49
49
from bzrlib import (
57
from bzrlib import symbol_versioning
57
58
from bzrlib.symbol_versioning import (
58
59
deprecated_function,
62
# sha and md5 modules are deprecated in python2.6 but hashlib is available as
64
if sys.version_info < (2, 5):
65
import md5 as _mod_md5
67
import sha as _mod_sha
77
from bzrlib import symbol_versioning
80
# Cross platform wall-clock time functionality with decent resolution.
81
# On Linux ``time.clock`` returns only CPU time. On Windows, ``time.time()``
82
# only has a resolution of ~15ms. Note that ``time.clock()`` is not
83
# synchronized with ``time.time()``, this is only meant to be used to find
84
# delta times by subtracting from another call to this function.
85
timer_func = time.time
86
if sys.platform == 'win32':
87
timer_func = time.clock
62
from bzrlib.trace import mutter
89
65
# On win32, O_BINARY is used to indicate the file should
90
66
# be opened in binary mode, rather than text mode.
91
67
# On other platforms, O_BINARY doesn't exist, because
92
68
# they always open in binary mode, so it is okay to
93
# OR with 0 on those platforms.
94
# O_NOINHERIT and O_TEXT exists only on win32 too.
69
# OR with 0 on those platforms
95
70
O_BINARY = getattr(os, 'O_BINARY', 0)
96
O_TEXT = getattr(os, 'O_TEXT', 0)
97
O_NOINHERIT = getattr(os, 'O_NOINHERIT', 0)
100
def get_unicode_argv():
102
user_encoding = get_user_encoding()
103
return [a.decode(user_encoding) for a in sys.argv[1:]]
104
except UnicodeDecodeError:
105
raise errors.BzrError(("Parameter '%r' is unsupported by the current "
109
73
def make_readonly(filename):
110
74
"""Make a filename read-only."""
111
mod = os.lstat(filename).st_mode
112
if not stat.S_ISLNK(mod):
114
os.chmod(filename, mod)
75
mod = os.stat(filename).st_mode
77
os.chmod(filename, mod)
117
80
def make_writable(filename):
118
mod = os.lstat(filename).st_mode
119
if not stat.S_ISLNK(mod):
121
os.chmod(filename, mod)
124
def minimum_path_selection(paths):
125
"""Return the smallset subset of paths which are outside paths.
127
:param paths: A container (and hence not None) of paths.
128
:return: A set of paths sufficient to include everything in paths via
129
is_inside, drawn from the paths parameter.
135
return path.split('/')
136
sorted_paths = sorted(list(paths), key=sort_key)
138
search_paths = [sorted_paths[0]]
139
for path in sorted_paths[1:]:
140
if not is_inside(search_paths[-1], path):
141
# This path is unique, add it
142
search_paths.append(path)
144
return set(search_paths)
81
mod = os.stat(filename).st_mode
83
os.chmod(filename, mod)
564
def pumpfile(from_file, to_file, read_length=-1, buff_size=32768,
565
report_activity=None, direction='read'):
566
"""Copy contents of one file to another.
568
The read_length can either be -1 to read to end-of-file (EOF) or
569
it can specify the maximum number of bytes to read.
571
The buff_size represents the maximum size for each read operation
572
performed on from_file.
574
:param report_activity: Call this as bytes are read, see
575
Transport._report_activity
576
:param direction: Will be passed to report_activity
578
:return: The number of bytes copied.
582
# read specified number of bytes
584
while read_length > 0:
585
num_bytes_to_read = min(read_length, buff_size)
587
block = from_file.read(num_bytes_to_read)
591
if report_activity is not None:
592
report_activity(len(block), direction)
595
actual_bytes_read = len(block)
596
read_length -= actual_bytes_read
597
length += actual_bytes_read
601
block = from_file.read(buff_size)
605
if report_activity is not None:
606
report_activity(len(block), direction)
612
def pump_string_file(bytes, file_handle, segment_size=None):
613
"""Write bytes to file_handle in many smaller writes.
615
:param bytes: The string to write.
616
:param file_handle: The file to write to.
618
# Write data in chunks rather than all at once, because very large
619
# writes fail on some platforms (e.g. Windows with SMB mounted
622
segment_size = 5242880 # 5MB
623
segments = range(len(bytes) / segment_size + 1)
624
write = file_handle.write
625
for segment_index in segments:
626
segment = buffer(bytes, segment_index * segment_size, segment_size)
535
def pumpfile(fromfile, tofile):
536
"""Copy contents of one file to another."""
539
b = fromfile.read(BUFSIZE)
630
545
def file_iterator(input_file, readsize=32768):
718
605
offset = datetime.fromtimestamp(t) - datetime.utcfromtimestamp(t)
719
606
return offset.days * 86400 + offset.seconds
721
weekdays = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun']
722
_default_format_by_weekday_num = [wd + " %Y-%m-%d %H:%M:%S" for wd in weekdays]
725
def format_date(t, offset=0, timezone='original', date_fmt=None,
609
def format_date(t, offset=0, timezone='original', date_fmt=None,
726
610
show_offset=True):
727
"""Return a formatted date string.
729
:param t: Seconds since the epoch.
730
:param offset: Timezone offset in seconds east of utc.
731
:param timezone: How to display the time: 'utc', 'original' for the
732
timezone specified by offset, or 'local' for the process's current
734
:param date_fmt: strftime format.
735
:param show_offset: Whether to append the timezone.
737
(date_fmt, tt, offset_str) = \
738
_format_date(t, offset, timezone, date_fmt, show_offset)
739
date_fmt = date_fmt.replace('%a', weekdays[tt[6]])
740
date_str = time.strftime(date_fmt, tt)
741
return date_str + offset_str
744
# Cache of formatted offset strings
748
def format_date_with_offset_in_original_timezone(t, offset=0,
749
_cache=_offset_cache):
750
"""Return a formatted date string in the original timezone.
752
This routine may be faster then format_date.
754
:param t: Seconds since the epoch.
755
:param offset: Timezone offset in seconds east of utc.
759
tt = time.gmtime(t + offset)
760
date_fmt = _default_format_by_weekday_num[tt[6]]
761
date_str = time.strftime(date_fmt, tt)
762
offset_str = _cache.get(offset, None)
763
if offset_str is None:
764
offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
765
_cache[offset] = offset_str
766
return date_str + offset_str
769
def format_local_date(t, offset=0, timezone='original', date_fmt=None,
771
"""Return an unicode date string formatted according to the current locale.
773
:param t: Seconds since the epoch.
774
:param offset: Timezone offset in seconds east of utc.
775
:param timezone: How to display the time: 'utc', 'original' for the
776
timezone specified by offset, or 'local' for the process's current
778
:param date_fmt: strftime format.
779
:param show_offset: Whether to append the timezone.
781
(date_fmt, tt, offset_str) = \
782
_format_date(t, offset, timezone, date_fmt, show_offset)
783
date_str = time.strftime(date_fmt, tt)
784
if not isinstance(date_str, unicode):
785
date_str = date_str.decode(get_user_encoding(), 'replace')
786
return date_str + offset_str
789
def _format_date(t, offset, timezone, date_fmt, show_offset):
611
## TODO: Perhaps a global option to use either universal or local time?
612
## Or perhaps just let people set $TZ?
613
assert isinstance(t, float)
790
615
if timezone == 'utc':
791
616
tt = time.gmtime(t)
753
assert isinstance(p, (list, tuple))
927
755
if (f == '..') or (f is None) or (f == ''):
928
756
raise errors.BzrError("sorry, %r not allowed in path" % f)
929
757
return pathjoin(*p)
932
def parent_directories(filename):
933
"""Return the list of parent directories, deepest first.
935
For example, parent_directories("a/b/c") -> ["a/b", "a"].
938
parts = splitpath(dirname(filename))
940
parents.append(joinpath(parts))
945
_extension_load_failures = []
948
def failed_to_load_extension(exception):
949
"""Handle failing to load a binary extension.
951
This should be called from the ImportError block guarding the attempt to
952
import the native extension. If this function returns, the pure-Python
953
implementation should be loaded instead::
956
>>> import bzrlib._fictional_extension_pyx
957
>>> except ImportError, e:
958
>>> bzrlib.osutils.failed_to_load_extension(e)
959
>>> import bzrlib._fictional_extension_py
961
# NB: This docstring is just an example, not a doctest, because doctest
962
# currently can't cope with the use of lazy imports in this namespace --
965
# This currently doesn't report the failure at the time it occurs, because
966
# they tend to happen very early in startup when we can't check config
967
# files etc, and also we want to report all failures but not spam the user
969
from bzrlib import trace
970
exception_str = str(exception)
971
if exception_str not in _extension_load_failures:
972
trace.mutter("failed to load compiled extension: %s" % exception_str)
973
_extension_load_failures.append(exception_str)
976
def report_extension_load_failures():
977
if not _extension_load_failures:
979
from bzrlib.config import GlobalConfig
980
if GlobalConfig().get_user_option_as_bool('ignore_missing_extensions'):
982
# the warnings framework should by default show this only once
983
from bzrlib.trace import warning
985
"bzr: warning: some compiled extensions could not be loaded; "
986
"see <https://answers.launchpad.net/bzr/+faq/703>")
987
# we no longer show the specific missing extensions here, because it makes
988
# the message too long and scary - see
989
# https://bugs.launchpad.net/bzr/+bug/430529
993
from bzrlib._chunks_to_lines_pyx import chunks_to_lines
994
except ImportError, e:
995
failed_to_load_extension(e)
996
from bzrlib._chunks_to_lines_py import chunks_to_lines
760
@deprecated_function(zero_nine)
761
def appendpath(p1, p2):
765
return pathjoin(p1, p2)
999
768
def split_lines(s):
1000
769
"""Split s into lines, but without removing the newline characters."""
1001
# Trivially convert a fulltext into a 'chunked' representation, and let
1002
# chunks_to_lines do the heavy lifting.
1003
if isinstance(s, str):
1004
# chunks_to_lines only supports 8-bit strings
1005
return chunks_to_lines([s])
1007
return _split_lines(s)
1010
def _split_lines(s):
1011
"""Split s into lines, but without removing the newline characters.
1013
This supports Unicode or plain string objects.
1015
770
lines = s.split('\n')
1016
771
result = [line + '\n' for line in lines[:-1]]
1138
846
os.path.commonprefix (python2.4) has a bad bug that it works just
1139
847
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
1140
848
avoids that problem.
1142
NOTE: `base` should not have a trailing slash otherwise you'll get
1143
PathNotChild exceptions regardless of `path`.
1146
if len(base) < MIN_ABS_PATHLENGTH:
1147
# must have space for e.g. a drive letter
1148
raise ValueError('%r is too short to calculate a relative path'
851
assert len(base) >= MIN_ABS_PATHLENGTH, ('Length of base must be equal or'
852
' exceed the platform minimum length (which is %d)' %
1151
855
rp = abspath(path)
1156
if len(head) <= len(base) and head != base:
1157
raise errors.PathNotChild(rp, base)
859
while len(head) >= len(base):
1158
860
if head == base:
1160
head, tail = split(head)
862
head, tail = os.path.split(head)
866
raise errors.PathNotChild(rp, base)
1165
return pathjoin(*reversed(s))
1170
def _cicp_canonical_relpath(base, path):
1171
"""Return the canonical path relative to base.
1173
Like relpath, but on case-insensitive-case-preserving file-systems, this
1174
will return the relpath as stored on the file-system rather than in the
1175
case specified in the input string, for all existing portions of the path.
1177
This will cause O(N) behaviour if called for every path in a tree; if you
1178
have a number of paths to convert, you should use canonical_relpaths().
1180
# TODO: it should be possible to optimize this for Windows by using the
1181
# win32 API FindFiles function to look for the specified name - but using
1182
# os.listdir() still gives us the correct, platform agnostic semantics in
1185
rel = relpath(base, path)
1186
# '.' will have been turned into ''
1190
abs_base = abspath(base)
1192
_listdir = os.listdir
1194
# use an explicit iterator so we can easily consume the rest on early exit.
1195
bit_iter = iter(rel.split('/'))
1196
for bit in bit_iter:
1199
next_entries = _listdir(current)
1200
except OSError: # enoent, eperm, etc
1201
# We can't find this in the filesystem, so just append the
1203
current = pathjoin(current, bit, *list(bit_iter))
1205
for look in next_entries:
1206
if lbit == look.lower():
1207
current = pathjoin(current, look)
1210
# got to the end, nothing matched, so we just return the
1211
# non-existing bits as they were specified (the filename may be
1212
# the target of a move, for example).
1213
current = pathjoin(current, bit, *list(bit_iter))
1215
return current[len(abs_base):].lstrip('/')
1217
# XXX - TODO - we need better detection/integration of case-insensitive
1218
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
1219
# filesystems), for example, so could probably benefit from the same basic
1220
# support there. For now though, only Windows and OSX get that support, and
1221
# they get it for *all* file-systems!
1222
if sys.platform in ('win32', 'darwin'):
1223
canonical_relpath = _cicp_canonical_relpath
1225
canonical_relpath = relpath
1227
def canonical_relpaths(base, paths):
1228
"""Create an iterable to canonicalize a sequence of relative paths.
1230
The intent is for this implementation to use a cache, vastly speeding
1231
up multiple transformations in the same directory.
1233
# but for now, we haven't optimized...
1234
return [canonical_relpath(base, p) for p in paths]
1236
874
def safe_unicode(unicode_or_utf8_string):
1237
875
"""Coerce unicode_or_utf8_string into unicode.
1239
877
If it is unicode, it is returned.
1240
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1241
wrapped in a BzrBadParameterNotUnicode exception.
878
Otherwise it is decoded from utf-8. If a decoding error
879
occurs, it is wrapped as a If the decoding fails, the exception is wrapped
880
as a BzrBadParameter exception.
1243
882
if isinstance(unicode_or_utf8_string, unicode):
1244
883
return unicode_or_utf8_string
1357
996
normalized_filename = _inaccessible_normalized_filename
1360
def set_signal_handler(signum, handler, restart_syscall=True):
1361
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1362
on platforms that support that.
1364
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1365
automatically restart (by calling `signal.siginterrupt(signum,
1366
False)`). May be ignored if the feature is not available on this
1367
platform or Python version.
1371
siginterrupt = signal.siginterrupt
1373
# This python implementation doesn't provide signal support, hence no
1376
except AttributeError:
1377
# siginterrupt doesn't exist on this platform, or for this version
1379
siginterrupt = lambda signum, flag: None
1381
def sig_handler(*args):
1382
# Python resets the siginterrupt flag when a signal is
1383
# received. <http://bugs.python.org/issue8354>
1384
# As a workaround for some cases, set it back the way we want it.
1385
siginterrupt(signum, False)
1386
# Now run the handler function passed to set_signal_handler.
1389
sig_handler = handler
1390
old_handler = signal.signal(signum, sig_handler)
1392
siginterrupt(signum, False)
1396
default_terminal_width = 80
1397
"""The default terminal width for ttys.
1399
This is defined so that higher levels can share a common fallback value when
1400
terminal_width() returns None.
1403
# Keep some state so that terminal_width can detect if _terminal_size has
1404
# returned a different size since the process started. See docstring and
1405
# comments of terminal_width for details.
1406
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
1407
_terminal_size_state = 'no_data'
1408
_first_terminal_size = None
1410
999
def terminal_width():
1411
"""Return terminal width.
1413
None is returned if the width can't established precisely.
1416
- if BZR_COLUMNS is set, returns its value
1417
- if there is no controlling terminal, returns None
1418
- query the OS, if the queried size has changed since the last query,
1420
- if COLUMNS is set, returns its value,
1421
- if the OS has a value (even though it's never changed), return its value.
1423
From there, we need to query the OS to get the size of the controlling
1426
On Unices we query the OS by:
1427
- get termios.TIOCGWINSZ
1428
- if an error occurs or a negative value is obtained, returns None
1430
On Windows we query the OS by:
1431
- win32utils.get_console_size() decides,
1432
- returns None on error (provided default value)
1434
# Note to implementors: if changing the rules for determining the width,
1435
# make sure you've considered the behaviour in these cases:
1436
# - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
1437
# - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
1439
# - (add more interesting cases here, if you find any)
1440
# Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
1441
# but we don't want to register a signal handler because it is impossible
1442
# to do so without risking EINTR errors in Python <= 2.6.5 (see
1443
# <http://bugs.python.org/issue8354>). Instead we check TIOCGWINSZ every
1444
# time so we can notice if the reported size has changed, which should have
1447
# If BZR_COLUMNS is set, take it, user is always right
1449
return int(os.environ['BZR_COLUMNS'])
1450
except (KeyError, ValueError):
1453
isatty = getattr(sys.stdout, 'isatty', None)
1454
if isatty is None or not isatty():
1455
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1459
width, height = os_size = _terminal_size(None, None)
1460
global _first_terminal_size, _terminal_size_state
1461
if _terminal_size_state == 'no_data':
1462
_first_terminal_size = os_size
1463
_terminal_size_state = 'unchanged'
1464
elif (_terminal_size_state == 'unchanged' and
1465
_first_terminal_size != os_size):
1466
_terminal_size_state = 'changed'
1468
# If the OS claims to know how wide the terminal is, and this value has
1469
# ever changed, use that.
1470
if _terminal_size_state == 'changed':
1471
if width is not None and width > 0:
1474
# If COLUMNS is set, use it.
1476
return int(os.environ['COLUMNS'])
1477
except (KeyError, ValueError):
1480
# Finally, use an unchanged size from the OS, if we have one.
1481
if _terminal_size_state == 'unchanged':
1482
if width is not None and width > 0:
1485
# The width could not be determined.
1489
def _win32_terminal_size(width, height):
1490
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1491
return width, height
1494
def _ioctl_terminal_size(width, height):
1000
"""Return estimated terminal width."""
1001
if sys.platform == 'win32':
1002
return win32utils.get_console_size()[0]
1496
1005
import struct, fcntl, termios
1497
1006
s = struct.pack('HHHH', 0, 0, 0, 0)
1498
1007
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1499
height, width = struct.unpack('HHHH', x)[0:2]
1500
except (IOError, AttributeError):
1008
width = struct.unpack('HHHH', x)[1]
1502
return width, height
1504
_terminal_size = None
1505
"""Returns the terminal size as (width, height).
1507
:param width: Default value for width.
1508
:param height: Default value for height.
1510
This is defined specifically for each OS and query the size of the controlling
1511
terminal. If any error occurs, the provided default values should be returned.
1513
if sys.platform == 'win32':
1514
_terminal_size = _win32_terminal_size
1516
_terminal_size = _ioctl_terminal_size
1013
width = int(os.environ['COLUMNS'])
1519
1022
def supports_executable():
1566
1069
raise errors.IllegalPath(path)
1569
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1571
def _is_error_enotdir(e):
1572
"""Check if this exception represents ENOTDIR.
1574
Unfortunately, python is very inconsistent about the exception
1575
here. The cases are:
1576
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1577
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1578
which is the windows error code.
1579
3) Windows, Python2.5 uses errno == EINVAL and
1580
winerror == ERROR_DIRECTORY
1582
:param e: An Exception object (expected to be OSError with an errno
1583
attribute, but we should be able to cope with anything)
1584
:return: True if this represents an ENOTDIR error. False otherwise.
1586
en = getattr(e, 'errno', None)
1587
if (en == errno.ENOTDIR
1588
or (sys.platform == 'win32'
1589
and (en == _WIN32_ERROR_DIRECTORY
1590
or (en == errno.EINVAL
1591
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1597
1072
def walkdirs(top, prefix=""):
1598
1073
"""Yield data about all the directories in a tree.
1600
1075
This yields all the data about the contents of a directory at a time.
1601
1076
After each directory has been yielded, if the caller has mutated the list
1602
1077
to exclude some directories, they are then not descended into.
1604
1079
The data yielded is of the form:
1605
1080
((directory-relpath, directory-path-from-top),
1606
[(relpath, basename, kind, lstat, path-from-top), ...]),
1081
[(directory-relpath, basename, kind, lstat, path-from-top), ...]),
1607
1082
- directory-relpath is the relative path of the directory being returned
1608
1083
with respect to top. prefix is prepended to this.
1609
- directory-path-from-root is the path including top for this directory.
1084
- directory-path-from-root is the path including top for this directory.
1610
1085
It is suitable for use with os functions.
1611
1086
- relpath is the relative path within the subtree being walked.
1612
1087
- basename is the basename of the path
1614
1089
present within the tree - but it may be recorded as versioned. See
1615
1090
versioned_kind.
1616
1091
- lstat is the stat data *if* the file was statted.
1617
- planned, not implemented:
1092
- planned, not implemented:
1618
1093
path_from_tree_root is the path from the root of the tree.
1620
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1095
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1621
1096
allows one to walk a subtree but get paths that are relative to a tree
1622
1097
rooted higher up.
1623
1098
:return: an iterator over the dirs.
1625
1100
#TODO there is a bit of a smell where the results of the directory-
1626
# summary in this, and the path from the root, may not agree
1101
# summary in this, and the path from the root, may not agree
1627
1102
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1628
1103
# potentially confusing output. We should make this more robust - but
1629
1104
# not at a speed cost. RBC 20060731
1630
1105
_lstat = os.lstat
1631
1106
_directory = _directory_kind
1632
1107
_listdir = os.listdir
1633
_kind_from_mode = file_kind_from_stat_mode
1108
_kind_from_mode = _formats.get
1634
1109
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1636
1111
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1645
1120
append = dirblock.append
1647
names = sorted(_listdir(top))
1649
if not _is_error_enotdir(e):
1653
abspath = top_slash + name
1654
statvalue = _lstat(abspath)
1655
kind = _kind_from_mode(statvalue.st_mode)
1656
append((relprefix + name, name, kind, statvalue, abspath))
1121
for name in sorted(_listdir(top)):
1122
abspath = top_slash + name
1123
statvalue = _lstat(abspath)
1124
kind = _kind_from_mode(statvalue.st_mode & 0170000, 'unknown')
1125
append((relprefix + name, name, kind, statvalue, abspath))
1657
1126
yield (relroot, top), dirblock
1659
1128
# push the user specified dirs from dirblock
1660
1129
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1663
class DirReader(object):
1664
"""An interface for reading directories."""
1666
def top_prefix_to_starting_dir(self, top, prefix=""):
1667
"""Converts top and prefix to a starting dir entry
1669
:param top: A utf8 path
1670
:param prefix: An optional utf8 path to prefix output relative paths
1672
:return: A tuple starting with prefix, and ending with the native
1675
raise NotImplementedError(self.top_prefix_to_starting_dir)
1677
def read_dir(self, prefix, top):
1678
"""Read a specific dir.
1680
:param prefix: A utf8 prefix to be preprended to the path basenames.
1681
:param top: A natively encoded path to read.
1682
:return: A list of the directories contents. Each item contains:
1683
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1685
raise NotImplementedError(self.read_dir)
1688
_selected_dir_reader = None
1691
1132
def _walkdirs_utf8(top, prefix=""):
1692
1133
"""Yield data about all the directories in a tree.
1702
1143
path-from-top might be unicode or utf8, but it is the correct path to
1703
1144
pass to os functions to affect the file in question. (such as os.lstat)
1705
global _selected_dir_reader
1706
if _selected_dir_reader is None:
1707
fs_encoding = _fs_enc.upper()
1708
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1709
# Win98 doesn't have unicode apis like FindFirstFileW
1710
# TODO: We possibly could support Win98 by falling back to the
1711
# original FindFirstFile, and using TCHAR instead of WCHAR,
1712
# but that gets a bit tricky, and requires custom compiling
1715
from bzrlib._walkdirs_win32 import Win32ReadDir
1716
_selected_dir_reader = Win32ReadDir()
1719
elif fs_encoding in ('UTF-8', 'US-ASCII', 'ANSI_X3.4-1968'):
1720
# ANSI_X3.4-1968 is a form of ASCII
1722
from bzrlib._readdir_pyx import UTF8DirReader
1723
_selected_dir_reader = UTF8DirReader()
1724
except ImportError, e:
1725
failed_to_load_extension(e)
1728
if _selected_dir_reader is None:
1729
# Fallback to the python version
1730
_selected_dir_reader = UnicodeDirReader()
1146
fs_encoding = sys.getfilesystemencoding()
1147
if (sys.platform == 'win32' or
1148
fs_encoding not in ('UTF-8', 'US-ASCII', 'ANSI_X3.4-1968')): # ascii
1149
return _walkdirs_unicode_to_utf8(top, prefix=prefix)
1151
return _walkdirs_fs_utf8(top, prefix=prefix)
1154
def _walkdirs_fs_utf8(top, prefix=""):
1155
"""See _walkdirs_utf8.
1157
This sub-function is called when we know the filesystem is already in utf8
1158
encoding. So we don't need to transcode filenames.
1161
_directory = _directory_kind
1162
_listdir = os.listdir
1163
_kind_from_mode = _formats.get
1732
1165
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1733
1166
# But we don't actually uses 1-3 in pending, so set them to None
1734
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1735
read_dir = _selected_dir_reader.read_dir
1736
_directory = _directory_kind
1167
pending = [(safe_utf8(prefix), None, None, None, safe_utf8(top))]
1738
relroot, _, _, _, top = pending[-1].pop()
1741
dirblock = sorted(read_dir(relroot, top))
1169
relroot, _, _, _, top = pending.pop()
1171
relprefix = relroot + '/'
1174
top_slash = top + '/'
1177
append = dirblock.append
1178
for name in sorted(_listdir(top)):
1179
abspath = top_slash + name
1180
statvalue = _lstat(abspath)
1181
kind = _kind_from_mode(statvalue.st_mode & 0170000, 'unknown')
1182
append((relprefix + name, name, kind, statvalue, abspath))
1742
1183
yield (relroot, top), dirblock
1743
1185
# push the user specified dirs from dirblock
1744
next = [d for d in reversed(dirblock) if d[2] == _directory]
1746
pending.append(next)
1749
class UnicodeDirReader(DirReader):
1750
"""A dir reader for non-utf8 file systems, which transcodes."""
1752
__slots__ = ['_utf8_encode']
1755
self._utf8_encode = codecs.getencoder('utf8')
1757
def top_prefix_to_starting_dir(self, top, prefix=""):
1758
"""See DirReader.top_prefix_to_starting_dir."""
1759
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1761
def read_dir(self, prefix, top):
1762
"""Read a single directory from a non-utf8 file system.
1764
top, and the abspath element in the output are unicode, all other paths
1765
are utf8. Local disk IO is done via unicode calls to listdir etc.
1767
This is currently the fallback code path when the filesystem encoding is
1768
not UTF-8. It may be better to implement an alternative so that we can
1769
safely handle paths that are not properly decodable in the current
1772
See DirReader.read_dir for details.
1774
_utf8_encode = self._utf8_encode
1776
_listdir = os.listdir
1777
_kind_from_mode = file_kind_from_stat_mode
1780
relprefix = prefix + '/'
1186
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1189
def _walkdirs_unicode_to_utf8(top, prefix=""):
1190
"""See _walkdirs_utf8
1192
Because Win32 has a Unicode api, all of the 'path-from-top' entries will be
1194
This is currently the fallback code path when the filesystem encoding is
1195
not UTF-8. It may be better to implement an alternative so that we can
1196
safely handle paths that are not properly decodable in the current
1199
_utf8_encode = codecs.getencoder('utf8')
1201
_directory = _directory_kind
1202
_listdir = os.listdir
1203
_kind_from_mode = _formats.get
1205
pending = [(safe_utf8(prefix), None, None, None, safe_unicode(top))]
1207
relroot, _, _, _, top = pending.pop()
1209
relprefix = relroot + '/'
1783
1212
top_slash = top + u'/'
1952
1346
return user_encoding
1955
def get_host_name():
1956
"""Return the current unicode host name.
1958
This is meant to be used in place of socket.gethostname() because that
1959
behaves inconsistently on different platforms.
1961
if sys.platform == "win32":
1963
return win32utils.get_host_name()
1966
return socket.gethostname().decode(get_user_encoding())
1969
# We must not read/write any more than 64k at a time from/to a socket so we
1970
# don't risk "no buffer space available" errors on some platforms. Windows in
1971
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
1973
MAX_SOCKET_CHUNK = 64 * 1024
1975
def read_bytes_from_socket(sock, report_activity=None,
1976
max_read_size=MAX_SOCKET_CHUNK):
1977
"""Read up to max_read_size of bytes from sock and notify of progress.
1979
Translates "Connection reset by peer" into file-like EOF (return an
1980
empty string rather than raise an error), and repeats the recv if
1981
interrupted by a signal.
1985
bytes = sock.recv(max_read_size)
1986
except socket.error, e:
1988
if eno == getattr(errno, "WSAECONNRESET", errno.ECONNRESET):
1989
# The connection was closed by the other side. Callers expect
1990
# an empty string to signal end-of-stream.
1992
elif eno == errno.EINTR:
1993
# Retry the interrupted recv.
1997
if report_activity is not None:
1998
report_activity(len(bytes), 'read')
2002
def recv_all(socket, count):
1349
def recv_all(socket, bytes):
2003
1350
"""Receive an exact number of bytes.
2005
1352
Regular Socket.recv() may return less than the requested number of bytes,
2006
depending on what's in the OS buffer. MSG_WAITALL is not available
1353
dependning on what's in the OS buffer. MSG_WAITALL is not available
2007
1354
on all platforms, but this should work everywhere. This will return
2008
1355
less than the requested amount if the remote end closes.
2010
1357
This isn't optimized and is intended mostly for use in testing.
2013
while len(b) < count:
2014
new = read_bytes_from_socket(socket, None, count - len(b))
1360
while len(b) < bytes:
1361
new = socket.recv(bytes - len(b))
2021
def send_all(sock, bytes, report_activity=None):
2022
"""Send all bytes on a socket.
2024
Breaks large blocks in smaller chunks to avoid buffering limitations on
2025
some platforms, and catches EINTR which may be thrown if the send is
2026
interrupted by a signal.
2028
This is preferred to socket.sendall(), because it avoids portability bugs
2029
and provides activity reporting.
2031
:param report_activity: Call this as bytes are read, see
2032
Transport._report_activity
2035
byte_count = len(bytes)
2036
while sent_total < byte_count:
2038
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2039
except socket.error, e:
2040
if e.args[0] != errno.EINTR:
2044
report_activity(sent, 'write')
2047
1367
def dereference_path(path):
2048
1368
"""Determine the real path to a file.
2056
1376
# The pathjoin for '.' is a workaround for Python bug #1213894.
2057
1377
# (initial path components aren't dereferenced)
2058
1378
return pathjoin(realpath(pathjoin('.', parent)), base)
2061
def supports_mapi():
2062
"""Return True if we can use MAPI to launch a mail client."""
2063
return sys.platform == "win32"
2066
def resource_string(package, resource_name):
2067
"""Load a resource from a package and return it as a string.
2069
Note: Only packages that start with bzrlib are currently supported.
2071
This is designed to be a lightweight implementation of resource
2072
loading in a way which is API compatible with the same API from
2074
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2075
If and when pkg_resources becomes a standard library, this routine
2078
# Check package name is within bzrlib
2079
if package == "bzrlib":
2080
resource_relpath = resource_name
2081
elif package.startswith("bzrlib."):
2082
package = package[len("bzrlib."):].replace('.', os.sep)
2083
resource_relpath = pathjoin(package, resource_name)
2085
raise errors.BzrError('resource package %s not in bzrlib' % package)
2087
# Map the resource to a file and read its contents
2088
base = dirname(bzrlib.__file__)
2089
if getattr(sys, 'frozen', None): # bzr.exe
2090
base = abspath(pathjoin(base, '..', '..'))
2091
f = file(pathjoin(base, resource_relpath), "rU")
2097
def file_kind_from_stat_mode_thunk(mode):
2098
global file_kind_from_stat_mode
2099
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2101
from bzrlib._readdir_pyx import UTF8DirReader
2102
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2103
except ImportError, e:
2104
# This is one time where we won't warn that an extension failed to
2105
# load. The extension is never available on Windows anyway.
2106
from bzrlib._readdir_py import (
2107
_kind_from_mode as file_kind_from_stat_mode
2109
return file_kind_from_stat_mode(mode)
2110
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2113
def file_kind(f, _lstat=os.lstat):
2115
return file_kind_from_stat_mode(_lstat(f).st_mode)
2117
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2118
raise errors.NoSuchFile(f)
2122
def until_no_eintr(f, *a, **kw):
2123
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2125
WARNING: you must be certain that it is safe to retry the call repeatedly
2126
if EINTR does occur. This is typically only true for low-level operations
2127
like os.read. If in any doubt, don't use this.
2129
Keep in mind that this is not a complete solution to EINTR. There is
2130
probably code in the Python standard library and other dependencies that
2131
may encounter EINTR if a signal arrives (and there is signal handler for
2132
that signal). So this function can reduce the impact for IO that bzrlib
2133
directly controls, but it is not a complete solution.
2135
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2139
except (IOError, OSError), e:
2140
if e.errno == errno.EINTR:
2145
def re_compile_checked(re_string, flags=0, where=""):
2146
"""Return a compiled re, or raise a sensible error.
2148
This should only be used when compiling user-supplied REs.
2150
:param re_string: Text form of regular expression.
2151
:param flags: eg re.IGNORECASE
2152
:param where: Message explaining to the user the context where
2153
it occurred, eg 'log search filter'.
2155
# from https://bugs.launchpad.net/bzr/+bug/251352
2157
re_obj = re.compile(re_string, flags)
2162
where = ' in ' + where
2163
# despite the name 'error' is a type
2164
raise errors.BzrCommandError('Invalid regular expression%s: %r: %s'
2165
% (where, re_string, e))
2168
if sys.platform == "win32":
2171
return msvcrt.getch()
2176
fd = sys.stdin.fileno()
2177
settings = termios.tcgetattr(fd)
2180
ch = sys.stdin.read(1)
2182
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2186
if sys.platform == 'linux2':
2187
def _local_concurrency():
2189
prefix = 'processor'
2190
for line in file('/proc/cpuinfo', 'rb'):
2191
if line.startswith(prefix):
2192
concurrency = int(line[line.find(':')+1:]) + 1
2194
elif sys.platform == 'darwin':
2195
def _local_concurrency():
2196
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2197
stdout=subprocess.PIPE).communicate()[0]
2198
elif sys.platform[0:7] == 'freebsd':
2199
def _local_concurrency():
2200
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2201
stdout=subprocess.PIPE).communicate()[0]
2202
elif sys.platform == 'sunos5':
2203
def _local_concurrency():
2204
return subprocess.Popen(['psrinfo', '-p',],
2205
stdout=subprocess.PIPE).communicate()[0]
2206
elif sys.platform == "win32":
2207
def _local_concurrency():
2208
# This appears to return the number of cores.
2209
return os.environ.get('NUMBER_OF_PROCESSORS')
2211
def _local_concurrency():
2216
_cached_local_concurrency = None
2218
def local_concurrency(use_cache=True):
2219
"""Return how many processes can be run concurrently.
2221
Rely on platform specific implementations and default to 1 (one) if
2222
anything goes wrong.
2224
global _cached_local_concurrency
2226
if _cached_local_concurrency is not None and use_cache:
2227
return _cached_local_concurrency
2229
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2230
if concurrency is None:
2232
concurrency = _local_concurrency()
2233
except (OSError, IOError):
2236
concurrency = int(concurrency)
2237
except (TypeError, ValueError):
2240
_cached_concurrency = concurrency
2244
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2245
"""A stream writer that doesn't decode str arguments."""
2247
def __init__(self, encode, stream, errors='strict'):
2248
codecs.StreamWriter.__init__(self, stream, errors)
2249
self.encode = encode
2251
def write(self, object):
2252
if type(object) is str:
2253
self.stream.write(object)
2255
data, _ = self.encode(object, self.errors)
2256
self.stream.write(data)
2258
if sys.platform == 'win32':
2259
def open_file(filename, mode='r', bufsize=-1):
2260
"""This function is used to override the ``open`` builtin.
2262
But it uses O_NOINHERIT flag so the file handle is not inherited by
2263
child processes. Deleting or renaming a closed file opened with this
2264
function is not blocking child processes.
2266
writing = 'w' in mode
2267
appending = 'a' in mode
2268
updating = '+' in mode
2269
binary = 'b' in mode
2272
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2273
# for flags for each modes.
2283
flags |= os.O_WRONLY
2284
flags |= os.O_CREAT | os.O_TRUNC
2289
flags |= os.O_WRONLY
2290
flags |= os.O_CREAT | os.O_APPEND
2295
flags |= os.O_RDONLY
2297
return os.fdopen(os.open(filename, flags), mode, bufsize)
2302
def getuser_unicode():
2303
"""Return the username as unicode.
2306
user_encoding = get_user_encoding()
2307
username = getpass.getuser().decode(user_encoding)
2308
except UnicodeDecodeError:
2309
raise errors.BzrError("Can't decode username as %s." % \
added
removed
bzrlib/osutils.py
