Mercurial > repos > miller-lab > genome_diversity
view cdblib.py @ 8:30771a097f14
changed max line length from 5,000 to 50,000
author | Richard Burhans <burhans@bx.psu.edu> |
---|---|
date | Wed, 20 Jun 2012 14:02:30 -0400 |
parents | 2c498d40ecde |
children |
line wrap: on
line source
#!/usr/bin/env python ''' Manipulate DJB's Constant Databases. These are 2 level disk-based hash tables that efficiently handle many keys, while remaining space-efficient. http://cr.yp.to/cdb.html When generated databases are only used with Python code, consider using hash() rather than djb_hash() for a tidy speedup. ''' from _struct import Struct from itertools import chain def py_djb_hash(s): '''Return the value of DJB's hash function for the given 8-bit string.''' h = 5381 for c in s: h = (((h << 5) + h) ^ ord(c)) & 0xffffffff return h try: from _cdblib import djb_hash except ImportError: djb_hash = py_djb_hash read_2_le4 = Struct('<LL').unpack write_2_le4 = Struct('<LL').pack class Reader(object): '''A dictionary-like object for reading a Constant Database accessed through a string or string-like sequence, such as mmap.mmap().''' def __init__(self, data, hashfn=djb_hash): '''Create an instance reading from a sequence and using hashfn to hash keys.''' if len(data) < 2048: raise IOError('CDB too small') self.data = data self.hashfn = hashfn self.index = [read_2_le4(data[i:i+8]) for i in xrange(0, 2048, 8)] self.table_start = min(p[0] for p in self.index) # Assume load load factor is 0.5 like official CDB. self.length = sum(p[1] >> 1 for p in self.index) def iteritems(self): '''Like dict.iteritems(). Items are returned in insertion order.''' pos = 2048 while pos < self.table_start: klen, dlen = read_2_le4(self.data[pos:pos+8]) pos += 8 key = self.data[pos:pos+klen] pos += klen data = self.data[pos:pos+dlen] pos += dlen yield key, data def items(self): '''Like dict.items().''' return list(self.iteritems()) def iterkeys(self): '''Like dict.iterkeys().''' return (p[0] for p in self.iteritems()) __iter__ = iterkeys def itervalues(self): '''Like dict.itervalues().''' return (p[1] for p in self.iteritems()) def keys(self): '''Like dict.keys().''' return [p[0] for p in self.iteritems()] def values(self): '''Like dict.values().''' return [p[1] for p in self.iteritems()] def __getitem__(self, key): '''Like dict.__getitem__().''' value = self.get(key) if value is None: raise KeyError(key) return value def has_key(self, key): '''Return True if key exists in the database.''' return self.get(key) is not None __contains__ = has_key def __len__(self): '''Return the number of records in the database.''' return self.length def gets(self, key): '''Yield values for key in insertion order.''' # Truncate to 32 bits and remove sign. h = self.hashfn(key) & 0xffffffff start, nslots = self.index[h & 0xff] if nslots: end = start + (nslots << 3) slot_off = start + (((h >> 8) % nslots) << 3) for pos in chain(xrange(slot_off, end, 8), xrange(start, slot_off, 8)): rec_h, rec_pos = read_2_le4(self.data[pos:pos+8]) if not rec_h: break elif rec_h == h: klen, dlen = read_2_le4(self.data[rec_pos:rec_pos+8]) rec_pos += 8 if self.data[rec_pos:rec_pos+klen] == key: rec_pos += klen yield self.data[rec_pos:rec_pos+dlen] def get(self, key, default=None): '''Get the first value for key, returning default if missing.''' # Avoid exception catch when handling default case; much faster. return chain(self.gets(key), (default,)).next() def getint(self, key, default=None, base=0): '''Get the first value for key converted it to an int, returning default if missing.''' value = self.get(key, default) if value is not default: return int(value, base) return value def getints(self, key, base=0): '''Yield values for key in insertion order after converting to int.''' return (int(v, base) for v in self.gets(key)) def getstring(self, key, default=None, encoding='utf-8'): '''Get the first value for key decoded as unicode, returning default if not found.''' value = self.get(key, default) if value is not default: return value.decode(encoding) return value def getstrings(self, key, encoding='utf-8'): '''Yield values for key in insertion order after decoding as unicode.''' return (v.decode(encoding) for v in self.gets(key)) class Writer(object): '''Object for building new Constant Databases, and writing them to a seekable file-like object.''' def __init__(self, fp, hashfn=djb_hash): '''Create an instance writing to a file-like object, using hashfn to hash keys.''' self.fp = fp self.hashfn = hashfn fp.write('\x00' * 2048) self._unordered = [[] for i in xrange(256)] def put(self, key, value=''): '''Write a string key/value pair to the output file.''' assert type(key) is str and type(value) is str pos = self.fp.tell() self.fp.write(write_2_le4(len(key), len(value))) self.fp.write(key) self.fp.write(value) h = self.hashfn(key) & 0xffffffff self._unordered[h & 0xff].append((h, pos)) def puts(self, key, values): '''Write more than one value for the same key to the output file. Equivalent to calling put() in a loop.''' for value in values: self.put(key, value) def putint(self, key, value): '''Write an integer as a base-10 string associated with the given key to the output file.''' self.put(key, str(value)) def putints(self, key, values): '''Write zero or more integers for the same key to the output file. Equivalent to calling putint() in a loop.''' self.puts(key, (str(value) for value in values)) def putstring(self, key, value, encoding='utf-8'): '''Write a unicode string associated with the given key to the output file after encoding it as UTF-8 or the given encoding.''' self.put(key, unicode.encode(value, encoding)) def putstrings(self, key, values, encoding='utf-8'): '''Write zero or more unicode strings to the output file. Equivalent to calling putstring() in a loop.''' self.puts(key, (unicode.encode(value, encoding) for value in values)) def finalize(self): '''Write the final hash tables to the output file, and write out its index. The output file remains open upon return.''' index = [] for tbl in self._unordered: length = len(tbl) << 1 ordered = [(0, 0)] * length for pair in tbl: where = (pair[0] >> 8) % length for i in chain(xrange(where, length), xrange(0, where)): if not ordered[i][0]: ordered[i] = pair break index.append((self.fp.tell(), length)) for pair in ordered: self.fp.write(write_2_le4(*pair)) self.fp.seek(0) for pair in index: self.fp.write(write_2_le4(*pair)) self.fp = None # prevent double finalize()