594
1138
os.path.commonprefix (python2.4) has a bad bug that it works just
595
1139
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
596
1140
avoids that problem.
1142
NOTE: `base` should not have a trailing slash otherwise you'll get
1143
PathNotChild exceptions regardless of `path`.
598
if sys.platform != "win32":
602
assert len(base) >= minlength, ('Length of base must be equal or exceed the'
603
' platform minimum length (which is %d)' % minlength)
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'
604
1151
rp = abspath(path)
608
while len(head) >= len(base):
1156
if len(head) <= len(base) and head != base:
1157
raise errors.PathNotChild(rp, base)
609
1158
if head == base:
611
head, tail = os.path.split(head)
1160
head, tail = split(head)
615
# XXX This should raise a NotChildPath exception, as its not tied
617
raise 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]
1237
def decode_filename(filename):
1238
"""Decode the filename using the filesystem encoding
1240
If it is unicode, it is returned.
1241
Otherwise it is decoded from the the filesystem's encoding. If decoding
1242
fails, a errors.BadFilenameEncoding exception is raised.
1244
if type(filename) is unicode:
1247
return filename.decode(_fs_enc)
1248
except UnicodeDecodeError:
1249
raise errors.BadFilenameEncoding(filename, _fs_enc)
625
1252
def safe_unicode(unicode_or_utf8_string):
626
1253
"""Coerce unicode_or_utf8_string into unicode.
628
1255
If it is unicode, it is returned.
629
Otherwise it is decoded from utf-8. If a decoding error
630
occurs, it is wrapped as a If the decoding fails, the exception is wrapped
631
as a BzrBadParameter exception.
1256
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1257
wrapped in a BzrBadParameterNotUnicode exception.
633
1259
if isinstance(unicode_or_utf8_string, unicode):
634
1260
return unicode_or_utf8_string
636
1262
return unicode_or_utf8_string.decode('utf8')
637
1263
except UnicodeDecodeError:
638
raise BzrBadParameterNotUnicode(unicode_or_utf8_string)
1264
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1267
def safe_utf8(unicode_or_utf8_string):
1268
"""Coerce unicode_or_utf8_string to a utf8 string.
1270
If it is a str, it is returned.
1271
If it is Unicode, it is encoded into a utf-8 string.
1273
if isinstance(unicode_or_utf8_string, str):
1274
# TODO: jam 20070209 This is overkill, and probably has an impact on
1275
# performance if we are dealing with lots of apis that want a
1278
# Make sure it is a valid utf-8 string
1279
unicode_or_utf8_string.decode('utf-8')
1280
except UnicodeDecodeError:
1281
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1282
return unicode_or_utf8_string
1283
return unicode_or_utf8_string.encode('utf-8')
1286
_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
1287
' Revision id generators should be creating utf8'
1291
def safe_revision_id(unicode_or_utf8_string, warn=True):
1292
"""Revision ids should now be utf8, but at one point they were unicode.
1294
:param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
1296
:param warn: Functions that are sanitizing user data can set warn=False
1297
:return: None or a utf8 revision id.
1299
if (unicode_or_utf8_string is None
1300
or unicode_or_utf8_string.__class__ == str):
1301
return unicode_or_utf8_string
1303
symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
1305
return cache_utf8.encode(unicode_or_utf8_string)
1308
_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
1309
' generators should be creating utf8 file ids.')
1312
def safe_file_id(unicode_or_utf8_string, warn=True):
1313
"""File ids should now be utf8, but at one point they were unicode.
1315
This is the same as safe_utf8, except it uses the cached encode functions
1316
to save a little bit of performance.
1318
:param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
1320
:param warn: Functions that are sanitizing user data can set warn=False
1321
:return: None or a utf8 file id.
1323
if (unicode_or_utf8_string is None
1324
or unicode_or_utf8_string.__class__ == str):
1325
return unicode_or_utf8_string
1327
symbol_versioning.warn(_file_id_warning, DeprecationWarning,
1329
return cache_utf8.encode(unicode_or_utf8_string)
1332
_platform_normalizes_filenames = False
1333
if sys.platform == 'darwin':
1334
_platform_normalizes_filenames = True
1337
def normalizes_filenames():
1338
"""Return True if this platform normalizes unicode filenames.
1342
return _platform_normalizes_filenames
1345
def _accessible_normalized_filename(path):
1346
"""Get the unicode normalized path, and if you can access the file.
1348
On platforms where the system normalizes filenames (Mac OSX),
1349
you can access a file by any path which will normalize correctly.
1350
On platforms where the system does not normalize filenames
1351
(everything else), you have to access a file by its exact path.
1353
Internally, bzr only supports NFC normalization, since that is
1354
the standard for XML documents.
1356
So return the normalized path, and a flag indicating if the file
1357
can be accessed by that path.
1360
return unicodedata.normalize('NFC', unicode(path)), True
1363
def _inaccessible_normalized_filename(path):
1364
__doc__ = _accessible_normalized_filename.__doc__
1366
normalized = unicodedata.normalize('NFC', unicode(path))
1367
return normalized, normalized == path
1370
if _platform_normalizes_filenames:
1371
normalized_filename = _accessible_normalized_filename
1373
normalized_filename = _inaccessible_normalized_filename
1376
def set_signal_handler(signum, handler, restart_syscall=True):
1377
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1378
on platforms that support that.
1380
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1381
automatically restart (by calling `signal.siginterrupt(signum,
1382
False)`). May be ignored if the feature is not available on this
1383
platform or Python version.
1387
siginterrupt = signal.siginterrupt
1389
# This python implementation doesn't provide signal support, hence no
1392
except AttributeError:
1393
# siginterrupt doesn't exist on this platform, or for this version
1395
siginterrupt = lambda signum, flag: None
1397
def sig_handler(*args):
1398
# Python resets the siginterrupt flag when a signal is
1399
# received. <http://bugs.python.org/issue8354>
1400
# As a workaround for some cases, set it back the way we want it.
1401
siginterrupt(signum, False)
1402
# Now run the handler function passed to set_signal_handler.
1405
sig_handler = handler
1406
old_handler = signal.signal(signum, sig_handler)
1408
siginterrupt(signum, False)
1412
default_terminal_width = 80
1413
"""The default terminal width for ttys.
1415
This is defined so that higher levels can share a common fallback value when
1416
terminal_width() returns None.
1419
# Keep some state so that terminal_width can detect if _terminal_size has
1420
# returned a different size since the process started. See docstring and
1421
# comments of terminal_width for details.
1422
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
1423
_terminal_size_state = 'no_data'
1424
_first_terminal_size = None
641
1426
def terminal_width():
642
"""Return estimated terminal width."""
644
# TODO: Do something smart on Windows?
646
# TODO: Is there anything that gets a better update when the window
647
# is resized while the program is running? We could use the Python termcap
1427
"""Return terminal width.
1429
None is returned if the width can't established precisely.
1432
- if BZR_COLUMNS is set, returns its value
1433
- if there is no controlling terminal, returns None
1434
- query the OS, if the queried size has changed since the last query,
1436
- if COLUMNS is set, returns its value,
1437
- if the OS has a value (even though it's never changed), return its value.
1439
From there, we need to query the OS to get the size of the controlling
1442
On Unices we query the OS by:
1443
- get termios.TIOCGWINSZ
1444
- if an error occurs or a negative value is obtained, returns None
1446
On Windows we query the OS by:
1447
- win32utils.get_console_size() decides,
1448
- returns None on error (provided default value)
1450
# Note to implementors: if changing the rules for determining the width,
1451
# make sure you've considered the behaviour in these cases:
1452
# - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
1453
# - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
1455
# - (add more interesting cases here, if you find any)
1456
# Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
1457
# but we don't want to register a signal handler because it is impossible
1458
# to do so without risking EINTR errors in Python <= 2.6.5 (see
1459
# <http://bugs.python.org/issue8354>). Instead we check TIOCGWINSZ every
1460
# time so we can notice if the reported size has changed, which should have
1463
# If BZR_COLUMNS is set, take it, user is always right
1464
# Except if they specified 0 in which case, impose no limit here
1466
width = int(os.environ['BZR_COLUMNS'])
1467
except (KeyError, ValueError):
1469
if width is not None:
1475
isatty = getattr(sys.stdout, 'isatty', None)
1476
if isatty is None or not isatty():
1477
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1481
width, height = os_size = _terminal_size(None, None)
1482
global _first_terminal_size, _terminal_size_state
1483
if _terminal_size_state == 'no_data':
1484
_first_terminal_size = os_size
1485
_terminal_size_state = 'unchanged'
1486
elif (_terminal_size_state == 'unchanged' and
1487
_first_terminal_size != os_size):
1488
_terminal_size_state = 'changed'
1490
# If the OS claims to know how wide the terminal is, and this value has
1491
# ever changed, use that.
1492
if _terminal_size_state == 'changed':
1493
if width is not None and width > 0:
1496
# If COLUMNS is set, use it.
650
1498
return int(os.environ['COLUMNS'])
651
except (IndexError, KeyError, ValueError):
1499
except (KeyError, ValueError):
1502
# Finally, use an unchanged size from the OS, if we have one.
1503
if _terminal_size_state == 'unchanged':
1504
if width is not None and width > 0:
1507
# The width could not be determined.
1511
def _win32_terminal_size(width, height):
1512
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1513
return width, height
1516
def _ioctl_terminal_size(width, height):
1518
import struct, fcntl, termios
1519
s = struct.pack('HHHH', 0, 0, 0, 0)
1520
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1521
height, width = struct.unpack('HHHH', x)[0:2]
1522
except (IOError, AttributeError):
1524
return width, height
1526
_terminal_size = None
1527
"""Returns the terminal size as (width, height).
1529
:param width: Default value for width.
1530
:param height: Default value for height.
1532
This is defined specifically for each OS and query the size of the controlling
1533
terminal. If any error occurs, the provided default values should be returned.
1535
if sys.platform == 'win32':
1536
_terminal_size = _win32_terminal_size
1538
_terminal_size = _ioctl_terminal_size
654
1541
def supports_executable():
655
1542
return sys.platform != "win32"
1545
def supports_posix_readonly():
1546
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1548
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1549
directory controls creation/deletion, etc.
1551
And under win32, readonly means that the directory itself cannot be
1552
deleted. The contents of a readonly directory can be changed, unlike POSIX
1553
where files in readonly directories cannot be added, deleted or renamed.
1555
return sys.platform != "win32"
1558
def set_or_unset_env(env_variable, value):
1559
"""Modify the environment, setting or removing the env_variable.
1561
:param env_variable: The environment variable in question
1562
:param value: The value to set the environment to. If None, then
1563
the variable will be removed.
1564
:return: The original value of the environment variable.
1566
orig_val = os.environ.get(env_variable)
1568
if orig_val is not None:
1569
del os.environ[env_variable]
1571
if isinstance(value, unicode):
1572
value = value.encode(get_user_encoding())
1573
os.environ[env_variable] = value
1577
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
1580
def check_legal_path(path):
1581
"""Check whether the supplied path is legal.
1582
This is only required on Windows, so we don't test on other platforms
1585
if sys.platform != "win32":
1587
if _validWin32PathRE.match(path) is None:
1588
raise errors.IllegalPath(path)
1591
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1593
def _is_error_enotdir(e):
1594
"""Check if this exception represents ENOTDIR.
1596
Unfortunately, python is very inconsistent about the exception
1597
here. The cases are:
1598
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1599
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1600
which is the windows error code.
1601
3) Windows, Python2.5 uses errno == EINVAL and
1602
winerror == ERROR_DIRECTORY
1604
:param e: An Exception object (expected to be OSError with an errno
1605
attribute, but we should be able to cope with anything)
1606
:return: True if this represents an ENOTDIR error. False otherwise.
1608
en = getattr(e, 'errno', None)
1609
if (en == errno.ENOTDIR
1610
or (sys.platform == 'win32'
1611
and (en == _WIN32_ERROR_DIRECTORY
1612
or (en == errno.EINVAL
1613
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1619
def walkdirs(top, prefix=""):
1620
"""Yield data about all the directories in a tree.
1622
This yields all the data about the contents of a directory at a time.
1623
After each directory has been yielded, if the caller has mutated the list
1624
to exclude some directories, they are then not descended into.
1626
The data yielded is of the form:
1627
((directory-relpath, directory-path-from-top),
1628
[(relpath, basename, kind, lstat, path-from-top), ...]),
1629
- directory-relpath is the relative path of the directory being returned
1630
with respect to top. prefix is prepended to this.
1631
- directory-path-from-root is the path including top for this directory.
1632
It is suitable for use with os functions.
1633
- relpath is the relative path within the subtree being walked.
1634
- basename is the basename of the path
1635
- kind is the kind of the file now. If unknown then the file is not
1636
present within the tree - but it may be recorded as versioned. See
1638
- lstat is the stat data *if* the file was statted.
1639
- planned, not implemented:
1640
path_from_tree_root is the path from the root of the tree.
1642
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1643
allows one to walk a subtree but get paths that are relative to a tree
1645
:return: an iterator over the dirs.
1647
#TODO there is a bit of a smell where the results of the directory-
1648
# summary in this, and the path from the root, may not agree
1649
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1650
# potentially confusing output. We should make this more robust - but
1651
# not at a speed cost. RBC 20060731
1653
_directory = _directory_kind
1654
_listdir = os.listdir
1655
_kind_from_mode = file_kind_from_stat_mode
1656
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1658
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1659
relroot, _, _, _, top = pending.pop()
1661
relprefix = relroot + u'/'
1664
top_slash = top + u'/'
1667
append = dirblock.append
1669
names = sorted(map(decode_filename, _listdir(top)))
1671
if not _is_error_enotdir(e):
1675
abspath = top_slash + name
1676
statvalue = _lstat(abspath)
1677
kind = _kind_from_mode(statvalue.st_mode)
1678
append((relprefix + name, name, kind, statvalue, abspath))
1679
yield (relroot, top), dirblock
1681
# push the user specified dirs from dirblock
1682
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1685
class DirReader(object):
1686
"""An interface for reading directories."""
1688
def top_prefix_to_starting_dir(self, top, prefix=""):
1689
"""Converts top and prefix to a starting dir entry
1691
:param top: A utf8 path
1692
:param prefix: An optional utf8 path to prefix output relative paths
1694
:return: A tuple starting with prefix, and ending with the native
1697
raise NotImplementedError(self.top_prefix_to_starting_dir)
1699
def read_dir(self, prefix, top):
1700
"""Read a specific dir.
1702
:param prefix: A utf8 prefix to be preprended to the path basenames.
1703
:param top: A natively encoded path to read.
1704
:return: A list of the directories contents. Each item contains:
1705
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1707
raise NotImplementedError(self.read_dir)
1710
_selected_dir_reader = None
1713
def _walkdirs_utf8(top, prefix=""):
1714
"""Yield data about all the directories in a tree.
1716
This yields the same information as walkdirs() only each entry is yielded
1717
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1718
are returned as exact byte-strings.
1720
:return: yields a tuple of (dir_info, [file_info])
1721
dir_info is (utf8_relpath, path-from-top)
1722
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1723
if top is an absolute path, path-from-top is also an absolute path.
1724
path-from-top might be unicode or utf8, but it is the correct path to
1725
pass to os functions to affect the file in question. (such as os.lstat)
1727
global _selected_dir_reader
1728
if _selected_dir_reader is None:
1729
fs_encoding = _fs_enc.upper()
1730
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1731
# Win98 doesn't have unicode apis like FindFirstFileW
1732
# TODO: We possibly could support Win98 by falling back to the
1733
# original FindFirstFile, and using TCHAR instead of WCHAR,
1734
# but that gets a bit tricky, and requires custom compiling
1737
from bzrlib._walkdirs_win32 import Win32ReadDir
1738
_selected_dir_reader = Win32ReadDir()
1741
elif fs_encoding in ('UTF-8', 'US-ASCII', 'ANSI_X3.4-1968'):
1742
# ANSI_X3.4-1968 is a form of ASCII
1744
from bzrlib._readdir_pyx import UTF8DirReader
1745
_selected_dir_reader = UTF8DirReader()
1746
except ImportError, e:
1747
failed_to_load_extension(e)
1750
if _selected_dir_reader is None:
1751
# Fallback to the python version
1752
_selected_dir_reader = UnicodeDirReader()
1754
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1755
# But we don't actually uses 1-3 in pending, so set them to None
1756
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1757
read_dir = _selected_dir_reader.read_dir
1758
_directory = _directory_kind
1760
relroot, _, _, _, top = pending[-1].pop()
1763
dirblock = sorted(read_dir(relroot, top))
1764
yield (relroot, top), dirblock
1765
# push the user specified dirs from dirblock
1766
next = [d for d in reversed(dirblock) if d[2] == _directory]
1768
pending.append(next)
1771
class UnicodeDirReader(DirReader):
1772
"""A dir reader for non-utf8 file systems, which transcodes."""
1774
__slots__ = ['_utf8_encode']
1777
self._utf8_encode = codecs.getencoder('utf8')
1779
def top_prefix_to_starting_dir(self, top, prefix=""):
1780
"""See DirReader.top_prefix_to_starting_dir."""
1781
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1783
def read_dir(self, prefix, top):
1784
"""Read a single directory from a non-utf8 file system.
1786
top, and the abspath element in the output are unicode, all other paths
1787
are utf8. Local disk IO is done via unicode calls to listdir etc.
1789
This is currently the fallback code path when the filesystem encoding is
1790
not UTF-8. It may be better to implement an alternative so that we can
1791
safely handle paths that are not properly decodable in the current
1794
See DirReader.read_dir for details.
1796
_utf8_encode = self._utf8_encode
1798
_listdir = os.listdir
1799
_kind_from_mode = file_kind_from_stat_mode
1802
relprefix = prefix + '/'
1805
top_slash = top + u'/'
1808
append = dirblock.append
1809
for name in sorted(_listdir(top)):
1811
name_utf8 = _utf8_encode(name)[0]
1812
except UnicodeDecodeError:
1813
raise errors.BadFilenameEncoding(
1814
_utf8_encode(relprefix)[0] + name, _fs_enc)
1815
abspath = top_slash + name
1816
statvalue = _lstat(abspath)
1817
kind = _kind_from_mode(statvalue.st_mode)
1818
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1822
def copy_tree(from_path, to_path, handlers={}):
1823
"""Copy all of the entries in from_path into to_path.
1825
:param from_path: The base directory to copy.
1826
:param to_path: The target directory. If it does not exist, it will
1828
:param handlers: A dictionary of functions, which takes a source and
1829
destinations for files, directories, etc.
1830
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1831
'file', 'directory', and 'symlink' should always exist.
1832
If they are missing, they will be replaced with 'os.mkdir()',
1833
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1835
# Now, just copy the existing cached tree to the new location
1836
# We use a cheap trick here.
1837
# Absolute paths are prefixed with the first parameter
1838
# relative paths are prefixed with the second.
1839
# So we can get both the source and target returned
1840
# without any extra work.
1842
def copy_dir(source, dest):
1845
def copy_link(source, dest):
1846
"""Copy the contents of a symlink"""
1847
link_to = os.readlink(source)
1848
os.symlink(link_to, dest)
1850
real_handlers = {'file':shutil.copy2,
1851
'symlink':copy_link,
1852
'directory':copy_dir,
1854
real_handlers.update(handlers)
1856
if not os.path.exists(to_path):
1857
real_handlers['directory'](from_path, to_path)
1859
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1860
for relpath, name, kind, st, abspath in entries:
1861
real_handlers[kind](abspath, relpath)
1864
def copy_ownership_from_path(dst, src=None):
1865
"""Copy usr/grp ownership from src file/dir to dst file/dir.
1867
If src is None, the containing directory is used as source. If chown
1868
fails, the error is ignored and a warning is printed.
1870
chown = getattr(os, 'chown', None)
1875
src = os.path.dirname(dst)
1881
chown(dst, s.st_uid, s.st_gid)
1884
'Unable to copy ownership from "%s" to "%s". '
1885
'You may want to set it manually.', src, dst)
1886
trace.log_exception_quietly()
1889
def path_prefix_key(path):
1890
"""Generate a prefix-order path key for path.
1892
This can be used to sort paths in the same way that walkdirs does.
1894
return (dirname(path) , path)
1897
def compare_paths_prefix_order(path_a, path_b):
1898
"""Compare path_a and path_b to generate the same order walkdirs uses."""
1899
key_a = path_prefix_key(path_a)
1900
key_b = path_prefix_key(path_b)
1901
return cmp(key_a, key_b)
1904
_cached_user_encoding = None
1907
def get_user_encoding(use_cache=True):
1908
"""Find out what the preferred user encoding is.
1910
This is generally the encoding that is used for command line parameters
1911
and file contents. This may be different from the terminal encoding
1912
or the filesystem encoding.
1914
:param use_cache: Enable cache for detected encoding.
1915
(This parameter is turned on by default,
1916
and required only for selftesting)
1918
:return: A string defining the preferred user encoding
1920
global _cached_user_encoding
1921
if _cached_user_encoding is not None and use_cache:
1922
return _cached_user_encoding
1924
if sys.platform == 'darwin':
1925
# python locale.getpreferredencoding() always return
1926
# 'mac-roman' on darwin. That's a lie.
1927
sys.platform = 'posix'
1929
if os.environ.get('LANG', None) is None:
1930
# If LANG is not set, we end up with 'ascii', which is bad
1931
# ('mac-roman' is more than ascii), so we set a default which
1932
# will give us UTF-8 (which appears to work in all cases on
1933
# OSX). Users are still free to override LANG of course, as
1934
# long as it give us something meaningful. This work-around
1935
# *may* not be needed with python 3k and/or OSX 10.5, but will
1936
# work with them too -- vila 20080908
1937
os.environ['LANG'] = 'en_US.UTF-8'
1940
sys.platform = 'darwin'
1945
user_encoding = locale.getpreferredencoding()
1946
except locale.Error, e:
1947
sys.stderr.write('bzr: warning: %s\n'
1948
' Could not determine what text encoding to use.\n'
1949
' This error usually means your Python interpreter\n'
1950
' doesn\'t support the locale set by $LANG (%s)\n'
1951
" Continuing with ascii encoding.\n"
1952
% (e, os.environ.get('LANG')))
1953
user_encoding = 'ascii'
1955
# Windows returns 'cp0' to indicate there is no code page. So we'll just
1956
# treat that as ASCII, and not support printing unicode characters to the
1959
# For python scripts run under vim, we get '', so also treat that as ASCII
1960
if user_encoding in (None, 'cp0', ''):
1961
user_encoding = 'ascii'
1965
codecs.lookup(user_encoding)
1967
sys.stderr.write('bzr: warning:'
1968
' unknown encoding %s.'
1969
' Continuing with ascii encoding.\n'
1972
user_encoding = 'ascii'
1975
_cached_user_encoding = user_encoding
1977
return user_encoding
1980
def get_diff_header_encoding():
1981
return get_terminal_encoding()
1984
def get_host_name():
1985
"""Return the current unicode host name.
1987
This is meant to be used in place of socket.gethostname() because that
1988
behaves inconsistently on different platforms.
1990
if sys.platform == "win32":
1992
return win32utils.get_host_name()
1995
return socket.gethostname().decode(get_user_encoding())
1998
# We must not read/write any more than 64k at a time from/to a socket so we
1999
# don't risk "no buffer space available" errors on some platforms. Windows in
2000
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
2002
MAX_SOCKET_CHUNK = 64 * 1024
2004
_end_of_stream_errors = [errno.ECONNRESET]
2005
for _eno in ['WSAECONNRESET', 'WSAECONNABORTED']:
2006
_eno = getattr(errno, _eno, None)
2007
if _eno is not None:
2008
_end_of_stream_errors.append(_eno)
2012
def read_bytes_from_socket(sock, report_activity=None,
2013
max_read_size=MAX_SOCKET_CHUNK):
2014
"""Read up to max_read_size of bytes from sock and notify of progress.
2016
Translates "Connection reset by peer" into file-like EOF (return an
2017
empty string rather than raise an error), and repeats the recv if
2018
interrupted by a signal.
2022
bytes = sock.recv(max_read_size)
2023
except socket.error, e:
2025
if eno in _end_of_stream_errors:
2026
# The connection was closed by the other side. Callers expect
2027
# an empty string to signal end-of-stream.
2029
elif eno == errno.EINTR:
2030
# Retry the interrupted recv.
2034
if report_activity is not None:
2035
report_activity(len(bytes), 'read')
2039
def recv_all(socket, count):
2040
"""Receive an exact number of bytes.
2042
Regular Socket.recv() may return less than the requested number of bytes,
2043
depending on what's in the OS buffer. MSG_WAITALL is not available
2044
on all platforms, but this should work everywhere. This will return
2045
less than the requested amount if the remote end closes.
2047
This isn't optimized and is intended mostly for use in testing.
2050
while len(b) < count:
2051
new = read_bytes_from_socket(socket, None, count - len(b))
2058
def send_all(sock, bytes, report_activity=None):
2059
"""Send all bytes on a socket.
2061
Breaks large blocks in smaller chunks to avoid buffering limitations on
2062
some platforms, and catches EINTR which may be thrown if the send is
2063
interrupted by a signal.
2065
This is preferred to socket.sendall(), because it avoids portability bugs
2066
and provides activity reporting.
2068
:param report_activity: Call this as bytes are read, see
2069
Transport._report_activity
2072
byte_count = len(bytes)
2073
while sent_total < byte_count:
2075
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2076
except socket.error, e:
2077
if e.args[0] != errno.EINTR:
2081
report_activity(sent, 'write')
2084
def connect_socket(address):
2085
# Slight variation of the socket.create_connection() function (provided by
2086
# python-2.6) that can fail if getaddrinfo returns an empty list. We also
2087
# provide it for previous python versions. Also, we don't use the timeout
2088
# parameter (provided by the python implementation) so we don't implement
2090
err = socket.error('getaddrinfo returns an empty list')
2091
host, port = address
2092
for res in socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM):
2093
af, socktype, proto, canonname, sa = res
2096
sock = socket.socket(af, socktype, proto)
2100
except socket.error, err:
2101
# 'err' is now the most recent error
2102
if sock is not None:
2107
def dereference_path(path):
2108
"""Determine the real path to a file.
2110
All parent elements are dereferenced. But the file itself is not
2112
:param path: The original path. May be absolute or relative.
2113
:return: the real path *to* the file
2115
parent, base = os.path.split(path)
2116
# The pathjoin for '.' is a workaround for Python bug #1213894.
2117
# (initial path components aren't dereferenced)
2118
return pathjoin(realpath(pathjoin('.', parent)), base)
2121
def supports_mapi():
2122
"""Return True if we can use MAPI to launch a mail client."""
2123
return sys.platform == "win32"
2126
def resource_string(package, resource_name):
2127
"""Load a resource from a package and return it as a string.
2129
Note: Only packages that start with bzrlib are currently supported.
2131
This is designed to be a lightweight implementation of resource
2132
loading in a way which is API compatible with the same API from
2134
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2135
If and when pkg_resources becomes a standard library, this routine
2138
# Check package name is within bzrlib
2139
if package == "bzrlib":
2140
resource_relpath = resource_name
2141
elif package.startswith("bzrlib."):
2142
package = package[len("bzrlib."):].replace('.', os.sep)
2143
resource_relpath = pathjoin(package, resource_name)
2145
raise errors.BzrError('resource package %s not in bzrlib' % package)
2147
# Map the resource to a file and read its contents
2148
base = dirname(bzrlib.__file__)
2149
if getattr(sys, 'frozen', None): # bzr.exe
2150
base = abspath(pathjoin(base, '..', '..'))
2151
f = file(pathjoin(base, resource_relpath), "rU")
2157
def file_kind_from_stat_mode_thunk(mode):
2158
global file_kind_from_stat_mode
2159
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2161
from bzrlib._readdir_pyx import UTF8DirReader
2162
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2163
except ImportError, e:
2164
# This is one time where we won't warn that an extension failed to
2165
# load. The extension is never available on Windows anyway.
2166
from bzrlib._readdir_py import (
2167
_kind_from_mode as file_kind_from_stat_mode
2169
return file_kind_from_stat_mode(mode)
2170
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2173
def file_kind(f, _lstat=os.lstat):
2175
return file_kind_from_stat_mode(_lstat(f).st_mode)
2177
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2178
raise errors.NoSuchFile(f)
2182
def until_no_eintr(f, *a, **kw):
2183
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2185
WARNING: you must be certain that it is safe to retry the call repeatedly
2186
if EINTR does occur. This is typically only true for low-level operations
2187
like os.read. If in any doubt, don't use this.
2189
Keep in mind that this is not a complete solution to EINTR. There is
2190
probably code in the Python standard library and other dependencies that
2191
may encounter EINTR if a signal arrives (and there is signal handler for
2192
that signal). So this function can reduce the impact for IO that bzrlib
2193
directly controls, but it is not a complete solution.
2195
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2199
except (IOError, OSError), e:
2200
if e.errno == errno.EINTR:
2205
@deprecated_function(deprecated_in((2, 2, 0)))
2206
def re_compile_checked(re_string, flags=0, where=""):
2207
"""Return a compiled re, or raise a sensible error.
2209
This should only be used when compiling user-supplied REs.
2211
:param re_string: Text form of regular expression.
2212
:param flags: eg re.IGNORECASE
2213
:param where: Message explaining to the user the context where
2214
it occurred, eg 'log search filter'.
2216
# from https://bugs.launchpad.net/bzr/+bug/251352
2218
re_obj = re.compile(re_string, flags)
2221
except errors.InvalidPattern, e:
2223
where = ' in ' + where
2224
# despite the name 'error' is a type
2225
raise errors.BzrCommandError('Invalid regular expression%s: %s'
2229
if sys.platform == "win32":
2232
return msvcrt.getch()
2237
fd = sys.stdin.fileno()
2238
settings = termios.tcgetattr(fd)
2241
ch = sys.stdin.read(1)
2243
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2247
if sys.platform == 'linux2':
2248
def _local_concurrency():
2250
prefix = 'processor'
2251
for line in file('/proc/cpuinfo', 'rb'):
2252
if line.startswith(prefix):
2253
concurrency = int(line[line.find(':')+1:]) + 1
2255
elif sys.platform == 'darwin':
2256
def _local_concurrency():
2257
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2258
stdout=subprocess.PIPE).communicate()[0]
2259
elif sys.platform[0:7] == 'freebsd':
2260
def _local_concurrency():
2261
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2262
stdout=subprocess.PIPE).communicate()[0]
2263
elif sys.platform == 'sunos5':
2264
def _local_concurrency():
2265
return subprocess.Popen(['psrinfo', '-p',],
2266
stdout=subprocess.PIPE).communicate()[0]
2267
elif sys.platform == "win32":
2268
def _local_concurrency():
2269
# This appears to return the number of cores.
2270
return os.environ.get('NUMBER_OF_PROCESSORS')
2272
def _local_concurrency():
2277
_cached_local_concurrency = None
2279
def local_concurrency(use_cache=True):
2280
"""Return how many processes can be run concurrently.
2282
Rely on platform specific implementations and default to 1 (one) if
2283
anything goes wrong.
2285
global _cached_local_concurrency
2287
if _cached_local_concurrency is not None and use_cache:
2288
return _cached_local_concurrency
2290
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2291
if concurrency is None:
2293
concurrency = _local_concurrency()
2294
except (OSError, IOError):
2297
concurrency = int(concurrency)
2298
except (TypeError, ValueError):
2301
_cached_concurrency = concurrency
2305
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2306
"""A stream writer that doesn't decode str arguments."""
2308
def __init__(self, encode, stream, errors='strict'):
2309
codecs.StreamWriter.__init__(self, stream, errors)
2310
self.encode = encode
2312
def write(self, object):
2313
if type(object) is str:
2314
self.stream.write(object)
2316
data, _ = self.encode(object, self.errors)
2317
self.stream.write(data)
2319
if sys.platform == 'win32':
2320
def open_file(filename, mode='r', bufsize=-1):
2321
"""This function is used to override the ``open`` builtin.
2323
But it uses O_NOINHERIT flag so the file handle is not inherited by
2324
child processes. Deleting or renaming a closed file opened with this
2325
function is not blocking child processes.
2327
writing = 'w' in mode
2328
appending = 'a' in mode
2329
updating = '+' in mode
2330
binary = 'b' in mode
2333
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2334
# for flags for each modes.
2344
flags |= os.O_WRONLY
2345
flags |= os.O_CREAT | os.O_TRUNC
2350
flags |= os.O_WRONLY
2351
flags |= os.O_CREAT | os.O_APPEND
2356
flags |= os.O_RDONLY
2358
return os.fdopen(os.open(filename, flags), mode, bufsize)
2363
def getuser_unicode():
2364
"""Return the username as unicode.
2367
user_encoding = get_user_encoding()
2368
username = getpass.getuser().decode(user_encoding)
2369
except UnicodeDecodeError:
2370
raise errors.BzrError("Can't decode username as %s." % \
2375
def available_backup_name(base, exists):
2376
"""Find a non-existing backup file name.
2378
This will *not* create anything, this only return a 'free' entry. This
2379
should be used for checking names in a directory below a locked
2380
tree/branch/repo to avoid race conditions. This is LBYL (Look Before You
2381
Leap) and generally discouraged.
2383
:param base: The base name.
2385
:param exists: A callable returning True if the path parameter exists.
2388
name = "%s.~%d~" % (base, counter)
2391
name = "%s.~%d~" % (base, counter)
2395
def set_fd_cloexec(fd):
2396
"""Set a Unix file descriptor's FD_CLOEXEC flag. Do nothing if platform
2397
support for this is not available.
2401
old = fcntl.fcntl(fd, fcntl.F_GETFD)
2402
fcntl.fcntl(fd, fcntl.F_SETFD, old | fcntl.FD_CLOEXEC)
2403
except (ImportError, AttributeError):
2404
# Either the fcntl module or specific constants are not present
2408
def find_executable_on_path(name):
2409
"""Finds an executable on the PATH.
2411
On Windows, this will try to append each extension in the PATHEXT
2412
environment variable to the name, if it cannot be found with the name
2415
:param name: The base name of the executable.
2416
:return: The path to the executable found or None.
2418
path = os.environ.get('PATH')
2421
path = path.split(os.pathsep)
2422
if sys.platform == 'win32':
2423
exts = os.environ.get('PATHEXT', '').split(os.pathsep)
2424
exts = [ext.lower() for ext in exts]
2425
base, ext = os.path.splitext(name)
2427
if ext.lower() not in exts:
2435
f = os.path.join(d, name) + ext
2436
if os.access(f, os.X_OK):
added
removed
bzrlib/osutils.py
