Show More
@@ -1,2691 +1,2691 b'' | |||
|
1 | 1 | # revlog.py - storage back-end for mercurial |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2005-2007 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 | """Storage back-end for Mercurial. |
|
9 | 9 | |
|
10 | 10 | This provides efficient delta storage with O(1) retrieve and append |
|
11 | 11 | and O(changes) merge between branches. |
|
12 | 12 | """ |
|
13 | 13 | |
|
14 | 14 | from __future__ import absolute_import |
|
15 | 15 | |
|
16 | 16 | import collections |
|
17 | 17 | import contextlib |
|
18 | 18 | import errno |
|
19 | 19 | import io |
|
20 | 20 | import os |
|
21 | 21 | import struct |
|
22 | 22 | import zlib |
|
23 | 23 | |
|
24 | 24 | # import stuff from node for others to import from revlog |
|
25 | 25 | from .node import ( |
|
26 | 26 | bin, |
|
27 | 27 | hex, |
|
28 | 28 | nullhex, |
|
29 | 29 | nullid, |
|
30 | 30 | nullrev, |
|
31 | 31 | short, |
|
32 | 32 | wdirfilenodeids, |
|
33 | 33 | wdirhex, |
|
34 | 34 | wdirid, |
|
35 | 35 | wdirrev, |
|
36 | 36 | ) |
|
37 | 37 | from .i18n import _ |
|
38 | 38 | from .revlogutils.constants import ( |
|
39 | 39 | FLAG_GENERALDELTA, |
|
40 | 40 | FLAG_INLINE_DATA, |
|
41 | 41 | REVIDX_DEFAULT_FLAGS, |
|
42 | 42 | REVIDX_ELLIPSIS, |
|
43 | 43 | REVIDX_EXTSTORED, |
|
44 | 44 | REVIDX_FLAGS_ORDER, |
|
45 | 45 | REVIDX_ISCENSORED, |
|
46 | 46 | REVIDX_KNOWN_FLAGS, |
|
47 | 47 | REVIDX_RAWTEXT_CHANGING_FLAGS, |
|
48 | 48 | REVLOGV0, |
|
49 | 49 | REVLOGV1, |
|
50 | 50 | REVLOGV1_FLAGS, |
|
51 | 51 | REVLOGV2, |
|
52 | 52 | REVLOGV2_FLAGS, |
|
53 | 53 | REVLOG_DEFAULT_FLAGS, |
|
54 | 54 | REVLOG_DEFAULT_FORMAT, |
|
55 | 55 | REVLOG_DEFAULT_VERSION, |
|
56 | 56 | ) |
|
57 | 57 | from .thirdparty import ( |
|
58 | 58 | attr, |
|
59 | 59 | ) |
|
60 | 60 | from . import ( |
|
61 | 61 | ancestor, |
|
62 | 62 | dagop, |
|
63 | 63 | error, |
|
64 | 64 | mdiff, |
|
65 | 65 | policy, |
|
66 | 66 | pycompat, |
|
67 | 67 | repository, |
|
68 | 68 | templatefilters, |
|
69 | 69 | util, |
|
70 | 70 | ) |
|
71 | 71 | from .revlogutils import ( |
|
72 | 72 | deltas as deltautil, |
|
73 | 73 | ) |
|
74 | 74 | from .utils import ( |
|
75 | 75 | interfaceutil, |
|
76 | 76 | storageutil, |
|
77 | 77 | stringutil, |
|
78 | 78 | ) |
|
79 | 79 | |
|
80 | 80 | # blanked usage of all the name to prevent pyflakes constraints |
|
81 | 81 | # We need these name available in the module for extensions. |
|
82 | 82 | REVLOGV0 |
|
83 | 83 | REVLOGV1 |
|
84 | 84 | REVLOGV2 |
|
85 | 85 | FLAG_INLINE_DATA |
|
86 | 86 | FLAG_GENERALDELTA |
|
87 | 87 | REVLOG_DEFAULT_FLAGS |
|
88 | 88 | REVLOG_DEFAULT_FORMAT |
|
89 | 89 | REVLOG_DEFAULT_VERSION |
|
90 | 90 | REVLOGV1_FLAGS |
|
91 | 91 | REVLOGV2_FLAGS |
|
92 | 92 | REVIDX_ISCENSORED |
|
93 | 93 | REVIDX_ELLIPSIS |
|
94 | 94 | REVIDX_EXTSTORED |
|
95 | 95 | REVIDX_DEFAULT_FLAGS |
|
96 | 96 | REVIDX_FLAGS_ORDER |
|
97 | 97 | REVIDX_KNOWN_FLAGS |
|
98 | 98 | REVIDX_RAWTEXT_CHANGING_FLAGS |
|
99 | 99 | |
|
100 | 100 | parsers = policy.importmod(r'parsers') |
|
101 | 101 | rustancestor = policy.importrust(r'ancestor') |
|
102 | 102 | rustdagop = policy.importrust(r'dagop') |
|
103 | 103 | |
|
104 | 104 | # Aliased for performance. |
|
105 | 105 | _zlibdecompress = zlib.decompress |
|
106 | 106 | |
|
107 | 107 | # max size of revlog with inline data |
|
108 | 108 | _maxinline = 131072 |
|
109 | 109 | _chunksize = 1048576 |
|
110 | 110 | |
|
111 | 111 | # Store flag processors (cf. 'addflagprocessor()' to register) |
|
112 | 112 | _flagprocessors = { |
|
113 | 113 | REVIDX_ISCENSORED: None, |
|
114 | 114 | } |
|
115 | 115 | |
|
116 | 116 | # Flag processors for REVIDX_ELLIPSIS. |
|
117 | 117 | def ellipsisreadprocessor(rl, text): |
|
118 | 118 | return text, False |
|
119 | 119 | |
|
120 | 120 | def ellipsiswriteprocessor(rl, text): |
|
121 | 121 | return text, False |
|
122 | 122 | |
|
123 | 123 | def ellipsisrawprocessor(rl, text): |
|
124 | 124 | return False |
|
125 | 125 | |
|
126 | 126 | ellipsisprocessor = ( |
|
127 | 127 | ellipsisreadprocessor, |
|
128 | 128 | ellipsiswriteprocessor, |
|
129 | 129 | ellipsisrawprocessor, |
|
130 | 130 | ) |
|
131 | 131 | |
|
132 | 132 | def addflagprocessor(flag, processor): |
|
133 | 133 | """Register a flag processor on a revision data flag. |
|
134 | 134 | |
|
135 | 135 | Invariant: |
|
136 | 136 | - Flags need to be defined in REVIDX_KNOWN_FLAGS and REVIDX_FLAGS_ORDER, |
|
137 | 137 | and REVIDX_RAWTEXT_CHANGING_FLAGS if they can alter rawtext. |
|
138 | 138 | - Only one flag processor can be registered on a specific flag. |
|
139 | 139 | - flagprocessors must be 3-tuples of functions (read, write, raw) with the |
|
140 | 140 | following signatures: |
|
141 | 141 | - (read) f(self, rawtext) -> text, bool |
|
142 | 142 | - (write) f(self, text) -> rawtext, bool |
|
143 | 143 | - (raw) f(self, rawtext) -> bool |
|
144 | 144 | "text" is presented to the user. "rawtext" is stored in revlog data, not |
|
145 | 145 | directly visible to the user. |
|
146 | 146 | The boolean returned by these transforms is used to determine whether |
|
147 | 147 | the returned text can be used for hash integrity checking. For example, |
|
148 | 148 | if "write" returns False, then "text" is used to generate hash. If |
|
149 | 149 | "write" returns True, that basically means "rawtext" returned by "write" |
|
150 | 150 | should be used to generate hash. Usually, "write" and "read" return |
|
151 | 151 | different booleans. And "raw" returns a same boolean as "write". |
|
152 | 152 | |
|
153 | 153 | Note: The 'raw' transform is used for changegroup generation and in some |
|
154 | 154 | debug commands. In this case the transform only indicates whether the |
|
155 | 155 | contents can be used for hash integrity checks. |
|
156 | 156 | """ |
|
157 | 157 | _insertflagprocessor(flag, processor, _flagprocessors) |
|
158 | 158 | |
|
159 | 159 | def _insertflagprocessor(flag, processor, flagprocessors): |
|
160 | 160 | if not flag & REVIDX_KNOWN_FLAGS: |
|
161 | 161 | msg = _("cannot register processor on unknown flag '%#x'.") % (flag) |
|
162 | 162 | raise error.ProgrammingError(msg) |
|
163 | 163 | if flag not in REVIDX_FLAGS_ORDER: |
|
164 | 164 | msg = _("flag '%#x' undefined in REVIDX_FLAGS_ORDER.") % (flag) |
|
165 | 165 | raise error.ProgrammingError(msg) |
|
166 | 166 | if flag in flagprocessors: |
|
167 | 167 | msg = _("cannot register multiple processors on flag '%#x'.") % (flag) |
|
168 | 168 | raise error.Abort(msg) |
|
169 | 169 | flagprocessors[flag] = processor |
|
170 | 170 | |
|
171 | 171 | def getoffset(q): |
|
172 | 172 | return int(q >> 16) |
|
173 | 173 | |
|
174 | 174 | def gettype(q): |
|
175 | 175 | return int(q & 0xFFFF) |
|
176 | 176 | |
|
177 | 177 | def offset_type(offset, type): |
|
178 | 178 | if (type & ~REVIDX_KNOWN_FLAGS) != 0: |
|
179 | 179 | raise ValueError('unknown revlog index flags') |
|
180 | 180 | return int(int(offset) << 16 | type) |
|
181 | 181 | |
|
182 | 182 | @attr.s(slots=True, frozen=True) |
|
183 | 183 | class _revisioninfo(object): |
|
184 | 184 | """Information about a revision that allows building its fulltext |
|
185 | 185 | node: expected hash of the revision |
|
186 | 186 | p1, p2: parent revs of the revision |
|
187 | 187 | btext: built text cache consisting of a one-element list |
|
188 | 188 | cachedelta: (baserev, uncompressed_delta) or None |
|
189 | 189 | flags: flags associated to the revision storage |
|
190 | 190 | |
|
191 | 191 | One of btext[0] or cachedelta must be set. |
|
192 | 192 | """ |
|
193 | 193 | node = attr.ib() |
|
194 | 194 | p1 = attr.ib() |
|
195 | 195 | p2 = attr.ib() |
|
196 | 196 | btext = attr.ib() |
|
197 | 197 | textlen = attr.ib() |
|
198 | 198 | cachedelta = attr.ib() |
|
199 | 199 | flags = attr.ib() |
|
200 | 200 | |
|
201 | 201 | @interfaceutil.implementer(repository.irevisiondelta) |
|
202 | 202 | @attr.s(slots=True) |
|
203 | 203 | class revlogrevisiondelta(object): |
|
204 | 204 | node = attr.ib() |
|
205 | 205 | p1node = attr.ib() |
|
206 | 206 | p2node = attr.ib() |
|
207 | 207 | basenode = attr.ib() |
|
208 | 208 | flags = attr.ib() |
|
209 | 209 | baserevisionsize = attr.ib() |
|
210 | 210 | revision = attr.ib() |
|
211 | 211 | delta = attr.ib() |
|
212 | 212 | linknode = attr.ib(default=None) |
|
213 | 213 | |
|
214 | 214 | @interfaceutil.implementer(repository.iverifyproblem) |
|
215 | 215 | @attr.s(frozen=True) |
|
216 | 216 | class revlogproblem(object): |
|
217 | 217 | warning = attr.ib(default=None) |
|
218 | 218 | error = attr.ib(default=None) |
|
219 | 219 | node = attr.ib(default=None) |
|
220 | 220 | |
|
221 | 221 | # index v0: |
|
222 | 222 | # 4 bytes: offset |
|
223 | 223 | # 4 bytes: compressed length |
|
224 | 224 | # 4 bytes: base rev |
|
225 | 225 | # 4 bytes: link rev |
|
226 | 226 | # 20 bytes: parent 1 nodeid |
|
227 | 227 | # 20 bytes: parent 2 nodeid |
|
228 | 228 | # 20 bytes: nodeid |
|
229 | 229 | indexformatv0 = struct.Struct(">4l20s20s20s") |
|
230 | 230 | indexformatv0_pack = indexformatv0.pack |
|
231 | 231 | indexformatv0_unpack = indexformatv0.unpack |
|
232 | 232 | |
|
233 | 233 | class revlogoldindex(list): |
|
234 | 234 | def __getitem__(self, i): |
|
235 | 235 | if i == -1: |
|
236 | 236 | return (0, 0, 0, -1, -1, -1, -1, nullid) |
|
237 | 237 | return list.__getitem__(self, i) |
|
238 | 238 | |
|
239 | 239 | class revlogoldio(object): |
|
240 | 240 | def __init__(self): |
|
241 | 241 | self.size = indexformatv0.size |
|
242 | 242 | |
|
243 | 243 | def parseindex(self, data, inline): |
|
244 | 244 | s = self.size |
|
245 | 245 | index = [] |
|
246 | 246 | nodemap = {nullid: nullrev} |
|
247 | 247 | n = off = 0 |
|
248 | 248 | l = len(data) |
|
249 | 249 | while off + s <= l: |
|
250 | 250 | cur = data[off:off + s] |
|
251 | 251 | off += s |
|
252 | 252 | e = indexformatv0_unpack(cur) |
|
253 | 253 | # transform to revlogv1 format |
|
254 | 254 | e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3], |
|
255 | 255 | nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6]) |
|
256 | 256 | index.append(e2) |
|
257 | 257 | nodemap[e[6]] = n |
|
258 | 258 | n += 1 |
|
259 | 259 | |
|
260 | 260 | return revlogoldindex(index), nodemap, None |
|
261 | 261 | |
|
262 | 262 | def packentry(self, entry, node, version, rev): |
|
263 | 263 | if gettype(entry[0]): |
|
264 | 264 | raise error.RevlogError(_('index entry flags need revlog ' |
|
265 | 265 | 'version 1')) |
|
266 | 266 | e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4], |
|
267 | 267 | node(entry[5]), node(entry[6]), entry[7]) |
|
268 | 268 | return indexformatv0_pack(*e2) |
|
269 | 269 | |
|
270 | 270 | # index ng: |
|
271 | 271 | # 6 bytes: offset |
|
272 | 272 | # 2 bytes: flags |
|
273 | 273 | # 4 bytes: compressed length |
|
274 | 274 | # 4 bytes: uncompressed length |
|
275 | 275 | # 4 bytes: base rev |
|
276 | 276 | # 4 bytes: link rev |
|
277 | 277 | # 4 bytes: parent 1 rev |
|
278 | 278 | # 4 bytes: parent 2 rev |
|
279 | 279 | # 32 bytes: nodeid |
|
280 | 280 | indexformatng = struct.Struct(">Qiiiiii20s12x") |
|
281 | 281 | indexformatng_pack = indexformatng.pack |
|
282 | 282 | versionformat = struct.Struct(">I") |
|
283 | 283 | versionformat_pack = versionformat.pack |
|
284 | 284 | versionformat_unpack = versionformat.unpack |
|
285 | 285 | |
|
286 | 286 | # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte |
|
287 | 287 | # signed integer) |
|
288 | 288 | _maxentrysize = 0x7fffffff |
|
289 | 289 | |
|
290 | 290 | class revlogio(object): |
|
291 | 291 | def __init__(self): |
|
292 | 292 | self.size = indexformatng.size |
|
293 | 293 | |
|
294 | 294 | def parseindex(self, data, inline): |
|
295 | 295 | # call the C implementation to parse the index data |
|
296 | 296 | index, cache = parsers.parse_index2(data, inline) |
|
297 | 297 | return index, getattr(index, 'nodemap', None), cache |
|
298 | 298 | |
|
299 | 299 | def packentry(self, entry, node, version, rev): |
|
300 | 300 | p = indexformatng_pack(*entry) |
|
301 | 301 | if rev == 0: |
|
302 | 302 | p = versionformat_pack(version) + p[4:] |
|
303 | 303 | return p |
|
304 | 304 | |
|
305 | 305 | class revlog(object): |
|
306 | 306 | """ |
|
307 | 307 | the underlying revision storage object |
|
308 | 308 | |
|
309 | 309 | A revlog consists of two parts, an index and the revision data. |
|
310 | 310 | |
|
311 | 311 | The index is a file with a fixed record size containing |
|
312 | 312 | information on each revision, including its nodeid (hash), the |
|
313 | 313 | nodeids of its parents, the position and offset of its data within |
|
314 | 314 | the data file, and the revision it's based on. Finally, each entry |
|
315 | 315 | contains a linkrev entry that can serve as a pointer to external |
|
316 | 316 | data. |
|
317 | 317 | |
|
318 | 318 | The revision data itself is a linear collection of data chunks. |
|
319 | 319 | Each chunk represents a revision and is usually represented as a |
|
320 | 320 | delta against the previous chunk. To bound lookup time, runs of |
|
321 | 321 | deltas are limited to about 2 times the length of the original |
|
322 | 322 | version data. This makes retrieval of a version proportional to |
|
323 | 323 | its size, or O(1) relative to the number of revisions. |
|
324 | 324 | |
|
325 | 325 | Both pieces of the revlog are written to in an append-only |
|
326 | 326 | fashion, which means we never need to rewrite a file to insert or |
|
327 | 327 | remove data, and can use some simple techniques to avoid the need |
|
328 | 328 | for locking while reading. |
|
329 | 329 | |
|
330 | 330 | If checkambig, indexfile is opened with checkambig=True at |
|
331 | 331 | writing, to avoid file stat ambiguity. |
|
332 | 332 | |
|
333 | 333 | If mmaplargeindex is True, and an mmapindexthreshold is set, the |
|
334 | 334 | index will be mmapped rather than read if it is larger than the |
|
335 | 335 | configured threshold. |
|
336 | 336 | |
|
337 | 337 | If censorable is True, the revlog can have censored revisions. |
|
338 | 338 | |
|
339 | 339 | If `upperboundcomp` is not None, this is the expected maximal gain from |
|
340 | 340 | compression for the data content. |
|
341 | 341 | """ |
|
342 | 342 | def __init__(self, opener, indexfile, datafile=None, checkambig=False, |
|
343 | 343 | mmaplargeindex=False, censorable=False, |
|
344 | 344 | upperboundcomp=None): |
|
345 | 345 | """ |
|
346 | 346 | create a revlog object |
|
347 | 347 | |
|
348 | 348 | opener is a function that abstracts the file opening operation |
|
349 | 349 | and can be used to implement COW semantics or the like. |
|
350 | 350 | |
|
351 | 351 | """ |
|
352 | 352 | self.upperboundcomp = upperboundcomp |
|
353 | 353 | self.indexfile = indexfile |
|
354 | 354 | self.datafile = datafile or (indexfile[:-2] + ".d") |
|
355 | 355 | self.opener = opener |
|
356 | 356 | # When True, indexfile is opened with checkambig=True at writing, to |
|
357 | 357 | # avoid file stat ambiguity. |
|
358 | 358 | self._checkambig = checkambig |
|
359 | 359 | self._mmaplargeindex = mmaplargeindex |
|
360 | 360 | self._censorable = censorable |
|
361 | 361 | # 3-tuple of (node, rev, text) for a raw revision. |
|
362 | 362 | self._revisioncache = None |
|
363 | 363 | # Maps rev to chain base rev. |
|
364 | 364 | self._chainbasecache = util.lrucachedict(100) |
|
365 | 365 | # 2-tuple of (offset, data) of raw data from the revlog at an offset. |
|
366 | 366 | self._chunkcache = (0, '') |
|
367 | 367 | # How much data to read and cache into the raw revlog data cache. |
|
368 | 368 | self._chunkcachesize = 65536 |
|
369 | 369 | self._maxchainlen = None |
|
370 | 370 | self._deltabothparents = True |
|
371 | 371 | self.index = [] |
|
372 | 372 | # Mapping of partial identifiers to full nodes. |
|
373 | 373 | self._pcache = {} |
|
374 | 374 | # Mapping of revision integer to full node. |
|
375 | 375 | self._nodecache = {nullid: nullrev} |
|
376 | 376 | self._nodepos = None |
|
377 | 377 | self._compengine = 'zlib' |
|
378 | 378 | self._compengineopts = {} |
|
379 | 379 | self._maxdeltachainspan = -1 |
|
380 | 380 | self._withsparseread = False |
|
381 | 381 | self._sparserevlog = False |
|
382 | 382 | self._srdensitythreshold = 0.50 |
|
383 | 383 | self._srmingapsize = 262144 |
|
384 | 384 | |
|
385 | 385 | # Make copy of flag processors so each revlog instance can support |
|
386 | 386 | # custom flags. |
|
387 | 387 | self._flagprocessors = dict(_flagprocessors) |
|
388 | 388 | |
|
389 | 389 | # 2-tuple of file handles being used for active writing. |
|
390 | 390 | self._writinghandles = None |
|
391 | 391 | |
|
392 | 392 | self._loadindex() |
|
393 | 393 | |
|
394 | 394 | def _loadindex(self): |
|
395 | 395 | mmapindexthreshold = None |
|
396 | 396 | opts = getattr(self.opener, 'options', {}) or {} |
|
397 | 397 | |
|
398 | 398 | if 'revlogv2' in opts: |
|
399 | 399 | newversionflags = REVLOGV2 | FLAG_INLINE_DATA |
|
400 | 400 | elif 'revlogv1' in opts: |
|
401 | 401 | newversionflags = REVLOGV1 | FLAG_INLINE_DATA |
|
402 | 402 | if 'generaldelta' in opts: |
|
403 | 403 | newversionflags |= FLAG_GENERALDELTA |
|
404 | 404 | elif getattr(self.opener, 'options', None) is not None: |
|
405 | 405 | # If options provided but no 'revlog*' found, the repository |
|
406 | 406 | # would have no 'requires' file in it, which means we have to |
|
407 | 407 | # stick to the old format. |
|
408 | 408 | newversionflags = REVLOGV0 |
|
409 | 409 | else: |
|
410 | 410 | newversionflags = REVLOG_DEFAULT_VERSION |
|
411 | 411 | |
|
412 | 412 | if 'chunkcachesize' in opts: |
|
413 | 413 | self._chunkcachesize = opts['chunkcachesize'] |
|
414 | 414 | if 'maxchainlen' in opts: |
|
415 | 415 | self._maxchainlen = opts['maxchainlen'] |
|
416 | 416 | if 'deltabothparents' in opts: |
|
417 | 417 | self._deltabothparents = opts['deltabothparents'] |
|
418 | 418 | self._lazydelta = bool(opts.get('lazydelta', True)) |
|
419 | 419 | self._lazydeltabase = False |
|
420 | 420 | if self._lazydelta: |
|
421 | 421 | self._lazydeltabase = bool(opts.get('lazydeltabase', False)) |
|
422 | 422 | if 'compengine' in opts: |
|
423 | 423 | self._compengine = opts['compengine'] |
|
424 | 424 | if 'zlib.level' in opts: |
|
425 | 425 | self._compengineopts['zlib.level'] = opts['zlib.level'] |
|
426 | 426 | if 'zstd.level' in opts: |
|
427 | 427 | self._compengineopts['zstd.level'] = opts['zstd.level'] |
|
428 | 428 | if 'maxdeltachainspan' in opts: |
|
429 | 429 | self._maxdeltachainspan = opts['maxdeltachainspan'] |
|
430 | 430 | if self._mmaplargeindex and 'mmapindexthreshold' in opts: |
|
431 | 431 | mmapindexthreshold = opts['mmapindexthreshold'] |
|
432 | 432 | self._sparserevlog = bool(opts.get('sparse-revlog', False)) |
|
433 | 433 | withsparseread = bool(opts.get('with-sparse-read', False)) |
|
434 | 434 | # sparse-revlog forces sparse-read |
|
435 | 435 | self._withsparseread = self._sparserevlog or withsparseread |
|
436 | 436 | if 'sparse-read-density-threshold' in opts: |
|
437 | 437 | self._srdensitythreshold = opts['sparse-read-density-threshold'] |
|
438 | 438 | if 'sparse-read-min-gap-size' in opts: |
|
439 | 439 | self._srmingapsize = opts['sparse-read-min-gap-size'] |
|
440 | 440 | if opts.get('enableellipsis'): |
|
441 | 441 | self._flagprocessors[REVIDX_ELLIPSIS] = ellipsisprocessor |
|
442 | 442 | |
|
443 | 443 | # revlog v0 doesn't have flag processors |
|
444 | 444 | for flag, processor in opts.get(b'flagprocessors', {}).iteritems(): |
|
445 | 445 | _insertflagprocessor(flag, processor, self._flagprocessors) |
|
446 | 446 | |
|
447 | 447 | if self._chunkcachesize <= 0: |
|
448 | 448 | raise error.RevlogError(_('revlog chunk cache size %r is not ' |
|
449 | 449 | 'greater than 0') % self._chunkcachesize) |
|
450 | 450 | elif self._chunkcachesize & (self._chunkcachesize - 1): |
|
451 | 451 | raise error.RevlogError(_('revlog chunk cache size %r is not a ' |
|
452 | 452 | 'power of 2') % self._chunkcachesize) |
|
453 | 453 | |
|
454 | 454 | indexdata = '' |
|
455 | 455 | self._initempty = True |
|
456 | 456 | try: |
|
457 | 457 | with self._indexfp() as f: |
|
458 | 458 | if (mmapindexthreshold is not None and |
|
459 | 459 | self.opener.fstat(f).st_size >= mmapindexthreshold): |
|
460 | 460 | # TODO: should .close() to release resources without |
|
461 | 461 | # relying on Python GC |
|
462 | 462 | indexdata = util.buffer(util.mmapread(f)) |
|
463 | 463 | else: |
|
464 | 464 | indexdata = f.read() |
|
465 | 465 | if len(indexdata) > 0: |
|
466 | 466 | versionflags = versionformat_unpack(indexdata[:4])[0] |
|
467 | 467 | self._initempty = False |
|
468 | 468 | else: |
|
469 | 469 | versionflags = newversionflags |
|
470 | 470 | except IOError as inst: |
|
471 | 471 | if inst.errno != errno.ENOENT: |
|
472 | 472 | raise |
|
473 | 473 | |
|
474 | 474 | versionflags = newversionflags |
|
475 | 475 | |
|
476 | 476 | self.version = versionflags |
|
477 | 477 | |
|
478 | 478 | flags = versionflags & ~0xFFFF |
|
479 | 479 | fmt = versionflags & 0xFFFF |
|
480 | 480 | |
|
481 | 481 | if fmt == REVLOGV0: |
|
482 | 482 | if flags: |
|
483 | 483 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' |
|
484 | 484 | 'revlog %s') % |
|
485 | 485 | (flags >> 16, fmt, self.indexfile)) |
|
486 | 486 | |
|
487 | 487 | self._inline = False |
|
488 | 488 | self._generaldelta = False |
|
489 | 489 | |
|
490 | 490 | elif fmt == REVLOGV1: |
|
491 | 491 | if flags & ~REVLOGV1_FLAGS: |
|
492 | 492 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' |
|
493 | 493 | 'revlog %s') % |
|
494 | 494 | (flags >> 16, fmt, self.indexfile)) |
|
495 | 495 | |
|
496 | 496 | self._inline = versionflags & FLAG_INLINE_DATA |
|
497 | 497 | self._generaldelta = versionflags & FLAG_GENERALDELTA |
|
498 | 498 | |
|
499 | 499 | elif fmt == REVLOGV2: |
|
500 | 500 | if flags & ~REVLOGV2_FLAGS: |
|
501 | 501 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' |
|
502 | 502 | 'revlog %s') % |
|
503 | 503 | (flags >> 16, fmt, self.indexfile)) |
|
504 | 504 | |
|
505 | 505 | self._inline = versionflags & FLAG_INLINE_DATA |
|
506 | 506 | # generaldelta implied by version 2 revlogs. |
|
507 | 507 | self._generaldelta = True |
|
508 | 508 | |
|
509 | 509 | else: |
|
510 | 510 | raise error.RevlogError(_('unknown version (%d) in revlog %s') % |
|
511 | 511 | (fmt, self.indexfile)) |
|
512 | 512 | # sparse-revlog can't be on without general-delta (issue6056) |
|
513 | 513 | if not self._generaldelta: |
|
514 | 514 | self._sparserevlog = False |
|
515 | 515 | |
|
516 | 516 | self._storedeltachains = True |
|
517 | 517 | |
|
518 | 518 | self._io = revlogio() |
|
519 | 519 | if self.version == REVLOGV0: |
|
520 | 520 | self._io = revlogoldio() |
|
521 | 521 | try: |
|
522 | 522 | d = self._io.parseindex(indexdata, self._inline) |
|
523 | 523 | except (ValueError, IndexError): |
|
524 | 524 | raise error.RevlogError(_("index %s is corrupted") % |
|
525 | 525 | self.indexfile) |
|
526 | 526 | self.index, nodemap, self._chunkcache = d |
|
527 | 527 | if nodemap is not None: |
|
528 | 528 | self.nodemap = self._nodecache = nodemap |
|
529 | 529 | if not self._chunkcache: |
|
530 | 530 | self._chunkclear() |
|
531 | 531 | # revnum -> (chain-length, sum-delta-length) |
|
532 | 532 | self._chaininfocache = {} |
|
533 | 533 | # revlog header -> revlog compressor |
|
534 | 534 | self._decompressors = {} |
|
535 | 535 | |
|
536 | 536 | @util.propertycache |
|
537 | 537 | def _compressor(self): |
|
538 | 538 | engine = util.compengines[self._compengine] |
|
539 | 539 | return engine.revlogcompressor(self._compengineopts) |
|
540 | 540 | |
|
541 | 541 | def _indexfp(self, mode='r'): |
|
542 | 542 | """file object for the revlog's index file""" |
|
543 | 543 | args = {r'mode': mode} |
|
544 | 544 | if mode != 'r': |
|
545 | 545 | args[r'checkambig'] = self._checkambig |
|
546 | 546 | if mode == 'w': |
|
547 | 547 | args[r'atomictemp'] = True |
|
548 | 548 | return self.opener(self.indexfile, **args) |
|
549 | 549 | |
|
550 | 550 | def _datafp(self, mode='r'): |
|
551 | 551 | """file object for the revlog's data file""" |
|
552 | 552 | return self.opener(self.datafile, mode=mode) |
|
553 | 553 | |
|
554 | 554 | @contextlib.contextmanager |
|
555 | 555 | def _datareadfp(self, existingfp=None): |
|
556 | 556 | """file object suitable to read data""" |
|
557 | 557 | # Use explicit file handle, if given. |
|
558 | 558 | if existingfp is not None: |
|
559 | 559 | yield existingfp |
|
560 | 560 | |
|
561 | 561 | # Use a file handle being actively used for writes, if available. |
|
562 | 562 | # There is some danger to doing this because reads will seek the |
|
563 | 563 | # file. However, _writeentry() performs a SEEK_END before all writes, |
|
564 | 564 | # so we should be safe. |
|
565 | 565 | elif self._writinghandles: |
|
566 | 566 | if self._inline: |
|
567 | 567 | yield self._writinghandles[0] |
|
568 | 568 | else: |
|
569 | 569 | yield self._writinghandles[1] |
|
570 | 570 | |
|
571 | 571 | # Otherwise open a new file handle. |
|
572 | 572 | else: |
|
573 | 573 | if self._inline: |
|
574 | 574 | func = self._indexfp |
|
575 | 575 | else: |
|
576 | 576 | func = self._datafp |
|
577 | 577 | with func() as fp: |
|
578 | 578 | yield fp |
|
579 | 579 | |
|
580 | 580 | def tip(self): |
|
581 | 581 | return self.node(len(self.index) - 1) |
|
582 | 582 | def __contains__(self, rev): |
|
583 | 583 | return 0 <= rev < len(self) |
|
584 | 584 | def __len__(self): |
|
585 | 585 | return len(self.index) |
|
586 | 586 | def __iter__(self): |
|
587 | 587 | return iter(pycompat.xrange(len(self))) |
|
588 | 588 | def revs(self, start=0, stop=None): |
|
589 | 589 | """iterate over all rev in this revlog (from start to stop)""" |
|
590 | 590 | return storageutil.iterrevs(len(self), start=start, stop=stop) |
|
591 | 591 | |
|
592 | 592 | @util.propertycache |
|
593 | 593 | def nodemap(self): |
|
594 | 594 | if self.index: |
|
595 | 595 | # populate mapping down to the initial node |
|
596 | 596 | node0 = self.index[0][7] # get around changelog filtering |
|
597 | 597 | self.rev(node0) |
|
598 | 598 | return self._nodecache |
|
599 | 599 | |
|
600 | 600 | def hasnode(self, node): |
|
601 | 601 | try: |
|
602 | 602 | self.rev(node) |
|
603 | 603 | return True |
|
604 | 604 | except KeyError: |
|
605 | 605 | return False |
|
606 | 606 | |
|
607 | 607 | def candelta(self, baserev, rev): |
|
608 | 608 | """whether two revisions (baserev, rev) can be delta-ed or not""" |
|
609 | 609 | # Disable delta if either rev requires a content-changing flag |
|
610 | 610 | # processor (ex. LFS). This is because such flag processor can alter |
|
611 | 611 | # the rawtext content that the delta will be based on, and two clients |
|
612 | 612 | # could have a same revlog node with different flags (i.e. different |
|
613 | 613 | # rawtext contents) and the delta could be incompatible. |
|
614 | 614 | if ((self.flags(baserev) & REVIDX_RAWTEXT_CHANGING_FLAGS) |
|
615 | 615 | or (self.flags(rev) & REVIDX_RAWTEXT_CHANGING_FLAGS)): |
|
616 | 616 | return False |
|
617 | 617 | return True |
|
618 | 618 | |
|
619 | 619 | def clearcaches(self): |
|
620 | 620 | self._revisioncache = None |
|
621 | 621 | self._chainbasecache.clear() |
|
622 | 622 | self._chunkcache = (0, '') |
|
623 | 623 | self._pcache = {} |
|
624 | 624 | |
|
625 | 625 | try: |
|
626 | 626 | # If we are using the native C version, you are in a fun case |
|
627 | 627 | # where self.index, self.nodemap and self._nodecaches is the same |
|
628 | 628 | # object. |
|
629 | 629 | self._nodecache.clearcaches() |
|
630 | 630 | except AttributeError: |
|
631 | 631 | self._nodecache = {nullid: nullrev} |
|
632 | 632 | self._nodepos = None |
|
633 | 633 | |
|
634 | 634 | def rev(self, node): |
|
635 | 635 | try: |
|
636 | 636 | return self._nodecache[node] |
|
637 | 637 | except TypeError: |
|
638 | 638 | raise |
|
639 | 639 | except error.RevlogError: |
|
640 | 640 | # parsers.c radix tree lookup failed |
|
641 | 641 | if node == wdirid or node in wdirfilenodeids: |
|
642 | 642 | raise error.WdirUnsupported |
|
643 | 643 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
644 | 644 | except KeyError: |
|
645 | 645 | # pure python cache lookup failed |
|
646 | 646 | n = self._nodecache |
|
647 | 647 | i = self.index |
|
648 | 648 | p = self._nodepos |
|
649 | 649 | if p is None: |
|
650 | 650 | p = len(i) - 1 |
|
651 | 651 | else: |
|
652 | 652 | assert p < len(i) |
|
653 | 653 | for r in pycompat.xrange(p, -1, -1): |
|
654 | 654 | v = i[r][7] |
|
655 | 655 | n[v] = r |
|
656 | 656 | if v == node: |
|
657 | 657 | self._nodepos = r - 1 |
|
658 | 658 | return r |
|
659 | 659 | if node == wdirid or node in wdirfilenodeids: |
|
660 | 660 | raise error.WdirUnsupported |
|
661 | 661 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
662 | 662 | |
|
663 | 663 | # Accessors for index entries. |
|
664 | 664 | |
|
665 | 665 | # First tuple entry is 8 bytes. First 6 bytes are offset. Last 2 bytes |
|
666 | 666 | # are flags. |
|
667 | 667 | def start(self, rev): |
|
668 | 668 | return int(self.index[rev][0] >> 16) |
|
669 | 669 | |
|
670 | 670 | def flags(self, rev): |
|
671 | 671 | return self.index[rev][0] & 0xFFFF |
|
672 | 672 | |
|
673 | 673 | def length(self, rev): |
|
674 | 674 | return self.index[rev][1] |
|
675 | 675 | |
|
676 | 676 | def rawsize(self, rev): |
|
677 | 677 | """return the length of the uncompressed text for a given revision""" |
|
678 | 678 | l = self.index[rev][2] |
|
679 | 679 | if l >= 0: |
|
680 | 680 | return l |
|
681 | 681 | |
|
682 | 682 | t = self.revision(rev, raw=True) |
|
683 | 683 | return len(t) |
|
684 | 684 | |
|
685 | 685 | def size(self, rev): |
|
686 | 686 | """length of non-raw text (processed by a "read" flag processor)""" |
|
687 | 687 | # fast path: if no "read" flag processor could change the content, |
|
688 | 688 | # size is rawsize. note: ELLIPSIS is known to not change the content. |
|
689 | 689 | flags = self.flags(rev) |
|
690 | 690 | if flags & (REVIDX_KNOWN_FLAGS ^ REVIDX_ELLIPSIS) == 0: |
|
691 | 691 | return self.rawsize(rev) |
|
692 | 692 | |
|
693 | 693 | return len(self.revision(rev, raw=False)) |
|
694 | 694 | |
|
695 | 695 | def chainbase(self, rev): |
|
696 | 696 | base = self._chainbasecache.get(rev) |
|
697 | 697 | if base is not None: |
|
698 | 698 | return base |
|
699 | 699 | |
|
700 | 700 | index = self.index |
|
701 | 701 | iterrev = rev |
|
702 | 702 | base = index[iterrev][3] |
|
703 | 703 | while base != iterrev: |
|
704 | 704 | iterrev = base |
|
705 | 705 | base = index[iterrev][3] |
|
706 | 706 | |
|
707 | 707 | self._chainbasecache[rev] = base |
|
708 | 708 | return base |
|
709 | 709 | |
|
710 | 710 | def linkrev(self, rev): |
|
711 | 711 | return self.index[rev][4] |
|
712 | 712 | |
|
713 | 713 | def parentrevs(self, rev): |
|
714 | 714 | try: |
|
715 | 715 | entry = self.index[rev] |
|
716 | 716 | except IndexError: |
|
717 | 717 | if rev == wdirrev: |
|
718 | 718 | raise error.WdirUnsupported |
|
719 | 719 | raise |
|
720 | 720 | |
|
721 | 721 | return entry[5], entry[6] |
|
722 | 722 | |
|
723 | 723 | # fast parentrevs(rev) where rev isn't filtered |
|
724 | 724 | _uncheckedparentrevs = parentrevs |
|
725 | 725 | |
|
726 | 726 | def node(self, rev): |
|
727 | 727 | try: |
|
728 | 728 | return self.index[rev][7] |
|
729 | 729 | except IndexError: |
|
730 | 730 | if rev == wdirrev: |
|
731 | 731 | raise error.WdirUnsupported |
|
732 | 732 | raise |
|
733 | 733 | |
|
734 | 734 | # Derived from index values. |
|
735 | 735 | |
|
736 | 736 | def end(self, rev): |
|
737 | 737 | return self.start(rev) + self.length(rev) |
|
738 | 738 | |
|
739 | 739 | def parents(self, node): |
|
740 | 740 | i = self.index |
|
741 | 741 | d = i[self.rev(node)] |
|
742 | 742 | return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline |
|
743 | 743 | |
|
744 | 744 | def chainlen(self, rev): |
|
745 | 745 | return self._chaininfo(rev)[0] |
|
746 | 746 | |
|
747 | 747 | def _chaininfo(self, rev): |
|
748 | 748 | chaininfocache = self._chaininfocache |
|
749 | 749 | if rev in chaininfocache: |
|
750 | 750 | return chaininfocache[rev] |
|
751 | 751 | index = self.index |
|
752 | 752 | generaldelta = self._generaldelta |
|
753 | 753 | iterrev = rev |
|
754 | 754 | e = index[iterrev] |
|
755 | 755 | clen = 0 |
|
756 | 756 | compresseddeltalen = 0 |
|
757 | 757 | while iterrev != e[3]: |
|
758 | 758 | clen += 1 |
|
759 | 759 | compresseddeltalen += e[1] |
|
760 | 760 | if generaldelta: |
|
761 | 761 | iterrev = e[3] |
|
762 | 762 | else: |
|
763 | 763 | iterrev -= 1 |
|
764 | 764 | if iterrev in chaininfocache: |
|
765 | 765 | t = chaininfocache[iterrev] |
|
766 | 766 | clen += t[0] |
|
767 | 767 | compresseddeltalen += t[1] |
|
768 | 768 | break |
|
769 | 769 | e = index[iterrev] |
|
770 | 770 | else: |
|
771 | 771 | # Add text length of base since decompressing that also takes |
|
772 | 772 | # work. For cache hits the length is already included. |
|
773 | 773 | compresseddeltalen += e[1] |
|
774 | 774 | r = (clen, compresseddeltalen) |
|
775 | 775 | chaininfocache[rev] = r |
|
776 | 776 | return r |
|
777 | 777 | |
|
778 | 778 | def _deltachain(self, rev, stoprev=None): |
|
779 | 779 | """Obtain the delta chain for a revision. |
|
780 | 780 | |
|
781 | 781 | ``stoprev`` specifies a revision to stop at. If not specified, we |
|
782 | 782 | stop at the base of the chain. |
|
783 | 783 | |
|
784 | 784 | Returns a 2-tuple of (chain, stopped) where ``chain`` is a list of |
|
785 | 785 | revs in ascending order and ``stopped`` is a bool indicating whether |
|
786 | 786 | ``stoprev`` was hit. |
|
787 | 787 | """ |
|
788 | 788 | # Try C implementation. |
|
789 | 789 | try: |
|
790 | 790 | return self.index.deltachain(rev, stoprev, self._generaldelta) |
|
791 | 791 | except AttributeError: |
|
792 | 792 | pass |
|
793 | 793 | |
|
794 | 794 | chain = [] |
|
795 | 795 | |
|
796 | 796 | # Alias to prevent attribute lookup in tight loop. |
|
797 | 797 | index = self.index |
|
798 | 798 | generaldelta = self._generaldelta |
|
799 | 799 | |
|
800 | 800 | iterrev = rev |
|
801 | 801 | e = index[iterrev] |
|
802 | 802 | while iterrev != e[3] and iterrev != stoprev: |
|
803 | 803 | chain.append(iterrev) |
|
804 | 804 | if generaldelta: |
|
805 | 805 | iterrev = e[3] |
|
806 | 806 | else: |
|
807 | 807 | iterrev -= 1 |
|
808 | 808 | e = index[iterrev] |
|
809 | 809 | |
|
810 | 810 | if iterrev == stoprev: |
|
811 | 811 | stopped = True |
|
812 | 812 | else: |
|
813 | 813 | chain.append(iterrev) |
|
814 | 814 | stopped = False |
|
815 | 815 | |
|
816 | 816 | chain.reverse() |
|
817 | 817 | return chain, stopped |
|
818 | 818 | |
|
819 | 819 | def ancestors(self, revs, stoprev=0, inclusive=False): |
|
820 | 820 | """Generate the ancestors of 'revs' in reverse revision order. |
|
821 | 821 | Does not generate revs lower than stoprev. |
|
822 | 822 | |
|
823 | 823 | See the documentation for ancestor.lazyancestors for more details.""" |
|
824 | 824 | |
|
825 | 825 | # first, make sure start revisions aren't filtered |
|
826 | 826 | revs = list(revs) |
|
827 | 827 | checkrev = self.node |
|
828 | 828 | for r in revs: |
|
829 | 829 | checkrev(r) |
|
830 | 830 | # and we're sure ancestors aren't filtered as well |
|
831 | 831 | |
|
832 | 832 | if rustancestor is not None: |
|
833 | 833 | lazyancestors = rustancestor.LazyAncestors |
|
834 | 834 | arg = self.index |
|
835 | 835 | elif util.safehasattr(parsers, 'rustlazyancestors'): |
|
836 | 836 | lazyancestors = ancestor.rustlazyancestors |
|
837 | 837 | arg = self.index |
|
838 | 838 | else: |
|
839 | 839 | lazyancestors = ancestor.lazyancestors |
|
840 | 840 | arg = self._uncheckedparentrevs |
|
841 | 841 | return lazyancestors(arg, revs, stoprev=stoprev, inclusive=inclusive) |
|
842 | 842 | |
|
843 | 843 | def descendants(self, revs): |
|
844 | 844 | return dagop.descendantrevs(revs, self.revs, self.parentrevs) |
|
845 | 845 | |
|
846 | 846 | def findcommonmissing(self, common=None, heads=None): |
|
847 | 847 | """Return a tuple of the ancestors of common and the ancestors of heads |
|
848 | 848 | that are not ancestors of common. In revset terminology, we return the |
|
849 | 849 | tuple: |
|
850 | 850 | |
|
851 | 851 | ::common, (::heads) - (::common) |
|
852 | 852 | |
|
853 | 853 | The list is sorted by revision number, meaning it is |
|
854 | 854 | topologically sorted. |
|
855 | 855 | |
|
856 | 856 | 'heads' and 'common' are both lists of node IDs. If heads is |
|
857 | 857 | not supplied, uses all of the revlog's heads. If common is not |
|
858 | 858 | supplied, uses nullid.""" |
|
859 | 859 | if common is None: |
|
860 | 860 | common = [nullid] |
|
861 | 861 | if heads is None: |
|
862 | 862 | heads = self.heads() |
|
863 | 863 | |
|
864 | 864 | common = [self.rev(n) for n in common] |
|
865 | 865 | heads = [self.rev(n) for n in heads] |
|
866 | 866 | |
|
867 | 867 | # we want the ancestors, but inclusive |
|
868 | 868 | class lazyset(object): |
|
869 | 869 | def __init__(self, lazyvalues): |
|
870 | 870 | self.addedvalues = set() |
|
871 | 871 | self.lazyvalues = lazyvalues |
|
872 | 872 | |
|
873 | 873 | def __contains__(self, value): |
|
874 | 874 | return value in self.addedvalues or value in self.lazyvalues |
|
875 | 875 | |
|
876 | 876 | def __iter__(self): |
|
877 | 877 | added = self.addedvalues |
|
878 | 878 | for r in added: |
|
879 | 879 | yield r |
|
880 | 880 | for r in self.lazyvalues: |
|
881 | 881 | if not r in added: |
|
882 | 882 | yield r |
|
883 | 883 | |
|
884 | 884 | def add(self, value): |
|
885 | 885 | self.addedvalues.add(value) |
|
886 | 886 | |
|
887 | 887 | def update(self, values): |
|
888 | 888 | self.addedvalues.update(values) |
|
889 | 889 | |
|
890 | 890 | has = lazyset(self.ancestors(common)) |
|
891 | 891 | has.add(nullrev) |
|
892 | 892 | has.update(common) |
|
893 | 893 | |
|
894 | 894 | # take all ancestors from heads that aren't in has |
|
895 | 895 | missing = set() |
|
896 | 896 | visit = collections.deque(r for r in heads if r not in has) |
|
897 | 897 | while visit: |
|
898 | 898 | r = visit.popleft() |
|
899 | 899 | if r in missing: |
|
900 | 900 | continue |
|
901 | 901 | else: |
|
902 | 902 | missing.add(r) |
|
903 | 903 | for p in self.parentrevs(r): |
|
904 | 904 | if p not in has: |
|
905 | 905 | visit.append(p) |
|
906 | 906 | missing = list(missing) |
|
907 | 907 | missing.sort() |
|
908 | 908 | return has, [self.node(miss) for miss in missing] |
|
909 | 909 | |
|
910 | 910 | def incrementalmissingrevs(self, common=None): |
|
911 | 911 | """Return an object that can be used to incrementally compute the |
|
912 | 912 | revision numbers of the ancestors of arbitrary sets that are not |
|
913 | 913 | ancestors of common. This is an ancestor.incrementalmissingancestors |
|
914 | 914 | object. |
|
915 | 915 | |
|
916 | 916 | 'common' is a list of revision numbers. If common is not supplied, uses |
|
917 | 917 | nullrev. |
|
918 | 918 | """ |
|
919 | 919 | if common is None: |
|
920 | 920 | common = [nullrev] |
|
921 | 921 | |
|
922 | 922 | if rustancestor is not None: |
|
923 | 923 | return rustancestor.MissingAncestors(self.index, common) |
|
924 | 924 | return ancestor.incrementalmissingancestors(self.parentrevs, common) |
|
925 | 925 | |
|
926 | 926 | def findmissingrevs(self, common=None, heads=None): |
|
927 | 927 | """Return the revision numbers of the ancestors of heads that |
|
928 | 928 | are not ancestors of common. |
|
929 | 929 | |
|
930 | 930 | More specifically, return a list of revision numbers corresponding to |
|
931 | 931 | nodes N such that every N satisfies the following constraints: |
|
932 | 932 | |
|
933 | 933 | 1. N is an ancestor of some node in 'heads' |
|
934 | 934 | 2. N is not an ancestor of any node in 'common' |
|
935 | 935 | |
|
936 | 936 | The list is sorted by revision number, meaning it is |
|
937 | 937 | topologically sorted. |
|
938 | 938 | |
|
939 | 939 | 'heads' and 'common' are both lists of revision numbers. If heads is |
|
940 | 940 | not supplied, uses all of the revlog's heads. If common is not |
|
941 | 941 | supplied, uses nullid.""" |
|
942 | 942 | if common is None: |
|
943 | 943 | common = [nullrev] |
|
944 | 944 | if heads is None: |
|
945 | 945 | heads = self.headrevs() |
|
946 | 946 | |
|
947 | 947 | inc = self.incrementalmissingrevs(common=common) |
|
948 | 948 | return inc.missingancestors(heads) |
|
949 | 949 | |
|
950 | 950 | def findmissing(self, common=None, heads=None): |
|
951 | 951 | """Return the ancestors of heads that are not ancestors of common. |
|
952 | 952 | |
|
953 | 953 | More specifically, return a list of nodes N such that every N |
|
954 | 954 | satisfies the following constraints: |
|
955 | 955 | |
|
956 | 956 | 1. N is an ancestor of some node in 'heads' |
|
957 | 957 | 2. N is not an ancestor of any node in 'common' |
|
958 | 958 | |
|
959 | 959 | The list is sorted by revision number, meaning it is |
|
960 | 960 | topologically sorted. |
|
961 | 961 | |
|
962 | 962 | 'heads' and 'common' are both lists of node IDs. If heads is |
|
963 | 963 | not supplied, uses all of the revlog's heads. If common is not |
|
964 | 964 | supplied, uses nullid.""" |
|
965 | 965 | if common is None: |
|
966 | 966 | common = [nullid] |
|
967 | 967 | if heads is None: |
|
968 | 968 | heads = self.heads() |
|
969 | 969 | |
|
970 | 970 | common = [self.rev(n) for n in common] |
|
971 | 971 | heads = [self.rev(n) for n in heads] |
|
972 | 972 | |
|
973 | 973 | inc = self.incrementalmissingrevs(common=common) |
|
974 | 974 | return [self.node(r) for r in inc.missingancestors(heads)] |
|
975 | 975 | |
|
976 | 976 | def nodesbetween(self, roots=None, heads=None): |
|
977 | 977 | """Return a topological path from 'roots' to 'heads'. |
|
978 | 978 | |
|
979 | 979 | Return a tuple (nodes, outroots, outheads) where 'nodes' is a |
|
980 | 980 | topologically sorted list of all nodes N that satisfy both of |
|
981 | 981 | these constraints: |
|
982 | 982 | |
|
983 | 983 | 1. N is a descendant of some node in 'roots' |
|
984 | 984 | 2. N is an ancestor of some node in 'heads' |
|
985 | 985 | |
|
986 | 986 | Every node is considered to be both a descendant and an ancestor |
|
987 | 987 | of itself, so every reachable node in 'roots' and 'heads' will be |
|
988 | 988 | included in 'nodes'. |
|
989 | 989 | |
|
990 | 990 | 'outroots' is the list of reachable nodes in 'roots', i.e., the |
|
991 | 991 | subset of 'roots' that is returned in 'nodes'. Likewise, |
|
992 | 992 | 'outheads' is the subset of 'heads' that is also in 'nodes'. |
|
993 | 993 | |
|
994 | 994 | 'roots' and 'heads' are both lists of node IDs. If 'roots' is |
|
995 | 995 | unspecified, uses nullid as the only root. If 'heads' is |
|
996 | 996 | unspecified, uses list of all of the revlog's heads.""" |
|
997 | 997 | nonodes = ([], [], []) |
|
998 | 998 | if roots is not None: |
|
999 | 999 | roots = list(roots) |
|
1000 | 1000 | if not roots: |
|
1001 | 1001 | return nonodes |
|
1002 | 1002 | lowestrev = min([self.rev(n) for n in roots]) |
|
1003 | 1003 | else: |
|
1004 | 1004 | roots = [nullid] # Everybody's a descendant of nullid |
|
1005 | 1005 | lowestrev = nullrev |
|
1006 | 1006 | if (lowestrev == nullrev) and (heads is None): |
|
1007 | 1007 | # We want _all_ the nodes! |
|
1008 | 1008 | return ([self.node(r) for r in self], [nullid], list(self.heads())) |
|
1009 | 1009 | if heads is None: |
|
1010 | 1010 | # All nodes are ancestors, so the latest ancestor is the last |
|
1011 | 1011 | # node. |
|
1012 | 1012 | highestrev = len(self) - 1 |
|
1013 | 1013 | # Set ancestors to None to signal that every node is an ancestor. |
|
1014 | 1014 | ancestors = None |
|
1015 | 1015 | # Set heads to an empty dictionary for later discovery of heads |
|
1016 | 1016 | heads = {} |
|
1017 | 1017 | else: |
|
1018 | 1018 | heads = list(heads) |
|
1019 | 1019 | if not heads: |
|
1020 | 1020 | return nonodes |
|
1021 | 1021 | ancestors = set() |
|
1022 | 1022 | # Turn heads into a dictionary so we can remove 'fake' heads. |
|
1023 | 1023 | # Also, later we will be using it to filter out the heads we can't |
|
1024 | 1024 | # find from roots. |
|
1025 | 1025 | heads = dict.fromkeys(heads, False) |
|
1026 | 1026 | # Start at the top and keep marking parents until we're done. |
|
1027 | 1027 | nodestotag = set(heads) |
|
1028 | 1028 | # Remember where the top was so we can use it as a limit later. |
|
1029 | 1029 | highestrev = max([self.rev(n) for n in nodestotag]) |
|
1030 | 1030 | while nodestotag: |
|
1031 | 1031 | # grab a node to tag |
|
1032 | 1032 | n = nodestotag.pop() |
|
1033 | 1033 | # Never tag nullid |
|
1034 | 1034 | if n == nullid: |
|
1035 | 1035 | continue |
|
1036 | 1036 | # A node's revision number represents its place in a |
|
1037 | 1037 | # topologically sorted list of nodes. |
|
1038 | 1038 | r = self.rev(n) |
|
1039 | 1039 | if r >= lowestrev: |
|
1040 | 1040 | if n not in ancestors: |
|
1041 | 1041 | # If we are possibly a descendant of one of the roots |
|
1042 | 1042 | # and we haven't already been marked as an ancestor |
|
1043 | 1043 | ancestors.add(n) # Mark as ancestor |
|
1044 | 1044 | # Add non-nullid parents to list of nodes to tag. |
|
1045 | 1045 | nodestotag.update([p for p in self.parents(n) if |
|
1046 | 1046 | p != nullid]) |
|
1047 | 1047 | elif n in heads: # We've seen it before, is it a fake head? |
|
1048 | 1048 | # So it is, real heads should not be the ancestors of |
|
1049 | 1049 | # any other heads. |
|
1050 | 1050 | heads.pop(n) |
|
1051 | 1051 | if not ancestors: |
|
1052 | 1052 | return nonodes |
|
1053 | 1053 | # Now that we have our set of ancestors, we want to remove any |
|
1054 | 1054 | # roots that are not ancestors. |
|
1055 | 1055 | |
|
1056 | 1056 | # If one of the roots was nullid, everything is included anyway. |
|
1057 | 1057 | if lowestrev > nullrev: |
|
1058 | 1058 | # But, since we weren't, let's recompute the lowest rev to not |
|
1059 | 1059 | # include roots that aren't ancestors. |
|
1060 | 1060 | |
|
1061 | 1061 | # Filter out roots that aren't ancestors of heads |
|
1062 | 1062 | roots = [root for root in roots if root in ancestors] |
|
1063 | 1063 | # Recompute the lowest revision |
|
1064 | 1064 | if roots: |
|
1065 | 1065 | lowestrev = min([self.rev(root) for root in roots]) |
|
1066 | 1066 | else: |
|
1067 | 1067 | # No more roots? Return empty list |
|
1068 | 1068 | return nonodes |
|
1069 | 1069 | else: |
|
1070 | 1070 | # We are descending from nullid, and don't need to care about |
|
1071 | 1071 | # any other roots. |
|
1072 | 1072 | lowestrev = nullrev |
|
1073 | 1073 | roots = [nullid] |
|
1074 | 1074 | # Transform our roots list into a set. |
|
1075 | 1075 | descendants = set(roots) |
|
1076 | 1076 | # Also, keep the original roots so we can filter out roots that aren't |
|
1077 | 1077 | # 'real' roots (i.e. are descended from other roots). |
|
1078 | 1078 | roots = descendants.copy() |
|
1079 | 1079 | # Our topologically sorted list of output nodes. |
|
1080 | 1080 | orderedout = [] |
|
1081 | 1081 | # Don't start at nullid since we don't want nullid in our output list, |
|
1082 | 1082 | # and if nullid shows up in descendants, empty parents will look like |
|
1083 | 1083 | # they're descendants. |
|
1084 | 1084 | for r in self.revs(start=max(lowestrev, 0), stop=highestrev + 1): |
|
1085 | 1085 | n = self.node(r) |
|
1086 | 1086 | isdescendant = False |
|
1087 | 1087 | if lowestrev == nullrev: # Everybody is a descendant of nullid |
|
1088 | 1088 | isdescendant = True |
|
1089 | 1089 | elif n in descendants: |
|
1090 | 1090 | # n is already a descendant |
|
1091 | 1091 | isdescendant = True |
|
1092 | 1092 | # This check only needs to be done here because all the roots |
|
1093 | 1093 | # will start being marked is descendants before the loop. |
|
1094 | 1094 | if n in roots: |
|
1095 | 1095 | # If n was a root, check if it's a 'real' root. |
|
1096 | 1096 | p = tuple(self.parents(n)) |
|
1097 | 1097 | # If any of its parents are descendants, it's not a root. |
|
1098 | 1098 | if (p[0] in descendants) or (p[1] in descendants): |
|
1099 | 1099 | roots.remove(n) |
|
1100 | 1100 | else: |
|
1101 | 1101 | p = tuple(self.parents(n)) |
|
1102 | 1102 | # A node is a descendant if either of its parents are |
|
1103 | 1103 | # descendants. (We seeded the dependents list with the roots |
|
1104 | 1104 | # up there, remember?) |
|
1105 | 1105 | if (p[0] in descendants) or (p[1] in descendants): |
|
1106 | 1106 | descendants.add(n) |
|
1107 | 1107 | isdescendant = True |
|
1108 | 1108 | if isdescendant and ((ancestors is None) or (n in ancestors)): |
|
1109 | 1109 | # Only include nodes that are both descendants and ancestors. |
|
1110 | 1110 | orderedout.append(n) |
|
1111 | 1111 | if (ancestors is not None) and (n in heads): |
|
1112 | 1112 | # We're trying to figure out which heads are reachable |
|
1113 | 1113 | # from roots. |
|
1114 | 1114 | # Mark this head as having been reached |
|
1115 | 1115 | heads[n] = True |
|
1116 | 1116 | elif ancestors is None: |
|
1117 | 1117 | # Otherwise, we're trying to discover the heads. |
|
1118 | 1118 | # Assume this is a head because if it isn't, the next step |
|
1119 | 1119 | # will eventually remove it. |
|
1120 | 1120 | heads[n] = True |
|
1121 | 1121 | # But, obviously its parents aren't. |
|
1122 | 1122 | for p in self.parents(n): |
|
1123 | 1123 | heads.pop(p, None) |
|
1124 | 1124 | heads = [head for head, flag in heads.iteritems() if flag] |
|
1125 | 1125 | roots = list(roots) |
|
1126 | 1126 | assert orderedout |
|
1127 | 1127 | assert roots |
|
1128 | 1128 | assert heads |
|
1129 | 1129 | return (orderedout, roots, heads) |
|
1130 | 1130 | |
|
1131 | 1131 | def headrevs(self, revs=None): |
|
1132 | 1132 | if revs is None: |
|
1133 | 1133 | try: |
|
1134 | 1134 | return self.index.headrevs() |
|
1135 | 1135 | except AttributeError: |
|
1136 | 1136 | return self._headrevs() |
|
1137 | 1137 | if rustdagop is not None: |
|
1138 | 1138 | return rustdagop.headrevs(self.index, revs) |
|
1139 | 1139 | return dagop.headrevs(revs, self._uncheckedparentrevs) |
|
1140 | 1140 | |
|
1141 | 1141 | def computephases(self, roots): |
|
1142 | 1142 | return self.index.computephasesmapsets(roots) |
|
1143 | 1143 | |
|
1144 | 1144 | def _headrevs(self): |
|
1145 | 1145 | count = len(self) |
|
1146 | 1146 | if not count: |
|
1147 | 1147 | return [nullrev] |
|
1148 | 1148 | # we won't iter over filtered rev so nobody is a head at start |
|
1149 | 1149 | ishead = [0] * (count + 1) |
|
1150 | 1150 | index = self.index |
|
1151 | 1151 | for r in self: |
|
1152 | 1152 | ishead[r] = 1 # I may be an head |
|
1153 | 1153 | e = index[r] |
|
1154 | 1154 | ishead[e[5]] = ishead[e[6]] = 0 # my parent are not |
|
1155 | 1155 | return [r for r, val in enumerate(ishead) if val] |
|
1156 | 1156 | |
|
1157 | 1157 | def heads(self, start=None, stop=None): |
|
1158 | 1158 | """return the list of all nodes that have no children |
|
1159 | 1159 | |
|
1160 | 1160 | if start is specified, only heads that are descendants of |
|
1161 | 1161 | start will be returned |
|
1162 | 1162 | if stop is specified, it will consider all the revs from stop |
|
1163 | 1163 | as if they had no children |
|
1164 | 1164 | """ |
|
1165 | 1165 | if start is None and stop is None: |
|
1166 | 1166 | if not len(self): |
|
1167 | 1167 | return [nullid] |
|
1168 | 1168 | return [self.node(r) for r in self.headrevs()] |
|
1169 | 1169 | |
|
1170 | 1170 | if start is None: |
|
1171 | 1171 | start = nullrev |
|
1172 | 1172 | else: |
|
1173 | 1173 | start = self.rev(start) |
|
1174 | 1174 | |
|
1175 | 1175 | stoprevs = set(self.rev(n) for n in stop or []) |
|
1176 | 1176 | |
|
1177 | 1177 | revs = dagop.headrevssubset(self.revs, self.parentrevs, startrev=start, |
|
1178 | 1178 | stoprevs=stoprevs) |
|
1179 | 1179 | |
|
1180 | 1180 | return [self.node(rev) for rev in revs] |
|
1181 | 1181 | |
|
1182 | 1182 | def children(self, node): |
|
1183 | 1183 | """find the children of a given node""" |
|
1184 | 1184 | c = [] |
|
1185 | 1185 | p = self.rev(node) |
|
1186 | 1186 | for r in self.revs(start=p + 1): |
|
1187 | 1187 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] |
|
1188 | 1188 | if prevs: |
|
1189 | 1189 | for pr in prevs: |
|
1190 | 1190 | if pr == p: |
|
1191 | 1191 | c.append(self.node(r)) |
|
1192 | 1192 | elif p == nullrev: |
|
1193 | 1193 | c.append(self.node(r)) |
|
1194 | 1194 | return c |
|
1195 | 1195 | |
|
1196 | 1196 | def commonancestorsheads(self, a, b): |
|
1197 | 1197 | """calculate all the heads of the common ancestors of nodes a and b""" |
|
1198 | 1198 | a, b = self.rev(a), self.rev(b) |
|
1199 | 1199 | ancs = self._commonancestorsheads(a, b) |
|
1200 | 1200 | return pycompat.maplist(self.node, ancs) |
|
1201 | 1201 | |
|
1202 | 1202 | def _commonancestorsheads(self, *revs): |
|
1203 | 1203 | """calculate all the heads of the common ancestors of revs""" |
|
1204 | 1204 | try: |
|
1205 | 1205 | ancs = self.index.commonancestorsheads(*revs) |
|
1206 | 1206 | except (AttributeError, OverflowError): # C implementation failed |
|
1207 | 1207 | ancs = ancestor.commonancestorsheads(self.parentrevs, *revs) |
|
1208 | 1208 | return ancs |
|
1209 | 1209 | |
|
1210 | 1210 | def isancestor(self, a, b): |
|
1211 | 1211 | """return True if node a is an ancestor of node b |
|
1212 | 1212 | |
|
1213 | 1213 | A revision is considered an ancestor of itself.""" |
|
1214 | 1214 | a, b = self.rev(a), self.rev(b) |
|
1215 | 1215 | return self.isancestorrev(a, b) |
|
1216 | 1216 | |
|
1217 | 1217 | def isancestorrev(self, a, b): |
|
1218 | 1218 | """return True if revision a is an ancestor of revision b |
|
1219 | 1219 | |
|
1220 | 1220 | A revision is considered an ancestor of itself. |
|
1221 | 1221 | |
|
1222 | 1222 | The implementation of this is trivial but the use of |
|
1223 | 1223 | reachableroots is not.""" |
|
1224 | 1224 | if a == nullrev: |
|
1225 | 1225 | return True |
|
1226 | 1226 | elif a == b: |
|
1227 | 1227 | return True |
|
1228 | 1228 | elif a > b: |
|
1229 | 1229 | return False |
|
1230 | 1230 | return bool(self.reachableroots(a, [b], [a], includepath=False)) |
|
1231 | 1231 | |
|
1232 | 1232 | def reachableroots(self, minroot, heads, roots, includepath=False): |
|
1233 | 1233 | """return (heads(::<roots> and <roots>::<heads>)) |
|
1234 | 1234 | |
|
1235 | 1235 | If includepath is True, return (<roots>::<heads>).""" |
|
1236 | 1236 | try: |
|
1237 | 1237 | return self.index.reachableroots2(minroot, heads, roots, |
|
1238 | 1238 | includepath) |
|
1239 | 1239 | except AttributeError: |
|
1240 | 1240 | return dagop._reachablerootspure(self.parentrevs, |
|
1241 | 1241 | minroot, roots, heads, includepath) |
|
1242 | 1242 | |
|
1243 | 1243 | def ancestor(self, a, b): |
|
1244 | 1244 | """calculate the "best" common ancestor of nodes a and b""" |
|
1245 | 1245 | |
|
1246 | 1246 | a, b = self.rev(a), self.rev(b) |
|
1247 | 1247 | try: |
|
1248 | 1248 | ancs = self.index.ancestors(a, b) |
|
1249 | 1249 | except (AttributeError, OverflowError): |
|
1250 | 1250 | ancs = ancestor.ancestors(self.parentrevs, a, b) |
|
1251 | 1251 | if ancs: |
|
1252 | 1252 | # choose a consistent winner when there's a tie |
|
1253 | 1253 | return min(map(self.node, ancs)) |
|
1254 | 1254 | return nullid |
|
1255 | 1255 | |
|
1256 | 1256 | def _match(self, id): |
|
1257 | 1257 | if isinstance(id, int): |
|
1258 | 1258 | # rev |
|
1259 | 1259 | return self.node(id) |
|
1260 | 1260 | if len(id) == 20: |
|
1261 | 1261 | # possibly a binary node |
|
1262 | 1262 | # odds of a binary node being all hex in ASCII are 1 in 10**25 |
|
1263 | 1263 | try: |
|
1264 | 1264 | node = id |
|
1265 | 1265 | self.rev(node) # quick search the index |
|
1266 | 1266 | return node |
|
1267 | 1267 | except error.LookupError: |
|
1268 | 1268 | pass # may be partial hex id |
|
1269 | 1269 | try: |
|
1270 | 1270 | # str(rev) |
|
1271 | 1271 | rev = int(id) |
|
1272 | 1272 | if "%d" % rev != id: |
|
1273 | 1273 | raise ValueError |
|
1274 | 1274 | if rev < 0: |
|
1275 | 1275 | rev = len(self) + rev |
|
1276 | 1276 | if rev < 0 or rev >= len(self): |
|
1277 | 1277 | raise ValueError |
|
1278 | 1278 | return self.node(rev) |
|
1279 | 1279 | except (ValueError, OverflowError): |
|
1280 | 1280 | pass |
|
1281 | 1281 | if len(id) == 40: |
|
1282 | 1282 | try: |
|
1283 | 1283 | # a full hex nodeid? |
|
1284 | 1284 | node = bin(id) |
|
1285 | 1285 | self.rev(node) |
|
1286 | 1286 | return node |
|
1287 | 1287 | except (TypeError, error.LookupError): |
|
1288 | 1288 | pass |
|
1289 | 1289 | |
|
1290 | 1290 | def _partialmatch(self, id): |
|
1291 | 1291 | # we don't care wdirfilenodeids as they should be always full hash |
|
1292 | 1292 | maybewdir = wdirhex.startswith(id) |
|
1293 | 1293 | try: |
|
1294 | 1294 | partial = self.index.partialmatch(id) |
|
1295 | 1295 | if partial and self.hasnode(partial): |
|
1296 | 1296 | if maybewdir: |
|
1297 | 1297 | # single 'ff...' match in radix tree, ambiguous with wdir |
|
1298 | 1298 | raise error.RevlogError |
|
1299 | 1299 | return partial |
|
1300 | 1300 | if maybewdir: |
|
1301 | 1301 | # no 'ff...' match in radix tree, wdir identified |
|
1302 | 1302 | raise error.WdirUnsupported |
|
1303 | 1303 | return None |
|
1304 | 1304 | except error.RevlogError: |
|
1305 | 1305 | # parsers.c radix tree lookup gave multiple matches |
|
1306 | 1306 | # fast path: for unfiltered changelog, radix tree is accurate |
|
1307 | 1307 | if not getattr(self, 'filteredrevs', None): |
|
1308 | 1308 | raise error.AmbiguousPrefixLookupError( |
|
1309 | 1309 | id, self.indexfile, _('ambiguous identifier')) |
|
1310 | 1310 | # fall through to slow path that filters hidden revisions |
|
1311 | 1311 | except (AttributeError, ValueError): |
|
1312 | 1312 | # we are pure python, or key was too short to search radix tree |
|
1313 | 1313 | pass |
|
1314 | 1314 | |
|
1315 | 1315 | if id in self._pcache: |
|
1316 | 1316 | return self._pcache[id] |
|
1317 | 1317 | |
|
1318 | 1318 | if len(id) <= 40: |
|
1319 | 1319 | try: |
|
1320 | 1320 | # hex(node)[:...] |
|
1321 | 1321 | l = len(id) // 2 # grab an even number of digits |
|
1322 | 1322 | prefix = bin(id[:l * 2]) |
|
1323 | 1323 | nl = [e[7] for e in self.index if e[7].startswith(prefix)] |
|
1324 | 1324 | nl = [n for n in nl if hex(n).startswith(id) and |
|
1325 | 1325 | self.hasnode(n)] |
|
1326 | 1326 | if nullhex.startswith(id): |
|
1327 | 1327 | nl.append(nullid) |
|
1328 | 1328 | if len(nl) > 0: |
|
1329 | 1329 | if len(nl) == 1 and not maybewdir: |
|
1330 | 1330 | self._pcache[id] = nl[0] |
|
1331 | 1331 | return nl[0] |
|
1332 | 1332 | raise error.AmbiguousPrefixLookupError( |
|
1333 | 1333 | id, self.indexfile, _('ambiguous identifier')) |
|
1334 | 1334 | if maybewdir: |
|
1335 | 1335 | raise error.WdirUnsupported |
|
1336 | 1336 | return None |
|
1337 | 1337 | except TypeError: |
|
1338 | 1338 | pass |
|
1339 | 1339 | |
|
1340 | 1340 | def lookup(self, id): |
|
1341 | 1341 | """locate a node based on: |
|
1342 | 1342 | - revision number or str(revision number) |
|
1343 | 1343 | - nodeid or subset of hex nodeid |
|
1344 | 1344 | """ |
|
1345 | 1345 | n = self._match(id) |
|
1346 | 1346 | if n is not None: |
|
1347 | 1347 | return n |
|
1348 | 1348 | n = self._partialmatch(id) |
|
1349 | 1349 | if n: |
|
1350 | 1350 | return n |
|
1351 | 1351 | |
|
1352 | 1352 | raise error.LookupError(id, self.indexfile, _('no match found')) |
|
1353 | 1353 | |
|
1354 | 1354 | def shortest(self, node, minlength=1): |
|
1355 | 1355 | """Find the shortest unambiguous prefix that matches node.""" |
|
1356 | 1356 | def isvalid(prefix): |
|
1357 | 1357 | try: |
|
1358 | node = self._partialmatch(prefix) | |
|
1358 | matchednode = self._partialmatch(prefix) | |
|
1359 | 1359 | except error.AmbiguousPrefixLookupError: |
|
1360 | 1360 | return False |
|
1361 | 1361 | except error.WdirUnsupported: |
|
1362 | 1362 | # single 'ff...' match |
|
1363 | 1363 | return True |
|
1364 | if node is None: | |
|
1364 | if matchednode is None: | |
|
1365 | 1365 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
1366 | 1366 | return True |
|
1367 | 1367 | |
|
1368 | 1368 | def maybewdir(prefix): |
|
1369 | 1369 | return all(c == 'f' for c in pycompat.iterbytestr(prefix)) |
|
1370 | 1370 | |
|
1371 | 1371 | hexnode = hex(node) |
|
1372 | 1372 | |
|
1373 | 1373 | def disambiguate(hexnode, minlength): |
|
1374 | 1374 | """Disambiguate against wdirid.""" |
|
1375 | 1375 | for length in range(minlength, 41): |
|
1376 | 1376 | prefix = hexnode[:length] |
|
1377 | 1377 | if not maybewdir(prefix): |
|
1378 | 1378 | return prefix |
|
1379 | 1379 | |
|
1380 | 1380 | if not getattr(self, 'filteredrevs', None): |
|
1381 | 1381 | try: |
|
1382 | 1382 | length = max(self.index.shortest(node), minlength) |
|
1383 | 1383 | return disambiguate(hexnode, length) |
|
1384 | 1384 | except error.RevlogError: |
|
1385 | 1385 | if node != wdirid: |
|
1386 | 1386 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
1387 | 1387 | except AttributeError: |
|
1388 | 1388 | # Fall through to pure code |
|
1389 | 1389 | pass |
|
1390 | 1390 | |
|
1391 | 1391 | if node == wdirid: |
|
1392 | 1392 | for length in range(minlength, 41): |
|
1393 | 1393 | prefix = hexnode[:length] |
|
1394 | 1394 | if isvalid(prefix): |
|
1395 | 1395 | return prefix |
|
1396 | 1396 | |
|
1397 | 1397 | for length in range(minlength, 41): |
|
1398 | 1398 | prefix = hexnode[:length] |
|
1399 | 1399 | if isvalid(prefix): |
|
1400 | 1400 | return disambiguate(hexnode, length) |
|
1401 | 1401 | |
|
1402 | 1402 | def cmp(self, node, text): |
|
1403 | 1403 | """compare text with a given file revision |
|
1404 | 1404 | |
|
1405 | 1405 | returns True if text is different than what is stored. |
|
1406 | 1406 | """ |
|
1407 | 1407 | p1, p2 = self.parents(node) |
|
1408 | 1408 | return storageutil.hashrevisionsha1(text, p1, p2) != node |
|
1409 | 1409 | |
|
1410 | 1410 | def _cachesegment(self, offset, data): |
|
1411 | 1411 | """Add a segment to the revlog cache. |
|
1412 | 1412 | |
|
1413 | 1413 | Accepts an absolute offset and the data that is at that location. |
|
1414 | 1414 | """ |
|
1415 | 1415 | o, d = self._chunkcache |
|
1416 | 1416 | # try to add to existing cache |
|
1417 | 1417 | if o + len(d) == offset and len(d) + len(data) < _chunksize: |
|
1418 | 1418 | self._chunkcache = o, d + data |
|
1419 | 1419 | else: |
|
1420 | 1420 | self._chunkcache = offset, data |
|
1421 | 1421 | |
|
1422 | 1422 | def _readsegment(self, offset, length, df=None): |
|
1423 | 1423 | """Load a segment of raw data from the revlog. |
|
1424 | 1424 | |
|
1425 | 1425 | Accepts an absolute offset, length to read, and an optional existing |
|
1426 | 1426 | file handle to read from. |
|
1427 | 1427 | |
|
1428 | 1428 | If an existing file handle is passed, it will be seeked and the |
|
1429 | 1429 | original seek position will NOT be restored. |
|
1430 | 1430 | |
|
1431 | 1431 | Returns a str or buffer of raw byte data. |
|
1432 | 1432 | |
|
1433 | 1433 | Raises if the requested number of bytes could not be read. |
|
1434 | 1434 | """ |
|
1435 | 1435 | # Cache data both forward and backward around the requested |
|
1436 | 1436 | # data, in a fixed size window. This helps speed up operations |
|
1437 | 1437 | # involving reading the revlog backwards. |
|
1438 | 1438 | cachesize = self._chunkcachesize |
|
1439 | 1439 | realoffset = offset & ~(cachesize - 1) |
|
1440 | 1440 | reallength = (((offset + length + cachesize) & ~(cachesize - 1)) |
|
1441 | 1441 | - realoffset) |
|
1442 | 1442 | with self._datareadfp(df) as df: |
|
1443 | 1443 | df.seek(realoffset) |
|
1444 | 1444 | d = df.read(reallength) |
|
1445 | 1445 | |
|
1446 | 1446 | self._cachesegment(realoffset, d) |
|
1447 | 1447 | if offset != realoffset or reallength != length: |
|
1448 | 1448 | startoffset = offset - realoffset |
|
1449 | 1449 | if len(d) - startoffset < length: |
|
1450 | 1450 | raise error.RevlogError( |
|
1451 | 1451 | _('partial read of revlog %s; expected %d bytes from ' |
|
1452 | 1452 | 'offset %d, got %d') % |
|
1453 | 1453 | (self.indexfile if self._inline else self.datafile, |
|
1454 | 1454 | length, realoffset, len(d) - startoffset)) |
|
1455 | 1455 | |
|
1456 | 1456 | return util.buffer(d, startoffset, length) |
|
1457 | 1457 | |
|
1458 | 1458 | if len(d) < length: |
|
1459 | 1459 | raise error.RevlogError( |
|
1460 | 1460 | _('partial read of revlog %s; expected %d bytes from offset ' |
|
1461 | 1461 | '%d, got %d') % |
|
1462 | 1462 | (self.indexfile if self._inline else self.datafile, |
|
1463 | 1463 | length, offset, len(d))) |
|
1464 | 1464 | |
|
1465 | 1465 | return d |
|
1466 | 1466 | |
|
1467 | 1467 | def _getsegment(self, offset, length, df=None): |
|
1468 | 1468 | """Obtain a segment of raw data from the revlog. |
|
1469 | 1469 | |
|
1470 | 1470 | Accepts an absolute offset, length of bytes to obtain, and an |
|
1471 | 1471 | optional file handle to the already-opened revlog. If the file |
|
1472 | 1472 | handle is used, it's original seek position will not be preserved. |
|
1473 | 1473 | |
|
1474 | 1474 | Requests for data may be returned from a cache. |
|
1475 | 1475 | |
|
1476 | 1476 | Returns a str or a buffer instance of raw byte data. |
|
1477 | 1477 | """ |
|
1478 | 1478 | o, d = self._chunkcache |
|
1479 | 1479 | l = len(d) |
|
1480 | 1480 | |
|
1481 | 1481 | # is it in the cache? |
|
1482 | 1482 | cachestart = offset - o |
|
1483 | 1483 | cacheend = cachestart + length |
|
1484 | 1484 | if cachestart >= 0 and cacheend <= l: |
|
1485 | 1485 | if cachestart == 0 and cacheend == l: |
|
1486 | 1486 | return d # avoid a copy |
|
1487 | 1487 | return util.buffer(d, cachestart, cacheend - cachestart) |
|
1488 | 1488 | |
|
1489 | 1489 | return self._readsegment(offset, length, df=df) |
|
1490 | 1490 | |
|
1491 | 1491 | def _getsegmentforrevs(self, startrev, endrev, df=None): |
|
1492 | 1492 | """Obtain a segment of raw data corresponding to a range of revisions. |
|
1493 | 1493 | |
|
1494 | 1494 | Accepts the start and end revisions and an optional already-open |
|
1495 | 1495 | file handle to be used for reading. If the file handle is read, its |
|
1496 | 1496 | seek position will not be preserved. |
|
1497 | 1497 | |
|
1498 | 1498 | Requests for data may be satisfied by a cache. |
|
1499 | 1499 | |
|
1500 | 1500 | Returns a 2-tuple of (offset, data) for the requested range of |
|
1501 | 1501 | revisions. Offset is the integer offset from the beginning of the |
|
1502 | 1502 | revlog and data is a str or buffer of the raw byte data. |
|
1503 | 1503 | |
|
1504 | 1504 | Callers will need to call ``self.start(rev)`` and ``self.length(rev)`` |
|
1505 | 1505 | to determine where each revision's data begins and ends. |
|
1506 | 1506 | """ |
|
1507 | 1507 | # Inlined self.start(startrev) & self.end(endrev) for perf reasons |
|
1508 | 1508 | # (functions are expensive). |
|
1509 | 1509 | index = self.index |
|
1510 | 1510 | istart = index[startrev] |
|
1511 | 1511 | start = int(istart[0] >> 16) |
|
1512 | 1512 | if startrev == endrev: |
|
1513 | 1513 | end = start + istart[1] |
|
1514 | 1514 | else: |
|
1515 | 1515 | iend = index[endrev] |
|
1516 | 1516 | end = int(iend[0] >> 16) + iend[1] |
|
1517 | 1517 | |
|
1518 | 1518 | if self._inline: |
|
1519 | 1519 | start += (startrev + 1) * self._io.size |
|
1520 | 1520 | end += (endrev + 1) * self._io.size |
|
1521 | 1521 | length = end - start |
|
1522 | 1522 | |
|
1523 | 1523 | return start, self._getsegment(start, length, df=df) |
|
1524 | 1524 | |
|
1525 | 1525 | def _chunk(self, rev, df=None): |
|
1526 | 1526 | """Obtain a single decompressed chunk for a revision. |
|
1527 | 1527 | |
|
1528 | 1528 | Accepts an integer revision and an optional already-open file handle |
|
1529 | 1529 | to be used for reading. If used, the seek position of the file will not |
|
1530 | 1530 | be preserved. |
|
1531 | 1531 | |
|
1532 | 1532 | Returns a str holding uncompressed data for the requested revision. |
|
1533 | 1533 | """ |
|
1534 | 1534 | return self.decompress(self._getsegmentforrevs(rev, rev, df=df)[1]) |
|
1535 | 1535 | |
|
1536 | 1536 | def _chunks(self, revs, df=None, targetsize=None): |
|
1537 | 1537 | """Obtain decompressed chunks for the specified revisions. |
|
1538 | 1538 | |
|
1539 | 1539 | Accepts an iterable of numeric revisions that are assumed to be in |
|
1540 | 1540 | ascending order. Also accepts an optional already-open file handle |
|
1541 | 1541 | to be used for reading. If used, the seek position of the file will |
|
1542 | 1542 | not be preserved. |
|
1543 | 1543 | |
|
1544 | 1544 | This function is similar to calling ``self._chunk()`` multiple times, |
|
1545 | 1545 | but is faster. |
|
1546 | 1546 | |
|
1547 | 1547 | Returns a list with decompressed data for each requested revision. |
|
1548 | 1548 | """ |
|
1549 | 1549 | if not revs: |
|
1550 | 1550 | return [] |
|
1551 | 1551 | start = self.start |
|
1552 | 1552 | length = self.length |
|
1553 | 1553 | inline = self._inline |
|
1554 | 1554 | iosize = self._io.size |
|
1555 | 1555 | buffer = util.buffer |
|
1556 | 1556 | |
|
1557 | 1557 | l = [] |
|
1558 | 1558 | ladd = l.append |
|
1559 | 1559 | |
|
1560 | 1560 | if not self._withsparseread: |
|
1561 | 1561 | slicedchunks = (revs,) |
|
1562 | 1562 | else: |
|
1563 | 1563 | slicedchunks = deltautil.slicechunk(self, revs, |
|
1564 | 1564 | targetsize=targetsize) |
|
1565 | 1565 | |
|
1566 | 1566 | for revschunk in slicedchunks: |
|
1567 | 1567 | firstrev = revschunk[0] |
|
1568 | 1568 | # Skip trailing revisions with empty diff |
|
1569 | 1569 | for lastrev in revschunk[::-1]: |
|
1570 | 1570 | if length(lastrev) != 0: |
|
1571 | 1571 | break |
|
1572 | 1572 | |
|
1573 | 1573 | try: |
|
1574 | 1574 | offset, data = self._getsegmentforrevs(firstrev, lastrev, df=df) |
|
1575 | 1575 | except OverflowError: |
|
1576 | 1576 | # issue4215 - we can't cache a run of chunks greater than |
|
1577 | 1577 | # 2G on Windows |
|
1578 | 1578 | return [self._chunk(rev, df=df) for rev in revschunk] |
|
1579 | 1579 | |
|
1580 | 1580 | decomp = self.decompress |
|
1581 | 1581 | for rev in revschunk: |
|
1582 | 1582 | chunkstart = start(rev) |
|
1583 | 1583 | if inline: |
|
1584 | 1584 | chunkstart += (rev + 1) * iosize |
|
1585 | 1585 | chunklength = length(rev) |
|
1586 | 1586 | ladd(decomp(buffer(data, chunkstart - offset, chunklength))) |
|
1587 | 1587 | |
|
1588 | 1588 | return l |
|
1589 | 1589 | |
|
1590 | 1590 | def _chunkclear(self): |
|
1591 | 1591 | """Clear the raw chunk cache.""" |
|
1592 | 1592 | self._chunkcache = (0, '') |
|
1593 | 1593 | |
|
1594 | 1594 | def deltaparent(self, rev): |
|
1595 | 1595 | """return deltaparent of the given revision""" |
|
1596 | 1596 | base = self.index[rev][3] |
|
1597 | 1597 | if base == rev: |
|
1598 | 1598 | return nullrev |
|
1599 | 1599 | elif self._generaldelta: |
|
1600 | 1600 | return base |
|
1601 | 1601 | else: |
|
1602 | 1602 | return rev - 1 |
|
1603 | 1603 | |
|
1604 | 1604 | def issnapshot(self, rev): |
|
1605 | 1605 | """tells whether rev is a snapshot |
|
1606 | 1606 | """ |
|
1607 | 1607 | if not self._sparserevlog: |
|
1608 | 1608 | return self.deltaparent(rev) == nullrev |
|
1609 | 1609 | elif util.safehasattr(self.index, 'issnapshot'): |
|
1610 | 1610 | # directly assign the method to cache the testing and access |
|
1611 | 1611 | self.issnapshot = self.index.issnapshot |
|
1612 | 1612 | return self.issnapshot(rev) |
|
1613 | 1613 | if rev == nullrev: |
|
1614 | 1614 | return True |
|
1615 | 1615 | entry = self.index[rev] |
|
1616 | 1616 | base = entry[3] |
|
1617 | 1617 | if base == rev: |
|
1618 | 1618 | return True |
|
1619 | 1619 | if base == nullrev: |
|
1620 | 1620 | return True |
|
1621 | 1621 | p1 = entry[5] |
|
1622 | 1622 | p2 = entry[6] |
|
1623 | 1623 | if base == p1 or base == p2: |
|
1624 | 1624 | return False |
|
1625 | 1625 | return self.issnapshot(base) |
|
1626 | 1626 | |
|
1627 | 1627 | def snapshotdepth(self, rev): |
|
1628 | 1628 | """number of snapshot in the chain before this one""" |
|
1629 | 1629 | if not self.issnapshot(rev): |
|
1630 | 1630 | raise error.ProgrammingError('revision %d not a snapshot') |
|
1631 | 1631 | return len(self._deltachain(rev)[0]) - 1 |
|
1632 | 1632 | |
|
1633 | 1633 | def revdiff(self, rev1, rev2): |
|
1634 | 1634 | """return or calculate a delta between two revisions |
|
1635 | 1635 | |
|
1636 | 1636 | The delta calculated is in binary form and is intended to be written to |
|
1637 | 1637 | revlog data directly. So this function needs raw revision data. |
|
1638 | 1638 | """ |
|
1639 | 1639 | if rev1 != nullrev and self.deltaparent(rev2) == rev1: |
|
1640 | 1640 | return bytes(self._chunk(rev2)) |
|
1641 | 1641 | |
|
1642 | 1642 | return mdiff.textdiff(self.revision(rev1, raw=True), |
|
1643 | 1643 | self.revision(rev2, raw=True)) |
|
1644 | 1644 | |
|
1645 | 1645 | def revision(self, nodeorrev, _df=None, raw=False): |
|
1646 | 1646 | """return an uncompressed revision of a given node or revision |
|
1647 | 1647 | number. |
|
1648 | 1648 | |
|
1649 | 1649 | _df - an existing file handle to read from. (internal-only) |
|
1650 | 1650 | raw - an optional argument specifying if the revision data is to be |
|
1651 | 1651 | treated as raw data when applying flag transforms. 'raw' should be set |
|
1652 | 1652 | to True when generating changegroups or in debug commands. |
|
1653 | 1653 | """ |
|
1654 | 1654 | if isinstance(nodeorrev, int): |
|
1655 | 1655 | rev = nodeorrev |
|
1656 | 1656 | node = self.node(rev) |
|
1657 | 1657 | else: |
|
1658 | 1658 | node = nodeorrev |
|
1659 | 1659 | rev = None |
|
1660 | 1660 | |
|
1661 | 1661 | cachedrev = None |
|
1662 | 1662 | flags = None |
|
1663 | 1663 | rawtext = None |
|
1664 | 1664 | if node == nullid: |
|
1665 | 1665 | return "" |
|
1666 | 1666 | if self._revisioncache: |
|
1667 | 1667 | if self._revisioncache[0] == node: |
|
1668 | 1668 | # _cache only stores rawtext |
|
1669 | 1669 | if raw: |
|
1670 | 1670 | return self._revisioncache[2] |
|
1671 | 1671 | # duplicated, but good for perf |
|
1672 | 1672 | if rev is None: |
|
1673 | 1673 | rev = self.rev(node) |
|
1674 | 1674 | if flags is None: |
|
1675 | 1675 | flags = self.flags(rev) |
|
1676 | 1676 | # no extra flags set, no flag processor runs, text = rawtext |
|
1677 | 1677 | if flags == REVIDX_DEFAULT_FLAGS: |
|
1678 | 1678 | return self._revisioncache[2] |
|
1679 | 1679 | # rawtext is reusable. need to run flag processor |
|
1680 | 1680 | rawtext = self._revisioncache[2] |
|
1681 | 1681 | |
|
1682 | 1682 | cachedrev = self._revisioncache[1] |
|
1683 | 1683 | |
|
1684 | 1684 | # look up what we need to read |
|
1685 | 1685 | if rawtext is None: |
|
1686 | 1686 | if rev is None: |
|
1687 | 1687 | rev = self.rev(node) |
|
1688 | 1688 | |
|
1689 | 1689 | chain, stopped = self._deltachain(rev, stoprev=cachedrev) |
|
1690 | 1690 | if stopped: |
|
1691 | 1691 | rawtext = self._revisioncache[2] |
|
1692 | 1692 | |
|
1693 | 1693 | # drop cache to save memory |
|
1694 | 1694 | self._revisioncache = None |
|
1695 | 1695 | |
|
1696 | 1696 | targetsize = None |
|
1697 | 1697 | rawsize = self.index[rev][2] |
|
1698 | 1698 | if 0 <= rawsize: |
|
1699 | 1699 | targetsize = 4 * rawsize |
|
1700 | 1700 | |
|
1701 | 1701 | bins = self._chunks(chain, df=_df, targetsize=targetsize) |
|
1702 | 1702 | if rawtext is None: |
|
1703 | 1703 | rawtext = bytes(bins[0]) |
|
1704 | 1704 | bins = bins[1:] |
|
1705 | 1705 | |
|
1706 | 1706 | rawtext = mdiff.patches(rawtext, bins) |
|
1707 | 1707 | self._revisioncache = (node, rev, rawtext) |
|
1708 | 1708 | |
|
1709 | 1709 | if flags is None: |
|
1710 | 1710 | if rev is None: |
|
1711 | 1711 | rev = self.rev(node) |
|
1712 | 1712 | flags = self.flags(rev) |
|
1713 | 1713 | |
|
1714 | 1714 | text, validatehash = self._processflags(rawtext, flags, 'read', raw=raw) |
|
1715 | 1715 | if validatehash: |
|
1716 | 1716 | self.checkhash(text, node, rev=rev) |
|
1717 | 1717 | |
|
1718 | 1718 | return text |
|
1719 | 1719 | |
|
1720 | 1720 | def hash(self, text, p1, p2): |
|
1721 | 1721 | """Compute a node hash. |
|
1722 | 1722 | |
|
1723 | 1723 | Available as a function so that subclasses can replace the hash |
|
1724 | 1724 | as needed. |
|
1725 | 1725 | """ |
|
1726 | 1726 | return storageutil.hashrevisionsha1(text, p1, p2) |
|
1727 | 1727 | |
|
1728 | 1728 | def _processflags(self, text, flags, operation, raw=False): |
|
1729 | 1729 | """Inspect revision data flags and applies transforms defined by |
|
1730 | 1730 | registered flag processors. |
|
1731 | 1731 | |
|
1732 | 1732 | ``text`` - the revision data to process |
|
1733 | 1733 | ``flags`` - the revision flags |
|
1734 | 1734 | ``operation`` - the operation being performed (read or write) |
|
1735 | 1735 | ``raw`` - an optional argument describing if the raw transform should be |
|
1736 | 1736 | applied. |
|
1737 | 1737 | |
|
1738 | 1738 | This method processes the flags in the order (or reverse order if |
|
1739 | 1739 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the |
|
1740 | 1740 | flag processors registered for present flags. The order of flags defined |
|
1741 | 1741 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. |
|
1742 | 1742 | |
|
1743 | 1743 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the |
|
1744 | 1744 | processed text and ``validatehash`` is a bool indicating whether the |
|
1745 | 1745 | returned text should be checked for hash integrity. |
|
1746 | 1746 | |
|
1747 | 1747 | Note: If the ``raw`` argument is set, it has precedence over the |
|
1748 | 1748 | operation and will only update the value of ``validatehash``. |
|
1749 | 1749 | """ |
|
1750 | 1750 | # fast path: no flag processors will run |
|
1751 | 1751 | if flags == 0: |
|
1752 | 1752 | return text, True |
|
1753 | 1753 | if not operation in ('read', 'write'): |
|
1754 | 1754 | raise error.ProgrammingError(_("invalid '%s' operation") % |
|
1755 | 1755 | operation) |
|
1756 | 1756 | # Check all flags are known. |
|
1757 | 1757 | if flags & ~REVIDX_KNOWN_FLAGS: |
|
1758 | 1758 | raise error.RevlogError(_("incompatible revision flag '%#x'") % |
|
1759 | 1759 | (flags & ~REVIDX_KNOWN_FLAGS)) |
|
1760 | 1760 | validatehash = True |
|
1761 | 1761 | # Depending on the operation (read or write), the order might be |
|
1762 | 1762 | # reversed due to non-commutative transforms. |
|
1763 | 1763 | orderedflags = REVIDX_FLAGS_ORDER |
|
1764 | 1764 | if operation == 'write': |
|
1765 | 1765 | orderedflags = reversed(orderedflags) |
|
1766 | 1766 | |
|
1767 | 1767 | for flag in orderedflags: |
|
1768 | 1768 | # If a flagprocessor has been registered for a known flag, apply the |
|
1769 | 1769 | # related operation transform and update result tuple. |
|
1770 | 1770 | if flag & flags: |
|
1771 | 1771 | vhash = True |
|
1772 | 1772 | |
|
1773 | 1773 | if flag not in self._flagprocessors: |
|
1774 | 1774 | message = _("missing processor for flag '%#x'") % (flag) |
|
1775 | 1775 | raise error.RevlogError(message) |
|
1776 | 1776 | |
|
1777 | 1777 | processor = self._flagprocessors[flag] |
|
1778 | 1778 | if processor is not None: |
|
1779 | 1779 | readtransform, writetransform, rawtransform = processor |
|
1780 | 1780 | |
|
1781 | 1781 | if raw: |
|
1782 | 1782 | vhash = rawtransform(self, text) |
|
1783 | 1783 | elif operation == 'read': |
|
1784 | 1784 | text, vhash = readtransform(self, text) |
|
1785 | 1785 | else: # write operation |
|
1786 | 1786 | text, vhash = writetransform(self, text) |
|
1787 | 1787 | validatehash = validatehash and vhash |
|
1788 | 1788 | |
|
1789 | 1789 | return text, validatehash |
|
1790 | 1790 | |
|
1791 | 1791 | def checkhash(self, text, node, p1=None, p2=None, rev=None): |
|
1792 | 1792 | """Check node hash integrity. |
|
1793 | 1793 | |
|
1794 | 1794 | Available as a function so that subclasses can extend hash mismatch |
|
1795 | 1795 | behaviors as needed. |
|
1796 | 1796 | """ |
|
1797 | 1797 | try: |
|
1798 | 1798 | if p1 is None and p2 is None: |
|
1799 | 1799 | p1, p2 = self.parents(node) |
|
1800 | 1800 | if node != self.hash(text, p1, p2): |
|
1801 | 1801 | # Clear the revision cache on hash failure. The revision cache |
|
1802 | 1802 | # only stores the raw revision and clearing the cache does have |
|
1803 | 1803 | # the side-effect that we won't have a cache hit when the raw |
|
1804 | 1804 | # revision data is accessed. But this case should be rare and |
|
1805 | 1805 | # it is extra work to teach the cache about the hash |
|
1806 | 1806 | # verification state. |
|
1807 | 1807 | if self._revisioncache and self._revisioncache[0] == node: |
|
1808 | 1808 | self._revisioncache = None |
|
1809 | 1809 | |
|
1810 | 1810 | revornode = rev |
|
1811 | 1811 | if revornode is None: |
|
1812 | 1812 | revornode = templatefilters.short(hex(node)) |
|
1813 | 1813 | raise error.RevlogError(_("integrity check failed on %s:%s") |
|
1814 | 1814 | % (self.indexfile, pycompat.bytestr(revornode))) |
|
1815 | 1815 | except error.RevlogError: |
|
1816 | 1816 | if self._censorable and storageutil.iscensoredtext(text): |
|
1817 | 1817 | raise error.CensoredNodeError(self.indexfile, node, text) |
|
1818 | 1818 | raise |
|
1819 | 1819 | |
|
1820 | 1820 | def _enforceinlinesize(self, tr, fp=None): |
|
1821 | 1821 | """Check if the revlog is too big for inline and convert if so. |
|
1822 | 1822 | |
|
1823 | 1823 | This should be called after revisions are added to the revlog. If the |
|
1824 | 1824 | revlog has grown too large to be an inline revlog, it will convert it |
|
1825 | 1825 | to use multiple index and data files. |
|
1826 | 1826 | """ |
|
1827 | 1827 | tiprev = len(self) - 1 |
|
1828 | 1828 | if (not self._inline or |
|
1829 | 1829 | (self.start(tiprev) + self.length(tiprev)) < _maxinline): |
|
1830 | 1830 | return |
|
1831 | 1831 | |
|
1832 | 1832 | trinfo = tr.find(self.indexfile) |
|
1833 | 1833 | if trinfo is None: |
|
1834 | 1834 | raise error.RevlogError(_("%s not found in the transaction") |
|
1835 | 1835 | % self.indexfile) |
|
1836 | 1836 | |
|
1837 | 1837 | trindex = trinfo[2] |
|
1838 | 1838 | if trindex is not None: |
|
1839 | 1839 | dataoff = self.start(trindex) |
|
1840 | 1840 | else: |
|
1841 | 1841 | # revlog was stripped at start of transaction, use all leftover data |
|
1842 | 1842 | trindex = len(self) - 1 |
|
1843 | 1843 | dataoff = self.end(tiprev) |
|
1844 | 1844 | |
|
1845 | 1845 | tr.add(self.datafile, dataoff) |
|
1846 | 1846 | |
|
1847 | 1847 | if fp: |
|
1848 | 1848 | fp.flush() |
|
1849 | 1849 | fp.close() |
|
1850 | 1850 | # We can't use the cached file handle after close(). So prevent |
|
1851 | 1851 | # its usage. |
|
1852 | 1852 | self._writinghandles = None |
|
1853 | 1853 | |
|
1854 | 1854 | with self._indexfp('r') as ifh, self._datafp('w') as dfh: |
|
1855 | 1855 | for r in self: |
|
1856 | 1856 | dfh.write(self._getsegmentforrevs(r, r, df=ifh)[1]) |
|
1857 | 1857 | |
|
1858 | 1858 | with self._indexfp('w') as fp: |
|
1859 | 1859 | self.version &= ~FLAG_INLINE_DATA |
|
1860 | 1860 | self._inline = False |
|
1861 | 1861 | io = self._io |
|
1862 | 1862 | for i in self: |
|
1863 | 1863 | e = io.packentry(self.index[i], self.node, self.version, i) |
|
1864 | 1864 | fp.write(e) |
|
1865 | 1865 | |
|
1866 | 1866 | # the temp file replace the real index when we exit the context |
|
1867 | 1867 | # manager |
|
1868 | 1868 | |
|
1869 | 1869 | tr.replace(self.indexfile, trindex * self._io.size) |
|
1870 | 1870 | self._chunkclear() |
|
1871 | 1871 | |
|
1872 | 1872 | def _nodeduplicatecallback(self, transaction, node): |
|
1873 | 1873 | """called when trying to add a node already stored. |
|
1874 | 1874 | """ |
|
1875 | 1875 | |
|
1876 | 1876 | def addrevision(self, text, transaction, link, p1, p2, cachedelta=None, |
|
1877 | 1877 | node=None, flags=REVIDX_DEFAULT_FLAGS, deltacomputer=None): |
|
1878 | 1878 | """add a revision to the log |
|
1879 | 1879 | |
|
1880 | 1880 | text - the revision data to add |
|
1881 | 1881 | transaction - the transaction object used for rollback |
|
1882 | 1882 | link - the linkrev data to add |
|
1883 | 1883 | p1, p2 - the parent nodeids of the revision |
|
1884 | 1884 | cachedelta - an optional precomputed delta |
|
1885 | 1885 | node - nodeid of revision; typically node is not specified, and it is |
|
1886 | 1886 | computed by default as hash(text, p1, p2), however subclasses might |
|
1887 | 1887 | use different hashing method (and override checkhash() in such case) |
|
1888 | 1888 | flags - the known flags to set on the revision |
|
1889 | 1889 | deltacomputer - an optional deltacomputer instance shared between |
|
1890 | 1890 | multiple calls |
|
1891 | 1891 | """ |
|
1892 | 1892 | if link == nullrev: |
|
1893 | 1893 | raise error.RevlogError(_("attempted to add linkrev -1 to %s") |
|
1894 | 1894 | % self.indexfile) |
|
1895 | 1895 | |
|
1896 | 1896 | if flags: |
|
1897 | 1897 | node = node or self.hash(text, p1, p2) |
|
1898 | 1898 | |
|
1899 | 1899 | rawtext, validatehash = self._processflags(text, flags, 'write') |
|
1900 | 1900 | |
|
1901 | 1901 | # If the flag processor modifies the revision data, ignore any provided |
|
1902 | 1902 | # cachedelta. |
|
1903 | 1903 | if rawtext != text: |
|
1904 | 1904 | cachedelta = None |
|
1905 | 1905 | |
|
1906 | 1906 | if len(rawtext) > _maxentrysize: |
|
1907 | 1907 | raise error.RevlogError( |
|
1908 | 1908 | _("%s: size of %d bytes exceeds maximum revlog storage of 2GiB") |
|
1909 | 1909 | % (self.indexfile, len(rawtext))) |
|
1910 | 1910 | |
|
1911 | 1911 | node = node or self.hash(rawtext, p1, p2) |
|
1912 | 1912 | if node in self.nodemap: |
|
1913 | 1913 | return node |
|
1914 | 1914 | |
|
1915 | 1915 | if validatehash: |
|
1916 | 1916 | self.checkhash(rawtext, node, p1=p1, p2=p2) |
|
1917 | 1917 | |
|
1918 | 1918 | return self.addrawrevision(rawtext, transaction, link, p1, p2, node, |
|
1919 | 1919 | flags, cachedelta=cachedelta, |
|
1920 | 1920 | deltacomputer=deltacomputer) |
|
1921 | 1921 | |
|
1922 | 1922 | def addrawrevision(self, rawtext, transaction, link, p1, p2, node, flags, |
|
1923 | 1923 | cachedelta=None, deltacomputer=None): |
|
1924 | 1924 | """add a raw revision with known flags, node and parents |
|
1925 | 1925 | useful when reusing a revision not stored in this revlog (ex: received |
|
1926 | 1926 | over wire, or read from an external bundle). |
|
1927 | 1927 | """ |
|
1928 | 1928 | dfh = None |
|
1929 | 1929 | if not self._inline: |
|
1930 | 1930 | dfh = self._datafp("a+") |
|
1931 | 1931 | ifh = self._indexfp("a+") |
|
1932 | 1932 | try: |
|
1933 | 1933 | return self._addrevision(node, rawtext, transaction, link, p1, p2, |
|
1934 | 1934 | flags, cachedelta, ifh, dfh, |
|
1935 | 1935 | deltacomputer=deltacomputer) |
|
1936 | 1936 | finally: |
|
1937 | 1937 | if dfh: |
|
1938 | 1938 | dfh.close() |
|
1939 | 1939 | ifh.close() |
|
1940 | 1940 | |
|
1941 | 1941 | def compress(self, data): |
|
1942 | 1942 | """Generate a possibly-compressed representation of data.""" |
|
1943 | 1943 | if not data: |
|
1944 | 1944 | return '', data |
|
1945 | 1945 | |
|
1946 | 1946 | compressed = self._compressor.compress(data) |
|
1947 | 1947 | |
|
1948 | 1948 | if compressed: |
|
1949 | 1949 | # The revlog compressor added the header in the returned data. |
|
1950 | 1950 | return '', compressed |
|
1951 | 1951 | |
|
1952 | 1952 | if data[0:1] == '\0': |
|
1953 | 1953 | return '', data |
|
1954 | 1954 | return 'u', data |
|
1955 | 1955 | |
|
1956 | 1956 | def decompress(self, data): |
|
1957 | 1957 | """Decompress a revlog chunk. |
|
1958 | 1958 | |
|
1959 | 1959 | The chunk is expected to begin with a header identifying the |
|
1960 | 1960 | format type so it can be routed to an appropriate decompressor. |
|
1961 | 1961 | """ |
|
1962 | 1962 | if not data: |
|
1963 | 1963 | return data |
|
1964 | 1964 | |
|
1965 | 1965 | # Revlogs are read much more frequently than they are written and many |
|
1966 | 1966 | # chunks only take microseconds to decompress, so performance is |
|
1967 | 1967 | # important here. |
|
1968 | 1968 | # |
|
1969 | 1969 | # We can make a few assumptions about revlogs: |
|
1970 | 1970 | # |
|
1971 | 1971 | # 1) the majority of chunks will be compressed (as opposed to inline |
|
1972 | 1972 | # raw data). |
|
1973 | 1973 | # 2) decompressing *any* data will likely by at least 10x slower than |
|
1974 | 1974 | # returning raw inline data. |
|
1975 | 1975 | # 3) we want to prioritize common and officially supported compression |
|
1976 | 1976 | # engines |
|
1977 | 1977 | # |
|
1978 | 1978 | # It follows that we want to optimize for "decompress compressed data |
|
1979 | 1979 | # when encoded with common and officially supported compression engines" |
|
1980 | 1980 | # case over "raw data" and "data encoded by less common or non-official |
|
1981 | 1981 | # compression engines." That is why we have the inline lookup first |
|
1982 | 1982 | # followed by the compengines lookup. |
|
1983 | 1983 | # |
|
1984 | 1984 | # According to `hg perfrevlogchunks`, this is ~0.5% faster for zlib |
|
1985 | 1985 | # compressed chunks. And this matters for changelog and manifest reads. |
|
1986 | 1986 | t = data[0:1] |
|
1987 | 1987 | |
|
1988 | 1988 | if t == 'x': |
|
1989 | 1989 | try: |
|
1990 | 1990 | return _zlibdecompress(data) |
|
1991 | 1991 | except zlib.error as e: |
|
1992 | 1992 | raise error.RevlogError(_('revlog decompress error: %s') % |
|
1993 | 1993 | stringutil.forcebytestr(e)) |
|
1994 | 1994 | # '\0' is more common than 'u' so it goes first. |
|
1995 | 1995 | elif t == '\0': |
|
1996 | 1996 | return data |
|
1997 | 1997 | elif t == 'u': |
|
1998 | 1998 | return util.buffer(data, 1) |
|
1999 | 1999 | |
|
2000 | 2000 | try: |
|
2001 | 2001 | compressor = self._decompressors[t] |
|
2002 | 2002 | except KeyError: |
|
2003 | 2003 | try: |
|
2004 | 2004 | engine = util.compengines.forrevlogheader(t) |
|
2005 | 2005 | compressor = engine.revlogcompressor(self._compengineopts) |
|
2006 | 2006 | self._decompressors[t] = compressor |
|
2007 | 2007 | except KeyError: |
|
2008 | 2008 | raise error.RevlogError(_('unknown compression type %r') % t) |
|
2009 | 2009 | |
|
2010 | 2010 | return compressor.decompress(data) |
|
2011 | 2011 | |
|
2012 | 2012 | def _addrevision(self, node, rawtext, transaction, link, p1, p2, flags, |
|
2013 | 2013 | cachedelta, ifh, dfh, alwayscache=False, |
|
2014 | 2014 | deltacomputer=None): |
|
2015 | 2015 | """internal function to add revisions to the log |
|
2016 | 2016 | |
|
2017 | 2017 | see addrevision for argument descriptions. |
|
2018 | 2018 | |
|
2019 | 2019 | note: "addrevision" takes non-raw text, "_addrevision" takes raw text. |
|
2020 | 2020 | |
|
2021 | 2021 | if "deltacomputer" is not provided or None, a defaultdeltacomputer will |
|
2022 | 2022 | be used. |
|
2023 | 2023 | |
|
2024 | 2024 | invariants: |
|
2025 | 2025 | - rawtext is optional (can be None); if not set, cachedelta must be set. |
|
2026 | 2026 | if both are set, they must correspond to each other. |
|
2027 | 2027 | """ |
|
2028 | 2028 | if node == nullid: |
|
2029 | 2029 | raise error.RevlogError(_("%s: attempt to add null revision") % |
|
2030 | 2030 | self.indexfile) |
|
2031 | 2031 | if node == wdirid or node in wdirfilenodeids: |
|
2032 | 2032 | raise error.RevlogError(_("%s: attempt to add wdir revision") % |
|
2033 | 2033 | self.indexfile) |
|
2034 | 2034 | |
|
2035 | 2035 | if self._inline: |
|
2036 | 2036 | fh = ifh |
|
2037 | 2037 | else: |
|
2038 | 2038 | fh = dfh |
|
2039 | 2039 | |
|
2040 | 2040 | btext = [rawtext] |
|
2041 | 2041 | |
|
2042 | 2042 | curr = len(self) |
|
2043 | 2043 | prev = curr - 1 |
|
2044 | 2044 | offset = self.end(prev) |
|
2045 | 2045 | p1r, p2r = self.rev(p1), self.rev(p2) |
|
2046 | 2046 | |
|
2047 | 2047 | # full versions are inserted when the needed deltas |
|
2048 | 2048 | # become comparable to the uncompressed text |
|
2049 | 2049 | if rawtext is None: |
|
2050 | 2050 | # need rawtext size, before changed by flag processors, which is |
|
2051 | 2051 | # the non-raw size. use revlog explicitly to avoid filelog's extra |
|
2052 | 2052 | # logic that might remove metadata size. |
|
2053 | 2053 | textlen = mdiff.patchedsize(revlog.size(self, cachedelta[0]), |
|
2054 | 2054 | cachedelta[1]) |
|
2055 | 2055 | else: |
|
2056 | 2056 | textlen = len(rawtext) |
|
2057 | 2057 | |
|
2058 | 2058 | if deltacomputer is None: |
|
2059 | 2059 | deltacomputer = deltautil.deltacomputer(self) |
|
2060 | 2060 | |
|
2061 | 2061 | revinfo = _revisioninfo(node, p1, p2, btext, textlen, cachedelta, flags) |
|
2062 | 2062 | |
|
2063 | 2063 | deltainfo = deltacomputer.finddeltainfo(revinfo, fh) |
|
2064 | 2064 | |
|
2065 | 2065 | e = (offset_type(offset, flags), deltainfo.deltalen, textlen, |
|
2066 | 2066 | deltainfo.base, link, p1r, p2r, node) |
|
2067 | 2067 | self.index.append(e) |
|
2068 | 2068 | self.nodemap[node] = curr |
|
2069 | 2069 | |
|
2070 | 2070 | # Reset the pure node cache start lookup offset to account for new |
|
2071 | 2071 | # revision. |
|
2072 | 2072 | if self._nodepos is not None: |
|
2073 | 2073 | self._nodepos = curr |
|
2074 | 2074 | |
|
2075 | 2075 | entry = self._io.packentry(e, self.node, self.version, curr) |
|
2076 | 2076 | self._writeentry(transaction, ifh, dfh, entry, deltainfo.data, |
|
2077 | 2077 | link, offset) |
|
2078 | 2078 | |
|
2079 | 2079 | rawtext = btext[0] |
|
2080 | 2080 | |
|
2081 | 2081 | if alwayscache and rawtext is None: |
|
2082 | 2082 | rawtext = deltacomputer.buildtext(revinfo, fh) |
|
2083 | 2083 | |
|
2084 | 2084 | if type(rawtext) == bytes: # only accept immutable objects |
|
2085 | 2085 | self._revisioncache = (node, curr, rawtext) |
|
2086 | 2086 | self._chainbasecache[curr] = deltainfo.chainbase |
|
2087 | 2087 | return node |
|
2088 | 2088 | |
|
2089 | 2089 | def _writeentry(self, transaction, ifh, dfh, entry, data, link, offset): |
|
2090 | 2090 | # Files opened in a+ mode have inconsistent behavior on various |
|
2091 | 2091 | # platforms. Windows requires that a file positioning call be made |
|
2092 | 2092 | # when the file handle transitions between reads and writes. See |
|
2093 | 2093 | # 3686fa2b8eee and the mixedfilemodewrapper in windows.py. On other |
|
2094 | 2094 | # platforms, Python or the platform itself can be buggy. Some versions |
|
2095 | 2095 | # of Solaris have been observed to not append at the end of the file |
|
2096 | 2096 | # if the file was seeked to before the end. See issue4943 for more. |
|
2097 | 2097 | # |
|
2098 | 2098 | # We work around this issue by inserting a seek() before writing. |
|
2099 | 2099 | # Note: This is likely not necessary on Python 3. However, because |
|
2100 | 2100 | # the file handle is reused for reads and may be seeked there, we need |
|
2101 | 2101 | # to be careful before changing this. |
|
2102 | 2102 | ifh.seek(0, os.SEEK_END) |
|
2103 | 2103 | if dfh: |
|
2104 | 2104 | dfh.seek(0, os.SEEK_END) |
|
2105 | 2105 | |
|
2106 | 2106 | curr = len(self) - 1 |
|
2107 | 2107 | if not self._inline: |
|
2108 | 2108 | transaction.add(self.datafile, offset) |
|
2109 | 2109 | transaction.add(self.indexfile, curr * len(entry)) |
|
2110 | 2110 | if data[0]: |
|
2111 | 2111 | dfh.write(data[0]) |
|
2112 | 2112 | dfh.write(data[1]) |
|
2113 | 2113 | ifh.write(entry) |
|
2114 | 2114 | else: |
|
2115 | 2115 | offset += curr * self._io.size |
|
2116 | 2116 | transaction.add(self.indexfile, offset, curr) |
|
2117 | 2117 | ifh.write(entry) |
|
2118 | 2118 | ifh.write(data[0]) |
|
2119 | 2119 | ifh.write(data[1]) |
|
2120 | 2120 | self._enforceinlinesize(transaction, ifh) |
|
2121 | 2121 | |
|
2122 | 2122 | def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None): |
|
2123 | 2123 | """ |
|
2124 | 2124 | add a delta group |
|
2125 | 2125 | |
|
2126 | 2126 | given a set of deltas, add them to the revision log. the |
|
2127 | 2127 | first delta is against its parent, which should be in our |
|
2128 | 2128 | log, the rest are against the previous delta. |
|
2129 | 2129 | |
|
2130 | 2130 | If ``addrevisioncb`` is defined, it will be called with arguments of |
|
2131 | 2131 | this revlog and the node that was added. |
|
2132 | 2132 | """ |
|
2133 | 2133 | |
|
2134 | 2134 | if self._writinghandles: |
|
2135 | 2135 | raise error.ProgrammingError('cannot nest addgroup() calls') |
|
2136 | 2136 | |
|
2137 | 2137 | nodes = [] |
|
2138 | 2138 | |
|
2139 | 2139 | r = len(self) |
|
2140 | 2140 | end = 0 |
|
2141 | 2141 | if r: |
|
2142 | 2142 | end = self.end(r - 1) |
|
2143 | 2143 | ifh = self._indexfp("a+") |
|
2144 | 2144 | isize = r * self._io.size |
|
2145 | 2145 | if self._inline: |
|
2146 | 2146 | transaction.add(self.indexfile, end + isize, r) |
|
2147 | 2147 | dfh = None |
|
2148 | 2148 | else: |
|
2149 | 2149 | transaction.add(self.indexfile, isize, r) |
|
2150 | 2150 | transaction.add(self.datafile, end) |
|
2151 | 2151 | dfh = self._datafp("a+") |
|
2152 | 2152 | def flush(): |
|
2153 | 2153 | if dfh: |
|
2154 | 2154 | dfh.flush() |
|
2155 | 2155 | ifh.flush() |
|
2156 | 2156 | |
|
2157 | 2157 | self._writinghandles = (ifh, dfh) |
|
2158 | 2158 | |
|
2159 | 2159 | try: |
|
2160 | 2160 | deltacomputer = deltautil.deltacomputer(self) |
|
2161 | 2161 | # loop through our set of deltas |
|
2162 | 2162 | for data in deltas: |
|
2163 | 2163 | node, p1, p2, linknode, deltabase, delta, flags = data |
|
2164 | 2164 | link = linkmapper(linknode) |
|
2165 | 2165 | flags = flags or REVIDX_DEFAULT_FLAGS |
|
2166 | 2166 | |
|
2167 | 2167 | nodes.append(node) |
|
2168 | 2168 | |
|
2169 | 2169 | if node in self.nodemap: |
|
2170 | 2170 | self._nodeduplicatecallback(transaction, node) |
|
2171 | 2171 | # this can happen if two branches make the same change |
|
2172 | 2172 | continue |
|
2173 | 2173 | |
|
2174 | 2174 | for p in (p1, p2): |
|
2175 | 2175 | if p not in self.nodemap: |
|
2176 | 2176 | raise error.LookupError(p, self.indexfile, |
|
2177 | 2177 | _('unknown parent')) |
|
2178 | 2178 | |
|
2179 | 2179 | if deltabase not in self.nodemap: |
|
2180 | 2180 | raise error.LookupError(deltabase, self.indexfile, |
|
2181 | 2181 | _('unknown delta base')) |
|
2182 | 2182 | |
|
2183 | 2183 | baserev = self.rev(deltabase) |
|
2184 | 2184 | |
|
2185 | 2185 | if baserev != nullrev and self.iscensored(baserev): |
|
2186 | 2186 | # if base is censored, delta must be full replacement in a |
|
2187 | 2187 | # single patch operation |
|
2188 | 2188 | hlen = struct.calcsize(">lll") |
|
2189 | 2189 | oldlen = self.rawsize(baserev) |
|
2190 | 2190 | newlen = len(delta) - hlen |
|
2191 | 2191 | if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen): |
|
2192 | 2192 | raise error.CensoredBaseError(self.indexfile, |
|
2193 | 2193 | self.node(baserev)) |
|
2194 | 2194 | |
|
2195 | 2195 | if not flags and self._peek_iscensored(baserev, delta, flush): |
|
2196 | 2196 | flags |= REVIDX_ISCENSORED |
|
2197 | 2197 | |
|
2198 | 2198 | # We assume consumers of addrevisioncb will want to retrieve |
|
2199 | 2199 | # the added revision, which will require a call to |
|
2200 | 2200 | # revision(). revision() will fast path if there is a cache |
|
2201 | 2201 | # hit. So, we tell _addrevision() to always cache in this case. |
|
2202 | 2202 | # We're only using addgroup() in the context of changegroup |
|
2203 | 2203 | # generation so the revision data can always be handled as raw |
|
2204 | 2204 | # by the flagprocessor. |
|
2205 | 2205 | self._addrevision(node, None, transaction, link, |
|
2206 | 2206 | p1, p2, flags, (baserev, delta), |
|
2207 | 2207 | ifh, dfh, |
|
2208 | 2208 | alwayscache=bool(addrevisioncb), |
|
2209 | 2209 | deltacomputer=deltacomputer) |
|
2210 | 2210 | |
|
2211 | 2211 | if addrevisioncb: |
|
2212 | 2212 | addrevisioncb(self, node) |
|
2213 | 2213 | |
|
2214 | 2214 | if not dfh and not self._inline: |
|
2215 | 2215 | # addrevision switched from inline to conventional |
|
2216 | 2216 | # reopen the index |
|
2217 | 2217 | ifh.close() |
|
2218 | 2218 | dfh = self._datafp("a+") |
|
2219 | 2219 | ifh = self._indexfp("a+") |
|
2220 | 2220 | self._writinghandles = (ifh, dfh) |
|
2221 | 2221 | finally: |
|
2222 | 2222 | self._writinghandles = None |
|
2223 | 2223 | |
|
2224 | 2224 | if dfh: |
|
2225 | 2225 | dfh.close() |
|
2226 | 2226 | ifh.close() |
|
2227 | 2227 | |
|
2228 | 2228 | return nodes |
|
2229 | 2229 | |
|
2230 | 2230 | def iscensored(self, rev): |
|
2231 | 2231 | """Check if a file revision is censored.""" |
|
2232 | 2232 | if not self._censorable: |
|
2233 | 2233 | return False |
|
2234 | 2234 | |
|
2235 | 2235 | return self.flags(rev) & REVIDX_ISCENSORED |
|
2236 | 2236 | |
|
2237 | 2237 | def _peek_iscensored(self, baserev, delta, flush): |
|
2238 | 2238 | """Quickly check if a delta produces a censored revision.""" |
|
2239 | 2239 | if not self._censorable: |
|
2240 | 2240 | return False |
|
2241 | 2241 | |
|
2242 | 2242 | return storageutil.deltaiscensored(delta, baserev, self.rawsize) |
|
2243 | 2243 | |
|
2244 | 2244 | def getstrippoint(self, minlink): |
|
2245 | 2245 | """find the minimum rev that must be stripped to strip the linkrev |
|
2246 | 2246 | |
|
2247 | 2247 | Returns a tuple containing the minimum rev and a set of all revs that |
|
2248 | 2248 | have linkrevs that will be broken by this strip. |
|
2249 | 2249 | """ |
|
2250 | 2250 | return storageutil.resolvestripinfo(minlink, len(self) - 1, |
|
2251 | 2251 | self.headrevs(), |
|
2252 | 2252 | self.linkrev, self.parentrevs) |
|
2253 | 2253 | |
|
2254 | 2254 | def strip(self, minlink, transaction): |
|
2255 | 2255 | """truncate the revlog on the first revision with a linkrev >= minlink |
|
2256 | 2256 | |
|
2257 | 2257 | This function is called when we're stripping revision minlink and |
|
2258 | 2258 | its descendants from the repository. |
|
2259 | 2259 | |
|
2260 | 2260 | We have to remove all revisions with linkrev >= minlink, because |
|
2261 | 2261 | the equivalent changelog revisions will be renumbered after the |
|
2262 | 2262 | strip. |
|
2263 | 2263 | |
|
2264 | 2264 | So we truncate the revlog on the first of these revisions, and |
|
2265 | 2265 | trust that the caller has saved the revisions that shouldn't be |
|
2266 | 2266 | removed and that it'll re-add them after this truncation. |
|
2267 | 2267 | """ |
|
2268 | 2268 | if len(self) == 0: |
|
2269 | 2269 | return |
|
2270 | 2270 | |
|
2271 | 2271 | rev, _ = self.getstrippoint(minlink) |
|
2272 | 2272 | if rev == len(self): |
|
2273 | 2273 | return |
|
2274 | 2274 | |
|
2275 | 2275 | # first truncate the files on disk |
|
2276 | 2276 | end = self.start(rev) |
|
2277 | 2277 | if not self._inline: |
|
2278 | 2278 | transaction.add(self.datafile, end) |
|
2279 | 2279 | end = rev * self._io.size |
|
2280 | 2280 | else: |
|
2281 | 2281 | end += rev * self._io.size |
|
2282 | 2282 | |
|
2283 | 2283 | transaction.add(self.indexfile, end) |
|
2284 | 2284 | |
|
2285 | 2285 | # then reset internal state in memory to forget those revisions |
|
2286 | 2286 | self._revisioncache = None |
|
2287 | 2287 | self._chaininfocache = {} |
|
2288 | 2288 | self._chunkclear() |
|
2289 | 2289 | for x in pycompat.xrange(rev, len(self)): |
|
2290 | 2290 | del self.nodemap[self.node(x)] |
|
2291 | 2291 | |
|
2292 | 2292 | del self.index[rev:-1] |
|
2293 | 2293 | self._nodepos = None |
|
2294 | 2294 | |
|
2295 | 2295 | def checksize(self): |
|
2296 | 2296 | """Check size of index and data files |
|
2297 | 2297 | |
|
2298 | 2298 | return a (dd, di) tuple. |
|
2299 | 2299 | - dd: extra bytes for the "data" file |
|
2300 | 2300 | - di: extra bytes for the "index" file |
|
2301 | 2301 | |
|
2302 | 2302 | A healthy revlog will return (0, 0). |
|
2303 | 2303 | """ |
|
2304 | 2304 | expected = 0 |
|
2305 | 2305 | if len(self): |
|
2306 | 2306 | expected = max(0, self.end(len(self) - 1)) |
|
2307 | 2307 | |
|
2308 | 2308 | try: |
|
2309 | 2309 | with self._datafp() as f: |
|
2310 | 2310 | f.seek(0, io.SEEK_END) |
|
2311 | 2311 | actual = f.tell() |
|
2312 | 2312 | dd = actual - expected |
|
2313 | 2313 | except IOError as inst: |
|
2314 | 2314 | if inst.errno != errno.ENOENT: |
|
2315 | 2315 | raise |
|
2316 | 2316 | dd = 0 |
|
2317 | 2317 | |
|
2318 | 2318 | try: |
|
2319 | 2319 | f = self.opener(self.indexfile) |
|
2320 | 2320 | f.seek(0, io.SEEK_END) |
|
2321 | 2321 | actual = f.tell() |
|
2322 | 2322 | f.close() |
|
2323 | 2323 | s = self._io.size |
|
2324 | 2324 | i = max(0, actual // s) |
|
2325 | 2325 | di = actual - (i * s) |
|
2326 | 2326 | if self._inline: |
|
2327 | 2327 | databytes = 0 |
|
2328 | 2328 | for r in self: |
|
2329 | 2329 | databytes += max(0, self.length(r)) |
|
2330 | 2330 | dd = 0 |
|
2331 | 2331 | di = actual - len(self) * s - databytes |
|
2332 | 2332 | except IOError as inst: |
|
2333 | 2333 | if inst.errno != errno.ENOENT: |
|
2334 | 2334 | raise |
|
2335 | 2335 | di = 0 |
|
2336 | 2336 | |
|
2337 | 2337 | return (dd, di) |
|
2338 | 2338 | |
|
2339 | 2339 | def files(self): |
|
2340 | 2340 | res = [self.indexfile] |
|
2341 | 2341 | if not self._inline: |
|
2342 | 2342 | res.append(self.datafile) |
|
2343 | 2343 | return res |
|
2344 | 2344 | |
|
2345 | 2345 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, |
|
2346 | 2346 | assumehaveparentrevisions=False, |
|
2347 | 2347 | deltamode=repository.CG_DELTAMODE_STD): |
|
2348 | 2348 | if nodesorder not in ('nodes', 'storage', 'linear', None): |
|
2349 | 2349 | raise error.ProgrammingError('unhandled value for nodesorder: %s' % |
|
2350 | 2350 | nodesorder) |
|
2351 | 2351 | |
|
2352 | 2352 | if nodesorder is None and not self._generaldelta: |
|
2353 | 2353 | nodesorder = 'storage' |
|
2354 | 2354 | |
|
2355 | 2355 | if (not self._storedeltachains and |
|
2356 | 2356 | deltamode != repository.CG_DELTAMODE_PREV): |
|
2357 | 2357 | deltamode = repository.CG_DELTAMODE_FULL |
|
2358 | 2358 | |
|
2359 | 2359 | return storageutil.emitrevisions( |
|
2360 | 2360 | self, nodes, nodesorder, revlogrevisiondelta, |
|
2361 | 2361 | deltaparentfn=self.deltaparent, |
|
2362 | 2362 | candeltafn=self.candelta, |
|
2363 | 2363 | rawsizefn=self.rawsize, |
|
2364 | 2364 | revdifffn=self.revdiff, |
|
2365 | 2365 | flagsfn=self.flags, |
|
2366 | 2366 | deltamode=deltamode, |
|
2367 | 2367 | revisiondata=revisiondata, |
|
2368 | 2368 | assumehaveparentrevisions=assumehaveparentrevisions) |
|
2369 | 2369 | |
|
2370 | 2370 | DELTAREUSEALWAYS = 'always' |
|
2371 | 2371 | DELTAREUSESAMEREVS = 'samerevs' |
|
2372 | 2372 | DELTAREUSENEVER = 'never' |
|
2373 | 2373 | |
|
2374 | 2374 | DELTAREUSEFULLADD = 'fulladd' |
|
2375 | 2375 | |
|
2376 | 2376 | DELTAREUSEALL = {'always', 'samerevs', 'never', 'fulladd'} |
|
2377 | 2377 | |
|
2378 | 2378 | def clone(self, tr, destrevlog, addrevisioncb=None, |
|
2379 | 2379 | deltareuse=DELTAREUSESAMEREVS, forcedeltabothparents=None): |
|
2380 | 2380 | """Copy this revlog to another, possibly with format changes. |
|
2381 | 2381 | |
|
2382 | 2382 | The destination revlog will contain the same revisions and nodes. |
|
2383 | 2383 | However, it may not be bit-for-bit identical due to e.g. delta encoding |
|
2384 | 2384 | differences. |
|
2385 | 2385 | |
|
2386 | 2386 | The ``deltareuse`` argument control how deltas from the existing revlog |
|
2387 | 2387 | are preserved in the destination revlog. The argument can have the |
|
2388 | 2388 | following values: |
|
2389 | 2389 | |
|
2390 | 2390 | DELTAREUSEALWAYS |
|
2391 | 2391 | Deltas will always be reused (if possible), even if the destination |
|
2392 | 2392 | revlog would not select the same revisions for the delta. This is the |
|
2393 | 2393 | fastest mode of operation. |
|
2394 | 2394 | DELTAREUSESAMEREVS |
|
2395 | 2395 | Deltas will be reused if the destination revlog would pick the same |
|
2396 | 2396 | revisions for the delta. This mode strikes a balance between speed |
|
2397 | 2397 | and optimization. |
|
2398 | 2398 | DELTAREUSENEVER |
|
2399 | 2399 | Deltas will never be reused. This is the slowest mode of execution. |
|
2400 | 2400 | This mode can be used to recompute deltas (e.g. if the diff/delta |
|
2401 | 2401 | algorithm changes). |
|
2402 | 2402 | |
|
2403 | 2403 | Delta computation can be slow, so the choice of delta reuse policy can |
|
2404 | 2404 | significantly affect run time. |
|
2405 | 2405 | |
|
2406 | 2406 | The default policy (``DELTAREUSESAMEREVS``) strikes a balance between |
|
2407 | 2407 | two extremes. Deltas will be reused if they are appropriate. But if the |
|
2408 | 2408 | delta could choose a better revision, it will do so. This means if you |
|
2409 | 2409 | are converting a non-generaldelta revlog to a generaldelta revlog, |
|
2410 | 2410 | deltas will be recomputed if the delta's parent isn't a parent of the |
|
2411 | 2411 | revision. |
|
2412 | 2412 | |
|
2413 | 2413 | In addition to the delta policy, the ``forcedeltabothparents`` |
|
2414 | 2414 | argument controls whether to force compute deltas against both parents |
|
2415 | 2415 | for merges. By default, the current default is used. |
|
2416 | 2416 | """ |
|
2417 | 2417 | if deltareuse not in self.DELTAREUSEALL: |
|
2418 | 2418 | raise ValueError(_('value for deltareuse invalid: %s') % deltareuse) |
|
2419 | 2419 | |
|
2420 | 2420 | if len(destrevlog): |
|
2421 | 2421 | raise ValueError(_('destination revlog is not empty')) |
|
2422 | 2422 | |
|
2423 | 2423 | if getattr(self, 'filteredrevs', None): |
|
2424 | 2424 | raise ValueError(_('source revlog has filtered revisions')) |
|
2425 | 2425 | if getattr(destrevlog, 'filteredrevs', None): |
|
2426 | 2426 | raise ValueError(_('destination revlog has filtered revisions')) |
|
2427 | 2427 | |
|
2428 | 2428 | # lazydelta and lazydeltabase controls whether to reuse a cached delta, |
|
2429 | 2429 | # if possible. |
|
2430 | 2430 | oldlazydelta = destrevlog._lazydelta |
|
2431 | 2431 | oldlazydeltabase = destrevlog._lazydeltabase |
|
2432 | 2432 | oldamd = destrevlog._deltabothparents |
|
2433 | 2433 | |
|
2434 | 2434 | try: |
|
2435 | 2435 | if deltareuse == self.DELTAREUSEALWAYS: |
|
2436 | 2436 | destrevlog._lazydeltabase = True |
|
2437 | 2437 | destrevlog._lazydelta = True |
|
2438 | 2438 | elif deltareuse == self.DELTAREUSESAMEREVS: |
|
2439 | 2439 | destrevlog._lazydeltabase = False |
|
2440 | 2440 | destrevlog._lazydelta = True |
|
2441 | 2441 | elif deltareuse == self.DELTAREUSENEVER: |
|
2442 | 2442 | destrevlog._lazydeltabase = False |
|
2443 | 2443 | destrevlog._lazydelta = False |
|
2444 | 2444 | |
|
2445 | 2445 | destrevlog._deltabothparents = forcedeltabothparents or oldamd |
|
2446 | 2446 | |
|
2447 | 2447 | deltacomputer = deltautil.deltacomputer(destrevlog) |
|
2448 | 2448 | index = self.index |
|
2449 | 2449 | for rev in self: |
|
2450 | 2450 | entry = index[rev] |
|
2451 | 2451 | |
|
2452 | 2452 | # Some classes override linkrev to take filtered revs into |
|
2453 | 2453 | # account. Use raw entry from index. |
|
2454 | 2454 | flags = entry[0] & 0xffff |
|
2455 | 2455 | linkrev = entry[4] |
|
2456 | 2456 | p1 = index[entry[5]][7] |
|
2457 | 2457 | p2 = index[entry[6]][7] |
|
2458 | 2458 | node = entry[7] |
|
2459 | 2459 | |
|
2460 | 2460 | # (Possibly) reuse the delta from the revlog if allowed and |
|
2461 | 2461 | # the revlog chunk is a delta. |
|
2462 | 2462 | cachedelta = None |
|
2463 | 2463 | rawtext = None |
|
2464 | 2464 | if destrevlog._lazydelta: |
|
2465 | 2465 | dp = self.deltaparent(rev) |
|
2466 | 2466 | if dp != nullrev: |
|
2467 | 2467 | cachedelta = (dp, bytes(self._chunk(rev))) |
|
2468 | 2468 | |
|
2469 | 2469 | if not cachedelta: |
|
2470 | 2470 | rawtext = self.revision(rev, raw=True) |
|
2471 | 2471 | |
|
2472 | 2472 | |
|
2473 | 2473 | if deltareuse == self.DELTAREUSEFULLADD: |
|
2474 | 2474 | destrevlog.addrevision(rawtext, tr, linkrev, p1, p2, |
|
2475 | 2475 | cachedelta=cachedelta, |
|
2476 | 2476 | node=node, flags=flags, |
|
2477 | 2477 | deltacomputer=deltacomputer) |
|
2478 | 2478 | else: |
|
2479 | 2479 | ifh = destrevlog.opener(destrevlog.indexfile, 'a+', |
|
2480 | 2480 | checkambig=False) |
|
2481 | 2481 | dfh = None |
|
2482 | 2482 | if not destrevlog._inline: |
|
2483 | 2483 | dfh = destrevlog.opener(destrevlog.datafile, 'a+') |
|
2484 | 2484 | try: |
|
2485 | 2485 | destrevlog._addrevision(node, rawtext, tr, linkrev, p1, |
|
2486 | 2486 | p2, flags, cachedelta, ifh, dfh, |
|
2487 | 2487 | deltacomputer=deltacomputer) |
|
2488 | 2488 | finally: |
|
2489 | 2489 | if dfh: |
|
2490 | 2490 | dfh.close() |
|
2491 | 2491 | ifh.close() |
|
2492 | 2492 | |
|
2493 | 2493 | if addrevisioncb: |
|
2494 | 2494 | addrevisioncb(self, rev, node) |
|
2495 | 2495 | finally: |
|
2496 | 2496 | destrevlog._lazydelta = oldlazydelta |
|
2497 | 2497 | destrevlog._lazydeltabase = oldlazydeltabase |
|
2498 | 2498 | destrevlog._deltabothparents = oldamd |
|
2499 | 2499 | |
|
2500 | 2500 | def censorrevision(self, tr, censornode, tombstone=b''): |
|
2501 | 2501 | if (self.version & 0xFFFF) == REVLOGV0: |
|
2502 | 2502 | raise error.RevlogError(_('cannot censor with version %d revlogs') % |
|
2503 | 2503 | self.version) |
|
2504 | 2504 | |
|
2505 | 2505 | censorrev = self.rev(censornode) |
|
2506 | 2506 | tombstone = storageutil.packmeta({b'censored': tombstone}, b'') |
|
2507 | 2507 | |
|
2508 | 2508 | if len(tombstone) > self.rawsize(censorrev): |
|
2509 | 2509 | raise error.Abort(_('censor tombstone must be no longer than ' |
|
2510 | 2510 | 'censored data')) |
|
2511 | 2511 | |
|
2512 | 2512 | # Rewriting the revlog in place is hard. Our strategy for censoring is |
|
2513 | 2513 | # to create a new revlog, copy all revisions to it, then replace the |
|
2514 | 2514 | # revlogs on transaction close. |
|
2515 | 2515 | |
|
2516 | 2516 | newindexfile = self.indexfile + b'.tmpcensored' |
|
2517 | 2517 | newdatafile = self.datafile + b'.tmpcensored' |
|
2518 | 2518 | |
|
2519 | 2519 | # This is a bit dangerous. We could easily have a mismatch of state. |
|
2520 | 2520 | newrl = revlog(self.opener, newindexfile, newdatafile, |
|
2521 | 2521 | censorable=True) |
|
2522 | 2522 | newrl.version = self.version |
|
2523 | 2523 | newrl._generaldelta = self._generaldelta |
|
2524 | 2524 | newrl._io = self._io |
|
2525 | 2525 | |
|
2526 | 2526 | for rev in self.revs(): |
|
2527 | 2527 | node = self.node(rev) |
|
2528 | 2528 | p1, p2 = self.parents(node) |
|
2529 | 2529 | |
|
2530 | 2530 | if rev == censorrev: |
|
2531 | 2531 | newrl.addrawrevision(tombstone, tr, self.linkrev(censorrev), |
|
2532 | 2532 | p1, p2, censornode, REVIDX_ISCENSORED) |
|
2533 | 2533 | |
|
2534 | 2534 | if newrl.deltaparent(rev) != nullrev: |
|
2535 | 2535 | raise error.Abort(_('censored revision stored as delta; ' |
|
2536 | 2536 | 'cannot censor'), |
|
2537 | 2537 | hint=_('censoring of revlogs is not ' |
|
2538 | 2538 | 'fully implemented; please report ' |
|
2539 | 2539 | 'this bug')) |
|
2540 | 2540 | continue |
|
2541 | 2541 | |
|
2542 | 2542 | if self.iscensored(rev): |
|
2543 | 2543 | if self.deltaparent(rev) != nullrev: |
|
2544 | 2544 | raise error.Abort(_('cannot censor due to censored ' |
|
2545 | 2545 | 'revision having delta stored')) |
|
2546 | 2546 | rawtext = self._chunk(rev) |
|
2547 | 2547 | else: |
|
2548 | 2548 | rawtext = self.revision(rev, raw=True) |
|
2549 | 2549 | |
|
2550 | 2550 | newrl.addrawrevision(rawtext, tr, self.linkrev(rev), p1, p2, node, |
|
2551 | 2551 | self.flags(rev)) |
|
2552 | 2552 | |
|
2553 | 2553 | tr.addbackup(self.indexfile, location='store') |
|
2554 | 2554 | if not self._inline: |
|
2555 | 2555 | tr.addbackup(self.datafile, location='store') |
|
2556 | 2556 | |
|
2557 | 2557 | self.opener.rename(newrl.indexfile, self.indexfile) |
|
2558 | 2558 | if not self._inline: |
|
2559 | 2559 | self.opener.rename(newrl.datafile, self.datafile) |
|
2560 | 2560 | |
|
2561 | 2561 | self.clearcaches() |
|
2562 | 2562 | self._loadindex() |
|
2563 | 2563 | |
|
2564 | 2564 | def verifyintegrity(self, state): |
|
2565 | 2565 | """Verifies the integrity of the revlog. |
|
2566 | 2566 | |
|
2567 | 2567 | Yields ``revlogproblem`` instances describing problems that are |
|
2568 | 2568 | found. |
|
2569 | 2569 | """ |
|
2570 | 2570 | dd, di = self.checksize() |
|
2571 | 2571 | if dd: |
|
2572 | 2572 | yield revlogproblem(error=_('data length off by %d bytes') % dd) |
|
2573 | 2573 | if di: |
|
2574 | 2574 | yield revlogproblem(error=_('index contains %d extra bytes') % di) |
|
2575 | 2575 | |
|
2576 | 2576 | version = self.version & 0xFFFF |
|
2577 | 2577 | |
|
2578 | 2578 | # The verifier tells us what version revlog we should be. |
|
2579 | 2579 | if version != state['expectedversion']: |
|
2580 | 2580 | yield revlogproblem( |
|
2581 | 2581 | warning=_("warning: '%s' uses revlog format %d; expected %d") % |
|
2582 | 2582 | (self.indexfile, version, state['expectedversion'])) |
|
2583 | 2583 | |
|
2584 | 2584 | state['skipread'] = set() |
|
2585 | 2585 | |
|
2586 | 2586 | for rev in self: |
|
2587 | 2587 | node = self.node(rev) |
|
2588 | 2588 | |
|
2589 | 2589 | # Verify contents. 4 cases to care about: |
|
2590 | 2590 | # |
|
2591 | 2591 | # common: the most common case |
|
2592 | 2592 | # rename: with a rename |
|
2593 | 2593 | # meta: file content starts with b'\1\n', the metadata |
|
2594 | 2594 | # header defined in filelog.py, but without a rename |
|
2595 | 2595 | # ext: content stored externally |
|
2596 | 2596 | # |
|
2597 | 2597 | # More formally, their differences are shown below: |
|
2598 | 2598 | # |
|
2599 | 2599 | # | common | rename | meta | ext |
|
2600 | 2600 | # ------------------------------------------------------- |
|
2601 | 2601 | # flags() | 0 | 0 | 0 | not 0 |
|
2602 | 2602 | # renamed() | False | True | False | ? |
|
2603 | 2603 | # rawtext[0:2]=='\1\n'| False | True | True | ? |
|
2604 | 2604 | # |
|
2605 | 2605 | # "rawtext" means the raw text stored in revlog data, which |
|
2606 | 2606 | # could be retrieved by "revision(rev, raw=True)". "text" |
|
2607 | 2607 | # mentioned below is "revision(rev, raw=False)". |
|
2608 | 2608 | # |
|
2609 | 2609 | # There are 3 different lengths stored physically: |
|
2610 | 2610 | # 1. L1: rawsize, stored in revlog index |
|
2611 | 2611 | # 2. L2: len(rawtext), stored in revlog data |
|
2612 | 2612 | # 3. L3: len(text), stored in revlog data if flags==0, or |
|
2613 | 2613 | # possibly somewhere else if flags!=0 |
|
2614 | 2614 | # |
|
2615 | 2615 | # L1 should be equal to L2. L3 could be different from them. |
|
2616 | 2616 | # "text" may or may not affect commit hash depending on flag |
|
2617 | 2617 | # processors (see revlog.addflagprocessor). |
|
2618 | 2618 | # |
|
2619 | 2619 | # | common | rename | meta | ext |
|
2620 | 2620 | # ------------------------------------------------- |
|
2621 | 2621 | # rawsize() | L1 | L1 | L1 | L1 |
|
2622 | 2622 | # size() | L1 | L2-LM | L1(*) | L1 (?) |
|
2623 | 2623 | # len(rawtext) | L2 | L2 | L2 | L2 |
|
2624 | 2624 | # len(text) | L2 | L2 | L2 | L3 |
|
2625 | 2625 | # len(read()) | L2 | L2-LM | L2-LM | L3 (?) |
|
2626 | 2626 | # |
|
2627 | 2627 | # LM: length of metadata, depending on rawtext |
|
2628 | 2628 | # (*): not ideal, see comment in filelog.size |
|
2629 | 2629 | # (?): could be "- len(meta)" if the resolved content has |
|
2630 | 2630 | # rename metadata |
|
2631 | 2631 | # |
|
2632 | 2632 | # Checks needed to be done: |
|
2633 | 2633 | # 1. length check: L1 == L2, in all cases. |
|
2634 | 2634 | # 2. hash check: depending on flag processor, we may need to |
|
2635 | 2635 | # use either "text" (external), or "rawtext" (in revlog). |
|
2636 | 2636 | |
|
2637 | 2637 | try: |
|
2638 | 2638 | skipflags = state.get('skipflags', 0) |
|
2639 | 2639 | if skipflags: |
|
2640 | 2640 | skipflags &= self.flags(rev) |
|
2641 | 2641 | |
|
2642 | 2642 | if skipflags: |
|
2643 | 2643 | state['skipread'].add(node) |
|
2644 | 2644 | else: |
|
2645 | 2645 | # Side-effect: read content and verify hash. |
|
2646 | 2646 | self.revision(node) |
|
2647 | 2647 | |
|
2648 | 2648 | l1 = self.rawsize(rev) |
|
2649 | 2649 | l2 = len(self.revision(node, raw=True)) |
|
2650 | 2650 | |
|
2651 | 2651 | if l1 != l2: |
|
2652 | 2652 | yield revlogproblem( |
|
2653 | 2653 | error=_('unpacked size is %d, %d expected') % (l2, l1), |
|
2654 | 2654 | node=node) |
|
2655 | 2655 | |
|
2656 | 2656 | except error.CensoredNodeError: |
|
2657 | 2657 | if state['erroroncensored']: |
|
2658 | 2658 | yield revlogproblem(error=_('censored file data'), |
|
2659 | 2659 | node=node) |
|
2660 | 2660 | state['skipread'].add(node) |
|
2661 | 2661 | except Exception as e: |
|
2662 | 2662 | yield revlogproblem( |
|
2663 | 2663 | error=_('unpacking %s: %s') % (short(node), |
|
2664 | 2664 | stringutil.forcebytestr(e)), |
|
2665 | 2665 | node=node) |
|
2666 | 2666 | state['skipread'].add(node) |
|
2667 | 2667 | |
|
2668 | 2668 | def storageinfo(self, exclusivefiles=False, sharedfiles=False, |
|
2669 | 2669 | revisionscount=False, trackedsize=False, |
|
2670 | 2670 | storedsize=False): |
|
2671 | 2671 | d = {} |
|
2672 | 2672 | |
|
2673 | 2673 | if exclusivefiles: |
|
2674 | 2674 | d['exclusivefiles'] = [(self.opener, self.indexfile)] |
|
2675 | 2675 | if not self._inline: |
|
2676 | 2676 | d['exclusivefiles'].append((self.opener, self.datafile)) |
|
2677 | 2677 | |
|
2678 | 2678 | if sharedfiles: |
|
2679 | 2679 | d['sharedfiles'] = [] |
|
2680 | 2680 | |
|
2681 | 2681 | if revisionscount: |
|
2682 | 2682 | d['revisionscount'] = len(self) |
|
2683 | 2683 | |
|
2684 | 2684 | if trackedsize: |
|
2685 | 2685 | d['trackedsize'] = sum(map(self.rawsize, iter(self))) |
|
2686 | 2686 | |
|
2687 | 2687 | if storedsize: |
|
2688 | 2688 | d['storedsize'] = sum(self.opener.stat(path).st_size |
|
2689 | 2689 | for path in self.files()) |
|
2690 | 2690 | |
|
2691 | 2691 | return d |
General Comments 0
You need to be logged in to leave comments.
Login now