1
# Copyright (C) 2005-2011 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25
from bzrlib.lazy_import import lazy_import
26
lazy_import(globals(), """
27
from datetime import datetime
32
# We need to import both shutil and rmtree as we export the later on posix
33
# and need the former on windows
35
from shutil import rmtree
38
# We need to import both tempfile and mkdtemp as we export the later on posix
39
# and need the former on windows
41
from tempfile import mkdtemp
51
from bzrlib.i18n import gettext
54
from bzrlib.symbol_versioning import (
66
from bzrlib import symbol_versioning, _fs_enc
69
# Cross platform wall-clock time functionality with decent resolution.
70
# On Linux ``time.clock`` returns only CPU time. On Windows, ``time.time()``
71
# only has a resolution of ~15ms. Note that ``time.clock()`` is not
72
# synchronized with ``time.time()``, this is only meant to be used to find
73
# delta times by subtracting from another call to this function.
74
timer_func = time.time
75
if sys.platform == 'win32':
76
timer_func = time.clock
78
# On win32, O_BINARY is used to indicate the file should
79
# be opened in binary mode, rather than text mode.
80
# On other platforms, O_BINARY doesn't exist, because
81
# they always open in binary mode, so it is okay to
82
# OR with 0 on those platforms.
83
# O_NOINHERIT and O_TEXT exists only on win32 too.
84
O_BINARY = getattr(os, 'O_BINARY', 0)
85
O_TEXT = getattr(os, 'O_TEXT', 0)
86
O_NOINHERIT = getattr(os, 'O_NOINHERIT', 0)
89
def get_unicode_argv():
91
user_encoding = get_user_encoding()
92
return [a.decode(user_encoding) for a in sys.argv[1:]]
93
except UnicodeDecodeError:
94
raise errors.BzrError(gettext("Parameter {0!r} encoding is unsupported by {1} "
95
"application locale.").format(a, user_encoding))
98
def make_readonly(filename):
99
"""Make a filename read-only."""
100
mod = os.lstat(filename).st_mode
101
if not stat.S_ISLNK(mod):
103
chmod_if_possible(filename, mod)
106
def make_writable(filename):
107
mod = os.lstat(filename).st_mode
108
if not stat.S_ISLNK(mod):
110
chmod_if_possible(filename, mod)
113
def chmod_if_possible(filename, mode):
114
# Set file mode if that can be safely done.
115
# Sometimes even on unix the filesystem won't allow it - see
116
# https://bugs.launchpad.net/bzr/+bug/606537
118
# It is probably faster to just do the chmod, rather than
119
# doing a stat, and then trying to compare
120
os.chmod(filename, mode)
121
except (IOError, OSError),e:
122
# Permission/access denied seems to commonly happen on smbfs; there's
123
# probably no point warning about it.
124
# <https://bugs.launchpad.net/bzr/+bug/606537>
125
if getattr(e, 'errno') in (errno.EPERM, errno.EACCES):
126
trace.mutter("ignore error on chmod of %r: %r" % (
132
def minimum_path_selection(paths):
133
"""Return the smallset subset of paths which are outside paths.
135
:param paths: A container (and hence not None) of paths.
136
:return: A set of paths sufficient to include everything in paths via
137
is_inside, drawn from the paths parameter.
143
return path.split('/')
144
sorted_paths = sorted(list(paths), key=sort_key)
146
search_paths = [sorted_paths[0]]
147
for path in sorted_paths[1:]:
148
if not is_inside(search_paths[-1], path):
149
# This path is unique, add it
150
search_paths.append(path)
152
return set(search_paths)
159
"""Return a quoted filename filename
161
This previously used backslash quoting, but that works poorly on
163
# TODO: I'm not really sure this is the best format either.x
165
if _QUOTE_RE is None:
166
_QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/\\_~-])')
168
if _QUOTE_RE.search(f):
174
_directory_kind = 'directory'
177
"""Return the current umask"""
178
# Assume that people aren't messing with the umask while running
179
# XXX: This is not thread safe, but there is no way to get the
180
# umask without setting it
188
_directory_kind: "/",
190
'tree-reference': '+',
194
def kind_marker(kind):
196
return _kind_marker_map[kind]
198
# Slightly faster than using .get(, '') when the common case is that
203
lexists = getattr(os.path, 'lexists', None)
207
stat = getattr(os, 'lstat', os.stat)
211
if e.errno == errno.ENOENT:
214
raise errors.BzrError(gettext("lstat/stat of ({0!r}): {1!r}").format(f, e))
217
def fancy_rename(old, new, rename_func, unlink_func):
218
"""A fancy rename, when you don't have atomic rename.
220
:param old: The old path, to rename from
221
:param new: The new path, to rename to
222
:param rename_func: The potentially non-atomic rename function
223
:param unlink_func: A way to delete the target file if the full rename
226
# sftp rename doesn't allow overwriting, so play tricks:
227
base = os.path.basename(new)
228
dirname = os.path.dirname(new)
229
# callers use different encodings for the paths so the following MUST
230
# respect that. We rely on python upcasting to unicode if new is unicode
231
# and keeping a str if not.
232
tmp_name = 'tmp.%s.%.9f.%d.%s' % (base, time.time(),
233
os.getpid(), rand_chars(10))
234
tmp_name = pathjoin(dirname, tmp_name)
236
# Rename the file out of the way, but keep track if it didn't exist
237
# We don't want to grab just any exception
238
# something like EACCES should prevent us from continuing
239
# The downside is that the rename_func has to throw an exception
240
# with an errno = ENOENT, or NoSuchFile
243
rename_func(new, tmp_name)
244
except (errors.NoSuchFile,), e:
247
# RBC 20060103 abstraction leakage: the paramiko SFTP clients rename
248
# function raises an IOError with errno is None when a rename fails.
249
# This then gets caught here.
250
if e.errno not in (None, errno.ENOENT, errno.ENOTDIR):
253
if (getattr(e, 'errno', None) is None
254
or e.errno not in (errno.ENOENT, errno.ENOTDIR)):
263
# This may throw an exception, in which case success will
265
rename_func(old, new)
267
except (IOError, OSError), e:
268
# source and target may be aliases of each other (e.g. on a
269
# case-insensitive filesystem), so we may have accidentally renamed
270
# source by when we tried to rename target
271
failure_exc = sys.exc_info()
272
if (file_existed and e.errno in (None, errno.ENOENT)
273
and old.lower() == new.lower()):
274
# source and target are the same file on a case-insensitive
275
# filesystem, so we don't generate an exception
279
# If the file used to exist, rename it back into place
280
# otherwise just delete it from the tmp location
282
unlink_func(tmp_name)
284
rename_func(tmp_name, new)
285
if failure_exc is not None:
287
raise failure_exc[0], failure_exc[1], failure_exc[2]
292
# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
293
# choke on a Unicode string containing a relative path if
294
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
296
def _posix_abspath(path):
297
# jam 20060426 rather than encoding to fsencoding
298
# copy posixpath.abspath, but use os.getcwdu instead
299
if not posixpath.isabs(path):
300
path = posixpath.join(getcwd(), path)
301
return _posix_normpath(path)
304
def _posix_realpath(path):
305
return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)
308
def _posix_normpath(path):
309
path = posixpath.normpath(path)
310
# Bug 861008: posixpath.normpath() returns a path normalized according to
311
# the POSIX standard, which stipulates (for compatibility reasons) that two
312
# leading slashes must not be simplified to one, and only if there are 3 or
313
# more should they be simplified as one. So we treat the leading 2 slashes
314
# as a special case here by simply removing the first slash, as we consider
315
# that breaking POSIX compatibility for this obscure feature is acceptable.
316
# This is not a paranoid precaution, as we notably get paths like this when
317
# the repo is hosted at the root of the filesystem, i.e. in "/".
318
if path.startswith('//'):
323
def _posix_path_from_environ(key):
324
"""Get unicode path from `key` in environment or None if not present
326
Note that posix systems use arbitrary byte strings for filesystem objects,
327
so a path that raises BadFilenameEncoding here may still be accessible.
329
val = os.environ.get(key, None)
333
return val.decode(_fs_enc)
334
except UnicodeDecodeError:
335
# GZ 2011-12-12:Ideally want to include `key` in the exception message
336
raise errors.BadFilenameEncoding(val, _fs_enc)
339
def _posix_getuser_unicode():
340
"""Get username from environment or password database as unicode"""
341
name = getpass.getuser()
342
user_encoding = get_user_encoding()
344
return name.decode(user_encoding)
345
except UnicodeDecodeError:
346
raise errors.BzrError("Encoding of username %r is unsupported by %s "
347
"application locale." % (name, user_encoding))
350
def _win32_fixdrive(path):
351
"""Force drive letters to be consistent.
353
win32 is inconsistent whether it returns lower or upper case
354
and even if it was consistent the user might type the other
355
so we force it to uppercase
356
running python.exe under cmd.exe return capital C:\\
357
running win32 python inside a cygwin shell returns lowercase c:\\
359
drive, path = ntpath.splitdrive(path)
360
return drive.upper() + path
363
def _win32_abspath(path):
364
# Real ntpath.abspath doesn't have a problem with a unicode cwd
365
return _win32_fixdrive(ntpath.abspath(unicode(path)).replace('\\', '/'))
368
def _win98_abspath(path):
369
"""Return the absolute version of a path.
370
Windows 98 safe implementation (python reimplementation
371
of Win32 API function GetFullPathNameW)
376
# \\HOST\path => //HOST/path
377
# //HOST/path => //HOST/path
378
# path => C:/cwd/path
381
# check for absolute path
382
drive = ntpath.splitdrive(path)[0]
383
if drive == '' and path[:2] not in('//','\\\\'):
385
# we cannot simply os.path.join cwd and path
386
# because os.path.join('C:','/path') produce '/path'
387
# and this is incorrect
388
if path[:1] in ('/','\\'):
389
cwd = ntpath.splitdrive(cwd)[0]
391
path = cwd + '\\' + path
392
return _win32_fixdrive(ntpath.normpath(path).replace('\\', '/'))
395
def _win32_realpath(path):
396
# Real ntpath.realpath doesn't have a problem with a unicode cwd
397
return _win32_fixdrive(ntpath.realpath(unicode(path)).replace('\\', '/'))
400
def _win32_pathjoin(*args):
401
return ntpath.join(*args).replace('\\', '/')
404
def _win32_normpath(path):
405
return _win32_fixdrive(ntpath.normpath(unicode(path)).replace('\\', '/'))
409
return _win32_fixdrive(os.getcwdu().replace('\\', '/'))
412
def _win32_mkdtemp(*args, **kwargs):
413
return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))
416
def _win32_rename(old, new):
417
"""We expect to be able to atomically replace 'new' with old.
419
On win32, if new exists, it must be moved out of the way first,
423
fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
425
if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
426
# If we try to rename a non-existant file onto cwd, we get
427
# EPERM or EACCES instead of ENOENT, this will raise ENOENT
428
# if the old path doesn't exist, sometimes we get EACCES
429
# On Linux, we seem to get EBUSY, on Mac we get EINVAL
435
return unicodedata.normalize('NFC', os.getcwdu())
438
# Default is to just use the python builtins, but these can be rebound on
439
# particular platforms.
440
abspath = _posix_abspath
441
realpath = _posix_realpath
442
pathjoin = os.path.join
443
normpath = _posix_normpath
444
path_from_environ = _posix_path_from_environ
445
getuser_unicode = _posix_getuser_unicode
448
dirname = os.path.dirname
449
basename = os.path.basename
450
split = os.path.split
451
splitext = os.path.splitext
452
# These were already lazily imported into local scope
453
# mkdtemp = tempfile.mkdtemp
454
# rmtree = shutil.rmtree
462
MIN_ABS_PATHLENGTH = 1
465
if sys.platform == 'win32':
466
if win32utils.winver == 'Windows 98':
467
abspath = _win98_abspath
469
abspath = _win32_abspath
470
realpath = _win32_realpath
471
pathjoin = _win32_pathjoin
472
normpath = _win32_normpath
473
getcwd = _win32_getcwd
474
mkdtemp = _win32_mkdtemp
475
rename = _win32_rename
477
from bzrlib import _walkdirs_win32
481
lstat = _walkdirs_win32.lstat
482
fstat = _walkdirs_win32.fstat
483
wrap_stat = _walkdirs_win32.wrap_stat
485
MIN_ABS_PATHLENGTH = 3
487
def _win32_delete_readonly(function, path, excinfo):
488
"""Error handler for shutil.rmtree function [for win32]
489
Helps to remove files and dirs marked as read-only.
491
exception = excinfo[1]
492
if function in (os.remove, os.rmdir) \
493
and isinstance(exception, OSError) \
494
and exception.errno == errno.EACCES:
500
def rmtree(path, ignore_errors=False, onerror=_win32_delete_readonly):
501
"""Replacer for shutil.rmtree: could remove readonly dirs/files"""
502
return shutil.rmtree(path, ignore_errors, onerror)
504
f = win32utils.get_unicode_argv # special function or None
507
path_from_environ = win32utils.get_environ_unicode
508
getuser_unicode = win32utils.get_user_name
510
elif sys.platform == 'darwin':
514
def get_terminal_encoding(trace=False):
515
"""Find the best encoding for printing to the screen.
517
This attempts to check both sys.stdout and sys.stdin to see
518
what encoding they are in, and if that fails it falls back to
519
osutils.get_user_encoding().
520
The problem is that on Windows, locale.getpreferredencoding()
521
is not the same encoding as that used by the console:
522
http://mail.python.org/pipermail/python-list/2003-May/162357.html
524
On my standard US Windows XP, the preferred encoding is
525
cp1252, but the console is cp437
527
:param trace: If True trace the selected encoding via mutter().
529
from bzrlib.trace import mutter
530
output_encoding = getattr(sys.stdout, 'encoding', None)
531
if not output_encoding:
532
input_encoding = getattr(sys.stdin, 'encoding', None)
533
if not input_encoding:
534
output_encoding = get_user_encoding()
536
mutter('encoding stdout as osutils.get_user_encoding() %r',
539
output_encoding = input_encoding
541
mutter('encoding stdout as sys.stdin encoding %r',
545
mutter('encoding stdout as sys.stdout encoding %r', output_encoding)
546
if output_encoding == 'cp0':
547
# invalid encoding (cp0 means 'no codepage' on Windows)
548
output_encoding = get_user_encoding()
550
mutter('cp0 is invalid encoding.'
551
' encoding stdout as osutils.get_user_encoding() %r',
555
codecs.lookup(output_encoding)
557
sys.stderr.write('bzr: warning:'
558
' unknown terminal encoding %s.\n'
559
' Using encoding %s instead.\n'
560
% (output_encoding, get_user_encoding())
562
output_encoding = get_user_encoding()
564
return output_encoding
567
def normalizepath(f):
568
if getattr(os.path, 'realpath', None) is not None:
572
[p,e] = os.path.split(f)
573
if e == "" or e == "." or e == "..":
576
return pathjoin(F(p), e)
580
"""True if f is an accessible directory."""
582
return stat.S_ISDIR(os.lstat(f)[stat.ST_MODE])
588
"""True if f is a regular file."""
590
return stat.S_ISREG(os.lstat(f)[stat.ST_MODE])
595
"""True if f is a symlink."""
597
return stat.S_ISLNK(os.lstat(f)[stat.ST_MODE])
601
def is_inside(dir, fname):
602
"""True if fname is inside dir.
604
The parameters should typically be passed to osutils.normpath first, so
605
that . and .. and repeated slashes are eliminated, and the separators
606
are canonical for the platform.
608
The empty string as a dir name is taken as top-of-tree and matches
611
# XXX: Most callers of this can actually do something smarter by
612
# looking at the inventory
622
return fname.startswith(dir)
625
def is_inside_any(dir_list, fname):
626
"""True if fname is inside any of given dirs."""
627
for dirname in dir_list:
628
if is_inside(dirname, fname):
633
def is_inside_or_parent_of_any(dir_list, fname):
634
"""True if fname is a child or a parent of any of the given files."""
635
for dirname in dir_list:
636
if is_inside(dirname, fname) or is_inside(fname, dirname):
641
def pumpfile(from_file, to_file, read_length=-1, buff_size=32768,
642
report_activity=None, direction='read'):
643
"""Copy contents of one file to another.
645
The read_length can either be -1 to read to end-of-file (EOF) or
646
it can specify the maximum number of bytes to read.
648
The buff_size represents the maximum size for each read operation
649
performed on from_file.
651
:param report_activity: Call this as bytes are read, see
652
Transport._report_activity
653
:param direction: Will be passed to report_activity
655
:return: The number of bytes copied.
659
# read specified number of bytes
661
while read_length > 0:
662
num_bytes_to_read = min(read_length, buff_size)
664
block = from_file.read(num_bytes_to_read)
668
if report_activity is not None:
669
report_activity(len(block), direction)
672
actual_bytes_read = len(block)
673
read_length -= actual_bytes_read
674
length += actual_bytes_read
678
block = from_file.read(buff_size)
682
if report_activity is not None:
683
report_activity(len(block), direction)
689
def pump_string_file(bytes, file_handle, segment_size=None):
690
"""Write bytes to file_handle in many smaller writes.
692
:param bytes: The string to write.
693
:param file_handle: The file to write to.
695
# Write data in chunks rather than all at once, because very large
696
# writes fail on some platforms (e.g. Windows with SMB mounted
699
segment_size = 5242880 # 5MB
700
segments = range(len(bytes) / segment_size + 1)
701
write = file_handle.write
702
for segment_index in segments:
703
segment = buffer(bytes, segment_index * segment_size, segment_size)
707
def file_iterator(input_file, readsize=32768):
709
b = input_file.read(readsize)
716
"""Calculate the hexdigest of an open file.
718
The file cursor should be already at the start.
730
def size_sha_file(f):
731
"""Calculate the size and hexdigest of an open file.
733
The file cursor should be already at the start and
734
the caller is responsible for closing the file afterwards.
745
return size, s.hexdigest()
748
def sha_file_by_name(fname):
749
"""Calculate the SHA1 of a file by reading the full text"""
751
f = os.open(fname, os.O_RDONLY | O_BINARY | O_NOINHERIT)
754
b = os.read(f, 1<<16)
762
def sha_strings(strings, _factory=sha):
763
"""Return the sha-1 of concatenation of strings"""
765
map(s.update, strings)
769
def sha_string(f, _factory=sha):
770
return _factory(f).hexdigest()
773
def fingerprint_file(f):
775
return {'size': len(b),
776
'sha1': sha(b).hexdigest()}
779
def compare_files(a, b):
780
"""Returns true if equal in contents"""
791
def local_time_offset(t=None):
792
"""Return offset of local zone from GMT, either at present or at time t."""
795
offset = datetime.fromtimestamp(t) - datetime.utcfromtimestamp(t)
796
return offset.days * 86400 + offset.seconds
798
weekdays = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun']
799
_default_format_by_weekday_num = [wd + " %Y-%m-%d %H:%M:%S" for wd in weekdays]
802
def format_date(t, offset=0, timezone='original', date_fmt=None,
804
"""Return a formatted date string.
806
:param t: Seconds since the epoch.
807
:param offset: Timezone offset in seconds east of utc.
808
:param timezone: How to display the time: 'utc', 'original' for the
809
timezone specified by offset, or 'local' for the process's current
811
:param date_fmt: strftime format.
812
:param show_offset: Whether to append the timezone.
814
(date_fmt, tt, offset_str) = \
815
_format_date(t, offset, timezone, date_fmt, show_offset)
816
date_fmt = date_fmt.replace('%a', weekdays[tt[6]])
817
date_str = time.strftime(date_fmt, tt)
818
return date_str + offset_str
821
# Cache of formatted offset strings
825
def format_date_with_offset_in_original_timezone(t, offset=0,
826
_cache=_offset_cache):
827
"""Return a formatted date string in the original timezone.
829
This routine may be faster then format_date.
831
:param t: Seconds since the epoch.
832
:param offset: Timezone offset in seconds east of utc.
836
tt = time.gmtime(t + offset)
837
date_fmt = _default_format_by_weekday_num[tt[6]]
838
date_str = time.strftime(date_fmt, tt)
839
offset_str = _cache.get(offset, None)
840
if offset_str is None:
841
offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
842
_cache[offset] = offset_str
843
return date_str + offset_str
846
def format_local_date(t, offset=0, timezone='original', date_fmt=None,
848
"""Return an unicode date string formatted according to the current locale.
850
:param t: Seconds since the epoch.
851
:param offset: Timezone offset in seconds east of utc.
852
:param timezone: How to display the time: 'utc', 'original' for the
853
timezone specified by offset, or 'local' for the process's current
855
:param date_fmt: strftime format.
856
:param show_offset: Whether to append the timezone.
858
(date_fmt, tt, offset_str) = \
859
_format_date(t, offset, timezone, date_fmt, show_offset)
860
date_str = time.strftime(date_fmt, tt)
861
if not isinstance(date_str, unicode):
862
date_str = date_str.decode(get_user_encoding(), 'replace')
863
return date_str + offset_str
866
def _format_date(t, offset, timezone, date_fmt, show_offset):
867
if timezone == 'utc':
870
elif timezone == 'original':
873
tt = time.gmtime(t + offset)
874
elif timezone == 'local':
875
tt = time.localtime(t)
876
offset = local_time_offset(t)
878
raise errors.UnsupportedTimezoneFormat(timezone)
880
date_fmt = "%a %Y-%m-%d %H:%M:%S"
882
offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
885
return (date_fmt, tt, offset_str)
888
def compact_date(when):
889
return time.strftime('%Y%m%d%H%M%S', time.gmtime(when))
892
def format_delta(delta):
893
"""Get a nice looking string for a time delta.
895
:param delta: The time difference in seconds, can be positive or negative.
896
positive indicates time in the past, negative indicates time in the
897
future. (usually time.time() - stored_time)
898
:return: String formatted to show approximate resolution
904
direction = 'in the future'
908
if seconds < 90: # print seconds up to 90 seconds
910
return '%d second %s' % (seconds, direction,)
912
return '%d seconds %s' % (seconds, direction)
914
minutes = int(seconds / 60)
915
seconds -= 60 * minutes
920
if minutes < 90: # print minutes, seconds up to 90 minutes
922
return '%d minute, %d second%s %s' % (
923
minutes, seconds, plural_seconds, direction)
925
return '%d minutes, %d second%s %s' % (
926
minutes, seconds, plural_seconds, direction)
928
hours = int(minutes / 60)
929
minutes -= 60 * hours
936
return '%d hour, %d minute%s %s' % (hours, minutes,
937
plural_minutes, direction)
938
return '%d hours, %d minute%s %s' % (hours, minutes,
939
plural_minutes, direction)
942
"""Return size of given open file."""
943
return os.fstat(f.fileno())[stat.ST_SIZE]
946
# Define rand_bytes based on platform.
948
# Python 2.4 and later have os.urandom,
949
# but it doesn't work on some arches
951
rand_bytes = os.urandom
952
except (NotImplementedError, AttributeError):
953
# If python doesn't have os.urandom, or it doesn't work,
954
# then try to first pull random data from /dev/urandom
956
rand_bytes = file('/dev/urandom', 'rb').read
957
# Otherwise, use this hack as a last resort
958
except (IOError, OSError):
959
# not well seeded, but better than nothing
964
s += chr(random.randint(0, 255))
969
ALNUM = '0123456789abcdefghijklmnopqrstuvwxyz'
971
"""Return a random string of num alphanumeric characters
973
The result only contains lowercase chars because it may be used on
974
case-insensitive filesystems.
977
for raw_byte in rand_bytes(num):
978
s += ALNUM[ord(raw_byte) % 36]
982
## TODO: We could later have path objects that remember their list
983
## decomposition (might be too tricksy though.)
986
"""Turn string into list of parts."""
987
# split on either delimiter because people might use either on
989
ps = re.split(r'[\\/]', p)
994
raise errors.BzrError(gettext("sorry, %r not allowed in path") % f)
995
elif (f == '.') or (f == ''):
1004
if (f == '..') or (f is None) or (f == ''):
1005
raise errors.BzrError(gettext("sorry, %r not allowed in path") % f)
1009
def parent_directories(filename):
1010
"""Return the list of parent directories, deepest first.
1012
For example, parent_directories("a/b/c") -> ["a/b", "a"].
1015
parts = splitpath(dirname(filename))
1017
parents.append(joinpath(parts))
1022
_extension_load_failures = []
1025
def failed_to_load_extension(exception):
1026
"""Handle failing to load a binary extension.
1028
This should be called from the ImportError block guarding the attempt to
1029
import the native extension. If this function returns, the pure-Python
1030
implementation should be loaded instead::
1033
>>> import bzrlib._fictional_extension_pyx
1034
>>> except ImportError, e:
1035
>>> bzrlib.osutils.failed_to_load_extension(e)
1036
>>> import bzrlib._fictional_extension_py
1038
# NB: This docstring is just an example, not a doctest, because doctest
1039
# currently can't cope with the use of lazy imports in this namespace --
1042
# This currently doesn't report the failure at the time it occurs, because
1043
# they tend to happen very early in startup when we can't check config
1044
# files etc, and also we want to report all failures but not spam the user
1046
exception_str = str(exception)
1047
if exception_str not in _extension_load_failures:
1048
trace.mutter("failed to load compiled extension: %s" % exception_str)
1049
_extension_load_failures.append(exception_str)
1052
def report_extension_load_failures():
1053
if not _extension_load_failures:
1055
if config.GlobalStack().get('ignore_missing_extensions'):
1057
# the warnings framework should by default show this only once
1058
from bzrlib.trace import warning
1060
"bzr: warning: some compiled extensions could not be loaded; "
1061
"see <https://answers.launchpad.net/bzr/+faq/703>")
1062
# we no longer show the specific missing extensions here, because it makes
1063
# the message too long and scary - see
1064
# https://bugs.launchpad.net/bzr/+bug/430529
1068
from bzrlib._chunks_to_lines_pyx import chunks_to_lines
1069
except ImportError, e:
1070
failed_to_load_extension(e)
1071
from bzrlib._chunks_to_lines_py import chunks_to_lines
1075
"""Split s into lines, but without removing the newline characters."""
1076
# Trivially convert a fulltext into a 'chunked' representation, and let
1077
# chunks_to_lines do the heavy lifting.
1078
if isinstance(s, str):
1079
# chunks_to_lines only supports 8-bit strings
1080
return chunks_to_lines([s])
1082
return _split_lines(s)
1085
def _split_lines(s):
1086
"""Split s into lines, but without removing the newline characters.
1088
This supports Unicode or plain string objects.
1090
lines = s.split('\n')
1091
result = [line + '\n' for line in lines[:-1]]
1093
result.append(lines[-1])
1097
def hardlinks_good():
1098
return sys.platform not in ('win32', 'cygwin', 'darwin')
1101
def link_or_copy(src, dest):
1102
"""Hardlink a file, or copy it if it can't be hardlinked."""
1103
if not hardlinks_good():
1104
shutil.copyfile(src, dest)
1108
except (OSError, IOError), e:
1109
if e.errno != errno.EXDEV:
1111
shutil.copyfile(src, dest)
1114
def delete_any(path):
1115
"""Delete a file, symlink or directory.
1117
Will delete even if readonly.
1120
_delete_file_or_dir(path)
1121
except (OSError, IOError), e:
1122
if e.errno in (errno.EPERM, errno.EACCES):
1123
# make writable and try again
1126
except (OSError, IOError):
1128
_delete_file_or_dir(path)
1133
def _delete_file_or_dir(path):
1134
# Look Before You Leap (LBYL) is appropriate here instead of Easier to Ask for
1135
# Forgiveness than Permission (EAFP) because:
1136
# - root can damage a solaris file system by using unlink,
1137
# - unlink raises different exceptions on different OSes (linux: EISDIR, win32:
1138
# EACCES, OSX: EPERM) when invoked on a directory.
1139
if isdir(path): # Takes care of symlinks
1146
if getattr(os, 'symlink', None) is not None:
1152
def has_hardlinks():
1153
if getattr(os, 'link', None) is not None:
1159
def host_os_dereferences_symlinks():
1160
return (has_symlinks()
1161
and sys.platform not in ('cygwin', 'win32'))
1164
def readlink(abspath):
1165
"""Return a string representing the path to which the symbolic link points.
1167
:param abspath: The link absolute unicode path.
1169
This his guaranteed to return the symbolic link in unicode in all python
1172
link = abspath.encode(_fs_enc)
1173
target = os.readlink(link)
1174
target = target.decode(_fs_enc)
1178
def contains_whitespace(s):
1179
"""True if there are any whitespace characters in s."""
1180
# string.whitespace can include '\xa0' in certain locales, because it is
1181
# considered "non-breaking-space" as part of ISO-8859-1. But it
1182
# 1) Isn't a breaking whitespace
1183
# 2) Isn't one of ' \t\r\n' which are characters we sometimes use as
1185
# 3) '\xa0' isn't unicode safe since it is >128.
1187
# This should *not* be a unicode set of characters in case the source
1188
# string is not a Unicode string. We can auto-up-cast the characters since
1189
# they are ascii, but we don't want to auto-up-cast the string in case it
1191
for ch in ' \t\n\r\v\f':
1198
def contains_linebreaks(s):
1199
"""True if there is any vertical whitespace in s."""
1207
def relpath(base, path):
1208
"""Return path relative to base, or raise PathNotChild exception.
1210
The path may be either an absolute path or a path relative to the
1211
current working directory.
1213
os.path.commonprefix (python2.4) has a bad bug that it works just
1214
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
1215
avoids that problem.
1217
NOTE: `base` should not have a trailing slash otherwise you'll get
1218
PathNotChild exceptions regardless of `path`.
1221
if len(base) < MIN_ABS_PATHLENGTH:
1222
# must have space for e.g. a drive letter
1223
raise ValueError(gettext('%r is too short to calculate a relative path')
1231
if len(head) <= len(base) and head != base:
1232
raise errors.PathNotChild(rp, base)
1235
head, tail = split(head)
1240
return pathjoin(*reversed(s))
1245
def _cicp_canonical_relpath(base, path):
1246
"""Return the canonical path relative to base.
1248
Like relpath, but on case-insensitive-case-preserving file-systems, this
1249
will return the relpath as stored on the file-system rather than in the
1250
case specified in the input string, for all existing portions of the path.
1252
This will cause O(N) behaviour if called for every path in a tree; if you
1253
have a number of paths to convert, you should use canonical_relpaths().
1255
# TODO: it should be possible to optimize this for Windows by using the
1256
# win32 API FindFiles function to look for the specified name - but using
1257
# os.listdir() still gives us the correct, platform agnostic semantics in
1260
rel = relpath(base, path)
1261
# '.' will have been turned into ''
1265
abs_base = abspath(base)
1267
_listdir = os.listdir
1269
# use an explicit iterator so we can easily consume the rest on early exit.
1270
bit_iter = iter(rel.split('/'))
1271
for bit in bit_iter:
1274
next_entries = _listdir(current)
1275
except OSError: # enoent, eperm, etc
1276
# We can't find this in the filesystem, so just append the
1278
current = pathjoin(current, bit, *list(bit_iter))
1280
for look in next_entries:
1281
if lbit == look.lower():
1282
current = pathjoin(current, look)
1285
# got to the end, nothing matched, so we just return the
1286
# non-existing bits as they were specified (the filename may be
1287
# the target of a move, for example).
1288
current = pathjoin(current, bit, *list(bit_iter))
1290
return current[len(abs_base):].lstrip('/')
1292
# XXX - TODO - we need better detection/integration of case-insensitive
1293
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
1294
# filesystems), for example, so could probably benefit from the same basic
1295
# support there. For now though, only Windows and OSX get that support, and
1296
# they get it for *all* file-systems!
1297
if sys.platform in ('win32', 'darwin'):
1298
canonical_relpath = _cicp_canonical_relpath
1300
canonical_relpath = relpath
1302
def canonical_relpaths(base, paths):
1303
"""Create an iterable to canonicalize a sequence of relative paths.
1305
The intent is for this implementation to use a cache, vastly speeding
1306
up multiple transformations in the same directory.
1308
# but for now, we haven't optimized...
1309
return [canonical_relpath(base, p) for p in paths]
1312
def decode_filename(filename):
1313
"""Decode the filename using the filesystem encoding
1315
If it is unicode, it is returned.
1316
Otherwise it is decoded from the the filesystem's encoding. If decoding
1317
fails, a errors.BadFilenameEncoding exception is raised.
1319
if type(filename) is unicode:
1322
return filename.decode(_fs_enc)
1323
except UnicodeDecodeError:
1324
raise errors.BadFilenameEncoding(filename, _fs_enc)
1327
def safe_unicode(unicode_or_utf8_string):
1328
"""Coerce unicode_or_utf8_string into unicode.
1330
If it is unicode, it is returned.
1331
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1332
wrapped in a BzrBadParameterNotUnicode exception.
1334
if isinstance(unicode_or_utf8_string, unicode):
1335
return unicode_or_utf8_string
1337
return unicode_or_utf8_string.decode('utf8')
1338
except UnicodeDecodeError:
1339
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1342
def safe_utf8(unicode_or_utf8_string):
1343
"""Coerce unicode_or_utf8_string to a utf8 string.
1345
If it is a str, it is returned.
1346
If it is Unicode, it is encoded into a utf-8 string.
1348
if isinstance(unicode_or_utf8_string, str):
1349
# TODO: jam 20070209 This is overkill, and probably has an impact on
1350
# performance if we are dealing with lots of apis that want a
1353
# Make sure it is a valid utf-8 string
1354
unicode_or_utf8_string.decode('utf-8')
1355
except UnicodeDecodeError:
1356
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1357
return unicode_or_utf8_string
1358
return unicode_or_utf8_string.encode('utf-8')
1361
_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
1362
' Revision id generators should be creating utf8'
1366
def safe_revision_id(unicode_or_utf8_string, warn=True):
1367
"""Revision ids should now be utf8, but at one point they were unicode.
1369
:param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
1371
:param warn: Functions that are sanitizing user data can set warn=False
1372
:return: None or a utf8 revision id.
1374
if (unicode_or_utf8_string is None
1375
or unicode_or_utf8_string.__class__ == str):
1376
return unicode_or_utf8_string
1378
symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
1380
return cache_utf8.encode(unicode_or_utf8_string)
1383
_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
1384
' generators should be creating utf8 file ids.')
1387
def safe_file_id(unicode_or_utf8_string, warn=True):
1388
"""File ids should now be utf8, but at one point they were unicode.
1390
This is the same as safe_utf8, except it uses the cached encode functions
1391
to save a little bit of performance.
1393
:param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
1395
:param warn: Functions that are sanitizing user data can set warn=False
1396
:return: None or a utf8 file id.
1398
if (unicode_or_utf8_string is None
1399
or unicode_or_utf8_string.__class__ == str):
1400
return unicode_or_utf8_string
1402
symbol_versioning.warn(_file_id_warning, DeprecationWarning,
1404
return cache_utf8.encode(unicode_or_utf8_string)
1407
_platform_normalizes_filenames = False
1408
if sys.platform == 'darwin':
1409
_platform_normalizes_filenames = True
1412
def normalizes_filenames():
1413
"""Return True if this platform normalizes unicode filenames.
1417
return _platform_normalizes_filenames
1420
def _accessible_normalized_filename(path):
1421
"""Get the unicode normalized path, and if you can access the file.
1423
On platforms where the system normalizes filenames (Mac OSX),
1424
you can access a file by any path which will normalize correctly.
1425
On platforms where the system does not normalize filenames
1426
(everything else), you have to access a file by its exact path.
1428
Internally, bzr only supports NFC normalization, since that is
1429
the standard for XML documents.
1431
So return the normalized path, and a flag indicating if the file
1432
can be accessed by that path.
1435
return unicodedata.normalize('NFC', unicode(path)), True
1438
def _inaccessible_normalized_filename(path):
1439
__doc__ = _accessible_normalized_filename.__doc__
1441
normalized = unicodedata.normalize('NFC', unicode(path))
1442
return normalized, normalized == path
1445
if _platform_normalizes_filenames:
1446
normalized_filename = _accessible_normalized_filename
1448
normalized_filename = _inaccessible_normalized_filename
1451
def set_signal_handler(signum, handler, restart_syscall=True):
1452
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1453
on platforms that support that.
1455
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1456
automatically restart (by calling `signal.siginterrupt(signum,
1457
False)`). May be ignored if the feature is not available on this
1458
platform or Python version.
1462
siginterrupt = signal.siginterrupt
1464
# This python implementation doesn't provide signal support, hence no
1467
except AttributeError:
1468
# siginterrupt doesn't exist on this platform, or for this version
1470
siginterrupt = lambda signum, flag: None
1472
def sig_handler(*args):
1473
# Python resets the siginterrupt flag when a signal is
1474
# received. <http://bugs.python.org/issue8354>
1475
# As a workaround for some cases, set it back the way we want it.
1476
siginterrupt(signum, False)
1477
# Now run the handler function passed to set_signal_handler.
1480
sig_handler = handler
1481
old_handler = signal.signal(signum, sig_handler)
1483
siginterrupt(signum, False)
1487
default_terminal_width = 80
1488
"""The default terminal width for ttys.
1490
This is defined so that higher levels can share a common fallback value when
1491
terminal_width() returns None.
1494
# Keep some state so that terminal_width can detect if _terminal_size has
1495
# returned a different size since the process started. See docstring and
1496
# comments of terminal_width for details.
1497
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
1498
_terminal_size_state = 'no_data'
1499
_first_terminal_size = None
1501
def terminal_width():
1502
"""Return terminal width.
1504
None is returned if the width can't established precisely.
1507
- if BZR_COLUMNS is set, returns its value
1508
- if there is no controlling terminal, returns None
1509
- query the OS, if the queried size has changed since the last query,
1511
- if COLUMNS is set, returns its value,
1512
- if the OS has a value (even though it's never changed), return its value.
1514
From there, we need to query the OS to get the size of the controlling
1517
On Unices we query the OS by:
1518
- get termios.TIOCGWINSZ
1519
- if an error occurs or a negative value is obtained, returns None
1521
On Windows we query the OS by:
1522
- win32utils.get_console_size() decides,
1523
- returns None on error (provided default value)
1525
# Note to implementors: if changing the rules for determining the width,
1526
# make sure you've considered the behaviour in these cases:
1527
# - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
1528
# - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
1530
# - (add more interesting cases here, if you find any)
1531
# Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
1532
# but we don't want to register a signal handler because it is impossible
1533
# to do so without risking EINTR errors in Python <= 2.6.5 (see
1534
# <http://bugs.python.org/issue8354>). Instead we check TIOCGWINSZ every
1535
# time so we can notice if the reported size has changed, which should have
1538
# If BZR_COLUMNS is set, take it, user is always right
1539
# Except if they specified 0 in which case, impose no limit here
1541
width = int(os.environ['BZR_COLUMNS'])
1542
except (KeyError, ValueError):
1544
if width is not None:
1550
isatty = getattr(sys.stdout, 'isatty', None)
1551
if isatty is None or not isatty():
1552
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1556
width, height = os_size = _terminal_size(None, None)
1557
global _first_terminal_size, _terminal_size_state
1558
if _terminal_size_state == 'no_data':
1559
_first_terminal_size = os_size
1560
_terminal_size_state = 'unchanged'
1561
elif (_terminal_size_state == 'unchanged' and
1562
_first_terminal_size != os_size):
1563
_terminal_size_state = 'changed'
1565
# If the OS claims to know how wide the terminal is, and this value has
1566
# ever changed, use that.
1567
if _terminal_size_state == 'changed':
1568
if width is not None and width > 0:
1571
# If COLUMNS is set, use it.
1573
return int(os.environ['COLUMNS'])
1574
except (KeyError, ValueError):
1577
# Finally, use an unchanged size from the OS, if we have one.
1578
if _terminal_size_state == 'unchanged':
1579
if width is not None and width > 0:
1582
# The width could not be determined.
1586
def _win32_terminal_size(width, height):
1587
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1588
return width, height
1591
def _ioctl_terminal_size(width, height):
1593
import struct, fcntl, termios
1594
s = struct.pack('HHHH', 0, 0, 0, 0)
1595
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1596
height, width = struct.unpack('HHHH', x)[0:2]
1597
except (IOError, AttributeError):
1599
return width, height
1601
_terminal_size = None
1602
"""Returns the terminal size as (width, height).
1604
:param width: Default value for width.
1605
:param height: Default value for height.
1607
This is defined specifically for each OS and query the size of the controlling
1608
terminal. If any error occurs, the provided default values should be returned.
1610
if sys.platform == 'win32':
1611
_terminal_size = _win32_terminal_size
1613
_terminal_size = _ioctl_terminal_size
1616
def supports_executable():
1617
return sys.platform != "win32"
1620
def supports_posix_readonly():
1621
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1623
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1624
directory controls creation/deletion, etc.
1626
And under win32, readonly means that the directory itself cannot be
1627
deleted. The contents of a readonly directory can be changed, unlike POSIX
1628
where files in readonly directories cannot be added, deleted or renamed.
1630
return sys.platform != "win32"
1633
def set_or_unset_env(env_variable, value):
1634
"""Modify the environment, setting or removing the env_variable.
1636
:param env_variable: The environment variable in question
1637
:param value: The value to set the environment to. If None, then
1638
the variable will be removed.
1639
:return: The original value of the environment variable.
1641
orig_val = os.environ.get(env_variable)
1643
if orig_val is not None:
1644
del os.environ[env_variable]
1646
if isinstance(value, unicode):
1647
value = value.encode(get_user_encoding())
1648
os.environ[env_variable] = value
1652
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
1655
def check_legal_path(path):
1656
"""Check whether the supplied path is legal.
1657
This is only required on Windows, so we don't test on other platforms
1660
if sys.platform != "win32":
1662
if _validWin32PathRE.match(path) is None:
1663
raise errors.IllegalPath(path)
1666
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1668
def _is_error_enotdir(e):
1669
"""Check if this exception represents ENOTDIR.
1671
Unfortunately, python is very inconsistent about the exception
1672
here. The cases are:
1673
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1674
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1675
which is the windows error code.
1676
3) Windows, Python2.5 uses errno == EINVAL and
1677
winerror == ERROR_DIRECTORY
1679
:param e: An Exception object (expected to be OSError with an errno
1680
attribute, but we should be able to cope with anything)
1681
:return: True if this represents an ENOTDIR error. False otherwise.
1683
en = getattr(e, 'errno', None)
1684
if (en == errno.ENOTDIR
1685
or (sys.platform == 'win32'
1686
and (en == _WIN32_ERROR_DIRECTORY
1687
or (en == errno.EINVAL
1688
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1694
def walkdirs(top, prefix=""):
1695
"""Yield data about all the directories in a tree.
1697
This yields all the data about the contents of a directory at a time.
1698
After each directory has been yielded, if the caller has mutated the list
1699
to exclude some directories, they are then not descended into.
1701
The data yielded is of the form:
1702
((directory-relpath, directory-path-from-top),
1703
[(relpath, basename, kind, lstat, path-from-top), ...]),
1704
- directory-relpath is the relative path of the directory being returned
1705
with respect to top. prefix is prepended to this.
1706
- directory-path-from-root is the path including top for this directory.
1707
It is suitable for use with os functions.
1708
- relpath is the relative path within the subtree being walked.
1709
- basename is the basename of the path
1710
- kind is the kind of the file now. If unknown then the file is not
1711
present within the tree - but it may be recorded as versioned. See
1713
- lstat is the stat data *if* the file was statted.
1714
- planned, not implemented:
1715
path_from_tree_root is the path from the root of the tree.
1717
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1718
allows one to walk a subtree but get paths that are relative to a tree
1720
:return: an iterator over the dirs.
1722
#TODO there is a bit of a smell where the results of the directory-
1723
# summary in this, and the path from the root, may not agree
1724
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1725
# potentially confusing output. We should make this more robust - but
1726
# not at a speed cost. RBC 20060731
1728
_directory = _directory_kind
1729
_listdir = os.listdir
1730
_kind_from_mode = file_kind_from_stat_mode
1731
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1733
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1734
relroot, _, _, _, top = pending.pop()
1736
relprefix = relroot + u'/'
1739
top_slash = top + u'/'
1742
append = dirblock.append
1744
names = sorted(map(decode_filename, _listdir(top)))
1746
if not _is_error_enotdir(e):
1750
abspath = top_slash + name
1751
statvalue = _lstat(abspath)
1752
kind = _kind_from_mode(statvalue.st_mode)
1753
append((relprefix + name, name, kind, statvalue, abspath))
1754
yield (relroot, top), dirblock
1756
# push the user specified dirs from dirblock
1757
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1760
class DirReader(object):
1761
"""An interface for reading directories."""
1763
def top_prefix_to_starting_dir(self, top, prefix=""):
1764
"""Converts top and prefix to a starting dir entry
1766
:param top: A utf8 path
1767
:param prefix: An optional utf8 path to prefix output relative paths
1769
:return: A tuple starting with prefix, and ending with the native
1772
raise NotImplementedError(self.top_prefix_to_starting_dir)
1774
def read_dir(self, prefix, top):
1775
"""Read a specific dir.
1777
:param prefix: A utf8 prefix to be preprended to the path basenames.
1778
:param top: A natively encoded path to read.
1779
:return: A list of the directories contents. Each item contains:
1780
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1782
raise NotImplementedError(self.read_dir)
1785
_selected_dir_reader = None
1788
def _walkdirs_utf8(top, prefix=""):
1789
"""Yield data about all the directories in a tree.
1791
This yields the same information as walkdirs() only each entry is yielded
1792
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1793
are returned as exact byte-strings.
1795
:return: yields a tuple of (dir_info, [file_info])
1796
dir_info is (utf8_relpath, path-from-top)
1797
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1798
if top is an absolute path, path-from-top is also an absolute path.
1799
path-from-top might be unicode or utf8, but it is the correct path to
1800
pass to os functions to affect the file in question. (such as os.lstat)
1802
global _selected_dir_reader
1803
if _selected_dir_reader is None:
1804
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1805
# Win98 doesn't have unicode apis like FindFirstFileW
1806
# TODO: We possibly could support Win98 by falling back to the
1807
# original FindFirstFile, and using TCHAR instead of WCHAR,
1808
# but that gets a bit tricky, and requires custom compiling
1811
from bzrlib._walkdirs_win32 import Win32ReadDir
1812
_selected_dir_reader = Win32ReadDir()
1815
elif _fs_enc in ('utf-8', 'ascii'):
1817
from bzrlib._readdir_pyx import UTF8DirReader
1818
_selected_dir_reader = UTF8DirReader()
1819
except ImportError, e:
1820
failed_to_load_extension(e)
1823
if _selected_dir_reader is None:
1824
# Fallback to the python version
1825
_selected_dir_reader = UnicodeDirReader()
1827
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1828
# But we don't actually uses 1-3 in pending, so set them to None
1829
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1830
read_dir = _selected_dir_reader.read_dir
1831
_directory = _directory_kind
1833
relroot, _, _, _, top = pending[-1].pop()
1836
dirblock = sorted(read_dir(relroot, top))
1837
yield (relroot, top), dirblock
1838
# push the user specified dirs from dirblock
1839
next = [d for d in reversed(dirblock) if d[2] == _directory]
1841
pending.append(next)
1844
class UnicodeDirReader(DirReader):
1845
"""A dir reader for non-utf8 file systems, which transcodes."""
1847
__slots__ = ['_utf8_encode']
1850
self._utf8_encode = codecs.getencoder('utf8')
1852
def top_prefix_to_starting_dir(self, top, prefix=""):
1853
"""See DirReader.top_prefix_to_starting_dir."""
1854
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1856
def read_dir(self, prefix, top):
1857
"""Read a single directory from a non-utf8 file system.
1859
top, and the abspath element in the output are unicode, all other paths
1860
are utf8. Local disk IO is done via unicode calls to listdir etc.
1862
This is currently the fallback code path when the filesystem encoding is
1863
not UTF-8. It may be better to implement an alternative so that we can
1864
safely handle paths that are not properly decodable in the current
1867
See DirReader.read_dir for details.
1869
_utf8_encode = self._utf8_encode
1871
_listdir = os.listdir
1872
_kind_from_mode = file_kind_from_stat_mode
1875
relprefix = prefix + '/'
1878
top_slash = top + u'/'
1881
append = dirblock.append
1882
for name in sorted(_listdir(top)):
1884
name_utf8 = _utf8_encode(name)[0]
1885
except UnicodeDecodeError:
1886
raise errors.BadFilenameEncoding(
1887
_utf8_encode(relprefix)[0] + name, _fs_enc)
1888
abspath = top_slash + name
1889
statvalue = _lstat(abspath)
1890
kind = _kind_from_mode(statvalue.st_mode)
1891
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1895
def copy_tree(from_path, to_path, handlers={}):
1896
"""Copy all of the entries in from_path into to_path.
1898
:param from_path: The base directory to copy.
1899
:param to_path: The target directory. If it does not exist, it will
1901
:param handlers: A dictionary of functions, which takes a source and
1902
destinations for files, directories, etc.
1903
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1904
'file', 'directory', and 'symlink' should always exist.
1905
If they are missing, they will be replaced with 'os.mkdir()',
1906
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1908
# Now, just copy the existing cached tree to the new location
1909
# We use a cheap trick here.
1910
# Absolute paths are prefixed with the first parameter
1911
# relative paths are prefixed with the second.
1912
# So we can get both the source and target returned
1913
# without any extra work.
1915
def copy_dir(source, dest):
1918
def copy_link(source, dest):
1919
"""Copy the contents of a symlink"""
1920
link_to = os.readlink(source)
1921
os.symlink(link_to, dest)
1923
real_handlers = {'file':shutil.copy2,
1924
'symlink':copy_link,
1925
'directory':copy_dir,
1927
real_handlers.update(handlers)
1929
if not os.path.exists(to_path):
1930
real_handlers['directory'](from_path, to_path)
1932
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1933
for relpath, name, kind, st, abspath in entries:
1934
real_handlers[kind](abspath, relpath)
1937
def copy_ownership_from_path(dst, src=None):
1938
"""Copy usr/grp ownership from src file/dir to dst file/dir.
1940
If src is None, the containing directory is used as source. If chown
1941
fails, the error is ignored and a warning is printed.
1943
chown = getattr(os, 'chown', None)
1948
src = os.path.dirname(dst)
1954
chown(dst, s.st_uid, s.st_gid)
1957
'Unable to copy ownership from "%s" to "%s". '
1958
'You may want to set it manually.', src, dst)
1959
trace.log_exception_quietly()
1962
def path_prefix_key(path):
1963
"""Generate a prefix-order path key for path.
1965
This can be used to sort paths in the same way that walkdirs does.
1967
return (dirname(path) , path)
1970
def compare_paths_prefix_order(path_a, path_b):
1971
"""Compare path_a and path_b to generate the same order walkdirs uses."""
1972
key_a = path_prefix_key(path_a)
1973
key_b = path_prefix_key(path_b)
1974
return cmp(key_a, key_b)
1977
_cached_user_encoding = None
1980
def get_user_encoding(use_cache=True):
1981
"""Find out what the preferred user encoding is.
1983
This is generally the encoding that is used for command line parameters
1984
and file contents. This may be different from the terminal encoding
1985
or the filesystem encoding.
1987
:param use_cache: Enable cache for detected encoding.
1988
(This parameter is turned on by default,
1989
and required only for selftesting)
1991
:return: A string defining the preferred user encoding
1993
global _cached_user_encoding
1994
if _cached_user_encoding is not None and use_cache:
1995
return _cached_user_encoding
1997
if sys.platform == 'darwin':
1998
# python locale.getpreferredencoding() always return
1999
# 'mac-roman' on darwin. That's a lie.
2000
sys.platform = 'posix'
2002
if os.environ.get('LANG', None) is None:
2003
# If LANG is not set, we end up with 'ascii', which is bad
2004
# ('mac-roman' is more than ascii), so we set a default which
2005
# will give us UTF-8 (which appears to work in all cases on
2006
# OSX). Users are still free to override LANG of course, as
2007
# long as it give us something meaningful. This work-around
2008
# *may* not be needed with python 3k and/or OSX 10.5, but will
2009
# work with them too -- vila 20080908
2010
os.environ['LANG'] = 'en_US.UTF-8'
2013
sys.platform = 'darwin'
2018
user_encoding = locale.getpreferredencoding()
2019
except locale.Error, e:
2020
sys.stderr.write('bzr: warning: %s\n'
2021
' Could not determine what text encoding to use.\n'
2022
' This error usually means your Python interpreter\n'
2023
' doesn\'t support the locale set by $LANG (%s)\n'
2024
" Continuing with ascii encoding.\n"
2025
% (e, os.environ.get('LANG')))
2026
user_encoding = 'ascii'
2028
# Windows returns 'cp0' to indicate there is no code page. So we'll just
2029
# treat that as ASCII, and not support printing unicode characters to the
2032
# For python scripts run under vim, we get '', so also treat that as ASCII
2033
if user_encoding in (None, 'cp0', ''):
2034
user_encoding = 'ascii'
2038
codecs.lookup(user_encoding)
2040
sys.stderr.write('bzr: warning:'
2041
' unknown encoding %s.'
2042
' Continuing with ascii encoding.\n'
2045
user_encoding = 'ascii'
2048
_cached_user_encoding = user_encoding
2050
return user_encoding
2053
def get_diff_header_encoding():
2054
return get_terminal_encoding()
2057
_message_encoding = None
2060
def get_message_encoding():
2061
"""Return the encoding used for messages
2063
While the message encoding is a general setting it should usually only be
2064
needed for decoding system error strings such as from OSError instances.
2066
global _message_encoding
2067
if _message_encoding is None:
2068
if os.name == "posix":
2070
# This is a process-global setting that can change, but should in
2071
# general just get set once at process startup then be constant.
2072
_message_encoding = locale.getlocale(locale.LC_MESSAGES)[1]
2074
# On windows want the result of GetACP() which this boils down to.
2075
_message_encoding = get_user_encoding()
2076
return _message_encoding or "ascii"
2079
def get_host_name():
2080
"""Return the current unicode host name.
2082
This is meant to be used in place of socket.gethostname() because that
2083
behaves inconsistently on different platforms.
2085
if sys.platform == "win32":
2087
return win32utils.get_host_name()
2090
return socket.gethostname().decode(get_user_encoding())
2093
# We must not read/write any more than 64k at a time from/to a socket so we
2094
# don't risk "no buffer space available" errors on some platforms. Windows in
2095
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
2097
MAX_SOCKET_CHUNK = 64 * 1024
2099
_end_of_stream_errors = [errno.ECONNRESET]
2100
for _eno in ['WSAECONNRESET', 'WSAECONNABORTED']:
2101
_eno = getattr(errno, _eno, None)
2102
if _eno is not None:
2103
_end_of_stream_errors.append(_eno)
2107
def read_bytes_from_socket(sock, report_activity=None,
2108
max_read_size=MAX_SOCKET_CHUNK):
2109
"""Read up to max_read_size of bytes from sock and notify of progress.
2111
Translates "Connection reset by peer" into file-like EOF (return an
2112
empty string rather than raise an error), and repeats the recv if
2113
interrupted by a signal.
2117
bytes = sock.recv(max_read_size)
2118
except socket.error, e:
2120
if eno in _end_of_stream_errors:
2121
# The connection was closed by the other side. Callers expect
2122
# an empty string to signal end-of-stream.
2124
elif eno == errno.EINTR:
2125
# Retry the interrupted recv.
2129
if report_activity is not None:
2130
report_activity(len(bytes), 'read')
2134
def recv_all(socket, count):
2135
"""Receive an exact number of bytes.
2137
Regular Socket.recv() may return less than the requested number of bytes,
2138
depending on what's in the OS buffer. MSG_WAITALL is not available
2139
on all platforms, but this should work everywhere. This will return
2140
less than the requested amount if the remote end closes.
2142
This isn't optimized and is intended mostly for use in testing.
2145
while len(b) < count:
2146
new = read_bytes_from_socket(socket, None, count - len(b))
2153
def send_all(sock, bytes, report_activity=None):
2154
"""Send all bytes on a socket.
2156
Breaks large blocks in smaller chunks to avoid buffering limitations on
2157
some platforms, and catches EINTR which may be thrown if the send is
2158
interrupted by a signal.
2160
This is preferred to socket.sendall(), because it avoids portability bugs
2161
and provides activity reporting.
2163
:param report_activity: Call this as bytes are read, see
2164
Transport._report_activity
2167
byte_count = len(bytes)
2168
while sent_total < byte_count:
2170
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2171
except socket.error, e:
2172
if e.args[0] != errno.EINTR:
2176
report_activity(sent, 'write')
2179
def connect_socket(address):
2180
# Slight variation of the socket.create_connection() function (provided by
2181
# python-2.6) that can fail if getaddrinfo returns an empty list. We also
2182
# provide it for previous python versions. Also, we don't use the timeout
2183
# parameter (provided by the python implementation) so we don't implement
2185
err = socket.error('getaddrinfo returns an empty list')
2186
host, port = address
2187
for res in socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM):
2188
af, socktype, proto, canonname, sa = res
2191
sock = socket.socket(af, socktype, proto)
2195
except socket.error, err:
2196
# 'err' is now the most recent error
2197
if sock is not None:
2202
def dereference_path(path):
2203
"""Determine the real path to a file.
2205
All parent elements are dereferenced. But the file itself is not
2207
:param path: The original path. May be absolute or relative.
2208
:return: the real path *to* the file
2210
parent, base = os.path.split(path)
2211
# The pathjoin for '.' is a workaround for Python bug #1213894.
2212
# (initial path components aren't dereferenced)
2213
return pathjoin(realpath(pathjoin('.', parent)), base)
2216
def supports_mapi():
2217
"""Return True if we can use MAPI to launch a mail client."""
2218
return sys.platform == "win32"
2221
def resource_string(package, resource_name):
2222
"""Load a resource from a package and return it as a string.
2224
Note: Only packages that start with bzrlib are currently supported.
2226
This is designed to be a lightweight implementation of resource
2227
loading in a way which is API compatible with the same API from
2229
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2230
If and when pkg_resources becomes a standard library, this routine
2233
# Check package name is within bzrlib
2234
if package == "bzrlib":
2235
resource_relpath = resource_name
2236
elif package.startswith("bzrlib."):
2237
package = package[len("bzrlib."):].replace('.', os.sep)
2238
resource_relpath = pathjoin(package, resource_name)
2240
raise errors.BzrError('resource package %s not in bzrlib' % package)
2242
# Map the resource to a file and read its contents
2243
base = dirname(bzrlib.__file__)
2244
if getattr(sys, 'frozen', None): # bzr.exe
2245
base = abspath(pathjoin(base, '..', '..'))
2246
f = file(pathjoin(base, resource_relpath), "rU")
2252
def file_kind_from_stat_mode_thunk(mode):
2253
global file_kind_from_stat_mode
2254
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2256
from bzrlib._readdir_pyx import UTF8DirReader
2257
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2258
except ImportError, e:
2259
# This is one time where we won't warn that an extension failed to
2260
# load. The extension is never available on Windows anyway.
2261
from bzrlib._readdir_py import (
2262
_kind_from_mode as file_kind_from_stat_mode
2264
return file_kind_from_stat_mode(mode)
2265
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2267
def file_stat(f, _lstat=os.lstat):
2272
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2273
raise errors.NoSuchFile(f)
2276
def file_kind(f, _lstat=os.lstat):
2277
stat_value = file_stat(f, _lstat)
2278
return file_kind_from_stat_mode(stat_value.st_mode)
2280
def until_no_eintr(f, *a, **kw):
2281
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2283
WARNING: you must be certain that it is safe to retry the call repeatedly
2284
if EINTR does occur. This is typically only true for low-level operations
2285
like os.read. If in any doubt, don't use this.
2287
Keep in mind that this is not a complete solution to EINTR. There is
2288
probably code in the Python standard library and other dependencies that
2289
may encounter EINTR if a signal arrives (and there is signal handler for
2290
that signal). So this function can reduce the impact for IO that bzrlib
2291
directly controls, but it is not a complete solution.
2293
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2297
except (IOError, OSError), e:
2298
if e.errno == errno.EINTR:
2303
@deprecated_function(deprecated_in((2, 2, 0)))
2304
def re_compile_checked(re_string, flags=0, where=""):
2305
"""Return a compiled re, or raise a sensible error.
2307
This should only be used when compiling user-supplied REs.
2309
:param re_string: Text form of regular expression.
2310
:param flags: eg re.IGNORECASE
2311
:param where: Message explaining to the user the context where
2312
it occurred, eg 'log search filter'.
2314
# from https://bugs.launchpad.net/bzr/+bug/251352
2316
re_obj = re.compile(re_string, flags)
2319
except errors.InvalidPattern, e:
2321
where = ' in ' + where
2322
# despite the name 'error' is a type
2323
raise errors.BzrCommandError('Invalid regular expression%s: %s'
2327
if sys.platform == "win32":
2330
return msvcrt.getch()
2335
fd = sys.stdin.fileno()
2336
settings = termios.tcgetattr(fd)
2339
ch = sys.stdin.read(1)
2341
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2344
if sys.platform.startswith('linux'):
2345
def _local_concurrency():
2347
return os.sysconf('SC_NPROCESSORS_ONLN')
2348
except (ValueError, OSError, AttributeError):
2350
elif sys.platform == 'darwin':
2351
def _local_concurrency():
2352
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2353
stdout=subprocess.PIPE).communicate()[0]
2354
elif "bsd" in sys.platform:
2355
def _local_concurrency():
2356
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2357
stdout=subprocess.PIPE).communicate()[0]
2358
elif sys.platform == 'sunos5':
2359
def _local_concurrency():
2360
return subprocess.Popen(['psrinfo', '-p',],
2361
stdout=subprocess.PIPE).communicate()[0]
2362
elif sys.platform == "win32":
2363
def _local_concurrency():
2364
# This appears to return the number of cores.
2365
return os.environ.get('NUMBER_OF_PROCESSORS')
2367
def _local_concurrency():
2372
_cached_local_concurrency = None
2374
def local_concurrency(use_cache=True):
2375
"""Return how many processes can be run concurrently.
2377
Rely on platform specific implementations and default to 1 (one) if
2378
anything goes wrong.
2380
global _cached_local_concurrency
2382
if _cached_local_concurrency is not None and use_cache:
2383
return _cached_local_concurrency
2385
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2386
if concurrency is None:
2388
import multiprocessing
2389
concurrency = multiprocessing.cpu_count()
2390
except (ImportError, NotImplementedError):
2391
# multiprocessing is only available on Python >= 2.6
2392
# and multiprocessing.cpu_count() isn't implemented on all
2395
concurrency = _local_concurrency()
2396
except (OSError, IOError):
2399
concurrency = int(concurrency)
2400
except (TypeError, ValueError):
2403
_cached_concurrency = concurrency
2407
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2408
"""A stream writer that doesn't decode str arguments."""
2410
def __init__(self, encode, stream, errors='strict'):
2411
codecs.StreamWriter.__init__(self, stream, errors)
2412
self.encode = encode
2414
def write(self, object):
2415
if type(object) is str:
2416
self.stream.write(object)
2418
data, _ = self.encode(object, self.errors)
2419
self.stream.write(data)
2421
if sys.platform == 'win32':
2422
def open_file(filename, mode='r', bufsize=-1):
2423
"""This function is used to override the ``open`` builtin.
2425
But it uses O_NOINHERIT flag so the file handle is not inherited by
2426
child processes. Deleting or renaming a closed file opened with this
2427
function is not blocking child processes.
2429
writing = 'w' in mode
2430
appending = 'a' in mode
2431
updating = '+' in mode
2432
binary = 'b' in mode
2435
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2436
# for flags for each modes.
2446
flags |= os.O_WRONLY
2447
flags |= os.O_CREAT | os.O_TRUNC
2452
flags |= os.O_WRONLY
2453
flags |= os.O_CREAT | os.O_APPEND
2458
flags |= os.O_RDONLY
2460
return os.fdopen(os.open(filename, flags), mode, bufsize)
2465
def available_backup_name(base, exists):
2466
"""Find a non-existing backup file name.
2468
This will *not* create anything, this only return a 'free' entry. This
2469
should be used for checking names in a directory below a locked
2470
tree/branch/repo to avoid race conditions. This is LBYL (Look Before You
2471
Leap) and generally discouraged.
2473
:param base: The base name.
2475
:param exists: A callable returning True if the path parameter exists.
2478
name = "%s.~%d~" % (base, counter)
2481
name = "%s.~%d~" % (base, counter)
2485
def set_fd_cloexec(fd):
2486
"""Set a Unix file descriptor's FD_CLOEXEC flag. Do nothing if platform
2487
support for this is not available.
2491
old = fcntl.fcntl(fd, fcntl.F_GETFD)
2492
fcntl.fcntl(fd, fcntl.F_SETFD, old | fcntl.FD_CLOEXEC)
2493
except (ImportError, AttributeError):
2494
# Either the fcntl module or specific constants are not present
2498
def find_executable_on_path(name):
2499
"""Finds an executable on the PATH.
2501
On Windows, this will try to append each extension in the PATHEXT
2502
environment variable to the name, if it cannot be found with the name
2505
:param name: The base name of the executable.
2506
:return: The path to the executable found or None.
2508
path = os.environ.get('PATH')
2511
path = path.split(os.pathsep)
2512
if sys.platform == 'win32':
2513
exts = os.environ.get('PATHEXT', '').split(os.pathsep)
2514
exts = [ext.lower() for ext in exts]
2515
base, ext = os.path.splitext(name)
2517
if ext.lower() not in exts:
2525
f = os.path.join(d, name) + ext
2526
if os.access(f, os.X_OK):
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
