# Copyright (C) 2008, 2009 Michael Trier ([email protected]) and contributors
#
# This module is part of GitPython and is released under the
# 3-Clause BSD License: https://opensource.org/license/bsd-3-clause/
from __future__ import annotations
__all__ = ["GitMeta", "Git"]
import contextlib
import io
import itertools
import logging
import os
import re
import signal
import subprocess
from subprocess import DEVNULL, PIPE, Popen
import sys
from textwrap import dedent
import threading
import warnings
from git.compat import defenc, force_bytes, safe_decode
from git.exc import (
CommandError,
GitCommandError,
GitCommandNotFound,
UnsafeOptionError,
UnsafeProtocolError,
)
from git.util import (
cygpath,
expand_path,
is_cygwin_git,
patch_env,
remove_password_if_present,
stream_copy,
)
# typing ---------------------------------------------------------------------------
from typing import (
Any,
AnyStr,
BinaryIO,
Callable,
Dict,
IO,
Iterator,
List,
Mapping,
Optional,
Sequence,
TYPE_CHECKING,
TextIO,
Tuple,
Union,
cast,
overload,
)
if sys.version_info >= (3, 10):
from typing import TypeAlias
else:
from typing_extensions import TypeAlias
from git.types import Literal, PathLike, TBD
if TYPE_CHECKING:
from git.diff import DiffIndex
from git.repo.base import Repo
# ---------------------------------------------------------------------------------
execute_kwargs = {
"istream",
"with_extended_output",
"with_exceptions",
"as_process",
"output_stream",
"stdout_as_string",
"kill_after_timeout",
"with_stdout",
"universal_newlines",
"shell",
"env",
"max_chunk_size",
"strip_newline_in_stdout",
}
_logger = logging.getLogger(__name__)
# ==============================================================================
## @name Utilities
# ------------------------------------------------------------------------------
# Documentation
## @{
def handle_process_output(
process: "Git.AutoInterrupt" | Popen,
stdout_handler: Union[
None,
Callable[[AnyStr], None],
Callable[[List[AnyStr]], None],
Callable[[bytes, "Repo", "DiffIndex"], None],
],
stderr_handler: Union[None, Callable[[AnyStr], None], Callable[[List[AnyStr]], None]],
finalizer: Union[None, Callable[[Union[Popen, "Git.AutoInterrupt"]], None]] = None,
decode_streams: bool = True,
kill_after_timeout: Union[None, float] = None,
) -> None:
R"""Register for notifications to learn that process output is ready to read, and
dispatch lines to the respective line handlers.
This function returns once the finalizer returns.
:param process:
:class:`subprocess.Popen` instance.
:param stdout_handler:
f(stdout_line_string), or ``None``.
:param stderr_handler:
f(stderr_line_string), or ``None``.
:param finalizer:
f(proc) - wait for proc to finish.
:param decode_streams:
Assume stdout/stderr streams are binary and decode them before pushing their
contents to handlers.
This defaults to ``True``. Set it to ``False`` if:
- ``universal_newlines == True``, as then streams are in text mode, or
- decoding must happen later, such as for :class:`~git.diff.Diff`\s.
:param kill_after_timeout:
:class:`float` or ``None``, Default = ``None``
To specify a timeout in seconds for the git command, after which the process
should be killed.
"""
# Use 2 "pump" threads and wait for both to finish.
def pump_stream(
cmdline: List[str],
name: str,
stream: Union[BinaryIO, TextIO],
is_decode: bool,
handler: Union[None, Callable[[Union[bytes, str]], None]],
) -> None:
try:
for line in stream:
if handler:
if is_decode:
assert isinstance(line, bytes)
line_str = line.decode(defenc)
handler(line_str)
else:
handler(line)
except Exception as ex:
_logger.error(f"Pumping {name!r} of cmd({remove_password_if_present(cmdline)}) failed due to: {ex!r}")
if "I/O operation on closed file" not in str(ex):
# Only reraise if the error was not due to the stream closing.
raise CommandError([f"<{name}-pump>"] + remove_password_if_present(cmdline), ex) from ex
finally:
stream.close()
if hasattr(process, "proc"):
process = cast("Git.AutoInterrupt", process)
cmdline: str | Tuple[str, ...] | List[str] = getattr(process.proc, "args", "")
p_stdout = process.proc.stdout if process.proc else None
p_stderr = process.proc.stderr if process.proc else None
else:
process = cast(Popen, process) # type: ignore[redundant-cast]
cmdline = getattr(process, "args", "")
p_stdout = process.stdout
p_stderr = process.stderr
if not isinstance(cmdline, (tuple, list)):
cmdline = cmdline.split()
pumps: List[Tuple[str, IO, Callable[..., None] | None]] = []
if p_stdout:
pumps.append(("stdout", p_stdout, stdout_handler))
if p_stderr:
pumps.append(("stderr", p_stderr, stderr_handler))
threads: List[threading.Thread] = []
for name, stream, handler in pumps:
t = threading.Thread(target=pump_stream, args=(cmdline, name, stream, decode_streams, handler))
t.daemon = True
t.start()
threads.append(t)
# FIXME: Why join? Will block if stdin needs feeding...
for t in threads:
t.join(timeout=kill_after_timeout)
if t.is_alive():
if isinstance(process, Git.AutoInterrupt):
process._terminate()
else: # Don't want to deal with the other case.
raise RuntimeError(
"Thread join() timed out in cmd.handle_process_output()."
f" kill_after_timeout={kill_after_timeout} seconds"
)
if stderr_handler:
error_str: Union[str, bytes] = (
f"error: process killed because it timed out. kill_after_timeout={kill_after_timeout} seconds"
)
if not decode_streams and isinstance(p_stderr, BinaryIO):
# Assume stderr_handler needs binary input.
error_str = cast(str, error_str)
error_str = error_str.encode()
# We ignore typing on the next line because mypy does not like the way
# we inferred that stderr takes str or bytes.
stderr_handler(error_str) # type: ignore[arg-type]
if finalizer:
finalizer(process)
safer_popen: Callable[..., Popen]
if sys.platform == "win32":
def _safer_popen_windows(
command: Union[str, Sequence[Any]],
*,
shell: bool = False,
env: Optional[Mapping[str, str]] = None,
**kwargs: Any,
) -> Popen:
"""Call :class:`subprocess.Popen` on Windows but don't include a CWD in the
search.
This avoids an untrusted search path condition where a file like ``git.exe`` in
a malicious repository would be run when GitPython operates on the repository.
The process using GitPython may have an untrusted repository's working tree as
its current working directory. Some operations may temporarily change to that
directory before running a subprocess. In addition, while by default GitPython
does not run external commands with a shell, it can be made to do so, in which
case the CWD of the subprocess, which GitPython usually sets to a repository
working tree, can itself be searched automatically by the shell. This wrapper
covers all those cases.
:note:
This currently works by setting the
:envvar:`NoDefaultCurrentDirectoryInExePath` environment variable during
subprocess creation. It also takes care of passing Windows-specific process
creation flags, but that is unrelated to path search.
:note:
The current implementation contains a race condition on :attr:`os.environ`.
GitPython isn't thread-safe, but a program using it on one thread should
ideally be able to mutate :attr:`os.environ` on another, without
unpredictable results. See comments in:
https://github.com/gitpython-developers/GitPython/pull/1650
"""
# CREATE_NEW_PROCESS_GROUP is needed for some ways of killing it afterwards.
# https://docs.python.org/3/library/subprocess.html#subprocess.Popen.send_signal
# https://docs.python.org/3/library/subprocess.html#subprocess.CREATE_NEW_PROCESS_GROUP
creationflags = subprocess.CREATE_NO_WINDOW | subprocess.CREATE_NEW_PROCESS_GROUP
# When using a shell, the shell is the direct subprocess, so the variable must
# be set in its environment, to affect its search behavior.
if shell:
# The original may be immutable, or the caller may reuse it. Mutate a copy.
env = {} if env is None else dict(env)
env["NoDefaultCurrentDirectoryInExePath"] = "1" # The "1" can be any value.
# When not using a shell, the current process does the search in a
# CreateProcessW API call, so the variable must be set in our environment. With
# a shell, that's unnecessary if https://github.com/python/cpython/issues/101283
# is patched. In Python versions where it is unpatched, in the rare case the
# ComSpec environment variable is unset, the search for the shell itself is
# unsafe. Setting NoDefaultCurrentDirectoryInExePath in all cases, as done here,
# is simpler and protects against that. (As above, the "1" can be any value.)
with patch_env("NoDefaultCurrentDirectoryInExePath", "1"):
return Popen(
command,
shell=shell,
env=env,
creationflags=creationflags,
**kwargs,
)
safer_popen = _safer_popen_windows
else:
safer_popen = Popen
def dashify(string: str) -> str:
return string.replace("_", "-")
def slots_to_dict(self: "Git", exclude: Sequence[str] = ()) -> Dict[str, Any]:
return {s: getattr(self, s) for s in self.__slots__ if s not in exclude}
def dict_to_slots_and__excluded_are_none(self: object, d: Mapping[str, Any], excluded: Sequence[str] = ()) -> None:
for k, v in d.items():
setattr(self, k, v)
for k in excluded:
setattr(self, k, None)
## -- End Utilities -- @}
class _AutoInterrupt:
"""Process wrapper that terminates the wrapped process on finalization.
This kills/interrupts the stored process instance once this instance goes out of
scope. It is used to prevent processes piling up in case iterators stop reading.
All attributes are wired through to the contained process object.
The wait method is overridden to perform automatic status code checking and possibly
raise.
"""
__slots__ = ("proc", "args", "status")
# If this is non-zero it will override any status code during _terminate, used
# to prevent race conditions in testing.
_status_code_if_terminate: int = 0
def __init__(self, proc: Union[None, subprocess.Popen], args: Any) -> None:
self.proc = proc
self.args = args
self.status: Union[int, None] = None
def _terminate(self) -> None:
"""Terminate the underlying process."""
if self.proc is None:
return
proc = self.proc
self.proc = None
if proc.stdin:
proc.stdin.close()
if proc.stdout:
proc.stdout.close()
if proc.stderr:
proc.stderr.close()
# Did the process finish already so we have a return code?
try:
if proc.poll() is not None:
self.status = self._status_code_if_terminate or proc.poll()
return
except OSError as ex:
_logger.info("Ignored error after process had died: %r", ex)
# It can be that nothing really exists anymore...
if os is None or getattr(os, "kill", None) is None:
return
# Try to kill it.
try:
proc.terminate()
status = proc.wait() # Ensure the process goes away.
self.status = self._status_code_if_terminate or status
except OSError as ex:
_logger.info("Ignored error after process had died: %r", ex)
# END exception handling
def __del__(self) -> None:
self._terminate()
def __getattr__(self, attr: str) -> Any:
return getattr(self.proc, attr)
# TODO: Bad choice to mimic `proc.wait()` but with different args.
def wait(self, stderr: Union[None, str, bytes] = b"") -> int:
"""Wait for the process and return its status code.
:param stderr:
Previously read value of stderr, in case stderr is already closed.
:warn:
May deadlock if output or error pipes are used and not handled separately.
:raise git.exc.GitCommandError:
If the return status is not 0.
"""
if stderr is None:
stderr_b = b""
stderr_b = force_bytes(data=stderr, encoding="utf-8")
status: Union[int, None]
if self.proc is not None:
status = self.proc.wait()
p_stderr = self.proc.stderr
else: # Assume the underlying proc was killed earlier or never existed.
status = self.status
p_stderr = None
def read_all_from_possibly_closed_stream(stream: Union[IO[bytes], None]) -> bytes:
if stream:
try:
return stderr_b + force_bytes(stream.read())
except (OSError, ValueError):
return stderr_b or b""
else:
return stderr_b or b""
# END status handling
if status != 0:
errstr = read_all_from_possibly_closed_stream(p_stderr)
_logger.debug("AutoInterrupt wait stderr: %r" % (errstr,))
raise GitCommandError(remove_password_if_present(self.args), status, errstr)
return status
_AutoInterrupt.__name__ = "AutoInterrupt"
_AutoInterrupt.__qualname__ = "Git.AutoInterrupt"
class _CatFileContentStream:
"""Object representing a sized read-only stream returning the contents of
an object.
This behaves like a stream, but counts the data read and simulates an empty stream
once our sized content region is empty.
If not all data are read to the end of the object's lifetime, we read the rest to
ensure the underlying stream continues to work.
"""
__slots__ = ("_stream", "_nbr", "_size")
def __init__(self, size: int, stream: IO[bytes]) -> None:
self._stream = stream
self._size = size
self._nbr = 0 # Number of bytes read.
# Special case: If the object is empty, has null bytes, get the final
# newline right away.
if size == 0:
stream.read(1)
# END handle empty streams
def read(self, size: int = -1) -> bytes:
bytes_left = self._size - self._nbr
if bytes_left == 0:
return b""
if size > -1:
# Ensure we don't try to read past our limit.
size = min(bytes_left, size)
else:
# They try to read all, make sure it's not more than what remains.
size = bytes_left
# END check early depletion
data = self._stream.read(size)
self._nbr += len(data)
# Check for depletion, read our final byte to make the stream usable by
# others.
if self._size - self._nbr == 0:
self._stream.read(1) # final newline
# END finish reading
return data
def readline(self, size: int = -1) -> bytes:
if self._nbr == self._size:
return b""
# Clamp size to lowest allowed value.
bytes_left = self._size - self._nbr
if size > -1:
size = min(bytes_left, size)
else:
size = bytes_left
# END handle size
data = self._stream.readline(size)
self._nbr += len(data)
# Handle final byte.
if self._size - self._nbr == 0:
self._stream.read(1)
# END finish reading
return data
def readlines(self, size: int = -1) -> List[bytes]:
if self._nbr == self._size:
return []
# Leave all additional logic to our readline method, we just check the size.
out = []
nbr = 0
while True:
line = self.readline()
if not line:
break
out.append(line)
if size > -1:
nbr += len(line)
if nbr > size:
break
# END handle size constraint
# END readline loop
return out
# skipcq: PYL-E0301
def __iter__(self) -> "Git.CatFileContentStream":
return self
def __next__(self) -> bytes:
line = self.readline()
if not line:
raise StopIteration
return line
next = __next__
def __del__(self) -> None:
bytes_left = self._size - self._nbr
if bytes_left:
# Read and discard - seeking is impossible within a stream.
# This includes any terminating newline.
self._stream.read(bytes_left + 1)
# END handle incomplete read
_CatFileContentStream.__name__ = "CatFileContentStream"
_CatFileContentStream.__qualname__ = "Git.CatFileContentStream"
_USE_SHELL_DEFAULT_MESSAGE = (
"Git.USE_SHELL is deprecated, because only its default value of False is safe. "
"It will be removed in a future release."
)
_USE_SHELL_DANGER_MESSAGE = (
"Setting Git.USE_SHELL to True is unsafe and insecure, as the effect of special "
"shell syntax cannot usually be accounted for. This can result in a command "
"injection vulnerability and arbitrary code execution. Git.USE_SHELL is deprecated "
"and will be removed in a future release."
)
def _warn_use_shell(*, extra_danger: bool) -> None:
warnings.warn(
_USE_SHELL_DANGER_MESSAGE if extra_danger else _USE_SHELL_DEFAULT_MESSAGE,
DeprecationWarning,
stacklevel=3,
)
class _GitMeta(type):
"""Metaclass for :class:`Git`.
This helps issue :class:`DeprecationWarning` if :attr:`Git.USE_SHELL` is used.
"""
def __getattribute(cls, name: str) -> Any:
if name == "USE_SHELL":
_warn_use_shell(extra_danger=False)
return super().__getattribute__(name)
def __setattr(cls, name: str, value: Any) -> Any:
if name == "USE_SHELL":
_warn_use_shell(extra_danger=value)
super().__setattr__(name, value)
if not TYPE_CHECKING:
# To preserve static checking for undefined/misspelled attributes while letting
# the methods' bodies be type-checked, these are defined as non-special methods,
# then bound to special names out of view of static type checkers. (The original
# names invoke name mangling (leading "__") to avoid confusion in other scopes.)
__getattribute__ = __getattribute
__setattr__ = __setattr
GitMeta = _GitMeta
"""Alias of :class:`Git`'s metaclass, whether it is :class:`type` or a custom metaclass.
Whether the :class:`Git` class has the default :class:`type` as its metaclass or uses a
custom metaclass is not documented and may change at any time. This statically checkable
metaclass alias is equivalent at runtime to ``type(Git)``. This should almost never be
used. Code that benefits from it is likely to be remain brittle even if it is used.
In view of the :class:`Git` class's intended use and :class:`Git` objects' dynamic
callable attributes representing git subcommands, it rarely makes sense to inherit from
:class:`Git` at all. Using :class:`Git` in multiple inheritance can be especially tricky
to do correctly. Attempting uses of :class:`Git` where its metaclass is relevant, such
as when a sibling class has an unrelated metaclass and a shared lower bound metaclass
might have to be introduced to solve a metaclass conflict, is not recommended.
:note:
The correct static type of the :class:`Git` class itself, and any subclasses, is
``Type[Git]``. (This can be written as ``type[Git]`` in Python 3.9 later.)
:class:`GitMeta` should never be used in any annotation where ``Type[Git]`` is
intended or otherwise possible to use. This alias is truly only for very rare and
inherently precarious situations where it is necessary to deal with the metaclass
explicitly.
"""
class Git(metaclass=_GitMeta):
"""The Git class manages communication with the Git binary.
It provides a convenient interface to calling the Git binary, such as in::
g = Git( git_dir )
g.init() # calls 'git init' program
rval = g.ls_files() # calls 'git ls-files' program
Debugging:
* Set the :envvar:`GIT_PYTHON_TRACE` environment variable to print each invocation
of the command to stdout.
* Set its value to ``full`` to see details about the returned values.
"""
__slots__ = (
"_working_dir",
"cat_file_all",
"cat_file_header",
"_version_info",
"_version_info_token",
"_git_options",
"_persistent_git_options",
"_environment",
)
_excluded_ = (
"cat_file_all",
"cat_file_header",
"_version_info",
"_version_info_token",
)
re_unsafe_protocol = re.compile(r"(.+)::.+")
def __getstate__(self) -> Dict[str, Any]:
return slots_to_dict(self, exclude=self._excluded_)
def __setstate__(self, d: Dict[str, Any]) -> None:
dict_to_slots_and__excluded_are_none(self, d, excluded=self._excluded_)
# CONFIGURATION
git_exec_name = "git"
"""Default git command that should work on Linux, Windows, and other systems."""
GIT_PYTHON_TRACE = os.environ.get("GIT_PYTHON_TRACE", False)
"""Enables debugging of GitPython's git commands."""
USE_SHELL: bool = False
"""Deprecated. If set to ``True``, a shell will be used when executing git commands.
Code that uses ``USE_SHELL = True`` or that passes ``shell=True`` to any GitPython
functions should be updated to use the default value of ``False`` instead. ``True``
is unsafe unless the effect of syntax treated specially by the shell is fully
considered and accounted for, which is not possible under most circumstances. As
detailed below, it is also no longer needed, even where it had been in the past.
It is in many if not most cases a command injection vulnerability for an application
to set :attr:`USE_SHELL` to ``True``. Any attacker who can cause a specially crafted
fragment of text to make its way into any part of any argument to any git command
(including paths, branch names, etc.) can cause the shell to read and write
arbitrary files and execute arbitrary commands. Innocent input may also accidentally
contain special shell syntax, leading to inadvertent malfunctions.
In addition, how a value of ``True`` interacts with some aspects of GitPython's
operation is not precisely specified and may change without warning, even before
GitPython 4.0.0 when :attr:`USE_SHELL` may be removed. This includes:
* Whether or how GitPython automatically customizes the shell environment.
* Whether, outside of Windows (where :class:`subprocess.Popen` supports lists of
separate arguments even when ``shell=True``), this can be used with any GitPython
functionality other than direct calls to the :meth:`execute` method.
* Whether any GitPython feature that runs git commands ever attempts to partially
sanitize data a shell may treat specially. Currently this is not done.
Prior to GitPython 2.0.8, this had a narrow purpose in suppressing console windows
in graphical Windows applications. In 2.0.8 and higher, it provides no benefit, as
GitPython solves that problem more robustly and safely by using the
``CREATE_NO_WINDOW`` process creation flag on Windows.
Because Windows path search differs subtly based on whether a shell is used, in rare
cases changing this from ``True`` to ``False`` may keep an unusual git "executable",
such as a batch file, from being found. To fix this, set the command name or full
path in the :envvar:`GIT_PYTHON_GIT_EXECUTABLE` environment variable or pass the
full path to :func:`git.refresh` (or invoke the script using a ``.exe`` shim).
Further reading:
* :meth:`Git.execute` (on the ``shell`` parameter).
* https://github.com/gitpython-developers/GitPython/commit/0d9390866f9ce42870d3116094cd49e0019a970a
* https://learn.microsoft.com/en-us/windows/win32/procthread/process-creation-flags
* https://github.com/python/cpython/issues/91558#issuecomment-1100942950
* https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-createprocessw
"""
_git_exec_env_var = "GIT_PYTHON_GIT_EXECUTABLE"
_refresh_env_var = "GIT_PYTHON_REFRESH"
GIT_PYTHON_GIT_EXECUTABLE = None
"""Provide the full path to the git executable. Otherwise it assumes git is in the
executable search path.
:note:
The git executable is actually found during the refresh step in the top level
``__init__``. It can also be changed by explicitly calling :func:`git.refresh`.
"""
_refresh_token = object() # Since None would match an initial _version_info_token.
@classmethod
def refresh(cls, path: Union[None, PathLike] = None) -> bool:
"""Update information about the git executable :class:`Git` objects will use.
Called by the :func:`git.refresh` function in the top level ``__init__``.
:param path:
Optional path to the git executable. If not absolute, it is resolved
immediately, relative to the current directory. (See note below.)
:note:
The top-level :func:`git.refresh` should be preferred because it calls this
method and may also update other state accordingly.
:note:
There are three different ways to specify the command that refreshing causes
to be used for git:
1. Pass no `path` argument and do not set the
:envvar:`GIT_PYTHON_GIT_EXECUTABLE` environment variable. The command
name ``git`` is used. It is looked up in a path search by the system, in
each command run (roughly similar to how git is found when running
``git`` commands manually). This is usually the desired behavior.
2. Pass no `path` argument but set the :envvar:`GIT_PYTHON_GIT_EXECUTABLE`
environment variable. The command given as the value of that variable is
used. This may be a simple command or an arbitrary path. It is looked up
in each command run. Setting :envvar:`GIT_PYTHON_GIT_EXECUTABLE` to
``git`` has the same effect as not setting it.
3. Pass a `path` argument. This path, if not absolute, is immediately
resolved, relative to the current directory. This resolution occurs at
the time of the refresh. When git commands are run, they are run using
that previously resolved path. If a `path` argument is passed, the
:envvar:`GIT_PYTHON_GIT_EXECUTABLE` environment variable is not
consulted.
:note:
Refreshing always sets the :attr:`Git.GIT_PYTHON_GIT_EXECUTABLE` class
attribute, which can be read on the :class:`Git` class or any of its
instances to check what command is used to run git. This attribute should
not be confused with the related :envvar:`GIT_PYTHON_GIT_EXECUTABLE`
environment variable. The class attribute is set no matter how refreshing is
performed.
"""
# Discern which path to refresh with.
if path is not None:
new_git = os.path.expanduser(path)
new_git = os.path.abspath(new_git)
else:
new_git = os.environ.get(cls._git_exec_env_var, cls.git_exec_name)
# Keep track of the old and new git executable path.
old_git = cls.GIT_PYTHON_GIT_EXECUTABLE
old_refresh_token = cls._refresh_token
cls.GIT_PYTHON_GIT_EXECUTABLE = new_git
cls._refresh_token = object()
# Test if the new git executable path is valid. A GitCommandNotFound error is
# raised by us. A PermissionError is raised if the git executable cannot be
# executed for whatever reason.
has_git = False
try:
cls().version()
has_git = True
except (GitCommandNotFound, PermissionError):
pass
# Warn or raise exception if test failed.
if not has_git:
err = (
dedent(
"""\
Bad git executable.
The git executable must be specified in one of the following ways:
- be included in your $PATH
- be set via $%s
- explicitly set via git.refresh(