cmd.py

# cmd.py
# Copyright (C) 2008, 2009 Michael Trier (mtrier@gmail.com) and contributors
#
# This module is part of GitPython and is released under
# the BSD License: http://www.opensource.org/licenses/bsd-license.php

import os
import os.path
import sys
import select
import logging
import threading
import errno
import mmap

from contextlib import contextmanager
from signal import SIGKILL
from subprocess import (
    call,
    Popen,
    PIPE
)


from .util import (
    LazyMixin,
    stream_copy,
    WaitGroup
)
from .exc import (
    GitCommandError,
    GitCommandNotFound
)
from git.compat import (
    string_types,
    defenc,
    PY3,
    bchr,
    # just to satisfy flake8 on py3
    unicode
)

execute_kwargs = ('istream', 'with_keep_cwd', 'with_extended_output',
                  'with_exceptions', 'as_process', 'stdout_as_string',
                  'output_stream', 'with_stdout', 'timeout')

log = logging.getLogger('git.cmd')
log.addHandler(logging.NullHandler())

__all__ = ('Git', )

if sys.platform != 'win32':
    WindowsError = OSError

if PY3:
    _bchr = bchr
else:
    def _bchr(c):
        return c
# get custom byte character handling


# ==============================================================================
## @name Utilities
# ------------------------------------------------------------------------------
# Documentation
## @{

def handle_process_output(process, stdout_handler, stderr_handler, finalizer):
    """Registers for notifications to lean that process output is ready to read, and dispatches lines to
    the respective line handlers. We are able to handle carriage returns in case progress is sent by that
    mean. For performance reasons, we only apply this to stderr.
    This function returns once the finalizer returns
    :return: result of finalizer
    :param process: subprocess.Popen instance
    :param stdout_handler: f(stdout_line_string), or None
    :param stderr_hanlder: f(stderr_line_string), or None
    :param finalizer: f(proc) - wait for proc to finish"""
    fdmap = {process.stdout.fileno(): (stdout_handler, [b'']),
             process.stderr.fileno(): (stderr_handler, [b''])}

    def _parse_lines_from_buffer(buf):
        line = b''
        bi = 0
        lb = len(buf)
        while bi < lb:
            char = _bchr(buf[bi])
            bi += 1

            if char in (b'\r', b'\n') and line:
                yield bi, line
                line = b''
            else:
                line += char
            # END process parsed line
        # END while file is not done reading
    # end

    def _read_lines_from_fno(fno, last_buf_list):
        buf = os.read(fno, mmap.PAGESIZE)
        buf = last_buf_list[0] + buf

        bi = 0
        for bi, line in _parse_lines_from_buffer(buf):
            yield line
        # for each line to parse from the buffer

        # keep remainder
        last_buf_list[0] = buf[bi:]

    def _dispatch_single_line(line, handler):
        line = line.decode(defenc)
        if line and handler:
            try:
                handler(line)
            except Exception:
                # Keep reading, have to pump the lines empty nontheless
                log.error("Line handler exception on line: %s", line, exc_info=True)
            # end
        # end dispatch helper
    # end single line helper

    def _dispatch_lines(fno, handler, buf_list):
        lc = 0
        for line in _read_lines_from_fno(fno, buf_list):
            _dispatch_single_line(line, handler)
            lc += 1
        # for each line
        return lc
    # end

    def _deplete_buffer(fno, handler, buf_list, wg=None):
        lc = 0
        while True:
            line_count = _dispatch_lines(fno, handler, buf_list)
            lc += line_count
            if line_count == 0:
                break
        # end deplete buffer

        if buf_list[0]:
            _dispatch_single_line(buf_list[0], handler)
            lc += 1
        # end

        if wg:
            wg.done()

        return lc
    # end

    if hasattr(select, 'poll'):
        # poll is preferred, as select is limited to file handles up to 1024 ... . This could otherwise be
        # an issue for us, as it matters how many handles our own process has
        poll = select.poll()
        READ_ONLY = select.POLLIN | select.POLLPRI | select.POLLHUP | select.POLLERR
        CLOSED = select.POLLHUP | select.POLLERR

        poll.register(process.stdout, READ_ONLY)
        poll.register(process.stderr, READ_ONLY)

        closed_streams = set()
        while True:
            # no timeout

            try:
                poll_result = poll.poll()
            except select.error as e:
                if e.args[0] == errno.EINTR:
                    continue
                raise
            # end handle poll exception

            for fd, result in poll_result:
                if result & CLOSED:
                    closed_streams.add(fd)
                else:
                    _dispatch_lines(fd, *fdmap[fd])
                # end handle closed stream
            # end for each poll-result tuple

            if len(closed_streams) == len(fdmap):
                break
            # end its all done
        # end endless loop

        # Depelete all remaining buffers
        for fno, (handler, buf_list) in fdmap.items():
            _deplete_buffer(fno, handler, buf_list)
        # end for each file handle

        for fno in fdmap.keys():
            poll.unregister(fno)
        # end don't forget to unregister !
    else:
        # Oh ... probably we are on windows. select.select() can only handle sockets, we have files
        # The only reliable way to do this now is to use threads and wait for both to finish
        # Since the finalizer is expected to wait, we don't have to introduce our own wait primitive
        # NO: It's not enough unfortunately, and we will have to sync the threads
        wg = WaitGroup()
        for fno, (handler, buf_list) in fdmap.items():
            wg.add(1)
            t = threading.Thread(target=lambda: _deplete_buffer(fno, handler, buf_list, wg))
            t.start()
        # end
        # NOTE: Just joining threads can possibly fail as there is a gap between .start() and when it's
        # actually started, which could make the wait() call to just return because the thread is not yet
        # active
        wg.wait()
    # end

    return finalizer(process)


def dashify(string):
    return string.replace('_', '-')

## -- End Utilities -- @}


class Git(LazyMixin):

    """
    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 GIT_PYTHON_TRACE environment variable 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",
                 "_git_options", "_environment")

    # CONFIGURATION
    # The size in bytes read from stdout when copying git's output to another stream
    max_chunk_size = 1024 * 64

    git_exec_name = "git"           # default that should work on linux and windows
    git_exec_name_win = "git.cmd"   # alternate command name, windows only

    # Enables debugging of GitPython's git commands
    GIT_PYTHON_TRACE = os.environ.get("GIT_PYTHON_TRACE", False)

    # Provide the full path to the git executable. Otherwise it assumes git is in the path
    _git_exec_env_var = "GIT_PYTHON_GIT_EXECUTABLE"
    GIT_PYTHON_GIT_EXECUTABLE = os.environ.get(_git_exec_env_var, git_exec_name)

    # If True, a shell will be used when executing git commands.
    # This should only be desirable on windows, see https://github.com/gitpython-developers/GitPython/pull/126
    # for more information
    # Override this value using `Git.USE_SHELL = True`
    USE_SHELL = False

    class AutoInterrupt(object):

        """Kill/Interrupt the stored process instance once this instance goes out of scope. It is
        used to prevent processes piling up in case iterators stop reading.
        Besides all attributes are wired through to the contained process object.

        The wait method was overridden to perform automatic status code checking
        and possibly raise."""
        __slots__ = ("proc", "args")

        def __init__(self, proc, args):
            self.proc = proc
            self.args = args

        def __del__(self):
            if self.proc is None:
                return

            proc = self.proc
            self.proc = None
            if proc.stdin:
                proc.stdin.close()
            proc.stdout.close()
            proc.stderr.close()

            # did the process finish already so we have a return code ?
            if proc.poll() is not None:
                return

            # can be that nothing really exists anymore ...
            if os is None:
                return

            # try to kill it
            try:
                os.kill(proc.pid, 2)   # interrupt signal
                proc.wait()    # ensure process goes away
            except (OSError, WindowsError):
                pass  # ignore error when process already died
            except AttributeError:
                # try windows
                # for some reason, providing None for stdout/stderr still prints something. This is why
                # we simply use the shell and redirect to nul. Its slower than CreateProcess, question
                # is whether we really want to see all these messages. Its annoying no matter what.
                call(("TASKKILL /F /T /PID %s 2>nul 1>nul" % str(proc.pid)), shell=True)
            # END exception handling

        def __getattr__(self, attr):
            return getattr(self.proc, attr)

        def wait(self):
            """Wait for the process and return its status code.

            :raise GitCommandError: if the return status is not 0"""
            status = self.proc.wait()
            if status != 0:
                raise GitCommandError(self.args, status, self.proc.stderr.read())
            # END status handling
            return status
    # END auto interrupt

    class CatFileContentStream(object):

        """Object representing a sized read-only stream returning the contents of
        an object.
        It 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 is read to the end of the objects's lifetime, we read the
        rest to assure the underlying stream continues to work"""

        __slots__ = ('_stream', '_nbr', '_size')

        def __init__(self, size, stream):
            self._stream = stream
            self._size = size
            self._nbr = 0           # num 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=-1):
            bytes_left = self._size - self._nbr
            if bytes_left == 0:
                return ''
            if size > -1:
                # assure we don't try to read past our limit
                size = min(bytes_left, size)
            else:
                # they try to read all, make sure its 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=-1):
            if self._nbr == self._size:
                return ''

            # 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=-1):
            if self._nbr == self._size:
                return list()

            # leave all additional logic to our readline method, we just check the size
            out = list()
            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

        def __iter__(self):
            return self

        def next(self):
            line = self.readline()
            if not line:
                raise StopIteration
            return line

        def __del__(self):
            bytes_left = self._size - self._nbr
            if bytes_left:
                # read and discard - seeking is impossible within a stream
                # includes terminating newline
                self._stream.read(bytes_left + 1)
            # END handle incomplete read

    def __init__(self, working_dir=None):
        """Initialize this instance with:

        :param working_dir:
           Git directory we should work in. If None, we always work in the current
           directory as returned by os.getcwd().
           It is meant to be the working tree directory if available, or the
           .git directory in case of bare repositories."""
        super(Git, self).__init__()
        self._working_dir = working_dir
        self._git_options = ()

        # Extra environment variables to pass to git commands
        self._environment = {}

        # cached command slots
        self.cat_file_header = None
        self.cat_file_all = None

    def __getattr__(self, name):
        """A convenience method as it allows to call the command as if it was
        an object.
        :return: Callable object that will execute call _call_process with your arguments."""
        if name[0] == '_':
            return LazyMixin.__getattr__(self, name)
        return lambda *args, **kwargs: self._call_process(name, *args, **kwargs)

    def _set_cache_(self, attr):
        if attr == '_version_info':
            # We only use the first 4 numbers, as everthing else could be strings in fact (on windows)
            version_numbers = self._call_process('version').split(' ')[2]
            self._version_info = tuple(int(n) for n in version_numbers.split('.')[:4] if n.isdigit())
        else:
            super(Git, self)._set_cache_(attr)
        # END handle version info

    @property
    def working_dir(self):
        """:return: Git directory we are working on"""
        return self._working_dir

    @property
    def version_info(self):
        """
        :return: tuple(int, int, int, int) tuple with integers representing the major, minor
            and additional version numbers as parsed from git version.
            This value is generated on demand and is cached"""
        return self._version_info

    def execute(self, command,
                istream=None,
                with_keep_cwd=False,
                with_extended_output=False,
                with_exceptions=True,
                as_process=False,
                output_stream=None,
                stdout_as_string=True,
                timeout=None,
                with_stdout=True,
                **subprocess_kwargs
                ):
        """Handles executing the command on the shell and consumes and returns
        the returned information (stdout)

        :param command:
            The command argument list to execute.
            It should be a string, or a sequence of program arguments. The
            program to execute is the first item in the args sequence or string.

        :param istream:
            Standard input filehandle passed to subprocess.Popen.

        :param with_keep_cwd:
            Whether to use the current working directory from os.getcwd().
            The cmd otherwise uses its own working_dir that it has been initialized
            with if possible.

        :param with_extended_output:
            Whether to return a (status, stdout, stderr) tuple.

        :param with_exceptions:
            Whether to raise an exception when git returns a non-zero status.

        :param as_process:
            Whether to return the created process instance directly from which
            streams can be read on demand. This will render with_extended_output and
            with_exceptions ineffective - the caller will have
            to deal with the details himself.
            It is important to note that the process will be placed into an AutoInterrupt
            wrapper that will interrupt the process once it goes out of scope. If you
            use the command in iterators, you should pass the whole process instance
            instead of a single stream.

        :param output_stream:
            If set to a file-like object, data produced by the git command will be
            output to the given stream directly.
            This feature only has any effect if as_process is False. Processes will
            always be created with a pipe due to issues with subprocess.
            This merely is a workaround as data will be copied from the
            output pipe to the given output stream directly.
            Judging from the implementation, you shouldn't use this flag !

        :param stdout_as_string:
            if False, the commands standard output will be bytes. Otherwise, it will be
            decoded into a string using the default encoding (usually utf-8).
            The latter can fail, if the output contains binary data.

        :param subprocess_kwargs:
            Keyword arguments to be passed to subprocess.Popen. Please note that
            some of the valid kwargs are already set by this method, the ones you
            specify may not be the same ones.

        :param with_stdout: If True, default True, we open stdout on the created process

        :param timeout:
            To specify a timeout in seconds for the git command, after which the process
            should be killed. This will have no effect if as_process is set to True. It is
            set to None by default and will let the process run until the timeout is
            explicitly specified.

        :return:
            * str(output) if extended_output = False (Default)
            * tuple(int(status), str(stdout), str(stderr)) if extended_output = True

            if ouput_stream is True, the stdout value will be your output stream:
            * output_stream if extended_output = False
            * tuple(int(status), output_stream, str(stderr)) if extended_output = True

            Note git is executed with LC_MESSAGES="C" to ensure consistent
            output regardless of system language.

        :raise GitCommandError:

        :note:
           If you add additional keyword arguments to the signature of this method,
           you must update the execute_kwargs tuple housed in this module."""
        if self.GIT_PYTHON_TRACE and (self.GIT_PYTHON_TRACE != 'full' or as_process):
            log.info(' '.join(command))

        # Allow the user to have the command executed in their working dir.
        if with_keep_cwd or self._working_dir is None:
            cwd = os.getcwd()
        else:
            cwd = self._working_dir

        # Start the process
        env = os.environ.copy()
        # Attempt to force all output to plain ascii english, which is what some parsing code
        # may expect.
        # According to stackoverflow (http://goo.gl/l74GC8), we are setting LANGUAGE as well
        # just to be sure.
        env["LANGUAGE"] = "C"
        env["LC_ALL"] = "C"
        env.update(self._environment)

        if sys.platform == 'win32':
            cmd_not_found_exception = WindowsError
            if timeout:
                raise GitCommandError('"timeout" feature is not supported on Windows.')
        else:
            if sys.version_info[0] > 2:
                cmd_not_found_exception = FileNotFoundError  # NOQA # this is defined, but flake8 doesn't know
            else:
                cmd_not_found_exception = OSError
        # end handle

        try:
            proc = Popen(command,
                         env=env,
                         cwd=cwd,
                         stdin=istream,
                         stderr=PIPE,
                         stdout=with_stdout and PIPE or None,
                         shell=self.USE_SHELL,
                         close_fds=(os.name == 'posix'),  # unsupported on windows
                         **subprocess_kwargs
                         )
        except cmd_not_found_exception as err:
            raise GitCommandNotFound(str(err))

        if as_process:
            return self.AutoInterrupt(proc, command)

        def _kill_process(pid):
            """ Callback method to kill a process. """
            p = Popen(['ps', '--ppid', str(pid)], stdout=PIPE)
            child_pids = []
            for line in p.stdout:
                if len(line.split()) > 0:
                    local_pid = (line.split())[0]
                    if local_pid.isdigit():
                        child_pids.append(int(local_pid))
            try:
                os.kill(pid, SIGKILL)
                for child_pid in child_pids:
                    os.kill(child_pid, SIGKILL)
                kill_check.set()    # tell the main routine that the process was killed
            except OSError:
                # It is possible that the process gets completed in the duration after timeout
                # happens and before we try to kill the process.
                pass
            return
        # end

        if timeout:
            kill_check = threading.Event()
            watchdog = threading.Timer(timeout, _kill_process, args=(proc.pid, ))

        # Wait for the process to return
        status = 0
        stdout_value = b''
        stderr_value = b''
        try:
            if output_stream is None:
                if timeout:
                    watchdog.start()
                stdout_value, stderr_value = proc.communicate()
                if timeout:
                    watchdog.cancel()
                    if kill_check.isSet():
                        stderr_value = 'Timeout: the command "%s" did not complete in %d ' \
                                       'secs.' % (" ".join(command), timeout)
                # strip trailing "\n"
                if stdout_value.endswith(b"\n"):
                    stdout_value = stdout_value[:-1]
                if stderr_value.endswith(b"\n"):
                    stderr_value = stderr_value[:-1]
                status = proc.returncode
            else:
                stream_copy(proc.stdout, output_stream, self.max_chunk_size)
                stdout_value = output_stream
                stderr_value = proc.stderr.read()
                # strip trailing "\n"
                if stderr_value.endswith(b"\n"):
                    stderr_value = stderr_value[:-1]
                status = proc.wait()
            # END stdout handling
        finally:
            proc.stdout.close()
            proc.stderr.close()

        if self.GIT_PYTHON_TRACE == 'full':
            cmdstr = " ".join(command)

            def as_text(stdout_value):
                return not output_stream and stdout_value.decode(defenc) or '<OUTPUT_STREAM>'
            # end

            if stderr_value:
                log.info("%s -> %d; stdout: '%s'; stderr: '%s'",
                         cmdstr, status, as_text(stdout_value), stderr_value.decode(defenc))
            elif stdout_value:
                log.info("%s -> %d; stdout: '%s'", cmdstr, status, as_text(stdout_value))
            else:
                log.info("%s -> %d", cmdstr, status)
        # END handle debug printing

        if with_exceptions and status != 0:
            if with_extended_output:
                raise GitCommandError(command, status, stderr_value, stdout_value)
            else:
                raise GitCommandError(command, status, stderr_value)

        if isinstance(stdout_value, bytes) and stdout_as_string:  # could also be output_stream
            stdout_value = stdout_value.decode(defenc)

        # Allow access to the command's status code
        if with_extended_output:
            return (status, stdout_value, stderr_value.decode(defenc))
        else:
            return stdout_value

    def environment(self):
        return self._environment

    def update_environment(self, **kwargs):
        """
        Set environment variables for future git invocations. Return all changed
        values in a format that can be passed back into this function to revert
        the changes:

        ``Examples``::

            old_env = self.update_environment(PWD='/tmp')
            self.update_environment(**old_env)

        :param kwargs: environment variables to use for git processes
        :return: dict that maps environment variables to their old values
        """
        old_env = {}
        for key, value in kwargs.items():
            # set value if it is None
            if value is not None:
                if key in self._environment:
                    old_env[key] = self._environment[key]
                else:
                    old_env[key] = None
                self._environment[key] = value
            # remove key from environment if its value is None
            elif key in self._environment:
                old_env[key] = self._environment[key]
                del self._environment[key]
        return old_env

    @contextmanager
    def custom_environment(self, **kwargs):
        """
        A context manager around the above ``update_environment`` method to restore the
        environment back to its previous state after operation.

        ``Examples``::

            with self.custom_environment(GIT_SSH='/bin/ssh_wrapper'):
                repo.remotes.origin.fetch()

        :param kwargs: see update_environment
        """
        old_env = self.update_environment(**kwargs)
        try:
            yield
        finally:
            self.update_environment(**old_env)

    def transform_kwargs(self, split_single_char_options=True, **kwargs):
        """Transforms Python style kwargs into git command line options."""
        args = list()
        for k, v in kwargs.items():
            if len(k) == 1:
                if v is True:
                    args.append("-%s" % k)
                elif type(v) is not bool:
                    if split_single_char_options:
                        args.extend(["-%s" % k, "%s" % v])
                    else:
                        args.append("-%s%s" % (k, v))
            else:
                if v is True:
                    args.append("--%s" % dashify(k))
                elif type(v) is not bool:
                    args.append("--%s=%s" % (dashify(k), v))
        return args

    @classmethod
    def __unpack_args(cls, arg_list):
        if not isinstance(arg_list, (list, tuple)):
            # This is just required for unicode conversion, as subprocess can't handle it
            # However, in any other case, passing strings (usually utf-8 encoded) is totally fine
            if not PY3 and isinstance(arg_list, unicode):
                return [arg_list.encode(defenc)]
            return [str(arg_list)]

        outlist = list()
        for arg in arg_list:
            if isinstance(arg_list, (list, tuple)):
                outlist.extend(cls.__unpack_args(arg))
            elif not PY3 and isinstance(arg_list, unicode):
                outlist.append(arg_list.encode(defenc))
            # END recursion
            else:
                outlist.append(str(arg))
        # END for each arg
        return outlist

    def __call__(self, **kwargs):
        """Specify command line options to the git executable
        for a subcommand call

        :param kwargs:
            is a dict of keyword arguments.
            these arguments are passed as in _call_process
            but will be passed to the git command rather than
            the subcommand.

        ``Examples``::
            git(work_tree='/tmp').difftool()"""
        self._git_options = self.transform_kwargs(
            split_single_char_options=True, **kwargs)
        return self

    def _call_process(self, method, *args, **kwargs):
        """Run the given git command with the specified arguments and return
        the result as a String

        :param method:
            is the command. Contained "_" characters will be converted to dashes,
            such as in 'ls_files' to call 'ls-files'.

        :param args:
            is the list of arguments. If None is included, it will be pruned.
            This allows your commands to call git more conveniently as None
            is realized as non-existent

        :param kwargs:
            is a dict of keyword arguments.
            This function accepts the same optional keyword arguments
            as execute().

        ``Examples``::
            git.rev_list('master', max_count=10, header=True)

        :return: Same as ``execute``"""
        # Handle optional arguments prior to calling transform_kwargs
        # otherwise these'll end up in args, which is bad.
        _kwargs = dict()
        for kwarg in execute_kwargs:
            try:
                _kwargs[kwarg] = kwargs.pop(kwarg)
            except KeyError:
                pass

        insert_after_this_arg = kwargs.pop('insert_kwargs_after', None)

        # Prepare the argument list
        opt_args = self.transform_kwargs(**kwargs)
        ext_args = self.__unpack_args([a for a in args if a is not None])

        if insert_after_this_arg is None:
            args = opt_args + ext_args
        else:
            try:
                index = ext_args.index(insert_after_this_arg)
            except ValueError:
                raise ValueError("Couldn't find argument '%s' in args %s to insert kwargs after"
                                 % (insert_after_this_arg, str(ext_args)))
            # end handle error
            args = ext_args[:index + 1] + opt_args + ext_args[index + 1:]
        # end handle kwargs

        def make_call():
            call = [self.GIT_PYTHON_GIT_EXECUTABLE]

            # add the git options, the reset to empty
            # to avoid side_effects
            call.extend(self._git_options)
            self._git_options = ()

            call.extend([dashify(method)])
            call.extend(args)
            return call
        # END utility to recreate call after changes

        if sys.platform == 'win32':
            try:
                try:
                    return self.execute(make_call(), **_kwargs)
                except WindowsError:
                    # did we switch to git.cmd already, or was it changed from default ? permanently fail
                    if self.GIT_PYTHON_GIT_EXECUTABLE != self.git_exec_name:
                        raise
                    # END handle overridden variable
                    type(self).GIT_PYTHON_GIT_EXECUTABLE = self.git_exec_name_win

                    try:
                        return self.execute(make_call(), **_kwargs)
                    finally:
                        import warnings
                        msg = "WARNING: Automatically switched to use git.cmd as git executable"
                        msg += ", which reduces performance by ~70%."
                        msg += "It is recommended to put git.exe into the PATH or to "
                        msg += "set the %s " % self._git_exec_env_var
                        msg += "environment variable to the executable's location"
                        warnings.warn(msg)
                    # END print of warning
                # END catch first failure
            except WindowsError:
                raise WindowsError("The system cannot find or execute the file at %r" % self.GIT_PYTHON_GIT_EXECUTABLE)
            # END provide better error message
        else:
            return self.execute(make_call(), **_kwargs)
        # END handle windows default installation

    def _parse_object_header(self, header_line):
        """
        :param header_line:
            <hex_sha> type_string size_as_int

        :return: (hex_sha, type_string, size_as_int)

        :raise ValueError: if the header contains indication for an error due to
            incorrect input sha"""
        tokens = header_line.split()
        if len(tokens) != 3:
            if not tokens:
                raise ValueError("SHA could not be resolved, git returned: %r" % (header_line.strip()))
            else:
                raise ValueError("SHA %s could not be resolved, git returned: %r" % (tokens[0], header_line.strip()))
            # END handle actual return value
        # END error handling

        if len(tokens[0]) != 40:
            raise ValueError("Failed to parse header: %r" % header_line)
        return (tokens[0], tokens[1], int(tokens[2]))

    def _prepare_ref(self, ref):
        # required for command to separate refs on stdin, as bytes
        refstr = ref
        if isinstance(ref, bytes):
            # Assume 40 bytes hexsha - bin-to-ascii for some reason returns bytes, not text
            refstr = ref.decode('ascii')
        elif not isinstance(ref, string_types):
            refstr = str(ref)               # could be ref-object

        if not refstr.endswith("\n"):
            refstr += "\n"
        return refstr.encode(defenc)

    def _get_persistent_cmd(self, attr_name, cmd_name, *args, **kwargs):
        cur_val = getattr(self, attr_name)
        if cur_val is not None:
            return cur_val

        options = {"istream": PIPE, "as_process": True}
        options.update(kwargs)

        cmd = self._call_process(cmd_name, *args, **options)
        setattr(self, attr_name, cmd)
        return cmd

    def __get_object_header(self, cmd, ref):
        cmd.stdin.write(self._prepare_ref(ref))
        cmd.stdin.flush()
        return self._parse_object_header(cmd.stdout.readline())

    def get_object_header(self, ref):
        """ Use this method to quickly examine the type and size of the object behind
        the given ref.

        :note: The method will only suffer from the costs of command invocation
            once and reuses the command in subsequent calls.

        :return: (hexsha, type_string, size_as_int)"""
        cmd = self._get_persistent_cmd("cat_file_header", "cat_file", batch_check=True)
        return self.__get_object_header(cmd, ref)

    def get_object_data(self, ref):
        """ As get_object_header, but returns object data as well
        :return: (hexsha, type_string, size_as_int,data_string)
        :note: not threadsafe"""
        hexsha, typename, size, stream = self.stream_object_data(ref)
        data = stream.read(size)
        del(stream)
        return (hexsha, typename, size, data)

    def stream_object_data(self, ref):
        """ As get_object_header, but returns the data as a stream

        :return: (hexsha, type_string, size_as_int, stream)
        :note: This method is not threadsafe, you need one independent Command instance per thread to be safe !"""
        cmd = self._get_persistent_cmd("cat_file_all", "cat_file", batch=True)
        hexsha, typename, size = self.__get_object_header(cmd, ref)
        return (hexsha, typename, size, self.CatFileContentStream(size, cmd.stdout))

    def clear_cache(self):
        """Clear all kinds of internal caches to release resources.

        Currently persistent commands will be interrupted.

        :return: self"""
        self.cat_file_all = None
        self.cat_file_header = None
        return self