59
147
# TODO: I'm not really sure this is the best format either.x
149
if _QUOTE_RE is None:
62
150
_QUOTE_RE = re.compile(r'([^a-zA-Z0-9.,:/\\_~-])')
64
152
if _QUOTE_RE.search(f):
65
153
return '"' + f + '"'
71
mode = os.lstat(f)[ST_MODE]
158
_directory_kind = 'directory'
161
"""Return the current umask"""
162
# Assume that people aren't messing with the umask while running
163
# XXX: This is not thread safe, but there is no way to get the
164
# umask without setting it
172
_directory_kind: "/",
174
'tree-reference': '+',
90
178
def kind_marker(kind):
180
return _kind_marker_map[kind]
182
# Slightly faster than using .get(, '') when the common case is that
93
elif kind == 'directory':
95
elif kind == 'symlink':
98
raise BzrError('invalid file kind %r' % kind)
102
if hasattr(os, 'lstat'):
108
if e.errno == errno.ENOENT:
111
raise BzrError("lstat/stat of (%r): %r" % (f, e))
187
lexists = getattr(os.path, 'lexists', None)
191
stat = getattr(os, 'lstat', os.stat)
195
if e.errno == errno.ENOENT:
198
raise errors.BzrError("lstat/stat of (%r): %r" % (f, e))
201
def fancy_rename(old, new, rename_func, unlink_func):
202
"""A fancy rename, when you don't have atomic rename.
204
:param old: The old path, to rename from
205
:param new: The new path, to rename to
206
:param rename_func: The potentially non-atomic rename function
207
:param unlink_func: A way to delete the target file if the full rename
210
# sftp rename doesn't allow overwriting, so play tricks:
211
base = os.path.basename(new)
212
dirname = os.path.dirname(new)
213
# callers use different encodings for the paths so the following MUST
214
# respect that. We rely on python upcasting to unicode if new is unicode
215
# and keeping a str if not.
216
tmp_name = 'tmp.%s.%.9f.%d.%s' % (base, time.time(),
217
os.getpid(), rand_chars(10))
218
tmp_name = pathjoin(dirname, tmp_name)
220
# Rename the file out of the way, but keep track if it didn't exist
221
# We don't want to grab just any exception
222
# something like EACCES should prevent us from continuing
223
# The downside is that the rename_func has to throw an exception
224
# with an errno = ENOENT, or NoSuchFile
227
rename_func(new, tmp_name)
228
except (errors.NoSuchFile,), e:
231
# RBC 20060103 abstraction leakage: the paramiko SFTP clients rename
232
# function raises an IOError with errno is None when a rename fails.
233
# This then gets caught here.
234
if e.errno not in (None, errno.ENOENT, errno.ENOTDIR):
237
if (getattr(e, 'errno', None) is None
238
or e.errno not in (errno.ENOENT, errno.ENOTDIR)):
247
# This may throw an exception, in which case success will
249
rename_func(old, new)
251
except (IOError, OSError), e:
252
# source and target may be aliases of each other (e.g. on a
253
# case-insensitive filesystem), so we may have accidentally renamed
254
# source by when we tried to rename target
255
failure_exc = sys.exc_info()
256
if (file_existed and e.errno in (None, errno.ENOENT)
257
and old.lower() == new.lower()):
258
# source and target are the same file on a case-insensitive
259
# filesystem, so we don't generate an exception
263
# If the file used to exist, rename it back into place
264
# otherwise just delete it from the tmp location
266
unlink_func(tmp_name)
268
rename_func(tmp_name, new)
269
if failure_exc is not None:
270
raise failure_exc[0], failure_exc[1], failure_exc[2]
273
# In Python 2.4.2 and older, os.path.abspath and os.path.realpath
274
# choke on a Unicode string containing a relative path if
275
# os.getcwd() returns a non-sys.getdefaultencoding()-encoded
277
_fs_enc = sys.getfilesystemencoding() or 'utf-8'
278
def _posix_abspath(path):
279
# jam 20060426 rather than encoding to fsencoding
280
# copy posixpath.abspath, but use os.getcwdu instead
281
if not posixpath.isabs(path):
282
path = posixpath.join(getcwd(), path)
283
return posixpath.normpath(path)
286
def _posix_realpath(path):
287
return posixpath.realpath(path.encode(_fs_enc)).decode(_fs_enc)
290
def _win32_fixdrive(path):
291
"""Force drive letters to be consistent.
293
win32 is inconsistent whether it returns lower or upper case
294
and even if it was consistent the user might type the other
295
so we force it to uppercase
296
running python.exe under cmd.exe return capital C:\\
297
running win32 python inside a cygwin shell returns lowercase c:\\
299
drive, path = _nt_splitdrive(path)
300
return drive.upper() + path
303
def _win32_abspath(path):
304
# Real _nt_abspath doesn't have a problem with a unicode cwd
305
return _win32_fixdrive(_nt_abspath(unicode(path)).replace('\\', '/'))
308
def _win98_abspath(path):
309
"""Return the absolute version of a path.
310
Windows 98 safe implementation (python reimplementation
311
of Win32 API function GetFullPathNameW)
316
# \\HOST\path => //HOST/path
317
# //HOST/path => //HOST/path
318
# path => C:/cwd/path
321
# check for absolute path
322
drive = _nt_splitdrive(path)[0]
323
if drive == '' and path[:2] not in('//','\\\\'):
325
# we cannot simply os.path.join cwd and path
326
# because os.path.join('C:','/path') produce '/path'
327
# and this is incorrect
328
if path[:1] in ('/','\\'):
329
cwd = _nt_splitdrive(cwd)[0]
331
path = cwd + '\\' + path
332
return _win32_fixdrive(_nt_normpath(path).replace('\\', '/'))
335
def _win32_realpath(path):
336
# Real _nt_realpath doesn't have a problem with a unicode cwd
337
return _win32_fixdrive(_nt_realpath(unicode(path)).replace('\\', '/'))
340
def _win32_pathjoin(*args):
341
return _nt_join(*args).replace('\\', '/')
344
def _win32_normpath(path):
345
return _win32_fixdrive(_nt_normpath(unicode(path)).replace('\\', '/'))
349
return _win32_fixdrive(os.getcwdu().replace('\\', '/'))
352
def _win32_mkdtemp(*args, **kwargs):
353
return _win32_fixdrive(tempfile.mkdtemp(*args, **kwargs).replace('\\', '/'))
356
def _win32_rename(old, new):
357
"""We expect to be able to atomically replace 'new' with old.
359
On win32, if new exists, it must be moved out of the way first,
363
fancy_rename(old, new, rename_func=os.rename, unlink_func=os.unlink)
365
if e.errno in (errno.EPERM, errno.EACCES, errno.EBUSY, errno.EINVAL):
366
# If we try to rename a non-existant file onto cwd, we get
367
# EPERM or EACCES instead of ENOENT, this will raise ENOENT
368
# if the old path doesn't exist, sometimes we get EACCES
369
# On Linux, we seem to get EBUSY, on Mac we get EINVAL
375
return unicodedata.normalize('NFC', os.getcwdu())
378
# Default is to just use the python builtins, but these can be rebound on
379
# particular platforms.
380
abspath = _posix_abspath
381
realpath = _posix_realpath
382
pathjoin = os.path.join
383
normpath = os.path.normpath
386
dirname = os.path.dirname
387
basename = os.path.basename
388
split = os.path.split
389
splitext = os.path.splitext
390
# These were already imported into local scope
391
# mkdtemp = tempfile.mkdtemp
392
# rmtree = shutil.rmtree
394
MIN_ABS_PATHLENGTH = 1
397
if sys.platform == 'win32':
398
if win32utils.winver == 'Windows 98':
399
abspath = _win98_abspath
401
abspath = _win32_abspath
402
realpath = _win32_realpath
403
pathjoin = _win32_pathjoin
404
normpath = _win32_normpath
405
getcwd = _win32_getcwd
406
mkdtemp = _win32_mkdtemp
407
rename = _win32_rename
409
MIN_ABS_PATHLENGTH = 3
411
def _win32_delete_readonly(function, path, excinfo):
412
"""Error handler for shutil.rmtree function [for win32]
413
Helps to remove files and dirs marked as read-only.
415
exception = excinfo[1]
416
if function in (os.remove, os.rmdir) \
417
and isinstance(exception, OSError) \
418
and exception.errno == errno.EACCES:
424
def rmtree(path, ignore_errors=False, onerror=_win32_delete_readonly):
425
"""Replacer for shutil.rmtree: could remove readonly dirs/files"""
426
return shutil.rmtree(path, ignore_errors, onerror)
428
f = win32utils.get_unicode_argv # special function or None
432
elif sys.platform == 'darwin':
436
def get_terminal_encoding():
437
"""Find the best encoding for printing to the screen.
439
This attempts to check both sys.stdout and sys.stdin to see
440
what encoding they are in, and if that fails it falls back to
441
osutils.get_user_encoding().
442
The problem is that on Windows, locale.getpreferredencoding()
443
is not the same encoding as that used by the console:
444
http://mail.python.org/pipermail/python-list/2003-May/162357.html
446
On my standard US Windows XP, the preferred encoding is
447
cp1252, but the console is cp437
449
from bzrlib.trace import mutter
450
output_encoding = getattr(sys.stdout, 'encoding', None)
451
if not output_encoding:
452
input_encoding = getattr(sys.stdin, 'encoding', None)
453
if not input_encoding:
454
output_encoding = get_user_encoding()
455
mutter('encoding stdout as osutils.get_user_encoding() %r',
458
output_encoding = input_encoding
459
mutter('encoding stdout as sys.stdin encoding %r', output_encoding)
461
mutter('encoding stdout as sys.stdout encoding %r', output_encoding)
462
if output_encoding == 'cp0':
463
# invalid encoding (cp0 means 'no codepage' on Windows)
464
output_encoding = get_user_encoding()
465
mutter('cp0 is invalid encoding.'
466
' encoding stdout as osutils.get_user_encoding() %r',
470
codecs.lookup(output_encoding)
472
sys.stderr.write('bzr: warning:'
473
' unknown terminal encoding %s.\n'
474
' Using encoding %s instead.\n'
475
% (output_encoding, get_user_encoding())
477
output_encoding = get_user_encoding()
479
return output_encoding
113
482
def normalizepath(f):
114
if hasattr(os.path, 'realpath'):
483
if getattr(os.path, 'realpath', None) is not None:
118
487
[p,e] = os.path.split(f)
119
488
if e == "" or e == "." or e == "..":
122
return os.path.join(F(p), e)
126
"""Copy a file to a backup.
128
Backups are named in GNU-style, with a ~ suffix.
130
If the file is already a backup, it's not copied.
136
if has_symlinks() and os.path.islink(fn):
137
target = os.readlink(fn)
138
os.symlink(target, bfn)
146
outf = file(bfn, 'wb')
491
return pathjoin(F(p), e)
1124
def relpath(base, path):
1125
"""Return path relative to base, or raise exception.
1127
The path may be either an absolute path or a path relative to the
1128
current working directory.
1130
os.path.commonprefix (python2.4) has a bad bug that it works just
1131
on string prefixes, assuming that '/u' is a prefix of '/u2'. This
1132
avoids that problem.
1135
if len(base) < MIN_ABS_PATHLENGTH:
1136
# must have space for e.g. a drive letter
1137
raise ValueError('%r is too short to calculate a relative path'
1145
if len(head) <= len(base) and head != base:
1146
raise errors.PathNotChild(rp, base)
1149
head, tail = split(head)
1154
return pathjoin(*reversed(s))
1159
def _cicp_canonical_relpath(base, path):
1160
"""Return the canonical path relative to base.
1162
Like relpath, but on case-insensitive-case-preserving file-systems, this
1163
will return the relpath as stored on the file-system rather than in the
1164
case specified in the input string, for all existing portions of the path.
1166
This will cause O(N) behaviour if called for every path in a tree; if you
1167
have a number of paths to convert, you should use canonical_relpaths().
1169
# TODO: it should be possible to optimize this for Windows by using the
1170
# win32 API FindFiles function to look for the specified name - but using
1171
# os.listdir() still gives us the correct, platform agnostic semantics in
1174
rel = relpath(base, path)
1175
# '.' will have been turned into ''
1179
abs_base = abspath(base)
1181
_listdir = os.listdir
1183
# use an explicit iterator so we can easily consume the rest on early exit.
1184
bit_iter = iter(rel.split('/'))
1185
for bit in bit_iter:
1188
next_entries = _listdir(current)
1189
except OSError: # enoent, eperm, etc
1190
# We can't find this in the filesystem, so just append the
1192
current = pathjoin(current, bit, *list(bit_iter))
1194
for look in next_entries:
1195
if lbit == look.lower():
1196
current = pathjoin(current, look)
1199
# got to the end, nothing matched, so we just return the
1200
# non-existing bits as they were specified (the filename may be
1201
# the target of a move, for example).
1202
current = pathjoin(current, bit, *list(bit_iter))
1204
return current[len(abs_base):].lstrip('/')
1206
# XXX - TODO - we need better detection/integration of case-insensitive
1207
# file-systems; Linux often sees FAT32 devices (or NFS-mounted OSX
1208
# filesystems), for example, so could probably benefit from the same basic
1209
# support there. For now though, only Windows and OSX get that support, and
1210
# they get it for *all* file-systems!
1211
if sys.platform in ('win32', 'darwin'):
1212
canonical_relpath = _cicp_canonical_relpath
1214
canonical_relpath = relpath
1216
def canonical_relpaths(base, paths):
1217
"""Create an iterable to canonicalize a sequence of relative paths.
1219
The intent is for this implementation to use a cache, vastly speeding
1220
up multiple transformations in the same directory.
1222
# but for now, we haven't optimized...
1223
return [canonical_relpath(base, p) for p in paths]
1225
def safe_unicode(unicode_or_utf8_string):
1226
"""Coerce unicode_or_utf8_string into unicode.
1228
If it is unicode, it is returned.
1229
Otherwise it is decoded from utf-8. If decoding fails, the exception is
1230
wrapped in a BzrBadParameterNotUnicode exception.
1232
if isinstance(unicode_or_utf8_string, unicode):
1233
return unicode_or_utf8_string
1235
return unicode_or_utf8_string.decode('utf8')
1236
except UnicodeDecodeError:
1237
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1240
def safe_utf8(unicode_or_utf8_string):
1241
"""Coerce unicode_or_utf8_string to a utf8 string.
1243
If it is a str, it is returned.
1244
If it is Unicode, it is encoded into a utf-8 string.
1246
if isinstance(unicode_or_utf8_string, str):
1247
# TODO: jam 20070209 This is overkill, and probably has an impact on
1248
# performance if we are dealing with lots of apis that want a
1251
# Make sure it is a valid utf-8 string
1252
unicode_or_utf8_string.decode('utf-8')
1253
except UnicodeDecodeError:
1254
raise errors.BzrBadParameterNotUnicode(unicode_or_utf8_string)
1255
return unicode_or_utf8_string
1256
return unicode_or_utf8_string.encode('utf-8')
1259
_revision_id_warning = ('Unicode revision ids were deprecated in bzr 0.15.'
1260
' Revision id generators should be creating utf8'
1264
def safe_revision_id(unicode_or_utf8_string, warn=True):
1265
"""Revision ids should now be utf8, but at one point they were unicode.
1267
:param unicode_or_utf8_string: A possibly Unicode revision_id. (can also be
1269
:param warn: Functions that are sanitizing user data can set warn=False
1270
:return: None or a utf8 revision id.
1272
if (unicode_or_utf8_string is None
1273
or unicode_or_utf8_string.__class__ == str):
1274
return unicode_or_utf8_string
1276
symbol_versioning.warn(_revision_id_warning, DeprecationWarning,
1278
return cache_utf8.encode(unicode_or_utf8_string)
1281
_file_id_warning = ('Unicode file ids were deprecated in bzr 0.15. File id'
1282
' generators should be creating utf8 file ids.')
1285
def safe_file_id(unicode_or_utf8_string, warn=True):
1286
"""File ids should now be utf8, but at one point they were unicode.
1288
This is the same as safe_utf8, except it uses the cached encode functions
1289
to save a little bit of performance.
1291
:param unicode_or_utf8_string: A possibly Unicode file_id. (can also be
1293
:param warn: Functions that are sanitizing user data can set warn=False
1294
:return: None or a utf8 file id.
1296
if (unicode_or_utf8_string is None
1297
or unicode_or_utf8_string.__class__ == str):
1298
return unicode_or_utf8_string
1300
symbol_versioning.warn(_file_id_warning, DeprecationWarning,
1302
return cache_utf8.encode(unicode_or_utf8_string)
1305
_platform_normalizes_filenames = False
1306
if sys.platform == 'darwin':
1307
_platform_normalizes_filenames = True
1310
def normalizes_filenames():
1311
"""Return True if this platform normalizes unicode filenames.
1313
Mac OSX does, Windows/Linux do not.
1315
return _platform_normalizes_filenames
1318
def _accessible_normalized_filename(path):
1319
"""Get the unicode normalized path, and if you can access the file.
1321
On platforms where the system normalizes filenames (Mac OSX),
1322
you can access a file by any path which will normalize correctly.
1323
On platforms where the system does not normalize filenames
1324
(Windows, Linux), you have to access a file by its exact path.
1326
Internally, bzr only supports NFC normalization, since that is
1327
the standard for XML documents.
1329
So return the normalized path, and a flag indicating if the file
1330
can be accessed by that path.
1333
return unicodedata.normalize('NFC', unicode(path)), True
1336
def _inaccessible_normalized_filename(path):
1337
__doc__ = _accessible_normalized_filename.__doc__
1339
normalized = unicodedata.normalize('NFC', unicode(path))
1340
return normalized, normalized == path
1343
if _platform_normalizes_filenames:
1344
normalized_filename = _accessible_normalized_filename
1346
normalized_filename = _inaccessible_normalized_filename
1349
default_terminal_width = 80
1350
"""The default terminal width for ttys.
1352
This is defined so that higher levels can share a common fallback value when
1353
terminal_width() returns None.
1357
def terminal_width():
1358
"""Return terminal width.
1360
None is returned if the width can't established precisely.
1363
- if BZR_COLUMNS is set, returns its value
1364
- if there is no controlling terminal, returns None
1365
- if COLUMNS is set, returns its value,
1367
From there, we need to query the OS to get the size of the controlling
1371
- get termios.TIOCGWINSZ
1372
- if an error occurs or a negative value is obtained, returns None
1376
- win32utils.get_console_size() decides,
1377
- returns None on error (provided default value)
1380
# If BZR_COLUMNS is set, take it, user is always right
1382
return int(os.environ['BZR_COLUMNS'])
1383
except (KeyError, ValueError):
1386
isatty = getattr(sys.stdout, 'isatty', None)
1387
if isatty is None or not isatty():
1388
# Don't guess, setting BZR_COLUMNS is the recommended way to override.
1391
# If COLUMNS is set, take it, the terminal knows better (even inside a
1392
# given terminal, the application can decide to set COLUMNS to a lower
1393
# value (splitted screen) or a bigger value (scroll bars))
1395
return int(os.environ['COLUMNS'])
1396
except (KeyError, ValueError):
1399
width, height = _terminal_size(None, None)
1401
# Consider invalid values as meaning no width
1407
def _win32_terminal_size(width, height):
1408
width, height = win32utils.get_console_size(defaultx=width, defaulty=height)
1409
return width, height
1412
def _ioctl_terminal_size(width, height):
1414
import struct, fcntl, termios
1415
s = struct.pack('HHHH', 0, 0, 0, 0)
1416
x = fcntl.ioctl(1, termios.TIOCGWINSZ, s)
1417
height, width = struct.unpack('HHHH', x)[0:2]
1418
except (IOError, AttributeError):
1420
return width, height
1422
_terminal_size = None
1423
"""Returns the terminal size as (width, height).
1425
:param width: Default value for width.
1426
:param height: Default value for height.
1428
This is defined specifically for each OS and query the size of the controlling
1429
terminal. If any error occurs, the provided default values should be returned.
1431
if sys.platform == 'win32':
1432
_terminal_size = _win32_terminal_size
1434
_terminal_size = _ioctl_terminal_size
1437
def _terminal_size_changed(signum, frame):
1438
"""Set COLUMNS upon receiving a SIGnal for WINdow size CHange."""
1439
width, height = _terminal_size(None, None)
1440
if width is not None:
1441
os.environ['COLUMNS'] = str(width)
1444
_registered_sigwinch = False
1446
def watch_sigwinch():
1447
"""Register for SIGWINCH, once and only once."""
1448
global _registered_sigwinch
1449
if not _registered_sigwinch:
1450
if sys.platform == 'win32':
1451
# Martin (gz) mentioned WINDOW_BUFFER_SIZE_RECORD from
1452
# ReadConsoleInput but I've no idea how to plug that in
1453
# the current design -- vila 20091216
1456
signal.signal(signal.SIGWINCH, _terminal_size_changed)
1457
_registered_sigwinch = True
1460
def supports_executable():
1461
return sys.platform != "win32"
1464
def supports_posix_readonly():
1465
"""Return True if 'readonly' has POSIX semantics, False otherwise.
1467
Notably, a win32 readonly file cannot be deleted, unlike POSIX where the
1468
directory controls creation/deletion, etc.
1470
And under win32, readonly means that the directory itself cannot be
1471
deleted. The contents of a readonly directory can be changed, unlike POSIX
1472
where files in readonly directories cannot be added, deleted or renamed.
1474
return sys.platform != "win32"
1477
def set_or_unset_env(env_variable, value):
1478
"""Modify the environment, setting or removing the env_variable.
1480
:param env_variable: The environment variable in question
1481
:param value: The value to set the environment to. If None, then
1482
the variable will be removed.
1483
:return: The original value of the environment variable.
1485
orig_val = os.environ.get(env_variable)
1487
if orig_val is not None:
1488
del os.environ[env_variable]
1490
if isinstance(value, unicode):
1491
value = value.encode(get_user_encoding())
1492
os.environ[env_variable] = value
1496
_validWin32PathRE = re.compile(r'^([A-Za-z]:[/\\])?[^:<>*"?\|]*$')
1499
def check_legal_path(path):
1500
"""Check whether the supplied path is legal.
1501
This is only required on Windows, so we don't test on other platforms
1504
if sys.platform != "win32":
1506
if _validWin32PathRE.match(path) is None:
1507
raise errors.IllegalPath(path)
1510
_WIN32_ERROR_DIRECTORY = 267 # Similar to errno.ENOTDIR
1512
def _is_error_enotdir(e):
1513
"""Check if this exception represents ENOTDIR.
1515
Unfortunately, python is very inconsistent about the exception
1516
here. The cases are:
1517
1) Linux, Mac OSX all versions seem to set errno == ENOTDIR
1518
2) Windows, Python2.4, uses errno == ERROR_DIRECTORY (267)
1519
which is the windows error code.
1520
3) Windows, Python2.5 uses errno == EINVAL and
1521
winerror == ERROR_DIRECTORY
1523
:param e: An Exception object (expected to be OSError with an errno
1524
attribute, but we should be able to cope with anything)
1525
:return: True if this represents an ENOTDIR error. False otherwise.
1527
en = getattr(e, 'errno', None)
1528
if (en == errno.ENOTDIR
1529
or (sys.platform == 'win32'
1530
and (en == _WIN32_ERROR_DIRECTORY
1531
or (en == errno.EINVAL
1532
and getattr(e, 'winerror', None) == _WIN32_ERROR_DIRECTORY)
1538
def walkdirs(top, prefix=""):
1539
"""Yield data about all the directories in a tree.
1541
This yields all the data about the contents of a directory at a time.
1542
After each directory has been yielded, if the caller has mutated the list
1543
to exclude some directories, they are then not descended into.
1545
The data yielded is of the form:
1546
((directory-relpath, directory-path-from-top),
1547
[(relpath, basename, kind, lstat, path-from-top), ...]),
1548
- directory-relpath is the relative path of the directory being returned
1549
with respect to top. prefix is prepended to this.
1550
- directory-path-from-root is the path including top for this directory.
1551
It is suitable for use with os functions.
1552
- relpath is the relative path within the subtree being walked.
1553
- basename is the basename of the path
1554
- kind is the kind of the file now. If unknown then the file is not
1555
present within the tree - but it may be recorded as versioned. See
1557
- lstat is the stat data *if* the file was statted.
1558
- planned, not implemented:
1559
path_from_tree_root is the path from the root of the tree.
1561
:param prefix: Prefix the relpaths that are yielded with 'prefix'. This
1562
allows one to walk a subtree but get paths that are relative to a tree
1564
:return: an iterator over the dirs.
1566
#TODO there is a bit of a smell where the results of the directory-
1567
# summary in this, and the path from the root, may not agree
1568
# depending on top and prefix - i.e. ./foo and foo as a pair leads to
1569
# potentially confusing output. We should make this more robust - but
1570
# not at a speed cost. RBC 20060731
1572
_directory = _directory_kind
1573
_listdir = os.listdir
1574
_kind_from_mode = file_kind_from_stat_mode
1575
pending = [(safe_unicode(prefix), "", _directory, None, safe_unicode(top))]
1577
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1578
relroot, _, _, _, top = pending.pop()
1580
relprefix = relroot + u'/'
1583
top_slash = top + u'/'
1586
append = dirblock.append
1588
names = sorted(_listdir(top))
1590
if not _is_error_enotdir(e):
1594
abspath = top_slash + name
1595
statvalue = _lstat(abspath)
1596
kind = _kind_from_mode(statvalue.st_mode)
1597
append((relprefix + name, name, kind, statvalue, abspath))
1598
yield (relroot, top), dirblock
1600
# push the user specified dirs from dirblock
1601
pending.extend(d for d in reversed(dirblock) if d[2] == _directory)
1604
class DirReader(object):
1605
"""An interface for reading directories."""
1607
def top_prefix_to_starting_dir(self, top, prefix=""):
1608
"""Converts top and prefix to a starting dir entry
1610
:param top: A utf8 path
1611
:param prefix: An optional utf8 path to prefix output relative paths
1613
:return: A tuple starting with prefix, and ending with the native
1616
raise NotImplementedError(self.top_prefix_to_starting_dir)
1618
def read_dir(self, prefix, top):
1619
"""Read a specific dir.
1621
:param prefix: A utf8 prefix to be preprended to the path basenames.
1622
:param top: A natively encoded path to read.
1623
:return: A list of the directories contents. Each item contains:
1624
(utf8_relpath, utf8_name, kind, lstatvalue, native_abspath)
1626
raise NotImplementedError(self.read_dir)
1629
_selected_dir_reader = None
1632
def _walkdirs_utf8(top, prefix=""):
1633
"""Yield data about all the directories in a tree.
1635
This yields the same information as walkdirs() only each entry is yielded
1636
in utf-8. On platforms which have a filesystem encoding of utf8 the paths
1637
are returned as exact byte-strings.
1639
:return: yields a tuple of (dir_info, [file_info])
1640
dir_info is (utf8_relpath, path-from-top)
1641
file_info is (utf8_relpath, utf8_name, kind, lstat, path-from-top)
1642
if top is an absolute path, path-from-top is also an absolute path.
1643
path-from-top might be unicode or utf8, but it is the correct path to
1644
pass to os functions to affect the file in question. (such as os.lstat)
1646
global _selected_dir_reader
1647
if _selected_dir_reader is None:
1648
fs_encoding = _fs_enc.upper()
1649
if sys.platform == "win32" and win32utils.winver == 'Windows NT':
1650
# Win98 doesn't have unicode apis like FindFirstFileW
1651
# TODO: We possibly could support Win98 by falling back to the
1652
# original FindFirstFile, and using TCHAR instead of WCHAR,
1653
# but that gets a bit tricky, and requires custom compiling
1656
from bzrlib._walkdirs_win32 import Win32ReadDir
1657
_selected_dir_reader = Win32ReadDir()
1660
elif fs_encoding in ('UTF-8', 'US-ASCII', 'ANSI_X3.4-1968'):
1661
# ANSI_X3.4-1968 is a form of ASCII
1663
from bzrlib._readdir_pyx import UTF8DirReader
1664
_selected_dir_reader = UTF8DirReader()
1665
except ImportError, e:
1666
failed_to_load_extension(e)
1669
if _selected_dir_reader is None:
1670
# Fallback to the python version
1671
_selected_dir_reader = UnicodeDirReader()
1673
# 0 - relpath, 1- basename, 2- kind, 3- stat, 4-toppath
1674
# But we don't actually uses 1-3 in pending, so set them to None
1675
pending = [[_selected_dir_reader.top_prefix_to_starting_dir(top, prefix)]]
1676
read_dir = _selected_dir_reader.read_dir
1677
_directory = _directory_kind
1679
relroot, _, _, _, top = pending[-1].pop()
1682
dirblock = sorted(read_dir(relroot, top))
1683
yield (relroot, top), dirblock
1684
# push the user specified dirs from dirblock
1685
next = [d for d in reversed(dirblock) if d[2] == _directory]
1687
pending.append(next)
1690
class UnicodeDirReader(DirReader):
1691
"""A dir reader for non-utf8 file systems, which transcodes."""
1693
__slots__ = ['_utf8_encode']
1696
self._utf8_encode = codecs.getencoder('utf8')
1698
def top_prefix_to_starting_dir(self, top, prefix=""):
1699
"""See DirReader.top_prefix_to_starting_dir."""
1700
return (safe_utf8(prefix), None, None, None, safe_unicode(top))
1702
def read_dir(self, prefix, top):
1703
"""Read a single directory from a non-utf8 file system.
1705
top, and the abspath element in the output are unicode, all other paths
1706
are utf8. Local disk IO is done via unicode calls to listdir etc.
1708
This is currently the fallback code path when the filesystem encoding is
1709
not UTF-8. It may be better to implement an alternative so that we can
1710
safely handle paths that are not properly decodable in the current
1713
See DirReader.read_dir for details.
1715
_utf8_encode = self._utf8_encode
1717
_listdir = os.listdir
1718
_kind_from_mode = file_kind_from_stat_mode
1721
relprefix = prefix + '/'
1724
top_slash = top + u'/'
1727
append = dirblock.append
1728
for name in sorted(_listdir(top)):
1730
name_utf8 = _utf8_encode(name)[0]
1731
except UnicodeDecodeError:
1732
raise errors.BadFilenameEncoding(
1733
_utf8_encode(relprefix)[0] + name, _fs_enc)
1734
abspath = top_slash + name
1735
statvalue = _lstat(abspath)
1736
kind = _kind_from_mode(statvalue.st_mode)
1737
append((relprefix + name_utf8, name_utf8, kind, statvalue, abspath))
1741
def copy_tree(from_path, to_path, handlers={}):
1742
"""Copy all of the entries in from_path into to_path.
1744
:param from_path: The base directory to copy.
1745
:param to_path: The target directory. If it does not exist, it will
1747
:param handlers: A dictionary of functions, which takes a source and
1748
destinations for files, directories, etc.
1749
It is keyed on the file kind, such as 'directory', 'symlink', or 'file'
1750
'file', 'directory', and 'symlink' should always exist.
1751
If they are missing, they will be replaced with 'os.mkdir()',
1752
'os.readlink() + os.symlink()', and 'shutil.copy2()', respectively.
1754
# Now, just copy the existing cached tree to the new location
1755
# We use a cheap trick here.
1756
# Absolute paths are prefixed with the first parameter
1757
# relative paths are prefixed with the second.
1758
# So we can get both the source and target returned
1759
# without any extra work.
1761
def copy_dir(source, dest):
1764
def copy_link(source, dest):
1765
"""Copy the contents of a symlink"""
1766
link_to = os.readlink(source)
1767
os.symlink(link_to, dest)
1769
real_handlers = {'file':shutil.copy2,
1770
'symlink':copy_link,
1771
'directory':copy_dir,
1773
real_handlers.update(handlers)
1775
if not os.path.exists(to_path):
1776
real_handlers['directory'](from_path, to_path)
1778
for dir_info, entries in walkdirs(from_path, prefix=to_path):
1779
for relpath, name, kind, st, abspath in entries:
1780
real_handlers[kind](abspath, relpath)
1783
def path_prefix_key(path):
1784
"""Generate a prefix-order path key for path.
1786
This can be used to sort paths in the same way that walkdirs does.
1788
return (dirname(path) , path)
1791
def compare_paths_prefix_order(path_a, path_b):
1792
"""Compare path_a and path_b to generate the same order walkdirs uses."""
1793
key_a = path_prefix_key(path_a)
1794
key_b = path_prefix_key(path_b)
1795
return cmp(key_a, key_b)
1798
_cached_user_encoding = None
1801
def get_user_encoding(use_cache=True):
1802
"""Find out what the preferred user encoding is.
1804
This is generally the encoding that is used for command line parameters
1805
and file contents. This may be different from the terminal encoding
1806
or the filesystem encoding.
1808
:param use_cache: Enable cache for detected encoding.
1809
(This parameter is turned on by default,
1810
and required only for selftesting)
1812
:return: A string defining the preferred user encoding
1814
global _cached_user_encoding
1815
if _cached_user_encoding is not None and use_cache:
1816
return _cached_user_encoding
1818
if sys.platform == 'darwin':
1819
# python locale.getpreferredencoding() always return
1820
# 'mac-roman' on darwin. That's a lie.
1821
sys.platform = 'posix'
1823
if os.environ.get('LANG', None) is None:
1824
# If LANG is not set, we end up with 'ascii', which is bad
1825
# ('mac-roman' is more than ascii), so we set a default which
1826
# will give us UTF-8 (which appears to work in all cases on
1827
# OSX). Users are still free to override LANG of course, as
1828
# long as it give us something meaningful. This work-around
1829
# *may* not be needed with python 3k and/or OSX 10.5, but will
1830
# work with them too -- vila 20080908
1831
os.environ['LANG'] = 'en_US.UTF-8'
1834
sys.platform = 'darwin'
1839
user_encoding = locale.getpreferredencoding()
1840
except locale.Error, e:
1841
sys.stderr.write('bzr: warning: %s\n'
1842
' Could not determine what text encoding to use.\n'
1843
' This error usually means your Python interpreter\n'
1844
' doesn\'t support the locale set by $LANG (%s)\n'
1845
" Continuing with ascii encoding.\n"
1846
% (e, os.environ.get('LANG')))
1847
user_encoding = 'ascii'
1849
# Windows returns 'cp0' to indicate there is no code page. So we'll just
1850
# treat that as ASCII, and not support printing unicode characters to the
1853
# For python scripts run under vim, we get '', so also treat that as ASCII
1854
if user_encoding in (None, 'cp0', ''):
1855
user_encoding = 'ascii'
1859
codecs.lookup(user_encoding)
1861
sys.stderr.write('bzr: warning:'
1862
' unknown encoding %s.'
1863
' Continuing with ascii encoding.\n'
1866
user_encoding = 'ascii'
1869
_cached_user_encoding = user_encoding
1871
return user_encoding
1874
def get_host_name():
1875
"""Return the current unicode host name.
1877
This is meant to be used in place of socket.gethostname() because that
1878
behaves inconsistently on different platforms.
1880
if sys.platform == "win32":
1882
return win32utils.get_host_name()
1885
return socket.gethostname().decode(get_user_encoding())
1888
def recv_all(socket, bytes):
1889
"""Receive an exact number of bytes.
1891
Regular Socket.recv() may return less than the requested number of bytes,
1892
dependning on what's in the OS buffer. MSG_WAITALL is not available
1893
on all platforms, but this should work everywhere. This will return
1894
less than the requested amount if the remote end closes.
1896
This isn't optimized and is intended mostly for use in testing.
1899
while len(b) < bytes:
1900
new = until_no_eintr(socket.recv, bytes - len(b))
1907
def send_all(socket, bytes, report_activity=None):
1908
"""Send all bytes on a socket.
1910
Regular socket.sendall() can give socket error 10053 on Windows. This
1911
implementation sends no more than 64k at a time, which avoids this problem.
1913
:param report_activity: Call this as bytes are read, see
1914
Transport._report_activity
1917
for pos in xrange(0, len(bytes), chunk_size):
1918
block = bytes[pos:pos+chunk_size]
1919
if report_activity is not None:
1920
report_activity(len(block), 'write')
1921
until_no_eintr(socket.sendall, block)
1924
def dereference_path(path):
1925
"""Determine the real path to a file.
1927
All parent elements are dereferenced. But the file itself is not
1929
:param path: The original path. May be absolute or relative.
1930
:return: the real path *to* the file
1932
parent, base = os.path.split(path)
1933
# The pathjoin for '.' is a workaround for Python bug #1213894.
1934
# (initial path components aren't dereferenced)
1935
return pathjoin(realpath(pathjoin('.', parent)), base)
1938
def supports_mapi():
1939
"""Return True if we can use MAPI to launch a mail client."""
1940
return sys.platform == "win32"
1943
def resource_string(package, resource_name):
1944
"""Load a resource from a package and return it as a string.
1946
Note: Only packages that start with bzrlib are currently supported.
1948
This is designed to be a lightweight implementation of resource
1949
loading in a way which is API compatible with the same API from
1951
http://peak.telecommunity.com/DevCenter/PkgResources#basic-resource-access.
1952
If and when pkg_resources becomes a standard library, this routine
1955
# Check package name is within bzrlib
1956
if package == "bzrlib":
1957
resource_relpath = resource_name
1958
elif package.startswith("bzrlib."):
1959
package = package[len("bzrlib."):].replace('.', os.sep)
1960
resource_relpath = pathjoin(package, resource_name)
1962
raise errors.BzrError('resource package %s not in bzrlib' % package)
1964
# Map the resource to a file and read its contents
1965
base = dirname(bzrlib.__file__)
1966
if getattr(sys, 'frozen', None): # bzr.exe
1967
base = abspath(pathjoin(base, '..', '..'))
1968
filename = pathjoin(base, resource_relpath)
1969
return open(filename, 'rU').read()
1972
def file_kind_from_stat_mode_thunk(mode):
1973
global file_kind_from_stat_mode
1974
if file_kind_from_stat_mode is file_kind_from_stat_mode_thunk:
1976
from bzrlib._readdir_pyx import UTF8DirReader
1977
file_kind_from_stat_mode = UTF8DirReader().kind_from_mode
1978
except ImportError, e:
1979
# This is one time where we won't warn that an extension failed to
1980
# load. The extension is never available on Windows anyway.
1981
from bzrlib._readdir_py import (
1982
_kind_from_mode as file_kind_from_stat_mode
1984
return file_kind_from_stat_mode(mode)
1985
file_kind_from_stat_mode = file_kind_from_stat_mode_thunk
1988
def file_kind(f, _lstat=os.lstat):
1990
return file_kind_from_stat_mode(_lstat(f).st_mode)
1992
if getattr(e, 'errno', None) in (errno.ENOENT, errno.ENOTDIR):
1993
raise errors.NoSuchFile(f)
1997
def until_no_eintr(f, *a, **kw):
1998
"""Run f(*a, **kw), retrying if an EINTR error occurs."""
1999
# Borrowed from Twisted's twisted.python.util.untilConcludes function.
2003
except (IOError, OSError), e:
2004
if e.errno == errno.EINTR:
2008
def re_compile_checked(re_string, flags=0, where=""):
2009
"""Return a compiled re, or raise a sensible error.
2011
This should only be used when compiling user-supplied REs.
2013
:param re_string: Text form of regular expression.
2014
:param flags: eg re.IGNORECASE
2015
:param where: Message explaining to the user the context where
2016
it occurred, eg 'log search filter'.
2018
# from https://bugs.launchpad.net/bzr/+bug/251352
2020
re_obj = re.compile(re_string, flags)
2025
where = ' in ' + where
2026
# despite the name 'error' is a type
2027
raise errors.BzrCommandError('Invalid regular expression%s: %r: %s'
2028
% (where, re_string, e))
2031
if sys.platform == "win32":
2034
return msvcrt.getch()
2039
fd = sys.stdin.fileno()
2040
settings = termios.tcgetattr(fd)
2043
ch = sys.stdin.read(1)
2045
termios.tcsetattr(fd, termios.TCSADRAIN, settings)
2049
if sys.platform == 'linux2':
2050
def _local_concurrency():
2052
prefix = 'processor'
2053
for line in file('/proc/cpuinfo', 'rb'):
2054
if line.startswith(prefix):
2055
concurrency = int(line[line.find(':')+1:]) + 1
2057
elif sys.platform == 'darwin':
2058
def _local_concurrency():
2059
return subprocess.Popen(['sysctl', '-n', 'hw.availcpu'],
2060
stdout=subprocess.PIPE).communicate()[0]
2061
elif sys.platform[0:7] == 'freebsd':
2062
def _local_concurrency():
2063
return subprocess.Popen(['sysctl', '-n', 'hw.ncpu'],
2064
stdout=subprocess.PIPE).communicate()[0]
2065
elif sys.platform == 'sunos5':
2066
def _local_concurrency():
2067
return subprocess.Popen(['psrinfo', '-p',],
2068
stdout=subprocess.PIPE).communicate()[0]
2069
elif sys.platform == "win32":
2070
def _local_concurrency():
2071
# This appears to return the number of cores.
2072
return os.environ.get('NUMBER_OF_PROCESSORS')
2074
def _local_concurrency():
2079
_cached_local_concurrency = None
2081
def local_concurrency(use_cache=True):
2082
"""Return how many processes can be run concurrently.
2084
Rely on platform specific implementations and default to 1 (one) if
2085
anything goes wrong.
2087
global _cached_local_concurrency
2089
if _cached_local_concurrency is not None and use_cache:
2090
return _cached_local_concurrency
2092
concurrency = os.environ.get('BZR_CONCURRENCY', None)
2093
if concurrency is None:
2095
concurrency = _local_concurrency()
2096
except (OSError, IOError):
2099
concurrency = int(concurrency)
2100
except (TypeError, ValueError):
2103
_cached_concurrency = concurrency
2107
class UnicodeOrBytesToBytesWriter(codecs.StreamWriter):
2108
"""A stream writer that doesn't decode str arguments."""
2110
def __init__(self, encode, stream, errors='strict'):
2111
codecs.StreamWriter.__init__(self, stream, errors)
2112
self.encode = encode
2114
def write(self, object):
2115
if type(object) is str:
2116
self.stream.write(object)
2118
data, _ = self.encode(object, self.errors)
2119
self.stream.write(data)
added
removed
bzrlib/osutils.py
