155
58
# TODO: I'm not really sure this is the best format either.x
157
if _QUOTE_RE is None:
158
61
_QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/\\_~-])')
160
63
if _QUOTE_RE.search(f):
161
64
return '"' + f + '"'
166
_directory_kind = 'directory'
169
"""Return the current umask"""
170
# Assume that people aren't messing with the umask while running
171
# XXX: This is not thread safe, but there is no way to get the
172
# umask without setting it
180
_directory_kind: "/",
182
'tree-reference': '+',
70
mode = os.lstat(f)[ST_MODE]
186
89
def kind_marker(kind):
188
return _kind_marker_map[kind]
190
# Slightly faster than using .get(, '') when the common case is that
195
lexists = getattr(os.path, 'lexists', None)
199
stat = getattr(os, 'lstat', os.stat)
203
if e.errno == errno.ENOENT:
206
raise errors.BzrError("lstat/stat of (%r): %r" % (f, e))
209
def fancy_rename(old, new, rename_func, unlink_func):
210
"""A fancy rename, when you don't have atomic rename.
212
:param old: The old path, to rename from
213
:param new: The new path, to rename to
214
:param rename_func: The potentially non-atomic rename function
215
:param unlink_func: A way to delete the target file if the full rename
218
# sftp rename doesn't allow overwriting, so play tricks:
219
base = os.path.basename(new)
220
dirname = os.path.dirname(new)
221
# callers use different encodings for the paths so the following MUST
222
# respect that. We rely on python upcasting to unicode if new is unicode
223
# and keeping a str if not.
224
tmp_name = 'tmp.%s.%.9f.%d.%s' % (base, time.time(),
225
os.getpid(), rand_chars(10))
226
tmp_name = pathjoin(dirname, tmp_name)
228
# Rename the file out of the way, but keep track if it didn't exist
229
# We don't want to grab just any exception
230
# something like EACCES should prevent us from continuing
231
# The downside is that the rename_func has to throw an exception
232
# with an errno = ENOENT, or NoSuchFile
235
rename_func(new, tmp_name)
236
except (errors.NoSuchFile,), e:
239
# RBC 20060103 abstraction leakage: the paramiko SFTP clients rename
240
# function raises an IOError with errno is None when a rename fails.
241
# This then gets caught here.
242
if e.errno not in (None, errno.ENOENT, errno.ENOTDIR):
245
if (getattr(e, 'errno', None) is None
246
or e.errno not in (errno.ENOENT, errno.ENOTDIR)):
255
# This may throw an exception, in which case success will
257
rename_func(old, new)
259
except (IOError, OSError), e:
260
# source and target may be aliases of each other (e.g. on a
261
# case-insensitive filesystem), so we may have accidentally renamed
262
# source by when we tried to rename target
263
failure_exc = sys.exc_info()
264
if (file_existed and e.errno in (None, errno.ENOENT)
265
and old.lower() == new.lower()):
266
# source and target are the same file on a case-insensitive
267
# filesystem, so we don't generate an exception
271
# If the file used to exist, rename it back into place
272
# otherwise just delete it from the tmp location
274
unlink_func(tmp_name)
276
rename_func(tmp_name, new)
277
if failure_exc is not None:
278
raise failure_exc[0], failure_exc[1], failure_exc[2]
281
# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
282
# choke on a Unicode string containing a relative path if
283
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
285
_fs_enc = sys.getfilesystemencoding() or 'utf-8'
286
def _posix_abspath(path):
287
# jam 20060426 rather than encoding to fsencoding
288
# copy posixpath.abspath, but use os.getcwdu instead
289
if not posixpath.isabs(path):
290
path = posixpath.join(getcwd(), path)
291
return posixpath.normpath(path)
294
def _posix_realpath(path):
295
return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)
298
def _win32_fixdrive(path):
299
"""Force drive letters to be consistent.
301
win32 is inconsistent whether it returns lower or upper case
302
and even if it was consistent the user might type the other
303
so we force it to uppercase
304
running python.exe under cmd.exe return capital C:\\
305
running win32 python inside a cygwin shell returns lowercase c:\\
307
drive, path = _nt_splitdrive(path)
308
return drive.upper() + path
311
def _win32_abspath(path):
312
# Real _nt_abspath doesn't have a problem with a unicode cwd
313
return _win32_fixdrive(_nt_abspath(unicode(path)).replace('\\', '/'))
316
def _win98_abspath(path):
317
"""Return the absolute version of a path.
318
Windows 98 safe implementation (python reimplementation
319
of Win32 API function GetFullPathNameW)
324
# \\HOST\path => //HOST/path
325
# //HOST/path => //HOST/path
326
# path => C:/cwd/path
329
# check for absolute path
330
drive = _nt_splitdrive(path)[0]
331
if drive == '' and path[:2] not in('//','\\\\'):
333
# we cannot simply os.path.join cwd and path
334
# because os.path.join('C:','/path') produce '/path'
335
# and this is incorrect
336
if path[:1] in ('/','\\'):
337
cwd = _nt_splitdrive(cwd)[0]
339
path = cwd + '\\' + path
340
return _win32_fixdrive(_nt_normpath(path).replace('\\', '/'))
343
def _win32_realpath(path):
344
# Real _nt_realpath doesn't have a problem with a unicode cwd
345
return _win32_fixdrive(_nt_realpath(unicode(path)).replace('\\', '/'))
348
def _win32_pathjoin(*args):
349
return _nt_join(*args).replace('\\', '/')
352
def _win32_normpath(path):
353
return _win32_fixdrive(_nt_normpath(unicode(path)).replace('\\', '/'))
357
return _win32_fixdrive(os.getcwdu().replace('\\', '/'))
360
def _win32_mkdtemp(*args, **kwargs):
361
return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))
364
def _add_rename_error_details(e, old, new):
365
new_e = OSError(e.errno, "failed to rename %s to %s: %s"
366
% (old, new, e.strerror))
368
new_e.to_filename = new
372
def _win32_rename(old, new):
373
"""We expect to be able to atomically replace 'new' with old.
375
On win32, if new exists, it must be moved out of the way first,
379
fancy_rename(old, new, rename_func=_wrapped_rename, unlink_func=os.unlink)
381
if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
382
# If we try to rename a non-existant file onto cwd, we get
383
# EPERM or EACCES instead of ENOENT, this will raise ENOENT
384
# if the old path doesn't exist, sometimes we get EACCES
385
# On Linux, we seem to get EBUSY, on Mac we get EINVAL
390
def _wrapped_rename(old, new):
391
"""Rename a file or directory"""
394
except (IOError, OSError), e:
395
# this is eventually called by all rename-like functions, so should
397
raise _add_rename_error_details(e, old, new)
401
return unicodedata.normalize('NFC', os.getcwdu())
404
# Default is to just use the python builtins, but these can be rebound on
405
# particular platforms.
406
abspath = _posix_abspath
407
realpath = _posix_realpath
408
pathjoin = os.path.join
409
normpath = os.path.normpath
410
rename = _wrapped_rename # overridden below on win32
412
dirname = os.path.dirname
413
basename = os.path.basename
414
split = os.path.split
415
splitext = os.path.splitext
416
# These were already imported into local scope
417
# mkdtemp = tempfile.mkdtemp
418
# rmtree = shutil.rmtree
420
MIN_ABS_PATHLENGTH = 1
423
if sys.platform == 'win32':
424
if win32utils.winver == 'Windows 98':
425
abspath = _win98_abspath
427
abspath = _win32_abspath
428
realpath = _win32_realpath
429
pathjoin = _win32_pathjoin
430
normpath = _win32_normpath
431
getcwd = _win32_getcwd
432
mkdtemp = _win32_mkdtemp
433
rename = _win32_rename
435
MIN_ABS_PATHLENGTH = 3
437
def _win32_delete_readonly(function, path, excinfo):
438
"""Error handler for shutil.rmtree function [for win32]
439
Helps to remove files and dirs marked as read-only.
441
exception = excinfo[1]
442
if function in (os.remove, os.rmdir) \
443
and isinstance(exception, OSError) \
444
and exception.errno == errno.EACCES:
450
def rmtree(path, ignore_errors=False, onerror=_win32_delete_readonly):
451
"""Replacer for shutil.rmtree: could remove readonly dirs/files"""
452
return shutil.rmtree(path, ignore_errors, onerror)
454
f = win32utils.get_unicode_argv # special function or None
458
elif sys.platform == 'darwin':
462
def get_terminal_encoding():
463
"""Find the best encoding for printing to the screen.
465
This attempts to check both sys.stdout and sys.stdin to see
466
what encoding they are in, and if that fails it falls back to
467
osutils.get_user_encoding().
468
The problem is that on Windows, locale.getpreferredencoding()
469
is not the same encoding as that used by the console:
470
http://mail.python.org/pipermail/python-list/2003-May/162357.html
472
On my standard US Windows XP, the preferred encoding is
473
cp1252, but the console is cp437
475
from bzrlib.trace import mutter
476
output_encoding = getattr(sys.stdout, 'encoding', None)
477
if not output_encoding:
478
input_encoding = getattr(sys.stdin, 'encoding', None)
479
if not input_encoding:
480
output_encoding = get_user_encoding()
481
mutter('encoding stdout as osutils.get_user_encoding() %r',
484
output_encoding = input_encoding
485
mutter('encoding stdout as sys.stdin encoding %r', output_encoding)
487
mutter('encoding stdout as sys.stdout encoding %r', output_encoding)
488
if output_encoding == 'cp0':
489
# invalid encoding (cp0 means 'no codepage' on Windows)
490
output_encoding = get_user_encoding()
491
mutter('cp0 is invalid encoding.'
492
' encoding stdout as osutils.get_user_encoding() %r',
496
codecs.lookup(output_encoding)
498
sys.stderr.write('bzr: warning:'
499
' unknown terminal encoding %s.\n'
500
' Using encoding %s instead.\n'
501
% (output_encoding, get_user_encoding())
503
output_encoding = get_user_encoding()
505
return output_encoding
92
elif kind == 'directory':
94
elif kind == 'symlink':
97
raise BzrError('invalid file kind %r' % kind)
101
if hasattr(os, 'lstat'):
107
if e.errno == errno.ENOENT:
110
raise BzrError("lstat/stat of (%r): %r" % (f, e))
508
112
def normalizepath(f):
509
if getattr(os.path, 'realpath', None) is not None:
113
if hasattr(os.path, 'realpath'):
513
117
[p,e] = os.path.split(f)
514
118
if e == "" or e == "." or e == "..":
517
return pathjoin(F(p), e)
121
return os.path.join(F(p), e)
125
"""Copy a file to a backup.
127
Backups are named in GNU-style, with a ~ suffix.
129
If the file is already a backup, it's not copied.
135
if has_symlinks() and os.path.islink(fn):
136
target = os.readlink(fn)
137
os.symlink(target, bfn)
145
outf = file(bfn, 'wb')
1150
447
def relpath(base, path):
1151
"""Return path relative to base, or raise PathNotChild exception.
448
"""Return path relative to base, or raise exception.
1153
450
The path may be either an absolute path or a path relative to the
1154
451
current working directory.
1156
453
os.path.commonprefix (python2.4) has a bad bug that it works just
1157
454
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
1158
avoids that problem.
1160
NOTE: `base` should not have a trailing slash otherwise you'll get
1161
PathNotChild exceptions regardless of `path`.
1164
if len(base) < MIN_ABS_PATHLENGTH:
1165
# must have space for e.g. a drive letter
1166
raise ValueError('%r is too short to calculate a relative path'
455
avoids that problem."""
456
rp = os.path.abspath(path)
1174
if len(head) <= len(base) and head != base:
1175
raise errors.PathNotChild(rp, base)
460
while len(head) >= len(base):
1176
461
if head == base:
1178
head, tail = split(head)
463
head, tail = os.path.split(head)
1183
return pathjoin(*reversed(s))
1188
def _cicp_canonical_relpath(base, path):
1189
"""Return the canonical path relative to base.
1191
Like relpath, but on case-insensitive-case-preserving file-systems, this
1192
will return the relpath as stored on the file-system rather than in the
1193
case specified in the input string, for all existing portions of the path.
1195
This will cause O(N) behaviour if called for every path in a tree; if you
1196
have a number of paths to convert, you should use canonical_relpaths().
1198
# TODO: it should be possible to optimize this for Windows by using the
1199
# win32 API FindFiles function to look for the specified name - but using
1200
# os.listdir() still gives us the correct, platform agnostic semantics in
1203
rel = relpath(base, path)
1204
# '.' will have been turned into ''
1208
abs_base = abspath(base)
1210
_listdir = os.listdir
1212
# use an explicit iterator so we can easily consume the rest on early exit.
1213
bit_iter = iter(rel.split('/'))
1214
for bit in bit_iter:
1217
next_entries = _listdir(current)
1218
except OSError: # enoent, eperm, etc
1219
# We can't find this in the filesystem, so just append the
1221
current = pathjoin(current, bit, *list(bit_iter))
1223
for look in next_entries:
1224
if lbit == look.lower():
1225
current = pathjoin(current, look)
1228
# got to the end, nothing matched, so we just return the
1229
# non-existing bits as they were specified (the filename may be
1230
# the target of a move, for example).
1231
current = pathjoin(current, bit, *list(bit_iter))
1233
return current[len(abs_base):].lstrip('/')
1235
# XXX - TODO - we need better detection/integration of case-insensitive
1236
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
1237
# filesystems), for example, so could probably benefit from the same basic
1238
# support there. For now though, only Windows and OSX get that support, and
1239
# they get it for *all* file-systems!
1240
if sys.platform in ('win32', 'darwin'):
1241
canonical_relpath = _cicp_canonical_relpath
1243
canonical_relpath = relpath
1245
def canonical_relpaths(base, paths):
1246
"""Create an iterable to canonicalize a sequence of relative paths.
1248
The intent is for this implementation to use a cache, vastly speeding
1249
up multiple transformations in the same directory.
1251
# but for now, we haven't optimized...
1252
return [canonical_relpath(base, p) for p in paths]
1254
def safe_unicode(unicode_or_utf8_string):
1255
"""Coerce unicode_or_utf8_string into unicode.
1257
If it is unicode, it is returned.
1258
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1259
wrapped in a BzrBadParameterNotUnicode exception.
1261
if isinstance(unicode_or_utf8_string, unicode):
1262
return unicode_or_utf8_string
1264
return unicode_or_utf8_string.decode('utf8')
1265
except UnicodeDecodeError:
1266
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1269
def safe_utf8(unicode_or_utf8_string):
1270
"""Coerce unicode_or_utf8_string to a utf8 string.
1272
If it is a str, it is returned.
1273
If it is Unicode, it is encoded into a utf-8 string.
1275
if isinstance(unicode_or_utf8_string, str):
1276
# TODO: jam 20070209 This is overkill, and probably has an impact on
1277
# performance if we are dealing with lots of apis that want a
1280
# Make sure it is a valid utf-8 string
1281
unicode_or_utf8_string.decode('utf-8')
1282
except UnicodeDecodeError:
1283
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1284
return unicode_or_utf8_string
1285
return unicode_or_utf8_string.encode('utf-8')
1288
_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
1289
' Revision id generators should be creating utf8'
1293
def safe_revision_id(unicode_or_utf8_string, warn=True):
1294
"""Revision ids should now be utf8, but at one point they were unicode.
1296
:param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
1298
:param warn: Functions that are sanitizing user data can set warn=False
1299
:return: None or a utf8 revision id.
1301
if (unicode_or_utf8_string is None
1302
or unicode_or_utf8_string.__class__ == str):
1303
return unicode_or_utf8_string
1305
symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
1307
return cache_utf8.encode(unicode_or_utf8_string)
1310
_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
1311
' generators should be creating utf8 file ids.')
1314
def safe_file_id(unicode_or_utf8_string, warn=True):
1315
"""File ids should now be utf8, but at one point they were unicode.
1317
This is the same as safe_utf8, except it uses the cached encode functions
1318
to save a little bit of performance.
1320
:param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
1322
:param warn: Functions that are sanitizing user data can set warn=False
1323
:return: None or a utf8 file id.
1325
if (unicode_or_utf8_string is None
1326
or unicode_or_utf8_string.__class__ == str):
1327
return unicode_or_utf8_string
1329
symbol_versioning.warn(_file_id_warning, DeprecationWarning,
1331
return cache_utf8.encode(unicode_or_utf8_string)
1334
_platform_normalizes_filenames = False
1335
if sys.platform == 'darwin':
1336
_platform_normalizes_filenames = True
1339
def normalizes_filenames():
1340
"""Return True if this platform normalizes unicode filenames.
1342
Mac OSX does, Windows/Linux do not.
1344
return _platform_normalizes_filenames
1347
def _accessible_normalized_filename(path):
1348
"""Get the unicode normalized path, and if you can access the file.
1350
On platforms where the system normalizes filenames (Mac OSX),
1351
you can access a file by any path which will normalize correctly.
1352
On platforms where the system does not normalize filenames
1353
(Windows, Linux), you have to access a file by its exact path.
1355
Internally, bzr only supports NFC normalization, since that is
1356
the standard for XML documents.
1358
So return the normalized path, and a flag indicating if the file
1359
can be accessed by that path.
1362
return unicodedata.normalize('NFC', unicode(path)), True
1365
def _inaccessible_normalized_filename(path):
1366
__doc__ = _accessible_normalized_filename.__doc__
1368
normalized = unicodedata.normalize('NFC', unicode(path))
1369
return normalized, normalized == path
1372
if _platform_normalizes_filenames:
1373
normalized_filename = _accessible_normalized_filename
1375
normalized_filename = _inaccessible_normalized_filename
1378
def set_signal_handler(signum, handler, restart_syscall=True):
1379
"""A wrapper for signal.signal that also calls siginterrupt(signum, False)
1380
on platforms that support that.
1382
:param restart_syscall: if set, allow syscalls interrupted by a signal to
1383
automatically restart (by calling `signal.siginterrupt(signum,
1384
False)`). May be ignored if the feature is not available on this
1385
platform or Python version.
1389
siginterrupt = signal.siginterrupt
1391
# This python implementation doesn't provide signal support, hence no
1394
except AttributeError:
1395
# siginterrupt doesn't exist on this platform, or for this version
1397
siginterrupt = lambda signum, flag: None
1399
def sig_handler(*args):
1400
# Python resets the siginterrupt flag when a signal is
1401
# received. <http://bugs.python.org/issue8354>
1402
# As a workaround for some cases, set it back the way we want it.
1403
siginterrupt(signum, False)
1404
# Now run the handler function passed to set_signal_handler.
1407
sig_handler = handler
1408
old_handler = signal.signal(signum, sig_handler)
1410
siginterrupt(signum, False)
1414
default_terminal_width = 80
1415
"""The default terminal width for ttys.
1417
This is defined so that higher levels can share a common fallback value when
1418
terminal_width() returns None.
1422
def terminal_width():
1423
"""Return terminal width.
1425
None is returned if the width can't established precisely.
1428
- if BZR_COLUMNS is set, returns its value
1429
- if there is no controlling terminal, returns None
1430
- if COLUMNS is set, returns its value,
1432
From there, we need to query the OS to get the size of the controlling
1436
- get termios.TIOCGWINSZ
1437
- if an error occurs or a negative value is obtained, returns None
1441
- win32utils.get_console_size() decides,
1442
- returns None on error (provided default value)
1445
# If BZR_COLUMNS is set, take it, user is always right
1447
return int(os.environ['BZR_COLUMNS'])
1448
except (KeyError, ValueError):
1451
isatty = getattr(sys.stdout, 'isatty', None)
1452
if isatty is None or not isatty():
1453
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1456
# If COLUMNS is set, take it, the terminal knows better (even inside a
1457
# given terminal, the application can decide to set COLUMNS to a lower
1458
# value (splitted screen) or a bigger value (scroll bars))
1460
return int(os.environ['COLUMNS'])
1461
except (KeyError, ValueError):
1464
width, height = _terminal_size(None, None)
1466
# Consider invalid values as meaning no width
1472
def _win32_terminal_size(width, height):
1473
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1474
return width, height
1477
def _ioctl_terminal_size(width, height):
1479
import struct, fcntl, termios
1480
s = struct.pack('HHHH', 0, 0, 0, 0)
1481
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1482
height, width = struct.unpack('HHHH', x)[0:2]
1483
except (IOError, AttributeError):
1485
return width, height
1487
_terminal_size = None
1488
"""Returns the terminal size as (width, height).
1490
:param width: Default value for width.
1491
:param height: Default value for height.
1493
This is defined specifically for each OS and query the size of the controlling
1494
terminal. If any error occurs, the provided default values should be returned.
1496
if sys.platform == 'win32':
1497
_terminal_size = _win32_terminal_size
1499
_terminal_size = _ioctl_terminal_size
1502
def _terminal_size_changed(signum, frame):
1503
"""Set COLUMNS upon receiving a SIGnal for WINdow size CHange."""
1504
width, height = _terminal_size(None, None)
1505
if width is not None:
1506
os.environ['COLUMNS'] = str(width)
1509
_registered_sigwinch = False
1510
def watch_sigwinch():
1511
"""Register for SIGWINCH, once and only once.
1513
Do nothing if the signal module is not available.
1515
global _registered_sigwinch
1516
if not _registered_sigwinch:
1519
if getattr(signal, "SIGWINCH", None) is not None:
1520
set_signal_handler(signal.SIGWINCH, _terminal_size_changed)
1522
# python doesn't provide signal support, nothing we can do about it
1524
_registered_sigwinch = True
1527
def supports_executable():
1528
return sys.platform != "win32"
1531
def supports_posix_readonly():
1532
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1534
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1535
directory controls creation/deletion, etc.
1537
And under win32, readonly means that the directory itself cannot be
1538
deleted. The contents of a readonly directory can be changed, unlike POSIX
1539
where files in readonly directories cannot be added, deleted or renamed.
1541
return sys.platform != "win32"
1544
def set_or_unset_env(env_variable, value):
1545
"""Modify the environment, setting or removing the env_variable.
1547
:param env_variable: The environment variable in question
1548
:param value: The value to set the environment to. If None, then
1549
the variable will be removed.
1550
:return: The original value of the environment variable.
1552
orig_val = os.environ.get(env_variable)
1554
if orig_val is not None:
1555
del os.environ[env_variable]
1557
if isinstance(value, unicode):
1558
value = value.encode(get_user_encoding())
1559
os.environ[env_variable] = value
1563
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
1566
def check_legal_path(path):
1567
"""Check whether the supplied path is legal.
1568
This is only required on Windows, so we don't test on other platforms
1571
if sys.platform != "win32":
1573
if _validWin32PathRE.match(path) is None:
1574
raise errors.IllegalPath(path)
1577
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1579
def _is_error_enotdir(e):
1580
"""Check if this exception represents ENOTDIR.
1582
Unfortunately, python is very inconsistent about the exception
1583
here. The cases are:
1584
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1585
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1586
which is the windows error code.
1587
3) Windows, Python2.5 uses errno == EINVAL and
1588
winerror == ERROR_DIRECTORY
1590
:param e: An Exception object (expected to be OSError with an errno
1591
attribute, but we should be able to cope with anything)
1592
:return: True if this represents an ENOTDIR error. False otherwise.
1594
en = getattr(e, 'errno', None)
1595
if (en == errno.ENOTDIR
1596
or (sys.platform == 'win32'
1597
and (en == _WIN32_ERROR_DIRECTORY
1598
or (en == errno.EINVAL
1599
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1605
def walkdirs(top, prefix=""):
1606
"""Yield data about all the directories in a tree.
1608
This yields all the data about the contents of a directory at a time.
1609
After each directory has been yielded, if the caller has mutated the list
1610
to exclude some directories, they are then not descended into.
1612
The data yielded is of the form:
1613
((directory-relpath, directory-path-from-top),
1614
[(relpath, basename, kind, lstat, path-from-top), ...]),
1615
- directory-relpath is the relative path of the directory being returned
1616
with respect to top. prefix is prepended to this.
1617
- directory-path-from-root is the path including top for this directory.
1618
It is suitable for use with os functions.
1619
- relpath is the relative path within the subtree being walked.
1620
- basename is the basename of the path
1621
- kind is the kind of the file now. If unknown then the file is not
1622
present within the tree - but it may be recorded as versioned. See
1624
- lstat is the stat data *if* the file was statted.
1625
- planned, not implemented:
1626
path_from_tree_root is the path from the root of the tree.
1628
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1629
allows one to walk a subtree but get paths that are relative to a tree
1631
:return: an iterator over the dirs.
1633
#TODO there is a bit of a smell where the results of the directory-
1634
# summary in this, and the path from the root, may not agree
1635
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1636
# potentially confusing output. We should make this more robust - but
1637
# not at a speed cost. RBC 20060731
1639
_directory = _directory_kind
1640
_listdir = os.listdir
1641
_kind_from_mode = file_kind_from_stat_mode
1642
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1644
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1645
relroot, _, _, _, top = pending.pop()
1647
relprefix = relroot + u'/'
1650
top_slash = top + u'/'
1653
append = dirblock.append
1655
names = sorted(_listdir(top))
1657
if not _is_error_enotdir(e):
1661
abspath = top_slash + name
1662
statvalue = _lstat(abspath)
1663
kind = _kind_from_mode(statvalue.st_mode)
1664
append((relprefix + name, name, kind, statvalue, abspath))
1665
yield (relroot, top), dirblock
1667
# push the user specified dirs from dirblock
1668
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1671
class DirReader(object):
1672
"""An interface for reading directories."""
1674
def top_prefix_to_starting_dir(self, top, prefix=""):
1675
"""Converts top and prefix to a starting dir entry
1677
:param top: A utf8 path
1678
:param prefix: An optional utf8 path to prefix output relative paths
1680
:return: A tuple starting with prefix, and ending with the native
1683
raise NotImplementedError(self.top_prefix_to_starting_dir)
1685
def read_dir(self, prefix, top):
1686
"""Read a specific dir.
1688
:param prefix: A utf8 prefix to be preprended to the path basenames.
1689
:param top: A natively encoded path to read.
1690
:return: A list of the directories contents. Each item contains:
1691
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1693
raise NotImplementedError(self.read_dir)
1696
_selected_dir_reader = None
1699
def _walkdirs_utf8(top, prefix=""):
1700
"""Yield data about all the directories in a tree.
1702
This yields the same information as walkdirs() only each entry is yielded
1703
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1704
are returned as exact byte-strings.
1706
:return: yields a tuple of (dir_info, [file_info])
1707
dir_info is (utf8_relpath, path-from-top)
1708
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1709
if top is an absolute path, path-from-top is also an absolute path.
1710
path-from-top might be unicode or utf8, but it is the correct path to
1711
pass to os functions to affect the file in question. (such as os.lstat)
1713
global _selected_dir_reader
1714
if _selected_dir_reader is None:
1715
fs_encoding = _fs_enc.upper()
1716
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1717
# Win98 doesn't have unicode apis like FindFirstFileW
1718
# TODO: We possibly could support Win98 by falling back to the
1719
# original FindFirstFile, and using TCHAR instead of WCHAR,
1720
# but that gets a bit tricky, and requires custom compiling
1723
from bzrlib._walkdirs_win32 import Win32ReadDir
1724
_selected_dir_reader = Win32ReadDir()
1727
elif fs_encoding in ('UTF-8', 'US-ASCII', 'ANSI_X3.4-1968'):
1728
# ANSI_X3.4-1968 is a form of ASCII
1730
from bzrlib._readdir_pyx import UTF8DirReader
1731
_selected_dir_reader = UTF8DirReader()
1732
except ImportError, e:
1733
failed_to_load_extension(e)
1736
if _selected_dir_reader is None:
1737
# Fallback to the python version
1738
_selected_dir_reader = UnicodeDirReader()
1740
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1741
# But we don't actually uses 1-3 in pending, so set them to None
1742
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1743
read_dir = _selected_dir_reader.read_dir
1744
_directory = _directory_kind
1746
relroot, _, _, _, top = pending[-1].pop()
1749
dirblock = sorted(read_dir(relroot, top))
1750
yield (relroot, top), dirblock
1751
# push the user specified dirs from dirblock
1752
next = [d for d in reversed(dirblock) if d[2] == _directory]
1754
pending.append(next)
1757
class UnicodeDirReader(DirReader):
1758
"""A dir reader for non-utf8 file systems, which transcodes."""
1760
__slots__ = ['_utf8_encode']
1763
self._utf8_encode = codecs.getencoder('utf8')
1765
def top_prefix_to_starting_dir(self, top, prefix=""):
1766
"""See DirReader.top_prefix_to_starting_dir."""
1767
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1769
def read_dir(self, prefix, top):
1770
"""Read a single directory from a non-utf8 file system.
1772
top, and the abspath element in the output are unicode, all other paths
1773
are utf8. Local disk IO is done via unicode calls to listdir etc.
1775
This is currently the fallback code path when the filesystem encoding is
1776
not UTF-8. It may be better to implement an alternative so that we can
1777
safely handle paths that are not properly decodable in the current
1780
See DirReader.read_dir for details.
1782
_utf8_encode = self._utf8_encode
1784
_listdir = os.listdir
1785
_kind_from_mode = file_kind_from_stat_mode
1788
relprefix = prefix + '/'
1791
top_slash = top + u'/'
1794
append = dirblock.append
1795
for name in sorted(_listdir(top)):
1797
name_utf8 = _utf8_encode(name)[0]
1798
except UnicodeDecodeError:
1799
raise errors.BadFilenameEncoding(
1800
_utf8_encode(relprefix)[0] + name, _fs_enc)
1801
abspath = top_slash + name
1802
statvalue = _lstat(abspath)
1803
kind = _kind_from_mode(statvalue.st_mode)
1804
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1808
def copy_tree(from_path, to_path, handlers={}):
1809
"""Copy all of the entries in from_path into to_path.
1811
:param from_path: The base directory to copy.
1812
:param to_path: The target directory. If it does not exist, it will
1814
:param handlers: A dictionary of functions, which takes a source and
1815
destinations for files, directories, etc.
1816
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1817
'file', 'directory', and 'symlink' should always exist.
1818
If they are missing, they will be replaced with 'os.mkdir()',
1819
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1821
# Now, just copy the existing cached tree to the new location
1822
# We use a cheap trick here.
1823
# Absolute paths are prefixed with the first parameter
1824
# relative paths are prefixed with the second.
1825
# So we can get both the source and target returned
1826
# without any extra work.
1828
def copy_dir(source, dest):
1831
def copy_link(source, dest):
1832
"""Copy the contents of a symlink"""
1833
link_to = os.readlink(source)
1834
os.symlink(link_to, dest)
1836
real_handlers = {'file':shutil.copy2,
1837
'symlink':copy_link,
1838
'directory':copy_dir,
1840
real_handlers.update(handlers)
1842
if not os.path.exists(to_path):
1843
real_handlers['directory'](from_path, to_path)
1845
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1846
for relpath, name, kind, st, abspath in entries:
1847
real_handlers[kind](abspath, relpath)
1850
def copy_ownership_from_path(dst, src=None):
1851
"""Copy usr/grp ownership from src file/dir to dst file/dir.
1853
If src is None, the containing directory is used as source. If chown
1854
fails, the error is ignored and a warning is printed.
1856
chown = getattr(os, 'chown', None)
1861
src = os.path.dirname(dst)
1867
chown(dst, s.st_uid, s.st_gid)
1869
trace.warning("Unable to copy ownership from '%s' to '%s': IOError: %s." % (src, dst, e))
1872
def path_prefix_key(path):
1873
"""Generate a prefix-order path key for path.
1875
This can be used to sort paths in the same way that walkdirs does.
1877
return (dirname(path) , path)
1880
def compare_paths_prefix_order(path_a, path_b):
1881
"""Compare path_a and path_b to generate the same order walkdirs uses."""
1882
key_a = path_prefix_key(path_a)
1883
key_b = path_prefix_key(path_b)
1884
return cmp(key_a, key_b)
1887
_cached_user_encoding = None
1890
def get_user_encoding(use_cache=True):
1891
"""Find out what the preferred user encoding is.
1893
This is generally the encoding that is used for command line parameters
1894
and file contents. This may be different from the terminal encoding
1895
or the filesystem encoding.
1897
:param use_cache: Enable cache for detected encoding.
1898
(This parameter is turned on by default,
1899
and required only for selftesting)
1901
:return: A string defining the preferred user encoding
1903
global _cached_user_encoding
1904
if _cached_user_encoding is not None and use_cache:
1905
return _cached_user_encoding
1907
if sys.platform == 'darwin':
1908
# python locale.getpreferredencoding() always return
1909
# 'mac-roman' on darwin. That's a lie.
1910
sys.platform = 'posix'
1912
if os.environ.get('LANG', None) is None:
1913
# If LANG is not set, we end up with 'ascii', which is bad
1914
# ('mac-roman' is more than ascii), so we set a default which
1915
# will give us UTF-8 (which appears to work in all cases on
1916
# OSX). Users are still free to override LANG of course, as
1917
# long as it give us something meaningful. This work-around
1918
# *may* not be needed with python 3k and/or OSX 10.5, but will
1919
# work with them too -- vila 20080908
1920
os.environ['LANG'] = 'en_US.UTF-8'
1923
sys.platform = 'darwin'
1928
user_encoding = locale.getpreferredencoding()
1929
except locale.Error, e:
1930
sys.stderr.write('bzr: warning: %s\n'
1931
' Could not determine what text encoding to use.\n'
1932
' This error usually means your Python interpreter\n'
1933
' doesn\'t support the locale set by $LANG (%s)\n'
1934
" Continuing with ascii encoding.\n"
1935
% (e, os.environ.get('LANG')))
1936
user_encoding = 'ascii'
1938
# Windows returns 'cp0' to indicate there is no code page. So we'll just
1939
# treat that as ASCII, and not support printing unicode characters to the
1942
# For python scripts run under vim, we get '', so also treat that as ASCII
1943
if user_encoding in (None, 'cp0', ''):
1944
user_encoding = 'ascii'
1948
codecs.lookup(user_encoding)
1950
sys.stderr.write('bzr: warning:'
1951
' unknown encoding %s.'
1952
' Continuing with ascii encoding.\n'
1955
user_encoding = 'ascii'
1958
_cached_user_encoding = user_encoding
1960
return user_encoding
1963
def get_host_name():
1964
"""Return the current unicode host name.
1966
This is meant to be used in place of socket.gethostname() because that
1967
behaves inconsistently on different platforms.
1969
if sys.platform == "win32":
1971
return win32utils.get_host_name()
1974
return socket.gethostname().decode(get_user_encoding())
1977
# We must not read/write any more than 64k at a time from/to a socket so we
1978
# don't risk "no buffer space available" errors on some platforms. Windows in
1979
# particular is likely to throw WSAECONNABORTED or WSAENOBUFS if given too much
1981
MAX_SOCKET_CHUNK = 64 * 1024
1983
def read_bytes_from_socket(sock, report_activity=None,
1984
max_read_size=MAX_SOCKET_CHUNK):
1985
"""Read up to max_read_size of bytes from sock and notify of progress.
1987
Translates "Connection reset by peer" into file-like EOF (return an
1988
empty string rather than raise an error), and repeats the recv if
1989
interrupted by a signal.
1993
bytes = sock.recv(max_read_size)
1994
except socket.error, e:
1996
if eno == getattr(errno, "WSAECONNRESET", errno.ECONNRESET):
1997
# The connection was closed by the other side. Callers expect
1998
# an empty string to signal end-of-stream.
2000
elif eno == errno.EINTR:
2001
# Retry the interrupted recv.
2005
if report_activity is not None:
2006
report_activity(len(bytes), 'read')
2010
def recv_all(socket, count):
2011
"""Receive an exact number of bytes.
2013
Regular Socket.recv() may return less than the requested number of bytes,
2014
depending on what's in the OS buffer. MSG_WAITALL is not available
2015
on all platforms, but this should work everywhere. This will return
2016
less than the requested amount if the remote end closes.
2018
This isn't optimized and is intended mostly for use in testing.
2021
while len(b) < count:
2022
new = read_bytes_from_socket(socket, None, count - len(b))
2029
def send_all(sock, bytes, report_activity=None):
2030
"""Send all bytes on a socket.
2032
Breaks large blocks in smaller chunks to avoid buffering limitations on
2033
some platforms, and catches EINTR which may be thrown if the send is
2034
interrupted by a signal.
2036
This is preferred to socket.sendall(), because it avoids portability bugs
2037
and provides activity reporting.
2039
:param report_activity: Call this as bytes are read, see
2040
Transport._report_activity
2043
byte_count = len(bytes)
2044
while sent_total < byte_count:
2046
sent = sock.send(buffer(bytes, sent_total, MAX_SOCKET_CHUNK))
2047
except socket.error, e:
2048
if e.args[0] != errno.EINTR:
2052
report_activity(sent, 'write')
2055
def dereference_path(path):
2056
"""Determine the real path to a file.
2058
All parent elements are dereferenced. But the file itself is not
2060
:param path: The original path. May be absolute or relative.
2061
:return: the real path *to* the file
2063
parent, base = os.path.split(path)
2064
# The pathjoin for '.' is a workaround for Python bug #1213894.
2065
# (initial path components aren't dereferenced)
2066
return pathjoin(realpath(pathjoin('.', parent)), base)
2069
def supports_mapi():
2070
"""Return True if we can use MAPI to launch a mail client."""
2071
return sys.platform == "win32"
2074
def resource_string(package, resource_name):
2075
"""Load a resource from a package and return it as a string.
2077
Note: Only packages that start with bzrlib are currently supported.
2079
This is designed to be a lightweight implementation of resource
2080
loading in a way which is API compatible with the same API from
2082
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
2083
If and when pkg_resources becomes a standard library, this routine
2086
# Check package name is within bzrlib
2087
if package == "bzrlib":
2088
resource_relpath = resource_name
2089
elif package.startswith("bzrlib."):
2090
package = package[len("bzrlib."):].replace('.', os.sep)
2091
resource_relpath = pathjoin(package, resource_name)
2093
raise errors.BzrError('resource package %s not in bzrlib' % package)
2095
# Map the resource to a file and read its contents
2096
base = dirname(bzrlib.__file__)
2097
if getattr(sys, 'frozen', None): # bzr.exe
2098
base = abspath(pathjoin(base, '..', '..'))
2099
filename = pathjoin(base, resource_relpath)
2100
return open(filename, 'rU').read()
2103
def file_kind_from_stat_mode_thunk(mode):
2104
global file_kind_from_stat_mode
2105
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
2107
from bzrlib._readdir_pyx import UTF8DirReader
2108
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
2109
except ImportError, e:
2110
# This is one time where we won't warn that an extension failed to
2111
# load. The extension is never available on Windows anyway.
2112
from bzrlib._readdir_py import (
2113
_kind_from_mode as file_kind_from_stat_mode
2115
return file_kind_from_stat_mode(mode)
2116
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
2119
def file_kind(f, _lstat=os.lstat):
2121
return file_kind_from_stat_mode(_lstat(f).st_mode)
2123
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
2124
raise errors.NoSuchFile(f)
2128
def until_no_eintr(f, *a, **kw):
2129
"""Run f(*a, **kw), retrying if an EINTR error occurs.
2131
WARNING: you must be certain that it is safe to retry the call repeatedly
2132
if EINTR does occur. This is typically only true for low-level operations
2133
like os.read. If in any doubt, don't use this.
2135
Keep in mind that this is not a complete solution to EINTR. There is
2136
probably code in the Python standard library and other dependencies that
2137
may encounter EINTR if a signal arrives (and there is signal handler for
2138
that signal). So this function can reduce the impact for IO that bzrlib
2139
directly controls, but it is not a complete solution.
2141
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2145
except (IOError, OSError), e:
2146
if e.errno == errno.EINTR:
2151
def re_compile_checked(re_string, flags=0, where=""):
2152
"""Return a compiled re, or raise a sensible error.
2154
This should only be used when compiling user-supplied REs.
2156
:param re_string: Text form of regular expression.
2157
:param flags: eg re.IGNORECASE
2158
:param where: Message explaining to the user the context where
2159
it occurred, eg 'log search filter'.
2161
# from https://bugs.launchpad.net/bzr/+bug/251352
2163
re_obj = re.compile(re_string, flags)
2168
where = ' in ' + where
2169
# despite the name 'error' is a type
2170
raise errors.BzrCommandError('Invalid regular expression%s: %r: %s'
2171
% (where, re_string, e))
2174
if sys.platform == "win32":
2177
return msvcrt.getch()
2182
fd = sys.stdin.fileno()
2183
settings = termios.tcgetattr(fd)
2186
ch = sys.stdin.read(1)
2188
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2192
if sys.platform == 'linux2':
2193
def _local_concurrency():
2195
prefix = 'processor'
2196
for line in file('/proc/cpuinfo', 'rb'):
2197
if line.startswith(prefix):
2198
concurrency = int(line[line.find(':')+1:]) + 1
2200
elif sys.platform == 'darwin':
2201
def _local_concurrency():
2202
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2203
stdout=subprocess.PIPE).communicate()[0]
2204
elif sys.platform[0:7] == 'freebsd':
2205
def _local_concurrency():
2206
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2207
stdout=subprocess.PIPE).communicate()[0]
2208
elif sys.platform == 'sunos5':
2209
def _local_concurrency():
2210
return subprocess.Popen(['psrinfo', '-p',],
2211
stdout=subprocess.PIPE).communicate()[0]
2212
elif sys.platform == "win32":
2213
def _local_concurrency():
2214
# This appears to return the number of cores.
2215
return os.environ.get('NUMBER_OF_PROCESSORS')
2217
def _local_concurrency():
2222
_cached_local_concurrency = None
2224
def local_concurrency(use_cache=True):
2225
"""Return how many processes can be run concurrently.
2227
Rely on platform specific implementations and default to 1 (one) if
2228
anything goes wrong.
2230
global _cached_local_concurrency
2232
if _cached_local_concurrency is not None and use_cache:
2233
return _cached_local_concurrency
2235
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2236
if concurrency is None:
2238
concurrency = _local_concurrency()
2239
except (OSError, IOError):
2242
concurrency = int(concurrency)
2243
except (TypeError, ValueError):
2246
_cached_concurrency = concurrency
2250
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2251
"""A stream writer that doesn't decode str arguments."""
2253
def __init__(self, encode, stream, errors='strict'):
2254
codecs.StreamWriter.__init__(self, stream, errors)
2255
self.encode = encode
2257
def write(self, object):
2258
if type(object) is str:
2259
self.stream.write(object)
2261
data, _ = self.encode(object, self.errors)
2262
self.stream.write(data)
2264
if sys.platform == 'win32':
2265
def open_file(filename, mode='r', bufsize=-1):
2266
"""This function is used to override the ``open`` builtin.
2268
But it uses O_NOINHERIT flag so the file handle is not inherited by
2269
child processes. Deleting or renaming a closed file opened with this
2270
function is not blocking child processes.
2272
writing = 'w' in mode
2273
appending = 'a' in mode
2274
updating = '+' in mode
2275
binary = 'b' in mode
2278
# see http://msdn.microsoft.com/en-us/library/yeby3zcb%28VS.71%29.aspx
2279
# for flags for each modes.
2289
flags |= os.O_WRONLY
2290
flags |= os.O_CREAT | os.O_TRUNC
2295
flags |= os.O_WRONLY
2296
flags |= os.O_CREAT | os.O_APPEND
2301
flags |= os.O_RDONLY
2303
return os.fdopen(os.open(filename, flags), mode, bufsize)
2308
def getuser_unicode():
2309
"""Return the username as unicode.
2312
user_encoding = get_user_encoding()
2313
username = getpass.getuser().decode(user_encoding)
2314
except UnicodeDecodeError:
2315
raise errors.BzrError("Can't decode username as %s." % \
467
# XXX This should raise a NotChildPath exception, as its not tied
469
raise NotBranchError("path %r is not within branch %r" % (rp, base))
471
return os.sep.join(s)
added
removed
bzrlib/osutils.py
