first commit

This commit is contained in:
Ayxan
2022-05-23 00:16:32 +04:00
commit d660f2a4ca
24786 changed files with 4428337 additions and 0 deletions

View File

@ -0,0 +1,391 @@
"""
Makes it possible to do the compiled analysis in a subprocess. This has two
goals:
1. Making it safer - Segfaults and RuntimeErrors as well as stdout/stderr can
be ignored and dealt with.
2. Make it possible to handle different Python versions as well as virtualenvs.
"""
import collections
import os
import sys
import queue
import subprocess
import traceback
import weakref
from functools import partial
from threading import Thread
from jedi._compatibility import pickle_dump, pickle_load
from jedi import debug
from jedi.cache import memoize_method
from jedi.inference.compiled.subprocess import functions
from jedi.inference.compiled.access import DirectObjectAccess, AccessPath, \
SignatureParam
from jedi.api.exceptions import InternalError
_MAIN_PATH = os.path.join(os.path.dirname(__file__), '__main__.py')
PICKLE_PROTOCOL = 4
def _GeneralizedPopen(*args, **kwargs):
if os.name == 'nt':
try:
# Was introduced in Python 3.7.
CREATE_NO_WINDOW = subprocess.CREATE_NO_WINDOW
except AttributeError:
CREATE_NO_WINDOW = 0x08000000
kwargs['creationflags'] = CREATE_NO_WINDOW
# The child process doesn't need file descriptors except 0, 1, 2.
# This is unix only.
kwargs['close_fds'] = 'posix' in sys.builtin_module_names
return subprocess.Popen(*args, **kwargs)
def _enqueue_output(out, queue_):
for line in iter(out.readline, b''):
queue_.put(line)
def _add_stderr_to_debug(stderr_queue):
while True:
# Try to do some error reporting from the subprocess and print its
# stderr contents.
try:
line = stderr_queue.get_nowait()
line = line.decode('utf-8', 'replace')
debug.warning('stderr output: %s' % line.rstrip('\n'))
except queue.Empty:
break
def _get_function(name):
return getattr(functions, name)
def _cleanup_process(process, thread):
try:
process.kill()
process.wait()
except OSError:
# Raised if the process is already killed.
pass
thread.join()
for stream in [process.stdin, process.stdout, process.stderr]:
try:
stream.close()
except OSError:
# Raised if the stream is broken.
pass
class _InferenceStateProcess:
def __init__(self, inference_state):
self._inference_state_weakref = weakref.ref(inference_state)
self._inference_state_id = id(inference_state)
self._handles = {}
def get_or_create_access_handle(self, obj):
id_ = id(obj)
try:
return self.get_access_handle(id_)
except KeyError:
access = DirectObjectAccess(self._inference_state_weakref(), obj)
handle = AccessHandle(self, access, id_)
self.set_access_handle(handle)
return handle
def get_access_handle(self, id_):
return self._handles[id_]
def set_access_handle(self, handle):
self._handles[handle.id] = handle
class InferenceStateSameProcess(_InferenceStateProcess):
"""
Basically just an easy access to functions.py. It has the same API
as InferenceStateSubprocess and does the same thing without using a subprocess.
This is necessary for the Interpreter process.
"""
def __getattr__(self, name):
return partial(_get_function(name), self._inference_state_weakref())
class InferenceStateSubprocess(_InferenceStateProcess):
def __init__(self, inference_state, compiled_subprocess):
super().__init__(inference_state)
self._used = False
self._compiled_subprocess = compiled_subprocess
def __getattr__(self, name):
func = _get_function(name)
def wrapper(*args, **kwargs):
self._used = True
result = self._compiled_subprocess.run(
self._inference_state_weakref(),
func,
args=args,
kwargs=kwargs,
)
# IMO it should be possible to create a hook in pickle.load to
# mess with the loaded objects. However it's extremely complicated
# to work around this so just do it with this call. ~ dave
return self._convert_access_handles(result)
return wrapper
def _convert_access_handles(self, obj):
if isinstance(obj, SignatureParam):
return SignatureParam(*self._convert_access_handles(tuple(obj)))
elif isinstance(obj, tuple):
return tuple(self._convert_access_handles(o) for o in obj)
elif isinstance(obj, list):
return [self._convert_access_handles(o) for o in obj]
elif isinstance(obj, AccessHandle):
try:
# Rewrite the access handle to one we're already having.
obj = self.get_access_handle(obj.id)
except KeyError:
obj.add_subprocess(self)
self.set_access_handle(obj)
elif isinstance(obj, AccessPath):
return AccessPath(self._convert_access_handles(obj.accesses))
return obj
def __del__(self):
if self._used and not self._compiled_subprocess.is_crashed:
self._compiled_subprocess.delete_inference_state(self._inference_state_id)
class CompiledSubprocess:
is_crashed = False
def __init__(self, executable, env_vars=None):
self._executable = executable
self._env_vars = env_vars
self._inference_state_deletion_queue = collections.deque()
self._cleanup_callable = lambda: None
def __repr__(self):
pid = os.getpid()
return '<%s _executable=%r, is_crashed=%r, pid=%r>' % (
self.__class__.__name__,
self._executable,
self.is_crashed,
pid,
)
@memoize_method
def _get_process(self):
debug.dbg('Start environment subprocess %s', self._executable)
parso_path = sys.modules['parso'].__file__
args = (
self._executable,
_MAIN_PATH,
os.path.dirname(os.path.dirname(parso_path)),
'.'.join(str(x) for x in sys.version_info[:3]),
)
process = _GeneralizedPopen(
args,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
env=self._env_vars
)
self._stderr_queue = queue.Queue()
self._stderr_thread = t = Thread(
target=_enqueue_output,
args=(process.stderr, self._stderr_queue)
)
t.daemon = True
t.start()
# Ensure the subprocess is properly cleaned up when the object
# is garbage collected.
self._cleanup_callable = weakref.finalize(self,
_cleanup_process,
process,
t)
return process
def run(self, inference_state, function, args=(), kwargs={}):
# Delete old inference_states.
while True:
try:
inference_state_id = self._inference_state_deletion_queue.pop()
except IndexError:
break
else:
self._send(inference_state_id, None)
assert callable(function)
return self._send(id(inference_state), function, args, kwargs)
def get_sys_path(self):
return self._send(None, functions.get_sys_path, (), {})
def _kill(self):
self.is_crashed = True
self._cleanup_callable()
def _send(self, inference_state_id, function, args=(), kwargs={}):
if self.is_crashed:
raise InternalError("The subprocess %s has crashed." % self._executable)
data = inference_state_id, function, args, kwargs
try:
pickle_dump(data, self._get_process().stdin, PICKLE_PROTOCOL)
except BrokenPipeError:
self._kill()
raise InternalError("The subprocess %s was killed. Maybe out of memory?"
% self._executable)
try:
is_exception, traceback, result = pickle_load(self._get_process().stdout)
except EOFError as eof_error:
try:
stderr = self._get_process().stderr.read().decode('utf-8', 'replace')
except Exception as exc:
stderr = '<empty/not available (%r)>' % exc
self._kill()
_add_stderr_to_debug(self._stderr_queue)
raise InternalError(
"The subprocess %s has crashed (%r, stderr=%s)." % (
self._executable,
eof_error,
stderr,
))
_add_stderr_to_debug(self._stderr_queue)
if is_exception:
# Replace the attribute error message with a the traceback. It's
# way more informative.
result.args = (traceback,)
raise result
return result
def delete_inference_state(self, inference_state_id):
"""
Currently we are not deleting inference_state instantly. They only get
deleted once the subprocess is used again. It would probably a better
solution to move all of this into a thread. However, the memory usage
of a single inference_state shouldn't be that high.
"""
# With an argument - the inference_state gets deleted.
self._inference_state_deletion_queue.append(inference_state_id)
class Listener:
def __init__(self):
self._inference_states = {}
# TODO refactor so we don't need to process anymore just handle
# controlling.
self._process = _InferenceStateProcess(Listener)
def _get_inference_state(self, function, inference_state_id):
from jedi.inference import InferenceState
try:
inference_state = self._inference_states[inference_state_id]
except KeyError:
from jedi import InterpreterEnvironment
inference_state = InferenceState(
# The project is not actually needed. Nothing should need to
# access it.
project=None,
environment=InterpreterEnvironment()
)
self._inference_states[inference_state_id] = inference_state
return inference_state
def _run(self, inference_state_id, function, args, kwargs):
if inference_state_id is None:
return function(*args, **kwargs)
elif function is None:
del self._inference_states[inference_state_id]
else:
inference_state = self._get_inference_state(function, inference_state_id)
# Exchange all handles
args = list(args)
for i, arg in enumerate(args):
if isinstance(arg, AccessHandle):
args[i] = inference_state.compiled_subprocess.get_access_handle(arg.id)
for key, value in kwargs.items():
if isinstance(value, AccessHandle):
kwargs[key] = inference_state.compiled_subprocess.get_access_handle(value.id)
return function(inference_state, *args, **kwargs)
def listen(self):
stdout = sys.stdout
# Mute stdout. Nobody should actually be able to write to it,
# because stdout is used for IPC.
sys.stdout = open(os.devnull, 'w')
stdin = sys.stdin
stdout = stdout.buffer
stdin = stdin.buffer
while True:
try:
payload = pickle_load(stdin)
except EOFError:
# It looks like the parent process closed.
# Don't make a big fuss here and just exit.
exit(0)
try:
result = False, None, self._run(*payload)
except Exception as e:
result = True, traceback.format_exc(), e
pickle_dump(result, stdout, PICKLE_PROTOCOL)
class AccessHandle:
def __init__(self, subprocess, access, id_):
self.access = access
self._subprocess = subprocess
self.id = id_
def add_subprocess(self, subprocess):
self._subprocess = subprocess
def __repr__(self):
try:
detail = self.access
except AttributeError:
detail = '#' + str(self.id)
return '<%s of %s>' % (self.__class__.__name__, detail)
def __getstate__(self):
return self.id
def __setstate__(self, state):
self.id = state
def __getattr__(self, name):
if name in ('id', 'access') or name.startswith('_'):
raise AttributeError("Something went wrong with unpickling")
# print('getattr', name, file=sys.stderr)
return partial(self._workaround, name)
def _workaround(self, name, *args, **kwargs):
"""
TODO Currently we're passing slice objects around. This should not
happen. They are also the only unhashable objects that we're passing
around.
"""
if args and isinstance(args[0], slice):
return self._subprocess.get_compiled_method_return(self.id, name, *args, **kwargs)
return self._cached_results(name, *args, **kwargs)
@memoize_method
def _cached_results(self, name, *args, **kwargs):
return self._subprocess.get_compiled_method_return(self.id, name, *args, **kwargs)

View File

@ -0,0 +1,40 @@
import os
import sys
from importlib.abc import MetaPathFinder
from importlib.machinery import PathFinder
# Remove the first entry, because it's simply a directory entry that equals
# this directory.
del sys.path[0]
def _get_paths():
# Get the path to jedi.
_d = os.path.dirname
_jedi_path = _d(_d(_d(_d(_d(__file__)))))
_parso_path = sys.argv[1]
# The paths are the directory that jedi and parso lie in.
return {'jedi': _jedi_path, 'parso': _parso_path}
class _ExactImporter(MetaPathFinder):
def __init__(self, path_dct):
self._path_dct = path_dct
def find_module(self, fullname, path=None):
if path is None and fullname in self._path_dct:
p = self._path_dct[fullname]
loader = PathFinder.find_module(fullname, path=[p])
return loader
return None
# Try to import jedi/parso.
sys.meta_path.insert(0, _ExactImporter(_get_paths()))
from jedi.inference.compiled import subprocess # noqa: E402
sys.meta_path.pop(0)
# Retrieve the pickle protocol.
host_sys_version = [int(x) for x in sys.argv[2].split('.')]
# And finally start the client.
subprocess.Listener().listen()

View File

@ -0,0 +1,255 @@
import sys
import os
import inspect
import importlib
import warnings
from pathlib import Path
from zipfile import ZipFile
from zipimport import zipimporter, ZipImportError
from importlib.machinery import all_suffixes
from jedi.inference.compiled import access
from jedi import debug
from jedi import parser_utils
from jedi.file_io import KnownContentFileIO, ZipFileIO
def get_sys_path():
return sys.path
def load_module(inference_state, **kwargs):
return access.load_module(inference_state, **kwargs)
def get_compiled_method_return(inference_state, id, attribute, *args, **kwargs):
handle = inference_state.compiled_subprocess.get_access_handle(id)
return getattr(handle.access, attribute)(*args, **kwargs)
def create_simple_object(inference_state, obj):
return access.create_access_path(inference_state, obj)
def get_module_info(inference_state, sys_path=None, full_name=None, **kwargs):
"""
Returns Tuple[Union[NamespaceInfo, FileIO, None], Optional[bool]]
"""
if sys_path is not None:
sys.path, temp = sys_path, sys.path
try:
return _find_module(full_name=full_name, **kwargs)
except ImportError:
return None, None
finally:
if sys_path is not None:
sys.path = temp
def get_builtin_module_names(inference_state):
return sys.builtin_module_names
def _test_raise_error(inference_state, exception_type):
"""
Raise an error to simulate certain problems for unit tests.
"""
raise exception_type
def _test_print(inference_state, stderr=None, stdout=None):
"""
Force some prints in the subprocesses. This exists for unit tests.
"""
if stderr is not None:
print(stderr, file=sys.stderr)
sys.stderr.flush()
if stdout is not None:
print(stdout)
sys.stdout.flush()
def _get_init_path(directory_path):
"""
The __init__ file can be searched in a directory. If found return it, else
None.
"""
for suffix in all_suffixes():
path = os.path.join(directory_path, '__init__' + suffix)
if os.path.exists(path):
return path
return None
def safe_literal_eval(inference_state, value):
return parser_utils.safe_literal_eval(value)
def iter_module_names(*args, **kwargs):
return list(_iter_module_names(*args, **kwargs))
def _iter_module_names(inference_state, paths):
# Python modules/packages
for path in paths:
try:
dir_entries = ((entry.name, entry.is_dir()) for entry in os.scandir(path))
except OSError:
try:
zip_import_info = zipimporter(path)
# Unfortunately, there is no public way to access zipimporter's
# private _files member. We therefore have to use a
# custom function to iterate over the files.
dir_entries = _zip_list_subdirectory(
zip_import_info.archive, zip_import_info.prefix)
except ZipImportError:
# The file might not exist or reading it might lead to an error.
debug.warning("Not possible to list directory: %s", path)
continue
for name, is_dir in dir_entries:
# First Namespaces then modules/stubs
if is_dir:
# pycache is obviously not an interesting namespace. Also the
# name must be a valid identifier.
if name != '__pycache__' and name.isidentifier():
yield name
else:
if name.endswith('.pyi'): # Stub files
modname = name[:-4]
else:
modname = inspect.getmodulename(name)
if modname and '.' not in modname:
if modname != '__init__':
yield modname
def _find_module(string, path=None, full_name=None, is_global_search=True):
"""
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.
"""
spec = None
loader = None
for finder in sys.meta_path:
if is_global_search and finder != importlib.machinery.PathFinder:
p = None
else:
p = path
try:
find_spec = finder.find_spec
except AttributeError:
# These are old-school clases that still have a different API, just
# ignore those.
continue
spec = find_spec(string, p)
if spec is not None:
loader = spec.loader
if loader is None and not spec.has_location:
# This is a namespace package.
full_name = string if not path else full_name
implicit_ns_info = ImplicitNSInfo(full_name, spec.submodule_search_locations._path)
return implicit_ns_info, True
break
return _find_module_py33(string, path, loader)
def _find_module_py33(string, path=None, loader=None, full_name=None, is_global_search=True):
loader = loader or importlib.machinery.PathFinder.find_module(string, path)
if loader is None and path is None: # Fallback to find builtins
try:
with warnings.catch_warnings(record=True):
# Mute "DeprecationWarning: Use importlib.util.find_spec()
# instead." While we should replace that in the future, it's
# probably good to wait until we deprecate Python 3.3, since
# it was added in Python 3.4 and find_loader hasn't been
# removed in 3.6.
loader = importlib.find_loader(string)
except ValueError as e:
# See #491. Importlib might raise a ValueError, to avoid this, we
# just raise an ImportError to fix the issue.
raise ImportError("Originally " + repr(e))
if loader is None:
raise ImportError("Couldn't find a loader for {}".format(string))
return _from_loader(loader, string)
def _from_loader(loader, string):
try:
is_package_method = loader.is_package
except AttributeError:
is_package = False
else:
is_package = is_package_method(string)
try:
get_filename = loader.get_filename
except AttributeError:
return None, is_package
else:
module_path = get_filename(string)
# To avoid unicode and read bytes, "overwrite" loader.get_source if
# possible.
try:
f = type(loader).get_source
except AttributeError:
raise ImportError("get_source was not defined on loader")
if f is not importlib.machinery.SourceFileLoader.get_source:
# Unfortunately we are reading unicode here, not bytes.
# It seems hard to get bytes, because the zip importer
# logic just unpacks the zip file and returns a file descriptor
# that we cannot as easily access. Therefore we just read it as
# a string in the cases where get_source was overwritten.
code = loader.get_source(string)
else:
code = _get_source(loader, string)
if code is None:
return None, is_package
if isinstance(loader, zipimporter):
return ZipFileIO(module_path, code, Path(loader.archive)), is_package
return KnownContentFileIO(module_path, code), is_package
def _get_source(loader, fullname):
"""
This method is here as a replacement for SourceLoader.get_source. That
method returns unicode, but we prefer bytes.
"""
path = loader.get_filename(fullname)
try:
return loader.get_data(path)
except OSError:
raise ImportError('source not available through get_data()',
name=fullname)
def _zip_list_subdirectory(zip_path, zip_subdir_path):
zip_file = ZipFile(zip_path)
zip_subdir_path = Path(zip_subdir_path)
zip_content_file_paths = zip_file.namelist()
for raw_file_name in zip_content_file_paths:
file_path = Path(raw_file_name)
if file_path.parent == zip_subdir_path:
file_path = file_path.relative_to(zip_subdir_path)
yield file_path.name, raw_file_name.endswith("/")
class ImplicitNSInfo:
"""Stores information returned from an implicit namespace spec"""
def __init__(self, name, paths):
self.name = name
self.paths = paths