Olivier Deprez | f4ef2d0 | 2021-04-20 13:36:24 +0200 | [diff] [blame^] | 1 | """A dumb and slow but simple dbm clone. |
| 2 | |
| 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). |
| 6 | |
| 7 | XXX TO DO: |
| 8 | |
| 9 | - seems to contain a bug when updating... |
| 10 | |
| 11 | - reclaim free space (currently, space once occupied by deleted or expanded |
| 12 | items is never reused) |
| 13 | |
| 14 | - support concurrent access (currently, if two processes take turns making |
| 15 | updates, they can mess up the index) |
| 16 | |
| 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) |
| 19 | |
| 20 | - support opening for read-only (flag = 'm') |
| 21 | |
| 22 | """ |
| 23 | |
| 24 | import ast as _ast |
| 25 | import io as _io |
| 26 | import os as _os |
| 27 | import collections.abc |
| 28 | |
| 29 | __all__ = ["error", "open"] |
| 30 | |
| 31 | _BLOCKSIZE = 512 |
| 32 | |
| 33 | error = OSError |
| 34 | |
| 35 | class _Database(collections.abc.MutableMapping): |
| 36 | |
| 37 | # The on-disk directory and data files can remain in mutually |
| 38 | # inconsistent states for an arbitrarily long time (see comments |
| 39 | # at the end of __setitem__). This is only repaired when _commit() |
| 40 | # gets called. One place _commit() gets called is from __del__(), |
| 41 | # and if that occurs at program shutdown time, module globals may |
| 42 | # already have gotten rebound to None. Since it's crucial that |
| 43 | # _commit() finish successfully, we can't ignore shutdown races |
| 44 | # here, and _commit() must not reference any globals. |
| 45 | _os = _os # for _commit() |
| 46 | _io = _io # for _commit() |
| 47 | |
| 48 | def __init__(self, filebasename, mode, flag='c'): |
| 49 | self._mode = mode |
| 50 | self._readonly = (flag == 'r') |
| 51 | |
| 52 | # The directory file is a text file. Each line looks like |
| 53 | # "%r, (%d, %d)\n" % (key, pos, siz) |
| 54 | # where key is the string key, pos is the offset into the dat |
| 55 | # file of the associated value's first byte, and siz is the number |
| 56 | # of bytes in the associated value. |
| 57 | self._dirfile = filebasename + '.dir' |
| 58 | |
| 59 | # The data file is a binary file pointed into by the directory |
| 60 | # file, and holds the values associated with keys. Each value |
| 61 | # begins at a _BLOCKSIZE-aligned byte offset, and is a raw |
| 62 | # binary 8-bit string value. |
| 63 | self._datfile = filebasename + '.dat' |
| 64 | self._bakfile = filebasename + '.bak' |
| 65 | |
| 66 | # The index is an in-memory dict, mirroring the directory file. |
| 67 | self._index = None # maps keys to (pos, siz) pairs |
| 68 | |
| 69 | # Handle the creation |
| 70 | self._create(flag) |
| 71 | self._update(flag) |
| 72 | |
| 73 | def _create(self, flag): |
| 74 | if flag == 'n': |
| 75 | for filename in (self._datfile, self._bakfile, self._dirfile): |
| 76 | try: |
| 77 | _os.remove(filename) |
| 78 | except OSError: |
| 79 | pass |
| 80 | # Mod by Jack: create data file if needed |
| 81 | try: |
| 82 | f = _io.open(self._datfile, 'r', encoding="Latin-1") |
| 83 | except OSError: |
| 84 | if flag not in ('c', 'n'): |
| 85 | raise |
| 86 | with _io.open(self._datfile, 'w', encoding="Latin-1") as f: |
| 87 | self._chmod(self._datfile) |
| 88 | else: |
| 89 | f.close() |
| 90 | |
| 91 | # Read directory file into the in-memory index dict. |
| 92 | def _update(self, flag): |
| 93 | self._modified = False |
| 94 | self._index = {} |
| 95 | try: |
| 96 | f = _io.open(self._dirfile, 'r', encoding="Latin-1") |
| 97 | except OSError: |
| 98 | if flag not in ('c', 'n'): |
| 99 | raise |
| 100 | self._modified = True |
| 101 | else: |
| 102 | with f: |
| 103 | for line in f: |
| 104 | line = line.rstrip() |
| 105 | key, pos_and_siz_pair = _ast.literal_eval(line) |
| 106 | key = key.encode('Latin-1') |
| 107 | self._index[key] = pos_and_siz_pair |
| 108 | |
| 109 | # Write the index dict to the directory file. The original directory |
| 110 | # file (if any) is renamed with a .bak extension first. If a .bak |
| 111 | # file currently exists, it's deleted. |
| 112 | def _commit(self): |
| 113 | # CAUTION: It's vital that _commit() succeed, and _commit() can |
| 114 | # be called from __del__(). Therefore we must never reference a |
| 115 | # global in this routine. |
| 116 | if self._index is None or not self._modified: |
| 117 | return # nothing to do |
| 118 | |
| 119 | try: |
| 120 | self._os.unlink(self._bakfile) |
| 121 | except OSError: |
| 122 | pass |
| 123 | |
| 124 | try: |
| 125 | self._os.rename(self._dirfile, self._bakfile) |
| 126 | except OSError: |
| 127 | pass |
| 128 | |
| 129 | with self._io.open(self._dirfile, 'w', encoding="Latin-1") as f: |
| 130 | self._chmod(self._dirfile) |
| 131 | for key, pos_and_siz_pair in self._index.items(): |
| 132 | # Use Latin-1 since it has no qualms with any value in any |
| 133 | # position; UTF-8, though, does care sometimes. |
| 134 | entry = "%r, %r\n" % (key.decode('Latin-1'), pos_and_siz_pair) |
| 135 | f.write(entry) |
| 136 | |
| 137 | sync = _commit |
| 138 | |
| 139 | def _verify_open(self): |
| 140 | if self._index is None: |
| 141 | raise error('DBM object has already been closed') |
| 142 | |
| 143 | def __getitem__(self, key): |
| 144 | if isinstance(key, str): |
| 145 | key = key.encode('utf-8') |
| 146 | self._verify_open() |
| 147 | pos, siz = self._index[key] # may raise KeyError |
| 148 | with _io.open(self._datfile, 'rb') as f: |
| 149 | f.seek(pos) |
| 150 | dat = f.read(siz) |
| 151 | return dat |
| 152 | |
| 153 | # Append val to the data file, starting at a _BLOCKSIZE-aligned |
| 154 | # offset. The data file is first padded with NUL bytes (if needed) |
| 155 | # to get to an aligned offset. Return pair |
| 156 | # (starting offset of val, len(val)) |
| 157 | def _addval(self, val): |
| 158 | with _io.open(self._datfile, 'rb+') as f: |
| 159 | f.seek(0, 2) |
| 160 | pos = int(f.tell()) |
| 161 | npos = ((pos + _BLOCKSIZE - 1) // _BLOCKSIZE) * _BLOCKSIZE |
| 162 | f.write(b'\0'*(npos-pos)) |
| 163 | pos = npos |
| 164 | f.write(val) |
| 165 | return (pos, len(val)) |
| 166 | |
| 167 | # Write val to the data file, starting at offset pos. The caller |
| 168 | # is responsible for ensuring that there's enough room starting at |
| 169 | # pos to hold val, without overwriting some other value. Return |
| 170 | # pair (pos, len(val)). |
| 171 | def _setval(self, pos, val): |
| 172 | with _io.open(self._datfile, 'rb+') as f: |
| 173 | f.seek(pos) |
| 174 | f.write(val) |
| 175 | return (pos, len(val)) |
| 176 | |
| 177 | # key is a new key whose associated value starts in the data file |
| 178 | # at offset pos and with length siz. Add an index record to |
| 179 | # the in-memory index dict, and append one to the directory file. |
| 180 | def _addkey(self, key, pos_and_siz_pair): |
| 181 | self._index[key] = pos_and_siz_pair |
| 182 | with _io.open(self._dirfile, 'a', encoding="Latin-1") as f: |
| 183 | self._chmod(self._dirfile) |
| 184 | f.write("%r, %r\n" % (key.decode("Latin-1"), pos_and_siz_pair)) |
| 185 | |
| 186 | def __setitem__(self, key, val): |
| 187 | if self._readonly: |
| 188 | raise error('The database is opened for reading only') |
| 189 | if isinstance(key, str): |
| 190 | key = key.encode('utf-8') |
| 191 | elif not isinstance(key, (bytes, bytearray)): |
| 192 | raise TypeError("keys must be bytes or strings") |
| 193 | if isinstance(val, str): |
| 194 | val = val.encode('utf-8') |
| 195 | elif not isinstance(val, (bytes, bytearray)): |
| 196 | raise TypeError("values must be bytes or strings") |
| 197 | self._verify_open() |
| 198 | self._modified = True |
| 199 | if key not in self._index: |
| 200 | self._addkey(key, self._addval(val)) |
| 201 | else: |
| 202 | # See whether the new value is small enough to fit in the |
| 203 | # (padded) space currently occupied by the old value. |
| 204 | pos, siz = self._index[key] |
| 205 | oldblocks = (siz + _BLOCKSIZE - 1) // _BLOCKSIZE |
| 206 | newblocks = (len(val) + _BLOCKSIZE - 1) // _BLOCKSIZE |
| 207 | if newblocks <= oldblocks: |
| 208 | self._index[key] = self._setval(pos, val) |
| 209 | else: |
| 210 | # The new value doesn't fit in the (padded) space used |
| 211 | # by the old value. The blocks used by the old value are |
| 212 | # forever lost. |
| 213 | self._index[key] = self._addval(val) |
| 214 | |
| 215 | # Note that _index may be out of synch with the directory |
| 216 | # file now: _setval() and _addval() don't update the directory |
| 217 | # file. This also means that the on-disk directory and data |
| 218 | # files are in a mutually inconsistent state, and they'll |
| 219 | # remain that way until _commit() is called. Note that this |
| 220 | # is a disaster (for the database) if the program crashes |
| 221 | # (so that _commit() never gets called). |
| 222 | |
| 223 | def __delitem__(self, key): |
| 224 | if self._readonly: |
| 225 | raise error('The database is opened for reading only') |
| 226 | if isinstance(key, str): |
| 227 | key = key.encode('utf-8') |
| 228 | self._verify_open() |
| 229 | self._modified = True |
| 230 | # The blocks used by the associated value are lost. |
| 231 | del self._index[key] |
| 232 | # XXX It's unclear why we do a _commit() here (the code always |
| 233 | # XXX has, so I'm not changing it). __setitem__ doesn't try to |
| 234 | # XXX keep the directory file in synch. Why should we? Or |
| 235 | # XXX why shouldn't __setitem__? |
| 236 | self._commit() |
| 237 | |
| 238 | def keys(self): |
| 239 | try: |
| 240 | return list(self._index) |
| 241 | except TypeError: |
| 242 | raise error('DBM object has already been closed') from None |
| 243 | |
| 244 | def items(self): |
| 245 | self._verify_open() |
| 246 | return [(key, self[key]) for key in self._index.keys()] |
| 247 | |
| 248 | def __contains__(self, key): |
| 249 | if isinstance(key, str): |
| 250 | key = key.encode('utf-8') |
| 251 | try: |
| 252 | return key in self._index |
| 253 | except TypeError: |
| 254 | if self._index is None: |
| 255 | raise error('DBM object has already been closed') from None |
| 256 | else: |
| 257 | raise |
| 258 | |
| 259 | def iterkeys(self): |
| 260 | try: |
| 261 | return iter(self._index) |
| 262 | except TypeError: |
| 263 | raise error('DBM object has already been closed') from None |
| 264 | __iter__ = iterkeys |
| 265 | |
| 266 | def __len__(self): |
| 267 | try: |
| 268 | return len(self._index) |
| 269 | except TypeError: |
| 270 | raise error('DBM object has already been closed') from None |
| 271 | |
| 272 | def close(self): |
| 273 | try: |
| 274 | self._commit() |
| 275 | finally: |
| 276 | self._index = self._datfile = self._dirfile = self._bakfile = None |
| 277 | |
| 278 | __del__ = close |
| 279 | |
| 280 | def _chmod(self, file): |
| 281 | self._os.chmod(file, self._mode) |
| 282 | |
| 283 | def __enter__(self): |
| 284 | return self |
| 285 | |
| 286 | def __exit__(self, *args): |
| 287 | self.close() |
| 288 | |
| 289 | |
| 290 | def open(file, flag='c', mode=0o666): |
| 291 | """Open the database file, filename, and return corresponding object. |
| 292 | |
| 293 | The flag argument, used to control how the database is opened in the |
| 294 | other DBM implementations, supports only the semantics of 'c' and 'n' |
| 295 | values. Other values will default to the semantics of 'c' value: |
| 296 | the database will always opened for update and will be created if it |
| 297 | does not exist. |
| 298 | |
| 299 | The optional mode argument is the UNIX mode of the file, used only when |
| 300 | the database has to be created. It defaults to octal code 0o666 (and |
| 301 | will be modified by the prevailing umask). |
| 302 | |
| 303 | """ |
| 304 | |
| 305 | # Modify mode depending on the umask |
| 306 | try: |
| 307 | um = _os.umask(0) |
| 308 | _os.umask(um) |
| 309 | except AttributeError: |
| 310 | pass |
| 311 | else: |
| 312 | # Turn off any bits that are set in the umask |
| 313 | mode = mode & (~um) |
| 314 | if flag not in ('r', 'w', 'c', 'n'): |
| 315 | raise ValueError("Flag must be one of 'r', 'w', 'c', or 'n'") |
| 316 | return _Database(file, mode, flag=flag) |