1
"""A dumb and slow but simple dbm clone.
3
For database spam, spam.dir contains the index (a text file),
4
spam.bak *may* contain a backup of the index (also a text file),
5
while spam.dat contains the data (a binary file).
9
- seems to contain a bug when updating...
11
- reclaim free space (currently, space once occupied by deleted or expanded
12
items is never reused)
14
- support concurrent access (currently, if two processes take turns making
15
updates, they can mess up the index)
17
- support efficient access to large databases (currently, the whole index
18
is read when the database is opened, and some updates rewrite the whole index)
20
- support opening for read-only (flag = 'm')
28
__all__ = ["error", "open"]
34
class _Database(collections.MutableMapping):
36
# The on-disk directory and data files can remain in mutually
37
# inconsistent states for an arbitrarily long time (see comments
38
# at the end of __setitem__). This is only repaired when _commit()
39
# gets called. One place _commit() gets called is from __del__(),
40
# and if that occurs at program shutdown time, module globals may
41
# already have gotten rebound to None. Since it's crucial that
42
# _commit() finish successfully, we can't ignore shutdown races
43
# here, and _commit() must not reference any globals.
44
_os = _os # for _commit()
45
_io = _io # for _commit()
47
def __init__(self, filebasename, mode):
50
# The directory file is a text file. Each line looks like
51
# "%r, (%d, %d)\n" % (key, pos, siz)
52
# where key is the string key, pos is the offset into the dat
53
# file of the associated value's first byte, and siz is the number
54
# of bytes in the associated value.
55
self._dirfile = filebasename + '.dir'
57
# The data file is a binary file pointed into by the directory
58
# file, and holds the values associated with keys. Each value
59
# begins at a _BLOCKSIZE-aligned byte offset, and is a raw
60
# binary 8-bit string value.
61
self._datfile = filebasename + '.dat'
62
self._bakfile = filebasename + '.bak'
64
# The index is an in-memory dict, mirroring the directory file.
65
self._index = None # maps keys to (pos, siz) pairs
67
# Mod by Jack: create data file if needed
69
f = _io.open(self._datfile, 'r', encoding="Latin-1")
71
f = _io.open(self._datfile, 'w', encoding="Latin-1")
72
self._chmod(self._datfile)
76
# Read directory file into the in-memory index dict.
80
f = _io.open(self._dirfile, 'r', encoding="Latin-1")
86
key, pos_and_siz_pair = eval(line)
87
key = key.encode('Latin-1')
88
self._index[key] = pos_and_siz_pair
91
# Write the index dict to the directory file. The original directory
92
# file (if any) is renamed with a .bak extension first. If a .bak
93
# file currently exists, it's deleted.
95
# CAUTION: It's vital that _commit() succeed, and _commit() can
96
# be called from __del__(). Therefore we must never reference a
97
# global in this routine.
98
if self._index is None:
99
return # nothing to do
102
self._os.unlink(self._bakfile)
103
except self._os.error:
107
self._os.rename(self._dirfile, self._bakfile)
108
except self._os.error:
111
f = self._io.open(self._dirfile, 'w', encoding="Latin-1")
112
self._chmod(self._dirfile)
113
for key, pos_and_siz_pair in self._index.items():
114
# Use Latin-1 since it has no qualms with any value in any
115
# position; UTF-8, though, does care sometimes.
116
f.write("%r, %r\n" % (key.decode('Latin-1'), pos_and_siz_pair))
121
def __getitem__(self, key):
122
if isinstance(key, str):
123
key = key.encode('utf-8')
124
pos, siz = self._index[key] # may raise KeyError
125
f = _io.open(self._datfile, 'rb')
131
# Append val to the data file, starting at a _BLOCKSIZE-aligned
132
# offset. The data file is first padded with NUL bytes (if needed)
133
# to get to an aligned offset. Return pair
134
# (starting offset of val, len(val))
135
def _addval(self, val):
136
f = _io.open(self._datfile, 'rb+')
139
npos = ((pos + _BLOCKSIZE - 1) // _BLOCKSIZE) * _BLOCKSIZE
140
f.write(b'\0'*(npos-pos))
144
return (pos, len(val))
146
# Write val to the data file, starting at offset pos. The caller
147
# is responsible for ensuring that there's enough room starting at
148
# pos to hold val, without overwriting some other value. Return
149
# pair (pos, len(val)).
150
def _setval(self, pos, val):
151
f = _io.open(self._datfile, 'rb+')
155
return (pos, len(val))
157
# key is a new key whose associated value starts in the data file
158
# at offset pos and with length siz. Add an index record to
159
# the in-memory index dict, and append one to the directory file.
160
def _addkey(self, key, pos_and_siz_pair):
161
self._index[key] = pos_and_siz_pair
162
f = _io.open(self._dirfile, 'a', encoding="Latin-1")
163
self._chmod(self._dirfile)
164
f.write("%r, %r\n" % (key.decode("Latin-1"), pos_and_siz_pair))
167
def __setitem__(self, key, val):
168
if isinstance(key, str):
169
key = key.encode('utf-8')
170
elif not isinstance(key, (bytes, bytearray)):
171
raise TypeError("keys must be bytes or strings")
172
if isinstance(val, str):
173
val = val.encode('utf-8')
174
elif not isinstance(val, (bytes, bytearray)):
175
raise TypeError("values must be bytes or strings")
176
if key not in self._index:
177
self._addkey(key, self._addval(val))
179
# See whether the new value is small enough to fit in the
180
# (padded) space currently occupied by the old value.
181
pos, siz = self._index[key]
182
oldblocks = (siz + _BLOCKSIZE - 1) // _BLOCKSIZE
183
newblocks = (len(val) + _BLOCKSIZE - 1) // _BLOCKSIZE
184
if newblocks <= oldblocks:
185
self._index[key] = self._setval(pos, val)
187
# The new value doesn't fit in the (padded) space used
188
# by the old value. The blocks used by the old value are
190
self._index[key] = self._addval(val)
192
# Note that _index may be out of synch with the directory
193
# file now: _setval() and _addval() don't update the directory
194
# file. This also means that the on-disk directory and data
195
# files are in a mutually inconsistent state, and they'll
196
# remain that way until _commit() is called. Note that this
197
# is a disaster (for the database) if the program crashes
198
# (so that _commit() never gets called).
200
def __delitem__(self, key):
201
if isinstance(key, str):
202
key = key.encode('utf-8')
203
# The blocks used by the associated value are lost.
205
# XXX It's unclear why we do a _commit() here (the code always
206
# XXX has, so I'm not changing it). _setitem__ doesn't try to
207
# XXX keep the directory file in synch. Why should we? Or
208
# XXX why shouldn't __setitem__?
212
return list(self._index.keys())
215
return [(key, self[key]) for key in self._index.keys()]
217
def __contains__(self, key):
218
if isinstance(key, str):
219
key = key.encode('utf-8')
220
return key in self._index
223
return iter(self._index.keys())
227
return len(self._index)
231
self._index = self._datfile = self._dirfile = self._bakfile = None
235
def _chmod (self, file):
236
if hasattr(self._os, 'chmod'):
237
self._os.chmod(file, self._mode)
240
def open(file, flag=None, mode=0o666):
241
"""Open the database file, filename, and return corresponding object.
243
The flag argument, used to control how the database is opened in the
244
other DBM implementations, is ignored in the dbm.dumb module; the
245
database is always opened for update, and will be created if it does
248
The optional mode argument is the UNIX mode of the file, used only when
249
the database has to be created. It defaults to octal code 0o666 (and
250
will be modified by the prevailing umask).
253
# flag argument is currently ignored
255
# Modify mode depending on the umask
259
except AttributeError:
262
# Turn off any bits that are set in the umask
265
return _Database(file, mode)