Lib/dumbdbm.py

   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 os as _os
  25 import __builtin__
  26 import UserDict
  27
  28 _open = __builtin__.open
  29
  30 _BLOCKSIZE = 512
  31
  32 error = IOError                         # For anydbm
  33
  34 class _Database(UserDict.DictMixin):
  35
  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     _open = _open   # for _commit()
  46
  47     def __init__(self, filebasename, mode):
  48         self._mode = mode
  49
  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 + _os.extsep + 'dir'
  56
  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 + _os.extsep + 'dat'
  62         self._bakfile = filebasename + _os.extsep + 'bak'
  63
  64         # The index is an in-memory dict, mirroring the directory file.
  65         self._index = None  # maps keys to (pos, siz) pairs
  66
  67         # Mod by Jack: create data file if needed
  68         try:
  69             f = _open(self._datfile, 'r')
  70         except IOError:
  71             f = _open(self._datfile, 'w', self._mode)
  72         f.close()
  73         self._update()
  74
  75     # Read directory file into the in-memory index dict.
  76     def _update(self):
  77         self._index = {}
  78         try:
  79             f = _open(self._dirfile)
  80         except IOError:
  81             pass
  82         else:
  83             for line in f:
  84                 key, pos_and_siz_pair = eval(line)
  85                 self._index[key] = pos_and_siz_pair
  86             f.close()
  87
  88     # Write the index dict to the directory file.  The original directory
  89     # file (if any) is renamed with a .bak extension first.  If a .bak
  90     # file currently exists, it's deleted.
  91     def _commit(self):
  92         # CAUTION:  It's vital that _commit() succeed, and _commit() can
  93         # be called from __del__().  Therefore we must never reference a
  94         # global in this routine.
  95         if self._index is None:
  96             return  # nothing to do
  97
  98         try:
  99             self._os.unlink(self._bakfile)
 100         except self._os.error:
 101             pass
 102
 103         try:
 104             self._os.rename(self._dirfile, self._bakfile)
 105         except self._os.error:
 106             pass
 107
 108         f = self._open(self._dirfile, 'w', self._mode)
 109         for key, pos_and_siz_pair in self._index.iteritems():
 110             f.write("%r, %r\n" % (key, pos_and_siz_pair))
 111         f.close()
 112
 113     sync = _commit
 114
 115     def __getitem__(self, key):
 116         pos, siz = self._index[key]     # may raise KeyError
 117         f = _open(self._datfile, 'rb')
 118         f.seek(pos)
 119         dat = f.read(siz)
 120         f.close()
 121         return dat
 122
 123     # Append val to the data file, starting at a _BLOCKSIZE-aligned
 124     # offset.  The data file is first padded with NUL bytes (if needed)
 125     # to get to an aligned offset.  Return pair
 126     #     (starting offset of val, len(val))
 127     def _addval(self, val):
 128         f = _open(self._datfile, 'rb+')
 129         f.seek(0, 2)
 130         pos = int(f.tell())
 131         npos = ((pos + _BLOCKSIZE - 1) // _BLOCKSIZE) * _BLOCKSIZE
 132         f.write('\0'*(npos-pos))
 133         pos = npos
 134         f.write(val)
 135         f.close()
 136         return (pos, len(val))
 137
 138     # Write val to the data file, starting at offset pos.  The caller
 139     # is responsible for ensuring that there's enough room starting at
 140     # pos to hold val, without overwriting some other value.  Return
 141     # pair (pos, len(val)).
 142     def _setval(self, pos, val):
 143         f = _open(self._datfile, 'rb+')
 144         f.seek(pos)
 145         f.write(val)
 146         f.close()
 147         return (pos, len(val))
 148
 149     # key is a new key whose associated value starts in the data file
 150     # at offset pos and with length siz.  Add an index record to
 151     # the in-memory index dict, and append one to the directory file.
 152     def _addkey(self, key, pos_and_siz_pair):
 153         self._index[key] = pos_and_siz_pair
 154         f = _open(self._dirfile, 'a', self._mode)
 155         f.write("%r, %r\n" % (key, pos_and_siz_pair))
 156         f.close()
 157
 158     def __setitem__(self, key, val):
 159         if not type(key) == type('') == type(val):
 160             raise TypeError, "keys and values must be strings"
 161         if key not in self._index:
 162             self._addkey(key, self._addval(val))
 163         else:
 164             # See whether the new value is small enough to fit in the
 165             # (padded) space currently occupied by the old value.
 166             pos, siz = self._index[key]
 167             oldblocks = (siz + _BLOCKSIZE - 1) // _BLOCKSIZE
 168             newblocks = (len(val) + _BLOCKSIZE - 1) // _BLOCKSIZE
 169             if newblocks <= oldblocks:
 170                 self._index[key] = self._setval(pos, val)
 171             else:
 172                 # The new value doesn't fit in the (padded) space used
 173                 # by the old value.  The blocks used by the old value are
 174                 # forever lost.
 175                 self._index[key] = self._addval(val)
 176
 177             # Note that _index may be out of synch with the directory
 178             # file now:  _setval() and _addval() don't update the directory
 179             # file.  This also means that the on-disk directory and data
 180             # files are in a mutually inconsistent state, and they'll
 181             # remain that way until _commit() is called.  Note that this
 182             # is a disaster (for the database) if the program crashes
 183             # (so that _commit() never gets called).
 184
 185     def __delitem__(self, key):
 186         # The blocks used by the associated value are lost.
 187         del self._index[key]
 188         # XXX It's unclear why we do a _commit() here (the code always
 189         # XXX has, so I'm not changing it).  _setitem__ doesn't try to
 190         # XXX keep the directory file in synch.  Why should we?  Or
 191         # XXX why shouldn't __setitem__?
 192         self._commit()
 193
 194     def keys(self):
 195         return self._index.keys()
 196
 197     def has_key(self, key):
 198         return key in self._index
 199
 200     def __contains__(self, key):
 201         return key in self._index
 202
 203     def iterkeys(self):
 204         return self._index.iterkeys()
 205     __iter__ = iterkeys
 206
 207     def __len__(self):
 208         return len(self._index)
 209
 210     def close(self):
 211         self._commit()
 212         self._index = self._datfile = self._dirfile = self._bakfile = None
 213
 214     __del__ = close
 215
 216
 217
 218 def open(file, flag=None, mode=0666):
 219     """Open the database file, filename, and return corresponding object.
 220
 221     The flag argument, used to control how the database is opened in the
 222     other DBM implementations, is ignored in the dumbdbm module; the
 223     database is always opened for update, and will be created if it does
 224     not exist.
 225
 226     The optional mode argument is the UNIX mode of the file, used only when
 227     the database has to be created.  It defaults to octal code 0666 (and
 228     will be modified by the prevailing umask).
 229
 230     """
 231     # flag argument is currently ignored
 232     return _Database(file, mode)