revlog.py
651 lines
| 20.5 KiB
| text/x-python
|
PythonLexer
/ mercurial / revlog.py
|
r1083 | """ | ||
revlog.py - storage back-end for mercurial | ||||
This provides efficient delta storage with O(1) retrieve and append | ||||
and O(changes) merge between branches | ||||
Copyright 2005 Matt Mackall <mpm@selenic.com> | ||||
This software may be used and distributed according to the terms | ||||
of the GNU General Public License, incorporated herein by reference. | ||||
""" | ||||
|
r0 | |||
|
r208 | import zlib, struct, sha, binascii, heapq | ||
|
r1099 | import mdiff | ||
|
r1089 | from node import * | ||
|
r36 | |||
|
r1091 | def hash(text, p1, p2): | ||
"""generate a hash from the given text and its parent hashes | ||||
This hash combines both the current file contents and its history | ||||
in a manner that makes it easy to distinguish nodes with the same | ||||
content in the revision graph. | ||||
""" | ||||
l = [p1, p2] | ||||
l.sort() | ||||
s = sha.new(l[0]) | ||||
s.update(l[1]) | ||||
s.update(text) | ||||
return s.digest() | ||||
|
r0 | def compress(text): | ||
|
r1083 | """ generate a possibly-compressed representation of text """ | ||
|
r112 | if not text: return text | ||
if len(text) < 44: | ||||
if text[0] == '\0': return text | ||||
return 'u' + text | ||||
bin = zlib.compress(text) | ||||
if len(bin) > len(text): | ||||
if text[0] == '\0': return text | ||||
return 'u' + text | ||||
return bin | ||||
|
r0 | |||
def decompress(bin): | ||||
|
r1083 | """ decompress the given input """ | ||
|
r112 | if not bin: return bin | ||
t = bin[0] | ||||
if t == '\0': return bin | ||||
if t == 'x': return zlib.decompress(bin) | ||||
if t == 'u': return bin[1:] | ||||
|
r1073 | raise RevlogError("unknown compression type %s" % t) | ||
|
r0 | |||
indexformat = ">4l20s20s20s" | ||||
|
r76 | class lazyparser: | ||
|
r1083 | """ | ||
this class avoids the need to parse the entirety of large indices | ||||
By default we parse and load 1000 entries at a time. | ||||
If no position is specified, we load the whole index, and replace | ||||
the lazy objects in revlog with the underlying objects for | ||||
efficiency in cases where we look at most of the nodes. | ||||
""" | ||||
|
r323 | def __init__(self, data, revlog): | ||
|
r76 | self.data = data | ||
self.s = struct.calcsize(indexformat) | ||||
self.l = len(data)/self.s | ||||
self.index = [None] * self.l | ||||
self.map = {nullid: -1} | ||||
|
r323 | self.all = 0 | ||
self.revlog = revlog | ||||
|
r76 | |||
|
r323 | def load(self, pos=None): | ||
if self.all: return | ||||
if pos is not None: | ||||
block = pos / 1000 | ||||
i = block * 1000 | ||||
end = min(self.l, i + 1000) | ||||
else: | ||||
self.all = 1 | ||||
i = 0 | ||||
end = self.l | ||||
self.revlog.index = self.index | ||||
self.revlog.nodemap = self.map | ||||
|
r515 | |||
|
r76 | while i < end: | ||
d = self.data[i * self.s: (i + 1) * self.s] | ||||
e = struct.unpack(indexformat, d) | ||||
self.index[i] = e | ||||
self.map[e[6]] = i | ||||
i += 1 | ||||
|
r515 | |||
|
r76 | class lazyindex: | ||
|
r1083 | """a lazy version of the index array""" | ||
|
r76 | def __init__(self, parser): | ||
self.p = parser | ||||
def __len__(self): | ||||
return len(self.p.index) | ||||
|
r115 | def load(self, pos): | ||
self.p.load(pos) | ||||
return self.p.index[pos] | ||||
|
r76 | def __getitem__(self, pos): | ||
|
r115 | return self.p.index[pos] or self.load(pos) | ||
|
r76 | def append(self, e): | ||
self.p.index.append(e) | ||||
|
r515 | |||
|
r76 | class lazymap: | ||
|
r1083 | """a lazy version of the node map""" | ||
|
r76 | def __init__(self, parser): | ||
self.p = parser | ||||
def load(self, key): | ||||
|
r323 | if self.p.all: return | ||
|
r76 | n = self.p.data.find(key) | ||
|
r1214 | if n < 0: | ||
raise KeyError(key) | ||||
|
r76 | pos = n / self.p.s | ||
self.p.load(pos) | ||||
def __contains__(self, key): | ||||
|
r323 | self.p.load() | ||
return key in self.p.map | ||||
|
r97 | def __iter__(self): | ||
|
r469 | yield nullid | ||
|
r97 | for i in xrange(self.p.l): | ||
try: | ||||
yield self.p.index[i][6] | ||||
except: | ||||
self.p.load(i) | ||||
yield self.p.index[i][6] | ||||
|
r76 | def __getitem__(self, key): | ||
try: | ||||
return self.p.map[key] | ||||
except KeyError: | ||||
|
r86 | try: | ||
self.load(key) | ||||
return self.p.map[key] | ||||
except KeyError: | ||||
raise KeyError("node " + hex(key)) | ||||
|
r76 | def __setitem__(self, key, val): | ||
self.p.map[key] = val | ||||
|
r1073 | class RevlogError(Exception): pass | ||
|
r0 | class revlog: | ||
|
r1083 | """ | ||
the underlying revision storage object | ||||
A revlog consists of two parts, an index and the revision data. | ||||
The index is a file with a fixed record size containing | ||||
information on each revision, includings its nodeid (hash), the | ||||
nodeids of its parents, the position and offset of its data within | ||||
the data file, and the revision it's based on. Finally, each entry | ||||
contains a linkrev entry that can serve as a pointer to external | ||||
data. | ||||
The revision data itself is a linear collection of data chunks. | ||||
Each chunk represents a revision and is usually represented as a | ||||
delta against the previous chunk. To bound lookup time, runs of | ||||
deltas are limited to about 2 times the length of the original | ||||
version data. This makes retrieval of a version proportional to | ||||
its size, or O(1) relative to the number of revisions. | ||||
Both pieces of the revlog are written to in an append-only | ||||
fashion, which means we never need to rewrite a file to insert or | ||||
remove data, and can use some simple techniques to avoid the need | ||||
for locking while reading. | ||||
""" | ||||
|
r0 | def __init__(self, opener, indexfile, datafile): | ||
|
r1083 | """ | ||
create a revlog object | ||||
opener is a function that abstracts the file opening operation | ||||
and can be used to implement COW semantics or the like. | ||||
""" | ||||
|
r0 | self.indexfile = indexfile | ||
self.datafile = datafile | ||||
self.opener = opener | ||||
self.cache = None | ||||
|
r116 | |||
|
r0 | try: | ||
i = self.opener(self.indexfile).read() | ||||
|
r73 | except IOError: | ||
|
r76 | i = "" | ||
|
r116 | |||
if len(i) > 10000: | ||||
# big index, let's parse it on demand | ||||
|
r323 | parser = lazyparser(i, self) | ||
|
r116 | self.index = lazyindex(parser) | ||
self.nodemap = lazymap(parser) | ||||
else: | ||||
s = struct.calcsize(indexformat) | ||||
l = len(i) / s | ||||
self.index = [None] * l | ||||
m = [None] * l | ||||
n = 0 | ||||
for f in xrange(0, len(i), s): | ||||
# offset, size, base, linkrev, p1, p2, nodeid | ||||
e = struct.unpack(indexformat, i[f:f + s]) | ||||
m[n] = (e[6], n) | ||||
self.index[n] = e | ||||
n += 1 | ||||
self.nodemap = dict(m) | ||||
self.nodemap[nullid] = -1 | ||||
|
r0 | |||
def tip(self): return self.node(len(self.index) - 1) | ||||
def count(self): return len(self.index) | ||||
|
r26 | def node(self, rev): return (rev < 0) and nullid or self.index[rev][6] | ||
|
r1201 | def rev(self, node): | ||
try: | ||||
return self.nodemap[node] | ||||
except KeyError: | ||||
|
r1214 | raise RevlogError('%s: no node %s' % (self.indexfile, hex(node))) | ||
|
r1201 | def linkrev(self, node): return self.index[self.rev(node)][3] | ||
|
r2 | def parents(self, node): | ||
if node == nullid: return (nullid, nullid) | ||||
|
r1201 | return self.index[self.rev(node)][4:6] | ||
|
r0 | |||
def start(self, rev): return self.index[rev][0] | ||||
def length(self, rev): return self.index[rev][1] | ||||
def end(self, rev): return self.start(rev) + self.length(rev) | ||||
def base(self, rev): return self.index[rev][2] | ||||
|
r1074 | def reachable(self, rev, stop=None): | ||
reachable = {} | ||||
visit = [rev] | ||||
reachable[rev] = 1 | ||||
if stop: | ||||
stopn = self.rev(stop) | ||||
else: | ||||
stopn = 0 | ||||
while visit: | ||||
n = visit.pop(0) | ||||
if n == stop: | ||||
continue | ||||
if n == nullid: | ||||
continue | ||||
for p in self.parents(n): | ||||
if self.rev(p) < stopn: | ||||
continue | ||||
if p not in reachable: | ||||
reachable[p] = 1 | ||||
visit.append(p) | ||||
return reachable | ||||
|
r902 | def heads(self, stop=None): | ||
|
r1083 | """return the list of all nodes that have no children""" | ||
|
r221 | p = {} | ||
h = [] | ||||
|
r902 | stoprev = 0 | ||
if stop and stop in self.nodemap: | ||||
stoprev = self.rev(stop) | ||||
|
r1083 | |||
|
r243 | for r in range(self.count() - 1, -1, -1): | ||
|
r221 | n = self.node(r) | ||
if n not in p: | ||||
h.append(n) | ||||
|
r902 | if n == stop: | ||
break | ||||
if r < stoprev: | ||||
break | ||||
|
r221 | for pn in self.parents(n): | ||
p[pn] = 1 | ||||
return h | ||||
|
r370 | |||
def children(self, node): | ||||
|
r1083 | """find the children of a given node""" | ||
|
r370 | c = [] | ||
p = self.rev(node) | ||||
for r in range(p + 1, self.count()): | ||||
n = self.node(r) | ||||
for pn in self.parents(n): | ||||
|
r854 | if pn == node: | ||
c.append(n) | ||||
|
r370 | continue | ||
elif pn == nullid: | ||||
continue | ||||
return c | ||||
|
r515 | |||
|
r36 | def lookup(self, id): | ||
|
r1083 | """locate a node based on revision number or subset of hex nodeid""" | ||
|
r36 | try: | ||
rev = int(id) | ||||
|
r469 | if str(rev) != id: raise ValueError | ||
if rev < 0: rev = self.count() + rev | ||||
|
r476 | if rev < 0 or rev >= self.count(): raise ValueError | ||
|
r36 | return self.node(rev) | ||
|
r469 | except (ValueError, OverflowError): | ||
|
r36 | c = [] | ||
for n in self.nodemap: | ||||
|
r469 | if hex(n).startswith(id): | ||
|
r36 | c.append(n) | ||
|
r1232 | if len(c) > 1: raise KeyError("Ambiguous identifier") | ||
if len(c) < 1: raise KeyError("No match found") | ||||
|
r36 | return c[0] | ||
|
r515 | |||
|
r36 | return None | ||
|
r0 | def diff(self, a, b): | ||
|
r1083 | """return a delta between two revisions""" | ||
|
r0 | return mdiff.textdiff(a, b) | ||
|
r73 | def patches(self, t, pl): | ||
|
r1083 | """apply a list of patches to a string""" | ||
|
r73 | return mdiff.patches(t, pl) | ||
|
r119 | def delta(self, node): | ||
|
r1083 | """return or calculate a delta between a node and its predecessor""" | ||
|
r119 | r = self.rev(node) | ||
b = self.base(r) | ||||
if r == b: | ||||
return self.diff(self.revision(self.node(r - 1)), | ||||
self.revision(node)) | ||||
else: | ||||
f = self.opener(self.datafile) | ||||
f.seek(self.start(r)) | ||||
data = f.read(self.length(r)) | ||||
return decompress(data) | ||||
|
r0 | def revision(self, node): | ||
|
r1083 | """return an uncompressed revision of a given""" | ||
|
r36 | if node == nullid: return "" | ||
|
r0 | if self.cache and self.cache[0] == node: return self.cache[2] | ||
|
r1083 | # look up what we need to read | ||
|
r0 | text = None | ||
rev = self.rev(node) | ||||
|
r117 | start, length, base, link, p1, p2, node = self.index[rev] | ||
end = start + length | ||||
if base != rev: start = self.start(base) | ||||
|
r0 | |||
|
r1083 | # do we have useful data cached? | ||
|
r0 | if self.cache and self.cache[1] >= base and self.cache[1] < rev: | ||
base = self.cache[1] | ||||
start = self.start(base + 1) | ||||
text = self.cache[2] | ||||
last = 0 | ||||
f = self.opener(self.datafile) | ||||
f.seek(start) | ||||
data = f.read(end - start) | ||||
|
r651 | if text is None: | ||
|
r0 | last = self.length(base) | ||
text = decompress(data[:last]) | ||||
|
r71 | bins = [] | ||
|
r64 | for r in xrange(base + 1, rev + 1): | ||
|
r0 | s = self.length(r) | ||
|
r71 | bins.append(decompress(data[last:last + s])) | ||
|
r0 | last = last + s | ||
|
r71 | text = mdiff.patches(text, bins) | ||
|
r26 | if node != hash(text, p1, p2): | ||
|
r1214 | raise RevlogError("integrity check failed on %s:%d" | ||
|
r98 | % (self.datafile, rev)) | ||
|
r0 | |||
self.cache = (node, rev, text) | ||||
|
r515 | return text | ||
|
r0 | |||
|
r644 | def addrevision(self, text, transaction, link, p1=None, p2=None, d=None): | ||
|
r1083 | """add a revision to the log | ||
text - the revision data to add | ||||
transaction - the transaction object used for rollback | ||||
link - the linkrev data to add | ||||
p1, p2 - the parent nodeids of the revision | ||||
d - an optional precomputed delta | ||||
""" | ||||
|
r0 | if text is None: text = "" | ||
if p1 is None: p1 = self.tip() | ||||
if p2 is None: p2 = nullid | ||||
node = hash(text, p1, p2) | ||||
|
r301 | if node in self.nodemap: | ||
return node | ||||
|
r0 | n = self.count() | ||
t = n - 1 | ||||
if n: | ||||
|
r64 | base = self.base(t) | ||
start = self.start(base) | ||||
|
r0 | end = self.end(t) | ||
|
r644 | if not d: | ||
prev = self.revision(self.tip()) | ||||
d = self.diff(prev, text) | ||||
|
r98 | data = compress(d) | ||
|
r64 | dist = end - start + len(data) | ||
|
r0 | |||
# full versions are inserted when the needed deltas | ||||
# become comparable to the uncompressed text | ||||
|
r64 | if not n or dist > len(text) * 2: | ||
|
r0 | data = compress(text) | ||
base = n | ||||
else: | ||||
base = self.base(t) | ||||
offset = 0 | ||||
if t >= 0: | ||||
offset = self.end(t) | ||||
e = (offset, len(data), base, link, p1, p2, node) | ||||
|
r515 | |||
|
r0 | self.index.append(e) | ||
self.nodemap[node] = n | ||||
entry = struct.pack(indexformat, *e) | ||||
|
r26 | transaction.add(self.datafile, e[0]) | ||
|
r0 | self.opener(self.datafile, "a").write(data) | ||
|
r41 | transaction.add(self.indexfile, n * len(entry)) | ||
|
r0 | self.opener(self.indexfile, "a").write(entry) | ||
self.cache = (node, n, text) | ||||
return node | ||||
def ancestor(self, a, b): | ||||
|
r1083 | """calculate the least common ancestor of nodes a and b""" | ||
|
r147 | # calculate the distance of every node from root | ||
dist = {nullid: 0} | ||||
for i in xrange(self.count()): | ||||
n = self.node(i) | ||||
p1, p2 = self.parents(n) | ||||
dist[n] = max(dist[p1], dist[p2]) + 1 | ||||
|
r515 | |||
|
r147 | # traverse ancestors in order of decreasing distance from root | ||
def ancestors(node): | ||||
# we store negative distances because heap returns smallest member | ||||
h = [(-dist[node], node)] | ||||
seen = {} | ||||
earliest = self.count() | ||||
while h: | ||||
d, n = heapq.heappop(h) | ||||
if n not in seen: | ||||
seen[n] = 1 | ||||
|
r381 | r = self.rev(n) | ||
yield (-d, r, n) | ||||
|
r147 | for p in self.parents(n): | ||
heapq.heappush(h, (-dist[p], p)) | ||||
|
r45 | |||
|
r147 | x = ancestors(a) | ||
y = ancestors(b) | ||||
lx = x.next() | ||||
ly = y.next() | ||||
|
r45 | |||
|
r147 | # increment each ancestor list until it is closer to root than | ||
# the other, or they match | ||||
while 1: | ||||
if lx == ly: | ||||
|
r381 | return lx[2] | ||
|
r147 | elif lx < ly: | ||
ly = y.next() | ||||
elif lx > ly: | ||||
lx = x.next() | ||||
|
r0 | |||
|
r46 | def group(self, linkmap): | ||
|
r1083 | """calculate a delta group | ||
|
r46 | |||
|
r1083 | Given a list of changeset revs, return a set of deltas and | ||
metadata corresponding to nodes. the first delta is | ||||
parent(nodes[0]) -> nodes[0] the receiver is guaranteed to | ||||
have this parent as it has all history before these | ||||
changesets. parent is parent[0] | ||||
""" | ||||
|
r46 | revs = [] | ||
needed = {} | ||||
# find file nodes/revs that match changeset revs | ||||
for i in xrange(0, self.count()): | ||||
if self.index[i][3] in linkmap: | ||||
revs.append(i) | ||||
needed[i] = 1 | ||||
# if we don't have any revisions touched by these changesets, bail | ||||
|
r192 | if not revs: | ||
yield struct.pack(">l", 0) | ||||
return | ||||
|
r46 | |||
# add the parent of the first rev | ||||
p = self.parents(self.node(revs[0]))[0] | ||||
revs.insert(0, self.rev(p)) | ||||
# for each delta that isn't contiguous in the log, we need to | ||||
# reconstruct the base, reconstruct the result, and then | ||||
# calculate the delta. We also need to do this where we've | ||||
# stored a full version and not a delta | ||||
for i in xrange(0, len(revs) - 1): | ||||
a, b = revs[i], revs[i + 1] | ||||
if a + 1 != b or self.base(b) == b: | ||||
for j in xrange(self.base(a), a + 1): | ||||
needed[j] = 1 | ||||
for j in xrange(self.base(b), b + 1): | ||||
needed[j] = 1 | ||||
# calculate spans to retrieve from datafile | ||||
needed = needed.keys() | ||||
needed.sort() | ||||
spans = [] | ||||
|
r192 | oo = -1 | ||
ol = 0 | ||||
|
r46 | for n in needed: | ||
if n < 0: continue | ||||
o = self.start(n) | ||||
l = self.length(n) | ||||
|
r192 | if oo + ol == o: # can we merge with the previous? | ||
nl = spans[-1][2] | ||||
nl.append((n, l)) | ||||
ol += l | ||||
spans[-1] = (oo, ol, nl) | ||||
|
r46 | else: | ||
|
r192 | oo = o | ||
ol = l | ||||
spans.append((oo, ol, [(n, l)])) | ||||
|
r46 | |||
# read spans in, divide up chunks | ||||
chunks = {} | ||||
|
r192 | for span in spans: | ||
|
r46 | # we reopen the file for each span to make http happy for now | ||
f = self.opener(self.datafile) | ||||
f.seek(span[0]) | ||||
data = f.read(span[1]) | ||||
# divide up the span | ||||
pos = 0 | ||||
for r, l in span[2]: | ||||
|
r192 | chunks[r] = decompress(data[pos: pos + l]) | ||
|
r46 | pos += l | ||
# helper to reconstruct intermediate versions | ||||
def construct(text, base, rev): | ||||
|
r192 | bins = [chunks[r] for r in xrange(base + 1, rev + 1)] | ||
|
r71 | return mdiff.patches(text, bins) | ||
|
r46 | |||
# build deltas | ||||
deltas = [] | ||||
|
r71 | for d in xrange(0, len(revs) - 1): | ||
|
r46 | a, b = revs[d], revs[d + 1] | ||
n = self.node(b) | ||||
|
r192 | |||
# do we need to construct a new delta? | ||||
|
r46 | if a + 1 != b or self.base(b) == b: | ||
if a >= 0: | ||||
base = self.base(a) | ||||
|
r192 | ta = chunks[self.base(a)] | ||
|
r46 | ta = construct(ta, base, a) | ||
else: | ||||
ta = "" | ||||
|
r515 | |||
|
r46 | base = self.base(b) | ||
if a > base: | ||||
base = a | ||||
tb = ta | ||||
else: | ||||
|
r192 | tb = chunks[self.base(b)] | ||
|
r46 | tb = construct(tb, base, b) | ||
d = self.diff(ta, tb) | ||||
else: | ||||
|
r192 | d = chunks[b] | ||
|
r46 | |||
p = self.parents(n) | ||||
meta = n + p[0] + p[1] + linkmap[self.linkrev(n)] | ||||
l = struct.pack(">l", len(meta) + len(d) + 4) | ||||
|
r192 | yield l | ||
yield meta | ||||
yield d | ||||
|
r46 | |||
|
r192 | yield struct.pack(">l", 0) | ||
|
r1062 | def addgroup(self, revs, linkmapper, transaction, unique=0): | ||
|
r1083 | """ | ||
add a delta group | ||||
|
r46 | |||
|
r1083 | given a set of deltas, add them to the revision log. the | ||
first delta is against its parent, which should be in our | ||||
log, the rest are against the previous delta. | ||||
""" | ||||
#track the base of the current delta log | ||||
|
r46 | r = self.count() | ||
t = r - 1 | ||||
|
r192 | node = nullid | ||
|
r515 | |||
|
r655 | base = prev = -1 | ||
|
r653 | start = end = measure = 0 | ||
|
r46 | if r: | ||
start = self.start(self.base(t)) | ||||
end = self.end(t) | ||||
measure = self.length(self.base(t)) | ||||
base = self.base(t) | ||||
prev = self.tip() | ||||
transaction.add(self.datafile, end) | ||||
transaction.add(self.indexfile, r * struct.calcsize(indexformat)) | ||||
dfh = self.opener(self.datafile, "a") | ||||
ifh = self.opener(self.indexfile, "a") | ||||
# loop through our set of deltas | ||||
|
r192 | chain = None | ||
for chunk in revs: | ||||
node, p1, p2, cs = struct.unpack("20s20s20s20s", chunk[:80]) | ||||
|
r94 | link = linkmapper(cs) | ||
|
r77 | if node in self.nodemap: | ||
|
r224 | # this can happen if two branches make the same change | ||
|
r1218 | # if unique: | ||
# raise RevlogError("already have %s" % hex(node[:4])) | ||||
|
r653 | chain = node | ||
|
r224 | continue | ||
|
r192 | delta = chunk[80:] | ||
if not chain: | ||||
# retrieve the parent revision of the delta chain | ||||
chain = p1 | ||||
if not chain in self.nodemap: | ||||
|
r1073 | raise RevlogError("unknown base %s" % short(chain[:4])) | ||
|
r46 | |||
# full versions are inserted when the needed deltas become | ||||
# comparable to the uncompressed text or when the previous | ||||
# version is not the one we have a delta against. We use | ||||
# the size of the previous full rev as a proxy for the | ||||
# current size. | ||||
if chain == prev: | ||||
cdelta = compress(delta) | ||||
if chain != prev or (end - start + len(cdelta)) > measure * 2: | ||||
# flush our writes here so we can read it in revision | ||||
dfh.flush() | ||||
ifh.flush() | ||||
|
r65 | text = self.revision(chain) | ||
|
r73 | text = self.patches(text, [delta]) | ||
|
r46 | chk = self.addrevision(text, transaction, link, p1, p2) | ||
if chk != node: | ||||
|
r1073 | raise RevlogError("consistency error adding group") | ||
|
r46 | measure = len(text) | ||
else: | ||||
e = (end, len(cdelta), self.base(t), link, p1, p2, node) | ||||
self.index.append(e) | ||||
self.nodemap[node] = r | ||||
dfh.write(cdelta) | ||||
ifh.write(struct.pack(indexformat, *e)) | ||||
|
r65 | t, r, chain, prev = r, r + 1, node, node | ||
|
r46 | start = self.start(self.base(t)) | ||
end = self.end(t) | ||||
dfh.close() | ||||
ifh.close() | ||||
return node | ||||