed,
'isinstance': builtins_isinstance,
},
'copy': {
'copy': _return_first_param,
'deepcopy': _return_first_param,
},
'json': {
'load': lambda *args: set(),
'loads': lambda *args: set(),
},
'collections': {
'namedtuple': collections_namedtuple,
},
}

View file

@ -0,0 +1,284 @@
import glob
import os
import sys
from jedi.evaluate.site import addsitedir
from jedi._compatibility import exec_function, unicode
from jedi.parser import tree
from jedi.parser import ParserWithRecovery
from jedi.evaluate.cache import memoize_default
from jedi import debug
from jedi import common
from jedi.parser.utils import load_parser, save_parser
def get_venv_path(venv):
"""Get sys.path for specified virtual environment."""
sys_path = _get_venv_path_dirs(venv)
with common.ignored(ValueError):
sys_path.remove('')
sys_path = _get_sys_path_with_egglinks(sys_path)
# As of now, get_venv_path_dirs does not scan built-in pythonpath and
# user-local site-packages, let's approximate them using path from Jedi
# interpreter.
return sys_path + sys.path
def _get_sys_path_with_egglinks(sys_path):
"""Find all paths including those referenced by egg-links.
Egg-link-referenced directories are inserted into path immediately before
the directory on which their links were found. Such directories are not
taken into consideration by normal import mechanism, but they are traversed
when doing pkg_resources.require.
"""
result = []
for p in sys_path:
# pkg_resources does not define a specific order for egg-link files
# using os.listdir to enumerate them, we're sorting them to have
# reproducible tests.
for egg_link in sorted(glob.glob(os.path.join(p, '*.egg-link'))):
with open(egg_link) as fd:
for line in fd:
line = line.strip()
if line:
result.append(os.path.join(p, line))
# pkg_resources package only interprets the first
# non-empty line in egg-link files.
break
result.append(p)
return result
def _get_venv_path_dirs(venv):
"""Get sys.path for venv without starting up the interpreter."""
venv = os.path.abspath(venv)
sitedir = _get_venv_sitepackages(venv)
sys_path = []
addsitedir(sys_path, sitedir)
return sys_path
def _get_venv_sitepackages(venv):
if os.name == 'nt':
p = os.path.join(venv, 'lib', 'site-packages')
else:
p = os.path.join(venv, 'lib', 'python%d.%d' % sys.version_info[:2],
'site-packages')
return p
def _execute_code(module_path, code):
c = "import os; from os.path import *; result=%s"
variables = {'__file__': module_path}
try:
exec_function(c % code, variables)
except Exception:
debug.warning('sys.path manipulation detected, but failed to evaluate.')
else:
try:
res = variables['result']
if isinstance(res, str):
return [os.path.abspath(res)]
except KeyError:
pass
return []
def _paths_from_assignment(evaluator, expr_stmt):
"""
Extracts the assigned strings from an assignment that looks as follows::
>>> sys.path[0:0] = ['module/path', 'another/module/path']
This function is in general pretty tolerant (and therefore 'buggy').
However, it's not a big issue usually to add more paths to Jedi's sys_path,
because it will only affect Jedi in very random situations and by adding
more paths than necessary, it usually benefits the general user.
"""
for assignee, operator in zip(expr_stmt.children[::2], expr_stmt.children[1::2]):
try:
assert operator in ['=', '+=']
assert tree.is_node(assignee, 'power', 'atom_expr') and \
len(assignee.children) > 1
c = assignee.children
assert c[0].type == 'name' and c[0].value == 'sys'
trailer = c[1]
assert trailer.children[0] == '.' and trailer.children[1].value == 'path'
# TODO Essentially we're not checking details on sys.path
# manipulation. Both assigment of the sys.path and changing/adding
# parts of the sys.path are the same: They get added to the current
# sys.path.
"""
execution = c[2]
assert execution.children[0] == '['
subscript = execution.children[1]
assert subscript.type == 'subscript'
assert ':' in subscript.children
"""
except AssertionError:
continue
from jedi.evaluate.iterable import py__iter__
from jedi.evaluate.precedence import is_string
types = evaluator.eval_element(expr_stmt)
for types in py__iter__(evaluator, types, expr_stmt):
for typ in types:
if is_string(typ):
yield typ.obj
def _paths_from_list_modifications(module_path, trailer1, trailer2):
""" extract the path from either "sys.path.append" or "sys.path.insert" """
# Guarantee that both are trailers, the first one a name and the second one
# a function execution with at least one param.
if not (tree.is_node(trailer1, 'trailer') and trailer1.children[0] == '.'
and tree.is_node(trailer2, 'trailer') and trailer2.children[0] == '('
and len(trailer2.children) == 3):
return []
name = trailer1.children[1].value
if name not in ['insert', 'append']:
return []
arg = trailer2.children[1]
if name == 'insert' and len(arg.children) in (3, 4): # Possible trailing comma.
arg = arg.children[2]
return _execute_code(module_path, arg.get_code())
def _check_module(evaluator, module):
"""
Detect sys.path modifications within module.
"""
def get_sys_path_powers(names):
for name in names:
power = name.parent.parent
if tree.is_node(power, 'power', 'atom_expr'):
c = power.children
if isinstance(c[0], tree.Name) and c[0].value == 'sys' \
and tree.is_node(c[1], 'trailer'):
n = c[1].children[1]
if isinstance(n, tree.Name) and n.value == 'path':
yield name, power
sys_path = list(evaluator.sys_path) # copy
try:
possible_names = module.used_names['path']
except KeyError:
# module.used_names is MergedNamesDict whose getitem never throws
# keyerror, this is superfluous.
pass
else:
for name, power in get_sys_path_powers(possible_names):
stmt = name.get_definition()
if len(power.children) >= 4:
sys_path.extend(_paths_from_list_modifications(module.path, *power.children[2:4]))
elif name.get_definition().type == 'expr_stmt':
sys_path.extend(_paths_from_assignment(evaluator, stmt))
return sys_path
@memoize_default(evaluator_is_first_arg=True, default=[])
def sys_path_with_modifications(evaluator, module):
if module.path is None:
# Support for modules without a path is bad, therefore return the
# normal path.
return list(evaluator.sys_path)
curdir = os.path.abspath(os.curdir)
#TODO why do we need a chdir?
with common.ignored(OSError):
os.chdir(os.path.dirname(module.path))
buildout_script_paths = set()
result = _check_module(evaluator, module)
result += _detect_django_path(module.path)
for buildout_script in _get_buildout_scripts(module.path):
for path in _get_paths_from_buildout_script(evaluator, buildout_script):
buildout_script_paths.add(path)
# cleanup, back to old directory
os.chdir(curdir)
return list(result) + list(buildout_script_paths)
def _get_paths_from_buildout_script(evaluator, buildout_script):
def load(buildout_script):
try:
with open(buildout_script, 'rb') as f:
source = common.source_to_unicode(f.read())
except IOError:
debug.dbg('Error trying to read buildout_script: %s', buildout_script)
return
p = ParserWithRecovery(evaluator.grammar, source, buildout_script)
save_parser(buildout_script, p)
return p.module
cached = load_parser(buildout_script)
module = cached and cached.module or load(buildout_script)
if not module:
return
for path in _check_module(evaluator, module):
yield path
def traverse_parents(path):
while True:
new = os.path.dirname(path)
if new == path:
return
path = new
yield path
def _get_parent_dir_with_file(path, filename):
for parent in traverse_parents(path):
if os.path.isfile(os.path.join(parent, filename)):
return parent
return None
def _detect_django_path(module_path):
""" Detects the path of the very well known Django library (if used) """
result = []
for parent in traverse_parents(module_path):
with common.ignored(IOError):
with open(parent + os.path.sep + 'manage.py'):
debug.dbg('Found django path: %s', module_path)
result.append(parent)
return result
def _get_buildout_scripts(module_path):
"""
if there is a 'buildout.cfg' file in one of the parent directories of the
given module it will return a list of all files in the buildout bin
directory that look like python files.
:param module_path: absolute path to the module.
:type module_path: str
"""
project_root = _get_parent_dir_with_file(module_path, 'buildout.cfg')
if not project_root:
return []
bin_path = os.path.join(project_root, 'bin')
if not os.path.exists(bin_path):
return []
extra_module_paths = []
for filename in os.listdir(bin_path):
try:
filepath = os.path.join(bin_path, filename)
with open(filepath, 'r') as f:
firstline = f.readline()
if firstline.startswith('#!') and 'python' in firstline:
extra_module_paths.append(filepath)
except (UnicodeDecodeError, IOError) as e:
# Probably a binary file; permission error or race cond. because file got deleted
# ignore
debug.warning(unicode(e))
continue
return extra_module_paths

View file

@ -0,0 +1,409 @@
"""
The ``Parser`` tries to convert the available Python code in an easy to read
format, something like an abstract syntax tree. The classes who represent this
tree, are sitting in the :mod:`jedi.parser.tree` module.
The Python module ``tokenize`` is a very important part in the ``Parser``,
because it splits the code into different words (tokens). Sometimes it looks a
bit messy. Sorry for that! You might ask now: "Why didn't you use the ``ast``
module for this? Well, ``ast`` does a very good job understanding proper Python
code, but fails to work as soon as there's a single line of broken code.
There's one important optimization that needs to be known: Statements are not
being parsed completely. ``Statement`` is just a representation of the tokens
within the statement. This lowers memory usage and cpu time and reduces the
complexity of the ``Parser`` (there's another parser sitting inside
``Statement``, which produces ``Array`` and ``Call``).
"""
import os
import re
from jedi.parser import tree as pt
from jedi.parser import tokenize
from jedi.parser.token import (DEDENT, INDENT, ENDMARKER, NEWLINE, NUMBER,
STRING)
from jedi.parser.pgen2.pgen import generate_grammar
from jedi.parser.pgen2.parse import PgenParser
OPERATOR_KEYWORDS = 'and', 'for', 'if', 'else', 'in', 'is', 'lambda', 'not', 'or'
# Not used yet. In the future I intend to add something like KeywordStatement
STATEMENT_KEYWORDS = 'assert', 'del', 'global', 'nonlocal', 'raise', \
'return', 'yield', 'pass', 'continue', 'break'
_loaded_grammars = {}
class ParseError(Exception):
"""
Signals you that the code you fed the Parser was not correct Python code.
"""
def load_grammar(version='3.4'):
# For now we only support two different Python syntax versions: The latest
# Python 3 and Python 2. This may change.
if version in ('3.2', '3.3'):
version = '3.4'
elif version == '2.6':
version = '2.7'
file = 'grammar' + version + '.txt'
global _loaded_grammars
path = os.path.join(os.path.dirname(__file__), file)
try:
return _loaded_grammars[path]
except KeyError:
return _loaded_grammars.setdefault(path, generate_grammar(path))
class ParserSyntaxError(object):
def __init__(self, message, position):
self.message = message
self.position = position
class Parser(object):
AST_MAPPING = {
'expr_stmt': pt.ExprStmt,
'classdef': pt.Class,
'funcdef': pt.Function,
'file_input': pt.Module,
'import_name': pt.ImportName,
'import_from': pt.ImportFrom,
'break_stmt': pt.KeywordStatement,
'continue_stmt': pt.KeywordStatement,
'return_stmt': pt.ReturnStmt,
'raise_stmt': pt.KeywordStatement,
'yield_expr': pt.YieldExpr,
'del_stmt': pt.KeywordStatement,
'pass_stmt': pt.KeywordStatement,
'global_stmt': pt.GlobalStmt,
'nonlocal_stmt': pt.KeywordStatement,
'print_stmt': pt.KeywordStatement,
'assert_stmt': pt.AssertStmt,
'if_stmt': pt.IfStmt,
'with_stmt': pt.WithStmt,
'for_stmt': pt.ForStmt,
'while_stmt': pt.WhileStmt,
'try_stmt': pt.TryStmt,
'comp_for': pt.CompFor,
'decorator': pt.Decorator,
'lambdef': pt.Lambda,
'old_lambdef': pt.Lambda,
'lambdef_nocond': pt.Lambda,
}
def __init__(self, grammar, source, start_symbol='file_input',
tokenizer=None, start_parsing=True):
# Todo Remove start_parsing (with False)
self._used_names = {}
self._scope_names_stack = [{}]
self._last_failed_start_pos = (0, 0)
self._global_names = []
# For the fast parser.
self.position_modifier = pt.PositionModifier()
self._added_newline = False
# The Python grammar needs a newline at the end of each statement.
if not source.endswith('\n') and start_symbol == 'file_input':
source += '\n'
self._added_newline = True
self.source = source
self._start_symbol = start_symbol
self._grammar = grammar
self._parsed = None
if start_parsing:
if tokenizer is None:
tokenizer = tokenize.source_tokens(source, use_exact_op_types=True)
self.parse(tokenizer)
def parse(self, tokenizer):
if self._parsed is not None:
return self._parsed
start_number = self._grammar.symbol2number[self._start_symbol]
pgen_parser = PgenParser(
self._grammar, self.convert_node, self.convert_leaf,
self.error_recovery, start_number
)
try:
self._parsed = pgen_parser.parse(tokenizer)
finally:
self.stack = pgen_parser.stack
if self._start_symbol == 'file_input' != self._parsed.type:
# If there's only one statement, we get back a non-module. That's
# not what we want, we want a module, so we add it here:
self._parsed = self.convert_node(self._grammar,
self._grammar.symbol2number['file_input'],
[self._parsed])
if self._added_newline:
self.remove_last_newline()
def get_parsed_node(self):
# TODO rename to get_root_node
return self._parsed
def error_recovery(self, grammar, stack, arcs, typ, value, start_pos, prefix,
add_token_callback):
raise ParseError
def convert_node(self, grammar, type, children):
"""
Convert raw node information to a Node instance.
This is passed to the parser driver which calls it whenever a reduction of a
grammar rule produces a new complete node, so that the tree is build
strictly bottom-up.
"""
symbol = grammar.number2symbol[type]
try:
new_node = Parser.AST_MAPPING[symbol](children)
except KeyError:
new_node = pt.Node(symbol, children)
# We need to check raw_node always, because the same node can be
# returned by convert multiple times.
if symbol == 'global_stmt':
self._global_names += new_node.get_global_names()
elif isinstance(new_node, pt.Lambda):
new_node.names_dict = self._scope_names_stack.pop()
elif isinstance(new_node, (pt.ClassOrFunc, pt.Module)) \
and symbol in ('funcdef', 'classdef', 'file_input'):
# scope_name_stack handling
scope_names = self._scope_names_stack.pop()
if isinstance(new_node, pt.ClassOrFunc):
n = new_node.name
scope_names[n.value].remove(n)
# Set the func name of the current node
arr = self._scope_names_stack[-1].setdefault(n.value, [])
arr.append(n)
new_node.names_dict = scope_names
elif isinstance(new_node, pt.CompFor):
# The name definitions of comprehenions shouldn't be part of the
# current scope. They are part of the comprehension scope.
for n in new_node.get_defined_names():
self._scope_names_stack[-1][n.value].remove(n)
return new_node
def convert_leaf(self, grammar, type, value, prefix, start_pos):
# print('leaf', repr(value), token.tok_name[type])
if type == tokenize.NAME:
if value in grammar.keywords:
if value in ('def', 'class', 'lambda'):
self._scope_names_stack.append({})
return pt.Keyword(self.position_modifier, value, start_pos, prefix)
else:
name = pt.Name(self.position_modifier, value, start_pos, prefix)
# Keep a listing of all used names
arr = self._used_names.setdefault(name.value, [])
arr.append(name)
arr = self._scope_names_stack[-1].setdefault(name.value, [])
arr.append(name)
return name
elif type == STRING:
return pt.String(self.position_modifier, value, start_pos, prefix)
elif type == NUMBER:
return pt.Number(self.position_modifier, value, start_pos, prefix)
elif type == NEWLINE:
return pt.Newline(self.position_modifier, value, start_pos, prefix)
elif type == INDENT:
return pt.Indent(self.position_modifier, value, start_pos, prefix)
elif type == DEDENT:
return pt.Dedent(self.position_modifier, value, start_pos, prefix)
elif type == ENDMARKER:
return pt.EndMarker(self.position_modifier, value, start_pos, prefix)
else:
return pt.Operator(self.position_modifier, value, start_pos, prefix)
def remove_last_newline(self):
"""
In all of this we need to work with _start_pos, because if we worked
with start_pos, we would need to check the position_modifier as well
(which is accounted for in the start_pos property).
"""
endmarker = self._parsed.children[-1]
# The newline is either in the endmarker as a prefix or the previous
# leaf as a newline token.
prefix = endmarker.prefix
if prefix.endswith('\n'):
endmarker.prefix = prefix = prefix[:-1]
last_end = 0
if '\n' not in prefix:
# Basically if the last line doesn't end with a newline. we
# have to add the previous line's end_position.
try:
last_end = endmarker.get_previous_leaf().end_pos[1]
except IndexError:
pass
last_line = re.sub('.*\n', '', prefix)
endmarker._start_pos = endmarker._start_pos[0] - 1, last_end + len(last_line)
else:
try:
newline = endmarker.get_previous_leaf()
except IndexError:
return # This means that the parser is empty.
while True:
if newline.value == '':
# Must be a DEDENT, just continue.
try:
newline = newline.get_previous_leaf()
except IndexError:
# If there's a statement that fails to be parsed, there
# will be no previous leaf. So just ignore it.
break
elif newline.value != '\n':
# TODO REMOVE, error recovery was simplified.
# This may happen if error correction strikes and removes
# a whole statement including '\n'.
break
else:
newline.value = ''
if self._last_failed_start_pos > newline._start_pos:
# It may be the case that there was a syntax error in a
# function. In that case error correction removes the
# right newline. So we use the previously assigned
# _last_failed_start_pos variable to account for that.
endmarker._start_pos = self._last_failed_start_pos
else:
endmarker._start_pos = newline._start_pos
break
class ParserWithRecovery(Parser):
"""
This class is used to parse a Python file, it then divides them into a
class structure of different scopes.
:param grammar: The grammar object of pgen2. Loaded by load_grammar.
:param source: The codebase for the parser. Must be unicode.
:param module_path: The path of the module in the file system, may be None.
:type module_path: str
"""
def __init__(self, grammar, source, module_path=None, tokenizer=None,
start_parsing=True):
self.syntax_errors = []
self._omit_dedent_list = []
self._indent_counter = 0
# TODO do print absolute import detection here.
# try:
# del python_grammar_no_print_statement.keywords["print"]
# except KeyError:
# pass # Doesn't exist in the Python 3 grammar.
# if self.options["print_function"]:
# python_grammar = pygram.python_grammar_no_print_statement
# else:
super(ParserWithRecovery, self).__init__(
grammar, source,
tokenizer=tokenizer,
start_parsing=start_parsing
)
if start_parsing:
self.module = self._parsed
self.module.used_names = self._used_names
self.module.path = module_path
self.module.global_names = self._global_names
def parse(self, tokenizer):
return super(ParserWithRecovery, self).parse(self._tokenize(self._tokenize(tokenizer)))
def error_recovery(self, grammar, stack, arcs, typ, value, start_pos, prefix,
add_token_callback):
"""
This parser is written in a dynamic way, meaning that this parser
allows using different grammars (even non-Python). However, error
recovery is purely written for Python.
"""
def current_suite(stack):
# For now just discard everything that is not a suite or
# file_input, if we detect an error.
for index, (dfa, state, (type_, nodes)) in reversed(list(enumerate(stack))):
# `suite` can sometimes be only simple_stmt, not stmt.
symbol = grammar.number2symbol[type_]
if symbol == 'file_input':
break
elif symbol == 'suite' and len(nodes) > 1:
# suites without an indent in them get discarded.
break
elif symbol == 'simple_stmt' and len(nodes) > 1:
# simple_stmt can just be turned into a Node, if there are
# enough statements. Ignore the rest after that.
break
return index, symbol, nodes
index, symbol, nodes = current_suite(stack)
if symbol == 'simple_stmt':
index -= 2
(_, _, (type_, suite_nodes)) = stack[index]
symbol = grammar.number2symbol[type_]
suite_nodes.append(pt.Node(symbol, list(nodes)))
# Remove
nodes[:] = []
nodes = suite_nodes
stack[index]
# print('err', token.tok_name[typ], repr(value), start_pos, len(stack), index)
if self._stack_removal(grammar, stack, arcs, index + 1, value, start_pos):
add_token_callback(typ, value, start_pos, prefix)
else:
if typ == INDENT:
# For every deleted INDENT we have to delete a DEDENT as well.
# Otherwise the parser will get into trouble and DEDENT too early.
self._omit_dedent_list.append(self._indent_counter)
else:
error_leaf = pt.ErrorLeaf(self.position_modifier, typ, value, start_pos, prefix)
stack[-1][2][1].append(error_leaf)
def _stack_removal(self, grammar, stack, arcs, start_index, value, start_pos):
failed_stack = []
found = False
all_nodes = []
for dfa, state, (typ, nodes) in stack[start_index:]:
if nodes:
found = True
if found:
symbol = grammar.number2symbol[typ]
failed_stack.append((symbol, nodes))
all_nodes += nodes
if nodes and nodes[0] in ('def', 'class', 'lambda'):
self._scope_names_stack.pop()
if failed_stack:
stack[start_index - 1][2][1].append(pt.ErrorNode(all_nodes))
self._last_failed_start_pos = start_pos
stack[start_index:] = []
return failed_stack
def _tokenize(self, tokenizer):
for typ, value, start_pos, prefix in tokenizer:
# print(tokenize.tok_name[typ], repr(value), start_pos, repr(prefix))
if typ == DEDENT:
# We need to count indents, because if we just omit any DEDENT,
# we might omit them in the wrong place.
o = self._omit_dedent_list
if o and o[-1] == self._indent_counter:
o.pop()
continue
self._indent_counter -= 1
elif typ == INDENT:
self._indent_counter += 1
yield typ, value, start_pos, prefix
def __repr__(self):
return "<%s: %s>" % (type(self).__name__, self.module)

View file

@ -0,0 +1,621 @@
"""
Basically a parser that is faster, because it tries to parse only parts and if
anything changes, it only reparses the changed parts. But because it's not
finished (and still not working as I want), I won't document it any further.
"""
import re
from itertools import chain
from jedi._compatibility import use_metaclass
from jedi import settings
from jedi.parser import ParserWithRecovery
from jedi.parser import tree
from jedi.parser.utils import underscore_memoization, parser_cache
from jedi import debug
from jedi.parser.tokenize import (source_tokens, NEWLINE,
ENDMARKER, INDENT, DEDENT)
FLOWS = 'if', 'else', 'elif', 'while', 'with', 'try', 'except', 'finally', 'for'
class FastModule(tree.Module):
type = 'file_input'
def __init__(self, module_path):
super(FastModule, self).__init__([])
self.modules = []
self.reset_caches()
self.names_dict = {}
self.path = module_path
def reset_caches(self):
self.modules = []
try:
del self._used_names # Remove the used names cache.
except AttributeError:
pass # It was never used.
@property
@underscore_memoization
def used_names(self):
return MergedNamesDict([m.used_names for m in self.modules])
@property
def global_names(self):
return [name for m in self.modules for name in m.global_names]
@property
def error_statements(self):
return [e for m in self.modules for e in m.error_statements]
def __repr__(self):
return "<fast.%s: %s@%s-%s>" % (type(self).__name__, self.name,
self.start_pos[0], self.end_pos[0])
# To avoid issues with with the `parser.ParserWithRecovery`, we need
# setters that do nothing, because if pickle comes along and sets those
# values.
@global_names.setter
def global_names(self, value):
pass
@error_statements.setter
def error_statements(self, value):
pass
@used_names.setter
def used_names(self, value):
pass
class MergedNamesDict(object):
def __init__(self, dicts):
self.dicts = dicts
def __iter__(self):
return iter(set(key for dct in self.dicts for key in dct))
def __getitem__(self, value):
return list(chain.from_iterable(dct.get(value, []) for dct in self.dicts))
def items(self):
dct = {}
for d in self.dicts:
for key, values in d.items():
try:
dct_values = dct[key]
dct_values += values
except KeyError:
dct[key] = list(values)
return dct.items()
def values(self):
lst = []
for dct in self.dicts:
lst += dct.values()
return lst
class CachedFastParser(type):
""" This is a metaclass for caching `FastParser`. """
def __call__(self, grammar, source, module_path=None):
if not settings.fast_parser:
return ParserWithRecovery(grammar, source, module_path)
pi = parser_cache.get(module_path, None)
if pi is None or isinstance(pi.parser, ParserWithRecovery):
p = super(CachedFastParser, self).__call__(grammar, source, module_path)
else:
p = pi.parser # pi is a `cache.ParserCacheItem`
p.update(source)
return p
class ParserNode(object):
def __init__(self, fast_module, parser, source):
self._fast_module = fast_module
self.parent = None
self._node_children = []
self.source = source
self.hash = hash(source)
self.parser = parser
if source:
self._end_pos = parser.module.end_pos
else:
self._end_pos = 1, 0
try:
# With fast_parser we have either 1 subscope or only statements.
self._content_scope = parser.module.subscopes[0]
# A parsed node's content will be in the first indent, because
# everything that's parsed is within this subscope.
self._is_class_or_def = True
except IndexError:
self._content_scope = parser.module
self._is_class_or_def = False
else:
self._rewrite_last_newline()
# We need to be able to reset the original children of a parser.
self._old_children = list(self._content_scope.children)
def is_root_node(self):
return self.parent is None
def _rewrite_last_newline(self):
"""
The ENDMARKER can contain a newline in the prefix. However this prefix
really belongs to the function - respectively to the next function or
parser node. If we don't rewrite that newline, we end up with a newline
in the wrong position, i.d. at the end of the file instead of in the
middle.
"""
c = self._content_scope.children
if tree.is_node(c[-1], 'suite'): # In a simple_stmt there's no DEDENT.
end_marker = self.parser.module.children[-1]
# Set the DEDENT prefix instead of the ENDMARKER.
c[-1].children[-1].prefix = end_marker.prefix
end_marker.prefix = ''
def __repr__(self):
module = self.parser.module
try:
return '<%s: %s-%s>' % (type(self).__name__, module.start_pos, module.end_pos)
except IndexError:
# There's no module yet.
return '<%s: empty>' % type(self).__name__
@property
def end_pos(self):
return self._end_pos[0] + self.parser.position_modifier.line, self._end_pos[1]
def reset_node(self):
"""
Removes changes that were applied in this class.
"""
self._node_children = []
scope = self._content_scope
scope.children = list(self._old_children)
try:
# This works if it's a MergedNamesDict.
# We are correcting it, because the MergedNamesDicts are artificial
# and can change after closing a node.
scope.names_dict = scope.names_dict.dicts[0]
except AttributeError:
pass
def close(self):
"""
Closes the current parser node. This means that after this no further
nodes should be added anymore.
"""
# We only need to replace the dict if multiple dictionaries are used:
if self._node_children:
dcts = [n.parser.module.names_dict for n in self._node_children]
# Need to insert the own node as well.
dcts.insert(0, self._content_scope.names_dict)
self._content_scope.names_dict = MergedNamesDict(dcts)
endmarker = self.parser.get_parsed_node().children[-1]
assert endmarker.type == 'endmarker'
last_parser = self._node_children[-1].parser
endmarker.start_pos = last_parser.get_parsed_node().end_pos
@property
def _indent(self):
if self.is_root_node():
return 0
return self.parser.module.children[0].start_pos[1]
def add_node(self, node, start_line, indent):
"""
Adding a node means adding a node that was either just parsed or one
that can be reused.
"""
# Content that is not a subscope can never be part of the current node,
# because it's basically a sister node, that sits next to it and not
# within it.
if (self._indent >= indent or not self._is_class_or_def) and \
not self.is_root_node():
self.close()
return self.parent.add_node(node, start_line, indent)
# Changing the line offsets is very important, because if they don't
# fit, all the start_pos values will be wrong.
m = node.parser.module
node.parser.position_modifier.line = start_line - 1
self._fast_module.modules.append(m)
node.parent = self
self._node_children.append(node)
# Insert parser objects into current structure. We only need to set the
# parents and children in a good way.
scope = self._content_scope
for child in m.children:
child.parent = scope
scope.children.append(child)
return node
def all_sub_nodes(self):
"""
Returns all nodes including nested ones.
"""
for n in self._node_children:
yield n
for y in n.all_sub_nodes():
yield y
@underscore_memoization # Should only happen once!
def remove_last_newline(self):
self.parser.remove_last_newline()
class FastParser(use_metaclass(CachedFastParser)):
_FLOWS_NEED_SPACE = 'if', 'elif', 'while', 'with', 'except', 'for'
_FLOWS_NEED_COLON = 'else', 'try', 'except', 'finally'
_keyword_re = re.compile('^[ \t]*(def |class |@|(?:%s)|(?:%s)\s*:)'
% ('|'.join(_FLOWS_NEED_SPACE),
'|'.join(_FLOWS_NEED_COLON)))
def __init__(self, grammar, source, module_path=None):
# set values like `tree.Module`.
self._grammar = grammar
self.module_path = module_path
self._reset_caches()
self.update(source)
def _reset_caches(self):
self.module = FastModule(self.module_path)
self.root_node = self.current_node = ParserNode(self.module, self, '')
def get_parsed_node(self):
return self.module
def update(self, source):
# Variables for testing purposes: It is important that the number of
# parsers used can be minimized. With these variables we can test
# against that.
self.number_parsers_used = 0
self.number_of_splits = 0
self.number_of_misses = 0
self.module.reset_caches()
self.source = source
try:
self._parse(source)
except:
# FastParser is cached, be careful with exceptions.
self._reset_caches()
raise
def _split_parts(self, source):
"""
Split the source code into different parts. This makes it possible to
parse each part seperately and therefore cache parts of the file and
not everything.
"""
def gen_part():
text = ''.join(current_lines)
del current_lines[:]
self.number_of_splits += 1
return text
def just_newlines(current_lines):
for line in current_lines:
line = line.lstrip('\t \n\r')
if line and line[0] != '#':
return False
return True
# Split only new lines. Distinction between \r\n is the tokenizer's
# job.
# It seems like there's no problem with form feed characters here,
# because we're not counting lines.
self._lines = source.splitlines(True)
current_lines = []
is_decorator = False
# Use -1, because that indent is always smaller than any other.
indent_list = [-1, 0]
new_indent = False
parentheses_level = 0
flow_indent = None
previous_line = None
# All things within flows are simply being ignored.
for i, l in enumerate(self._lines):
# Handle backslash newline escaping.
if l.endswith('\\\n') or l.endswith('\\\r\n'):
if previous_line is not None:
previous_line += l
else:
previous_line = l
continue
if previous_line is not None:
l = previous_line + l
previous_line = None
# check for dedents
s = l.lstrip('\t \n\r')
indent = len(l) - len(s)
if not s or s[0] == '#':
current_lines.append(l) # Just ignore comments and blank lines
continue
if new_indent and not parentheses_level:
if indent > indent_list[-2]:
# Set the actual indent, not just the random old indent + 1.
indent_list[-1] = indent
new_indent = False
while indent < indent_list[-1]: # -> dedent
indent_list.pop()
# This automatically resets the flow_indent if there was a
# dedent or a flow just on one line (with one simple_stmt).
new_indent = False
if flow_indent is None and current_lines and not parentheses_level:
yield gen_part()
flow_indent = None
# Check lines for functions/classes and split the code there.
if flow_indent is None:
m = self._keyword_re.match(l)
if m:
# Strip whitespace and colon from flows as a check.
if m.group(1).strip(' \t\r\n:') in FLOWS:
if not parentheses_level:
flow_indent = indent
else:
if not is_decorator and not just_newlines(current_lines):
yield gen_part()
is_decorator = '@' == m.group(1)
if not is_decorator:
parentheses_level = 0
# The new indent needs to be higher
indent_list.append(indent + 1)
new_indent = True
elif is_decorator:
is_decorator = False
parentheses_level = \
max(0, (l.count('(') + l.count('[') + l.count('{') -
l.count(')') - l.count(']') - l.count('}')))
current_lines.append(l)
if previous_line is not None:
current_lines.append(previous_line)
if current_lines:
yield gen_part()
def _parse(self, source):
""" :type source: str """
added_newline = False
if not source or source[-1] != '\n':
# To be compatible with Pythons grammar, we need a newline at the
# end. The parser would handle it, but since the fast parser abuses
# the normal parser in various ways, we need to care for this
# ourselves.
source += '\n'
added_newline = True
next_code_part_end_line = code_part_end_line = 1
start = 0
nodes = list(self.root_node.all_sub_nodes())
# Now we can reset the node, because we have all the old nodes.
self.root_node.reset_node()
self.current_node = self.root_node
last_end_line = 1
for code_part in self._split_parts(source):
next_code_part_end_line += code_part.count('\n')
# If the last code part parsed isn't equal to the current end_pos,
# we know that the parser went further (`def` start in a
# docstring). So just parse the next part.
if code_part_end_line == last_end_line:
self._parse_part(code_part, source[start:], code_part_end_line, nodes)
else:
self.number_of_misses += 1
# Means that some lines where not fully parsed. Parse it now.
# This is a very rare case. Should only happens with very
# strange code bits.
while last_end_line < next_code_part_end_line:
code_part_end_line = last_end_line
# We could calculate the src in a more complicated way to
# make caching here possible as well. However, this is
# complicated and error-prone. Since this is not very often
# called - just ignore it.
src = ''.join(self._lines[code_part_end_line - 1:])
self._parse_part(code_part, src, code_part_end_line, nodes)
last_end_line = self.current_node.end_pos[0]
debug.dbg("While parsing %s, starting with line %s wasn't included in split.",
self.module_path, code_part_end_line)
#assert code_part_end_line > last_end_line
# This means that the parser parsed faster than the last given
# `code_part`.
debug.dbg('While parsing %s, line %s slowed down the fast parser.',
self.module_path, code_part_end_line)
code_part_end_line = next_code_part_end_line
start += len(code_part)
last_end_line = self.current_node.end_pos[0]
if added_newline:
self.current_node.remove_last_newline()
# Now that the for loop is finished, we still want to close all nodes.
node = self.current_node
while node is not None:
node.close()
node = node.parent
debug.dbg('Parsed %s, with %s parsers in %s splits.'
% (self.module_path, self.number_parsers_used,
self.number_of_splits))
def _parse_part(self, source, parser_code, code_part_end_line, nodes):
"""
Side effect: Alters the list of nodes.
"""
h = hash(source)
for index, node in enumerate(nodes):
if node.hash == h and node.source == source:
node.reset_node()
nodes.remove(node)
parser_code = source
break
else:
tokenizer = FastTokenizer(parser_code)
self.number_parsers_used += 1
p = ParserWithRecovery(self._grammar, parser_code, self.module_path, tokenizer=tokenizer)
end = code_part_end_line - 1 + p.module.end_pos[0]
used_lines = self._lines[code_part_end_line - 1:end - 1]
code_part_actually_used = ''.join(used_lines)
node = ParserNode(self.module, p, code_part_actually_used)
indent = len(parser_code) - len(parser_code.lstrip('\t '))
self.current_node.add_node(node, code_part_end_line, indent)
self.current_node = node
class FastTokenizer(object):
"""
Breaks when certain conditions are met, i.e. a new function or class opens.
"""
def __init__(self, source):
self.source = source
self._gen = source_tokens(source, use_exact_op_types=True)
self._closed = False
# fast parser options
self.current = self.previous = NEWLINE, '', (0, 0)
self._in_flow = False
self._is_decorator = False
self._first_stmt = True
self._parentheses_level = 0
self._indent_counter = 0
self._flow_indent_counter = 0
self._returned_endmarker = False
self._expect_indent = False
def __iter__(self):
return self
def next(self):
""" Python 2 Compatibility """
return self.__next__()
def __next__(self):
if self._closed:
return self._finish_dedents()
typ, value, start_pos, prefix = current = next(self._gen)
if typ == ENDMARKER:
self._closed = True
self._returned_endmarker = True
return current
self.previous = self.current
self.current = current
if typ == INDENT:
self._indent_counter += 1
if not self._expect_indent and not self._first_stmt and not self._in_flow:
# This does not mean that there is an actual flow, it means
# that the INDENT is syntactically wrong.
self._flow_indent_counter = self._indent_counter - 1
self._in_flow = True
self._expect_indent = False
elif typ == DEDENT:
self._indent_counter -= 1
if self._in_flow:
if self._indent_counter == self._flow_indent_counter:
self._in_flow = False
else:
self._closed = True
return current
previous_type = self.previous[0]
if value in ('def', 'class') and self._parentheses_level:
# Account for the fact that an open parentheses before a function
# will reset the parentheses counter, but new lines before will
# still be ignored. So check the prefix.
# TODO what about flow parentheses counter resets in the tokenizer?
self._parentheses_level = 0
# We need to simulate a newline before the indent, because the
# open parentheses ignored them.
if re.search('\n\s*', prefix):
previous_type = NEWLINE
# Parentheses ignore the indentation rules. The other three stand for
# new lines.
if previous_type in (NEWLINE, INDENT, DEDENT) \
and not self._parentheses_level and typ not in (INDENT, DEDENT):
if not self._in_flow:
if value in FLOWS:
self._flow_indent_counter = self._indent_counter
self._first_stmt = False
elif value in ('def', 'class', '@'):
# The values here are exactly the same check as in
# _split_parts, but this time with tokenize and therefore
# precise.
if not self._first_stmt and not self._is_decorator:
return self._close()
self._is_decorator = '@' == value
if not self._is_decorator:
self._first_stmt = False
self._expect_indent = True
elif self._expect_indent:
return self._close()
else:
self._first_stmt = False
if value in '([{' and value:
self._parentheses_level += 1
elif value in ')]}' and value:
# Ignore closing parentheses, because they are all
# irrelevant for the indentation.
self._parentheses_level = max(self._parentheses_level - 1, 0)
return current
def _close(self):
if self._first_stmt:
# Continue like nothing has happened, because we want to enter
# the first class/function.
if self.current[1] != '@':
self._first_stmt = False
return self.current
else:
self._closed = True
return self._finish_dedents()
def _finish_dedents(self):
if self._indent_counter:
self._indent_counter -= 1
return DEDENT, '', self.current[2], ''
elif not self._returned_endmarker:
self._returned_endmarker = True
return ENDMARKER, '', self.current[2], self._get_prefix()
else:
raise StopIteration
def _get_prefix(self):
"""
We're using the current prefix for the endmarker to not loose any
information. However we care about "lost" lines. The prefix of the
current line (indent) will always be included in the current line.
"""
cur = self.current
while cur[0] == DEDENT:
cur = next(self._gen)
prefix = cur[3]
# \Z for the end of the string. $ is bugged, because it has the
# same behavior with or without re.MULTILINE.
return re.sub(r'[^\n]+\Z', '', prefix)

View file

@ -0,0 +1,152 @@
# Grammar for 2to3. This grammar supports Python 2.x and 3.x.
# Note: Changing the grammar specified in this file will most likely
# require corresponding changes in the parser module
# (../Modules/parsermodule.c). If you can't make the changes to
# that module yourself, please co-ordinate the required changes
# with someone who can; ask around on python-dev for help. Fred
# Drake <fdrake@acm.org> will probably be listening there.
# NOTE WELL: You should also follow all the steps listed in PEP 306,
# "How to Change Python's Grammar"
# Start symbols for the grammar:
# file_input is a module or sequence of commands read from an input file;
# single_input is a single interactive statement;
# eval_input is the input for the eval() and input() functions.
# NB: compound_stmt in single_input is followed by extra NEWLINE!
file_input: (NEWLINE | stmt)* ENDMARKER
single_input: NEWLINE | simple_stmt | compound_stmt NEWLINE
eval_input: testlist NEWLINE* ENDMARKER
decorator: '@' dotted_name [ '(' [arglist] ')' ] NEWLINE
decorators: decorator+
decorated: decorators (classdef | funcdef)
funcdef: 'def' NAME parameters ['->' test] ':' suite
parameters: '(' [typedargslist] ')'
typedargslist: ((tfpdef ['=' test] ',')*
('*' [tname] (',' tname ['=' test])* [',' '**' tname] | '**' tname)
| tfpdef ['=' test] (',' tfpdef ['=' test])* [','])
tname: NAME [':' test]
tfpdef: tname | '(' tfplist ')'
tfplist: tfpdef (',' tfpdef)* [',']
varargslist: ((vfpdef ['=' test] ',')*
('*' [vname] (',' vname ['=' test])* [',' '**' vname] | '**' vname)
| vfpdef ['=' test] (',' vfpdef ['=' test])* [','])
vname: NAME
vfpdef: vname | '(' vfplist ')'
vfplist: vfpdef (',' vfpdef)* [',']
stmt: simple_stmt | compound_stmt
simple_stmt: small_stmt (';' small_stmt)* [';'] NEWLINE
small_stmt: (expr_stmt | print_stmt | del_stmt | pass_stmt | flow_stmt |
import_stmt | global_stmt | exec_stmt | assert_stmt)
expr_stmt: testlist_star_expr (augassign (yield_expr|testlist) |
('=' (yield_expr|testlist_star_expr))*)
testlist_star_expr: (test|star_expr) (',' (test|star_expr))* [',']
augassign: ('+=' | '-=' | '*=' | '/=' | '%=' | '&=' | '|=' | '^=' |
'<<=' | '>>=' | '**=' | '//=')
# For normal assignments, additional restrictions enforced by the interpreter
print_stmt: 'print' ( [ test (',' test)* [','] ] |
'>>' test [ (',' test)+ [','] ] )
del_stmt: 'del' exprlist
pass_stmt: 'pass'
flow_stmt: break_stmt | continue_stmt | return_stmt | raise_stmt | yield_stmt
break_stmt: 'break'
continue_stmt: 'continue'
return_stmt: 'return' [testlist]
yield_stmt: yield_expr
raise_stmt: 'raise' [test [',' test [',' test]]]
import_stmt: import_name | import_from
import_name: 'import' dotted_as_names
# note below: the ('.' | '...') is necessary because '...' is tokenized as ELLIPSIS
import_from: ('from' (('.' | '...')* dotted_name | ('.' | '...')+)
'import' ('*' | '(' import_as_names ')' | import_as_names))
import_as_name: NAME ['as' NAME]
dotted_as_name: dotted_name ['as' NAME]
import_as_names: import_as_name (',' import_as_name)* [',']
dotted_as_names: dotted_as_name (',' dotted_as_name)*
dotted_name: NAME ('.' NAME)*
global_stmt: 'global' NAME (',' NAME)*
exec_stmt: 'exec' expr ['in' test [',' test]]
assert_stmt: 'assert' test [',' test]
compound_stmt: if_stmt | while_stmt | for_stmt | try_stmt | with_stmt | funcdef | classdef | decorated
if_stmt: 'if' test ':' suite ('elif' test ':' suite)* ['else' ':' suite]
while_stmt: 'while' test ':' suite ['else' ':' suite]
for_stmt: 'for' exprlist 'in' testlist ':' suite ['else' ':' suite]
try_stmt: ('try' ':' suite
((except_clause ':' suite)+
['else' ':' suite]
['finally' ':' suite] |
'finally' ':' suite))
with_stmt: 'with' with_item (',' with_item)* ':' suite
with_item: test ['as' expr]
with_var: 'as' expr
# NB compile.c makes sure that the default except clause is last
except_clause: 'except' [test [(',' | 'as') test]]
# Edit by David Halter: The stmt is now optional. This reflects how Jedi allows
# classes and functions to be empty, which is beneficial for autocompletion.
suite: simple_stmt | NEWLINE INDENT stmt* DEDENT
# Backward compatibility cruft to support:
# [ x for x in lambda: True, lambda: False if x() ]
# even while also allowing:
# lambda x: 5 if x else 2
# (But not a mix of the two)
testlist_safe: old_test [(',' old_test)+ [',']]
old_test: or_test | old_lambdef
old_lambdef: 'lambda' [varargslist] ':' old_test
test: or_test ['if' or_test 'else' test] | lambdef
or_test: and_test ('or' and_test)*
and_test: not_test ('and' not_test)*
not_test: 'not' not_test | comparison
comparison: expr (comp_op expr)*
comp_op: '<'|'>'|'=='|'>='|'<='|'<>'|'!='|'in'|'not' 'in'|'is'|'is' 'not'
star_expr: '*' expr
expr: xor_expr ('|' xor_expr)*
xor_expr: and_expr ('^' and_expr)*
and_expr: shift_expr ('&' shift_expr)*
shift_expr: arith_expr (('<<'|'>>') arith_expr)*
arith_expr: term (('+'|'-') term)*
term: factor (('*'|'/'|'%'|'//') factor)*
factor: ('+'|'-'|'~') factor | power
power: atom trailer* ['**' factor]
atom: ('(' [yield_expr|testlist_comp] ')' |
'[' [testlist_comp] ']' |
'{' [dictorsetmaker] '}' |
'`' testlist1 '`' |
NAME | NUMBER | STRING+ | '.' '.' '.')
# Modification by David Halter, remove `testlist_gexp` and `listmaker`
testlist_comp: (test|star_expr) ( comp_for | (',' (test|star_expr))* [','] )
lambdef: 'lambda' [varargslist] ':' test
trailer: '(' [arglist] ')' | '[' subscriptlist ']' | '.' NAME
subscriptlist: subscript (',' subscript)* [',']
subscript: test | [test] ':' [test] [sliceop]
sliceop: ':' [test]
exprlist: (expr|star_expr) (',' (expr|star_expr))* [',']
testlist: test (',' test)* [',']
# Modification by David Halter, dictsetmaker -> dictorsetmaker (so that it's
# the same as in the 3.4 grammar).
dictorsetmaker: ( (test ':' test (comp_for | (',' test ':' test)* [','])) |
(test (comp_for | (',' test)* [','])) )
classdef: 'class' NAME ['(' [arglist] ')'] ':' suite
arglist: (argument ',')* (argument [',']
|'*' test (',' argument)* [',' '**' test]
|'**' test)
argument: test [comp_for] | test '=' test # Really [keyword '='] test
comp_iter: comp_for | comp_if
comp_for: 'for' exprlist 'in' testlist_safe [comp_iter]
comp_if: 'if' old_test [comp_iter]
testlist1: test (',' test)*
# not used in grammar, but may appear in "node" passed from Parser to Compiler
encoding_decl: NAME
yield_expr: 'yield' [testlist]

View file

@ -0,0 +1,135 @@
# Grammar for Python
# Note: Changing the grammar specified in this file will most likely
# require corresponding changes in the parser module
# (../Modules/parsermodule.c). If you can't make the changes to
# that module yourself, please co-ordinate the required changes
# with someone who can; ask around on python-dev for help. Fred
# Drake <fdrake@acm.org> will probably be listening there.
# NOTE WELL: You should also follow all the steps listed in PEP 306,
# "How to Change Python's Grammar"
# Start symbols for the grammar:
# single_input is a single interactive statement;
# file_input is a module or sequence of commands read from an input file;
# eval_input is the input for the eval() functions.
# NB: compound_stmt in single_input is followed by extra NEWLINE!
file_input: (NEWLINE | stmt)* ENDMARKER
single_input: NEWLINE | simple_stmt | compound_stmt NEWLINE
eval_input: testlist NEWLINE* ENDMARKER
decorator: '@' dotted_name [ '(' [arglist] ')' ] NEWLINE
decorators: decorator+
decorated: decorators (classdef | funcdef)
funcdef: 'def' NAME parameters ['->' test] ':' suite
parameters: '(' [typedargslist] ')'
typedargslist: (tfpdef ['=' test] (',' tfpdef ['=' test])* [','
['*' [tfpdef] (',' tfpdef ['=' test])* [',' '**' tfpdef] | '**' tfpdef]]
| '*' [tfpdef] (',' tfpdef ['=' test])* [',' '**' tfpdef] | '**' tfpdef)
tfpdef: NAME [':' test]
varargslist: (vfpdef ['=' test] (',' vfpdef ['=' test])* [','
['*' [vfpdef] (',' vfpdef ['=' test])* [',' '**' vfpdef] | '**' vfpdef]]
| '*' [vfpdef] (',' vfpdef ['=' test])* [',' '**' vfpdef] | '**' vfpdef)
vfpdef: NAME
stmt: simple_stmt | compound_stmt
simple_stmt: small_stmt (';' small_stmt)* [';'] NEWLINE
small_stmt: (expr_stmt | del_stmt | pass_stmt | flow_stmt |
import_stmt | global_stmt | nonlocal_stmt | assert_stmt)
expr_stmt: testlist_star_expr (augassign (yield_expr|testlist) |
('=' (yield_expr|testlist_star_expr))*)
testlist_star_expr: (test|star_expr) (',' (test|star_expr))* [',']
augassign: ('+=' | '-=' | '*=' | '/=' | '%=' | '&=' | '|=' | '^=' |
'<<=' | '>>=' | '**=' | '//=')
# For normal assignments, additional restrictions enforced by the interpreter
del_stmt: 'del' exprlist
pass_stmt: 'pass'
flow_stmt: break_stmt | continue_stmt | return_stmt | raise_stmt | yield_stmt
break_stmt: 'break'
continue_stmt: 'continue'
return_stmt: 'return' [testlist]
yield_stmt: yield_expr
raise_stmt: 'raise' [test ['from' test]]
import_stmt: import_name | import_from
import_name: 'import' dotted_as_names
# note below: the ('.' | '...') is necessary because '...' is tokenized as ELLIPSIS
import_from: ('from' (('.' | '...')* dotted_name | ('.' | '...')+)
'import' ('*' | '(' import_as_names ')' | import_as_names))
import_as_name: NAME ['as' NAME]
dotted_as_name: dotted_name ['as' NAME]
import_as_names: import_as_name (',' import_as_name)* [',']
dotted_as_names: dotted_as_name (',' dotted_as_name)*
dotted_name: NAME ('.' NAME)*
global_stmt: 'global' NAME (',' NAME)*
nonlocal_stmt: 'nonlocal' NAME (',' NAME)*
assert_stmt: 'assert' test [',' test]
compound_stmt: if_stmt | while_stmt | for_stmt | try_stmt | with_stmt | funcdef | classdef | decorated
if_stmt: 'if' test ':' suite ('elif' test ':' suite)* ['else' ':' suite]
while_stmt: 'while' test ':' suite ['else' ':' suite]
for_stmt: 'for' exprlist 'in' testlist ':' suite ['else' ':' suite]
try_stmt: ('try' ':' suite
((except_clause ':' suite)+
['else' ':' suite]
['finally' ':' suite] |
'finally' ':' suite))
with_stmt: 'with' with_item (',' with_item)* ':' suite
with_item: test ['as' expr]
# NB compile.c makes sure that the default except clause is last
except_clause: 'except' [test ['as' NAME]]
# Edit by David Halter: The stmt is now optional. This reflects how Jedi allows
# classes and functions to be empty, which is beneficial for autocompletion.
suite: simple_stmt | NEWLINE INDENT stmt* DEDENT
test: or_test ['if' or_test 'else' test] | lambdef
test_nocond: or_test | lambdef_nocond
lambdef: 'lambda' [varargslist] ':' test
lambdef_nocond: 'lambda' [varargslist] ':' test_nocond
or_test: and_test ('or' and_test)*
and_test: not_test ('and' not_test)*
not_test: 'not' not_test | comparison
comparison: expr (comp_op expr)*
# <> isn't actually a valid comparison operator in Python. It's here for the
# sake of a __future__ import described in PEP 401
comp_op: '<'|'>'|'=='|'>='|'<='|'<>'|'!='|'in'|'not' 'in'|'is'|'is' 'not'
star_expr: '*' expr
expr: xor_expr ('|' xor_expr)*
xor_expr: and_expr ('^' and_expr)*
and_expr: shift_expr ('&' shift_expr)*
shift_expr: arith_expr (('<<'|'>>') arith_expr)*
arith_expr: term (('+'|'-') term)*
term: factor (('*'|'/'|'%'|'//') factor)*
factor: ('+'|'-'|'~') factor | power
power: atom trailer* ['**' factor]
atom: ('(' [yield_expr|testlist_comp] ')' |
'[' [testlist_comp] ']' |
'{' [dictorsetmaker] '}' |
NAME | NUMBER | STRING+ | '...' | 'None' | 'True' | 'False')
testlist_comp: (test|star_expr) ( comp_for | (',' (test|star_expr))* [','] )
trailer: '(' [arglist] ')' | '[' subscriptlist ']' | '.' NAME
subscriptlist: subscript (',' subscript)* [',']
subscript: test | [test] ':' [test] [sliceop]
sliceop: ':' [test]
exprlist: (expr|star_expr) (',' (expr|star_expr))* [',']
testlist: test (',' test)* [',']
dictorsetmaker: ( (test ':' test (comp_for | (',' test ':' test)* [','])) |
(test (comp_for | (',' test)* [','])) )
classdef: 'class' NAME ['(' [arglist] ')'] ':' suite
arglist: (argument ',')* (argument [',']
|'*' test (',' argument)* [',' '**' test]
|'**' test)
# The reason that keywords are test nodes instead of NAME is that using NAME
# results in an ambiguity. ast.c makes sure it's a NAME.
argument: test [comp_for] | test '=' test # Really [keyword '='] test
comp_iter: comp_for | comp_if
comp_for: 'for' exprlist 'in' or_test [comp_iter]
comp_if: 'if' test_nocond [comp_iter]
# not used in grammar, but may appear in "node" passed from Parser to Compiler
encoding_decl: NAME
yield_expr: 'yield' [yield_arg]
yield_arg: 'from' test | testlist

View file

@ -0,0 +1,154 @@
# Grammar for Python
# Note: Changing the grammar specified in this file will most likely
# require corresponding changes in the parser module
# (../Modules/parsermodule.c). If you can't make the changes to
# that module yourself, please co-ordinate the required changes
# with someone who can; ask around on python-dev for help. Fred
# Drake <fdrake@acm.org> will probably be listening there.
# NOTE WELL: You should also follow all the steps listed at
# https://docs.python.org/devguide/grammar.html
# Start symbols for the grammar:
# single_input is a single interactive statement;
# file_input is a module or sequence of commands read from an input file;
# eval_input is the input for the eval() functions.
# NB: compound_stmt in single_input is followed by extra NEWLINE!
file_input: (NEWLINE | stmt)* ENDMARKER
single_input: NEWLINE | simple_stmt | compound_stmt NEWLINE
eval_input: testlist NEWLINE* ENDMARKER
decorator: '@' dotted_name [ '(' [arglist] ')' ] NEWLINE
decorators: decorator+
decorated: decorators (classdef | funcdef | async_funcdef)
# NOTE: Reinoud Elhorst, using ASYNC/AWAIT keywords instead of tokens
# skipping python3.5 compatibility, in favour of 3.7 solution
async_funcdef: 'async' funcdef
funcdef: 'def' NAME parameters ['->' test] ':' suite
parameters: '(' [typedargslist] ')'
typedargslist: (tfpdef ['=' test] (',' tfpdef ['=' test])* [','
['*' [tfpdef] (',' tfpdef ['=' test])* [',' '**' tfpdef] | '**' tfpdef]]
| '*' [tfpdef] (',' tfpdef ['=' test])* [',' '**' tfpdef] | '**' tfpdef)
tfpdef: NAME [':' test]
varargslist: (vfpdef ['=' test] (',' vfpdef ['=' test])* [','
['*' [vfpdef] (',' vfpdef ['=' test])* [',' '**' vfpdef] | '**' vfpdef]]
| '*' [vfpdef] (',' vfpdef ['=' test])* [',' '**' vfpdef] | '**' vfpdef)
vfpdef: NAME
stmt: simple_stmt | compound_stmt
simple_stmt: small_stmt (';' small_stmt)* [';'] NEWLINE
small_stmt: (expr_stmt | del_stmt | pass_stmt | flow_stmt |
import_stmt | global_stmt | nonlocal_stmt | assert_stmt)
expr_stmt: testlist_star_expr (augassign (yield_expr|testlist) |
('=' (yield_expr|testlist_star_expr))*)
testlist_star_expr: (test|star_expr) (',' (test|star_expr))* [',']
augassign: ('+=' | '-=' | '*=' | '@=' | '/=' | '%=' | '&=' | '|=' | '^=' |
'<<=' | '>>=' | '**=' | '//=')
# For normal assignments, additional restrictions enforced by the interpreter
del_stmt: 'del' exprlist
pass_stmt: 'pass'
flow_stmt: break_stmt | continue_stmt | return_stmt | raise_stmt | yield_stmt
break_stmt: 'break'
continue_stmt: 'continue'
return_stmt: 'return' [testlist]
yield_stmt: yield_expr
raise_stmt: 'raise' [test ['from' test]]
import_stmt: import_name | import_from
import_name: 'import' dotted_as_names
# note below: the ('.' | '...') is necessary because '...' is tokenized as ELLIPSIS
import_from: ('from' (('.' | '...')* dotted_name | ('.' | '...')+)
'import' ('*' | '(' import_as_names ')' | import_as_names))
import_as_name: NAME ['as' NAME]
dotted_as_name: dotted_name ['as' NAME]
import_as_names: import_as_name (',' import_as_name)* [',']
dotted_as_names: dotted_as_name (',' dotted_as_name)*
dotted_name: NAME ('.' NAME)*
global_stmt: 'global' NAME (',' NAME)*
nonlocal_stmt: 'nonlocal' NAME (',' NAME)*
assert_stmt: 'assert' test [',' test]
compound_stmt: if_stmt | while_stmt | for_stmt | try_stmt | with_stmt | funcdef | classdef | decorated | async_stmt
async_stmt: 'async' (funcdef | with_stmt | for_stmt)
if_stmt: 'if' test ':' suite ('elif' test ':' suite)* ['else' ':' suite]
while_stmt: 'while' test ':' suite ['else' ':' suite]
for_stmt: 'for' exprlist 'in' testlist ':' suite ['else' ':' suite]
try_stmt: ('try' ':' suite
((except_clause ':' suite)+
['else' ':' suite]
['finally' ':' suite] |
'finally' ':' suite))
with_stmt: 'with' with_item (',' with_item)* ':' suite
with_item: test ['as' expr]
# NB compile.c makes sure that the default except clause is last
except_clause: 'except' [test ['as' NAME]]
# Edit by David Halter: The stmt is now optional. This reflects how Jedi allows
# classes and functions to be empty, which is beneficial for autocompletion.
suite: simple_stmt | NEWLINE INDENT stmt* DEDENT
test: or_test ['if' or_test 'else' test] | lambdef
test_nocond: or_test | lambdef_nocond
lambdef: 'lambda' [varargslist] ':' test
lambdef_nocond: 'lambda' [varargslist] ':' test_nocond
or_test: and_test ('or' and_test)*
and_test: not_test ('and' not_test)*
not_test: 'not' not_test | comparison
comparison: expr (comp_op expr)*
# <> isn't actually a valid comparison operator in Python. It's here for the
# sake of a __future__ import described in PEP 401 (which really works :-)
comp_op: '<'|'>'|'=='|'>='|'<='|'<>'|'!='|'in'|'not' 'in'|'is'|'is' 'not'
star_expr: '*' expr
expr: xor_expr ('|' xor_expr)*
xor_expr: and_expr ('^' and_expr)*
and_expr: shift_expr ('&' shift_expr)*
shift_expr: arith_expr (('<<'|'>>') arith_expr)*
arith_expr: term (('+'|'-') term)*
term: factor (('*'|'@'|'/'|'%'|'//') factor)*
factor: ('+'|'-'|'~') factor | power
power: atom_expr ['**' factor]
atom_expr: ['await'] atom trailer*
atom: ('(' [yield_expr|testlist_comp] ')' |
'[' [testlist_comp] ']' |
'{' [dictorsetmaker] '}' |
NAME | NUMBER | STRING+ | '...' | 'None' | 'True' | 'False')
testlist_comp: (test|star_expr) ( comp_for | (',' (test|star_expr))* [','] )
trailer: '(' [arglist] ')' | '[' subscriptlist ']' | '.' NAME
subscriptlist: subscript (',' subscript)* [',']
subscript: test | [test] ':' [test] [sliceop]
sliceop: ':' [test]
exprlist: (expr|star_expr) (',' (expr|star_expr))* [',']
testlist: test (',' test)* [',']
dictorsetmaker: ( ((test ':' test | '**' expr)
(comp_for | (',' (test ':' test | '**' expr))* [','])) |
((test | star_expr)
(comp_for | (',' (test | star_expr))* [','])) )
classdef: 'class' NAME ['(' [arglist] ')'] ':' suite
arglist: argument (',' argument)* [',']
# The reason that keywords are test nodes instead of NAME is that using NAME
# results in an ambiguity. ast.c makes sure it's a NAME.
# "test '=' test" is really "keyword '=' test", but we have no such token.
# These need to be in a single rule to avoid grammar that is ambiguous
# to our LL(1) parser. Even though 'test' includes '*expr' in star_expr,
# we explicitly match '*' here, too, to give it proper precedence.
# Illegal combinations and orderings are blocked in ast.c:
# multiple (test comp_for) arguements are blocked; keyword unpackings
# that precede iterable unpackings are blocked; etc.
argument: ( test [comp_for] |
test '=' test |
'**' test |
'*' test )
comp_iter: comp_for | comp_if
comp_for: 'for' exprlist 'in' or_test [comp_iter]
comp_if: 'if' test_nocond [comp_iter]
# not used in grammar, but may appear in "node" passed from Parser to Compiler
encoding_decl: NAME
yield_expr: 'yield' [yield_arg]
yield_arg: 'from' test | testlist

View file

@ -0,0 +1,8 @@
# Copyright 2004-2005 Elemental Security, Inc. All Rights Reserved.
# Licensed to PSF under a Contributor Agreement.
# Modifications:
# Copyright 2006 Google, Inc. All Rights Reserved.
# Licensed to PSF under a Contributor Agreement.
# Copyright 2014 David Halter. Integration into Jedi.
# Modifications are dual-licensed: MIT and PSF.

View file

@ -0,0 +1,125 @@
# Copyright 2004-2005 Elemental Security, Inc. All Rights Reserved.
# Licensed to PSF under a Contributor Agreement.
# Modifications:
# Copyright 2014 David Halter. Integration into Jedi.
# Modifications are dual-licensed: MIT and PSF.
"""This module defines the data structures used to represent a grammar.
These are a bit arcane because they are derived from the data
structures used by Python's 'pgen' parser generator.
There's also a table here mapping operators to their names in the
token module; the Python tokenize module reports all operators as the
fallback token code OP, but the parser needs the actual token code.
"""
# Python imports
import pickle
class Grammar(object):
"""Pgen parsing tables conversion class.
Once initialized, this class supplies the grammar tables for the
parsing engine implemented by parse.py. The parsing engine
accesses the instance variables directly. The class here does not
provide initialization of the tables; several subclasses exist to
do this (see the conv and pgen modules).
The load() method reads the tables from a pickle file, which is
much faster than the other ways offered by subclasses. The pickle
file is written by calling dump() (after loading the grammar
tables using a subclass). The report() method prints a readable
representation of the tables to stdout, for debugging.
The instance variables are as follows:
symbol2number -- a dict mapping symbol names to numbers. Symbol
numbers are always 256 or higher, to distinguish
them from token numbers, which are between 0 and
255 (inclusive).
number2symbol -- a dict mapping numbers to symbol names;
these two are each other's inverse.
states -- a list of DFAs, where each DFA is a list of
states, each state is a list of arcs, and each
arc is a (i, j) pair where i is a label and j is
a state number. The DFA number is the index into
this list. (This name is slightly confusing.)
Final states are represented by a special arc of
the form (0, j) where j is its own state number.
dfas -- a dict mapping symbol numbers to (DFA, first)
pairs, where DFA is an item from the states list
above, and first is a set of tokens that can
begin this grammar rule (represented by a dict
whose values are always 1).
labels -- a list of (x, y) pairs where x is either a token
number or a symbol number, and y is either None
or a string; the strings are keywords. The label
number is the index in this list; label numbers
are used to mark state transitions (arcs) in the
DFAs.
start -- the number of the grammar's start symbol.
keywords -- a dict mapping keyword strings to arc labels.
tokens -- a dict mapping token numbers to arc labels.
"""
def __init__(self):
self.symbol2number = {}
self.number2symbol = {}
self.states = []
self.dfas = {}
self.labels = [(0, "EMPTY")]
self.keywords = {}
self.tokens = {}
self.symbol2label = {}
self.start = 256
def dump(self, filename):
"""Dump the grammar tables to a pickle file."""
with open(filename, "wb") as f:
pickle.dump(self.__dict__, f, 2)
def load(self, filename):
"""Load the grammar tables from a pickle file."""
with open(filename, "rb") as f:
d = pickle.load(f)
self.__dict__.update(d)
def copy(self):
"""
Copy the grammar.
"""
new = self.__class__()
for dict_attr in ("symbol2number", "number2symbol", "dfas", "keywords",
"tokens", "symbol2label"):
setattr(new, dict_attr, getattr(self, dict_attr).copy())
new.labels = self.labels[:]
new.states = self.states[:]
new.start = self.start
return new
def report(self):
"""Dump the grammar tables to standard output, for debugging."""
from pprint import pprint
print("s2n")
pprint(self.symbol2number)
print("n2s")
pprint(self.number2symbol)
print("states")
pprint(self.states)
print("dfas")
pprint(self.dfas)
print("labels")
pprint(self.labels)
print("start", self.start)

View file

@ -0,0 +1,213 @@
# Copyright 2004-2005 Elemental Security, Inc. All Rights Reserved.
# Licensed to PSF under a Contributor Agreement.
# Modifications:
# Copyright 2014 David Halter. Integration into Jedi.
# Modifications are dual-licensed: MIT and PSF.
"""
Parser engine for the grammar tables generated by pgen.
The grammar table must be loaded first.
See Parser/parser.c in the Python distribution for additional info on
how this parsing engine works.
"""
# Local imports
from jedi.parser import tokenize
class InternalParseError(Exception):
"""
Exception to signal the parser is stuck and error recovery didn't help.
Basically this shouldn't happen. It's a sign that something is really
wrong.
"""
def __init__(self, msg, type, value, start_pos):
Exception.__init__(self, "%s: type=%r, value=%r, start_pos=%r" %
(msg, tokenize.tok_name[type], value, start_pos))
self.msg = msg
self.type = type
self.value = value
self.start_pos = start_pos
def token_to_ilabel(grammar, type_, value):
# Map from token to label
if type_ == tokenize.NAME:
# Check for reserved words (keywords)
try:
return grammar.keywords[value]
except KeyError:
pass
try:
return grammar.tokens[type_]
except KeyError:
return None
class PgenParser(object):
"""Parser engine.
The proper usage sequence is:
p = Parser(grammar, [converter]) # create instance
p.setup([start]) # prepare for parsing
<for each input token>:
if p.addtoken(...): # parse a token
break
root = p.rootnode # root of abstract syntax tree
A Parser instance may be reused by calling setup() repeatedly.
A Parser instance contains state pertaining to the current token
sequence, and should not be used concurrently by different threads
to parse separate token sequences.
See driver.py for how to get input tokens by tokenizing a file or
string.
Parsing is complete when addtoken() returns True; the root of the
abstract syntax tree can then be retrieved from the rootnode
instance variable. When a syntax error occurs, error_recovery()
is called. There is no error recovery; the parser cannot be used
after a syntax error was reported (but it can be reinitialized by
calling setup()).
"""
def __init__(self, grammar, convert_node, convert_leaf, error_recovery, start):
"""Constructor.
The grammar argument is a grammar.Grammar instance; see the
grammar module for more information.
The parser is not ready yet for parsing; you must call the
setup() method to get it started.
The optional convert argument is a function mapping concrete
syntax tree nodes to abstract syntax tree nodes. If not
given, no conversion is done and the syntax tree produced is
the concrete syntax tree. If given, it must be a function of
two arguments, the first being the grammar (a grammar.Grammar
instance), and the second being the concrete syntax tree node
to be converted. The syntax tree is converted from the bottom
up.
A concrete syntax tree node is a (type, nodes) tuple, where
type is the node type (a token or symbol number) and nodes
is a list of children for symbols, and None for tokens.
An abstract syntax tree node may be anything; this is entirely
up to the converter function.
"""
self.grammar = grammar
self.convert_node = convert_node
self.convert_leaf = convert_leaf
# Each stack entry is a tuple: (dfa, state, node).
# A node is a tuple: (type, children),
# where children is a list of nodes or None
newnode = (start, [])
stackentry = (self.grammar.dfas[start], 0, newnode)
self.stack = [stackentry]
self.rootnode = None
self.error_recovery = error_recovery
def parse(self, tokenizer):
for type_, value, start_pos, prefix in tokenizer:
if self.addtoken(type_, value, start_pos, prefix):
break
else:
# We never broke out -- EOF is too soon -- Unfinished statement.
# However, the error recovery might have added the token again, if
# the stack is empty, we're fine.
if self.stack:
raise InternalParseError("incomplete input", type_, value, start_pos)
return self.rootnode
def addtoken(self, type_, value, start_pos, prefix):
"""Add a token; return True if this is the end of the program."""
ilabel = token_to_ilabel(self.grammar, type_, value)
# Loop until the token is shifted; may raise exceptions
while True:
dfa, state, node = self.stack[-1]
states, first = dfa
arcs = states[state]
# Look for a state with this label
for i, newstate in arcs:
t, v = self.grammar.labels[i]
if ilabel == i:
# Look it up in the list of labels
assert t < 256
# Shift a token; we're done with it
self.shift(type_, value, newstate, prefix, start_pos)
# Pop while we are in an accept-only state
state = newstate
while states[state] == [(0, state)]:
self.pop()
if not self.stack:
# Done parsing!
return True
dfa, state, node = self.stack[-1]
states, first = dfa
# Done with this token
return False
elif t >= 256:
# See if it's a symbol and if we're in its first set
itsdfa = self.grammar.dfas[t]
itsstates, itsfirst = itsdfa
if ilabel in itsfirst:
# Push a symbol
self.push(t, itsdfa, newstate)
break # To continue the outer while loop
else:
if (0, state) in arcs:
# An accepting state, pop it and try something else
self.pop()
if not self.stack:
# Done parsing, but another token is input
raise InternalParseError("too much input", type_, value, start_pos)
else:
self.error_recovery(self.grammar, self.stack, arcs, type_,
value, start_pos, prefix, self.addtoken)
break
def shift(self, type_, value, newstate, prefix, start_pos):
"""Shift a token. (Internal)"""
dfa, state, node = self.stack[-1]
newnode = self.convert_leaf(self.grammar, type_, value, prefix, start_pos)
node[-1].append(newnode)
self.stack[-1] = (dfa, newstate, node)
def push(self, type_, newdfa, newstate):
"""Push a nonterminal. (Internal)"""
dfa, state, node = self.stack[-1]
newnode = (type_, [])
self.stack[-1] = (dfa, newstate, node)
self.stack.append((newdfa, 0, newnode))
def pop(self):
"""Pop a nonterminal. (Internal)"""
popdfa, popstate, (type_, children) = self.stack.pop()
# If there's exactly one child, return that child instead of creating a
# new node. We still create expr_stmt and file_input though, because a
# lot of Jedi depends on its logic.
if len(children) == 1:
newnode = children[0]
else:
newnode = self.convert_node(self.grammar, type_, children)
try:
# Equal to:
# dfa, state, node = self.stack[-1]
# symbol, children = node
self.stack[-1][2][1].append(newnode)
except IndexError:
# Stack is empty, set the rootnode.
self.rootnode = newnode

View file

@ -0,0 +1,394 @@
# Copyright 2004-2005 Elemental Security, Inc. All Rights Reserved.
# Licensed to PSF under a Contributor Agreement.
# Modifications:
# Copyright 2014 David Halter. Integration into Jedi.
# Modifications are dual-licensed: MIT and PSF.
# Pgen imports
from . import grammar
from jedi.parser import token
from jedi.parser import tokenize
class ParserGenerator(object):
def __init__(self, filename, stream=None):
close_stream = None
if stream is None:
stream = open(filename)
close_stream = stream.close
self.filename = filename
self.stream = stream
self.generator = tokenize.generate_tokens(stream.readline)
self.gettoken() # Initialize lookahead
self.dfas, self.startsymbol = self.parse()
if close_stream is not None:
close_stream()
self.first = {} # map from symbol name to set of tokens
self.addfirstsets()
def make_grammar(self):
c = grammar.Grammar()
names = list(self.dfas.keys())
names.sort()
names.remove(self.startsymbol)
names.insert(0, self.startsymbol)
for name in names:
i = 256 + len(c.symbol2number)
c.symbol2number[name] = i
c.number2symbol[i] = name
for name in names:
dfa = self.dfas[name]
states = []
for state in dfa:
arcs = []
for label, next in state.arcs.items():
arcs.append((self.make_label(c, label), dfa.index(next)))
if state.isfinal:
arcs.append((0, dfa.index(state)))
states.append(arcs)
c.states.append(states)
c.dfas[c.symbol2number[name]] = (states, self.make_first(c, name))
c.start = c.symbol2number[self.startsymbol]
return c
def make_first(self, c, name):
rawfirst = self.first[name]
first = {}
for label in rawfirst:
ilabel = self.make_label(c, label)
##assert ilabel not in first # XXX failed on <> ... !=
first[ilabel] = 1
return first
def make_label(self, c, label):
# XXX Maybe this should be a method on a subclass of converter?
ilabel = len(c.labels)
if label[0].isalpha():
# Either a symbol name or a named token
if label in c.symbol2number:
# A symbol name (a non-terminal)
if label in c.symbol2label:
return c.symbol2label[label]
else:
c.labels.append((c.symbol2number[label], None))
c.symbol2label[label] = ilabel
return ilabel
else:
# A named token (NAME, NUMBER, STRING)
itoken = getattr(token, label, None)
assert isinstance(itoken, int), label
assert itoken in token.tok_name, label
if itoken in c.tokens:
return c.tokens[itoken]
else:
c.labels.append((itoken, None))
c.tokens[itoken] = ilabel
return ilabel
else:
# Either a keyword or an operator
assert label[0] in ('"', "'"), label
value = eval(label)
if value[0].isalpha():
# A keyword
if value in c.keywords:
return c.keywords[value]
else:
c.labels.append((token.NAME, value))
c.keywords[value] = ilabel
return ilabel
else:
# An operator (any non-numeric token)
itoken = token.opmap[value] # Fails if unknown token
if itoken in c.tokens:
return c.tokens[itoken]
else:
c.labels.append((itoken, None))
c.tokens[itoken] = ilabel
return ilabel
def addfirstsets(self):
names = list(self.dfas.keys())
names.sort()
for name in names:
if name not in self.first:
self.calcfirst(name)
#print name, self.first[name].keys()
def calcfirst(self, name):
dfa = self.dfas[name]
self.first[name] = None # dummy to detect left recursion
state = dfa[0]
totalset = {}
overlapcheck = {}
for label, next in state.arcs.items():
if label in self.dfas:
if label in self.first:
fset = self.first[label]
if fset is None:
raise ValueError("recursion for rule %r" % name)
else:
self.calcfirst(label)
fset = self.first[label]
totalset.update(fset)
overlapcheck[label] = fset
else:
totalset[label] = 1
overlapcheck[label] = {label: 1}
inverse = {}
for label, itsfirst in overlapcheck.items():
for symbol in itsfirst:
if symbol in inverse:
raise ValueError("rule %s is ambiguous; %s is in the"
" first sets of %s as well as %s" %
(name, symbol, label, inverse[symbol]))
inverse[symbol] = label
self.first[name] = totalset
def parse(self):
dfas = {}
startsymbol = None
# MSTART: (NEWLINE | RULE)* ENDMARKER
while self.type != token.ENDMARKER:
while self.type == token.NEWLINE:
self.gettoken()
# RULE: NAME ':' RHS NEWLINE
name = self.expect(token.NAME)
self.expect(token.OP, ":")
a, z = self.parse_rhs()
self.expect(token.NEWLINE)
#self.dump_nfa(name, a, z)
dfa = self.make_dfa(a, z)
#self.dump_dfa(name, dfa)
# oldlen = len(dfa)
self.simplify_dfa(dfa)
# newlen = len(dfa)
dfas[name] = dfa
#print name, oldlen, newlen
if startsymbol is None:
startsymbol = name
return dfas, startsymbol
def make_dfa(self, start, finish):
# To turn an NFA into a DFA, we define the states of the DFA
# to correspond to *sets* of states of the NFA. Then do some
# state reduction. Let's represent sets as dicts with 1 for
# values.
assert isinstance(start, NFAState)
assert isinstance(finish, NFAState)
def closure(state):
base = {}
addclosure(state, base)
return base
def addclosure(state, base):
assert isinstance(state, NFAState)
if state in base:
return
base[state] = 1
for label, next in state.arcs:
if label is None:
addclosure(next, base)
states = [DFAState(closure(start), finish)]
for state in states: # NB states grows while we're iterating
arcs = {}
for nfastate in state.nfaset:
for label, next in nfastate.arcs:
if label is not None:
addclosure(next, arcs.setdefault(label, {}))
for label, nfaset in arcs.items():
for st in states:
if st.nfaset == nfaset:
break
else:
st = DFAState(nfaset, finish)
states.append(st)
state.addarc(st, label)
return states # List of DFAState instances; first one is start
def dump_nfa(self, name, start, finish):
print("Dump of NFA for", name)
todo = [start]
for i, state in enumerate(todo):
print(" State", i, state is finish and "(final)" or "")
for label, next in state.arcs:
if next in todo:
j = todo.index(next)
else:
j = len(todo)
todo.append(next)
if label is None:
print(" -> %d" % j)
else:
print(" %s -> %d" % (label, j))
def dump_dfa(self, name, dfa):
print("Dump of DFA for", name)
for i, state in enumerate(dfa):
print(" State", i, state.isfinal and "(final)" or "")
for label, next in state.arcs.items():
print(" %s -> %d" % (label, dfa.index(next)))
def simplify_dfa(self, dfa):
# This is not theoretically optimal, but works well enough.
# Algorithm: repeatedly look for two states that have the same
# set of arcs (same labels pointing to the same nodes) and
# unify them, until things stop changing.
# dfa is a list of DFAState instances
changes = True
while changes:
changes = False
for i, state_i in enumerate(dfa):
for j in range(i + 1, len(dfa)):
state_j = dfa[j]
if state_i == state_j:
#print " unify", i, j
del dfa[j]
for state in dfa:
state.unifystate(state_j, state_i)
changes = True
break
def parse_rhs(self):
# RHS: ALT ('|' ALT)*
a, z = self.parse_alt()
if self.value != "|":
return a, z
else:
aa = NFAState()
zz = NFAState()
aa.addarc(a)
z.addarc(zz)
while self.value == "|":
self.gettoken()
a, z = self.parse_alt()
aa.addarc(a)
z.addarc(zz)
return aa, zz
def parse_alt(self):
# ALT: ITEM+
a, b = self.parse_item()
while (self.value in ("(", "[") or
self.type in (token.NAME, token.STRING)):
c, d = self.parse_item()
b.addarc(c)
b = d
return a, b
def parse_item(self):
# ITEM: '[' RHS ']' | ATOM ['+' | '*']
if self.value == "[":
self.gettoken()
a, z = self.parse_rhs()
self.expect(token.OP, "]")
a.addarc(z)
return a, z
else:
a, z = self.parse_atom()
value = self.value
if value not in ("+", "*"):
return a, z
self.gettoken()
z.addarc(a)
if value == "+":
return a, z
else:
return a, a
def parse_atom(self):
# ATOM: '(' RHS ')' | NAME | STRING
if self.value == "(":
self.gettoken()
a, z = self.parse_rhs()
self.expect(token.OP, ")")
return a, z
elif self.type in (token.NAME, token.STRING):
a = NFAState()
z = NFAState()
a.addarc(z, self.value)
self.gettoken()
return a, z
else:
self.raise_error("expected (...) or NAME or STRING, got %s/%s",
self.type, self.value)
def expect(self, type, value=None):
if self.type != type or (value is not None and self.value != value):
self.raise_error("expected %s/%s, got %s/%s",
type, value, self.type, self.value)
value = self.value
self.gettoken()
return value
def gettoken(self):
tup = next(self.generator)
while tup[0] in (token.COMMENT, token.NL):
tup = next(self.generator)
self.type, self.value, self.begin, prefix = tup
#print tokenize.tok_name[self.type], repr(self.value)
def raise_error(self, msg, *args):
if args:
try:
msg = msg % args
except:
msg = " ".join([msg] + list(map(str, args)))
line = open(self.filename).readlines()[self.begin[0]]
raise SyntaxError(msg, (self.filename, self.begin[0],
self.begin[1], line))
class NFAState(object):
def __init__(self):
self.arcs = [] # list of (label, NFAState) pairs
def addarc(self, next, label=None):
assert label is None or isinstance(label, str)
assert isinstance(next, NFAState)
self.arcs.append((label, next))
class DFAState(object):
def __init__(self, nfaset, final):
assert isinstance(nfaset, dict)
assert isinstance(next(iter(nfaset)), NFAState)
assert isinstance(final, NFAState)
self.nfaset = nfaset
self.isfinal = final in nfaset
self.arcs = {} # map from label to DFAState
def addarc(self, next, label):
assert isinstance(label, str)
assert label not in self.arcs
assert isinstance(next, DFAState)
self.arcs[label] = next
def unifystate(self, old, new):
for label, next in self.arcs.items():
if next is old:
self.arcs[label] = new
def __eq__(self, other):
# Equality test -- ignore the nfaset instance variable
assert isinstance(other, DFAState)
if self.isfinal != other.isfinal:
return False
# Can't just return self.arcs == other.arcs, because that
# would invoke this method recursively, with cycles...
if len(self.arcs) != len(other.arcs):
return False
for label, next in self.arcs.items():
if next is not other.arcs.get(label):
return False
return True
__hash__ = None # For Py3 compatibility.
def generate_grammar(filename="Grammar.txt"):
p = ParserGenerator(filename)
return p.make_grammar()

View file

@ -0,0 +1,90 @@
from __future__ import absolute_import
from jedi._compatibility import is_py3, is_py35
from token import *
COMMENT = N_TOKENS
tok_name[COMMENT] = 'COMMENT'
N_TOKENS += 1
NL = N_TOKENS
tok_name[NL] = 'NL'
N_TOKENS += 1
if is_py3:
BACKQUOTE = N_TOKENS
tok_name[BACKQUOTE] = 'BACKQUOTE'
N_TOKENS += 1
else:
RARROW = N_TOKENS
tok_name[RARROW] = 'RARROW'
N_TOKENS += 1
ELLIPSIS = N_TOKENS
tok_name[ELLIPSIS] = 'ELLIPSIS'
N_TOKENS += 1
if not is_py35:
ATEQUAL = N_TOKENS
tok_name[ATEQUAL] = 'ATEQUAL'
N_TOKENS += 1
# Map from operator to number (since tokenize doesn't do this)
opmap_raw = """\
( LPAR
) RPAR
[ LSQB
] RSQB
: COLON
, COMMA
; SEMI
+ PLUS
- MINUS
* STAR
/ SLASH
| VBAR
& AMPER
< LESS
> GREATER
= EQUAL
. DOT
% PERCENT
` BACKQUOTE
{ LBRACE
} RBRACE
@ AT
== EQEQUAL
!= NOTEQUAL
<> NOTEQUAL
<= LESSEQUAL
>= GREATEREQUAL
~ TILDE
^ CIRCUMFLEX
<< LEFTSHIFT
>> RIGHTSHIFT
** DOUBLESTAR
+= PLUSEQUAL
-= MINEQUAL
*= STAREQUAL
/= SLASHEQUAL
%= PERCENTEQUAL
&= AMPEREQUAL
|= VBAREQUAL
@= ATEQUAL
^= CIRCUMFLEXEQUAL
<<= LEFTSHIFTEQUAL
>>= RIGHTSHIFTEQUAL
**= DOUBLESTAREQUAL
// DOUBLESLASH
//= DOUBLESLASHEQUAL
-> RARROW
... ELLIPSIS
"""
opmap = {}
for line in opmap_raw.splitlines():
op, name = line.split()
opmap[op] = globals()[name]

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