195
288
unlink_func(tmp_name)
197
290
rename_func(tmp_name, new)
291
if failure_exc is not None:
293
raise failure_exc[0], failure_exc[1], failure_exc[2]
200
298
# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
201
299
# choke on a Unicode string containing a relative path if
202
300
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
204
_fs_enc = sys.getfilesystemencoding()
205
302
def _posix_abspath(path):
206
return os.path.abspath(path.encode(_fs_enc)).decode(_fs_enc)
207
# jam 20060426 This is another possibility which mimics
208
# os.path.abspath, only uses unicode characters instead
209
# if not os.path.isabs(path):
210
# return os.path.join(os.getcwdu(), path)
303
# jam 20060426 rather than encoding to fsencoding
304
# copy posixpath.abspath, but use os.getcwdu instead
305
if not posixpath.isabs(path):
306
path = posixpath.join(getcwd(), path)
307
return _posix_normpath(path)
214
310
def _posix_realpath(path):
215
return os.path.realpath(path.encode(_fs_enc)).decode(_fs_enc)
311
return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)
314
def _posix_normpath(path):
315
path = posixpath.normpath(path)
316
# Bug 861008: posixpath.normpath() returns a path normalized according to
317
# the POSIX standard, which stipulates (for compatibility reasons) that two
318
# leading slashes must not be simplified to one, and only if there are 3 or
319
# more should they be simplified as one. So we treat the leading 2 slashes
320
# as a special case here by simply removing the first slash, as we consider
321
# that breaking POSIX compatibility for this obscure feature is acceptable.
322
# This is not a paranoid precaution, as we notably get paths like this when
323
# the repo is hosted at the root of the filesystem, i.e. in "/".
324
if path.startswith('//'):
329
def _posix_path_from_environ(key):
330
"""Get unicode path from `key` in environment or None if not present
332
Note that posix systems use arbitrary byte strings for filesystem objects,
333
so a path that raises BadFilenameEncoding here may still be accessible.
335
val = os.environ.get(key, None)
339
return val.decode(_fs_enc)
340
except UnicodeDecodeError:
341
# GZ 2011-12-12:Ideally want to include `key` in the exception message
342
raise errors.BadFilenameEncoding(val, _fs_enc)
345
def _posix_get_home_dir():
346
"""Get the home directory of the current user as a unicode path"""
347
path = posixpath.expanduser("~")
349
return path.decode(_fs_enc)
350
except UnicodeDecodeError:
351
raise errors.BadFilenameEncoding(path, _fs_enc)
354
def _posix_getuser_unicode():
355
"""Get username from environment or password database as unicode"""
356
name = getpass.getuser()
357
user_encoding = get_user_encoding()
359
return name.decode(user_encoding)
360
except UnicodeDecodeError:
361
raise errors.BzrError("Encoding of username %r is unsupported by %s "
362
"application locale." % (name, user_encoding))
365
def _win32_fixdrive(path):
366
"""Force drive letters to be consistent.
368
win32 is inconsistent whether it returns lower or upper case
369
and even if it was consistent the user might type the other
370
so we force it to uppercase
371
running python.exe under cmd.exe return capital C:\\
372
running win32 python inside a cygwin shell returns lowercase c:\\
374
drive, path = ntpath.splitdrive(path)
375
return drive.upper() + path
218
378
def _win32_abspath(path):
219
return _nt_abspath(path.encode(_fs_enc)).decode(_fs_enc).replace('\\', '/')
379
# Real ntpath.abspath doesn't have a problem with a unicode cwd
380
return _win32_fixdrive(ntpath.abspath(unicode(path)).replace('\\', '/'))
383
def _win98_abspath(path):
384
"""Return the absolute version of a path.
385
Windows 98 safe implementation (python reimplementation
386
of Win32 API function GetFullPathNameW)
391
# \\HOST\path => //HOST/path
392
# //HOST/path => //HOST/path
393
# path => C:/cwd/path
396
# check for absolute path
397
drive = ntpath.splitdrive(path)[0]
398
if drive == '' and path[:2] not in('//','\\\\'):
400
# we cannot simply os.path.join cwd and path
401
# because os.path.join('C:','/path') produce '/path'
402
# and this is incorrect
403
if path[:1] in ('/','\\'):
404
cwd = ntpath.splitdrive(cwd)[0]
406
path = cwd + '\\' + path
407
return _win32_fixdrive(ntpath.normpath(path).replace('\\', '/'))
222
410
def _win32_realpath(path):
223
return _nt_realpath(path.encode(_fs_enc)).decode(_fs_enc).replace('\\', '/')
411
# Real ntpath.realpath doesn't have a problem with a unicode cwd
412
return _win32_fixdrive(ntpath.realpath(unicode(path)).replace('\\', '/'))
226
415
def _win32_pathjoin(*args):
227
return _nt_join(*args).replace('\\', '/')
416
return ntpath.join(*args).replace('\\', '/')
230
419
def _win32_normpath(path):
231
return _nt_normpath(path).replace('\\', '/')
420
return _win32_fixdrive(ntpath.normpath(unicode(path)).replace('\\', '/'))
234
423
def _win32_getcwd():
235
return os.getcwdu().replace('\\', '/')
424
return _win32_fixdrive(os.getcwdu().replace('\\', '/'))
238
427
def _win32_mkdtemp(*args, **kwargs):
239
return tempfile.mkdtemp(*args, **kwargs).replace('\\', '/')
428
return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))
242
431
def _win32_rename(old, new):
243
fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
432
"""We expect to be able to atomically replace 'new' with old.
434
On win32, if new exists, it must be moved out of the way first,
438
fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
440
if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
441
# If we try to rename a non-existant file onto cwd, we get
442
# EPERM or EACCES instead of ENOENT, this will raise ENOENT
443
# if the old path doesn't exist, sometimes we get EACCES
444
# On Linux, we seem to get EBUSY, on Mac we get EINVAL
450
return unicodedata.normalize('NFC', os.getcwdu())
453
def _rename_wrap_exception(rename_func):
454
"""Adds extra information to any exceptions that come from rename().
456
The exception has an updated message and 'old_filename' and 'new_filename'
460
def _rename_wrapper(old, new):
462
rename_func(old, new)
464
detailed_error = OSError(e.errno, e.strerror +
465
" [occurred when renaming '%s' to '%s']" %
467
detailed_error.old_filename = old
468
detailed_error.new_filename = new
471
return _rename_wrapper
473
# Default rename wraps os.rename()
474
rename = _rename_wrap_exception(os.rename)
246
476
# Default is to just use the python builtins, but these can be rebound on
247
477
# particular platforms.
248
478
abspath = _posix_abspath
249
479
realpath = _posix_realpath
250
480
pathjoin = os.path.join
251
normpath = os.path.normpath
481
normpath = _posix_normpath
482
path_from_environ = _posix_path_from_environ
483
_get_home_dir = _posix_get_home_dir
484
getuser_unicode = _posix_getuser_unicode
252
485
getcwd = os.getcwdu
253
mkdtemp = tempfile.mkdtemp
255
486
dirname = os.path.dirname
256
487
basename = os.path.basename
257
rmtree = shutil.rmtree
488
split = os.path.split
489
splitext = os.path.splitext
490
# These were already lazily imported into local scope
491
# mkdtemp = tempfile.mkdtemp
492
# rmtree = shutil.rmtree
259
500
MIN_ABS_PATHLENGTH = 1
262
503
if sys.platform == 'win32':
263
abspath = _win32_abspath
504
if win32utils.winver == 'Windows 98':
505
abspath = _win98_abspath
507
abspath = _win32_abspath
264
508
realpath = _win32_realpath
265
509
pathjoin = _win32_pathjoin
266
510
normpath = _win32_normpath
267
511
getcwd = _win32_getcwd
268
512
mkdtemp = _win32_mkdtemp
269
rename = _win32_rename
513
rename = _rename_wrap_exception(_win32_rename)
515
from bzrlib import _walkdirs_win32
519
lstat = _walkdirs_win32.lstat
520
fstat = _walkdirs_win32.fstat
521
wrap_stat = _walkdirs_win32.wrap_stat
271
523
MIN_ABS_PATHLENGTH = 3
727
1448
def normalizes_filenames():
728
1449
"""Return True if this platform normalizes unicode filenames.
730
Mac OSX does, Windows/Linux do not.
732
1453
return _platform_normalizes_filenames
1456
def _accessible_normalized_filename(path):
1457
"""Get the unicode normalized path, and if you can access the file.
1459
On platforms where the system normalizes filenames (Mac OSX),
1460
you can access a file by any path which will normalize correctly.
1461
On platforms where the system does not normalize filenames
1462
(everything else), you have to access a file by its exact path.
1464
Internally, bzr only supports NFC normalization, since that is
1465
the standard for XML documents.
1467
So return the normalized path, and a flag indicating if the file
1468
can be accessed by that path.
1471
return unicodedata.normalize('NFC', unicode(path)), True
1474
def _inaccessible_normalized_filename(path):
1475
__doc__ = _accessible_normalized_filename.__doc__
1477
normalized = unicodedata.normalize('NFC', unicode(path))
1478
return normalized, normalized == path
735
1481
if _platform_normalizes_filenames:
736
def unicode_filename(path):
737
"""Make sure 'path' is a properly normalized filename.
739
On platforms where the system normalizes filenames (Mac OSX),
740
you can access a file by any path which will normalize
742
Internally, bzr only supports NFC/NFKC normalization, since
743
that is the standard for XML documents.
744
So we return an normalized path, and indicate this has been
747
:return: (path, is_normalized) Return a path which can
748
access the file, and whether or not this path is
751
return unicodedata.normalize('NFKC', path), True
1482
normalized_filename = _accessible_normalized_filename
753
def unicode_filename(path):
754
"""Make sure 'path' is a properly normalized filename.
756
On platforms where the system does not normalize filenames
757
(Windows, Linux), you have to access a file by its exact path.
758
Internally, bzr only supports NFC/NFKC normalization, since
759
that is the standard for XML documents.
760
So we return the original path, and indicate if this is
763
:return: (path, is_normalized) Return a path which can
764
access the file, and whether or not this path is
767
return path, unicodedata.normalize('NFKC', path) == path
1484
normalized_filename = _inaccessible_normalized_filename
1487
def set_signal_handler(signum, handler, restart_syscall=True):
1488
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1489
on platforms that support that.
1491
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1492
automatically restart (by calling `signal.siginterrupt(signum,
1493
False)`). May be ignored if the feature is not available on this
1494
platform or Python version.
1498
siginterrupt = signal.siginterrupt
1500
# This python implementation doesn't provide signal support, hence no
1503
except AttributeError:
1504
# siginterrupt doesn't exist on this platform, or for this version
1506
siginterrupt = lambda signum, flag: None
1508
def sig_handler(*args):
1509
# Python resets the siginterrupt flag when a signal is
1510
# received. <http://bugs.python.org/issue8354>
1511
# As a workaround for some cases, set it back the way we want it.
1512
siginterrupt(signum, False)
1513
# Now run the handler function passed to set_signal_handler.
1516
sig_handler = handler
1517
old_handler = signal.signal(signum, sig_handler)
1519
siginterrupt(signum, False)
1523
default_terminal_width = 80
1524
"""The default terminal width for ttys.
1526
This is defined so that higher levels can share a common fallback value when
1527
terminal_width() returns None.
1530
# Keep some state so that terminal_width can detect if _terminal_size has
1531
# returned a different size since the process started. See docstring and
1532
# comments of terminal_width for details.
1533
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
1534
_terminal_size_state = 'no_data'
1535
_first_terminal_size = None
770
1537
def terminal_width():
771
"""Return estimated terminal width."""
772
if sys.platform == 'win32':
773
import bzrlib.win32console
774
return bzrlib.win32console.get_console_size()[0]
1538
"""Return terminal width.
1540
None is returned if the width can't established precisely.
1543
- if BZR_COLUMNS is set, returns its value
1544
- if there is no controlling terminal, returns None
1545
- query the OS, if the queried size has changed since the last query,
1547
- if COLUMNS is set, returns its value,
1548
- if the OS has a value (even though it's never changed), return its value.
1550
From there, we need to query the OS to get the size of the controlling
1553
On Unices we query the OS by:
1554
- get termios.TIOCGWINSZ
1555
- if an error occurs or a negative value is obtained, returns None
1557
On Windows we query the OS by:
1558
- win32utils.get_console_size() decides,
1559
- returns None on error (provided default value)
1561
# Note to implementors: if changing the rules for determining the width,
1562
# make sure you've considered the behaviour in these cases:
1563
# - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
1564
# - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
1566
# - (add more interesting cases here, if you find any)
1567
# Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
1568
# but we don't want to register a signal handler because it is impossible
1569
# to do so without risking EINTR errors in Python <= 2.6.5 (see
1570
# <http://bugs.python.org/issue8354>). Instead we check TIOCGWINSZ every
1571
# time so we can notice if the reported size has changed, which should have
1574
# If BZR_COLUMNS is set, take it, user is always right
1575
# Except if they specified 0 in which case, impose no limit here
1577
width = int(os.environ['BZR_COLUMNS'])
1578
except (KeyError, ValueError):
1580
if width is not None:
1586
isatty = getattr(sys.stdout, 'isatty', None)
1587
if isatty is None or not isatty():
1588
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1592
width, height = os_size = _terminal_size(None, None)
1593
global _first_terminal_size, _terminal_size_state
1594
if _terminal_size_state == 'no_data':
1595
_first_terminal_size = os_size
1596
_terminal_size_state = 'unchanged'
1597
elif (_terminal_size_state == 'unchanged' and
1598
_first_terminal_size != os_size):
1599
_terminal_size_state = 'changed'
1601
# If the OS claims to know how wide the terminal is, and this value has
1602
# ever changed, use that.
1603
if _terminal_size_state == 'changed':
1604
if width is not None and width > 0:
1607
# If COLUMNS is set, use it.
1609
return int(os.environ['COLUMNS'])
1610
except (KeyError, ValueError):
1613
# Finally, use an unchanged size from the OS, if we have one.
1614
if _terminal_size_state == 'unchanged':
1615
if width is not None and width > 0:
1618
# The width could not be determined.
1622
def _win32_terminal_size(width, height):
1623
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1624
return width, height
1627
def _ioctl_terminal_size(width, height):
777
1629
import struct, fcntl, termios
778
1630
s = struct.pack('HHHH', 0, 0, 0, 0)
779
1631
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
780
width = struct.unpack('HHHH', x)[1]
1632
height, width = struct.unpack('HHHH', x)[0:2]
1633
except (IOError, AttributeError):
785
width = int(os.environ['COLUMNS'])
1635
return width, height
1637
_terminal_size = None
1638
"""Returns the terminal size as (width, height).
1640
:param width: Default value for width.
1641
:param height: Default value for height.
1643
This is defined specifically for each OS and query the size of the controlling
1644
terminal. If any error occurs, the provided default values should be returned.
1646
if sys.platform == 'win32':
1647
_terminal_size = _win32_terminal_size
1649
_terminal_size = _ioctl_terminal_size
793
1652
def supports_executable():
794
1653
return sys.platform != "win32"
1656
def supports_posix_readonly():
1657
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1659
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1660
directory controls creation/deletion, etc.
1662
And under win32, readonly means that the directory itself cannot be
1663
deleted. The contents of a readonly directory can be changed, unlike POSIX
1664
where files in readonly directories cannot be added, deleted or renamed.
1666
return sys.platform != "win32"
1669
def set_or_unset_env(env_variable, value):
1670
"""Modify the environment, setting or removing the env_variable.
1672
:param env_variable: The environment variable in question
1673
:param value: The value to set the environment to. If None, then
1674
the variable will be removed.
1675
:return: The original value of the environment variable.
1677
orig_val = os.environ.get(env_variable)
1679
if orig_val is not None:
1680
del os.environ[env_variable]
1682
if isinstance(value, unicode):
1683
value = value.encode(get_user_encoding())
1684
os.environ[env_variable] = value
797
1688
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
800
1691
def check_legal_path(path):
801
"""Check whether the supplied path is legal.
1692
"""Check whether the supplied path is legal.
802
1693
This is only required on Windows, so we don't test on other platforms
805
1696
if sys.platform != "win32":
807
1698
if _validWin32PathRE.match(path) is None:
808
raise IllegalPath(path)
1699
raise errors.IllegalPath(path)
1702
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1704
def _is_error_enotdir(e):
1705
"""Check if this exception represents ENOTDIR.
1707
Unfortunately, python is very inconsistent about the exception
1708
here. The cases are:
1709
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1710
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1711
which is the windows error code.
1712
3) Windows, Python2.5 uses errno == EINVAL and
1713
winerror == ERROR_DIRECTORY
1715
:param e: An Exception object (expected to be OSError with an errno
1716
attribute, but we should be able to cope with anything)
1717
:return: True if this represents an ENOTDIR error. False otherwise.
1719
en = getattr(e, 'errno', None)
1720
if (en == errno.ENOTDIR
1721
or (sys.platform == 'win32'
1722
and (en == _WIN32_ERROR_DIRECTORY
1723
or (en == errno.EINVAL
1724
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
811
1730
def walkdirs(top, prefix=""):
812
1731
"""Yield data about all the directories in a tree.
814
1733
This yields all the data about the contents of a directory at a time.
815
1734
After each directory has been yielded, if the caller has mutated the list
816
1735
to exclude some directories, they are then not descended into.
818
1737
The data yielded is of the form:
819
[(relpath, basename, kind, lstat, path_from_top), ...]
1738
((directory-relpath, directory-path-from-top),
1739
[(relpath, basename, kind, lstat, path-from-top), ...]),
1740
- directory-relpath is the relative path of the directory being returned
1741
with respect to top. prefix is prepended to this.
1742
- directory-path-from-root is the path including top for this directory.
1743
It is suitable for use with os functions.
1744
- relpath is the relative path within the subtree being walked.
1745
- basename is the basename of the path
1746
- kind is the kind of the file now. If unknown then the file is not
1747
present within the tree - but it may be recorded as versioned. See
1749
- lstat is the stat data *if* the file was statted.
1750
- planned, not implemented:
1751
path_from_tree_root is the path from the root of the tree.
821
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1753
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
822
1754
allows one to walk a subtree but get paths that are relative to a tree
823
1755
rooted higher up.
824
1756
:return: an iterator over the dirs.
1758
#TODO there is a bit of a smell where the results of the directory-
1759
# summary in this, and the path from the root, may not agree
1760
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1761
# potentially confusing output. We should make this more robust - but
1762
# not at a speed cost. RBC 20060731
828
1764
_directory = _directory_kind
830
pending = [(prefix, "", _directory, None, top)]
1765
_listdir = os.listdir
1766
_kind_from_mode = file_kind_from_stat_mode
1767
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
833
currentdir = pending.pop()
834
1769
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
837
relroot = currentdir[0] + '/'
1770
relroot, _, _, _, top = pending.pop()
1772
relprefix = relroot + u'/'
1775
top_slash = top + u'/'
1778
append = dirblock.append
1780
names = sorted(map(decode_filename, _listdir(top)))
1782
if not _is_error_enotdir(e):
1786
abspath = top_slash + name
1787
statvalue = _lstat(abspath)
1788
kind = _kind_from_mode(statvalue.st_mode)
1789
append((relprefix + name, name, kind, statvalue, abspath))
1790
yield (relroot, top), dirblock
1792
# push the user specified dirs from dirblock
1793
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1796
class DirReader(object):
1797
"""An interface for reading directories."""
1799
def top_prefix_to_starting_dir(self, top, prefix=""):
1800
"""Converts top and prefix to a starting dir entry
1802
:param top: A utf8 path
1803
:param prefix: An optional utf8 path to prefix output relative paths
1805
:return: A tuple starting with prefix, and ending with the native
1808
raise NotImplementedError(self.top_prefix_to_starting_dir)
1810
def read_dir(self, prefix, top):
1811
"""Read a specific dir.
1813
:param prefix: A utf8 prefix to be preprended to the path basenames.
1814
:param top: A natively encoded path to read.
1815
:return: A list of the directories contents. Each item contains:
1816
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1818
raise NotImplementedError(self.read_dir)
1821
_selected_dir_reader = None
1824
def _walkdirs_utf8(top, prefix=""):
1825
"""Yield data about all the directories in a tree.
1827
This yields the same information as walkdirs() only each entry is yielded
1828
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1829
are returned as exact byte-strings.
1831
:return: yields a tuple of (dir_info, [file_info])
1832
dir_info is (utf8_relpath, path-from-top)
1833
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1834
if top is an absolute path, path-from-top is also an absolute path.
1835
path-from-top might be unicode or utf8, but it is the correct path to
1836
pass to os functions to affect the file in question. (such as os.lstat)
1838
global _selected_dir_reader
1839
if _selected_dir_reader is None:
1840
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1841
# Win98 doesn't have unicode apis like FindFirstFileW
1842
# TODO: We possibly could support Win98 by falling back to the
1843
# original FindFirstFile, and using TCHAR instead of WCHAR,
1844
# but that gets a bit tricky, and requires custom compiling
1847
from bzrlib._walkdirs_win32 import Win32ReadDir
1848
_selected_dir_reader = Win32ReadDir()
1851
elif _fs_enc in ('utf-8', 'ascii'):
1853
from bzrlib._readdir_pyx import UTF8DirReader
1854
_selected_dir_reader = UTF8DirReader()
1855
except ImportError, e:
1856
failed_to_load_extension(e)
1859
if _selected_dir_reader is None:
1860
# Fallback to the python version
1861
_selected_dir_reader = UnicodeDirReader()
1863
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1864
# But we don't actually uses 1-3 in pending, so set them to None
1865
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1866
read_dir = _selected_dir_reader.read_dir
1867
_directory = _directory_kind
1869
relroot, _, _, _, top = pending[-1].pop()
1872
dirblock = sorted(read_dir(relroot, top))
1873
yield (relroot, top), dirblock
1874
# push the user specified dirs from dirblock
1875
next = [d for d in reversed(dirblock) if d[2] == _directory]
1877
pending.append(next)
1880
class UnicodeDirReader(DirReader):
1881
"""A dir reader for non-utf8 file systems, which transcodes."""
1883
__slots__ = ['_utf8_encode']
1886
self._utf8_encode = codecs.getencoder('utf8')
1888
def top_prefix_to_starting_dir(self, top, prefix=""):
1889
"""See DirReader.top_prefix_to_starting_dir."""
1890
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1892
def read_dir(self, prefix, top):
1893
"""Read a single directory from a non-utf8 file system.
1895
top, and the abspath element in the output are unicode, all other paths
1896
are utf8. Local disk IO is done via unicode calls to listdir etc.
1898
This is currently the fallback code path when the filesystem encoding is
1899
not UTF-8. It may be better to implement an alternative so that we can
1900
safely handle paths that are not properly decodable in the current
1903
See DirReader.read_dir for details.
1905
_utf8_encode = self._utf8_encode
1907
_listdir = os.listdir
1908
_kind_from_mode = file_kind_from_stat_mode
1911
relprefix = prefix + '/'
1914
top_slash = top + u'/'
1917
append = dirblock.append
840
1918
for name in sorted(_listdir(top)):
841
abspath = top + '/' + name
842
statvalue = lstat(abspath)
843
dirblock.append ((relroot + name, name, file_kind_from_stat_mode(statvalue.st_mode), statvalue, abspath))
845
# push the user specified dirs from dirblock
846
for dir in reversed(dirblock):
847
if dir[2] == _directory:
1920
name_utf8 = _utf8_encode(name)[0]
1921
except UnicodeDecodeError:
1922
raise errors.BadFilenameEncoding(
1923
_utf8_encode(relprefix)[0] + name, _fs_enc)
1924
abspath = top_slash + name
1925
statvalue = _lstat(abspath)
1926
kind = _kind_from_mode(statvalue.st_mode)
1927
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1931
def copy_tree(from_path, to_path, handlers={}):
1932
"""Copy all of the entries in from_path into to_path.
1934
:param from_path: The base directory to copy.
1935
:param to_path: The target directory. If it does not exist, it will
1937
:param handlers: A dictionary of functions, which takes a source and
1938
destinations for files, directories, etc.
1939
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1940
'file', 'directory', and 'symlink' should always exist.
1941
If they are missing, they will be replaced with 'os.mkdir()',
1942
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1944
# Now, just copy the existing cached tree to the new location
1945
# We use a cheap trick here.
1946
# Absolute paths are prefixed with the first parameter
1947
# relative paths are prefixed with the second.
1948
# So we can get both the source and target returned
1949
# without any extra work.
1951
def copy_dir(source, dest):
1954
def copy_link(source, dest):
1955
"""Copy the contents of a symlink"""
1956
link_to = os.readlink(source)
1957
os.symlink(link_to, dest)
1959
real_handlers = {'file':shutil.copy2,
1960
'symlink':copy_link,
1961
'directory':copy_dir,
1963
real_handlers.update(handlers)
1965
if not os.path.exists(to_path):
1966
real_handlers['directory'](from_path, to_path)
1968
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1969
for relpath, name, kind, st, abspath in entries:
1970
real_handlers[kind](abspath, relpath)
1973
def copy_ownership_from_path(dst, src=None):
1974
"""Copy usr/grp ownership from src file/dir to dst file/dir.
1976
If src is None, the containing directory is used as source. If chown
1977
fails, the error is ignored and a warning is printed.
1979
chown = getattr(os, 'chown', None)
1984
src = os.path.dirname(dst)
1990
chown(dst, s.st_uid, s.st_gid)
1993
'Unable to copy ownership from "%s" to "%s". '
1994
'You may want to set it manually.', src, dst)
1995
trace.log_exception_quietly()
1998
def path_prefix_key(path):
1999
"""Generate a prefix-order path key for path.
2001
This can be used to sort paths in the same way that walkdirs does.
2003
return (dirname(path) , path)
2006
def compare_paths_prefix_order(path_a, path_b):
2007
"""Compare path_a and path_b to generate the same order walkdirs uses."""
2008
key_a = path_prefix_key(path_a)
2009
key_b = path_prefix_key(path_b)
2010
return cmp(key_a, key_b)
2013
_cached_user_encoding = None
2016
def get_user_encoding(use_cache=DEPRECATED_PARAMETER):
2017
"""Find out what the preferred user encoding is.
2019
This is generally the encoding that is used for command line parameters
2020
and file contents. This may be different from the terminal encoding
2021
or the filesystem encoding.
2023
:return: A string defining the preferred user encoding
2025
global _cached_user_encoding
2026
if deprecated_passed(use_cache):
2027
warn_deprecated("use_cache should only have been used for tests",
2028
DeprecationWarning, stacklevel=2)
2029
if _cached_user_encoding is not None:
2030
return _cached_user_encoding
2032
if os.name == 'posix' and getattr(locale, 'CODESET', None) is not None:
2033
# Use the existing locale settings and call nl_langinfo directly
2034
# rather than going through getpreferredencoding. This avoids
2035
# <http://bugs.python.org/issue6202> on OSX Python 2.6 and the
2036
# possibility of the setlocale call throwing an error.
2037
user_encoding = locale.nl_langinfo(locale.CODESET)
2039
# GZ 2011-12-19: On windows could call GetACP directly instead.
2040
user_encoding = locale.getpreferredencoding(False)
2043
user_encoding = codecs.lookup(user_encoding).name
2045
if user_encoding not in ("", "cp0"):
2046
sys.stderr.write('bzr: warning:'
2047
' unknown encoding %s.'
2048
' Continuing with ascii encoding.\n'
2051
user_encoding = 'ascii'
2053
# Get 'ascii' when setlocale has not been called or LANG=C or unset.
2054
if user_encoding == 'ascii':
2055
if sys.platform == 'darwin':
2056
# OSX is special-cased in Python to have a UTF-8 filesystem
2057
# encoding and previously had LANG set here if not present.
2058
user_encoding = 'utf-8'
2059
# GZ 2011-12-19: Maybe UTF-8 should be the default in this case
2060
# for some other posix platforms as well.
2062
_cached_user_encoding = user_encoding
2063
return user_encoding
2066
def get_diff_header_encoding():
2067
return get_terminal_encoding()
2070
def get_host_name():
2071
"""Return the current unicode host name.
2073
This is meant to be used in place of socket.gethostname() because that
2074
behaves inconsistently on different platforms.
2076
if sys.platform == "win32":
2077
return win32utils.get_host_name()
2080
return socket.gethostname().decode(get_user_encoding())
2083
# We must not read/write any more than 64k at a time from/to a socket so we
2084
# don't risk "no buffer space available" errors on some platforms. Windows in
2085
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
2087
MAX_SOCKET_CHUNK = 64 * 1024
2089
_end_of_stream_errors = [errno.ECONNRESET, errno.EPIPE, errno.EINVAL]
2090
for _eno in ['WSAECONNRESET', 'WSAECONNABORTED']:
2091
_eno = getattr(errno, _eno, None)
2092
if _eno is not None:
2093
_end_of_stream_errors.append(_eno)
2097
def read_bytes_from_socket(sock, report_activity=None,
2098
max_read_size=MAX_SOCKET_CHUNK):
2099
"""Read up to max_read_size of bytes from sock and notify of progress.
2101
Translates "Connection reset by peer" into file-like EOF (return an
2102
empty string rather than raise an error), and repeats the recv if
2103
interrupted by a signal.
2107
bytes = sock.recv(max_read_size)
2108
except socket.error, e:
2110
if eno in _end_of_stream_errors:
2111
# The connection was closed by the other side. Callers expect
2112
# an empty string to signal end-of-stream.
2114
elif eno == errno.EINTR:
2115
# Retry the interrupted recv.
2119
if report_activity is not None:
2120
report_activity(len(bytes), 'read')
2124
def recv_all(socket, count):
2125
"""Receive an exact number of bytes.
2127
Regular Socket.recv() may return less than the requested number of bytes,
2128
depending on what's in the OS buffer. MSG_WAITALL is not available
2129
on all platforms, but this should work everywhere. This will return
2130
less than the requested amount if the remote end closes.
2132
This isn't optimized and is intended mostly for use in testing.
2135
while len(b) < count:
2136
new = read_bytes_from_socket(socket, None, count - len(b))
2143
def send_all(sock, bytes, report_activity=None):
2144
"""Send all bytes on a socket.
2146
Breaks large blocks in smaller chunks to avoid buffering limitations on
2147
some platforms, and catches EINTR which may be thrown if the send is
2148
interrupted by a signal.
2150
This is preferred to socket.sendall(), because it avoids portability bugs
2151
and provides activity reporting.
2153
:param report_activity: Call this as bytes are read, see
2154
Transport._report_activity
2157
byte_count = len(bytes)
2158
while sent_total < byte_count:
2160
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2161
except (socket.error, IOError), e:
2162
if e.args[0] in _end_of_stream_errors:
2163
raise errors.ConnectionReset(
2164
"Error trying to write to socket", e)
2165
if e.args[0] != errno.EINTR:
2169
raise errors.ConnectionReset('Sending to %s returned 0 bytes'
2172
if report_activity is not None:
2173
report_activity(sent, 'write')
2176
def connect_socket(address):
2177
# Slight variation of the socket.create_connection() function (provided by
2178
# python-2.6) that can fail if getaddrinfo returns an empty list. We also
2179
# provide it for previous python versions. Also, we don't use the timeout
2180
# parameter (provided by the python implementation) so we don't implement
2182
err = socket.error('getaddrinfo returns an empty list')
2183
host, port = address
2184
for res in socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM):
2185
af, socktype, proto, canonname, sa = res
2188
sock = socket.socket(af, socktype, proto)
2192
except socket.error, err:
2193
# 'err' is now the most recent error
2194
if sock is not None:
2199
def dereference_path(path):
2200
"""Determine the real path to a file.
2202
All parent elements are dereferenced. But the file itself is not
2204
:param path: The original path. May be absolute or relative.
2205
:return: the real path *to* the file
2207
parent, base = os.path.split(path)
2208
# The pathjoin for '.' is a workaround for Python bug #1213894.
2209
# (initial path components aren't dereferenced)
2210
return pathjoin(realpath(pathjoin('.', parent)), base)
2213
def supports_mapi():
2214
"""Return True if we can use MAPI to launch a mail client."""
2215
return sys.platform == "win32"
2218
def resource_string(package, resource_name):
2219
"""Load a resource from a package and return it as a string.
2221
Note: Only packages that start with bzrlib are currently supported.
2223
This is designed to be a lightweight implementation of resource
2224
loading in a way which is API compatible with the same API from
2226
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2227
If and when pkg_resources becomes a standard library, this routine
2230
# Check package name is within bzrlib
2231
if package == "bzrlib":
2232
resource_relpath = resource_name
2233
elif package.startswith("bzrlib."):
2234
package = package[len("bzrlib."):].replace('.', os.sep)
2235
resource_relpath = pathjoin(package, resource_name)
2237
raise errors.BzrError('resource package %s not in bzrlib' % package)
2239
# Map the resource to a file and read its contents
2240
base = dirname(bzrlib.__file__)
2241
if getattr(sys, 'frozen', None): # bzr.exe
2242
base = abspath(pathjoin(base, '..', '..'))
2243
f = file(pathjoin(base, resource_relpath), "rU")
2249
def file_kind_from_stat_mode_thunk(mode):
2250
global file_kind_from_stat_mode
2251
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2253
from bzrlib._readdir_pyx import UTF8DirReader
2254
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2255
except ImportError, e:
2256
# This is one time where we won't warn that an extension failed to
2257
# load. The extension is never available on Windows anyway.
2258
from bzrlib._readdir_py import (
2259
_kind_from_mode as file_kind_from_stat_mode
2261
return file_kind_from_stat_mode(mode)
2262
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2264
def file_stat(f, _lstat=os.lstat):
2269
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2270
raise errors.NoSuchFile(f)
2273
def file_kind(f, _lstat=os.lstat):
2274
stat_value = file_stat(f, _lstat)
2275
return file_kind_from_stat_mode(stat_value.st_mode)
2277
def until_no_eintr(f, *a, **kw):
2278
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2280
WARNING: you must be certain that it is safe to retry the call repeatedly
2281
if EINTR does occur. This is typically only true for low-level operations
2282
like os.read. If in any doubt, don't use this.
2284
Keep in mind that this is not a complete solution to EINTR. There is
2285
probably code in the Python standard library and other dependencies that
2286
may encounter EINTR if a signal arrives (and there is signal handler for
2287
that signal). So this function can reduce the impact for IO that bzrlib
2288
directly controls, but it is not a complete solution.
2290
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2294
except (IOError, OSError), e:
2295
if e.errno == errno.EINTR:
2300
@deprecated_function(deprecated_in((2, 2, 0)))
2301
def re_compile_checked(re_string, flags=0, where=""):
2302
"""Return a compiled re, or raise a sensible error.
2304
This should only be used when compiling user-supplied REs.
2306
:param re_string: Text form of regular expression.
2307
:param flags: eg re.IGNORECASE
2308
:param where: Message explaining to the user the context where
2309
it occurred, eg 'log search filter'.
2311
# from https://bugs.launchpad.net/bzr/+bug/251352
2313
re_obj = re.compile(re_string, flags)
2316
except errors.InvalidPattern, e:
2318
where = ' in ' + where
2319
# despite the name 'error' is a type
2320
raise errors.BzrCommandError('Invalid regular expression%s: %s'
2324
if sys.platform == "win32":
2327
return msvcrt.getch()
2332
fd = sys.stdin.fileno()
2333
settings = termios.tcgetattr(fd)
2336
ch = sys.stdin.read(1)
2338
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2341
if sys.platform.startswith('linux'):
2342
def _local_concurrency():
2344
return os.sysconf('SC_NPROCESSORS_ONLN')
2345
except (ValueError, OSError, AttributeError):
2347
elif sys.platform == 'darwin':
2348
def _local_concurrency():
2349
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2350
stdout=subprocess.PIPE).communicate()[0]
2351
elif "bsd" in sys.platform:
2352
def _local_concurrency():
2353
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2354
stdout=subprocess.PIPE).communicate()[0]
2355
elif sys.platform == 'sunos5':
2356
def _local_concurrency():
2357
return subprocess.Popen(['psrinfo', '-p',],
2358
stdout=subprocess.PIPE).communicate()[0]
2359
elif sys.platform == "win32":
2360
def _local_concurrency():
2361
# This appears to return the number of cores.
2362
return os.environ.get('NUMBER_OF_PROCESSORS')
2364
def _local_concurrency():
2369
_cached_local_concurrency = None
2371
def local_concurrency(use_cache=True):
2372
"""Return how many processes can be run concurrently.
2374
Rely on platform specific implementations and default to 1 (one) if
2375
anything goes wrong.
2377
global _cached_local_concurrency
2379
if _cached_local_concurrency is not None and use_cache:
2380
return _cached_local_concurrency
2382
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2383
if concurrency is None:
2385
import multiprocessing
2386
concurrency = multiprocessing.cpu_count()
2387
except (ImportError, NotImplementedError):
2388
# multiprocessing is only available on Python >= 2.6
2389
# and multiprocessing.cpu_count() isn't implemented on all
2392
concurrency = _local_concurrency()
2393
except (OSError, IOError):
2396
concurrency = int(concurrency)
2397
except (TypeError, ValueError):
2400
_cached_concurrency = concurrency
2404
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2405
"""A stream writer that doesn't decode str arguments."""
2407
def __init__(self, encode, stream, errors='strict'):
2408
codecs.StreamWriter.__init__(self, stream, errors)
2409
self.encode = encode
2411
def write(self, object):
2412
if type(object) is str:
2413
self.stream.write(object)
2415
data, _ = self.encode(object, self.errors)
2416
self.stream.write(data)
2418
if sys.platform == 'win32':
2419
def open_file(filename, mode='r', bufsize=-1):
2420
"""This function is used to override the ``open`` builtin.
2422
But it uses O_NOINHERIT flag so the file handle is not inherited by
2423
child processes. Deleting or renaming a closed file opened with this
2424
function is not blocking child processes.
2426
writing = 'w' in mode
2427
appending = 'a' in mode
2428
updating = '+' in mode
2429
binary = 'b' in mode
2432
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2433
# for flags for each modes.
2443
flags |= os.O_WRONLY
2444
flags |= os.O_CREAT | os.O_TRUNC
2449
flags |= os.O_WRONLY
2450
flags |= os.O_CREAT | os.O_APPEND
2455
flags |= os.O_RDONLY
2457
return os.fdopen(os.open(filename, flags), mode, bufsize)
2462
def available_backup_name(base, exists):
2463
"""Find a non-existing backup file name.
2465
This will *not* create anything, this only return a 'free' entry. This
2466
should be used for checking names in a directory below a locked
2467
tree/branch/repo to avoid race conditions. This is LBYL (Look Before You
2468
Leap) and generally discouraged.
2470
:param base: The base name.
2472
:param exists: A callable returning True if the path parameter exists.
2475
name = "%s.~%d~" % (base, counter)
2478
name = "%s.~%d~" % (base, counter)
2482
def set_fd_cloexec(fd):
2483
"""Set a Unix file descriptor's FD_CLOEXEC flag. Do nothing if platform
2484
support for this is not available.
2488
old = fcntl.fcntl(fd, fcntl.F_GETFD)
2489
fcntl.fcntl(fd, fcntl.F_SETFD, old | fcntl.FD_CLOEXEC)
2490
except (ImportError, AttributeError):
2491
# Either the fcntl module or specific constants are not present
2495
def find_executable_on_path(name):
2496
"""Finds an executable on the PATH.
2498
On Windows, this will try to append each extension in the PATHEXT
2499
environment variable to the name, if it cannot be found with the name
2502
:param name: The base name of the executable.
2503
:return: The path to the executable found or None.
2505
if sys.platform == 'win32':
2506
exts = os.environ.get('PATHEXT', '').split(os.pathsep)
2507
exts = [ext.lower() for ext in exts]
2508
base, ext = os.path.splitext(name)
2510
if ext.lower() not in exts:
2516
path = os.environ.get('PATH')
2517
if path is not None:
2518
path = path.split(os.pathsep)
2521
f = os.path.join(d, name) + ext
2522
if os.access(f, os.X_OK):
2524
if sys.platform == 'win32':
2525
app_path = win32utils.get_app_path(name)
2526
if app_path != name:
2531
def _posix_is_local_pid_dead(pid):
2532
"""True if pid doesn't correspond to live process on this machine"""
2534
# Special meaning of unix kill: just check if it's there.
2537
if e.errno == errno.ESRCH:
2538
# On this machine, and really not found: as sure as we can be
2541
elif e.errno == errno.EPERM:
2542
# exists, though not ours
2545
mutter("os.kill(%d, 0) failed: %s" % (pid, e))
2546
# Don't really know.
2549
# Exists and our process: not dead.
2552
if sys.platform == "win32":
2553
is_local_pid_dead = win32utils.is_local_pid_dead
2555
is_local_pid_dead = _posix_is_local_pid_dead
2558
def fdatasync(fileno):
2559
"""Flush file contents to disk if possible.
2561
:param fileno: Integer OS file handle.
2562
:raises TransportNotPossible: If flushing to disk is not possible.
2564
fn = getattr(os, 'fdatasync', getattr(os, 'fsync', None))
2569
def ensure_empty_directory_exists(path, exception_class):
2570
"""Make sure a local directory exists and is empty.
2572
If it does not exist, it is created. If it exists and is not empty, an
2573
instance of exception_class is raised.
2578
if e.errno != errno.EEXIST:
2580
if os.listdir(path) != []:
2581
raise exception_class(path)
2584
def is_environment_error(evalue):
2585
"""True if exception instance is due to a process environment issue
2587
This includes OSError and IOError, but also other errors that come from
2588
the operating system or core libraries but are not subclasses of those.
2590
if isinstance(evalue, (EnvironmentError, select.error)):
2592
if sys.platform == "win32" and win32utils._is_pywintypes_error(evalue):
added
removed
bzrlib/osutils.py
