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
17
from __future__ import absolute_import
27
from bzrlib.lazy_import import lazy_import
28
lazy_import(globals(), """
29
from datetime import datetime
34
# We need to import both shutil and rmtree as we export the later on posix
35
# and need the former on windows
37
from shutil import rmtree
40
# We need to import both tempfile and mkdtemp as we export the later on posix
41
# and need the former on windows
43
from tempfile import mkdtemp
53
from bzrlib.i18n import gettext
56
from bzrlib.symbol_versioning import (
68
from bzrlib import symbol_versioning, _fs_enc
71
# Cross platform wall-clock time functionality with decent resolution.
72
# On Linux ``time.clock`` returns only CPU time. On Windows, ``time.time()``
73
# only has a resolution of ~15ms. Note that ``time.clock()`` is not
74
# synchronized with ``time.time()``, this is only meant to be used to find
75
# delta times by subtracting from another call to this function.
76
timer_func = time.time
77
if sys.platform == 'win32':
78
timer_func = time.clock
80
# On win32, O_BINARY is used to indicate the file should
81
# be opened in binary mode, rather than text mode.
82
# On other platforms, O_BINARY doesn't exist, because
83
# they always open in binary mode, so it is okay to
84
# OR with 0 on those platforms.
85
# O_NOINHERIT and O_TEXT exists only on win32 too.
86
O_BINARY = getattr(os, 'O_BINARY', 0)
87
O_TEXT = getattr(os, 'O_TEXT', 0)
88
O_NOINHERIT = getattr(os, 'O_NOINHERIT', 0)
91
def get_unicode_argv():
93
user_encoding = get_user_encoding()
94
return [a.decode(user_encoding) for a in sys.argv[1:]]
95
except UnicodeDecodeError:
96
raise errors.BzrError(gettext("Parameter {0!r} encoding is unsupported by {1} "
97
"application locale.").format(a, user_encoding))
100
def make_readonly(filename):
101
"""Make a filename read-only."""
102
mod = os.lstat(filename).st_mode
103
if not stat.S_ISLNK(mod):
105
chmod_if_possible(filename, mod)
108
def make_writable(filename):
109
mod = os.lstat(filename).st_mode
110
if not stat.S_ISLNK(mod):
112
chmod_if_possible(filename, mod)
115
def chmod_if_possible(filename, mode):
116
# Set file mode if that can be safely done.
117
# Sometimes even on unix the filesystem won't allow it - see
118
# https://bugs.launchpad.net/bzr/+bug/606537
120
# It is probably faster to just do the chmod, rather than
121
# doing a stat, and then trying to compare
122
os.chmod(filename, mode)
123
except (IOError, OSError),e:
124
# Permission/access denied seems to commonly happen on smbfs; there's
125
# probably no point warning about it.
126
# <https://bugs.launchpad.net/bzr/+bug/606537>
127
if getattr(e, 'errno') in (errno.EPERM, errno.EACCES):
128
trace.mutter("ignore error on chmod of %r: %r" % (
134
def minimum_path_selection(paths):
135
"""Return the smallset subset of paths which are outside paths.
137
:param paths: A container (and hence not None) of paths.
138
:return: A set of paths sufficient to include everything in paths via
139
is_inside, drawn from the paths parameter.
145
return path.split('/')
146
sorted_paths = sorted(list(paths), key=sort_key)
148
search_paths = [sorted_paths[0]]
149
for path in sorted_paths[1:]:
150
if not is_inside(search_paths[-1], path):
151
# This path is unique, add it
152
search_paths.append(path)
154
return set(search_paths)
161
"""Return a quoted filename filename
163
This previously used backslash quoting, but that works poorly on
165
# TODO: I'm not really sure this is the best format either.x
167
if _QUOTE_RE is None:
168
_QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/\\_~-])')
170
if _QUOTE_RE.search(f):
176
_directory_kind = 'directory'
179
"""Return the current umask"""
180
# Assume that people aren't messing with the umask while running
181
# XXX: This is not thread safe, but there is no way to get the
182
# umask without setting it
190
_directory_kind: "/",
192
'tree-reference': '+',
196
def kind_marker(kind):
198
return _kind_marker_map[kind]
200
# Slightly faster than using .get(, '') when the common case is that
205
lexists = getattr(os.path, 'lexists', None)
209
stat = getattr(os, 'lstat', os.stat)
213
if e.errno == errno.ENOENT:
216
raise errors.BzrError(gettext("lstat/stat of ({0!r}): {1!r}").format(f, e))
219
def fancy_rename(old, new, rename_func, unlink_func):
220
"""A fancy rename, when you don't have atomic rename.
222
:param old: The old path, to rename from
223
:param new: The new path, to rename to
224
:param rename_func: The potentially non-atomic rename function
225
:param unlink_func: A way to delete the target file if the full rename
228
# sftp rename doesn't allow overwriting, so play tricks:
229
base = os.path.basename(new)
230
dirname = os.path.dirname(new)
231
# callers use different encodings for the paths so the following MUST
232
# respect that. We rely on python upcasting to unicode if new is unicode
233
# and keeping a str if not.
234
tmp_name = 'tmp.%s.%.9f.%d.%s' % (base, time.time(),
235
os.getpid(), rand_chars(10))
236
tmp_name = pathjoin(dirname, tmp_name)
238
# Rename the file out of the way, but keep track if it didn't exist
239
# We don't want to grab just any exception
240
# something like EACCES should prevent us from continuing
241
# The downside is that the rename_func has to throw an exception
242
# with an errno = ENOENT, or NoSuchFile
245
rename_func(new, tmp_name)
246
except (errors.NoSuchFile,), e:
249
# RBC 20060103 abstraction leakage: the paramiko SFTP clients rename
250
# function raises an IOError with errno is None when a rename fails.
251
# This then gets caught here.
252
if e.errno not in (None, errno.ENOENT, errno.ENOTDIR):
255
if (getattr(e, 'errno', None) is None
256
or e.errno not in (errno.ENOENT, errno.ENOTDIR)):
265
# This may throw an exception, in which case success will
267
rename_func(old, new)
269
except (IOError, OSError), e:
270
# source and target may be aliases of each other (e.g. on a
271
# case-insensitive filesystem), so we may have accidentally renamed
272
# source by when we tried to rename target
273
failure_exc = sys.exc_info()
274
if (file_existed and e.errno in (None, errno.ENOENT)
275
and old.lower() == new.lower()):
276
# source and target are the same file on a case-insensitive
277
# filesystem, so we don't generate an exception
281
# If the file used to exist, rename it back into place
282
# otherwise just delete it from the tmp location
284
unlink_func(tmp_name)
286
rename_func(tmp_name, new)
287
if failure_exc is not None:
289
raise failure_exc[0], failure_exc[1], failure_exc[2]
294
# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
295
# choke on a Unicode string containing a relative path if
296
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
298
def _posix_abspath(path):
299
# jam 20060426 rather than encoding to fsencoding
300
# copy posixpath.abspath, but use os.getcwdu instead
301
if not posixpath.isabs(path):
302
path = posixpath.join(getcwd(), path)
303
return _posix_normpath(path)
306
def _posix_realpath(path):
307
return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)
310
def _posix_normpath(path):
311
path = posixpath.normpath(path)
312
# Bug 861008: posixpath.normpath() returns a path normalized according to
313
# the POSIX standard, which stipulates (for compatibility reasons) that two
314
# leading slashes must not be simplified to one, and only if there are 3 or
315
# more should they be simplified as one. So we treat the leading 2 slashes
316
# as a special case here by simply removing the first slash, as we consider
317
# that breaking POSIX compatibility for this obscure feature is acceptable.
318
# This is not a paranoid precaution, as we notably get paths like this when
319
# the repo is hosted at the root of the filesystem, i.e. in "/".
320
if path.startswith('//'):
325
def _posix_path_from_environ(key):
326
"""Get unicode path from `key` in environment or None if not present
328
Note that posix systems use arbitrary byte strings for filesystem objects,
329
so a path that raises BadFilenameEncoding here may still be accessible.
331
val = os.environ.get(key, None)
335
return val.decode(_fs_enc)
336
except UnicodeDecodeError:
337
# GZ 2011-12-12:Ideally want to include `key` in the exception message
338
raise errors.BadFilenameEncoding(val, _fs_enc)
341
def _posix_getuser_unicode():
342
"""Get username from environment or password database as unicode"""
343
name = getpass.getuser()
344
user_encoding = get_user_encoding()
346
return name.decode(user_encoding)
347
except UnicodeDecodeError:
348
raise errors.BzrError("Encoding of username %r is unsupported by %s "
349
"application locale." % (name, user_encoding))
352
def _win32_fixdrive(path):
353
"""Force drive letters to be consistent.
355
win32 is inconsistent whether it returns lower or upper case
356
and even if it was consistent the user might type the other
357
so we force it to uppercase
358
running python.exe under cmd.exe return capital C:\\
359
running win32 python inside a cygwin shell returns lowercase c:\\
361
drive, path = ntpath.splitdrive(path)
362
return drive.upper() + path
365
def _win32_abspath(path):
366
# Real ntpath.abspath doesn't have a problem with a unicode cwd
367
return _win32_fixdrive(ntpath.abspath(unicode(path)).replace('\\', '/'))
370
def _win98_abspath(path):
371
"""Return the absolute version of a path.
372
Windows 98 safe implementation (python reimplementation
373
of Win32 API function GetFullPathNameW)
378
# \\HOST\path => //HOST/path
379
# //HOST/path => //HOST/path
380
# path => C:/cwd/path
383
# check for absolute path
384
drive = ntpath.splitdrive(path)[0]
385
if drive == '' and path[:2] not in('//','\\\\'):
387
# we cannot simply os.path.join cwd and path
388
# because os.path.join('C:','/path') produce '/path'
389
# and this is incorrect
390
if path[:1] in ('/','\\'):
391
cwd = ntpath.splitdrive(cwd)[0]
393
path = cwd + '\\' + path
394
return _win32_fixdrive(ntpath.normpath(path).replace('\\', '/'))
397
def _win32_realpath(path):
398
# Real ntpath.realpath doesn't have a problem with a unicode cwd
399
return _win32_fixdrive(ntpath.realpath(unicode(path)).replace('\\', '/'))
402
def _win32_pathjoin(*args):
403
return ntpath.join(*args).replace('\\', '/')
406
def _win32_normpath(path):
407
return _win32_fixdrive(ntpath.normpath(unicode(path)).replace('\\', '/'))
411
return _win32_fixdrive(os.getcwdu().replace('\\', '/'))
414
def _win32_mkdtemp(*args, **kwargs):
415
return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))
418
def _win32_rename(old, new):
419
"""We expect to be able to atomically replace 'new' with old.
421
On win32, if new exists, it must be moved out of the way first,
425
fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
427
if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
428
# If we try to rename a non-existant file onto cwd, we get
429
# EPERM or EACCES instead of ENOENT, this will raise ENOENT
430
# if the old path doesn't exist, sometimes we get EACCES
431
# On Linux, we seem to get EBUSY, on Mac we get EINVAL
437
return unicodedata.normalize('NFC', os.getcwdu())
440
# Default is to just use the python builtins, but these can be rebound on
441
# particular platforms.
442
abspath = _posix_abspath
443
realpath = _posix_realpath
444
pathjoin = os.path.join
445
normpath = _posix_normpath
446
path_from_environ = _posix_path_from_environ
447
getuser_unicode = _posix_getuser_unicode
450
dirname = os.path.dirname
451
basename = os.path.basename
452
split = os.path.split
453
splitext = os.path.splitext
454
# These were already lazily imported into local scope
455
# mkdtemp = tempfile.mkdtemp
456
# rmtree = shutil.rmtree
464
MIN_ABS_PATHLENGTH = 1
467
if sys.platform == 'win32':
468
if win32utils.winver == 'Windows 98':
469
abspath = _win98_abspath
471
abspath = _win32_abspath
472
realpath = _win32_realpath
473
pathjoin = _win32_pathjoin
474
normpath = _win32_normpath
475
getcwd = _win32_getcwd
476
mkdtemp = _win32_mkdtemp
477
rename = _win32_rename
479
from bzrlib import _walkdirs_win32
483
lstat = _walkdirs_win32.lstat
484
fstat = _walkdirs_win32.fstat
485
wrap_stat = _walkdirs_win32.wrap_stat
487
MIN_ABS_PATHLENGTH = 3
489
def _win32_delete_readonly(function, path, excinfo):
490
"""Error handler for shutil.rmtree function [for win32]
491
Helps to remove files and dirs marked as read-only.
493
exception = excinfo[1]
494
if function in (os.remove, os.rmdir) \
495
and isinstance(exception, OSError) \
496
and exception.errno == errno.EACCES:
502
def rmtree(path, ignore_errors=False, onerror=_win32_delete_readonly):
503
"""Replacer for shutil.rmtree: could remove readonly dirs/files"""
504
return shutil.rmtree(path, ignore_errors, onerror)
506
f = win32utils.get_unicode_argv # special function or None
509
path_from_environ = win32utils.get_environ_unicode
510
getuser_unicode = win32utils.get_user_name
512
elif sys.platform == 'darwin':
516
def get_terminal_encoding(trace=False):
517
"""Find the best encoding for printing to the screen.
519
This attempts to check both sys.stdout and sys.stdin to see
520
what encoding they are in, and if that fails it falls back to
521
osutils.get_user_encoding().
522
The problem is that on Windows, locale.getpreferredencoding()
523
is not the same encoding as that used by the console:
524
http://mail.python.org/pipermail/python-list/2003-May/162357.html
526
On my standard US Windows XP, the preferred encoding is
527
cp1252, but the console is cp437
529
:param trace: If True trace the selected encoding via mutter().
531
from bzrlib.trace import mutter
532
output_encoding = getattr(sys.stdout, 'encoding', None)
533
if not output_encoding:
534
input_encoding = getattr(sys.stdin, 'encoding', None)
535
if not input_encoding:
536
output_encoding = get_user_encoding()
538
mutter('encoding stdout as osutils.get_user_encoding() %r',
541
output_encoding = input_encoding
543
mutter('encoding stdout as sys.stdin encoding %r',
547
mutter('encoding stdout as sys.stdout encoding %r', output_encoding)
548
if output_encoding == 'cp0':
549
# invalid encoding (cp0 means 'no codepage' on Windows)
550
output_encoding = get_user_encoding()
552
mutter('cp0 is invalid encoding.'
553
' encoding stdout as osutils.get_user_encoding() %r',
557
codecs.lookup(output_encoding)
559
sys.stderr.write('bzr: warning:'
560
' unknown terminal encoding %s.\n'
561
' Using encoding %s instead.\n'
562
% (output_encoding, get_user_encoding())
564
output_encoding = get_user_encoding()
566
return output_encoding
569
def normalizepath(f):
570
if getattr(os.path, 'realpath', None) is not None:
574
[p,e] = os.path.split(f)
575
if e == "" or e == "." or e == "..":
578
return pathjoin(F(p), e)
582
"""True if f is an accessible directory."""
584
return stat.S_ISDIR(os.lstat(f)[stat.ST_MODE])
590
"""True if f is a regular file."""
592
return stat.S_ISREG(os.lstat(f)[stat.ST_MODE])
597
"""True if f is a symlink."""
599
return stat.S_ISLNK(os.lstat(f)[stat.ST_MODE])
603
def is_inside(dir, fname):
604
"""True if fname is inside dir.
606
The parameters should typically be passed to osutils.normpath first, so
607
that . and .. and repeated slashes are eliminated, and the separators
608
are canonical for the platform.
610
The empty string as a dir name is taken as top-of-tree and matches
613
# XXX: Most callers of this can actually do something smarter by
614
# looking at the inventory
624
return fname.startswith(dir)
627
def is_inside_any(dir_list, fname):
628
"""True if fname is inside any of given dirs."""
629
for dirname in dir_list:
630
if is_inside(dirname, fname):
635
def is_inside_or_parent_of_any(dir_list, fname):
636
"""True if fname is a child or a parent of any of the given files."""
637
for dirname in dir_list:
638
if is_inside(dirname, fname) or is_inside(fname, dirname):
643
def pumpfile(from_file, to_file, read_length=-1, buff_size=32768,
644
report_activity=None, direction='read'):
645
"""Copy contents of one file to another.
647
The read_length can either be -1 to read to end-of-file (EOF) or
648
it can specify the maximum number of bytes to read.
650
The buff_size represents the maximum size for each read operation
651
performed on from_file.
653
:param report_activity: Call this as bytes are read, see
654
Transport._report_activity
655
:param direction: Will be passed to report_activity
657
:return: The number of bytes copied.
661
# read specified number of bytes
663
while read_length > 0:
664
num_bytes_to_read = min(read_length, buff_size)
666
block = from_file.read(num_bytes_to_read)
670
if report_activity is not None:
671
report_activity(len(block), direction)
674
actual_bytes_read = len(block)
675
read_length -= actual_bytes_read
676
length += actual_bytes_read
680
block = from_file.read(buff_size)
684
if report_activity is not None:
685
report_activity(len(block), direction)
691
def pump_string_file(bytes, file_handle, segment_size=None):
692
"""Write bytes to file_handle in many smaller writes.
694
:param bytes: The string to write.
695
:param file_handle: The file to write to.
697
# Write data in chunks rather than all at once, because very large
698
# writes fail on some platforms (e.g. Windows with SMB mounted
701
segment_size = 5242880 # 5MB
702
segments = range(len(bytes) / segment_size + 1)
703
write = file_handle.write
704
for segment_index in segments:
705
segment = buffer(bytes, segment_index * segment_size, segment_size)
709
def file_iterator(input_file, readsize=32768):
711
b = input_file.read(readsize)
718
"""Calculate the hexdigest of an open file.
720
The file cursor should be already at the start.
732
def size_sha_file(f):
733
"""Calculate the size and hexdigest of an open file.
735
The file cursor should be already at the start and
736
the caller is responsible for closing the file afterwards.
747
return size, s.hexdigest()
750
def sha_file_by_name(fname):
751
"""Calculate the SHA1 of a file by reading the full text"""
753
f = os.open(fname, os.O_RDONLY | O_BINARY | O_NOINHERIT)
756
b = os.read(f, 1<<16)
764
def sha_strings(strings, _factory=sha):
765
"""Return the sha-1 of concatenation of strings"""
767
map(s.update, strings)
771
def sha_string(f, _factory=sha):
772
return _factory(f).hexdigest()
775
def fingerprint_file(f):
777
return {'size': len(b),
778
'sha1': sha(b).hexdigest()}
781
def compare_files(a, b):
782
"""Returns true if equal in contents"""
793
def local_time_offset(t=None):
794
"""Return offset of local zone from GMT, either at present or at time t."""
797
offset = datetime.fromtimestamp(t) - datetime.utcfromtimestamp(t)
798
return offset.days * 86400 + offset.seconds
800
weekdays = ['Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat', 'Sun']
801
_default_format_by_weekday_num = [wd + " %Y-%m-%d %H:%M:%S" for wd in weekdays]
804
def format_date(t, offset=0, timezone='original', date_fmt=None,
806
"""Return a formatted date string.
808
:param t: Seconds since the epoch.
809
:param offset: Timezone offset in seconds east of utc.
810
:param timezone: How to display the time: 'utc', 'original' for the
811
timezone specified by offset, or 'local' for the process's current
813
:param date_fmt: strftime format.
814
:param show_offset: Whether to append the timezone.
816
(date_fmt, tt, offset_str) = \
817
_format_date(t, offset, timezone, date_fmt, show_offset)
818
date_fmt = date_fmt.replace('%a', weekdays[tt[6]])
819
date_str = time.strftime(date_fmt, tt)
820
return date_str + offset_str
823
# Cache of formatted offset strings
827
def format_date_with_offset_in_original_timezone(t, offset=0,
828
_cache=_offset_cache):
829
"""Return a formatted date string in the original timezone.
831
This routine may be faster then format_date.
833
:param t: Seconds since the epoch.
834
:param offset: Timezone offset in seconds east of utc.
838
tt = time.gmtime(t + offset)
839
date_fmt = _default_format_by_weekday_num[tt[6]]
840
date_str = time.strftime(date_fmt, tt)
841
offset_str = _cache.get(offset, None)
842
if offset_str is None:
843
offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
844
_cache[offset] = offset_str
845
return date_str + offset_str
848
def format_local_date(t, offset=0, timezone='original', date_fmt=None,
850
"""Return an unicode date string formatted according to the current locale.
852
:param t: Seconds since the epoch.
853
:param offset: Timezone offset in seconds east of utc.
854
:param timezone: How to display the time: 'utc', 'original' for the
855
timezone specified by offset, or 'local' for the process's current
857
:param date_fmt: strftime format.
858
:param show_offset: Whether to append the timezone.
860
(date_fmt, tt, offset_str) = \
861
_format_date(t, offset, timezone, date_fmt, show_offset)
862
date_str = time.strftime(date_fmt, tt)
863
if not isinstance(date_str, unicode):
864
date_str = date_str.decode(get_user_encoding(), 'replace')
865
return date_str + offset_str
868
def _format_date(t, offset, timezone, date_fmt, show_offset):
869
if timezone == 'utc':
872
elif timezone == 'original':
875
tt = time.gmtime(t + offset)
876
elif timezone == 'local':
877
tt = time.localtime(t)
878
offset = local_time_offset(t)
880
raise errors.UnsupportedTimezoneFormat(timezone)
882
date_fmt = "%a %Y-%m-%d %H:%M:%S"
884
offset_str = ' %+03d%02d' % (offset / 3600, (offset / 60) % 60)
887
return (date_fmt, tt, offset_str)
890
def compact_date(when):
891
return time.strftime('%Y%m%d%H%M%S', time.gmtime(when))
894
def format_delta(delta):
895
"""Get a nice looking string for a time delta.
897
:param delta: The time difference in seconds, can be positive or negative.
898
positive indicates time in the past, negative indicates time in the
899
future. (usually time.time() - stored_time)
900
:return: String formatted to show approximate resolution
906
direction = 'in the future'
910
if seconds < 90: # print seconds up to 90 seconds
912
return '%d second %s' % (seconds, direction,)
914
return '%d seconds %s' % (seconds, direction)
916
minutes = int(seconds / 60)
917
seconds -= 60 * minutes
922
if minutes < 90: # print minutes, seconds up to 90 minutes
924
return '%d minute, %d second%s %s' % (
925
minutes, seconds, plural_seconds, direction)
927
return '%d minutes, %d second%s %s' % (
928
minutes, seconds, plural_seconds, direction)
930
hours = int(minutes / 60)
931
minutes -= 60 * hours
938
return '%d hour, %d minute%s %s' % (hours, minutes,
939
plural_minutes, direction)
940
return '%d hours, %d minute%s %s' % (hours, minutes,
941
plural_minutes, direction)
944
"""Return size of given open file."""
945
return os.fstat(f.fileno())[stat.ST_SIZE]
948
# Define rand_bytes based on platform.
950
# Python 2.4 and later have os.urandom,
951
# but it doesn't work on some arches
953
rand_bytes = os.urandom
954
except (NotImplementedError, AttributeError):
955
# If python doesn't have os.urandom, or it doesn't work,
956
# then try to first pull random data from /dev/urandom
958
rand_bytes = file('/dev/urandom', 'rb').read
959
# Otherwise, use this hack as a last resort
960
except (IOError, OSError):
961
# not well seeded, but better than nothing
966
s += chr(random.randint(0, 255))
971
ALNUM = '0123456789abcdefghijklmnopqrstuvwxyz'
973
"""Return a random string of num alphanumeric characters
975
The result only contains lowercase chars because it may be used on
976
case-insensitive filesystems.
979
for raw_byte in rand_bytes(num):
980
s += ALNUM[ord(raw_byte) % 36]
984
## TODO: We could later have path objects that remember their list
985
## decomposition (might be too tricksy though.)
988
"""Turn string into list of parts."""
989
# split on either delimiter because people might use either on
991
ps = re.split(r'[\\/]', p)
996
raise errors.BzrError(gettext("sorry, %r not allowed in path") % f)
997
elif (f == '.') or (f == ''):
1006
if (f == '..') or (f is None) or (f == ''):
1007
raise errors.BzrError(gettext("sorry, %r not allowed in path") % f)
1011
def parent_directories(filename):
1012
"""Return the list of parent directories, deepest first.
1014
For example, parent_directories("a/b/c") -> ["a/b", "a"].
1017
parts = splitpath(dirname(filename))
1019
parents.append(joinpath(parts))
1024
_extension_load_failures = []
1027
def failed_to_load_extension(exception):
1028
"""Handle failing to load a binary extension.
1030
This should be called from the ImportError block guarding the attempt to
1031
import the native extension. If this function returns, the pure-Python
1032
implementation should be loaded instead::
1035
>>> import bzrlib._fictional_extension_pyx
1036
>>> except ImportError, e:
1037
>>> bzrlib.osutils.failed_to_load_extension(e)
1038
>>> import bzrlib._fictional_extension_py
1040
# NB: This docstring is just an example, not a doctest, because doctest
1041
# currently can't cope with the use of lazy imports in this namespace --
1044
# This currently doesn't report the failure at the time it occurs, because
1045
# they tend to happen very early in startup when we can't check config
1046
# files etc, and also we want to report all failures but not spam the user
1048
exception_str = str(exception)
1049
if exception_str not in _extension_load_failures:
1050
trace.mutter("failed to load compiled extension: %s" % exception_str)
1051
_extension_load_failures.append(exception_str)
1054
def report_extension_load_failures():
1055
if not _extension_load_failures:
1057
if config.GlobalStack().get('ignore_missing_extensions'):
1059
# the warnings framework should by default show this only once
1060
from bzrlib.trace import warning
1062
"bzr: warning: some compiled extensions could not be loaded; "
1063
"see <https://answers.launchpad.net/bzr/+faq/703>")
1064
# we no longer show the specific missing extensions here, because it makes
1065
# the message too long and scary - see
1066
# https://bugs.launchpad.net/bzr/+bug/430529
1070
from bzrlib._chunks_to_lines_pyx import chunks_to_lines
1071
except ImportError, e:
1072
failed_to_load_extension(e)
1073
from bzrlib._chunks_to_lines_py import chunks_to_lines
1077
"""Split s into lines, but without removing the newline characters."""
1078
# Trivially convert a fulltext into a 'chunked' representation, and let
1079
# chunks_to_lines do the heavy lifting.
1080
if isinstance(s, str):
1081
# chunks_to_lines only supports 8-bit strings
1082
return chunks_to_lines([s])
1084
return _split_lines(s)
1087
def _split_lines(s):
1088
"""Split s into lines, but without removing the newline characters.
1090
This supports Unicode or plain string objects.
1092
lines = s.split('\n')
1093
result = [line + '\n' for line in lines[:-1]]
1095
result.append(lines[-1])
1099
def hardlinks_good():
1100
return sys.platform not in ('win32', 'cygwin', 'darwin')
1103
def link_or_copy(src, dest):
1104
"""Hardlink a file, or copy it if it can't be hardlinked."""
1105
if not hardlinks_good():
1106
shutil.copyfile(src, dest)
1110
except (OSError, IOError), e:
1111
if e.errno != errno.EXDEV:
1113
shutil.copyfile(src, dest)
1116
def delete_any(path):
1117
"""Delete a file, symlink or directory.
1119
Will delete even if readonly.
1122
_delete_file_or_dir(path)
1123
except (OSError, IOError), e:
1124
if e.errno in (errno.EPERM, errno.EACCES):
1125
# make writable and try again
1128
except (OSError, IOError):
1130
_delete_file_or_dir(path)
1135
def _delete_file_or_dir(path):
1136
# Look Before You Leap (LBYL) is appropriate here instead of Easier to Ask for
1137
# Forgiveness than Permission (EAFP) because:
1138
# - root can damage a solaris file system by using unlink,
1139
# - unlink raises different exceptions on different OSes (linux: EISDIR, win32:
1140
# EACCES, OSX: EPERM) when invoked on a directory.
1141
if isdir(path): # Takes care of symlinks
1148
if getattr(os, 'symlink', None) is not None:
1154
def has_hardlinks():
1155
if getattr(os, 'link', None) is not None:
1161
def host_os_dereferences_symlinks():
1162
return (has_symlinks()
1163
and sys.platform not in ('cygwin', 'win32'))
1166
def readlink(abspath):
1167
"""Return a string representing the path to which the symbolic link points.
1169
:param abspath: The link absolute unicode path.
1171
This his guaranteed to return the symbolic link in unicode in all python
1174
link = abspath.encode(_fs_enc)
1175
target = os.readlink(link)
1176
target = target.decode(_fs_enc)
1180
def contains_whitespace(s):
1181
"""True if there are any whitespace characters in s."""
1182
# string.whitespace can include '\xa0' in certain locales, because it is
1183
# considered "non-breaking-space" as part of ISO-8859-1. But it
1184
# 1) Isn't a breaking whitespace
1185
# 2) Isn't one of ' \t\r\n' which are characters we sometimes use as
1187
# 3) '\xa0' isn't unicode safe since it is >128.
1189
# This should *not* be a unicode set of characters in case the source
1190
# string is not a Unicode string. We can auto-up-cast the characters since
1191
# they are ascii, but we don't want to auto-up-cast the string in case it
1193
for ch in ' \t\n\r\v\f':
1200
def contains_linebreaks(s):
1201
"""True if there is any vertical whitespace in s."""
1209
def relpath(base, path):
1210
"""Return path relative to base, or raise PathNotChild exception.
1212
The path may be either an absolute path or a path relative to the
1213
current working directory.
1215
os.path.commonprefix (python2.4) has a bad bug that it works just
1216
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
1217
avoids that problem.
1219
NOTE: `base` should not have a trailing slash otherwise you'll get
1220
PathNotChild exceptions regardless of `path`.
1223
if len(base) < MIN_ABS_PATHLENGTH:
1224
# must have space for e.g. a drive letter
1225
raise ValueError(gettext('%r is too short to calculate a relative path')
1233
if len(head) <= len(base) and head != base:
1234
raise errors.PathNotChild(rp, base)
1237
head, tail = split(head)
1242
return pathjoin(*reversed(s))
1247
def _cicp_canonical_relpath(base, path):
1248
"""Return the canonical path relative to base.
1250
Like relpath, but on case-insensitive-case-preserving file-systems, this
1251
will return the relpath as stored on the file-system rather than in the
1252
case specified in the input string, for all existing portions of the path.
1254
This will cause O(N) behaviour if called for every path in a tree; if you
1255
have a number of paths to convert, you should use canonical_relpaths().
1257
# TODO: it should be possible to optimize this for Windows by using the
1258
# win32 API FindFiles function to look for the specified name - but using
1259
# os.listdir() still gives us the correct, platform agnostic semantics in
1262
rel = relpath(base, path)
1263
# '.' will have been turned into ''
1267
abs_base = abspath(base)
1269
_listdir = os.listdir
1271
# use an explicit iterator so we can easily consume the rest on early exit.
1272
bit_iter = iter(rel.split('/'))
1273
for bit in bit_iter:
1276
next_entries = _listdir(current)
1277
except OSError: # enoent, eperm, etc
1278
# We can't find this in the filesystem, so just append the
1280
current = pathjoin(current, bit, *list(bit_iter))
1282
for look in next_entries:
1283
if lbit == look.lower():
1284
current = pathjoin(current, look)
1287
# got to the end, nothing matched, so we just return the
1288
# non-existing bits as they were specified (the filename may be
1289
# the target of a move, for example).
1290
current = pathjoin(current, bit, *list(bit_iter))
1292
return current[len(abs_base):].lstrip('/')
1294
# XXX - TODO - we need better detection/integration of case-insensitive
1295
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
1296
# filesystems), for example, so could probably benefit from the same basic
1297
# support there. For now though, only Windows and OSX get that support, and
1298
# they get it for *all* file-systems!
1299
if sys.platform in ('win32', 'darwin'):
1300
canonical_relpath = _cicp_canonical_relpath
1302
canonical_relpath = relpath
1304
def canonical_relpaths(base, paths):
1305
"""Create an iterable to canonicalize a sequence of relative paths.
1307
The intent is for this implementation to use a cache, vastly speeding
1308
up multiple transformations in the same directory.
1310
# but for now, we haven't optimized...
1311
return [canonical_relpath(base, p) for p in paths]
1314
def decode_filename(filename):
1315
"""Decode the filename using the filesystem encoding
1317
If it is unicode, it is returned.
1318
Otherwise it is decoded from the the filesystem's encoding. If decoding
1319
fails, a errors.BadFilenameEncoding exception is raised.
1321
if type(filename) is unicode:
1324
return filename.decode(_fs_enc)
1325
except UnicodeDecodeError:
1326
raise errors.BadFilenameEncoding(filename, _fs_enc)
1329
def safe_unicode(unicode_or_utf8_string):
1330
"""Coerce unicode_or_utf8_string into unicode.
1332
If it is unicode, it is returned.
1333
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1334
wrapped in a BzrBadParameterNotUnicode exception.
1336
if isinstance(unicode_or_utf8_string, unicode):
1337
return unicode_or_utf8_string
1339
return unicode_or_utf8_string.decode('utf8')
1340
except UnicodeDecodeError:
1341
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1344
def safe_utf8(unicode_or_utf8_string):
1345
"""Coerce unicode_or_utf8_string to a utf8 string.
1347
If it is a str, it is returned.
1348
If it is Unicode, it is encoded into a utf-8 string.
1350
if isinstance(unicode_or_utf8_string, str):
1351
# TODO: jam 20070209 This is overkill, and probably has an impact on
1352
# performance if we are dealing with lots of apis that want a
1355
# Make sure it is a valid utf-8 string
1356
unicode_or_utf8_string.decode('utf-8')
1357
except UnicodeDecodeError:
1358
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1359
return unicode_or_utf8_string
1360
return unicode_or_utf8_string.encode('utf-8')
1363
_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
1364
' Revision id generators should be creating utf8'
1368
def safe_revision_id(unicode_or_utf8_string, warn=True):
1369
"""Revision ids should now be utf8, but at one point they were unicode.
1371
:param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
1373
:param warn: Functions that are sanitizing user data can set warn=False
1374
:return: None or a utf8 revision id.
1376
if (unicode_or_utf8_string is None
1377
or unicode_or_utf8_string.__class__ == str):
1378
return unicode_or_utf8_string
1380
symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
1382
return cache_utf8.encode(unicode_or_utf8_string)
1385
_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
1386
' generators should be creating utf8 file ids.')
1389
def safe_file_id(unicode_or_utf8_string, warn=True):
1390
"""File ids should now be utf8, but at one point they were unicode.
1392
This is the same as safe_utf8, except it uses the cached encode functions
1393
to save a little bit of performance.
1395
:param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
1397
:param warn: Functions that are sanitizing user data can set warn=False
1398
:return: None or a utf8 file id.
1400
if (unicode_or_utf8_string is None
1401
or unicode_or_utf8_string.__class__ == str):
1402
return unicode_or_utf8_string
1404
symbol_versioning.warn(_file_id_warning, DeprecationWarning,
1406
return cache_utf8.encode(unicode_or_utf8_string)
1409
_platform_normalizes_filenames = False
1410
if sys.platform == 'darwin':
1411
_platform_normalizes_filenames = True
1414
def normalizes_filenames():
1415
"""Return True if this platform normalizes unicode filenames.
1419
return _platform_normalizes_filenames
1422
def _accessible_normalized_filename(path):
1423
"""Get the unicode normalized path, and if you can access the file.
1425
On platforms where the system normalizes filenames (Mac OSX),
1426
you can access a file by any path which will normalize correctly.
1427
On platforms where the system does not normalize filenames
1428
(everything else), you have to access a file by its exact path.
1430
Internally, bzr only supports NFC normalization, since that is
1431
the standard for XML documents.
1433
So return the normalized path, and a flag indicating if the file
1434
can be accessed by that path.
1437
return unicodedata.normalize('NFC', unicode(path)), True
1440
def _inaccessible_normalized_filename(path):
1441
__doc__ = _accessible_normalized_filename.__doc__
1443
normalized = unicodedata.normalize('NFC', unicode(path))
1444
return normalized, normalized == path
1447
if _platform_normalizes_filenames:
1448
normalized_filename = _accessible_normalized_filename
1450
normalized_filename = _inaccessible_normalized_filename
1453
def set_signal_handler(signum, handler, restart_syscall=True):
1454
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1455
on platforms that support that.
1457
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1458
automatically restart (by calling `signal.siginterrupt(signum,
1459
False)`). May be ignored if the feature is not available on this
1460
platform or Python version.
1464
siginterrupt = signal.siginterrupt
1466
# This python implementation doesn't provide signal support, hence no
1469
except AttributeError:
1470
# siginterrupt doesn't exist on this platform, or for this version
1472
siginterrupt = lambda signum, flag: None
1474
def sig_handler(*args):
1475
# Python resets the siginterrupt flag when a signal is
1476
# received. <http://bugs.python.org/issue8354>
1477
# As a workaround for some cases, set it back the way we want it.
1478
siginterrupt(signum, False)
1479
# Now run the handler function passed to set_signal_handler.
1482
sig_handler = handler
1483
old_handler = signal.signal(signum, sig_handler)
1485
siginterrupt(signum, False)
1489
default_terminal_width = 80
1490
"""The default terminal width for ttys.
1492
This is defined so that higher levels can share a common fallback value when
1493
terminal_width() returns None.
1496
# Keep some state so that terminal_width can detect if _terminal_size has
1497
# returned a different size since the process started. See docstring and
1498
# comments of terminal_width for details.
1499
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
1500
_terminal_size_state = 'no_data'
1501
_first_terminal_size = None
1503
def terminal_width():
1504
"""Return terminal width.
1506
None is returned if the width can't established precisely.
1509
- if BZR_COLUMNS is set, returns its value
1510
- if there is no controlling terminal, returns None
1511
- query the OS, if the queried size has changed since the last query,
1513
- if COLUMNS is set, returns its value,
1514
- if the OS has a value (even though it's never changed), return its value.
1516
From there, we need to query the OS to get the size of the controlling
1519
On Unices we query the OS by:
1520
- get termios.TIOCGWINSZ
1521
- if an error occurs or a negative value is obtained, returns None
1523
On Windows we query the OS by:
1524
- win32utils.get_console_size() decides,
1525
- returns None on error (provided default value)
1527
# Note to implementors: if changing the rules for determining the width,
1528
# make sure you've considered the behaviour in these cases:
1529
# - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
1530
# - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
1532
# - (add more interesting cases here, if you find any)
1533
# Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
1534
# but we don't want to register a signal handler because it is impossible
1535
# to do so without risking EINTR errors in Python <= 2.6.5 (see
1536
# <http://bugs.python.org/issue8354>). Instead we check TIOCGWINSZ every
1537
# time so we can notice if the reported size has changed, which should have
1540
# If BZR_COLUMNS is set, take it, user is always right
1541
# Except if they specified 0 in which case, impose no limit here
1543
width = int(os.environ['BZR_COLUMNS'])
1544
except (KeyError, ValueError):
1546
if width is not None:
1552
isatty = getattr(sys.stdout, 'isatty', None)
1553
if isatty is None or not isatty():
1554
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1558
width, height = os_size = _terminal_size(None, None)
1559
global _first_terminal_size, _terminal_size_state
1560
if _terminal_size_state == 'no_data':
1561
_first_terminal_size = os_size
1562
_terminal_size_state = 'unchanged'
1563
elif (_terminal_size_state == 'unchanged' and
1564
_first_terminal_size != os_size):
1565
_terminal_size_state = 'changed'
1567
# If the OS claims to know how wide the terminal is, and this value has
1568
# ever changed, use that.
1569
if _terminal_size_state == 'changed':
1570
if width is not None and width > 0:
1573
# If COLUMNS is set, use it.
1575
return int(os.environ['COLUMNS'])
1576
except (KeyError, ValueError):
1579
# Finally, use an unchanged size from the OS, if we have one.
1580
if _terminal_size_state == 'unchanged':
1581
if width is not None and width > 0:
1584
# The width could not be determined.
1588
def _win32_terminal_size(width, height):
1589
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1590
return width, height
1593
def _ioctl_terminal_size(width, height):
1595
import struct, fcntl, termios
1596
s = struct.pack('HHHH', 0, 0, 0, 0)
1597
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1598
height, width = struct.unpack('HHHH', x)[0:2]
1599
except (IOError, AttributeError):
1601
return width, height
1603
_terminal_size = None
1604
"""Returns the terminal size as (width, height).
1606
:param width: Default value for width.
1607
:param height: Default value for height.
1609
This is defined specifically for each OS and query the size of the controlling
1610
terminal. If any error occurs, the provided default values should be returned.
1612
if sys.platform == 'win32':
1613
_terminal_size = _win32_terminal_size
1615
_terminal_size = _ioctl_terminal_size
1618
def supports_executable():
1619
return sys.platform != "win32"
1622
def supports_posix_readonly():
1623
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1625
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1626
directory controls creation/deletion, etc.
1628
And under win32, readonly means that the directory itself cannot be
1629
deleted. The contents of a readonly directory can be changed, unlike POSIX
1630
where files in readonly directories cannot be added, deleted or renamed.
1632
return sys.platform != "win32"
1635
def set_or_unset_env(env_variable, value):
1636
"""Modify the environment, setting or removing the env_variable.
1638
:param env_variable: The environment variable in question
1639
:param value: The value to set the environment to. If None, then
1640
the variable will be removed.
1641
:return: The original value of the environment variable.
1643
orig_val = os.environ.get(env_variable)
1645
if orig_val is not None:
1646
del os.environ[env_variable]
1648
if isinstance(value, unicode):
1649
value = value.encode(get_user_encoding())
1650
os.environ[env_variable] = value
1654
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
1657
def check_legal_path(path):
1658
"""Check whether the supplied path is legal.
1659
This is only required on Windows, so we don't test on other platforms
1662
if sys.platform != "win32":
1664
if _validWin32PathRE.match(path) is None:
1665
raise errors.IllegalPath(path)
1668
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1670
def _is_error_enotdir(e):
1671
"""Check if this exception represents ENOTDIR.
1673
Unfortunately, python is very inconsistent about the exception
1674
here. The cases are:
1675
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1676
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1677
which is the windows error code.
1678
3) Windows, Python2.5 uses errno == EINVAL and
1679
winerror == ERROR_DIRECTORY
1681
:param e: An Exception object (expected to be OSError with an errno
1682
attribute, but we should be able to cope with anything)
1683
:return: True if this represents an ENOTDIR error. False otherwise.
1685
en = getattr(e, 'errno', None)
1686
if (en == errno.ENOTDIR
1687
or (sys.platform == 'win32'
1688
and (en == _WIN32_ERROR_DIRECTORY
1689
or (en == errno.EINVAL
1690
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1696
def walkdirs(top, prefix=""):
1697
"""Yield data about all the directories in a tree.
1699
This yields all the data about the contents of a directory at a time.
1700
After each directory has been yielded, if the caller has mutated the list
1701
to exclude some directories, they are then not descended into.
1703
The data yielded is of the form:
1704
((directory-relpath, directory-path-from-top),
1705
[(relpath, basename, kind, lstat, path-from-top), ...]),
1706
- directory-relpath is the relative path of the directory being returned
1707
with respect to top. prefix is prepended to this.
1708
- directory-path-from-root is the path including top for this directory.
1709
It is suitable for use with os functions.
1710
- relpath is the relative path within the subtree being walked.
1711
- basename is the basename of the path
1712
- kind is the kind of the file now. If unknown then the file is not
1713
present within the tree - but it may be recorded as versioned. See
1715
- lstat is the stat data *if* the file was statted.
1716
- planned, not implemented:
1717
path_from_tree_root is the path from the root of the tree.
1719
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1720
allows one to walk a subtree but get paths that are relative to a tree
1722
:return: an iterator over the dirs.
1724
#TODO there is a bit of a smell where the results of the directory-
1725
# summary in this, and the path from the root, may not agree
1726
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1727
# potentially confusing output. We should make this more robust - but
1728
# not at a speed cost. RBC 20060731
1730
_directory = _directory_kind
1731
_listdir = os.listdir
1732
_kind_from_mode = file_kind_from_stat_mode
1733
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1735
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1736
relroot, _, _, _, top = pending.pop()
1738
relprefix = relroot + u'/'
1741
top_slash = top + u'/'
1744
append = dirblock.append
1746
names = sorted(map(decode_filename, _listdir(top)))
1748
if not _is_error_enotdir(e):
1752
abspath = top_slash + name
1753
statvalue = _lstat(abspath)
1754
kind = _kind_from_mode(statvalue.st_mode)
1755
append((relprefix + name, name, kind, statvalue, abspath))
1756
yield (relroot, top), dirblock
1758
# push the user specified dirs from dirblock
1759
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1762
class DirReader(object):
1763
"""An interface for reading directories."""
1765
def top_prefix_to_starting_dir(self, top, prefix=""):
1766
"""Converts top and prefix to a starting dir entry
1768
:param top: A utf8 path
1769
:param prefix: An optional utf8 path to prefix output relative paths
1771
:return: A tuple starting with prefix, and ending with the native
1774
raise NotImplementedError(self.top_prefix_to_starting_dir)
1776
def read_dir(self, prefix, top):
1777
"""Read a specific dir.
1779
:param prefix: A utf8 prefix to be preprended to the path basenames.
1780
:param top: A natively encoded path to read.
1781
:return: A list of the directories contents. Each item contains:
1782
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1784
raise NotImplementedError(self.read_dir)
1787
_selected_dir_reader = None
1790
def _walkdirs_utf8(top, prefix=""):
1791
"""Yield data about all the directories in a tree.
1793
This yields the same information as walkdirs() only each entry is yielded
1794
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1795
are returned as exact byte-strings.
1797
:return: yields a tuple of (dir_info, [file_info])
1798
dir_info is (utf8_relpath, path-from-top)
1799
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1800
if top is an absolute path, path-from-top is also an absolute path.
1801
path-from-top might be unicode or utf8, but it is the correct path to
1802
pass to os functions to affect the file in question. (such as os.lstat)
1804
global _selected_dir_reader
1805
if _selected_dir_reader is None:
1806
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1807
# Win98 doesn't have unicode apis like FindFirstFileW
1808
# TODO: We possibly could support Win98 by falling back to the
1809
# original FindFirstFile, and using TCHAR instead of WCHAR,
1810
# but that gets a bit tricky, and requires custom compiling
1813
from bzrlib._walkdirs_win32 import Win32ReadDir
1814
_selected_dir_reader = Win32ReadDir()
1817
elif _fs_enc in ('utf-8', 'ascii'):
1819
from bzrlib._readdir_pyx import UTF8DirReader
1820
_selected_dir_reader = UTF8DirReader()
1821
except ImportError, e:
1822
failed_to_load_extension(e)
1825
if _selected_dir_reader is None:
1826
# Fallback to the python version
1827
_selected_dir_reader = UnicodeDirReader()
1829
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1830
# But we don't actually uses 1-3 in pending, so set them to None
1831
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1832
read_dir = _selected_dir_reader.read_dir
1833
_directory = _directory_kind
1835
relroot, _, _, _, top = pending[-1].pop()
1838
dirblock = sorted(read_dir(relroot, top))
1839
yield (relroot, top), dirblock
1840
# push the user specified dirs from dirblock
1841
next = [d for d in reversed(dirblock) if d[2] == _directory]
1843
pending.append(next)
1846
class UnicodeDirReader(DirReader):
1847
"""A dir reader for non-utf8 file systems, which transcodes."""
1849
__slots__ = ['_utf8_encode']
1852
self._utf8_encode = codecs.getencoder('utf8')
1854
def top_prefix_to_starting_dir(self, top, prefix=""):
1855
"""See DirReader.top_prefix_to_starting_dir."""
1856
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1858
def read_dir(self, prefix, top):
1859
"""Read a single directory from a non-utf8 file system.
1861
top, and the abspath element in the output are unicode, all other paths
1862
are utf8. Local disk IO is done via unicode calls to listdir etc.
1864
This is currently the fallback code path when the filesystem encoding is
1865
not UTF-8. It may be better to implement an alternative so that we can
1866
safely handle paths that are not properly decodable in the current
1869
See DirReader.read_dir for details.
1871
_utf8_encode = self._utf8_encode
1873
_listdir = os.listdir
1874
_kind_from_mode = file_kind_from_stat_mode
1877
relprefix = prefix + '/'
1880
top_slash = top + u'/'
1883
append = dirblock.append
1884
for name in sorted(_listdir(top)):
1886
name_utf8 = _utf8_encode(name)[0]
1887
except UnicodeDecodeError:
1888
raise errors.BadFilenameEncoding(
1889
_utf8_encode(relprefix)[0] + name, _fs_enc)
1890
abspath = top_slash + name
1891
statvalue = _lstat(abspath)
1892
kind = _kind_from_mode(statvalue.st_mode)
1893
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1897
def copy_tree(from_path, to_path, handlers={}):
1898
"""Copy all of the entries in from_path into to_path.
1900
:param from_path: The base directory to copy.
1901
:param to_path: The target directory. If it does not exist, it will
1903
:param handlers: A dictionary of functions, which takes a source and
1904
destinations for files, directories, etc.
1905
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1906
'file', 'directory', and 'symlink' should always exist.
1907
If they are missing, they will be replaced with 'os.mkdir()',
1908
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1910
# Now, just copy the existing cached tree to the new location
1911
# We use a cheap trick here.
1912
# Absolute paths are prefixed with the first parameter
1913
# relative paths are prefixed with the second.
1914
# So we can get both the source and target returned
1915
# without any extra work.
1917
def copy_dir(source, dest):
1920
def copy_link(source, dest):
1921
"""Copy the contents of a symlink"""
1922
link_to = os.readlink(source)
1923
os.symlink(link_to, dest)
1925
real_handlers = {'file':shutil.copy2,
1926
'symlink':copy_link,
1927
'directory':copy_dir,
1929
real_handlers.update(handlers)
1931
if not os.path.exists(to_path):
1932
real_handlers['directory'](from_path, to_path)
1934
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1935
for relpath, name, kind, st, abspath in entries:
1936
real_handlers[kind](abspath, relpath)
1939
def copy_ownership_from_path(dst, src=None):
1940
"""Copy usr/grp ownership from src file/dir to dst file/dir.
1942
If src is None, the containing directory is used as source. If chown
1943
fails, the error is ignored and a warning is printed.
1945
chown = getattr(os, 'chown', None)
1950
src = os.path.dirname(dst)
1956
chown(dst, s.st_uid, s.st_gid)
1959
'Unable to copy ownership from "%s" to "%s". '
1960
'You may want to set it manually.', src, dst)
1961
trace.log_exception_quietly()
1964
def path_prefix_key(path):
1965
"""Generate a prefix-order path key for path.
1967
This can be used to sort paths in the same way that walkdirs does.
1969
return (dirname(path) , path)
1972
def compare_paths_prefix_order(path_a, path_b):
1973
"""Compare path_a and path_b to generate the same order walkdirs uses."""
1974
key_a = path_prefix_key(path_a)
1975
key_b = path_prefix_key(path_b)
1976
return cmp(key_a, key_b)
1979
_cached_user_encoding = None
1982
def get_user_encoding(use_cache=True):
1983
"""Find out what the preferred user encoding is.
1985
This is generally the encoding that is used for command line parameters
1986
and file contents. This may be different from the terminal encoding
1987
or the filesystem encoding.
1989
:param use_cache: Enable cache for detected encoding.
1990
(This parameter is turned on by default,
1991
and required only for selftesting)
1993
:return: A string defining the preferred user encoding
1995
global _cached_user_encoding
1996
if _cached_user_encoding is not None and use_cache:
1997
return _cached_user_encoding
1999
if sys.platform == 'darwin':
2000
# python locale.getpreferredencoding() always return
2001
# 'mac-roman' on darwin. That's a lie.
2002
sys.platform = 'posix'
2004
if os.environ.get('LANG', None) is None:
2005
# If LANG is not set, we end up with 'ascii', which is bad
2006
# ('mac-roman' is more than ascii), so we set a default which
2007
# will give us UTF-8 (which appears to work in all cases on
2008
# OSX). Users are still free to override LANG of course, as
2009
# long as it give us something meaningful. This work-around
2010
# *may* not be needed with python 3k and/or OSX 10.5, but will
2011
# work with them too -- vila 20080908
2012
os.environ['LANG'] = 'en_US.UTF-8'
2015
sys.platform = 'darwin'
2020
user_encoding = locale.getpreferredencoding()
2021
except locale.Error, e:
2022
sys.stderr.write('bzr: warning: %s\n'
2023
' Could not determine what text encoding to use.\n'
2024
' This error usually means your Python interpreter\n'
2025
' doesn\'t support the locale set by $LANG (%s)\n'
2026
" Continuing with ascii encoding.\n"
2027
% (e, os.environ.get('LANG')))
2028
user_encoding = 'ascii'
2030
# Windows returns 'cp0' to indicate there is no code page. So we'll just
2031
# treat that as ASCII, and not support printing unicode characters to the
2034
# For python scripts run under vim, we get '', so also treat that as ASCII
2035
if user_encoding in (None, 'cp0', ''):
2036
user_encoding = 'ascii'
2040
codecs.lookup(user_encoding)
2042
sys.stderr.write('bzr: warning:'
2043
' unknown encoding %s.'
2044
' Continuing with ascii encoding.\n'
2047
user_encoding = 'ascii'
2050
_cached_user_encoding = user_encoding
2052
return user_encoding
2055
def get_diff_header_encoding():
2056
return get_terminal_encoding()
2059
_message_encoding = None
2062
def get_message_encoding():
2063
"""Return the encoding used for messages
2065
While the message encoding is a general setting it should usually only be
2066
needed for decoding system error strings such as from OSError instances.
2068
global _message_encoding
2069
if _message_encoding is None:
2070
if os.name == "posix":
2072
# This is a process-global setting that can change, but should in
2073
# general just get set once at process startup then be constant.
2074
_message_encoding = locale.getlocale(locale.LC_MESSAGES)[1]
2076
# On windows want the result of GetACP() which this boils down to.
2077
_message_encoding = get_user_encoding()
2078
return _message_encoding or "ascii"
2081
def get_host_name():
2082
"""Return the current unicode host name.
2084
This is meant to be used in place of socket.gethostname() because that
2085
behaves inconsistently on different platforms.
2087
if sys.platform == "win32":
2089
return win32utils.get_host_name()
2092
return socket.gethostname().decode(get_user_encoding())
2095
# We must not read/write any more than 64k at a time from/to a socket so we
2096
# don't risk "no buffer space available" errors on some platforms. Windows in
2097
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
2099
MAX_SOCKET_CHUNK = 64 * 1024
2101
_end_of_stream_errors = [errno.ECONNRESET]
2102
for _eno in ['WSAECONNRESET', 'WSAECONNABORTED']:
2103
_eno = getattr(errno, _eno, None)
2104
if _eno is not None:
2105
_end_of_stream_errors.append(_eno)
2109
def read_bytes_from_socket(sock, report_activity=None,
2110
max_read_size=MAX_SOCKET_CHUNK):
2111
"""Read up to max_read_size of bytes from sock and notify of progress.
2113
Translates "Connection reset by peer" into file-like EOF (return an
2114
empty string rather than raise an error), and repeats the recv if
2115
interrupted by a signal.
2119
bytes = sock.recv(max_read_size)
2120
except socket.error, e:
2122
if eno in _end_of_stream_errors:
2123
# The connection was closed by the other side. Callers expect
2124
# an empty string to signal end-of-stream.
2126
elif eno == errno.EINTR:
2127
# Retry the interrupted recv.
2131
if report_activity is not None:
2132
report_activity(len(bytes), 'read')
2136
def recv_all(socket, count):
2137
"""Receive an exact number of bytes.
2139
Regular Socket.recv() may return less than the requested number of bytes,
2140
depending on what's in the OS buffer. MSG_WAITALL is not available
2141
on all platforms, but this should work everywhere. This will return
2142
less than the requested amount if the remote end closes.
2144
This isn't optimized and is intended mostly for use in testing.
2147
while len(b) < count:
2148
new = read_bytes_from_socket(socket, None, count - len(b))
2155
def send_all(sock, bytes, report_activity=None):
2156
"""Send all bytes on a socket.
2158
Breaks large blocks in smaller chunks to avoid buffering limitations on
2159
some platforms, and catches EINTR which may be thrown if the send is
2160
interrupted by a signal.
2162
This is preferred to socket.sendall(), because it avoids portability bugs
2163
and provides activity reporting.
2165
:param report_activity: Call this as bytes are read, see
2166
Transport._report_activity
2169
byte_count = len(bytes)
2170
while sent_total < byte_count:
2172
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2173
except socket.error, e:
2174
if e.args[0] != errno.EINTR:
2178
report_activity(sent, 'write')
2181
def connect_socket(address):
2182
# Slight variation of the socket.create_connection() function (provided by
2183
# python-2.6) that can fail if getaddrinfo returns an empty list. We also
2184
# provide it for previous python versions. Also, we don't use the timeout
2185
# parameter (provided by the python implementation) so we don't implement
2187
err = socket.error('getaddrinfo returns an empty list')
2188
host, port = address
2189
for res in socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM):
2190
af, socktype, proto, canonname, sa = res
2193
sock = socket.socket(af, socktype, proto)
2197
except socket.error, err:
2198
# 'err' is now the most recent error
2199
if sock is not None:
2204
def dereference_path(path):
2205
"""Determine the real path to a file.
2207
All parent elements are dereferenced. But the file itself is not
2209
:param path: The original path. May be absolute or relative.
2210
:return: the real path *to* the file
2212
parent, base = os.path.split(path)
2213
# The pathjoin for '.' is a workaround for Python bug #1213894.
2214
# (initial path components aren't dereferenced)
2215
return pathjoin(realpath(pathjoin('.', parent)), base)
2218
def supports_mapi():
2219
"""Return True if we can use MAPI to launch a mail client."""
2220
return sys.platform == "win32"
2223
def resource_string(package, resource_name):
2224
"""Load a resource from a package and return it as a string.
2226
Note: Only packages that start with bzrlib are currently supported.
2228
This is designed to be a lightweight implementation of resource
2229
loading in a way which is API compatible with the same API from
2231
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2232
If and when pkg_resources becomes a standard library, this routine
2235
# Check package name is within bzrlib
2236
if package == "bzrlib":
2237
resource_relpath = resource_name
2238
elif package.startswith("bzrlib."):
2239
package = package[len("bzrlib."):].replace('.', os.sep)
2240
resource_relpath = pathjoin(package, resource_name)
2242
raise errors.BzrError('resource package %s not in bzrlib' % package)
2244
# Map the resource to a file and read its contents
2245
base = dirname(bzrlib.__file__)
2246
if getattr(sys, 'frozen', None): # bzr.exe
2247
base = abspath(pathjoin(base, '..', '..'))
2248
f = file(pathjoin(base, resource_relpath), "rU")
2254
def file_kind_from_stat_mode_thunk(mode):
2255
global file_kind_from_stat_mode
2256
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2258
from bzrlib._readdir_pyx import UTF8DirReader
2259
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2260
except ImportError, e:
2261
# This is one time where we won't warn that an extension failed to
2262
# load. The extension is never available on Windows anyway.
2263
from bzrlib._readdir_py import (
2264
_kind_from_mode as file_kind_from_stat_mode
2266
return file_kind_from_stat_mode(mode)
2267
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2269
def file_stat(f, _lstat=os.lstat):
2274
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2275
raise errors.NoSuchFile(f)
2278
def file_kind(f, _lstat=os.lstat):
2279
stat_value = file_stat(f, _lstat)
2280
return file_kind_from_stat_mode(stat_value.st_mode)
2282
def until_no_eintr(f, *a, **kw):
2283
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2285
WARNING: you must be certain that it is safe to retry the call repeatedly
2286
if EINTR does occur. This is typically only true for low-level operations
2287
like os.read. If in any doubt, don't use this.
2289
Keep in mind that this is not a complete solution to EINTR. There is
2290
probably code in the Python standard library and other dependencies that
2291
may encounter EINTR if a signal arrives (and there is signal handler for
2292
that signal). So this function can reduce the impact for IO that bzrlib
2293
directly controls, but it is not a complete solution.
2295
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2299
except (IOError, OSError), e:
2300
if e.errno == errno.EINTR:
2305
@deprecated_function(deprecated_in((2, 2, 0)))
2306
def re_compile_checked(re_string, flags=0, where=""):
2307
"""Return a compiled re, or raise a sensible error.
2309
This should only be used when compiling user-supplied REs.
2311
:param re_string: Text form of regular expression.
2312
:param flags: eg re.IGNORECASE
2313
:param where: Message explaining to the user the context where
2314
it occurred, eg 'log search filter'.
2316
# from https://bugs.launchpad.net/bzr/+bug/251352
2318
re_obj = re.compile(re_string, flags)
2321
except errors.InvalidPattern, e:
2323
where = ' in ' + where
2324
# despite the name 'error' is a type
2325
raise errors.BzrCommandError('Invalid regular expression%s: %s'
2329
if sys.platform == "win32":
2332
return msvcrt.getch()
2337
fd = sys.stdin.fileno()
2338
settings = termios.tcgetattr(fd)
2341
ch = sys.stdin.read(1)
2343
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2346
if sys.platform.startswith('linux'):
2347
def _local_concurrency():
2349
return os.sysconf('SC_NPROCESSORS_ONLN')
2350
except (ValueError, OSError, AttributeError):
2352
elif sys.platform == 'darwin':
2353
def _local_concurrency():
2354
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2355
stdout=subprocess.PIPE).communicate()[0]
2356
elif "bsd" in sys.platform:
2357
def _local_concurrency():
2358
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2359
stdout=subprocess.PIPE).communicate()[0]
2360
elif sys.platform == 'sunos5':
2361
def _local_concurrency():
2362
return subprocess.Popen(['psrinfo', '-p',],
2363
stdout=subprocess.PIPE).communicate()[0]
2364
elif sys.platform == "win32":
2365
def _local_concurrency():
2366
# This appears to return the number of cores.
2367
return os.environ.get('NUMBER_OF_PROCESSORS')
2369
def _local_concurrency():
2374
_cached_local_concurrency = None
2376
def local_concurrency(use_cache=True):
2377
"""Return how many processes can be run concurrently.
2379
Rely on platform specific implementations and default to 1 (one) if
2380
anything goes wrong.
2382
global _cached_local_concurrency
2384
if _cached_local_concurrency is not None and use_cache:
2385
return _cached_local_concurrency
2387
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2388
if concurrency is None:
2390
import multiprocessing
2391
concurrency = multiprocessing.cpu_count()
2392
except (ImportError, NotImplementedError):
2393
# multiprocessing is only available on Python >= 2.6
2394
# and multiprocessing.cpu_count() isn't implemented on all
2397
concurrency = _local_concurrency()
2398
except (OSError, IOError):
2401
concurrency = int(concurrency)
2402
except (TypeError, ValueError):
2405
_cached_concurrency = concurrency
2409
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2410
"""A stream writer that doesn't decode str arguments."""
2412
def __init__(self, encode, stream, errors='strict'):
2413
codecs.StreamWriter.__init__(self, stream, errors)
2414
self.encode = encode
2416
def write(self, object):
2417
if type(object) is str:
2418
self.stream.write(object)
2420
data, _ = self.encode(object, self.errors)
2421
self.stream.write(data)
2423
if sys.platform == 'win32':
2424
def open_file(filename, mode='r', bufsize=-1):
2425
"""This function is used to override the ``open`` builtin.
2427
But it uses O_NOINHERIT flag so the file handle is not inherited by
2428
child processes. Deleting or renaming a closed file opened with this
2429
function is not blocking child processes.
2431
writing = 'w' in mode
2432
appending = 'a' in mode
2433
updating = '+' in mode
2434
binary = 'b' in mode
2437
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2438
# for flags for each modes.
2448
flags |= os.O_WRONLY
2449
flags |= os.O_CREAT | os.O_TRUNC
2454
flags |= os.O_WRONLY
2455
flags |= os.O_CREAT | os.O_APPEND
2460
flags |= os.O_RDONLY
2462
return os.fdopen(os.open(filename, flags), mode, bufsize)
2467
def available_backup_name(base, exists):
2468
"""Find a non-existing backup file name.
2470
This will *not* create anything, this only return a 'free' entry. This
2471
should be used for checking names in a directory below a locked
2472
tree/branch/repo to avoid race conditions. This is LBYL (Look Before You
2473
Leap) and generally discouraged.
2475
:param base: The base name.
2477
:param exists: A callable returning True if the path parameter exists.
2480
name = "%s.~%d~" % (base, counter)
2483
name = "%s.~%d~" % (base, counter)
2487
def set_fd_cloexec(fd):
2488
"""Set a Unix file descriptor's FD_CLOEXEC flag. Do nothing if platform
2489
support for this is not available.
2493
old = fcntl.fcntl(fd, fcntl.F_GETFD)
2494
fcntl.fcntl(fd, fcntl.F_SETFD, old | fcntl.FD_CLOEXEC)
2495
except (ImportError, AttributeError):
2496
# Either the fcntl module or specific constants are not present
2500
def find_executable_on_path(name):
2501
"""Finds an executable on the PATH.
2503
On Windows, this will try to append each extension in the PATHEXT
2504
environment variable to the name, if it cannot be found with the name
2507
:param name: The base name of the executable.
2508
:return: The path to the executable found or None.
2510
path = os.environ.get('PATH')
2513
path = path.split(os.pathsep)
2514
if sys.platform == 'win32':
2515
exts = os.environ.get('PATHEXT', '').split(os.pathsep)
2516
exts = [ext.lower() for ext in exts]
2517
base, ext = os.path.splitext(name)
2519
if ext.lower() not in exts:
2527
f = os.path.join(d, name) + ext
2528
if os.access(f, os.X_OK):
2533
def _posix_is_local_pid_dead(pid):
2534
"""True if pid doesn't correspond to live process on this machine"""
2536
# Special meaning of unix kill: just check if it's there.
2539
if e.errno == errno.ESRCH:
2540
# On this machine, and really not found: as sure as we can be
2543
elif e.errno == errno.EPERM:
2544
# exists, though not ours
2547
mutter("os.kill(%d, 0) failed: %s" % (pid, e))
2548
# Don't really know.
2551
# Exists and our process: not dead.
2554
if sys.platform == "win32":
2555
is_local_pid_dead = win32utils.is_local_pid_dead
2557
is_local_pid_dead = _posix_is_local_pid_dead
2560
def fdatasync(fileno):
2561
"""Flush file contents to disk if possible.
2563
:param fileno: Integer OS file handle.
2564
:raises TransportNotPossible: If flushing to disk is not possible.
2566
fn = getattr(os, 'fdatasync', getattr(os, 'fsync', None))
2571
def ensure_empty_directory_exists(path, exception_class):
2572
"""Make sure a local directory exists and is empty.
2574
If it does not exist, it is created. If it exists and is not empty, an
2575
instance of exception_class is raised.
2580
if e.errno != errno.EEXIST:
2582
if os.listdir(path) != []:
2583
raise exception_class(path)
2586
def is_environment_error(evalue):
2587
"""True if exception instance is due to a process environment issue
2589
This includes OSError and IOError, but also other errors that come from
2590
the operating system or core libraries but are not subclasses of those.
2592
if isinstance(evalue, (EnvironmentError, select.error)):
2594
if sys.platform == "win32" and win32utils._is_pywintypes_error(evalue):
added
removed
bzrlib/osutils.py
