Show More
@@ -1,1396 +1,1398 b'' | |||||
1 | # revlog.py - storage back-end for mercurial |
|
1 | # revlog.py - storage back-end for mercurial | |
2 | # |
|
2 | # | |
3 | # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> |
|
3 | # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> | |
4 | # |
|
4 | # | |
5 | # This software may be used and distributed according to the terms of the |
|
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. |
|
6 | # GNU General Public License version 2 or any later version. | |
7 |
|
7 | |||
8 | """Storage back-end for Mercurial. |
|
8 | """Storage back-end for Mercurial. | |
9 |
|
9 | |||
10 | This provides efficient delta storage with O(1) retrieve and append |
|
10 | This provides efficient delta storage with O(1) retrieve and append | |
11 | and O(changes) merge between branches. |
|
11 | and O(changes) merge between branches. | |
12 | """ |
|
12 | """ | |
13 |
|
13 | |||
14 | # import stuff from node for others to import from revlog |
|
14 | # import stuff from node for others to import from revlog | |
15 | from node import bin, hex, nullid, nullrev, short #@UnusedImport |
|
15 | from node import bin, hex, nullid, nullrev, short #@UnusedImport | |
16 | from i18n import _ |
|
16 | from i18n import _ | |
17 | import changegroup, ancestor, mdiff, parsers, error, util |
|
17 | import changegroup, ancestor, mdiff, parsers, error, util | |
18 | import struct, zlib, errno |
|
18 | import struct, zlib, errno | |
19 |
|
19 | |||
20 | _pack = struct.pack |
|
20 | _pack = struct.pack | |
21 | _unpack = struct.unpack |
|
21 | _unpack = struct.unpack | |
22 | _compress = zlib.compress |
|
22 | _compress = zlib.compress | |
23 | _decompress = zlib.decompress |
|
23 | _decompress = zlib.decompress | |
24 | _sha = util.sha1 |
|
24 | _sha = util.sha1 | |
25 |
|
25 | |||
26 | # revlog flags |
|
26 | # revlog flags | |
27 | REVLOGV0 = 0 |
|
27 | REVLOGV0 = 0 | |
28 | REVLOGNG = 1 |
|
28 | REVLOGNG = 1 | |
29 | REVLOGNGINLINEDATA = (1 << 16) |
|
29 | REVLOGNGINLINEDATA = (1 << 16) | |
30 | REVLOG_DEFAULT_FLAGS = REVLOGNGINLINEDATA |
|
30 | REVLOG_DEFAULT_FLAGS = REVLOGNGINLINEDATA | |
31 | REVLOG_DEFAULT_FORMAT = REVLOGNG |
|
31 | REVLOG_DEFAULT_FORMAT = REVLOGNG | |
32 | REVLOG_DEFAULT_VERSION = REVLOG_DEFAULT_FORMAT | REVLOG_DEFAULT_FLAGS |
|
32 | REVLOG_DEFAULT_VERSION = REVLOG_DEFAULT_FORMAT | REVLOG_DEFAULT_FLAGS | |
33 |
|
33 | |||
34 | _prereadsize = 1048576 |
|
34 | _prereadsize = 1048576 | |
35 |
|
35 | |||
|
36 | _maxinline = 131072 # max size of revlog with inline data | |||
|
37 | ||||
36 | RevlogError = error.RevlogError |
|
38 | RevlogError = error.RevlogError | |
37 | LookupError = error.LookupError |
|
39 | LookupError = error.LookupError | |
38 |
|
40 | |||
39 | def getoffset(q): |
|
41 | def getoffset(q): | |
40 | return int(q >> 16) |
|
42 | return int(q >> 16) | |
41 |
|
43 | |||
42 | def gettype(q): |
|
44 | def gettype(q): | |
43 | return int(q & 0xFFFF) |
|
45 | return int(q & 0xFFFF) | |
44 |
|
46 | |||
45 | def offset_type(offset, type): |
|
47 | def offset_type(offset, type): | |
46 | return long(long(offset) << 16 | type) |
|
48 | return long(long(offset) << 16 | type) | |
47 |
|
49 | |||
48 | nullhash = _sha(nullid) |
|
50 | nullhash = _sha(nullid) | |
49 |
|
51 | |||
50 | def hash(text, p1, p2): |
|
52 | def hash(text, p1, p2): | |
51 | """generate a hash from the given text and its parent hashes |
|
53 | """generate a hash from the given text and its parent hashes | |
52 |
|
54 | |||
53 | This hash combines both the current file contents and its history |
|
55 | This hash combines both the current file contents and its history | |
54 | in a manner that makes it easy to distinguish nodes with the same |
|
56 | in a manner that makes it easy to distinguish nodes with the same | |
55 | content in the revision graph. |
|
57 | content in the revision graph. | |
56 | """ |
|
58 | """ | |
57 | # As of now, if one of the parent node is null, p2 is null |
|
59 | # As of now, if one of the parent node is null, p2 is null | |
58 | if p2 == nullid: |
|
60 | if p2 == nullid: | |
59 | # deep copy of a hash is faster than creating one |
|
61 | # deep copy of a hash is faster than creating one | |
60 | s = nullhash.copy() |
|
62 | s = nullhash.copy() | |
61 | s.update(p1) |
|
63 | s.update(p1) | |
62 | else: |
|
64 | else: | |
63 | # none of the parent nodes are nullid |
|
65 | # none of the parent nodes are nullid | |
64 | l = [p1, p2] |
|
66 | l = [p1, p2] | |
65 | l.sort() |
|
67 | l.sort() | |
66 | s = _sha(l[0]) |
|
68 | s = _sha(l[0]) | |
67 | s.update(l[1]) |
|
69 | s.update(l[1]) | |
68 | s.update(text) |
|
70 | s.update(text) | |
69 | return s.digest() |
|
71 | return s.digest() | |
70 |
|
72 | |||
71 | def compress(text): |
|
73 | def compress(text): | |
72 | """ generate a possibly-compressed representation of text """ |
|
74 | """ generate a possibly-compressed representation of text """ | |
73 | if not text: |
|
75 | if not text: | |
74 | return ("", text) |
|
76 | return ("", text) | |
75 | l = len(text) |
|
77 | l = len(text) | |
76 | bin = None |
|
78 | bin = None | |
77 | if l < 44: |
|
79 | if l < 44: | |
78 | pass |
|
80 | pass | |
79 | elif l > 1000000: |
|
81 | elif l > 1000000: | |
80 | # zlib makes an internal copy, thus doubling memory usage for |
|
82 | # zlib makes an internal copy, thus doubling memory usage for | |
81 | # large files, so lets do this in pieces |
|
83 | # large files, so lets do this in pieces | |
82 | z = zlib.compressobj() |
|
84 | z = zlib.compressobj() | |
83 | p = [] |
|
85 | p = [] | |
84 | pos = 0 |
|
86 | pos = 0 | |
85 | while pos < l: |
|
87 | while pos < l: | |
86 | pos2 = pos + 2**20 |
|
88 | pos2 = pos + 2**20 | |
87 | p.append(z.compress(text[pos:pos2])) |
|
89 | p.append(z.compress(text[pos:pos2])) | |
88 | pos = pos2 |
|
90 | pos = pos2 | |
89 | p.append(z.flush()) |
|
91 | p.append(z.flush()) | |
90 | if sum(map(len, p)) < l: |
|
92 | if sum(map(len, p)) < l: | |
91 | bin = "".join(p) |
|
93 | bin = "".join(p) | |
92 | else: |
|
94 | else: | |
93 | bin = _compress(text) |
|
95 | bin = _compress(text) | |
94 | if bin is None or len(bin) > l: |
|
96 | if bin is None or len(bin) > l: | |
95 | if text[0] == '\0': |
|
97 | if text[0] == '\0': | |
96 | return ("", text) |
|
98 | return ("", text) | |
97 | return ('u', text) |
|
99 | return ('u', text) | |
98 | return ("", bin) |
|
100 | return ("", bin) | |
99 |
|
101 | |||
100 | def decompress(bin): |
|
102 | def decompress(bin): | |
101 | """ decompress the given input """ |
|
103 | """ decompress the given input """ | |
102 | if not bin: |
|
104 | if not bin: | |
103 | return bin |
|
105 | return bin | |
104 | t = bin[0] |
|
106 | t = bin[0] | |
105 | if t == '\0': |
|
107 | if t == '\0': | |
106 | return bin |
|
108 | return bin | |
107 | if t == 'x': |
|
109 | if t == 'x': | |
108 | return _decompress(bin) |
|
110 | return _decompress(bin) | |
109 | if t == 'u': |
|
111 | if t == 'u': | |
110 | return bin[1:] |
|
112 | return bin[1:] | |
111 | raise RevlogError(_("unknown compression type %r") % t) |
|
113 | raise RevlogError(_("unknown compression type %r") % t) | |
112 |
|
114 | |||
113 | class lazyparser(object): |
|
115 | class lazyparser(object): | |
114 | """ |
|
116 | """ | |
115 | this class avoids the need to parse the entirety of large indices |
|
117 | this class avoids the need to parse the entirety of large indices | |
116 | """ |
|
118 | """ | |
117 |
|
119 | |||
118 | # lazyparser is not safe to use on windows if win32 extensions not |
|
120 | # lazyparser is not safe to use on windows if win32 extensions not | |
119 | # available. it keeps file handle open, which make it not possible |
|
121 | # available. it keeps file handle open, which make it not possible | |
120 | # to break hardlinks on local cloned repos. |
|
122 | # to break hardlinks on local cloned repos. | |
121 |
|
123 | |||
122 | def __init__(self, dataf): |
|
124 | def __init__(self, dataf): | |
123 | try: |
|
125 | try: | |
124 | size = util.fstat(dataf).st_size |
|
126 | size = util.fstat(dataf).st_size | |
125 | except AttributeError: |
|
127 | except AttributeError: | |
126 | size = 0 |
|
128 | size = 0 | |
127 | self.dataf = dataf |
|
129 | self.dataf = dataf | |
128 | self.s = struct.calcsize(indexformatng) |
|
130 | self.s = struct.calcsize(indexformatng) | |
129 | self.datasize = size |
|
131 | self.datasize = size | |
130 | self.l = size / self.s |
|
132 | self.l = size / self.s | |
131 | self.index = [None] * self.l |
|
133 | self.index = [None] * self.l | |
132 | self.map = {nullid: nullrev} |
|
134 | self.map = {nullid: nullrev} | |
133 | self.allmap = 0 |
|
135 | self.allmap = 0 | |
134 | self.all = 0 |
|
136 | self.all = 0 | |
135 | self.mapfind_count = 0 |
|
137 | self.mapfind_count = 0 | |
136 |
|
138 | |||
137 | def loadmap(self): |
|
139 | def loadmap(self): | |
138 | """ |
|
140 | """ | |
139 | during a commit, we need to make sure the rev being added is |
|
141 | during a commit, we need to make sure the rev being added is | |
140 | not a duplicate. This requires loading the entire index, |
|
142 | not a duplicate. This requires loading the entire index, | |
141 | which is fairly slow. loadmap can load up just the node map, |
|
143 | which is fairly slow. loadmap can load up just the node map, | |
142 | which takes much less time. |
|
144 | which takes much less time. | |
143 | """ |
|
145 | """ | |
144 | if self.allmap: |
|
146 | if self.allmap: | |
145 | return |
|
147 | return | |
146 | end = self.datasize |
|
148 | end = self.datasize | |
147 | self.allmap = 1 |
|
149 | self.allmap = 1 | |
148 | cur = 0 |
|
150 | cur = 0 | |
149 | count = 0 |
|
151 | count = 0 | |
150 | blocksize = self.s * 256 |
|
152 | blocksize = self.s * 256 | |
151 | self.dataf.seek(0) |
|
153 | self.dataf.seek(0) | |
152 | while cur < end: |
|
154 | while cur < end: | |
153 | data = self.dataf.read(blocksize) |
|
155 | data = self.dataf.read(blocksize) | |
154 | off = 0 |
|
156 | off = 0 | |
155 | for x in xrange(256): |
|
157 | for x in xrange(256): | |
156 | n = data[off + ngshaoffset:off + ngshaoffset + 20] |
|
158 | n = data[off + ngshaoffset:off + ngshaoffset + 20] | |
157 | self.map[n] = count |
|
159 | self.map[n] = count | |
158 | count += 1 |
|
160 | count += 1 | |
159 | if count >= self.l: |
|
161 | if count >= self.l: | |
160 | break |
|
162 | break | |
161 | off += self.s |
|
163 | off += self.s | |
162 | cur += blocksize |
|
164 | cur += blocksize | |
163 |
|
165 | |||
164 | def loadblock(self, blockstart, blocksize, data=None): |
|
166 | def loadblock(self, blockstart, blocksize, data=None): | |
165 | if self.all: |
|
167 | if self.all: | |
166 | return |
|
168 | return | |
167 | if data is None: |
|
169 | if data is None: | |
168 | self.dataf.seek(blockstart) |
|
170 | self.dataf.seek(blockstart) | |
169 | if blockstart + blocksize > self.datasize: |
|
171 | if blockstart + blocksize > self.datasize: | |
170 | # the revlog may have grown since we've started running, |
|
172 | # the revlog may have grown since we've started running, | |
171 | # but we don't have space in self.index for more entries. |
|
173 | # but we don't have space in self.index for more entries. | |
172 | # limit blocksize so that we don't get too much data. |
|
174 | # limit blocksize so that we don't get too much data. | |
173 | blocksize = max(self.datasize - blockstart, 0) |
|
175 | blocksize = max(self.datasize - blockstart, 0) | |
174 | data = self.dataf.read(blocksize) |
|
176 | data = self.dataf.read(blocksize) | |
175 | lend = len(data) / self.s |
|
177 | lend = len(data) / self.s | |
176 | i = blockstart / self.s |
|
178 | i = blockstart / self.s | |
177 | off = 0 |
|
179 | off = 0 | |
178 | # lazyindex supports __delitem__ |
|
180 | # lazyindex supports __delitem__ | |
179 | if lend > len(self.index) - i: |
|
181 | if lend > len(self.index) - i: | |
180 | lend = len(self.index) - i |
|
182 | lend = len(self.index) - i | |
181 | for x in xrange(lend): |
|
183 | for x in xrange(lend): | |
182 | if self.index[i + x] is None: |
|
184 | if self.index[i + x] is None: | |
183 | b = data[off : off + self.s] |
|
185 | b = data[off : off + self.s] | |
184 | self.index[i + x] = b |
|
186 | self.index[i + x] = b | |
185 | n = b[ngshaoffset:ngshaoffset + 20] |
|
187 | n = b[ngshaoffset:ngshaoffset + 20] | |
186 | self.map[n] = i + x |
|
188 | self.map[n] = i + x | |
187 | off += self.s |
|
189 | off += self.s | |
188 |
|
190 | |||
189 | def findnode(self, node): |
|
191 | def findnode(self, node): | |
190 | """search backwards through the index file for a specific node""" |
|
192 | """search backwards through the index file for a specific node""" | |
191 | if self.allmap: |
|
193 | if self.allmap: | |
192 | return None |
|
194 | return None | |
193 |
|
195 | |||
194 | # hg log will cause many many searches for the manifest |
|
196 | # hg log will cause many many searches for the manifest | |
195 | # nodes. After we get called a few times, just load the whole |
|
197 | # nodes. After we get called a few times, just load the whole | |
196 | # thing. |
|
198 | # thing. | |
197 | if self.mapfind_count > 8: |
|
199 | if self.mapfind_count > 8: | |
198 | self.loadmap() |
|
200 | self.loadmap() | |
199 | if node in self.map: |
|
201 | if node in self.map: | |
200 | return node |
|
202 | return node | |
201 | return None |
|
203 | return None | |
202 | self.mapfind_count += 1 |
|
204 | self.mapfind_count += 1 | |
203 | last = self.l - 1 |
|
205 | last = self.l - 1 | |
204 | while self.index[last] != None: |
|
206 | while self.index[last] != None: | |
205 | if last == 0: |
|
207 | if last == 0: | |
206 | self.all = 1 |
|
208 | self.all = 1 | |
207 | self.allmap = 1 |
|
209 | self.allmap = 1 | |
208 | return None |
|
210 | return None | |
209 | last -= 1 |
|
211 | last -= 1 | |
210 | end = (last + 1) * self.s |
|
212 | end = (last + 1) * self.s | |
211 | blocksize = self.s * 256 |
|
213 | blocksize = self.s * 256 | |
212 | while end >= 0: |
|
214 | while end >= 0: | |
213 | start = max(end - blocksize, 0) |
|
215 | start = max(end - blocksize, 0) | |
214 | self.dataf.seek(start) |
|
216 | self.dataf.seek(start) | |
215 | data = self.dataf.read(end - start) |
|
217 | data = self.dataf.read(end - start) | |
216 | findend = end - start |
|
218 | findend = end - start | |
217 | while True: |
|
219 | while True: | |
218 | # we're searching backwards, so we have to make sure |
|
220 | # we're searching backwards, so we have to make sure | |
219 | # we don't find a changeset where this node is a parent |
|
221 | # we don't find a changeset where this node is a parent | |
220 | off = data.find(node, 0, findend) |
|
222 | off = data.find(node, 0, findend) | |
221 | findend = off |
|
223 | findend = off | |
222 | if off >= 0: |
|
224 | if off >= 0: | |
223 | i = off / self.s |
|
225 | i = off / self.s | |
224 | off = i * self.s |
|
226 | off = i * self.s | |
225 | n = data[off + ngshaoffset:off + ngshaoffset + 20] |
|
227 | n = data[off + ngshaoffset:off + ngshaoffset + 20] | |
226 | if n == node: |
|
228 | if n == node: | |
227 | self.map[n] = i + start / self.s |
|
229 | self.map[n] = i + start / self.s | |
228 | return node |
|
230 | return node | |
229 | else: |
|
231 | else: | |
230 | break |
|
232 | break | |
231 | end -= blocksize |
|
233 | end -= blocksize | |
232 | return None |
|
234 | return None | |
233 |
|
235 | |||
234 | def loadindex(self, i=None, end=None): |
|
236 | def loadindex(self, i=None, end=None): | |
235 | if self.all: |
|
237 | if self.all: | |
236 | return |
|
238 | return | |
237 | all = False |
|
239 | all = False | |
238 | if i is None: |
|
240 | if i is None: | |
239 | blockstart = 0 |
|
241 | blockstart = 0 | |
240 | blocksize = (65536 / self.s) * self.s |
|
242 | blocksize = (65536 / self.s) * self.s | |
241 | end = self.datasize |
|
243 | end = self.datasize | |
242 | all = True |
|
244 | all = True | |
243 | else: |
|
245 | else: | |
244 | if end: |
|
246 | if end: | |
245 | blockstart = i * self.s |
|
247 | blockstart = i * self.s | |
246 | end = end * self.s |
|
248 | end = end * self.s | |
247 | blocksize = end - blockstart |
|
249 | blocksize = end - blockstart | |
248 | else: |
|
250 | else: | |
249 | blockstart = (i & ~1023) * self.s |
|
251 | blockstart = (i & ~1023) * self.s | |
250 | blocksize = self.s * 1024 |
|
252 | blocksize = self.s * 1024 | |
251 | end = blockstart + blocksize |
|
253 | end = blockstart + blocksize | |
252 | while blockstart < end: |
|
254 | while blockstart < end: | |
253 | self.loadblock(blockstart, blocksize) |
|
255 | self.loadblock(blockstart, blocksize) | |
254 | blockstart += blocksize |
|
256 | blockstart += blocksize | |
255 | if all: |
|
257 | if all: | |
256 | self.all = True |
|
258 | self.all = True | |
257 |
|
259 | |||
258 | class lazyindex(object): |
|
260 | class lazyindex(object): | |
259 | """a lazy version of the index array""" |
|
261 | """a lazy version of the index array""" | |
260 | def __init__(self, parser): |
|
262 | def __init__(self, parser): | |
261 | self.p = parser |
|
263 | self.p = parser | |
262 | def __len__(self): |
|
264 | def __len__(self): | |
263 | return len(self.p.index) |
|
265 | return len(self.p.index) | |
264 | def load(self, pos): |
|
266 | def load(self, pos): | |
265 | if pos < 0: |
|
267 | if pos < 0: | |
266 | pos += len(self.p.index) |
|
268 | pos += len(self.p.index) | |
267 | self.p.loadindex(pos) |
|
269 | self.p.loadindex(pos) | |
268 | return self.p.index[pos] |
|
270 | return self.p.index[pos] | |
269 | def __getitem__(self, pos): |
|
271 | def __getitem__(self, pos): | |
270 | return _unpack(indexformatng, self.p.index[pos] or self.load(pos)) |
|
272 | return _unpack(indexformatng, self.p.index[pos] or self.load(pos)) | |
271 | def __setitem__(self, pos, item): |
|
273 | def __setitem__(self, pos, item): | |
272 | self.p.index[pos] = _pack(indexformatng, *item) |
|
274 | self.p.index[pos] = _pack(indexformatng, *item) | |
273 | def __delitem__(self, pos): |
|
275 | def __delitem__(self, pos): | |
274 | del self.p.index[pos] |
|
276 | del self.p.index[pos] | |
275 | def insert(self, pos, e): |
|
277 | def insert(self, pos, e): | |
276 | self.p.index.insert(pos, _pack(indexformatng, *e)) |
|
278 | self.p.index.insert(pos, _pack(indexformatng, *e)) | |
277 | def append(self, e): |
|
279 | def append(self, e): | |
278 | self.p.index.append(_pack(indexformatng, *e)) |
|
280 | self.p.index.append(_pack(indexformatng, *e)) | |
279 |
|
281 | |||
280 | class lazymap(object): |
|
282 | class lazymap(object): | |
281 | """a lazy version of the node map""" |
|
283 | """a lazy version of the node map""" | |
282 | def __init__(self, parser): |
|
284 | def __init__(self, parser): | |
283 | self.p = parser |
|
285 | self.p = parser | |
284 | def load(self, key): |
|
286 | def load(self, key): | |
285 | n = self.p.findnode(key) |
|
287 | n = self.p.findnode(key) | |
286 | if n is None: |
|
288 | if n is None: | |
287 | raise KeyError(key) |
|
289 | raise KeyError(key) | |
288 | def __contains__(self, key): |
|
290 | def __contains__(self, key): | |
289 | if key in self.p.map: |
|
291 | if key in self.p.map: | |
290 | return True |
|
292 | return True | |
291 | self.p.loadmap() |
|
293 | self.p.loadmap() | |
292 | return key in self.p.map |
|
294 | return key in self.p.map | |
293 | def __iter__(self): |
|
295 | def __iter__(self): | |
294 | yield nullid |
|
296 | yield nullid | |
295 | for i in xrange(self.p.l): |
|
297 | for i in xrange(self.p.l): | |
296 | ret = self.p.index[i] |
|
298 | ret = self.p.index[i] | |
297 | if not ret: |
|
299 | if not ret: | |
298 | self.p.loadindex(i) |
|
300 | self.p.loadindex(i) | |
299 | ret = self.p.index[i] |
|
301 | ret = self.p.index[i] | |
300 | if isinstance(ret, str): |
|
302 | if isinstance(ret, str): | |
301 | ret = _unpack(indexformatng, ret) |
|
303 | ret = _unpack(indexformatng, ret) | |
302 | yield ret[7] |
|
304 | yield ret[7] | |
303 | def __getitem__(self, key): |
|
305 | def __getitem__(self, key): | |
304 | try: |
|
306 | try: | |
305 | return self.p.map[key] |
|
307 | return self.p.map[key] | |
306 | except KeyError: |
|
308 | except KeyError: | |
307 | try: |
|
309 | try: | |
308 | self.load(key) |
|
310 | self.load(key) | |
309 | return self.p.map[key] |
|
311 | return self.p.map[key] | |
310 | except KeyError: |
|
312 | except KeyError: | |
311 | raise KeyError("node " + hex(key)) |
|
313 | raise KeyError("node " + hex(key)) | |
312 | def __setitem__(self, key, val): |
|
314 | def __setitem__(self, key, val): | |
313 | self.p.map[key] = val |
|
315 | self.p.map[key] = val | |
314 | def __delitem__(self, key): |
|
316 | def __delitem__(self, key): | |
315 | del self.p.map[key] |
|
317 | del self.p.map[key] | |
316 |
|
318 | |||
317 | indexformatv0 = ">4l20s20s20s" |
|
319 | indexformatv0 = ">4l20s20s20s" | |
318 | v0shaoffset = 56 |
|
320 | v0shaoffset = 56 | |
319 |
|
321 | |||
320 | class revlogoldio(object): |
|
322 | class revlogoldio(object): | |
321 | def __init__(self): |
|
323 | def __init__(self): | |
322 | self.size = struct.calcsize(indexformatv0) |
|
324 | self.size = struct.calcsize(indexformatv0) | |
323 |
|
325 | |||
324 | def parseindex(self, fp, data, inline): |
|
326 | def parseindex(self, fp, data, inline): | |
325 | s = self.size |
|
327 | s = self.size | |
326 | index = [] |
|
328 | index = [] | |
327 | nodemap = {nullid: nullrev} |
|
329 | nodemap = {nullid: nullrev} | |
328 | n = off = 0 |
|
330 | n = off = 0 | |
329 | if len(data) == _prereadsize: |
|
331 | if len(data) == _prereadsize: | |
330 | data += fp.read() # read the rest |
|
332 | data += fp.read() # read the rest | |
331 | l = len(data) |
|
333 | l = len(data) | |
332 | while off + s <= l: |
|
334 | while off + s <= l: | |
333 | cur = data[off:off + s] |
|
335 | cur = data[off:off + s] | |
334 | off += s |
|
336 | off += s | |
335 | e = _unpack(indexformatv0, cur) |
|
337 | e = _unpack(indexformatv0, cur) | |
336 | # transform to revlogv1 format |
|
338 | # transform to revlogv1 format | |
337 | e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3], |
|
339 | e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3], | |
338 | nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6]) |
|
340 | nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6]) | |
339 | index.append(e2) |
|
341 | index.append(e2) | |
340 | nodemap[e[6]] = n |
|
342 | nodemap[e[6]] = n | |
341 | n += 1 |
|
343 | n += 1 | |
342 |
|
344 | |||
343 | return index, nodemap, None |
|
345 | return index, nodemap, None | |
344 |
|
346 | |||
345 | def packentry(self, entry, node, version, rev): |
|
347 | def packentry(self, entry, node, version, rev): | |
346 | if gettype(entry[0]): |
|
348 | if gettype(entry[0]): | |
347 | raise RevlogError(_("index entry flags need RevlogNG")) |
|
349 | raise RevlogError(_("index entry flags need RevlogNG")) | |
348 | e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4], |
|
350 | e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4], | |
349 | node(entry[5]), node(entry[6]), entry[7]) |
|
351 | node(entry[5]), node(entry[6]), entry[7]) | |
350 | return _pack(indexformatv0, *e2) |
|
352 | return _pack(indexformatv0, *e2) | |
351 |
|
353 | |||
352 | # index ng: |
|
354 | # index ng: | |
353 | # 6 bytes offset |
|
355 | # 6 bytes offset | |
354 | # 2 bytes flags |
|
356 | # 2 bytes flags | |
355 | # 4 bytes compressed length |
|
357 | # 4 bytes compressed length | |
356 | # 4 bytes uncompressed length |
|
358 | # 4 bytes uncompressed length | |
357 | # 4 bytes: base rev |
|
359 | # 4 bytes: base rev | |
358 | # 4 bytes link rev |
|
360 | # 4 bytes link rev | |
359 | # 4 bytes parent 1 rev |
|
361 | # 4 bytes parent 1 rev | |
360 | # 4 bytes parent 2 rev |
|
362 | # 4 bytes parent 2 rev | |
361 | # 32 bytes: nodeid |
|
363 | # 32 bytes: nodeid | |
362 | indexformatng = ">Qiiiiii20s12x" |
|
364 | indexformatng = ">Qiiiiii20s12x" | |
363 | ngshaoffset = 32 |
|
365 | ngshaoffset = 32 | |
364 | versionformat = ">I" |
|
366 | versionformat = ">I" | |
365 |
|
367 | |||
366 | class revlogio(object): |
|
368 | class revlogio(object): | |
367 | def __init__(self): |
|
369 | def __init__(self): | |
368 | self.size = struct.calcsize(indexformatng) |
|
370 | self.size = struct.calcsize(indexformatng) | |
369 |
|
371 | |||
370 | def parseindex(self, fp, data, inline): |
|
372 | def parseindex(self, fp, data, inline): | |
371 | if len(data) == _prereadsize: |
|
373 | if len(data) == _prereadsize: | |
372 | if util.openhardlinks() and not inline: |
|
374 | if util.openhardlinks() and not inline: | |
373 | # big index, let's parse it on demand |
|
375 | # big index, let's parse it on demand | |
374 | parser = lazyparser(fp) |
|
376 | parser = lazyparser(fp) | |
375 | index = lazyindex(parser) |
|
377 | index = lazyindex(parser) | |
376 | nodemap = lazymap(parser) |
|
378 | nodemap = lazymap(parser) | |
377 | e = list(index[0]) |
|
379 | e = list(index[0]) | |
378 | type = gettype(e[0]) |
|
380 | type = gettype(e[0]) | |
379 | e[0] = offset_type(0, type) |
|
381 | e[0] = offset_type(0, type) | |
380 | index[0] = e |
|
382 | index[0] = e | |
381 | return index, nodemap, None |
|
383 | return index, nodemap, None | |
382 | else: |
|
384 | else: | |
383 | data += fp.read() |
|
385 | data += fp.read() | |
384 |
|
386 | |||
385 | # call the C implementation to parse the index data |
|
387 | # call the C implementation to parse the index data | |
386 | index, nodemap, cache = parsers.parse_index(data, inline) |
|
388 | index, nodemap, cache = parsers.parse_index(data, inline) | |
387 | return index, nodemap, cache |
|
389 | return index, nodemap, cache | |
388 |
|
390 | |||
389 | def packentry(self, entry, node, version, rev): |
|
391 | def packentry(self, entry, node, version, rev): | |
390 | p = _pack(indexformatng, *entry) |
|
392 | p = _pack(indexformatng, *entry) | |
391 | if rev == 0: |
|
393 | if rev == 0: | |
392 | p = _pack(versionformat, version) + p[4:] |
|
394 | p = _pack(versionformat, version) + p[4:] | |
393 | return p |
|
395 | return p | |
394 |
|
396 | |||
395 | class revlog(object): |
|
397 | class revlog(object): | |
396 | """ |
|
398 | """ | |
397 | the underlying revision storage object |
|
399 | the underlying revision storage object | |
398 |
|
400 | |||
399 | A revlog consists of two parts, an index and the revision data. |
|
401 | A revlog consists of two parts, an index and the revision data. | |
400 |
|
402 | |||
401 | The index is a file with a fixed record size containing |
|
403 | The index is a file with a fixed record size containing | |
402 | information on each revision, including its nodeid (hash), the |
|
404 | information on each revision, including its nodeid (hash), the | |
403 | nodeids of its parents, the position and offset of its data within |
|
405 | nodeids of its parents, the position and offset of its data within | |
404 | the data file, and the revision it's based on. Finally, each entry |
|
406 | the data file, and the revision it's based on. Finally, each entry | |
405 | contains a linkrev entry that can serve as a pointer to external |
|
407 | contains a linkrev entry that can serve as a pointer to external | |
406 | data. |
|
408 | data. | |
407 |
|
409 | |||
408 | The revision data itself is a linear collection of data chunks. |
|
410 | The revision data itself is a linear collection of data chunks. | |
409 | Each chunk represents a revision and is usually represented as a |
|
411 | Each chunk represents a revision and is usually represented as a | |
410 | delta against the previous chunk. To bound lookup time, runs of |
|
412 | delta against the previous chunk. To bound lookup time, runs of | |
411 | deltas are limited to about 2 times the length of the original |
|
413 | deltas are limited to about 2 times the length of the original | |
412 | version data. This makes retrieval of a version proportional to |
|
414 | version data. This makes retrieval of a version proportional to | |
413 | its size, or O(1) relative to the number of revisions. |
|
415 | its size, or O(1) relative to the number of revisions. | |
414 |
|
416 | |||
415 | Both pieces of the revlog are written to in an append-only |
|
417 | Both pieces of the revlog are written to in an append-only | |
416 | fashion, which means we never need to rewrite a file to insert or |
|
418 | fashion, which means we never need to rewrite a file to insert or | |
417 | remove data, and can use some simple techniques to avoid the need |
|
419 | remove data, and can use some simple techniques to avoid the need | |
418 | for locking while reading. |
|
420 | for locking while reading. | |
419 | """ |
|
421 | """ | |
420 | def __init__(self, opener, indexfile): |
|
422 | def __init__(self, opener, indexfile): | |
421 | """ |
|
423 | """ | |
422 | create a revlog object |
|
424 | create a revlog object | |
423 |
|
425 | |||
424 | opener is a function that abstracts the file opening operation |
|
426 | opener is a function that abstracts the file opening operation | |
425 | and can be used to implement COW semantics or the like. |
|
427 | and can be used to implement COW semantics or the like. | |
426 | """ |
|
428 | """ | |
427 | self.indexfile = indexfile |
|
429 | self.indexfile = indexfile | |
428 | self.datafile = indexfile[:-2] + ".d" |
|
430 | self.datafile = indexfile[:-2] + ".d" | |
429 | self.opener = opener |
|
431 | self.opener = opener | |
430 | self._cache = None |
|
432 | self._cache = None | |
431 | self._chunkcache = (0, '') |
|
433 | self._chunkcache = (0, '') | |
432 | self.nodemap = {nullid: nullrev} |
|
434 | self.nodemap = {nullid: nullrev} | |
433 | self.index = [] |
|
435 | self.index = [] | |
434 |
|
436 | |||
435 | v = REVLOG_DEFAULT_VERSION |
|
437 | v = REVLOG_DEFAULT_VERSION | |
436 | if hasattr(opener, 'options') and 'defversion' in opener.options: |
|
438 | if hasattr(opener, 'options') and 'defversion' in opener.options: | |
437 | v = opener.options['defversion'] |
|
439 | v = opener.options['defversion'] | |
438 | if v & REVLOGNG: |
|
440 | if v & REVLOGNG: | |
439 | v |= REVLOGNGINLINEDATA |
|
441 | v |= REVLOGNGINLINEDATA | |
440 |
|
442 | |||
441 | i = '' |
|
443 | i = '' | |
442 | try: |
|
444 | try: | |
443 | f = self.opener(self.indexfile) |
|
445 | f = self.opener(self.indexfile) | |
444 | i = f.read(_prereadsize) |
|
446 | i = f.read(_prereadsize) | |
445 | if len(i) > 0: |
|
447 | if len(i) > 0: | |
446 | v = struct.unpack(versionformat, i[:4])[0] |
|
448 | v = struct.unpack(versionformat, i[:4])[0] | |
447 | except IOError, inst: |
|
449 | except IOError, inst: | |
448 | if inst.errno != errno.ENOENT: |
|
450 | if inst.errno != errno.ENOENT: | |
449 | raise |
|
451 | raise | |
450 |
|
452 | |||
451 | self.version = v |
|
453 | self.version = v | |
452 | self._inline = v & REVLOGNGINLINEDATA |
|
454 | self._inline = v & REVLOGNGINLINEDATA | |
453 | flags = v & ~0xFFFF |
|
455 | flags = v & ~0xFFFF | |
454 | fmt = v & 0xFFFF |
|
456 | fmt = v & 0xFFFF | |
455 | if fmt == REVLOGV0 and flags: |
|
457 | if fmt == REVLOGV0 and flags: | |
456 | raise RevlogError(_("index %s unknown flags %#04x for format v0") |
|
458 | raise RevlogError(_("index %s unknown flags %#04x for format v0") | |
457 | % (self.indexfile, flags >> 16)) |
|
459 | % (self.indexfile, flags >> 16)) | |
458 | elif fmt == REVLOGNG and flags & ~REVLOGNGINLINEDATA: |
|
460 | elif fmt == REVLOGNG and flags & ~REVLOGNGINLINEDATA: | |
459 | raise RevlogError(_("index %s unknown flags %#04x for revlogng") |
|
461 | raise RevlogError(_("index %s unknown flags %#04x for revlogng") | |
460 | % (self.indexfile, flags >> 16)) |
|
462 | % (self.indexfile, flags >> 16)) | |
461 | elif fmt > REVLOGNG: |
|
463 | elif fmt > REVLOGNG: | |
462 | raise RevlogError(_("index %s unknown format %d") |
|
464 | raise RevlogError(_("index %s unknown format %d") | |
463 | % (self.indexfile, fmt)) |
|
465 | % (self.indexfile, fmt)) | |
464 |
|
466 | |||
465 | self._io = revlogio() |
|
467 | self._io = revlogio() | |
466 | if self.version == REVLOGV0: |
|
468 | if self.version == REVLOGV0: | |
467 | self._io = revlogoldio() |
|
469 | self._io = revlogoldio() | |
468 | if i: |
|
470 | if i: | |
469 | try: |
|
471 | try: | |
470 | d = self._io.parseindex(f, i, self._inline) |
|
472 | d = self._io.parseindex(f, i, self._inline) | |
471 | except (ValueError, IndexError): |
|
473 | except (ValueError, IndexError): | |
472 | raise RevlogError(_("index %s is corrupted") % (self.indexfile)) |
|
474 | raise RevlogError(_("index %s is corrupted") % (self.indexfile)) | |
473 | self.index, self.nodemap, self._chunkcache = d |
|
475 | self.index, self.nodemap, self._chunkcache = d | |
474 | if not self._chunkcache: |
|
476 | if not self._chunkcache: | |
475 | self._chunkclear() |
|
477 | self._chunkclear() | |
476 |
|
478 | |||
477 | # add the magic null revision at -1 (if it hasn't been done already) |
|
479 | # add the magic null revision at -1 (if it hasn't been done already) | |
478 | if (self.index == [] or isinstance(self.index, lazyindex) or |
|
480 | if (self.index == [] or isinstance(self.index, lazyindex) or | |
479 | self.index[-1][7] != nullid) : |
|
481 | self.index[-1][7] != nullid) : | |
480 | self.index.append((0, 0, 0, -1, -1, -1, -1, nullid)) |
|
482 | self.index.append((0, 0, 0, -1, -1, -1, -1, nullid)) | |
481 |
|
483 | |||
482 | def _loadindex(self, start, end): |
|
484 | def _loadindex(self, start, end): | |
483 | """load a block of indexes all at once from the lazy parser""" |
|
485 | """load a block of indexes all at once from the lazy parser""" | |
484 | if isinstance(self.index, lazyindex): |
|
486 | if isinstance(self.index, lazyindex): | |
485 | self.index.p.loadindex(start, end) |
|
487 | self.index.p.loadindex(start, end) | |
486 |
|
488 | |||
487 | def _loadindexmap(self): |
|
489 | def _loadindexmap(self): | |
488 | """loads both the map and the index from the lazy parser""" |
|
490 | """loads both the map and the index from the lazy parser""" | |
489 | if isinstance(self.index, lazyindex): |
|
491 | if isinstance(self.index, lazyindex): | |
490 | p = self.index.p |
|
492 | p = self.index.p | |
491 | p.loadindex() |
|
493 | p.loadindex() | |
492 | self.nodemap = p.map |
|
494 | self.nodemap = p.map | |
493 |
|
495 | |||
494 | def _loadmap(self): |
|
496 | def _loadmap(self): | |
495 | """loads the map from the lazy parser""" |
|
497 | """loads the map from the lazy parser""" | |
496 | if isinstance(self.nodemap, lazymap): |
|
498 | if isinstance(self.nodemap, lazymap): | |
497 | self.nodemap.p.loadmap() |
|
499 | self.nodemap.p.loadmap() | |
498 | self.nodemap = self.nodemap.p.map |
|
500 | self.nodemap = self.nodemap.p.map | |
499 |
|
501 | |||
500 | def tip(self): |
|
502 | def tip(self): | |
501 | return self.node(len(self.index) - 2) |
|
503 | return self.node(len(self.index) - 2) | |
502 | def __len__(self): |
|
504 | def __len__(self): | |
503 | return len(self.index) - 1 |
|
505 | return len(self.index) - 1 | |
504 | def __iter__(self): |
|
506 | def __iter__(self): | |
505 | for i in xrange(len(self)): |
|
507 | for i in xrange(len(self)): | |
506 | yield i |
|
508 | yield i | |
507 | def rev(self, node): |
|
509 | def rev(self, node): | |
508 | try: |
|
510 | try: | |
509 | return self.nodemap[node] |
|
511 | return self.nodemap[node] | |
510 | except KeyError: |
|
512 | except KeyError: | |
511 | raise LookupError(node, self.indexfile, _('no node')) |
|
513 | raise LookupError(node, self.indexfile, _('no node')) | |
512 | def node(self, rev): |
|
514 | def node(self, rev): | |
513 | return self.index[rev][7] |
|
515 | return self.index[rev][7] | |
514 | def linkrev(self, rev): |
|
516 | def linkrev(self, rev): | |
515 | return self.index[rev][4] |
|
517 | return self.index[rev][4] | |
516 | def parents(self, node): |
|
518 | def parents(self, node): | |
517 | i = self.index |
|
519 | i = self.index | |
518 | d = i[self.rev(node)] |
|
520 | d = i[self.rev(node)] | |
519 | return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline |
|
521 | return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline | |
520 | def parentrevs(self, rev): |
|
522 | def parentrevs(self, rev): | |
521 | return self.index[rev][5:7] |
|
523 | return self.index[rev][5:7] | |
522 | def start(self, rev): |
|
524 | def start(self, rev): | |
523 | return int(self.index[rev][0] >> 16) |
|
525 | return int(self.index[rev][0] >> 16) | |
524 | def end(self, rev): |
|
526 | def end(self, rev): | |
525 | return self.start(rev) + self.length(rev) |
|
527 | return self.start(rev) + self.length(rev) | |
526 | def length(self, rev): |
|
528 | def length(self, rev): | |
527 | return self.index[rev][1] |
|
529 | return self.index[rev][1] | |
528 | def base(self, rev): |
|
530 | def base(self, rev): | |
529 | return self.index[rev][3] |
|
531 | return self.index[rev][3] | |
530 |
|
532 | |||
531 | def size(self, rev): |
|
533 | def size(self, rev): | |
532 | """return the length of the uncompressed text for a given revision""" |
|
534 | """return the length of the uncompressed text for a given revision""" | |
533 | l = self.index[rev][2] |
|
535 | l = self.index[rev][2] | |
534 | if l >= 0: |
|
536 | if l >= 0: | |
535 | return l |
|
537 | return l | |
536 |
|
538 | |||
537 | t = self.revision(self.node(rev)) |
|
539 | t = self.revision(self.node(rev)) | |
538 | return len(t) |
|
540 | return len(t) | |
539 |
|
541 | |||
540 | def reachable(self, node, stop=None): |
|
542 | def reachable(self, node, stop=None): | |
541 | """return the set of all nodes ancestral to a given node, including |
|
543 | """return the set of all nodes ancestral to a given node, including | |
542 | the node itself, stopping when stop is matched""" |
|
544 | the node itself, stopping when stop is matched""" | |
543 | reachable = set((node,)) |
|
545 | reachable = set((node,)) | |
544 | visit = [node] |
|
546 | visit = [node] | |
545 | if stop: |
|
547 | if stop: | |
546 | stopn = self.rev(stop) |
|
548 | stopn = self.rev(stop) | |
547 | else: |
|
549 | else: | |
548 | stopn = 0 |
|
550 | stopn = 0 | |
549 | while visit: |
|
551 | while visit: | |
550 | n = visit.pop(0) |
|
552 | n = visit.pop(0) | |
551 | if n == stop: |
|
553 | if n == stop: | |
552 | continue |
|
554 | continue | |
553 | if n == nullid: |
|
555 | if n == nullid: | |
554 | continue |
|
556 | continue | |
555 | for p in self.parents(n): |
|
557 | for p in self.parents(n): | |
556 | if self.rev(p) < stopn: |
|
558 | if self.rev(p) < stopn: | |
557 | continue |
|
559 | continue | |
558 | if p not in reachable: |
|
560 | if p not in reachable: | |
559 | reachable.add(p) |
|
561 | reachable.add(p) | |
560 | visit.append(p) |
|
562 | visit.append(p) | |
561 | return reachable |
|
563 | return reachable | |
562 |
|
564 | |||
563 | def ancestors(self, *revs): |
|
565 | def ancestors(self, *revs): | |
564 | """Generate the ancestors of 'revs' in reverse topological order. |
|
566 | """Generate the ancestors of 'revs' in reverse topological order. | |
565 |
|
567 | |||
566 | Yield a sequence of revision numbers starting with the parents |
|
568 | Yield a sequence of revision numbers starting with the parents | |
567 | of each revision in revs, i.e., each revision is *not* considered |
|
569 | of each revision in revs, i.e., each revision is *not* considered | |
568 | an ancestor of itself. Results are in breadth-first order: |
|
570 | an ancestor of itself. Results are in breadth-first order: | |
569 | parents of each rev in revs, then parents of those, etc. Result |
|
571 | parents of each rev in revs, then parents of those, etc. Result | |
570 | does not include the null revision.""" |
|
572 | does not include the null revision.""" | |
571 | visit = list(revs) |
|
573 | visit = list(revs) | |
572 | seen = set([nullrev]) |
|
574 | seen = set([nullrev]) | |
573 | while visit: |
|
575 | while visit: | |
574 | for parent in self.parentrevs(visit.pop(0)): |
|
576 | for parent in self.parentrevs(visit.pop(0)): | |
575 | if parent not in seen: |
|
577 | if parent not in seen: | |
576 | visit.append(parent) |
|
578 | visit.append(parent) | |
577 | seen.add(parent) |
|
579 | seen.add(parent) | |
578 | yield parent |
|
580 | yield parent | |
579 |
|
581 | |||
580 | def descendants(self, *revs): |
|
582 | def descendants(self, *revs): | |
581 | """Generate the descendants of 'revs' in revision order. |
|
583 | """Generate the descendants of 'revs' in revision order. | |
582 |
|
584 | |||
583 | Yield a sequence of revision numbers starting with a child of |
|
585 | Yield a sequence of revision numbers starting with a child of | |
584 | some rev in revs, i.e., each revision is *not* considered a |
|
586 | some rev in revs, i.e., each revision is *not* considered a | |
585 | descendant of itself. Results are ordered by revision number (a |
|
587 | descendant of itself. Results are ordered by revision number (a | |
586 | topological sort).""" |
|
588 | topological sort).""" | |
587 | seen = set(revs) |
|
589 | seen = set(revs) | |
588 | for i in xrange(min(revs) + 1, len(self)): |
|
590 | for i in xrange(min(revs) + 1, len(self)): | |
589 | for x in self.parentrevs(i): |
|
591 | for x in self.parentrevs(i): | |
590 | if x != nullrev and x in seen: |
|
592 | if x != nullrev and x in seen: | |
591 | seen.add(i) |
|
593 | seen.add(i) | |
592 | yield i |
|
594 | yield i | |
593 | break |
|
595 | break | |
594 |
|
596 | |||
595 | def findmissing(self, common=None, heads=None): |
|
597 | def findmissing(self, common=None, heads=None): | |
596 | """Return the ancestors of heads that are not ancestors of common. |
|
598 | """Return the ancestors of heads that are not ancestors of common. | |
597 |
|
599 | |||
598 | More specifically, return a list of nodes N such that every N |
|
600 | More specifically, return a list of nodes N such that every N | |
599 | satisfies the following constraints: |
|
601 | satisfies the following constraints: | |
600 |
|
602 | |||
601 | 1. N is an ancestor of some node in 'heads' |
|
603 | 1. N is an ancestor of some node in 'heads' | |
602 | 2. N is not an ancestor of any node in 'common' |
|
604 | 2. N is not an ancestor of any node in 'common' | |
603 |
|
605 | |||
604 | The list is sorted by revision number, meaning it is |
|
606 | The list is sorted by revision number, meaning it is | |
605 | topologically sorted. |
|
607 | topologically sorted. | |
606 |
|
608 | |||
607 | 'heads' and 'common' are both lists of node IDs. If heads is |
|
609 | 'heads' and 'common' are both lists of node IDs. If heads is | |
608 | not supplied, uses all of the revlog's heads. If common is not |
|
610 | not supplied, uses all of the revlog's heads. If common is not | |
609 | supplied, uses nullid.""" |
|
611 | supplied, uses nullid.""" | |
610 | if common is None: |
|
612 | if common is None: | |
611 | common = [nullid] |
|
613 | common = [nullid] | |
612 | if heads is None: |
|
614 | if heads is None: | |
613 | heads = self.heads() |
|
615 | heads = self.heads() | |
614 |
|
616 | |||
615 | common = [self.rev(n) for n in common] |
|
617 | common = [self.rev(n) for n in common] | |
616 | heads = [self.rev(n) for n in heads] |
|
618 | heads = [self.rev(n) for n in heads] | |
617 |
|
619 | |||
618 | # we want the ancestors, but inclusive |
|
620 | # we want the ancestors, but inclusive | |
619 | has = set(self.ancestors(*common)) |
|
621 | has = set(self.ancestors(*common)) | |
620 | has.add(nullrev) |
|
622 | has.add(nullrev) | |
621 | has.update(common) |
|
623 | has.update(common) | |
622 |
|
624 | |||
623 | # take all ancestors from heads that aren't in has |
|
625 | # take all ancestors from heads that aren't in has | |
624 | missing = set() |
|
626 | missing = set() | |
625 | visit = [r for r in heads if r not in has] |
|
627 | visit = [r for r in heads if r not in has] | |
626 | while visit: |
|
628 | while visit: | |
627 | r = visit.pop(0) |
|
629 | r = visit.pop(0) | |
628 | if r in missing: |
|
630 | if r in missing: | |
629 | continue |
|
631 | continue | |
630 | else: |
|
632 | else: | |
631 | missing.add(r) |
|
633 | missing.add(r) | |
632 | for p in self.parentrevs(r): |
|
634 | for p in self.parentrevs(r): | |
633 | if p not in has: |
|
635 | if p not in has: | |
634 | visit.append(p) |
|
636 | visit.append(p) | |
635 | missing = list(missing) |
|
637 | missing = list(missing) | |
636 | missing.sort() |
|
638 | missing.sort() | |
637 | return [self.node(r) for r in missing] |
|
639 | return [self.node(r) for r in missing] | |
638 |
|
640 | |||
639 | def nodesbetween(self, roots=None, heads=None): |
|
641 | def nodesbetween(self, roots=None, heads=None): | |
640 | """Return a topological path from 'roots' to 'heads'. |
|
642 | """Return a topological path from 'roots' to 'heads'. | |
641 |
|
643 | |||
642 | Return a tuple (nodes, outroots, outheads) where 'nodes' is a |
|
644 | Return a tuple (nodes, outroots, outheads) where 'nodes' is a | |
643 | topologically sorted list of all nodes N that satisfy both of |
|
645 | topologically sorted list of all nodes N that satisfy both of | |
644 | these constraints: |
|
646 | these constraints: | |
645 |
|
647 | |||
646 | 1. N is a descendant of some node in 'roots' |
|
648 | 1. N is a descendant of some node in 'roots' | |
647 | 2. N is an ancestor of some node in 'heads' |
|
649 | 2. N is an ancestor of some node in 'heads' | |
648 |
|
650 | |||
649 | Every node is considered to be both a descendant and an ancestor |
|
651 | Every node is considered to be both a descendant and an ancestor | |
650 | of itself, so every reachable node in 'roots' and 'heads' will be |
|
652 | of itself, so every reachable node in 'roots' and 'heads' will be | |
651 | included in 'nodes'. |
|
653 | included in 'nodes'. | |
652 |
|
654 | |||
653 | 'outroots' is the list of reachable nodes in 'roots', i.e., the |
|
655 | 'outroots' is the list of reachable nodes in 'roots', i.e., the | |
654 | subset of 'roots' that is returned in 'nodes'. Likewise, |
|
656 | subset of 'roots' that is returned in 'nodes'. Likewise, | |
655 | 'outheads' is the subset of 'heads' that is also in 'nodes'. |
|
657 | 'outheads' is the subset of 'heads' that is also in 'nodes'. | |
656 |
|
658 | |||
657 | 'roots' and 'heads' are both lists of node IDs. If 'roots' is |
|
659 | 'roots' and 'heads' are both lists of node IDs. If 'roots' is | |
658 | unspecified, uses nullid as the only root. If 'heads' is |
|
660 | unspecified, uses nullid as the only root. If 'heads' is | |
659 | unspecified, uses list of all of the revlog's heads.""" |
|
661 | unspecified, uses list of all of the revlog's heads.""" | |
660 | nonodes = ([], [], []) |
|
662 | nonodes = ([], [], []) | |
661 | if roots is not None: |
|
663 | if roots is not None: | |
662 | roots = list(roots) |
|
664 | roots = list(roots) | |
663 | if not roots: |
|
665 | if not roots: | |
664 | return nonodes |
|
666 | return nonodes | |
665 | lowestrev = min([self.rev(n) for n in roots]) |
|
667 | lowestrev = min([self.rev(n) for n in roots]) | |
666 | else: |
|
668 | else: | |
667 | roots = [nullid] # Everybody's a descendent of nullid |
|
669 | roots = [nullid] # Everybody's a descendent of nullid | |
668 | lowestrev = nullrev |
|
670 | lowestrev = nullrev | |
669 | if (lowestrev == nullrev) and (heads is None): |
|
671 | if (lowestrev == nullrev) and (heads is None): | |
670 | # We want _all_ the nodes! |
|
672 | # We want _all_ the nodes! | |
671 | return ([self.node(r) for r in self], [nullid], list(self.heads())) |
|
673 | return ([self.node(r) for r in self], [nullid], list(self.heads())) | |
672 | if heads is None: |
|
674 | if heads is None: | |
673 | # All nodes are ancestors, so the latest ancestor is the last |
|
675 | # All nodes are ancestors, so the latest ancestor is the last | |
674 | # node. |
|
676 | # node. | |
675 | highestrev = len(self) - 1 |
|
677 | highestrev = len(self) - 1 | |
676 | # Set ancestors to None to signal that every node is an ancestor. |
|
678 | # Set ancestors to None to signal that every node is an ancestor. | |
677 | ancestors = None |
|
679 | ancestors = None | |
678 | # Set heads to an empty dictionary for later discovery of heads |
|
680 | # Set heads to an empty dictionary for later discovery of heads | |
679 | heads = {} |
|
681 | heads = {} | |
680 | else: |
|
682 | else: | |
681 | heads = list(heads) |
|
683 | heads = list(heads) | |
682 | if not heads: |
|
684 | if not heads: | |
683 | return nonodes |
|
685 | return nonodes | |
684 | ancestors = set() |
|
686 | ancestors = set() | |
685 | # Turn heads into a dictionary so we can remove 'fake' heads. |
|
687 | # Turn heads into a dictionary so we can remove 'fake' heads. | |
686 | # Also, later we will be using it to filter out the heads we can't |
|
688 | # Also, later we will be using it to filter out the heads we can't | |
687 | # find from roots. |
|
689 | # find from roots. | |
688 | heads = dict.fromkeys(heads, 0) |
|
690 | heads = dict.fromkeys(heads, 0) | |
689 | # Start at the top and keep marking parents until we're done. |
|
691 | # Start at the top and keep marking parents until we're done. | |
690 | nodestotag = set(heads) |
|
692 | nodestotag = set(heads) | |
691 | # Remember where the top was so we can use it as a limit later. |
|
693 | # Remember where the top was so we can use it as a limit later. | |
692 | highestrev = max([self.rev(n) for n in nodestotag]) |
|
694 | highestrev = max([self.rev(n) for n in nodestotag]) | |
693 | while nodestotag: |
|
695 | while nodestotag: | |
694 | # grab a node to tag |
|
696 | # grab a node to tag | |
695 | n = nodestotag.pop() |
|
697 | n = nodestotag.pop() | |
696 | # Never tag nullid |
|
698 | # Never tag nullid | |
697 | if n == nullid: |
|
699 | if n == nullid: | |
698 | continue |
|
700 | continue | |
699 | # A node's revision number represents its place in a |
|
701 | # A node's revision number represents its place in a | |
700 | # topologically sorted list of nodes. |
|
702 | # topologically sorted list of nodes. | |
701 | r = self.rev(n) |
|
703 | r = self.rev(n) | |
702 | if r >= lowestrev: |
|
704 | if r >= lowestrev: | |
703 | if n not in ancestors: |
|
705 | if n not in ancestors: | |
704 | # If we are possibly a descendent of one of the roots |
|
706 | # If we are possibly a descendent of one of the roots | |
705 | # and we haven't already been marked as an ancestor |
|
707 | # and we haven't already been marked as an ancestor | |
706 | ancestors.add(n) # Mark as ancestor |
|
708 | ancestors.add(n) # Mark as ancestor | |
707 | # Add non-nullid parents to list of nodes to tag. |
|
709 | # Add non-nullid parents to list of nodes to tag. | |
708 | nodestotag.update([p for p in self.parents(n) if |
|
710 | nodestotag.update([p for p in self.parents(n) if | |
709 | p != nullid]) |
|
711 | p != nullid]) | |
710 | elif n in heads: # We've seen it before, is it a fake head? |
|
712 | elif n in heads: # We've seen it before, is it a fake head? | |
711 | # So it is, real heads should not be the ancestors of |
|
713 | # So it is, real heads should not be the ancestors of | |
712 | # any other heads. |
|
714 | # any other heads. | |
713 | heads.pop(n) |
|
715 | heads.pop(n) | |
714 | if not ancestors: |
|
716 | if not ancestors: | |
715 | return nonodes |
|
717 | return nonodes | |
716 | # Now that we have our set of ancestors, we want to remove any |
|
718 | # Now that we have our set of ancestors, we want to remove any | |
717 | # roots that are not ancestors. |
|
719 | # roots that are not ancestors. | |
718 |
|
720 | |||
719 | # If one of the roots was nullid, everything is included anyway. |
|
721 | # If one of the roots was nullid, everything is included anyway. | |
720 | if lowestrev > nullrev: |
|
722 | if lowestrev > nullrev: | |
721 | # But, since we weren't, let's recompute the lowest rev to not |
|
723 | # But, since we weren't, let's recompute the lowest rev to not | |
722 | # include roots that aren't ancestors. |
|
724 | # include roots that aren't ancestors. | |
723 |
|
725 | |||
724 | # Filter out roots that aren't ancestors of heads |
|
726 | # Filter out roots that aren't ancestors of heads | |
725 | roots = [n for n in roots if n in ancestors] |
|
727 | roots = [n for n in roots if n in ancestors] | |
726 | # Recompute the lowest revision |
|
728 | # Recompute the lowest revision | |
727 | if roots: |
|
729 | if roots: | |
728 | lowestrev = min([self.rev(n) for n in roots]) |
|
730 | lowestrev = min([self.rev(n) for n in roots]) | |
729 | else: |
|
731 | else: | |
730 | # No more roots? Return empty list |
|
732 | # No more roots? Return empty list | |
731 | return nonodes |
|
733 | return nonodes | |
732 | else: |
|
734 | else: | |
733 | # We are descending from nullid, and don't need to care about |
|
735 | # We are descending from nullid, and don't need to care about | |
734 | # any other roots. |
|
736 | # any other roots. | |
735 | lowestrev = nullrev |
|
737 | lowestrev = nullrev | |
736 | roots = [nullid] |
|
738 | roots = [nullid] | |
737 | # Transform our roots list into a set. |
|
739 | # Transform our roots list into a set. | |
738 | descendents = set(roots) |
|
740 | descendents = set(roots) | |
739 | # Also, keep the original roots so we can filter out roots that aren't |
|
741 | # Also, keep the original roots so we can filter out roots that aren't | |
740 | # 'real' roots (i.e. are descended from other roots). |
|
742 | # 'real' roots (i.e. are descended from other roots). | |
741 | roots = descendents.copy() |
|
743 | roots = descendents.copy() | |
742 | # Our topologically sorted list of output nodes. |
|
744 | # Our topologically sorted list of output nodes. | |
743 | orderedout = [] |
|
745 | orderedout = [] | |
744 | # Don't start at nullid since we don't want nullid in our output list, |
|
746 | # Don't start at nullid since we don't want nullid in our output list, | |
745 | # and if nullid shows up in descedents, empty parents will look like |
|
747 | # and if nullid shows up in descedents, empty parents will look like | |
746 | # they're descendents. |
|
748 | # they're descendents. | |
747 | for r in xrange(max(lowestrev, 0), highestrev + 1): |
|
749 | for r in xrange(max(lowestrev, 0), highestrev + 1): | |
748 | n = self.node(r) |
|
750 | n = self.node(r) | |
749 | isdescendent = False |
|
751 | isdescendent = False | |
750 | if lowestrev == nullrev: # Everybody is a descendent of nullid |
|
752 | if lowestrev == nullrev: # Everybody is a descendent of nullid | |
751 | isdescendent = True |
|
753 | isdescendent = True | |
752 | elif n in descendents: |
|
754 | elif n in descendents: | |
753 | # n is already a descendent |
|
755 | # n is already a descendent | |
754 | isdescendent = True |
|
756 | isdescendent = True | |
755 | # This check only needs to be done here because all the roots |
|
757 | # This check only needs to be done here because all the roots | |
756 | # will start being marked is descendents before the loop. |
|
758 | # will start being marked is descendents before the loop. | |
757 | if n in roots: |
|
759 | if n in roots: | |
758 | # If n was a root, check if it's a 'real' root. |
|
760 | # If n was a root, check if it's a 'real' root. | |
759 | p = tuple(self.parents(n)) |
|
761 | p = tuple(self.parents(n)) | |
760 | # If any of its parents are descendents, it's not a root. |
|
762 | # If any of its parents are descendents, it's not a root. | |
761 | if (p[0] in descendents) or (p[1] in descendents): |
|
763 | if (p[0] in descendents) or (p[1] in descendents): | |
762 | roots.remove(n) |
|
764 | roots.remove(n) | |
763 | else: |
|
765 | else: | |
764 | p = tuple(self.parents(n)) |
|
766 | p = tuple(self.parents(n)) | |
765 | # A node is a descendent if either of its parents are |
|
767 | # A node is a descendent if either of its parents are | |
766 | # descendents. (We seeded the dependents list with the roots |
|
768 | # descendents. (We seeded the dependents list with the roots | |
767 | # up there, remember?) |
|
769 | # up there, remember?) | |
768 | if (p[0] in descendents) or (p[1] in descendents): |
|
770 | if (p[0] in descendents) or (p[1] in descendents): | |
769 | descendents.add(n) |
|
771 | descendents.add(n) | |
770 | isdescendent = True |
|
772 | isdescendent = True | |
771 | if isdescendent and ((ancestors is None) or (n in ancestors)): |
|
773 | if isdescendent and ((ancestors is None) or (n in ancestors)): | |
772 | # Only include nodes that are both descendents and ancestors. |
|
774 | # Only include nodes that are both descendents and ancestors. | |
773 | orderedout.append(n) |
|
775 | orderedout.append(n) | |
774 | if (ancestors is not None) and (n in heads): |
|
776 | if (ancestors is not None) and (n in heads): | |
775 | # We're trying to figure out which heads are reachable |
|
777 | # We're trying to figure out which heads are reachable | |
776 | # from roots. |
|
778 | # from roots. | |
777 | # Mark this head as having been reached |
|
779 | # Mark this head as having been reached | |
778 | heads[n] = 1 |
|
780 | heads[n] = 1 | |
779 | elif ancestors is None: |
|
781 | elif ancestors is None: | |
780 | # Otherwise, we're trying to discover the heads. |
|
782 | # Otherwise, we're trying to discover the heads. | |
781 | # Assume this is a head because if it isn't, the next step |
|
783 | # Assume this is a head because if it isn't, the next step | |
782 | # will eventually remove it. |
|
784 | # will eventually remove it. | |
783 | heads[n] = 1 |
|
785 | heads[n] = 1 | |
784 | # But, obviously its parents aren't. |
|
786 | # But, obviously its parents aren't. | |
785 | for p in self.parents(n): |
|
787 | for p in self.parents(n): | |
786 | heads.pop(p, None) |
|
788 | heads.pop(p, None) | |
787 | heads = [n for n in heads.iterkeys() if heads[n] != 0] |
|
789 | heads = [n for n in heads.iterkeys() if heads[n] != 0] | |
788 | roots = list(roots) |
|
790 | roots = list(roots) | |
789 | assert orderedout |
|
791 | assert orderedout | |
790 | assert roots |
|
792 | assert roots | |
791 | assert heads |
|
793 | assert heads | |
792 | return (orderedout, roots, heads) |
|
794 | return (orderedout, roots, heads) | |
793 |
|
795 | |||
794 | def heads(self, start=None, stop=None): |
|
796 | def heads(self, start=None, stop=None): | |
795 | """return the list of all nodes that have no children |
|
797 | """return the list of all nodes that have no children | |
796 |
|
798 | |||
797 | if start is specified, only heads that are descendants of |
|
799 | if start is specified, only heads that are descendants of | |
798 | start will be returned |
|
800 | start will be returned | |
799 | if stop is specified, it will consider all the revs from stop |
|
801 | if stop is specified, it will consider all the revs from stop | |
800 | as if they had no children |
|
802 | as if they had no children | |
801 | """ |
|
803 | """ | |
802 | if start is None and stop is None: |
|
804 | if start is None and stop is None: | |
803 | count = len(self) |
|
805 | count = len(self) | |
804 | if not count: |
|
806 | if not count: | |
805 | return [nullid] |
|
807 | return [nullid] | |
806 | ishead = [1] * (count + 1) |
|
808 | ishead = [1] * (count + 1) | |
807 | index = self.index |
|
809 | index = self.index | |
808 | for r in xrange(count): |
|
810 | for r in xrange(count): | |
809 | e = index[r] |
|
811 | e = index[r] | |
810 | ishead[e[5]] = ishead[e[6]] = 0 |
|
812 | ishead[e[5]] = ishead[e[6]] = 0 | |
811 | return [self.node(r) for r in xrange(count) if ishead[r]] |
|
813 | return [self.node(r) for r in xrange(count) if ishead[r]] | |
812 |
|
814 | |||
813 | if start is None: |
|
815 | if start is None: | |
814 | start = nullid |
|
816 | start = nullid | |
815 | if stop is None: |
|
817 | if stop is None: | |
816 | stop = [] |
|
818 | stop = [] | |
817 | stoprevs = set([self.rev(n) for n in stop]) |
|
819 | stoprevs = set([self.rev(n) for n in stop]) | |
818 | startrev = self.rev(start) |
|
820 | startrev = self.rev(start) | |
819 | reachable = set((startrev,)) |
|
821 | reachable = set((startrev,)) | |
820 | heads = set((startrev,)) |
|
822 | heads = set((startrev,)) | |
821 |
|
823 | |||
822 | parentrevs = self.parentrevs |
|
824 | parentrevs = self.parentrevs | |
823 | for r in xrange(startrev + 1, len(self)): |
|
825 | for r in xrange(startrev + 1, len(self)): | |
824 | for p in parentrevs(r): |
|
826 | for p in parentrevs(r): | |
825 | if p in reachable: |
|
827 | if p in reachable: | |
826 | if r not in stoprevs: |
|
828 | if r not in stoprevs: | |
827 | reachable.add(r) |
|
829 | reachable.add(r) | |
828 | heads.add(r) |
|
830 | heads.add(r) | |
829 | if p in heads and p not in stoprevs: |
|
831 | if p in heads and p not in stoprevs: | |
830 | heads.remove(p) |
|
832 | heads.remove(p) | |
831 |
|
833 | |||
832 | return [self.node(r) for r in heads] |
|
834 | return [self.node(r) for r in heads] | |
833 |
|
835 | |||
834 | def children(self, node): |
|
836 | def children(self, node): | |
835 | """find the children of a given node""" |
|
837 | """find the children of a given node""" | |
836 | c = [] |
|
838 | c = [] | |
837 | p = self.rev(node) |
|
839 | p = self.rev(node) | |
838 | for r in range(p + 1, len(self)): |
|
840 | for r in range(p + 1, len(self)): | |
839 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] |
|
841 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] | |
840 | if prevs: |
|
842 | if prevs: | |
841 | for pr in prevs: |
|
843 | for pr in prevs: | |
842 | if pr == p: |
|
844 | if pr == p: | |
843 | c.append(self.node(r)) |
|
845 | c.append(self.node(r)) | |
844 | elif p == nullrev: |
|
846 | elif p == nullrev: | |
845 | c.append(self.node(r)) |
|
847 | c.append(self.node(r)) | |
846 | return c |
|
848 | return c | |
847 |
|
849 | |||
848 | def descendant(self, start, end): |
|
850 | def descendant(self, start, end): | |
849 | for i in self.descendants(start): |
|
851 | for i in self.descendants(start): | |
850 | if i == end: |
|
852 | if i == end: | |
851 | return True |
|
853 | return True | |
852 | elif i > end: |
|
854 | elif i > end: | |
853 | break |
|
855 | break | |
854 | return False |
|
856 | return False | |
855 |
|
857 | |||
856 | def ancestor(self, a, b): |
|
858 | def ancestor(self, a, b): | |
857 | """calculate the least common ancestor of nodes a and b""" |
|
859 | """calculate the least common ancestor of nodes a and b""" | |
858 |
|
860 | |||
859 | # fast path, check if it is a descendant |
|
861 | # fast path, check if it is a descendant | |
860 | a, b = self.rev(a), self.rev(b) |
|
862 | a, b = self.rev(a), self.rev(b) | |
861 | start, end = sorted((a, b)) |
|
863 | start, end = sorted((a, b)) | |
862 | if self.descendant(start, end): |
|
864 | if self.descendant(start, end): | |
863 | return self.node(start) |
|
865 | return self.node(start) | |
864 |
|
866 | |||
865 | def parents(rev): |
|
867 | def parents(rev): | |
866 | return [p for p in self.parentrevs(rev) if p != nullrev] |
|
868 | return [p for p in self.parentrevs(rev) if p != nullrev] | |
867 |
|
869 | |||
868 | c = ancestor.ancestor(a, b, parents) |
|
870 | c = ancestor.ancestor(a, b, parents) | |
869 | if c is None: |
|
871 | if c is None: | |
870 | return nullid |
|
872 | return nullid | |
871 |
|
873 | |||
872 | return self.node(c) |
|
874 | return self.node(c) | |
873 |
|
875 | |||
874 | def _match(self, id): |
|
876 | def _match(self, id): | |
875 | if isinstance(id, (long, int)): |
|
877 | if isinstance(id, (long, int)): | |
876 | # rev |
|
878 | # rev | |
877 | return self.node(id) |
|
879 | return self.node(id) | |
878 | if len(id) == 20: |
|
880 | if len(id) == 20: | |
879 | # possibly a binary node |
|
881 | # possibly a binary node | |
880 | # odds of a binary node being all hex in ASCII are 1 in 10**25 |
|
882 | # odds of a binary node being all hex in ASCII are 1 in 10**25 | |
881 | try: |
|
883 | try: | |
882 | node = id |
|
884 | node = id | |
883 | self.rev(node) # quick search the index |
|
885 | self.rev(node) # quick search the index | |
884 | return node |
|
886 | return node | |
885 | except LookupError: |
|
887 | except LookupError: | |
886 | pass # may be partial hex id |
|
888 | pass # may be partial hex id | |
887 | try: |
|
889 | try: | |
888 | # str(rev) |
|
890 | # str(rev) | |
889 | rev = int(id) |
|
891 | rev = int(id) | |
890 | if str(rev) != id: |
|
892 | if str(rev) != id: | |
891 | raise ValueError |
|
893 | raise ValueError | |
892 | if rev < 0: |
|
894 | if rev < 0: | |
893 | rev = len(self) + rev |
|
895 | rev = len(self) + rev | |
894 | if rev < 0 or rev >= len(self): |
|
896 | if rev < 0 or rev >= len(self): | |
895 | raise ValueError |
|
897 | raise ValueError | |
896 | return self.node(rev) |
|
898 | return self.node(rev) | |
897 | except (ValueError, OverflowError): |
|
899 | except (ValueError, OverflowError): | |
898 | pass |
|
900 | pass | |
899 | if len(id) == 40: |
|
901 | if len(id) == 40: | |
900 | try: |
|
902 | try: | |
901 | # a full hex nodeid? |
|
903 | # a full hex nodeid? | |
902 | node = bin(id) |
|
904 | node = bin(id) | |
903 | self.rev(node) |
|
905 | self.rev(node) | |
904 | return node |
|
906 | return node | |
905 | except (TypeError, LookupError): |
|
907 | except (TypeError, LookupError): | |
906 | pass |
|
908 | pass | |
907 |
|
909 | |||
908 | def _partialmatch(self, id): |
|
910 | def _partialmatch(self, id): | |
909 | if len(id) < 40: |
|
911 | if len(id) < 40: | |
910 | try: |
|
912 | try: | |
911 | # hex(node)[:...] |
|
913 | # hex(node)[:...] | |
912 | l = len(id) // 2 # grab an even number of digits |
|
914 | l = len(id) // 2 # grab an even number of digits | |
913 | bin_id = bin(id[:l * 2]) |
|
915 | bin_id = bin(id[:l * 2]) | |
914 | nl = [n for n in self.nodemap if n[:l] == bin_id] |
|
916 | nl = [n for n in self.nodemap if n[:l] == bin_id] | |
915 | nl = [n for n in nl if hex(n).startswith(id)] |
|
917 | nl = [n for n in nl if hex(n).startswith(id)] | |
916 | if len(nl) > 0: |
|
918 | if len(nl) > 0: | |
917 | if len(nl) == 1: |
|
919 | if len(nl) == 1: | |
918 | return nl[0] |
|
920 | return nl[0] | |
919 | raise LookupError(id, self.indexfile, |
|
921 | raise LookupError(id, self.indexfile, | |
920 | _('ambiguous identifier')) |
|
922 | _('ambiguous identifier')) | |
921 | return None |
|
923 | return None | |
922 | except TypeError: |
|
924 | except TypeError: | |
923 | pass |
|
925 | pass | |
924 |
|
926 | |||
925 | def lookup(self, id): |
|
927 | def lookup(self, id): | |
926 | """locate a node based on: |
|
928 | """locate a node based on: | |
927 | - revision number or str(revision number) |
|
929 | - revision number or str(revision number) | |
928 | - nodeid or subset of hex nodeid |
|
930 | - nodeid or subset of hex nodeid | |
929 | """ |
|
931 | """ | |
930 | n = self._match(id) |
|
932 | n = self._match(id) | |
931 | if n is not None: |
|
933 | if n is not None: | |
932 | return n |
|
934 | return n | |
933 | n = self._partialmatch(id) |
|
935 | n = self._partialmatch(id) | |
934 | if n: |
|
936 | if n: | |
935 | return n |
|
937 | return n | |
936 |
|
938 | |||
937 | raise LookupError(id, self.indexfile, _('no match found')) |
|
939 | raise LookupError(id, self.indexfile, _('no match found')) | |
938 |
|
940 | |||
939 | def cmp(self, node, text): |
|
941 | def cmp(self, node, text): | |
940 | """compare text with a given file revision""" |
|
942 | """compare text with a given file revision""" | |
941 | p1, p2 = self.parents(node) |
|
943 | p1, p2 = self.parents(node) | |
942 | return hash(text, p1, p2) != node |
|
944 | return hash(text, p1, p2) != node | |
943 |
|
945 | |||
944 | def _addchunk(self, offset, data): |
|
946 | def _addchunk(self, offset, data): | |
945 | o, d = self._chunkcache |
|
947 | o, d = self._chunkcache | |
946 | # try to add to existing cache |
|
948 | # try to add to existing cache | |
947 | if o + len(d) == offset and len(d) + len(data) < _prereadsize: |
|
949 | if o + len(d) == offset and len(d) + len(data) < _prereadsize: | |
948 | self._chunkcache = o, d + data |
|
950 | self._chunkcache = o, d + data | |
949 | else: |
|
951 | else: | |
950 | self._chunkcache = offset, data |
|
952 | self._chunkcache = offset, data | |
951 |
|
953 | |||
952 | def _loadchunk(self, offset, length): |
|
954 | def _loadchunk(self, offset, length): | |
953 | if self._inline: |
|
955 | if self._inline: | |
954 | df = self.opener(self.indexfile) |
|
956 | df = self.opener(self.indexfile) | |
955 | else: |
|
957 | else: | |
956 | df = self.opener(self.datafile) |
|
958 | df = self.opener(self.datafile) | |
957 |
|
959 | |||
958 | readahead = max(65536, length) |
|
960 | readahead = max(65536, length) | |
959 | df.seek(offset) |
|
961 | df.seek(offset) | |
960 | d = df.read(readahead) |
|
962 | d = df.read(readahead) | |
961 | self._addchunk(offset, d) |
|
963 | self._addchunk(offset, d) | |
962 | if readahead > length: |
|
964 | if readahead > length: | |
963 | return d[:length] |
|
965 | return d[:length] | |
964 | return d |
|
966 | return d | |
965 |
|
967 | |||
966 | def _getchunk(self, offset, length): |
|
968 | def _getchunk(self, offset, length): | |
967 | o, d = self._chunkcache |
|
969 | o, d = self._chunkcache | |
968 | l = len(d) |
|
970 | l = len(d) | |
969 |
|
971 | |||
970 | # is it in the cache? |
|
972 | # is it in the cache? | |
971 | cachestart = offset - o |
|
973 | cachestart = offset - o | |
972 | cacheend = cachestart + length |
|
974 | cacheend = cachestart + length | |
973 | if cachestart >= 0 and cacheend <= l: |
|
975 | if cachestart >= 0 and cacheend <= l: | |
974 | if cachestart == 0 and cacheend == l: |
|
976 | if cachestart == 0 and cacheend == l: | |
975 | return d # avoid a copy |
|
977 | return d # avoid a copy | |
976 | return d[cachestart:cacheend] |
|
978 | return d[cachestart:cacheend] | |
977 |
|
979 | |||
978 | return self._loadchunk(offset, length) |
|
980 | return self._loadchunk(offset, length) | |
979 |
|
981 | |||
980 | def _chunkraw(self, startrev, endrev): |
|
982 | def _chunkraw(self, startrev, endrev): | |
981 | start = self.start(startrev) |
|
983 | start = self.start(startrev) | |
982 | length = self.end(endrev) - start |
|
984 | length = self.end(endrev) - start | |
983 | if self._inline: |
|
985 | if self._inline: | |
984 | start += (startrev + 1) * self._io.size |
|
986 | start += (startrev + 1) * self._io.size | |
985 | return self._getchunk(start, length) |
|
987 | return self._getchunk(start, length) | |
986 |
|
988 | |||
987 | def _chunk(self, rev): |
|
989 | def _chunk(self, rev): | |
988 | return decompress(self._chunkraw(rev, rev)) |
|
990 | return decompress(self._chunkraw(rev, rev)) | |
989 |
|
991 | |||
990 | def _chunkclear(self): |
|
992 | def _chunkclear(self): | |
991 | self._chunkcache = (0, '') |
|
993 | self._chunkcache = (0, '') | |
992 |
|
994 | |||
993 | def revdiff(self, rev1, rev2): |
|
995 | def revdiff(self, rev1, rev2): | |
994 | """return or calculate a delta between two revisions""" |
|
996 | """return or calculate a delta between two revisions""" | |
995 | if rev1 + 1 == rev2 and self.base(rev1) == self.base(rev2): |
|
997 | if rev1 + 1 == rev2 and self.base(rev1) == self.base(rev2): | |
996 | return self._chunk(rev2) |
|
998 | return self._chunk(rev2) | |
997 |
|
999 | |||
998 | return mdiff.textdiff(self.revision(self.node(rev1)), |
|
1000 | return mdiff.textdiff(self.revision(self.node(rev1)), | |
999 | self.revision(self.node(rev2))) |
|
1001 | self.revision(self.node(rev2))) | |
1000 |
|
1002 | |||
1001 | def revision(self, node): |
|
1003 | def revision(self, node): | |
1002 | """return an uncompressed revision of a given node""" |
|
1004 | """return an uncompressed revision of a given node""" | |
1003 | if node == nullid: |
|
1005 | if node == nullid: | |
1004 | return "" |
|
1006 | return "" | |
1005 | if self._cache and self._cache[0] == node: |
|
1007 | if self._cache and self._cache[0] == node: | |
1006 | return self._cache[2] |
|
1008 | return self._cache[2] | |
1007 |
|
1009 | |||
1008 | # look up what we need to read |
|
1010 | # look up what we need to read | |
1009 | text = None |
|
1011 | text = None | |
1010 | rev = self.rev(node) |
|
1012 | rev = self.rev(node) | |
1011 | base = self.base(rev) |
|
1013 | base = self.base(rev) | |
1012 |
|
1014 | |||
1013 | # check rev flags |
|
1015 | # check rev flags | |
1014 | if self.index[rev][0] & 0xFFFF: |
|
1016 | if self.index[rev][0] & 0xFFFF: | |
1015 | raise RevlogError(_('incompatible revision flag %x') % |
|
1017 | raise RevlogError(_('incompatible revision flag %x') % | |
1016 | (self.index[rev][0] & 0xFFFF)) |
|
1018 | (self.index[rev][0] & 0xFFFF)) | |
1017 |
|
1019 | |||
1018 | # do we have useful data cached? |
|
1020 | # do we have useful data cached? | |
1019 | if self._cache and self._cache[1] >= base and self._cache[1] < rev: |
|
1021 | if self._cache and self._cache[1] >= base and self._cache[1] < rev: | |
1020 | base = self._cache[1] |
|
1022 | base = self._cache[1] | |
1021 | text = self._cache[2] |
|
1023 | text = self._cache[2] | |
1022 |
|
1024 | |||
1023 | self._loadindex(base, rev + 1) |
|
1025 | self._loadindex(base, rev + 1) | |
1024 | self._chunkraw(base, rev) |
|
1026 | self._chunkraw(base, rev) | |
1025 | if text is None: |
|
1027 | if text is None: | |
1026 | text = self._chunk(base) |
|
1028 | text = self._chunk(base) | |
1027 |
|
1029 | |||
1028 | bins = [self._chunk(r) for r in xrange(base + 1, rev + 1)] |
|
1030 | bins = [self._chunk(r) for r in xrange(base + 1, rev + 1)] | |
1029 | text = mdiff.patches(text, bins) |
|
1031 | text = mdiff.patches(text, bins) | |
1030 | p1, p2 = self.parents(node) |
|
1032 | p1, p2 = self.parents(node) | |
1031 | if node != hash(text, p1, p2): |
|
1033 | if node != hash(text, p1, p2): | |
1032 | raise RevlogError(_("integrity check failed on %s:%d") |
|
1034 | raise RevlogError(_("integrity check failed on %s:%d") | |
1033 | % (self.indexfile, rev)) |
|
1035 | % (self.indexfile, rev)) | |
1034 |
|
1036 | |||
1035 | self._cache = (node, rev, text) |
|
1037 | self._cache = (node, rev, text) | |
1036 | return text |
|
1038 | return text | |
1037 |
|
1039 | |||
1038 | def checkinlinesize(self, tr, fp=None): |
|
1040 | def checkinlinesize(self, tr, fp=None): | |
1039 |
if not self._inline or (self.start(-2) + self.length(-2)) < |
|
1041 | if not self._inline or (self.start(-2) + self.length(-2)) < _maxinline: | |
1040 | return |
|
1042 | return | |
1041 |
|
1043 | |||
1042 | trinfo = tr.find(self.indexfile) |
|
1044 | trinfo = tr.find(self.indexfile) | |
1043 | if trinfo is None: |
|
1045 | if trinfo is None: | |
1044 | raise RevlogError(_("%s not found in the transaction") |
|
1046 | raise RevlogError(_("%s not found in the transaction") | |
1045 | % self.indexfile) |
|
1047 | % self.indexfile) | |
1046 |
|
1048 | |||
1047 | trindex = trinfo[2] |
|
1049 | trindex = trinfo[2] | |
1048 | dataoff = self.start(trindex) |
|
1050 | dataoff = self.start(trindex) | |
1049 |
|
1051 | |||
1050 | tr.add(self.datafile, dataoff) |
|
1052 | tr.add(self.datafile, dataoff) | |
1051 |
|
1053 | |||
1052 | if fp: |
|
1054 | if fp: | |
1053 | fp.flush() |
|
1055 | fp.flush() | |
1054 | fp.close() |
|
1056 | fp.close() | |
1055 |
|
1057 | |||
1056 | df = self.opener(self.datafile, 'w') |
|
1058 | df = self.opener(self.datafile, 'w') | |
1057 | try: |
|
1059 | try: | |
1058 | for r in self: |
|
1060 | for r in self: | |
1059 | df.write(self._chunkraw(r, r)) |
|
1061 | df.write(self._chunkraw(r, r)) | |
1060 | finally: |
|
1062 | finally: | |
1061 | df.close() |
|
1063 | df.close() | |
1062 |
|
1064 | |||
1063 | fp = self.opener(self.indexfile, 'w', atomictemp=True) |
|
1065 | fp = self.opener(self.indexfile, 'w', atomictemp=True) | |
1064 | self.version &= ~(REVLOGNGINLINEDATA) |
|
1066 | self.version &= ~(REVLOGNGINLINEDATA) | |
1065 | self._inline = False |
|
1067 | self._inline = False | |
1066 | for i in self: |
|
1068 | for i in self: | |
1067 | e = self._io.packentry(self.index[i], self.node, self.version, i) |
|
1069 | e = self._io.packentry(self.index[i], self.node, self.version, i) | |
1068 | fp.write(e) |
|
1070 | fp.write(e) | |
1069 |
|
1071 | |||
1070 | # if we don't call rename, the temp file will never replace the |
|
1072 | # if we don't call rename, the temp file will never replace the | |
1071 | # real index |
|
1073 | # real index | |
1072 | fp.rename() |
|
1074 | fp.rename() | |
1073 |
|
1075 | |||
1074 | tr.replace(self.indexfile, trindex * self._io.size) |
|
1076 | tr.replace(self.indexfile, trindex * self._io.size) | |
1075 | self._chunkclear() |
|
1077 | self._chunkclear() | |
1076 |
|
1078 | |||
1077 | def addrevision(self, text, transaction, link, p1, p2, d=None): |
|
1079 | def addrevision(self, text, transaction, link, p1, p2, d=None): | |
1078 | """add a revision to the log |
|
1080 | """add a revision to the log | |
1079 |
|
1081 | |||
1080 | text - the revision data to add |
|
1082 | text - the revision data to add | |
1081 | transaction - the transaction object used for rollback |
|
1083 | transaction - the transaction object used for rollback | |
1082 | link - the linkrev data to add |
|
1084 | link - the linkrev data to add | |
1083 | p1, p2 - the parent nodeids of the revision |
|
1085 | p1, p2 - the parent nodeids of the revision | |
1084 | d - an optional precomputed delta |
|
1086 | d - an optional precomputed delta | |
1085 | """ |
|
1087 | """ | |
1086 | dfh = None |
|
1088 | dfh = None | |
1087 | if not self._inline: |
|
1089 | if not self._inline: | |
1088 | dfh = self.opener(self.datafile, "a") |
|
1090 | dfh = self.opener(self.datafile, "a") | |
1089 | ifh = self.opener(self.indexfile, "a+") |
|
1091 | ifh = self.opener(self.indexfile, "a+") | |
1090 | try: |
|
1092 | try: | |
1091 | return self._addrevision(text, transaction, link, p1, p2, d, ifh, dfh) |
|
1093 | return self._addrevision(text, transaction, link, p1, p2, d, ifh, dfh) | |
1092 | finally: |
|
1094 | finally: | |
1093 | if dfh: |
|
1095 | if dfh: | |
1094 | dfh.close() |
|
1096 | dfh.close() | |
1095 | ifh.close() |
|
1097 | ifh.close() | |
1096 |
|
1098 | |||
1097 | def _addrevision(self, text, transaction, link, p1, p2, d, ifh, dfh): |
|
1099 | def _addrevision(self, text, transaction, link, p1, p2, d, ifh, dfh): | |
1098 | node = hash(text, p1, p2) |
|
1100 | node = hash(text, p1, p2) | |
1099 | if node in self.nodemap: |
|
1101 | if node in self.nodemap: | |
1100 | return node |
|
1102 | return node | |
1101 |
|
1103 | |||
1102 | curr = len(self) |
|
1104 | curr = len(self) | |
1103 | prev = curr - 1 |
|
1105 | prev = curr - 1 | |
1104 | base = self.base(prev) |
|
1106 | base = self.base(prev) | |
1105 | offset = self.end(prev) |
|
1107 | offset = self.end(prev) | |
1106 |
|
1108 | |||
1107 | if curr: |
|
1109 | if curr: | |
1108 | if not d: |
|
1110 | if not d: | |
1109 | ptext = self.revision(self.node(prev)) |
|
1111 | ptext = self.revision(self.node(prev)) | |
1110 | d = mdiff.textdiff(ptext, text) |
|
1112 | d = mdiff.textdiff(ptext, text) | |
1111 | data = compress(d) |
|
1113 | data = compress(d) | |
1112 | l = len(data[1]) + len(data[0]) |
|
1114 | l = len(data[1]) + len(data[0]) | |
1113 | dist = l + offset - self.start(base) |
|
1115 | dist = l + offset - self.start(base) | |
1114 |
|
1116 | |||
1115 | # full versions are inserted when the needed deltas |
|
1117 | # full versions are inserted when the needed deltas | |
1116 | # become comparable to the uncompressed text |
|
1118 | # become comparable to the uncompressed text | |
1117 | if not curr or dist > len(text) * 2: |
|
1119 | if not curr or dist > len(text) * 2: | |
1118 | data = compress(text) |
|
1120 | data = compress(text) | |
1119 | l = len(data[1]) + len(data[0]) |
|
1121 | l = len(data[1]) + len(data[0]) | |
1120 | base = curr |
|
1122 | base = curr | |
1121 |
|
1123 | |||
1122 | e = (offset_type(offset, 0), l, len(text), |
|
1124 | e = (offset_type(offset, 0), l, len(text), | |
1123 | base, link, self.rev(p1), self.rev(p2), node) |
|
1125 | base, link, self.rev(p1), self.rev(p2), node) | |
1124 | self.index.insert(-1, e) |
|
1126 | self.index.insert(-1, e) | |
1125 | self.nodemap[node] = curr |
|
1127 | self.nodemap[node] = curr | |
1126 |
|
1128 | |||
1127 | entry = self._io.packentry(e, self.node, self.version, curr) |
|
1129 | entry = self._io.packentry(e, self.node, self.version, curr) | |
1128 | if not self._inline: |
|
1130 | if not self._inline: | |
1129 | transaction.add(self.datafile, offset) |
|
1131 | transaction.add(self.datafile, offset) | |
1130 | transaction.add(self.indexfile, curr * len(entry)) |
|
1132 | transaction.add(self.indexfile, curr * len(entry)) | |
1131 | if data[0]: |
|
1133 | if data[0]: | |
1132 | dfh.write(data[0]) |
|
1134 | dfh.write(data[0]) | |
1133 | dfh.write(data[1]) |
|
1135 | dfh.write(data[1]) | |
1134 | dfh.flush() |
|
1136 | dfh.flush() | |
1135 | ifh.write(entry) |
|
1137 | ifh.write(entry) | |
1136 | else: |
|
1138 | else: | |
1137 | offset += curr * self._io.size |
|
1139 | offset += curr * self._io.size | |
1138 | transaction.add(self.indexfile, offset, curr) |
|
1140 | transaction.add(self.indexfile, offset, curr) | |
1139 | ifh.write(entry) |
|
1141 | ifh.write(entry) | |
1140 | ifh.write(data[0]) |
|
1142 | ifh.write(data[0]) | |
1141 | ifh.write(data[1]) |
|
1143 | ifh.write(data[1]) | |
1142 | self.checkinlinesize(transaction, ifh) |
|
1144 | self.checkinlinesize(transaction, ifh) | |
1143 |
|
1145 | |||
1144 | if type(text) == str: # only accept immutable objects |
|
1146 | if type(text) == str: # only accept immutable objects | |
1145 | self._cache = (node, curr, text) |
|
1147 | self._cache = (node, curr, text) | |
1146 | return node |
|
1148 | return node | |
1147 |
|
1149 | |||
1148 | def group(self, nodelist, lookup, infocollect=None): |
|
1150 | def group(self, nodelist, lookup, infocollect=None): | |
1149 | """Calculate a delta group, yielding a sequence of changegroup chunks |
|
1151 | """Calculate a delta group, yielding a sequence of changegroup chunks | |
1150 | (strings). |
|
1152 | (strings). | |
1151 |
|
1153 | |||
1152 | Given a list of changeset revs, return a set of deltas and |
|
1154 | Given a list of changeset revs, return a set of deltas and | |
1153 | metadata corresponding to nodes. the first delta is |
|
1155 | metadata corresponding to nodes. the first delta is | |
1154 | parent(nodes[0]) -> nodes[0] the receiver is guaranteed to |
|
1156 | parent(nodes[0]) -> nodes[0] the receiver is guaranteed to | |
1155 | have this parent as it has all history before these |
|
1157 | have this parent as it has all history before these | |
1156 | changesets. parent is parent[0] |
|
1158 | changesets. parent is parent[0] | |
1157 | """ |
|
1159 | """ | |
1158 |
|
1160 | |||
1159 | revs = [self.rev(n) for n in nodelist] |
|
1161 | revs = [self.rev(n) for n in nodelist] | |
1160 |
|
1162 | |||
1161 | # if we don't have any revisions touched by these changesets, bail |
|
1163 | # if we don't have any revisions touched by these changesets, bail | |
1162 | if not revs: |
|
1164 | if not revs: | |
1163 | yield changegroup.closechunk() |
|
1165 | yield changegroup.closechunk() | |
1164 | return |
|
1166 | return | |
1165 |
|
1167 | |||
1166 | # add the parent of the first rev |
|
1168 | # add the parent of the first rev | |
1167 | p = self.parentrevs(revs[0])[0] |
|
1169 | p = self.parentrevs(revs[0])[0] | |
1168 | revs.insert(0, p) |
|
1170 | revs.insert(0, p) | |
1169 |
|
1171 | |||
1170 | # build deltas |
|
1172 | # build deltas | |
1171 | for d in xrange(len(revs) - 1): |
|
1173 | for d in xrange(len(revs) - 1): | |
1172 | a, b = revs[d], revs[d + 1] |
|
1174 | a, b = revs[d], revs[d + 1] | |
1173 | nb = self.node(b) |
|
1175 | nb = self.node(b) | |
1174 |
|
1176 | |||
1175 | if infocollect is not None: |
|
1177 | if infocollect is not None: | |
1176 | infocollect(nb) |
|
1178 | infocollect(nb) | |
1177 |
|
1179 | |||
1178 | p = self.parents(nb) |
|
1180 | p = self.parents(nb) | |
1179 | meta = nb + p[0] + p[1] + lookup(nb) |
|
1181 | meta = nb + p[0] + p[1] + lookup(nb) | |
1180 | if a == -1: |
|
1182 | if a == -1: | |
1181 | d = self.revision(nb) |
|
1183 | d = self.revision(nb) | |
1182 | meta += mdiff.trivialdiffheader(len(d)) |
|
1184 | meta += mdiff.trivialdiffheader(len(d)) | |
1183 | else: |
|
1185 | else: | |
1184 | d = self.revdiff(a, b) |
|
1186 | d = self.revdiff(a, b) | |
1185 | yield changegroup.chunkheader(len(meta) + len(d)) |
|
1187 | yield changegroup.chunkheader(len(meta) + len(d)) | |
1186 | yield meta |
|
1188 | yield meta | |
1187 | if len(d) > 2**20: |
|
1189 | if len(d) > 2**20: | |
1188 | pos = 0 |
|
1190 | pos = 0 | |
1189 | while pos < len(d): |
|
1191 | while pos < len(d): | |
1190 | pos2 = pos + 2 ** 18 |
|
1192 | pos2 = pos + 2 ** 18 | |
1191 | yield d[pos:pos2] |
|
1193 | yield d[pos:pos2] | |
1192 | pos = pos2 |
|
1194 | pos = pos2 | |
1193 | else: |
|
1195 | else: | |
1194 | yield d |
|
1196 | yield d | |
1195 |
|
1197 | |||
1196 | yield changegroup.closechunk() |
|
1198 | yield changegroup.closechunk() | |
1197 |
|
1199 | |||
1198 | def addgroup(self, revs, linkmapper, transaction): |
|
1200 | def addgroup(self, revs, linkmapper, transaction): | |
1199 | """ |
|
1201 | """ | |
1200 | add a delta group |
|
1202 | add a delta group | |
1201 |
|
1203 | |||
1202 | given a set of deltas, add them to the revision log. the |
|
1204 | given a set of deltas, add them to the revision log. the | |
1203 | first delta is against its parent, which should be in our |
|
1205 | first delta is against its parent, which should be in our | |
1204 | log, the rest are against the previous delta. |
|
1206 | log, the rest are against the previous delta. | |
1205 | """ |
|
1207 | """ | |
1206 |
|
1208 | |||
1207 | #track the base of the current delta log |
|
1209 | #track the base of the current delta log | |
1208 | r = len(self) |
|
1210 | r = len(self) | |
1209 | t = r - 1 |
|
1211 | t = r - 1 | |
1210 | node = None |
|
1212 | node = None | |
1211 |
|
1213 | |||
1212 | base = prev = nullrev |
|
1214 | base = prev = nullrev | |
1213 | start = end = textlen = 0 |
|
1215 | start = end = textlen = 0 | |
1214 | if r: |
|
1216 | if r: | |
1215 | end = self.end(t) |
|
1217 | end = self.end(t) | |
1216 |
|
1218 | |||
1217 | ifh = self.opener(self.indexfile, "a+") |
|
1219 | ifh = self.opener(self.indexfile, "a+") | |
1218 | isize = r * self._io.size |
|
1220 | isize = r * self._io.size | |
1219 | if self._inline: |
|
1221 | if self._inline: | |
1220 | transaction.add(self.indexfile, end + isize, r) |
|
1222 | transaction.add(self.indexfile, end + isize, r) | |
1221 | dfh = None |
|
1223 | dfh = None | |
1222 | else: |
|
1224 | else: | |
1223 | transaction.add(self.indexfile, isize, r) |
|
1225 | transaction.add(self.indexfile, isize, r) | |
1224 | transaction.add(self.datafile, end) |
|
1226 | transaction.add(self.datafile, end) | |
1225 | dfh = self.opener(self.datafile, "a") |
|
1227 | dfh = self.opener(self.datafile, "a") | |
1226 |
|
1228 | |||
1227 | try: |
|
1229 | try: | |
1228 | # loop through our set of deltas |
|
1230 | # loop through our set of deltas | |
1229 | chain = None |
|
1231 | chain = None | |
1230 | for chunk in revs: |
|
1232 | for chunk in revs: | |
1231 | node, p1, p2, cs = struct.unpack("20s20s20s20s", chunk[:80]) |
|
1233 | node, p1, p2, cs = struct.unpack("20s20s20s20s", chunk[:80]) | |
1232 | link = linkmapper(cs) |
|
1234 | link = linkmapper(cs) | |
1233 | if node in self.nodemap: |
|
1235 | if node in self.nodemap: | |
1234 | # this can happen if two branches make the same change |
|
1236 | # this can happen if two branches make the same change | |
1235 | chain = node |
|
1237 | chain = node | |
1236 | continue |
|
1238 | continue | |
1237 | delta = buffer(chunk, 80) |
|
1239 | delta = buffer(chunk, 80) | |
1238 | del chunk |
|
1240 | del chunk | |
1239 |
|
1241 | |||
1240 | for p in (p1, p2): |
|
1242 | for p in (p1, p2): | |
1241 | if not p in self.nodemap: |
|
1243 | if not p in self.nodemap: | |
1242 | raise LookupError(p, self.indexfile, _('unknown parent')) |
|
1244 | raise LookupError(p, self.indexfile, _('unknown parent')) | |
1243 |
|
1245 | |||
1244 | if not chain: |
|
1246 | if not chain: | |
1245 | # retrieve the parent revision of the delta chain |
|
1247 | # retrieve the parent revision of the delta chain | |
1246 | chain = p1 |
|
1248 | chain = p1 | |
1247 | if not chain in self.nodemap: |
|
1249 | if not chain in self.nodemap: | |
1248 | raise LookupError(chain, self.indexfile, _('unknown base')) |
|
1250 | raise LookupError(chain, self.indexfile, _('unknown base')) | |
1249 |
|
1251 | |||
1250 | # full versions are inserted when the needed deltas become |
|
1252 | # full versions are inserted when the needed deltas become | |
1251 | # comparable to the uncompressed text or when the previous |
|
1253 | # comparable to the uncompressed text or when the previous | |
1252 | # version is not the one we have a delta against. We use |
|
1254 | # version is not the one we have a delta against. We use | |
1253 | # the size of the previous full rev as a proxy for the |
|
1255 | # the size of the previous full rev as a proxy for the | |
1254 | # current size. |
|
1256 | # current size. | |
1255 |
|
1257 | |||
1256 | if chain == prev: |
|
1258 | if chain == prev: | |
1257 | cdelta = compress(delta) |
|
1259 | cdelta = compress(delta) | |
1258 | cdeltalen = len(cdelta[0]) + len(cdelta[1]) |
|
1260 | cdeltalen = len(cdelta[0]) + len(cdelta[1]) | |
1259 | textlen = mdiff.patchedsize(textlen, delta) |
|
1261 | textlen = mdiff.patchedsize(textlen, delta) | |
1260 |
|
1262 | |||
1261 | if chain != prev or (end - start + cdeltalen) > textlen * 2: |
|
1263 | if chain != prev or (end - start + cdeltalen) > textlen * 2: | |
1262 | # flush our writes here so we can read it in revision |
|
1264 | # flush our writes here so we can read it in revision | |
1263 | if dfh: |
|
1265 | if dfh: | |
1264 | dfh.flush() |
|
1266 | dfh.flush() | |
1265 | ifh.flush() |
|
1267 | ifh.flush() | |
1266 | text = self.revision(chain) |
|
1268 | text = self.revision(chain) | |
1267 | if len(text) == 0: |
|
1269 | if len(text) == 0: | |
1268 | # skip over trivial delta header |
|
1270 | # skip over trivial delta header | |
1269 | text = buffer(delta, 12) |
|
1271 | text = buffer(delta, 12) | |
1270 | else: |
|
1272 | else: | |
1271 | text = mdiff.patches(text, [delta]) |
|
1273 | text = mdiff.patches(text, [delta]) | |
1272 | del delta |
|
1274 | del delta | |
1273 | chk = self._addrevision(text, transaction, link, p1, p2, None, |
|
1275 | chk = self._addrevision(text, transaction, link, p1, p2, None, | |
1274 | ifh, dfh) |
|
1276 | ifh, dfh) | |
1275 | if not dfh and not self._inline: |
|
1277 | if not dfh and not self._inline: | |
1276 | # addrevision switched from inline to conventional |
|
1278 | # addrevision switched from inline to conventional | |
1277 | # reopen the index |
|
1279 | # reopen the index | |
1278 | dfh = self.opener(self.datafile, "a") |
|
1280 | dfh = self.opener(self.datafile, "a") | |
1279 | ifh = self.opener(self.indexfile, "a") |
|
1281 | ifh = self.opener(self.indexfile, "a") | |
1280 | if chk != node: |
|
1282 | if chk != node: | |
1281 | raise RevlogError(_("consistency error adding group")) |
|
1283 | raise RevlogError(_("consistency error adding group")) | |
1282 | textlen = len(text) |
|
1284 | textlen = len(text) | |
1283 | else: |
|
1285 | else: | |
1284 | e = (offset_type(end, 0), cdeltalen, textlen, base, |
|
1286 | e = (offset_type(end, 0), cdeltalen, textlen, base, | |
1285 | link, self.rev(p1), self.rev(p2), node) |
|
1287 | link, self.rev(p1), self.rev(p2), node) | |
1286 | self.index.insert(-1, e) |
|
1288 | self.index.insert(-1, e) | |
1287 | self.nodemap[node] = r |
|
1289 | self.nodemap[node] = r | |
1288 | entry = self._io.packentry(e, self.node, self.version, r) |
|
1290 | entry = self._io.packentry(e, self.node, self.version, r) | |
1289 | if self._inline: |
|
1291 | if self._inline: | |
1290 | ifh.write(entry) |
|
1292 | ifh.write(entry) | |
1291 | ifh.write(cdelta[0]) |
|
1293 | ifh.write(cdelta[0]) | |
1292 | ifh.write(cdelta[1]) |
|
1294 | ifh.write(cdelta[1]) | |
1293 | self.checkinlinesize(transaction, ifh) |
|
1295 | self.checkinlinesize(transaction, ifh) | |
1294 | if not self._inline: |
|
1296 | if not self._inline: | |
1295 | dfh = self.opener(self.datafile, "a") |
|
1297 | dfh = self.opener(self.datafile, "a") | |
1296 | ifh = self.opener(self.indexfile, "a") |
|
1298 | ifh = self.opener(self.indexfile, "a") | |
1297 | else: |
|
1299 | else: | |
1298 | dfh.write(cdelta[0]) |
|
1300 | dfh.write(cdelta[0]) | |
1299 | dfh.write(cdelta[1]) |
|
1301 | dfh.write(cdelta[1]) | |
1300 | ifh.write(entry) |
|
1302 | ifh.write(entry) | |
1301 |
|
1303 | |||
1302 | t, r, chain, prev = r, r + 1, node, node |
|
1304 | t, r, chain, prev = r, r + 1, node, node | |
1303 | base = self.base(t) |
|
1305 | base = self.base(t) | |
1304 | start = self.start(base) |
|
1306 | start = self.start(base) | |
1305 | end = self.end(t) |
|
1307 | end = self.end(t) | |
1306 | finally: |
|
1308 | finally: | |
1307 | if dfh: |
|
1309 | if dfh: | |
1308 | dfh.close() |
|
1310 | dfh.close() | |
1309 | ifh.close() |
|
1311 | ifh.close() | |
1310 |
|
1312 | |||
1311 | return node |
|
1313 | return node | |
1312 |
|
1314 | |||
1313 | def strip(self, minlink, transaction): |
|
1315 | def strip(self, minlink, transaction): | |
1314 | """truncate the revlog on the first revision with a linkrev >= minlink |
|
1316 | """truncate the revlog on the first revision with a linkrev >= minlink | |
1315 |
|
1317 | |||
1316 | This function is called when we're stripping revision minlink and |
|
1318 | This function is called when we're stripping revision minlink and | |
1317 | its descendants from the repository. |
|
1319 | its descendants from the repository. | |
1318 |
|
1320 | |||
1319 | We have to remove all revisions with linkrev >= minlink, because |
|
1321 | We have to remove all revisions with linkrev >= minlink, because | |
1320 | the equivalent changelog revisions will be renumbered after the |
|
1322 | the equivalent changelog revisions will be renumbered after the | |
1321 | strip. |
|
1323 | strip. | |
1322 |
|
1324 | |||
1323 | So we truncate the revlog on the first of these revisions, and |
|
1325 | So we truncate the revlog on the first of these revisions, and | |
1324 | trust that the caller has saved the revisions that shouldn't be |
|
1326 | trust that the caller has saved the revisions that shouldn't be | |
1325 | removed and that it'll readd them after this truncation. |
|
1327 | removed and that it'll readd them after this truncation. | |
1326 | """ |
|
1328 | """ | |
1327 | if len(self) == 0: |
|
1329 | if len(self) == 0: | |
1328 | return |
|
1330 | return | |
1329 |
|
1331 | |||
1330 | if isinstance(self.index, lazyindex): |
|
1332 | if isinstance(self.index, lazyindex): | |
1331 | self._loadindexmap() |
|
1333 | self._loadindexmap() | |
1332 |
|
1334 | |||
1333 | for rev in self: |
|
1335 | for rev in self: | |
1334 | if self.index[rev][4] >= minlink: |
|
1336 | if self.index[rev][4] >= minlink: | |
1335 | break |
|
1337 | break | |
1336 | else: |
|
1338 | else: | |
1337 | return |
|
1339 | return | |
1338 |
|
1340 | |||
1339 | # first truncate the files on disk |
|
1341 | # first truncate the files on disk | |
1340 | end = self.start(rev) |
|
1342 | end = self.start(rev) | |
1341 | if not self._inline: |
|
1343 | if not self._inline: | |
1342 | transaction.add(self.datafile, end) |
|
1344 | transaction.add(self.datafile, end) | |
1343 | end = rev * self._io.size |
|
1345 | end = rev * self._io.size | |
1344 | else: |
|
1346 | else: | |
1345 | end += rev * self._io.size |
|
1347 | end += rev * self._io.size | |
1346 |
|
1348 | |||
1347 | transaction.add(self.indexfile, end) |
|
1349 | transaction.add(self.indexfile, end) | |
1348 |
|
1350 | |||
1349 | # then reset internal state in memory to forget those revisions |
|
1351 | # then reset internal state in memory to forget those revisions | |
1350 | self._cache = None |
|
1352 | self._cache = None | |
1351 | self._chunkclear() |
|
1353 | self._chunkclear() | |
1352 | for x in xrange(rev, len(self)): |
|
1354 | for x in xrange(rev, len(self)): | |
1353 | del self.nodemap[self.node(x)] |
|
1355 | del self.nodemap[self.node(x)] | |
1354 |
|
1356 | |||
1355 | del self.index[rev:-1] |
|
1357 | del self.index[rev:-1] | |
1356 |
|
1358 | |||
1357 | def checksize(self): |
|
1359 | def checksize(self): | |
1358 | expected = 0 |
|
1360 | expected = 0 | |
1359 | if len(self): |
|
1361 | if len(self): | |
1360 | expected = max(0, self.end(len(self) - 1)) |
|
1362 | expected = max(0, self.end(len(self) - 1)) | |
1361 |
|
1363 | |||
1362 | try: |
|
1364 | try: | |
1363 | f = self.opener(self.datafile) |
|
1365 | f = self.opener(self.datafile) | |
1364 | f.seek(0, 2) |
|
1366 | f.seek(0, 2) | |
1365 | actual = f.tell() |
|
1367 | actual = f.tell() | |
1366 | dd = actual - expected |
|
1368 | dd = actual - expected | |
1367 | except IOError, inst: |
|
1369 | except IOError, inst: | |
1368 | if inst.errno != errno.ENOENT: |
|
1370 | if inst.errno != errno.ENOENT: | |
1369 | raise |
|
1371 | raise | |
1370 | dd = 0 |
|
1372 | dd = 0 | |
1371 |
|
1373 | |||
1372 | try: |
|
1374 | try: | |
1373 | f = self.opener(self.indexfile) |
|
1375 | f = self.opener(self.indexfile) | |
1374 | f.seek(0, 2) |
|
1376 | f.seek(0, 2) | |
1375 | actual = f.tell() |
|
1377 | actual = f.tell() | |
1376 | s = self._io.size |
|
1378 | s = self._io.size | |
1377 | i = max(0, actual // s) |
|
1379 | i = max(0, actual // s) | |
1378 | di = actual - (i * s) |
|
1380 | di = actual - (i * s) | |
1379 | if self._inline: |
|
1381 | if self._inline: | |
1380 | databytes = 0 |
|
1382 | databytes = 0 | |
1381 | for r in self: |
|
1383 | for r in self: | |
1382 | databytes += max(0, self.length(r)) |
|
1384 | databytes += max(0, self.length(r)) | |
1383 | dd = 0 |
|
1385 | dd = 0 | |
1384 | di = actual - len(self) * s - databytes |
|
1386 | di = actual - len(self) * s - databytes | |
1385 | except IOError, inst: |
|
1387 | except IOError, inst: | |
1386 | if inst.errno != errno.ENOENT: |
|
1388 | if inst.errno != errno.ENOENT: | |
1387 | raise |
|
1389 | raise | |
1388 | di = 0 |
|
1390 | di = 0 | |
1389 |
|
1391 | |||
1390 | return (dd, di) |
|
1392 | return (dd, di) | |
1391 |
|
1393 | |||
1392 | def files(self): |
|
1394 | def files(self): | |
1393 | res = [self.indexfile] |
|
1395 | res = [self.indexfile] | |
1394 | if not self._inline: |
|
1396 | if not self._inline: | |
1395 | res.append(self.datafile) |
|
1397 | res.append(self.datafile) | |
1396 | return res |
|
1398 | return res |
General Comments 0
You need to be logged in to leave comments.
Login now