Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.11/site-packages/git/cmd.py: 60%
Shortcuts on this page
r m x toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
1# Copyright (C) 2008, 2009 Michael Trier (mtrier@gmail.com) and contributors
2#
3# This module is part of GitPython and is released under the
4# 3-Clause BSD License: https://opensource.org/license/bsd-3-clause/
6from __future__ import annotations
8__all__ = ["GitMeta", "Git"]
10import contextlib
11import io
12import itertools
13import logging
14import os
15import re
16import signal
17import subprocess
18from subprocess import DEVNULL, PIPE, Popen
19import sys
20from textwrap import dedent
21import threading
22import warnings
24from git.compat import defenc, force_bytes, safe_decode
25from git.exc import (
26 CommandError,
27 GitCommandError,
28 GitCommandNotFound,
29 UnsafeOptionError,
30 UnsafeProtocolError,
31)
32from git.util import (
33 cygpath,
34 expand_path,
35 is_cygwin_git,
36 patch_env,
37 remove_password_if_present,
38 stream_copy,
39)
41# typing ---------------------------------------------------------------------------
43from typing import (
44 Any,
45 AnyStr,
46 BinaryIO,
47 Callable,
48 Dict,
49 IO,
50 Iterator,
51 List,
52 Mapping,
53 Optional,
54 Sequence,
55 TYPE_CHECKING,
56 TextIO,
57 Tuple,
58 Union,
59 cast,
60 overload,
61)
63if sys.version_info >= (3, 10):
64 from typing import TypeAlias
65else:
66 from typing_extensions import TypeAlias
68from git.types import Literal, PathLike, TBD
70if TYPE_CHECKING:
71 from git.diff import DiffIndex
72 from git.repo.base import Repo
74# ---------------------------------------------------------------------------------
76execute_kwargs = {
77 "istream",
78 "with_extended_output",
79 "with_exceptions",
80 "as_process",
81 "output_stream",
82 "stdout_as_string",
83 "kill_after_timeout",
84 "with_stdout",
85 "universal_newlines",
86 "shell",
87 "env",
88 "max_chunk_size",
89 "strip_newline_in_stdout",
90}
92_logger = logging.getLogger(__name__)
95# ==============================================================================
96## @name Utilities
97# ------------------------------------------------------------------------------
98# Documentation
99## @{
102def handle_process_output(
103 process: "Git.AutoInterrupt" | Popen,
104 stdout_handler: Union[
105 None,
106 Callable[[AnyStr], None],
107 Callable[[List[AnyStr]], None],
108 Callable[[bytes, "Repo", "DiffIndex"], None],
109 ],
110 stderr_handler: Union[None, Callable[[AnyStr], None], Callable[[List[AnyStr]], None]],
111 finalizer: Union[None, Callable[[Union[Popen, "Git.AutoInterrupt"]], None]] = None,
112 decode_streams: bool = True,
113 kill_after_timeout: Union[None, float] = None,
114) -> None:
115 R"""Register for notifications to learn that process output is ready to read, and
116 dispatch lines to the respective line handlers.
118 This function returns once the finalizer returns.
120 :param process:
121 :class:`subprocess.Popen` instance.
123 :param stdout_handler:
124 f(stdout_line_string), or ``None``.
126 :param stderr_handler:
127 f(stderr_line_string), or ``None``.
129 :param finalizer:
130 f(proc) - wait for proc to finish.
132 :param decode_streams:
133 Assume stdout/stderr streams are binary and decode them before pushing their
134 contents to handlers.
136 This defaults to ``True``. Set it to ``False`` if:
138 - ``universal_newlines == True``, as then streams are in text mode, or
139 - decoding must happen later, such as for :class:`~git.diff.Diff`\s.
141 :param kill_after_timeout:
142 :class:`float` or ``None``, Default = ``None``
144 To specify a timeout in seconds for the git command, after which the process
145 should be killed.
146 """
148 # Use 2 "pump" threads and wait for both to finish.
149 def pump_stream(
150 cmdline: List[str],
151 name: str,
152 stream: Union[BinaryIO, TextIO],
153 is_decode: bool,
154 handler: Union[None, Callable[[Union[bytes, str]], None]],
155 ) -> None:
156 try:
157 for line in stream:
158 if handler:
159 if is_decode:
160 assert isinstance(line, bytes)
161 line_str = line.decode(defenc)
162 handler(line_str)
163 else:
164 handler(line)
166 except Exception as ex:
167 _logger.error(f"Pumping {name!r} of cmd({remove_password_if_present(cmdline)}) failed due to: {ex!r}")
168 if "I/O operation on closed file" not in str(ex):
169 # Only reraise if the error was not due to the stream closing.
170 raise CommandError([f"<{name}-pump>"] + remove_password_if_present(cmdline), ex) from ex
171 finally:
172 stream.close()
174 if hasattr(process, "proc"):
175 process = cast("Git.AutoInterrupt", process)
176 cmdline: str | Tuple[str, ...] | List[str] = getattr(process.proc, "args", "")
177 p_stdout = process.proc.stdout if process.proc else None
178 p_stderr = process.proc.stderr if process.proc else None
179 else:
180 process = cast(Popen, process) # type: ignore[redundant-cast]
181 cmdline = getattr(process, "args", "")
182 p_stdout = process.stdout
183 p_stderr = process.stderr
185 if not isinstance(cmdline, (tuple, list)):
186 cmdline = cmdline.split()
188 pumps: List[Tuple[str, IO, Callable[..., None] | None]] = []
189 if p_stdout:
190 pumps.append(("stdout", p_stdout, stdout_handler))
191 if p_stderr:
192 pumps.append(("stderr", p_stderr, stderr_handler))
194 threads: List[threading.Thread] = []
196 for name, stream, handler in pumps:
197 t = threading.Thread(target=pump_stream, args=(cmdline, name, stream, decode_streams, handler))
198 t.daemon = True
199 t.start()
200 threads.append(t)
202 # FIXME: Why join? Will block if stdin needs feeding...
203 for t in threads:
204 t.join(timeout=kill_after_timeout)
205 if t.is_alive():
206 if isinstance(process, Git.AutoInterrupt):
207 process._terminate()
208 else: # Don't want to deal with the other case.
209 raise RuntimeError(
210 "Thread join() timed out in cmd.handle_process_output()."
211 f" kill_after_timeout={kill_after_timeout} seconds"
212 )
213 if stderr_handler:
214 error_str: Union[str, bytes] = (
215 f"error: process killed because it timed out. kill_after_timeout={kill_after_timeout} seconds"
216 )
217 if not decode_streams and isinstance(p_stderr, BinaryIO):
218 # Assume stderr_handler needs binary input.
219 error_str = cast(str, error_str)
220 error_str = error_str.encode()
221 # We ignore typing on the next line because mypy does not like the way
222 # we inferred that stderr takes str or bytes.
223 stderr_handler(error_str) # type: ignore[arg-type]
225 if finalizer:
226 finalizer(process)
229safer_popen: Callable[..., Popen]
231if sys.platform == "win32":
233 def _safer_popen_windows(
234 command: Union[str, Sequence[Any]],
235 *,
236 shell: bool = False,
237 env: Optional[Mapping[str, str]] = None,
238 **kwargs: Any,
239 ) -> Popen:
240 """Call :class:`subprocess.Popen` on Windows but don't include a CWD in the
241 search.
243 This avoids an untrusted search path condition where a file like ``git.exe`` in
244 a malicious repository would be run when GitPython operates on the repository.
245 The process using GitPython may have an untrusted repository's working tree as
246 its current working directory. Some operations may temporarily change to that
247 directory before running a subprocess. In addition, while by default GitPython
248 does not run external commands with a shell, it can be made to do so, in which
249 case the CWD of the subprocess, which GitPython usually sets to a repository
250 working tree, can itself be searched automatically by the shell. This wrapper
251 covers all those cases.
253 :note:
254 This currently works by setting the
255 :envvar:`NoDefaultCurrentDirectoryInExePath` environment variable during
256 subprocess creation. It also takes care of passing Windows-specific process
257 creation flags, but that is unrelated to path search.
259 :note:
260 The current implementation contains a race condition on :attr:`os.environ`.
261 GitPython isn't thread-safe, but a program using it on one thread should
262 ideally be able to mutate :attr:`os.environ` on another, without
263 unpredictable results. See comments in:
264 https://github.com/gitpython-developers/GitPython/pull/1650
265 """
266 # CREATE_NEW_PROCESS_GROUP is needed for some ways of killing it afterwards.
267 # https://docs.python.org/3/library/subprocess.html#subprocess.Popen.send_signal
268 # https://docs.python.org/3/library/subprocess.html#subprocess.CREATE_NEW_PROCESS_GROUP
269 creationflags = subprocess.CREATE_NO_WINDOW | subprocess.CREATE_NEW_PROCESS_GROUP
271 # When using a shell, the shell is the direct subprocess, so the variable must
272 # be set in its environment, to affect its search behavior.
273 if shell:
274 # The original may be immutable, or the caller may reuse it. Mutate a copy.
275 env = {} if env is None else dict(env)
276 env["NoDefaultCurrentDirectoryInExePath"] = "1" # The "1" can be any value.
278 # When not using a shell, the current process does the search in a
279 # CreateProcessW API call, so the variable must be set in our environment. With
280 # a shell, that's unnecessary if https://github.com/python/cpython/issues/101283
281 # is patched. In Python versions where it is unpatched, in the rare case the
282 # ComSpec environment variable is unset, the search for the shell itself is
283 # unsafe. Setting NoDefaultCurrentDirectoryInExePath in all cases, as done here,
284 # is simpler and protects against that. (As above, the "1" can be any value.)
285 with patch_env("NoDefaultCurrentDirectoryInExePath", "1"):
286 return Popen(
287 command,
288 shell=shell,
289 env=env,
290 creationflags=creationflags,
291 **kwargs,
292 )
294 safer_popen = _safer_popen_windows
295else:
296 safer_popen = Popen
299def dashify(string: str) -> str:
300 return string.replace("_", "-")
303def slots_to_dict(self: "Git", exclude: Sequence[str] = ()) -> Dict[str, Any]:
304 return {s: getattr(self, s) for s in self.__slots__ if s not in exclude}
307def dict_to_slots_and__excluded_are_none(self: object, d: Mapping[str, Any], excluded: Sequence[str] = ()) -> None:
308 for k, v in d.items():
309 setattr(self, k, v)
310 for k in excluded:
311 setattr(self, k, None)
314## -- End Utilities -- @}
317class _AutoInterrupt:
318 """Process wrapper that terminates the wrapped process on finalization.
320 This kills/interrupts the stored process instance once this instance goes out of
321 scope. It is used to prevent processes piling up in case iterators stop reading.
323 All attributes are wired through to the contained process object.
325 The wait method is overridden to perform automatic status code checking and possibly
326 raise.
327 """
329 __slots__ = ("proc", "args", "status")
331 # If this is non-zero it will override any status code during _terminate, used
332 # to prevent race conditions in testing.
333 _status_code_if_terminate: int = 0
335 def __init__(self, proc: Union[None, subprocess.Popen], args: Any) -> None:
336 self.proc = proc
337 self.args = args
338 self.status: Union[int, None] = None
340 def _terminate(self) -> None:
341 """Terminate the underlying process."""
342 if self.proc is None:
343 return
345 proc = self.proc
346 self.proc = None
347 if proc.stdin:
348 proc.stdin.close()
349 if proc.stdout:
350 proc.stdout.close()
351 if proc.stderr:
352 proc.stderr.close()
353 # Did the process finish already so we have a return code?
354 try:
355 if proc.poll() is not None:
356 self.status = self._status_code_if_terminate or proc.poll()
357 return
358 except OSError as ex:
359 _logger.info("Ignored error after process had died: %r", ex)
361 # It can be that nothing really exists anymore...
362 if os is None or getattr(os, "kill", None) is None:
363 return
365 # Try to kill it.
366 try:
367 proc.terminate()
368 status = proc.wait() # Ensure the process goes away.
370 self.status = self._status_code_if_terminate or status
371 except (OSError, AttributeError) as ex:
372 # On interpreter shutdown (notably on Windows), parts of the stdlib used by
373 # subprocess can already be torn down (e.g. `subprocess._winapi` becomes None),
374 # which can cause AttributeError during terminate(). In that case, we prefer
375 # to silently ignore to avoid noisy "Exception ignored in: __del__" messages.
376 _logger.info("Ignored error while terminating process: %r", ex)
377 # END exception handling
379 def __del__(self) -> None:
380 self._terminate()
382 def __getattr__(self, attr: str) -> Any:
383 return getattr(self.proc, attr)
385 # TODO: Bad choice to mimic `proc.wait()` but with different args.
386 def wait(self, stderr: Union[None, str, bytes] = b"") -> int:
387 """Wait for the process and return its status code.
389 :param stderr:
390 Previously read value of stderr, in case stderr is already closed.
392 :warn:
393 May deadlock if output or error pipes are used and not handled separately.
395 :raise git.exc.GitCommandError:
396 If the return status is not 0.
397 """
398 if stderr is None:
399 stderr_b = b""
400 stderr_b = force_bytes(data=stderr, encoding="utf-8")
401 status: Union[int, None]
402 if self.proc is not None:
403 status = self.proc.wait()
404 p_stderr = self.proc.stderr
405 else: # Assume the underlying proc was killed earlier or never existed.
406 status = self.status
407 p_stderr = None
409 def read_all_from_possibly_closed_stream(stream: Union[IO[bytes], None]) -> bytes:
410 if stream:
411 try:
412 return stderr_b + force_bytes(stream.read())
413 except (OSError, ValueError):
414 return stderr_b or b""
415 else:
416 return stderr_b or b""
418 # END status handling
420 if status != 0:
421 errstr = read_all_from_possibly_closed_stream(p_stderr)
422 _logger.debug("AutoInterrupt wait stderr: %r" % (errstr,))
423 raise GitCommandError(remove_password_if_present(self.args), status, errstr)
424 return status
427_AutoInterrupt.__name__ = "AutoInterrupt"
428_AutoInterrupt.__qualname__ = "Git.AutoInterrupt"
431class _CatFileContentStream:
432 """Object representing a sized read-only stream returning the contents of
433 an object.
435 This behaves like a stream, but counts the data read and simulates an empty stream
436 once our sized content region is empty.
438 If not all data are read to the end of the object's lifetime, we read the rest to
439 ensure the underlying stream continues to work.
440 """
442 __slots__ = ("_stream", "_nbr", "_size")
444 def __init__(self, size: int, stream: IO[bytes]) -> None:
445 self._stream = stream
446 self._size = size
447 self._nbr = 0 # Number of bytes read.
449 # Special case: If the object is empty, has null bytes, get the final
450 # newline right away.
451 if size == 0:
452 stream.read(1)
453 # END handle empty streams
455 def read(self, size: int = -1) -> bytes:
456 bytes_left = self._size - self._nbr
457 if bytes_left == 0:
458 return b""
459 if size > -1:
460 # Ensure we don't try to read past our limit.
461 size = min(bytes_left, size)
462 else:
463 # They try to read all, make sure it's not more than what remains.
464 size = bytes_left
465 # END check early depletion
466 data = self._stream.read(size)
467 self._nbr += len(data)
469 # Check for depletion, read our final byte to make the stream usable by
470 # others.
471 if self._size - self._nbr == 0:
472 self._stream.read(1) # final newline
473 # END finish reading
474 return data
476 def readline(self, size: int = -1) -> bytes:
477 if self._nbr == self._size:
478 return b""
480 # Clamp size to lowest allowed value.
481 bytes_left = self._size - self._nbr
482 if size > -1:
483 size = min(bytes_left, size)
484 else:
485 size = bytes_left
486 # END handle size
488 data = self._stream.readline(size)
489 self._nbr += len(data)
491 # Handle final byte.
492 if self._size - self._nbr == 0:
493 self._stream.read(1)
494 # END finish reading
496 return data
498 def readlines(self, size: int = -1) -> List[bytes]:
499 if self._nbr == self._size:
500 return []
502 # Leave all additional logic to our readline method, we just check the size.
503 out = []
504 nbr = 0
505 while True:
506 line = self.readline()
507 if not line:
508 break
509 out.append(line)
510 if size > -1:
511 nbr += len(line)
512 if nbr > size:
513 break
514 # END handle size constraint
515 # END readline loop
516 return out
518 # skipcq: PYL-E0301
519 def __iter__(self) -> "Git.CatFileContentStream":
520 return self
522 def __next__(self) -> bytes:
523 line = self.readline()
524 if not line:
525 raise StopIteration
527 return line
529 next = __next__
531 def __del__(self) -> None:
532 bytes_left = self._size - self._nbr
533 if bytes_left:
534 # Read and discard - seeking is impossible within a stream.
535 # This includes any terminating newline.
536 self._stream.read(bytes_left + 1)
537 # END handle incomplete read
540_CatFileContentStream.__name__ = "CatFileContentStream"
541_CatFileContentStream.__qualname__ = "Git.CatFileContentStream"
544_USE_SHELL_DEFAULT_MESSAGE = (
545 "Git.USE_SHELL is deprecated, because only its default value of False is safe. "
546 "It will be removed in a future release."
547)
549_USE_SHELL_DANGER_MESSAGE = (
550 "Setting Git.USE_SHELL to True is unsafe and insecure, as the effect of special "
551 "shell syntax cannot usually be accounted for. This can result in a command "
552 "injection vulnerability and arbitrary code execution. Git.USE_SHELL is deprecated "
553 "and will be removed in a future release."
554)
557def _warn_use_shell(*, extra_danger: bool) -> None:
558 warnings.warn(
559 _USE_SHELL_DANGER_MESSAGE if extra_danger else _USE_SHELL_DEFAULT_MESSAGE,
560 DeprecationWarning,
561 stacklevel=3,
562 )
565class _GitMeta(type):
566 """Metaclass for :class:`Git`.
568 This helps issue :class:`DeprecationWarning` if :attr:`Git.USE_SHELL` is used.
569 """
571 def __getattribute(cls, name: str) -> Any:
572 if name == "USE_SHELL":
573 _warn_use_shell(extra_danger=False)
574 return super().__getattribute__(name)
576 def __setattr(cls, name: str, value: Any) -> Any:
577 if name == "USE_SHELL":
578 _warn_use_shell(extra_danger=value)
579 super().__setattr__(name, value)
581 if not TYPE_CHECKING:
582 # To preserve static checking for undefined/misspelled attributes while letting
583 # the methods' bodies be type-checked, these are defined as non-special methods,
584 # then bound to special names out of view of static type checkers. (The original
585 # names invoke name mangling (leading "__") to avoid confusion in other scopes.)
586 __getattribute__ = __getattribute
587 __setattr__ = __setattr
590GitMeta = _GitMeta
591"""Alias of :class:`Git`'s metaclass, whether it is :class:`type` or a custom metaclass.
593Whether the :class:`Git` class has the default :class:`type` as its metaclass or uses a
594custom metaclass is not documented and may change at any time. This statically checkable
595metaclass alias is equivalent at runtime to ``type(Git)``. This should almost never be
596used. Code that benefits from it is likely to be remain brittle even if it is used.
598In view of the :class:`Git` class's intended use and :class:`Git` objects' dynamic
599callable attributes representing git subcommands, it rarely makes sense to inherit from
600:class:`Git` at all. Using :class:`Git` in multiple inheritance can be especially tricky
601to do correctly. Attempting uses of :class:`Git` where its metaclass is relevant, such
602as when a sibling class has an unrelated metaclass and a shared lower bound metaclass
603might have to be introduced to solve a metaclass conflict, is not recommended.
605:note:
606 The correct static type of the :class:`Git` class itself, and any subclasses, is
607 ``Type[Git]``. (This can be written as ``type[Git]`` in Python 3.9 later.)
609 :class:`GitMeta` should never be used in any annotation where ``Type[Git]`` is
610 intended or otherwise possible to use. This alias is truly only for very rare and
611 inherently precarious situations where it is necessary to deal with the metaclass
612 explicitly.
613"""
616class Git(metaclass=_GitMeta):
617 """The Git class manages communication with the Git binary.
619 It provides a convenient interface to calling the Git binary, such as in::
621 g = Git( git_dir )
622 g.init() # calls 'git init' program
623 rval = g.ls_files() # calls 'git ls-files' program
625 Debugging:
627 * Set the :envvar:`GIT_PYTHON_TRACE` environment variable to print each invocation
628 of the command to stdout.
629 * Set its value to ``full`` to see details about the returned values.
630 """
632 __slots__ = (
633 "_working_dir",
634 "cat_file_all",
635 "cat_file_header",
636 "_version_info",
637 "_version_info_token",
638 "_git_options",
639 "_persistent_git_options",
640 "_environment",
641 )
643 _excluded_ = (
644 "cat_file_all",
645 "cat_file_header",
646 "_version_info",
647 "_version_info_token",
648 )
650 re_unsafe_protocol = re.compile(r"(.+)::.+")
652 unsafe_git_ls_remote_options = [
653 # This option allows arbitrary command execution in git-ls-remote.
654 "--upload-pack",
655 "--exec",
656 ]
658 unsafe_git_pathspec_from_file_options = [
659 # Reads pathspecs from a caller-controlled file. Some commands include an
660 # unmatched pathspec in their error output, which can disclose the file.
661 "--pathspec-from-file",
662 ]
664 def __getstate__(self) -> Dict[str, Any]:
665 return slots_to_dict(self, exclude=self._excluded_)
667 def __setstate__(self, d: Dict[str, Any]) -> None:
668 dict_to_slots_and__excluded_are_none(self, d, excluded=self._excluded_)
670 # CONFIGURATION
672 git_exec_name = "git"
673 """Default git command that should work on Linux, Windows, and other systems."""
675 GIT_PYTHON_TRACE = os.environ.get("GIT_PYTHON_TRACE", False)
676 """Enables debugging of GitPython's git commands."""
678 USE_SHELL: bool = False
679 """Deprecated. If set to ``True``, a shell will be used when executing git commands.
681 Code that uses ``USE_SHELL = True`` or that passes ``shell=True`` to any GitPython
682 functions should be updated to use the default value of ``False`` instead. ``True``
683 is unsafe unless the effect of syntax treated specially by the shell is fully
684 considered and accounted for, which is not possible under most circumstances. As
685 detailed below, it is also no longer needed, even where it had been in the past.
687 It is in many if not most cases a command injection vulnerability for an application
688 to set :attr:`USE_SHELL` to ``True``. Any attacker who can cause a specially crafted
689 fragment of text to make its way into any part of any argument to any git command
690 (including paths, branch names, etc.) can cause the shell to read and write
691 arbitrary files and execute arbitrary commands. Innocent input may also accidentally
692 contain special shell syntax, leading to inadvertent malfunctions.
694 In addition, how a value of ``True`` interacts with some aspects of GitPython's
695 operation is not precisely specified and may change without warning, even before
696 GitPython 4.0.0 when :attr:`USE_SHELL` may be removed. This includes:
698 * Whether or how GitPython automatically customizes the shell environment.
700 * Whether, outside of Windows (where :class:`subprocess.Popen` supports lists of
701 separate arguments even when ``shell=True``), this can be used with any GitPython
702 functionality other than direct calls to the :meth:`execute` method.
704 * Whether any GitPython feature that runs git commands ever attempts to partially
705 sanitize data a shell may treat specially. Currently this is not done.
707 Prior to GitPython 2.0.8, this had a narrow purpose in suppressing console windows
708 in graphical Windows applications. In 2.0.8 and higher, it provides no benefit, as
709 GitPython solves that problem more robustly and safely by using the
710 ``CREATE_NO_WINDOW`` process creation flag on Windows.
712 Because Windows path search differs subtly based on whether a shell is used, in rare
713 cases changing this from ``True`` to ``False`` may keep an unusual git "executable",
714 such as a batch file, from being found. To fix this, set the command name or full
715 path in the :envvar:`GIT_PYTHON_GIT_EXECUTABLE` environment variable or pass the
716 full path to :func:`git.refresh` (or invoke the script using a ``.exe`` shim).
718 Further reading:
720 * :meth:`Git.execute` (on the ``shell`` parameter).
721 * https://github.com/gitpython-developers/GitPython/commit/0d9390866f9ce42870d3116094cd49e0019a970a
722 * https://learn.microsoft.com/en-us/windows/win32/procthread/process-creation-flags
723 * https://github.com/python/cpython/issues/91558#issuecomment-1100942950
724 * https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-createprocessw
725 """
727 _git_exec_env_var = "GIT_PYTHON_GIT_EXECUTABLE"
728 _refresh_env_var = "GIT_PYTHON_REFRESH"
730 GIT_PYTHON_GIT_EXECUTABLE = None
731 """Provide the full path to the git executable. Otherwise it assumes git is in the
732 executable search path.
734 :note:
735 The git executable is actually found during the refresh step in the top level
736 ``__init__``. It can also be changed by explicitly calling :func:`git.refresh`.
737 """
739 _refresh_token = object() # Since None would match an initial _version_info_token.
741 @classmethod
742 def refresh(cls, path: Union[None, PathLike] = None) -> bool:
743 """Update information about the git executable :class:`Git` objects will use.
745 Called by the :func:`git.refresh` function in the top level ``__init__``.
747 :param path:
748 Optional path to the git executable. If not absolute, it is resolved
749 immediately, relative to the current directory. (See note below.)
751 :note:
752 The top-level :func:`git.refresh` should be preferred because it calls this
753 method and may also update other state accordingly.
755 :note:
756 There are three different ways to specify the command that refreshing causes
757 to be used for git:
759 1. Pass no `path` argument and do not set the
760 :envvar:`GIT_PYTHON_GIT_EXECUTABLE` environment variable. The command
761 name ``git`` is used. It is looked up in a path search by the system, in
762 each command run (roughly similar to how git is found when running
763 ``git`` commands manually). This is usually the desired behavior.
765 2. Pass no `path` argument but set the :envvar:`GIT_PYTHON_GIT_EXECUTABLE`
766 environment variable. The command given as the value of that variable is
767 used. This may be a simple command or an arbitrary path. It is looked up
768 in each command run. Setting :envvar:`GIT_PYTHON_GIT_EXECUTABLE` to
769 ``git`` has the same effect as not setting it.
771 3. Pass a `path` argument. This path, if not absolute, is immediately
772 resolved, relative to the current directory. This resolution occurs at
773 the time of the refresh. When git commands are run, they are run using
774 that previously resolved path. If a `path` argument is passed, the
775 :envvar:`GIT_PYTHON_GIT_EXECUTABLE` environment variable is not
776 consulted.
778 :note:
779 Refreshing always sets the :attr:`Git.GIT_PYTHON_GIT_EXECUTABLE` class
780 attribute, which can be read on the :class:`Git` class or any of its
781 instances to check what command is used to run git. This attribute should
782 not be confused with the related :envvar:`GIT_PYTHON_GIT_EXECUTABLE`
783 environment variable. The class attribute is set no matter how refreshing is
784 performed.
785 """
786 # Discern which path to refresh with.
787 if path is not None:
788 new_git = os.path.expanduser(path)
789 new_git = os.path.abspath(new_git)
790 else:
791 new_git = os.environ.get(cls._git_exec_env_var, cls.git_exec_name)
793 # Keep track of the old and new git executable path.
794 old_git = cls.GIT_PYTHON_GIT_EXECUTABLE
795 old_refresh_token = cls._refresh_token
796 cls.GIT_PYTHON_GIT_EXECUTABLE = new_git
797 cls._refresh_token = object()
799 # Test if the new git executable path is valid. A GitCommandNotFound error is
800 # raised by us. A PermissionError is raised if the git executable cannot be
801 # executed for whatever reason.
802 has_git = False
803 try:
804 cls().version()
805 has_git = True
806 except (GitCommandNotFound, PermissionError):
807 pass
809 # Warn or raise exception if test failed.
810 if not has_git:
811 err = (
812 dedent(
813 """\
814 Bad git executable.
815 The git executable must be specified in one of the following ways:
816 - be included in your $PATH
817 - be set via $%s
818 - explicitly set via git.refresh(<full-path-to-git-executable>)
819 """
820 )
821 % cls._git_exec_env_var
822 )
824 # Revert to whatever the old_git was.
825 cls.GIT_PYTHON_GIT_EXECUTABLE = old_git
826 cls._refresh_token = old_refresh_token
828 if old_git is None:
829 # On the first refresh (when GIT_PYTHON_GIT_EXECUTABLE is None) we only
830 # are quiet, warn, or error depending on the GIT_PYTHON_REFRESH value.
832 # Determine what the user wants to happen during the initial refresh. We
833 # expect GIT_PYTHON_REFRESH to either be unset or be one of the
834 # following values:
835 #
836 # 0|q|quiet|s|silence|silent|n|none
837 # 1|w|warn|warning|l|log
838 # 2|r|raise|e|error|exception
840 mode = os.environ.get(cls._refresh_env_var, "raise").lower()
842 quiet = ["quiet", "q", "silence", "s", "silent", "none", "n", "0"]
843 warn = ["warn", "w", "warning", "log", "l", "1"]
844 error = ["error", "e", "exception", "raise", "r", "2"]
846 if mode in quiet:
847 pass
848 elif mode in warn or mode in error:
849 err = dedent(
850 """\
851 %s
852 All git commands will error until this is rectified.
854 This initial message can be silenced or aggravated in the future by setting the
855 $%s environment variable. Use one of the following values:
856 - %s: for no message or exception
857 - %s: for a warning message (logging level CRITICAL, displayed by default)
858 - %s: for a raised exception
860 Example:
861 export %s=%s
862 """
863 ) % (
864 err,
865 cls._refresh_env_var,
866 "|".join(quiet),
867 "|".join(warn),
868 "|".join(error),
869 cls._refresh_env_var,
870 quiet[0],
871 )
873 if mode in warn:
874 _logger.critical(err)
875 else:
876 raise ImportError(err)
877 else:
878 err = dedent(
879 """\
880 %s environment variable has been set but it has been set with an invalid value.
882 Use only the following values:
883 - %s: for no message or exception
884 - %s: for a warning message (logging level CRITICAL, displayed by default)
885 - %s: for a raised exception
886 """
887 ) % (
888 cls._refresh_env_var,
889 "|".join(quiet),
890 "|".join(warn),
891 "|".join(error),
892 )
893 raise ImportError(err)
895 # We get here if this was the initial refresh and the refresh mode was
896 # not error. Go ahead and set the GIT_PYTHON_GIT_EXECUTABLE such that we
897 # discern the difference between the first refresh at import time
898 # and subsequent calls to git.refresh or this refresh method.
899 cls.GIT_PYTHON_GIT_EXECUTABLE = cls.git_exec_name
900 else:
901 # After the first refresh (when GIT_PYTHON_GIT_EXECUTABLE is no longer
902 # None) we raise an exception.
903 raise GitCommandNotFound(new_git, err)
905 return has_git
907 @classmethod
908 def is_cygwin(cls) -> bool:
909 return is_cygwin_git(cls.GIT_PYTHON_GIT_EXECUTABLE)
911 @overload
912 @classmethod
913 def polish_url(cls, url: str, is_cygwin: Literal[False] = ..., expand_vars: bool = ...) -> str: ...
915 @overload
916 @classmethod
917 def polish_url(cls, url: str, is_cygwin: Union[None, bool] = None, expand_vars: bool = True) -> str: ...
919 @classmethod
920 def polish_url(cls, url: str, is_cygwin: Union[None, bool] = None, expand_vars: bool = True) -> PathLike:
921 """Remove any backslashes from URLs to be written in config files.
923 Windows might create config files containing paths with backslashes, but git
924 stops liking them as it will escape the backslashes. Hence we undo the escaping
925 just to be sure.
927 :param expand_vars:
928 Expand environment variables and an initial ``~``. Disable this for values
929 obtained from an untrusted source, such as remote URLs.
930 """
931 if is_cygwin is None:
932 is_cygwin = cls.is_cygwin()
934 if is_cygwin:
935 url = cygpath(url, expand_vars=expand_vars)
936 else:
937 if expand_vars:
938 url = os.path.expandvars(url)
939 if url.startswith("~"):
940 url = os.path.expanduser(url)
941 url = url.replace("\\\\", "\\").replace("\\", "/")
942 return url
944 @classmethod
945 def check_unsafe_protocols(cls, url: str) -> None:
946 """Check for unsafe protocols.
948 Apart from the usual protocols (http, git, ssh), Git allows "remote helpers"
949 that have the form ``<transport>::<address>``. One of these helpers (``ext::``)
950 can be used to invoke any arbitrary command.
952 See:
954 - https://git-scm.com/docs/gitremote-helpers
955 - https://git-scm.com/docs/git-remote-ext
956 """
957 match = cls.re_unsafe_protocol.match(url)
958 if match:
959 protocol = match.group(1)
960 raise UnsafeProtocolError(
961 f"The `{protocol}::` protocol looks suspicious, use `allow_unsafe_protocols=True` to allow it."
962 )
964 @classmethod
965 def _canonicalize_option_name(cls, option: str) -> str:
966 """Return the option name used for unsafe-option checks.
968 Examples:
969 ``"--upload-pack=/tmp/helper"`` -> ``"upload-pack"``
970 ``"upload_pack"`` -> ``"upload-pack"``
971 ``"--config core.filemode=false"`` -> ``"config"``
972 """
973 option_name = option.lstrip("-").split("=", 1)[0]
974 option_tokens = option_name.split(None, 1)
975 if not option_tokens:
976 return ""
977 return dashify(option_tokens[0])
979 @classmethod
980 def check_unsafe_options(
981 cls, options: List[str], unsafe_options: List[str], clusterable_short_options: str = "46flnqsv"
982 ) -> None:
983 """Raise :class:`~git.exc.UnsafeOptionError` for blocked option spellings.
985 In addition to exact matches, this rejects abbreviated long options accepted
986 by Git (for example, ``--upl`` for ``--upload-pack``) and unsafe short options
987 whose values are joined to the same token, including after clusterable flags
988 (for example, ``-uVALUE`` and ``-fuVALUE``).
990 A list containing only bare names is treated as normalized keyword arguments,
991 so multi-character names such as ``upload_p`` are checked as long-option
992 abbreviations. If any item starts with ``-``, the list is treated as tokenized
993 command-line input: bare items can be option values and are not checked as
994 abbreviations. Thus ``["--origin", "upload"]`` is allowed. Single-dash options
995 use short-option parsing rather than broad prefix matching, preserving safe
996 attached values such as ``-oupstream`` and ``-bcurrent``.
998 Some options passed to ``git <command>`` can execute arbitrary commands and
999 are therefore blocked by default unless the caller explicitly allows them.
1000 """
1001 # Options can be of the form `foo`, `--foo`, `--foo bar`, or `--foo=bar`.
1002 # Git accepts any unambiguous prefix of a long option, so an abbreviated
1003 # spelling such as `--upl` for `--upload-pack` must be rejected too. An
1004 # option is unsafe if its canonical name is a prefix of any blocked
1005 # option's canonical name. Only long options and multi-character kwargs
1006 # can be abbreviations; single-character short options remain exact-match
1007 # only.
1008 canonical_unsafe_options = {cls._canonicalize_option_name(option): option for option in unsafe_options}
1009 unsafe_short_options = {
1010 canonical: option
1011 for canonical, option in canonical_unsafe_options.items()
1012 if option.startswith("-") and not option.startswith("--") and len(canonical) == 1
1013 }
1014 # These value-less Git flags can be clustered before another short option
1015 # (for example, ``-fuVALUE``). Stop at any other character because it may
1016 # begin an attached value, as ``o`` does in the safe option ``-oupstream``.
1017 clusterable_short_options_set = frozenset(clusterable_short_options)
1018 options_are_kwargs = all(not option.startswith("-") for option in options)
1019 for option in options:
1020 candidate = cls._canonicalize_option_name(option)
1021 if not candidate:
1022 continue
1023 unsafe_option = canonical_unsafe_options.get(candidate)
1024 if unsafe_option is not None:
1025 raise UnsafeOptionError(f"{unsafe_option} is not allowed, use `allow_unsafe_options=True` to allow it.")
1026 option_token = option.split("=", 1)[0].split(None, 1)[0]
1027 if option_token.startswith("-") and not option_token.startswith("--"):
1028 for option_char in option_token[1:]:
1029 unsafe_option = unsafe_short_options.get(option_char)
1030 if unsafe_option is not None:
1031 raise UnsafeOptionError(
1032 f"{unsafe_option} is not allowed, use `allow_unsafe_options=True` to allow it."
1033 )
1034 if option_char not in clusterable_short_options_set:
1035 break
1036 if not (option.startswith("--") or (options_are_kwargs and len(candidate) > 1)):
1037 continue
1038 for canonical, unsafe_option in canonical_unsafe_options.items():
1039 if canonical.startswith(candidate):
1040 raise UnsafeOptionError(
1041 f"{unsafe_option} is not allowed, use `allow_unsafe_options=True` to allow it."
1042 )
1044 @classmethod
1045 def _option_candidates(cls, args: Sequence[Any] = (), kwargs: Optional[Mapping[str, Any]] = None) -> List[str]:
1046 """Collect possible option spellings before command-line transformation."""
1047 options = [
1048 option for option in cls._unpack_args([arg for arg in args if arg is not None]) if option.startswith("-")
1049 ]
1050 if kwargs:
1051 split_single_char_options = kwargs.get("split_single_char_options", True)
1052 for key, value in kwargs.items():
1053 values = value if isinstance(value, (list, tuple)) else (value,)
1054 if any(value is True or (value is not False and value is not None) for value in values):
1055 key = str(key)
1056 if len(key) != 1:
1057 options.append(f"--{dashify(key)}")
1058 elif split_single_char_options:
1059 options.append(f"-{key}")
1060 options.extend(
1061 str(value)
1062 for value in values
1063 if value is not True and value not in (False, None) and str(value).startswith("-")
1064 )
1065 else:
1066 options.extend(
1067 f"-{key}" if value is True else f"-{key}{value}"
1068 for value in values
1069 if value is True or (value is not False and value is not None)
1070 )
1071 return options
1073 AutoInterrupt: TypeAlias = _AutoInterrupt
1075 CatFileContentStream: TypeAlias = _CatFileContentStream
1077 def __init__(self, working_dir: Union[None, PathLike] = None) -> None:
1078 """Initialize this instance with:
1080 :param working_dir:
1081 Git directory we should work in. If ``None``, we always work in the current
1082 directory as returned by :func:`os.getcwd`.
1083 This is meant to be the working tree directory if available, or the
1084 ``.git`` directory in case of bare repositories.
1085 """
1086 super().__init__()
1087 self._working_dir = expand_path(working_dir)
1088 self._git_options: Union[List[str], Tuple[str, ...]] = ()
1089 self._persistent_git_options: List[str] = []
1091 # Extra environment variables to pass to git commands
1092 self._environment: Dict[str, str] = {}
1094 # Cached version slots
1095 self._version_info: Union[Tuple[int, ...], None] = None
1096 self._version_info_token: object = None
1098 # Cached command slots
1099 self.cat_file_header: Union[None, TBD] = None
1100 self.cat_file_all: Union[None, TBD] = None
1102 def __getattribute__(self, name: str) -> Any:
1103 if name == "USE_SHELL":
1104 _warn_use_shell(extra_danger=False)
1105 return super().__getattribute__(name)
1107 def __getattr__(self, name: str) -> Any:
1108 """A convenience method as it allows to call the command as if it was an object.
1110 :return:
1111 Callable object that will execute call :meth:`_call_process` with your
1112 arguments.
1113 """
1114 if name.startswith("_"):
1115 return super().__getattribute__(name)
1116 return lambda *args, **kwargs: self._call_process(name, *args, **kwargs)
1118 def set_persistent_git_options(self, **kwargs: Any) -> None:
1119 """Specify command line options to the git executable for subsequent
1120 subcommand calls.
1122 :param kwargs:
1123 A dict of keyword arguments.
1124 These arguments are passed as in :meth:`_call_process`, but will be passed
1125 to the git command rather than the subcommand.
1126 """
1128 self._persistent_git_options = self.transform_kwargs(split_single_char_options=True, **kwargs)
1130 def ls_remote(
1131 self,
1132 *args: Any,
1133 allow_unsafe_options: bool = False,
1134 **kwargs: Any,
1135 ) -> Union[str, bytes, Tuple[int, Union[str, bytes], str], "Git.AutoInterrupt"]:
1136 """List references in a remote repository.
1138 :param allow_unsafe_options:
1139 Allow unsafe options, like ``--upload-pack`` or ``--exec``.
1140 """
1141 if not allow_unsafe_options:
1142 candidate_options = self._option_candidates(args, kwargs)
1143 Git.check_unsafe_options(options=candidate_options, unsafe_options=self.unsafe_git_ls_remote_options)
1144 return self._call_process("ls_remote", *args, **kwargs)
1146 @property
1147 def working_dir(self) -> Union[None, PathLike]:
1148 """:return: Git directory we are working on"""
1149 return self._working_dir
1151 @property
1152 def version_info(self) -> Tuple[int, ...]:
1153 """
1154 :return: Tuple with integers representing the major, minor and additional
1155 version numbers as parsed from :manpage:`git-version(1)`. Up to four fields
1156 are used.
1158 This value is generated on demand and is cached.
1159 """
1160 # Refreshing is global, but version_info caching is per-instance.
1161 refresh_token = self._refresh_token # Copy token in case of concurrent refresh.
1163 # Use the cached version if obtained after the most recent refresh.
1164 if self._version_info_token is refresh_token:
1165 assert self._version_info is not None, "Bug: corrupted token-check state"
1166 return self._version_info
1168 # Run "git version" and parse it.
1169 process_version = self._call_process("version")
1170 version_string = process_version.split(" ")[2]
1171 version_fields = version_string.split(".")[:4]
1172 leading_numeric_fields = itertools.takewhile(str.isdigit, version_fields)
1173 self._version_info = tuple(map(int, leading_numeric_fields))
1175 # This value will be considered valid until the next refresh.
1176 self._version_info_token = refresh_token
1177 return self._version_info
1179 @overload
1180 def execute(
1181 self,
1182 command: Union[str, Sequence[Any]],
1183 *,
1184 as_process: Literal[True],
1185 ) -> "AutoInterrupt": ...
1187 @overload
1188 def execute(
1189 self,
1190 command: Union[str, Sequence[Any]],
1191 *,
1192 as_process: Literal[False] = False,
1193 stdout_as_string: Literal[True],
1194 ) -> Union[str, Tuple[int, str, str]]: ...
1196 @overload
1197 def execute(
1198 self,
1199 command: Union[str, Sequence[Any]],
1200 *,
1201 as_process: Literal[False] = False,
1202 stdout_as_string: Literal[False] = False,
1203 ) -> Union[bytes, Tuple[int, bytes, str]]: ...
1205 @overload
1206 def execute(
1207 self,
1208 command: Union[str, Sequence[Any]],
1209 *,
1210 with_extended_output: Literal[False],
1211 as_process: Literal[False],
1212 stdout_as_string: Literal[True],
1213 ) -> str: ...
1215 @overload
1216 def execute(
1217 self,
1218 command: Union[str, Sequence[Any]],
1219 *,
1220 with_extended_output: Literal[False],
1221 as_process: Literal[False],
1222 stdout_as_string: Literal[False],
1223 ) -> bytes: ...
1225 def execute(
1226 self,
1227 command: Union[str, Sequence[Any]],
1228 istream: Union[None, BinaryIO] = None,
1229 with_extended_output: bool = False,
1230 with_exceptions: bool = True,
1231 as_process: bool = False,
1232 output_stream: Union[None, BinaryIO] = None,
1233 stdout_as_string: bool = True,
1234 kill_after_timeout: Union[None, float] = None,
1235 with_stdout: bool = True,
1236 universal_newlines: bool = False,
1237 shell: Union[None, bool] = None,
1238 env: Union[None, Mapping[str, str]] = None,
1239 max_chunk_size: int = io.DEFAULT_BUFFER_SIZE,
1240 strip_newline_in_stdout: bool = True,
1241 **subprocess_kwargs: Any,
1242 ) -> Union[str, bytes, Tuple[int, Union[str, bytes], str], AutoInterrupt]:
1243 R"""Handle executing the command, and consume and return the returned
1244 information (stdout).
1246 :param command:
1247 The command to execute. A sequence of program arguments is recommended.
1248 A string is also accepted, but its meaning is strongly platform-dependent.
1250 By default, a shell is not used. On Unix-like systems, a string is the whole
1251 program name (so ``"git log -n 1"`` raises :class:`GitCommandNotFound`). On
1252 Windows, the program parses the arguments itself, so multi-word strings can
1253 work but are not portable.
1255 Avoid ``shell=True`` (and :attr:`Git.USE_SHELL`): this runs the command in
1256 a shell, which is generally unsafe. The shell interprets metacharacters
1257 such as ``;``, ``|``, ``&``, ``$(...)``, ``$VAR``, ``%VAR%``, and ``^``
1258 (depending on the platform) as syntax. Any untrusted text in the command
1259 can then execute arbitrary OS commands. See :attr:`Git.USE_SHELL`.
1261 Producing a sequence automatically by :func:`shlex.split` and passing it
1262 as the command is far safer than ``shell=True``. But :func:`shlex.split`
1263 parses POSIX shell syntax on all systems, and the result is still unsafe
1264 for anything but *fixed, fully trusted* strings. Do not use it on strings
1265 built by interpolating values: whitespace or quoting in an untrusted value
1266 can still inject arguments. For input derived in any way from untrusted
1267 data, build the argument sequence yourself, while ensuring each argument
1268 is fully sanitized.
1270 :param istream:
1271 Standard input filehandle passed to :class:`subprocess.Popen`.
1273 :param with_extended_output:
1274 Whether to return a (status, stdout, stderr) tuple.
1276 :param with_exceptions:
1277 Whether to raise an exception when git returns a non-zero status.
1279 :param as_process:
1280 Whether to return the created process instance directly from which
1281 streams can be read on demand. This will render `with_extended_output`
1282 and `with_exceptions` ineffective - the caller will have to deal with
1283 the details. It is important to note that the process will be placed
1284 into an :class:`AutoInterrupt` wrapper that will interrupt the process
1285 once it goes out of scope. If you use the command in iterators, you
1286 should pass the whole process instance instead of a single stream.
1288 :param output_stream:
1289 If set to a file-like object, data produced by the git command will be
1290 copied to the given stream instead of being returned as a string.
1291 This feature only has any effect if `as_process` is ``False``.
1293 :param stdout_as_string:
1294 If ``False``, the command's standard output will be bytes. Otherwise, it
1295 will be decoded into a string using the default encoding (usually UTF-8).
1296 The latter can fail, if the output contains binary data.
1298 :param kill_after_timeout:
1299 Specifies a timeout in seconds for the git command, after which the process
1300 should be killed. This will have no effect if `as_process` is set to
1301 ``True``. It is set to ``None`` by default and will let the process run
1302 until the timeout is explicitly specified. Uses of this feature should be
1303 carefully considered, due to the following limitations:
1305 1. This feature is not supported at all on Windows.
1306 2. Effectiveness may vary by operating system. ``ps --ppid`` is used to
1307 enumerate child processes, which is available on most GNU/Linux systems
1308 but not most others.
1309 3. Deeper descendants do not receive signals, though they may sometimes
1310 terminate as a consequence of their parent processes being killed.
1311 4. `kill_after_timeout` uses ``SIGKILL``, which can have negative side
1312 effects on a repository. For example, stale locks in case of
1313 :manpage:`git-gc(1)` could render the repository incapable of accepting
1314 changes until the lock is manually removed.
1316 :param with_stdout:
1317 If ``True``, default ``True``, we open stdout on the created process.
1319 :param universal_newlines:
1320 If ``True``, pipes will be opened as text, and lines are split at all known
1321 line endings.
1323 :param shell:
1324 Whether to invoke commands through a shell
1325 (see :class:`Popen(..., shell=True) <subprocess.Popen>`).
1326 If this is not ``None``, it overrides :attr:`USE_SHELL`.
1328 Passing ``shell=True`` to this or any other GitPython function should be
1329 avoided, as it is unsafe under most circumstances. This is because it is
1330 typically not feasible to fully consider and account for the effect of shell
1331 expansions, especially when passing ``shell=True`` to other methods that
1332 forward it to :meth:`Git.execute`. Passing ``shell=True`` is also no longer
1333 needed (nor useful) to work around any known operating system specific
1334 issues.
1336 On Unix-like systems, when migrating away from passing string commands with
1337 ``shell=True``, :func:`shlex.split` may serve as a transitional step in rare
1338 cases, with extreme care. (Drop ``shell=True`` and pass the resulting
1339 sequence as the command.) See the `command` parameter above on the risks.
1341 :param env:
1342 A dictionary of environment variables to be passed to
1343 :class:`subprocess.Popen`.
1345 :param max_chunk_size:
1346 Maximum number of bytes in one chunk of data passed to the `output_stream`
1347 in one invocation of its ``write()`` method. If the given number is not
1348 positive then the default value is used.
1350 :param strip_newline_in_stdout:
1351 Whether to strip the trailing ``\n`` of the command stdout.
1353 :param subprocess_kwargs:
1354 Keyword arguments to be passed to :class:`subprocess.Popen`. Please note
1355 that some of the valid kwargs are already set by this method; the ones you
1356 specify may not be the same ones.
1358 :return:
1359 * str(output), if `extended_output` is ``False`` (Default)
1360 * tuple(int(status), str(stdout), str(stderr)),
1361 if `extended_output` is ``True``
1363 If `output_stream` is ``True``, the stdout value will be your output stream:
1365 * output_stream, if `extended_output` is ``False``
1366 * tuple(int(status), output_stream, str(stderr)),
1367 if `extended_output` is ``True``
1369 Note that git is executed with ``LC_MESSAGES="C"`` to ensure consistent
1370 output regardless of system language.
1372 :raise git.exc.GitCommandError:
1374 :note:
1375 If you add additional keyword arguments to the signature of this method, you
1376 must update the ``execute_kwargs`` variable housed in this module.
1377 """
1378 # Remove password for the command if present.
1379 redacted_command = remove_password_if_present(command)
1380 if self.GIT_PYTHON_TRACE and (self.GIT_PYTHON_TRACE != "full" or as_process):
1381 _logger.info(" ".join(redacted_command))
1383 # Allow the user to have the command executed in their working dir.
1384 try:
1385 cwd = self._working_dir or os.getcwd() # type: Optional[PathLike]
1386 if not os.access(str(cwd), os.X_OK):
1387 cwd = None
1388 except FileNotFoundError:
1389 cwd = None
1391 # Start the process.
1392 inline_env = env
1393 env = os.environ.copy()
1394 # Attempt to force all output to plain ASCII English, which is what some parsing
1395 # code may expect.
1396 # According to https://askubuntu.com/a/311796, we are setting LANGUAGE as well
1397 # just to be sure.
1398 env["LANGUAGE"] = "C"
1399 env["LC_ALL"] = "C"
1400 env.update(self._environment)
1401 if inline_env is not None:
1402 env.update(inline_env)
1404 if sys.platform == "win32":
1405 if kill_after_timeout is not None:
1406 raise GitCommandError(
1407 redacted_command,
1408 '"kill_after_timeout" feature is not supported on Windows.',
1409 )
1410 cmd_not_found_exception = OSError
1411 else:
1412 cmd_not_found_exception = FileNotFoundError
1413 # END handle
1415 stdout_sink = PIPE if with_stdout else getattr(subprocess, "DEVNULL", None) or open(os.devnull, "wb")
1416 if shell is None:
1417 # Get the value of USE_SHELL with no deprecation warning. Do this without
1418 # warnings.catch_warnings, to avoid a race condition with application code
1419 # configuring warnings. The value could be looked up in type(self).__dict__
1420 # or Git.__dict__, but those can break under some circumstances. This works
1421 # the same as self.USE_SHELL in more situations; see Git.__getattribute__.
1422 shell = super().__getattribute__("USE_SHELL")
1423 _logger.debug(
1424 "Popen(%s, cwd=%s, stdin=%s, shell=%s, universal_newlines=%s)",
1425 redacted_command,
1426 cwd,
1427 "<valid stream>" if istream else "None",
1428 shell,
1429 universal_newlines,
1430 )
1431 try:
1432 proc = safer_popen(
1433 command,
1434 env=env,
1435 cwd=cwd,
1436 bufsize=-1,
1437 stdin=(istream or DEVNULL),
1438 stderr=PIPE,
1439 stdout=stdout_sink,
1440 shell=shell,
1441 universal_newlines=universal_newlines,
1442 encoding=defenc if universal_newlines else None,
1443 **subprocess_kwargs,
1444 )
1445 except cmd_not_found_exception as err:
1446 raise GitCommandNotFound(redacted_command, err) from err
1447 else:
1448 # Replace with a typeguard for Popen[bytes]?
1449 proc.stdout = cast(BinaryIO, proc.stdout)
1450 proc.stderr = cast(BinaryIO, proc.stderr)
1452 if as_process:
1453 return self.AutoInterrupt(proc, command)
1455 watchdog: Optional[threading.Timer] = None
1456 kill_check: Optional[threading.Event] = None
1457 timeout_error: Optional[Callable[[], Union[str, bytes]]] = None
1459 if sys.platform != "win32" and kill_after_timeout is not None:
1460 # Help mypy figure out this is not None even when used inside communicate().
1461 timeout = kill_after_timeout
1463 def kill_process(pid: int) -> None:
1464 """Callback to kill a process.
1466 This callback implementation would be ineffective and unsafe on Windows.
1467 """
1468 p = Popen(["ps", "--ppid", str(pid)], stdout=PIPE)
1469 child_pids = []
1470 if p.stdout is not None:
1471 for line in p.stdout:
1472 if len(line.split()) > 0:
1473 local_pid = (line.split())[0]
1474 if local_pid.isdigit():
1475 child_pids.append(int(local_pid))
1476 try:
1477 os.kill(pid, signal.SIGKILL)
1478 for child_pid in child_pids:
1479 try:
1480 os.kill(child_pid, signal.SIGKILL)
1481 except OSError:
1482 pass
1483 # Tell the main routine that the process was killed.
1484 assert kill_check is not None
1485 kill_check.set()
1486 except OSError:
1487 # It is possible that the process gets completed in the duration
1488 # after timeout happens and before we try to kill the process.
1489 pass
1490 return
1492 def make_timeout_error() -> Union[str, bytes]:
1493 err = f'Timeout: the command "{" ".join(redacted_command)}" did not complete in {timeout:g} secs.'
1494 return err if universal_newlines else err.encode(defenc)
1496 def communicate() -> Tuple[AnyStr, AnyStr]:
1497 assert watchdog is not None
1498 assert kill_check is not None
1499 watchdog.start()
1500 out, err = proc.communicate()
1501 watchdog.cancel()
1502 if kill_check.is_set():
1503 err = make_timeout_error()
1504 return out, err
1506 # END helpers
1508 kill_check = threading.Event()
1509 watchdog = threading.Timer(timeout, kill_process, args=(proc.pid,))
1510 timeout_error = make_timeout_error
1511 else:
1512 communicate = proc.communicate
1514 # Wait for the process to return.
1515 status = 0
1516 stdout_value: Union[str, bytes] = b""
1517 stderr_value: Union[str, bytes] = b""
1518 newline = "\n" if universal_newlines else b"\n"
1519 try:
1520 if output_stream is None:
1521 stdout_value, stderr_value = communicate()
1522 # Strip trailing "\n".
1523 if stdout_value is not None and stdout_value.endswith(newline) and strip_newline_in_stdout: # type: ignore[arg-type]
1524 stdout_value = stdout_value[:-1]
1525 if stderr_value is not None and stderr_value.endswith(newline): # type: ignore[arg-type]
1526 stderr_value = stderr_value[:-1]
1528 status = proc.returncode
1529 else:
1530 max_chunk_size = max_chunk_size if max_chunk_size and max_chunk_size > 0 else io.DEFAULT_BUFFER_SIZE
1531 if watchdog is not None:
1532 watchdog.start()
1533 try:
1534 if proc.stdout is not None:
1535 stream_copy(proc.stdout, output_stream, max_chunk_size)
1536 stdout_value = proc.stdout.read()
1537 if proc.stderr is not None:
1538 stderr_value = proc.stderr.read()
1539 status = proc.wait()
1540 finally:
1541 if watchdog is not None:
1542 watchdog.cancel()
1543 # Strip trailing "\n".
1544 if stderr_value is not None and stderr_value.endswith(newline): # type: ignore[arg-type]
1545 stderr_value = stderr_value[:-1]
1546 if kill_check is not None and kill_check.is_set():
1547 assert timeout_error is not None
1548 stderr_value = timeout_error()
1549 # END stdout handling
1550 finally:
1551 if proc.stdout is not None:
1552 proc.stdout.close()
1553 if proc.stderr is not None:
1554 proc.stderr.close()
1556 if self.GIT_PYTHON_TRACE == "full":
1557 cmdstr = " ".join(redacted_command)
1559 def as_text(stdout_value: Union[bytes, str]) -> str:
1560 return not output_stream and safe_decode(stdout_value) or "<OUTPUT_STREAM>"
1562 # END as_text
1564 if stderr_value:
1565 _logger.info(
1566 "%s -> %d; stdout: '%s'; stderr: '%s'",
1567 cmdstr,
1568 status,
1569 as_text(stdout_value),
1570 safe_decode(stderr_value),
1571 )
1572 elif stdout_value:
1573 _logger.info("%s -> %d; stdout: '%s'", cmdstr, status, as_text(stdout_value))
1574 else:
1575 _logger.info("%s -> %d", cmdstr, status)
1576 # END handle debug printing
1578 if with_exceptions and status != 0:
1579 raise GitCommandError(redacted_command, status, stderr_value, stdout_value)
1581 if isinstance(stdout_value, bytes) and stdout_as_string: # Could also be output_stream.
1582 stdout_value = safe_decode(stdout_value)
1584 # Allow access to the command's status code.
1585 if with_extended_output:
1586 return (status, stdout_value, safe_decode(stderr_value))
1587 else:
1588 return stdout_value
1590 def environment(self) -> Dict[str, str]:
1591 return self._environment
1593 def update_environment(self, **kwargs: Any) -> Dict[str, Union[str, None]]:
1594 """Set environment variables for future git invocations. Return all changed
1595 values in a format that can be passed back into this function to revert the
1596 changes.
1598 Examples::
1600 old_env = self.update_environment(PWD='/tmp')
1601 self.update_environment(**old_env)
1603 :param kwargs:
1604 Environment variables to use for git processes.
1606 :return:
1607 Dict that maps environment variables to their old values
1608 """
1609 old_env = {}
1610 for key, value in kwargs.items():
1611 # Set value if it is None.
1612 if value is not None:
1613 old_env[key] = self._environment.get(key)
1614 self._environment[key] = value
1615 # Remove key from environment if its value is None.
1616 elif key in self._environment:
1617 old_env[key] = self._environment[key]
1618 del self._environment[key]
1619 return old_env
1621 @contextlib.contextmanager
1622 def custom_environment(self, **kwargs: Any) -> Iterator[None]:
1623 """A context manager around the above :meth:`update_environment` method to
1624 restore the environment back to its previous state after operation.
1626 Examples::
1628 with self.custom_environment(GIT_SSH='/bin/ssh_wrapper'):
1629 repo.remotes.origin.fetch()
1631 :param kwargs:
1632 See :meth:`update_environment`.
1633 """
1634 old_env = self.update_environment(**kwargs)
1635 try:
1636 yield
1637 finally:
1638 self.update_environment(**old_env)
1640 def transform_kwarg(self, name: str, value: Any, split_single_char_options: bool) -> List[str]:
1641 if len(name) == 1:
1642 if value is True:
1643 return ["-%s" % name]
1644 elif value not in (False, None):
1645 if split_single_char_options:
1646 return ["-%s" % name, "%s" % value]
1647 else:
1648 return ["-%s%s" % (name, value)]
1649 else:
1650 if value is True:
1651 return ["--%s" % dashify(name)]
1652 elif value is not False and value is not None:
1653 return ["--%s=%s" % (dashify(name), value)]
1654 return []
1656 def transform_kwargs(self, split_single_char_options: bool = True, **kwargs: Any) -> List[str]:
1657 """Transform Python-style kwargs into git command line options."""
1658 args = []
1659 for k, v in kwargs.items():
1660 if isinstance(v, (list, tuple)):
1661 for value in v:
1662 args += self.transform_kwarg(k, value, split_single_char_options)
1663 else:
1664 args += self.transform_kwarg(k, v, split_single_char_options)
1665 return args
1667 @classmethod
1668 def _unpack_args(cls, arg_list: Sequence[Any]) -> List[str]:
1669 outlist = []
1670 if isinstance(arg_list, (list, tuple)):
1671 for arg in arg_list:
1672 outlist.extend(cls._unpack_args(arg))
1673 else:
1674 outlist.append(str(arg_list))
1676 return outlist
1678 def __call__(self, **kwargs: Any) -> "Git":
1679 """Specify command line options to the git executable for a subcommand call.
1681 :param kwargs:
1682 A dict of keyword arguments.
1683 These arguments are passed as in :meth:`_call_process`, but will be passed
1684 to the git command rather than the subcommand.
1686 Examples::
1688 git(work_tree='/tmp').difftool()
1689 """
1690 self._git_options = self.transform_kwargs(split_single_char_options=True, **kwargs)
1691 return self
1693 @overload
1694 def _call_process(
1695 self, method: str, *args: None, **kwargs: None
1696 ) -> str: ... # If no args were given, execute the call with all defaults.
1698 @overload
1699 def _call_process(
1700 self,
1701 method: str,
1702 istream: int,
1703 as_process: Literal[True],
1704 *args: Any,
1705 **kwargs: Any,
1706 ) -> "Git.AutoInterrupt": ...
1708 @overload
1709 def _call_process(
1710 self, method: str, *args: Any, **kwargs: Any
1711 ) -> Union[str, bytes, Tuple[int, Union[str, bytes], str], "Git.AutoInterrupt"]: ...
1713 def _call_process(
1714 self, method: str, *args: Any, **kwargs: Any
1715 ) -> Union[str, bytes, Tuple[int, Union[str, bytes], str], "Git.AutoInterrupt"]:
1716 """Run the given git command with the specified arguments and return the result
1717 as a string.
1719 :param method:
1720 The command. Contained ``_`` characters will be converted to hyphens, such
1721 as in ``ls_files`` to call ``ls-files``.
1723 :param args:
1724 The list of arguments. If ``None`` is included, it will be pruned.
1725 This allows your commands to call git more conveniently, as ``None`` is
1726 realized as non-existent.
1728 :param kwargs:
1729 Contains key-values for the following:
1731 - The :meth:`execute()` kwds, as listed in ``execute_kwargs``.
1732 - "Command options" to be converted by :meth:`transform_kwargs`.
1733 - The ``insert_kwargs_after`` key which its value must match one of
1734 ``*args``.
1736 It also contains any command options, to be appended after the matched arg.
1738 Examples::
1740 git.rev_list('master', max_count=10, header=True)
1742 turns into::
1744 git rev-list --max-count=10 --header master
1746 :return:
1747 Same as :meth:`execute`. If no args are given, used :meth:`execute`'s
1748 default (especially ``as_process = False``, ``stdout_as_string = True``) and
1749 return :class:`str`.
1750 """
1751 # Handle optional arguments prior to calling transform_kwargs.
1752 # Otherwise these'll end up in args, which is bad.
1753 exec_kwargs = {k: v for k, v in kwargs.items() if k in execute_kwargs}
1754 opts_kwargs = {k: v for k, v in kwargs.items() if k not in execute_kwargs}
1756 insert_after_this_arg = opts_kwargs.pop("insert_kwargs_after", None)
1758 # Prepare the argument list.
1760 opt_args = self.transform_kwargs(**opts_kwargs)
1761 ext_args = self._unpack_args([a for a in args if a is not None])
1763 if insert_after_this_arg is None:
1764 args_list = opt_args + ext_args
1765 else:
1766 try:
1767 index = ext_args.index(insert_after_this_arg)
1768 except ValueError as err:
1769 raise ValueError(
1770 "Couldn't find argument '%s' in args %s to insert cmd options after"
1771 % (insert_after_this_arg, str(ext_args))
1772 ) from err
1773 # END handle error
1774 args_list = ext_args[: index + 1] + opt_args + ext_args[index + 1 :]
1775 # END handle opts_kwargs
1777 call = [self.GIT_PYTHON_GIT_EXECUTABLE]
1779 # Add persistent git options.
1780 call.extend(self._persistent_git_options)
1782 # Add the git options, then reset to empty to avoid side effects.
1783 call.extend(self._git_options)
1784 self._git_options = ()
1786 call.append(dashify(method))
1787 call.extend(args_list)
1789 return self.execute(call, **exec_kwargs)
1791 def _parse_object_header(self, header_line: str) -> Tuple[str, str, int]:
1792 """
1793 :param header_line:
1794 A line of the form::
1796 <hex_sha> type_string size_as_int
1798 :return:
1799 (hex_sha, type_string, size_as_int)
1801 :raise ValueError:
1802 If the header contains indication for an error due to incorrect input sha.
1803 """
1804 tokens = header_line.split()
1805 if len(tokens) != 3:
1806 if not tokens:
1807 err_msg = (
1808 f"SHA is empty, possible dubious ownership in the repository "
1809 f"""at {self._working_dir}.\n If this is unintended run:\n\n """
1810 f""" "git config --global --add safe.directory {self._working_dir}" """
1811 )
1812 raise ValueError(err_msg)
1813 else:
1814 raise ValueError("SHA %s could not be resolved, git returned: %r" % (tokens[0], header_line.strip()))
1815 # END handle actual return value
1816 # END error handling
1818 if len(tokens[0]) != 40:
1819 raise ValueError("Failed to parse header: %r" % header_line)
1820 return (tokens[0], tokens[1], int(tokens[2]))
1822 def _prepare_ref(self, ref: AnyStr) -> bytes:
1823 # Required for command to separate refs on stdin, as bytes.
1824 if isinstance(ref, bytes):
1825 # Assume 40 bytes hexsha - bin-to-ascii for some reason returns bytes, not text.
1826 refstr: str = ref.decode("ascii")
1827 elif not isinstance(ref, str):
1828 refstr = str(ref) # Could be ref-object.
1829 else:
1830 refstr = ref
1832 if not refstr.endswith("\n"):
1833 refstr += "\n"
1834 return refstr.encode(defenc)
1836 def _get_persistent_cmd(self, attr_name: str, cmd_name: str, *args: Any, **kwargs: Any) -> "Git.AutoInterrupt":
1837 cur_val = getattr(self, attr_name)
1838 if cur_val is not None:
1839 return cur_val
1841 options = {"istream": PIPE, "as_process": True}
1842 options.update(kwargs)
1844 cmd = self._call_process(cmd_name, *args, **options)
1845 setattr(self, attr_name, cmd)
1846 cmd = cast("Git.AutoInterrupt", cmd)
1847 return cmd
1849 def __get_object_header(self, cmd: "Git.AutoInterrupt", ref: AnyStr) -> Tuple[str, str, int]:
1850 if cmd.stdin and cmd.stdout:
1851 cmd.stdin.write(self._prepare_ref(ref))
1852 cmd.stdin.flush()
1853 return self._parse_object_header(cmd.stdout.readline())
1854 else:
1855 raise ValueError("cmd stdin was empty")
1857 def get_object_header(self, ref: str) -> Tuple[str, str, int]:
1858 """Use this method to quickly examine the type and size of the object behind the
1859 given ref.
1861 :note:
1862 The method will only suffer from the costs of command invocation once and
1863 reuses the command in subsequent calls.
1865 :return:
1866 (hexsha, type_string, size_as_int)
1867 """
1868 cmd = self._get_persistent_cmd("cat_file_header", "cat_file", batch_check=True)
1869 return self.__get_object_header(cmd, ref)
1871 def get_object_data(self, ref: str) -> Tuple[str, str, int, bytes]:
1872 """Similar to :meth:`get_object_header`, but returns object data as well.
1874 :return:
1875 (hexsha, type_string, size_as_int, data_string)
1877 :note:
1878 Not threadsafe.
1879 """
1880 hexsha, typename, size, stream = self.stream_object_data(ref)
1881 data = stream.read(size)
1882 del stream
1883 return (hexsha, typename, size, data)
1885 def stream_object_data(self, ref: str) -> Tuple[str, str, int, "Git.CatFileContentStream"]:
1886 """Similar to :meth:`get_object_data`, but returns the data as a stream.
1888 :return:
1889 (hexsha, type_string, size_as_int, stream)
1891 :note:
1892 This method is not threadsafe. You need one independent :class:`Git`
1893 instance per thread to be safe!
1894 """
1895 cmd = self._get_persistent_cmd("cat_file_all", "cat_file", batch=True)
1896 hexsha, typename, size = self.__get_object_header(cmd, ref)
1897 cmd_stdout = cmd.stdout if cmd.stdout is not None else io.BytesIO()
1898 return (hexsha, typename, size, self.CatFileContentStream(size, cmd_stdout))
1900 def clear_cache(self) -> "Git":
1901 """Clear all kinds of internal caches to release resources.
1903 Currently persistent commands will be interrupted.
1905 :return:
1906 self
1907 """
1908 for cmd in (self.cat_file_all, self.cat_file_header):
1909 if cmd:
1910 cmd.__del__()
1912 self.cat_file_all = None
1913 self.cat_file_header = None
1914 return self