/Lib/dumbdbm.py

http://unladen-swallow.googlecode.com/ · Python · 250 lines · 138 code · 30 blank · 82 comment · 21 complexity · ae1058e892af134aa1d3fe41887246ff MD5 · raw file

  1. """A dumb and slow but simple dbm clone.
  2. For database spam, spam.dir contains the index (a text file),
  3. spam.bak *may* contain a backup of the index (also a text file),
  4. while spam.dat contains the data (a binary file).
  5. XXX TO DO:
  6. - seems to contain a bug when updating...
  7. - reclaim free space (currently, space once occupied by deleted or expanded
  8. items is never reused)
  9. - support concurrent access (currently, if two processes take turns making
  10. updates, they can mess up the index)
  11. - support efficient access to large databases (currently, the whole index
  12. is read when the database is opened, and some updates rewrite the whole index)
  13. - support opening for read-only (flag = 'm')
  14. """
  15. import os as _os
  16. import __builtin__
  17. import UserDict
  18. _open = __builtin__.open
  19. _BLOCKSIZE = 512
  20. error = IOError # For anydbm
  21. class _Database(UserDict.DictMixin):
  22. # The on-disk directory and data files can remain in mutually
  23. # inconsistent states for an arbitrarily long time (see comments
  24. # at the end of __setitem__). This is only repaired when _commit()
  25. # gets called. One place _commit() gets called is from __del__(),
  26. # and if that occurs at program shutdown time, module globals may
  27. # already have gotten rebound to None. Since it's crucial that
  28. # _commit() finish successfully, we can't ignore shutdown races
  29. # here, and _commit() must not reference any globals.
  30. _os = _os # for _commit()
  31. _open = _open # for _commit()
  32. def __init__(self, filebasename, mode):
  33. self._mode = mode
  34. # The directory file is a text file. Each line looks like
  35. # "%r, (%d, %d)\n" % (key, pos, siz)
  36. # where key is the string key, pos is the offset into the dat
  37. # file of the associated value's first byte, and siz is the number
  38. # of bytes in the associated value.
  39. self._dirfile = filebasename + _os.extsep + 'dir'
  40. # The data file is a binary file pointed into by the directory
  41. # file, and holds the values associated with keys. Each value
  42. # begins at a _BLOCKSIZE-aligned byte offset, and is a raw
  43. # binary 8-bit string value.
  44. self._datfile = filebasename + _os.extsep + 'dat'
  45. self._bakfile = filebasename + _os.extsep + 'bak'
  46. # The index is an in-memory dict, mirroring the directory file.
  47. self._index = None # maps keys to (pos, siz) pairs
  48. # Mod by Jack: create data file if needed
  49. try:
  50. f = _open(self._datfile, 'r')
  51. except IOError:
  52. f = _open(self._datfile, 'w')
  53. self._chmod(self._datfile)
  54. f.close()
  55. self._update()
  56. # Read directory file into the in-memory index dict.
  57. def _update(self):
  58. self._index = {}
  59. try:
  60. f = _open(self._dirfile)
  61. except IOError:
  62. pass
  63. else:
  64. for line in f:
  65. line = line.rstrip()
  66. key, pos_and_siz_pair = eval(line)
  67. self._index[key] = pos_and_siz_pair
  68. f.close()
  69. # Write the index dict to the directory file. The original directory
  70. # file (if any) is renamed with a .bak extension first. If a .bak
  71. # file currently exists, it's deleted.
  72. def _commit(self):
  73. # CAUTION: It's vital that _commit() succeed, and _commit() can
  74. # be called from __del__(). Therefore we must never reference a
  75. # global in this routine.
  76. if self._index is None:
  77. return # nothing to do
  78. try:
  79. self._os.unlink(self._bakfile)
  80. except self._os.error:
  81. pass
  82. try:
  83. self._os.rename(self._dirfile, self._bakfile)
  84. except self._os.error:
  85. pass
  86. f = self._open(self._dirfile, 'w')
  87. self._chmod(self._dirfile)
  88. for key, pos_and_siz_pair in self._index.iteritems():
  89. f.write("%r, %r\n" % (key, pos_and_siz_pair))
  90. f.close()
  91. sync = _commit
  92. def __getitem__(self, key):
  93. pos, siz = self._index[key] # may raise KeyError
  94. f = _open(self._datfile, 'rb')
  95. f.seek(pos)
  96. dat = f.read(siz)
  97. f.close()
  98. return dat
  99. # Append val to the data file, starting at a _BLOCKSIZE-aligned
  100. # offset. The data file is first padded with NUL bytes (if needed)
  101. # to get to an aligned offset. Return pair
  102. # (starting offset of val, len(val))
  103. def _addval(self, val):
  104. f = _open(self._datfile, 'rb+')
  105. f.seek(0, 2)
  106. pos = int(f.tell())
  107. npos = ((pos + _BLOCKSIZE - 1) // _BLOCKSIZE) * _BLOCKSIZE
  108. f.write('\0'*(npos-pos))
  109. pos = npos
  110. f.write(val)
  111. f.close()
  112. return (pos, len(val))
  113. # Write val to the data file, starting at offset pos. The caller
  114. # is responsible for ensuring that there's enough room starting at
  115. # pos to hold val, without overwriting some other value. Return
  116. # pair (pos, len(val)).
  117. def _setval(self, pos, val):
  118. f = _open(self._datfile, 'rb+')
  119. f.seek(pos)
  120. f.write(val)
  121. f.close()
  122. return (pos, len(val))
  123. # key is a new key whose associated value starts in the data file
  124. # at offset pos and with length siz. Add an index record to
  125. # the in-memory index dict, and append one to the directory file.
  126. def _addkey(self, key, pos_and_siz_pair):
  127. self._index[key] = pos_and_siz_pair
  128. f = _open(self._dirfile, 'a')
  129. self._chmod(self._dirfile)
  130. f.write("%r, %r\n" % (key, pos_and_siz_pair))
  131. f.close()
  132. def __setitem__(self, key, val):
  133. if not type(key) == type('') == type(val):
  134. raise TypeError, "keys and values must be strings"
  135. if key not in self._index:
  136. self._addkey(key, self._addval(val))
  137. else:
  138. # See whether the new value is small enough to fit in the
  139. # (padded) space currently occupied by the old value.
  140. pos, siz = self._index[key]
  141. oldblocks = (siz + _BLOCKSIZE - 1) // _BLOCKSIZE
  142. newblocks = (len(val) + _BLOCKSIZE - 1) // _BLOCKSIZE
  143. if newblocks <= oldblocks:
  144. self._index[key] = self._setval(pos, val)
  145. else:
  146. # The new value doesn't fit in the (padded) space used
  147. # by the old value. The blocks used by the old value are
  148. # forever lost.
  149. self._index[key] = self._addval(val)
  150. # Note that _index may be out of synch with the directory
  151. # file now: _setval() and _addval() don't update the directory
  152. # file. This also means that the on-disk directory and data
  153. # files are in a mutually inconsistent state, and they'll
  154. # remain that way until _commit() is called. Note that this
  155. # is a disaster (for the database) if the program crashes
  156. # (so that _commit() never gets called).
  157. def __delitem__(self, key):
  158. # The blocks used by the associated value are lost.
  159. del self._index[key]
  160. # XXX It's unclear why we do a _commit() here (the code always
  161. # XXX has, so I'm not changing it). _setitem__ doesn't try to
  162. # XXX keep the directory file in synch. Why should we? Or
  163. # XXX why shouldn't __setitem__?
  164. self._commit()
  165. def keys(self):
  166. return self._index.keys()
  167. def has_key(self, key):
  168. return key in self._index
  169. def __contains__(self, key):
  170. return key in self._index
  171. def iterkeys(self):
  172. return self._index.iterkeys()
  173. __iter__ = iterkeys
  174. def __len__(self):
  175. return len(self._index)
  176. def close(self):
  177. self._commit()
  178. self._index = self._datfile = self._dirfile = self._bakfile = None
  179. __del__ = close
  180. def _chmod (self, file):
  181. if hasattr(self._os, 'chmod'):
  182. self._os.chmod(file, self._mode)
  183. def open(file, flag=None, mode=0666):
  184. """Open the database file, filename, and return corresponding object.
  185. The flag argument, used to control how the database is opened in the
  186. other DBM implementations, is ignored in the dumbdbm module; the
  187. database is always opened for update, and will be created if it does
  188. not exist.
  189. The optional mode argument is the UNIX mode of the file, used only when
  190. the database has to be created. It defaults to octal code 0666 (and
  191. will be modified by the prevailing umask).
  192. """
  193. # flag argument is currently ignored
  194. # Modify mode depending on the umask
  195. try:
  196. um = _os.umask(0)
  197. _os.umask(um)
  198. except AttributeError:
  199. pass
  200. else:
  201. # Turn off any bits that are set in the umask
  202. mode = mode & (~um)
  203. return _Database(file, mode)