~bzr-pqm/bzr/bzr.dev

# Copyright (C) 2005-2010 Canonical Ltd
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

import errno
import os
import re
import stat
import sys
import time
import codecs

from bzrlib.lazy_import import lazy_import
lazy_import(globals(), """
from datetime import datetime
import getpass
import ntpath
import posixpath
# We need to import both shutil and rmtree as we export the later on posix
# and need the former on windows
import shutil
from shutil import rmtree
import socket
import subprocess
# We need to import both tempfile and mkdtemp as we export the later on posix
# and need the former on windows
import tempfile
from tempfile import mkdtemp
import unicodedata

from bzrlib import (
    cache_utf8,
    errors,
    trace,
    win32utils,
    )
""")

from bzrlib.symbol_versioning import (
    deprecated_function,
    deprecated_in,
    )

# sha and md5 modules are deprecated in python2.6 but hashlib is available as
# of 2.5
if sys.version_info < (2, 5):
    import md5 as _mod_md5
    md5 = _mod_md5.new
    import sha as _mod_sha
    sha = _mod_sha.new
else:
    from hashlib import (
        md5,
        sha1 as sha,
        )


import bzrlib
from bzrlib import symbol_versioning


# Cross platform wall-clock time functionality with decent resolution.
# On Linux ``time.clock`` returns only CPU time. On Windows, ``time.time()``
# only has a resolution of ~15ms. Note that ``time.clock()`` is not
# synchronized with ``time.time()``, this is only meant to be used to find
# delta times by subtracting from another call to this function.
timer_func = time.time
if sys.platform == 'win32':
    timer_func = time.clock

# On win32, O_BINARY is used to indicate the file should
# be opened in binary mode, rather than text mode.
# On other platforms, O_BINARY doesn't exist, because
# they always open in binary mode, so it is okay to
# OR with 0 on those platforms.
# O_NOINHERIT and O_TEXT exists only on win32 too.
O_BINARY = getattr(os, 'O_BINARY', 0)
O_TEXT = getattr(os, 'O_TEXT', 0)
O_NOINHERIT = getattr(os, 'O_NOINHERIT', 0)


def get_unicode_argv():
    try:
        user_encoding = get_user_encoding()
        return [a.decode(user_encoding) for a in sys.argv[1:]]
    except UnicodeDecodeError:
        raise errors.BzrError(("Parameter '%r' is unsupported by the current "
                                                            "encoding." % a))


def make_readonly(filename):
    """Make a filename read-only."""
    mod = os.lstat(filename).st_mode
    if not stat.S_ISLNK(mod):
        mod = mod & 0777555
        os.chmod(filename, mod)


def make_writable(filename):
    mod = os.lstat(filename).st_mode
    if not stat.S_ISLNK(mod):
        mod = mod | 0200
        os.chmod(filename, mod)


def minimum_path_selection(paths):
    """Return the smallset subset of paths which are outside paths.

    :param paths: A container (and hence not None) of paths.
    :return: A set of paths sufficient to include everything in paths via
        is_inside, drawn from the paths parameter.
    """
    if len(paths) < 2:
        return set(paths)

    def sort_key(path):
        return path.split('/')
    sorted_paths = sorted(list(paths), key=sort_key)

    search_paths = [sorted_paths[0]]
    for path in sorted_paths[1:]:
        if not is_inside(search_paths[-1], path):
            # This path is unique, add it
            search_paths.append(path)

    return set(search_paths)


_QUOTE_RE = None


def quotefn(f):
    """Return a quoted filename filename

    This previously used backslash quoting, but that works poorly on
    Windows."""
    # TODO: I'm not really sure this is the best format either.x
    global _QUOTE_RE
    if _QUOTE_RE is None:
        _QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/\\_~-])')

    if _QUOTE_RE.search(f):
        return '"' + f + '"'
    else:
        return f


_directory_kind = 'directory'

def get_umask():
    """Return the current umask"""
    # Assume that people aren't messing with the umask while running
    # XXX: This is not thread safe, but there is no way to get the
    #      umask without setting it
    umask = os.umask(0)
    os.umask(umask)
    return umask


_kind_marker_map = {
    "file": "",
    _directory_kind: "/",
    "symlink": "@",
    'tree-reference': '+',
}


def kind_marker(kind):
    try:
        return _kind_marker_map[kind]
    except KeyError:
        # Slightly faster than using .get(, '') when the common case is that
        # kind will be found
        return ''


lexists = getattr(os.path, 'lexists', None)
if lexists is None:
    def lexists(f):
        try:
            stat = getattr(os, 'lstat', os.stat)
            stat(f)
            return True
        except OSError, e:
            if e.errno == errno.ENOENT:
                return False;
            else:
                raise errors.BzrError("lstat/stat of (%r): %r" % (f, e))


def fancy_rename(old, new, rename_func, unlink_func):
    """A fancy rename, when you don't have atomic rename.

    :param old: The old path, to rename from
    :param new: The new path, to rename to
    :param rename_func: The potentially non-atomic rename function
    :param unlink_func: A way to delete the target file if the full rename
        succeeds
    """
    # sftp rename doesn't allow overwriting, so play tricks:
    base = os.path.basename(new)
    dirname = os.path.dirname(new)
    # callers use different encodings for the paths so the following MUST
    # respect that. We rely on python upcasting to unicode if new is unicode
    # and keeping a str if not.
    tmp_name = 'tmp.%s.%.9f.%d.%s' % (base, time.time(),
                                      os.getpid(), rand_chars(10))
    tmp_name = pathjoin(dirname, tmp_name)

    # Rename the file out of the way, but keep track if it didn't exist
    # We don't want to grab just any exception
    # something like EACCES should prevent us from continuing
    # The downside is that the rename_func has to throw an exception
    # with an errno = ENOENT, or NoSuchFile
    file_existed = False
    try:
        rename_func(new, tmp_name)
    except (errors.NoSuchFile,), e:
        pass
    except IOError, e:
        # RBC 20060103 abstraction leakage: the paramiko SFTP clients rename
        # function raises an IOError with errno is None when a rename fails.
        # This then gets caught here.
        if e.errno not in (None, errno.ENOENT, errno.ENOTDIR):
            raise
    except Exception, e:
        if (getattr(e, 'errno', None) is None
            or e.errno not in (errno.ENOENT, errno.ENOTDIR)):
            raise
    else:
        file_existed = True

    failure_exc = None
    success = False
    try:
        try:
            # This may throw an exception, in which case success will
            # not be set.
            rename_func(old, new)
            success = True
        except (IOError, OSError), e:
            # source and target may be aliases of each other (e.g. on a
            # case-insensitive filesystem), so we may have accidentally renamed
            # source by when we tried to rename target
            failure_exc = sys.exc_info()
            if (file_existed and e.errno in (None, errno.ENOENT)
                and old.lower() == new.lower()):
                # source and target are the same file on a case-insensitive
                # filesystem, so we don't generate an exception
                failure_exc = None
    finally:
        if file_existed:
            # If the file used to exist, rename it back into place
            # otherwise just delete it from the tmp location
            if success:
                unlink_func(tmp_name)
            else:
                rename_func(tmp_name, new)
    if failure_exc is not None:
        raise failure_exc[0], failure_exc[1], failure_exc[2]


# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
# choke on a Unicode string containing a relative path if
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
# string.
_fs_enc = sys.getfilesystemencoding() or 'utf-8'
def _posix_abspath(path):
    # jam 20060426 rather than encoding to fsencoding
    # copy posixpath.abspath, but use os.getcwdu instead
    if not posixpath.isabs(path):
        path = posixpath.join(getcwd(), path)
    return posixpath.normpath(path)


def _posix_realpath(path):
    return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)


def _win32_fixdrive(path):
    """Force drive letters to be consistent.

    win32 is inconsistent whether it returns lower or upper case
    and even if it was consistent the user might type the other
    so we force it to uppercase
    running python.exe under cmd.exe return capital C:\\
    running win32 python inside a cygwin shell returns lowercase c:\\
    """
    drive, path = ntpath.splitdrive(path)
    return drive.upper() + path


def _win32_abspath(path):
    # Real ntpath.abspath doesn't have a problem with a unicode cwd
    return _win32_fixdrive(ntpath.abspath(unicode(path)).replace('\\', '/'))


def _win98_abspath(path):
    """Return the absolute version of a path.
    Windows 98 safe implementation (python reimplementation
    of Win32 API function GetFullPathNameW)
    """
    # Corner cases:
    #   C:\path     => C:/path
    #   C:/path     => C:/path
    #   \\HOST\path => //HOST/path
    #   //HOST/path => //HOST/path
    #   path        => C:/cwd/path
    #   /path       => C:/path
    path = unicode(path)
    # check for absolute path
    drive = ntpath.splitdrive(path)[0]
    if drive == '' and path[:2] not in('//','\\\\'):
        cwd = os.getcwdu()
        # we cannot simply os.path.join cwd and path
        # because os.path.join('C:','/path') produce '/path'
        # and this is incorrect
        if path[:1] in ('/','\\'):
            cwd = ntpath.splitdrive(cwd)[0]
            path = path[1:]
        path = cwd + '\\' + path
    return _win32_fixdrive(ntpath.normpath(path).replace('\\', '/'))


def _win32_realpath(path):
    # Real ntpath.realpath doesn't have a problem with a unicode cwd
    return _win32_fixdrive(ntpath.realpath(unicode(path)).replace('\\', '/'))


def _win32_pathjoin(*args):
    return ntpath.join(*args).replace('\\', '/')


def _win32_normpath(path):
    return _win32_fixdrive(ntpath.normpath(unicode(path)).replace('\\', '/'))


def _win32_getcwd():
    return _win32_fixdrive(os.getcwdu().replace('\\', '/'))


def _win32_mkdtemp(*args, **kwargs):
    return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))


def _win32_rename(old, new):
    """We expect to be able to atomically replace 'new' with old.

    On win32, if new exists, it must be moved out of the way first,
    and then deleted.
    """
    try:
        fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
    except OSError, e:
        if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
            # If we try to rename a non-existant file onto cwd, we get
            # EPERM or EACCES instead of ENOENT, this will raise ENOENT
            # if the old path doesn't exist, sometimes we get EACCES
            # On Linux, we seem to get EBUSY, on Mac we get EINVAL
            os.lstat(old)
        raise


def _mac_getcwd():
    return unicodedata.normalize('NFC', os.getcwdu())


# Default is to just use the python builtins, but these can be rebound on
# particular platforms.
abspath = _posix_abspath
realpath = _posix_realpath
pathjoin = os.path.join
normpath = os.path.normpath
getcwd = os.getcwdu
rename = os.rename
dirname = os.path.dirname
basename = os.path.basename
split = os.path.split
splitext = os.path.splitext
# These were already lazily imported into local scope
# mkdtemp = tempfile.mkdtemp
# rmtree = shutil.rmtree

MIN_ABS_PATHLENGTH = 1


if sys.platform == 'win32':
    if win32utils.winver == 'Windows 98':
        abspath = _win98_abspath
    else:
        abspath = _win32_abspath
    realpath = _win32_realpath
    pathjoin = _win32_pathjoin
    normpath = _win32_normpath
    getcwd = _win32_getcwd
    mkdtemp = _win32_mkdtemp
    rename = _win32_rename

    MIN_ABS_PATHLENGTH = 3

    def _win32_delete_readonly(function, path, excinfo):
        """Error handler for shutil.rmtree function [for win32]
        Helps to remove files and dirs marked as read-only.
        """
        exception = excinfo[1]
        if function in (os.remove, os.rmdir) \
            and isinstance(exception, OSError) \
            and exception.errno == errno.EACCES:
            make_writable(path)
            function(path)
        else:
            raise

    def rmtree(path, ignore_errors=False, onerror=_win32_delete_readonly):
        """Replacer for shutil.rmtree: could remove readonly dirs/files"""
        return shutil.rmtree(path, ignore_errors, onerror)

    f = win32utils.get_unicode_argv     # special function or None
    if f is not None:
        get_unicode_argv = f

elif sys.platform == 'darwin':
    getcwd = _mac_getcwd


def get_terminal_encoding(trace=False):
    """Find the best encoding for printing to the screen.

    This attempts to check both sys.stdout and sys.stdin to see
    what encoding they are in, and if that fails it falls back to
    osutils.get_user_encoding().
    The problem is that on Windows, locale.getpreferredencoding()
    is not the same encoding as that used by the console:
    http://mail.python.org/pipermail/python-list/2003-May/162357.html

    On my standard US Windows XP, the preferred encoding is
    cp1252, but the console is cp437

    :param trace: If True trace the selected encoding via mutter().
    """
    from bzrlib.trace import mutter
    output_encoding = getattr(sys.stdout, 'encoding', None)
    if not output_encoding:
        input_encoding = getattr(sys.stdin, 'encoding', None)
        if not input_encoding:
            output_encoding = get_user_encoding()
            if trace:
                mutter('encoding stdout as osutils.get_user_encoding() %r',
                   output_encoding)
        else:
            output_encoding = input_encoding
            if trace:
                mutter('encoding stdout as sys.stdin encoding %r',
                    output_encoding)
    else:
        if trace:
            mutter('encoding stdout as sys.stdout encoding %r', output_encoding)
    if output_encoding == 'cp0':
        # invalid encoding (cp0 means 'no codepage' on Windows)
        output_encoding = get_user_encoding()
        if trace:
            mutter('cp0 is invalid encoding.'
               ' encoding stdout as osutils.get_user_encoding() %r',
               output_encoding)
    # check encoding
    try:
        codecs.lookup(output_encoding)
    except LookupError:
        sys.stderr.write('bzr: warning:'
                         ' unknown terminal encoding %s.\n'
                         '  Using encoding %s instead.\n'
                         % (output_encoding, get_user_encoding())
                        )
        output_encoding = get_user_encoding()

    return output_encoding


def normalizepath(f):
    if getattr(os.path, 'realpath', None) is not None:
        F = realpath
    else:
        F = abspath
    [p,e] = os.path.split(f)
    if e == "" or e == "." or e == "..":
        return F(f)
    else:
        return pathjoin(F(p), e)


def isdir(f):
    """True if f is an accessible directory."""
    try:
        return stat.S_ISDIR(os.lstat(f)[stat.ST_MODE])
    except OSError:
        return False


def isfile(f):
    """True if f is a regular file."""
    try:
        return stat.S_ISREG(os.lstat(f)[stat.ST_MODE])
    except OSError:
        return False

def islink(f):
    """True if f is a symlink."""
    try:
        return stat.S_ISLNK(os.lstat(f)[stat.ST_MODE])
    except OSError:
        return False

def is_inside(dir, fname):
    """True if fname is inside dir.

    The parameters should typically be passed to osutils.normpath first, so
    that . and .. and repeated slashes are eliminated, and the separators
    are canonical for the platform.

    The empty string as a dir name is taken as top-of-tree and matches
    everything.
    """
    # XXX: Most callers of this can actually do something smarter by
    # looking at the inventory
    if dir == fname:
        return True

    if dir == '':
        return True

    if dir[-1] != '/':
        dir += '/'

    return fname.startswith(dir)


def is_inside_any(dir_list, fname):
    """True if fname is inside any of given dirs."""
    for dirname in dir_list:
        if is_inside(dirname, fname):
            return True
    return False


def is_inside_or_parent_of_any(dir_list, fname):
    """True if fname is a child or a parent of any of the given files."""
    for dirname in dir_list:
        if is_inside(dirname, fname) or is_inside(fname, dirname):
            return True
    return False


def pumpfile(from_file, to_file, read_length=-1, buff_size=32768,
             report_activity=None, direction='read'):
    """Copy contents of one file to another.

    The read_length can either be -1 to read to end-of-file (EOF) or
    it can specify the maximum number of bytes to read.

    The buff_size represents the maximum size for each read operation
    performed on from_file.

    :param report_activity: Call this as bytes are read, see
        Transport._report_activity
    :param direction: Will be passed to report_activity

    :return: The number of bytes copied.
    """
    length = 0
    if read_length >= 0:
        # read specified number of bytes

        while read_length > 0:
            num_bytes_to_read = min(read_length, buff_size)

            block = from_file.read(num_bytes_to_read)
            if not block:
                # EOF reached
                break
            if report_activity is not None:
                report_activity(len(block), direction)
            to_file.write(block)

            actual_bytes_read = len(block)
            read_length -= actual_bytes_read
            length += actual_bytes_read
    else:
        # read to EOF
        while True:
            block = from_file.read(buff_size)
            if not block:
                # EOF reached
                break
            if report_activity is not None:
                report_activity(len(block), direction)
            to_file.write(block)
            length += len(block)
    return length


def pump_string_file(bytes, file_handle, segment_size=None):
    """Write bytes to file_handle in many smaller writes.

    :param bytes: The string to write.
    :param file_handle: The file to write to.
    """
    # Write data in chunks rather than all at once, because very large
    # writes fail on some platforms (e.g. Windows with SMB  mounted
    # drives).
    if not segment_size:
        segment_size = 5242880 # 5MB
    segments = range(len(bytes) / segment_size + 1)
    write = file_handle.write
    for segment_index in segments:
        segment = buffer(bytes, segment_index * segment_size, segment_size)
        write(segment)


def file_iterator(input_file, readsize=32768):
    while True:
        b = input_file.read(readsize)
        if len(b) == 0:
            break
        yield b


def sha_file(f):
    """Calculate the hexdigest of an open file.

    The file cursor should be already at the start.
    """
    s = sha()
    BUFSIZE = 128<<10
    while True:
        b = f.read(BUFSIZE)
        if not b:
            break
        s.update(b)
    return s.hexdigest()


def size_sha_file(f):
    """Calculate the size and hexdigest of an open file.

    The file cursor should be already at the start and
    the caller is responsible for closing the file afterwards.
    """
    size = 0
    s = sha()
    BUFSIZE = 128<<10
    while True:
        b = f.read(BUFSIZE)
        if not b:
            break
        size += len(b)
        s.update(b)
    return size, s.hexdigest()


def sha_file_by_name(fname):
    """Calculate the SHA1 of a file by reading the full text"""
    s = sha()
    f = os.open(fname, os.O_RDONLY | O_BINARY | O_NOINHERIT)
    try:
        while True:
            b = os.read(f, 1<<16)
            if not b:
                return s.hexdigest()
            s.update(b)
    finally:
        os.close(f)


def sha_strings(strings, _factory=sha):
    """Return the sha-1 of concatenation of strings"""
    s = _factory()
    map(s.update, strings)
    return s.hexdigest()


def sha_string(f, _factory=sha):
    return _factory(f).hexdigest()


def fingerprint_file(f):
    b = f.read()
    return {'size': len(b),
            'sha1': sha(b).hexdigest()}


def compare_files(a, b):
    """Returns true if equal in contents"""
    BUFSIZE = 4096
    while True:
        ai = a.read(BUFSIZE)
        bi = b.read(BUFSIZE)
        if ai != bi:
            return False
        if ai == '':
            return True


def local_time_offset(t=None):
    """Return offset of local zone from GMT, either at present or at time t."""
    if t is None:
        t = time.time()
    offset = datetime.fromtimestamp(t) - datetime.utcfromtimestamp(t)
    return offset.days * 86400 + offset.seconds

weekdays = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun']
_default_format_by_weekday_num = [wd + " %Y-%m-%d %H:%M:%S" for wd in weekdays]


def format_date(t, offset=0, timezone='original', date_fmt=None,
                show_offset=True):
    """Return a formatted date string.

    :param t: Seconds since the epoch.
    :param offset: Timezone offset in seconds east of utc.
    :param timezone: How to display the time: 'utc', 'original' for the
         timezone specified by offset, or 'local' for the process's current
         timezone.
    :param date_fmt: strftime format.
    :param show_offset: Whether to append the timezone.
    """
    (date_fmt, tt, offset_str) = \
               _format_date(t, offset, timezone, date_fmt, show_offset)
    date_fmt = date_fmt.replace('%a', weekdays[tt[6]])
    date_str = time.strftime(date_fmt, tt)
    return date_str + offset_str


# Cache of formatted offset strings
_offset_cache = {}


def format_date_with_offset_in_original_timezone(t, offset=0,
    _cache=_offset_cache):
    """Return a formatted date string in the original timezone.

    This routine may be faster then format_date.

    :param t: Seconds since the epoch.
    :param offset: Timezone offset in seconds east of utc.
    """
    if offset is None:
        offset = 0
    tt = time.gmtime(t + offset)
    date_fmt = _default_format_by_weekday_num[tt[6]]
    date_str = time.strftime(date_fmt, tt)
    offset_str = _cache.get(offset, None)
    if offset_str is None:
        offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
        _cache[offset] = offset_str
    return date_str + offset_str


def format_local_date(t, offset=0, timezone='original', date_fmt=None,
                      show_offset=True):
    """Return an unicode date string formatted according to the current locale.

    :param t: Seconds since the epoch.
    :param offset: Timezone offset in seconds east of utc.
    :param timezone: How to display the time: 'utc', 'original' for the
         timezone specified by offset, or 'local' for the process's current
         timezone.
    :param date_fmt: strftime format.
    :param show_offset: Whether to append the timezone.
    """
    (date_fmt, tt, offset_str) = \
               _format_date(t, offset, timezone, date_fmt, show_offset)
    date_str = time.strftime(date_fmt, tt)
    if not isinstance(date_str, unicode):
        date_str = date_str.decode(get_user_encoding(), 'replace')
    return date_str + offset_str


def _format_date(t, offset, timezone, date_fmt, show_offset):
    if timezone == 'utc':
        tt = time.gmtime(t)
        offset = 0
    elif timezone == 'original':
        if offset is None:
            offset = 0
        tt = time.gmtime(t + offset)
    elif timezone == 'local':
        tt = time.localtime(t)
        offset = local_time_offset(t)
    else:
        raise errors.UnsupportedTimezoneFormat(timezone)
    if date_fmt is None:
        date_fmt = "%a %Y-%m-%d %H:%M:%S"
    if show_offset:
        offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
    else:
        offset_str = ''
    return (date_fmt, tt, offset_str)


def compact_date(when):
    return time.strftime('%Y%m%d%H%M%S', time.gmtime(when))


def format_delta(delta):
    """Get a nice looking string for a time delta.

    :param delta: The time difference in seconds, can be positive or negative.
        positive indicates time in the past, negative indicates time in the
        future. (usually time.time() - stored_time)
    :return: String formatted to show approximate resolution
    """
    delta = int(delta)
    if delta >= 0:
        direction = 'ago'
    else:
        direction = 'in the future'
        delta = -delta

    seconds = delta
    if seconds < 90: # print seconds up to 90 seconds
        if seconds == 1:
            return '%d second %s' % (seconds, direction,)
        else:
            return '%d seconds %s' % (seconds, direction)

    minutes = int(seconds / 60)
    seconds -= 60 * minutes
    if seconds == 1:
        plural_seconds = ''
    else:
        plural_seconds = 's'
    if minutes < 90: # print minutes, seconds up to 90 minutes
        if minutes == 1:
            return '%d minute, %d second%s %s' % (
                    minutes, seconds, plural_seconds, direction)
        else:
            return '%d minutes, %d second%s %s' % (
                    minutes, seconds, plural_seconds, direction)

    hours = int(minutes / 60)
    minutes -= 60 * hours
    if minutes == 1:
        plural_minutes = ''
    else:
        plural_minutes = 's'

    if hours == 1:
        return '%d hour, %d minute%s %s' % (hours, minutes,
                                            plural_minutes, direction)
    return '%d hours, %d minute%s %s' % (hours, minutes,
                                         plural_minutes, direction)

def filesize(f):
    """Return size of given open file."""
    return os.fstat(f.fileno())[stat.ST_SIZE]


# Define rand_bytes based on platform.
try:
    # Python 2.4 and later have os.urandom,
    # but it doesn't work on some arches
    os.urandom(1)
    rand_bytes = os.urandom
except (NotImplementedError, AttributeError):
    # If python doesn't have os.urandom, or it doesn't work,
    # then try to first pull random data from /dev/urandom
    try:
        rand_bytes = file('/dev/urandom', 'rb').read
    # Otherwise, use this hack as a last resort
    except (IOError, OSError):
        # not well seeded, but better than nothing
        def rand_bytes(n):
            import random
            s = ''
            while n:
                s += chr(random.randint(0, 255))
                n -= 1
            return s


ALNUM = '0123456789abcdefghijklmnopqrstuvwxyz'
def rand_chars(num):
    """Return a random string of num alphanumeric characters

    The result only contains lowercase chars because it may be used on
    case-insensitive filesystems.
    """
    s = ''
    for raw_byte in rand_bytes(num):
        s += ALNUM[ord(raw_byte) % 36]
    return s


## TODO: We could later have path objects that remember their list
## decomposition (might be too tricksy though.)

def splitpath(p):
    """Turn string into list of parts."""
    # split on either delimiter because people might use either on
    # Windows
    ps = re.split(r'[\\/]', p)

    rps = []
    for f in ps:
        if f == '..':
            raise errors.BzrError("sorry, %r not allowed in path" % f)
        elif (f == '.') or (f == ''):
            pass
        else:
            rps.append(f)
    return rps


def joinpath(p):
    for f in p:
        if (f == '..') or (f is None) or (f == ''):
            raise errors.BzrError("sorry, %r not allowed in path" % f)
    return pathjoin(*p)


def parent_directories(filename):
    """Return the list of parent directories, deepest first.

    For example, parent_directories("a/b/c") -> ["a/b", "a"].
    """
    parents = []
    parts = splitpath(dirname(filename))
    while parts:
        parents.append(joinpath(parts))
        parts.pop()
    return parents


_extension_load_failures = []


def failed_to_load_extension(exception):
    """Handle failing to load a binary extension.

    This should be called from the ImportError block guarding the attempt to
    import the native extension.  If this function returns, the pure-Python
    implementation should be loaded instead::

    >>> try:
    >>>     import bzrlib._fictional_extension_pyx
    >>> except ImportError, e:
    >>>     bzrlib.osutils.failed_to_load_extension(e)
    >>>     import bzrlib._fictional_extension_py
    """
    # NB: This docstring is just an example, not a doctest, because doctest
    # currently can't cope with the use of lazy imports in this namespace --
    # mbp 20090729

    # This currently doesn't report the failure at the time it occurs, because
    # they tend to happen very early in startup when we can't check config
    # files etc, and also we want to report all failures but not spam the user
    # with 10 warnings.
    from bzrlib import trace
    exception_str = str(exception)
    if exception_str not in _extension_load_failures:
        trace.mutter("failed to load compiled extension: %s" % exception_str)
        _extension_load_failures.append(exception_str)


def report_extension_load_failures():
    if not _extension_load_failures:
        return
    from bzrlib.config import GlobalConfig
    if GlobalConfig().get_user_option_as_bool('ignore_missing_extensions'):
        return
    # the warnings framework should by default show this only once
    from bzrlib.trace import warning
    warning(
        "bzr: warning: some compiled extensions could not be loaded; "
        "see <https://answers.launchpad.net/bzr/+faq/703>")
    # we no longer show the specific missing extensions here, because it makes
    # the message too long and scary - see
    # https://bugs.launchpad.net/bzr/+bug/430529


try:
    from bzrlib._chunks_to_lines_pyx import chunks_to_lines
except ImportError, e:
    failed_to_load_extension(e)
    from bzrlib._chunks_to_lines_py import chunks_to_lines


def split_lines(s):
    """Split s into lines, but without removing the newline characters."""
    # Trivially convert a fulltext into a 'chunked' representation, and let
    # chunks_to_lines do the heavy lifting.
    if isinstance(s, str):
        # chunks_to_lines only supports 8-bit strings
        return chunks_to_lines([s])
    else:
        return _split_lines(s)


def _split_lines(s):
    """Split s into lines, but without removing the newline characters.

    This supports Unicode or plain string objects.
    """
    lines = s.split('\n')
    result = [line + '\n' for line in lines[:-1]]
    if lines[-1]:
        result.append(lines[-1])
    return result


def hardlinks_good():
    return sys.platform not in ('win32', 'cygwin', 'darwin')


def link_or_copy(src, dest):
    """Hardlink a file, or copy it if it can't be hardlinked."""
    if not hardlinks_good():
        shutil.copyfile(src, dest)
        return
    try:
        os.link(src, dest)
    except (OSError, IOError), e:
        if e.errno != errno.EXDEV:
            raise
        shutil.copyfile(src, dest)


def delete_any(path):
    """Delete a file, symlink or directory.

    Will delete even if readonly.
    """
    try:
       _delete_file_or_dir(path)
    except (OSError, IOError), e:
        if e.errno in (errno.EPERM, errno.EACCES):
            # make writable and try again
            try:
                make_writable(path)
            except (OSError, IOError):
                pass
            _delete_file_or_dir(path)
        else:
            raise


def _delete_file_or_dir(path):
    # Look Before You Leap (LBYL) is appropriate here instead of Easier to Ask for
    # Forgiveness than Permission (EAFP) because:
    # - root can damage a solaris file system by using unlink,
    # - unlink raises different exceptions on different OSes (linux: EISDIR, win32:
    #   EACCES, OSX: EPERM) when invoked on a directory.
    if isdir(path): # Takes care of symlinks
        os.rmdir(path)
    else:
        os.unlink(path)


def has_symlinks():
    if getattr(os, 'symlink', None) is not None:
        return True
    else:
        return False


def has_hardlinks():
    if getattr(os, 'link', None) is not None:
        return True
    else:
        return False


def host_os_dereferences_symlinks():
    return (has_symlinks()
            and sys.platform not in ('cygwin', 'win32'))


def readlink(abspath):
    """Return a string representing the path to which the symbolic link points.

    :param abspath: The link absolute unicode path.

    This his guaranteed to return the symbolic link in unicode in all python
    versions.
    """
    link = abspath.encode(_fs_enc)
    target = os.readlink(link)
    target = target.decode(_fs_enc)
    return target


def contains_whitespace(s):
    """True if there are any whitespace characters in s."""
    # string.whitespace can include '\xa0' in certain locales, because it is
    # considered "non-breaking-space" as part of ISO-8859-1. But it
    # 1) Isn't a breaking whitespace
    # 2) Isn't one of ' \t\r\n' which are characters we sometimes use as
    #    separators
    # 3) '\xa0' isn't unicode safe since it is >128.

    # This should *not* be a unicode set of characters in case the source
    # string is not a Unicode string. We can auto-up-cast the characters since
    # they are ascii, but we don't want to auto-up-cast the string in case it
    # is utf-8
    for ch in ' \t\n\r\v\f':
        if ch in s:
            return True
    else:
        return False


def contains_linebreaks(s):
    """True if there is any vertical whitespace in s."""
    for ch in '\f\n\r':
        if ch in s:
            return True
    else:
        return False


def relpath(base, path):
    """Return path relative to base, or raise PathNotChild exception.

    The path may be either an absolute path or a path relative to the
    current working directory.

    os.path.commonprefix (python2.4) has a bad bug that it works just
    on string prefixes, assuming that '/u' is a prefix of '/u2'.  This
    avoids that problem.

    NOTE: `base` should not have a trailing slash otherwise you'll get
    PathNotChild exceptions regardless of `path`.
    """

    if len(base) < MIN_ABS_PATHLENGTH:
        # must have space for e.g. a drive letter
        raise ValueError('%r is too short to calculate a relative path'
            % (base,))

    rp = abspath(path)

    s = []
    head = rp
    while True:
        if len(head) <= len(base) and head != base:
            raise errors.PathNotChild(rp, base)
        if head == base:
            break
        head, tail = split(head)
        if tail:
            s.append(tail)

    if s:
        return pathjoin(*reversed(s))
    else:
        return ''


def _cicp_canonical_relpath(base, path):
    """Return the canonical path relative to base.

    Like relpath, but on case-insensitive-case-preserving file-systems, this
    will return the relpath as stored on the file-system rather than in the
    case specified in the input string, for all existing portions of the path.

    This will cause O(N) behaviour if called for every path in a tree; if you
    have a number of paths to convert, you should use canonical_relpaths().
    """
    # TODO: it should be possible to optimize this for Windows by using the
    # win32 API FindFiles function to look for the specified name - but using
    # os.listdir() still gives us the correct, platform agnostic semantics in
    # the short term.

    rel = relpath(base, path)
    # '.' will have been turned into ''
    if not rel:
        return rel

    abs_base = abspath(base)
    current = abs_base
    _listdir = os.listdir

    # use an explicit iterator so we can easily consume the rest on early exit.
    bit_iter = iter(rel.split('/'))
    for bit in bit_iter:
        lbit = bit.lower()
        try:
            next_entries = _listdir(current)
        except OSError: # enoent, eperm, etc
            # We can't find this in the filesystem, so just append the
            # remaining bits.
            current = pathjoin(current, bit, *list(bit_iter))
            break
        for look in next_entries:
            if lbit == look.lower():
                current = pathjoin(current, look)
                break
        else:
            # got to the end, nothing matched, so we just return the
            # non-existing bits as they were specified (the filename may be
            # the target of a move, for example).
            current = pathjoin(current, bit, *list(bit_iter))
            break
    return current[len(abs_base):].lstrip('/')

# XXX - TODO - we need better detection/integration of case-insensitive
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
# filesystems), for example, so could probably benefit from the same basic
# support there.  For now though, only Windows and OSX get that support, and
# they get it for *all* file-systems!
if sys.platform in ('win32', 'darwin'):
    canonical_relpath = _cicp_canonical_relpath
else:
    canonical_relpath = relpath

def canonical_relpaths(base, paths):
    """Create an iterable to canonicalize a sequence of relative paths.

    The intent is for this implementation to use a cache, vastly speeding
    up multiple transformations in the same directory.
    """
    # but for now, we haven't optimized...
    return [canonical_relpath(base, p) for p in paths]


def decode_filename(filename):
    """Decode the filename using the filesystem encoding

    If it is unicode, it is returned.
    Otherwise it is decoded from the the filesystem's encoding. If decoding
    fails, a errors.BadFilenameEncoding exception is raised.
    """
    if type(filename) is unicode:
        return filename
    try:
        return filename.decode(_fs_enc)
    except UnicodeDecodeError:
        raise errors.BadFilenameEncoding(filename, _fs_enc)


def safe_unicode(unicode_or_utf8_string):
    """Coerce unicode_or_utf8_string into unicode.

    If it is unicode, it is returned.
    Otherwise it is decoded from utf-8. If decoding fails, the exception is
    wrapped in a BzrBadParameterNotUnicode exception.
    """
    if isinstance(unicode_or_utf8_string, unicode):
        return unicode_or_utf8_string
    try:
        return unicode_or_utf8_string.decode('utf8')
    except UnicodeDecodeError:
        raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)


def safe_utf8(unicode_or_utf8_string):
    """Coerce unicode_or_utf8_string to a utf8 string.

    If it is a str, it is returned.
    If it is Unicode, it is encoded into a utf-8 string.
    """
    if isinstance(unicode_or_utf8_string, str):
        # TODO: jam 20070209 This is overkill, and probably has an impact on
        #       performance if we are dealing with lots of apis that want a
        #       utf-8 revision id
        try:
            # Make sure it is a valid utf-8 string
            unicode_or_utf8_string.decode('utf-8')
        except UnicodeDecodeError:
            raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
        return unicode_or_utf8_string
    return unicode_or_utf8_string.encode('utf-8')


_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
                        ' Revision id generators should be creating utf8'
                        ' revision ids.')


def safe_revision_id(unicode_or_utf8_string, warn=True):
    """Revision ids should now be utf8, but at one point they were unicode.

    :param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
        utf8 or None).
    :param warn: Functions that are sanitizing user data can set warn=False
    :return: None or a utf8 revision id.
    """
    if (unicode_or_utf8_string is None
        or unicode_or_utf8_string.__class__ == str):
        return unicode_or_utf8_string
    if warn:
        symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
                               stacklevel=2)
    return cache_utf8.encode(unicode_or_utf8_string)


_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
                    ' generators should be creating utf8 file ids.')


def safe_file_id(unicode_or_utf8_string, warn=True):
    """File ids should now be utf8, but at one point they were unicode.

    This is the same as safe_utf8, except it uses the cached encode functions
    to save a little bit of performance.

    :param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
        utf8 or None).
    :param warn: Functions that are sanitizing user data can set warn=False
    :return: None or a utf8 file id.
    """
    if (unicode_or_utf8_string is None
        or unicode_or_utf8_string.__class__ == str):
        return unicode_or_utf8_string
    if warn:
        symbol_versioning.warn(_file_id_warning, DeprecationWarning,
                               stacklevel=2)
    return cache_utf8.encode(unicode_or_utf8_string)


_platform_normalizes_filenames = False
if sys.platform == 'darwin':
    _platform_normalizes_filenames = True


def normalizes_filenames():
    """Return True if this platform normalizes unicode filenames.

    Only Mac OSX.
    """
    return _platform_normalizes_filenames


def _accessible_normalized_filename(path):
    """Get the unicode normalized path, and if you can access the file.

    On platforms where the system normalizes filenames (Mac OSX),
    you can access a file by any path which will normalize correctly.
    On platforms where the system does not normalize filenames
    (everything else), you have to access a file by its exact path.

    Internally, bzr only supports NFC normalization, since that is
    the standard for XML documents.

    So return the normalized path, and a flag indicating if the file
    can be accessed by that path.
    """

    return unicodedata.normalize('NFC', unicode(path)), True


def _inaccessible_normalized_filename(path):
    __doc__ = _accessible_normalized_filename.__doc__

    normalized = unicodedata.normalize('NFC', unicode(path))
    return normalized, normalized == path


if _platform_normalizes_filenames:
    normalized_filename = _accessible_normalized_filename
else:
    normalized_filename = _inaccessible_normalized_filename


def set_signal_handler(signum, handler, restart_syscall=True):
    """A wrapper for signal.signal that also calls siginterrupt(signum, False)
    on platforms that support that.

    :param restart_syscall: if set, allow syscalls interrupted by a signal to
        automatically restart (by calling `signal.siginterrupt(signum,
        False)`).  May be ignored if the feature is not available on this
        platform or Python version.
    """
    try:
        import signal
        siginterrupt = signal.siginterrupt
    except ImportError:
        # This python implementation doesn't provide signal support, hence no
        # handler exists
        return None
    except AttributeError:
        # siginterrupt doesn't exist on this platform, or for this version
        # of Python.
        siginterrupt = lambda signum, flag: None
    if restart_syscall:
        def sig_handler(*args):
            # Python resets the siginterrupt flag when a signal is
            # received.  <http://bugs.python.org/issue8354>
            # As a workaround for some cases, set it back the way we want it.
            siginterrupt(signum, False)
            # Now run the handler function passed to set_signal_handler.
            handler(*args)
    else:
        sig_handler = handler
    old_handler = signal.signal(signum, sig_handler)
    if restart_syscall:
        siginterrupt(signum, False)
    return old_handler


default_terminal_width = 80
"""The default terminal width for ttys.

This is defined so that higher levels can share a common fallback value when
terminal_width() returns None.
"""

# Keep some state so that terminal_width can detect if _terminal_size has
# returned a different size since the process started.  See docstring and
# comments of terminal_width for details.
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
_terminal_size_state = 'no_data'
_first_terminal_size = None

def terminal_width():
    """Return terminal width.

    None is returned if the width can't established precisely.

    The rules are:
    - if BZR_COLUMNS is set, returns its value
    - if there is no controlling terminal, returns None
    - query the OS, if the queried size has changed since the last query,
      return its value,
    - if COLUMNS is set, returns its value,
    - if the OS has a value (even though it's never changed), return its value.

    From there, we need to query the OS to get the size of the controlling
    terminal.

    On Unices we query the OS by:
    - get termios.TIOCGWINSZ
    - if an error occurs or a negative value is obtained, returns None

    On Windows we query the OS by:
    - win32utils.get_console_size() decides,
    - returns None on error (provided default value)
    """
    # Note to implementors: if changing the rules for determining the width,
    # make sure you've considered the behaviour in these cases:
    #  - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
    #  - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
    #    0,0.
    #  - (add more interesting cases here, if you find any)
    # Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
    # but we don't want to register a signal handler because it is impossible
    # to do so without risking EINTR errors in Python <= 2.6.5 (see
    # <http://bugs.python.org/issue8354>).  Instead we check TIOCGWINSZ every
    # time so we can notice if the reported size has changed, which should have
    # a similar effect.

    # If BZR_COLUMNS is set, take it, user is always right
    try:
        return int(os.environ['BZR_COLUMNS'])
    except (KeyError, ValueError):
        pass

    isatty = getattr(sys.stdout, 'isatty', None)
    if isatty is None or not isatty():
        # Don't guess, setting BZR_COLUMNS is the recommended way to override.
        return None

    # Query the OS
    width, height = os_size = _terminal_size(None, None)
    global _first_terminal_size, _terminal_size_state
    if _terminal_size_state == 'no_data':
        _first_terminal_size = os_size
        _terminal_size_state = 'unchanged'
    elif (_terminal_size_state == 'unchanged' and
          _first_terminal_size != os_size):
        _terminal_size_state = 'changed'

    # If the OS claims to know how wide the terminal is, and this value has
    # ever changed, use that.
    if _terminal_size_state == 'changed':
        if width is not None and width > 0:
            return width

    # If COLUMNS is set, use it.
    try:
        return int(os.environ['COLUMNS'])
    except (KeyError, ValueError):
        pass

    # Finally, use an unchanged size from the OS, if we have one.
    if _terminal_size_state == 'unchanged':
        if width is not None and width > 0:
            return width

    # The width could not be determined.
    return None


def _win32_terminal_size(width, height):
    width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
    return width, height


def _ioctl_terminal_size(width, height):
    try:
        import struct, fcntl, termios
        s = struct.pack('HHHH', 0, 0, 0, 0)
        x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
        height, width = struct.unpack('HHHH', x)[0:2]
    except (IOError, AttributeError):
        pass
    return width, height

_terminal_size = None
"""Returns the terminal size as (width, height).

:param width: Default value for width.
:param height: Default value for height.

This is defined specifically for each OS and query the size of the controlling
terminal. If any error occurs, the provided default values should be returned.
"""
if sys.platform == 'win32':
    _terminal_size = _win32_terminal_size
else:
    _terminal_size = _ioctl_terminal_size


def supports_executable():
    return sys.platform != "win32"


def supports_posix_readonly():
    """Return True if 'readonly' has POSIX semantics, False otherwise.

    Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
    directory controls creation/deletion, etc.

    And under win32, readonly means that the directory itself cannot be
    deleted.  The contents of a readonly directory can be changed, unlike POSIX
    where files in readonly directories cannot be added, deleted or renamed.
    """
    return sys.platform != "win32"


def set_or_unset_env(env_variable, value):
    """Modify the environment, setting or removing the env_variable.

    :param env_variable: The environment variable in question
    :param value: The value to set the environment to. If None, then
        the variable will be removed.
    :return: The original value of the environment variable.
    """
    orig_val = os.environ.get(env_variable)
    if value is None:
        if orig_val is not None:
            del os.environ[env_variable]
    else:
        if isinstance(value, unicode):
            value = value.encode(get_user_encoding())
        os.environ[env_variable] = value
    return orig_val


_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')


def check_legal_path(path):
    """Check whether the supplied path is legal.
    This is only required on Windows, so we don't test on other platforms
    right now.
    """
    if sys.platform != "win32":
        return
    if _validWin32PathRE.match(path) is None:
        raise errors.IllegalPath(path)


_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR

def _is_error_enotdir(e):
    """Check if this exception represents ENOTDIR.

    Unfortunately, python is very inconsistent about the exception
    here. The cases are:
      1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
      2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
         which is the windows error code.
      3) Windows, Python2.5 uses errno == EINVAL and
         winerror == ERROR_DIRECTORY

    :param e: An Exception object (expected to be OSError with an errno
        attribute, but we should be able to cope with anything)
    :return: True if this represents an ENOTDIR error. False otherwise.
    """
    en = getattr(e, 'errno', None)
    if (en == errno.ENOTDIR
        or (sys.platform == 'win32'
            and (en == _WIN32_ERROR_DIRECTORY
                 or (en == errno.EINVAL
                     and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
        ))):
        return True
    return False


def walkdirs(top, prefix=""):
    """Yield data about all the directories in a tree.

    This yields all the data about the contents of a directory at a time.
    After each directory has been yielded, if the caller has mutated the list
    to exclude some directories, they are then not descended into.

    The data yielded is of the form:
    ((directory-relpath, directory-path-from-top),
    [(relpath, basename, kind, lstat, path-from-top), ...]),
     - directory-relpath is the relative path of the directory being returned
       with respect to top. prefix is prepended to this.
     - directory-path-from-root is the path including top for this directory.
       It is suitable for use with os functions.
     - relpath is the relative path within the subtree being walked.
     - basename is the basename of the path
     - kind is the kind of the file now. If unknown then the file is not
       present within the tree - but it may be recorded as versioned. See
       versioned_kind.
     - lstat is the stat data *if* the file was statted.
     - planned, not implemented:
       path_from_tree_root is the path from the root of the tree.

    :param prefix: Prefix the relpaths that are yielded with 'prefix'. This
        allows one to walk a subtree but get paths that are relative to a tree
        rooted higher up.
    :return: an iterator over the dirs.
    """
    #TODO there is a bit of a smell where the results of the directory-
    # summary in this, and the path from the root, may not agree
    # depending on top and prefix - i.e. ./foo and foo as a pair leads to
    # potentially confusing output. We should make this more robust - but
    # not at a speed cost. RBC 20060731
    _lstat = os.lstat
    _directory = _directory_kind
    _listdir = os.listdir
    _kind_from_mode = file_kind_from_stat_mode
    pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
    while pending:
        # 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
        relroot, _, _, _, top = pending.pop()
        if relroot:
            relprefix = relroot + u'/'
        else:
            relprefix = ''
        top_slash = top + u'/'

        dirblock = []
        append = dirblock.append
        try:
            names = sorted(map(decode_filename, _listdir(top)))
        except OSError, e:
            if not _is_error_enotdir(e):
                raise
        else:
            for name in names:
                abspath = top_slash + name
                statvalue = _lstat(abspath)
                kind = _kind_from_mode(statvalue.st_mode)
                append((relprefix + name, name, kind, statvalue, abspath))
        yield (relroot, top), dirblock

        # push the user specified dirs from dirblock
        pending.extend(d for d in reversed(dirblock) if d[2] == _directory)


class DirReader(object):
    """An interface for reading directories."""

    def top_prefix_to_starting_dir(self, top, prefix=""):
        """Converts top and prefix to a starting dir entry

        :param top: A utf8 path
        :param prefix: An optional utf8 path to prefix output relative paths
            with.
        :return: A tuple starting with prefix, and ending with the native
            encoding of top.
        """
        raise NotImplementedError(self.top_prefix_to_starting_dir)

    def read_dir(self, prefix, top):
        """Read a specific dir.

        :param prefix: A utf8 prefix to be preprended to the path basenames.
        :param top: A natively encoded path to read.
        :return: A list of the directories contents. Each item contains:
            (utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
        """
        raise NotImplementedError(self.read_dir)


_selected_dir_reader = None


def _walkdirs_utf8(top, prefix=""):
    """Yield data about all the directories in a tree.

    This yields the same information as walkdirs() only each entry is yielded
    in utf-8. On platforms which have a filesystem encoding of utf8 the paths
    are returned as exact byte-strings.

    :return: yields a tuple of (dir_info, [file_info])
        dir_info is (utf8_relpath, path-from-top)
        file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
        if top is an absolute path, path-from-top is also an absolute path.
        path-from-top might be unicode or utf8, but it is the correct path to
        pass to os functions to affect the file in question. (such as os.lstat)
    """
    global _selected_dir_reader
    if _selected_dir_reader is None:
        fs_encoding = _fs_enc.upper()
        if sys.platform == "win32" and win32utils.winver == 'Windows NT':
            # Win98 doesn't have unicode apis like FindFirstFileW
            # TODO: We possibly could support Win98 by falling back to the
            #       original FindFirstFile, and using TCHAR instead of WCHAR,
            #       but that gets a bit tricky, and requires custom compiling
            #       for win98 anyway.
            try:
                from bzrlib._walkdirs_win32 import Win32ReadDir
                _selected_dir_reader = Win32ReadDir()
            except ImportError:
                pass
        elif fs_encoding in ('UTF-8', 'US-ASCII', 'ANSI_X3.4-1968'):
            # ANSI_X3.4-1968 is a form of ASCII
            try:
                from bzrlib._readdir_pyx import UTF8DirReader
                _selected_dir_reader = UTF8DirReader()
            except ImportError, e:
                failed_to_load_extension(e)
                pass

    if _selected_dir_reader is None:
        # Fallback to the python version
        _selected_dir_reader = UnicodeDirReader()

    # 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
    # But we don't actually uses 1-3 in pending, so set them to None
    pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
    read_dir = _selected_dir_reader.read_dir
    _directory = _directory_kind
    while pending:
        relroot, _, _, _, top = pending[-1].pop()
        if not pending[-1]:
            pending.pop()
        dirblock = sorted(read_dir(relroot, top))
        yield (relroot, top), dirblock
        # push the user specified dirs from dirblock
        next = [d for d in reversed(dirblock) if d[2] == _directory]
        if next:
            pending.append(next)


class UnicodeDirReader(DirReader):
    """A dir reader for non-utf8 file systems, which transcodes."""

    __slots__ = ['_utf8_encode']

    def __init__(self):
        self._utf8_encode = codecs.getencoder('utf8')

    def top_prefix_to_starting_dir(self, top, prefix=""):
        """See DirReader.top_prefix_to_starting_dir."""
        return (safe_utf8(prefix), None, None, None, safe_unicode(top))

    def read_dir(self, prefix, top):
        """Read a single directory from a non-utf8 file system.

        top, and the abspath element in the output are unicode, all other paths
        are utf8. Local disk IO is done via unicode calls to listdir etc.

        This is currently the fallback code path when the filesystem encoding is
        not UTF-8. It may be better to implement an alternative so that we can
        safely handle paths that are not properly decodable in the current
        encoding.

        See DirReader.read_dir for details.
        """
        _utf8_encode = self._utf8_encode
        _lstat = os.lstat
        _listdir = os.listdir
        _kind_from_mode = file_kind_from_stat_mode

        if prefix:
            relprefix = prefix + '/'
        else:
            relprefix = ''
        top_slash = top + u'/'

        dirblock = []
        append = dirblock.append
        for name in sorted(_listdir(top)):
            try:
                name_utf8 = _utf8_encode(name)[0]
            except UnicodeDecodeError:
                raise errors.BadFilenameEncoding(
                    _utf8_encode(relprefix)[0] + name, _fs_enc)
            abspath = top_slash + name
            statvalue = _lstat(abspath)
            kind = _kind_from_mode(statvalue.st_mode)
            append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
        return dirblock


def copy_tree(from_path, to_path, handlers={}):
    """Copy all of the entries in from_path into to_path.

    :param from_path: The base directory to copy.
    :param to_path: The target directory. If it does not exist, it will
        be created.
    :param handlers: A dictionary of functions, which takes a source and
        destinations for files, directories, etc.
        It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
        'file', 'directory', and 'symlink' should always exist.
        If they are missing, they will be replaced with 'os.mkdir()',
        'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
    """
    # Now, just copy the existing cached tree to the new location
    # We use a cheap trick here.
    # Absolute paths are prefixed with the first parameter
    # relative paths are prefixed with the second.
    # So we can get both the source and target returned
    # without any extra work.

    def copy_dir(source, dest):
        os.mkdir(dest)

    def copy_link(source, dest):
        """Copy the contents of a symlink"""
        link_to = os.readlink(source)
        os.symlink(link_to, dest)

    real_handlers = {'file':shutil.copy2,
                     'symlink':copy_link,
                     'directory':copy_dir,
                    }
    real_handlers.update(handlers)

    if not os.path.exists(to_path):
        real_handlers['directory'](from_path, to_path)

    for dir_info, entries in walkdirs(from_path, prefix=to_path):
        for relpath, name, kind, st, abspath in entries:
            real_handlers[kind](abspath, relpath)


def copy_ownership_from_path(dst, src=None):
    """Copy usr/grp ownership from src file/dir to dst file/dir.

    If src is None, the containing directory is used as source. If chown
    fails, the error is ignored and a warning is printed.
    """
    chown = getattr(os, 'chown', None)
    if chown is None:
        return

    if src == None:
        src = os.path.dirname(dst)
        if src == '':
            src = '.'

    try:
        s = os.stat(src)
        chown(dst, s.st_uid, s.st_gid)
    except OSError, e:
        trace.warning("Unable to copy ownership from '%s' to '%s': IOError: %s." % (src, dst, e))


def path_prefix_key(path):
    """Generate a prefix-order path key for path.

    This can be used to sort paths in the same way that walkdirs does.
    """
    return (dirname(path) , path)


def compare_paths_prefix_order(path_a, path_b):
    """Compare path_a and path_b to generate the same order walkdirs uses."""
    key_a = path_prefix_key(path_a)
    key_b = path_prefix_key(path_b)
    return cmp(key_a, key_b)


_cached_user_encoding = None


def get_user_encoding(use_cache=True):
    """Find out what the preferred user encoding is.

    This is generally the encoding that is used for command line parameters
    and file contents. This may be different from the terminal encoding
    or the filesystem encoding.

    :param  use_cache:  Enable cache for detected encoding.
                        (This parameter is turned on by default,
                        and required only for selftesting)

    :return: A string defining the preferred user encoding
    """
    global _cached_user_encoding
    if _cached_user_encoding is not None and use_cache:
        return _cached_user_encoding

    if sys.platform == 'darwin':
        # python locale.getpreferredencoding() always return
        # 'mac-roman' on darwin. That's a lie.
        sys.platform = 'posix'
        try:
            if os.environ.get('LANG', None) is None:
                # If LANG is not set, we end up with 'ascii', which is bad
                # ('mac-roman' is more than ascii), so we set a default which
                # will give us UTF-8 (which appears to work in all cases on
                # OSX). Users are still free to override LANG of course, as
                # long as it give us something meaningful. This work-around
                # *may* not be needed with python 3k and/or OSX 10.5, but will
                # work with them too -- vila 20080908
                os.environ['LANG'] = 'en_US.UTF-8'
            import locale
        finally:
            sys.platform = 'darwin'
    else:
        import locale

    try:
        user_encoding = locale.getpreferredencoding()
    except locale.Error, e:
        sys.stderr.write('bzr: warning: %s\n'
                         '  Could not determine what text encoding to use.\n'
                         '  This error usually means your Python interpreter\n'
                         '  doesn\'t support the locale set by $LANG (%s)\n'
                         "  Continuing with ascii encoding.\n"
                         % (e, os.environ.get('LANG')))
        user_encoding = 'ascii'

    # Windows returns 'cp0' to indicate there is no code page. So we'll just
    # treat that as ASCII, and not support printing unicode characters to the
    # console.
    #
    # For python scripts run under vim, we get '', so also treat that as ASCII
    if user_encoding in (None, 'cp0', ''):
        user_encoding = 'ascii'
    else:
        # check encoding
        try:
            codecs.lookup(user_encoding)
        except LookupError:
            sys.stderr.write('bzr: warning:'
                             ' unknown encoding %s.'
                             ' Continuing with ascii encoding.\n'
                             % user_encoding
                            )
            user_encoding = 'ascii'

    if use_cache:
        _cached_user_encoding = user_encoding

    return user_encoding


def get_diff_header_encoding():
    return get_terminal_encoding()


def get_host_name():
    """Return the current unicode host name.

    This is meant to be used in place of socket.gethostname() because that
    behaves inconsistently on different platforms.
    """
    if sys.platform == "win32":
        import win32utils
        return win32utils.get_host_name()
    else:
        import socket
        return socket.gethostname().decode(get_user_encoding())


# We must not read/write any more than 64k at a time from/to a socket so we
# don't risk "no buffer space available" errors on some platforms.  Windows in
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
# data at once.
MAX_SOCKET_CHUNK = 64 * 1024

def read_bytes_from_socket(sock, report_activity=None,
        max_read_size=MAX_SOCKET_CHUNK):
    """Read up to max_read_size of bytes from sock and notify of progress.

    Translates "Connection reset by peer" into file-like EOF (return an
    empty string rather than raise an error), and repeats the recv if
    interrupted by a signal.
    """
    while 1:
        try:
            bytes = sock.recv(max_read_size)
        except socket.error, e:
            eno = e.args[0]
            if eno == getattr(errno, "WSAECONNRESET", errno.ECONNRESET):
                # The connection was closed by the other side.  Callers expect
                # an empty string to signal end-of-stream.
                return ""
            elif eno == errno.EINTR:
                # Retry the interrupted recv.
                continue
            raise
        else:
            if report_activity is not None:
                report_activity(len(bytes), 'read')
            return bytes


def recv_all(socket, count):
    """Receive an exact number of bytes.

    Regular Socket.recv() may return less than the requested number of bytes,
    depending on what's in the OS buffer.  MSG_WAITALL is not available
    on all platforms, but this should work everywhere.  This will return
    less than the requested amount if the remote end closes.

    This isn't optimized and is intended mostly for use in testing.
    """
    b = ''
    while len(b) < count:
        new = read_bytes_from_socket(socket, None, count - len(b))
        if new == '':
            break # eof
        b += new
    return b


def send_all(sock, bytes, report_activity=None):
    """Send all bytes on a socket.

    Breaks large blocks in smaller chunks to avoid buffering limitations on
    some platforms, and catches EINTR which may be thrown if the send is
    interrupted by a signal.

    This is preferred to socket.sendall(), because it avoids portability bugs
    and provides activity reporting.

    :param report_activity: Call this as bytes are read, see
        Transport._report_activity
    """
    sent_total = 0
    byte_count = len(bytes)
    while sent_total < byte_count:
        try:
            sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
        except socket.error, e:
            if e.args[0] != errno.EINTR:
                raise
        else:
            sent_total += sent
            report_activity(sent, 'write')


def dereference_path(path):
    """Determine the real path to a file.

    All parent elements are dereferenced.  But the file itself is not
    dereferenced.
    :param path: The original path.  May be absolute or relative.
    :return: the real path *to* the file
    """
    parent, base = os.path.split(path)
    # The pathjoin for '.' is a workaround for Python bug #1213894.
    # (initial path components aren't dereferenced)
    return pathjoin(realpath(pathjoin('.', parent)), base)


def supports_mapi():
    """Return True if we can use MAPI to launch a mail client."""
    return sys.platform == "win32"


def resource_string(package, resource_name):
    """Load a resource from a package and return it as a string.

    Note: Only packages that start with bzrlib are currently supported.

    This is designed to be a lightweight implementation of resource
    loading in a way which is API compatible with the same API from
    pkg_resources. See
    http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
    If and when pkg_resources becomes a standard library, this routine
    can delegate to it.
    """
    # Check package name is within bzrlib
    if package == "bzrlib":
        resource_relpath = resource_name
    elif package.startswith("bzrlib."):
        package = package[len("bzrlib."):].replace('.', os.sep)
        resource_relpath = pathjoin(package, resource_name)
    else:
        raise errors.BzrError('resource package %s not in bzrlib' % package)

    # Map the resource to a file and read its contents
    base = dirname(bzrlib.__file__)
    if getattr(sys, 'frozen', None):    # bzr.exe
        base = abspath(pathjoin(base, '..', '..'))
    f = file(pathjoin(base, resource_relpath), "rU")
    try:
        return f.read()
    finally:
        f.close()

def file_kind_from_stat_mode_thunk(mode):
    global file_kind_from_stat_mode
    if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
        try:
            from bzrlib._readdir_pyx import UTF8DirReader
            file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
        except ImportError, e:
            # This is one time where we won't warn that an extension failed to
            # load. The extension is never available on Windows anyway.
            from bzrlib._readdir_py import (
                _kind_from_mode as file_kind_from_stat_mode
                )
    return file_kind_from_stat_mode(mode)
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk


def file_kind(f, _lstat=os.lstat):
    try:
        return file_kind_from_stat_mode(_lstat(f).st_mode)
    except OSError, e:
        if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
            raise errors.NoSuchFile(f)
        raise


def until_no_eintr(f, *a, **kw):
    """Run f(*a, **kw), retrying if an EINTR error occurs.

    WARNING: you must be certain that it is safe to retry the call repeatedly
    if EINTR does occur.  This is typically only true for low-level operations
    like os.read.  If in any doubt, don't use this.

    Keep in mind that this is not a complete solution to EINTR.  There is
    probably code in the Python standard library and other dependencies that
    may encounter EINTR if a signal arrives (and there is signal handler for
    that signal).  So this function can reduce the impact for IO that bzrlib
    directly controls, but it is not a complete solution.
    """
    # Borrowed from Twisted's twisted.python.util.untilConcludes function.
    while True:
        try:
            return f(*a, **kw)
        except (IOError, OSError), e:
            if e.errno == errno.EINTR:
                continue
            raise


def re_compile_checked(re_string, flags=0, where=""):
    """Return a compiled re, or raise a sensible error.

    This should only be used when compiling user-supplied REs.

    :param re_string: Text form of regular expression.
    :param flags: eg re.IGNORECASE
    :param where: Message explaining to the user the context where
        it occurred, eg 'log search filter'.
    """
    # from https://bugs.launchpad.net/bzr/+bug/251352
    try:
        re_obj = re.compile(re_string, flags)
        re_obj.search("")
        return re_obj
    except re.error, e:
        if where:
            where = ' in ' + where
        # despite the name 'error' is a type
        raise errors.BzrCommandError('Invalid regular expression%s: %r: %s'
            % (where, re_string, e))


if sys.platform == "win32":
    import msvcrt
    def getchar():
        return msvcrt.getch()
else:
    import tty
    import termios
    def getchar():
        fd = sys.stdin.fileno()
        settings = termios.tcgetattr(fd)
        try:
            tty.setraw(fd)
            ch = sys.stdin.read(1)
        finally:
            termios.tcsetattr(fd, termios.TCSADRAIN, settings)
        return ch


if sys.platform == 'linux2':
    def _local_concurrency():
        concurrency = None
        prefix = 'processor'
        for line in file('/proc/cpuinfo', 'rb'):
            if line.startswith(prefix):
                concurrency = int(line[line.find(':')+1:]) + 1
        return concurrency
elif sys.platform == 'darwin':
    def _local_concurrency():
        return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
                                stdout=subprocess.PIPE).communicate()[0]
elif sys.platform[0:7] == 'freebsd':
    def _local_concurrency():
        return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
                                stdout=subprocess.PIPE).communicate()[0]
elif sys.platform == 'sunos5':
    def _local_concurrency():
        return subprocess.Popen(['psrinfo', '-p',],
                                stdout=subprocess.PIPE).communicate()[0]
elif sys.platform == "win32":
    def _local_concurrency():
        # This appears to return the number of cores.
        return os.environ.get('NUMBER_OF_PROCESSORS')
else:
    def _local_concurrency():
        # Who knows ?
        return None


_cached_local_concurrency = None

def local_concurrency(use_cache=True):
    """Return how many processes can be run concurrently.

    Rely on platform specific implementations and default to 1 (one) if
    anything goes wrong.
    """
    global _cached_local_concurrency

    if _cached_local_concurrency is not None and use_cache:
        return _cached_local_concurrency

    concurrency = os.environ.get('BZR_CONCURRENCY', None)
    if concurrency is None:
        try:
            concurrency = _local_concurrency()
        except (OSError, IOError):
            pass
    try:
        concurrency = int(concurrency)
    except (TypeError, ValueError):
        concurrency = 1
    if use_cache:
        _cached_concurrency = concurrency
    return concurrency


class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
    """A stream writer that doesn't decode str arguments."""

    def __init__(self, encode, stream, errors='strict'):
        codecs.StreamWriter.__init__(self, stream, errors)
        self.encode = encode

    def write(self, object):
        if type(object) is str:
            self.stream.write(object)
        else:
            data, _ = self.encode(object, self.errors)
            self.stream.write(data)

if sys.platform == 'win32':
    def open_file(filename, mode='r', bufsize=-1):
        """This function is used to override the ``open`` builtin.

        But it uses O_NOINHERIT flag so the file handle is not inherited by
        child processes.  Deleting or renaming a closed file opened with this
        function is not blocking child processes.
        """
        writing = 'w' in mode
        appending = 'a' in mode
        updating = '+' in mode
        binary = 'b' in mode

        flags = O_NOINHERIT
        # see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
        # for flags for each modes.
        if binary:
            flags |= O_BINARY
        else:
            flags |= O_TEXT

        if writing:
            if updating:
                flags |= os.O_RDWR
            else:
                flags |= os.O_WRONLY
            flags |= os.O_CREAT | os.O_TRUNC
        elif appending:
            if updating:
                flags |= os.O_RDWR
            else:
                flags |= os.O_WRONLY
            flags |= os.O_CREAT | os.O_APPEND
        else: #reading
            if updating:
                flags |= os.O_RDWR
            else:
                flags |= os.O_RDONLY

        return os.fdopen(os.open(filename, flags), mode, bufsize)
else:
    open_file = open


def getuser_unicode():
    """Return the username as unicode.
    """
    try:
        user_encoding = get_user_encoding()
        username = getpass.getuser().decode(user_encoding)
    except UnicodeDecodeError:
        raise errors.BzrError("Can't decode username as %s." % \
                user_encoding)
    return username