Show More
@@ -1,1744 +1,1844 b'' | |||
|
1 | 1 | # hg.py - repository classes for mercurial |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2005 Matt Mackall <mpm@selenic.com> |
|
4 | 4 | # |
|
5 | 5 | # This software may be used and distributed according to the terms |
|
6 | 6 | # of the GNU General Public License, incorporated herein by reference. |
|
7 | 7 | |
|
8 | 8 | import sys, struct, os |
|
9 | 9 | import util |
|
10 | 10 | from revlog import * |
|
11 | 11 | from demandload import * |
|
12 | 12 | demandload(globals(), "re lock urllib urllib2 transaction time socket") |
|
13 | 13 | demandload(globals(), "tempfile httprangereader bdiff") |
|
14 | demandload(globals(), "bisect") | |
|
14 | 15 | |
|
15 | 16 | class filelog(revlog): |
|
16 | 17 | def __init__(self, opener, path): |
|
17 | 18 | revlog.__init__(self, opener, |
|
18 | 19 | os.path.join("data", path + ".i"), |
|
19 | 20 | os.path.join("data", path + ".d")) |
|
20 | 21 | |
|
21 | 22 | def read(self, node): |
|
22 | 23 | t = self.revision(node) |
|
23 | 24 | if t[:2] != '\1\n': |
|
24 | 25 | return t |
|
25 | 26 | s = t.find('\1\n', 2) |
|
26 | 27 | return t[s+2:] |
|
27 | 28 | |
|
28 | 29 | def readmeta(self, node): |
|
29 | 30 | t = self.revision(node) |
|
30 | 31 | if t[:2] != '\1\n': |
|
31 | 32 | return t |
|
32 | 33 | s = t.find('\1\n', 2) |
|
33 | 34 | mt = t[2:s] |
|
34 | 35 | for l in mt.splitlines(): |
|
35 | 36 | k, v = l.split(": ", 1) |
|
36 | 37 | m[k] = v |
|
37 | 38 | return m |
|
38 | 39 | |
|
39 | 40 | def add(self, text, meta, transaction, link, p1=None, p2=None): |
|
40 | 41 | if meta or text[:2] == '\1\n': |
|
41 | 42 | mt = "" |
|
42 | 43 | if meta: |
|
43 | 44 | mt = [ "%s: %s\n" % (k, v) for k,v in meta.items() ] |
|
44 | 45 | text = "\1\n" + "".join(mt) + "\1\n" + text |
|
45 | 46 | return self.addrevision(text, transaction, link, p1, p2) |
|
46 | 47 | |
|
47 | 48 | def annotate(self, node): |
|
48 | 49 | |
|
49 | 50 | def decorate(text, rev): |
|
50 | 51 | return ([rev] * len(text.splitlines()), text) |
|
51 | 52 | |
|
52 | 53 | def pair(parent, child): |
|
53 | 54 | for a1, a2, b1, b2 in bdiff.blocks(parent[1], child[1]): |
|
54 | 55 | child[0][b1:b2] = parent[0][a1:a2] |
|
55 | 56 | return child |
|
56 | 57 | |
|
57 | 58 | # find all ancestors |
|
58 | 59 | needed = {node:1} |
|
59 | 60 | visit = [node] |
|
60 | 61 | while visit: |
|
61 | 62 | n = visit.pop(0) |
|
62 | 63 | for p in self.parents(n): |
|
63 | 64 | if p not in needed: |
|
64 | 65 | needed[p] = 1 |
|
65 | 66 | visit.append(p) |
|
66 | 67 | else: |
|
67 | 68 | # count how many times we'll use this |
|
68 | 69 | needed[p] += 1 |
|
69 | 70 | |
|
70 | 71 | # sort by revision which is a topological order |
|
71 | 72 | visit = [ (self.rev(n), n) for n in needed.keys() ] |
|
72 | 73 | visit.sort() |
|
73 | 74 | hist = {} |
|
74 | 75 | |
|
75 | 76 | for r,n in visit: |
|
76 | 77 | curr = decorate(self.read(n), self.linkrev(n)) |
|
77 | 78 | for p in self.parents(n): |
|
78 | 79 | if p != nullid: |
|
79 | 80 | curr = pair(hist[p], curr) |
|
80 | 81 | # trim the history of unneeded revs |
|
81 | 82 | needed[p] -= 1 |
|
82 | 83 | if not needed[p]: |
|
83 | 84 | del hist[p] |
|
84 | 85 | hist[n] = curr |
|
85 | 86 | |
|
86 | 87 | return zip(hist[n][0], hist[n][1].splitlines(1)) |
|
87 | 88 | |
|
88 | 89 | class manifest(revlog): |
|
89 | 90 | def __init__(self, opener): |
|
90 | 91 | self.mapcache = None |
|
91 | 92 | self.listcache = None |
|
92 | 93 | self.addlist = None |
|
93 | 94 | revlog.__init__(self, opener, "00manifest.i", "00manifest.d") |
|
94 | 95 | |
|
95 | 96 | def read(self, node): |
|
96 | 97 | if node == nullid: return {} # don't upset local cache |
|
97 | 98 | if self.mapcache and self.mapcache[0] == node: |
|
98 | 99 | return self.mapcache[1] |
|
99 | 100 | text = self.revision(node) |
|
100 | 101 | map = {} |
|
101 | 102 | flag = {} |
|
102 | 103 | self.listcache = (text, text.splitlines(1)) |
|
103 | 104 | for l in self.listcache[1]: |
|
104 | 105 | (f, n) = l.split('\0') |
|
105 | 106 | map[f] = bin(n[:40]) |
|
106 | 107 | flag[f] = (n[40:-1] == "x") |
|
107 | 108 | self.mapcache = (node, map, flag) |
|
108 | 109 | return map |
|
109 | 110 | |
|
110 | 111 | def readflags(self, node): |
|
111 | 112 | if node == nullid: return {} # don't upset local cache |
|
112 | 113 | if not self.mapcache or self.mapcache[0] != node: |
|
113 | 114 | self.read(node) |
|
114 | 115 | return self.mapcache[2] |
|
115 | 116 | |
|
116 | 117 | def diff(self, a, b): |
|
117 | 118 | # this is sneaky, as we're not actually using a and b |
|
118 | 119 | if self.listcache and self.addlist and self.listcache[0] == a: |
|
119 | 120 | d = mdiff.diff(self.listcache[1], self.addlist, 1) |
|
120 | 121 | if mdiff.patch(a, d) != b: |
|
121 | 122 | sys.stderr.write("*** sortdiff failed, falling back ***\n") |
|
122 | 123 | return mdiff.textdiff(a, b) |
|
123 | 124 | return d |
|
124 | 125 | else: |
|
125 | 126 | return mdiff.textdiff(a, b) |
|
126 | 127 | |
|
127 | def add(self, map, flags, transaction, link, p1=None, p2=None): | |
|
128 | files = map.keys() | |
|
129 | files.sort() | |
|
128 | def add(self, map, flags, transaction, link, p1=None, p2=None,changed=None): | |
|
129 | # directly generate the mdiff delta from the data collected during | |
|
130 | # the bisect loop below | |
|
131 | def gendelta(delta): | |
|
132 | i = 0 | |
|
133 | result = [] | |
|
134 | while i < len(delta): | |
|
135 | start = delta[i][2] | |
|
136 | end = delta[i][3] | |
|
137 | l = delta[i][4] | |
|
138 | if l == None: | |
|
139 | l = "" | |
|
140 | while i < len(delta) - 1 and start <= delta[i+1][2] and end >= delta[i+1][2]: | |
|
141 | if delta[i+1][3] > end: | |
|
142 | end = delta[i+1][3] | |
|
143 | if delta[i+1][4]: | |
|
144 | l += delta[i+1][4] | |
|
145 | i += 1 | |
|
146 | result.append(struct.pack(">lll", start, end, len(l)) + l) | |
|
147 | i += 1 | |
|
148 | return result | |
|
149 | ||
|
150 | # apply the changes collected during the bisect loop to our addlist | |
|
151 | def addlistdelta(addlist, delta): | |
|
152 | # apply the deltas to the addlist. start from the bottom up | |
|
153 | # so changes to the offsets don't mess things up. | |
|
154 | i = len(delta) | |
|
155 | while i > 0: | |
|
156 | i -= 1 | |
|
157 | start = delta[i][0] | |
|
158 | end = delta[i][1] | |
|
159 | if delta[i][4]: | |
|
160 | addlist[start:end] = [delta[i][4]] | |
|
161 | else: | |
|
162 | del addlist[start:end] | |
|
163 | return addlist | |
|
164 | ||
|
165 | # calculate the byte offset of the start of each line in the | |
|
166 | # manifest | |
|
167 | def calcoffsets(addlist): | |
|
168 | offsets = [0] * (len(addlist) + 1) | |
|
169 | offset = 0 | |
|
170 | i = 0 | |
|
171 | while i < len(addlist): | |
|
172 | offsets[i] = offset | |
|
173 | offset += len(addlist[i]) | |
|
174 | i += 1 | |
|
175 | offsets[i] = offset | |
|
176 | return offsets | |
|
130 | 177 | |
|
131 | self.addlist = ["%s\000%s%s\n" % | |
|
132 | (f, hex(map[f]), flags[f] and "x" or '') | |
|
133 | for f in files] | |
|
178 | # if we're using the listcache, make sure it is valid and | |
|
179 | # parented by the same node we're diffing against | |
|
180 | if not changed or not self.listcache or not p1 or self.mapcache[0] != p1: | |
|
181 | files = map.keys() | |
|
182 | files.sort() | |
|
183 | ||
|
184 | self.addlist = ["%s\000%s%s\n" % | |
|
185 | (f, hex(map[f]), flags[f] and "x" or '') | |
|
186 | for f in files] | |
|
187 | cachedelta = None | |
|
188 | else: | |
|
189 | addlist = self.listcache[1] | |
|
190 | ||
|
191 | # find the starting offset for each line in the add list | |
|
192 | offsets = calcoffsets(addlist) | |
|
193 | ||
|
194 | # combine the changed lists into one list for sorting | |
|
195 | work = [[x, 0] for x in changed[0]] | |
|
196 | work[len(work):] = [[x, 1] for x in changed[1]] | |
|
197 | work.sort() | |
|
198 | ||
|
199 | delta = [] | |
|
200 | bs = 0 | |
|
201 | ||
|
202 | for w in work: | |
|
203 | f = w[0] | |
|
204 | # bs will either be the index of the item or the insertion point | |
|
205 | bs = bisect.bisect(addlist, f, bs) | |
|
206 | if bs < len(addlist): | |
|
207 | fn = addlist[bs][:addlist[bs].index('\0')] | |
|
208 | else: | |
|
209 | fn = None | |
|
210 | if w[1] == 0: | |
|
211 | l = "%s\000%s%s\n" % (f, hex(map[f]), flags[f] and "x" or '') | |
|
212 | else: | |
|
213 | l = None | |
|
214 | start = bs | |
|
215 | if fn != f: | |
|
216 | # item not found, insert a new one | |
|
217 | end = bs | |
|
218 | if w[1] == 1: | |
|
219 | sys.stderr.write("failed to remove %s from manifest" % f) | |
|
220 | sys.exit(1) | |
|
221 | else: | |
|
222 | # item is found, replace/delete the existing line | |
|
223 | end = bs + 1 | |
|
224 | delta.append([start, end, offsets[start], offsets[end], l]) | |
|
225 | ||
|
226 | self.addlist = addlistdelta(addlist, delta) | |
|
227 | if self.mapcache[0] == self.tip(): | |
|
228 | cachedelta = "".join(gendelta(delta)) | |
|
229 | else: | |
|
230 | cachedelta = None | |
|
231 | ||
|
134 | 232 | text = "".join(self.addlist) |
|
135 | ||
|
136 | n = self.addrevision(text, transaction, link, p1, p2) | |
|
233 | if cachedelta and mdiff.patch(self.listcache[0], cachedelta) != text: | |
|
234 | sys.stderr.write("manifest delta failure") | |
|
235 | sys.exit(1) | |
|
236 | n = self.addrevision(text, transaction, link, p1, p2, cachedelta) | |
|
137 | 237 | self.mapcache = (n, map, flags) |
|
138 | 238 | self.listcache = (text, self.addlist) |
|
139 | 239 | self.addlist = None |
|
140 | 240 | |
|
141 | 241 | return n |
|
142 | 242 | |
|
143 | 243 | class changelog(revlog): |
|
144 | 244 | def __init__(self, opener): |
|
145 | 245 | revlog.__init__(self, opener, "00changelog.i", "00changelog.d") |
|
146 | 246 | |
|
147 | 247 | def extract(self, text): |
|
148 | 248 | if not text: |
|
149 | 249 | return (nullid, "", "0", [], "") |
|
150 | 250 | last = text.index("\n\n") |
|
151 | 251 | desc = text[last + 2:] |
|
152 | 252 | l = text[:last].splitlines() |
|
153 | 253 | manifest = bin(l[0]) |
|
154 | 254 | user = l[1] |
|
155 | 255 | date = l[2] |
|
156 | 256 | files = l[3:] |
|
157 | 257 | return (manifest, user, date, files, desc) |
|
158 | 258 | |
|
159 | 259 | def read(self, node): |
|
160 | 260 | return self.extract(self.revision(node)) |
|
161 | 261 | |
|
162 | 262 | def add(self, manifest, list, desc, transaction, p1=None, p2=None, |
|
163 | 263 | user=None, date=None): |
|
164 | 264 | date = date or "%d %d" % (time.time(), time.timezone) |
|
165 | 265 | list.sort() |
|
166 | 266 | l = [hex(manifest), user, date] + list + ["", desc] |
|
167 | 267 | text = "\n".join(l) |
|
168 | 268 | return self.addrevision(text, transaction, self.count(), p1, p2) |
|
169 | 269 | |
|
170 | 270 | class dirstate: |
|
171 | 271 | def __init__(self, opener, ui, root): |
|
172 | 272 | self.opener = opener |
|
173 | 273 | self.root = root |
|
174 | 274 | self.dirty = 0 |
|
175 | 275 | self.ui = ui |
|
176 | 276 | self.map = None |
|
177 | 277 | self.pl = None |
|
178 | 278 | self.copies = {} |
|
179 | 279 | |
|
180 | 280 | def __del__(self): |
|
181 | 281 | if self.dirty: |
|
182 | 282 | self.write() |
|
183 | 283 | |
|
184 | 284 | def __getitem__(self, key): |
|
185 | 285 | try: |
|
186 | 286 | return self.map[key] |
|
187 | 287 | except TypeError: |
|
188 | 288 | self.read() |
|
189 | 289 | return self[key] |
|
190 | 290 | |
|
191 | 291 | def __contains__(self, key): |
|
192 | 292 | if not self.map: self.read() |
|
193 | 293 | return key in self.map |
|
194 | 294 | |
|
195 | 295 | def parents(self): |
|
196 | 296 | if not self.pl: |
|
197 | 297 | self.read() |
|
198 | 298 | return self.pl |
|
199 | 299 | |
|
200 | 300 | def setparents(self, p1, p2 = nullid): |
|
201 | 301 | self.dirty = 1 |
|
202 | 302 | self.pl = p1, p2 |
|
203 | 303 | |
|
204 | 304 | def state(self, key): |
|
205 | 305 | try: |
|
206 | 306 | return self[key][0] |
|
207 | 307 | except KeyError: |
|
208 | 308 | return "?" |
|
209 | 309 | |
|
210 | 310 | def read(self): |
|
211 | 311 | if self.map is not None: return self.map |
|
212 | 312 | |
|
213 | 313 | self.map = {} |
|
214 | 314 | self.pl = [nullid, nullid] |
|
215 | 315 | try: |
|
216 | 316 | st = self.opener("dirstate").read() |
|
217 | 317 | if not st: return |
|
218 | 318 | except: return |
|
219 | 319 | |
|
220 | 320 | self.pl = [st[:20], st[20: 40]] |
|
221 | 321 | |
|
222 | 322 | pos = 40 |
|
223 | 323 | while pos < len(st): |
|
224 | 324 | e = struct.unpack(">cllll", st[pos:pos+17]) |
|
225 | 325 | l = e[4] |
|
226 | 326 | pos += 17 |
|
227 | 327 | f = st[pos:pos + l] |
|
228 | 328 | if '\0' in f: |
|
229 | 329 | f, c = f.split('\0') |
|
230 | 330 | self.copies[f] = c |
|
231 | 331 | self.map[f] = e[:4] |
|
232 | 332 | pos += l |
|
233 | 333 | |
|
234 | 334 | def copy(self, source, dest): |
|
235 | 335 | self.read() |
|
236 | 336 | self.dirty = 1 |
|
237 | 337 | self.copies[dest] = source |
|
238 | 338 | |
|
239 | 339 | def copied(self, file): |
|
240 | 340 | return self.copies.get(file, None) |
|
241 | 341 | |
|
242 | 342 | def update(self, files, state): |
|
243 | 343 | ''' current states: |
|
244 | 344 | n normal |
|
245 | 345 | m needs merging |
|
246 | 346 | r marked for removal |
|
247 | 347 | a marked for addition''' |
|
248 | 348 | |
|
249 | 349 | if not files: return |
|
250 | 350 | self.read() |
|
251 | 351 | self.dirty = 1 |
|
252 | 352 | for f in files: |
|
253 | 353 | if state == "r": |
|
254 | 354 | self.map[f] = ('r', 0, 0, 0) |
|
255 | 355 | else: |
|
256 | 356 | s = os.stat(os.path.join(self.root, f)) |
|
257 | 357 | self.map[f] = (state, s.st_mode, s.st_size, s.st_mtime) |
|
258 | 358 | |
|
259 | 359 | def forget(self, files): |
|
260 | 360 | if not files: return |
|
261 | 361 | self.read() |
|
262 | 362 | self.dirty = 1 |
|
263 | 363 | for f in files: |
|
264 | 364 | try: |
|
265 | 365 | del self.map[f] |
|
266 | 366 | except KeyError: |
|
267 | 367 | self.ui.warn("not in dirstate: %s!\n" % f) |
|
268 | 368 | pass |
|
269 | 369 | |
|
270 | 370 | def clear(self): |
|
271 | 371 | self.map = {} |
|
272 | 372 | self.dirty = 1 |
|
273 | 373 | |
|
274 | 374 | def write(self): |
|
275 | 375 | st = self.opener("dirstate", "w") |
|
276 | 376 | st.write("".join(self.pl)) |
|
277 | 377 | for f, e in self.map.items(): |
|
278 | 378 | c = self.copied(f) |
|
279 | 379 | if c: |
|
280 | 380 | f = f + "\0" + c |
|
281 | 381 | e = struct.pack(">cllll", e[0], e[1], e[2], e[3], len(f)) |
|
282 | 382 | st.write(e + f) |
|
283 | 383 | self.dirty = 0 |
|
284 | 384 | |
|
285 | 385 | def changes(self, files, ignore): |
|
286 | 386 | self.read() |
|
287 | 387 | dc = self.map.copy() |
|
288 | 388 | lookup, changed, added, unknown = [], [], [], [] |
|
289 | 389 | |
|
290 | 390 | # compare all files by default |
|
291 | 391 | if not files: files = [self.root] |
|
292 | 392 | |
|
293 | 393 | # recursive generator of all files listed |
|
294 | 394 | def walk(files): |
|
295 | 395 | for f in util.unique(files): |
|
296 | 396 | f = os.path.join(self.root, f) |
|
297 | 397 | if os.path.isdir(f): |
|
298 | 398 | for dir, subdirs, fl in os.walk(f): |
|
299 | 399 | d = dir[len(self.root) + 1:] |
|
300 | 400 | if ".hg" in subdirs: subdirs.remove(".hg") |
|
301 | 401 | for fn in fl: |
|
302 | 402 | fn = util.pconvert(os.path.join(d, fn)) |
|
303 | 403 | yield fn |
|
304 | 404 | else: |
|
305 | 405 | yield f[len(self.root) + 1:] |
|
306 | 406 | |
|
307 | 407 | for fn in util.unique(walk(files)): |
|
308 | 408 | try: s = os.stat(os.path.join(self.root, fn)) |
|
309 | 409 | except: continue |
|
310 | 410 | |
|
311 | 411 | if fn in dc: |
|
312 | 412 | c = dc[fn] |
|
313 | 413 | del dc[fn] |
|
314 | 414 | |
|
315 | 415 | if c[0] == 'm': |
|
316 | 416 | changed.append(fn) |
|
317 | 417 | elif c[0] == 'a': |
|
318 | 418 | added.append(fn) |
|
319 | 419 | elif c[0] == 'r': |
|
320 | 420 | unknown.append(fn) |
|
321 | 421 | elif c[2] != s.st_size or (c[1] ^ s.st_mode) & 0100: |
|
322 | 422 | changed.append(fn) |
|
323 | 423 | elif c[1] != s.st_mode or c[3] != s.st_mtime: |
|
324 | 424 | lookup.append(fn) |
|
325 | 425 | else: |
|
326 | 426 | if not ignore(fn): unknown.append(fn) |
|
327 | 427 | |
|
328 | 428 | return (lookup, changed, added, dc.keys(), unknown) |
|
329 | 429 | |
|
330 | 430 | # used to avoid circular references so destructors work |
|
331 | 431 | def opener(base): |
|
332 | 432 | p = base |
|
333 | 433 | def o(path, mode="r"): |
|
334 | 434 | if p[:7] == "http://": |
|
335 | 435 | f = os.path.join(p, urllib.quote(path)) |
|
336 | 436 | return httprangereader.httprangereader(f) |
|
337 | 437 | |
|
338 | 438 | f = os.path.join(p, path) |
|
339 | 439 | |
|
340 | 440 | mode += "b" # for that other OS |
|
341 | 441 | |
|
342 | 442 | if mode[0] != "r": |
|
343 | 443 | try: |
|
344 | 444 | s = os.stat(f) |
|
345 | 445 | except OSError: |
|
346 | 446 | d = os.path.dirname(f) |
|
347 | 447 | if not os.path.isdir(d): |
|
348 | 448 | os.makedirs(d) |
|
349 | 449 | else: |
|
350 | 450 | if s.st_nlink > 1: |
|
351 | 451 | file(f + ".tmp", "wb").write(file(f, "rb").read()) |
|
352 | 452 | util.rename(f+".tmp", f) |
|
353 | 453 | |
|
354 | 454 | return file(f, mode) |
|
355 | 455 | |
|
356 | 456 | return o |
|
357 | 457 | |
|
358 | 458 | class RepoError(Exception): pass |
|
359 | 459 | |
|
360 | 460 | class localrepository: |
|
361 | 461 | def __init__(self, ui, path=None, create=0): |
|
362 | 462 | self.remote = 0 |
|
363 | 463 | if path and path[:7] == "http://": |
|
364 | 464 | self.remote = 1 |
|
365 | 465 | self.path = path |
|
366 | 466 | else: |
|
367 | 467 | if not path: |
|
368 | 468 | p = os.getcwd() |
|
369 | 469 | while not os.path.isdir(os.path.join(p, ".hg")): |
|
370 | 470 | oldp = p |
|
371 | 471 | p = os.path.dirname(p) |
|
372 | 472 | if p == oldp: raise RepoError("no repo found") |
|
373 | 473 | path = p |
|
374 | 474 | self.path = os.path.join(path, ".hg") |
|
375 | 475 | |
|
376 | 476 | if not create and not os.path.isdir(self.path): |
|
377 | 477 | raise RepoError("repository %s not found" % self.path) |
|
378 | 478 | |
|
379 | 479 | self.root = path |
|
380 | 480 | self.ui = ui |
|
381 | 481 | |
|
382 | 482 | if create: |
|
383 | 483 | os.mkdir(self.path) |
|
384 | 484 | os.mkdir(self.join("data")) |
|
385 | 485 | |
|
386 | 486 | self.opener = opener(self.path) |
|
387 | 487 | self.wopener = opener(self.root) |
|
388 | 488 | self.manifest = manifest(self.opener) |
|
389 | 489 | self.changelog = changelog(self.opener) |
|
390 | 490 | self.ignorelist = None |
|
391 | 491 | self.tagscache = None |
|
392 | 492 | self.nodetagscache = None |
|
393 | 493 | |
|
394 | 494 | if not self.remote: |
|
395 | 495 | self.dirstate = dirstate(self.opener, ui, self.root) |
|
396 | 496 | try: |
|
397 | 497 | self.ui.readconfig(self.opener("hgrc")) |
|
398 | 498 | except IOError: pass |
|
399 | 499 | |
|
400 | 500 | def ignore(self, f): |
|
401 | 501 | if self.ignorelist is None: |
|
402 | 502 | self.ignorelist = [] |
|
403 | 503 | try: |
|
404 | 504 | l = file(self.wjoin(".hgignore")) |
|
405 | 505 | for pat in l: |
|
406 | 506 | if pat != "\n": |
|
407 | 507 | self.ignorelist.append(re.compile(util.pconvert(pat[:-1]))) |
|
408 | 508 | except IOError: pass |
|
409 | 509 | for pat in self.ignorelist: |
|
410 | 510 | if pat.search(f): return True |
|
411 | 511 | return False |
|
412 | 512 | |
|
413 | 513 | def hook(self, name, **args): |
|
414 | 514 | s = self.ui.config("hooks", name) |
|
415 | 515 | if s: |
|
416 | 516 | self.ui.note("running hook %s: %s\n" % (name, s)) |
|
417 | 517 | old = {} |
|
418 | 518 | for k, v in args.items(): |
|
419 | 519 | k = k.upper() |
|
420 | 520 | old[k] = os.environ.get(k, None) |
|
421 | 521 | os.environ[k] = v |
|
422 | 522 | |
|
423 | 523 | r = os.system(s) |
|
424 | 524 | |
|
425 | 525 | for k, v in old.items(): |
|
426 | 526 | if v != None: |
|
427 | 527 | os.environ[k] = v |
|
428 | 528 | else: |
|
429 | 529 | del os.environ[k] |
|
430 | 530 | |
|
431 | 531 | if r: |
|
432 | 532 | self.ui.warn("abort: %s hook failed with status %d!\n" % |
|
433 | 533 | (name, r)) |
|
434 | 534 | return False |
|
435 | 535 | return True |
|
436 | 536 | |
|
437 | 537 | def tags(self): |
|
438 | 538 | '''return a mapping of tag to node''' |
|
439 | 539 | if not self.tagscache: |
|
440 | 540 | self.tagscache = {} |
|
441 | 541 | def addtag(self, k, n): |
|
442 | 542 | try: |
|
443 | 543 | bin_n = bin(n) |
|
444 | 544 | except TypeError: |
|
445 | 545 | bin_n = '' |
|
446 | 546 | self.tagscache[k.strip()] = bin_n |
|
447 | 547 | |
|
448 | 548 | try: |
|
449 | 549 | # read each head of the tags file, ending with the tip |
|
450 | 550 | # and add each tag found to the map, with "newer" ones |
|
451 | 551 | # taking precedence |
|
452 | 552 | fl = self.file(".hgtags") |
|
453 | 553 | h = fl.heads() |
|
454 | 554 | h.reverse() |
|
455 | 555 | for r in h: |
|
456 | 556 | for l in fl.revision(r).splitlines(): |
|
457 | 557 | if l: |
|
458 | 558 | n, k = l.split(" ", 1) |
|
459 | 559 | addtag(self, k, n) |
|
460 | 560 | except KeyError: |
|
461 | 561 | pass |
|
462 | 562 | |
|
463 | 563 | try: |
|
464 | 564 | f = self.opener("localtags") |
|
465 | 565 | for l in f: |
|
466 | 566 | n, k = l.split(" ", 1) |
|
467 | 567 | addtag(self, k, n) |
|
468 | 568 | except IOError: |
|
469 | 569 | pass |
|
470 | 570 | |
|
471 | 571 | self.tagscache['tip'] = self.changelog.tip() |
|
472 | 572 | |
|
473 | 573 | return self.tagscache |
|
474 | 574 | |
|
475 | 575 | def tagslist(self): |
|
476 | 576 | '''return a list of tags ordered by revision''' |
|
477 | 577 | l = [] |
|
478 | 578 | for t, n in self.tags().items(): |
|
479 | 579 | try: |
|
480 | 580 | r = self.changelog.rev(n) |
|
481 | 581 | except: |
|
482 | 582 | r = -2 # sort to the beginning of the list if unknown |
|
483 | 583 | l.append((r,t,n)) |
|
484 | 584 | l.sort() |
|
485 | 585 | return [(t,n) for r,t,n in l] |
|
486 | 586 | |
|
487 | 587 | def nodetags(self, node): |
|
488 | 588 | '''return the tags associated with a node''' |
|
489 | 589 | if not self.nodetagscache: |
|
490 | 590 | self.nodetagscache = {} |
|
491 | 591 | for t,n in self.tags().items(): |
|
492 | 592 | self.nodetagscache.setdefault(n,[]).append(t) |
|
493 | 593 | return self.nodetagscache.get(node, []) |
|
494 | 594 | |
|
495 | 595 | def lookup(self, key): |
|
496 | 596 | try: |
|
497 | 597 | return self.tags()[key] |
|
498 | 598 | except KeyError: |
|
499 | 599 | return self.changelog.lookup(key) |
|
500 | 600 | |
|
501 | 601 | def dev(self): |
|
502 | 602 | if self.remote: return -1 |
|
503 | 603 | return os.stat(self.path).st_dev |
|
504 | 604 | |
|
505 | 605 | def join(self, f): |
|
506 | 606 | return os.path.join(self.path, f) |
|
507 | 607 | |
|
508 | 608 | def wjoin(self, f): |
|
509 | 609 | return os.path.join(self.root, f) |
|
510 | 610 | |
|
511 | 611 | def file(self, f): |
|
512 | 612 | if f[0] == '/': f = f[1:] |
|
513 | 613 | return filelog(self.opener, f) |
|
514 | 614 | |
|
515 | 615 | def getcwd(self): |
|
516 | 616 | cwd = os.getcwd() |
|
517 | 617 | if cwd == self.root: return '' |
|
518 | 618 | return cwd[len(self.root) + 1:] |
|
519 | 619 | |
|
520 | 620 | def wfile(self, f, mode='r'): |
|
521 | 621 | return self.wopener(f, mode) |
|
522 | 622 | |
|
523 | 623 | def transaction(self): |
|
524 | 624 | # save dirstate for undo |
|
525 | 625 | try: |
|
526 | 626 | ds = self.opener("dirstate").read() |
|
527 | 627 | except IOError: |
|
528 | 628 | ds = "" |
|
529 | 629 | self.opener("undo.dirstate", "w").write(ds) |
|
530 | 630 | |
|
531 | 631 | return transaction.transaction(self.ui.warn, |
|
532 | 632 | self.opener, self.join("journal"), |
|
533 | 633 | self.join("undo")) |
|
534 | 634 | |
|
535 | 635 | def recover(self): |
|
536 | 636 | lock = self.lock() |
|
537 | 637 | if os.path.exists(self.join("journal")): |
|
538 | 638 | self.ui.status("rolling back interrupted transaction\n") |
|
539 | 639 | return transaction.rollback(self.opener, self.join("journal")) |
|
540 | 640 | else: |
|
541 | 641 | self.ui.warn("no interrupted transaction available\n") |
|
542 | 642 | |
|
543 | 643 | def undo(self): |
|
544 | 644 | lock = self.lock() |
|
545 | 645 | if os.path.exists(self.join("undo")): |
|
546 | 646 | self.ui.status("rolling back last transaction\n") |
|
547 | 647 | transaction.rollback(self.opener, self.join("undo")) |
|
548 | 648 | self.dirstate = None |
|
549 | 649 | util.rename(self.join("undo.dirstate"), self.join("dirstate")) |
|
550 | 650 | self.dirstate = dirstate(self.opener, self.ui, self.root) |
|
551 | 651 | else: |
|
552 | 652 | self.ui.warn("no undo information available\n") |
|
553 | 653 | |
|
554 | 654 | def lock(self, wait = 1): |
|
555 | 655 | try: |
|
556 | 656 | return lock.lock(self.join("lock"), 0) |
|
557 | 657 | except lock.LockHeld, inst: |
|
558 | 658 | if wait: |
|
559 | 659 | self.ui.warn("waiting for lock held by %s\n" % inst.args[0]) |
|
560 | 660 | return lock.lock(self.join("lock"), wait) |
|
561 | 661 | raise inst |
|
562 | 662 | |
|
563 | 663 | def rawcommit(self, files, text, user, date, p1=None, p2=None): |
|
564 | 664 | orig_parent = self.dirstate.parents()[0] or nullid |
|
565 | 665 | p1 = p1 or self.dirstate.parents()[0] or nullid |
|
566 | 666 | p2 = p2 or self.dirstate.parents()[1] or nullid |
|
567 | 667 | c1 = self.changelog.read(p1) |
|
568 | 668 | c2 = self.changelog.read(p2) |
|
569 | 669 | m1 = self.manifest.read(c1[0]) |
|
570 | 670 | mf1 = self.manifest.readflags(c1[0]) |
|
571 | 671 | m2 = self.manifest.read(c2[0]) |
|
572 | 672 | |
|
573 | 673 | if orig_parent == p1: |
|
574 | 674 | update_dirstate = 1 |
|
575 | 675 | else: |
|
576 | 676 | update_dirstate = 0 |
|
577 | 677 | |
|
578 | 678 | tr = self.transaction() |
|
579 | 679 | mm = m1.copy() |
|
580 | 680 | mfm = mf1.copy() |
|
581 | 681 | linkrev = self.changelog.count() |
|
582 | 682 | for f in files: |
|
583 | 683 | try: |
|
584 | 684 | t = self.wfile(f).read() |
|
585 | 685 | tm = util.is_exec(self.wjoin(f), mfm.get(f, False)) |
|
586 | 686 | r = self.file(f) |
|
587 | 687 | mfm[f] = tm |
|
588 | 688 | mm[f] = r.add(t, {}, tr, linkrev, |
|
589 | 689 | m1.get(f, nullid), m2.get(f, nullid)) |
|
590 | 690 | if update_dirstate: |
|
591 | 691 | self.dirstate.update([f], "n") |
|
592 | 692 | except IOError: |
|
593 | 693 | try: |
|
594 | 694 | del mm[f] |
|
595 | 695 | del mfm[f] |
|
596 | 696 | if update_dirstate: |
|
597 | 697 | self.dirstate.forget([f]) |
|
598 | 698 | except: |
|
599 | 699 | # deleted from p2? |
|
600 | 700 | pass |
|
601 | 701 | |
|
602 | 702 | mnode = self.manifest.add(mm, mfm, tr, linkrev, c1[0], c2[0]) |
|
603 | 703 | user = user or self.ui.username() |
|
604 | 704 | n = self.changelog.add(mnode, files, text, tr, p1, p2, user, date) |
|
605 | 705 | tr.close() |
|
606 | 706 | if update_dirstate: |
|
607 | 707 | self.dirstate.setparents(n, nullid) |
|
608 | 708 | |
|
609 | 709 | def commit(self, files = None, text = "", user = None, date = None): |
|
610 | 710 | commit = [] |
|
611 | 711 | remove = [] |
|
612 | 712 | if files: |
|
613 | 713 | for f in files: |
|
614 | 714 | s = self.dirstate.state(f) |
|
615 | 715 | if s in 'nmai': |
|
616 | 716 | commit.append(f) |
|
617 | 717 | elif s == 'r': |
|
618 | 718 | remove.append(f) |
|
619 | 719 | else: |
|
620 | 720 | self.ui.warn("%s not tracked!\n" % f) |
|
621 | 721 | else: |
|
622 | 722 | (c, a, d, u) = self.changes(None, None) |
|
623 | 723 | commit = c + a |
|
624 | 724 | remove = d |
|
625 | 725 | |
|
626 | 726 | if not commit and not remove: |
|
627 | 727 | self.ui.status("nothing changed\n") |
|
628 | 728 | return |
|
629 | 729 | |
|
630 | 730 | if not self.hook("precommit"): |
|
631 | 731 | return 1 |
|
632 | 732 | |
|
633 | 733 | p1, p2 = self.dirstate.parents() |
|
634 | 734 | c1 = self.changelog.read(p1) |
|
635 | 735 | c2 = self.changelog.read(p2) |
|
636 | 736 | m1 = self.manifest.read(c1[0]) |
|
637 | 737 | mf1 = self.manifest.readflags(c1[0]) |
|
638 | 738 | m2 = self.manifest.read(c2[0]) |
|
639 | 739 | lock = self.lock() |
|
640 | 740 | tr = self.transaction() |
|
641 | 741 | |
|
642 | 742 | # check in files |
|
643 | 743 | new = {} |
|
644 | 744 | linkrev = self.changelog.count() |
|
645 | 745 | commit.sort() |
|
646 | 746 | for f in commit: |
|
647 | 747 | self.ui.note(f + "\n") |
|
648 | 748 | try: |
|
649 | 749 | mf1[f] = util.is_exec(self.wjoin(f), mf1.get(f, False)) |
|
650 | 750 | t = self.wfile(f).read() |
|
651 | 751 | except IOError: |
|
652 | 752 | self.warn("trouble committing %s!\n" % f) |
|
653 | 753 | raise |
|
654 | 754 | |
|
655 | 755 | meta = {} |
|
656 | 756 | cp = self.dirstate.copied(f) |
|
657 | 757 | if cp: |
|
658 | 758 | meta["copy"] = cp |
|
659 | 759 | meta["copyrev"] = hex(m1.get(cp, m2.get(cp, nullid))) |
|
660 | 760 | self.ui.debug(" %s: copy %s:%s\n" % (f, cp, meta["copyrev"])) |
|
661 | 761 | |
|
662 | 762 | r = self.file(f) |
|
663 | 763 | fp1 = m1.get(f, nullid) |
|
664 | 764 | fp2 = m2.get(f, nullid) |
|
665 | 765 | new[f] = r.add(t, meta, tr, linkrev, fp1, fp2) |
|
666 | 766 | |
|
667 | 767 | # update manifest |
|
668 | 768 | m1.update(new) |
|
669 | 769 | for f in remove: |
|
670 | 770 | if f in m1: |
|
671 | 771 | del m1[f] |
|
672 | mn = self.manifest.add(m1, mf1, tr, linkrev, c1[0], c2[0]) | |
|
772 | mn = self.manifest.add(m1, mf1, tr, linkrev, c1[0], c2[0], (new,remove)) | |
|
673 | 773 | |
|
674 | 774 | # add changeset |
|
675 | 775 | new = new.keys() |
|
676 | 776 | new.sort() |
|
677 | 777 | |
|
678 | 778 | if not text: |
|
679 | 779 | edittext = "\n" + "HG: manifest hash %s\n" % hex(mn) |
|
680 | 780 | edittext += "".join(["HG: changed %s\n" % f for f in new]) |
|
681 | 781 | edittext += "".join(["HG: removed %s\n" % f for f in remove]) |
|
682 | 782 | edittext = self.ui.edit(edittext) |
|
683 | 783 | if not edittext.rstrip(): |
|
684 | 784 | return 1 |
|
685 | 785 | text = edittext |
|
686 | 786 | |
|
687 | 787 | user = user or self.ui.username() |
|
688 | 788 | n = self.changelog.add(mn, new, text, tr, p1, p2, user, date) |
|
689 | 789 | |
|
690 | 790 | if not self.hook("commit", node=hex(n)): |
|
691 | 791 | return 1 |
|
692 | 792 | |
|
693 | 793 | tr.close() |
|
694 | 794 | |
|
695 | 795 | self.dirstate.setparents(n) |
|
696 | 796 | self.dirstate.update(new, "n") |
|
697 | 797 | self.dirstate.forget(remove) |
|
698 | 798 | |
|
699 | 799 | def changes(self, node1, node2, files=None): |
|
700 | 800 | mf2, u = None, [] |
|
701 | 801 | |
|
702 | 802 | def fcmp(fn, mf): |
|
703 | 803 | t1 = self.wfile(fn).read() |
|
704 | 804 | t2 = self.file(fn).revision(mf[fn]) |
|
705 | 805 | return cmp(t1, t2) |
|
706 | 806 | |
|
707 | 807 | # are we comparing the working directory? |
|
708 | 808 | if not node2: |
|
709 | 809 | l, c, a, d, u = self.dirstate.changes(files, self.ignore) |
|
710 | 810 | |
|
711 | 811 | # are we comparing working dir against its parent? |
|
712 | 812 | if not node1: |
|
713 | 813 | if l: |
|
714 | 814 | # do a full compare of any files that might have changed |
|
715 | 815 | change = self.changelog.read(self.dirstate.parents()[0]) |
|
716 | 816 | mf2 = self.manifest.read(change[0]) |
|
717 | 817 | for f in l: |
|
718 | 818 | if fcmp(f, mf2): |
|
719 | 819 | c.append(f) |
|
720 | 820 | |
|
721 | 821 | for l in c, a, d, u: |
|
722 | 822 | l.sort() |
|
723 | 823 | |
|
724 | 824 | return (c, a, d, u) |
|
725 | 825 | |
|
726 | 826 | # are we comparing working dir against non-tip? |
|
727 | 827 | # generate a pseudo-manifest for the working dir |
|
728 | 828 | if not node2: |
|
729 | 829 | if not mf2: |
|
730 | 830 | change = self.changelog.read(self.dirstate.parents()[0]) |
|
731 | 831 | mf2 = self.manifest.read(change[0]).copy() |
|
732 | 832 | for f in a + c + l: |
|
733 | 833 | mf2[f] = "" |
|
734 | 834 | for f in d: |
|
735 | 835 | if f in mf2: del mf2[f] |
|
736 | 836 | else: |
|
737 | 837 | change = self.changelog.read(node2) |
|
738 | 838 | mf2 = self.manifest.read(change[0]) |
|
739 | 839 | |
|
740 | 840 | # flush lists from dirstate before comparing manifests |
|
741 | 841 | c, a = [], [] |
|
742 | 842 | |
|
743 | 843 | change = self.changelog.read(node1) |
|
744 | 844 | mf1 = self.manifest.read(change[0]).copy() |
|
745 | 845 | |
|
746 | 846 | for fn in mf2: |
|
747 | 847 | if mf1.has_key(fn): |
|
748 | 848 | if mf1[fn] != mf2[fn]: |
|
749 | 849 | if mf2[fn] != "" or fcmp(fn, mf1): |
|
750 | 850 | c.append(fn) |
|
751 | 851 | del mf1[fn] |
|
752 | 852 | else: |
|
753 | 853 | a.append(fn) |
|
754 | 854 | |
|
755 | 855 | d = mf1.keys() |
|
756 | 856 | |
|
757 | 857 | for l in c, a, d, u: |
|
758 | 858 | l.sort() |
|
759 | 859 | |
|
760 | 860 | return (c, a, d, u) |
|
761 | 861 | |
|
762 | 862 | def add(self, list): |
|
763 | 863 | for f in list: |
|
764 | 864 | p = self.wjoin(f) |
|
765 | 865 | if not os.path.exists(p): |
|
766 | 866 | self.ui.warn("%s does not exist!\n" % f) |
|
767 | 867 | elif not os.path.isfile(p): |
|
768 | 868 | self.ui.warn("%s not added: mercurial only supports files currently\n" % f) |
|
769 | 869 | elif self.dirstate.state(f) == 'n': |
|
770 | 870 | self.ui.warn("%s already tracked!\n" % f) |
|
771 | 871 | else: |
|
772 | 872 | self.dirstate.update([f], "a") |
|
773 | 873 | |
|
774 | 874 | def forget(self, list): |
|
775 | 875 | for f in list: |
|
776 | 876 | if self.dirstate.state(f) not in 'ai': |
|
777 | 877 | self.ui.warn("%s not added!\n" % f) |
|
778 | 878 | else: |
|
779 | 879 | self.dirstate.forget([f]) |
|
780 | 880 | |
|
781 | 881 | def remove(self, list): |
|
782 | 882 | for f in list: |
|
783 | 883 | p = self.wjoin(f) |
|
784 | 884 | if os.path.exists(p): |
|
785 | 885 | self.ui.warn("%s still exists!\n" % f) |
|
786 | 886 | elif self.dirstate.state(f) == 'a': |
|
787 | 887 | self.ui.warn("%s never committed!\n" % f) |
|
788 | 888 | self.dirstate.forget(f) |
|
789 | 889 | elif f not in self.dirstate: |
|
790 | 890 | self.ui.warn("%s not tracked!\n" % f) |
|
791 | 891 | else: |
|
792 | 892 | self.dirstate.update([f], "r") |
|
793 | 893 | |
|
794 | 894 | def copy(self, source, dest): |
|
795 | 895 | p = self.wjoin(dest) |
|
796 | 896 | if not os.path.exists(dest): |
|
797 | 897 | self.ui.warn("%s does not exist!\n" % dest) |
|
798 | 898 | elif not os.path.isfile(dest): |
|
799 | 899 | self.ui.warn("copy failed: %s is not a file\n" % dest) |
|
800 | 900 | else: |
|
801 | 901 | if self.dirstate.state(dest) == '?': |
|
802 | 902 | self.dirstate.update([dest], "a") |
|
803 | 903 | self.dirstate.copy(source, dest) |
|
804 | 904 | |
|
805 | 905 | def heads(self): |
|
806 | 906 | return self.changelog.heads() |
|
807 | 907 | |
|
808 | 908 | def branches(self, nodes): |
|
809 | 909 | if not nodes: nodes = [self.changelog.tip()] |
|
810 | 910 | b = [] |
|
811 | 911 | for n in nodes: |
|
812 | 912 | t = n |
|
813 | 913 | while n: |
|
814 | 914 | p = self.changelog.parents(n) |
|
815 | 915 | if p[1] != nullid or p[0] == nullid: |
|
816 | 916 | b.append((t, n, p[0], p[1])) |
|
817 | 917 | break |
|
818 | 918 | n = p[0] |
|
819 | 919 | return b |
|
820 | 920 | |
|
821 | 921 | def between(self, pairs): |
|
822 | 922 | r = [] |
|
823 | 923 | |
|
824 | 924 | for top, bottom in pairs: |
|
825 | 925 | n, l, i = top, [], 0 |
|
826 | 926 | f = 1 |
|
827 | 927 | |
|
828 | 928 | while n != bottom: |
|
829 | 929 | p = self.changelog.parents(n)[0] |
|
830 | 930 | if i == f: |
|
831 | 931 | l.append(n) |
|
832 | 932 | f = f * 2 |
|
833 | 933 | n = p |
|
834 | 934 | i += 1 |
|
835 | 935 | |
|
836 | 936 | r.append(l) |
|
837 | 937 | |
|
838 | 938 | return r |
|
839 | 939 | |
|
840 | 940 | def newer(self, nodes): |
|
841 | 941 | m = {} |
|
842 | 942 | nl = [] |
|
843 | 943 | pm = {} |
|
844 | 944 | cl = self.changelog |
|
845 | 945 | t = l = cl.count() |
|
846 | 946 | |
|
847 | 947 | # find the lowest numbered node |
|
848 | 948 | for n in nodes: |
|
849 | 949 | l = min(l, cl.rev(n)) |
|
850 | 950 | m[n] = 1 |
|
851 | 951 | |
|
852 | 952 | for i in xrange(l, t): |
|
853 | 953 | n = cl.node(i) |
|
854 | 954 | if n in m: # explicitly listed |
|
855 | 955 | pm[n] = 1 |
|
856 | 956 | nl.append(n) |
|
857 | 957 | continue |
|
858 | 958 | for p in cl.parents(n): |
|
859 | 959 | if p in pm: # parent listed |
|
860 | 960 | pm[n] = 1 |
|
861 | 961 | nl.append(n) |
|
862 | 962 | break |
|
863 | 963 | |
|
864 | 964 | return nl |
|
865 | 965 | |
|
866 | 966 | def findincoming(self, remote, base={}): |
|
867 | 967 | m = self.changelog.nodemap |
|
868 | 968 | search = [] |
|
869 | 969 | fetch = [] |
|
870 | 970 | seen = {} |
|
871 | 971 | seenbranch = {} |
|
872 | 972 | |
|
873 | 973 | # assume we're closer to the tip than the root |
|
874 | 974 | # and start by examining the heads |
|
875 | 975 | self.ui.status("searching for changes\n") |
|
876 | 976 | heads = remote.heads() |
|
877 | 977 | unknown = [] |
|
878 | 978 | for h in heads: |
|
879 | 979 | if h not in m: |
|
880 | 980 | unknown.append(h) |
|
881 | 981 | else: |
|
882 | 982 | base[h] = 1 |
|
883 | 983 | |
|
884 | 984 | if not unknown: |
|
885 | 985 | return None |
|
886 | 986 | |
|
887 | 987 | rep = {} |
|
888 | 988 | reqcnt = 0 |
|
889 | 989 | |
|
890 | 990 | # search through remote branches |
|
891 | 991 | # a 'branch' here is a linear segment of history, with four parts: |
|
892 | 992 | # head, root, first parent, second parent |
|
893 | 993 | # (a branch always has two parents (or none) by definition) |
|
894 | 994 | unknown = remote.branches(unknown) |
|
895 | 995 | while unknown: |
|
896 | 996 | r = [] |
|
897 | 997 | while unknown: |
|
898 | 998 | n = unknown.pop(0) |
|
899 | 999 | if n[0] in seen: |
|
900 | 1000 | continue |
|
901 | 1001 | |
|
902 | 1002 | self.ui.debug("examining %s:%s\n" % (short(n[0]), short(n[1]))) |
|
903 | 1003 | if n[0] == nullid: |
|
904 | 1004 | break |
|
905 | 1005 | if n in seenbranch: |
|
906 | 1006 | self.ui.debug("branch already found\n") |
|
907 | 1007 | continue |
|
908 | 1008 | if n[1] and n[1] in m: # do we know the base? |
|
909 | 1009 | self.ui.debug("found incomplete branch %s:%s\n" |
|
910 | 1010 | % (short(n[0]), short(n[1]))) |
|
911 | 1011 | search.append(n) # schedule branch range for scanning |
|
912 | 1012 | seenbranch[n] = 1 |
|
913 | 1013 | else: |
|
914 | 1014 | if n[1] not in seen and n[1] not in fetch: |
|
915 | 1015 | if n[2] in m and n[3] in m: |
|
916 | 1016 | self.ui.debug("found new changeset %s\n" % |
|
917 | 1017 | short(n[1])) |
|
918 | 1018 | fetch.append(n[1]) # earliest unknown |
|
919 | 1019 | base[n[2]] = 1 # latest known |
|
920 | 1020 | continue |
|
921 | 1021 | |
|
922 | 1022 | for a in n[2:4]: |
|
923 | 1023 | if a not in rep: |
|
924 | 1024 | r.append(a) |
|
925 | 1025 | rep[a] = 1 |
|
926 | 1026 | |
|
927 | 1027 | seen[n[0]] = 1 |
|
928 | 1028 | |
|
929 | 1029 | if r: |
|
930 | 1030 | reqcnt += 1 |
|
931 | 1031 | self.ui.debug("request %d: %s\n" % |
|
932 | 1032 | (reqcnt, " ".join(map(short, r)))) |
|
933 | 1033 | for p in range(0, len(r), 10): |
|
934 | 1034 | for b in remote.branches(r[p:p+10]): |
|
935 | 1035 | self.ui.debug("received %s:%s\n" % |
|
936 | 1036 | (short(b[0]), short(b[1]))) |
|
937 | 1037 | if b[0] not in m and b[0] not in seen: |
|
938 | 1038 | unknown.append(b) |
|
939 | 1039 | |
|
940 | 1040 | # do binary search on the branches we found |
|
941 | 1041 | while search: |
|
942 | 1042 | n = search.pop(0) |
|
943 | 1043 | reqcnt += 1 |
|
944 | 1044 | l = remote.between([(n[0], n[1])])[0] |
|
945 | 1045 | l.append(n[1]) |
|
946 | 1046 | p = n[0] |
|
947 | 1047 | f = 1 |
|
948 | 1048 | for i in l: |
|
949 | 1049 | self.ui.debug("narrowing %d:%d %s\n" % (f, len(l), short(i))) |
|
950 | 1050 | if i in m: |
|
951 | 1051 | if f <= 2: |
|
952 | 1052 | self.ui.debug("found new branch changeset %s\n" % |
|
953 | 1053 | short(p)) |
|
954 | 1054 | fetch.append(p) |
|
955 | 1055 | base[i] = 1 |
|
956 | 1056 | else: |
|
957 | 1057 | self.ui.debug("narrowed branch search to %s:%s\n" |
|
958 | 1058 | % (short(p), short(i))) |
|
959 | 1059 | search.append((p, i)) |
|
960 | 1060 | break |
|
961 | 1061 | p, f = i, f * 2 |
|
962 | 1062 | |
|
963 | 1063 | # sanity check our fetch list |
|
964 | 1064 | for f in fetch: |
|
965 | 1065 | if f in m: |
|
966 | 1066 | raise RepoError("already have changeset " + short(f[:4])) |
|
967 | 1067 | |
|
968 | 1068 | if base.keys() == [nullid]: |
|
969 | 1069 | self.ui.warn("warning: pulling from an unrelated repository!\n") |
|
970 | 1070 | |
|
971 | 1071 | self.ui.note("adding new changesets starting at " + |
|
972 | 1072 | " ".join([short(f) for f in fetch]) + "\n") |
|
973 | 1073 | |
|
974 | 1074 | self.ui.debug("%d total queries\n" % reqcnt) |
|
975 | 1075 | |
|
976 | 1076 | return fetch |
|
977 | 1077 | |
|
978 | 1078 | def findoutgoing(self, remote): |
|
979 | 1079 | base = {} |
|
980 | 1080 | self.findincoming(remote, base) |
|
981 | 1081 | remain = dict.fromkeys(self.changelog.nodemap) |
|
982 | 1082 | |
|
983 | 1083 | # prune everything remote has from the tree |
|
984 | 1084 | del remain[nullid] |
|
985 | 1085 | remove = base.keys() |
|
986 | 1086 | while remove: |
|
987 | 1087 | n = remove.pop(0) |
|
988 | 1088 | if n in remain: |
|
989 | 1089 | del remain[n] |
|
990 | 1090 | for p in self.changelog.parents(n): |
|
991 | 1091 | remove.append(p) |
|
992 | 1092 | |
|
993 | 1093 | # find every node whose parents have been pruned |
|
994 | 1094 | subset = [] |
|
995 | 1095 | for n in remain: |
|
996 | 1096 | p1, p2 = self.changelog.parents(n) |
|
997 | 1097 | if p1 not in remain and p2 not in remain: |
|
998 | 1098 | subset.append(n) |
|
999 | 1099 | |
|
1000 | 1100 | # this is the set of all roots we have to push |
|
1001 | 1101 | return subset |
|
1002 | 1102 | |
|
1003 | 1103 | def pull(self, remote): |
|
1004 | 1104 | lock = self.lock() |
|
1005 | 1105 | |
|
1006 | 1106 | # if we have an empty repo, fetch everything |
|
1007 | 1107 | if self.changelog.tip() == nullid: |
|
1008 | 1108 | self.ui.status("requesting all changes\n") |
|
1009 | 1109 | fetch = [nullid] |
|
1010 | 1110 | else: |
|
1011 | 1111 | fetch = self.findincoming(remote) |
|
1012 | 1112 | |
|
1013 | 1113 | if not fetch: |
|
1014 | 1114 | self.ui.status("no changes found\n") |
|
1015 | 1115 | return 1 |
|
1016 | 1116 | |
|
1017 | 1117 | cg = remote.changegroup(fetch) |
|
1018 | 1118 | return self.addchangegroup(cg) |
|
1019 | 1119 | |
|
1020 | 1120 | def push(self, remote): |
|
1021 | 1121 | lock = remote.lock() |
|
1022 | 1122 | update = self.findoutgoing(remote) |
|
1023 | 1123 | if not update: |
|
1024 | 1124 | self.ui.status("no changes found\n") |
|
1025 | 1125 | return 1 |
|
1026 | 1126 | |
|
1027 | 1127 | cg = self.changegroup(update) |
|
1028 | 1128 | return remote.addchangegroup(cg) |
|
1029 | 1129 | |
|
1030 | 1130 | def changegroup(self, basenodes): |
|
1031 | 1131 | class genread: |
|
1032 | 1132 | def __init__(self, generator): |
|
1033 | 1133 | self.g = generator |
|
1034 | 1134 | self.buf = "" |
|
1035 | 1135 | def read(self, l): |
|
1036 | 1136 | while l > len(self.buf): |
|
1037 | 1137 | try: |
|
1038 | 1138 | self.buf += self.g.next() |
|
1039 | 1139 | except StopIteration: |
|
1040 | 1140 | break |
|
1041 | 1141 | d, self.buf = self.buf[:l], self.buf[l:] |
|
1042 | 1142 | return d |
|
1043 | 1143 | |
|
1044 | 1144 | def gengroup(): |
|
1045 | 1145 | nodes = self.newer(basenodes) |
|
1046 | 1146 | |
|
1047 | 1147 | # construct the link map |
|
1048 | 1148 | linkmap = {} |
|
1049 | 1149 | for n in nodes: |
|
1050 | 1150 | linkmap[self.changelog.rev(n)] = n |
|
1051 | 1151 | |
|
1052 | 1152 | # construct a list of all changed files |
|
1053 | 1153 | changed = {} |
|
1054 | 1154 | for n in nodes: |
|
1055 | 1155 | c = self.changelog.read(n) |
|
1056 | 1156 | for f in c[3]: |
|
1057 | 1157 | changed[f] = 1 |
|
1058 | 1158 | changed = changed.keys() |
|
1059 | 1159 | changed.sort() |
|
1060 | 1160 | |
|
1061 | 1161 | # the changegroup is changesets + manifests + all file revs |
|
1062 | 1162 | revs = [ self.changelog.rev(n) for n in nodes ] |
|
1063 | 1163 | |
|
1064 | 1164 | for y in self.changelog.group(linkmap): yield y |
|
1065 | 1165 | for y in self.manifest.group(linkmap): yield y |
|
1066 | 1166 | for f in changed: |
|
1067 | 1167 | yield struct.pack(">l", len(f) + 4) + f |
|
1068 | 1168 | g = self.file(f).group(linkmap) |
|
1069 | 1169 | for y in g: |
|
1070 | 1170 | yield y |
|
1071 | 1171 | |
|
1072 | 1172 | yield struct.pack(">l", 0) |
|
1073 | 1173 | |
|
1074 | 1174 | return genread(gengroup()) |
|
1075 | 1175 | |
|
1076 | 1176 | def addchangegroup(self, source): |
|
1077 | 1177 | |
|
1078 | 1178 | def getchunk(): |
|
1079 | 1179 | d = source.read(4) |
|
1080 | 1180 | if not d: return "" |
|
1081 | 1181 | l = struct.unpack(">l", d)[0] |
|
1082 | 1182 | if l <= 4: return "" |
|
1083 | 1183 | return source.read(l - 4) |
|
1084 | 1184 | |
|
1085 | 1185 | def getgroup(): |
|
1086 | 1186 | while 1: |
|
1087 | 1187 | c = getchunk() |
|
1088 | 1188 | if not c: break |
|
1089 | 1189 | yield c |
|
1090 | 1190 | |
|
1091 | 1191 | def csmap(x): |
|
1092 | 1192 | self.ui.debug("add changeset %s\n" % short(x)) |
|
1093 | 1193 | return self.changelog.count() |
|
1094 | 1194 | |
|
1095 | 1195 | def revmap(x): |
|
1096 | 1196 | return self.changelog.rev(x) |
|
1097 | 1197 | |
|
1098 | 1198 | if not source: return |
|
1099 | 1199 | changesets = files = revisions = 0 |
|
1100 | 1200 | |
|
1101 | 1201 | tr = self.transaction() |
|
1102 | 1202 | |
|
1103 | 1203 | # pull off the changeset group |
|
1104 | 1204 | self.ui.status("adding changesets\n") |
|
1105 | 1205 | co = self.changelog.tip() |
|
1106 | 1206 | cn = self.changelog.addgroup(getgroup(), csmap, tr, 1) # unique |
|
1107 | 1207 | changesets = self.changelog.rev(cn) - self.changelog.rev(co) |
|
1108 | 1208 | |
|
1109 | 1209 | # pull off the manifest group |
|
1110 | 1210 | self.ui.status("adding manifests\n") |
|
1111 | 1211 | mm = self.manifest.tip() |
|
1112 | 1212 | mo = self.manifest.addgroup(getgroup(), revmap, tr) |
|
1113 | 1213 | |
|
1114 | 1214 | # process the files |
|
1115 | 1215 | self.ui.status("adding file revisions\n") |
|
1116 | 1216 | while 1: |
|
1117 | 1217 | f = getchunk() |
|
1118 | 1218 | if not f: break |
|
1119 | 1219 | self.ui.debug("adding %s revisions\n" % f) |
|
1120 | 1220 | fl = self.file(f) |
|
1121 | 1221 | o = fl.count() |
|
1122 | 1222 | n = fl.addgroup(getgroup(), revmap, tr) |
|
1123 | 1223 | revisions += fl.count() - o |
|
1124 | 1224 | files += 1 |
|
1125 | 1225 | |
|
1126 | 1226 | self.ui.status(("modified %d files, added %d changesets" + |
|
1127 | 1227 | " and %d new revisions\n") |
|
1128 | 1228 | % (files, changesets, revisions)) |
|
1129 | 1229 | |
|
1130 | 1230 | tr.close() |
|
1131 | 1231 | return |
|
1132 | 1232 | |
|
1133 | 1233 | def update(self, node, allow=False, force=False, choose=None, |
|
1134 | 1234 | moddirstate=True): |
|
1135 | 1235 | pl = self.dirstate.parents() |
|
1136 | 1236 | if not force and pl[1] != nullid: |
|
1137 | 1237 | self.ui.warn("aborting: outstanding uncommitted merges\n") |
|
1138 | 1238 | return |
|
1139 | 1239 | |
|
1140 | 1240 | p1, p2 = pl[0], node |
|
1141 | 1241 | pa = self.changelog.ancestor(p1, p2) |
|
1142 | 1242 | m1n = self.changelog.read(p1)[0] |
|
1143 | 1243 | m2n = self.changelog.read(p2)[0] |
|
1144 | 1244 | man = self.manifest.ancestor(m1n, m2n) |
|
1145 | 1245 | m1 = self.manifest.read(m1n) |
|
1146 | 1246 | mf1 = self.manifest.readflags(m1n) |
|
1147 | 1247 | m2 = self.manifest.read(m2n) |
|
1148 | 1248 | mf2 = self.manifest.readflags(m2n) |
|
1149 | 1249 | ma = self.manifest.read(man) |
|
1150 | 1250 | mfa = self.manifest.readflags(man) |
|
1151 | 1251 | |
|
1152 | 1252 | (c, a, d, u) = self.changes(None, None) |
|
1153 | 1253 | |
|
1154 | 1254 | # is this a jump, or a merge? i.e. is there a linear path |
|
1155 | 1255 | # from p1 to p2? |
|
1156 | 1256 | linear_path = (pa == p1 or pa == p2) |
|
1157 | 1257 | |
|
1158 | 1258 | # resolve the manifest to determine which files |
|
1159 | 1259 | # we care about merging |
|
1160 | 1260 | self.ui.note("resolving manifests\n") |
|
1161 | 1261 | self.ui.debug(" ancestor %s local %s remote %s\n" % |
|
1162 | 1262 | (short(man), short(m1n), short(m2n))) |
|
1163 | 1263 | |
|
1164 | 1264 | merge = {} |
|
1165 | 1265 | get = {} |
|
1166 | 1266 | remove = [] |
|
1167 | 1267 | mark = {} |
|
1168 | 1268 | |
|
1169 | 1269 | # construct a working dir manifest |
|
1170 | 1270 | mw = m1.copy() |
|
1171 | 1271 | mfw = mf1.copy() |
|
1172 | 1272 | umap = dict.fromkeys(u) |
|
1173 | 1273 | |
|
1174 | 1274 | for f in a + c + u: |
|
1175 | 1275 | mw[f] = "" |
|
1176 | 1276 | mfw[f] = util.is_exec(self.wjoin(f), mfw.get(f, False)) |
|
1177 | 1277 | |
|
1178 | 1278 | for f in d: |
|
1179 | 1279 | if f in mw: del mw[f] |
|
1180 | 1280 | |
|
1181 | 1281 | # If we're jumping between revisions (as opposed to merging), |
|
1182 | 1282 | # and if neither the working directory nor the target rev has |
|
1183 | 1283 | # the file, then we need to remove it from the dirstate, to |
|
1184 | 1284 | # prevent the dirstate from listing the file when it is no |
|
1185 | 1285 | # longer in the manifest. |
|
1186 | 1286 | if moddirstate and linear_path and f not in m2: |
|
1187 | 1287 | self.dirstate.forget((f,)) |
|
1188 | 1288 | |
|
1189 | 1289 | # Compare manifests |
|
1190 | 1290 | for f, n in mw.iteritems(): |
|
1191 | 1291 | if choose and not choose(f): continue |
|
1192 | 1292 | if f in m2: |
|
1193 | 1293 | s = 0 |
|
1194 | 1294 | |
|
1195 | 1295 | # is the wfile new since m1, and match m2? |
|
1196 | 1296 | if f not in m1: |
|
1197 | 1297 | t1 = self.wfile(f).read() |
|
1198 | 1298 | t2 = self.file(f).revision(m2[f]) |
|
1199 | 1299 | if cmp(t1, t2) == 0: |
|
1200 | 1300 | mark[f] = 1 |
|
1201 | 1301 | n = m2[f] |
|
1202 | 1302 | del t1, t2 |
|
1203 | 1303 | |
|
1204 | 1304 | # are files different? |
|
1205 | 1305 | if n != m2[f]: |
|
1206 | 1306 | a = ma.get(f, nullid) |
|
1207 | 1307 | # are both different from the ancestor? |
|
1208 | 1308 | if n != a and m2[f] != a: |
|
1209 | 1309 | self.ui.debug(" %s versions differ, resolve\n" % f) |
|
1210 | 1310 | # merge executable bits |
|
1211 | 1311 | # "if we changed or they changed, change in merge" |
|
1212 | 1312 | a, b, c = mfa.get(f, 0), mfw[f], mf2[f] |
|
1213 | 1313 | mode = ((a^b) | (a^c)) ^ a |
|
1214 | 1314 | merge[f] = (m1.get(f, nullid), m2[f], mode) |
|
1215 | 1315 | s = 1 |
|
1216 | 1316 | # are we clobbering? |
|
1217 | 1317 | # is remote's version newer? |
|
1218 | 1318 | # or are we going back in time? |
|
1219 | 1319 | elif force or m2[f] != a or (p2 == pa and mw[f] == m1[f]): |
|
1220 | 1320 | self.ui.debug(" remote %s is newer, get\n" % f) |
|
1221 | 1321 | get[f] = m2[f] |
|
1222 | 1322 | s = 1 |
|
1223 | 1323 | else: |
|
1224 | 1324 | mark[f] = 1 |
|
1225 | 1325 | elif f in umap: |
|
1226 | 1326 | # this unknown file is the same as the checkout |
|
1227 | 1327 | get[f] = m2[f] |
|
1228 | 1328 | |
|
1229 | 1329 | if not s and mfw[f] != mf2[f]: |
|
1230 | 1330 | if force: |
|
1231 | 1331 | self.ui.debug(" updating permissions for %s\n" % f) |
|
1232 | 1332 | util.set_exec(self.wjoin(f), mf2[f]) |
|
1233 | 1333 | else: |
|
1234 | 1334 | a, b, c = mfa.get(f, 0), mfw[f], mf2[f] |
|
1235 | 1335 | mode = ((a^b) | (a^c)) ^ a |
|
1236 | 1336 | if mode != b: |
|
1237 | 1337 | self.ui.debug(" updating permissions for %s\n" % f) |
|
1238 | 1338 | util.set_exec(self.wjoin(f), mode) |
|
1239 | 1339 | mark[f] = 1 |
|
1240 | 1340 | del m2[f] |
|
1241 | 1341 | elif f in ma: |
|
1242 | 1342 | if n != ma[f]: |
|
1243 | 1343 | r = "d" |
|
1244 | 1344 | if not force and (linear_path or allow): |
|
1245 | 1345 | r = self.ui.prompt( |
|
1246 | 1346 | (" local changed %s which remote deleted\n" % f) + |
|
1247 | 1347 | "(k)eep or (d)elete?", "[kd]", "k") |
|
1248 | 1348 | if r == "d": |
|
1249 | 1349 | remove.append(f) |
|
1250 | 1350 | else: |
|
1251 | 1351 | self.ui.debug("other deleted %s\n" % f) |
|
1252 | 1352 | remove.append(f) # other deleted it |
|
1253 | 1353 | else: |
|
1254 | 1354 | if n == m1.get(f, nullid): # same as parent |
|
1255 | 1355 | if p2 == pa: # going backwards? |
|
1256 | 1356 | self.ui.debug("remote deleted %s\n" % f) |
|
1257 | 1357 | remove.append(f) |
|
1258 | 1358 | else: |
|
1259 | 1359 | self.ui.debug("local created %s, keeping\n" % f) |
|
1260 | 1360 | else: |
|
1261 | 1361 | self.ui.debug("working dir created %s, keeping\n" % f) |
|
1262 | 1362 | |
|
1263 | 1363 | for f, n in m2.iteritems(): |
|
1264 | 1364 | if choose and not choose(f): continue |
|
1265 | 1365 | if f[0] == "/": continue |
|
1266 | 1366 | if f in ma and n != ma[f]: |
|
1267 | 1367 | r = "k" |
|
1268 | 1368 | if not force and (linear_path or allow): |
|
1269 | 1369 | r = self.ui.prompt( |
|
1270 | 1370 | ("remote changed %s which local deleted\n" % f) + |
|
1271 | 1371 | "(k)eep or (d)elete?", "[kd]", "k") |
|
1272 | 1372 | if r == "k": get[f] = n |
|
1273 | 1373 | elif f not in ma: |
|
1274 | 1374 | self.ui.debug("remote created %s\n" % f) |
|
1275 | 1375 | get[f] = n |
|
1276 | 1376 | else: |
|
1277 | 1377 | self.ui.debug("local deleted %s\n" % f) |
|
1278 | 1378 | |
|
1279 | 1379 | del mw, m1, m2, ma |
|
1280 | 1380 | |
|
1281 | 1381 | if force: |
|
1282 | 1382 | for f in merge: |
|
1283 | 1383 | get[f] = merge[f][1] |
|
1284 | 1384 | merge = {} |
|
1285 | 1385 | |
|
1286 | 1386 | if linear_path: |
|
1287 | 1387 | # we don't need to do any magic, just jump to the new rev |
|
1288 | 1388 | mode = 'n' |
|
1289 | 1389 | p1, p2 = p2, nullid |
|
1290 | 1390 | else: |
|
1291 | 1391 | if not allow: |
|
1292 | 1392 | self.ui.status("this update spans a branch" + |
|
1293 | 1393 | " affecting the following files:\n") |
|
1294 | 1394 | fl = merge.keys() + get.keys() |
|
1295 | 1395 | fl.sort() |
|
1296 | 1396 | for f in fl: |
|
1297 | 1397 | cf = "" |
|
1298 | 1398 | if f in merge: cf = " (resolve)" |
|
1299 | 1399 | self.ui.status(" %s%s\n" % (f, cf)) |
|
1300 | 1400 | self.ui.warn("aborting update spanning branches!\n") |
|
1301 | 1401 | self.ui.status("(use update -m to perform a branch merge)\n") |
|
1302 | 1402 | return 1 |
|
1303 | 1403 | # we have to remember what files we needed to get/change |
|
1304 | 1404 | # because any file that's different from either one of its |
|
1305 | 1405 | # parents must be in the changeset |
|
1306 | 1406 | mode = 'm' |
|
1307 | 1407 | if moddirstate: |
|
1308 | 1408 | self.dirstate.update(mark.keys(), "m") |
|
1309 | 1409 | |
|
1310 | 1410 | if moddirstate: |
|
1311 | 1411 | self.dirstate.setparents(p1, p2) |
|
1312 | 1412 | |
|
1313 | 1413 | # get the files we don't need to change |
|
1314 | 1414 | files = get.keys() |
|
1315 | 1415 | files.sort() |
|
1316 | 1416 | for f in files: |
|
1317 | 1417 | if f[0] == "/": continue |
|
1318 | 1418 | self.ui.note("getting %s\n" % f) |
|
1319 | 1419 | t = self.file(f).read(get[f]) |
|
1320 | 1420 | try: |
|
1321 | 1421 | self.wfile(f, "w").write(t) |
|
1322 | 1422 | except IOError: |
|
1323 | 1423 | os.makedirs(os.path.dirname(self.wjoin(f))) |
|
1324 | 1424 | self.wfile(f, "w").write(t) |
|
1325 | 1425 | util.set_exec(self.wjoin(f), mf2[f]) |
|
1326 | 1426 | if moddirstate: |
|
1327 | 1427 | self.dirstate.update([f], mode) |
|
1328 | 1428 | |
|
1329 | 1429 | # merge the tricky bits |
|
1330 | 1430 | files = merge.keys() |
|
1331 | 1431 | files.sort() |
|
1332 | 1432 | for f in files: |
|
1333 | 1433 | self.ui.status("merging %s\n" % f) |
|
1334 | 1434 | m, o, flag = merge[f] |
|
1335 | 1435 | self.merge3(f, m, o) |
|
1336 | 1436 | util.set_exec(self.wjoin(f), flag) |
|
1337 | 1437 | if moddirstate: |
|
1338 | 1438 | self.dirstate.update([f], 'm') |
|
1339 | 1439 | |
|
1340 | 1440 | for f in remove: |
|
1341 | 1441 | self.ui.note("removing %s\n" % f) |
|
1342 | 1442 | os.unlink(f) |
|
1343 | 1443 | # try removing directories that might now be empty |
|
1344 | 1444 | try: os.removedirs(os.path.dirname(f)) |
|
1345 | 1445 | except: pass |
|
1346 | 1446 | if moddirstate: |
|
1347 | 1447 | if mode == 'n': |
|
1348 | 1448 | self.dirstate.forget(remove) |
|
1349 | 1449 | else: |
|
1350 | 1450 | self.dirstate.update(remove, 'r') |
|
1351 | 1451 | |
|
1352 | 1452 | def merge3(self, fn, my, other): |
|
1353 | 1453 | """perform a 3-way merge in the working directory""" |
|
1354 | 1454 | |
|
1355 | 1455 | def temp(prefix, node): |
|
1356 | 1456 | pre = "%s~%s." % (os.path.basename(fn), prefix) |
|
1357 | 1457 | (fd, name) = tempfile.mkstemp("", pre) |
|
1358 | 1458 | f = os.fdopen(fd, "wb") |
|
1359 | 1459 | f.write(fl.revision(node)) |
|
1360 | 1460 | f.close() |
|
1361 | 1461 | return name |
|
1362 | 1462 | |
|
1363 | 1463 | fl = self.file(fn) |
|
1364 | 1464 | base = fl.ancestor(my, other) |
|
1365 | 1465 | a = self.wjoin(fn) |
|
1366 | 1466 | b = temp("base", base) |
|
1367 | 1467 | c = temp("other", other) |
|
1368 | 1468 | |
|
1369 | 1469 | self.ui.note("resolving %s\n" % fn) |
|
1370 | 1470 | self.ui.debug("file %s: other %s ancestor %s\n" % |
|
1371 | 1471 | (fn, short(other), short(base))) |
|
1372 | 1472 | |
|
1373 | 1473 | cmd = self.ui.config("ui", "merge") or \ |
|
1374 | 1474 | os.environ.get("HGMERGE", "hgmerge") |
|
1375 | 1475 | r = os.system("%s %s %s %s" % (cmd, a, b, c)) |
|
1376 | 1476 | if r: |
|
1377 | 1477 | self.ui.warn("merging %s failed!\n" % fn) |
|
1378 | 1478 | |
|
1379 | 1479 | os.unlink(b) |
|
1380 | 1480 | os.unlink(c) |
|
1381 | 1481 | |
|
1382 | 1482 | def verify(self): |
|
1383 | 1483 | filelinkrevs = {} |
|
1384 | 1484 | filenodes = {} |
|
1385 | 1485 | changesets = revisions = files = 0 |
|
1386 | 1486 | errors = 0 |
|
1387 | 1487 | |
|
1388 | 1488 | seen = {} |
|
1389 | 1489 | self.ui.status("checking changesets\n") |
|
1390 | 1490 | for i in range(self.changelog.count()): |
|
1391 | 1491 | changesets += 1 |
|
1392 | 1492 | n = self.changelog.node(i) |
|
1393 | 1493 | if n in seen: |
|
1394 | 1494 | self.ui.warn("duplicate changeset at revision %d\n" % i) |
|
1395 | 1495 | errors += 1 |
|
1396 | 1496 | seen[n] = 1 |
|
1397 | 1497 | |
|
1398 | 1498 | for p in self.changelog.parents(n): |
|
1399 | 1499 | if p not in self.changelog.nodemap: |
|
1400 | 1500 | self.ui.warn("changeset %s has unknown parent %s\n" % |
|
1401 | 1501 | (short(n), short(p))) |
|
1402 | 1502 | errors += 1 |
|
1403 | 1503 | try: |
|
1404 | 1504 | changes = self.changelog.read(n) |
|
1405 | 1505 | except Exception, inst: |
|
1406 | 1506 | self.ui.warn("unpacking changeset %s: %s\n" % (short(n), inst)) |
|
1407 | 1507 | errors += 1 |
|
1408 | 1508 | |
|
1409 | 1509 | for f in changes[3]: |
|
1410 | 1510 | filelinkrevs.setdefault(f, []).append(i) |
|
1411 | 1511 | |
|
1412 | 1512 | seen = {} |
|
1413 | 1513 | self.ui.status("checking manifests\n") |
|
1414 | 1514 | for i in range(self.manifest.count()): |
|
1415 | 1515 | n = self.manifest.node(i) |
|
1416 | 1516 | if n in seen: |
|
1417 | 1517 | self.ui.warn("duplicate manifest at revision %d\n" % i) |
|
1418 | 1518 | errors += 1 |
|
1419 | 1519 | seen[n] = 1 |
|
1420 | 1520 | |
|
1421 | 1521 | for p in self.manifest.parents(n): |
|
1422 | 1522 | if p not in self.manifest.nodemap: |
|
1423 | 1523 | self.ui.warn("manifest %s has unknown parent %s\n" % |
|
1424 | 1524 | (short(n), short(p))) |
|
1425 | 1525 | errors += 1 |
|
1426 | 1526 | |
|
1427 | 1527 | try: |
|
1428 | 1528 | delta = mdiff.patchtext(self.manifest.delta(n)) |
|
1429 | 1529 | except KeyboardInterrupt: |
|
1430 | 1530 | self.ui.warn("aborted") |
|
1431 | 1531 | sys.exit(0) |
|
1432 | 1532 | except Exception, inst: |
|
1433 | 1533 | self.ui.warn("unpacking manifest %s: %s\n" |
|
1434 | 1534 | % (short(n), inst)) |
|
1435 | 1535 | errors += 1 |
|
1436 | 1536 | |
|
1437 | 1537 | ff = [ l.split('\0') for l in delta.splitlines() ] |
|
1438 | 1538 | for f, fn in ff: |
|
1439 | 1539 | filenodes.setdefault(f, {})[bin(fn[:40])] = 1 |
|
1440 | 1540 | |
|
1441 | 1541 | self.ui.status("crosschecking files in changesets and manifests\n") |
|
1442 | 1542 | for f in filenodes: |
|
1443 | 1543 | if f not in filelinkrevs: |
|
1444 | 1544 | self.ui.warn("file %s in manifest but not in changesets\n" % f) |
|
1445 | 1545 | errors += 1 |
|
1446 | 1546 | |
|
1447 | 1547 | for f in filelinkrevs: |
|
1448 | 1548 | if f not in filenodes: |
|
1449 | 1549 | self.ui.warn("file %s in changeset but not in manifest\n" % f) |
|
1450 | 1550 | errors += 1 |
|
1451 | 1551 | |
|
1452 | 1552 | self.ui.status("checking files\n") |
|
1453 | 1553 | ff = filenodes.keys() |
|
1454 | 1554 | ff.sort() |
|
1455 | 1555 | for f in ff: |
|
1456 | 1556 | if f == "/dev/null": continue |
|
1457 | 1557 | files += 1 |
|
1458 | 1558 | fl = self.file(f) |
|
1459 | 1559 | nodes = { nullid: 1 } |
|
1460 | 1560 | seen = {} |
|
1461 | 1561 | for i in range(fl.count()): |
|
1462 | 1562 | revisions += 1 |
|
1463 | 1563 | n = fl.node(i) |
|
1464 | 1564 | |
|
1465 | 1565 | if n in seen: |
|
1466 | 1566 | self.ui.warn("%s: duplicate revision %d\n" % (f, i)) |
|
1467 | 1567 | errors += 1 |
|
1468 | 1568 | |
|
1469 | 1569 | if n not in filenodes[f]: |
|
1470 | 1570 | self.ui.warn("%s: %d:%s not in manifests\n" |
|
1471 | 1571 | % (f, i, short(n))) |
|
1472 | 1572 | errors += 1 |
|
1473 | 1573 | else: |
|
1474 | 1574 | del filenodes[f][n] |
|
1475 | 1575 | |
|
1476 | 1576 | flr = fl.linkrev(n) |
|
1477 | 1577 | if flr not in filelinkrevs[f]: |
|
1478 | 1578 | self.ui.warn("%s:%s points to unexpected changeset %d\n" |
|
1479 | 1579 | % (f, short(n), fl.linkrev(n))) |
|
1480 | 1580 | errors += 1 |
|
1481 | 1581 | else: |
|
1482 | 1582 | filelinkrevs[f].remove(flr) |
|
1483 | 1583 | |
|
1484 | 1584 | # verify contents |
|
1485 | 1585 | try: |
|
1486 | 1586 | t = fl.read(n) |
|
1487 | 1587 | except Exception, inst: |
|
1488 | 1588 | self.ui.warn("unpacking file %s %s: %s\n" |
|
1489 | 1589 | % (f, short(n), inst)) |
|
1490 | 1590 | errors += 1 |
|
1491 | 1591 | |
|
1492 | 1592 | # verify parents |
|
1493 | 1593 | (p1, p2) = fl.parents(n) |
|
1494 | 1594 | if p1 not in nodes: |
|
1495 | 1595 | self.ui.warn("file %s:%s unknown parent 1 %s" % |
|
1496 | 1596 | (f, short(n), short(p1))) |
|
1497 | 1597 | errors += 1 |
|
1498 | 1598 | if p2 not in nodes: |
|
1499 | 1599 | self.ui.warn("file %s:%s unknown parent 2 %s" % |
|
1500 | 1600 | (f, short(n), short(p1))) |
|
1501 | 1601 | errors += 1 |
|
1502 | 1602 | nodes[n] = 1 |
|
1503 | 1603 | |
|
1504 | 1604 | # cross-check |
|
1505 | 1605 | for node in filenodes[f]: |
|
1506 | 1606 | self.ui.warn("node %s in manifests not in %s\n" |
|
1507 | 1607 | % (hex(n), f)) |
|
1508 | 1608 | errors += 1 |
|
1509 | 1609 | |
|
1510 | 1610 | self.ui.status("%d files, %d changesets, %d total revisions\n" % |
|
1511 | 1611 | (files, changesets, revisions)) |
|
1512 | 1612 | |
|
1513 | 1613 | if errors: |
|
1514 | 1614 | self.ui.warn("%d integrity errors encountered!\n" % errors) |
|
1515 | 1615 | return 1 |
|
1516 | 1616 | |
|
1517 | 1617 | class httprepository: |
|
1518 | 1618 | def __init__(self, ui, path): |
|
1519 | 1619 | self.url = path |
|
1520 | 1620 | self.ui = ui |
|
1521 | 1621 | no_list = [ "localhost", "127.0.0.1" ] |
|
1522 | 1622 | host = ui.config("http_proxy", "host") |
|
1523 | 1623 | if host is None: |
|
1524 | 1624 | host = os.environ.get("http_proxy") |
|
1525 | 1625 | if host and host.startswith('http://'): |
|
1526 | 1626 | host = host[7:] |
|
1527 | 1627 | user = ui.config("http_proxy", "user") |
|
1528 | 1628 | passwd = ui.config("http_proxy", "passwd") |
|
1529 | 1629 | no = ui.config("http_proxy", "no") |
|
1530 | 1630 | if no is None: |
|
1531 | 1631 | no = os.environ.get("no_proxy") |
|
1532 | 1632 | if no: |
|
1533 | 1633 | no_list = no_list + no.split(",") |
|
1534 | 1634 | |
|
1535 | 1635 | no_proxy = 0 |
|
1536 | 1636 | for h in no_list: |
|
1537 | 1637 | if (path.startswith("http://" + h + "/") or |
|
1538 | 1638 | path.startswith("http://" + h + ":") or |
|
1539 | 1639 | path == "http://" + h): |
|
1540 | 1640 | no_proxy = 1 |
|
1541 | 1641 | |
|
1542 | 1642 | # Note: urllib2 takes proxy values from the environment and those will |
|
1543 | 1643 | # take precedence |
|
1544 | 1644 | for env in ["HTTP_PROXY", "http_proxy", "no_proxy"]: |
|
1545 | 1645 | if os.environ.has_key(env): |
|
1546 | 1646 | del os.environ[env] |
|
1547 | 1647 | |
|
1548 | 1648 | proxy_handler = urllib2.BaseHandler() |
|
1549 | 1649 | if host and not no_proxy: |
|
1550 | 1650 | proxy_handler = urllib2.ProxyHandler({"http" : "http://" + host}) |
|
1551 | 1651 | |
|
1552 | 1652 | authinfo = None |
|
1553 | 1653 | if user and passwd: |
|
1554 | 1654 | passmgr = urllib2.HTTPPasswordMgrWithDefaultRealm() |
|
1555 | 1655 | passmgr.add_password(None, host, user, passwd) |
|
1556 | 1656 | authinfo = urllib2.ProxyBasicAuthHandler(passmgr) |
|
1557 | 1657 | |
|
1558 | 1658 | opener = urllib2.build_opener(proxy_handler, authinfo) |
|
1559 | 1659 | urllib2.install_opener(opener) |
|
1560 | 1660 | |
|
1561 | 1661 | def dev(self): |
|
1562 | 1662 | return -1 |
|
1563 | 1663 | |
|
1564 | 1664 | def do_cmd(self, cmd, **args): |
|
1565 | 1665 | self.ui.debug("sending %s command\n" % cmd) |
|
1566 | 1666 | q = {"cmd": cmd} |
|
1567 | 1667 | q.update(args) |
|
1568 | 1668 | qs = urllib.urlencode(q) |
|
1569 | 1669 | cu = "%s?%s" % (self.url, qs) |
|
1570 | 1670 | return urllib2.urlopen(cu) |
|
1571 | 1671 | |
|
1572 | 1672 | def heads(self): |
|
1573 | 1673 | d = self.do_cmd("heads").read() |
|
1574 | 1674 | try: |
|
1575 | 1675 | return map(bin, d[:-1].split(" ")) |
|
1576 | 1676 | except: |
|
1577 | 1677 | self.ui.warn("unexpected response:\n" + d[:400] + "\n...\n") |
|
1578 | 1678 | raise |
|
1579 | 1679 | |
|
1580 | 1680 | def branches(self, nodes): |
|
1581 | 1681 | n = " ".join(map(hex, nodes)) |
|
1582 | 1682 | d = self.do_cmd("branches", nodes=n).read() |
|
1583 | 1683 | try: |
|
1584 | 1684 | br = [ tuple(map(bin, b.split(" "))) for b in d.splitlines() ] |
|
1585 | 1685 | return br |
|
1586 | 1686 | except: |
|
1587 | 1687 | self.ui.warn("unexpected response:\n" + d[:400] + "\n...\n") |
|
1588 | 1688 | raise |
|
1589 | 1689 | |
|
1590 | 1690 | def between(self, pairs): |
|
1591 | 1691 | n = "\n".join(["-".join(map(hex, p)) for p in pairs]) |
|
1592 | 1692 | d = self.do_cmd("between", pairs=n).read() |
|
1593 | 1693 | try: |
|
1594 | 1694 | p = [ l and map(bin, l.split(" ")) or [] for l in d.splitlines() ] |
|
1595 | 1695 | return p |
|
1596 | 1696 | except: |
|
1597 | 1697 | self.ui.warn("unexpected response:\n" + d[:400] + "\n...\n") |
|
1598 | 1698 | raise |
|
1599 | 1699 | |
|
1600 | 1700 | def changegroup(self, nodes): |
|
1601 | 1701 | n = " ".join(map(hex, nodes)) |
|
1602 | 1702 | f = self.do_cmd("changegroup", roots=n) |
|
1603 | 1703 | bytes = 0 |
|
1604 | 1704 | |
|
1605 | 1705 | class zread: |
|
1606 | 1706 | def __init__(self, f): |
|
1607 | 1707 | self.zd = zlib.decompressobj() |
|
1608 | 1708 | self.f = f |
|
1609 | 1709 | self.buf = "" |
|
1610 | 1710 | def read(self, l): |
|
1611 | 1711 | while l > len(self.buf): |
|
1612 | 1712 | r = f.read(4096) |
|
1613 | 1713 | if r: |
|
1614 | 1714 | self.buf += self.zd.decompress(r) |
|
1615 | 1715 | else: |
|
1616 | 1716 | self.buf += self.zd.flush() |
|
1617 | 1717 | break |
|
1618 | 1718 | d, self.buf = self.buf[:l], self.buf[l:] |
|
1619 | 1719 | return d |
|
1620 | 1720 | |
|
1621 | 1721 | return zread(f) |
|
1622 | 1722 | |
|
1623 | 1723 | class remotelock: |
|
1624 | 1724 | def __init__(self, repo): |
|
1625 | 1725 | self.repo = repo |
|
1626 | 1726 | def release(self): |
|
1627 | 1727 | self.repo.unlock() |
|
1628 | 1728 | self.repo = None |
|
1629 | 1729 | def __del__(self): |
|
1630 | 1730 | if self.repo: |
|
1631 | 1731 | self.release() |
|
1632 | 1732 | |
|
1633 | 1733 | class sshrepository: |
|
1634 | 1734 | def __init__(self, ui, path): |
|
1635 | 1735 | self.url = path |
|
1636 | 1736 | self.ui = ui |
|
1637 | 1737 | |
|
1638 | 1738 | m = re.match(r'ssh://(([^@]+)@)?([^:/]+)(:(\d+))?(/(.*))?', path) |
|
1639 | 1739 | if not m: |
|
1640 | 1740 | raise RepoError("couldn't parse destination %s\n" % path) |
|
1641 | 1741 | |
|
1642 | 1742 | self.user = m.group(2) |
|
1643 | 1743 | self.host = m.group(3) |
|
1644 | 1744 | self.port = m.group(5) |
|
1645 | 1745 | self.path = m.group(7) |
|
1646 | 1746 | |
|
1647 | 1747 | args = self.user and ("%s@%s" % (self.user, self.host)) or self.host |
|
1648 | 1748 | args = self.port and ("%s -p %s") % (args, self.port) or args |
|
1649 | 1749 | path = self.path or "" |
|
1650 | 1750 | |
|
1651 | 1751 | cmd = "ssh %s 'hg -R %s serve --stdio'" |
|
1652 | 1752 | cmd = cmd % (args, path) |
|
1653 | 1753 | |
|
1654 | 1754 | self.pipeo, self.pipei = os.popen2(cmd) |
|
1655 | 1755 | |
|
1656 | 1756 | def __del__(self): |
|
1657 | 1757 | self.pipeo.close() |
|
1658 | 1758 | self.pipei.close() |
|
1659 | 1759 | |
|
1660 | 1760 | def dev(self): |
|
1661 | 1761 | return -1 |
|
1662 | 1762 | |
|
1663 | 1763 | def do_cmd(self, cmd, **args): |
|
1664 | 1764 | self.ui.debug("sending %s command\n" % cmd) |
|
1665 | 1765 | self.pipeo.write("%s\n" % cmd) |
|
1666 | 1766 | for k, v in args.items(): |
|
1667 | 1767 | self.pipeo.write("%s %d\n" % (k, len(v))) |
|
1668 | 1768 | self.pipeo.write(v) |
|
1669 | 1769 | self.pipeo.flush() |
|
1670 | 1770 | |
|
1671 | 1771 | return self.pipei |
|
1672 | 1772 | |
|
1673 | 1773 | def call(self, cmd, **args): |
|
1674 | 1774 | r = self.do_cmd(cmd, **args) |
|
1675 | 1775 | l = int(r.readline()) |
|
1676 | 1776 | return r.read(l) |
|
1677 | 1777 | |
|
1678 | 1778 | def lock(self): |
|
1679 | 1779 | self.call("lock") |
|
1680 | 1780 | return remotelock(self) |
|
1681 | 1781 | |
|
1682 | 1782 | def unlock(self): |
|
1683 | 1783 | self.call("unlock") |
|
1684 | 1784 | |
|
1685 | 1785 | def heads(self): |
|
1686 | 1786 | d = self.call("heads") |
|
1687 | 1787 | try: |
|
1688 | 1788 | return map(bin, d[:-1].split(" ")) |
|
1689 | 1789 | except: |
|
1690 | 1790 | self.ui.warn("unexpected response:\n" + d[:400] + "\n...\n") |
|
1691 | 1791 | raise |
|
1692 | 1792 | |
|
1693 | 1793 | def branches(self, nodes): |
|
1694 | 1794 | n = " ".join(map(hex, nodes)) |
|
1695 | 1795 | d = self.call("branches", nodes=n) |
|
1696 | 1796 | try: |
|
1697 | 1797 | br = [ tuple(map(bin, b.split(" "))) for b in d.splitlines() ] |
|
1698 | 1798 | return br |
|
1699 | 1799 | except: |
|
1700 | 1800 | self.ui.warn("unexpected response:\n" + d[:400] + "\n...\n") |
|
1701 | 1801 | raise |
|
1702 | 1802 | |
|
1703 | 1803 | def between(self, pairs): |
|
1704 | 1804 | n = "\n".join(["-".join(map(hex, p)) for p in pairs]) |
|
1705 | 1805 | d = self.call("between", pairs=n) |
|
1706 | 1806 | try: |
|
1707 | 1807 | p = [ l and map(bin, l.split(" ")) or [] for l in d.splitlines() ] |
|
1708 | 1808 | return p |
|
1709 | 1809 | except: |
|
1710 | 1810 | self.ui.warn("unexpected response:\n" + d[:400] + "\n...\n") |
|
1711 | 1811 | raise |
|
1712 | 1812 | |
|
1713 | 1813 | def changegroup(self, nodes): |
|
1714 | 1814 | n = " ".join(map(hex, nodes)) |
|
1715 | 1815 | f = self.do_cmd("changegroup", roots=n) |
|
1716 | 1816 | return self.pipei |
|
1717 | 1817 | |
|
1718 | 1818 | def addchangegroup(self, cg): |
|
1719 | 1819 | d = self.call("addchangegroup") |
|
1720 | 1820 | if d: |
|
1721 | 1821 | raise RepoError("push refused: %s", d) |
|
1722 | 1822 | |
|
1723 | 1823 | while 1: |
|
1724 | 1824 | d = cg.read(4096) |
|
1725 | 1825 | if not d: break |
|
1726 | 1826 | self.pipeo.write(d) |
|
1727 | 1827 | |
|
1728 | 1828 | self.pipeo.flush() |
|
1729 | 1829 | |
|
1730 | 1830 | l = int(self.pipei.readline()) |
|
1731 | 1831 | return self.pipei.read(l) |
|
1732 | 1832 | |
|
1733 | 1833 | def repository(ui, path=None, create=0): |
|
1734 | 1834 | if path: |
|
1735 | 1835 | if path.startswith("http://"): |
|
1736 | 1836 | return httprepository(ui, path) |
|
1737 | 1837 | if path.startswith("hg://"): |
|
1738 | 1838 | return httprepository(ui, path.replace("hg://", "http://")) |
|
1739 | 1839 | if path.startswith("old-http://"): |
|
1740 | 1840 | return localrepository(ui, path.replace("old-http://", "http://")) |
|
1741 | 1841 | if path.startswith("ssh://"): |
|
1742 | 1842 | return sshrepository(ui, path) |
|
1743 | 1843 | |
|
1744 | 1844 | return localrepository(ui, path, create) |
@@ -1,541 +1,542 b'' | |||
|
1 | 1 | # revlog.py - storage back-end for mercurial |
|
2 | 2 | # |
|
3 | 3 | # This provides efficient delta storage with O(1) retrieve and append |
|
4 | 4 | # and O(changes) merge between branches |
|
5 | 5 | # |
|
6 | 6 | # Copyright 2005 Matt Mackall <mpm@selenic.com> |
|
7 | 7 | # |
|
8 | 8 | # This software may be used and distributed according to the terms |
|
9 | 9 | # of the GNU General Public License, incorporated herein by reference. |
|
10 | 10 | |
|
11 | 11 | import zlib, struct, sha, binascii, heapq |
|
12 | 12 | from mercurial import mdiff |
|
13 | 13 | |
|
14 | 14 | def hex(node): return binascii.hexlify(node) |
|
15 | 15 | def bin(node): return binascii.unhexlify(node) |
|
16 | 16 | def short(node): return hex(node[:6]) |
|
17 | 17 | |
|
18 | 18 | def compress(text): |
|
19 | 19 | if not text: return text |
|
20 | 20 | if len(text) < 44: |
|
21 | 21 | if text[0] == '\0': return text |
|
22 | 22 | return 'u' + text |
|
23 | 23 | bin = zlib.compress(text) |
|
24 | 24 | if len(bin) > len(text): |
|
25 | 25 | if text[0] == '\0': return text |
|
26 | 26 | return 'u' + text |
|
27 | 27 | return bin |
|
28 | 28 | |
|
29 | 29 | def decompress(bin): |
|
30 | 30 | if not bin: return bin |
|
31 | 31 | t = bin[0] |
|
32 | 32 | if t == '\0': return bin |
|
33 | 33 | if t == 'x': return zlib.decompress(bin) |
|
34 | 34 | if t == 'u': return bin[1:] |
|
35 | 35 | raise "unknown compression type %s" % t |
|
36 | 36 | |
|
37 | 37 | def hash(text, p1, p2): |
|
38 | 38 | l = [p1, p2] |
|
39 | 39 | l.sort() |
|
40 | 40 | s = sha.new(l[0]) |
|
41 | 41 | s.update(l[1]) |
|
42 | 42 | s.update(text) |
|
43 | 43 | return s.digest() |
|
44 | 44 | |
|
45 | 45 | nullid = "\0" * 20 |
|
46 | 46 | indexformat = ">4l20s20s20s" |
|
47 | 47 | |
|
48 | 48 | class lazyparser: |
|
49 | 49 | def __init__(self, data, revlog): |
|
50 | 50 | self.data = data |
|
51 | 51 | self.s = struct.calcsize(indexformat) |
|
52 | 52 | self.l = len(data)/self.s |
|
53 | 53 | self.index = [None] * self.l |
|
54 | 54 | self.map = {nullid: -1} |
|
55 | 55 | self.all = 0 |
|
56 | 56 | self.revlog = revlog |
|
57 | 57 | |
|
58 | 58 | def load(self, pos=None): |
|
59 | 59 | if self.all: return |
|
60 | 60 | if pos is not None: |
|
61 | 61 | block = pos / 1000 |
|
62 | 62 | i = block * 1000 |
|
63 | 63 | end = min(self.l, i + 1000) |
|
64 | 64 | else: |
|
65 | 65 | self.all = 1 |
|
66 | 66 | i = 0 |
|
67 | 67 | end = self.l |
|
68 | 68 | self.revlog.index = self.index |
|
69 | 69 | self.revlog.nodemap = self.map |
|
70 | 70 | |
|
71 | 71 | while i < end: |
|
72 | 72 | d = self.data[i * self.s: (i + 1) * self.s] |
|
73 | 73 | e = struct.unpack(indexformat, d) |
|
74 | 74 | self.index[i] = e |
|
75 | 75 | self.map[e[6]] = i |
|
76 | 76 | i += 1 |
|
77 | 77 | |
|
78 | 78 | class lazyindex: |
|
79 | 79 | def __init__(self, parser): |
|
80 | 80 | self.p = parser |
|
81 | 81 | def __len__(self): |
|
82 | 82 | return len(self.p.index) |
|
83 | 83 | def load(self, pos): |
|
84 | 84 | self.p.load(pos) |
|
85 | 85 | return self.p.index[pos] |
|
86 | 86 | def __getitem__(self, pos): |
|
87 | 87 | return self.p.index[pos] or self.load(pos) |
|
88 | 88 | def append(self, e): |
|
89 | 89 | self.p.index.append(e) |
|
90 | 90 | |
|
91 | 91 | class lazymap: |
|
92 | 92 | def __init__(self, parser): |
|
93 | 93 | self.p = parser |
|
94 | 94 | def load(self, key): |
|
95 | 95 | if self.p.all: return |
|
96 | 96 | n = self.p.data.find(key) |
|
97 | 97 | if n < 0: raise KeyError("node " + hex(key)) |
|
98 | 98 | pos = n / self.p.s |
|
99 | 99 | self.p.load(pos) |
|
100 | 100 | def __contains__(self, key): |
|
101 | 101 | self.p.load() |
|
102 | 102 | return key in self.p.map |
|
103 | 103 | def __iter__(self): |
|
104 | 104 | yield nullid |
|
105 | 105 | for i in xrange(self.p.l): |
|
106 | 106 | try: |
|
107 | 107 | yield self.p.index[i][6] |
|
108 | 108 | except: |
|
109 | 109 | self.p.load(i) |
|
110 | 110 | yield self.p.index[i][6] |
|
111 | 111 | def __getitem__(self, key): |
|
112 | 112 | try: |
|
113 | 113 | return self.p.map[key] |
|
114 | 114 | except KeyError: |
|
115 | 115 | try: |
|
116 | 116 | self.load(key) |
|
117 | 117 | return self.p.map[key] |
|
118 | 118 | except KeyError: |
|
119 | 119 | raise KeyError("node " + hex(key)) |
|
120 | 120 | def __setitem__(self, key, val): |
|
121 | 121 | self.p.map[key] = val |
|
122 | 122 | |
|
123 | 123 | class revlog: |
|
124 | 124 | def __init__(self, opener, indexfile, datafile): |
|
125 | 125 | self.indexfile = indexfile |
|
126 | 126 | self.datafile = datafile |
|
127 | 127 | self.opener = opener |
|
128 | 128 | self.cache = None |
|
129 | 129 | |
|
130 | 130 | try: |
|
131 | 131 | i = self.opener(self.indexfile).read() |
|
132 | 132 | except IOError: |
|
133 | 133 | i = "" |
|
134 | 134 | |
|
135 | 135 | if len(i) > 10000: |
|
136 | 136 | # big index, let's parse it on demand |
|
137 | 137 | parser = lazyparser(i, self) |
|
138 | 138 | self.index = lazyindex(parser) |
|
139 | 139 | self.nodemap = lazymap(parser) |
|
140 | 140 | else: |
|
141 | 141 | s = struct.calcsize(indexformat) |
|
142 | 142 | l = len(i) / s |
|
143 | 143 | self.index = [None] * l |
|
144 | 144 | m = [None] * l |
|
145 | 145 | |
|
146 | 146 | n = 0 |
|
147 | 147 | for f in xrange(0, len(i), s): |
|
148 | 148 | # offset, size, base, linkrev, p1, p2, nodeid |
|
149 | 149 | e = struct.unpack(indexformat, i[f:f + s]) |
|
150 | 150 | m[n] = (e[6], n) |
|
151 | 151 | self.index[n] = e |
|
152 | 152 | n += 1 |
|
153 | 153 | |
|
154 | 154 | self.nodemap = dict(m) |
|
155 | 155 | self.nodemap[nullid] = -1 |
|
156 | 156 | |
|
157 | 157 | def tip(self): return self.node(len(self.index) - 1) |
|
158 | 158 | def count(self): return len(self.index) |
|
159 | 159 | def node(self, rev): return (rev < 0) and nullid or self.index[rev][6] |
|
160 | 160 | def rev(self, node): return self.nodemap[node] |
|
161 | 161 | def linkrev(self, node): return self.index[self.nodemap[node]][3] |
|
162 | 162 | def parents(self, node): |
|
163 | 163 | if node == nullid: return (nullid, nullid) |
|
164 | 164 | return self.index[self.nodemap[node]][4:6] |
|
165 | 165 | |
|
166 | 166 | def start(self, rev): return self.index[rev][0] |
|
167 | 167 | def length(self, rev): return self.index[rev][1] |
|
168 | 168 | def end(self, rev): return self.start(rev) + self.length(rev) |
|
169 | 169 | def base(self, rev): return self.index[rev][2] |
|
170 | 170 | |
|
171 | 171 | def heads(self): |
|
172 | 172 | p = {} |
|
173 | 173 | h = [] |
|
174 | 174 | for r in range(self.count() - 1, -1, -1): |
|
175 | 175 | n = self.node(r) |
|
176 | 176 | if n not in p: |
|
177 | 177 | h.append(n) |
|
178 | 178 | for pn in self.parents(n): |
|
179 | 179 | p[pn] = 1 |
|
180 | 180 | return h |
|
181 | 181 | |
|
182 | 182 | def children(self, node): |
|
183 | 183 | c = [] |
|
184 | 184 | p = self.rev(node) |
|
185 | 185 | for r in range(p + 1, self.count()): |
|
186 | 186 | n = self.node(r) |
|
187 | 187 | for pn in self.parents(n): |
|
188 | 188 | if pn == p: |
|
189 | 189 | c.append(p) |
|
190 | 190 | continue |
|
191 | 191 | elif pn == nullid: |
|
192 | 192 | continue |
|
193 | 193 | return c |
|
194 | 194 | |
|
195 | 195 | def lookup(self, id): |
|
196 | 196 | try: |
|
197 | 197 | rev = int(id) |
|
198 | 198 | if str(rev) != id: raise ValueError |
|
199 | 199 | if rev < 0: rev = self.count() + rev |
|
200 | 200 | if rev < 0 or rev >= self.count(): raise ValueError |
|
201 | 201 | return self.node(rev) |
|
202 | 202 | except (ValueError, OverflowError): |
|
203 | 203 | c = [] |
|
204 | 204 | for n in self.nodemap: |
|
205 | 205 | if hex(n).startswith(id): |
|
206 | 206 | c.append(n) |
|
207 | 207 | if len(c) > 1: raise KeyError("Ambiguous identifier") |
|
208 | 208 | if len(c) < 1: raise KeyError("No match found") |
|
209 | 209 | return c[0] |
|
210 | 210 | |
|
211 | 211 | return None |
|
212 | 212 | |
|
213 | 213 | def diff(self, a, b): |
|
214 | 214 | return mdiff.textdiff(a, b) |
|
215 | 215 | |
|
216 | 216 | def patches(self, t, pl): |
|
217 | 217 | return mdiff.patches(t, pl) |
|
218 | 218 | |
|
219 | 219 | def delta(self, node): |
|
220 | 220 | r = self.rev(node) |
|
221 | 221 | b = self.base(r) |
|
222 | 222 | if r == b: |
|
223 | 223 | return self.diff(self.revision(self.node(r - 1)), |
|
224 | 224 | self.revision(node)) |
|
225 | 225 | else: |
|
226 | 226 | f = self.opener(self.datafile) |
|
227 | 227 | f.seek(self.start(r)) |
|
228 | 228 | data = f.read(self.length(r)) |
|
229 | 229 | return decompress(data) |
|
230 | 230 | |
|
231 | 231 | def revision(self, node): |
|
232 | 232 | if node == nullid: return "" |
|
233 | 233 | if self.cache and self.cache[0] == node: return self.cache[2] |
|
234 | 234 | |
|
235 | 235 | text = None |
|
236 | 236 | rev = self.rev(node) |
|
237 | 237 | start, length, base, link, p1, p2, node = self.index[rev] |
|
238 | 238 | end = start + length |
|
239 | 239 | if base != rev: start = self.start(base) |
|
240 | 240 | |
|
241 | 241 | if self.cache and self.cache[1] >= base and self.cache[1] < rev: |
|
242 | 242 | base = self.cache[1] |
|
243 | 243 | start = self.start(base + 1) |
|
244 | 244 | text = self.cache[2] |
|
245 | 245 | last = 0 |
|
246 | 246 | |
|
247 | 247 | f = self.opener(self.datafile) |
|
248 | 248 | f.seek(start) |
|
249 | 249 | data = f.read(end - start) |
|
250 | 250 | |
|
251 | 251 | if not text: |
|
252 | 252 | last = self.length(base) |
|
253 | 253 | text = decompress(data[:last]) |
|
254 | 254 | |
|
255 | 255 | bins = [] |
|
256 | 256 | for r in xrange(base + 1, rev + 1): |
|
257 | 257 | s = self.length(r) |
|
258 | 258 | bins.append(decompress(data[last:last + s])) |
|
259 | 259 | last = last + s |
|
260 | 260 | |
|
261 | 261 | text = mdiff.patches(text, bins) |
|
262 | 262 | |
|
263 | 263 | if node != hash(text, p1, p2): |
|
264 | 264 | raise IOError("integrity check failed on %s:%d" |
|
265 | 265 | % (self.datafile, rev)) |
|
266 | 266 | |
|
267 | 267 | self.cache = (node, rev, text) |
|
268 | 268 | return text |
|
269 | 269 | |
|
270 | def addrevision(self, text, transaction, link, p1=None, p2=None): | |
|
270 | def addrevision(self, text, transaction, link, p1=None, p2=None, d=None): | |
|
271 | 271 | if text is None: text = "" |
|
272 | 272 | if p1 is None: p1 = self.tip() |
|
273 | 273 | if p2 is None: p2 = nullid |
|
274 | 274 | |
|
275 | 275 | node = hash(text, p1, p2) |
|
276 | 276 | |
|
277 | 277 | if node in self.nodemap: |
|
278 | 278 | return node |
|
279 | 279 | |
|
280 | 280 | n = self.count() |
|
281 | 281 | t = n - 1 |
|
282 | 282 | |
|
283 | 283 | if n: |
|
284 | 284 | base = self.base(t) |
|
285 | 285 | start = self.start(base) |
|
286 | 286 | end = self.end(t) |
|
287 | prev = self.revision(self.tip()) | |
|
288 | d = self.diff(prev, text) | |
|
287 | if not d: | |
|
288 | prev = self.revision(self.tip()) | |
|
289 | d = self.diff(prev, text) | |
|
289 | 290 | data = compress(d) |
|
290 | 291 | dist = end - start + len(data) |
|
291 | 292 | |
|
292 | 293 | # full versions are inserted when the needed deltas |
|
293 | 294 | # become comparable to the uncompressed text |
|
294 | 295 | if not n or dist > len(text) * 2: |
|
295 | 296 | data = compress(text) |
|
296 | 297 | base = n |
|
297 | 298 | else: |
|
298 | 299 | base = self.base(t) |
|
299 | 300 | |
|
300 | 301 | offset = 0 |
|
301 | 302 | if t >= 0: |
|
302 | 303 | offset = self.end(t) |
|
303 | 304 | |
|
304 | 305 | e = (offset, len(data), base, link, p1, p2, node) |
|
305 | 306 | |
|
306 | 307 | self.index.append(e) |
|
307 | 308 | self.nodemap[node] = n |
|
308 | 309 | entry = struct.pack(indexformat, *e) |
|
309 | 310 | |
|
310 | 311 | transaction.add(self.datafile, e[0]) |
|
311 | 312 | self.opener(self.datafile, "a").write(data) |
|
312 | 313 | transaction.add(self.indexfile, n * len(entry)) |
|
313 | 314 | self.opener(self.indexfile, "a").write(entry) |
|
314 | 315 | |
|
315 | 316 | self.cache = (node, n, text) |
|
316 | 317 | return node |
|
317 | 318 | |
|
318 | 319 | def ancestor(self, a, b): |
|
319 | 320 | # calculate the distance of every node from root |
|
320 | 321 | dist = {nullid: 0} |
|
321 | 322 | for i in xrange(self.count()): |
|
322 | 323 | n = self.node(i) |
|
323 | 324 | p1, p2 = self.parents(n) |
|
324 | 325 | dist[n] = max(dist[p1], dist[p2]) + 1 |
|
325 | 326 | |
|
326 | 327 | # traverse ancestors in order of decreasing distance from root |
|
327 | 328 | def ancestors(node): |
|
328 | 329 | # we store negative distances because heap returns smallest member |
|
329 | 330 | h = [(-dist[node], node)] |
|
330 | 331 | seen = {} |
|
331 | 332 | earliest = self.count() |
|
332 | 333 | while h: |
|
333 | 334 | d, n = heapq.heappop(h) |
|
334 | 335 | if n not in seen: |
|
335 | 336 | seen[n] = 1 |
|
336 | 337 | r = self.rev(n) |
|
337 | 338 | yield (-d, r, n) |
|
338 | 339 | for p in self.parents(n): |
|
339 | 340 | heapq.heappush(h, (-dist[p], p)) |
|
340 | 341 | |
|
341 | 342 | x = ancestors(a) |
|
342 | 343 | y = ancestors(b) |
|
343 | 344 | lx = x.next() |
|
344 | 345 | ly = y.next() |
|
345 | 346 | |
|
346 | 347 | # increment each ancestor list until it is closer to root than |
|
347 | 348 | # the other, or they match |
|
348 | 349 | while 1: |
|
349 | 350 | if lx == ly: |
|
350 | 351 | return lx[2] |
|
351 | 352 | elif lx < ly: |
|
352 | 353 | ly = y.next() |
|
353 | 354 | elif lx > ly: |
|
354 | 355 | lx = x.next() |
|
355 | 356 | |
|
356 | 357 | def group(self, linkmap): |
|
357 | 358 | # given a list of changeset revs, return a set of deltas and |
|
358 | 359 | # metadata corresponding to nodes. the first delta is |
|
359 | 360 | # parent(nodes[0]) -> nodes[0] the receiver is guaranteed to |
|
360 | 361 | # have this parent as it has all history before these |
|
361 | 362 | # changesets. parent is parent[0] |
|
362 | 363 | |
|
363 | 364 | revs = [] |
|
364 | 365 | needed = {} |
|
365 | 366 | |
|
366 | 367 | # find file nodes/revs that match changeset revs |
|
367 | 368 | for i in xrange(0, self.count()): |
|
368 | 369 | if self.index[i][3] in linkmap: |
|
369 | 370 | revs.append(i) |
|
370 | 371 | needed[i] = 1 |
|
371 | 372 | |
|
372 | 373 | # if we don't have any revisions touched by these changesets, bail |
|
373 | 374 | if not revs: |
|
374 | 375 | yield struct.pack(">l", 0) |
|
375 | 376 | return |
|
376 | 377 | |
|
377 | 378 | # add the parent of the first rev |
|
378 | 379 | p = self.parents(self.node(revs[0]))[0] |
|
379 | 380 | revs.insert(0, self.rev(p)) |
|
380 | 381 | |
|
381 | 382 | # for each delta that isn't contiguous in the log, we need to |
|
382 | 383 | # reconstruct the base, reconstruct the result, and then |
|
383 | 384 | # calculate the delta. We also need to do this where we've |
|
384 | 385 | # stored a full version and not a delta |
|
385 | 386 | for i in xrange(0, len(revs) - 1): |
|
386 | 387 | a, b = revs[i], revs[i + 1] |
|
387 | 388 | if a + 1 != b or self.base(b) == b: |
|
388 | 389 | for j in xrange(self.base(a), a + 1): |
|
389 | 390 | needed[j] = 1 |
|
390 | 391 | for j in xrange(self.base(b), b + 1): |
|
391 | 392 | needed[j] = 1 |
|
392 | 393 | |
|
393 | 394 | # calculate spans to retrieve from datafile |
|
394 | 395 | needed = needed.keys() |
|
395 | 396 | needed.sort() |
|
396 | 397 | spans = [] |
|
397 | 398 | oo = -1 |
|
398 | 399 | ol = 0 |
|
399 | 400 | for n in needed: |
|
400 | 401 | if n < 0: continue |
|
401 | 402 | o = self.start(n) |
|
402 | 403 | l = self.length(n) |
|
403 | 404 | if oo + ol == o: # can we merge with the previous? |
|
404 | 405 | nl = spans[-1][2] |
|
405 | 406 | nl.append((n, l)) |
|
406 | 407 | ol += l |
|
407 | 408 | spans[-1] = (oo, ol, nl) |
|
408 | 409 | else: |
|
409 | 410 | oo = o |
|
410 | 411 | ol = l |
|
411 | 412 | spans.append((oo, ol, [(n, l)])) |
|
412 | 413 | |
|
413 | 414 | # read spans in, divide up chunks |
|
414 | 415 | chunks = {} |
|
415 | 416 | for span in spans: |
|
416 | 417 | # we reopen the file for each span to make http happy for now |
|
417 | 418 | f = self.opener(self.datafile) |
|
418 | 419 | f.seek(span[0]) |
|
419 | 420 | data = f.read(span[1]) |
|
420 | 421 | |
|
421 | 422 | # divide up the span |
|
422 | 423 | pos = 0 |
|
423 | 424 | for r, l in span[2]: |
|
424 | 425 | chunks[r] = decompress(data[pos: pos + l]) |
|
425 | 426 | pos += l |
|
426 | 427 | |
|
427 | 428 | # helper to reconstruct intermediate versions |
|
428 | 429 | def construct(text, base, rev): |
|
429 | 430 | bins = [chunks[r] for r in xrange(base + 1, rev + 1)] |
|
430 | 431 | return mdiff.patches(text, bins) |
|
431 | 432 | |
|
432 | 433 | # build deltas |
|
433 | 434 | deltas = [] |
|
434 | 435 | for d in xrange(0, len(revs) - 1): |
|
435 | 436 | a, b = revs[d], revs[d + 1] |
|
436 | 437 | n = self.node(b) |
|
437 | 438 | |
|
438 | 439 | # do we need to construct a new delta? |
|
439 | 440 | if a + 1 != b or self.base(b) == b: |
|
440 | 441 | if a >= 0: |
|
441 | 442 | base = self.base(a) |
|
442 | 443 | ta = chunks[self.base(a)] |
|
443 | 444 | ta = construct(ta, base, a) |
|
444 | 445 | else: |
|
445 | 446 | ta = "" |
|
446 | 447 | |
|
447 | 448 | base = self.base(b) |
|
448 | 449 | if a > base: |
|
449 | 450 | base = a |
|
450 | 451 | tb = ta |
|
451 | 452 | else: |
|
452 | 453 | tb = chunks[self.base(b)] |
|
453 | 454 | tb = construct(tb, base, b) |
|
454 | 455 | d = self.diff(ta, tb) |
|
455 | 456 | else: |
|
456 | 457 | d = chunks[b] |
|
457 | 458 | |
|
458 | 459 | p = self.parents(n) |
|
459 | 460 | meta = n + p[0] + p[1] + linkmap[self.linkrev(n)] |
|
460 | 461 | l = struct.pack(">l", len(meta) + len(d) + 4) |
|
461 | 462 | yield l |
|
462 | 463 | yield meta |
|
463 | 464 | yield d |
|
464 | 465 | |
|
465 | 466 | yield struct.pack(">l", 0) |
|
466 | 467 | |
|
467 | 468 | def addgroup(self, revs, linkmapper, transaction, unique = 0): |
|
468 | 469 | # given a set of deltas, add them to the revision log. the |
|
469 | 470 | # first delta is against its parent, which should be in our |
|
470 | 471 | # log, the rest are against the previous delta. |
|
471 | 472 | |
|
472 | 473 | # track the base of the current delta log |
|
473 | 474 | r = self.count() |
|
474 | 475 | t = r - 1 |
|
475 | 476 | node = nullid |
|
476 | 477 | |
|
477 | 478 | base = prev = -1 |
|
478 | 479 | start = end = 0 |
|
479 | 480 | if r: |
|
480 | 481 | start = self.start(self.base(t)) |
|
481 | 482 | end = self.end(t) |
|
482 | 483 | measure = self.length(self.base(t)) |
|
483 | 484 | base = self.base(t) |
|
484 | 485 | prev = self.tip() |
|
485 | 486 | |
|
486 | 487 | transaction.add(self.datafile, end) |
|
487 | 488 | transaction.add(self.indexfile, r * struct.calcsize(indexformat)) |
|
488 | 489 | dfh = self.opener(self.datafile, "a") |
|
489 | 490 | ifh = self.opener(self.indexfile, "a") |
|
490 | 491 | |
|
491 | 492 | # loop through our set of deltas |
|
492 | 493 | chain = None |
|
493 | 494 | for chunk in revs: |
|
494 | 495 | node, p1, p2, cs = struct.unpack("20s20s20s20s", chunk[:80]) |
|
495 | 496 | link = linkmapper(cs) |
|
496 | 497 | if node in self.nodemap: |
|
497 | 498 | # this can happen if two branches make the same change |
|
498 | 499 | if unique: |
|
499 | 500 | raise "already have %s" % hex(node[:4]) |
|
500 | 501 | continue |
|
501 | 502 | delta = chunk[80:] |
|
502 | 503 | |
|
503 | 504 | if not chain: |
|
504 | 505 | # retrieve the parent revision of the delta chain |
|
505 | 506 | chain = p1 |
|
506 | 507 | if not chain in self.nodemap: |
|
507 | 508 | raise "unknown base %s" % short(chain[:4]) |
|
508 | 509 | |
|
509 | 510 | # full versions are inserted when the needed deltas become |
|
510 | 511 | # comparable to the uncompressed text or when the previous |
|
511 | 512 | # version is not the one we have a delta against. We use |
|
512 | 513 | # the size of the previous full rev as a proxy for the |
|
513 | 514 | # current size. |
|
514 | 515 | |
|
515 | 516 | if chain == prev: |
|
516 | 517 | cdelta = compress(delta) |
|
517 | 518 | |
|
518 | 519 | if chain != prev or (end - start + len(cdelta)) > measure * 2: |
|
519 | 520 | # flush our writes here so we can read it in revision |
|
520 | 521 | dfh.flush() |
|
521 | 522 | ifh.flush() |
|
522 | 523 | text = self.revision(chain) |
|
523 | 524 | text = self.patches(text, [delta]) |
|
524 | 525 | chk = self.addrevision(text, transaction, link, p1, p2) |
|
525 | 526 | if chk != node: |
|
526 | 527 | raise "consistency error adding group" |
|
527 | 528 | measure = len(text) |
|
528 | 529 | else: |
|
529 | 530 | e = (end, len(cdelta), self.base(t), link, p1, p2, node) |
|
530 | 531 | self.index.append(e) |
|
531 | 532 | self.nodemap[node] = r |
|
532 | 533 | dfh.write(cdelta) |
|
533 | 534 | ifh.write(struct.pack(indexformat, *e)) |
|
534 | 535 | |
|
535 | 536 | t, r, chain, prev = r, r + 1, node, node |
|
536 | 537 | start = self.start(self.base(t)) |
|
537 | 538 | end = self.end(t) |
|
538 | 539 | |
|
539 | 540 | dfh.close() |
|
540 | 541 | ifh.close() |
|
541 | 542 | return node |
General Comments 0
You need to be logged in to leave comments.
Login now