16249
|
1 |
# pvec.py - probabilistic vector clocks for Mercurial
|
|
2 |
#
|
|
3 |
# Copyright 2012 Matt Mackall <mpm@selenic.com>
|
|
4 |
#
|
|
5 |
# This software may be used and distributed according to the terms of the
|
|
6 |
# GNU General Public License version 2 or any later version.
|
|
7 |
|
|
8 |
'''
|
|
9 |
A "pvec" is a changeset property based on the theory of vector clocks
|
|
10 |
that can be compared to discover relatedness without consulting a
|
|
11 |
graph. This can be useful for tasks like determining how a
|
|
12 |
disconnected patch relates to a repository.
|
|
13 |
|
|
14 |
Currently a pvec consist of 448 bits, of which 24 are 'depth' and the
|
|
15 |
remainder are a bit vector. It is represented as a 70-character base85
|
|
16 |
string.
|
|
17 |
|
|
18 |
Construction:
|
|
19 |
|
|
20 |
- a root changeset has a depth of 0 and a bit vector based on its hash
|
|
21 |
- a normal commit has a changeset where depth is increased by one and
|
|
22 |
one bit vector bit is flipped based on its hash
|
|
23 |
- a merge changeset pvec is constructed by copying changes from one pvec into
|
|
24 |
the other to balance its depth
|
|
25 |
|
|
26 |
Properties:
|
|
27 |
|
|
28 |
- for linear changes, difference in depth is always <= hamming distance
|
|
29 |
- otherwise, changes are probably divergent
|
|
30 |
- when hamming distance is < 200, we can reliably detect when pvecs are near
|
|
31 |
|
|
32 |
Issues:
|
|
33 |
|
|
34 |
- hamming distance ceases to work over distances of ~ 200
|
|
35 |
- detecting divergence is less accurate when the common ancestor is very close
|
|
36 |
to either revision or total distance is high
|
|
37 |
- this could probably be improved by modeling the relation between
|
|
38 |
delta and hdist
|
|
39 |
|
|
40 |
Uses:
|
|
41 |
|
|
42 |
- a patch pvec can be used to locate the nearest available common ancestor for
|
|
43 |
resolving conflicts
|
|
44 |
- ordering of patches can be established without a DAG
|
|
45 |
- two head pvecs can be compared to determine whether push/pull/merge is needed
|
|
46 |
and approximately how many changesets are involved
|
|
47 |
- can be used to find a heuristic divergence measure between changesets on
|
|
48 |
different branches
|
|
49 |
'''
|
|
50 |
|
|
51 |
import base85, util
|
|
52 |
from node import nullrev
|
|
53 |
|
|
54 |
_size = 448 # 70 chars b85-encoded
|
|
55 |
_bytes = _size / 8
|
|
56 |
_depthbits = 24
|
|
57 |
_depthbytes = _depthbits / 8
|
|
58 |
_vecbytes = _bytes - _depthbytes
|
|
59 |
_vecbits = _vecbytes * 8
|
|
60 |
_radius = (_vecbits - 30) / 2 # high probability vecs are related
|
|
61 |
|
|
62 |
def _bin(bs):
|
|
63 |
'''convert a bytestring to a long'''
|
|
64 |
v = 0
|
|
65 |
for b in bs:
|
|
66 |
v = v * 256 + ord(b)
|
|
67 |
return v
|
|
68 |
|
|
69 |
def _str(v, l):
|
|
70 |
bs = ""
|
|
71 |
for p in xrange(l):
|
|
72 |
bs = chr(v & 255) + bs
|
|
73 |
v >>= 8
|
|
74 |
return bs
|
|
75 |
|
|
76 |
def _split(b):
|
|
77 |
'''depth and bitvec'''
|
|
78 |
return _bin(b[:_depthbytes]), _bin(b[_depthbytes:])
|
|
79 |
|
|
80 |
def _join(depth, bitvec):
|
|
81 |
return _str(depth, _depthbytes) + _str(bitvec, _vecbytes)
|
|
82 |
|
|
83 |
def _hweight(x):
|
|
84 |
c = 0
|
|
85 |
while x:
|
|
86 |
if x & 1:
|
|
87 |
c += 1
|
|
88 |
x >>= 1
|
|
89 |
return c
|
|
90 |
_htab = [_hweight(x) for x in xrange(256)]
|
|
91 |
|
|
92 |
def _hamming(a, b):
|
|
93 |
'''find the hamming distance between two longs'''
|
|
94 |
d = a ^ b
|
|
95 |
c = 0
|
|
96 |
while d:
|
|
97 |
c += _htab[d & 0xff]
|
|
98 |
d >>= 8
|
|
99 |
return c
|
|
100 |
|
|
101 |
def _mergevec(x, y, c):
|
|
102 |
# Ideally, this function would be x ^ y ^ ancestor, but finding
|
|
103 |
# ancestors is a nuisance. So instead we find the minimal number
|
|
104 |
# of changes to balance the depth and hamming distance
|
|
105 |
|
|
106 |
d1, v1 = x
|
|
107 |
d2, v2 = y
|
|
108 |
if d1 < d2:
|
|
109 |
d1, d2, v1, v2 = d2, d1, v2, v1
|
|
110 |
|
|
111 |
hdist = _hamming(v1, v2)
|
|
112 |
ddist = d1 - d2
|
|
113 |
v = v1
|
|
114 |
m = v1 ^ v2 # mask of different bits
|
|
115 |
i = 1
|
|
116 |
|
|
117 |
if hdist > ddist:
|
|
118 |
# if delta = 10 and hdist = 100, then we need to go up 55 steps
|
|
119 |
# to the ancestor and down 45
|
|
120 |
changes = (hdist - ddist + 1) / 2
|
|
121 |
else:
|
|
122 |
# must make at least one change
|
|
123 |
changes = 1
|
|
124 |
depth = d1 + changes
|
|
125 |
|
|
126 |
# copy changes from v2
|
|
127 |
if m:
|
|
128 |
while changes:
|
|
129 |
if m & i:
|
|
130 |
v ^= i
|
|
131 |
changes -= 1
|
|
132 |
i <<= 1
|
|
133 |
else:
|
|
134 |
v = _flipbit(v, c)
|
|
135 |
|
|
136 |
return depth, v
|
|
137 |
|
|
138 |
def _flipbit(v, node):
|
|
139 |
# converting bit strings to longs is slow
|
|
140 |
bit = (hash(node) & 0xffffffff) % _vecbits
|
|
141 |
return v ^ (1<<bit)
|
|
142 |
|
|
143 |
def ctxpvec(ctx):
|
|
144 |
'''construct a pvec for ctx while filling in the cache'''
|
|
145 |
r = ctx._repo
|
|
146 |
if not util.safehasattr(r, "_pveccache"):
|
|
147 |
r._pveccache = {}
|
|
148 |
pvc = r._pveccache
|
|
149 |
if ctx.rev() not in pvc:
|
|
150 |
cl = r.changelog
|
|
151 |
for n in xrange(ctx.rev() + 1):
|
|
152 |
if n not in pvc:
|
|
153 |
node = cl.node(n)
|
|
154 |
p1, p2 = cl.parentrevs(n)
|
|
155 |
if p1 == nullrev:
|
|
156 |
# start with a 'random' vector at root
|
|
157 |
pvc[n] = (0, _bin((node * 3)[:_vecbytes]))
|
|
158 |
elif p2 == nullrev:
|
|
159 |
d, v = pvc[p1]
|
|
160 |
pvc[n] = (d + 1, _flipbit(v, node))
|
|
161 |
else:
|
|
162 |
pvc[n] = _mergevec(pvc[p1], pvc[p2], node)
|
|
163 |
bs = _join(*pvc[ctx.rev()])
|
|
164 |
return pvec(base85.b85encode(bs))
|
|
165 |
|
|
166 |
class pvec(object):
|
|
167 |
def __init__(self, hashorctx):
|
|
168 |
if isinstance(hashorctx, str):
|
|
169 |
self._bs = hashorctx
|
|
170 |
self._depth, self._vec = _split(base85.b85decode(hashorctx))
|
|
171 |
else:
|
|
172 |
self._vec = ctxpvec(ctx)
|
|
173 |
|
|
174 |
def __str__(self):
|
|
175 |
return self._bs
|
|
176 |
|
|
177 |
def __eq__(self, b):
|
|
178 |
return self._vec == b._vec and self._depth == b._depth
|
|
179 |
|
|
180 |
def __lt__(self, b):
|
|
181 |
delta = b._depth - self._depth
|
|
182 |
if delta < 0:
|
|
183 |
return False # always correct
|
|
184 |
if _hamming(self._vec, b._vec) > delta:
|
|
185 |
return False
|
|
186 |
return True
|
|
187 |
|
|
188 |
def __gt__(self, b):
|
|
189 |
return b < self
|
|
190 |
|
|
191 |
def __or__(self, b):
|
|
192 |
delta = abs(b._depth - self._depth)
|
|
193 |
if _hamming(self._vec, b._vec) <= delta:
|
|
194 |
return False
|
|
195 |
return True
|
|
196 |
|
|
197 |
def __sub__(self, b):
|
|
198 |
if self | b:
|
|
199 |
raise ValueError("concurrent pvecs")
|
|
200 |
return self._depth - b._depth
|
|
201 |
|
|
202 |
def distance(self, b):
|
|
203 |
d = abs(b._depth - self._depth)
|
|
204 |
h = _hamming(self._vec, b._vec)
|
|
205 |
return max(d, h)
|
|
206 |
|
|
207 |
def near(self, b):
|
|
208 |
dist = abs(b.depth - self._depth)
|
|
209 |
if dist > _radius or _hamming(self._vec, b._vec) > _radius:
|
|
210 |
return False
|