author | Yuya Nishihara <yuya@tcha.org> |
Sat, 17 Aug 2019 18:28:55 +0900 | |
changeset 42803 | 8ab1ae7f1cf4 |
parent 38783 | e7aa113b14f7 |
child 43076 | 2372284d9457 |
permissions | -rw-r--r-- |
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 |
||
27501
983e93d88193
pvec: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents:
24339
diff
changeset
|
51 |
from __future__ import absolute_import |
983e93d88193
pvec: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents:
24339
diff
changeset
|
52 |
|
983e93d88193
pvec: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents:
24339
diff
changeset
|
53 |
from .node import nullrev |
983e93d88193
pvec: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents:
24339
diff
changeset
|
54 |
from . import ( |
38783
e7aa113b14f7
global: use pycompat.xrange()
Gregory Szorc <gregory.szorc@gmail.com>
parents:
32201
diff
changeset
|
55 |
pycompat, |
27501
983e93d88193
pvec: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents:
24339
diff
changeset
|
56 |
util, |
983e93d88193
pvec: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents:
24339
diff
changeset
|
57 |
) |
16249 | 58 |
|
59 |
_size = 448 # 70 chars b85-encoded |
|
60 |
_bytes = _size / 8 |
|
61 |
_depthbits = 24 |
|
62 |
_depthbytes = _depthbits / 8 |
|
63 |
_vecbytes = _bytes - _depthbytes |
|
64 |
_vecbits = _vecbytes * 8 |
|
17424
e7cfe3587ea4
fix trivial spelling errors
Mads Kiilerich <mads@kiilerich.com>
parents:
16249
diff
changeset
|
65 |
_radius = (_vecbits - 30) / 2 # high probability vectors are related |
16249 | 66 |
|
67 |
def _bin(bs): |
|
68 |
'''convert a bytestring to a long''' |
|
69 |
v = 0 |
|
70 |
for b in bs: |
|
71 |
v = v * 256 + ord(b) |
|
72 |
return v |
|
73 |
||
74 |
def _str(v, l): |
|
75 |
bs = "" |
|
38783
e7aa113b14f7
global: use pycompat.xrange()
Gregory Szorc <gregory.szorc@gmail.com>
parents:
32201
diff
changeset
|
76 |
for p in pycompat.xrange(l): |
16249 | 77 |
bs = chr(v & 255) + bs |
78 |
v >>= 8 |
|
79 |
return bs |
|
80 |
||
81 |
def _split(b): |
|
82 |
'''depth and bitvec''' |
|
83 |
return _bin(b[:_depthbytes]), _bin(b[_depthbytes:]) |
|
84 |
||
85 |
def _join(depth, bitvec): |
|
86 |
return _str(depth, _depthbytes) + _str(bitvec, _vecbytes) |
|
87 |
||
88 |
def _hweight(x): |
|
89 |
c = 0 |
|
90 |
while x: |
|
91 |
if x & 1: |
|
92 |
c += 1 |
|
93 |
x >>= 1 |
|
94 |
return c |
|
38783
e7aa113b14f7
global: use pycompat.xrange()
Gregory Szorc <gregory.szorc@gmail.com>
parents:
32201
diff
changeset
|
95 |
_htab = [_hweight(x) for x in pycompat.xrange(256)] |
16249 | 96 |
|
97 |
def _hamming(a, b): |
|
98 |
'''find the hamming distance between two longs''' |
|
99 |
d = a ^ b |
|
100 |
c = 0 |
|
101 |
while d: |
|
102 |
c += _htab[d & 0xff] |
|
103 |
d >>= 8 |
|
104 |
return c |
|
105 |
||
106 |
def _mergevec(x, y, c): |
|
107 |
# Ideally, this function would be x ^ y ^ ancestor, but finding |
|
108 |
# ancestors is a nuisance. So instead we find the minimal number |
|
109 |
# of changes to balance the depth and hamming distance |
|
110 |
||
111 |
d1, v1 = x |
|
112 |
d2, v2 = y |
|
113 |
if d1 < d2: |
|
114 |
d1, d2, v1, v2 = d2, d1, v2, v1 |
|
115 |
||
116 |
hdist = _hamming(v1, v2) |
|
117 |
ddist = d1 - d2 |
|
118 |
v = v1 |
|
119 |
m = v1 ^ v2 # mask of different bits |
|
120 |
i = 1 |
|
121 |
||
122 |
if hdist > ddist: |
|
123 |
# if delta = 10 and hdist = 100, then we need to go up 55 steps |
|
124 |
# to the ancestor and down 45 |
|
125 |
changes = (hdist - ddist + 1) / 2 |
|
126 |
else: |
|
127 |
# must make at least one change |
|
128 |
changes = 1 |
|
129 |
depth = d1 + changes |
|
130 |
||
131 |
# copy changes from v2 |
|
132 |
if m: |
|
133 |
while changes: |
|
134 |
if m & i: |
|
135 |
v ^= i |
|
136 |
changes -= 1 |
|
137 |
i <<= 1 |
|
138 |
else: |
|
139 |
v = _flipbit(v, c) |
|
140 |
||
141 |
return depth, v |
|
142 |
||
143 |
def _flipbit(v, node): |
|
144 |
# converting bit strings to longs is slow |
|
145 |
bit = (hash(node) & 0xffffffff) % _vecbits |
|
146 |
return v ^ (1<<bit) |
|
147 |
||
148 |
def ctxpvec(ctx): |
|
149 |
'''construct a pvec for ctx while filling in the cache''' |
|
24339
bcc319d936a3
pvec: replace 'ctx._repo' with 'ctx.repo()'
Matt Harbison <matt_harbison@yahoo.com>
parents:
18918
diff
changeset
|
150 |
r = ctx.repo() |
16249 | 151 |
if not util.safehasattr(r, "_pveccache"): |
152 |
r._pveccache = {} |
|
153 |
pvc = r._pveccache |
|
154 |
if ctx.rev() not in pvc: |
|
155 |
cl = r.changelog |
|
38783
e7aa113b14f7
global: use pycompat.xrange()
Gregory Szorc <gregory.szorc@gmail.com>
parents:
32201
diff
changeset
|
156 |
for n in pycompat.xrange(ctx.rev() + 1): |
16249 | 157 |
if n not in pvc: |
158 |
node = cl.node(n) |
|
159 |
p1, p2 = cl.parentrevs(n) |
|
160 |
if p1 == nullrev: |
|
161 |
# start with a 'random' vector at root |
|
162 |
pvc[n] = (0, _bin((node * 3)[:_vecbytes])) |
|
163 |
elif p2 == nullrev: |
|
164 |
d, v = pvc[p1] |
|
165 |
pvc[n] = (d + 1, _flipbit(v, node)) |
|
166 |
else: |
|
167 |
pvc[n] = _mergevec(pvc[p1], pvc[p2], node) |
|
168 |
bs = _join(*pvc[ctx.rev()]) |
|
32201
4462a981e8df
base85: proxy through util module
Yuya Nishihara <yuya@tcha.org>
parents:
27501
diff
changeset
|
169 |
return pvec(util.b85encode(bs)) |
16249 | 170 |
|
171 |
class pvec(object): |
|
172 |
def __init__(self, hashorctx): |
|
173 |
if isinstance(hashorctx, str): |
|
174 |
self._bs = hashorctx |
|
32201
4462a981e8df
base85: proxy through util module
Yuya Nishihara <yuya@tcha.org>
parents:
27501
diff
changeset
|
175 |
self._depth, self._vec = _split(util.b85decode(hashorctx)) |
16249 | 176 |
else: |
18918
5093d2a87ff6
pvec: use the correct name for an identifier
Bryan O'Sullivan <bryano@fb.com>
parents:
17424
diff
changeset
|
177 |
self._vec = ctxpvec(hashorctx) |
16249 | 178 |
|
179 |
def __str__(self): |
|
180 |
return self._bs |
|
181 |
||
182 |
def __eq__(self, b): |
|
183 |
return self._vec == b._vec and self._depth == b._depth |
|
184 |
||
185 |
def __lt__(self, b): |
|
186 |
delta = b._depth - self._depth |
|
187 |
if delta < 0: |
|
188 |
return False # always correct |
|
189 |
if _hamming(self._vec, b._vec) > delta: |
|
190 |
return False |
|
191 |
return True |
|
192 |
||
193 |
def __gt__(self, b): |
|
194 |
return b < self |
|
195 |
||
196 |
def __or__(self, b): |
|
197 |
delta = abs(b._depth - self._depth) |
|
198 |
if _hamming(self._vec, b._vec) <= delta: |
|
199 |
return False |
|
200 |
return True |
|
201 |
||
202 |
def __sub__(self, b): |
|
203 |
if self | b: |
|
204 |
raise ValueError("concurrent pvecs") |
|
205 |
return self._depth - b._depth |
|
206 |
||
207 |
def distance(self, b): |
|
208 |
d = abs(b._depth - self._depth) |
|
209 |
h = _hamming(self._vec, b._vec) |
|
210 |
return max(d, h) |
|
211 |
||
212 |
def near(self, b): |
|
213 |
dist = abs(b.depth - self._depth) |
|
214 |
if dist > _radius or _hamming(self._vec, b._vec) > _radius: |
|
215 |
return False |