Show More
@@ -1,1487 +1,1461 b'' | |||
|
1 | 1 | # changegroup.py - Mercurial changegroup manipulation functions |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2006 Matt Mackall <mpm@selenic.com> |
|
4 | 4 | # |
|
5 | 5 | # This software may be used and distributed according to the terms of the |
|
6 | 6 | # GNU General Public License version 2 or any later version. |
|
7 | 7 | |
|
8 | 8 | from __future__ import absolute_import |
|
9 | 9 | |
|
10 | 10 | import os |
|
11 | 11 | import struct |
|
12 | 12 | import weakref |
|
13 | 13 | |
|
14 | 14 | from .i18n import _ |
|
15 | 15 | from .node import ( |
|
16 | 16 | hex, |
|
17 | 17 | nullid, |
|
18 | 18 | nullrev, |
|
19 | 19 | short, |
|
20 | 20 | ) |
|
21 | 21 | |
|
22 | 22 | from .thirdparty import ( |
|
23 | 23 | attr, |
|
24 | 24 | ) |
|
25 | 25 | |
|
26 | 26 | from . import ( |
|
27 | 27 | dagutil, |
|
28 | 28 | error, |
|
29 | 29 | match as matchmod, |
|
30 | 30 | mdiff, |
|
31 | 31 | phases, |
|
32 | 32 | pycompat, |
|
33 | 33 | repository, |
|
34 | 34 | revlog, |
|
35 | 35 | util, |
|
36 | 36 | ) |
|
37 | 37 | |
|
38 | 38 | from .utils import ( |
|
39 | 39 | stringutil, |
|
40 | 40 | ) |
|
41 | 41 | |
|
42 | 42 | _CHANGEGROUPV1_DELTA_HEADER = struct.Struct("20s20s20s20s") |
|
43 | 43 | _CHANGEGROUPV2_DELTA_HEADER = struct.Struct("20s20s20s20s20s") |
|
44 | 44 | _CHANGEGROUPV3_DELTA_HEADER = struct.Struct(">20s20s20s20s20sH") |
|
45 | 45 | |
|
46 | 46 | LFS_REQUIREMENT = 'lfs' |
|
47 | 47 | |
|
48 | 48 | readexactly = util.readexactly |
|
49 | 49 | |
|
50 | 50 | def getchunk(stream): |
|
51 | 51 | """return the next chunk from stream as a string""" |
|
52 | 52 | d = readexactly(stream, 4) |
|
53 | 53 | l = struct.unpack(">l", d)[0] |
|
54 | 54 | if l <= 4: |
|
55 | 55 | if l: |
|
56 | 56 | raise error.Abort(_("invalid chunk length %d") % l) |
|
57 | 57 | return "" |
|
58 | 58 | return readexactly(stream, l - 4) |
|
59 | 59 | |
|
60 | 60 | def chunkheader(length): |
|
61 | 61 | """return a changegroup chunk header (string)""" |
|
62 | 62 | return struct.pack(">l", length + 4) |
|
63 | 63 | |
|
64 | 64 | def closechunk(): |
|
65 | 65 | """return a changegroup chunk header (string) for a zero-length chunk""" |
|
66 | 66 | return struct.pack(">l", 0) |
|
67 | 67 | |
|
68 | 68 | def _fileheader(path): |
|
69 | 69 | """Obtain a changegroup chunk header for a named path.""" |
|
70 | 70 | return chunkheader(len(path)) + path |
|
71 | 71 | |
|
72 | 72 | def writechunks(ui, chunks, filename, vfs=None): |
|
73 | 73 | """Write chunks to a file and return its filename. |
|
74 | 74 | |
|
75 | 75 | The stream is assumed to be a bundle file. |
|
76 | 76 | Existing files will not be overwritten. |
|
77 | 77 | If no filename is specified, a temporary file is created. |
|
78 | 78 | """ |
|
79 | 79 | fh = None |
|
80 | 80 | cleanup = None |
|
81 | 81 | try: |
|
82 | 82 | if filename: |
|
83 | 83 | if vfs: |
|
84 | 84 | fh = vfs.open(filename, "wb") |
|
85 | 85 | else: |
|
86 | 86 | # Increase default buffer size because default is usually |
|
87 | 87 | # small (4k is common on Linux). |
|
88 | 88 | fh = open(filename, "wb", 131072) |
|
89 | 89 | else: |
|
90 | 90 | fd, filename = pycompat.mkstemp(prefix="hg-bundle-", suffix=".hg") |
|
91 | 91 | fh = os.fdopen(fd, r"wb") |
|
92 | 92 | cleanup = filename |
|
93 | 93 | for c in chunks: |
|
94 | 94 | fh.write(c) |
|
95 | 95 | cleanup = None |
|
96 | 96 | return filename |
|
97 | 97 | finally: |
|
98 | 98 | if fh is not None: |
|
99 | 99 | fh.close() |
|
100 | 100 | if cleanup is not None: |
|
101 | 101 | if filename and vfs: |
|
102 | 102 | vfs.unlink(cleanup) |
|
103 | 103 | else: |
|
104 | 104 | os.unlink(cleanup) |
|
105 | 105 | |
|
106 | 106 | class cg1unpacker(object): |
|
107 | 107 | """Unpacker for cg1 changegroup streams. |
|
108 | 108 | |
|
109 | 109 | A changegroup unpacker handles the framing of the revision data in |
|
110 | 110 | the wire format. Most consumers will want to use the apply() |
|
111 | 111 | method to add the changes from the changegroup to a repository. |
|
112 | 112 | |
|
113 | 113 | If you're forwarding a changegroup unmodified to another consumer, |
|
114 | 114 | use getchunks(), which returns an iterator of changegroup |
|
115 | 115 | chunks. This is mostly useful for cases where you need to know the |
|
116 | 116 | data stream has ended by observing the end of the changegroup. |
|
117 | 117 | |
|
118 | 118 | deltachunk() is useful only if you're applying delta data. Most |
|
119 | 119 | consumers should prefer apply() instead. |
|
120 | 120 | |
|
121 | 121 | A few other public methods exist. Those are used only for |
|
122 | 122 | bundlerepo and some debug commands - their use is discouraged. |
|
123 | 123 | """ |
|
124 | 124 | deltaheader = _CHANGEGROUPV1_DELTA_HEADER |
|
125 | 125 | deltaheadersize = deltaheader.size |
|
126 | 126 | version = '01' |
|
127 | 127 | _grouplistcount = 1 # One list of files after the manifests |
|
128 | 128 | |
|
129 | 129 | def __init__(self, fh, alg, extras=None): |
|
130 | 130 | if alg is None: |
|
131 | 131 | alg = 'UN' |
|
132 | 132 | if alg not in util.compengines.supportedbundletypes: |
|
133 | 133 | raise error.Abort(_('unknown stream compression type: %s') |
|
134 | 134 | % alg) |
|
135 | 135 | if alg == 'BZ': |
|
136 | 136 | alg = '_truncatedBZ' |
|
137 | 137 | |
|
138 | 138 | compengine = util.compengines.forbundletype(alg) |
|
139 | 139 | self._stream = compengine.decompressorreader(fh) |
|
140 | 140 | self._type = alg |
|
141 | 141 | self.extras = extras or {} |
|
142 | 142 | self.callback = None |
|
143 | 143 | |
|
144 | 144 | # These methods (compressed, read, seek, tell) all appear to only |
|
145 | 145 | # be used by bundlerepo, but it's a little hard to tell. |
|
146 | 146 | def compressed(self): |
|
147 | 147 | return self._type is not None and self._type != 'UN' |
|
148 | 148 | def read(self, l): |
|
149 | 149 | return self._stream.read(l) |
|
150 | 150 | def seek(self, pos): |
|
151 | 151 | return self._stream.seek(pos) |
|
152 | 152 | def tell(self): |
|
153 | 153 | return self._stream.tell() |
|
154 | 154 | def close(self): |
|
155 | 155 | return self._stream.close() |
|
156 | 156 | |
|
157 | 157 | def _chunklength(self): |
|
158 | 158 | d = readexactly(self._stream, 4) |
|
159 | 159 | l = struct.unpack(">l", d)[0] |
|
160 | 160 | if l <= 4: |
|
161 | 161 | if l: |
|
162 | 162 | raise error.Abort(_("invalid chunk length %d") % l) |
|
163 | 163 | return 0 |
|
164 | 164 | if self.callback: |
|
165 | 165 | self.callback() |
|
166 | 166 | return l - 4 |
|
167 | 167 | |
|
168 | 168 | def changelogheader(self): |
|
169 | 169 | """v10 does not have a changelog header chunk""" |
|
170 | 170 | return {} |
|
171 | 171 | |
|
172 | 172 | def manifestheader(self): |
|
173 | 173 | """v10 does not have a manifest header chunk""" |
|
174 | 174 | return {} |
|
175 | 175 | |
|
176 | 176 | def filelogheader(self): |
|
177 | 177 | """return the header of the filelogs chunk, v10 only has the filename""" |
|
178 | 178 | l = self._chunklength() |
|
179 | 179 | if not l: |
|
180 | 180 | return {} |
|
181 | 181 | fname = readexactly(self._stream, l) |
|
182 | 182 | return {'filename': fname} |
|
183 | 183 | |
|
184 | 184 | def _deltaheader(self, headertuple, prevnode): |
|
185 | 185 | node, p1, p2, cs = headertuple |
|
186 | 186 | if prevnode is None: |
|
187 | 187 | deltabase = p1 |
|
188 | 188 | else: |
|
189 | 189 | deltabase = prevnode |
|
190 | 190 | flags = 0 |
|
191 | 191 | return node, p1, p2, deltabase, cs, flags |
|
192 | 192 | |
|
193 | 193 | def deltachunk(self, prevnode): |
|
194 | 194 | l = self._chunklength() |
|
195 | 195 | if not l: |
|
196 | 196 | return {} |
|
197 | 197 | headerdata = readexactly(self._stream, self.deltaheadersize) |
|
198 | 198 | header = self.deltaheader.unpack(headerdata) |
|
199 | 199 | delta = readexactly(self._stream, l - self.deltaheadersize) |
|
200 | 200 | node, p1, p2, deltabase, cs, flags = self._deltaheader(header, prevnode) |
|
201 | 201 | return (node, p1, p2, cs, deltabase, delta, flags) |
|
202 | 202 | |
|
203 | 203 | def getchunks(self): |
|
204 | 204 | """returns all the chunks contains in the bundle |
|
205 | 205 | |
|
206 | 206 | Used when you need to forward the binary stream to a file or another |
|
207 | 207 | network API. To do so, it parse the changegroup data, otherwise it will |
|
208 | 208 | block in case of sshrepo because it don't know the end of the stream. |
|
209 | 209 | """ |
|
210 | 210 | # For changegroup 1 and 2, we expect 3 parts: changelog, manifestlog, |
|
211 | 211 | # and a list of filelogs. For changegroup 3, we expect 4 parts: |
|
212 | 212 | # changelog, manifestlog, a list of tree manifestlogs, and a list of |
|
213 | 213 | # filelogs. |
|
214 | 214 | # |
|
215 | 215 | # Changelog and manifestlog parts are terminated with empty chunks. The |
|
216 | 216 | # tree and file parts are a list of entry sections. Each entry section |
|
217 | 217 | # is a series of chunks terminating in an empty chunk. The list of these |
|
218 | 218 | # entry sections is terminated in yet another empty chunk, so we know |
|
219 | 219 | # we've reached the end of the tree/file list when we reach an empty |
|
220 | 220 | # chunk that was proceeded by no non-empty chunks. |
|
221 | 221 | |
|
222 | 222 | parts = 0 |
|
223 | 223 | while parts < 2 + self._grouplistcount: |
|
224 | 224 | noentries = True |
|
225 | 225 | while True: |
|
226 | 226 | chunk = getchunk(self) |
|
227 | 227 | if not chunk: |
|
228 | 228 | # The first two empty chunks represent the end of the |
|
229 | 229 | # changelog and the manifestlog portions. The remaining |
|
230 | 230 | # empty chunks represent either A) the end of individual |
|
231 | 231 | # tree or file entries in the file list, or B) the end of |
|
232 | 232 | # the entire list. It's the end of the entire list if there |
|
233 | 233 | # were no entries (i.e. noentries is True). |
|
234 | 234 | if parts < 2: |
|
235 | 235 | parts += 1 |
|
236 | 236 | elif noentries: |
|
237 | 237 | parts += 1 |
|
238 | 238 | break |
|
239 | 239 | noentries = False |
|
240 | 240 | yield chunkheader(len(chunk)) |
|
241 | 241 | pos = 0 |
|
242 | 242 | while pos < len(chunk): |
|
243 | 243 | next = pos + 2**20 |
|
244 | 244 | yield chunk[pos:next] |
|
245 | 245 | pos = next |
|
246 | 246 | yield closechunk() |
|
247 | 247 | |
|
248 | 248 | def _unpackmanifests(self, repo, revmap, trp, prog): |
|
249 | 249 | self.callback = prog.increment |
|
250 | 250 | # no need to check for empty manifest group here: |
|
251 | 251 | # if the result of the merge of 1 and 2 is the same in 3 and 4, |
|
252 | 252 | # no new manifest will be created and the manifest group will |
|
253 | 253 | # be empty during the pull |
|
254 | 254 | self.manifestheader() |
|
255 | 255 | deltas = self.deltaiter() |
|
256 | 256 | repo.manifestlog.addgroup(deltas, revmap, trp) |
|
257 | 257 | prog.complete() |
|
258 | 258 | self.callback = None |
|
259 | 259 | |
|
260 | 260 | def apply(self, repo, tr, srctype, url, targetphase=phases.draft, |
|
261 | 261 | expectedtotal=None): |
|
262 | 262 | """Add the changegroup returned by source.read() to this repo. |
|
263 | 263 | srctype is a string like 'push', 'pull', or 'unbundle'. url is |
|
264 | 264 | the URL of the repo where this changegroup is coming from. |
|
265 | 265 | |
|
266 | 266 | Return an integer summarizing the change to this repo: |
|
267 | 267 | - nothing changed or no source: 0 |
|
268 | 268 | - more heads than before: 1+added heads (2..n) |
|
269 | 269 | - fewer heads than before: -1-removed heads (-2..-n) |
|
270 | 270 | - number of heads stays the same: 1 |
|
271 | 271 | """ |
|
272 | 272 | repo = repo.unfiltered() |
|
273 | 273 | def csmap(x): |
|
274 | 274 | repo.ui.debug("add changeset %s\n" % short(x)) |
|
275 | 275 | return len(cl) |
|
276 | 276 | |
|
277 | 277 | def revmap(x): |
|
278 | 278 | return cl.rev(x) |
|
279 | 279 | |
|
280 | 280 | changesets = files = revisions = 0 |
|
281 | 281 | |
|
282 | 282 | try: |
|
283 | 283 | # The transaction may already carry source information. In this |
|
284 | 284 | # case we use the top level data. We overwrite the argument |
|
285 | 285 | # because we need to use the top level value (if they exist) |
|
286 | 286 | # in this function. |
|
287 | 287 | srctype = tr.hookargs.setdefault('source', srctype) |
|
288 | 288 | url = tr.hookargs.setdefault('url', url) |
|
289 | 289 | repo.hook('prechangegroup', |
|
290 | 290 | throw=True, **pycompat.strkwargs(tr.hookargs)) |
|
291 | 291 | |
|
292 | 292 | # write changelog data to temp files so concurrent readers |
|
293 | 293 | # will not see an inconsistent view |
|
294 | 294 | cl = repo.changelog |
|
295 | 295 | cl.delayupdate(tr) |
|
296 | 296 | oldheads = set(cl.heads()) |
|
297 | 297 | |
|
298 | 298 | trp = weakref.proxy(tr) |
|
299 | 299 | # pull off the changeset group |
|
300 | 300 | repo.ui.status(_("adding changesets\n")) |
|
301 | 301 | clstart = len(cl) |
|
302 | 302 | progress = repo.ui.makeprogress(_('changesets'), unit=_('chunks'), |
|
303 | 303 | total=expectedtotal) |
|
304 | 304 | self.callback = progress.increment |
|
305 | 305 | |
|
306 | 306 | efiles = set() |
|
307 | 307 | def onchangelog(cl, node): |
|
308 | 308 | efiles.update(cl.readfiles(node)) |
|
309 | 309 | |
|
310 | 310 | self.changelogheader() |
|
311 | 311 | deltas = self.deltaiter() |
|
312 | 312 | cgnodes = cl.addgroup(deltas, csmap, trp, addrevisioncb=onchangelog) |
|
313 | 313 | efiles = len(efiles) |
|
314 | 314 | |
|
315 | 315 | if not cgnodes: |
|
316 | 316 | repo.ui.develwarn('applied empty changegroup', |
|
317 | 317 | config='warn-empty-changegroup') |
|
318 | 318 | clend = len(cl) |
|
319 | 319 | changesets = clend - clstart |
|
320 | 320 | progress.complete() |
|
321 | 321 | self.callback = None |
|
322 | 322 | |
|
323 | 323 | # pull off the manifest group |
|
324 | 324 | repo.ui.status(_("adding manifests\n")) |
|
325 | 325 | # We know that we'll never have more manifests than we had |
|
326 | 326 | # changesets. |
|
327 | 327 | progress = repo.ui.makeprogress(_('manifests'), unit=_('chunks'), |
|
328 | 328 | total=changesets) |
|
329 | 329 | self._unpackmanifests(repo, revmap, trp, progress) |
|
330 | 330 | |
|
331 | 331 | needfiles = {} |
|
332 | 332 | if repo.ui.configbool('server', 'validate'): |
|
333 | 333 | cl = repo.changelog |
|
334 | 334 | ml = repo.manifestlog |
|
335 | 335 | # validate incoming csets have their manifests |
|
336 | 336 | for cset in pycompat.xrange(clstart, clend): |
|
337 | 337 | mfnode = cl.changelogrevision(cset).manifest |
|
338 | 338 | mfest = ml[mfnode].readdelta() |
|
339 | 339 | # store file cgnodes we must see |
|
340 | 340 | for f, n in mfest.iteritems(): |
|
341 | 341 | needfiles.setdefault(f, set()).add(n) |
|
342 | 342 | |
|
343 | 343 | # process the files |
|
344 | 344 | repo.ui.status(_("adding file changes\n")) |
|
345 | 345 | newrevs, newfiles = _addchangegroupfiles( |
|
346 | 346 | repo, self, revmap, trp, efiles, needfiles) |
|
347 | 347 | revisions += newrevs |
|
348 | 348 | files += newfiles |
|
349 | 349 | |
|
350 | 350 | deltaheads = 0 |
|
351 | 351 | if oldheads: |
|
352 | 352 | heads = cl.heads() |
|
353 | 353 | deltaheads = len(heads) - len(oldheads) |
|
354 | 354 | for h in heads: |
|
355 | 355 | if h not in oldheads and repo[h].closesbranch(): |
|
356 | 356 | deltaheads -= 1 |
|
357 | 357 | htext = "" |
|
358 | 358 | if deltaheads: |
|
359 | 359 | htext = _(" (%+d heads)") % deltaheads |
|
360 | 360 | |
|
361 | 361 | repo.ui.status(_("added %d changesets" |
|
362 | 362 | " with %d changes to %d files%s\n") |
|
363 | 363 | % (changesets, revisions, files, htext)) |
|
364 | 364 | repo.invalidatevolatilesets() |
|
365 | 365 | |
|
366 | 366 | if changesets > 0: |
|
367 | 367 | if 'node' not in tr.hookargs: |
|
368 | 368 | tr.hookargs['node'] = hex(cl.node(clstart)) |
|
369 | 369 | tr.hookargs['node_last'] = hex(cl.node(clend - 1)) |
|
370 | 370 | hookargs = dict(tr.hookargs) |
|
371 | 371 | else: |
|
372 | 372 | hookargs = dict(tr.hookargs) |
|
373 | 373 | hookargs['node'] = hex(cl.node(clstart)) |
|
374 | 374 | hookargs['node_last'] = hex(cl.node(clend - 1)) |
|
375 | 375 | repo.hook('pretxnchangegroup', |
|
376 | 376 | throw=True, **pycompat.strkwargs(hookargs)) |
|
377 | 377 | |
|
378 | 378 | added = [cl.node(r) for r in pycompat.xrange(clstart, clend)] |
|
379 | 379 | phaseall = None |
|
380 | 380 | if srctype in ('push', 'serve'): |
|
381 | 381 | # Old servers can not push the boundary themselves. |
|
382 | 382 | # New servers won't push the boundary if changeset already |
|
383 | 383 | # exists locally as secret |
|
384 | 384 | # |
|
385 | 385 | # We should not use added here but the list of all change in |
|
386 | 386 | # the bundle |
|
387 | 387 | if repo.publishing(): |
|
388 | 388 | targetphase = phaseall = phases.public |
|
389 | 389 | else: |
|
390 | 390 | # closer target phase computation |
|
391 | 391 | |
|
392 | 392 | # Those changesets have been pushed from the |
|
393 | 393 | # outside, their phases are going to be pushed |
|
394 | 394 | # alongside. Therefor `targetphase` is |
|
395 | 395 | # ignored. |
|
396 | 396 | targetphase = phaseall = phases.draft |
|
397 | 397 | if added: |
|
398 | 398 | phases.registernew(repo, tr, targetphase, added) |
|
399 | 399 | if phaseall is not None: |
|
400 | 400 | phases.advanceboundary(repo, tr, phaseall, cgnodes) |
|
401 | 401 | |
|
402 | 402 | if changesets > 0: |
|
403 | 403 | |
|
404 | 404 | def runhooks(): |
|
405 | 405 | # These hooks run when the lock releases, not when the |
|
406 | 406 | # transaction closes. So it's possible for the changelog |
|
407 | 407 | # to have changed since we last saw it. |
|
408 | 408 | if clstart >= len(repo): |
|
409 | 409 | return |
|
410 | 410 | |
|
411 | 411 | repo.hook("changegroup", **pycompat.strkwargs(hookargs)) |
|
412 | 412 | |
|
413 | 413 | for n in added: |
|
414 | 414 | args = hookargs.copy() |
|
415 | 415 | args['node'] = hex(n) |
|
416 | 416 | del args['node_last'] |
|
417 | 417 | repo.hook("incoming", **pycompat.strkwargs(args)) |
|
418 | 418 | |
|
419 | 419 | newheads = [h for h in repo.heads() |
|
420 | 420 | if h not in oldheads] |
|
421 | 421 | repo.ui.log("incoming", |
|
422 | 422 | "%d incoming changes - new heads: %s\n", |
|
423 | 423 | len(added), |
|
424 | 424 | ', '.join([hex(c[:6]) for c in newheads])) |
|
425 | 425 | |
|
426 | 426 | tr.addpostclose('changegroup-runhooks-%020i' % clstart, |
|
427 | 427 | lambda tr: repo._afterlock(runhooks)) |
|
428 | 428 | finally: |
|
429 | 429 | repo.ui.flush() |
|
430 | 430 | # never return 0 here: |
|
431 | 431 | if deltaheads < 0: |
|
432 | 432 | ret = deltaheads - 1 |
|
433 | 433 | else: |
|
434 | 434 | ret = deltaheads + 1 |
|
435 | 435 | return ret |
|
436 | 436 | |
|
437 | 437 | def deltaiter(self): |
|
438 | 438 | """ |
|
439 | 439 | returns an iterator of the deltas in this changegroup |
|
440 | 440 | |
|
441 | 441 | Useful for passing to the underlying storage system to be stored. |
|
442 | 442 | """ |
|
443 | 443 | chain = None |
|
444 | 444 | for chunkdata in iter(lambda: self.deltachunk(chain), {}): |
|
445 | 445 | # Chunkdata: (node, p1, p2, cs, deltabase, delta, flags) |
|
446 | 446 | yield chunkdata |
|
447 | 447 | chain = chunkdata[0] |
|
448 | 448 | |
|
449 | 449 | class cg2unpacker(cg1unpacker): |
|
450 | 450 | """Unpacker for cg2 streams. |
|
451 | 451 | |
|
452 | 452 | cg2 streams add support for generaldelta, so the delta header |
|
453 | 453 | format is slightly different. All other features about the data |
|
454 | 454 | remain the same. |
|
455 | 455 | """ |
|
456 | 456 | deltaheader = _CHANGEGROUPV2_DELTA_HEADER |
|
457 | 457 | deltaheadersize = deltaheader.size |
|
458 | 458 | version = '02' |
|
459 | 459 | |
|
460 | 460 | def _deltaheader(self, headertuple, prevnode): |
|
461 | 461 | node, p1, p2, deltabase, cs = headertuple |
|
462 | 462 | flags = 0 |
|
463 | 463 | return node, p1, p2, deltabase, cs, flags |
|
464 | 464 | |
|
465 | 465 | class cg3unpacker(cg2unpacker): |
|
466 | 466 | """Unpacker for cg3 streams. |
|
467 | 467 | |
|
468 | 468 | cg3 streams add support for exchanging treemanifests and revlog |
|
469 | 469 | flags. It adds the revlog flags to the delta header and an empty chunk |
|
470 | 470 | separating manifests and files. |
|
471 | 471 | """ |
|
472 | 472 | deltaheader = _CHANGEGROUPV3_DELTA_HEADER |
|
473 | 473 | deltaheadersize = deltaheader.size |
|
474 | 474 | version = '03' |
|
475 | 475 | _grouplistcount = 2 # One list of manifests and one list of files |
|
476 | 476 | |
|
477 | 477 | def _deltaheader(self, headertuple, prevnode): |
|
478 | 478 | node, p1, p2, deltabase, cs, flags = headertuple |
|
479 | 479 | return node, p1, p2, deltabase, cs, flags |
|
480 | 480 | |
|
481 | 481 | def _unpackmanifests(self, repo, revmap, trp, prog): |
|
482 | 482 | super(cg3unpacker, self)._unpackmanifests(repo, revmap, trp, prog) |
|
483 | 483 | for chunkdata in iter(self.filelogheader, {}): |
|
484 | 484 | # If we get here, there are directory manifests in the changegroup |
|
485 | 485 | d = chunkdata["filename"] |
|
486 | 486 | repo.ui.debug("adding %s revisions\n" % d) |
|
487 | 487 | dirlog = repo.manifestlog._revlog.dirlog(d) |
|
488 | 488 | deltas = self.deltaiter() |
|
489 | 489 | if not dirlog.addgroup(deltas, revmap, trp): |
|
490 | 490 | raise error.Abort(_("received dir revlog group is empty")) |
|
491 | 491 | |
|
492 | 492 | class headerlessfixup(object): |
|
493 | 493 | def __init__(self, fh, h): |
|
494 | 494 | self._h = h |
|
495 | 495 | self._fh = fh |
|
496 | 496 | def read(self, n): |
|
497 | 497 | if self._h: |
|
498 | 498 | d, self._h = self._h[:n], self._h[n:] |
|
499 | 499 | if len(d) < n: |
|
500 | 500 | d += readexactly(self._fh, n - len(d)) |
|
501 | 501 | return d |
|
502 | 502 | return readexactly(self._fh, n) |
|
503 | 503 | |
|
504 | 504 | @attr.s(slots=True, frozen=True) |
|
505 | 505 | class revisiondelta(object): |
|
506 | 506 | """Describes a delta entry in a changegroup. |
|
507 | 507 | |
|
508 | 508 | Captured data is sufficient to serialize the delta into multiple |
|
509 | 509 | formats. |
|
510 | 510 | |
|
511 | 511 | ``revision`` and ``delta`` are mutually exclusive. |
|
512 | 512 | """ |
|
513 | 513 | # 20 byte node of this revision. |
|
514 | 514 | node = attr.ib() |
|
515 | 515 | # 20 byte nodes of parent revisions. |
|
516 | 516 | p1node = attr.ib() |
|
517 | 517 | p2node = attr.ib() |
|
518 | 518 | # 20 byte node of node this delta is against. |
|
519 | 519 | basenode = attr.ib() |
|
520 | 520 | # 20 byte node of changeset revision this delta is associated with. |
|
521 | 521 | linknode = attr.ib() |
|
522 | 522 | # 2 bytes of flags to apply to revision data. |
|
523 | 523 | flags = attr.ib() |
|
524 | 524 | # Size of base revision this delta is against. May be None if |
|
525 | 525 | # basenode is nullid. |
|
526 | 526 | baserevisionsize = attr.ib() |
|
527 | 527 | # Raw fulltext revision data. |
|
528 | 528 | revision = attr.ib() |
|
529 | 529 | # Delta between the basenode and node. |
|
530 | 530 | delta = attr.ib() |
|
531 | 531 | |
|
532 | 532 | def _revisiondeltatochunks(delta, headerfn): |
|
533 | 533 | """Serialize a revisiondelta to changegroup chunks.""" |
|
534 | 534 | |
|
535 | 535 | # The captured revision delta may be encoded as a delta against |
|
536 | 536 | # a base revision or as a full revision. The changegroup format |
|
537 | 537 | # requires that everything on the wire be deltas. So for full |
|
538 | 538 | # revisions, we need to invent a header that says to rewrite |
|
539 | 539 | # data. |
|
540 | 540 | |
|
541 | 541 | if delta.delta is not None: |
|
542 | 542 | prefix, data = b'', delta.delta |
|
543 | 543 | elif delta.basenode == nullid: |
|
544 | 544 | data = delta.revision |
|
545 | 545 | prefix = mdiff.trivialdiffheader(len(data)) |
|
546 | 546 | else: |
|
547 | 547 | data = delta.revision |
|
548 | 548 | prefix = mdiff.replacediffheader(delta.baserevisionsize, |
|
549 | 549 | len(data)) |
|
550 | 550 | |
|
551 | 551 | meta = headerfn(delta) |
|
552 | 552 | |
|
553 | 553 | yield chunkheader(len(meta) + len(prefix) + len(data)) |
|
554 | 554 | yield meta |
|
555 | 555 | if prefix: |
|
556 | 556 | yield prefix |
|
557 | 557 | yield data |
|
558 | 558 | |
|
559 | 559 | def _sortnodesnormal(store, nodes, reorder): |
|
560 | 560 | """Sort nodes for changegroup generation and turn into revnums.""" |
|
561 | 561 | # for generaldelta revlogs, we linearize the revs; this will both be |
|
562 | 562 | # much quicker and generate a much smaller bundle |
|
563 | 563 | if (store._generaldelta and reorder is None) or reorder: |
|
564 | 564 | dag = dagutil.revlogdag(store) |
|
565 | 565 | return dag.linearize(set(store.rev(n) for n in nodes)) |
|
566 | 566 | else: |
|
567 | 567 | return sorted([store.rev(n) for n in nodes]) |
|
568 | 568 | |
|
569 | 569 | def _sortnodesellipsis(store, nodes, cl, lookup): |
|
570 | 570 | """Sort nodes for changegroup generation and turn into revnums.""" |
|
571 | 571 | # Ellipses serving mode. |
|
572 | 572 | # |
|
573 | 573 | # In a perfect world, we'd generate better ellipsis-ified graphs |
|
574 | 574 | # for non-changelog revlogs. In practice, we haven't started doing |
|
575 | 575 | # that yet, so the resulting DAGs for the manifestlog and filelogs |
|
576 | 576 | # are actually full of bogus parentage on all the ellipsis |
|
577 | 577 | # nodes. This has the side effect that, while the contents are |
|
578 | 578 | # correct, the individual DAGs might be completely out of whack in |
|
579 | 579 | # a case like 882681bc3166 and its ancestors (back about 10 |
|
580 | 580 | # revisions or so) in the main hg repo. |
|
581 | 581 | # |
|
582 | 582 | # The one invariant we *know* holds is that the new (potentially |
|
583 | 583 | # bogus) DAG shape will be valid if we order the nodes in the |
|
584 | 584 | # order that they're introduced in dramatis personae by the |
|
585 | 585 | # changelog, so what we do is we sort the non-changelog histories |
|
586 | 586 | # by the order in which they are used by the changelog. |
|
587 | 587 | key = lambda n: cl.rev(lookup(n)) |
|
588 | 588 | return [store.rev(n) for n in sorted(nodes, key=key)] |
|
589 | 589 | |
|
590 |
def _revisiondeltanormal(store, rev, prev, linknode, deltaparent |
|
|
590 | def _revisiondeltanormal(store, rev, prev, linknode, forcedeltaparentprev): | |
|
591 | 591 | """Construct a revision delta for non-ellipses changegroup generation.""" |
|
592 | 592 | node = store.node(rev) |
|
593 | 593 | p1, p2 = store.parentrevs(rev) |
|
594 | base = deltaparentfn(store, rev, p1, p2, prev) | |
|
594 | ||
|
595 | if forcedeltaparentprev: | |
|
596 | base = prev | |
|
597 | else: | |
|
598 | dp = store.deltaparent(rev) | |
|
599 | ||
|
600 | if dp == nullrev and store.storedeltachains: | |
|
601 | # Avoid sending full revisions when delta parent is null. Pick prev | |
|
602 | # in that case. It's tempting to pick p1 in this case, as p1 will | |
|
603 | # be smaller in the common case. However, computing a delta against | |
|
604 | # p1 may require resolving the raw text of p1, which could be | |
|
605 | # expensive. The revlog caches should have prev cached, meaning | |
|
606 | # less CPU for changegroup generation. There is likely room to add | |
|
607 | # a flag and/or config option to control this behavior. | |
|
608 | base = prev | |
|
609 | elif dp == nullrev: | |
|
610 | # revlog is configured to use full snapshot for a reason, | |
|
611 | # stick to full snapshot. | |
|
612 | base = nullrev | |
|
613 | elif dp not in (p1, p2, prev): | |
|
614 | # Pick prev when we can't be sure remote has the base revision. | |
|
615 | base = prev | |
|
616 | else: | |
|
617 | base = dp | |
|
618 | ||
|
619 | if base != nullrev and not store.candelta(base, rev): | |
|
620 | base = nullrev | |
|
595 | 621 | |
|
596 | 622 | revision = None |
|
597 | 623 | delta = None |
|
598 | 624 | baserevisionsize = None |
|
599 | 625 | |
|
600 | 626 | if store.iscensored(base) or store.iscensored(rev): |
|
601 | 627 | try: |
|
602 | 628 | revision = store.revision(node, raw=True) |
|
603 | 629 | except error.CensoredNodeError as e: |
|
604 | 630 | revision = e.tombstone |
|
605 | 631 | |
|
606 | 632 | if base != nullrev: |
|
607 | 633 | baserevisionsize = store.rawsize(base) |
|
608 | 634 | |
|
609 | 635 | elif base == nullrev: |
|
610 | 636 | revision = store.revision(node, raw=True) |
|
611 | 637 | else: |
|
612 | 638 | delta = store.revdiff(base, rev) |
|
613 | 639 | |
|
614 | 640 | p1n, p2n = store.parents(node) |
|
615 | 641 | |
|
616 | 642 | return revisiondelta( |
|
617 | 643 | node=node, |
|
618 | 644 | p1node=p1n, |
|
619 | 645 | p2node=p2n, |
|
620 | 646 | basenode=store.node(base), |
|
621 | 647 | linknode=linknode, |
|
622 | 648 | flags=store.flags(rev), |
|
623 | 649 | baserevisionsize=baserevisionsize, |
|
624 | 650 | revision=revision, |
|
625 | 651 | delta=delta, |
|
626 | 652 | ) |
|
627 | 653 | |
|
628 | 654 | def _revisiondeltanarrow(cl, store, ischangelog, rev, linkrev, |
|
629 | 655 | linknode, clrevtolocalrev, fullclnodes, |
|
630 | 656 | precomputedellipsis): |
|
631 | 657 | linkparents = precomputedellipsis[linkrev] |
|
632 | 658 | def local(clrev): |
|
633 | 659 | """Turn a changelog revnum into a local revnum. |
|
634 | 660 | |
|
635 | 661 | The ellipsis dag is stored as revnums on the changelog, |
|
636 | 662 | but when we're producing ellipsis entries for |
|
637 | 663 | non-changelog revlogs, we need to turn those numbers into |
|
638 | 664 | something local. This does that for us, and during the |
|
639 | 665 | changelog sending phase will also expand the stored |
|
640 | 666 | mappings as needed. |
|
641 | 667 | """ |
|
642 | 668 | if clrev == nullrev: |
|
643 | 669 | return nullrev |
|
644 | 670 | |
|
645 | 671 | if ischangelog: |
|
646 | 672 | return clrev |
|
647 | 673 | |
|
648 | 674 | # Walk the ellipsis-ized changelog breadth-first looking for a |
|
649 | 675 | # change that has been linked from the current revlog. |
|
650 | 676 | # |
|
651 | 677 | # For a flat manifest revlog only a single step should be necessary |
|
652 | 678 | # as all relevant changelog entries are relevant to the flat |
|
653 | 679 | # manifest. |
|
654 | 680 | # |
|
655 | 681 | # For a filelog or tree manifest dirlog however not every changelog |
|
656 | 682 | # entry will have been relevant, so we need to skip some changelog |
|
657 | 683 | # nodes even after ellipsis-izing. |
|
658 | 684 | walk = [clrev] |
|
659 | 685 | while walk: |
|
660 | 686 | p = walk[0] |
|
661 | 687 | walk = walk[1:] |
|
662 | 688 | if p in clrevtolocalrev: |
|
663 | 689 | return clrevtolocalrev[p] |
|
664 | 690 | elif p in fullclnodes: |
|
665 | 691 | walk.extend([pp for pp in cl.parentrevs(p) |
|
666 | 692 | if pp != nullrev]) |
|
667 | 693 | elif p in precomputedellipsis: |
|
668 | 694 | walk.extend([pp for pp in precomputedellipsis[p] |
|
669 | 695 | if pp != nullrev]) |
|
670 | 696 | else: |
|
671 | 697 | # In this case, we've got an ellipsis with parents |
|
672 | 698 | # outside the current bundle (likely an |
|
673 | 699 | # incremental pull). We "know" that we can use the |
|
674 | 700 | # value of this same revlog at whatever revision |
|
675 | 701 | # is pointed to by linknode. "Know" is in scare |
|
676 | 702 | # quotes because I haven't done enough examination |
|
677 | 703 | # of edge cases to convince myself this is really |
|
678 | 704 | # a fact - it works for all the (admittedly |
|
679 | 705 | # thorough) cases in our testsuite, but I would be |
|
680 | 706 | # somewhat unsurprised to find a case in the wild |
|
681 | 707 | # where this breaks down a bit. That said, I don't |
|
682 | 708 | # know if it would hurt anything. |
|
683 | 709 | for i in pycompat.xrange(rev, 0, -1): |
|
684 | 710 | if store.linkrev(i) == clrev: |
|
685 | 711 | return i |
|
686 | 712 | # We failed to resolve a parent for this node, so |
|
687 | 713 | # we crash the changegroup construction. |
|
688 | 714 | raise error.Abort( |
|
689 | 715 | 'unable to resolve parent while packing %r %r' |
|
690 | 716 | ' for changeset %r' % (store.indexfile, rev, clrev)) |
|
691 | 717 | |
|
692 | 718 | return nullrev |
|
693 | 719 | |
|
694 | 720 | if not linkparents or ( |
|
695 | 721 | store.parentrevs(rev) == (nullrev, nullrev)): |
|
696 | 722 | p1, p2 = nullrev, nullrev |
|
697 | 723 | elif len(linkparents) == 1: |
|
698 | 724 | p1, = sorted(local(p) for p in linkparents) |
|
699 | 725 | p2 = nullrev |
|
700 | 726 | else: |
|
701 | 727 | p1, p2 = sorted(local(p) for p in linkparents) |
|
702 | 728 | |
|
703 | 729 | n = store.node(rev) |
|
704 | 730 | p1n, p2n = store.node(p1), store.node(p2) |
|
705 | 731 | flags = store.flags(rev) |
|
706 | 732 | flags |= revlog.REVIDX_ELLIPSIS |
|
707 | 733 | |
|
708 | 734 | # TODO: try and actually send deltas for ellipsis data blocks |
|
709 | 735 | |
|
710 | 736 | return revisiondelta( |
|
711 | 737 | node=n, |
|
712 | 738 | p1node=p1n, |
|
713 | 739 | p2node=p2n, |
|
714 | 740 | basenode=nullid, |
|
715 | 741 | linknode=linknode, |
|
716 | 742 | flags=flags, |
|
717 | 743 | baserevisionsize=None, |
|
718 | 744 | revision=store.revision(n), |
|
719 | 745 | delta=None, |
|
720 | 746 | ) |
|
721 | 747 | |
|
722 |
def deltagroup(repo, revs, store, ischangelog, lookup, deltaparent |
|
|
748 | def deltagroup(repo, revs, store, ischangelog, lookup, forcedeltaparentprev, | |
|
723 | 749 | units=None, |
|
724 | 750 | ellipses=False, clrevtolocalrev=None, fullclnodes=None, |
|
725 | 751 | precomputedellipsis=None): |
|
726 | 752 | """Calculate deltas for a set of revisions. |
|
727 | 753 | |
|
728 | 754 | Is a generator of ``revisiondelta`` instances. |
|
729 | 755 | |
|
730 | 756 | If units is not None, progress detail will be generated, units specifies |
|
731 | 757 | the type of revlog that is touched (changelog, manifest, etc.). |
|
732 | 758 | """ |
|
733 | 759 | if not revs: |
|
734 | 760 | return |
|
735 | 761 | |
|
736 | 762 | cl = repo.changelog |
|
737 | 763 | |
|
738 | 764 | # Add the parent of the first rev. |
|
739 | 765 | revs.insert(0, store.parentrevs(revs[0])[0]) |
|
740 | 766 | |
|
741 | 767 | # build deltas |
|
742 | 768 | progress = None |
|
743 | 769 | if units is not None: |
|
744 | 770 | progress = repo.ui.makeprogress(_('bundling'), unit=units, |
|
745 | 771 | total=(len(revs) - 1)) |
|
746 | 772 | |
|
747 | 773 | for i in pycompat.xrange(len(revs) - 1): |
|
748 | 774 | if progress: |
|
749 | 775 | progress.update(i + 1) |
|
750 | 776 | |
|
751 | 777 | prev = revs[i] |
|
752 | 778 | curr = revs[i + 1] |
|
753 | 779 | |
|
754 | 780 | linknode = lookup(store.node(curr)) |
|
755 | 781 | |
|
756 | 782 | if ellipses: |
|
757 | 783 | linkrev = cl.rev(linknode) |
|
758 | 784 | clrevtolocalrev[linkrev] = curr |
|
759 | 785 | |
|
760 | 786 | # This is a node to send in full, because the changeset it |
|
761 | 787 | # corresponds to was a full changeset. |
|
762 | 788 | if linknode in fullclnodes: |
|
763 | 789 | delta = _revisiondeltanormal(store, curr, prev, linknode, |
|
764 |
deltaparent |
|
|
790 | forcedeltaparentprev) | |
|
765 | 791 | elif linkrev not in precomputedellipsis: |
|
766 | 792 | delta = None |
|
767 | 793 | else: |
|
768 | 794 | delta = _revisiondeltanarrow( |
|
769 | 795 | cl, store, ischangelog, curr, linkrev, linknode, |
|
770 | 796 | clrevtolocalrev, fullclnodes, |
|
771 | 797 | precomputedellipsis) |
|
772 | 798 | else: |
|
773 | 799 | delta = _revisiondeltanormal(store, curr, prev, linknode, |
|
774 |
deltaparent |
|
|
800 | forcedeltaparentprev) | |
|
775 | 801 | |
|
776 | 802 | if delta: |
|
777 | 803 | yield delta |
|
778 | 804 | |
|
779 | 805 | if progress: |
|
780 | 806 | progress.complete() |
|
781 | 807 | |
|
782 | 808 | class cgpacker(object): |
|
783 | 809 | def __init__(self, repo, filematcher, version, allowreorder, |
|
784 |
|
|
|
810 | builddeltaheader, manifestsend, | |
|
811 | forcedeltaparentprev=False, | |
|
785 | 812 | bundlecaps=None, ellipses=False, |
|
786 | 813 | shallow=False, ellipsisroots=None, fullnodes=None): |
|
787 | 814 | """Given a source repo, construct a bundler. |
|
788 | 815 | |
|
789 | 816 | filematcher is a matcher that matches on files to include in the |
|
790 | 817 | changegroup. Used to facilitate sparse changegroups. |
|
791 | 818 | |
|
792 | 819 | allowreorder controls whether reordering of revisions is allowed. |
|
793 | 820 | This value is used when ``bundle.reorder`` is ``auto`` or isn't |
|
794 | 821 | set. |
|
795 | 822 | |
|
796 |
deltaparent |
|
|
797 | a specific revision. | |
|
823 | forcedeltaparentprev indicates whether delta parents must be against | |
|
824 | the previous revision in a delta group. This should only be used for | |
|
825 | compatibility with changegroup version 1. | |
|
798 | 826 | |
|
799 | 827 | builddeltaheader is a callable that constructs the header for a group |
|
800 | 828 | delta. |
|
801 | 829 | |
|
802 | 830 | manifestsend is a chunk to send after manifests have been fully emitted. |
|
803 | 831 | |
|
804 | 832 | ellipses indicates whether ellipsis serving mode is enabled. |
|
805 | 833 | |
|
806 | 834 | bundlecaps is optional and can be used to specify the set of |
|
807 | 835 | capabilities which can be used to build the bundle. While bundlecaps is |
|
808 | 836 | unused in core Mercurial, extensions rely on this feature to communicate |
|
809 | 837 | capabilities to customize the changegroup packer. |
|
810 | 838 | |
|
811 | 839 | shallow indicates whether shallow data might be sent. The packer may |
|
812 | 840 | need to pack file contents not introduced by the changes being packed. |
|
813 | 841 | |
|
814 | 842 | fullnodes is the set of changelog nodes which should not be ellipsis |
|
815 | 843 | nodes. We store this rather than the set of nodes that should be |
|
816 | 844 | ellipsis because for very large histories we expect this to be |
|
817 | 845 | significantly smaller. |
|
818 | 846 | """ |
|
819 | 847 | assert filematcher |
|
820 | 848 | self._filematcher = filematcher |
|
821 | 849 | |
|
822 | 850 | self.version = version |
|
823 |
self._deltaparent |
|
|
851 | self._forcedeltaparentprev = forcedeltaparentprev | |
|
824 | 852 | self._builddeltaheader = builddeltaheader |
|
825 | 853 | self._manifestsend = manifestsend |
|
826 | 854 | self._ellipses = ellipses |
|
827 | 855 | |
|
828 | 856 | # Set of capabilities we can use to build the bundle. |
|
829 | 857 | if bundlecaps is None: |
|
830 | 858 | bundlecaps = set() |
|
831 | 859 | self._bundlecaps = bundlecaps |
|
832 | 860 | self._isshallow = shallow |
|
833 | 861 | self._fullclnodes = fullnodes |
|
834 | 862 | |
|
835 | 863 | # Maps ellipsis revs to their roots at the changelog level. |
|
836 | 864 | self._precomputedellipsis = ellipsisroots |
|
837 | 865 | |
|
838 | 866 | # experimental config: bundle.reorder |
|
839 | 867 | reorder = repo.ui.config('bundle', 'reorder') |
|
840 | 868 | if reorder == 'auto': |
|
841 | 869 | self._reorder = allowreorder |
|
842 | 870 | else: |
|
843 | 871 | self._reorder = stringutil.parsebool(reorder) |
|
844 | 872 | |
|
845 | 873 | self._repo = repo |
|
846 | 874 | |
|
847 | 875 | if self._repo.ui.verbose and not self._repo.ui.debugflag: |
|
848 | 876 | self._verbosenote = self._repo.ui.note |
|
849 | 877 | else: |
|
850 | 878 | self._verbosenote = lambda s: None |
|
851 | 879 | |
|
852 | 880 | def generate(self, commonrevs, clnodes, fastpathlinkrev, source): |
|
853 | 881 | """Yield a sequence of changegroup byte chunks.""" |
|
854 | 882 | |
|
855 | 883 | repo = self._repo |
|
856 | 884 | cl = repo.changelog |
|
857 | 885 | |
|
858 | 886 | self._verbosenote(_('uncompressed size of bundle content:\n')) |
|
859 | 887 | size = 0 |
|
860 | 888 | |
|
861 | 889 | clstate, deltas = self._generatechangelog(cl, clnodes) |
|
862 | 890 | for delta in deltas: |
|
863 | 891 | for chunk in _revisiondeltatochunks(delta, self._builddeltaheader): |
|
864 | 892 | size += len(chunk) |
|
865 | 893 | yield chunk |
|
866 | 894 | |
|
867 | 895 | close = closechunk() |
|
868 | 896 | size += len(close) |
|
869 | 897 | yield closechunk() |
|
870 | 898 | |
|
871 | 899 | self._verbosenote(_('%8.i (changelog)\n') % size) |
|
872 | 900 | |
|
873 | 901 | clrevorder = clstate['clrevorder'] |
|
874 | 902 | mfs = clstate['mfs'] |
|
875 | 903 | changedfiles = clstate['changedfiles'] |
|
876 | 904 | |
|
877 | 905 | # We need to make sure that the linkrev in the changegroup refers to |
|
878 | 906 | # the first changeset that introduced the manifest or file revision. |
|
879 | 907 | # The fastpath is usually safer than the slowpath, because the filelogs |
|
880 | 908 | # are walked in revlog order. |
|
881 | 909 | # |
|
882 | 910 | # When taking the slowpath with reorder=None and the manifest revlog |
|
883 | 911 | # uses generaldelta, the manifest may be walked in the "wrong" order. |
|
884 | 912 | # Without 'clrevorder', we would get an incorrect linkrev (see fix in |
|
885 | 913 | # cc0ff93d0c0c). |
|
886 | 914 | # |
|
887 | 915 | # When taking the fastpath, we are only vulnerable to reordering |
|
888 | 916 | # of the changelog itself. The changelog never uses generaldelta, so |
|
889 | 917 | # it is only reordered when reorder=True. To handle this case, we |
|
890 | 918 | # simply take the slowpath, which already has the 'clrevorder' logic. |
|
891 | 919 | # This was also fixed in cc0ff93d0c0c. |
|
892 | 920 | fastpathlinkrev = fastpathlinkrev and not self._reorder |
|
893 | 921 | # Treemanifests don't work correctly with fastpathlinkrev |
|
894 | 922 | # either, because we don't discover which directory nodes to |
|
895 | 923 | # send along with files. This could probably be fixed. |
|
896 | 924 | fastpathlinkrev = fastpathlinkrev and ( |
|
897 | 925 | 'treemanifest' not in repo.requirements) |
|
898 | 926 | |
|
899 | 927 | fnodes = {} # needed file nodes |
|
900 | 928 | |
|
901 | 929 | size = 0 |
|
902 | 930 | it = self.generatemanifests( |
|
903 | 931 | commonrevs, clrevorder, fastpathlinkrev, mfs, fnodes, source, |
|
904 | 932 | clstate['clrevtomanifestrev']) |
|
905 | 933 | |
|
906 | 934 | for dir, deltas in it: |
|
907 | 935 | if dir: |
|
908 | 936 | assert self.version == b'03' |
|
909 | 937 | chunk = _fileheader(dir) |
|
910 | 938 | size += len(chunk) |
|
911 | 939 | yield chunk |
|
912 | 940 | |
|
913 | 941 | for delta in deltas: |
|
914 | 942 | chunks = _revisiondeltatochunks(delta, self._builddeltaheader) |
|
915 | 943 | for chunk in chunks: |
|
916 | 944 | size += len(chunk) |
|
917 | 945 | yield chunk |
|
918 | 946 | |
|
919 | 947 | close = closechunk() |
|
920 | 948 | size += len(close) |
|
921 | 949 | yield close |
|
922 | 950 | |
|
923 | 951 | self._verbosenote(_('%8.i (manifests)\n') % size) |
|
924 | 952 | yield self._manifestsend |
|
925 | 953 | |
|
926 | 954 | mfdicts = None |
|
927 | 955 | if self._ellipses and self._isshallow: |
|
928 | 956 | mfdicts = [(self._repo.manifestlog[n].read(), lr) |
|
929 | 957 | for (n, lr) in mfs.iteritems()] |
|
930 | 958 | |
|
931 | 959 | mfs.clear() |
|
932 | 960 | clrevs = set(cl.rev(x) for x in clnodes) |
|
933 | 961 | |
|
934 | 962 | it = self.generatefiles(changedfiles, commonrevs, |
|
935 | 963 | source, mfdicts, fastpathlinkrev, |
|
936 | 964 | fnodes, clrevs) |
|
937 | 965 | |
|
938 | 966 | for path, deltas in it: |
|
939 | 967 | h = _fileheader(path) |
|
940 | 968 | size = len(h) |
|
941 | 969 | yield h |
|
942 | 970 | |
|
943 | 971 | for delta in deltas: |
|
944 | 972 | chunks = _revisiondeltatochunks(delta, self._builddeltaheader) |
|
945 | 973 | for chunk in chunks: |
|
946 | 974 | size += len(chunk) |
|
947 | 975 | yield chunk |
|
948 | 976 | |
|
949 | 977 | close = closechunk() |
|
950 | 978 | size += len(close) |
|
951 | 979 | yield close |
|
952 | 980 | |
|
953 | 981 | self._verbosenote(_('%8.i %s\n') % (size, path)) |
|
954 | 982 | |
|
955 | 983 | yield closechunk() |
|
956 | 984 | |
|
957 | 985 | if clnodes: |
|
958 | 986 | repo.hook('outgoing', node=hex(clnodes[0]), source=source) |
|
959 | 987 | |
|
960 | 988 | def _generatechangelog(self, cl, nodes): |
|
961 | 989 | """Generate data for changelog chunks. |
|
962 | 990 | |
|
963 | 991 | Returns a 2-tuple of a dict containing state and an iterable of |
|
964 | 992 | byte chunks. The state will not be fully populated until the |
|
965 | 993 | chunk stream has been fully consumed. |
|
966 | 994 | """ |
|
967 | 995 | clrevorder = {} |
|
968 | 996 | mfs = {} # needed manifests |
|
969 | 997 | mfl = self._repo.manifestlog |
|
970 | 998 | # TODO violates storage abstraction. |
|
971 | 999 | mfrevlog = mfl._revlog |
|
972 | 1000 | changedfiles = set() |
|
973 | 1001 | clrevtomanifestrev = {} |
|
974 | 1002 | |
|
975 | 1003 | # Callback for the changelog, used to collect changed files and |
|
976 | 1004 | # manifest nodes. |
|
977 | 1005 | # Returns the linkrev node (identity in the changelog case). |
|
978 | 1006 | def lookupcl(x): |
|
979 | 1007 | c = cl.read(x) |
|
980 | 1008 | clrevorder[x] = len(clrevorder) |
|
981 | 1009 | |
|
982 | 1010 | if self._ellipses: |
|
983 | 1011 | # Only update mfs if x is going to be sent. Otherwise we |
|
984 | 1012 | # end up with bogus linkrevs specified for manifests and |
|
985 | 1013 | # we skip some manifest nodes that we should otherwise |
|
986 | 1014 | # have sent. |
|
987 | 1015 | if (x in self._fullclnodes |
|
988 | 1016 | or cl.rev(x) in self._precomputedellipsis): |
|
989 | 1017 | n = c[0] |
|
990 | 1018 | # Record the first changeset introducing this manifest |
|
991 | 1019 | # version. |
|
992 | 1020 | mfs.setdefault(n, x) |
|
993 | 1021 | # Set this narrow-specific dict so we have the lowest |
|
994 | 1022 | # manifest revnum to look up for this cl revnum. (Part of |
|
995 | 1023 | # mapping changelog ellipsis parents to manifest ellipsis |
|
996 | 1024 | # parents) |
|
997 | 1025 | clrevtomanifestrev.setdefault(cl.rev(x), mfrevlog.rev(n)) |
|
998 | 1026 | # We can't trust the changed files list in the changeset if the |
|
999 | 1027 | # client requested a shallow clone. |
|
1000 | 1028 | if self._isshallow: |
|
1001 | 1029 | changedfiles.update(mfl[c[0]].read().keys()) |
|
1002 | 1030 | else: |
|
1003 | 1031 | changedfiles.update(c[3]) |
|
1004 | 1032 | else: |
|
1005 | 1033 | |
|
1006 | 1034 | n = c[0] |
|
1007 | 1035 | # record the first changeset introducing this manifest version |
|
1008 | 1036 | mfs.setdefault(n, x) |
|
1009 | 1037 | # Record a complete list of potentially-changed files in |
|
1010 | 1038 | # this manifest. |
|
1011 | 1039 | changedfiles.update(c[3]) |
|
1012 | 1040 | |
|
1013 | 1041 | return x |
|
1014 | 1042 | |
|
1015 | 1043 | # Changelog doesn't benefit from reordering revisions. So send out |
|
1016 | 1044 | # revisions in store order. |
|
1017 | 1045 | revs = sorted(cl.rev(n) for n in nodes) |
|
1018 | 1046 | |
|
1019 | 1047 | state = { |
|
1020 | 1048 | 'clrevorder': clrevorder, |
|
1021 | 1049 | 'mfs': mfs, |
|
1022 | 1050 | 'changedfiles': changedfiles, |
|
1023 | 1051 | 'clrevtomanifestrev': clrevtomanifestrev, |
|
1024 | 1052 | } |
|
1025 | 1053 | |
|
1026 | 1054 | gen = deltagroup( |
|
1027 | 1055 | self._repo, revs, cl, True, lookupcl, |
|
1028 |
self._deltaparent |
|
|
1056 | self._forcedeltaparentprev, | |
|
1029 | 1057 | ellipses=self._ellipses, |
|
1030 | 1058 | units=_('changesets'), |
|
1031 | 1059 | clrevtolocalrev={}, |
|
1032 | 1060 | fullclnodes=self._fullclnodes, |
|
1033 | 1061 | precomputedellipsis=self._precomputedellipsis) |
|
1034 | 1062 | |
|
1035 | 1063 | return state, gen |
|
1036 | 1064 | |
|
1037 | 1065 | def generatemanifests(self, commonrevs, clrevorder, fastpathlinkrev, mfs, |
|
1038 | 1066 | fnodes, source, clrevtolocalrev): |
|
1039 | 1067 | """Returns an iterator of changegroup chunks containing manifests. |
|
1040 | 1068 | |
|
1041 | 1069 | `source` is unused here, but is used by extensions like remotefilelog to |
|
1042 | 1070 | change what is sent based in pulls vs pushes, etc. |
|
1043 | 1071 | """ |
|
1044 | 1072 | repo = self._repo |
|
1045 | 1073 | cl = repo.changelog |
|
1046 | 1074 | mfl = repo.manifestlog |
|
1047 | 1075 | dirlog = mfl._revlog.dirlog |
|
1048 | 1076 | tmfnodes = {'': mfs} |
|
1049 | 1077 | |
|
1050 | 1078 | # Callback for the manifest, used to collect linkrevs for filelog |
|
1051 | 1079 | # revisions. |
|
1052 | 1080 | # Returns the linkrev node (collected in lookupcl). |
|
1053 | 1081 | def makelookupmflinknode(dir, nodes): |
|
1054 | 1082 | if fastpathlinkrev: |
|
1055 | 1083 | assert not dir |
|
1056 | 1084 | return mfs.__getitem__ |
|
1057 | 1085 | |
|
1058 | 1086 | def lookupmflinknode(x): |
|
1059 | 1087 | """Callback for looking up the linknode for manifests. |
|
1060 | 1088 | |
|
1061 | 1089 | Returns the linkrev node for the specified manifest. |
|
1062 | 1090 | |
|
1063 | 1091 | SIDE EFFECT: |
|
1064 | 1092 | |
|
1065 | 1093 | 1) fclnodes gets populated with the list of relevant |
|
1066 | 1094 | file nodes if we're not using fastpathlinkrev |
|
1067 | 1095 | 2) When treemanifests are in use, collects treemanifest nodes |
|
1068 | 1096 | to send |
|
1069 | 1097 | |
|
1070 | 1098 | Note that this means manifests must be completely sent to |
|
1071 | 1099 | the client before you can trust the list of files and |
|
1072 | 1100 | treemanifests to send. |
|
1073 | 1101 | """ |
|
1074 | 1102 | clnode = nodes[x] |
|
1075 | 1103 | mdata = mfl.get(dir, x).readfast(shallow=True) |
|
1076 | 1104 | for p, n, fl in mdata.iterentries(): |
|
1077 | 1105 | if fl == 't': # subdirectory manifest |
|
1078 | 1106 | subdir = dir + p + '/' |
|
1079 | 1107 | tmfclnodes = tmfnodes.setdefault(subdir, {}) |
|
1080 | 1108 | tmfclnode = tmfclnodes.setdefault(n, clnode) |
|
1081 | 1109 | if clrevorder[clnode] < clrevorder[tmfclnode]: |
|
1082 | 1110 | tmfclnodes[n] = clnode |
|
1083 | 1111 | else: |
|
1084 | 1112 | f = dir + p |
|
1085 | 1113 | fclnodes = fnodes.setdefault(f, {}) |
|
1086 | 1114 | fclnode = fclnodes.setdefault(n, clnode) |
|
1087 | 1115 | if clrevorder[clnode] < clrevorder[fclnode]: |
|
1088 | 1116 | fclnodes[n] = clnode |
|
1089 | 1117 | return clnode |
|
1090 | 1118 | return lookupmflinknode |
|
1091 | 1119 | |
|
1092 | 1120 | while tmfnodes: |
|
1093 | 1121 | dir, nodes = tmfnodes.popitem() |
|
1094 | 1122 | store = dirlog(dir) |
|
1095 | 1123 | |
|
1096 | 1124 | if not self._filematcher.visitdir(store._dir[:-1] or '.'): |
|
1097 | 1125 | prunednodes = [] |
|
1098 | 1126 | else: |
|
1099 | 1127 | frev, flr = store.rev, store.linkrev |
|
1100 | 1128 | prunednodes = [n for n in nodes |
|
1101 | 1129 | if flr(frev(n)) not in commonrevs] |
|
1102 | 1130 | |
|
1103 | 1131 | if dir and not prunednodes: |
|
1104 | 1132 | continue |
|
1105 | 1133 | |
|
1106 | 1134 | lookupfn = makelookupmflinknode(dir, nodes) |
|
1107 | 1135 | |
|
1108 | 1136 | if self._ellipses: |
|
1109 | 1137 | revs = _sortnodesellipsis(store, prunednodes, cl, |
|
1110 | 1138 | lookupfn) |
|
1111 | 1139 | else: |
|
1112 | 1140 | revs = _sortnodesnormal(store, prunednodes, |
|
1113 | 1141 | self._reorder) |
|
1114 | 1142 | |
|
1115 | 1143 | deltas = deltagroup( |
|
1116 | 1144 | self._repo, revs, store, False, lookupfn, |
|
1117 |
self._deltaparent |
|
|
1145 | self._forcedeltaparentprev, | |
|
1118 | 1146 | ellipses=self._ellipses, |
|
1119 | 1147 | units=_('manifests'), |
|
1120 | 1148 | clrevtolocalrev=clrevtolocalrev, |
|
1121 | 1149 | fullclnodes=self._fullclnodes, |
|
1122 | 1150 | precomputedellipsis=self._precomputedellipsis) |
|
1123 | 1151 | |
|
1124 | 1152 | yield dir, deltas |
|
1125 | 1153 | |
|
1126 | 1154 | # The 'source' parameter is useful for extensions |
|
1127 | 1155 | def generatefiles(self, changedfiles, commonrevs, source, |
|
1128 | 1156 | mfdicts, fastpathlinkrev, fnodes, clrevs): |
|
1129 | 1157 | changedfiles = list(filter(self._filematcher, changedfiles)) |
|
1130 | 1158 | |
|
1131 | 1159 | if not fastpathlinkrev: |
|
1132 | 1160 | def normallinknodes(unused, fname): |
|
1133 | 1161 | return fnodes.get(fname, {}) |
|
1134 | 1162 | else: |
|
1135 | 1163 | cln = self._repo.changelog.node |
|
1136 | 1164 | |
|
1137 | 1165 | def normallinknodes(store, fname): |
|
1138 | 1166 | flinkrev = store.linkrev |
|
1139 | 1167 | fnode = store.node |
|
1140 | 1168 | revs = ((r, flinkrev(r)) for r in store) |
|
1141 | 1169 | return dict((fnode(r), cln(lr)) |
|
1142 | 1170 | for r, lr in revs if lr in clrevs) |
|
1143 | 1171 | |
|
1144 | 1172 | clrevtolocalrev = {} |
|
1145 | 1173 | |
|
1146 | 1174 | if self._isshallow: |
|
1147 | 1175 | # In a shallow clone, the linknodes callback needs to also include |
|
1148 | 1176 | # those file nodes that are in the manifests we sent but weren't |
|
1149 | 1177 | # introduced by those manifests. |
|
1150 | 1178 | commonctxs = [self._repo[c] for c in commonrevs] |
|
1151 | 1179 | clrev = self._repo.changelog.rev |
|
1152 | 1180 | |
|
1153 | 1181 | # Defining this function has a side-effect of overriding the |
|
1154 | 1182 | # function of the same name that was passed in as an argument. |
|
1155 | 1183 | # TODO have caller pass in appropriate function. |
|
1156 | 1184 | def linknodes(flog, fname): |
|
1157 | 1185 | for c in commonctxs: |
|
1158 | 1186 | try: |
|
1159 | 1187 | fnode = c.filenode(fname) |
|
1160 | 1188 | clrevtolocalrev[c.rev()] = flog.rev(fnode) |
|
1161 | 1189 | except error.ManifestLookupError: |
|
1162 | 1190 | pass |
|
1163 | 1191 | links = normallinknodes(flog, fname) |
|
1164 | 1192 | if len(links) != len(mfdicts): |
|
1165 | 1193 | for mf, lr in mfdicts: |
|
1166 | 1194 | fnode = mf.get(fname, None) |
|
1167 | 1195 | if fnode in links: |
|
1168 | 1196 | links[fnode] = min(links[fnode], lr, key=clrev) |
|
1169 | 1197 | elif fnode: |
|
1170 | 1198 | links[fnode] = lr |
|
1171 | 1199 | return links |
|
1172 | 1200 | else: |
|
1173 | 1201 | linknodes = normallinknodes |
|
1174 | 1202 | |
|
1175 | 1203 | repo = self._repo |
|
1176 | 1204 | cl = repo.changelog |
|
1177 | 1205 | progress = repo.ui.makeprogress(_('bundling'), unit=_('files'), |
|
1178 | 1206 | total=len(changedfiles)) |
|
1179 | 1207 | for i, fname in enumerate(sorted(changedfiles)): |
|
1180 | 1208 | filerevlog = repo.file(fname) |
|
1181 | 1209 | if not filerevlog: |
|
1182 | 1210 | raise error.Abort(_("empty or missing file data for %s") % |
|
1183 | 1211 | fname) |
|
1184 | 1212 | |
|
1185 | 1213 | clrevtolocalrev.clear() |
|
1186 | 1214 | |
|
1187 | 1215 | linkrevnodes = linknodes(filerevlog, fname) |
|
1188 | 1216 | # Lookup for filenodes, we collected the linkrev nodes above in the |
|
1189 | 1217 | # fastpath case and with lookupmf in the slowpath case. |
|
1190 | 1218 | def lookupfilelog(x): |
|
1191 | 1219 | return linkrevnodes[x] |
|
1192 | 1220 | |
|
1193 | 1221 | frev, flr = filerevlog.rev, filerevlog.linkrev |
|
1194 | 1222 | filenodes = [n for n in linkrevnodes |
|
1195 | 1223 | if flr(frev(n)) not in commonrevs] |
|
1196 | 1224 | |
|
1197 | 1225 | if filenodes: |
|
1198 | 1226 | if self._ellipses: |
|
1199 | 1227 | revs = _sortnodesellipsis(filerevlog, filenodes, |
|
1200 | 1228 | cl, lookupfilelog) |
|
1201 | 1229 | else: |
|
1202 | 1230 | revs = _sortnodesnormal(filerevlog, filenodes, |
|
1203 | 1231 | self._reorder) |
|
1204 | 1232 | |
|
1205 | 1233 | progress.update(i + 1, item=fname) |
|
1206 | 1234 | |
|
1207 | 1235 | deltas = deltagroup( |
|
1208 | 1236 | self._repo, revs, filerevlog, False, lookupfilelog, |
|
1209 |
self._deltaparent |
|
|
1237 | self._forcedeltaparentprev, | |
|
1210 | 1238 | ellipses=self._ellipses, |
|
1211 | 1239 | clrevtolocalrev=clrevtolocalrev, |
|
1212 | 1240 | fullclnodes=self._fullclnodes, |
|
1213 | 1241 | precomputedellipsis=self._precomputedellipsis) |
|
1214 | 1242 | |
|
1215 | 1243 | yield fname, deltas |
|
1216 | 1244 | |
|
1217 | 1245 | progress.complete() |
|
1218 | 1246 | |
|
1219 | def _deltaparentprev(store, rev, p1, p2, prev): | |
|
1220 | """Resolve a delta parent to the previous revision. | |
|
1221 | ||
|
1222 | Used for version 1 changegroups, which don't support generaldelta. | |
|
1223 | """ | |
|
1224 | return prev | |
|
1225 | ||
|
1226 | def _deltaparentgeneraldelta(store, rev, p1, p2, prev): | |
|
1227 | """Resolve a delta parent when general deltas are supported.""" | |
|
1228 | dp = store.deltaparent(rev) | |
|
1229 | if dp == nullrev and store.storedeltachains: | |
|
1230 | # Avoid sending full revisions when delta parent is null. Pick prev | |
|
1231 | # in that case. It's tempting to pick p1 in this case, as p1 will | |
|
1232 | # be smaller in the common case. However, computing a delta against | |
|
1233 | # p1 may require resolving the raw text of p1, which could be | |
|
1234 | # expensive. The revlog caches should have prev cached, meaning | |
|
1235 | # less CPU for changegroup generation. There is likely room to add | |
|
1236 | # a flag and/or config option to control this behavior. | |
|
1237 | base = prev | |
|
1238 | elif dp == nullrev: | |
|
1239 | # revlog is configured to use full snapshot for a reason, | |
|
1240 | # stick to full snapshot. | |
|
1241 | base = nullrev | |
|
1242 | elif dp not in (p1, p2, prev): | |
|
1243 | # Pick prev when we can't be sure remote has the base revision. | |
|
1244 | return prev | |
|
1245 | else: | |
|
1246 | base = dp | |
|
1247 | ||
|
1248 | if base != nullrev and not store.candelta(base, rev): | |
|
1249 | base = nullrev | |
|
1250 | ||
|
1251 | return base | |
|
1252 | ||
|
1253 | def _deltaparentellipses(store, rev, p1, p2, prev): | |
|
1254 | """Resolve a delta parent when in ellipses mode.""" | |
|
1255 | # TODO: send better deltas when in narrow mode. | |
|
1256 | # | |
|
1257 | # changegroup.group() loops over revisions to send, | |
|
1258 | # including revisions we'll skip. What this means is that | |
|
1259 | # `prev` will be a potentially useless delta base for all | |
|
1260 | # ellipsis nodes, as the client likely won't have it. In | |
|
1261 | # the future we should do bookkeeping about which nodes | |
|
1262 | # have been sent to the client, and try to be | |
|
1263 | # significantly smarter about delta bases. This is | |
|
1264 | # slightly tricky because this same code has to work for | |
|
1265 | # all revlogs, and we don't have the linkrev/linknode here. | |
|
1266 | return p1 | |
|
1267 | ||
|
1268 | 1247 | def _makecg1packer(repo, filematcher, bundlecaps, ellipses=False, |
|
1269 | 1248 | shallow=False, ellipsisroots=None, fullnodes=None): |
|
1270 | 1249 | builddeltaheader = lambda d: _CHANGEGROUPV1_DELTA_HEADER.pack( |
|
1271 | 1250 | d.node, d.p1node, d.p2node, d.linknode) |
|
1272 | 1251 | |
|
1273 | 1252 | return cgpacker(repo, filematcher, b'01', |
|
1274 | deltaparentfn=_deltaparentprev, | |
|
1275 | 1253 | allowreorder=None, |
|
1276 | 1254 | builddeltaheader=builddeltaheader, |
|
1277 | 1255 | manifestsend=b'', |
|
1256 | forcedeltaparentprev=True, | |
|
1278 | 1257 | bundlecaps=bundlecaps, |
|
1279 | 1258 | ellipses=ellipses, |
|
1280 | 1259 | shallow=shallow, |
|
1281 | 1260 | ellipsisroots=ellipsisroots, |
|
1282 | 1261 | fullnodes=fullnodes) |
|
1283 | 1262 | |
|
1284 | 1263 | def _makecg2packer(repo, filematcher, bundlecaps, ellipses=False, |
|
1285 | 1264 | shallow=False, ellipsisroots=None, fullnodes=None): |
|
1286 | 1265 | builddeltaheader = lambda d: _CHANGEGROUPV2_DELTA_HEADER.pack( |
|
1287 | 1266 | d.node, d.p1node, d.p2node, d.basenode, d.linknode) |
|
1288 | 1267 | |
|
1289 | 1268 | # Since generaldelta is directly supported by cg2, reordering |
|
1290 | 1269 | # generally doesn't help, so we disable it by default (treating |
|
1291 | 1270 | # bundle.reorder=auto just like bundle.reorder=False). |
|
1292 | 1271 | return cgpacker(repo, filematcher, b'02', |
|
1293 | deltaparentfn=_deltaparentgeneraldelta, | |
|
1294 | 1272 | allowreorder=False, |
|
1295 | 1273 | builddeltaheader=builddeltaheader, |
|
1296 | 1274 | manifestsend=b'', |
|
1297 | 1275 | bundlecaps=bundlecaps, |
|
1298 | 1276 | ellipses=ellipses, |
|
1299 | 1277 | shallow=shallow, |
|
1300 | 1278 | ellipsisroots=ellipsisroots, |
|
1301 | 1279 | fullnodes=fullnodes) |
|
1302 | 1280 | |
|
1303 | 1281 | def _makecg3packer(repo, filematcher, bundlecaps, ellipses=False, |
|
1304 | 1282 | shallow=False, ellipsisroots=None, fullnodes=None): |
|
1305 | 1283 | builddeltaheader = lambda d: _CHANGEGROUPV3_DELTA_HEADER.pack( |
|
1306 | 1284 | d.node, d.p1node, d.p2node, d.basenode, d.linknode, d.flags) |
|
1307 | 1285 | |
|
1308 | deltaparentfn = (_deltaparentellipses if ellipses | |
|
1309 | else _deltaparentgeneraldelta) | |
|
1310 | ||
|
1311 | 1286 | return cgpacker(repo, filematcher, b'03', |
|
1312 | deltaparentfn=deltaparentfn, | |
|
1313 | 1287 | allowreorder=False, |
|
1314 | 1288 | builddeltaheader=builddeltaheader, |
|
1315 | 1289 | manifestsend=closechunk(), |
|
1316 | 1290 | bundlecaps=bundlecaps, |
|
1317 | 1291 | ellipses=ellipses, |
|
1318 | 1292 | shallow=shallow, |
|
1319 | 1293 | ellipsisroots=ellipsisroots, |
|
1320 | 1294 | fullnodes=fullnodes) |
|
1321 | 1295 | |
|
1322 | 1296 | _packermap = {'01': (_makecg1packer, cg1unpacker), |
|
1323 | 1297 | # cg2 adds support for exchanging generaldelta |
|
1324 | 1298 | '02': (_makecg2packer, cg2unpacker), |
|
1325 | 1299 | # cg3 adds support for exchanging revlog flags and treemanifests |
|
1326 | 1300 | '03': (_makecg3packer, cg3unpacker), |
|
1327 | 1301 | } |
|
1328 | 1302 | |
|
1329 | 1303 | def allsupportedversions(repo): |
|
1330 | 1304 | versions = set(_packermap.keys()) |
|
1331 | 1305 | if not (repo.ui.configbool('experimental', 'changegroup3') or |
|
1332 | 1306 | repo.ui.configbool('experimental', 'treemanifest') or |
|
1333 | 1307 | 'treemanifest' in repo.requirements): |
|
1334 | 1308 | versions.discard('03') |
|
1335 | 1309 | return versions |
|
1336 | 1310 | |
|
1337 | 1311 | # Changegroup versions that can be applied to the repo |
|
1338 | 1312 | def supportedincomingversions(repo): |
|
1339 | 1313 | return allsupportedversions(repo) |
|
1340 | 1314 | |
|
1341 | 1315 | # Changegroup versions that can be created from the repo |
|
1342 | 1316 | def supportedoutgoingversions(repo): |
|
1343 | 1317 | versions = allsupportedversions(repo) |
|
1344 | 1318 | if 'treemanifest' in repo.requirements: |
|
1345 | 1319 | # Versions 01 and 02 support only flat manifests and it's just too |
|
1346 | 1320 | # expensive to convert between the flat manifest and tree manifest on |
|
1347 | 1321 | # the fly. Since tree manifests are hashed differently, all of history |
|
1348 | 1322 | # would have to be converted. Instead, we simply don't even pretend to |
|
1349 | 1323 | # support versions 01 and 02. |
|
1350 | 1324 | versions.discard('01') |
|
1351 | 1325 | versions.discard('02') |
|
1352 | 1326 | if repository.NARROW_REQUIREMENT in repo.requirements: |
|
1353 | 1327 | # Versions 01 and 02 don't support revlog flags, and we need to |
|
1354 | 1328 | # support that for stripping and unbundling to work. |
|
1355 | 1329 | versions.discard('01') |
|
1356 | 1330 | versions.discard('02') |
|
1357 | 1331 | if LFS_REQUIREMENT in repo.requirements: |
|
1358 | 1332 | # Versions 01 and 02 don't support revlog flags, and we need to |
|
1359 | 1333 | # mark LFS entries with REVIDX_EXTSTORED. |
|
1360 | 1334 | versions.discard('01') |
|
1361 | 1335 | versions.discard('02') |
|
1362 | 1336 | |
|
1363 | 1337 | return versions |
|
1364 | 1338 | |
|
1365 | 1339 | def localversion(repo): |
|
1366 | 1340 | # Finds the best version to use for bundles that are meant to be used |
|
1367 | 1341 | # locally, such as those from strip and shelve, and temporary bundles. |
|
1368 | 1342 | return max(supportedoutgoingversions(repo)) |
|
1369 | 1343 | |
|
1370 | 1344 | def safeversion(repo): |
|
1371 | 1345 | # Finds the smallest version that it's safe to assume clients of the repo |
|
1372 | 1346 | # will support. For example, all hg versions that support generaldelta also |
|
1373 | 1347 | # support changegroup 02. |
|
1374 | 1348 | versions = supportedoutgoingversions(repo) |
|
1375 | 1349 | if 'generaldelta' in repo.requirements: |
|
1376 | 1350 | versions.discard('01') |
|
1377 | 1351 | assert versions |
|
1378 | 1352 | return min(versions) |
|
1379 | 1353 | |
|
1380 | 1354 | def getbundler(version, repo, bundlecaps=None, filematcher=None, |
|
1381 | 1355 | ellipses=False, shallow=False, ellipsisroots=None, |
|
1382 | 1356 | fullnodes=None): |
|
1383 | 1357 | assert version in supportedoutgoingversions(repo) |
|
1384 | 1358 | |
|
1385 | 1359 | if filematcher is None: |
|
1386 | 1360 | filematcher = matchmod.alwaysmatcher(repo.root, '') |
|
1387 | 1361 | |
|
1388 | 1362 | if version == '01' and not filematcher.always(): |
|
1389 | 1363 | raise error.ProgrammingError('version 01 changegroups do not support ' |
|
1390 | 1364 | 'sparse file matchers') |
|
1391 | 1365 | |
|
1392 | 1366 | if ellipses and version in (b'01', b'02'): |
|
1393 | 1367 | raise error.Abort( |
|
1394 | 1368 | _('ellipsis nodes require at least cg3 on client and server, ' |
|
1395 | 1369 | 'but negotiated version %s') % version) |
|
1396 | 1370 | |
|
1397 | 1371 | # Requested files could include files not in the local store. So |
|
1398 | 1372 | # filter those out. |
|
1399 | 1373 | filematcher = matchmod.intersectmatchers(repo.narrowmatch(), |
|
1400 | 1374 | filematcher) |
|
1401 | 1375 | |
|
1402 | 1376 | fn = _packermap[version][0] |
|
1403 | 1377 | return fn(repo, filematcher, bundlecaps, ellipses=ellipses, |
|
1404 | 1378 | shallow=shallow, ellipsisroots=ellipsisroots, |
|
1405 | 1379 | fullnodes=fullnodes) |
|
1406 | 1380 | |
|
1407 | 1381 | def getunbundler(version, fh, alg, extras=None): |
|
1408 | 1382 | return _packermap[version][1](fh, alg, extras=extras) |
|
1409 | 1383 | |
|
1410 | 1384 | def _changegroupinfo(repo, nodes, source): |
|
1411 | 1385 | if repo.ui.verbose or source == 'bundle': |
|
1412 | 1386 | repo.ui.status(_("%d changesets found\n") % len(nodes)) |
|
1413 | 1387 | if repo.ui.debugflag: |
|
1414 | 1388 | repo.ui.debug("list of changesets:\n") |
|
1415 | 1389 | for node in nodes: |
|
1416 | 1390 | repo.ui.debug("%s\n" % hex(node)) |
|
1417 | 1391 | |
|
1418 | 1392 | def makechangegroup(repo, outgoing, version, source, fastpath=False, |
|
1419 | 1393 | bundlecaps=None): |
|
1420 | 1394 | cgstream = makestream(repo, outgoing, version, source, |
|
1421 | 1395 | fastpath=fastpath, bundlecaps=bundlecaps) |
|
1422 | 1396 | return getunbundler(version, util.chunkbuffer(cgstream), None, |
|
1423 | 1397 | {'clcount': len(outgoing.missing) }) |
|
1424 | 1398 | |
|
1425 | 1399 | def makestream(repo, outgoing, version, source, fastpath=False, |
|
1426 | 1400 | bundlecaps=None, filematcher=None): |
|
1427 | 1401 | bundler = getbundler(version, repo, bundlecaps=bundlecaps, |
|
1428 | 1402 | filematcher=filematcher) |
|
1429 | 1403 | |
|
1430 | 1404 | repo = repo.unfiltered() |
|
1431 | 1405 | commonrevs = outgoing.common |
|
1432 | 1406 | csets = outgoing.missing |
|
1433 | 1407 | heads = outgoing.missingheads |
|
1434 | 1408 | # We go through the fast path if we get told to, or if all (unfiltered |
|
1435 | 1409 | # heads have been requested (since we then know there all linkrevs will |
|
1436 | 1410 | # be pulled by the client). |
|
1437 | 1411 | heads.sort() |
|
1438 | 1412 | fastpathlinkrev = fastpath or ( |
|
1439 | 1413 | repo.filtername is None and heads == sorted(repo.heads())) |
|
1440 | 1414 | |
|
1441 | 1415 | repo.hook('preoutgoing', throw=True, source=source) |
|
1442 | 1416 | _changegroupinfo(repo, csets, source) |
|
1443 | 1417 | return bundler.generate(commonrevs, csets, fastpathlinkrev, source) |
|
1444 | 1418 | |
|
1445 | 1419 | def _addchangegroupfiles(repo, source, revmap, trp, expectedfiles, needfiles): |
|
1446 | 1420 | revisions = 0 |
|
1447 | 1421 | files = 0 |
|
1448 | 1422 | progress = repo.ui.makeprogress(_('files'), unit=_('files'), |
|
1449 | 1423 | total=expectedfiles) |
|
1450 | 1424 | for chunkdata in iter(source.filelogheader, {}): |
|
1451 | 1425 | files += 1 |
|
1452 | 1426 | f = chunkdata["filename"] |
|
1453 | 1427 | repo.ui.debug("adding %s revisions\n" % f) |
|
1454 | 1428 | progress.increment() |
|
1455 | 1429 | fl = repo.file(f) |
|
1456 | 1430 | o = len(fl) |
|
1457 | 1431 | try: |
|
1458 | 1432 | deltas = source.deltaiter() |
|
1459 | 1433 | if not fl.addgroup(deltas, revmap, trp): |
|
1460 | 1434 | raise error.Abort(_("received file revlog group is empty")) |
|
1461 | 1435 | except error.CensoredBaseError as e: |
|
1462 | 1436 | raise error.Abort(_("received delta base is censored: %s") % e) |
|
1463 | 1437 | revisions += len(fl) - o |
|
1464 | 1438 | if f in needfiles: |
|
1465 | 1439 | needs = needfiles[f] |
|
1466 | 1440 | for new in pycompat.xrange(o, len(fl)): |
|
1467 | 1441 | n = fl.node(new) |
|
1468 | 1442 | if n in needs: |
|
1469 | 1443 | needs.remove(n) |
|
1470 | 1444 | else: |
|
1471 | 1445 | raise error.Abort( |
|
1472 | 1446 | _("received spurious file revlog entry")) |
|
1473 | 1447 | if not needs: |
|
1474 | 1448 | del needfiles[f] |
|
1475 | 1449 | progress.complete() |
|
1476 | 1450 | |
|
1477 | 1451 | for f, needs in needfiles.iteritems(): |
|
1478 | 1452 | fl = repo.file(f) |
|
1479 | 1453 | for n in needs: |
|
1480 | 1454 | try: |
|
1481 | 1455 | fl.rev(n) |
|
1482 | 1456 | except error.LookupError: |
|
1483 | 1457 | raise error.Abort( |
|
1484 | 1458 | _('missing file data for %s:%s - run hg verify') % |
|
1485 | 1459 | (f, hex(n))) |
|
1486 | 1460 | |
|
1487 | 1461 | return revisions, files |
General Comments 0
You need to be logged in to leave comments.
Login now