1
# Copyright (C) 2005-2011 Canonical Ltd
1
# Bazaar-NG -- distributed version control
3
# Copyright (C) 2005 by Canonical Ltd
3
5
# This program is free software; you can redistribute it and/or modify
4
6
# it under the terms of the GNU General Public License as published by
5
7
# the Free Software Foundation; either version 2 of the License, or
6
8
# (at your option) any later version.
8
10
# This program is distributed in the hope that it will be useful,
9
11
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
12
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
13
# GNU General Public License for more details.
13
15
# You should have received a copy of the GNU General Public License
14
16
# 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
35
# We need to import both shutil and rmtree as we export the later on posix
36
# and need the former on windows
38
from shutil import rmtree
41
# We need to import both tempfile and mkdtemp as we export the later on posix
42
# and need the former on windows
44
from tempfile import mkdtemp
54
from bzrlib.i18n import gettext
57
from bzrlib.symbol_versioning import (
62
warn as warn_deprecated,
17
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19
import os, types, re, time, errno
20
from stat import S_ISREG, S_ISDIR, S_ISLNK, ST_MODE, ST_SIZE
22
from errors import bailout, BzrError
23
from trace import mutter
72
from bzrlib import symbol_versioning, _fs_enc
75
# Cross platform wall-clock time functionality with decent resolution.
76
# On Linux ``time.clock`` returns only CPU time. On Windows, ``time.time()``
77
# only has a resolution of ~15ms. Note that ``time.clock()`` is not
78
# synchronized with ``time.time()``, this is only meant to be used to find
79
# delta times by subtracting from another call to this function.
80
timer_func = time.time
81
if sys.platform == 'win32':
82
timer_func = time.clock
84
# On win32, O_BINARY is used to indicate the file should
85
# be opened in binary mode, rather than text mode.
86
# On other platforms, O_BINARY doesn't exist, because
87
# they always open in binary mode, so it is okay to
88
# OR with 0 on those platforms.
89
# O_NOINHERIT and O_TEXT exists only on win32 too.
90
O_BINARY = getattr(os, 'O_BINARY', 0)
91
O_TEXT = getattr(os, 'O_TEXT', 0)
92
O_NOINHERIT = getattr(os, 'O_NOINHERIT', 0)
95
def get_unicode_argv():
97
user_encoding = get_user_encoding()
98
return [a.decode(user_encoding) for a in sys.argv[1:]]
99
except UnicodeDecodeError:
100
raise errors.BzrError(gettext("Parameter {0!r} encoding is unsupported by {1} "
101
"application locale.").format(a, user_encoding))
104
26
def make_readonly(filename):
105
27
"""Make a filename read-only."""
106
mod = os.lstat(filename).st_mode
107
if not stat.S_ISLNK(mod):
109
chmod_if_possible(filename, mod)
28
# TODO: probably needs to be fixed for windows
29
mod = os.stat(filename).st_mode
31
os.chmod(filename, mod)
112
34
def make_writable(filename):
113
mod = os.lstat(filename).st_mode
114
if not stat.S_ISLNK(mod):
116
chmod_if_possible(filename, mod)
119
def chmod_if_possible(filename, mode):
120
# Set file mode if that can be safely done.
121
# Sometimes even on unix the filesystem won't allow it - see
122
# https://bugs.launchpad.net/bzr/+bug/606537
124
# It is probably faster to just do the chmod, rather than
125
# doing a stat, and then trying to compare
126
os.chmod(filename, mode)
127
except (IOError, OSError),e:
128
# Permission/access denied seems to commonly happen on smbfs; there's
129
# probably no point warning about it.
130
# <https://bugs.launchpad.net/bzr/+bug/606537>
131
if getattr(e, 'errno') in (errno.EPERM, errno.EACCES):
132
trace.mutter("ignore error on chmod of %r: %r" % (
138
def minimum_path_selection(paths):
139
"""Return the smallset subset of paths which are outside paths.
141
:param paths: A container (and hence not None) of paths.
142
:return: A set of paths sufficient to include everything in paths via
143
is_inside, drawn from the paths parameter.
149
return path.split('/')
150
sorted_paths = sorted(list(paths), key=sort_key)
152
search_paths = [sorted_paths[0]]
153
for path in sorted_paths[1:]:
154
if not is_inside(search_paths[-1], path):
155
# This path is unique, add it
156
search_paths.append(path)
158
return set(search_paths)
35
mod = os.stat(filename).st_mode
37
os.chmod(filename, mod)
40
_QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/_~-])')
165
"""Return a quoted filename filename
167
This previously used backslash quoting, but that works poorly on
169
# TODO: I'm not really sure this is the best format either.x
171
if _QUOTE_RE is None:
172
_QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/\\_~-])')
174
if _QUOTE_RE.search(f):
180
_directory_kind = 'directory'
183
"""Return the current umask"""
184
# Assume that people aren't messing with the umask while running
185
# XXX: This is not thread safe, but there is no way to get the
186
# umask without setting it
194
_directory_kind: "/",
196
'tree-reference': '+',
200
def kind_marker(kind):
202
return _kind_marker_map[kind]
204
# Slightly faster than using .get(, '') when the common case is that
209
lexists = getattr(os.path, 'lexists', None)
213
stat = getattr(os, 'lstat', os.stat)
217
if e.errno == errno.ENOENT:
220
raise errors.BzrError(gettext("lstat/stat of ({0!r}): {1!r}").format(f, e))
223
def fancy_rename(old, new, rename_func, unlink_func):
224
"""A fancy rename, when you don't have atomic rename.
226
:param old: The old path, to rename from
227
:param new: The new path, to rename to
228
:param rename_func: The potentially non-atomic rename function
229
:param unlink_func: A way to delete the target file if the full rename
232
# sftp rename doesn't allow overwriting, so play tricks:
233
base = os.path.basename(new)
234
dirname = os.path.dirname(new)
235
# callers use different encodings for the paths so the following MUST
236
# respect that. We rely on python upcasting to unicode if new is unicode
237
# and keeping a str if not.
238
tmp_name = 'tmp.%s.%.9f.%d.%s' % (base, time.time(),
239
os.getpid(), rand_chars(10))
240
tmp_name = pathjoin(dirname, tmp_name)
242
# Rename the file out of the way, but keep track if it didn't exist
243
# We don't want to grab just any exception
244
# something like EACCES should prevent us from continuing
245
# The downside is that the rename_func has to throw an exception
246
# with an errno = ENOENT, or NoSuchFile
249
rename_func(new, tmp_name)
250
except (errors.NoSuchFile,), e:
253
# RBC 20060103 abstraction leakage: the paramiko SFTP clients rename
254
# function raises an IOError with errno is None when a rename fails.
255
# This then gets caught here.
256
if e.errno not in (None, errno.ENOENT, errno.ENOTDIR):
259
if (getattr(e, 'errno', None) is None
260
or e.errno not in (errno.ENOENT, errno.ENOTDIR)):
269
# This may throw an exception, in which case success will
271
rename_func(old, new)
273
except (IOError, OSError), e:
274
# source and target may be aliases of each other (e.g. on a
275
# case-insensitive filesystem), so we may have accidentally renamed
276
# source by when we tried to rename target
277
failure_exc = sys.exc_info()
278
if (file_existed and e.errno in (None, errno.ENOENT)
279
and old.lower() == new.lower()):
280
# source and target are the same file on a case-insensitive
281
# filesystem, so we don't generate an exception
285
# If the file used to exist, rename it back into place
286
# otherwise just delete it from the tmp location
288
unlink_func(tmp_name)
290
rename_func(tmp_name, new)
291
if failure_exc is not None:
293
raise failure_exc[0], failure_exc[1], failure_exc[2]
298
# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
299
# choke on a Unicode string containing a relative path if
300
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
302
def _posix_abspath(path):
303
# jam 20060426 rather than encoding to fsencoding
304
# copy posixpath.abspath, but use os.getcwdu instead
305
if not posixpath.isabs(path):
306
path = posixpath.join(getcwd(), path)
307
return _posix_normpath(path)
310
def _posix_realpath(path):
311
return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)
314
def _posix_normpath(path):
315
path = posixpath.normpath(path)
316
# Bug 861008: posixpath.normpath() returns a path normalized according to
317
# the POSIX standard, which stipulates (for compatibility reasons) that two
318
# leading slashes must not be simplified to one, and only if there are 3 or
319
# more should they be simplified as one. So we treat the leading 2 slashes
320
# as a special case here by simply removing the first slash, as we consider
321
# that breaking POSIX compatibility for this obscure feature is acceptable.
322
# This is not a paranoid precaution, as we notably get paths like this when
323
# the repo is hosted at the root of the filesystem, i.e. in "/".
324
if path.startswith('//'):
329
def _posix_path_from_environ(key):
330
"""Get unicode path from `key` in environment or None if not present
332
Note that posix systems use arbitrary byte strings for filesystem objects,
333
so a path that raises BadFilenameEncoding here may still be accessible.
335
val = os.environ.get(key, None)
339
return val.decode(_fs_enc)
340
except UnicodeDecodeError:
341
# GZ 2011-12-12:Ideally want to include `key` in the exception message
342
raise errors.BadFilenameEncoding(val, _fs_enc)
345
def _posix_get_home_dir():
346
"""Get the home directory of the current user as a unicode path"""
347
path = posixpath.expanduser("~")
349
return path.decode(_fs_enc)
350
except UnicodeDecodeError:
351
raise errors.BadFilenameEncoding(path, _fs_enc)
354
def _posix_getuser_unicode():
355
"""Get username from environment or password database as unicode"""
356
name = getpass.getuser()
357
user_encoding = get_user_encoding()
359
return name.decode(user_encoding)
360
except UnicodeDecodeError:
361
raise errors.BzrError("Encoding of username %r is unsupported by %s "
362
"application locale." % (name, user_encoding))
365
def _win32_fixdrive(path):
366
"""Force drive letters to be consistent.
368
win32 is inconsistent whether it returns lower or upper case
369
and even if it was consistent the user might type the other
370
so we force it to uppercase
371
running python.exe under cmd.exe return capital C:\\
372
running win32 python inside a cygwin shell returns lowercase c:\\
374
drive, path = ntpath.splitdrive(path)
375
return drive.upper() + path
378
def _win32_abspath(path):
379
# Real ntpath.abspath doesn't have a problem with a unicode cwd
380
return _win32_fixdrive(ntpath.abspath(unicode(path)).replace('\\', '/'))
383
def _win98_abspath(path):
384
"""Return the absolute version of a path.
385
Windows 98 safe implementation (python reimplementation
386
of Win32 API function GetFullPathNameW)
391
# \\HOST\path => //HOST/path
392
# //HOST/path => //HOST/path
393
# path => C:/cwd/path
396
# check for absolute path
397
drive = ntpath.splitdrive(path)[0]
398
if drive == '' and path[:2] not in('//','\\\\'):
400
# we cannot simply os.path.join cwd and path
401
# because os.path.join('C:','/path') produce '/path'
402
# and this is incorrect
403
if path[:1] in ('/','\\'):
404
cwd = ntpath.splitdrive(cwd)[0]
406
path = cwd + '\\' + path
407
return _win32_fixdrive(ntpath.normpath(path).replace('\\', '/'))
410
def _win32_realpath(path):
411
# Real ntpath.realpath doesn't have a problem with a unicode cwd
412
return _win32_fixdrive(ntpath.realpath(unicode(path)).replace('\\', '/'))
415
def _win32_pathjoin(*args):
416
return ntpath.join(*args).replace('\\', '/')
419
def _win32_normpath(path):
420
return _win32_fixdrive(ntpath.normpath(unicode(path)).replace('\\', '/'))
424
return _win32_fixdrive(os.getcwdu().replace('\\', '/'))
427
def _win32_mkdtemp(*args, **kwargs):
428
return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))
431
def _win32_rename(old, new):
432
"""We expect to be able to atomically replace 'new' with old.
434
On win32, if new exists, it must be moved out of the way first,
438
fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
440
if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
441
# If we try to rename a non-existant file onto cwd, we get
442
# EPERM or EACCES instead of ENOENT, this will raise ENOENT
443
# if the old path doesn't exist, sometimes we get EACCES
444
# On Linux, we seem to get EBUSY, on Mac we get EINVAL
450
return unicodedata.normalize('NFC', os.getcwdu())
453
# Default is to just use the python builtins, but these can be rebound on
454
# particular platforms.
455
abspath = _posix_abspath
456
realpath = _posix_realpath
457
pathjoin = os.path.join
458
normpath = _posix_normpath
459
path_from_environ = _posix_path_from_environ
460
_get_home_dir = _posix_get_home_dir
461
getuser_unicode = _posix_getuser_unicode
464
dirname = os.path.dirname
465
basename = os.path.basename
466
split = os.path.split
467
splitext = os.path.splitext
468
# These were already lazily imported into local scope
469
# mkdtemp = tempfile.mkdtemp
470
# rmtree = shutil.rmtree
478
MIN_ABS_PATHLENGTH = 1
481
if sys.platform == 'win32':
482
if win32utils.winver == 'Windows 98':
483
abspath = _win98_abspath
485
abspath = _win32_abspath
486
realpath = _win32_realpath
487
pathjoin = _win32_pathjoin
488
normpath = _win32_normpath
489
getcwd = _win32_getcwd
490
mkdtemp = _win32_mkdtemp
491
rename = _win32_rename
493
from bzrlib import _walkdirs_win32
497
lstat = _walkdirs_win32.lstat
498
fstat = _walkdirs_win32.fstat
499
wrap_stat = _walkdirs_win32.wrap_stat
501
MIN_ABS_PATHLENGTH = 3
503
def _win32_delete_readonly(function, path, excinfo):
504
"""Error handler for shutil.rmtree function [for win32]
505
Helps to remove files and dirs marked as read-only.
507
exception = excinfo[1]
508
if function in (os.remove, os.rmdir) \
509
and isinstance(exception, OSError) \
510
and exception.errno == errno.EACCES:
516
def rmtree(path, ignore_errors=False, onerror=_win32_delete_readonly):
517
"""Replacer for shutil.rmtree: could remove readonly dirs/files"""
518
return shutil.rmtree(path, ignore_errors, onerror)
520
f = win32utils.get_unicode_argv # special function or None
523
path_from_environ = win32utils.get_environ_unicode
524
_get_home_dir = win32utils.get_home_location
525
getuser_unicode = win32utils.get_user_name
527
elif sys.platform == 'darwin':
531
def get_terminal_encoding(trace=False):
532
"""Find the best encoding for printing to the screen.
534
This attempts to check both sys.stdout and sys.stdin to see
535
what encoding they are in, and if that fails it falls back to
536
osutils.get_user_encoding().
537
The problem is that on Windows, locale.getpreferredencoding()
538
is not the same encoding as that used by the console:
539
http://mail.python.org/pipermail/python-list/2003-May/162357.html
541
On my standard US Windows XP, the preferred encoding is
542
cp1252, but the console is cp437
544
:param trace: If True trace the selected encoding via mutter().
546
from bzrlib.trace import mutter
547
output_encoding = getattr(sys.stdout, 'encoding', None)
548
if not output_encoding:
549
input_encoding = getattr(sys.stdin, 'encoding', None)
550
if not input_encoding:
551
output_encoding = get_user_encoding()
553
mutter('encoding stdout as osutils.get_user_encoding() %r',
556
output_encoding = input_encoding
558
mutter('encoding stdout as sys.stdin encoding %r',
562
mutter('encoding stdout as sys.stdout encoding %r', output_encoding)
563
if output_encoding == 'cp0':
564
# invalid encoding (cp0 means 'no codepage' on Windows)
565
output_encoding = get_user_encoding()
567
mutter('cp0 is invalid encoding.'
568
' encoding stdout as osutils.get_user_encoding() %r',
572
codecs.lookup(output_encoding)
574
sys.stderr.write('bzr: warning:'
575
' unknown terminal encoding %s.\n'
576
' Using encoding %s instead.\n'
577
% (output_encoding, get_user_encoding())
579
output_encoding = get_user_encoding()
581
return output_encoding
584
def normalizepath(f):
585
if getattr(os.path, 'realpath', None) is not None:
589
[p,e] = os.path.split(f)
590
if e == "" or e == "." or e == "..":
593
return pathjoin(F(p), e)
42
"""Return shell-quoted filename"""
43
## We could be a bit more terse by using double-quotes etc
44
f = _QUOTE_RE.sub(r'\\\1', f)
51
mode = os.lstat(f)[ST_MODE]
59
raise BzrError("can't handle file kind with mode %o of %r" % (mode, f))
597
64
"""True if f is an accessible directory."""
599
return stat.S_ISDIR(os.lstat(f)[stat.ST_MODE])
66
return S_ISDIR(os.lstat(f)[ST_MODE])
605
73
"""True if f is a regular file."""
607
return stat.S_ISREG(os.lstat(f)[stat.ST_MODE])
75
return S_ISREG(os.lstat(f)[ST_MODE])
612
"""True if f is a symlink."""
80
def pumpfile(fromfile, tofile):
81
"""Copy contents of one file to another."""
82
tofile.write(fromfile.read())
86
"""Return a new UUID"""
88
## XXX: Could alternatively read /proc/sys/kernel/random/uuid on
89
## Linux, but we need something portable for other systems;
90
## preferably an implementation in Python.
614
return stat.S_ISLNK(os.lstat(f)[stat.ST_MODE])
618
def is_inside(dir, fname):
619
"""True if fname is inside dir.
621
The parameters should typically be passed to osutils.normpath first, so
622
that . and .. and repeated slashes are eliminated, and the separators
623
are canonical for the platform.
625
The empty string as a dir name is taken as top-of-tree and matches
628
# XXX: Most callers of this can actually do something smarter by
629
# looking at the inventory
639
return fname.startswith(dir)
642
def is_inside_any(dir_list, fname):
643
"""True if fname is inside any of given dirs."""
644
for dirname in dir_list:
645
if is_inside(dirname, fname):
650
def is_inside_or_parent_of_any(dir_list, fname):
651
"""True if fname is a child or a parent of any of the given files."""
652
for dirname in dir_list:
653
if is_inside(dirname, fname) or is_inside(fname, dirname):
658
def pumpfile(from_file, to_file, read_length=-1, buff_size=32768,
659
report_activity=None, direction='read'):
660
"""Copy contents of one file to another.
662
The read_length can either be -1 to read to end-of-file (EOF) or
663
it can specify the maximum number of bytes to read.
665
The buff_size represents the maximum size for each read operation
666
performed on from_file.
668
:param report_activity: Call this as bytes are read, see
669
Transport._report_activity
670
:param direction: Will be passed to report_activity
672
:return: The number of bytes copied.
676
# read specified number of bytes
678
while read_length > 0:
679
num_bytes_to_read = min(read_length, buff_size)
681
block = from_file.read(num_bytes_to_read)
685
if report_activity is not None:
686
report_activity(len(block), direction)
689
actual_bytes_read = len(block)
690
read_length -= actual_bytes_read
691
length += actual_bytes_read
92
return chomp(file('/proc/sys/kernel/random/uuid').readline())
94
return chomp(os.popen('uuidgen').readline())
98
if s and (s[-1] == '\n'):
695
block = from_file.read(buff_size)
699
if report_activity is not None:
700
report_activity(len(block), direction)
706
def pump_string_file(bytes, file_handle, segment_size=None):
707
"""Write bytes to file_handle in many smaller writes.
709
:param bytes: The string to write.
710
:param file_handle: The file to write to.
712
# Write data in chunks rather than all at once, because very large
713
# writes fail on some platforms (e.g. Windows with SMB mounted
716
segment_size = 5242880 # 5MB
717
segments = range(len(bytes) / segment_size + 1)
718
write = file_handle.write
719
for segment_index in segments:
720
segment = buffer(bytes, segment_index * segment_size, segment_size)
724
def file_iterator(input_file, readsize=32768):
726
b = input_file.read(readsize)
733
"""Calculate the hexdigest of an open file.
735
The file cursor should be already at the start.
747
def size_sha_file(f):
748
"""Calculate the size and hexdigest of an open file.
750
The file cursor should be already at the start and
751
the caller is responsible for closing the file afterwards.
762
return size, s.hexdigest()
765
def sha_file_by_name(fname):
766
"""Calculate the SHA1 of a file by reading the full text"""
768
f = os.open(fname, os.O_RDONLY | O_BINARY | O_NOINHERIT)
771
b = os.read(f, 1<<16)
779
def sha_strings(strings, _factory=sha):
780
"""Return the sha-1 of concatenation of strings"""
782
map(s.update, strings)
786
def sha_string(f, _factory=sha):
787
return _factory(f).hexdigest()
106
## TODO: Maybe read in chunks to handle big files
107
if hasattr(f, 'tell'):
790
122
def fingerprint_file(f):
792
return {'size': len(b),
793
'sha1': sha(b).hexdigest()}
128
return {'size': size,
129
'sha1': s.hexdigest()}
133
"""Return per-user configuration directory.
135
By default this is ~/.bzr.conf/
137
TODO: Global option --config-dir to override this.
139
return os.path.expanduser("~/.bzr.conf")
143
"""Calculate automatic user identification.
145
Returns (realname, email).
147
Only used when none is set in the environment or the id file.
149
This previously used the FQDN as the default domain, but that can
150
be very slow on machines where DNS is broken. So now we simply
155
# XXX: Any good way to get real user name on win32?
160
w = pwd.getpwuid(uid)
161
gecos = w.pw_gecos.decode(bzrlib.user_encoding)
162
username = w.pw_name.decode(bzrlib.user_encoding)
163
comma = gecos.find(',')
167
realname = gecos[:comma]
173
realname = username = getpass.getuser().decode(bzrlib.user_encoding)
175
return realname, (username + '@' + socket.gethostname())
179
"""Return the full user id from a file or environment variable.
181
TODO: Allow taking this from a file in the branch directory too
182
for per-branch ids."""
183
v = os.environ.get('BZREMAIL')
185
return v.decode(bzrlib.user_encoding)
188
return (open(os.path.join(config_dir(), "email"))
190
.decode(bzrlib.user_encoding)
193
if e.errno != errno.ENOENT:
196
v = os.environ.get('EMAIL')
198
return v.decode(bzrlib.user_encoding)
204
"""Return email-style username.
206
Something similar to 'Martin Pool <mbp@sourcefrog.net>'
208
TODO: Check it's reasonably well-formed.
214
name, email = _auto_user_id()
216
return '%s <%s>' % (name, email)
221
_EMAIL_RE = re.compile(r'[\w+.-]+@[\w+.-]+')
223
"""Return just the email component of a username."""
226
m = _EMAIL_RE.search(e)
228
bailout("%r doesn't seem to contain a reasonable email address" % e)
231
return _auto_user_id()[1]
796
235
def compare_files(a, b):
797
236
"""Returns true if equal in contents"""
237
# TODO: don't read the whole thing in one go.
800
240
ai = a.read(BUFSIZE)
1008
raise errors.BzrError(gettext("sorry, %r not allowed in path") % f)
330
bailout("sorry, %r not allowed in path" % f)
1009
331
elif (f == '.') or (f == ''):
1016
337
def joinpath(p):
338
assert isinstance(p, list)
1018
if (f == '..') or (f is None) or (f == ''):
1019
raise errors.BzrError(gettext("sorry, %r not allowed in path") % f)
1023
def parent_directories(filename):
1024
"""Return the list of parent directories, deepest first.
1026
For example, parent_directories("a/b/c") -> ["a/b", "a"].
1029
parts = splitpath(dirname(filename))
1031
parents.append(joinpath(parts))
1036
_extension_load_failures = []
1039
def failed_to_load_extension(exception):
1040
"""Handle failing to load a binary extension.
1042
This should be called from the ImportError block guarding the attempt to
1043
import the native extension. If this function returns, the pure-Python
1044
implementation should be loaded instead::
1047
>>> import bzrlib._fictional_extension_pyx
1048
>>> except ImportError, e:
1049
>>> bzrlib.osutils.failed_to_load_extension(e)
1050
>>> import bzrlib._fictional_extension_py
1052
# NB: This docstring is just an example, not a doctest, because doctest
1053
# currently can't cope with the use of lazy imports in this namespace --
1056
# This currently doesn't report the failure at the time it occurs, because
1057
# they tend to happen very early in startup when we can't check config
1058
# files etc, and also we want to report all failures but not spam the user
1060
exception_str = str(exception)
1061
if exception_str not in _extension_load_failures:
1062
trace.mutter("failed to load compiled extension: %s" % exception_str)
1063
_extension_load_failures.append(exception_str)
1066
def report_extension_load_failures():
1067
if not _extension_load_failures:
1069
if config.GlobalStack().get('ignore_missing_extensions'):
1071
# the warnings framework should by default show this only once
1072
from bzrlib.trace import warning
1074
"bzr: warning: some compiled extensions could not be loaded; "
1075
"see <https://answers.launchpad.net/bzr/+faq/703>")
1076
# we no longer show the specific missing extensions here, because it makes
1077
# the message too long and scary - see
1078
# https://bugs.launchpad.net/bzr/+bug/430529
1082
from bzrlib._chunks_to_lines_pyx import chunks_to_lines
1083
except ImportError, e:
1084
failed_to_load_extension(e)
1085
from bzrlib._chunks_to_lines_py import chunks_to_lines
1089
"""Split s into lines, but without removing the newline characters."""
1090
# Trivially convert a fulltext into a 'chunked' representation, and let
1091
# chunks_to_lines do the heavy lifting.
1092
if isinstance(s, str):
1093
# chunks_to_lines only supports 8-bit strings
1094
return chunks_to_lines([s])
1096
return _split_lines(s)
1099
def _split_lines(s):
1100
"""Split s into lines, but without removing the newline characters.
1102
This supports Unicode or plain string objects.
1104
lines = s.split('\n')
1105
result = [line + '\n' for line in lines[:-1]]
1107
result.append(lines[-1])
1111
def hardlinks_good():
1112
return sys.platform not in ('win32', 'cygwin', 'darwin')
1115
def link_or_copy(src, dest):
1116
"""Hardlink a file, or copy it if it can't be hardlinked."""
1117
if not hardlinks_good():
1118
shutil.copyfile(src, dest)
1122
except (OSError, IOError), e:
1123
if e.errno != errno.EXDEV:
1125
shutil.copyfile(src, dest)
1128
def delete_any(path):
1129
"""Delete a file, symlink or directory.
1131
Will delete even if readonly.
1134
_delete_file_or_dir(path)
1135
except (OSError, IOError), e:
1136
if e.errno in (errno.EPERM, errno.EACCES):
1137
# make writable and try again
1140
except (OSError, IOError):
1142
_delete_file_or_dir(path)
1147
def _delete_file_or_dir(path):
1148
# Look Before You Leap (LBYL) is appropriate here instead of Easier to Ask for
1149
# Forgiveness than Permission (EAFP) because:
1150
# - root can damage a solaris file system by using unlink,
1151
# - unlink raises different exceptions on different OSes (linux: EISDIR, win32:
1152
# EACCES, OSX: EPERM) when invoked on a directory.
1153
if isdir(path): # Takes care of symlinks
1160
if getattr(os, 'symlink', None) is not None:
1166
def has_hardlinks():
1167
if getattr(os, 'link', None) is not None:
1173
def host_os_dereferences_symlinks():
1174
return (has_symlinks()
1175
and sys.platform not in ('cygwin', 'win32'))
1178
def readlink(abspath):
1179
"""Return a string representing the path to which the symbolic link points.
1181
:param abspath: The link absolute unicode path.
1183
This his guaranteed to return the symbolic link in unicode in all python
1186
link = abspath.encode(_fs_enc)
1187
target = os.readlink(link)
1188
target = target.decode(_fs_enc)
1192
def contains_whitespace(s):
1193
"""True if there are any whitespace characters in s."""
1194
# string.whitespace can include '\xa0' in certain locales, because it is
1195
# considered "non-breaking-space" as part of ISO-8859-1. But it
1196
# 1) Isn't a breaking whitespace
1197
# 2) Isn't one of ' \t\r\n' which are characters we sometimes use as
1199
# 3) '\xa0' isn't unicode safe since it is >128.
1201
# This should *not* be a unicode set of characters in case the source
1202
# string is not a Unicode string. We can auto-up-cast the characters since
1203
# they are ascii, but we don't want to auto-up-cast the string in case it
1205
for ch in ' \t\n\r\v\f':
1212
def contains_linebreaks(s):
1213
"""True if there is any vertical whitespace in s."""
1221
def relpath(base, path):
1222
"""Return path relative to base, or raise PathNotChild exception.
1224
The path may be either an absolute path or a path relative to the
1225
current working directory.
1227
os.path.commonprefix (python2.4) has a bad bug that it works just
1228
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
1229
avoids that problem.
1231
NOTE: `base` should not have a trailing slash otherwise you'll get
1232
PathNotChild exceptions regardless of `path`.
1235
if len(base) < MIN_ABS_PATHLENGTH:
1236
# must have space for e.g. a drive letter
1237
raise ValueError(gettext('%r is too short to calculate a relative path')
1245
if len(head) <= len(base) and head != base:
1246
raise errors.PathNotChild(rp, base)
1249
head, tail = split(head)
1254
return pathjoin(*reversed(s))
1259
def _cicp_canonical_relpath(base, path):
1260
"""Return the canonical path relative to base.
1262
Like relpath, but on case-insensitive-case-preserving file-systems, this
1263
will return the relpath as stored on the file-system rather than in the
1264
case specified in the input string, for all existing portions of the path.
1266
This will cause O(N) behaviour if called for every path in a tree; if you
1267
have a number of paths to convert, you should use canonical_relpaths().
1269
# TODO: it should be possible to optimize this for Windows by using the
1270
# win32 API FindFiles function to look for the specified name - but using
1271
# os.listdir() still gives us the correct, platform agnostic semantics in
1274
rel = relpath(base, path)
1275
# '.' will have been turned into ''
1279
abs_base = abspath(base)
1281
_listdir = os.listdir
1283
# use an explicit iterator so we can easily consume the rest on early exit.
1284
bit_iter = iter(rel.split('/'))
1285
for bit in bit_iter:
1288
next_entries = _listdir(current)
1289
except OSError: # enoent, eperm, etc
1290
# We can't find this in the filesystem, so just append the
1292
current = pathjoin(current, bit, *list(bit_iter))
1294
for look in next_entries:
1295
if lbit == look.lower():
1296
current = pathjoin(current, look)
1299
# got to the end, nothing matched, so we just return the
1300
# non-existing bits as they were specified (the filename may be
1301
# the target of a move, for example).
1302
current = pathjoin(current, bit, *list(bit_iter))
1304
return current[len(abs_base):].lstrip('/')
1306
# XXX - TODO - we need better detection/integration of case-insensitive
1307
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
1308
# filesystems), for example, so could probably benefit from the same basic
1309
# support there. For now though, only Windows and OSX get that support, and
1310
# they get it for *all* file-systems!
1311
if sys.platform in ('win32', 'darwin'):
1312
canonical_relpath = _cicp_canonical_relpath
1314
canonical_relpath = relpath
1316
def canonical_relpaths(base, paths):
1317
"""Create an iterable to canonicalize a sequence of relative paths.
1319
The intent is for this implementation to use a cache, vastly speeding
1320
up multiple transformations in the same directory.
1322
# but for now, we haven't optimized...
1323
return [canonical_relpath(base, p) for p in paths]
1326
def decode_filename(filename):
1327
"""Decode the filename using the filesystem encoding
1329
If it is unicode, it is returned.
1330
Otherwise it is decoded from the the filesystem's encoding. If decoding
1331
fails, a errors.BadFilenameEncoding exception is raised.
1333
if type(filename) is unicode:
1336
return filename.decode(_fs_enc)
1337
except UnicodeDecodeError:
1338
raise errors.BadFilenameEncoding(filename, _fs_enc)
1341
def safe_unicode(unicode_or_utf8_string):
1342
"""Coerce unicode_or_utf8_string into unicode.
1344
If it is unicode, it is returned.
1345
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1346
wrapped in a BzrBadParameterNotUnicode exception.
1348
if isinstance(unicode_or_utf8_string, unicode):
1349
return unicode_or_utf8_string
1351
return unicode_or_utf8_string.decode('utf8')
1352
except UnicodeDecodeError:
1353
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1356
def safe_utf8(unicode_or_utf8_string):
1357
"""Coerce unicode_or_utf8_string to a utf8 string.
1359
If it is a str, it is returned.
1360
If it is Unicode, it is encoded into a utf-8 string.
1362
if isinstance(unicode_or_utf8_string, str):
1363
# TODO: jam 20070209 This is overkill, and probably has an impact on
1364
# performance if we are dealing with lots of apis that want a
1367
# Make sure it is a valid utf-8 string
1368
unicode_or_utf8_string.decode('utf-8')
1369
except UnicodeDecodeError:
1370
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1371
return unicode_or_utf8_string
1372
return unicode_or_utf8_string.encode('utf-8')
1375
_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
1376
' Revision id generators should be creating utf8'
1380
def safe_revision_id(unicode_or_utf8_string, warn=True):
1381
"""Revision ids should now be utf8, but at one point they were unicode.
1383
:param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
1385
:param warn: Functions that are sanitizing user data can set warn=False
1386
:return: None or a utf8 revision id.
1388
if (unicode_or_utf8_string is None
1389
or unicode_or_utf8_string.__class__ == str):
1390
return unicode_or_utf8_string
1392
symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
1394
return cache_utf8.encode(unicode_or_utf8_string)
1397
_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
1398
' generators should be creating utf8 file ids.')
1401
def safe_file_id(unicode_or_utf8_string, warn=True):
1402
"""File ids should now be utf8, but at one point they were unicode.
1404
This is the same as safe_utf8, except it uses the cached encode functions
1405
to save a little bit of performance.
1407
:param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
1409
:param warn: Functions that are sanitizing user data can set warn=False
1410
:return: None or a utf8 file id.
1412
if (unicode_or_utf8_string is None
1413
or unicode_or_utf8_string.__class__ == str):
1414
return unicode_or_utf8_string
1416
symbol_versioning.warn(_file_id_warning, DeprecationWarning,
1418
return cache_utf8.encode(unicode_or_utf8_string)
1421
_platform_normalizes_filenames = False
1422
if sys.platform == 'darwin':
1423
_platform_normalizes_filenames = True
1426
def normalizes_filenames():
1427
"""Return True if this platform normalizes unicode filenames.
1431
return _platform_normalizes_filenames
1434
def _accessible_normalized_filename(path):
1435
"""Get the unicode normalized path, and if you can access the file.
1437
On platforms where the system normalizes filenames (Mac OSX),
1438
you can access a file by any path which will normalize correctly.
1439
On platforms where the system does not normalize filenames
1440
(everything else), you have to access a file by its exact path.
1442
Internally, bzr only supports NFC normalization, since that is
1443
the standard for XML documents.
1445
So return the normalized path, and a flag indicating if the file
1446
can be accessed by that path.
1449
return unicodedata.normalize('NFC', unicode(path)), True
1452
def _inaccessible_normalized_filename(path):
1453
__doc__ = _accessible_normalized_filename.__doc__
1455
normalized = unicodedata.normalize('NFC', unicode(path))
1456
return normalized, normalized == path
1459
if _platform_normalizes_filenames:
1460
normalized_filename = _accessible_normalized_filename
1462
normalized_filename = _inaccessible_normalized_filename
1465
def set_signal_handler(signum, handler, restart_syscall=True):
1466
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1467
on platforms that support that.
1469
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1470
automatically restart (by calling `signal.siginterrupt(signum,
1471
False)`). May be ignored if the feature is not available on this
1472
platform or Python version.
1476
siginterrupt = signal.siginterrupt
1478
# This python implementation doesn't provide signal support, hence no
1481
except AttributeError:
1482
# siginterrupt doesn't exist on this platform, or for this version
1484
siginterrupt = lambda signum, flag: None
1486
def sig_handler(*args):
1487
# Python resets the siginterrupt flag when a signal is
1488
# received. <http://bugs.python.org/issue8354>
1489
# As a workaround for some cases, set it back the way we want it.
1490
siginterrupt(signum, False)
1491
# Now run the handler function passed to set_signal_handler.
1494
sig_handler = handler
1495
old_handler = signal.signal(signum, sig_handler)
1497
siginterrupt(signum, False)
1501
default_terminal_width = 80
1502
"""The default terminal width for ttys.
1504
This is defined so that higher levels can share a common fallback value when
1505
terminal_width() returns None.
1508
# Keep some state so that terminal_width can detect if _terminal_size has
1509
# returned a different size since the process started. See docstring and
1510
# comments of terminal_width for details.
1511
# _terminal_size_state has 3 possible values: no_data, unchanged, and changed.
1512
_terminal_size_state = 'no_data'
1513
_first_terminal_size = None
1515
def terminal_width():
1516
"""Return terminal width.
1518
None is returned if the width can't established precisely.
1521
- if BZR_COLUMNS is set, returns its value
1522
- if there is no controlling terminal, returns None
1523
- query the OS, if the queried size has changed since the last query,
1525
- if COLUMNS is set, returns its value,
1526
- if the OS has a value (even though it's never changed), return its value.
1528
From there, we need to query the OS to get the size of the controlling
1531
On Unices we query the OS by:
1532
- get termios.TIOCGWINSZ
1533
- if an error occurs or a negative value is obtained, returns None
1535
On Windows we query the OS by:
1536
- win32utils.get_console_size() decides,
1537
- returns None on error (provided default value)
1539
# Note to implementors: if changing the rules for determining the width,
1540
# make sure you've considered the behaviour in these cases:
1541
# - M-x shell in emacs, where $COLUMNS is set and TIOCGWINSZ returns 0,0.
1542
# - bzr log | less, in bash, where $COLUMNS not set and TIOCGWINSZ returns
1544
# - (add more interesting cases here, if you find any)
1545
# Some programs implement "Use $COLUMNS (if set) until SIGWINCH occurs",
1546
# but we don't want to register a signal handler because it is impossible
1547
# to do so without risking EINTR errors in Python <= 2.6.5 (see
1548
# <http://bugs.python.org/issue8354>). Instead we check TIOCGWINSZ every
1549
# time so we can notice if the reported size has changed, which should have
1552
# If BZR_COLUMNS is set, take it, user is always right
1553
# Except if they specified 0 in which case, impose no limit here
1555
width = int(os.environ['BZR_COLUMNS'])
1556
except (KeyError, ValueError):
1558
if width is not None:
1564
isatty = getattr(sys.stdout, 'isatty', None)
1565
if isatty is None or not isatty():
1566
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1570
width, height = os_size = _terminal_size(None, None)
1571
global _first_terminal_size, _terminal_size_state
1572
if _terminal_size_state == 'no_data':
1573
_first_terminal_size = os_size
1574
_terminal_size_state = 'unchanged'
1575
elif (_terminal_size_state == 'unchanged' and
1576
_first_terminal_size != os_size):
1577
_terminal_size_state = 'changed'
1579
# If the OS claims to know how wide the terminal is, and this value has
1580
# ever changed, use that.
1581
if _terminal_size_state == 'changed':
1582
if width is not None and width > 0:
1585
# If COLUMNS is set, use it.
1587
return int(os.environ['COLUMNS'])
1588
except (KeyError, ValueError):
1591
# Finally, use an unchanged size from the OS, if we have one.
1592
if _terminal_size_state == 'unchanged':
1593
if width is not None and width > 0:
1596
# The width could not be determined.
1600
def _win32_terminal_size(width, height):
1601
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1602
return width, height
1605
def _ioctl_terminal_size(width, height):
1607
import struct, fcntl, termios
1608
s = struct.pack('HHHH', 0, 0, 0, 0)
1609
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1610
height, width = struct.unpack('HHHH', x)[0:2]
1611
except (IOError, AttributeError):
1613
return width, height
1615
_terminal_size = None
1616
"""Returns the terminal size as (width, height).
1618
:param width: Default value for width.
1619
:param height: Default value for height.
1621
This is defined specifically for each OS and query the size of the controlling
1622
terminal. If any error occurs, the provided default values should be returned.
1624
if sys.platform == 'win32':
1625
_terminal_size = _win32_terminal_size
1627
_terminal_size = _ioctl_terminal_size
1630
def supports_executable():
1631
return sys.platform != "win32"
1634
def supports_posix_readonly():
1635
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1637
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1638
directory controls creation/deletion, etc.
1640
And under win32, readonly means that the directory itself cannot be
1641
deleted. The contents of a readonly directory can be changed, unlike POSIX
1642
where files in readonly directories cannot be added, deleted or renamed.
1644
return sys.platform != "win32"
1647
def set_or_unset_env(env_variable, value):
1648
"""Modify the environment, setting or removing the env_variable.
1650
:param env_variable: The environment variable in question
1651
:param value: The value to set the environment to. If None, then
1652
the variable will be removed.
1653
:return: The original value of the environment variable.
1655
orig_val = os.environ.get(env_variable)
1657
if orig_val is not None:
1658
del os.environ[env_variable]
1660
if isinstance(value, unicode):
1661
value = value.encode(get_user_encoding())
1662
os.environ[env_variable] = value
1666
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
1669
def check_legal_path(path):
1670
"""Check whether the supplied path is legal.
1671
This is only required on Windows, so we don't test on other platforms
1674
if sys.platform != "win32":
1676
if _validWin32PathRE.match(path) is None:
1677
raise errors.IllegalPath(path)
1680
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1682
def _is_error_enotdir(e):
1683
"""Check if this exception represents ENOTDIR.
1685
Unfortunately, python is very inconsistent about the exception
1686
here. The cases are:
1687
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1688
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1689
which is the windows error code.
1690
3) Windows, Python2.5 uses errno == EINVAL and
1691
winerror == ERROR_DIRECTORY
1693
:param e: An Exception object (expected to be OSError with an errno
1694
attribute, but we should be able to cope with anything)
1695
:return: True if this represents an ENOTDIR error. False otherwise.
1697
en = getattr(e, 'errno', None)
1698
if (en == errno.ENOTDIR
1699
or (sys.platform == 'win32'
1700
and (en == _WIN32_ERROR_DIRECTORY
1701
or (en == errno.EINVAL
1702
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1708
def walkdirs(top, prefix=""):
1709
"""Yield data about all the directories in a tree.
1711
This yields all the data about the contents of a directory at a time.
1712
After each directory has been yielded, if the caller has mutated the list
1713
to exclude some directories, they are then not descended into.
1715
The data yielded is of the form:
1716
((directory-relpath, directory-path-from-top),
1717
[(relpath, basename, kind, lstat, path-from-top), ...]),
1718
- directory-relpath is the relative path of the directory being returned
1719
with respect to top. prefix is prepended to this.
1720
- directory-path-from-root is the path including top for this directory.
1721
It is suitable for use with os functions.
1722
- relpath is the relative path within the subtree being walked.
1723
- basename is the basename of the path
1724
- kind is the kind of the file now. If unknown then the file is not
1725
present within the tree - but it may be recorded as versioned. See
1727
- lstat is the stat data *if* the file was statted.
1728
- planned, not implemented:
1729
path_from_tree_root is the path from the root of the tree.
1731
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1732
allows one to walk a subtree but get paths that are relative to a tree
1734
:return: an iterator over the dirs.
1736
#TODO there is a bit of a smell where the results of the directory-
1737
# summary in this, and the path from the root, may not agree
1738
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1739
# potentially confusing output. We should make this more robust - but
1740
# not at a speed cost. RBC 20060731
1742
_directory = _directory_kind
1743
_listdir = os.listdir
1744
_kind_from_mode = file_kind_from_stat_mode
1745
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1747
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1748
relroot, _, _, _, top = pending.pop()
1750
relprefix = relroot + u'/'
1753
top_slash = top + u'/'
1756
append = dirblock.append
1758
names = sorted(map(decode_filename, _listdir(top)))
1760
if not _is_error_enotdir(e):
1764
abspath = top_slash + name
1765
statvalue = _lstat(abspath)
1766
kind = _kind_from_mode(statvalue.st_mode)
1767
append((relprefix + name, name, kind, statvalue, abspath))
1768
yield (relroot, top), dirblock
1770
# push the user specified dirs from dirblock
1771
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1774
class DirReader(object):
1775
"""An interface for reading directories."""
1777
def top_prefix_to_starting_dir(self, top, prefix=""):
1778
"""Converts top and prefix to a starting dir entry
1780
:param top: A utf8 path
1781
:param prefix: An optional utf8 path to prefix output relative paths
1783
:return: A tuple starting with prefix, and ending with the native
1786
raise NotImplementedError(self.top_prefix_to_starting_dir)
1788
def read_dir(self, prefix, top):
1789
"""Read a specific dir.
1791
:param prefix: A utf8 prefix to be preprended to the path basenames.
1792
:param top: A natively encoded path to read.
1793
:return: A list of the directories contents. Each item contains:
1794
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1796
raise NotImplementedError(self.read_dir)
1799
_selected_dir_reader = None
1802
def _walkdirs_utf8(top, prefix=""):
1803
"""Yield data about all the directories in a tree.
1805
This yields the same information as walkdirs() only each entry is yielded
1806
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1807
are returned as exact byte-strings.
1809
:return: yields a tuple of (dir_info, [file_info])
1810
dir_info is (utf8_relpath, path-from-top)
1811
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1812
if top is an absolute path, path-from-top is also an absolute path.
1813
path-from-top might be unicode or utf8, but it is the correct path to
1814
pass to os functions to affect the file in question. (such as os.lstat)
1816
global _selected_dir_reader
1817
if _selected_dir_reader is None:
1818
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1819
# Win98 doesn't have unicode apis like FindFirstFileW
1820
# TODO: We possibly could support Win98 by falling back to the
1821
# original FindFirstFile, and using TCHAR instead of WCHAR,
1822
# but that gets a bit tricky, and requires custom compiling
1825
from bzrlib._walkdirs_win32 import Win32ReadDir
1826
_selected_dir_reader = Win32ReadDir()
1829
elif _fs_enc in ('utf-8', 'ascii'):
1831
from bzrlib._readdir_pyx import UTF8DirReader
1832
_selected_dir_reader = UTF8DirReader()
1833
except ImportError, e:
1834
failed_to_load_extension(e)
1837
if _selected_dir_reader is None:
1838
# Fallback to the python version
1839
_selected_dir_reader = UnicodeDirReader()
1841
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1842
# But we don't actually uses 1-3 in pending, so set them to None
1843
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1844
read_dir = _selected_dir_reader.read_dir
1845
_directory = _directory_kind
1847
relroot, _, _, _, top = pending[-1].pop()
1850
dirblock = sorted(read_dir(relroot, top))
1851
yield (relroot, top), dirblock
1852
# push the user specified dirs from dirblock
1853
next = [d for d in reversed(dirblock) if d[2] == _directory]
1855
pending.append(next)
1858
class UnicodeDirReader(DirReader):
1859
"""A dir reader for non-utf8 file systems, which transcodes."""
1861
__slots__ = ['_utf8_encode']
1864
self._utf8_encode = codecs.getencoder('utf8')
1866
def top_prefix_to_starting_dir(self, top, prefix=""):
1867
"""See DirReader.top_prefix_to_starting_dir."""
1868
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1870
def read_dir(self, prefix, top):
1871
"""Read a single directory from a non-utf8 file system.
1873
top, and the abspath element in the output are unicode, all other paths
1874
are utf8. Local disk IO is done via unicode calls to listdir etc.
1876
This is currently the fallback code path when the filesystem encoding is
1877
not UTF-8. It may be better to implement an alternative so that we can
1878
safely handle paths that are not properly decodable in the current
1881
See DirReader.read_dir for details.
1883
_utf8_encode = self._utf8_encode
1885
_listdir = os.listdir
1886
_kind_from_mode = file_kind_from_stat_mode
1889
relprefix = prefix + '/'
1892
top_slash = top + u'/'
1895
append = dirblock.append
1896
for name in sorted(_listdir(top)):
1898
name_utf8 = _utf8_encode(name)[0]
1899
except UnicodeDecodeError:
1900
raise errors.BadFilenameEncoding(
1901
_utf8_encode(relprefix)[0] + name, _fs_enc)
1902
abspath = top_slash + name
1903
statvalue = _lstat(abspath)
1904
kind = _kind_from_mode(statvalue.st_mode)
1905
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1909
def copy_tree(from_path, to_path, handlers={}):
1910
"""Copy all of the entries in from_path into to_path.
1912
:param from_path: The base directory to copy.
1913
:param to_path: The target directory. If it does not exist, it will
1915
:param handlers: A dictionary of functions, which takes a source and
1916
destinations for files, directories, etc.
1917
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1918
'file', 'directory', and 'symlink' should always exist.
1919
If they are missing, they will be replaced with 'os.mkdir()',
1920
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1922
# Now, just copy the existing cached tree to the new location
1923
# We use a cheap trick here.
1924
# Absolute paths are prefixed with the first parameter
1925
# relative paths are prefixed with the second.
1926
# So we can get both the source and target returned
1927
# without any extra work.
1929
def copy_dir(source, dest):
1932
def copy_link(source, dest):
1933
"""Copy the contents of a symlink"""
1934
link_to = os.readlink(source)
1935
os.symlink(link_to, dest)
1937
real_handlers = {'file':shutil.copy2,
1938
'symlink':copy_link,
1939
'directory':copy_dir,
1941
real_handlers.update(handlers)
1943
if not os.path.exists(to_path):
1944
real_handlers['directory'](from_path, to_path)
1946
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1947
for relpath, name, kind, st, abspath in entries:
1948
real_handlers[kind](abspath, relpath)
1951
def copy_ownership_from_path(dst, src=None):
1952
"""Copy usr/grp ownership from src file/dir to dst file/dir.
1954
If src is None, the containing directory is used as source. If chown
1955
fails, the error is ignored and a warning is printed.
1957
chown = getattr(os, 'chown', None)
1962
src = os.path.dirname(dst)
1968
chown(dst, s.st_uid, s.st_gid)
1971
'Unable to copy ownership from "%s" to "%s". '
1972
'You may want to set it manually.', src, dst)
1973
trace.log_exception_quietly()
1976
def path_prefix_key(path):
1977
"""Generate a prefix-order path key for path.
1979
This can be used to sort paths in the same way that walkdirs does.
1981
return (dirname(path) , path)
1984
def compare_paths_prefix_order(path_a, path_b):
1985
"""Compare path_a and path_b to generate the same order walkdirs uses."""
1986
key_a = path_prefix_key(path_a)
1987
key_b = path_prefix_key(path_b)
1988
return cmp(key_a, key_b)
1991
_cached_user_encoding = None
1994
def get_user_encoding(use_cache=DEPRECATED_PARAMETER):
1995
"""Find out what the preferred user encoding is.
1997
This is generally the encoding that is used for command line parameters
1998
and file contents. This may be different from the terminal encoding
1999
or the filesystem encoding.
2001
:return: A string defining the preferred user encoding
2003
global _cached_user_encoding
2004
if deprecated_passed(use_cache):
2005
warn_deprecated("use_cache should only have been used for tests",
2006
DeprecationWarning, stacklevel=2)
2007
if _cached_user_encoding is not None:
2008
return _cached_user_encoding
2010
if os.name == 'posix' and getattr(locale, 'CODESET', None) is not None:
2011
# Use the existing locale settings and call nl_langinfo directly
2012
# rather than going through getpreferredencoding. This avoids
2013
# <http://bugs.python.org/issue6202> on OSX Python 2.6 and the
2014
# possibility of the setlocale call throwing an error.
2015
user_encoding = locale.nl_langinfo(locale.CODESET)
2017
# GZ 2011-12-19: On windows could call GetACP directly instead.
2018
user_encoding = locale.getpreferredencoding(False)
2021
user_encoding = codecs.lookup(user_encoding).name
2023
if user_encoding not in ("", "cp0"):
2024
sys.stderr.write('bzr: warning:'
2025
' unknown encoding %s.'
2026
' Continuing with ascii encoding.\n'
2029
user_encoding = 'ascii'
2031
# Get 'ascii' when setlocale has not been called or LANG=C or unset.
2032
if user_encoding == 'ascii':
2033
if sys.platform == 'darwin':
2034
# OSX is special-cased in Python to have a UTF-8 filesystem
2035
# encoding and previously had LANG set here if not present.
2036
user_encoding = 'utf-8'
2037
# GZ 2011-12-19: Maybe UTF-8 should be the default in this case
2038
# for some other posix platforms as well.
2040
_cached_user_encoding = user_encoding
2041
return user_encoding
2044
def get_diff_header_encoding():
2045
return get_terminal_encoding()
2048
def get_host_name():
2049
"""Return the current unicode host name.
2051
This is meant to be used in place of socket.gethostname() because that
2052
behaves inconsistently on different platforms.
2054
if sys.platform == "win32":
2055
return win32utils.get_host_name()
2058
return socket.gethostname().decode(get_user_encoding())
2061
# We must not read/write any more than 64k at a time from/to a socket so we
2062
# don't risk "no buffer space available" errors on some platforms. Windows in
2063
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
2065
MAX_SOCKET_CHUNK = 64 * 1024
2067
_end_of_stream_errors = [errno.ECONNRESET]
2068
for _eno in ['WSAECONNRESET', 'WSAECONNABORTED']:
2069
_eno = getattr(errno, _eno, None)
2070
if _eno is not None:
2071
_end_of_stream_errors.append(_eno)
2075
def read_bytes_from_socket(sock, report_activity=None,
2076
max_read_size=MAX_SOCKET_CHUNK):
2077
"""Read up to max_read_size of bytes from sock and notify of progress.
2079
Translates "Connection reset by peer" into file-like EOF (return an
2080
empty string rather than raise an error), and repeats the recv if
2081
interrupted by a signal.
2085
bytes = sock.recv(max_read_size)
2086
except socket.error, e:
2088
if eno in _end_of_stream_errors:
2089
# The connection was closed by the other side. Callers expect
2090
# an empty string to signal end-of-stream.
2092
elif eno == errno.EINTR:
2093
# Retry the interrupted recv.
2097
if report_activity is not None:
2098
report_activity(len(bytes), 'read')
2102
def recv_all(socket, count):
2103
"""Receive an exact number of bytes.
2105
Regular Socket.recv() may return less than the requested number of bytes,
2106
depending on what's in the OS buffer. MSG_WAITALL is not available
2107
on all platforms, but this should work everywhere. This will return
2108
less than the requested amount if the remote end closes.
2110
This isn't optimized and is intended mostly for use in testing.
2113
while len(b) < count:
2114
new = read_bytes_from_socket(socket, None, count - len(b))
2121
def send_all(sock, bytes, report_activity=None):
2122
"""Send all bytes on a socket.
2124
Breaks large blocks in smaller chunks to avoid buffering limitations on
2125
some platforms, and catches EINTR which may be thrown if the send is
2126
interrupted by a signal.
2128
This is preferred to socket.sendall(), because it avoids portability bugs
2129
and provides activity reporting.
2131
:param report_activity: Call this as bytes are read, see
2132
Transport._report_activity
2135
byte_count = len(bytes)
2136
while sent_total < byte_count:
2138
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2139
except socket.error, e:
2140
if e.args[0] != errno.EINTR:
2144
report_activity(sent, 'write')
2147
def connect_socket(address):
2148
# Slight variation of the socket.create_connection() function (provided by
2149
# python-2.6) that can fail if getaddrinfo returns an empty list. We also
2150
# provide it for previous python versions. Also, we don't use the timeout
2151
# parameter (provided by the python implementation) so we don't implement
2153
err = socket.error('getaddrinfo returns an empty list')
2154
host, port = address
2155
for res in socket.getaddrinfo(host, port, 0, socket.SOCK_STREAM):
2156
af, socktype, proto, canonname, sa = res
2159
sock = socket.socket(af, socktype, proto)
2163
except socket.error, err:
2164
# 'err' is now the most recent error
2165
if sock is not None:
2170
def dereference_path(path):
2171
"""Determine the real path to a file.
2173
All parent elements are dereferenced. But the file itself is not
2175
:param path: The original path. May be absolute or relative.
2176
:return: the real path *to* the file
2178
parent, base = os.path.split(path)
2179
# The pathjoin for '.' is a workaround for Python bug #1213894.
2180
# (initial path components aren't dereferenced)
2181
return pathjoin(realpath(pathjoin('.', parent)), base)
2184
def supports_mapi():
2185
"""Return True if we can use MAPI to launch a mail client."""
2186
return sys.platform == "win32"
2189
def resource_string(package, resource_name):
2190
"""Load a resource from a package and return it as a string.
2192
Note: Only packages that start with bzrlib are currently supported.
2194
This is designed to be a lightweight implementation of resource
2195
loading in a way which is API compatible with the same API from
2197
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2198
If and when pkg_resources becomes a standard library, this routine
2201
# Check package name is within bzrlib
2202
if package == "bzrlib":
2203
resource_relpath = resource_name
2204
elif package.startswith("bzrlib."):
2205
package = package[len("bzrlib."):].replace('.', os.sep)
2206
resource_relpath = pathjoin(package, resource_name)
2208
raise errors.BzrError('resource package %s not in bzrlib' % package)
2210
# Map the resource to a file and read its contents
2211
base = dirname(bzrlib.__file__)
2212
if getattr(sys, 'frozen', None): # bzr.exe
2213
base = abspath(pathjoin(base, '..', '..'))
2214
f = file(pathjoin(base, resource_relpath), "rU")
2220
def file_kind_from_stat_mode_thunk(mode):
2221
global file_kind_from_stat_mode
2222
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2224
from bzrlib._readdir_pyx import UTF8DirReader
2225
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2226
except ImportError, e:
2227
# This is one time where we won't warn that an extension failed to
2228
# load. The extension is never available on Windows anyway.
2229
from bzrlib._readdir_py import (
2230
_kind_from_mode as file_kind_from_stat_mode
2232
return file_kind_from_stat_mode(mode)
2233
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2235
def file_stat(f, _lstat=os.lstat):
2240
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2241
raise errors.NoSuchFile(f)
2244
def file_kind(f, _lstat=os.lstat):
2245
stat_value = file_stat(f, _lstat)
2246
return file_kind_from_stat_mode(stat_value.st_mode)
2248
def until_no_eintr(f, *a, **kw):
2249
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2251
WARNING: you must be certain that it is safe to retry the call repeatedly
2252
if EINTR does occur. This is typically only true for low-level operations
2253
like os.read. If in any doubt, don't use this.
2255
Keep in mind that this is not a complete solution to EINTR. There is
2256
probably code in the Python standard library and other dependencies that
2257
may encounter EINTR if a signal arrives (and there is signal handler for
2258
that signal). So this function can reduce the impact for IO that bzrlib
2259
directly controls, but it is not a complete solution.
2261
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2265
except (IOError, OSError), e:
2266
if e.errno == errno.EINTR:
2271
@deprecated_function(deprecated_in((2, 2, 0)))
2272
def re_compile_checked(re_string, flags=0, where=""):
2273
"""Return a compiled re, or raise a sensible error.
2275
This should only be used when compiling user-supplied REs.
2277
:param re_string: Text form of regular expression.
2278
:param flags: eg re.IGNORECASE
2279
:param where: Message explaining to the user the context where
2280
it occurred, eg 'log search filter'.
2282
# from https://bugs.launchpad.net/bzr/+bug/251352
2284
re_obj = re.compile(re_string, flags)
2287
except errors.InvalidPattern, e:
2289
where = ' in ' + where
2290
# despite the name 'error' is a type
2291
raise errors.BzrCommandError('Invalid regular expression%s: %s'
2295
if sys.platform == "win32":
2298
return msvcrt.getch()
2303
fd = sys.stdin.fileno()
2304
settings = termios.tcgetattr(fd)
2307
ch = sys.stdin.read(1)
2309
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2312
if sys.platform.startswith('linux'):
2313
def _local_concurrency():
2315
return os.sysconf('SC_NPROCESSORS_ONLN')
2316
except (ValueError, OSError, AttributeError):
2318
elif sys.platform == 'darwin':
2319
def _local_concurrency():
2320
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2321
stdout=subprocess.PIPE).communicate()[0]
2322
elif "bsd" in sys.platform:
2323
def _local_concurrency():
2324
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2325
stdout=subprocess.PIPE).communicate()[0]
2326
elif sys.platform == 'sunos5':
2327
def _local_concurrency():
2328
return subprocess.Popen(['psrinfo', '-p',],
2329
stdout=subprocess.PIPE).communicate()[0]
2330
elif sys.platform == "win32":
2331
def _local_concurrency():
2332
# This appears to return the number of cores.
2333
return os.environ.get('NUMBER_OF_PROCESSORS')
2335
def _local_concurrency():
2340
_cached_local_concurrency = None
2342
def local_concurrency(use_cache=True):
2343
"""Return how many processes can be run concurrently.
2345
Rely on platform specific implementations and default to 1 (one) if
2346
anything goes wrong.
2348
global _cached_local_concurrency
2350
if _cached_local_concurrency is not None and use_cache:
2351
return _cached_local_concurrency
2353
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2354
if concurrency is None:
2356
import multiprocessing
2357
concurrency = multiprocessing.cpu_count()
2358
except (ImportError, NotImplementedError):
2359
# multiprocessing is only available on Python >= 2.6
2360
# and multiprocessing.cpu_count() isn't implemented on all
2363
concurrency = _local_concurrency()
2364
except (OSError, IOError):
2367
concurrency = int(concurrency)
2368
except (TypeError, ValueError):
2371
_cached_concurrency = concurrency
2375
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2376
"""A stream writer that doesn't decode str arguments."""
2378
def __init__(self, encode, stream, errors='strict'):
2379
codecs.StreamWriter.__init__(self, stream, errors)
2380
self.encode = encode
2382
def write(self, object):
2383
if type(object) is str:
2384
self.stream.write(object)
2386
data, _ = self.encode(object, self.errors)
2387
self.stream.write(data)
2389
if sys.platform == 'win32':
2390
def open_file(filename, mode='r', bufsize=-1):
2391
"""This function is used to override the ``open`` builtin.
2393
But it uses O_NOINHERIT flag so the file handle is not inherited by
2394
child processes. Deleting or renaming a closed file opened with this
2395
function is not blocking child processes.
2397
writing = 'w' in mode
2398
appending = 'a' in mode
2399
updating = '+' in mode
2400
binary = 'b' in mode
2403
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2404
# for flags for each modes.
2414
flags |= os.O_WRONLY
2415
flags |= os.O_CREAT | os.O_TRUNC
2420
flags |= os.O_WRONLY
2421
flags |= os.O_CREAT | os.O_APPEND
2426
flags |= os.O_RDONLY
2428
return os.fdopen(os.open(filename, flags), mode, bufsize)
2433
def available_backup_name(base, exists):
2434
"""Find a non-existing backup file name.
2436
This will *not* create anything, this only return a 'free' entry. This
2437
should be used for checking names in a directory below a locked
2438
tree/branch/repo to avoid race conditions. This is LBYL (Look Before You
2439
Leap) and generally discouraged.
2441
:param base: The base name.
2443
:param exists: A callable returning True if the path parameter exists.
2446
name = "%s.~%d~" % (base, counter)
2449
name = "%s.~%d~" % (base, counter)
2453
def set_fd_cloexec(fd):
2454
"""Set a Unix file descriptor's FD_CLOEXEC flag. Do nothing if platform
2455
support for this is not available.
2459
old = fcntl.fcntl(fd, fcntl.F_GETFD)
2460
fcntl.fcntl(fd, fcntl.F_SETFD, old | fcntl.FD_CLOEXEC)
2461
except (ImportError, AttributeError):
2462
# Either the fcntl module or specific constants are not present
2466
def find_executable_on_path(name):
2467
"""Finds an executable on the PATH.
2469
On Windows, this will try to append each extension in the PATHEXT
2470
environment variable to the name, if it cannot be found with the name
2473
:param name: The base name of the executable.
2474
:return: The path to the executable found or None.
2476
path = os.environ.get('PATH')
2479
path = path.split(os.pathsep)
2480
if sys.platform == 'win32':
2481
exts = os.environ.get('PATHEXT', '').split(os.pathsep)
2482
exts = [ext.lower() for ext in exts]
2483
base, ext = os.path.splitext(name)
2485
if ext.lower() not in exts:
2493
f = os.path.join(d, name) + ext
2494
if os.access(f, os.X_OK):
2499
def _posix_is_local_pid_dead(pid):
2500
"""True if pid doesn't correspond to live process on this machine"""
2502
# Special meaning of unix kill: just check if it's there.
2505
if e.errno == errno.ESRCH:
2506
# On this machine, and really not found: as sure as we can be
2509
elif e.errno == errno.EPERM:
2510
# exists, though not ours
2513
mutter("os.kill(%d, 0) failed: %s" % (pid, e))
2514
# Don't really know.
2517
# Exists and our process: not dead.
2520
if sys.platform == "win32":
2521
is_local_pid_dead = win32utils.is_local_pid_dead
2523
is_local_pid_dead = _posix_is_local_pid_dead
2526
def fdatasync(fileno):
2527
"""Flush file contents to disk if possible.
2529
:param fileno: Integer OS file handle.
2530
:raises TransportNotPossible: If flushing to disk is not possible.
2532
fn = getattr(os, 'fdatasync', getattr(os, 'fsync', None))
2537
def ensure_empty_directory_exists(path, exception_class):
2538
"""Make sure a local directory exists and is empty.
2540
If it does not exist, it is created. If it exists and is not empty, an
2541
instance of exception_class is raised.
2546
if e.errno != errno.EEXIST:
2548
if os.listdir(path) != []:
2549
raise exception_class(path)
2552
def is_environment_error(evalue):
2553
"""True if exception instance is due to a process environment issue
2555
This includes OSError and IOError, but also other errors that come from
2556
the operating system or core libraries but are not subclasses of those.
2558
if isinstance(evalue, (EnvironmentError, select.error)):
2560
if sys.platform == "win32" and win32utils._is_pywintypes_error(evalue):
340
if (f == '..') or (f == None) or (f == ''):
341
bailout("sorry, %r not allowed in path" % f)
342
return os.path.join(*p)
345
def appendpath(p1, p2):
349
return os.path.join(p1, p2)
352
def extern_command(cmd, ignore_errors = False):
353
mutter('external command: %s' % `cmd`)
355
if not ignore_errors:
356
bailout('command failed')
added
removed
bzrlib/osutils.py
