Show More
@@ -1,459 +1,459 b'' | |||||
1 | # remotefilelog.py - filelog implementation where filelog history is stored |
|
1 | # remotefilelog.py - filelog implementation where filelog history is stored | |
2 | # remotely |
|
2 | # remotely | |
3 | # |
|
3 | # | |
4 | # Copyright 2013 Facebook, Inc. |
|
4 | # Copyright 2013 Facebook, Inc. | |
5 | # |
|
5 | # | |
6 | # This software may be used and distributed according to the terms of the |
|
6 | # This software may be used and distributed according to the terms of the | |
7 | # GNU General Public License version 2 or any later version. |
|
7 | # GNU General Public License version 2 or any later version. | |
8 | from __future__ import absolute_import |
|
8 | from __future__ import absolute_import | |
9 |
|
9 | |||
10 | import collections |
|
10 | import collections | |
11 | import os |
|
11 | import os | |
12 |
|
12 | |||
13 | from mercurial.node import ( |
|
13 | from mercurial.node import ( | |
14 | bin, |
|
14 | bin, | |
15 | nullid, |
|
15 | nullid, | |
16 | wdirfilenodeids, |
|
16 | wdirfilenodeids, | |
17 | wdirid, |
|
17 | wdirid, | |
18 | ) |
|
18 | ) | |
19 | from mercurial.i18n import _ |
|
19 | from mercurial.i18n import _ | |
20 | from mercurial import ( |
|
20 | from mercurial import ( | |
21 | ancestor, |
|
21 | ancestor, | |
22 | error, |
|
22 | error, | |
23 | mdiff, |
|
23 | mdiff, | |
24 | revlog, |
|
24 | revlog, | |
25 | util, |
|
25 | util, | |
26 | ) |
|
26 | ) | |
27 | from mercurial.utils import storageutil |
|
27 | from mercurial.utils import storageutil | |
28 | from mercurial.revlogutils import flagutil |
|
28 | from mercurial.revlogutils import flagutil | |
29 |
|
29 | |||
30 | from . import ( |
|
30 | from . import ( | |
31 | constants, |
|
31 | constants, | |
32 | fileserverclient, |
|
32 | fileserverclient, | |
33 | shallowutil, |
|
33 | shallowutil, | |
34 | ) |
|
34 | ) | |
35 |
|
35 | |||
36 | class remotefilelognodemap(object): |
|
36 | class remotefilelognodemap(object): | |
37 | def __init__(self, filename, store): |
|
37 | def __init__(self, filename, store): | |
38 | self._filename = filename |
|
38 | self._filename = filename | |
39 | self._store = store |
|
39 | self._store = store | |
40 |
|
40 | |||
41 | def __contains__(self, node): |
|
41 | def __contains__(self, node): | |
42 | missing = self._store.getmissing([(self._filename, node)]) |
|
42 | missing = self._store.getmissing([(self._filename, node)]) | |
43 | return not bool(missing) |
|
43 | return not bool(missing) | |
44 |
|
44 | |||
45 | def __get__(self, node): |
|
45 | def __get__(self, node): | |
46 | if node not in self: |
|
46 | if node not in self: | |
47 | raise KeyError(node) |
|
47 | raise KeyError(node) | |
48 | return node |
|
48 | return node | |
49 |
|
49 | |||
50 | class remotefilelog(flagutil.flagprocessorsmixin): |
|
50 | class remotefilelog(object): | |
51 |
|
51 | |||
52 | _generaldelta = True |
|
52 | _generaldelta = True | |
53 | _flagserrorclass = error.RevlogError |
|
53 | _flagserrorclass = error.RevlogError | |
54 |
|
54 | |||
55 | def __init__(self, opener, path, repo): |
|
55 | def __init__(self, opener, path, repo): | |
56 | self.opener = opener |
|
56 | self.opener = opener | |
57 | self.filename = path |
|
57 | self.filename = path | |
58 | self.repo = repo |
|
58 | self.repo = repo | |
59 | self.nodemap = remotefilelognodemap(self.filename, repo.contentstore) |
|
59 | self.nodemap = remotefilelognodemap(self.filename, repo.contentstore) | |
60 |
|
60 | |||
61 | self.version = 1 |
|
61 | self.version = 1 | |
62 |
|
62 | |||
63 | self._flagprocessors = dict(flagutil.flagprocessors) |
|
63 | self._flagprocessors = dict(flagutil.flagprocessors) | |
64 |
|
64 | |||
65 | def read(self, node): |
|
65 | def read(self, node): | |
66 | """returns the file contents at this node""" |
|
66 | """returns the file contents at this node""" | |
67 | t = self.revision(node) |
|
67 | t = self.revision(node) | |
68 | if not t.startswith('\1\n'): |
|
68 | if not t.startswith('\1\n'): | |
69 | return t |
|
69 | return t | |
70 | s = t.index('\1\n', 2) |
|
70 | s = t.index('\1\n', 2) | |
71 | return t[s + 2:] |
|
71 | return t[s + 2:] | |
72 |
|
72 | |||
73 | def add(self, text, meta, transaction, linknode, p1=None, p2=None): |
|
73 | def add(self, text, meta, transaction, linknode, p1=None, p2=None): | |
74 | # hash with the metadata, like in vanilla filelogs |
|
74 | # hash with the metadata, like in vanilla filelogs | |
75 | hashtext = shallowutil.createrevlogtext(text, meta.get('copy'), |
|
75 | hashtext = shallowutil.createrevlogtext(text, meta.get('copy'), | |
76 | meta.get('copyrev')) |
|
76 | meta.get('copyrev')) | |
77 | node = storageutil.hashrevisionsha1(hashtext, p1, p2) |
|
77 | node = storageutil.hashrevisionsha1(hashtext, p1, p2) | |
78 | return self.addrevision(hashtext, transaction, linknode, p1, p2, |
|
78 | return self.addrevision(hashtext, transaction, linknode, p1, p2, | |
79 | node=node) |
|
79 | node=node) | |
80 |
|
80 | |||
81 | def _createfileblob(self, text, meta, flags, p1, p2, node, linknode): |
|
81 | def _createfileblob(self, text, meta, flags, p1, p2, node, linknode): | |
82 | # text passed to "_createfileblob" does not include filelog metadata |
|
82 | # text passed to "_createfileblob" does not include filelog metadata | |
83 | header = shallowutil.buildfileblobheader(len(text), flags) |
|
83 | header = shallowutil.buildfileblobheader(len(text), flags) | |
84 | data = "%s\0%s" % (header, text) |
|
84 | data = "%s\0%s" % (header, text) | |
85 |
|
85 | |||
86 | realp1 = p1 |
|
86 | realp1 = p1 | |
87 | copyfrom = "" |
|
87 | copyfrom = "" | |
88 | if meta and 'copy' in meta: |
|
88 | if meta and 'copy' in meta: | |
89 | copyfrom = meta['copy'] |
|
89 | copyfrom = meta['copy'] | |
90 | realp1 = bin(meta['copyrev']) |
|
90 | realp1 = bin(meta['copyrev']) | |
91 |
|
91 | |||
92 | data += "%s%s%s%s%s\0" % (node, realp1, p2, linknode, copyfrom) |
|
92 | data += "%s%s%s%s%s\0" % (node, realp1, p2, linknode, copyfrom) | |
93 |
|
93 | |||
94 | visited = set() |
|
94 | visited = set() | |
95 |
|
95 | |||
96 | pancestors = {} |
|
96 | pancestors = {} | |
97 | queue = [] |
|
97 | queue = [] | |
98 | if realp1 != nullid: |
|
98 | if realp1 != nullid: | |
99 | p1flog = self |
|
99 | p1flog = self | |
100 | if copyfrom: |
|
100 | if copyfrom: | |
101 | p1flog = remotefilelog(self.opener, copyfrom, self.repo) |
|
101 | p1flog = remotefilelog(self.opener, copyfrom, self.repo) | |
102 |
|
102 | |||
103 | pancestors.update(p1flog.ancestormap(realp1)) |
|
103 | pancestors.update(p1flog.ancestormap(realp1)) | |
104 | queue.append(realp1) |
|
104 | queue.append(realp1) | |
105 | visited.add(realp1) |
|
105 | visited.add(realp1) | |
106 | if p2 != nullid: |
|
106 | if p2 != nullid: | |
107 | pancestors.update(self.ancestormap(p2)) |
|
107 | pancestors.update(self.ancestormap(p2)) | |
108 | queue.append(p2) |
|
108 | queue.append(p2) | |
109 | visited.add(p2) |
|
109 | visited.add(p2) | |
110 |
|
110 | |||
111 | ancestortext = "" |
|
111 | ancestortext = "" | |
112 |
|
112 | |||
113 | # add the ancestors in topological order |
|
113 | # add the ancestors in topological order | |
114 | while queue: |
|
114 | while queue: | |
115 | c = queue.pop(0) |
|
115 | c = queue.pop(0) | |
116 | pa1, pa2, ancestorlinknode, pacopyfrom = pancestors[c] |
|
116 | pa1, pa2, ancestorlinknode, pacopyfrom = pancestors[c] | |
117 |
|
117 | |||
118 | pacopyfrom = pacopyfrom or '' |
|
118 | pacopyfrom = pacopyfrom or '' | |
119 | ancestortext += "%s%s%s%s%s\0" % ( |
|
119 | ancestortext += "%s%s%s%s%s\0" % ( | |
120 | c, pa1, pa2, ancestorlinknode, pacopyfrom) |
|
120 | c, pa1, pa2, ancestorlinknode, pacopyfrom) | |
121 |
|
121 | |||
122 | if pa1 != nullid and pa1 not in visited: |
|
122 | if pa1 != nullid and pa1 not in visited: | |
123 | queue.append(pa1) |
|
123 | queue.append(pa1) | |
124 | visited.add(pa1) |
|
124 | visited.add(pa1) | |
125 | if pa2 != nullid and pa2 not in visited: |
|
125 | if pa2 != nullid and pa2 not in visited: | |
126 | queue.append(pa2) |
|
126 | queue.append(pa2) | |
127 | visited.add(pa2) |
|
127 | visited.add(pa2) | |
128 |
|
128 | |||
129 | data += ancestortext |
|
129 | data += ancestortext | |
130 |
|
130 | |||
131 | return data |
|
131 | return data | |
132 |
|
132 | |||
133 | def addrevision(self, text, transaction, linknode, p1, p2, cachedelta=None, |
|
133 | def addrevision(self, text, transaction, linknode, p1, p2, cachedelta=None, | |
134 | node=None, flags=revlog.REVIDX_DEFAULT_FLAGS, |
|
134 | node=None, flags=revlog.REVIDX_DEFAULT_FLAGS, | |
135 | sidedata=None): |
|
135 | sidedata=None): | |
136 | # text passed to "addrevision" includes hg filelog metadata header |
|
136 | # text passed to "addrevision" includes hg filelog metadata header | |
137 | if node is None: |
|
137 | if node is None: | |
138 | node = storageutil.hashrevisionsha1(text, p1, p2) |
|
138 | node = storageutil.hashrevisionsha1(text, p1, p2) | |
139 | if sidedata is None: |
|
139 | if sidedata is None: | |
140 | sidedata = {} |
|
140 | sidedata = {} | |
141 |
|
141 | |||
142 | meta, metaoffset = storageutil.parsemeta(text) |
|
142 | meta, metaoffset = storageutil.parsemeta(text) | |
143 | rawtext, validatehash = flagutil.processflagswrite(self, text, flags, |
|
143 | rawtext, validatehash = flagutil.processflagswrite(self, text, flags, | |
144 | sidedata=sidedata) |
|
144 | sidedata=sidedata) | |
145 | return self.addrawrevision(rawtext, transaction, linknode, p1, p2, |
|
145 | return self.addrawrevision(rawtext, transaction, linknode, p1, p2, | |
146 | node, flags, cachedelta, |
|
146 | node, flags, cachedelta, | |
147 | _metatuple=(meta, metaoffset)) |
|
147 | _metatuple=(meta, metaoffset)) | |
148 |
|
148 | |||
149 | def addrawrevision(self, rawtext, transaction, linknode, p1, p2, node, |
|
149 | def addrawrevision(self, rawtext, transaction, linknode, p1, p2, node, | |
150 | flags, cachedelta=None, _metatuple=None): |
|
150 | flags, cachedelta=None, _metatuple=None): | |
151 | if _metatuple: |
|
151 | if _metatuple: | |
152 | # _metatuple: used by "addrevision" internally by remotefilelog |
|
152 | # _metatuple: used by "addrevision" internally by remotefilelog | |
153 | # meta was parsed confidently |
|
153 | # meta was parsed confidently | |
154 | meta, metaoffset = _metatuple |
|
154 | meta, metaoffset = _metatuple | |
155 | else: |
|
155 | else: | |
156 | # not from self.addrevision, but something else (repo._filecommit) |
|
156 | # not from self.addrevision, but something else (repo._filecommit) | |
157 | # calls addrawrevision directly. remotefilelog needs to get and |
|
157 | # calls addrawrevision directly. remotefilelog needs to get and | |
158 | # strip filelog metadata. |
|
158 | # strip filelog metadata. | |
159 | # we don't have confidence about whether rawtext contains filelog |
|
159 | # we don't have confidence about whether rawtext contains filelog | |
160 | # metadata or not (flag processor could replace it), so we just |
|
160 | # metadata or not (flag processor could replace it), so we just | |
161 | # parse it as best-effort. |
|
161 | # parse it as best-effort. | |
162 | # in LFS (flags != 0)'s case, the best way is to call LFS code to |
|
162 | # in LFS (flags != 0)'s case, the best way is to call LFS code to | |
163 | # get the meta information, instead of storageutil.parsemeta. |
|
163 | # get the meta information, instead of storageutil.parsemeta. | |
164 | meta, metaoffset = storageutil.parsemeta(rawtext) |
|
164 | meta, metaoffset = storageutil.parsemeta(rawtext) | |
165 | if flags != 0: |
|
165 | if flags != 0: | |
166 | # when flags != 0, be conservative and do not mangle rawtext, since |
|
166 | # when flags != 0, be conservative and do not mangle rawtext, since | |
167 | # a read flag processor expects the text not being mangled at all. |
|
167 | # a read flag processor expects the text not being mangled at all. | |
168 | metaoffset = 0 |
|
168 | metaoffset = 0 | |
169 | if metaoffset: |
|
169 | if metaoffset: | |
170 | # remotefilelog fileblob stores copy metadata in its ancestortext, |
|
170 | # remotefilelog fileblob stores copy metadata in its ancestortext, | |
171 | # not its main blob. so we need to remove filelog metadata |
|
171 | # not its main blob. so we need to remove filelog metadata | |
172 | # (containing copy information) from text. |
|
172 | # (containing copy information) from text. | |
173 | blobtext = rawtext[metaoffset:] |
|
173 | blobtext = rawtext[metaoffset:] | |
174 | else: |
|
174 | else: | |
175 | blobtext = rawtext |
|
175 | blobtext = rawtext | |
176 | data = self._createfileblob(blobtext, meta, flags, p1, p2, node, |
|
176 | data = self._createfileblob(blobtext, meta, flags, p1, p2, node, | |
177 | linknode) |
|
177 | linknode) | |
178 | self.repo.contentstore.addremotefilelognode(self.filename, node, data) |
|
178 | self.repo.contentstore.addremotefilelognode(self.filename, node, data) | |
179 |
|
179 | |||
180 | return node |
|
180 | return node | |
181 |
|
181 | |||
182 | def renamed(self, node): |
|
182 | def renamed(self, node): | |
183 | ancestors = self.repo.metadatastore.getancestors(self.filename, node) |
|
183 | ancestors = self.repo.metadatastore.getancestors(self.filename, node) | |
184 | p1, p2, linknode, copyfrom = ancestors[node] |
|
184 | p1, p2, linknode, copyfrom = ancestors[node] | |
185 | if copyfrom: |
|
185 | if copyfrom: | |
186 | return (copyfrom, p1) |
|
186 | return (copyfrom, p1) | |
187 |
|
187 | |||
188 | return False |
|
188 | return False | |
189 |
|
189 | |||
190 | def size(self, node): |
|
190 | def size(self, node): | |
191 | """return the size of a given revision""" |
|
191 | """return the size of a given revision""" | |
192 | return len(self.read(node)) |
|
192 | return len(self.read(node)) | |
193 |
|
193 | |||
194 | rawsize = size |
|
194 | rawsize = size | |
195 |
|
195 | |||
196 | def cmp(self, node, text): |
|
196 | def cmp(self, node, text): | |
197 | """compare text with a given file revision |
|
197 | """compare text with a given file revision | |
198 |
|
198 | |||
199 | returns True if text is different than what is stored. |
|
199 | returns True if text is different than what is stored. | |
200 | """ |
|
200 | """ | |
201 |
|
201 | |||
202 | if node == nullid: |
|
202 | if node == nullid: | |
203 | return True |
|
203 | return True | |
204 |
|
204 | |||
205 | nodetext = self.read(node) |
|
205 | nodetext = self.read(node) | |
206 | return nodetext != text |
|
206 | return nodetext != text | |
207 |
|
207 | |||
208 | def __nonzero__(self): |
|
208 | def __nonzero__(self): | |
209 | return True |
|
209 | return True | |
210 |
|
210 | |||
211 | __bool__ = __nonzero__ |
|
211 | __bool__ = __nonzero__ | |
212 |
|
212 | |||
213 | def __len__(self): |
|
213 | def __len__(self): | |
214 | if self.filename == '.hgtags': |
|
214 | if self.filename == '.hgtags': | |
215 | # The length of .hgtags is used to fast path tag checking. |
|
215 | # The length of .hgtags is used to fast path tag checking. | |
216 | # remotefilelog doesn't support .hgtags since the entire .hgtags |
|
216 | # remotefilelog doesn't support .hgtags since the entire .hgtags | |
217 | # history is needed. Use the excludepattern setting to make |
|
217 | # history is needed. Use the excludepattern setting to make | |
218 | # .hgtags a normal filelog. |
|
218 | # .hgtags a normal filelog. | |
219 | return 0 |
|
219 | return 0 | |
220 |
|
220 | |||
221 | raise RuntimeError("len not supported") |
|
221 | raise RuntimeError("len not supported") | |
222 |
|
222 | |||
223 | def empty(self): |
|
223 | def empty(self): | |
224 | return False |
|
224 | return False | |
225 |
|
225 | |||
226 | def flags(self, node): |
|
226 | def flags(self, node): | |
227 | if isinstance(node, int): |
|
227 | if isinstance(node, int): | |
228 | raise error.ProgrammingError( |
|
228 | raise error.ProgrammingError( | |
229 | 'remotefilelog does not accept integer rev for flags') |
|
229 | 'remotefilelog does not accept integer rev for flags') | |
230 | store = self.repo.contentstore |
|
230 | store = self.repo.contentstore | |
231 | return store.getmeta(self.filename, node).get(constants.METAKEYFLAG, 0) |
|
231 | return store.getmeta(self.filename, node).get(constants.METAKEYFLAG, 0) | |
232 |
|
232 | |||
233 | def parents(self, node): |
|
233 | def parents(self, node): | |
234 | if node == nullid: |
|
234 | if node == nullid: | |
235 | return nullid, nullid |
|
235 | return nullid, nullid | |
236 |
|
236 | |||
237 | ancestormap = self.repo.metadatastore.getancestors(self.filename, node) |
|
237 | ancestormap = self.repo.metadatastore.getancestors(self.filename, node) | |
238 | p1, p2, linknode, copyfrom = ancestormap[node] |
|
238 | p1, p2, linknode, copyfrom = ancestormap[node] | |
239 | if copyfrom: |
|
239 | if copyfrom: | |
240 | p1 = nullid |
|
240 | p1 = nullid | |
241 |
|
241 | |||
242 | return p1, p2 |
|
242 | return p1, p2 | |
243 |
|
243 | |||
244 | def parentrevs(self, rev): |
|
244 | def parentrevs(self, rev): | |
245 | # TODO(augie): this is a node and should be a rev, but for now |
|
245 | # TODO(augie): this is a node and should be a rev, but for now | |
246 | # nothing in core seems to actually break. |
|
246 | # nothing in core seems to actually break. | |
247 | return self.parents(rev) |
|
247 | return self.parents(rev) | |
248 |
|
248 | |||
249 | def linknode(self, node): |
|
249 | def linknode(self, node): | |
250 | ancestormap = self.repo.metadatastore.getancestors(self.filename, node) |
|
250 | ancestormap = self.repo.metadatastore.getancestors(self.filename, node) | |
251 | p1, p2, linknode, copyfrom = ancestormap[node] |
|
251 | p1, p2, linknode, copyfrom = ancestormap[node] | |
252 | return linknode |
|
252 | return linknode | |
253 |
|
253 | |||
254 | def linkrev(self, node): |
|
254 | def linkrev(self, node): | |
255 | return self.repo.unfiltered().changelog.rev(self.linknode(node)) |
|
255 | return self.repo.unfiltered().changelog.rev(self.linknode(node)) | |
256 |
|
256 | |||
257 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, |
|
257 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, | |
258 | assumehaveparentrevisions=False, deltaprevious=False, |
|
258 | assumehaveparentrevisions=False, deltaprevious=False, | |
259 | deltamode=None): |
|
259 | deltamode=None): | |
260 | # we don't use any of these parameters here |
|
260 | # we don't use any of these parameters here | |
261 | del nodesorder, revisiondata, assumehaveparentrevisions, deltaprevious |
|
261 | del nodesorder, revisiondata, assumehaveparentrevisions, deltaprevious | |
262 | del deltamode |
|
262 | del deltamode | |
263 | prevnode = None |
|
263 | prevnode = None | |
264 | for node in nodes: |
|
264 | for node in nodes: | |
265 | p1, p2 = self.parents(node) |
|
265 | p1, p2 = self.parents(node) | |
266 | if prevnode is None: |
|
266 | if prevnode is None: | |
267 | basenode = prevnode = p1 |
|
267 | basenode = prevnode = p1 | |
268 | if basenode == node: |
|
268 | if basenode == node: | |
269 | basenode = nullid |
|
269 | basenode = nullid | |
270 | if basenode != nullid: |
|
270 | if basenode != nullid: | |
271 | revision = None |
|
271 | revision = None | |
272 | delta = self.revdiff(basenode, node) |
|
272 | delta = self.revdiff(basenode, node) | |
273 | else: |
|
273 | else: | |
274 | revision = self.rawdata(node) |
|
274 | revision = self.rawdata(node) | |
275 | delta = None |
|
275 | delta = None | |
276 | yield revlog.revlogrevisiondelta( |
|
276 | yield revlog.revlogrevisiondelta( | |
277 | node=node, |
|
277 | node=node, | |
278 | p1node=p1, |
|
278 | p1node=p1, | |
279 | p2node=p2, |
|
279 | p2node=p2, | |
280 | linknode=self.linknode(node), |
|
280 | linknode=self.linknode(node), | |
281 | basenode=basenode, |
|
281 | basenode=basenode, | |
282 | flags=self.flags(node), |
|
282 | flags=self.flags(node), | |
283 | baserevisionsize=None, |
|
283 | baserevisionsize=None, | |
284 | revision=revision, |
|
284 | revision=revision, | |
285 | delta=delta, |
|
285 | delta=delta, | |
286 | ) |
|
286 | ) | |
287 |
|
287 | |||
288 | def revdiff(self, node1, node2): |
|
288 | def revdiff(self, node1, node2): | |
289 | return mdiff.textdiff(self.rawdata(node1), |
|
289 | return mdiff.textdiff(self.rawdata(node1), | |
290 | self.rawdata(node2)) |
|
290 | self.rawdata(node2)) | |
291 |
|
291 | |||
292 | def lookup(self, node): |
|
292 | def lookup(self, node): | |
293 | if len(node) == 40: |
|
293 | if len(node) == 40: | |
294 | node = bin(node) |
|
294 | node = bin(node) | |
295 | if len(node) != 20: |
|
295 | if len(node) != 20: | |
296 | raise error.LookupError(node, self.filename, |
|
296 | raise error.LookupError(node, self.filename, | |
297 | _('invalid lookup input')) |
|
297 | _('invalid lookup input')) | |
298 |
|
298 | |||
299 | return node |
|
299 | return node | |
300 |
|
300 | |||
301 | def rev(self, node): |
|
301 | def rev(self, node): | |
302 | # This is a hack to make TortoiseHG work. |
|
302 | # This is a hack to make TortoiseHG work. | |
303 | return node |
|
303 | return node | |
304 |
|
304 | |||
305 | def node(self, rev): |
|
305 | def node(self, rev): | |
306 | # This is a hack. |
|
306 | # This is a hack. | |
307 | if isinstance(rev, int): |
|
307 | if isinstance(rev, int): | |
308 | raise error.ProgrammingError( |
|
308 | raise error.ProgrammingError( | |
309 | 'remotefilelog does not convert integer rev to node') |
|
309 | 'remotefilelog does not convert integer rev to node') | |
310 | return rev |
|
310 | return rev | |
311 |
|
311 | |||
312 | def _processflags(self, text, flags, operation, raw=False): |
|
312 | def _processflags(self, text, flags, operation, raw=False): | |
313 | """deprecated entry point to access flag processors""" |
|
313 | """deprecated entry point to access flag processors""" | |
314 | msg = ('_processflag(...) use the specialized variant') |
|
314 | msg = ('_processflag(...) use the specialized variant') | |
315 | util.nouideprecwarn(msg, '5.2', stacklevel=2) |
|
315 | util.nouideprecwarn(msg, '5.2', stacklevel=2) | |
316 | if raw: |
|
316 | if raw: | |
317 | return text, flagutil.processflagsraw(self, text, flags) |
|
317 | return text, flagutil.processflagsraw(self, text, flags) | |
318 | elif operation == 'read': |
|
318 | elif operation == 'read': | |
319 | return flagutil.processflagsread(self, text, flags) |
|
319 | return flagutil.processflagsread(self, text, flags) | |
320 | else: # write operation |
|
320 | else: # write operation | |
321 | return flagutil.processflagswrite(self, text, flags) |
|
321 | return flagutil.processflagswrite(self, text, flags) | |
322 |
|
322 | |||
323 | def revision(self, node, raw=False): |
|
323 | def revision(self, node, raw=False): | |
324 | """returns the revlog contents at this node. |
|
324 | """returns the revlog contents at this node. | |
325 | this includes the meta data traditionally included in file revlogs. |
|
325 | this includes the meta data traditionally included in file revlogs. | |
326 | this is generally only used for bundling and communicating with vanilla |
|
326 | this is generally only used for bundling and communicating with vanilla | |
327 | hg clients. |
|
327 | hg clients. | |
328 | """ |
|
328 | """ | |
329 | if node == nullid: |
|
329 | if node == nullid: | |
330 | return "" |
|
330 | return "" | |
331 | if len(node) != 20: |
|
331 | if len(node) != 20: | |
332 | raise error.LookupError(node, self.filename, |
|
332 | raise error.LookupError(node, self.filename, | |
333 | _('invalid revision input')) |
|
333 | _('invalid revision input')) | |
334 | if node == wdirid or node in wdirfilenodeids: |
|
334 | if node == wdirid or node in wdirfilenodeids: | |
335 | raise error.WdirUnsupported |
|
335 | raise error.WdirUnsupported | |
336 |
|
336 | |||
337 | store = self.repo.contentstore |
|
337 | store = self.repo.contentstore | |
338 | rawtext = store.get(self.filename, node) |
|
338 | rawtext = store.get(self.filename, node) | |
339 | if raw: |
|
339 | if raw: | |
340 | return rawtext |
|
340 | return rawtext | |
341 | flags = store.getmeta(self.filename, node).get(constants.METAKEYFLAG, 0) |
|
341 | flags = store.getmeta(self.filename, node).get(constants.METAKEYFLAG, 0) | |
342 | if flags == 0: |
|
342 | if flags == 0: | |
343 | return rawtext |
|
343 | return rawtext | |
344 | return flagutil.processflagsread(self, rawtext, flags)[0] |
|
344 | return flagutil.processflagsread(self, rawtext, flags)[0] | |
345 |
|
345 | |||
346 | def rawdata(self, node): |
|
346 | def rawdata(self, node): | |
347 | return self.revision(node, raw=False) |
|
347 | return self.revision(node, raw=False) | |
348 |
|
348 | |||
349 | def _read(self, id): |
|
349 | def _read(self, id): | |
350 | """reads the raw file blob from disk, cache, or server""" |
|
350 | """reads the raw file blob from disk, cache, or server""" | |
351 | fileservice = self.repo.fileservice |
|
351 | fileservice = self.repo.fileservice | |
352 | localcache = fileservice.localcache |
|
352 | localcache = fileservice.localcache | |
353 | cachekey = fileserverclient.getcachekey(self.repo.name, self.filename, |
|
353 | cachekey = fileserverclient.getcachekey(self.repo.name, self.filename, | |
354 | id) |
|
354 | id) | |
355 | try: |
|
355 | try: | |
356 | return localcache.read(cachekey) |
|
356 | return localcache.read(cachekey) | |
357 | except KeyError: |
|
357 | except KeyError: | |
358 | pass |
|
358 | pass | |
359 |
|
359 | |||
360 | localkey = fileserverclient.getlocalkey(self.filename, id) |
|
360 | localkey = fileserverclient.getlocalkey(self.filename, id) | |
361 | localpath = os.path.join(self.localpath, localkey) |
|
361 | localpath = os.path.join(self.localpath, localkey) | |
362 | try: |
|
362 | try: | |
363 | return shallowutil.readfile(localpath) |
|
363 | return shallowutil.readfile(localpath) | |
364 | except IOError: |
|
364 | except IOError: | |
365 | pass |
|
365 | pass | |
366 |
|
366 | |||
367 | fileservice.prefetch([(self.filename, id)]) |
|
367 | fileservice.prefetch([(self.filename, id)]) | |
368 | try: |
|
368 | try: | |
369 | return localcache.read(cachekey) |
|
369 | return localcache.read(cachekey) | |
370 | except KeyError: |
|
370 | except KeyError: | |
371 | pass |
|
371 | pass | |
372 |
|
372 | |||
373 | raise error.LookupError(id, self.filename, _('no node')) |
|
373 | raise error.LookupError(id, self.filename, _('no node')) | |
374 |
|
374 | |||
375 | def ancestormap(self, node): |
|
375 | def ancestormap(self, node): | |
376 | return self.repo.metadatastore.getancestors(self.filename, node) |
|
376 | return self.repo.metadatastore.getancestors(self.filename, node) | |
377 |
|
377 | |||
378 | def ancestor(self, a, b): |
|
378 | def ancestor(self, a, b): | |
379 | if a == nullid or b == nullid: |
|
379 | if a == nullid or b == nullid: | |
380 | return nullid |
|
380 | return nullid | |
381 |
|
381 | |||
382 | revmap, parentfunc = self._buildrevgraph(a, b) |
|
382 | revmap, parentfunc = self._buildrevgraph(a, b) | |
383 | nodemap = dict(((v, k) for (k, v) in revmap.iteritems())) |
|
383 | nodemap = dict(((v, k) for (k, v) in revmap.iteritems())) | |
384 |
|
384 | |||
385 | ancs = ancestor.ancestors(parentfunc, revmap[a], revmap[b]) |
|
385 | ancs = ancestor.ancestors(parentfunc, revmap[a], revmap[b]) | |
386 | if ancs: |
|
386 | if ancs: | |
387 | # choose a consistent winner when there's a tie |
|
387 | # choose a consistent winner when there's a tie | |
388 | return min(map(nodemap.__getitem__, ancs)) |
|
388 | return min(map(nodemap.__getitem__, ancs)) | |
389 | return nullid |
|
389 | return nullid | |
390 |
|
390 | |||
391 | def commonancestorsheads(self, a, b): |
|
391 | def commonancestorsheads(self, a, b): | |
392 | """calculate all the heads of the common ancestors of nodes a and b""" |
|
392 | """calculate all the heads of the common ancestors of nodes a and b""" | |
393 |
|
393 | |||
394 | if a == nullid or b == nullid: |
|
394 | if a == nullid or b == nullid: | |
395 | return nullid |
|
395 | return nullid | |
396 |
|
396 | |||
397 | revmap, parentfunc = self._buildrevgraph(a, b) |
|
397 | revmap, parentfunc = self._buildrevgraph(a, b) | |
398 | nodemap = dict(((v, k) for (k, v) in revmap.iteritems())) |
|
398 | nodemap = dict(((v, k) for (k, v) in revmap.iteritems())) | |
399 |
|
399 | |||
400 | ancs = ancestor.commonancestorsheads(parentfunc, revmap[a], revmap[b]) |
|
400 | ancs = ancestor.commonancestorsheads(parentfunc, revmap[a], revmap[b]) | |
401 | return map(nodemap.__getitem__, ancs) |
|
401 | return map(nodemap.__getitem__, ancs) | |
402 |
|
402 | |||
403 | def _buildrevgraph(self, a, b): |
|
403 | def _buildrevgraph(self, a, b): | |
404 | """Builds a numeric revision graph for the given two nodes. |
|
404 | """Builds a numeric revision graph for the given two nodes. | |
405 | Returns a node->rev map and a rev->[revs] parent function. |
|
405 | Returns a node->rev map and a rev->[revs] parent function. | |
406 | """ |
|
406 | """ | |
407 | amap = self.ancestormap(a) |
|
407 | amap = self.ancestormap(a) | |
408 | bmap = self.ancestormap(b) |
|
408 | bmap = self.ancestormap(b) | |
409 |
|
409 | |||
410 | # Union the two maps |
|
410 | # Union the two maps | |
411 | parentsmap = collections.defaultdict(list) |
|
411 | parentsmap = collections.defaultdict(list) | |
412 | allparents = set() |
|
412 | allparents = set() | |
413 | for mapping in (amap, bmap): |
|
413 | for mapping in (amap, bmap): | |
414 | for node, pdata in mapping.iteritems(): |
|
414 | for node, pdata in mapping.iteritems(): | |
415 | parents = parentsmap[node] |
|
415 | parents = parentsmap[node] | |
416 | p1, p2, linknode, copyfrom = pdata |
|
416 | p1, p2, linknode, copyfrom = pdata | |
417 | # Don't follow renames (copyfrom). |
|
417 | # Don't follow renames (copyfrom). | |
418 | # remotefilectx.ancestor does that. |
|
418 | # remotefilectx.ancestor does that. | |
419 | if p1 != nullid and not copyfrom: |
|
419 | if p1 != nullid and not copyfrom: | |
420 | parents.append(p1) |
|
420 | parents.append(p1) | |
421 | allparents.add(p1) |
|
421 | allparents.add(p1) | |
422 | if p2 != nullid: |
|
422 | if p2 != nullid: | |
423 | parents.append(p2) |
|
423 | parents.append(p2) | |
424 | allparents.add(p2) |
|
424 | allparents.add(p2) | |
425 |
|
425 | |||
426 | # Breadth first traversal to build linkrev graph |
|
426 | # Breadth first traversal to build linkrev graph | |
427 | parentrevs = collections.defaultdict(list) |
|
427 | parentrevs = collections.defaultdict(list) | |
428 | revmap = {} |
|
428 | revmap = {} | |
429 | queue = collections.deque(((None, n) for n in parentsmap |
|
429 | queue = collections.deque(((None, n) for n in parentsmap | |
430 | if n not in allparents)) |
|
430 | if n not in allparents)) | |
431 | while queue: |
|
431 | while queue: | |
432 | prevrev, current = queue.pop() |
|
432 | prevrev, current = queue.pop() | |
433 | if current in revmap: |
|
433 | if current in revmap: | |
434 | if prevrev: |
|
434 | if prevrev: | |
435 | parentrevs[prevrev].append(revmap[current]) |
|
435 | parentrevs[prevrev].append(revmap[current]) | |
436 | continue |
|
436 | continue | |
437 |
|
437 | |||
438 | # Assign linkrevs in reverse order, so start at |
|
438 | # Assign linkrevs in reverse order, so start at | |
439 | # len(parentsmap) and work backwards. |
|
439 | # len(parentsmap) and work backwards. | |
440 | currentrev = len(parentsmap) - len(revmap) - 1 |
|
440 | currentrev = len(parentsmap) - len(revmap) - 1 | |
441 | revmap[current] = currentrev |
|
441 | revmap[current] = currentrev | |
442 |
|
442 | |||
443 | if prevrev: |
|
443 | if prevrev: | |
444 | parentrevs[prevrev].append(currentrev) |
|
444 | parentrevs[prevrev].append(currentrev) | |
445 |
|
445 | |||
446 | for parent in parentsmap.get(current): |
|
446 | for parent in parentsmap.get(current): | |
447 | queue.appendleft((currentrev, parent)) |
|
447 | queue.appendleft((currentrev, parent)) | |
448 |
|
448 | |||
449 | return revmap, parentrevs.__getitem__ |
|
449 | return revmap, parentrevs.__getitem__ | |
450 |
|
450 | |||
451 | def strip(self, minlink, transaction): |
|
451 | def strip(self, minlink, transaction): | |
452 | pass |
|
452 | pass | |
453 |
|
453 | |||
454 | # misc unused things |
|
454 | # misc unused things | |
455 | def files(self): |
|
455 | def files(self): | |
456 | return [] |
|
456 | return [] | |
457 |
|
457 | |||
458 | def checksize(self): |
|
458 | def checksize(self): | |
459 | return 0, 0 |
|
459 | return 0, 0 |
@@ -1,2653 +1,2653 b'' | |||||
1 | # revlog.py - storage back-end for mercurial |
|
1 | # revlog.py - storage back-end for mercurial | |
2 | # |
|
2 | # | |
3 | # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> |
|
3 | # Copyright 2005-2007 Matt Mackall <mpm@selenic.com> | |
4 | # |
|
4 | # | |
5 | # This software may be used and distributed according to the terms of the |
|
5 | # This software may be used and distributed according to the terms of the | |
6 | # GNU General Public License version 2 or any later version. |
|
6 | # GNU General Public License version 2 or any later version. | |
7 |
|
7 | |||
8 | """Storage back-end for Mercurial. |
|
8 | """Storage back-end for Mercurial. | |
9 |
|
9 | |||
10 | This provides efficient delta storage with O(1) retrieve and append |
|
10 | This provides efficient delta storage with O(1) retrieve and append | |
11 | and O(changes) merge between branches. |
|
11 | and O(changes) merge between branches. | |
12 | """ |
|
12 | """ | |
13 |
|
13 | |||
14 | from __future__ import absolute_import |
|
14 | from __future__ import absolute_import | |
15 |
|
15 | |||
16 | import collections |
|
16 | import collections | |
17 | import contextlib |
|
17 | import contextlib | |
18 | import errno |
|
18 | import errno | |
19 | import io |
|
19 | import io | |
20 | import os |
|
20 | import os | |
21 | import struct |
|
21 | import struct | |
22 | import zlib |
|
22 | import zlib | |
23 |
|
23 | |||
24 | # import stuff from node for others to import from revlog |
|
24 | # import stuff from node for others to import from revlog | |
25 | from .node import ( |
|
25 | from .node import ( | |
26 | bin, |
|
26 | bin, | |
27 | hex, |
|
27 | hex, | |
28 | nullhex, |
|
28 | nullhex, | |
29 | nullid, |
|
29 | nullid, | |
30 | nullrev, |
|
30 | nullrev, | |
31 | short, |
|
31 | short, | |
32 | wdirfilenodeids, |
|
32 | wdirfilenodeids, | |
33 | wdirhex, |
|
33 | wdirhex, | |
34 | wdirid, |
|
34 | wdirid, | |
35 | wdirrev, |
|
35 | wdirrev, | |
36 | ) |
|
36 | ) | |
37 | from .i18n import _ |
|
37 | from .i18n import _ | |
38 | from .revlogutils.constants import ( |
|
38 | from .revlogutils.constants import ( | |
39 | FLAG_GENERALDELTA, |
|
39 | FLAG_GENERALDELTA, | |
40 | FLAG_INLINE_DATA, |
|
40 | FLAG_INLINE_DATA, | |
41 | REVLOGV0, |
|
41 | REVLOGV0, | |
42 | REVLOGV1, |
|
42 | REVLOGV1, | |
43 | REVLOGV1_FLAGS, |
|
43 | REVLOGV1_FLAGS, | |
44 | REVLOGV2, |
|
44 | REVLOGV2, | |
45 | REVLOGV2_FLAGS, |
|
45 | REVLOGV2_FLAGS, | |
46 | REVLOG_DEFAULT_FLAGS, |
|
46 | REVLOG_DEFAULT_FLAGS, | |
47 | REVLOG_DEFAULT_FORMAT, |
|
47 | REVLOG_DEFAULT_FORMAT, | |
48 | REVLOG_DEFAULT_VERSION, |
|
48 | REVLOG_DEFAULT_VERSION, | |
49 | ) |
|
49 | ) | |
50 | from .revlogutils.flagutil import ( |
|
50 | from .revlogutils.flagutil import ( | |
51 | REVIDX_DEFAULT_FLAGS, |
|
51 | REVIDX_DEFAULT_FLAGS, | |
52 | REVIDX_ELLIPSIS, |
|
52 | REVIDX_ELLIPSIS, | |
53 | REVIDX_EXTSTORED, |
|
53 | REVIDX_EXTSTORED, | |
54 | REVIDX_FLAGS_ORDER, |
|
54 | REVIDX_FLAGS_ORDER, | |
55 | REVIDX_ISCENSORED, |
|
55 | REVIDX_ISCENSORED, | |
56 | REVIDX_RAWTEXT_CHANGING_FLAGS, |
|
56 | REVIDX_RAWTEXT_CHANGING_FLAGS, | |
57 | ) |
|
57 | ) | |
58 | from .thirdparty import ( |
|
58 | from .thirdparty import ( | |
59 | attr, |
|
59 | attr, | |
60 | ) |
|
60 | ) | |
61 | from . import ( |
|
61 | from . import ( | |
62 | ancestor, |
|
62 | ancestor, | |
63 | dagop, |
|
63 | dagop, | |
64 | error, |
|
64 | error, | |
65 | mdiff, |
|
65 | mdiff, | |
66 | policy, |
|
66 | policy, | |
67 | pycompat, |
|
67 | pycompat, | |
68 | templatefilters, |
|
68 | templatefilters, | |
69 | util, |
|
69 | util, | |
70 | ) |
|
70 | ) | |
71 | from .interfaces import ( |
|
71 | from .interfaces import ( | |
72 | repository, |
|
72 | repository, | |
73 | util as interfaceutil, |
|
73 | util as interfaceutil, | |
74 | ) |
|
74 | ) | |
75 | from .revlogutils import ( |
|
75 | from .revlogutils import ( | |
76 | deltas as deltautil, |
|
76 | deltas as deltautil, | |
77 | flagutil, |
|
77 | flagutil, | |
78 | ) |
|
78 | ) | |
79 | from .utils import ( |
|
79 | from .utils import ( | |
80 | storageutil, |
|
80 | storageutil, | |
81 | stringutil, |
|
81 | stringutil, | |
82 | ) |
|
82 | ) | |
83 |
|
83 | |||
84 | # blanked usage of all the name to prevent pyflakes constraints |
|
84 | # blanked usage of all the name to prevent pyflakes constraints | |
85 | # We need these name available in the module for extensions. |
|
85 | # We need these name available in the module for extensions. | |
86 | REVLOGV0 |
|
86 | REVLOGV0 | |
87 | REVLOGV1 |
|
87 | REVLOGV1 | |
88 | REVLOGV2 |
|
88 | REVLOGV2 | |
89 | FLAG_INLINE_DATA |
|
89 | FLAG_INLINE_DATA | |
90 | FLAG_GENERALDELTA |
|
90 | FLAG_GENERALDELTA | |
91 | REVLOG_DEFAULT_FLAGS |
|
91 | REVLOG_DEFAULT_FLAGS | |
92 | REVLOG_DEFAULT_FORMAT |
|
92 | REVLOG_DEFAULT_FORMAT | |
93 | REVLOG_DEFAULT_VERSION |
|
93 | REVLOG_DEFAULT_VERSION | |
94 | REVLOGV1_FLAGS |
|
94 | REVLOGV1_FLAGS | |
95 | REVLOGV2_FLAGS |
|
95 | REVLOGV2_FLAGS | |
96 | REVIDX_ISCENSORED |
|
96 | REVIDX_ISCENSORED | |
97 | REVIDX_ELLIPSIS |
|
97 | REVIDX_ELLIPSIS | |
98 | REVIDX_EXTSTORED |
|
98 | REVIDX_EXTSTORED | |
99 | REVIDX_DEFAULT_FLAGS |
|
99 | REVIDX_DEFAULT_FLAGS | |
100 | REVIDX_FLAGS_ORDER |
|
100 | REVIDX_FLAGS_ORDER | |
101 | REVIDX_RAWTEXT_CHANGING_FLAGS |
|
101 | REVIDX_RAWTEXT_CHANGING_FLAGS | |
102 |
|
102 | |||
103 | parsers = policy.importmod(r'parsers') |
|
103 | parsers = policy.importmod(r'parsers') | |
104 | rustancestor = policy.importrust(r'ancestor') |
|
104 | rustancestor = policy.importrust(r'ancestor') | |
105 | rustdagop = policy.importrust(r'dagop') |
|
105 | rustdagop = policy.importrust(r'dagop') | |
106 |
|
106 | |||
107 | # Aliased for performance. |
|
107 | # Aliased for performance. | |
108 | _zlibdecompress = zlib.decompress |
|
108 | _zlibdecompress = zlib.decompress | |
109 |
|
109 | |||
110 | # max size of revlog with inline data |
|
110 | # max size of revlog with inline data | |
111 | _maxinline = 131072 |
|
111 | _maxinline = 131072 | |
112 | _chunksize = 1048576 |
|
112 | _chunksize = 1048576 | |
113 |
|
113 | |||
114 | # Flag processors for REVIDX_ELLIPSIS. |
|
114 | # Flag processors for REVIDX_ELLIPSIS. | |
115 | def ellipsisreadprocessor(rl, text): |
|
115 | def ellipsisreadprocessor(rl, text): | |
116 | return text, False, {} |
|
116 | return text, False, {} | |
117 |
|
117 | |||
118 | def ellipsiswriteprocessor(rl, text, sidedata): |
|
118 | def ellipsiswriteprocessor(rl, text, sidedata): | |
119 | return text, False |
|
119 | return text, False | |
120 |
|
120 | |||
121 | def ellipsisrawprocessor(rl, text): |
|
121 | def ellipsisrawprocessor(rl, text): | |
122 | return False |
|
122 | return False | |
123 |
|
123 | |||
124 | ellipsisprocessor = ( |
|
124 | ellipsisprocessor = ( | |
125 | ellipsisreadprocessor, |
|
125 | ellipsisreadprocessor, | |
126 | ellipsiswriteprocessor, |
|
126 | ellipsiswriteprocessor, | |
127 | ellipsisrawprocessor, |
|
127 | ellipsisrawprocessor, | |
128 | ) |
|
128 | ) | |
129 |
|
129 | |||
130 | def getoffset(q): |
|
130 | def getoffset(q): | |
131 | return int(q >> 16) |
|
131 | return int(q >> 16) | |
132 |
|
132 | |||
133 | def gettype(q): |
|
133 | def gettype(q): | |
134 | return int(q & 0xFFFF) |
|
134 | return int(q & 0xFFFF) | |
135 |
|
135 | |||
136 | def offset_type(offset, type): |
|
136 | def offset_type(offset, type): | |
137 | if (type & ~flagutil.REVIDX_KNOWN_FLAGS) != 0: |
|
137 | if (type & ~flagutil.REVIDX_KNOWN_FLAGS) != 0: | |
138 | raise ValueError('unknown revlog index flags') |
|
138 | raise ValueError('unknown revlog index flags') | |
139 | return int(int(offset) << 16 | type) |
|
139 | return int(int(offset) << 16 | type) | |
140 |
|
140 | |||
141 | @attr.s(slots=True, frozen=True) |
|
141 | @attr.s(slots=True, frozen=True) | |
142 | class _revisioninfo(object): |
|
142 | class _revisioninfo(object): | |
143 | """Information about a revision that allows building its fulltext |
|
143 | """Information about a revision that allows building its fulltext | |
144 | node: expected hash of the revision |
|
144 | node: expected hash of the revision | |
145 | p1, p2: parent revs of the revision |
|
145 | p1, p2: parent revs of the revision | |
146 | btext: built text cache consisting of a one-element list |
|
146 | btext: built text cache consisting of a one-element list | |
147 | cachedelta: (baserev, uncompressed_delta) or None |
|
147 | cachedelta: (baserev, uncompressed_delta) or None | |
148 | flags: flags associated to the revision storage |
|
148 | flags: flags associated to the revision storage | |
149 |
|
149 | |||
150 | One of btext[0] or cachedelta must be set. |
|
150 | One of btext[0] or cachedelta must be set. | |
151 | """ |
|
151 | """ | |
152 | node = attr.ib() |
|
152 | node = attr.ib() | |
153 | p1 = attr.ib() |
|
153 | p1 = attr.ib() | |
154 | p2 = attr.ib() |
|
154 | p2 = attr.ib() | |
155 | btext = attr.ib() |
|
155 | btext = attr.ib() | |
156 | textlen = attr.ib() |
|
156 | textlen = attr.ib() | |
157 | cachedelta = attr.ib() |
|
157 | cachedelta = attr.ib() | |
158 | flags = attr.ib() |
|
158 | flags = attr.ib() | |
159 |
|
159 | |||
160 | @interfaceutil.implementer(repository.irevisiondelta) |
|
160 | @interfaceutil.implementer(repository.irevisiondelta) | |
161 | @attr.s(slots=True) |
|
161 | @attr.s(slots=True) | |
162 | class revlogrevisiondelta(object): |
|
162 | class revlogrevisiondelta(object): | |
163 | node = attr.ib() |
|
163 | node = attr.ib() | |
164 | p1node = attr.ib() |
|
164 | p1node = attr.ib() | |
165 | p2node = attr.ib() |
|
165 | p2node = attr.ib() | |
166 | basenode = attr.ib() |
|
166 | basenode = attr.ib() | |
167 | flags = attr.ib() |
|
167 | flags = attr.ib() | |
168 | baserevisionsize = attr.ib() |
|
168 | baserevisionsize = attr.ib() | |
169 | revision = attr.ib() |
|
169 | revision = attr.ib() | |
170 | delta = attr.ib() |
|
170 | delta = attr.ib() | |
171 | linknode = attr.ib(default=None) |
|
171 | linknode = attr.ib(default=None) | |
172 |
|
172 | |||
173 | @interfaceutil.implementer(repository.iverifyproblem) |
|
173 | @interfaceutil.implementer(repository.iverifyproblem) | |
174 | @attr.s(frozen=True) |
|
174 | @attr.s(frozen=True) | |
175 | class revlogproblem(object): |
|
175 | class revlogproblem(object): | |
176 | warning = attr.ib(default=None) |
|
176 | warning = attr.ib(default=None) | |
177 | error = attr.ib(default=None) |
|
177 | error = attr.ib(default=None) | |
178 | node = attr.ib(default=None) |
|
178 | node = attr.ib(default=None) | |
179 |
|
179 | |||
180 | # index v0: |
|
180 | # index v0: | |
181 | # 4 bytes: offset |
|
181 | # 4 bytes: offset | |
182 | # 4 bytes: compressed length |
|
182 | # 4 bytes: compressed length | |
183 | # 4 bytes: base rev |
|
183 | # 4 bytes: base rev | |
184 | # 4 bytes: link rev |
|
184 | # 4 bytes: link rev | |
185 | # 20 bytes: parent 1 nodeid |
|
185 | # 20 bytes: parent 1 nodeid | |
186 | # 20 bytes: parent 2 nodeid |
|
186 | # 20 bytes: parent 2 nodeid | |
187 | # 20 bytes: nodeid |
|
187 | # 20 bytes: nodeid | |
188 | indexformatv0 = struct.Struct(">4l20s20s20s") |
|
188 | indexformatv0 = struct.Struct(">4l20s20s20s") | |
189 | indexformatv0_pack = indexformatv0.pack |
|
189 | indexformatv0_pack = indexformatv0.pack | |
190 | indexformatv0_unpack = indexformatv0.unpack |
|
190 | indexformatv0_unpack = indexformatv0.unpack | |
191 |
|
191 | |||
192 | class revlogoldindex(list): |
|
192 | class revlogoldindex(list): | |
193 | def __getitem__(self, i): |
|
193 | def __getitem__(self, i): | |
194 | if i == -1: |
|
194 | if i == -1: | |
195 | return (0, 0, 0, -1, -1, -1, -1, nullid) |
|
195 | return (0, 0, 0, -1, -1, -1, -1, nullid) | |
196 | return list.__getitem__(self, i) |
|
196 | return list.__getitem__(self, i) | |
197 |
|
197 | |||
198 | class revlogoldio(object): |
|
198 | class revlogoldio(object): | |
199 | def __init__(self): |
|
199 | def __init__(self): | |
200 | self.size = indexformatv0.size |
|
200 | self.size = indexformatv0.size | |
201 |
|
201 | |||
202 | def parseindex(self, data, inline): |
|
202 | def parseindex(self, data, inline): | |
203 | s = self.size |
|
203 | s = self.size | |
204 | index = [] |
|
204 | index = [] | |
205 | nodemap = {nullid: nullrev} |
|
205 | nodemap = {nullid: nullrev} | |
206 | n = off = 0 |
|
206 | n = off = 0 | |
207 | l = len(data) |
|
207 | l = len(data) | |
208 | while off + s <= l: |
|
208 | while off + s <= l: | |
209 | cur = data[off:off + s] |
|
209 | cur = data[off:off + s] | |
210 | off += s |
|
210 | off += s | |
211 | e = indexformatv0_unpack(cur) |
|
211 | e = indexformatv0_unpack(cur) | |
212 | # transform to revlogv1 format |
|
212 | # transform to revlogv1 format | |
213 | e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3], |
|
213 | e2 = (offset_type(e[0], 0), e[1], -1, e[2], e[3], | |
214 | nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6]) |
|
214 | nodemap.get(e[4], nullrev), nodemap.get(e[5], nullrev), e[6]) | |
215 | index.append(e2) |
|
215 | index.append(e2) | |
216 | nodemap[e[6]] = n |
|
216 | nodemap[e[6]] = n | |
217 | n += 1 |
|
217 | n += 1 | |
218 |
|
218 | |||
219 | return revlogoldindex(index), nodemap, None |
|
219 | return revlogoldindex(index), nodemap, None | |
220 |
|
220 | |||
221 | def packentry(self, entry, node, version, rev): |
|
221 | def packentry(self, entry, node, version, rev): | |
222 | if gettype(entry[0]): |
|
222 | if gettype(entry[0]): | |
223 | raise error.RevlogError(_('index entry flags need revlog ' |
|
223 | raise error.RevlogError(_('index entry flags need revlog ' | |
224 | 'version 1')) |
|
224 | 'version 1')) | |
225 | e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4], |
|
225 | e2 = (getoffset(entry[0]), entry[1], entry[3], entry[4], | |
226 | node(entry[5]), node(entry[6]), entry[7]) |
|
226 | node(entry[5]), node(entry[6]), entry[7]) | |
227 | return indexformatv0_pack(*e2) |
|
227 | return indexformatv0_pack(*e2) | |
228 |
|
228 | |||
229 | # index ng: |
|
229 | # index ng: | |
230 | # 6 bytes: offset |
|
230 | # 6 bytes: offset | |
231 | # 2 bytes: flags |
|
231 | # 2 bytes: flags | |
232 | # 4 bytes: compressed length |
|
232 | # 4 bytes: compressed length | |
233 | # 4 bytes: uncompressed length |
|
233 | # 4 bytes: uncompressed length | |
234 | # 4 bytes: base rev |
|
234 | # 4 bytes: base rev | |
235 | # 4 bytes: link rev |
|
235 | # 4 bytes: link rev | |
236 | # 4 bytes: parent 1 rev |
|
236 | # 4 bytes: parent 1 rev | |
237 | # 4 bytes: parent 2 rev |
|
237 | # 4 bytes: parent 2 rev | |
238 | # 32 bytes: nodeid |
|
238 | # 32 bytes: nodeid | |
239 | indexformatng = struct.Struct(">Qiiiiii20s12x") |
|
239 | indexformatng = struct.Struct(">Qiiiiii20s12x") | |
240 | indexformatng_pack = indexformatng.pack |
|
240 | indexformatng_pack = indexformatng.pack | |
241 | versionformat = struct.Struct(">I") |
|
241 | versionformat = struct.Struct(">I") | |
242 | versionformat_pack = versionformat.pack |
|
242 | versionformat_pack = versionformat.pack | |
243 | versionformat_unpack = versionformat.unpack |
|
243 | versionformat_unpack = versionformat.unpack | |
244 |
|
244 | |||
245 | # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte |
|
245 | # corresponds to uncompressed length of indexformatng (2 gigs, 4-byte | |
246 | # signed integer) |
|
246 | # signed integer) | |
247 | _maxentrysize = 0x7fffffff |
|
247 | _maxentrysize = 0x7fffffff | |
248 |
|
248 | |||
249 | class revlogio(object): |
|
249 | class revlogio(object): | |
250 | def __init__(self): |
|
250 | def __init__(self): | |
251 | self.size = indexformatng.size |
|
251 | self.size = indexformatng.size | |
252 |
|
252 | |||
253 | def parseindex(self, data, inline): |
|
253 | def parseindex(self, data, inline): | |
254 | # call the C implementation to parse the index data |
|
254 | # call the C implementation to parse the index data | |
255 | index, cache = parsers.parse_index2(data, inline) |
|
255 | index, cache = parsers.parse_index2(data, inline) | |
256 | return index, getattr(index, 'nodemap', None), cache |
|
256 | return index, getattr(index, 'nodemap', None), cache | |
257 |
|
257 | |||
258 | def packentry(self, entry, node, version, rev): |
|
258 | def packentry(self, entry, node, version, rev): | |
259 | p = indexformatng_pack(*entry) |
|
259 | p = indexformatng_pack(*entry) | |
260 | if rev == 0: |
|
260 | if rev == 0: | |
261 | p = versionformat_pack(version) + p[4:] |
|
261 | p = versionformat_pack(version) + p[4:] | |
262 | return p |
|
262 | return p | |
263 |
|
263 | |||
264 | class revlog(flagutil.flagprocessorsmixin): |
|
264 | class revlog(object): | |
265 | """ |
|
265 | """ | |
266 | the underlying revision storage object |
|
266 | the underlying revision storage object | |
267 |
|
267 | |||
268 | A revlog consists of two parts, an index and the revision data. |
|
268 | A revlog consists of two parts, an index and the revision data. | |
269 |
|
269 | |||
270 | The index is a file with a fixed record size containing |
|
270 | The index is a file with a fixed record size containing | |
271 | information on each revision, including its nodeid (hash), the |
|
271 | information on each revision, including its nodeid (hash), the | |
272 | nodeids of its parents, the position and offset of its data within |
|
272 | nodeids of its parents, the position and offset of its data within | |
273 | the data file, and the revision it's based on. Finally, each entry |
|
273 | the data file, and the revision it's based on. Finally, each entry | |
274 | contains a linkrev entry that can serve as a pointer to external |
|
274 | contains a linkrev entry that can serve as a pointer to external | |
275 | data. |
|
275 | data. | |
276 |
|
276 | |||
277 | The revision data itself is a linear collection of data chunks. |
|
277 | The revision data itself is a linear collection of data chunks. | |
278 | Each chunk represents a revision and is usually represented as a |
|
278 | Each chunk represents a revision and is usually represented as a | |
279 | delta against the previous chunk. To bound lookup time, runs of |
|
279 | delta against the previous chunk. To bound lookup time, runs of | |
280 | deltas are limited to about 2 times the length of the original |
|
280 | deltas are limited to about 2 times the length of the original | |
281 | version data. This makes retrieval of a version proportional to |
|
281 | version data. This makes retrieval of a version proportional to | |
282 | its size, or O(1) relative to the number of revisions. |
|
282 | its size, or O(1) relative to the number of revisions. | |
283 |
|
283 | |||
284 | Both pieces of the revlog are written to in an append-only |
|
284 | Both pieces of the revlog are written to in an append-only | |
285 | fashion, which means we never need to rewrite a file to insert or |
|
285 | fashion, which means we never need to rewrite a file to insert or | |
286 | remove data, and can use some simple techniques to avoid the need |
|
286 | remove data, and can use some simple techniques to avoid the need | |
287 | for locking while reading. |
|
287 | for locking while reading. | |
288 |
|
288 | |||
289 | If checkambig, indexfile is opened with checkambig=True at |
|
289 | If checkambig, indexfile is opened with checkambig=True at | |
290 | writing, to avoid file stat ambiguity. |
|
290 | writing, to avoid file stat ambiguity. | |
291 |
|
291 | |||
292 | If mmaplargeindex is True, and an mmapindexthreshold is set, the |
|
292 | If mmaplargeindex is True, and an mmapindexthreshold is set, the | |
293 | index will be mmapped rather than read if it is larger than the |
|
293 | index will be mmapped rather than read if it is larger than the | |
294 | configured threshold. |
|
294 | configured threshold. | |
295 |
|
295 | |||
296 | If censorable is True, the revlog can have censored revisions. |
|
296 | If censorable is True, the revlog can have censored revisions. | |
297 |
|
297 | |||
298 | If `upperboundcomp` is not None, this is the expected maximal gain from |
|
298 | If `upperboundcomp` is not None, this is the expected maximal gain from | |
299 | compression for the data content. |
|
299 | compression for the data content. | |
300 | """ |
|
300 | """ | |
301 |
|
301 | |||
302 | _flagserrorclass = error.RevlogError |
|
302 | _flagserrorclass = error.RevlogError | |
303 |
|
303 | |||
304 | def __init__(self, opener, indexfile, datafile=None, checkambig=False, |
|
304 | def __init__(self, opener, indexfile, datafile=None, checkambig=False, | |
305 | mmaplargeindex=False, censorable=False, |
|
305 | mmaplargeindex=False, censorable=False, | |
306 | upperboundcomp=None): |
|
306 | upperboundcomp=None): | |
307 | """ |
|
307 | """ | |
308 | create a revlog object |
|
308 | create a revlog object | |
309 |
|
309 | |||
310 | opener is a function that abstracts the file opening operation |
|
310 | opener is a function that abstracts the file opening operation | |
311 | and can be used to implement COW semantics or the like. |
|
311 | and can be used to implement COW semantics or the like. | |
312 |
|
312 | |||
313 | """ |
|
313 | """ | |
314 | self.upperboundcomp = upperboundcomp |
|
314 | self.upperboundcomp = upperboundcomp | |
315 | self.indexfile = indexfile |
|
315 | self.indexfile = indexfile | |
316 | self.datafile = datafile or (indexfile[:-2] + ".d") |
|
316 | self.datafile = datafile or (indexfile[:-2] + ".d") | |
317 | self.opener = opener |
|
317 | self.opener = opener | |
318 | # When True, indexfile is opened with checkambig=True at writing, to |
|
318 | # When True, indexfile is opened with checkambig=True at writing, to | |
319 | # avoid file stat ambiguity. |
|
319 | # avoid file stat ambiguity. | |
320 | self._checkambig = checkambig |
|
320 | self._checkambig = checkambig | |
321 | self._mmaplargeindex = mmaplargeindex |
|
321 | self._mmaplargeindex = mmaplargeindex | |
322 | self._censorable = censorable |
|
322 | self._censorable = censorable | |
323 | # 3-tuple of (node, rev, text) for a raw revision. |
|
323 | # 3-tuple of (node, rev, text) for a raw revision. | |
324 | self._revisioncache = None |
|
324 | self._revisioncache = None | |
325 | # Maps rev to chain base rev. |
|
325 | # Maps rev to chain base rev. | |
326 | self._chainbasecache = util.lrucachedict(100) |
|
326 | self._chainbasecache = util.lrucachedict(100) | |
327 | # 2-tuple of (offset, data) of raw data from the revlog at an offset. |
|
327 | # 2-tuple of (offset, data) of raw data from the revlog at an offset. | |
328 | self._chunkcache = (0, '') |
|
328 | self._chunkcache = (0, '') | |
329 | # How much data to read and cache into the raw revlog data cache. |
|
329 | # How much data to read and cache into the raw revlog data cache. | |
330 | self._chunkcachesize = 65536 |
|
330 | self._chunkcachesize = 65536 | |
331 | self._maxchainlen = None |
|
331 | self._maxchainlen = None | |
332 | self._deltabothparents = True |
|
332 | self._deltabothparents = True | |
333 | self.index = [] |
|
333 | self.index = [] | |
334 | # Mapping of partial identifiers to full nodes. |
|
334 | # Mapping of partial identifiers to full nodes. | |
335 | self._pcache = {} |
|
335 | self._pcache = {} | |
336 | # Mapping of revision integer to full node. |
|
336 | # Mapping of revision integer to full node. | |
337 | self._nodecache = {nullid: nullrev} |
|
337 | self._nodecache = {nullid: nullrev} | |
338 | self._nodepos = None |
|
338 | self._nodepos = None | |
339 | self._compengine = 'zlib' |
|
339 | self._compengine = 'zlib' | |
340 | self._compengineopts = {} |
|
340 | self._compengineopts = {} | |
341 | self._maxdeltachainspan = -1 |
|
341 | self._maxdeltachainspan = -1 | |
342 | self._withsparseread = False |
|
342 | self._withsparseread = False | |
343 | self._sparserevlog = False |
|
343 | self._sparserevlog = False | |
344 | self._srdensitythreshold = 0.50 |
|
344 | self._srdensitythreshold = 0.50 | |
345 | self._srmingapsize = 262144 |
|
345 | self._srmingapsize = 262144 | |
346 |
|
346 | |||
347 | # Make copy of flag processors so each revlog instance can support |
|
347 | # Make copy of flag processors so each revlog instance can support | |
348 | # custom flags. |
|
348 | # custom flags. | |
349 | self._flagprocessors = dict(flagutil.flagprocessors) |
|
349 | self._flagprocessors = dict(flagutil.flagprocessors) | |
350 |
|
350 | |||
351 | # 2-tuple of file handles being used for active writing. |
|
351 | # 2-tuple of file handles being used for active writing. | |
352 | self._writinghandles = None |
|
352 | self._writinghandles = None | |
353 |
|
353 | |||
354 | self._loadindex() |
|
354 | self._loadindex() | |
355 |
|
355 | |||
356 | def _loadindex(self): |
|
356 | def _loadindex(self): | |
357 | mmapindexthreshold = None |
|
357 | mmapindexthreshold = None | |
358 | opts = getattr(self.opener, 'options', {}) or {} |
|
358 | opts = getattr(self.opener, 'options', {}) or {} | |
359 |
|
359 | |||
360 | if 'revlogv2' in opts: |
|
360 | if 'revlogv2' in opts: | |
361 | newversionflags = REVLOGV2 | FLAG_INLINE_DATA |
|
361 | newversionflags = REVLOGV2 | FLAG_INLINE_DATA | |
362 | elif 'revlogv1' in opts: |
|
362 | elif 'revlogv1' in opts: | |
363 | newversionflags = REVLOGV1 | FLAG_INLINE_DATA |
|
363 | newversionflags = REVLOGV1 | FLAG_INLINE_DATA | |
364 | if 'generaldelta' in opts: |
|
364 | if 'generaldelta' in opts: | |
365 | newversionflags |= FLAG_GENERALDELTA |
|
365 | newversionflags |= FLAG_GENERALDELTA | |
366 | elif getattr(self.opener, 'options', None) is not None: |
|
366 | elif getattr(self.opener, 'options', None) is not None: | |
367 | # If options provided but no 'revlog*' found, the repository |
|
367 | # If options provided but no 'revlog*' found, the repository | |
368 | # would have no 'requires' file in it, which means we have to |
|
368 | # would have no 'requires' file in it, which means we have to | |
369 | # stick to the old format. |
|
369 | # stick to the old format. | |
370 | newversionflags = REVLOGV0 |
|
370 | newversionflags = REVLOGV0 | |
371 | else: |
|
371 | else: | |
372 | newversionflags = REVLOG_DEFAULT_VERSION |
|
372 | newversionflags = REVLOG_DEFAULT_VERSION | |
373 |
|
373 | |||
374 | if 'chunkcachesize' in opts: |
|
374 | if 'chunkcachesize' in opts: | |
375 | self._chunkcachesize = opts['chunkcachesize'] |
|
375 | self._chunkcachesize = opts['chunkcachesize'] | |
376 | if 'maxchainlen' in opts: |
|
376 | if 'maxchainlen' in opts: | |
377 | self._maxchainlen = opts['maxchainlen'] |
|
377 | self._maxchainlen = opts['maxchainlen'] | |
378 | if 'deltabothparents' in opts: |
|
378 | if 'deltabothparents' in opts: | |
379 | self._deltabothparents = opts['deltabothparents'] |
|
379 | self._deltabothparents = opts['deltabothparents'] | |
380 | self._lazydelta = bool(opts.get('lazydelta', True)) |
|
380 | self._lazydelta = bool(opts.get('lazydelta', True)) | |
381 | self._lazydeltabase = False |
|
381 | self._lazydeltabase = False | |
382 | if self._lazydelta: |
|
382 | if self._lazydelta: | |
383 | self._lazydeltabase = bool(opts.get('lazydeltabase', False)) |
|
383 | self._lazydeltabase = bool(opts.get('lazydeltabase', False)) | |
384 | if 'compengine' in opts: |
|
384 | if 'compengine' in opts: | |
385 | self._compengine = opts['compengine'] |
|
385 | self._compengine = opts['compengine'] | |
386 | if 'zlib.level' in opts: |
|
386 | if 'zlib.level' in opts: | |
387 | self._compengineopts['zlib.level'] = opts['zlib.level'] |
|
387 | self._compengineopts['zlib.level'] = opts['zlib.level'] | |
388 | if 'zstd.level' in opts: |
|
388 | if 'zstd.level' in opts: | |
389 | self._compengineopts['zstd.level'] = opts['zstd.level'] |
|
389 | self._compengineopts['zstd.level'] = opts['zstd.level'] | |
390 | if 'maxdeltachainspan' in opts: |
|
390 | if 'maxdeltachainspan' in opts: | |
391 | self._maxdeltachainspan = opts['maxdeltachainspan'] |
|
391 | self._maxdeltachainspan = opts['maxdeltachainspan'] | |
392 | if self._mmaplargeindex and 'mmapindexthreshold' in opts: |
|
392 | if self._mmaplargeindex and 'mmapindexthreshold' in opts: | |
393 | mmapindexthreshold = opts['mmapindexthreshold'] |
|
393 | mmapindexthreshold = opts['mmapindexthreshold'] | |
394 | self._sparserevlog = bool(opts.get('sparse-revlog', False)) |
|
394 | self._sparserevlog = bool(opts.get('sparse-revlog', False)) | |
395 | withsparseread = bool(opts.get('with-sparse-read', False)) |
|
395 | withsparseread = bool(opts.get('with-sparse-read', False)) | |
396 | # sparse-revlog forces sparse-read |
|
396 | # sparse-revlog forces sparse-read | |
397 | self._withsparseread = self._sparserevlog or withsparseread |
|
397 | self._withsparseread = self._sparserevlog or withsparseread | |
398 | if 'sparse-read-density-threshold' in opts: |
|
398 | if 'sparse-read-density-threshold' in opts: | |
399 | self._srdensitythreshold = opts['sparse-read-density-threshold'] |
|
399 | self._srdensitythreshold = opts['sparse-read-density-threshold'] | |
400 | if 'sparse-read-min-gap-size' in opts: |
|
400 | if 'sparse-read-min-gap-size' in opts: | |
401 | self._srmingapsize = opts['sparse-read-min-gap-size'] |
|
401 | self._srmingapsize = opts['sparse-read-min-gap-size'] | |
402 | if opts.get('enableellipsis'): |
|
402 | if opts.get('enableellipsis'): | |
403 | self._flagprocessors[REVIDX_ELLIPSIS] = ellipsisprocessor |
|
403 | self._flagprocessors[REVIDX_ELLIPSIS] = ellipsisprocessor | |
404 |
|
404 | |||
405 | # revlog v0 doesn't have flag processors |
|
405 | # revlog v0 doesn't have flag processors | |
406 | for flag, processor in opts.get(b'flagprocessors', {}).iteritems(): |
|
406 | for flag, processor in opts.get(b'flagprocessors', {}).iteritems(): | |
407 | flagutil.insertflagprocessor(flag, processor, self._flagprocessors) |
|
407 | flagutil.insertflagprocessor(flag, processor, self._flagprocessors) | |
408 |
|
408 | |||
409 | if self._chunkcachesize <= 0: |
|
409 | if self._chunkcachesize <= 0: | |
410 | raise error.RevlogError(_('revlog chunk cache size %r is not ' |
|
410 | raise error.RevlogError(_('revlog chunk cache size %r is not ' | |
411 | 'greater than 0') % self._chunkcachesize) |
|
411 | 'greater than 0') % self._chunkcachesize) | |
412 | elif self._chunkcachesize & (self._chunkcachesize - 1): |
|
412 | elif self._chunkcachesize & (self._chunkcachesize - 1): | |
413 | raise error.RevlogError(_('revlog chunk cache size %r is not a ' |
|
413 | raise error.RevlogError(_('revlog chunk cache size %r is not a ' | |
414 | 'power of 2') % self._chunkcachesize) |
|
414 | 'power of 2') % self._chunkcachesize) | |
415 |
|
415 | |||
416 | indexdata = '' |
|
416 | indexdata = '' | |
417 | self._initempty = True |
|
417 | self._initempty = True | |
418 | try: |
|
418 | try: | |
419 | with self._indexfp() as f: |
|
419 | with self._indexfp() as f: | |
420 | if (mmapindexthreshold is not None and |
|
420 | if (mmapindexthreshold is not None and | |
421 | self.opener.fstat(f).st_size >= mmapindexthreshold): |
|
421 | self.opener.fstat(f).st_size >= mmapindexthreshold): | |
422 | # TODO: should .close() to release resources without |
|
422 | # TODO: should .close() to release resources without | |
423 | # relying on Python GC |
|
423 | # relying on Python GC | |
424 | indexdata = util.buffer(util.mmapread(f)) |
|
424 | indexdata = util.buffer(util.mmapread(f)) | |
425 | else: |
|
425 | else: | |
426 | indexdata = f.read() |
|
426 | indexdata = f.read() | |
427 | if len(indexdata) > 0: |
|
427 | if len(indexdata) > 0: | |
428 | versionflags = versionformat_unpack(indexdata[:4])[0] |
|
428 | versionflags = versionformat_unpack(indexdata[:4])[0] | |
429 | self._initempty = False |
|
429 | self._initempty = False | |
430 | else: |
|
430 | else: | |
431 | versionflags = newversionflags |
|
431 | versionflags = newversionflags | |
432 | except IOError as inst: |
|
432 | except IOError as inst: | |
433 | if inst.errno != errno.ENOENT: |
|
433 | if inst.errno != errno.ENOENT: | |
434 | raise |
|
434 | raise | |
435 |
|
435 | |||
436 | versionflags = newversionflags |
|
436 | versionflags = newversionflags | |
437 |
|
437 | |||
438 | self.version = versionflags |
|
438 | self.version = versionflags | |
439 |
|
439 | |||
440 | flags = versionflags & ~0xFFFF |
|
440 | flags = versionflags & ~0xFFFF | |
441 | fmt = versionflags & 0xFFFF |
|
441 | fmt = versionflags & 0xFFFF | |
442 |
|
442 | |||
443 | if fmt == REVLOGV0: |
|
443 | if fmt == REVLOGV0: | |
444 | if flags: |
|
444 | if flags: | |
445 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' |
|
445 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' | |
446 | 'revlog %s') % |
|
446 | 'revlog %s') % | |
447 | (flags >> 16, fmt, self.indexfile)) |
|
447 | (flags >> 16, fmt, self.indexfile)) | |
448 |
|
448 | |||
449 | self._inline = False |
|
449 | self._inline = False | |
450 | self._generaldelta = False |
|
450 | self._generaldelta = False | |
451 |
|
451 | |||
452 | elif fmt == REVLOGV1: |
|
452 | elif fmt == REVLOGV1: | |
453 | if flags & ~REVLOGV1_FLAGS: |
|
453 | if flags & ~REVLOGV1_FLAGS: | |
454 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' |
|
454 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' | |
455 | 'revlog %s') % |
|
455 | 'revlog %s') % | |
456 | (flags >> 16, fmt, self.indexfile)) |
|
456 | (flags >> 16, fmt, self.indexfile)) | |
457 |
|
457 | |||
458 | self._inline = versionflags & FLAG_INLINE_DATA |
|
458 | self._inline = versionflags & FLAG_INLINE_DATA | |
459 | self._generaldelta = versionflags & FLAG_GENERALDELTA |
|
459 | self._generaldelta = versionflags & FLAG_GENERALDELTA | |
460 |
|
460 | |||
461 | elif fmt == REVLOGV2: |
|
461 | elif fmt == REVLOGV2: | |
462 | if flags & ~REVLOGV2_FLAGS: |
|
462 | if flags & ~REVLOGV2_FLAGS: | |
463 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' |
|
463 | raise error.RevlogError(_('unknown flags (%#04x) in version %d ' | |
464 | 'revlog %s') % |
|
464 | 'revlog %s') % | |
465 | (flags >> 16, fmt, self.indexfile)) |
|
465 | (flags >> 16, fmt, self.indexfile)) | |
466 |
|
466 | |||
467 | self._inline = versionflags & FLAG_INLINE_DATA |
|
467 | self._inline = versionflags & FLAG_INLINE_DATA | |
468 | # generaldelta implied by version 2 revlogs. |
|
468 | # generaldelta implied by version 2 revlogs. | |
469 | self._generaldelta = True |
|
469 | self._generaldelta = True | |
470 |
|
470 | |||
471 | else: |
|
471 | else: | |
472 | raise error.RevlogError(_('unknown version (%d) in revlog %s') % |
|
472 | raise error.RevlogError(_('unknown version (%d) in revlog %s') % | |
473 | (fmt, self.indexfile)) |
|
473 | (fmt, self.indexfile)) | |
474 | # sparse-revlog can't be on without general-delta (issue6056) |
|
474 | # sparse-revlog can't be on without general-delta (issue6056) | |
475 | if not self._generaldelta: |
|
475 | if not self._generaldelta: | |
476 | self._sparserevlog = False |
|
476 | self._sparserevlog = False | |
477 |
|
477 | |||
478 | self._storedeltachains = True |
|
478 | self._storedeltachains = True | |
479 |
|
479 | |||
480 | self._io = revlogio() |
|
480 | self._io = revlogio() | |
481 | if self.version == REVLOGV0: |
|
481 | if self.version == REVLOGV0: | |
482 | self._io = revlogoldio() |
|
482 | self._io = revlogoldio() | |
483 | try: |
|
483 | try: | |
484 | d = self._io.parseindex(indexdata, self._inline) |
|
484 | d = self._io.parseindex(indexdata, self._inline) | |
485 | except (ValueError, IndexError): |
|
485 | except (ValueError, IndexError): | |
486 | raise error.RevlogError(_("index %s is corrupted") % |
|
486 | raise error.RevlogError(_("index %s is corrupted") % | |
487 | self.indexfile) |
|
487 | self.indexfile) | |
488 | self.index, nodemap, self._chunkcache = d |
|
488 | self.index, nodemap, self._chunkcache = d | |
489 | if nodemap is not None: |
|
489 | if nodemap is not None: | |
490 | self.nodemap = self._nodecache = nodemap |
|
490 | self.nodemap = self._nodecache = nodemap | |
491 | if not self._chunkcache: |
|
491 | if not self._chunkcache: | |
492 | self._chunkclear() |
|
492 | self._chunkclear() | |
493 | # revnum -> (chain-length, sum-delta-length) |
|
493 | # revnum -> (chain-length, sum-delta-length) | |
494 | self._chaininfocache = {} |
|
494 | self._chaininfocache = {} | |
495 | # revlog header -> revlog compressor |
|
495 | # revlog header -> revlog compressor | |
496 | self._decompressors = {} |
|
496 | self._decompressors = {} | |
497 |
|
497 | |||
498 | @util.propertycache |
|
498 | @util.propertycache | |
499 | def _compressor(self): |
|
499 | def _compressor(self): | |
500 | engine = util.compengines[self._compengine] |
|
500 | engine = util.compengines[self._compengine] | |
501 | return engine.revlogcompressor(self._compengineopts) |
|
501 | return engine.revlogcompressor(self._compengineopts) | |
502 |
|
502 | |||
503 | def _indexfp(self, mode='r'): |
|
503 | def _indexfp(self, mode='r'): | |
504 | """file object for the revlog's index file""" |
|
504 | """file object for the revlog's index file""" | |
505 | args = {r'mode': mode} |
|
505 | args = {r'mode': mode} | |
506 | if mode != 'r': |
|
506 | if mode != 'r': | |
507 | args[r'checkambig'] = self._checkambig |
|
507 | args[r'checkambig'] = self._checkambig | |
508 | if mode == 'w': |
|
508 | if mode == 'w': | |
509 | args[r'atomictemp'] = True |
|
509 | args[r'atomictemp'] = True | |
510 | return self.opener(self.indexfile, **args) |
|
510 | return self.opener(self.indexfile, **args) | |
511 |
|
511 | |||
512 | def _datafp(self, mode='r'): |
|
512 | def _datafp(self, mode='r'): | |
513 | """file object for the revlog's data file""" |
|
513 | """file object for the revlog's data file""" | |
514 | return self.opener(self.datafile, mode=mode) |
|
514 | return self.opener(self.datafile, mode=mode) | |
515 |
|
515 | |||
516 | @contextlib.contextmanager |
|
516 | @contextlib.contextmanager | |
517 | def _datareadfp(self, existingfp=None): |
|
517 | def _datareadfp(self, existingfp=None): | |
518 | """file object suitable to read data""" |
|
518 | """file object suitable to read data""" | |
519 | # Use explicit file handle, if given. |
|
519 | # Use explicit file handle, if given. | |
520 | if existingfp is not None: |
|
520 | if existingfp is not None: | |
521 | yield existingfp |
|
521 | yield existingfp | |
522 |
|
522 | |||
523 | # Use a file handle being actively used for writes, if available. |
|
523 | # Use a file handle being actively used for writes, if available. | |
524 | # There is some danger to doing this because reads will seek the |
|
524 | # There is some danger to doing this because reads will seek the | |
525 | # file. However, _writeentry() performs a SEEK_END before all writes, |
|
525 | # file. However, _writeentry() performs a SEEK_END before all writes, | |
526 | # so we should be safe. |
|
526 | # so we should be safe. | |
527 | elif self._writinghandles: |
|
527 | elif self._writinghandles: | |
528 | if self._inline: |
|
528 | if self._inline: | |
529 | yield self._writinghandles[0] |
|
529 | yield self._writinghandles[0] | |
530 | else: |
|
530 | else: | |
531 | yield self._writinghandles[1] |
|
531 | yield self._writinghandles[1] | |
532 |
|
532 | |||
533 | # Otherwise open a new file handle. |
|
533 | # Otherwise open a new file handle. | |
534 | else: |
|
534 | else: | |
535 | if self._inline: |
|
535 | if self._inline: | |
536 | func = self._indexfp |
|
536 | func = self._indexfp | |
537 | else: |
|
537 | else: | |
538 | func = self._datafp |
|
538 | func = self._datafp | |
539 | with func() as fp: |
|
539 | with func() as fp: | |
540 | yield fp |
|
540 | yield fp | |
541 |
|
541 | |||
542 | def tip(self): |
|
542 | def tip(self): | |
543 | return self.node(len(self.index) - 1) |
|
543 | return self.node(len(self.index) - 1) | |
544 | def __contains__(self, rev): |
|
544 | def __contains__(self, rev): | |
545 | return 0 <= rev < len(self) |
|
545 | return 0 <= rev < len(self) | |
546 | def __len__(self): |
|
546 | def __len__(self): | |
547 | return len(self.index) |
|
547 | return len(self.index) | |
548 | def __iter__(self): |
|
548 | def __iter__(self): | |
549 | return iter(pycompat.xrange(len(self))) |
|
549 | return iter(pycompat.xrange(len(self))) | |
550 | def revs(self, start=0, stop=None): |
|
550 | def revs(self, start=0, stop=None): | |
551 | """iterate over all rev in this revlog (from start to stop)""" |
|
551 | """iterate over all rev in this revlog (from start to stop)""" | |
552 | return storageutil.iterrevs(len(self), start=start, stop=stop) |
|
552 | return storageutil.iterrevs(len(self), start=start, stop=stop) | |
553 |
|
553 | |||
554 | @util.propertycache |
|
554 | @util.propertycache | |
555 | def nodemap(self): |
|
555 | def nodemap(self): | |
556 | if self.index: |
|
556 | if self.index: | |
557 | # populate mapping down to the initial node |
|
557 | # populate mapping down to the initial node | |
558 | node0 = self.index[0][7] # get around changelog filtering |
|
558 | node0 = self.index[0][7] # get around changelog filtering | |
559 | self.rev(node0) |
|
559 | self.rev(node0) | |
560 | return self._nodecache |
|
560 | return self._nodecache | |
561 |
|
561 | |||
562 | def hasnode(self, node): |
|
562 | def hasnode(self, node): | |
563 | try: |
|
563 | try: | |
564 | self.rev(node) |
|
564 | self.rev(node) | |
565 | return True |
|
565 | return True | |
566 | except KeyError: |
|
566 | except KeyError: | |
567 | return False |
|
567 | return False | |
568 |
|
568 | |||
569 | def candelta(self, baserev, rev): |
|
569 | def candelta(self, baserev, rev): | |
570 | """whether two revisions (baserev, rev) can be delta-ed or not""" |
|
570 | """whether two revisions (baserev, rev) can be delta-ed or not""" | |
571 | # Disable delta if either rev requires a content-changing flag |
|
571 | # Disable delta if either rev requires a content-changing flag | |
572 | # processor (ex. LFS). This is because such flag processor can alter |
|
572 | # processor (ex. LFS). This is because such flag processor can alter | |
573 | # the rawtext content that the delta will be based on, and two clients |
|
573 | # the rawtext content that the delta will be based on, and two clients | |
574 | # could have a same revlog node with different flags (i.e. different |
|
574 | # could have a same revlog node with different flags (i.e. different | |
575 | # rawtext contents) and the delta could be incompatible. |
|
575 | # rawtext contents) and the delta could be incompatible. | |
576 | if ((self.flags(baserev) & REVIDX_RAWTEXT_CHANGING_FLAGS) |
|
576 | if ((self.flags(baserev) & REVIDX_RAWTEXT_CHANGING_FLAGS) | |
577 | or (self.flags(rev) & REVIDX_RAWTEXT_CHANGING_FLAGS)): |
|
577 | or (self.flags(rev) & REVIDX_RAWTEXT_CHANGING_FLAGS)): | |
578 | return False |
|
578 | return False | |
579 | return True |
|
579 | return True | |
580 |
|
580 | |||
581 | def clearcaches(self): |
|
581 | def clearcaches(self): | |
582 | self._revisioncache = None |
|
582 | self._revisioncache = None | |
583 | self._chainbasecache.clear() |
|
583 | self._chainbasecache.clear() | |
584 | self._chunkcache = (0, '') |
|
584 | self._chunkcache = (0, '') | |
585 | self._pcache = {} |
|
585 | self._pcache = {} | |
586 |
|
586 | |||
587 | try: |
|
587 | try: | |
588 | # If we are using the native C version, you are in a fun case |
|
588 | # If we are using the native C version, you are in a fun case | |
589 | # where self.index, self.nodemap and self._nodecaches is the same |
|
589 | # where self.index, self.nodemap and self._nodecaches is the same | |
590 | # object. |
|
590 | # object. | |
591 | self._nodecache.clearcaches() |
|
591 | self._nodecache.clearcaches() | |
592 | except AttributeError: |
|
592 | except AttributeError: | |
593 | self._nodecache = {nullid: nullrev} |
|
593 | self._nodecache = {nullid: nullrev} | |
594 | self._nodepos = None |
|
594 | self._nodepos = None | |
595 |
|
595 | |||
596 | def rev(self, node): |
|
596 | def rev(self, node): | |
597 | try: |
|
597 | try: | |
598 | return self._nodecache[node] |
|
598 | return self._nodecache[node] | |
599 | except TypeError: |
|
599 | except TypeError: | |
600 | raise |
|
600 | raise | |
601 | except error.RevlogError: |
|
601 | except error.RevlogError: | |
602 | # parsers.c radix tree lookup failed |
|
602 | # parsers.c radix tree lookup failed | |
603 | if node == wdirid or node in wdirfilenodeids: |
|
603 | if node == wdirid or node in wdirfilenodeids: | |
604 | raise error.WdirUnsupported |
|
604 | raise error.WdirUnsupported | |
605 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
605 | raise error.LookupError(node, self.indexfile, _('no node')) | |
606 | except KeyError: |
|
606 | except KeyError: | |
607 | # pure python cache lookup failed |
|
607 | # pure python cache lookup failed | |
608 | n = self._nodecache |
|
608 | n = self._nodecache | |
609 | i = self.index |
|
609 | i = self.index | |
610 | p = self._nodepos |
|
610 | p = self._nodepos | |
611 | if p is None: |
|
611 | if p is None: | |
612 | p = len(i) - 1 |
|
612 | p = len(i) - 1 | |
613 | else: |
|
613 | else: | |
614 | assert p < len(i) |
|
614 | assert p < len(i) | |
615 | for r in pycompat.xrange(p, -1, -1): |
|
615 | for r in pycompat.xrange(p, -1, -1): | |
616 | v = i[r][7] |
|
616 | v = i[r][7] | |
617 | n[v] = r |
|
617 | n[v] = r | |
618 | if v == node: |
|
618 | if v == node: | |
619 | self._nodepos = r - 1 |
|
619 | self._nodepos = r - 1 | |
620 | return r |
|
620 | return r | |
621 | if node == wdirid or node in wdirfilenodeids: |
|
621 | if node == wdirid or node in wdirfilenodeids: | |
622 | raise error.WdirUnsupported |
|
622 | raise error.WdirUnsupported | |
623 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
623 | raise error.LookupError(node, self.indexfile, _('no node')) | |
624 |
|
624 | |||
625 | # Accessors for index entries. |
|
625 | # Accessors for index entries. | |
626 |
|
626 | |||
627 | # First tuple entry is 8 bytes. First 6 bytes are offset. Last 2 bytes |
|
627 | # First tuple entry is 8 bytes. First 6 bytes are offset. Last 2 bytes | |
628 | # are flags. |
|
628 | # are flags. | |
629 | def start(self, rev): |
|
629 | def start(self, rev): | |
630 | return int(self.index[rev][0] >> 16) |
|
630 | return int(self.index[rev][0] >> 16) | |
631 |
|
631 | |||
632 | def flags(self, rev): |
|
632 | def flags(self, rev): | |
633 | return self.index[rev][0] & 0xFFFF |
|
633 | return self.index[rev][0] & 0xFFFF | |
634 |
|
634 | |||
635 | def length(self, rev): |
|
635 | def length(self, rev): | |
636 | return self.index[rev][1] |
|
636 | return self.index[rev][1] | |
637 |
|
637 | |||
638 | def rawsize(self, rev): |
|
638 | def rawsize(self, rev): | |
639 | """return the length of the uncompressed text for a given revision""" |
|
639 | """return the length of the uncompressed text for a given revision""" | |
640 | l = self.index[rev][2] |
|
640 | l = self.index[rev][2] | |
641 | if l >= 0: |
|
641 | if l >= 0: | |
642 | return l |
|
642 | return l | |
643 |
|
643 | |||
644 | t = self.rawdata(rev) |
|
644 | t = self.rawdata(rev) | |
645 | return len(t) |
|
645 | return len(t) | |
646 |
|
646 | |||
647 | def size(self, rev): |
|
647 | def size(self, rev): | |
648 | """length of non-raw text (processed by a "read" flag processor)""" |
|
648 | """length of non-raw text (processed by a "read" flag processor)""" | |
649 | # fast path: if no "read" flag processor could change the content, |
|
649 | # fast path: if no "read" flag processor could change the content, | |
650 | # size is rawsize. note: ELLIPSIS is known to not change the content. |
|
650 | # size is rawsize. note: ELLIPSIS is known to not change the content. | |
651 | flags = self.flags(rev) |
|
651 | flags = self.flags(rev) | |
652 | if flags & (flagutil.REVIDX_KNOWN_FLAGS ^ REVIDX_ELLIPSIS) == 0: |
|
652 | if flags & (flagutil.REVIDX_KNOWN_FLAGS ^ REVIDX_ELLIPSIS) == 0: | |
653 | return self.rawsize(rev) |
|
653 | return self.rawsize(rev) | |
654 |
|
654 | |||
655 | return len(self.revision(rev, raw=False)) |
|
655 | return len(self.revision(rev, raw=False)) | |
656 |
|
656 | |||
657 | def chainbase(self, rev): |
|
657 | def chainbase(self, rev): | |
658 | base = self._chainbasecache.get(rev) |
|
658 | base = self._chainbasecache.get(rev) | |
659 | if base is not None: |
|
659 | if base is not None: | |
660 | return base |
|
660 | return base | |
661 |
|
661 | |||
662 | index = self.index |
|
662 | index = self.index | |
663 | iterrev = rev |
|
663 | iterrev = rev | |
664 | base = index[iterrev][3] |
|
664 | base = index[iterrev][3] | |
665 | while base != iterrev: |
|
665 | while base != iterrev: | |
666 | iterrev = base |
|
666 | iterrev = base | |
667 | base = index[iterrev][3] |
|
667 | base = index[iterrev][3] | |
668 |
|
668 | |||
669 | self._chainbasecache[rev] = base |
|
669 | self._chainbasecache[rev] = base | |
670 | return base |
|
670 | return base | |
671 |
|
671 | |||
672 | def linkrev(self, rev): |
|
672 | def linkrev(self, rev): | |
673 | return self.index[rev][4] |
|
673 | return self.index[rev][4] | |
674 |
|
674 | |||
675 | def parentrevs(self, rev): |
|
675 | def parentrevs(self, rev): | |
676 | try: |
|
676 | try: | |
677 | entry = self.index[rev] |
|
677 | entry = self.index[rev] | |
678 | except IndexError: |
|
678 | except IndexError: | |
679 | if rev == wdirrev: |
|
679 | if rev == wdirrev: | |
680 | raise error.WdirUnsupported |
|
680 | raise error.WdirUnsupported | |
681 | raise |
|
681 | raise | |
682 |
|
682 | |||
683 | return entry[5], entry[6] |
|
683 | return entry[5], entry[6] | |
684 |
|
684 | |||
685 | # fast parentrevs(rev) where rev isn't filtered |
|
685 | # fast parentrevs(rev) where rev isn't filtered | |
686 | _uncheckedparentrevs = parentrevs |
|
686 | _uncheckedparentrevs = parentrevs | |
687 |
|
687 | |||
688 | def node(self, rev): |
|
688 | def node(self, rev): | |
689 | try: |
|
689 | try: | |
690 | return self.index[rev][7] |
|
690 | return self.index[rev][7] | |
691 | except IndexError: |
|
691 | except IndexError: | |
692 | if rev == wdirrev: |
|
692 | if rev == wdirrev: | |
693 | raise error.WdirUnsupported |
|
693 | raise error.WdirUnsupported | |
694 | raise |
|
694 | raise | |
695 |
|
695 | |||
696 | # Derived from index values. |
|
696 | # Derived from index values. | |
697 |
|
697 | |||
698 | def end(self, rev): |
|
698 | def end(self, rev): | |
699 | return self.start(rev) + self.length(rev) |
|
699 | return self.start(rev) + self.length(rev) | |
700 |
|
700 | |||
701 | def parents(self, node): |
|
701 | def parents(self, node): | |
702 | i = self.index |
|
702 | i = self.index | |
703 | d = i[self.rev(node)] |
|
703 | d = i[self.rev(node)] | |
704 | return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline |
|
704 | return i[d[5]][7], i[d[6]][7] # map revisions to nodes inline | |
705 |
|
705 | |||
706 | def chainlen(self, rev): |
|
706 | def chainlen(self, rev): | |
707 | return self._chaininfo(rev)[0] |
|
707 | return self._chaininfo(rev)[0] | |
708 |
|
708 | |||
709 | def _chaininfo(self, rev): |
|
709 | def _chaininfo(self, rev): | |
710 | chaininfocache = self._chaininfocache |
|
710 | chaininfocache = self._chaininfocache | |
711 | if rev in chaininfocache: |
|
711 | if rev in chaininfocache: | |
712 | return chaininfocache[rev] |
|
712 | return chaininfocache[rev] | |
713 | index = self.index |
|
713 | index = self.index | |
714 | generaldelta = self._generaldelta |
|
714 | generaldelta = self._generaldelta | |
715 | iterrev = rev |
|
715 | iterrev = rev | |
716 | e = index[iterrev] |
|
716 | e = index[iterrev] | |
717 | clen = 0 |
|
717 | clen = 0 | |
718 | compresseddeltalen = 0 |
|
718 | compresseddeltalen = 0 | |
719 | while iterrev != e[3]: |
|
719 | while iterrev != e[3]: | |
720 | clen += 1 |
|
720 | clen += 1 | |
721 | compresseddeltalen += e[1] |
|
721 | compresseddeltalen += e[1] | |
722 | if generaldelta: |
|
722 | if generaldelta: | |
723 | iterrev = e[3] |
|
723 | iterrev = e[3] | |
724 | else: |
|
724 | else: | |
725 | iterrev -= 1 |
|
725 | iterrev -= 1 | |
726 | if iterrev in chaininfocache: |
|
726 | if iterrev in chaininfocache: | |
727 | t = chaininfocache[iterrev] |
|
727 | t = chaininfocache[iterrev] | |
728 | clen += t[0] |
|
728 | clen += t[0] | |
729 | compresseddeltalen += t[1] |
|
729 | compresseddeltalen += t[1] | |
730 | break |
|
730 | break | |
731 | e = index[iterrev] |
|
731 | e = index[iterrev] | |
732 | else: |
|
732 | else: | |
733 | # Add text length of base since decompressing that also takes |
|
733 | # Add text length of base since decompressing that also takes | |
734 | # work. For cache hits the length is already included. |
|
734 | # work. For cache hits the length is already included. | |
735 | compresseddeltalen += e[1] |
|
735 | compresseddeltalen += e[1] | |
736 | r = (clen, compresseddeltalen) |
|
736 | r = (clen, compresseddeltalen) | |
737 | chaininfocache[rev] = r |
|
737 | chaininfocache[rev] = r | |
738 | return r |
|
738 | return r | |
739 |
|
739 | |||
740 | def _deltachain(self, rev, stoprev=None): |
|
740 | def _deltachain(self, rev, stoprev=None): | |
741 | """Obtain the delta chain for a revision. |
|
741 | """Obtain the delta chain for a revision. | |
742 |
|
742 | |||
743 | ``stoprev`` specifies a revision to stop at. If not specified, we |
|
743 | ``stoprev`` specifies a revision to stop at. If not specified, we | |
744 | stop at the base of the chain. |
|
744 | stop at the base of the chain. | |
745 |
|
745 | |||
746 | Returns a 2-tuple of (chain, stopped) where ``chain`` is a list of |
|
746 | Returns a 2-tuple of (chain, stopped) where ``chain`` is a list of | |
747 | revs in ascending order and ``stopped`` is a bool indicating whether |
|
747 | revs in ascending order and ``stopped`` is a bool indicating whether | |
748 | ``stoprev`` was hit. |
|
748 | ``stoprev`` was hit. | |
749 | """ |
|
749 | """ | |
750 | # Try C implementation. |
|
750 | # Try C implementation. | |
751 | try: |
|
751 | try: | |
752 | return self.index.deltachain(rev, stoprev, self._generaldelta) |
|
752 | return self.index.deltachain(rev, stoprev, self._generaldelta) | |
753 | except AttributeError: |
|
753 | except AttributeError: | |
754 | pass |
|
754 | pass | |
755 |
|
755 | |||
756 | chain = [] |
|
756 | chain = [] | |
757 |
|
757 | |||
758 | # Alias to prevent attribute lookup in tight loop. |
|
758 | # Alias to prevent attribute lookup in tight loop. | |
759 | index = self.index |
|
759 | index = self.index | |
760 | generaldelta = self._generaldelta |
|
760 | generaldelta = self._generaldelta | |
761 |
|
761 | |||
762 | iterrev = rev |
|
762 | iterrev = rev | |
763 | e = index[iterrev] |
|
763 | e = index[iterrev] | |
764 | while iterrev != e[3] and iterrev != stoprev: |
|
764 | while iterrev != e[3] and iterrev != stoprev: | |
765 | chain.append(iterrev) |
|
765 | chain.append(iterrev) | |
766 | if generaldelta: |
|
766 | if generaldelta: | |
767 | iterrev = e[3] |
|
767 | iterrev = e[3] | |
768 | else: |
|
768 | else: | |
769 | iterrev -= 1 |
|
769 | iterrev -= 1 | |
770 | e = index[iterrev] |
|
770 | e = index[iterrev] | |
771 |
|
771 | |||
772 | if iterrev == stoprev: |
|
772 | if iterrev == stoprev: | |
773 | stopped = True |
|
773 | stopped = True | |
774 | else: |
|
774 | else: | |
775 | chain.append(iterrev) |
|
775 | chain.append(iterrev) | |
776 | stopped = False |
|
776 | stopped = False | |
777 |
|
777 | |||
778 | chain.reverse() |
|
778 | chain.reverse() | |
779 | return chain, stopped |
|
779 | return chain, stopped | |
780 |
|
780 | |||
781 | def ancestors(self, revs, stoprev=0, inclusive=False): |
|
781 | def ancestors(self, revs, stoprev=0, inclusive=False): | |
782 | """Generate the ancestors of 'revs' in reverse revision order. |
|
782 | """Generate the ancestors of 'revs' in reverse revision order. | |
783 | Does not generate revs lower than stoprev. |
|
783 | Does not generate revs lower than stoprev. | |
784 |
|
784 | |||
785 | See the documentation for ancestor.lazyancestors for more details.""" |
|
785 | See the documentation for ancestor.lazyancestors for more details.""" | |
786 |
|
786 | |||
787 | # first, make sure start revisions aren't filtered |
|
787 | # first, make sure start revisions aren't filtered | |
788 | revs = list(revs) |
|
788 | revs = list(revs) | |
789 | checkrev = self.node |
|
789 | checkrev = self.node | |
790 | for r in revs: |
|
790 | for r in revs: | |
791 | checkrev(r) |
|
791 | checkrev(r) | |
792 | # and we're sure ancestors aren't filtered as well |
|
792 | # and we're sure ancestors aren't filtered as well | |
793 |
|
793 | |||
794 | if rustancestor is not None: |
|
794 | if rustancestor is not None: | |
795 | lazyancestors = rustancestor.LazyAncestors |
|
795 | lazyancestors = rustancestor.LazyAncestors | |
796 | arg = self.index |
|
796 | arg = self.index | |
797 | elif util.safehasattr(parsers, 'rustlazyancestors'): |
|
797 | elif util.safehasattr(parsers, 'rustlazyancestors'): | |
798 | lazyancestors = ancestor.rustlazyancestors |
|
798 | lazyancestors = ancestor.rustlazyancestors | |
799 | arg = self.index |
|
799 | arg = self.index | |
800 | else: |
|
800 | else: | |
801 | lazyancestors = ancestor.lazyancestors |
|
801 | lazyancestors = ancestor.lazyancestors | |
802 | arg = self._uncheckedparentrevs |
|
802 | arg = self._uncheckedparentrevs | |
803 | return lazyancestors(arg, revs, stoprev=stoprev, inclusive=inclusive) |
|
803 | return lazyancestors(arg, revs, stoprev=stoprev, inclusive=inclusive) | |
804 |
|
804 | |||
805 | def descendants(self, revs): |
|
805 | def descendants(self, revs): | |
806 | return dagop.descendantrevs(revs, self.revs, self.parentrevs) |
|
806 | return dagop.descendantrevs(revs, self.revs, self.parentrevs) | |
807 |
|
807 | |||
808 | def findcommonmissing(self, common=None, heads=None): |
|
808 | def findcommonmissing(self, common=None, heads=None): | |
809 | """Return a tuple of the ancestors of common and the ancestors of heads |
|
809 | """Return a tuple of the ancestors of common and the ancestors of heads | |
810 | that are not ancestors of common. In revset terminology, we return the |
|
810 | that are not ancestors of common. In revset terminology, we return the | |
811 | tuple: |
|
811 | tuple: | |
812 |
|
812 | |||
813 | ::common, (::heads) - (::common) |
|
813 | ::common, (::heads) - (::common) | |
814 |
|
814 | |||
815 | The list is sorted by revision number, meaning it is |
|
815 | The list is sorted by revision number, meaning it is | |
816 | topologically sorted. |
|
816 | topologically sorted. | |
817 |
|
817 | |||
818 | 'heads' and 'common' are both lists of node IDs. If heads is |
|
818 | 'heads' and 'common' are both lists of node IDs. If heads is | |
819 | not supplied, uses all of the revlog's heads. If common is not |
|
819 | not supplied, uses all of the revlog's heads. If common is not | |
820 | supplied, uses nullid.""" |
|
820 | supplied, uses nullid.""" | |
821 | if common is None: |
|
821 | if common is None: | |
822 | common = [nullid] |
|
822 | common = [nullid] | |
823 | if heads is None: |
|
823 | if heads is None: | |
824 | heads = self.heads() |
|
824 | heads = self.heads() | |
825 |
|
825 | |||
826 | common = [self.rev(n) for n in common] |
|
826 | common = [self.rev(n) for n in common] | |
827 | heads = [self.rev(n) for n in heads] |
|
827 | heads = [self.rev(n) for n in heads] | |
828 |
|
828 | |||
829 | # we want the ancestors, but inclusive |
|
829 | # we want the ancestors, but inclusive | |
830 | class lazyset(object): |
|
830 | class lazyset(object): | |
831 | def __init__(self, lazyvalues): |
|
831 | def __init__(self, lazyvalues): | |
832 | self.addedvalues = set() |
|
832 | self.addedvalues = set() | |
833 | self.lazyvalues = lazyvalues |
|
833 | self.lazyvalues = lazyvalues | |
834 |
|
834 | |||
835 | def __contains__(self, value): |
|
835 | def __contains__(self, value): | |
836 | return value in self.addedvalues or value in self.lazyvalues |
|
836 | return value in self.addedvalues or value in self.lazyvalues | |
837 |
|
837 | |||
838 | def __iter__(self): |
|
838 | def __iter__(self): | |
839 | added = self.addedvalues |
|
839 | added = self.addedvalues | |
840 | for r in added: |
|
840 | for r in added: | |
841 | yield r |
|
841 | yield r | |
842 | for r in self.lazyvalues: |
|
842 | for r in self.lazyvalues: | |
843 | if not r in added: |
|
843 | if not r in added: | |
844 | yield r |
|
844 | yield r | |
845 |
|
845 | |||
846 | def add(self, value): |
|
846 | def add(self, value): | |
847 | self.addedvalues.add(value) |
|
847 | self.addedvalues.add(value) | |
848 |
|
848 | |||
849 | def update(self, values): |
|
849 | def update(self, values): | |
850 | self.addedvalues.update(values) |
|
850 | self.addedvalues.update(values) | |
851 |
|
851 | |||
852 | has = lazyset(self.ancestors(common)) |
|
852 | has = lazyset(self.ancestors(common)) | |
853 | has.add(nullrev) |
|
853 | has.add(nullrev) | |
854 | has.update(common) |
|
854 | has.update(common) | |
855 |
|
855 | |||
856 | # take all ancestors from heads that aren't in has |
|
856 | # take all ancestors from heads that aren't in has | |
857 | missing = set() |
|
857 | missing = set() | |
858 | visit = collections.deque(r for r in heads if r not in has) |
|
858 | visit = collections.deque(r for r in heads if r not in has) | |
859 | while visit: |
|
859 | while visit: | |
860 | r = visit.popleft() |
|
860 | r = visit.popleft() | |
861 | if r in missing: |
|
861 | if r in missing: | |
862 | continue |
|
862 | continue | |
863 | else: |
|
863 | else: | |
864 | missing.add(r) |
|
864 | missing.add(r) | |
865 | for p in self.parentrevs(r): |
|
865 | for p in self.parentrevs(r): | |
866 | if p not in has: |
|
866 | if p not in has: | |
867 | visit.append(p) |
|
867 | visit.append(p) | |
868 | missing = list(missing) |
|
868 | missing = list(missing) | |
869 | missing.sort() |
|
869 | missing.sort() | |
870 | return has, [self.node(miss) for miss in missing] |
|
870 | return has, [self.node(miss) for miss in missing] | |
871 |
|
871 | |||
872 | def incrementalmissingrevs(self, common=None): |
|
872 | def incrementalmissingrevs(self, common=None): | |
873 | """Return an object that can be used to incrementally compute the |
|
873 | """Return an object that can be used to incrementally compute the | |
874 | revision numbers of the ancestors of arbitrary sets that are not |
|
874 | revision numbers of the ancestors of arbitrary sets that are not | |
875 | ancestors of common. This is an ancestor.incrementalmissingancestors |
|
875 | ancestors of common. This is an ancestor.incrementalmissingancestors | |
876 | object. |
|
876 | object. | |
877 |
|
877 | |||
878 | 'common' is a list of revision numbers. If common is not supplied, uses |
|
878 | 'common' is a list of revision numbers. If common is not supplied, uses | |
879 | nullrev. |
|
879 | nullrev. | |
880 | """ |
|
880 | """ | |
881 | if common is None: |
|
881 | if common is None: | |
882 | common = [nullrev] |
|
882 | common = [nullrev] | |
883 |
|
883 | |||
884 | if rustancestor is not None: |
|
884 | if rustancestor is not None: | |
885 | return rustancestor.MissingAncestors(self.index, common) |
|
885 | return rustancestor.MissingAncestors(self.index, common) | |
886 | return ancestor.incrementalmissingancestors(self.parentrevs, common) |
|
886 | return ancestor.incrementalmissingancestors(self.parentrevs, common) | |
887 |
|
887 | |||
888 | def findmissingrevs(self, common=None, heads=None): |
|
888 | def findmissingrevs(self, common=None, heads=None): | |
889 | """Return the revision numbers of the ancestors of heads that |
|
889 | """Return the revision numbers of the ancestors of heads that | |
890 | are not ancestors of common. |
|
890 | are not ancestors of common. | |
891 |
|
891 | |||
892 | More specifically, return a list of revision numbers corresponding to |
|
892 | More specifically, return a list of revision numbers corresponding to | |
893 | nodes N such that every N satisfies the following constraints: |
|
893 | nodes N such that every N satisfies the following constraints: | |
894 |
|
894 | |||
895 | 1. N is an ancestor of some node in 'heads' |
|
895 | 1. N is an ancestor of some node in 'heads' | |
896 | 2. N is not an ancestor of any node in 'common' |
|
896 | 2. N is not an ancestor of any node in 'common' | |
897 |
|
897 | |||
898 | The list is sorted by revision number, meaning it is |
|
898 | The list is sorted by revision number, meaning it is | |
899 | topologically sorted. |
|
899 | topologically sorted. | |
900 |
|
900 | |||
901 | 'heads' and 'common' are both lists of revision numbers. If heads is |
|
901 | 'heads' and 'common' are both lists of revision numbers. If heads is | |
902 | not supplied, uses all of the revlog's heads. If common is not |
|
902 | not supplied, uses all of the revlog's heads. If common is not | |
903 | supplied, uses nullid.""" |
|
903 | supplied, uses nullid.""" | |
904 | if common is None: |
|
904 | if common is None: | |
905 | common = [nullrev] |
|
905 | common = [nullrev] | |
906 | if heads is None: |
|
906 | if heads is None: | |
907 | heads = self.headrevs() |
|
907 | heads = self.headrevs() | |
908 |
|
908 | |||
909 | inc = self.incrementalmissingrevs(common=common) |
|
909 | inc = self.incrementalmissingrevs(common=common) | |
910 | return inc.missingancestors(heads) |
|
910 | return inc.missingancestors(heads) | |
911 |
|
911 | |||
912 | def findmissing(self, common=None, heads=None): |
|
912 | def findmissing(self, common=None, heads=None): | |
913 | """Return the ancestors of heads that are not ancestors of common. |
|
913 | """Return the ancestors of heads that are not ancestors of common. | |
914 |
|
914 | |||
915 | More specifically, return a list of nodes N such that every N |
|
915 | More specifically, return a list of nodes N such that every N | |
916 | satisfies the following constraints: |
|
916 | satisfies the following constraints: | |
917 |
|
917 | |||
918 | 1. N is an ancestor of some node in 'heads' |
|
918 | 1. N is an ancestor of some node in 'heads' | |
919 | 2. N is not an ancestor of any node in 'common' |
|
919 | 2. N is not an ancestor of any node in 'common' | |
920 |
|
920 | |||
921 | The list is sorted by revision number, meaning it is |
|
921 | The list is sorted by revision number, meaning it is | |
922 | topologically sorted. |
|
922 | topologically sorted. | |
923 |
|
923 | |||
924 | 'heads' and 'common' are both lists of node IDs. If heads is |
|
924 | 'heads' and 'common' are both lists of node IDs. If heads is | |
925 | not supplied, uses all of the revlog's heads. If common is not |
|
925 | not supplied, uses all of the revlog's heads. If common is not | |
926 | supplied, uses nullid.""" |
|
926 | supplied, uses nullid.""" | |
927 | if common is None: |
|
927 | if common is None: | |
928 | common = [nullid] |
|
928 | common = [nullid] | |
929 | if heads is None: |
|
929 | if heads is None: | |
930 | heads = self.heads() |
|
930 | heads = self.heads() | |
931 |
|
931 | |||
932 | common = [self.rev(n) for n in common] |
|
932 | common = [self.rev(n) for n in common] | |
933 | heads = [self.rev(n) for n in heads] |
|
933 | heads = [self.rev(n) for n in heads] | |
934 |
|
934 | |||
935 | inc = self.incrementalmissingrevs(common=common) |
|
935 | inc = self.incrementalmissingrevs(common=common) | |
936 | return [self.node(r) for r in inc.missingancestors(heads)] |
|
936 | return [self.node(r) for r in inc.missingancestors(heads)] | |
937 |
|
937 | |||
938 | def nodesbetween(self, roots=None, heads=None): |
|
938 | def nodesbetween(self, roots=None, heads=None): | |
939 | """Return a topological path from 'roots' to 'heads'. |
|
939 | """Return a topological path from 'roots' to 'heads'. | |
940 |
|
940 | |||
941 | Return a tuple (nodes, outroots, outheads) where 'nodes' is a |
|
941 | Return a tuple (nodes, outroots, outheads) where 'nodes' is a | |
942 | topologically sorted list of all nodes N that satisfy both of |
|
942 | topologically sorted list of all nodes N that satisfy both of | |
943 | these constraints: |
|
943 | these constraints: | |
944 |
|
944 | |||
945 | 1. N is a descendant of some node in 'roots' |
|
945 | 1. N is a descendant of some node in 'roots' | |
946 | 2. N is an ancestor of some node in 'heads' |
|
946 | 2. N is an ancestor of some node in 'heads' | |
947 |
|
947 | |||
948 | Every node is considered to be both a descendant and an ancestor |
|
948 | Every node is considered to be both a descendant and an ancestor | |
949 | of itself, so every reachable node in 'roots' and 'heads' will be |
|
949 | of itself, so every reachable node in 'roots' and 'heads' will be | |
950 | included in 'nodes'. |
|
950 | included in 'nodes'. | |
951 |
|
951 | |||
952 | 'outroots' is the list of reachable nodes in 'roots', i.e., the |
|
952 | 'outroots' is the list of reachable nodes in 'roots', i.e., the | |
953 | subset of 'roots' that is returned in 'nodes'. Likewise, |
|
953 | subset of 'roots' that is returned in 'nodes'. Likewise, | |
954 | 'outheads' is the subset of 'heads' that is also in 'nodes'. |
|
954 | 'outheads' is the subset of 'heads' that is also in 'nodes'. | |
955 |
|
955 | |||
956 | 'roots' and 'heads' are both lists of node IDs. If 'roots' is |
|
956 | 'roots' and 'heads' are both lists of node IDs. If 'roots' is | |
957 | unspecified, uses nullid as the only root. If 'heads' is |
|
957 | unspecified, uses nullid as the only root. If 'heads' is | |
958 | unspecified, uses list of all of the revlog's heads.""" |
|
958 | unspecified, uses list of all of the revlog's heads.""" | |
959 | nonodes = ([], [], []) |
|
959 | nonodes = ([], [], []) | |
960 | if roots is not None: |
|
960 | if roots is not None: | |
961 | roots = list(roots) |
|
961 | roots = list(roots) | |
962 | if not roots: |
|
962 | if not roots: | |
963 | return nonodes |
|
963 | return nonodes | |
964 | lowestrev = min([self.rev(n) for n in roots]) |
|
964 | lowestrev = min([self.rev(n) for n in roots]) | |
965 | else: |
|
965 | else: | |
966 | roots = [nullid] # Everybody's a descendant of nullid |
|
966 | roots = [nullid] # Everybody's a descendant of nullid | |
967 | lowestrev = nullrev |
|
967 | lowestrev = nullrev | |
968 | if (lowestrev == nullrev) and (heads is None): |
|
968 | if (lowestrev == nullrev) and (heads is None): | |
969 | # We want _all_ the nodes! |
|
969 | # We want _all_ the nodes! | |
970 | return ([self.node(r) for r in self], [nullid], list(self.heads())) |
|
970 | return ([self.node(r) for r in self], [nullid], list(self.heads())) | |
971 | if heads is None: |
|
971 | if heads is None: | |
972 | # All nodes are ancestors, so the latest ancestor is the last |
|
972 | # All nodes are ancestors, so the latest ancestor is the last | |
973 | # node. |
|
973 | # node. | |
974 | highestrev = len(self) - 1 |
|
974 | highestrev = len(self) - 1 | |
975 | # Set ancestors to None to signal that every node is an ancestor. |
|
975 | # Set ancestors to None to signal that every node is an ancestor. | |
976 | ancestors = None |
|
976 | ancestors = None | |
977 | # Set heads to an empty dictionary for later discovery of heads |
|
977 | # Set heads to an empty dictionary for later discovery of heads | |
978 | heads = {} |
|
978 | heads = {} | |
979 | else: |
|
979 | else: | |
980 | heads = list(heads) |
|
980 | heads = list(heads) | |
981 | if not heads: |
|
981 | if not heads: | |
982 | return nonodes |
|
982 | return nonodes | |
983 | ancestors = set() |
|
983 | ancestors = set() | |
984 | # Turn heads into a dictionary so we can remove 'fake' heads. |
|
984 | # Turn heads into a dictionary so we can remove 'fake' heads. | |
985 | # Also, later we will be using it to filter out the heads we can't |
|
985 | # Also, later we will be using it to filter out the heads we can't | |
986 | # find from roots. |
|
986 | # find from roots. | |
987 | heads = dict.fromkeys(heads, False) |
|
987 | heads = dict.fromkeys(heads, False) | |
988 | # Start at the top and keep marking parents until we're done. |
|
988 | # Start at the top and keep marking parents until we're done. | |
989 | nodestotag = set(heads) |
|
989 | nodestotag = set(heads) | |
990 | # Remember where the top was so we can use it as a limit later. |
|
990 | # Remember where the top was so we can use it as a limit later. | |
991 | highestrev = max([self.rev(n) for n in nodestotag]) |
|
991 | highestrev = max([self.rev(n) for n in nodestotag]) | |
992 | while nodestotag: |
|
992 | while nodestotag: | |
993 | # grab a node to tag |
|
993 | # grab a node to tag | |
994 | n = nodestotag.pop() |
|
994 | n = nodestotag.pop() | |
995 | # Never tag nullid |
|
995 | # Never tag nullid | |
996 | if n == nullid: |
|
996 | if n == nullid: | |
997 | continue |
|
997 | continue | |
998 | # A node's revision number represents its place in a |
|
998 | # A node's revision number represents its place in a | |
999 | # topologically sorted list of nodes. |
|
999 | # topologically sorted list of nodes. | |
1000 | r = self.rev(n) |
|
1000 | r = self.rev(n) | |
1001 | if r >= lowestrev: |
|
1001 | if r >= lowestrev: | |
1002 | if n not in ancestors: |
|
1002 | if n not in ancestors: | |
1003 | # If we are possibly a descendant of one of the roots |
|
1003 | # If we are possibly a descendant of one of the roots | |
1004 | # and we haven't already been marked as an ancestor |
|
1004 | # and we haven't already been marked as an ancestor | |
1005 | ancestors.add(n) # Mark as ancestor |
|
1005 | ancestors.add(n) # Mark as ancestor | |
1006 | # Add non-nullid parents to list of nodes to tag. |
|
1006 | # Add non-nullid parents to list of nodes to tag. | |
1007 | nodestotag.update([p for p in self.parents(n) if |
|
1007 | nodestotag.update([p for p in self.parents(n) if | |
1008 | p != nullid]) |
|
1008 | p != nullid]) | |
1009 | elif n in heads: # We've seen it before, is it a fake head? |
|
1009 | elif n in heads: # We've seen it before, is it a fake head? | |
1010 | # So it is, real heads should not be the ancestors of |
|
1010 | # So it is, real heads should not be the ancestors of | |
1011 | # any other heads. |
|
1011 | # any other heads. | |
1012 | heads.pop(n) |
|
1012 | heads.pop(n) | |
1013 | if not ancestors: |
|
1013 | if not ancestors: | |
1014 | return nonodes |
|
1014 | return nonodes | |
1015 | # Now that we have our set of ancestors, we want to remove any |
|
1015 | # Now that we have our set of ancestors, we want to remove any | |
1016 | # roots that are not ancestors. |
|
1016 | # roots that are not ancestors. | |
1017 |
|
1017 | |||
1018 | # If one of the roots was nullid, everything is included anyway. |
|
1018 | # If one of the roots was nullid, everything is included anyway. | |
1019 | if lowestrev > nullrev: |
|
1019 | if lowestrev > nullrev: | |
1020 | # But, since we weren't, let's recompute the lowest rev to not |
|
1020 | # But, since we weren't, let's recompute the lowest rev to not | |
1021 | # include roots that aren't ancestors. |
|
1021 | # include roots that aren't ancestors. | |
1022 |
|
1022 | |||
1023 | # Filter out roots that aren't ancestors of heads |
|
1023 | # Filter out roots that aren't ancestors of heads | |
1024 | roots = [root for root in roots if root in ancestors] |
|
1024 | roots = [root for root in roots if root in ancestors] | |
1025 | # Recompute the lowest revision |
|
1025 | # Recompute the lowest revision | |
1026 | if roots: |
|
1026 | if roots: | |
1027 | lowestrev = min([self.rev(root) for root in roots]) |
|
1027 | lowestrev = min([self.rev(root) for root in roots]) | |
1028 | else: |
|
1028 | else: | |
1029 | # No more roots? Return empty list |
|
1029 | # No more roots? Return empty list | |
1030 | return nonodes |
|
1030 | return nonodes | |
1031 | else: |
|
1031 | else: | |
1032 | # We are descending from nullid, and don't need to care about |
|
1032 | # We are descending from nullid, and don't need to care about | |
1033 | # any other roots. |
|
1033 | # any other roots. | |
1034 | lowestrev = nullrev |
|
1034 | lowestrev = nullrev | |
1035 | roots = [nullid] |
|
1035 | roots = [nullid] | |
1036 | # Transform our roots list into a set. |
|
1036 | # Transform our roots list into a set. | |
1037 | descendants = set(roots) |
|
1037 | descendants = set(roots) | |
1038 | # Also, keep the original roots so we can filter out roots that aren't |
|
1038 | # Also, keep the original roots so we can filter out roots that aren't | |
1039 | # 'real' roots (i.e. are descended from other roots). |
|
1039 | # 'real' roots (i.e. are descended from other roots). | |
1040 | roots = descendants.copy() |
|
1040 | roots = descendants.copy() | |
1041 | # Our topologically sorted list of output nodes. |
|
1041 | # Our topologically sorted list of output nodes. | |
1042 | orderedout = [] |
|
1042 | orderedout = [] | |
1043 | # Don't start at nullid since we don't want nullid in our output list, |
|
1043 | # Don't start at nullid since we don't want nullid in our output list, | |
1044 | # and if nullid shows up in descendants, empty parents will look like |
|
1044 | # and if nullid shows up in descendants, empty parents will look like | |
1045 | # they're descendants. |
|
1045 | # they're descendants. | |
1046 | for r in self.revs(start=max(lowestrev, 0), stop=highestrev + 1): |
|
1046 | for r in self.revs(start=max(lowestrev, 0), stop=highestrev + 1): | |
1047 | n = self.node(r) |
|
1047 | n = self.node(r) | |
1048 | isdescendant = False |
|
1048 | isdescendant = False | |
1049 | if lowestrev == nullrev: # Everybody is a descendant of nullid |
|
1049 | if lowestrev == nullrev: # Everybody is a descendant of nullid | |
1050 | isdescendant = True |
|
1050 | isdescendant = True | |
1051 | elif n in descendants: |
|
1051 | elif n in descendants: | |
1052 | # n is already a descendant |
|
1052 | # n is already a descendant | |
1053 | isdescendant = True |
|
1053 | isdescendant = True | |
1054 | # This check only needs to be done here because all the roots |
|
1054 | # This check only needs to be done here because all the roots | |
1055 | # will start being marked is descendants before the loop. |
|
1055 | # will start being marked is descendants before the loop. | |
1056 | if n in roots: |
|
1056 | if n in roots: | |
1057 | # If n was a root, check if it's a 'real' root. |
|
1057 | # If n was a root, check if it's a 'real' root. | |
1058 | p = tuple(self.parents(n)) |
|
1058 | p = tuple(self.parents(n)) | |
1059 | # If any of its parents are descendants, it's not a root. |
|
1059 | # If any of its parents are descendants, it's not a root. | |
1060 | if (p[0] in descendants) or (p[1] in descendants): |
|
1060 | if (p[0] in descendants) or (p[1] in descendants): | |
1061 | roots.remove(n) |
|
1061 | roots.remove(n) | |
1062 | else: |
|
1062 | else: | |
1063 | p = tuple(self.parents(n)) |
|
1063 | p = tuple(self.parents(n)) | |
1064 | # A node is a descendant if either of its parents are |
|
1064 | # A node is a descendant if either of its parents are | |
1065 | # descendants. (We seeded the dependents list with the roots |
|
1065 | # descendants. (We seeded the dependents list with the roots | |
1066 | # up there, remember?) |
|
1066 | # up there, remember?) | |
1067 | if (p[0] in descendants) or (p[1] in descendants): |
|
1067 | if (p[0] in descendants) or (p[1] in descendants): | |
1068 | descendants.add(n) |
|
1068 | descendants.add(n) | |
1069 | isdescendant = True |
|
1069 | isdescendant = True | |
1070 | if isdescendant and ((ancestors is None) or (n in ancestors)): |
|
1070 | if isdescendant and ((ancestors is None) or (n in ancestors)): | |
1071 | # Only include nodes that are both descendants and ancestors. |
|
1071 | # Only include nodes that are both descendants and ancestors. | |
1072 | orderedout.append(n) |
|
1072 | orderedout.append(n) | |
1073 | if (ancestors is not None) and (n in heads): |
|
1073 | if (ancestors is not None) and (n in heads): | |
1074 | # We're trying to figure out which heads are reachable |
|
1074 | # We're trying to figure out which heads are reachable | |
1075 | # from roots. |
|
1075 | # from roots. | |
1076 | # Mark this head as having been reached |
|
1076 | # Mark this head as having been reached | |
1077 | heads[n] = True |
|
1077 | heads[n] = True | |
1078 | elif ancestors is None: |
|
1078 | elif ancestors is None: | |
1079 | # Otherwise, we're trying to discover the heads. |
|
1079 | # Otherwise, we're trying to discover the heads. | |
1080 | # Assume this is a head because if it isn't, the next step |
|
1080 | # Assume this is a head because if it isn't, the next step | |
1081 | # will eventually remove it. |
|
1081 | # will eventually remove it. | |
1082 | heads[n] = True |
|
1082 | heads[n] = True | |
1083 | # But, obviously its parents aren't. |
|
1083 | # But, obviously its parents aren't. | |
1084 | for p in self.parents(n): |
|
1084 | for p in self.parents(n): | |
1085 | heads.pop(p, None) |
|
1085 | heads.pop(p, None) | |
1086 | heads = [head for head, flag in heads.iteritems() if flag] |
|
1086 | heads = [head for head, flag in heads.iteritems() if flag] | |
1087 | roots = list(roots) |
|
1087 | roots = list(roots) | |
1088 | assert orderedout |
|
1088 | assert orderedout | |
1089 | assert roots |
|
1089 | assert roots | |
1090 | assert heads |
|
1090 | assert heads | |
1091 | return (orderedout, roots, heads) |
|
1091 | return (orderedout, roots, heads) | |
1092 |
|
1092 | |||
1093 | def headrevs(self, revs=None): |
|
1093 | def headrevs(self, revs=None): | |
1094 | if revs is None: |
|
1094 | if revs is None: | |
1095 | try: |
|
1095 | try: | |
1096 | return self.index.headrevs() |
|
1096 | return self.index.headrevs() | |
1097 | except AttributeError: |
|
1097 | except AttributeError: | |
1098 | return self._headrevs() |
|
1098 | return self._headrevs() | |
1099 | if rustdagop is not None: |
|
1099 | if rustdagop is not None: | |
1100 | return rustdagop.headrevs(self.index, revs) |
|
1100 | return rustdagop.headrevs(self.index, revs) | |
1101 | return dagop.headrevs(revs, self._uncheckedparentrevs) |
|
1101 | return dagop.headrevs(revs, self._uncheckedparentrevs) | |
1102 |
|
1102 | |||
1103 | def computephases(self, roots): |
|
1103 | def computephases(self, roots): | |
1104 | return self.index.computephasesmapsets(roots) |
|
1104 | return self.index.computephasesmapsets(roots) | |
1105 |
|
1105 | |||
1106 | def _headrevs(self): |
|
1106 | def _headrevs(self): | |
1107 | count = len(self) |
|
1107 | count = len(self) | |
1108 | if not count: |
|
1108 | if not count: | |
1109 | return [nullrev] |
|
1109 | return [nullrev] | |
1110 | # we won't iter over filtered rev so nobody is a head at start |
|
1110 | # we won't iter over filtered rev so nobody is a head at start | |
1111 | ishead = [0] * (count + 1) |
|
1111 | ishead = [0] * (count + 1) | |
1112 | index = self.index |
|
1112 | index = self.index | |
1113 | for r in self: |
|
1113 | for r in self: | |
1114 | ishead[r] = 1 # I may be an head |
|
1114 | ishead[r] = 1 # I may be an head | |
1115 | e = index[r] |
|
1115 | e = index[r] | |
1116 | ishead[e[5]] = ishead[e[6]] = 0 # my parent are not |
|
1116 | ishead[e[5]] = ishead[e[6]] = 0 # my parent are not | |
1117 | return [r for r, val in enumerate(ishead) if val] |
|
1117 | return [r for r, val in enumerate(ishead) if val] | |
1118 |
|
1118 | |||
1119 | def heads(self, start=None, stop=None): |
|
1119 | def heads(self, start=None, stop=None): | |
1120 | """return the list of all nodes that have no children |
|
1120 | """return the list of all nodes that have no children | |
1121 |
|
1121 | |||
1122 | if start is specified, only heads that are descendants of |
|
1122 | if start is specified, only heads that are descendants of | |
1123 | start will be returned |
|
1123 | start will be returned | |
1124 | if stop is specified, it will consider all the revs from stop |
|
1124 | if stop is specified, it will consider all the revs from stop | |
1125 | as if they had no children |
|
1125 | as if they had no children | |
1126 | """ |
|
1126 | """ | |
1127 | if start is None and stop is None: |
|
1127 | if start is None and stop is None: | |
1128 | if not len(self): |
|
1128 | if not len(self): | |
1129 | return [nullid] |
|
1129 | return [nullid] | |
1130 | return [self.node(r) for r in self.headrevs()] |
|
1130 | return [self.node(r) for r in self.headrevs()] | |
1131 |
|
1131 | |||
1132 | if start is None: |
|
1132 | if start is None: | |
1133 | start = nullrev |
|
1133 | start = nullrev | |
1134 | else: |
|
1134 | else: | |
1135 | start = self.rev(start) |
|
1135 | start = self.rev(start) | |
1136 |
|
1136 | |||
1137 | stoprevs = set(self.rev(n) for n in stop or []) |
|
1137 | stoprevs = set(self.rev(n) for n in stop or []) | |
1138 |
|
1138 | |||
1139 | revs = dagop.headrevssubset(self.revs, self.parentrevs, startrev=start, |
|
1139 | revs = dagop.headrevssubset(self.revs, self.parentrevs, startrev=start, | |
1140 | stoprevs=stoprevs) |
|
1140 | stoprevs=stoprevs) | |
1141 |
|
1141 | |||
1142 | return [self.node(rev) for rev in revs] |
|
1142 | return [self.node(rev) for rev in revs] | |
1143 |
|
1143 | |||
1144 | def children(self, node): |
|
1144 | def children(self, node): | |
1145 | """find the children of a given node""" |
|
1145 | """find the children of a given node""" | |
1146 | c = [] |
|
1146 | c = [] | |
1147 | p = self.rev(node) |
|
1147 | p = self.rev(node) | |
1148 | for r in self.revs(start=p + 1): |
|
1148 | for r in self.revs(start=p + 1): | |
1149 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] |
|
1149 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] | |
1150 | if prevs: |
|
1150 | if prevs: | |
1151 | for pr in prevs: |
|
1151 | for pr in prevs: | |
1152 | if pr == p: |
|
1152 | if pr == p: | |
1153 | c.append(self.node(r)) |
|
1153 | c.append(self.node(r)) | |
1154 | elif p == nullrev: |
|
1154 | elif p == nullrev: | |
1155 | c.append(self.node(r)) |
|
1155 | c.append(self.node(r)) | |
1156 | return c |
|
1156 | return c | |
1157 |
|
1157 | |||
1158 | def commonancestorsheads(self, a, b): |
|
1158 | def commonancestorsheads(self, a, b): | |
1159 | """calculate all the heads of the common ancestors of nodes a and b""" |
|
1159 | """calculate all the heads of the common ancestors of nodes a and b""" | |
1160 | a, b = self.rev(a), self.rev(b) |
|
1160 | a, b = self.rev(a), self.rev(b) | |
1161 | ancs = self._commonancestorsheads(a, b) |
|
1161 | ancs = self._commonancestorsheads(a, b) | |
1162 | return pycompat.maplist(self.node, ancs) |
|
1162 | return pycompat.maplist(self.node, ancs) | |
1163 |
|
1163 | |||
1164 | def _commonancestorsheads(self, *revs): |
|
1164 | def _commonancestorsheads(self, *revs): | |
1165 | """calculate all the heads of the common ancestors of revs""" |
|
1165 | """calculate all the heads of the common ancestors of revs""" | |
1166 | try: |
|
1166 | try: | |
1167 | ancs = self.index.commonancestorsheads(*revs) |
|
1167 | ancs = self.index.commonancestorsheads(*revs) | |
1168 | except (AttributeError, OverflowError): # C implementation failed |
|
1168 | except (AttributeError, OverflowError): # C implementation failed | |
1169 | ancs = ancestor.commonancestorsheads(self.parentrevs, *revs) |
|
1169 | ancs = ancestor.commonancestorsheads(self.parentrevs, *revs) | |
1170 | return ancs |
|
1170 | return ancs | |
1171 |
|
1171 | |||
1172 | def isancestor(self, a, b): |
|
1172 | def isancestor(self, a, b): | |
1173 | """return True if node a is an ancestor of node b |
|
1173 | """return True if node a is an ancestor of node b | |
1174 |
|
1174 | |||
1175 | A revision is considered an ancestor of itself.""" |
|
1175 | A revision is considered an ancestor of itself.""" | |
1176 | a, b = self.rev(a), self.rev(b) |
|
1176 | a, b = self.rev(a), self.rev(b) | |
1177 | return self.isancestorrev(a, b) |
|
1177 | return self.isancestorrev(a, b) | |
1178 |
|
1178 | |||
1179 | def isancestorrev(self, a, b): |
|
1179 | def isancestorrev(self, a, b): | |
1180 | """return True if revision a is an ancestor of revision b |
|
1180 | """return True if revision a is an ancestor of revision b | |
1181 |
|
1181 | |||
1182 | A revision is considered an ancestor of itself. |
|
1182 | A revision is considered an ancestor of itself. | |
1183 |
|
1183 | |||
1184 | The implementation of this is trivial but the use of |
|
1184 | The implementation of this is trivial but the use of | |
1185 | reachableroots is not.""" |
|
1185 | reachableroots is not.""" | |
1186 | if a == nullrev: |
|
1186 | if a == nullrev: | |
1187 | return True |
|
1187 | return True | |
1188 | elif a == b: |
|
1188 | elif a == b: | |
1189 | return True |
|
1189 | return True | |
1190 | elif a > b: |
|
1190 | elif a > b: | |
1191 | return False |
|
1191 | return False | |
1192 | return bool(self.reachableroots(a, [b], [a], includepath=False)) |
|
1192 | return bool(self.reachableroots(a, [b], [a], includepath=False)) | |
1193 |
|
1193 | |||
1194 | def reachableroots(self, minroot, heads, roots, includepath=False): |
|
1194 | def reachableroots(self, minroot, heads, roots, includepath=False): | |
1195 | """return (heads(::<roots> and <roots>::<heads>)) |
|
1195 | """return (heads(::<roots> and <roots>::<heads>)) | |
1196 |
|
1196 | |||
1197 | If includepath is True, return (<roots>::<heads>).""" |
|
1197 | If includepath is True, return (<roots>::<heads>).""" | |
1198 | try: |
|
1198 | try: | |
1199 | return self.index.reachableroots2(minroot, heads, roots, |
|
1199 | return self.index.reachableroots2(minroot, heads, roots, | |
1200 | includepath) |
|
1200 | includepath) | |
1201 | except AttributeError: |
|
1201 | except AttributeError: | |
1202 | return dagop._reachablerootspure(self.parentrevs, |
|
1202 | return dagop._reachablerootspure(self.parentrevs, | |
1203 | minroot, roots, heads, includepath) |
|
1203 | minroot, roots, heads, includepath) | |
1204 |
|
1204 | |||
1205 | def ancestor(self, a, b): |
|
1205 | def ancestor(self, a, b): | |
1206 | """calculate the "best" common ancestor of nodes a and b""" |
|
1206 | """calculate the "best" common ancestor of nodes a and b""" | |
1207 |
|
1207 | |||
1208 | a, b = self.rev(a), self.rev(b) |
|
1208 | a, b = self.rev(a), self.rev(b) | |
1209 | try: |
|
1209 | try: | |
1210 | ancs = self.index.ancestors(a, b) |
|
1210 | ancs = self.index.ancestors(a, b) | |
1211 | except (AttributeError, OverflowError): |
|
1211 | except (AttributeError, OverflowError): | |
1212 | ancs = ancestor.ancestors(self.parentrevs, a, b) |
|
1212 | ancs = ancestor.ancestors(self.parentrevs, a, b) | |
1213 | if ancs: |
|
1213 | if ancs: | |
1214 | # choose a consistent winner when there's a tie |
|
1214 | # choose a consistent winner when there's a tie | |
1215 | return min(map(self.node, ancs)) |
|
1215 | return min(map(self.node, ancs)) | |
1216 | return nullid |
|
1216 | return nullid | |
1217 |
|
1217 | |||
1218 | def _match(self, id): |
|
1218 | def _match(self, id): | |
1219 | if isinstance(id, int): |
|
1219 | if isinstance(id, int): | |
1220 | # rev |
|
1220 | # rev | |
1221 | return self.node(id) |
|
1221 | return self.node(id) | |
1222 | if len(id) == 20: |
|
1222 | if len(id) == 20: | |
1223 | # possibly a binary node |
|
1223 | # possibly a binary node | |
1224 | # odds of a binary node being all hex in ASCII are 1 in 10**25 |
|
1224 | # odds of a binary node being all hex in ASCII are 1 in 10**25 | |
1225 | try: |
|
1225 | try: | |
1226 | node = id |
|
1226 | node = id | |
1227 | self.rev(node) # quick search the index |
|
1227 | self.rev(node) # quick search the index | |
1228 | return node |
|
1228 | return node | |
1229 | except error.LookupError: |
|
1229 | except error.LookupError: | |
1230 | pass # may be partial hex id |
|
1230 | pass # may be partial hex id | |
1231 | try: |
|
1231 | try: | |
1232 | # str(rev) |
|
1232 | # str(rev) | |
1233 | rev = int(id) |
|
1233 | rev = int(id) | |
1234 | if "%d" % rev != id: |
|
1234 | if "%d" % rev != id: | |
1235 | raise ValueError |
|
1235 | raise ValueError | |
1236 | if rev < 0: |
|
1236 | if rev < 0: | |
1237 | rev = len(self) + rev |
|
1237 | rev = len(self) + rev | |
1238 | if rev < 0 or rev >= len(self): |
|
1238 | if rev < 0 or rev >= len(self): | |
1239 | raise ValueError |
|
1239 | raise ValueError | |
1240 | return self.node(rev) |
|
1240 | return self.node(rev) | |
1241 | except (ValueError, OverflowError): |
|
1241 | except (ValueError, OverflowError): | |
1242 | pass |
|
1242 | pass | |
1243 | if len(id) == 40: |
|
1243 | if len(id) == 40: | |
1244 | try: |
|
1244 | try: | |
1245 | # a full hex nodeid? |
|
1245 | # a full hex nodeid? | |
1246 | node = bin(id) |
|
1246 | node = bin(id) | |
1247 | self.rev(node) |
|
1247 | self.rev(node) | |
1248 | return node |
|
1248 | return node | |
1249 | except (TypeError, error.LookupError): |
|
1249 | except (TypeError, error.LookupError): | |
1250 | pass |
|
1250 | pass | |
1251 |
|
1251 | |||
1252 | def _partialmatch(self, id): |
|
1252 | def _partialmatch(self, id): | |
1253 | # we don't care wdirfilenodeids as they should be always full hash |
|
1253 | # we don't care wdirfilenodeids as they should be always full hash | |
1254 | maybewdir = wdirhex.startswith(id) |
|
1254 | maybewdir = wdirhex.startswith(id) | |
1255 | try: |
|
1255 | try: | |
1256 | partial = self.index.partialmatch(id) |
|
1256 | partial = self.index.partialmatch(id) | |
1257 | if partial and self.hasnode(partial): |
|
1257 | if partial and self.hasnode(partial): | |
1258 | if maybewdir: |
|
1258 | if maybewdir: | |
1259 | # single 'ff...' match in radix tree, ambiguous with wdir |
|
1259 | # single 'ff...' match in radix tree, ambiguous with wdir | |
1260 | raise error.RevlogError |
|
1260 | raise error.RevlogError | |
1261 | return partial |
|
1261 | return partial | |
1262 | if maybewdir: |
|
1262 | if maybewdir: | |
1263 | # no 'ff...' match in radix tree, wdir identified |
|
1263 | # no 'ff...' match in radix tree, wdir identified | |
1264 | raise error.WdirUnsupported |
|
1264 | raise error.WdirUnsupported | |
1265 | return None |
|
1265 | return None | |
1266 | except error.RevlogError: |
|
1266 | except error.RevlogError: | |
1267 | # parsers.c radix tree lookup gave multiple matches |
|
1267 | # parsers.c radix tree lookup gave multiple matches | |
1268 | # fast path: for unfiltered changelog, radix tree is accurate |
|
1268 | # fast path: for unfiltered changelog, radix tree is accurate | |
1269 | if not getattr(self, 'filteredrevs', None): |
|
1269 | if not getattr(self, 'filteredrevs', None): | |
1270 | raise error.AmbiguousPrefixLookupError( |
|
1270 | raise error.AmbiguousPrefixLookupError( | |
1271 | id, self.indexfile, _('ambiguous identifier')) |
|
1271 | id, self.indexfile, _('ambiguous identifier')) | |
1272 | # fall through to slow path that filters hidden revisions |
|
1272 | # fall through to slow path that filters hidden revisions | |
1273 | except (AttributeError, ValueError): |
|
1273 | except (AttributeError, ValueError): | |
1274 | # we are pure python, or key was too short to search radix tree |
|
1274 | # we are pure python, or key was too short to search radix tree | |
1275 | pass |
|
1275 | pass | |
1276 |
|
1276 | |||
1277 | if id in self._pcache: |
|
1277 | if id in self._pcache: | |
1278 | return self._pcache[id] |
|
1278 | return self._pcache[id] | |
1279 |
|
1279 | |||
1280 | if len(id) <= 40: |
|
1280 | if len(id) <= 40: | |
1281 | try: |
|
1281 | try: | |
1282 | # hex(node)[:...] |
|
1282 | # hex(node)[:...] | |
1283 | l = len(id) // 2 # grab an even number of digits |
|
1283 | l = len(id) // 2 # grab an even number of digits | |
1284 | prefix = bin(id[:l * 2]) |
|
1284 | prefix = bin(id[:l * 2]) | |
1285 | nl = [e[7] for e in self.index if e[7].startswith(prefix)] |
|
1285 | nl = [e[7] for e in self.index if e[7].startswith(prefix)] | |
1286 | nl = [n for n in nl if hex(n).startswith(id) and |
|
1286 | nl = [n for n in nl if hex(n).startswith(id) and | |
1287 | self.hasnode(n)] |
|
1287 | self.hasnode(n)] | |
1288 | if nullhex.startswith(id): |
|
1288 | if nullhex.startswith(id): | |
1289 | nl.append(nullid) |
|
1289 | nl.append(nullid) | |
1290 | if len(nl) > 0: |
|
1290 | if len(nl) > 0: | |
1291 | if len(nl) == 1 and not maybewdir: |
|
1291 | if len(nl) == 1 and not maybewdir: | |
1292 | self._pcache[id] = nl[0] |
|
1292 | self._pcache[id] = nl[0] | |
1293 | return nl[0] |
|
1293 | return nl[0] | |
1294 | raise error.AmbiguousPrefixLookupError( |
|
1294 | raise error.AmbiguousPrefixLookupError( | |
1295 | id, self.indexfile, _('ambiguous identifier')) |
|
1295 | id, self.indexfile, _('ambiguous identifier')) | |
1296 | if maybewdir: |
|
1296 | if maybewdir: | |
1297 | raise error.WdirUnsupported |
|
1297 | raise error.WdirUnsupported | |
1298 | return None |
|
1298 | return None | |
1299 | except TypeError: |
|
1299 | except TypeError: | |
1300 | pass |
|
1300 | pass | |
1301 |
|
1301 | |||
1302 | def lookup(self, id): |
|
1302 | def lookup(self, id): | |
1303 | """locate a node based on: |
|
1303 | """locate a node based on: | |
1304 | - revision number or str(revision number) |
|
1304 | - revision number or str(revision number) | |
1305 | - nodeid or subset of hex nodeid |
|
1305 | - nodeid or subset of hex nodeid | |
1306 | """ |
|
1306 | """ | |
1307 | n = self._match(id) |
|
1307 | n = self._match(id) | |
1308 | if n is not None: |
|
1308 | if n is not None: | |
1309 | return n |
|
1309 | return n | |
1310 | n = self._partialmatch(id) |
|
1310 | n = self._partialmatch(id) | |
1311 | if n: |
|
1311 | if n: | |
1312 | return n |
|
1312 | return n | |
1313 |
|
1313 | |||
1314 | raise error.LookupError(id, self.indexfile, _('no match found')) |
|
1314 | raise error.LookupError(id, self.indexfile, _('no match found')) | |
1315 |
|
1315 | |||
1316 | def shortest(self, node, minlength=1): |
|
1316 | def shortest(self, node, minlength=1): | |
1317 | """Find the shortest unambiguous prefix that matches node.""" |
|
1317 | """Find the shortest unambiguous prefix that matches node.""" | |
1318 | def isvalid(prefix): |
|
1318 | def isvalid(prefix): | |
1319 | try: |
|
1319 | try: | |
1320 | matchednode = self._partialmatch(prefix) |
|
1320 | matchednode = self._partialmatch(prefix) | |
1321 | except error.AmbiguousPrefixLookupError: |
|
1321 | except error.AmbiguousPrefixLookupError: | |
1322 | return False |
|
1322 | return False | |
1323 | except error.WdirUnsupported: |
|
1323 | except error.WdirUnsupported: | |
1324 | # single 'ff...' match |
|
1324 | # single 'ff...' match | |
1325 | return True |
|
1325 | return True | |
1326 | if matchednode is None: |
|
1326 | if matchednode is None: | |
1327 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
1327 | raise error.LookupError(node, self.indexfile, _('no node')) | |
1328 | return True |
|
1328 | return True | |
1329 |
|
1329 | |||
1330 | def maybewdir(prefix): |
|
1330 | def maybewdir(prefix): | |
1331 | return all(c == 'f' for c in pycompat.iterbytestr(prefix)) |
|
1331 | return all(c == 'f' for c in pycompat.iterbytestr(prefix)) | |
1332 |
|
1332 | |||
1333 | hexnode = hex(node) |
|
1333 | hexnode = hex(node) | |
1334 |
|
1334 | |||
1335 | def disambiguate(hexnode, minlength): |
|
1335 | def disambiguate(hexnode, minlength): | |
1336 | """Disambiguate against wdirid.""" |
|
1336 | """Disambiguate against wdirid.""" | |
1337 | for length in range(minlength, 41): |
|
1337 | for length in range(minlength, 41): | |
1338 | prefix = hexnode[:length] |
|
1338 | prefix = hexnode[:length] | |
1339 | if not maybewdir(prefix): |
|
1339 | if not maybewdir(prefix): | |
1340 | return prefix |
|
1340 | return prefix | |
1341 |
|
1341 | |||
1342 | if not getattr(self, 'filteredrevs', None): |
|
1342 | if not getattr(self, 'filteredrevs', None): | |
1343 | try: |
|
1343 | try: | |
1344 | length = max(self.index.shortest(node), minlength) |
|
1344 | length = max(self.index.shortest(node), minlength) | |
1345 | return disambiguate(hexnode, length) |
|
1345 | return disambiguate(hexnode, length) | |
1346 | except error.RevlogError: |
|
1346 | except error.RevlogError: | |
1347 | if node != wdirid: |
|
1347 | if node != wdirid: | |
1348 | raise error.LookupError(node, self.indexfile, _('no node')) |
|
1348 | raise error.LookupError(node, self.indexfile, _('no node')) | |
1349 | except AttributeError: |
|
1349 | except AttributeError: | |
1350 | # Fall through to pure code |
|
1350 | # Fall through to pure code | |
1351 | pass |
|
1351 | pass | |
1352 |
|
1352 | |||
1353 | if node == wdirid: |
|
1353 | if node == wdirid: | |
1354 | for length in range(minlength, 41): |
|
1354 | for length in range(minlength, 41): | |
1355 | prefix = hexnode[:length] |
|
1355 | prefix = hexnode[:length] | |
1356 | if isvalid(prefix): |
|
1356 | if isvalid(prefix): | |
1357 | return prefix |
|
1357 | return prefix | |
1358 |
|
1358 | |||
1359 | for length in range(minlength, 41): |
|
1359 | for length in range(minlength, 41): | |
1360 | prefix = hexnode[:length] |
|
1360 | prefix = hexnode[:length] | |
1361 | if isvalid(prefix): |
|
1361 | if isvalid(prefix): | |
1362 | return disambiguate(hexnode, length) |
|
1362 | return disambiguate(hexnode, length) | |
1363 |
|
1363 | |||
1364 | def cmp(self, node, text): |
|
1364 | def cmp(self, node, text): | |
1365 | """compare text with a given file revision |
|
1365 | """compare text with a given file revision | |
1366 |
|
1366 | |||
1367 | returns True if text is different than what is stored. |
|
1367 | returns True if text is different than what is stored. | |
1368 | """ |
|
1368 | """ | |
1369 | p1, p2 = self.parents(node) |
|
1369 | p1, p2 = self.parents(node) | |
1370 | return storageutil.hashrevisionsha1(text, p1, p2) != node |
|
1370 | return storageutil.hashrevisionsha1(text, p1, p2) != node | |
1371 |
|
1371 | |||
1372 | def _cachesegment(self, offset, data): |
|
1372 | def _cachesegment(self, offset, data): | |
1373 | """Add a segment to the revlog cache. |
|
1373 | """Add a segment to the revlog cache. | |
1374 |
|
1374 | |||
1375 | Accepts an absolute offset and the data that is at that location. |
|
1375 | Accepts an absolute offset and the data that is at that location. | |
1376 | """ |
|
1376 | """ | |
1377 | o, d = self._chunkcache |
|
1377 | o, d = self._chunkcache | |
1378 | # try to add to existing cache |
|
1378 | # try to add to existing cache | |
1379 | if o + len(d) == offset and len(d) + len(data) < _chunksize: |
|
1379 | if o + len(d) == offset and len(d) + len(data) < _chunksize: | |
1380 | self._chunkcache = o, d + data |
|
1380 | self._chunkcache = o, d + data | |
1381 | else: |
|
1381 | else: | |
1382 | self._chunkcache = offset, data |
|
1382 | self._chunkcache = offset, data | |
1383 |
|
1383 | |||
1384 | def _readsegment(self, offset, length, df=None): |
|
1384 | def _readsegment(self, offset, length, df=None): | |
1385 | """Load a segment of raw data from the revlog. |
|
1385 | """Load a segment of raw data from the revlog. | |
1386 |
|
1386 | |||
1387 | Accepts an absolute offset, length to read, and an optional existing |
|
1387 | Accepts an absolute offset, length to read, and an optional existing | |
1388 | file handle to read from. |
|
1388 | file handle to read from. | |
1389 |
|
1389 | |||
1390 | If an existing file handle is passed, it will be seeked and the |
|
1390 | If an existing file handle is passed, it will be seeked and the | |
1391 | original seek position will NOT be restored. |
|
1391 | original seek position will NOT be restored. | |
1392 |
|
1392 | |||
1393 | Returns a str or buffer of raw byte data. |
|
1393 | Returns a str or buffer of raw byte data. | |
1394 |
|
1394 | |||
1395 | Raises if the requested number of bytes could not be read. |
|
1395 | Raises if the requested number of bytes could not be read. | |
1396 | """ |
|
1396 | """ | |
1397 | # Cache data both forward and backward around the requested |
|
1397 | # Cache data both forward and backward around the requested | |
1398 | # data, in a fixed size window. This helps speed up operations |
|
1398 | # data, in a fixed size window. This helps speed up operations | |
1399 | # involving reading the revlog backwards. |
|
1399 | # involving reading the revlog backwards. | |
1400 | cachesize = self._chunkcachesize |
|
1400 | cachesize = self._chunkcachesize | |
1401 | realoffset = offset & ~(cachesize - 1) |
|
1401 | realoffset = offset & ~(cachesize - 1) | |
1402 | reallength = (((offset + length + cachesize) & ~(cachesize - 1)) |
|
1402 | reallength = (((offset + length + cachesize) & ~(cachesize - 1)) | |
1403 | - realoffset) |
|
1403 | - realoffset) | |
1404 | with self._datareadfp(df) as df: |
|
1404 | with self._datareadfp(df) as df: | |
1405 | df.seek(realoffset) |
|
1405 | df.seek(realoffset) | |
1406 | d = df.read(reallength) |
|
1406 | d = df.read(reallength) | |
1407 |
|
1407 | |||
1408 | self._cachesegment(realoffset, d) |
|
1408 | self._cachesegment(realoffset, d) | |
1409 | if offset != realoffset or reallength != length: |
|
1409 | if offset != realoffset or reallength != length: | |
1410 | startoffset = offset - realoffset |
|
1410 | startoffset = offset - realoffset | |
1411 | if len(d) - startoffset < length: |
|
1411 | if len(d) - startoffset < length: | |
1412 | raise error.RevlogError( |
|
1412 | raise error.RevlogError( | |
1413 | _('partial read of revlog %s; expected %d bytes from ' |
|
1413 | _('partial read of revlog %s; expected %d bytes from ' | |
1414 | 'offset %d, got %d') % |
|
1414 | 'offset %d, got %d') % | |
1415 | (self.indexfile if self._inline else self.datafile, |
|
1415 | (self.indexfile if self._inline else self.datafile, | |
1416 | length, realoffset, len(d) - startoffset)) |
|
1416 | length, realoffset, len(d) - startoffset)) | |
1417 |
|
1417 | |||
1418 | return util.buffer(d, startoffset, length) |
|
1418 | return util.buffer(d, startoffset, length) | |
1419 |
|
1419 | |||
1420 | if len(d) < length: |
|
1420 | if len(d) < length: | |
1421 | raise error.RevlogError( |
|
1421 | raise error.RevlogError( | |
1422 | _('partial read of revlog %s; expected %d bytes from offset ' |
|
1422 | _('partial read of revlog %s; expected %d bytes from offset ' | |
1423 | '%d, got %d') % |
|
1423 | '%d, got %d') % | |
1424 | (self.indexfile if self._inline else self.datafile, |
|
1424 | (self.indexfile if self._inline else self.datafile, | |
1425 | length, offset, len(d))) |
|
1425 | length, offset, len(d))) | |
1426 |
|
1426 | |||
1427 | return d |
|
1427 | return d | |
1428 |
|
1428 | |||
1429 | def _getsegment(self, offset, length, df=None): |
|
1429 | def _getsegment(self, offset, length, df=None): | |
1430 | """Obtain a segment of raw data from the revlog. |
|
1430 | """Obtain a segment of raw data from the revlog. | |
1431 |
|
1431 | |||
1432 | Accepts an absolute offset, length of bytes to obtain, and an |
|
1432 | Accepts an absolute offset, length of bytes to obtain, and an | |
1433 | optional file handle to the already-opened revlog. If the file |
|
1433 | optional file handle to the already-opened revlog. If the file | |
1434 | handle is used, it's original seek position will not be preserved. |
|
1434 | handle is used, it's original seek position will not be preserved. | |
1435 |
|
1435 | |||
1436 | Requests for data may be returned from a cache. |
|
1436 | Requests for data may be returned from a cache. | |
1437 |
|
1437 | |||
1438 | Returns a str or a buffer instance of raw byte data. |
|
1438 | Returns a str or a buffer instance of raw byte data. | |
1439 | """ |
|
1439 | """ | |
1440 | o, d = self._chunkcache |
|
1440 | o, d = self._chunkcache | |
1441 | l = len(d) |
|
1441 | l = len(d) | |
1442 |
|
1442 | |||
1443 | # is it in the cache? |
|
1443 | # is it in the cache? | |
1444 | cachestart = offset - o |
|
1444 | cachestart = offset - o | |
1445 | cacheend = cachestart + length |
|
1445 | cacheend = cachestart + length | |
1446 | if cachestart >= 0 and cacheend <= l: |
|
1446 | if cachestart >= 0 and cacheend <= l: | |
1447 | if cachestart == 0 and cacheend == l: |
|
1447 | if cachestart == 0 and cacheend == l: | |
1448 | return d # avoid a copy |
|
1448 | return d # avoid a copy | |
1449 | return util.buffer(d, cachestart, cacheend - cachestart) |
|
1449 | return util.buffer(d, cachestart, cacheend - cachestart) | |
1450 |
|
1450 | |||
1451 | return self._readsegment(offset, length, df=df) |
|
1451 | return self._readsegment(offset, length, df=df) | |
1452 |
|
1452 | |||
1453 | def _getsegmentforrevs(self, startrev, endrev, df=None): |
|
1453 | def _getsegmentforrevs(self, startrev, endrev, df=None): | |
1454 | """Obtain a segment of raw data corresponding to a range of revisions. |
|
1454 | """Obtain a segment of raw data corresponding to a range of revisions. | |
1455 |
|
1455 | |||
1456 | Accepts the start and end revisions and an optional already-open |
|
1456 | Accepts the start and end revisions and an optional already-open | |
1457 | file handle to be used for reading. If the file handle is read, its |
|
1457 | file handle to be used for reading. If the file handle is read, its | |
1458 | seek position will not be preserved. |
|
1458 | seek position will not be preserved. | |
1459 |
|
1459 | |||
1460 | Requests for data may be satisfied by a cache. |
|
1460 | Requests for data may be satisfied by a cache. | |
1461 |
|
1461 | |||
1462 | Returns a 2-tuple of (offset, data) for the requested range of |
|
1462 | Returns a 2-tuple of (offset, data) for the requested range of | |
1463 | revisions. Offset is the integer offset from the beginning of the |
|
1463 | revisions. Offset is the integer offset from the beginning of the | |
1464 | revlog and data is a str or buffer of the raw byte data. |
|
1464 | revlog and data is a str or buffer of the raw byte data. | |
1465 |
|
1465 | |||
1466 | Callers will need to call ``self.start(rev)`` and ``self.length(rev)`` |
|
1466 | Callers will need to call ``self.start(rev)`` and ``self.length(rev)`` | |
1467 | to determine where each revision's data begins and ends. |
|
1467 | to determine where each revision's data begins and ends. | |
1468 | """ |
|
1468 | """ | |
1469 | # Inlined self.start(startrev) & self.end(endrev) for perf reasons |
|
1469 | # Inlined self.start(startrev) & self.end(endrev) for perf reasons | |
1470 | # (functions are expensive). |
|
1470 | # (functions are expensive). | |
1471 | index = self.index |
|
1471 | index = self.index | |
1472 | istart = index[startrev] |
|
1472 | istart = index[startrev] | |
1473 | start = int(istart[0] >> 16) |
|
1473 | start = int(istart[0] >> 16) | |
1474 | if startrev == endrev: |
|
1474 | if startrev == endrev: | |
1475 | end = start + istart[1] |
|
1475 | end = start + istart[1] | |
1476 | else: |
|
1476 | else: | |
1477 | iend = index[endrev] |
|
1477 | iend = index[endrev] | |
1478 | end = int(iend[0] >> 16) + iend[1] |
|
1478 | end = int(iend[0] >> 16) + iend[1] | |
1479 |
|
1479 | |||
1480 | if self._inline: |
|
1480 | if self._inline: | |
1481 | start += (startrev + 1) * self._io.size |
|
1481 | start += (startrev + 1) * self._io.size | |
1482 | end += (endrev + 1) * self._io.size |
|
1482 | end += (endrev + 1) * self._io.size | |
1483 | length = end - start |
|
1483 | length = end - start | |
1484 |
|
1484 | |||
1485 | return start, self._getsegment(start, length, df=df) |
|
1485 | return start, self._getsegment(start, length, df=df) | |
1486 |
|
1486 | |||
1487 | def _chunk(self, rev, df=None): |
|
1487 | def _chunk(self, rev, df=None): | |
1488 | """Obtain a single decompressed chunk for a revision. |
|
1488 | """Obtain a single decompressed chunk for a revision. | |
1489 |
|
1489 | |||
1490 | Accepts an integer revision and an optional already-open file handle |
|
1490 | Accepts an integer revision and an optional already-open file handle | |
1491 | to be used for reading. If used, the seek position of the file will not |
|
1491 | to be used for reading. If used, the seek position of the file will not | |
1492 | be preserved. |
|
1492 | be preserved. | |
1493 |
|
1493 | |||
1494 | Returns a str holding uncompressed data for the requested revision. |
|
1494 | Returns a str holding uncompressed data for the requested revision. | |
1495 | """ |
|
1495 | """ | |
1496 | return self.decompress(self._getsegmentforrevs(rev, rev, df=df)[1]) |
|
1496 | return self.decompress(self._getsegmentforrevs(rev, rev, df=df)[1]) | |
1497 |
|
1497 | |||
1498 | def _chunks(self, revs, df=None, targetsize=None): |
|
1498 | def _chunks(self, revs, df=None, targetsize=None): | |
1499 | """Obtain decompressed chunks for the specified revisions. |
|
1499 | """Obtain decompressed chunks for the specified revisions. | |
1500 |
|
1500 | |||
1501 | Accepts an iterable of numeric revisions that are assumed to be in |
|
1501 | Accepts an iterable of numeric revisions that are assumed to be in | |
1502 | ascending order. Also accepts an optional already-open file handle |
|
1502 | ascending order. Also accepts an optional already-open file handle | |
1503 | to be used for reading. If used, the seek position of the file will |
|
1503 | to be used for reading. If used, the seek position of the file will | |
1504 | not be preserved. |
|
1504 | not be preserved. | |
1505 |
|
1505 | |||
1506 | This function is similar to calling ``self._chunk()`` multiple times, |
|
1506 | This function is similar to calling ``self._chunk()`` multiple times, | |
1507 | but is faster. |
|
1507 | but is faster. | |
1508 |
|
1508 | |||
1509 | Returns a list with decompressed data for each requested revision. |
|
1509 | Returns a list with decompressed data for each requested revision. | |
1510 | """ |
|
1510 | """ | |
1511 | if not revs: |
|
1511 | if not revs: | |
1512 | return [] |
|
1512 | return [] | |
1513 | start = self.start |
|
1513 | start = self.start | |
1514 | length = self.length |
|
1514 | length = self.length | |
1515 | inline = self._inline |
|
1515 | inline = self._inline | |
1516 | iosize = self._io.size |
|
1516 | iosize = self._io.size | |
1517 | buffer = util.buffer |
|
1517 | buffer = util.buffer | |
1518 |
|
1518 | |||
1519 | l = [] |
|
1519 | l = [] | |
1520 | ladd = l.append |
|
1520 | ladd = l.append | |
1521 |
|
1521 | |||
1522 | if not self._withsparseread: |
|
1522 | if not self._withsparseread: | |
1523 | slicedchunks = (revs,) |
|
1523 | slicedchunks = (revs,) | |
1524 | else: |
|
1524 | else: | |
1525 | slicedchunks = deltautil.slicechunk(self, revs, |
|
1525 | slicedchunks = deltautil.slicechunk(self, revs, | |
1526 | targetsize=targetsize) |
|
1526 | targetsize=targetsize) | |
1527 |
|
1527 | |||
1528 | for revschunk in slicedchunks: |
|
1528 | for revschunk in slicedchunks: | |
1529 | firstrev = revschunk[0] |
|
1529 | firstrev = revschunk[0] | |
1530 | # Skip trailing revisions with empty diff |
|
1530 | # Skip trailing revisions with empty diff | |
1531 | for lastrev in revschunk[::-1]: |
|
1531 | for lastrev in revschunk[::-1]: | |
1532 | if length(lastrev) != 0: |
|
1532 | if length(lastrev) != 0: | |
1533 | break |
|
1533 | break | |
1534 |
|
1534 | |||
1535 | try: |
|
1535 | try: | |
1536 | offset, data = self._getsegmentforrevs(firstrev, lastrev, df=df) |
|
1536 | offset, data = self._getsegmentforrevs(firstrev, lastrev, df=df) | |
1537 | except OverflowError: |
|
1537 | except OverflowError: | |
1538 | # issue4215 - we can't cache a run of chunks greater than |
|
1538 | # issue4215 - we can't cache a run of chunks greater than | |
1539 | # 2G on Windows |
|
1539 | # 2G on Windows | |
1540 | return [self._chunk(rev, df=df) for rev in revschunk] |
|
1540 | return [self._chunk(rev, df=df) for rev in revschunk] | |
1541 |
|
1541 | |||
1542 | decomp = self.decompress |
|
1542 | decomp = self.decompress | |
1543 | for rev in revschunk: |
|
1543 | for rev in revschunk: | |
1544 | chunkstart = start(rev) |
|
1544 | chunkstart = start(rev) | |
1545 | if inline: |
|
1545 | if inline: | |
1546 | chunkstart += (rev + 1) * iosize |
|
1546 | chunkstart += (rev + 1) * iosize | |
1547 | chunklength = length(rev) |
|
1547 | chunklength = length(rev) | |
1548 | ladd(decomp(buffer(data, chunkstart - offset, chunklength))) |
|
1548 | ladd(decomp(buffer(data, chunkstart - offset, chunklength))) | |
1549 |
|
1549 | |||
1550 | return l |
|
1550 | return l | |
1551 |
|
1551 | |||
1552 | def _chunkclear(self): |
|
1552 | def _chunkclear(self): | |
1553 | """Clear the raw chunk cache.""" |
|
1553 | """Clear the raw chunk cache.""" | |
1554 | self._chunkcache = (0, '') |
|
1554 | self._chunkcache = (0, '') | |
1555 |
|
1555 | |||
1556 | def deltaparent(self, rev): |
|
1556 | def deltaparent(self, rev): | |
1557 | """return deltaparent of the given revision""" |
|
1557 | """return deltaparent of the given revision""" | |
1558 | base = self.index[rev][3] |
|
1558 | base = self.index[rev][3] | |
1559 | if base == rev: |
|
1559 | if base == rev: | |
1560 | return nullrev |
|
1560 | return nullrev | |
1561 | elif self._generaldelta: |
|
1561 | elif self._generaldelta: | |
1562 | return base |
|
1562 | return base | |
1563 | else: |
|
1563 | else: | |
1564 | return rev - 1 |
|
1564 | return rev - 1 | |
1565 |
|
1565 | |||
1566 | def issnapshot(self, rev): |
|
1566 | def issnapshot(self, rev): | |
1567 | """tells whether rev is a snapshot |
|
1567 | """tells whether rev is a snapshot | |
1568 | """ |
|
1568 | """ | |
1569 | if not self._sparserevlog: |
|
1569 | if not self._sparserevlog: | |
1570 | return self.deltaparent(rev) == nullrev |
|
1570 | return self.deltaparent(rev) == nullrev | |
1571 | elif util.safehasattr(self.index, 'issnapshot'): |
|
1571 | elif util.safehasattr(self.index, 'issnapshot'): | |
1572 | # directly assign the method to cache the testing and access |
|
1572 | # directly assign the method to cache the testing and access | |
1573 | self.issnapshot = self.index.issnapshot |
|
1573 | self.issnapshot = self.index.issnapshot | |
1574 | return self.issnapshot(rev) |
|
1574 | return self.issnapshot(rev) | |
1575 | if rev == nullrev: |
|
1575 | if rev == nullrev: | |
1576 | return True |
|
1576 | return True | |
1577 | entry = self.index[rev] |
|
1577 | entry = self.index[rev] | |
1578 | base = entry[3] |
|
1578 | base = entry[3] | |
1579 | if base == rev: |
|
1579 | if base == rev: | |
1580 | return True |
|
1580 | return True | |
1581 | if base == nullrev: |
|
1581 | if base == nullrev: | |
1582 | return True |
|
1582 | return True | |
1583 | p1 = entry[5] |
|
1583 | p1 = entry[5] | |
1584 | p2 = entry[6] |
|
1584 | p2 = entry[6] | |
1585 | if base == p1 or base == p2: |
|
1585 | if base == p1 or base == p2: | |
1586 | return False |
|
1586 | return False | |
1587 | return self.issnapshot(base) |
|
1587 | return self.issnapshot(base) | |
1588 |
|
1588 | |||
1589 | def snapshotdepth(self, rev): |
|
1589 | def snapshotdepth(self, rev): | |
1590 | """number of snapshot in the chain before this one""" |
|
1590 | """number of snapshot in the chain before this one""" | |
1591 | if not self.issnapshot(rev): |
|
1591 | if not self.issnapshot(rev): | |
1592 | raise error.ProgrammingError('revision %d not a snapshot') |
|
1592 | raise error.ProgrammingError('revision %d not a snapshot') | |
1593 | return len(self._deltachain(rev)[0]) - 1 |
|
1593 | return len(self._deltachain(rev)[0]) - 1 | |
1594 |
|
1594 | |||
1595 | def revdiff(self, rev1, rev2): |
|
1595 | def revdiff(self, rev1, rev2): | |
1596 | """return or calculate a delta between two revisions |
|
1596 | """return or calculate a delta between two revisions | |
1597 |
|
1597 | |||
1598 | The delta calculated is in binary form and is intended to be written to |
|
1598 | The delta calculated is in binary form and is intended to be written to | |
1599 | revlog data directly. So this function needs raw revision data. |
|
1599 | revlog data directly. So this function needs raw revision data. | |
1600 | """ |
|
1600 | """ | |
1601 | if rev1 != nullrev and self.deltaparent(rev2) == rev1: |
|
1601 | if rev1 != nullrev and self.deltaparent(rev2) == rev1: | |
1602 | return bytes(self._chunk(rev2)) |
|
1602 | return bytes(self._chunk(rev2)) | |
1603 |
|
1603 | |||
1604 | return mdiff.textdiff(self.rawdata(rev1), |
|
1604 | return mdiff.textdiff(self.rawdata(rev1), | |
1605 | self.rawdata(rev2)) |
|
1605 | self.rawdata(rev2)) | |
1606 |
|
1606 | |||
1607 | def _processflags(self, text, flags, operation, raw=False): |
|
1607 | def _processflags(self, text, flags, operation, raw=False): | |
1608 | """deprecated entry point to access flag processors""" |
|
1608 | """deprecated entry point to access flag processors""" | |
1609 | msg = ('_processflag(...) use the specialized variant') |
|
1609 | msg = ('_processflag(...) use the specialized variant') | |
1610 | util.nouideprecwarn(msg, '5.2', stacklevel=2) |
|
1610 | util.nouideprecwarn(msg, '5.2', stacklevel=2) | |
1611 | if raw: |
|
1611 | if raw: | |
1612 | return text, flagutil.processflagsraw(self, text, flags) |
|
1612 | return text, flagutil.processflagsraw(self, text, flags) | |
1613 | elif operation == 'read': |
|
1613 | elif operation == 'read': | |
1614 | return flagutil.processflagsread(self, text, flags) |
|
1614 | return flagutil.processflagsread(self, text, flags) | |
1615 | else: # write operation |
|
1615 | else: # write operation | |
1616 | return flagutil.processflagswrite(self, text, flags) |
|
1616 | return flagutil.processflagswrite(self, text, flags) | |
1617 |
|
1617 | |||
1618 | def revision(self, nodeorrev, _df=None, raw=False): |
|
1618 | def revision(self, nodeorrev, _df=None, raw=False): | |
1619 | """return an uncompressed revision of a given node or revision |
|
1619 | """return an uncompressed revision of a given node or revision | |
1620 | number. |
|
1620 | number. | |
1621 |
|
1621 | |||
1622 | _df - an existing file handle to read from. (internal-only) |
|
1622 | _df - an existing file handle to read from. (internal-only) | |
1623 | raw - an optional argument specifying if the revision data is to be |
|
1623 | raw - an optional argument specifying if the revision data is to be | |
1624 | treated as raw data when applying flag transforms. 'raw' should be set |
|
1624 | treated as raw data when applying flag transforms. 'raw' should be set | |
1625 | to True when generating changegroups or in debug commands. |
|
1625 | to True when generating changegroups or in debug commands. | |
1626 | """ |
|
1626 | """ | |
1627 | if raw: |
|
1627 | if raw: | |
1628 | msg = ('revlog.revision(..., raw=True) is deprecated, ' |
|
1628 | msg = ('revlog.revision(..., raw=True) is deprecated, ' | |
1629 | 'use revlog.rawdata(...)') |
|
1629 | 'use revlog.rawdata(...)') | |
1630 | util.nouideprecwarn(msg, '5.2', stacklevel=2) |
|
1630 | util.nouideprecwarn(msg, '5.2', stacklevel=2) | |
1631 | return self._revisiondata(nodeorrev, _df, raw=raw)[0] |
|
1631 | return self._revisiondata(nodeorrev, _df, raw=raw)[0] | |
1632 |
|
1632 | |||
1633 | def sidedata(self, nodeorrev, _df=None): |
|
1633 | def sidedata(self, nodeorrev, _df=None): | |
1634 | """a map of extra data related to the changeset but not part of the hash |
|
1634 | """a map of extra data related to the changeset but not part of the hash | |
1635 |
|
1635 | |||
1636 | This function currently return a dictionary. However, more advanced |
|
1636 | This function currently return a dictionary. However, more advanced | |
1637 | mapping object will likely be used in the future for a more |
|
1637 | mapping object will likely be used in the future for a more | |
1638 | efficient/lazy code. |
|
1638 | efficient/lazy code. | |
1639 | """ |
|
1639 | """ | |
1640 | return self._revisiondata(nodeorrev, _df)[1] |
|
1640 | return self._revisiondata(nodeorrev, _df)[1] | |
1641 |
|
1641 | |||
1642 | def _revisiondata(self, nodeorrev, _df=None, raw=False): |
|
1642 | def _revisiondata(self, nodeorrev, _df=None, raw=False): | |
1643 | # deal with <nodeorrev> argument type |
|
1643 | # deal with <nodeorrev> argument type | |
1644 | if isinstance(nodeorrev, int): |
|
1644 | if isinstance(nodeorrev, int): | |
1645 | rev = nodeorrev |
|
1645 | rev = nodeorrev | |
1646 | node = self.node(rev) |
|
1646 | node = self.node(rev) | |
1647 | else: |
|
1647 | else: | |
1648 | node = nodeorrev |
|
1648 | node = nodeorrev | |
1649 | rev = None |
|
1649 | rev = None | |
1650 |
|
1650 | |||
1651 | # fast path the special `nullid` rev |
|
1651 | # fast path the special `nullid` rev | |
1652 | if node == nullid: |
|
1652 | if node == nullid: | |
1653 | return "", {} |
|
1653 | return "", {} | |
1654 |
|
1654 | |||
1655 | # The text as stored inside the revlog. Might be the revision or might |
|
1655 | # The text as stored inside the revlog. Might be the revision or might | |
1656 | # need to be processed to retrieve the revision. |
|
1656 | # need to be processed to retrieve the revision. | |
1657 | rawtext = None |
|
1657 | rawtext = None | |
1658 |
|
1658 | |||
1659 | rev, rawtext, validated = self._rawtext(node, rev, _df=_df) |
|
1659 | rev, rawtext, validated = self._rawtext(node, rev, _df=_df) | |
1660 |
|
1660 | |||
1661 | if raw and validated: |
|
1661 | if raw and validated: | |
1662 | # if we don't want to process the raw text and that raw |
|
1662 | # if we don't want to process the raw text and that raw | |
1663 | # text is cached, we can exit early. |
|
1663 | # text is cached, we can exit early. | |
1664 | return rawtext, {} |
|
1664 | return rawtext, {} | |
1665 | if rev is None: |
|
1665 | if rev is None: | |
1666 | rev = self.rev(node) |
|
1666 | rev = self.rev(node) | |
1667 | # the revlog's flag for this revision |
|
1667 | # the revlog's flag for this revision | |
1668 | # (usually alter its state or content) |
|
1668 | # (usually alter its state or content) | |
1669 | flags = self.flags(rev) |
|
1669 | flags = self.flags(rev) | |
1670 |
|
1670 | |||
1671 | if validated and flags == REVIDX_DEFAULT_FLAGS: |
|
1671 | if validated and flags == REVIDX_DEFAULT_FLAGS: | |
1672 | # no extra flags set, no flag processor runs, text = rawtext |
|
1672 | # no extra flags set, no flag processor runs, text = rawtext | |
1673 | return rawtext, {} |
|
1673 | return rawtext, {} | |
1674 |
|
1674 | |||
1675 | sidedata = {} |
|
1675 | sidedata = {} | |
1676 | if raw: |
|
1676 | if raw: | |
1677 | validatehash = flagutil.processflagsraw(self, rawtext, flags) |
|
1677 | validatehash = flagutil.processflagsraw(self, rawtext, flags) | |
1678 | text = rawtext |
|
1678 | text = rawtext | |
1679 | else: |
|
1679 | else: | |
1680 | r = flagutil.processflagsread(self, rawtext, flags) |
|
1680 | r = flagutil.processflagsread(self, rawtext, flags) | |
1681 | text, validatehash, sidedata = r |
|
1681 | text, validatehash, sidedata = r | |
1682 | if validatehash: |
|
1682 | if validatehash: | |
1683 | self.checkhash(text, node, rev=rev) |
|
1683 | self.checkhash(text, node, rev=rev) | |
1684 | if not validated: |
|
1684 | if not validated: | |
1685 | self._revisioncache = (node, rev, rawtext) |
|
1685 | self._revisioncache = (node, rev, rawtext) | |
1686 |
|
1686 | |||
1687 | return text, sidedata |
|
1687 | return text, sidedata | |
1688 |
|
1688 | |||
1689 | def _rawtext(self, node, rev, _df=None): |
|
1689 | def _rawtext(self, node, rev, _df=None): | |
1690 | """return the possibly unvalidated rawtext for a revision |
|
1690 | """return the possibly unvalidated rawtext for a revision | |
1691 |
|
1691 | |||
1692 | returns (rev, rawtext, validated) |
|
1692 | returns (rev, rawtext, validated) | |
1693 | """ |
|
1693 | """ | |
1694 |
|
1694 | |||
1695 | # revision in the cache (could be useful to apply delta) |
|
1695 | # revision in the cache (could be useful to apply delta) | |
1696 | cachedrev = None |
|
1696 | cachedrev = None | |
1697 | # An intermediate text to apply deltas to |
|
1697 | # An intermediate text to apply deltas to | |
1698 | basetext = None |
|
1698 | basetext = None | |
1699 |
|
1699 | |||
1700 | # Check if we have the entry in cache |
|
1700 | # Check if we have the entry in cache | |
1701 | # The cache entry looks like (node, rev, rawtext) |
|
1701 | # The cache entry looks like (node, rev, rawtext) | |
1702 | if self._revisioncache: |
|
1702 | if self._revisioncache: | |
1703 | if self._revisioncache[0] == node: |
|
1703 | if self._revisioncache[0] == node: | |
1704 | return (rev, self._revisioncache[2], True) |
|
1704 | return (rev, self._revisioncache[2], True) | |
1705 | cachedrev = self._revisioncache[1] |
|
1705 | cachedrev = self._revisioncache[1] | |
1706 |
|
1706 | |||
1707 | if rev is None: |
|
1707 | if rev is None: | |
1708 | rev = self.rev(node) |
|
1708 | rev = self.rev(node) | |
1709 |
|
1709 | |||
1710 | chain, stopped = self._deltachain(rev, stoprev=cachedrev) |
|
1710 | chain, stopped = self._deltachain(rev, stoprev=cachedrev) | |
1711 | if stopped: |
|
1711 | if stopped: | |
1712 | basetext = self._revisioncache[2] |
|
1712 | basetext = self._revisioncache[2] | |
1713 |
|
1713 | |||
1714 | # drop cache to save memory, the caller is expected to |
|
1714 | # drop cache to save memory, the caller is expected to | |
1715 | # update self._revisioncache after validating the text |
|
1715 | # update self._revisioncache after validating the text | |
1716 | self._revisioncache = None |
|
1716 | self._revisioncache = None | |
1717 |
|
1717 | |||
1718 | targetsize = None |
|
1718 | targetsize = None | |
1719 | rawsize = self.index[rev][2] |
|
1719 | rawsize = self.index[rev][2] | |
1720 | if 0 <= rawsize: |
|
1720 | if 0 <= rawsize: | |
1721 | targetsize = 4 * rawsize |
|
1721 | targetsize = 4 * rawsize | |
1722 |
|
1722 | |||
1723 | bins = self._chunks(chain, df=_df, targetsize=targetsize) |
|
1723 | bins = self._chunks(chain, df=_df, targetsize=targetsize) | |
1724 | if basetext is None: |
|
1724 | if basetext is None: | |
1725 | basetext = bytes(bins[0]) |
|
1725 | basetext = bytes(bins[0]) | |
1726 | bins = bins[1:] |
|
1726 | bins = bins[1:] | |
1727 |
|
1727 | |||
1728 | rawtext = mdiff.patches(basetext, bins) |
|
1728 | rawtext = mdiff.patches(basetext, bins) | |
1729 | del basetext # let us have a chance to free memory early |
|
1729 | del basetext # let us have a chance to free memory early | |
1730 | return (rev, rawtext, False) |
|
1730 | return (rev, rawtext, False) | |
1731 |
|
1731 | |||
1732 | def rawdata(self, nodeorrev, _df=None): |
|
1732 | def rawdata(self, nodeorrev, _df=None): | |
1733 | """return an uncompressed raw data of a given node or revision number. |
|
1733 | """return an uncompressed raw data of a given node or revision number. | |
1734 |
|
1734 | |||
1735 | _df - an existing file handle to read from. (internal-only) |
|
1735 | _df - an existing file handle to read from. (internal-only) | |
1736 | """ |
|
1736 | """ | |
1737 | return self._revisiondata(nodeorrev, _df, raw=True)[0] |
|
1737 | return self._revisiondata(nodeorrev, _df, raw=True)[0] | |
1738 |
|
1738 | |||
1739 | def hash(self, text, p1, p2): |
|
1739 | def hash(self, text, p1, p2): | |
1740 | """Compute a node hash. |
|
1740 | """Compute a node hash. | |
1741 |
|
1741 | |||
1742 | Available as a function so that subclasses can replace the hash |
|
1742 | Available as a function so that subclasses can replace the hash | |
1743 | as needed. |
|
1743 | as needed. | |
1744 | """ |
|
1744 | """ | |
1745 | return storageutil.hashrevisionsha1(text, p1, p2) |
|
1745 | return storageutil.hashrevisionsha1(text, p1, p2) | |
1746 |
|
1746 | |||
1747 | def checkhash(self, text, node, p1=None, p2=None, rev=None): |
|
1747 | def checkhash(self, text, node, p1=None, p2=None, rev=None): | |
1748 | """Check node hash integrity. |
|
1748 | """Check node hash integrity. | |
1749 |
|
1749 | |||
1750 | Available as a function so that subclasses can extend hash mismatch |
|
1750 | Available as a function so that subclasses can extend hash mismatch | |
1751 | behaviors as needed. |
|
1751 | behaviors as needed. | |
1752 | """ |
|
1752 | """ | |
1753 | try: |
|
1753 | try: | |
1754 | if p1 is None and p2 is None: |
|
1754 | if p1 is None and p2 is None: | |
1755 | p1, p2 = self.parents(node) |
|
1755 | p1, p2 = self.parents(node) | |
1756 | if node != self.hash(text, p1, p2): |
|
1756 | if node != self.hash(text, p1, p2): | |
1757 | # Clear the revision cache on hash failure. The revision cache |
|
1757 | # Clear the revision cache on hash failure. The revision cache | |
1758 | # only stores the raw revision and clearing the cache does have |
|
1758 | # only stores the raw revision and clearing the cache does have | |
1759 | # the side-effect that we won't have a cache hit when the raw |
|
1759 | # the side-effect that we won't have a cache hit when the raw | |
1760 | # revision data is accessed. But this case should be rare and |
|
1760 | # revision data is accessed. But this case should be rare and | |
1761 | # it is extra work to teach the cache about the hash |
|
1761 | # it is extra work to teach the cache about the hash | |
1762 | # verification state. |
|
1762 | # verification state. | |
1763 | if self._revisioncache and self._revisioncache[0] == node: |
|
1763 | if self._revisioncache and self._revisioncache[0] == node: | |
1764 | self._revisioncache = None |
|
1764 | self._revisioncache = None | |
1765 |
|
1765 | |||
1766 | revornode = rev |
|
1766 | revornode = rev | |
1767 | if revornode is None: |
|
1767 | if revornode is None: | |
1768 | revornode = templatefilters.short(hex(node)) |
|
1768 | revornode = templatefilters.short(hex(node)) | |
1769 | raise error.RevlogError(_("integrity check failed on %s:%s") |
|
1769 | raise error.RevlogError(_("integrity check failed on %s:%s") | |
1770 | % (self.indexfile, pycompat.bytestr(revornode))) |
|
1770 | % (self.indexfile, pycompat.bytestr(revornode))) | |
1771 | except error.RevlogError: |
|
1771 | except error.RevlogError: | |
1772 | if self._censorable and storageutil.iscensoredtext(text): |
|
1772 | if self._censorable and storageutil.iscensoredtext(text): | |
1773 | raise error.CensoredNodeError(self.indexfile, node, text) |
|
1773 | raise error.CensoredNodeError(self.indexfile, node, text) | |
1774 | raise |
|
1774 | raise | |
1775 |
|
1775 | |||
1776 | def _enforceinlinesize(self, tr, fp=None): |
|
1776 | def _enforceinlinesize(self, tr, fp=None): | |
1777 | """Check if the revlog is too big for inline and convert if so. |
|
1777 | """Check if the revlog is too big for inline and convert if so. | |
1778 |
|
1778 | |||
1779 | This should be called after revisions are added to the revlog. If the |
|
1779 | This should be called after revisions are added to the revlog. If the | |
1780 | revlog has grown too large to be an inline revlog, it will convert it |
|
1780 | revlog has grown too large to be an inline revlog, it will convert it | |
1781 | to use multiple index and data files. |
|
1781 | to use multiple index and data files. | |
1782 | """ |
|
1782 | """ | |
1783 | tiprev = len(self) - 1 |
|
1783 | tiprev = len(self) - 1 | |
1784 | if (not self._inline or |
|
1784 | if (not self._inline or | |
1785 | (self.start(tiprev) + self.length(tiprev)) < _maxinline): |
|
1785 | (self.start(tiprev) + self.length(tiprev)) < _maxinline): | |
1786 | return |
|
1786 | return | |
1787 |
|
1787 | |||
1788 | trinfo = tr.find(self.indexfile) |
|
1788 | trinfo = tr.find(self.indexfile) | |
1789 | if trinfo is None: |
|
1789 | if trinfo is None: | |
1790 | raise error.RevlogError(_("%s not found in the transaction") |
|
1790 | raise error.RevlogError(_("%s not found in the transaction") | |
1791 | % self.indexfile) |
|
1791 | % self.indexfile) | |
1792 |
|
1792 | |||
1793 | trindex = trinfo[2] |
|
1793 | trindex = trinfo[2] | |
1794 | if trindex is not None: |
|
1794 | if trindex is not None: | |
1795 | dataoff = self.start(trindex) |
|
1795 | dataoff = self.start(trindex) | |
1796 | else: |
|
1796 | else: | |
1797 | # revlog was stripped at start of transaction, use all leftover data |
|
1797 | # revlog was stripped at start of transaction, use all leftover data | |
1798 | trindex = len(self) - 1 |
|
1798 | trindex = len(self) - 1 | |
1799 | dataoff = self.end(tiprev) |
|
1799 | dataoff = self.end(tiprev) | |
1800 |
|
1800 | |||
1801 | tr.add(self.datafile, dataoff) |
|
1801 | tr.add(self.datafile, dataoff) | |
1802 |
|
1802 | |||
1803 | if fp: |
|
1803 | if fp: | |
1804 | fp.flush() |
|
1804 | fp.flush() | |
1805 | fp.close() |
|
1805 | fp.close() | |
1806 | # We can't use the cached file handle after close(). So prevent |
|
1806 | # We can't use the cached file handle after close(). So prevent | |
1807 | # its usage. |
|
1807 | # its usage. | |
1808 | self._writinghandles = None |
|
1808 | self._writinghandles = None | |
1809 |
|
1809 | |||
1810 | with self._indexfp('r') as ifh, self._datafp('w') as dfh: |
|
1810 | with self._indexfp('r') as ifh, self._datafp('w') as dfh: | |
1811 | for r in self: |
|
1811 | for r in self: | |
1812 | dfh.write(self._getsegmentforrevs(r, r, df=ifh)[1]) |
|
1812 | dfh.write(self._getsegmentforrevs(r, r, df=ifh)[1]) | |
1813 |
|
1813 | |||
1814 | with self._indexfp('w') as fp: |
|
1814 | with self._indexfp('w') as fp: | |
1815 | self.version &= ~FLAG_INLINE_DATA |
|
1815 | self.version &= ~FLAG_INLINE_DATA | |
1816 | self._inline = False |
|
1816 | self._inline = False | |
1817 | io = self._io |
|
1817 | io = self._io | |
1818 | for i in self: |
|
1818 | for i in self: | |
1819 | e = io.packentry(self.index[i], self.node, self.version, i) |
|
1819 | e = io.packentry(self.index[i], self.node, self.version, i) | |
1820 | fp.write(e) |
|
1820 | fp.write(e) | |
1821 |
|
1821 | |||
1822 | # the temp file replace the real index when we exit the context |
|
1822 | # the temp file replace the real index when we exit the context | |
1823 | # manager |
|
1823 | # manager | |
1824 |
|
1824 | |||
1825 | tr.replace(self.indexfile, trindex * self._io.size) |
|
1825 | tr.replace(self.indexfile, trindex * self._io.size) | |
1826 | self._chunkclear() |
|
1826 | self._chunkclear() | |
1827 |
|
1827 | |||
1828 | def _nodeduplicatecallback(self, transaction, node): |
|
1828 | def _nodeduplicatecallback(self, transaction, node): | |
1829 | """called when trying to add a node already stored. |
|
1829 | """called when trying to add a node already stored. | |
1830 | """ |
|
1830 | """ | |
1831 |
|
1831 | |||
1832 | def addrevision(self, text, transaction, link, p1, p2, cachedelta=None, |
|
1832 | def addrevision(self, text, transaction, link, p1, p2, cachedelta=None, | |
1833 | node=None, flags=REVIDX_DEFAULT_FLAGS, deltacomputer=None, |
|
1833 | node=None, flags=REVIDX_DEFAULT_FLAGS, deltacomputer=None, | |
1834 | sidedata=None): |
|
1834 | sidedata=None): | |
1835 | """add a revision to the log |
|
1835 | """add a revision to the log | |
1836 |
|
1836 | |||
1837 | text - the revision data to add |
|
1837 | text - the revision data to add | |
1838 | transaction - the transaction object used for rollback |
|
1838 | transaction - the transaction object used for rollback | |
1839 | link - the linkrev data to add |
|
1839 | link - the linkrev data to add | |
1840 | p1, p2 - the parent nodeids of the revision |
|
1840 | p1, p2 - the parent nodeids of the revision | |
1841 | cachedelta - an optional precomputed delta |
|
1841 | cachedelta - an optional precomputed delta | |
1842 | node - nodeid of revision; typically node is not specified, and it is |
|
1842 | node - nodeid of revision; typically node is not specified, and it is | |
1843 | computed by default as hash(text, p1, p2), however subclasses might |
|
1843 | computed by default as hash(text, p1, p2), however subclasses might | |
1844 | use different hashing method (and override checkhash() in such case) |
|
1844 | use different hashing method (and override checkhash() in such case) | |
1845 | flags - the known flags to set on the revision |
|
1845 | flags - the known flags to set on the revision | |
1846 | deltacomputer - an optional deltacomputer instance shared between |
|
1846 | deltacomputer - an optional deltacomputer instance shared between | |
1847 | multiple calls |
|
1847 | multiple calls | |
1848 | """ |
|
1848 | """ | |
1849 | if link == nullrev: |
|
1849 | if link == nullrev: | |
1850 | raise error.RevlogError(_("attempted to add linkrev -1 to %s") |
|
1850 | raise error.RevlogError(_("attempted to add linkrev -1 to %s") | |
1851 | % self.indexfile) |
|
1851 | % self.indexfile) | |
1852 |
|
1852 | |||
1853 | if sidedata is None: |
|
1853 | if sidedata is None: | |
1854 | sidedata = {} |
|
1854 | sidedata = {} | |
1855 |
|
1855 | |||
1856 | if flags: |
|
1856 | if flags: | |
1857 | node = node or self.hash(text, p1, p2) |
|
1857 | node = node or self.hash(text, p1, p2) | |
1858 |
|
1858 | |||
1859 | rawtext, validatehash = flagutil.processflagswrite(self, text, flags, |
|
1859 | rawtext, validatehash = flagutil.processflagswrite(self, text, flags, | |
1860 | sidedata=sidedata) |
|
1860 | sidedata=sidedata) | |
1861 |
|
1861 | |||
1862 | # If the flag processor modifies the revision data, ignore any provided |
|
1862 | # If the flag processor modifies the revision data, ignore any provided | |
1863 | # cachedelta. |
|
1863 | # cachedelta. | |
1864 | if rawtext != text: |
|
1864 | if rawtext != text: | |
1865 | cachedelta = None |
|
1865 | cachedelta = None | |
1866 |
|
1866 | |||
1867 | if len(rawtext) > _maxentrysize: |
|
1867 | if len(rawtext) > _maxentrysize: | |
1868 | raise error.RevlogError( |
|
1868 | raise error.RevlogError( | |
1869 | _("%s: size of %d bytes exceeds maximum revlog storage of 2GiB") |
|
1869 | _("%s: size of %d bytes exceeds maximum revlog storage of 2GiB") | |
1870 | % (self.indexfile, len(rawtext))) |
|
1870 | % (self.indexfile, len(rawtext))) | |
1871 |
|
1871 | |||
1872 | node = node or self.hash(rawtext, p1, p2) |
|
1872 | node = node or self.hash(rawtext, p1, p2) | |
1873 | if node in self.nodemap: |
|
1873 | if node in self.nodemap: | |
1874 | return node |
|
1874 | return node | |
1875 |
|
1875 | |||
1876 | if validatehash: |
|
1876 | if validatehash: | |
1877 | self.checkhash(rawtext, node, p1=p1, p2=p2) |
|
1877 | self.checkhash(rawtext, node, p1=p1, p2=p2) | |
1878 |
|
1878 | |||
1879 | return self.addrawrevision(rawtext, transaction, link, p1, p2, node, |
|
1879 | return self.addrawrevision(rawtext, transaction, link, p1, p2, node, | |
1880 | flags, cachedelta=cachedelta, |
|
1880 | flags, cachedelta=cachedelta, | |
1881 | deltacomputer=deltacomputer) |
|
1881 | deltacomputer=deltacomputer) | |
1882 |
|
1882 | |||
1883 | def addrawrevision(self, rawtext, transaction, link, p1, p2, node, flags, |
|
1883 | def addrawrevision(self, rawtext, transaction, link, p1, p2, node, flags, | |
1884 | cachedelta=None, deltacomputer=None): |
|
1884 | cachedelta=None, deltacomputer=None): | |
1885 | """add a raw revision with known flags, node and parents |
|
1885 | """add a raw revision with known flags, node and parents | |
1886 | useful when reusing a revision not stored in this revlog (ex: received |
|
1886 | useful when reusing a revision not stored in this revlog (ex: received | |
1887 | over wire, or read from an external bundle). |
|
1887 | over wire, or read from an external bundle). | |
1888 | """ |
|
1888 | """ | |
1889 | dfh = None |
|
1889 | dfh = None | |
1890 | if not self._inline: |
|
1890 | if not self._inline: | |
1891 | dfh = self._datafp("a+") |
|
1891 | dfh = self._datafp("a+") | |
1892 | ifh = self._indexfp("a+") |
|
1892 | ifh = self._indexfp("a+") | |
1893 | try: |
|
1893 | try: | |
1894 | return self._addrevision(node, rawtext, transaction, link, p1, p2, |
|
1894 | return self._addrevision(node, rawtext, transaction, link, p1, p2, | |
1895 | flags, cachedelta, ifh, dfh, |
|
1895 | flags, cachedelta, ifh, dfh, | |
1896 | deltacomputer=deltacomputer) |
|
1896 | deltacomputer=deltacomputer) | |
1897 | finally: |
|
1897 | finally: | |
1898 | if dfh: |
|
1898 | if dfh: | |
1899 | dfh.close() |
|
1899 | dfh.close() | |
1900 | ifh.close() |
|
1900 | ifh.close() | |
1901 |
|
1901 | |||
1902 | def compress(self, data): |
|
1902 | def compress(self, data): | |
1903 | """Generate a possibly-compressed representation of data.""" |
|
1903 | """Generate a possibly-compressed representation of data.""" | |
1904 | if not data: |
|
1904 | if not data: | |
1905 | return '', data |
|
1905 | return '', data | |
1906 |
|
1906 | |||
1907 | compressed = self._compressor.compress(data) |
|
1907 | compressed = self._compressor.compress(data) | |
1908 |
|
1908 | |||
1909 | if compressed: |
|
1909 | if compressed: | |
1910 | # The revlog compressor added the header in the returned data. |
|
1910 | # The revlog compressor added the header in the returned data. | |
1911 | return '', compressed |
|
1911 | return '', compressed | |
1912 |
|
1912 | |||
1913 | if data[0:1] == '\0': |
|
1913 | if data[0:1] == '\0': | |
1914 | return '', data |
|
1914 | return '', data | |
1915 | return 'u', data |
|
1915 | return 'u', data | |
1916 |
|
1916 | |||
1917 | def decompress(self, data): |
|
1917 | def decompress(self, data): | |
1918 | """Decompress a revlog chunk. |
|
1918 | """Decompress a revlog chunk. | |
1919 |
|
1919 | |||
1920 | The chunk is expected to begin with a header identifying the |
|
1920 | The chunk is expected to begin with a header identifying the | |
1921 | format type so it can be routed to an appropriate decompressor. |
|
1921 | format type so it can be routed to an appropriate decompressor. | |
1922 | """ |
|
1922 | """ | |
1923 | if not data: |
|
1923 | if not data: | |
1924 | return data |
|
1924 | return data | |
1925 |
|
1925 | |||
1926 | # Revlogs are read much more frequently than they are written and many |
|
1926 | # Revlogs are read much more frequently than they are written and many | |
1927 | # chunks only take microseconds to decompress, so performance is |
|
1927 | # chunks only take microseconds to decompress, so performance is | |
1928 | # important here. |
|
1928 | # important here. | |
1929 | # |
|
1929 | # | |
1930 | # We can make a few assumptions about revlogs: |
|
1930 | # We can make a few assumptions about revlogs: | |
1931 | # |
|
1931 | # | |
1932 | # 1) the majority of chunks will be compressed (as opposed to inline |
|
1932 | # 1) the majority of chunks will be compressed (as opposed to inline | |
1933 | # raw data). |
|
1933 | # raw data). | |
1934 | # 2) decompressing *any* data will likely by at least 10x slower than |
|
1934 | # 2) decompressing *any* data will likely by at least 10x slower than | |
1935 | # returning raw inline data. |
|
1935 | # returning raw inline data. | |
1936 | # 3) we want to prioritize common and officially supported compression |
|
1936 | # 3) we want to prioritize common and officially supported compression | |
1937 | # engines |
|
1937 | # engines | |
1938 | # |
|
1938 | # | |
1939 | # It follows that we want to optimize for "decompress compressed data |
|
1939 | # It follows that we want to optimize for "decompress compressed data | |
1940 | # when encoded with common and officially supported compression engines" |
|
1940 | # when encoded with common and officially supported compression engines" | |
1941 | # case over "raw data" and "data encoded by less common or non-official |
|
1941 | # case over "raw data" and "data encoded by less common or non-official | |
1942 | # compression engines." That is why we have the inline lookup first |
|
1942 | # compression engines." That is why we have the inline lookup first | |
1943 | # followed by the compengines lookup. |
|
1943 | # followed by the compengines lookup. | |
1944 | # |
|
1944 | # | |
1945 | # According to `hg perfrevlogchunks`, this is ~0.5% faster for zlib |
|
1945 | # According to `hg perfrevlogchunks`, this is ~0.5% faster for zlib | |
1946 | # compressed chunks. And this matters for changelog and manifest reads. |
|
1946 | # compressed chunks. And this matters for changelog and manifest reads. | |
1947 | t = data[0:1] |
|
1947 | t = data[0:1] | |
1948 |
|
1948 | |||
1949 | if t == 'x': |
|
1949 | if t == 'x': | |
1950 | try: |
|
1950 | try: | |
1951 | return _zlibdecompress(data) |
|
1951 | return _zlibdecompress(data) | |
1952 | except zlib.error as e: |
|
1952 | except zlib.error as e: | |
1953 | raise error.RevlogError(_('revlog decompress error: %s') % |
|
1953 | raise error.RevlogError(_('revlog decompress error: %s') % | |
1954 | stringutil.forcebytestr(e)) |
|
1954 | stringutil.forcebytestr(e)) | |
1955 | # '\0' is more common than 'u' so it goes first. |
|
1955 | # '\0' is more common than 'u' so it goes first. | |
1956 | elif t == '\0': |
|
1956 | elif t == '\0': | |
1957 | return data |
|
1957 | return data | |
1958 | elif t == 'u': |
|
1958 | elif t == 'u': | |
1959 | return util.buffer(data, 1) |
|
1959 | return util.buffer(data, 1) | |
1960 |
|
1960 | |||
1961 | try: |
|
1961 | try: | |
1962 | compressor = self._decompressors[t] |
|
1962 | compressor = self._decompressors[t] | |
1963 | except KeyError: |
|
1963 | except KeyError: | |
1964 | try: |
|
1964 | try: | |
1965 | engine = util.compengines.forrevlogheader(t) |
|
1965 | engine = util.compengines.forrevlogheader(t) | |
1966 | compressor = engine.revlogcompressor(self._compengineopts) |
|
1966 | compressor = engine.revlogcompressor(self._compengineopts) | |
1967 | self._decompressors[t] = compressor |
|
1967 | self._decompressors[t] = compressor | |
1968 | except KeyError: |
|
1968 | except KeyError: | |
1969 | raise error.RevlogError(_('unknown compression type %r') % t) |
|
1969 | raise error.RevlogError(_('unknown compression type %r') % t) | |
1970 |
|
1970 | |||
1971 | return compressor.decompress(data) |
|
1971 | return compressor.decompress(data) | |
1972 |
|
1972 | |||
1973 | def _addrevision(self, node, rawtext, transaction, link, p1, p2, flags, |
|
1973 | def _addrevision(self, node, rawtext, transaction, link, p1, p2, flags, | |
1974 | cachedelta, ifh, dfh, alwayscache=False, |
|
1974 | cachedelta, ifh, dfh, alwayscache=False, | |
1975 | deltacomputer=None): |
|
1975 | deltacomputer=None): | |
1976 | """internal function to add revisions to the log |
|
1976 | """internal function to add revisions to the log | |
1977 |
|
1977 | |||
1978 | see addrevision for argument descriptions. |
|
1978 | see addrevision for argument descriptions. | |
1979 |
|
1979 | |||
1980 | note: "addrevision" takes non-raw text, "_addrevision" takes raw text. |
|
1980 | note: "addrevision" takes non-raw text, "_addrevision" takes raw text. | |
1981 |
|
1981 | |||
1982 | if "deltacomputer" is not provided or None, a defaultdeltacomputer will |
|
1982 | if "deltacomputer" is not provided or None, a defaultdeltacomputer will | |
1983 | be used. |
|
1983 | be used. | |
1984 |
|
1984 | |||
1985 | invariants: |
|
1985 | invariants: | |
1986 | - rawtext is optional (can be None); if not set, cachedelta must be set. |
|
1986 | - rawtext is optional (can be None); if not set, cachedelta must be set. | |
1987 | if both are set, they must correspond to each other. |
|
1987 | if both are set, they must correspond to each other. | |
1988 | """ |
|
1988 | """ | |
1989 | if node == nullid: |
|
1989 | if node == nullid: | |
1990 | raise error.RevlogError(_("%s: attempt to add null revision") % |
|
1990 | raise error.RevlogError(_("%s: attempt to add null revision") % | |
1991 | self.indexfile) |
|
1991 | self.indexfile) | |
1992 | if node == wdirid or node in wdirfilenodeids: |
|
1992 | if node == wdirid or node in wdirfilenodeids: | |
1993 | raise error.RevlogError(_("%s: attempt to add wdir revision") % |
|
1993 | raise error.RevlogError(_("%s: attempt to add wdir revision") % | |
1994 | self.indexfile) |
|
1994 | self.indexfile) | |
1995 |
|
1995 | |||
1996 | if self._inline: |
|
1996 | if self._inline: | |
1997 | fh = ifh |
|
1997 | fh = ifh | |
1998 | else: |
|
1998 | else: | |
1999 | fh = dfh |
|
1999 | fh = dfh | |
2000 |
|
2000 | |||
2001 | btext = [rawtext] |
|
2001 | btext = [rawtext] | |
2002 |
|
2002 | |||
2003 | curr = len(self) |
|
2003 | curr = len(self) | |
2004 | prev = curr - 1 |
|
2004 | prev = curr - 1 | |
2005 | offset = self.end(prev) |
|
2005 | offset = self.end(prev) | |
2006 | p1r, p2r = self.rev(p1), self.rev(p2) |
|
2006 | p1r, p2r = self.rev(p1), self.rev(p2) | |
2007 |
|
2007 | |||
2008 | # full versions are inserted when the needed deltas |
|
2008 | # full versions are inserted when the needed deltas | |
2009 | # become comparable to the uncompressed text |
|
2009 | # become comparable to the uncompressed text | |
2010 | if rawtext is None: |
|
2010 | if rawtext is None: | |
2011 | # need rawtext size, before changed by flag processors, which is |
|
2011 | # need rawtext size, before changed by flag processors, which is | |
2012 | # the non-raw size. use revlog explicitly to avoid filelog's extra |
|
2012 | # the non-raw size. use revlog explicitly to avoid filelog's extra | |
2013 | # logic that might remove metadata size. |
|
2013 | # logic that might remove metadata size. | |
2014 | textlen = mdiff.patchedsize(revlog.size(self, cachedelta[0]), |
|
2014 | textlen = mdiff.patchedsize(revlog.size(self, cachedelta[0]), | |
2015 | cachedelta[1]) |
|
2015 | cachedelta[1]) | |
2016 | else: |
|
2016 | else: | |
2017 | textlen = len(rawtext) |
|
2017 | textlen = len(rawtext) | |
2018 |
|
2018 | |||
2019 | if deltacomputer is None: |
|
2019 | if deltacomputer is None: | |
2020 | deltacomputer = deltautil.deltacomputer(self) |
|
2020 | deltacomputer = deltautil.deltacomputer(self) | |
2021 |
|
2021 | |||
2022 | revinfo = _revisioninfo(node, p1, p2, btext, textlen, cachedelta, flags) |
|
2022 | revinfo = _revisioninfo(node, p1, p2, btext, textlen, cachedelta, flags) | |
2023 |
|
2023 | |||
2024 | deltainfo = deltacomputer.finddeltainfo(revinfo, fh) |
|
2024 | deltainfo = deltacomputer.finddeltainfo(revinfo, fh) | |
2025 |
|
2025 | |||
2026 | e = (offset_type(offset, flags), deltainfo.deltalen, textlen, |
|
2026 | e = (offset_type(offset, flags), deltainfo.deltalen, textlen, | |
2027 | deltainfo.base, link, p1r, p2r, node) |
|
2027 | deltainfo.base, link, p1r, p2r, node) | |
2028 | self.index.append(e) |
|
2028 | self.index.append(e) | |
2029 | self.nodemap[node] = curr |
|
2029 | self.nodemap[node] = curr | |
2030 |
|
2030 | |||
2031 | # Reset the pure node cache start lookup offset to account for new |
|
2031 | # Reset the pure node cache start lookup offset to account for new | |
2032 | # revision. |
|
2032 | # revision. | |
2033 | if self._nodepos is not None: |
|
2033 | if self._nodepos is not None: | |
2034 | self._nodepos = curr |
|
2034 | self._nodepos = curr | |
2035 |
|
2035 | |||
2036 | entry = self._io.packentry(e, self.node, self.version, curr) |
|
2036 | entry = self._io.packentry(e, self.node, self.version, curr) | |
2037 | self._writeentry(transaction, ifh, dfh, entry, deltainfo.data, |
|
2037 | self._writeentry(transaction, ifh, dfh, entry, deltainfo.data, | |
2038 | link, offset) |
|
2038 | link, offset) | |
2039 |
|
2039 | |||
2040 | rawtext = btext[0] |
|
2040 | rawtext = btext[0] | |
2041 |
|
2041 | |||
2042 | if alwayscache and rawtext is None: |
|
2042 | if alwayscache and rawtext is None: | |
2043 | rawtext = deltacomputer.buildtext(revinfo, fh) |
|
2043 | rawtext = deltacomputer.buildtext(revinfo, fh) | |
2044 |
|
2044 | |||
2045 | if type(rawtext) == bytes: # only accept immutable objects |
|
2045 | if type(rawtext) == bytes: # only accept immutable objects | |
2046 | self._revisioncache = (node, curr, rawtext) |
|
2046 | self._revisioncache = (node, curr, rawtext) | |
2047 | self._chainbasecache[curr] = deltainfo.chainbase |
|
2047 | self._chainbasecache[curr] = deltainfo.chainbase | |
2048 | return node |
|
2048 | return node | |
2049 |
|
2049 | |||
2050 | def _writeentry(self, transaction, ifh, dfh, entry, data, link, offset): |
|
2050 | def _writeentry(self, transaction, ifh, dfh, entry, data, link, offset): | |
2051 | # Files opened in a+ mode have inconsistent behavior on various |
|
2051 | # Files opened in a+ mode have inconsistent behavior on various | |
2052 | # platforms. Windows requires that a file positioning call be made |
|
2052 | # platforms. Windows requires that a file positioning call be made | |
2053 | # when the file handle transitions between reads and writes. See |
|
2053 | # when the file handle transitions between reads and writes. See | |
2054 | # 3686fa2b8eee and the mixedfilemodewrapper in windows.py. On other |
|
2054 | # 3686fa2b8eee and the mixedfilemodewrapper in windows.py. On other | |
2055 | # platforms, Python or the platform itself can be buggy. Some versions |
|
2055 | # platforms, Python or the platform itself can be buggy. Some versions | |
2056 | # of Solaris have been observed to not append at the end of the file |
|
2056 | # of Solaris have been observed to not append at the end of the file | |
2057 | # if the file was seeked to before the end. See issue4943 for more. |
|
2057 | # if the file was seeked to before the end. See issue4943 for more. | |
2058 | # |
|
2058 | # | |
2059 | # We work around this issue by inserting a seek() before writing. |
|
2059 | # We work around this issue by inserting a seek() before writing. | |
2060 | # Note: This is likely not necessary on Python 3. However, because |
|
2060 | # Note: This is likely not necessary on Python 3. However, because | |
2061 | # the file handle is reused for reads and may be seeked there, we need |
|
2061 | # the file handle is reused for reads and may be seeked there, we need | |
2062 | # to be careful before changing this. |
|
2062 | # to be careful before changing this. | |
2063 | ifh.seek(0, os.SEEK_END) |
|
2063 | ifh.seek(0, os.SEEK_END) | |
2064 | if dfh: |
|
2064 | if dfh: | |
2065 | dfh.seek(0, os.SEEK_END) |
|
2065 | dfh.seek(0, os.SEEK_END) | |
2066 |
|
2066 | |||
2067 | curr = len(self) - 1 |
|
2067 | curr = len(self) - 1 | |
2068 | if not self._inline: |
|
2068 | if not self._inline: | |
2069 | transaction.add(self.datafile, offset) |
|
2069 | transaction.add(self.datafile, offset) | |
2070 | transaction.add(self.indexfile, curr * len(entry)) |
|
2070 | transaction.add(self.indexfile, curr * len(entry)) | |
2071 | if data[0]: |
|
2071 | if data[0]: | |
2072 | dfh.write(data[0]) |
|
2072 | dfh.write(data[0]) | |
2073 | dfh.write(data[1]) |
|
2073 | dfh.write(data[1]) | |
2074 | ifh.write(entry) |
|
2074 | ifh.write(entry) | |
2075 | else: |
|
2075 | else: | |
2076 | offset += curr * self._io.size |
|
2076 | offset += curr * self._io.size | |
2077 | transaction.add(self.indexfile, offset, curr) |
|
2077 | transaction.add(self.indexfile, offset, curr) | |
2078 | ifh.write(entry) |
|
2078 | ifh.write(entry) | |
2079 | ifh.write(data[0]) |
|
2079 | ifh.write(data[0]) | |
2080 | ifh.write(data[1]) |
|
2080 | ifh.write(data[1]) | |
2081 | self._enforceinlinesize(transaction, ifh) |
|
2081 | self._enforceinlinesize(transaction, ifh) | |
2082 |
|
2082 | |||
2083 | def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None): |
|
2083 | def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None): | |
2084 | """ |
|
2084 | """ | |
2085 | add a delta group |
|
2085 | add a delta group | |
2086 |
|
2086 | |||
2087 | given a set of deltas, add them to the revision log. the |
|
2087 | given a set of deltas, add them to the revision log. the | |
2088 | first delta is against its parent, which should be in our |
|
2088 | first delta is against its parent, which should be in our | |
2089 | log, the rest are against the previous delta. |
|
2089 | log, the rest are against the previous delta. | |
2090 |
|
2090 | |||
2091 | If ``addrevisioncb`` is defined, it will be called with arguments of |
|
2091 | If ``addrevisioncb`` is defined, it will be called with arguments of | |
2092 | this revlog and the node that was added. |
|
2092 | this revlog and the node that was added. | |
2093 | """ |
|
2093 | """ | |
2094 |
|
2094 | |||
2095 | if self._writinghandles: |
|
2095 | if self._writinghandles: | |
2096 | raise error.ProgrammingError('cannot nest addgroup() calls') |
|
2096 | raise error.ProgrammingError('cannot nest addgroup() calls') | |
2097 |
|
2097 | |||
2098 | nodes = [] |
|
2098 | nodes = [] | |
2099 |
|
2099 | |||
2100 | r = len(self) |
|
2100 | r = len(self) | |
2101 | end = 0 |
|
2101 | end = 0 | |
2102 | if r: |
|
2102 | if r: | |
2103 | end = self.end(r - 1) |
|
2103 | end = self.end(r - 1) | |
2104 | ifh = self._indexfp("a+") |
|
2104 | ifh = self._indexfp("a+") | |
2105 | isize = r * self._io.size |
|
2105 | isize = r * self._io.size | |
2106 | if self._inline: |
|
2106 | if self._inline: | |
2107 | transaction.add(self.indexfile, end + isize, r) |
|
2107 | transaction.add(self.indexfile, end + isize, r) | |
2108 | dfh = None |
|
2108 | dfh = None | |
2109 | else: |
|
2109 | else: | |
2110 | transaction.add(self.indexfile, isize, r) |
|
2110 | transaction.add(self.indexfile, isize, r) | |
2111 | transaction.add(self.datafile, end) |
|
2111 | transaction.add(self.datafile, end) | |
2112 | dfh = self._datafp("a+") |
|
2112 | dfh = self._datafp("a+") | |
2113 | def flush(): |
|
2113 | def flush(): | |
2114 | if dfh: |
|
2114 | if dfh: | |
2115 | dfh.flush() |
|
2115 | dfh.flush() | |
2116 | ifh.flush() |
|
2116 | ifh.flush() | |
2117 |
|
2117 | |||
2118 | self._writinghandles = (ifh, dfh) |
|
2118 | self._writinghandles = (ifh, dfh) | |
2119 |
|
2119 | |||
2120 | try: |
|
2120 | try: | |
2121 | deltacomputer = deltautil.deltacomputer(self) |
|
2121 | deltacomputer = deltautil.deltacomputer(self) | |
2122 | # loop through our set of deltas |
|
2122 | # loop through our set of deltas | |
2123 | for data in deltas: |
|
2123 | for data in deltas: | |
2124 | node, p1, p2, linknode, deltabase, delta, flags = data |
|
2124 | node, p1, p2, linknode, deltabase, delta, flags = data | |
2125 | link = linkmapper(linknode) |
|
2125 | link = linkmapper(linknode) | |
2126 | flags = flags or REVIDX_DEFAULT_FLAGS |
|
2126 | flags = flags or REVIDX_DEFAULT_FLAGS | |
2127 |
|
2127 | |||
2128 | nodes.append(node) |
|
2128 | nodes.append(node) | |
2129 |
|
2129 | |||
2130 | if node in self.nodemap: |
|
2130 | if node in self.nodemap: | |
2131 | self._nodeduplicatecallback(transaction, node) |
|
2131 | self._nodeduplicatecallback(transaction, node) | |
2132 | # this can happen if two branches make the same change |
|
2132 | # this can happen if two branches make the same change | |
2133 | continue |
|
2133 | continue | |
2134 |
|
2134 | |||
2135 | for p in (p1, p2): |
|
2135 | for p in (p1, p2): | |
2136 | if p not in self.nodemap: |
|
2136 | if p not in self.nodemap: | |
2137 | raise error.LookupError(p, self.indexfile, |
|
2137 | raise error.LookupError(p, self.indexfile, | |
2138 | _('unknown parent')) |
|
2138 | _('unknown parent')) | |
2139 |
|
2139 | |||
2140 | if deltabase not in self.nodemap: |
|
2140 | if deltabase not in self.nodemap: | |
2141 | raise error.LookupError(deltabase, self.indexfile, |
|
2141 | raise error.LookupError(deltabase, self.indexfile, | |
2142 | _('unknown delta base')) |
|
2142 | _('unknown delta base')) | |
2143 |
|
2143 | |||
2144 | baserev = self.rev(deltabase) |
|
2144 | baserev = self.rev(deltabase) | |
2145 |
|
2145 | |||
2146 | if baserev != nullrev and self.iscensored(baserev): |
|
2146 | if baserev != nullrev and self.iscensored(baserev): | |
2147 | # if base is censored, delta must be full replacement in a |
|
2147 | # if base is censored, delta must be full replacement in a | |
2148 | # single patch operation |
|
2148 | # single patch operation | |
2149 | hlen = struct.calcsize(">lll") |
|
2149 | hlen = struct.calcsize(">lll") | |
2150 | oldlen = self.rawsize(baserev) |
|
2150 | oldlen = self.rawsize(baserev) | |
2151 | newlen = len(delta) - hlen |
|
2151 | newlen = len(delta) - hlen | |
2152 | if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen): |
|
2152 | if delta[:hlen] != mdiff.replacediffheader(oldlen, newlen): | |
2153 | raise error.CensoredBaseError(self.indexfile, |
|
2153 | raise error.CensoredBaseError(self.indexfile, | |
2154 | self.node(baserev)) |
|
2154 | self.node(baserev)) | |
2155 |
|
2155 | |||
2156 | if not flags and self._peek_iscensored(baserev, delta, flush): |
|
2156 | if not flags and self._peek_iscensored(baserev, delta, flush): | |
2157 | flags |= REVIDX_ISCENSORED |
|
2157 | flags |= REVIDX_ISCENSORED | |
2158 |
|
2158 | |||
2159 | # We assume consumers of addrevisioncb will want to retrieve |
|
2159 | # We assume consumers of addrevisioncb will want to retrieve | |
2160 | # the added revision, which will require a call to |
|
2160 | # the added revision, which will require a call to | |
2161 | # revision(). revision() will fast path if there is a cache |
|
2161 | # revision(). revision() will fast path if there is a cache | |
2162 | # hit. So, we tell _addrevision() to always cache in this case. |
|
2162 | # hit. So, we tell _addrevision() to always cache in this case. | |
2163 | # We're only using addgroup() in the context of changegroup |
|
2163 | # We're only using addgroup() in the context of changegroup | |
2164 | # generation so the revision data can always be handled as raw |
|
2164 | # generation so the revision data can always be handled as raw | |
2165 | # by the flagprocessor. |
|
2165 | # by the flagprocessor. | |
2166 | self._addrevision(node, None, transaction, link, |
|
2166 | self._addrevision(node, None, transaction, link, | |
2167 | p1, p2, flags, (baserev, delta), |
|
2167 | p1, p2, flags, (baserev, delta), | |
2168 | ifh, dfh, |
|
2168 | ifh, dfh, | |
2169 | alwayscache=bool(addrevisioncb), |
|
2169 | alwayscache=bool(addrevisioncb), | |
2170 | deltacomputer=deltacomputer) |
|
2170 | deltacomputer=deltacomputer) | |
2171 |
|
2171 | |||
2172 | if addrevisioncb: |
|
2172 | if addrevisioncb: | |
2173 | addrevisioncb(self, node) |
|
2173 | addrevisioncb(self, node) | |
2174 |
|
2174 | |||
2175 | if not dfh and not self._inline: |
|
2175 | if not dfh and not self._inline: | |
2176 | # addrevision switched from inline to conventional |
|
2176 | # addrevision switched from inline to conventional | |
2177 | # reopen the index |
|
2177 | # reopen the index | |
2178 | ifh.close() |
|
2178 | ifh.close() | |
2179 | dfh = self._datafp("a+") |
|
2179 | dfh = self._datafp("a+") | |
2180 | ifh = self._indexfp("a+") |
|
2180 | ifh = self._indexfp("a+") | |
2181 | self._writinghandles = (ifh, dfh) |
|
2181 | self._writinghandles = (ifh, dfh) | |
2182 | finally: |
|
2182 | finally: | |
2183 | self._writinghandles = None |
|
2183 | self._writinghandles = None | |
2184 |
|
2184 | |||
2185 | if dfh: |
|
2185 | if dfh: | |
2186 | dfh.close() |
|
2186 | dfh.close() | |
2187 | ifh.close() |
|
2187 | ifh.close() | |
2188 |
|
2188 | |||
2189 | return nodes |
|
2189 | return nodes | |
2190 |
|
2190 | |||
2191 | def iscensored(self, rev): |
|
2191 | def iscensored(self, rev): | |
2192 | """Check if a file revision is censored.""" |
|
2192 | """Check if a file revision is censored.""" | |
2193 | if not self._censorable: |
|
2193 | if not self._censorable: | |
2194 | return False |
|
2194 | return False | |
2195 |
|
2195 | |||
2196 | return self.flags(rev) & REVIDX_ISCENSORED |
|
2196 | return self.flags(rev) & REVIDX_ISCENSORED | |
2197 |
|
2197 | |||
2198 | def _peek_iscensored(self, baserev, delta, flush): |
|
2198 | def _peek_iscensored(self, baserev, delta, flush): | |
2199 | """Quickly check if a delta produces a censored revision.""" |
|
2199 | """Quickly check if a delta produces a censored revision.""" | |
2200 | if not self._censorable: |
|
2200 | if not self._censorable: | |
2201 | return False |
|
2201 | return False | |
2202 |
|
2202 | |||
2203 | return storageutil.deltaiscensored(delta, baserev, self.rawsize) |
|
2203 | return storageutil.deltaiscensored(delta, baserev, self.rawsize) | |
2204 |
|
2204 | |||
2205 | def getstrippoint(self, minlink): |
|
2205 | def getstrippoint(self, minlink): | |
2206 | """find the minimum rev that must be stripped to strip the linkrev |
|
2206 | """find the minimum rev that must be stripped to strip the linkrev | |
2207 |
|
2207 | |||
2208 | Returns a tuple containing the minimum rev and a set of all revs that |
|
2208 | Returns a tuple containing the minimum rev and a set of all revs that | |
2209 | have linkrevs that will be broken by this strip. |
|
2209 | have linkrevs that will be broken by this strip. | |
2210 | """ |
|
2210 | """ | |
2211 | return storageutil.resolvestripinfo(minlink, len(self) - 1, |
|
2211 | return storageutil.resolvestripinfo(minlink, len(self) - 1, | |
2212 | self.headrevs(), |
|
2212 | self.headrevs(), | |
2213 | self.linkrev, self.parentrevs) |
|
2213 | self.linkrev, self.parentrevs) | |
2214 |
|
2214 | |||
2215 | def strip(self, minlink, transaction): |
|
2215 | def strip(self, minlink, transaction): | |
2216 | """truncate the revlog on the first revision with a linkrev >= minlink |
|
2216 | """truncate the revlog on the first revision with a linkrev >= minlink | |
2217 |
|
2217 | |||
2218 | This function is called when we're stripping revision minlink and |
|
2218 | This function is called when we're stripping revision minlink and | |
2219 | its descendants from the repository. |
|
2219 | its descendants from the repository. | |
2220 |
|
2220 | |||
2221 | We have to remove all revisions with linkrev >= minlink, because |
|
2221 | We have to remove all revisions with linkrev >= minlink, because | |
2222 | the equivalent changelog revisions will be renumbered after the |
|
2222 | the equivalent changelog revisions will be renumbered after the | |
2223 | strip. |
|
2223 | strip. | |
2224 |
|
2224 | |||
2225 | So we truncate the revlog on the first of these revisions, and |
|
2225 | So we truncate the revlog on the first of these revisions, and | |
2226 | trust that the caller has saved the revisions that shouldn't be |
|
2226 | trust that the caller has saved the revisions that shouldn't be | |
2227 | removed and that it'll re-add them after this truncation. |
|
2227 | removed and that it'll re-add them after this truncation. | |
2228 | """ |
|
2228 | """ | |
2229 | if len(self) == 0: |
|
2229 | if len(self) == 0: | |
2230 | return |
|
2230 | return | |
2231 |
|
2231 | |||
2232 | rev, _ = self.getstrippoint(minlink) |
|
2232 | rev, _ = self.getstrippoint(minlink) | |
2233 | if rev == len(self): |
|
2233 | if rev == len(self): | |
2234 | return |
|
2234 | return | |
2235 |
|
2235 | |||
2236 | # first truncate the files on disk |
|
2236 | # first truncate the files on disk | |
2237 | end = self.start(rev) |
|
2237 | end = self.start(rev) | |
2238 | if not self._inline: |
|
2238 | if not self._inline: | |
2239 | transaction.add(self.datafile, end) |
|
2239 | transaction.add(self.datafile, end) | |
2240 | end = rev * self._io.size |
|
2240 | end = rev * self._io.size | |
2241 | else: |
|
2241 | else: | |
2242 | end += rev * self._io.size |
|
2242 | end += rev * self._io.size | |
2243 |
|
2243 | |||
2244 | transaction.add(self.indexfile, end) |
|
2244 | transaction.add(self.indexfile, end) | |
2245 |
|
2245 | |||
2246 | # then reset internal state in memory to forget those revisions |
|
2246 | # then reset internal state in memory to forget those revisions | |
2247 | self._revisioncache = None |
|
2247 | self._revisioncache = None | |
2248 | self._chaininfocache = {} |
|
2248 | self._chaininfocache = {} | |
2249 | self._chunkclear() |
|
2249 | self._chunkclear() | |
2250 | for x in pycompat.xrange(rev, len(self)): |
|
2250 | for x in pycompat.xrange(rev, len(self)): | |
2251 | del self.nodemap[self.node(x)] |
|
2251 | del self.nodemap[self.node(x)] | |
2252 |
|
2252 | |||
2253 | del self.index[rev:-1] |
|
2253 | del self.index[rev:-1] | |
2254 | self._nodepos = None |
|
2254 | self._nodepos = None | |
2255 |
|
2255 | |||
2256 | def checksize(self): |
|
2256 | def checksize(self): | |
2257 | """Check size of index and data files |
|
2257 | """Check size of index and data files | |
2258 |
|
2258 | |||
2259 | return a (dd, di) tuple. |
|
2259 | return a (dd, di) tuple. | |
2260 | - dd: extra bytes for the "data" file |
|
2260 | - dd: extra bytes for the "data" file | |
2261 | - di: extra bytes for the "index" file |
|
2261 | - di: extra bytes for the "index" file | |
2262 |
|
2262 | |||
2263 | A healthy revlog will return (0, 0). |
|
2263 | A healthy revlog will return (0, 0). | |
2264 | """ |
|
2264 | """ | |
2265 | expected = 0 |
|
2265 | expected = 0 | |
2266 | if len(self): |
|
2266 | if len(self): | |
2267 | expected = max(0, self.end(len(self) - 1)) |
|
2267 | expected = max(0, self.end(len(self) - 1)) | |
2268 |
|
2268 | |||
2269 | try: |
|
2269 | try: | |
2270 | with self._datafp() as f: |
|
2270 | with self._datafp() as f: | |
2271 | f.seek(0, io.SEEK_END) |
|
2271 | f.seek(0, io.SEEK_END) | |
2272 | actual = f.tell() |
|
2272 | actual = f.tell() | |
2273 | dd = actual - expected |
|
2273 | dd = actual - expected | |
2274 | except IOError as inst: |
|
2274 | except IOError as inst: | |
2275 | if inst.errno != errno.ENOENT: |
|
2275 | if inst.errno != errno.ENOENT: | |
2276 | raise |
|
2276 | raise | |
2277 | dd = 0 |
|
2277 | dd = 0 | |
2278 |
|
2278 | |||
2279 | try: |
|
2279 | try: | |
2280 | f = self.opener(self.indexfile) |
|
2280 | f = self.opener(self.indexfile) | |
2281 | f.seek(0, io.SEEK_END) |
|
2281 | f.seek(0, io.SEEK_END) | |
2282 | actual = f.tell() |
|
2282 | actual = f.tell() | |
2283 | f.close() |
|
2283 | f.close() | |
2284 | s = self._io.size |
|
2284 | s = self._io.size | |
2285 | i = max(0, actual // s) |
|
2285 | i = max(0, actual // s) | |
2286 | di = actual - (i * s) |
|
2286 | di = actual - (i * s) | |
2287 | if self._inline: |
|
2287 | if self._inline: | |
2288 | databytes = 0 |
|
2288 | databytes = 0 | |
2289 | for r in self: |
|
2289 | for r in self: | |
2290 | databytes += max(0, self.length(r)) |
|
2290 | databytes += max(0, self.length(r)) | |
2291 | dd = 0 |
|
2291 | dd = 0 | |
2292 | di = actual - len(self) * s - databytes |
|
2292 | di = actual - len(self) * s - databytes | |
2293 | except IOError as inst: |
|
2293 | except IOError as inst: | |
2294 | if inst.errno != errno.ENOENT: |
|
2294 | if inst.errno != errno.ENOENT: | |
2295 | raise |
|
2295 | raise | |
2296 | di = 0 |
|
2296 | di = 0 | |
2297 |
|
2297 | |||
2298 | return (dd, di) |
|
2298 | return (dd, di) | |
2299 |
|
2299 | |||
2300 | def files(self): |
|
2300 | def files(self): | |
2301 | res = [self.indexfile] |
|
2301 | res = [self.indexfile] | |
2302 | if not self._inline: |
|
2302 | if not self._inline: | |
2303 | res.append(self.datafile) |
|
2303 | res.append(self.datafile) | |
2304 | return res |
|
2304 | return res | |
2305 |
|
2305 | |||
2306 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, |
|
2306 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, | |
2307 | assumehaveparentrevisions=False, |
|
2307 | assumehaveparentrevisions=False, | |
2308 | deltamode=repository.CG_DELTAMODE_STD): |
|
2308 | deltamode=repository.CG_DELTAMODE_STD): | |
2309 | if nodesorder not in ('nodes', 'storage', 'linear', None): |
|
2309 | if nodesorder not in ('nodes', 'storage', 'linear', None): | |
2310 | raise error.ProgrammingError('unhandled value for nodesorder: %s' % |
|
2310 | raise error.ProgrammingError('unhandled value for nodesorder: %s' % | |
2311 | nodesorder) |
|
2311 | nodesorder) | |
2312 |
|
2312 | |||
2313 | if nodesorder is None and not self._generaldelta: |
|
2313 | if nodesorder is None and not self._generaldelta: | |
2314 | nodesorder = 'storage' |
|
2314 | nodesorder = 'storage' | |
2315 |
|
2315 | |||
2316 | if (not self._storedeltachains and |
|
2316 | if (not self._storedeltachains and | |
2317 | deltamode != repository.CG_DELTAMODE_PREV): |
|
2317 | deltamode != repository.CG_DELTAMODE_PREV): | |
2318 | deltamode = repository.CG_DELTAMODE_FULL |
|
2318 | deltamode = repository.CG_DELTAMODE_FULL | |
2319 |
|
2319 | |||
2320 | return storageutil.emitrevisions( |
|
2320 | return storageutil.emitrevisions( | |
2321 | self, nodes, nodesorder, revlogrevisiondelta, |
|
2321 | self, nodes, nodesorder, revlogrevisiondelta, | |
2322 | deltaparentfn=self.deltaparent, |
|
2322 | deltaparentfn=self.deltaparent, | |
2323 | candeltafn=self.candelta, |
|
2323 | candeltafn=self.candelta, | |
2324 | rawsizefn=self.rawsize, |
|
2324 | rawsizefn=self.rawsize, | |
2325 | revdifffn=self.revdiff, |
|
2325 | revdifffn=self.revdiff, | |
2326 | flagsfn=self.flags, |
|
2326 | flagsfn=self.flags, | |
2327 | deltamode=deltamode, |
|
2327 | deltamode=deltamode, | |
2328 | revisiondata=revisiondata, |
|
2328 | revisiondata=revisiondata, | |
2329 | assumehaveparentrevisions=assumehaveparentrevisions) |
|
2329 | assumehaveparentrevisions=assumehaveparentrevisions) | |
2330 |
|
2330 | |||
2331 | DELTAREUSEALWAYS = 'always' |
|
2331 | DELTAREUSEALWAYS = 'always' | |
2332 | DELTAREUSESAMEREVS = 'samerevs' |
|
2332 | DELTAREUSESAMEREVS = 'samerevs' | |
2333 | DELTAREUSENEVER = 'never' |
|
2333 | DELTAREUSENEVER = 'never' | |
2334 |
|
2334 | |||
2335 | DELTAREUSEFULLADD = 'fulladd' |
|
2335 | DELTAREUSEFULLADD = 'fulladd' | |
2336 |
|
2336 | |||
2337 | DELTAREUSEALL = {'always', 'samerevs', 'never', 'fulladd'} |
|
2337 | DELTAREUSEALL = {'always', 'samerevs', 'never', 'fulladd'} | |
2338 |
|
2338 | |||
2339 | def clone(self, tr, destrevlog, addrevisioncb=None, |
|
2339 | def clone(self, tr, destrevlog, addrevisioncb=None, | |
2340 | deltareuse=DELTAREUSESAMEREVS, forcedeltabothparents=None): |
|
2340 | deltareuse=DELTAREUSESAMEREVS, forcedeltabothparents=None): | |
2341 | """Copy this revlog to another, possibly with format changes. |
|
2341 | """Copy this revlog to another, possibly with format changes. | |
2342 |
|
2342 | |||
2343 | The destination revlog will contain the same revisions and nodes. |
|
2343 | The destination revlog will contain the same revisions and nodes. | |
2344 | However, it may not be bit-for-bit identical due to e.g. delta encoding |
|
2344 | However, it may not be bit-for-bit identical due to e.g. delta encoding | |
2345 | differences. |
|
2345 | differences. | |
2346 |
|
2346 | |||
2347 | The ``deltareuse`` argument control how deltas from the existing revlog |
|
2347 | The ``deltareuse`` argument control how deltas from the existing revlog | |
2348 | are preserved in the destination revlog. The argument can have the |
|
2348 | are preserved in the destination revlog. The argument can have the | |
2349 | following values: |
|
2349 | following values: | |
2350 |
|
2350 | |||
2351 | DELTAREUSEALWAYS |
|
2351 | DELTAREUSEALWAYS | |
2352 | Deltas will always be reused (if possible), even if the destination |
|
2352 | Deltas will always be reused (if possible), even if the destination | |
2353 | revlog would not select the same revisions for the delta. This is the |
|
2353 | revlog would not select the same revisions for the delta. This is the | |
2354 | fastest mode of operation. |
|
2354 | fastest mode of operation. | |
2355 | DELTAREUSESAMEREVS |
|
2355 | DELTAREUSESAMEREVS | |
2356 | Deltas will be reused if the destination revlog would pick the same |
|
2356 | Deltas will be reused if the destination revlog would pick the same | |
2357 | revisions for the delta. This mode strikes a balance between speed |
|
2357 | revisions for the delta. This mode strikes a balance between speed | |
2358 | and optimization. |
|
2358 | and optimization. | |
2359 | DELTAREUSENEVER |
|
2359 | DELTAREUSENEVER | |
2360 | Deltas will never be reused. This is the slowest mode of execution. |
|
2360 | Deltas will never be reused. This is the slowest mode of execution. | |
2361 | This mode can be used to recompute deltas (e.g. if the diff/delta |
|
2361 | This mode can be used to recompute deltas (e.g. if the diff/delta | |
2362 | algorithm changes). |
|
2362 | algorithm changes). | |
2363 |
|
2363 | |||
2364 | Delta computation can be slow, so the choice of delta reuse policy can |
|
2364 | Delta computation can be slow, so the choice of delta reuse policy can | |
2365 | significantly affect run time. |
|
2365 | significantly affect run time. | |
2366 |
|
2366 | |||
2367 | The default policy (``DELTAREUSESAMEREVS``) strikes a balance between |
|
2367 | The default policy (``DELTAREUSESAMEREVS``) strikes a balance between | |
2368 | two extremes. Deltas will be reused if they are appropriate. But if the |
|
2368 | two extremes. Deltas will be reused if they are appropriate. But if the | |
2369 | delta could choose a better revision, it will do so. This means if you |
|
2369 | delta could choose a better revision, it will do so. This means if you | |
2370 | are converting a non-generaldelta revlog to a generaldelta revlog, |
|
2370 | are converting a non-generaldelta revlog to a generaldelta revlog, | |
2371 | deltas will be recomputed if the delta's parent isn't a parent of the |
|
2371 | deltas will be recomputed if the delta's parent isn't a parent of the | |
2372 | revision. |
|
2372 | revision. | |
2373 |
|
2373 | |||
2374 | In addition to the delta policy, the ``forcedeltabothparents`` |
|
2374 | In addition to the delta policy, the ``forcedeltabothparents`` | |
2375 | argument controls whether to force compute deltas against both parents |
|
2375 | argument controls whether to force compute deltas against both parents | |
2376 | for merges. By default, the current default is used. |
|
2376 | for merges. By default, the current default is used. | |
2377 | """ |
|
2377 | """ | |
2378 | if deltareuse not in self.DELTAREUSEALL: |
|
2378 | if deltareuse not in self.DELTAREUSEALL: | |
2379 | raise ValueError(_('value for deltareuse invalid: %s') % deltareuse) |
|
2379 | raise ValueError(_('value for deltareuse invalid: %s') % deltareuse) | |
2380 |
|
2380 | |||
2381 | if len(destrevlog): |
|
2381 | if len(destrevlog): | |
2382 | raise ValueError(_('destination revlog is not empty')) |
|
2382 | raise ValueError(_('destination revlog is not empty')) | |
2383 |
|
2383 | |||
2384 | if getattr(self, 'filteredrevs', None): |
|
2384 | if getattr(self, 'filteredrevs', None): | |
2385 | raise ValueError(_('source revlog has filtered revisions')) |
|
2385 | raise ValueError(_('source revlog has filtered revisions')) | |
2386 | if getattr(destrevlog, 'filteredrevs', None): |
|
2386 | if getattr(destrevlog, 'filteredrevs', None): | |
2387 | raise ValueError(_('destination revlog has filtered revisions')) |
|
2387 | raise ValueError(_('destination revlog has filtered revisions')) | |
2388 |
|
2388 | |||
2389 | # lazydelta and lazydeltabase controls whether to reuse a cached delta, |
|
2389 | # lazydelta and lazydeltabase controls whether to reuse a cached delta, | |
2390 | # if possible. |
|
2390 | # if possible. | |
2391 | oldlazydelta = destrevlog._lazydelta |
|
2391 | oldlazydelta = destrevlog._lazydelta | |
2392 | oldlazydeltabase = destrevlog._lazydeltabase |
|
2392 | oldlazydeltabase = destrevlog._lazydeltabase | |
2393 | oldamd = destrevlog._deltabothparents |
|
2393 | oldamd = destrevlog._deltabothparents | |
2394 |
|
2394 | |||
2395 | try: |
|
2395 | try: | |
2396 | if deltareuse == self.DELTAREUSEALWAYS: |
|
2396 | if deltareuse == self.DELTAREUSEALWAYS: | |
2397 | destrevlog._lazydeltabase = True |
|
2397 | destrevlog._lazydeltabase = True | |
2398 | destrevlog._lazydelta = True |
|
2398 | destrevlog._lazydelta = True | |
2399 | elif deltareuse == self.DELTAREUSESAMEREVS: |
|
2399 | elif deltareuse == self.DELTAREUSESAMEREVS: | |
2400 | destrevlog._lazydeltabase = False |
|
2400 | destrevlog._lazydeltabase = False | |
2401 | destrevlog._lazydelta = True |
|
2401 | destrevlog._lazydelta = True | |
2402 | elif deltareuse == self.DELTAREUSENEVER: |
|
2402 | elif deltareuse == self.DELTAREUSENEVER: | |
2403 | destrevlog._lazydeltabase = False |
|
2403 | destrevlog._lazydeltabase = False | |
2404 | destrevlog._lazydelta = False |
|
2404 | destrevlog._lazydelta = False | |
2405 |
|
2405 | |||
2406 | destrevlog._deltabothparents = forcedeltabothparents or oldamd |
|
2406 | destrevlog._deltabothparents = forcedeltabothparents or oldamd | |
2407 |
|
2407 | |||
2408 | deltacomputer = deltautil.deltacomputer(destrevlog) |
|
2408 | deltacomputer = deltautil.deltacomputer(destrevlog) | |
2409 | index = self.index |
|
2409 | index = self.index | |
2410 | for rev in self: |
|
2410 | for rev in self: | |
2411 | entry = index[rev] |
|
2411 | entry = index[rev] | |
2412 |
|
2412 | |||
2413 | # Some classes override linkrev to take filtered revs into |
|
2413 | # Some classes override linkrev to take filtered revs into | |
2414 | # account. Use raw entry from index. |
|
2414 | # account. Use raw entry from index. | |
2415 | flags = entry[0] & 0xffff |
|
2415 | flags = entry[0] & 0xffff | |
2416 | linkrev = entry[4] |
|
2416 | linkrev = entry[4] | |
2417 | p1 = index[entry[5]][7] |
|
2417 | p1 = index[entry[5]][7] | |
2418 | p2 = index[entry[6]][7] |
|
2418 | p2 = index[entry[6]][7] | |
2419 | node = entry[7] |
|
2419 | node = entry[7] | |
2420 |
|
2420 | |||
2421 | # (Possibly) reuse the delta from the revlog if allowed and |
|
2421 | # (Possibly) reuse the delta from the revlog if allowed and | |
2422 | # the revlog chunk is a delta. |
|
2422 | # the revlog chunk is a delta. | |
2423 | cachedelta = None |
|
2423 | cachedelta = None | |
2424 | rawtext = None |
|
2424 | rawtext = None | |
2425 | if (deltareuse != self.DELTAREUSEFULLADD |
|
2425 | if (deltareuse != self.DELTAREUSEFULLADD | |
2426 | and destrevlog._lazydelta): |
|
2426 | and destrevlog._lazydelta): | |
2427 | dp = self.deltaparent(rev) |
|
2427 | dp = self.deltaparent(rev) | |
2428 | if dp != nullrev: |
|
2428 | if dp != nullrev: | |
2429 | cachedelta = (dp, bytes(self._chunk(rev))) |
|
2429 | cachedelta = (dp, bytes(self._chunk(rev))) | |
2430 |
|
2430 | |||
2431 | if not cachedelta: |
|
2431 | if not cachedelta: | |
2432 | rawtext = self.rawdata(rev) |
|
2432 | rawtext = self.rawdata(rev) | |
2433 |
|
2433 | |||
2434 |
|
2434 | |||
2435 | if deltareuse == self.DELTAREUSEFULLADD: |
|
2435 | if deltareuse == self.DELTAREUSEFULLADD: | |
2436 | destrevlog.addrevision(rawtext, tr, linkrev, p1, p2, |
|
2436 | destrevlog.addrevision(rawtext, tr, linkrev, p1, p2, | |
2437 | cachedelta=cachedelta, |
|
2437 | cachedelta=cachedelta, | |
2438 | node=node, flags=flags, |
|
2438 | node=node, flags=flags, | |
2439 | deltacomputer=deltacomputer) |
|
2439 | deltacomputer=deltacomputer) | |
2440 | else: |
|
2440 | else: | |
2441 | ifh = destrevlog.opener(destrevlog.indexfile, 'a+', |
|
2441 | ifh = destrevlog.opener(destrevlog.indexfile, 'a+', | |
2442 | checkambig=False) |
|
2442 | checkambig=False) | |
2443 | dfh = None |
|
2443 | dfh = None | |
2444 | if not destrevlog._inline: |
|
2444 | if not destrevlog._inline: | |
2445 | dfh = destrevlog.opener(destrevlog.datafile, 'a+') |
|
2445 | dfh = destrevlog.opener(destrevlog.datafile, 'a+') | |
2446 | try: |
|
2446 | try: | |
2447 | destrevlog._addrevision(node, rawtext, tr, linkrev, p1, |
|
2447 | destrevlog._addrevision(node, rawtext, tr, linkrev, p1, | |
2448 | p2, flags, cachedelta, ifh, dfh, |
|
2448 | p2, flags, cachedelta, ifh, dfh, | |
2449 | deltacomputer=deltacomputer) |
|
2449 | deltacomputer=deltacomputer) | |
2450 | finally: |
|
2450 | finally: | |
2451 | if dfh: |
|
2451 | if dfh: | |
2452 | dfh.close() |
|
2452 | dfh.close() | |
2453 | ifh.close() |
|
2453 | ifh.close() | |
2454 |
|
2454 | |||
2455 | if addrevisioncb: |
|
2455 | if addrevisioncb: | |
2456 | addrevisioncb(self, rev, node) |
|
2456 | addrevisioncb(self, rev, node) | |
2457 | finally: |
|
2457 | finally: | |
2458 | destrevlog._lazydelta = oldlazydelta |
|
2458 | destrevlog._lazydelta = oldlazydelta | |
2459 | destrevlog._lazydeltabase = oldlazydeltabase |
|
2459 | destrevlog._lazydeltabase = oldlazydeltabase | |
2460 | destrevlog._deltabothparents = oldamd |
|
2460 | destrevlog._deltabothparents = oldamd | |
2461 |
|
2461 | |||
2462 | def censorrevision(self, tr, censornode, tombstone=b''): |
|
2462 | def censorrevision(self, tr, censornode, tombstone=b''): | |
2463 | if (self.version & 0xFFFF) == REVLOGV0: |
|
2463 | if (self.version & 0xFFFF) == REVLOGV0: | |
2464 | raise error.RevlogError(_('cannot censor with version %d revlogs') % |
|
2464 | raise error.RevlogError(_('cannot censor with version %d revlogs') % | |
2465 | self.version) |
|
2465 | self.version) | |
2466 |
|
2466 | |||
2467 | censorrev = self.rev(censornode) |
|
2467 | censorrev = self.rev(censornode) | |
2468 | tombstone = storageutil.packmeta({b'censored': tombstone}, b'') |
|
2468 | tombstone = storageutil.packmeta({b'censored': tombstone}, b'') | |
2469 |
|
2469 | |||
2470 | if len(tombstone) > self.rawsize(censorrev): |
|
2470 | if len(tombstone) > self.rawsize(censorrev): | |
2471 | raise error.Abort(_('censor tombstone must be no longer than ' |
|
2471 | raise error.Abort(_('censor tombstone must be no longer than ' | |
2472 | 'censored data')) |
|
2472 | 'censored data')) | |
2473 |
|
2473 | |||
2474 | # Rewriting the revlog in place is hard. Our strategy for censoring is |
|
2474 | # Rewriting the revlog in place is hard. Our strategy for censoring is | |
2475 | # to create a new revlog, copy all revisions to it, then replace the |
|
2475 | # to create a new revlog, copy all revisions to it, then replace the | |
2476 | # revlogs on transaction close. |
|
2476 | # revlogs on transaction close. | |
2477 |
|
2477 | |||
2478 | newindexfile = self.indexfile + b'.tmpcensored' |
|
2478 | newindexfile = self.indexfile + b'.tmpcensored' | |
2479 | newdatafile = self.datafile + b'.tmpcensored' |
|
2479 | newdatafile = self.datafile + b'.tmpcensored' | |
2480 |
|
2480 | |||
2481 | # This is a bit dangerous. We could easily have a mismatch of state. |
|
2481 | # This is a bit dangerous. We could easily have a mismatch of state. | |
2482 | newrl = revlog(self.opener, newindexfile, newdatafile, |
|
2482 | newrl = revlog(self.opener, newindexfile, newdatafile, | |
2483 | censorable=True) |
|
2483 | censorable=True) | |
2484 | newrl.version = self.version |
|
2484 | newrl.version = self.version | |
2485 | newrl._generaldelta = self._generaldelta |
|
2485 | newrl._generaldelta = self._generaldelta | |
2486 | newrl._io = self._io |
|
2486 | newrl._io = self._io | |
2487 |
|
2487 | |||
2488 | for rev in self.revs(): |
|
2488 | for rev in self.revs(): | |
2489 | node = self.node(rev) |
|
2489 | node = self.node(rev) | |
2490 | p1, p2 = self.parents(node) |
|
2490 | p1, p2 = self.parents(node) | |
2491 |
|
2491 | |||
2492 | if rev == censorrev: |
|
2492 | if rev == censorrev: | |
2493 | newrl.addrawrevision(tombstone, tr, self.linkrev(censorrev), |
|
2493 | newrl.addrawrevision(tombstone, tr, self.linkrev(censorrev), | |
2494 | p1, p2, censornode, REVIDX_ISCENSORED) |
|
2494 | p1, p2, censornode, REVIDX_ISCENSORED) | |
2495 |
|
2495 | |||
2496 | if newrl.deltaparent(rev) != nullrev: |
|
2496 | if newrl.deltaparent(rev) != nullrev: | |
2497 | raise error.Abort(_('censored revision stored as delta; ' |
|
2497 | raise error.Abort(_('censored revision stored as delta; ' | |
2498 | 'cannot censor'), |
|
2498 | 'cannot censor'), | |
2499 | hint=_('censoring of revlogs is not ' |
|
2499 | hint=_('censoring of revlogs is not ' | |
2500 | 'fully implemented; please report ' |
|
2500 | 'fully implemented; please report ' | |
2501 | 'this bug')) |
|
2501 | 'this bug')) | |
2502 | continue |
|
2502 | continue | |
2503 |
|
2503 | |||
2504 | if self.iscensored(rev): |
|
2504 | if self.iscensored(rev): | |
2505 | if self.deltaparent(rev) != nullrev: |
|
2505 | if self.deltaparent(rev) != nullrev: | |
2506 | raise error.Abort(_('cannot censor due to censored ' |
|
2506 | raise error.Abort(_('cannot censor due to censored ' | |
2507 | 'revision having delta stored')) |
|
2507 | 'revision having delta stored')) | |
2508 | rawtext = self._chunk(rev) |
|
2508 | rawtext = self._chunk(rev) | |
2509 | else: |
|
2509 | else: | |
2510 | rawtext = self.rawdata(rev) |
|
2510 | rawtext = self.rawdata(rev) | |
2511 |
|
2511 | |||
2512 | newrl.addrawrevision(rawtext, tr, self.linkrev(rev), p1, p2, node, |
|
2512 | newrl.addrawrevision(rawtext, tr, self.linkrev(rev), p1, p2, node, | |
2513 | self.flags(rev)) |
|
2513 | self.flags(rev)) | |
2514 |
|
2514 | |||
2515 | tr.addbackup(self.indexfile, location='store') |
|
2515 | tr.addbackup(self.indexfile, location='store') | |
2516 | if not self._inline: |
|
2516 | if not self._inline: | |
2517 | tr.addbackup(self.datafile, location='store') |
|
2517 | tr.addbackup(self.datafile, location='store') | |
2518 |
|
2518 | |||
2519 | self.opener.rename(newrl.indexfile, self.indexfile) |
|
2519 | self.opener.rename(newrl.indexfile, self.indexfile) | |
2520 | if not self._inline: |
|
2520 | if not self._inline: | |
2521 | self.opener.rename(newrl.datafile, self.datafile) |
|
2521 | self.opener.rename(newrl.datafile, self.datafile) | |
2522 |
|
2522 | |||
2523 | self.clearcaches() |
|
2523 | self.clearcaches() | |
2524 | self._loadindex() |
|
2524 | self._loadindex() | |
2525 |
|
2525 | |||
2526 | def verifyintegrity(self, state): |
|
2526 | def verifyintegrity(self, state): | |
2527 | """Verifies the integrity of the revlog. |
|
2527 | """Verifies the integrity of the revlog. | |
2528 |
|
2528 | |||
2529 | Yields ``revlogproblem`` instances describing problems that are |
|
2529 | Yields ``revlogproblem`` instances describing problems that are | |
2530 | found. |
|
2530 | found. | |
2531 | """ |
|
2531 | """ | |
2532 | dd, di = self.checksize() |
|
2532 | dd, di = self.checksize() | |
2533 | if dd: |
|
2533 | if dd: | |
2534 | yield revlogproblem(error=_('data length off by %d bytes') % dd) |
|
2534 | yield revlogproblem(error=_('data length off by %d bytes') % dd) | |
2535 | if di: |
|
2535 | if di: | |
2536 | yield revlogproblem(error=_('index contains %d extra bytes') % di) |
|
2536 | yield revlogproblem(error=_('index contains %d extra bytes') % di) | |
2537 |
|
2537 | |||
2538 | version = self.version & 0xFFFF |
|
2538 | version = self.version & 0xFFFF | |
2539 |
|
2539 | |||
2540 | # The verifier tells us what version revlog we should be. |
|
2540 | # The verifier tells us what version revlog we should be. | |
2541 | if version != state['expectedversion']: |
|
2541 | if version != state['expectedversion']: | |
2542 | yield revlogproblem( |
|
2542 | yield revlogproblem( | |
2543 | warning=_("warning: '%s' uses revlog format %d; expected %d") % |
|
2543 | warning=_("warning: '%s' uses revlog format %d; expected %d") % | |
2544 | (self.indexfile, version, state['expectedversion'])) |
|
2544 | (self.indexfile, version, state['expectedversion'])) | |
2545 |
|
2545 | |||
2546 | state['skipread'] = set() |
|
2546 | state['skipread'] = set() | |
2547 |
|
2547 | |||
2548 | for rev in self: |
|
2548 | for rev in self: | |
2549 | node = self.node(rev) |
|
2549 | node = self.node(rev) | |
2550 |
|
2550 | |||
2551 | # Verify contents. 4 cases to care about: |
|
2551 | # Verify contents. 4 cases to care about: | |
2552 | # |
|
2552 | # | |
2553 | # common: the most common case |
|
2553 | # common: the most common case | |
2554 | # rename: with a rename |
|
2554 | # rename: with a rename | |
2555 | # meta: file content starts with b'\1\n', the metadata |
|
2555 | # meta: file content starts with b'\1\n', the metadata | |
2556 | # header defined in filelog.py, but without a rename |
|
2556 | # header defined in filelog.py, but without a rename | |
2557 | # ext: content stored externally |
|
2557 | # ext: content stored externally | |
2558 | # |
|
2558 | # | |
2559 | # More formally, their differences are shown below: |
|
2559 | # More formally, their differences are shown below: | |
2560 | # |
|
2560 | # | |
2561 | # | common | rename | meta | ext |
|
2561 | # | common | rename | meta | ext | |
2562 | # ------------------------------------------------------- |
|
2562 | # ------------------------------------------------------- | |
2563 | # flags() | 0 | 0 | 0 | not 0 |
|
2563 | # flags() | 0 | 0 | 0 | not 0 | |
2564 | # renamed() | False | True | False | ? |
|
2564 | # renamed() | False | True | False | ? | |
2565 | # rawtext[0:2]=='\1\n'| False | True | True | ? |
|
2565 | # rawtext[0:2]=='\1\n'| False | True | True | ? | |
2566 | # |
|
2566 | # | |
2567 | # "rawtext" means the raw text stored in revlog data, which |
|
2567 | # "rawtext" means the raw text stored in revlog data, which | |
2568 | # could be retrieved by "rawdata(rev)". "text" |
|
2568 | # could be retrieved by "rawdata(rev)". "text" | |
2569 | # mentioned below is "revision(rev)". |
|
2569 | # mentioned below is "revision(rev)". | |
2570 | # |
|
2570 | # | |
2571 | # There are 3 different lengths stored physically: |
|
2571 | # There are 3 different lengths stored physically: | |
2572 | # 1. L1: rawsize, stored in revlog index |
|
2572 | # 1. L1: rawsize, stored in revlog index | |
2573 | # 2. L2: len(rawtext), stored in revlog data |
|
2573 | # 2. L2: len(rawtext), stored in revlog data | |
2574 | # 3. L3: len(text), stored in revlog data if flags==0, or |
|
2574 | # 3. L3: len(text), stored in revlog data if flags==0, or | |
2575 | # possibly somewhere else if flags!=0 |
|
2575 | # possibly somewhere else if flags!=0 | |
2576 | # |
|
2576 | # | |
2577 | # L1 should be equal to L2. L3 could be different from them. |
|
2577 | # L1 should be equal to L2. L3 could be different from them. | |
2578 | # "text" may or may not affect commit hash depending on flag |
|
2578 | # "text" may or may not affect commit hash depending on flag | |
2579 | # processors (see flagutil.addflagprocessor). |
|
2579 | # processors (see flagutil.addflagprocessor). | |
2580 | # |
|
2580 | # | |
2581 | # | common | rename | meta | ext |
|
2581 | # | common | rename | meta | ext | |
2582 | # ------------------------------------------------- |
|
2582 | # ------------------------------------------------- | |
2583 | # rawsize() | L1 | L1 | L1 | L1 |
|
2583 | # rawsize() | L1 | L1 | L1 | L1 | |
2584 | # size() | L1 | L2-LM | L1(*) | L1 (?) |
|
2584 | # size() | L1 | L2-LM | L1(*) | L1 (?) | |
2585 | # len(rawtext) | L2 | L2 | L2 | L2 |
|
2585 | # len(rawtext) | L2 | L2 | L2 | L2 | |
2586 | # len(text) | L2 | L2 | L2 | L3 |
|
2586 | # len(text) | L2 | L2 | L2 | L3 | |
2587 | # len(read()) | L2 | L2-LM | L2-LM | L3 (?) |
|
2587 | # len(read()) | L2 | L2-LM | L2-LM | L3 (?) | |
2588 | # |
|
2588 | # | |
2589 | # LM: length of metadata, depending on rawtext |
|
2589 | # LM: length of metadata, depending on rawtext | |
2590 | # (*): not ideal, see comment in filelog.size |
|
2590 | # (*): not ideal, see comment in filelog.size | |
2591 | # (?): could be "- len(meta)" if the resolved content has |
|
2591 | # (?): could be "- len(meta)" if the resolved content has | |
2592 | # rename metadata |
|
2592 | # rename metadata | |
2593 | # |
|
2593 | # | |
2594 | # Checks needed to be done: |
|
2594 | # Checks needed to be done: | |
2595 | # 1. length check: L1 == L2, in all cases. |
|
2595 | # 1. length check: L1 == L2, in all cases. | |
2596 | # 2. hash check: depending on flag processor, we may need to |
|
2596 | # 2. hash check: depending on flag processor, we may need to | |
2597 | # use either "text" (external), or "rawtext" (in revlog). |
|
2597 | # use either "text" (external), or "rawtext" (in revlog). | |
2598 |
|
2598 | |||
2599 | try: |
|
2599 | try: | |
2600 | skipflags = state.get('skipflags', 0) |
|
2600 | skipflags = state.get('skipflags', 0) | |
2601 | if skipflags: |
|
2601 | if skipflags: | |
2602 | skipflags &= self.flags(rev) |
|
2602 | skipflags &= self.flags(rev) | |
2603 |
|
2603 | |||
2604 | if skipflags: |
|
2604 | if skipflags: | |
2605 | state['skipread'].add(node) |
|
2605 | state['skipread'].add(node) | |
2606 | else: |
|
2606 | else: | |
2607 | # Side-effect: read content and verify hash. |
|
2607 | # Side-effect: read content and verify hash. | |
2608 | self.revision(node) |
|
2608 | self.revision(node) | |
2609 |
|
2609 | |||
2610 | l1 = self.rawsize(rev) |
|
2610 | l1 = self.rawsize(rev) | |
2611 | l2 = len(self.rawdata(node)) |
|
2611 | l2 = len(self.rawdata(node)) | |
2612 |
|
2612 | |||
2613 | if l1 != l2: |
|
2613 | if l1 != l2: | |
2614 | yield revlogproblem( |
|
2614 | yield revlogproblem( | |
2615 | error=_('unpacked size is %d, %d expected') % (l2, l1), |
|
2615 | error=_('unpacked size is %d, %d expected') % (l2, l1), | |
2616 | node=node) |
|
2616 | node=node) | |
2617 |
|
2617 | |||
2618 | except error.CensoredNodeError: |
|
2618 | except error.CensoredNodeError: | |
2619 | if state['erroroncensored']: |
|
2619 | if state['erroroncensored']: | |
2620 | yield revlogproblem(error=_('censored file data'), |
|
2620 | yield revlogproblem(error=_('censored file data'), | |
2621 | node=node) |
|
2621 | node=node) | |
2622 | state['skipread'].add(node) |
|
2622 | state['skipread'].add(node) | |
2623 | except Exception as e: |
|
2623 | except Exception as e: | |
2624 | yield revlogproblem( |
|
2624 | yield revlogproblem( | |
2625 | error=_('unpacking %s: %s') % (short(node), |
|
2625 | error=_('unpacking %s: %s') % (short(node), | |
2626 | stringutil.forcebytestr(e)), |
|
2626 | stringutil.forcebytestr(e)), | |
2627 | node=node) |
|
2627 | node=node) | |
2628 | state['skipread'].add(node) |
|
2628 | state['skipread'].add(node) | |
2629 |
|
2629 | |||
2630 | def storageinfo(self, exclusivefiles=False, sharedfiles=False, |
|
2630 | def storageinfo(self, exclusivefiles=False, sharedfiles=False, | |
2631 | revisionscount=False, trackedsize=False, |
|
2631 | revisionscount=False, trackedsize=False, | |
2632 | storedsize=False): |
|
2632 | storedsize=False): | |
2633 | d = {} |
|
2633 | d = {} | |
2634 |
|
2634 | |||
2635 | if exclusivefiles: |
|
2635 | if exclusivefiles: | |
2636 | d['exclusivefiles'] = [(self.opener, self.indexfile)] |
|
2636 | d['exclusivefiles'] = [(self.opener, self.indexfile)] | |
2637 | if not self._inline: |
|
2637 | if not self._inline: | |
2638 | d['exclusivefiles'].append((self.opener, self.datafile)) |
|
2638 | d['exclusivefiles'].append((self.opener, self.datafile)) | |
2639 |
|
2639 | |||
2640 | if sharedfiles: |
|
2640 | if sharedfiles: | |
2641 | d['sharedfiles'] = [] |
|
2641 | d['sharedfiles'] = [] | |
2642 |
|
2642 | |||
2643 | if revisionscount: |
|
2643 | if revisionscount: | |
2644 | d['revisionscount'] = len(self) |
|
2644 | d['revisionscount'] = len(self) | |
2645 |
|
2645 | |||
2646 | if trackedsize: |
|
2646 | if trackedsize: | |
2647 | d['trackedsize'] = sum(map(self.rawsize, iter(self))) |
|
2647 | d['trackedsize'] = sum(map(self.rawsize, iter(self))) | |
2648 |
|
2648 | |||
2649 | if storedsize: |
|
2649 | if storedsize: | |
2650 | d['storedsize'] = sum(self.opener.stat(path).st_size |
|
2650 | d['storedsize'] = sum(self.opener.stat(path).st_size | |
2651 | for path in self.files()) |
|
2651 | for path in self.files()) | |
2652 |
|
2652 | |||
2653 | return d |
|
2653 | return d |
@@ -1,193 +1,185 b'' | |||||
1 | # flagutils.py - code to deal with revlog flags and their processors |
|
1 | # flagutils.py - code to deal with revlog flags and their processors | |
2 | # |
|
2 | # | |
3 | # Copyright 2016 Remi Chaintron <remi@fb.com> |
|
3 | # Copyright 2016 Remi Chaintron <remi@fb.com> | |
4 | # Copyright 2016-2019 Pierre-Yves David <pierre-yves.david@ens-lyon.org> |
|
4 | # Copyright 2016-2019 Pierre-Yves David <pierre-yves.david@ens-lyon.org> | |
5 | # |
|
5 | # | |
6 | # This software may be used and distributed according to the terms of the |
|
6 | # This software may be used and distributed according to the terms of the | |
7 | # GNU General Public License version 2 or any later version. |
|
7 | # GNU General Public License version 2 or any later version. | |
8 |
|
8 | |||
9 | from __future__ import absolute_import |
|
9 | from __future__ import absolute_import | |
10 |
|
10 | |||
11 | from ..i18n import _ |
|
11 | from ..i18n import _ | |
12 |
|
12 | |||
13 | from .constants import ( |
|
13 | from .constants import ( | |
14 | REVIDX_DEFAULT_FLAGS, |
|
14 | REVIDX_DEFAULT_FLAGS, | |
15 | REVIDX_ELLIPSIS, |
|
15 | REVIDX_ELLIPSIS, | |
16 | REVIDX_EXTSTORED, |
|
16 | REVIDX_EXTSTORED, | |
17 | REVIDX_FLAGS_ORDER, |
|
17 | REVIDX_FLAGS_ORDER, | |
18 | REVIDX_ISCENSORED, |
|
18 | REVIDX_ISCENSORED, | |
19 | REVIDX_RAWTEXT_CHANGING_FLAGS, |
|
19 | REVIDX_RAWTEXT_CHANGING_FLAGS, | |
20 | ) |
|
20 | ) | |
21 |
|
21 | |||
22 | from .. import ( |
|
22 | from .. import ( | |
23 | error, |
|
23 | error, | |
24 | util |
|
24 | util | |
25 | ) |
|
25 | ) | |
26 |
|
26 | |||
27 | # blanked usage of all the name to prevent pyflakes constraints |
|
27 | # blanked usage of all the name to prevent pyflakes constraints | |
28 | # We need these name available in the module for extensions. |
|
28 | # We need these name available in the module for extensions. | |
29 | REVIDX_ISCENSORED |
|
29 | REVIDX_ISCENSORED | |
30 | REVIDX_ELLIPSIS |
|
30 | REVIDX_ELLIPSIS | |
31 | REVIDX_EXTSTORED |
|
31 | REVIDX_EXTSTORED | |
32 | REVIDX_DEFAULT_FLAGS |
|
32 | REVIDX_DEFAULT_FLAGS | |
33 | REVIDX_FLAGS_ORDER |
|
33 | REVIDX_FLAGS_ORDER | |
34 | REVIDX_RAWTEXT_CHANGING_FLAGS |
|
34 | REVIDX_RAWTEXT_CHANGING_FLAGS | |
35 |
|
35 | |||
36 | REVIDX_KNOWN_FLAGS = util.bitsfrom(REVIDX_FLAGS_ORDER) |
|
36 | REVIDX_KNOWN_FLAGS = util.bitsfrom(REVIDX_FLAGS_ORDER) | |
37 |
|
37 | |||
38 | # Store flag processors (cf. 'addflagprocessor()' to register) |
|
38 | # Store flag processors (cf. 'addflagprocessor()' to register) | |
39 | flagprocessors = { |
|
39 | flagprocessors = { | |
40 | REVIDX_ISCENSORED: None, |
|
40 | REVIDX_ISCENSORED: None, | |
41 | } |
|
41 | } | |
42 |
|
42 | |||
43 | def addflagprocessor(flag, processor): |
|
43 | def addflagprocessor(flag, processor): | |
44 | """Register a flag processor on a revision data flag. |
|
44 | """Register a flag processor on a revision data flag. | |
45 |
|
45 | |||
46 | Invariant: |
|
46 | Invariant: | |
47 | - Flags need to be defined in REVIDX_KNOWN_FLAGS and REVIDX_FLAGS_ORDER, |
|
47 | - Flags need to be defined in REVIDX_KNOWN_FLAGS and REVIDX_FLAGS_ORDER, | |
48 | and REVIDX_RAWTEXT_CHANGING_FLAGS if they can alter rawtext. |
|
48 | and REVIDX_RAWTEXT_CHANGING_FLAGS if they can alter rawtext. | |
49 | - Only one flag processor can be registered on a specific flag. |
|
49 | - Only one flag processor can be registered on a specific flag. | |
50 | - flagprocessors must be 3-tuples of functions (read, write, raw) with the |
|
50 | - flagprocessors must be 3-tuples of functions (read, write, raw) with the | |
51 | following signatures: |
|
51 | following signatures: | |
52 | - (read) f(self, rawtext) -> text, bool |
|
52 | - (read) f(self, rawtext) -> text, bool | |
53 | - (write) f(self, text) -> rawtext, bool |
|
53 | - (write) f(self, text) -> rawtext, bool | |
54 | - (raw) f(self, rawtext) -> bool |
|
54 | - (raw) f(self, rawtext) -> bool | |
55 | "text" is presented to the user. "rawtext" is stored in revlog data, not |
|
55 | "text" is presented to the user. "rawtext" is stored in revlog data, not | |
56 | directly visible to the user. |
|
56 | directly visible to the user. | |
57 | The boolean returned by these transforms is used to determine whether |
|
57 | The boolean returned by these transforms is used to determine whether | |
58 | the returned text can be used for hash integrity checking. For example, |
|
58 | the returned text can be used for hash integrity checking. For example, | |
59 | if "write" returns False, then "text" is used to generate hash. If |
|
59 | if "write" returns False, then "text" is used to generate hash. If | |
60 | "write" returns True, that basically means "rawtext" returned by "write" |
|
60 | "write" returns True, that basically means "rawtext" returned by "write" | |
61 | should be used to generate hash. Usually, "write" and "read" return |
|
61 | should be used to generate hash. Usually, "write" and "read" return | |
62 | different booleans. And "raw" returns a same boolean as "write". |
|
62 | different booleans. And "raw" returns a same boolean as "write". | |
63 |
|
63 | |||
64 | Note: The 'raw' transform is used for changegroup generation and in some |
|
64 | Note: The 'raw' transform is used for changegroup generation and in some | |
65 | debug commands. In this case the transform only indicates whether the |
|
65 | debug commands. In this case the transform only indicates whether the | |
66 | contents can be used for hash integrity checks. |
|
66 | contents can be used for hash integrity checks. | |
67 | """ |
|
67 | """ | |
68 | insertflagprocessor(flag, processor, flagprocessors) |
|
68 | insertflagprocessor(flag, processor, flagprocessors) | |
69 |
|
69 | |||
70 | def insertflagprocessor(flag, processor, flagprocessors): |
|
70 | def insertflagprocessor(flag, processor, flagprocessors): | |
71 | if not flag & REVIDX_KNOWN_FLAGS: |
|
71 | if not flag & REVIDX_KNOWN_FLAGS: | |
72 | msg = _("cannot register processor on unknown flag '%#x'.") % (flag) |
|
72 | msg = _("cannot register processor on unknown flag '%#x'.") % (flag) | |
73 | raise error.ProgrammingError(msg) |
|
73 | raise error.ProgrammingError(msg) | |
74 | if flag not in REVIDX_FLAGS_ORDER: |
|
74 | if flag not in REVIDX_FLAGS_ORDER: | |
75 | msg = _("flag '%#x' undefined in REVIDX_FLAGS_ORDER.") % (flag) |
|
75 | msg = _("flag '%#x' undefined in REVIDX_FLAGS_ORDER.") % (flag) | |
76 | raise error.ProgrammingError(msg) |
|
76 | raise error.ProgrammingError(msg) | |
77 | if flag in flagprocessors: |
|
77 | if flag in flagprocessors: | |
78 | msg = _("cannot register multiple processors on flag '%#x'.") % (flag) |
|
78 | msg = _("cannot register multiple processors on flag '%#x'.") % (flag) | |
79 | raise error.Abort(msg) |
|
79 | raise error.Abort(msg) | |
80 | flagprocessors[flag] = processor |
|
80 | flagprocessors[flag] = processor | |
81 |
|
81 | |||
82 | class flagprocessorsmixin(object): |
|
|||
83 | """basic mixin to support revlog flag processing |
|
|||
84 |
|
||||
85 | Make sure the `_flagprocessors` attribute is set at ``__init__`` time. |
|
|||
86 |
|
||||
87 | See the documentation of the ``_processflags`` method for details. |
|
|||
88 | """ |
|
|||
89 |
|
||||
90 | def processflagswrite(revlog, text, flags, sidedata): |
|
82 | def processflagswrite(revlog, text, flags, sidedata): | |
91 | """Inspect revision data flags and applies write transformations defined |
|
83 | """Inspect revision data flags and applies write transformations defined | |
92 | by registered flag processors. |
|
84 | by registered flag processors. | |
93 |
|
85 | |||
94 | ``text`` - the revision data to process |
|
86 | ``text`` - the revision data to process | |
95 | ``flags`` - the revision flags |
|
87 | ``flags`` - the revision flags | |
96 |
|
88 | |||
97 | This method processes the flags in the order (or reverse order if |
|
89 | This method processes the flags in the order (or reverse order if | |
98 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the |
|
90 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the | |
99 | flag processors registered for present flags. The order of flags defined |
|
91 | flag processors registered for present flags. The order of flags defined | |
100 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. |
|
92 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. | |
101 |
|
93 | |||
102 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the |
|
94 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the | |
103 | processed text and ``validatehash`` is a bool indicating whether the |
|
95 | processed text and ``validatehash`` is a bool indicating whether the | |
104 | returned text should be checked for hash integrity. |
|
96 | returned text should be checked for hash integrity. | |
105 | """ |
|
97 | """ | |
106 | return _processflagsfunc(revlog, text, flags, 'write', |
|
98 | return _processflagsfunc(revlog, text, flags, 'write', | |
107 | sidedata=sidedata)[:2] |
|
99 | sidedata=sidedata)[:2] | |
108 |
|
100 | |||
109 | def processflagsread(revlog, text, flags): |
|
101 | def processflagsread(revlog, text, flags): | |
110 | """Inspect revision data flags and applies read transformations defined |
|
102 | """Inspect revision data flags and applies read transformations defined | |
111 | by registered flag processors. |
|
103 | by registered flag processors. | |
112 |
|
104 | |||
113 | ``text`` - the revision data to process |
|
105 | ``text`` - the revision data to process | |
114 | ``flags`` - the revision flags |
|
106 | ``flags`` - the revision flags | |
115 | ``raw`` - an optional argument describing if the raw transform should be |
|
107 | ``raw`` - an optional argument describing if the raw transform should be | |
116 | applied. |
|
108 | applied. | |
117 |
|
109 | |||
118 | This method processes the flags in the order (or reverse order if |
|
110 | This method processes the flags in the order (or reverse order if | |
119 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the |
|
111 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the | |
120 | flag processors registered for present flags. The order of flags defined |
|
112 | flag processors registered for present flags. The order of flags defined | |
121 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. |
|
113 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. | |
122 |
|
114 | |||
123 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the |
|
115 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the | |
124 | processed text and ``validatehash`` is a bool indicating whether the |
|
116 | processed text and ``validatehash`` is a bool indicating whether the | |
125 | returned text should be checked for hash integrity. |
|
117 | returned text should be checked for hash integrity. | |
126 | """ |
|
118 | """ | |
127 | return _processflagsfunc(revlog, text, flags, 'read') |
|
119 | return _processflagsfunc(revlog, text, flags, 'read') | |
128 |
|
120 | |||
129 | def processflagsraw(revlog, text, flags): |
|
121 | def processflagsraw(revlog, text, flags): | |
130 | """Inspect revision data flags to check is the content hash should be |
|
122 | """Inspect revision data flags to check is the content hash should be | |
131 | validated. |
|
123 | validated. | |
132 |
|
124 | |||
133 | ``text`` - the revision data to process |
|
125 | ``text`` - the revision data to process | |
134 | ``flags`` - the revision flags |
|
126 | ``flags`` - the revision flags | |
135 |
|
127 | |||
136 | This method processes the flags in the order (or reverse order if |
|
128 | This method processes the flags in the order (or reverse order if | |
137 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the |
|
129 | ``operation`` is 'write') defined by REVIDX_FLAGS_ORDER, applying the | |
138 | flag processors registered for present flags. The order of flags defined |
|
130 | flag processors registered for present flags. The order of flags defined | |
139 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. |
|
131 | in REVIDX_FLAGS_ORDER needs to be stable to allow non-commutativity. | |
140 |
|
132 | |||
141 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the |
|
133 | Returns a 2-tuple of ``(text, validatehash)`` where ``text`` is the | |
142 | processed text and ``validatehash`` is a bool indicating whether the |
|
134 | processed text and ``validatehash`` is a bool indicating whether the | |
143 | returned text should be checked for hash integrity. |
|
135 | returned text should be checked for hash integrity. | |
144 | """ |
|
136 | """ | |
145 | return _processflagsfunc(revlog, text, flags, 'raw')[1] |
|
137 | return _processflagsfunc(revlog, text, flags, 'raw')[1] | |
146 |
|
138 | |||
147 | def _processflagsfunc(revlog, text, flags, operation, sidedata=None): |
|
139 | def _processflagsfunc(revlog, text, flags, operation, sidedata=None): | |
148 | """internal function to process flag on a revlog |
|
140 | """internal function to process flag on a revlog | |
149 |
|
141 | |||
150 | This function is private to this module, code should never needs to call it |
|
142 | This function is private to this module, code should never needs to call it | |
151 | directly.""" |
|
143 | directly.""" | |
152 | # fast path: no flag processors will run |
|
144 | # fast path: no flag processors will run | |
153 | if flags == 0: |
|
145 | if flags == 0: | |
154 | return text, True, {} |
|
146 | return text, True, {} | |
155 | if operation not in ('read', 'write', 'raw'): |
|
147 | if operation not in ('read', 'write', 'raw'): | |
156 | raise error.ProgrammingError(_("invalid '%s' operation") % |
|
148 | raise error.ProgrammingError(_("invalid '%s' operation") % | |
157 | operation) |
|
149 | operation) | |
158 | # Check all flags are known. |
|
150 | # Check all flags are known. | |
159 | if flags & ~REVIDX_KNOWN_FLAGS: |
|
151 | if flags & ~REVIDX_KNOWN_FLAGS: | |
160 | raise revlog._flagserrorclass(_("incompatible revision flag '%#x'") % |
|
152 | raise revlog._flagserrorclass(_("incompatible revision flag '%#x'") % | |
161 | (flags & ~REVIDX_KNOWN_FLAGS)) |
|
153 | (flags & ~REVIDX_KNOWN_FLAGS)) | |
162 | validatehash = True |
|
154 | validatehash = True | |
163 | # Depending on the operation (read or write), the order might be |
|
155 | # Depending on the operation (read or write), the order might be | |
164 | # reversed due to non-commutative transforms. |
|
156 | # reversed due to non-commutative transforms. | |
165 | orderedflags = REVIDX_FLAGS_ORDER |
|
157 | orderedflags = REVIDX_FLAGS_ORDER | |
166 | if operation == 'write': |
|
158 | if operation == 'write': | |
167 | orderedflags = reversed(orderedflags) |
|
159 | orderedflags = reversed(orderedflags) | |
168 |
|
160 | |||
169 | outsidedata = {} |
|
161 | outsidedata = {} | |
170 | for flag in orderedflags: |
|
162 | for flag in orderedflags: | |
171 | # If a flagprocessor has been registered for a known flag, apply the |
|
163 | # If a flagprocessor has been registered for a known flag, apply the | |
172 | # related operation transform and update result tuple. |
|
164 | # related operation transform and update result tuple. | |
173 | if flag & flags: |
|
165 | if flag & flags: | |
174 | vhash = True |
|
166 | vhash = True | |
175 |
|
167 | |||
176 | if flag not in revlog._flagprocessors: |
|
168 | if flag not in revlog._flagprocessors: | |
177 | message = _("missing processor for flag '%#x'") % (flag) |
|
169 | message = _("missing processor for flag '%#x'") % (flag) | |
178 | raise revlog._flagserrorclass(message) |
|
170 | raise revlog._flagserrorclass(message) | |
179 |
|
171 | |||
180 | processor = revlog._flagprocessors[flag] |
|
172 | processor = revlog._flagprocessors[flag] | |
181 | if processor is not None: |
|
173 | if processor is not None: | |
182 | readtransform, writetransform, rawtransform = processor |
|
174 | readtransform, writetransform, rawtransform = processor | |
183 |
|
175 | |||
184 | if operation == 'raw': |
|
176 | if operation == 'raw': | |
185 | vhash = rawtransform(revlog, text) |
|
177 | vhash = rawtransform(revlog, text) | |
186 | elif operation == 'read': |
|
178 | elif operation == 'read': | |
187 | text, vhash, s = readtransform(revlog, text) |
|
179 | text, vhash, s = readtransform(revlog, text) | |
188 | outsidedata.update(s) |
|
180 | outsidedata.update(s) | |
189 | else: # write operation |
|
181 | else: # write operation | |
190 | text, vhash = writetransform(revlog, text, sidedata) |
|
182 | text, vhash = writetransform(revlog, text, sidedata) | |
191 | validatehash = validatehash and vhash |
|
183 | validatehash = validatehash and vhash | |
192 |
|
184 | |||
193 | return text, validatehash, outsidedata |
|
185 | return text, validatehash, outsidedata |
@@ -1,683 +1,683 b'' | |||||
1 | # simplestorerepo.py - Extension that swaps in alternate repository storage. |
|
1 | # simplestorerepo.py - Extension that swaps in alternate repository storage. | |
2 | # |
|
2 | # | |
3 | # Copyright 2018 Gregory Szorc <gregory.szorc@gmail.com> |
|
3 | # Copyright 2018 Gregory Szorc <gregory.szorc@gmail.com> | |
4 | # |
|
4 | # | |
5 | # This software may be used and distributed according to the terms of the |
|
5 | # This software may be used and distributed according to the terms of the | |
6 | # GNU General Public License version 2 or any later version. |
|
6 | # GNU General Public License version 2 or any later version. | |
7 |
|
7 | |||
8 | # To use this with the test suite: |
|
8 | # To use this with the test suite: | |
9 | # |
|
9 | # | |
10 | # $ HGREPOFEATURES="simplestore" ./run-tests.py \ |
|
10 | # $ HGREPOFEATURES="simplestore" ./run-tests.py \ | |
11 | # --extra-config-opt extensions.simplestore=`pwd`/simplestorerepo.py |
|
11 | # --extra-config-opt extensions.simplestore=`pwd`/simplestorerepo.py | |
12 |
|
12 | |||
13 | from __future__ import absolute_import |
|
13 | from __future__ import absolute_import | |
14 |
|
14 | |||
15 | import stat |
|
15 | import stat | |
16 |
|
16 | |||
17 | from mercurial.i18n import _ |
|
17 | from mercurial.i18n import _ | |
18 | from mercurial.node import ( |
|
18 | from mercurial.node import ( | |
19 | bin, |
|
19 | bin, | |
20 | hex, |
|
20 | hex, | |
21 | nullid, |
|
21 | nullid, | |
22 | nullrev, |
|
22 | nullrev, | |
23 | ) |
|
23 | ) | |
24 | from mercurial.thirdparty import ( |
|
24 | from mercurial.thirdparty import ( | |
25 | attr, |
|
25 | attr, | |
26 | ) |
|
26 | ) | |
27 | from mercurial import ( |
|
27 | from mercurial import ( | |
28 | ancestor, |
|
28 | ancestor, | |
29 | bundlerepo, |
|
29 | bundlerepo, | |
30 | error, |
|
30 | error, | |
31 | extensions, |
|
31 | extensions, | |
32 | localrepo, |
|
32 | localrepo, | |
33 | mdiff, |
|
33 | mdiff, | |
34 | pycompat, |
|
34 | pycompat, | |
35 | revlog, |
|
35 | revlog, | |
36 | store, |
|
36 | store, | |
37 | verify, |
|
37 | verify, | |
38 | ) |
|
38 | ) | |
39 | from mercurial.interfaces import ( |
|
39 | from mercurial.interfaces import ( | |
40 | repository, |
|
40 | repository, | |
41 | util as interfaceutil, |
|
41 | util as interfaceutil, | |
42 | ) |
|
42 | ) | |
43 | from mercurial.utils import ( |
|
43 | from mercurial.utils import ( | |
44 | cborutil, |
|
44 | cborutil, | |
45 | storageutil, |
|
45 | storageutil, | |
46 | ) |
|
46 | ) | |
47 | from mercurial.revlogutils import ( |
|
47 | from mercurial.revlogutils import ( | |
48 | flagutil, |
|
48 | flagutil, | |
49 | ) |
|
49 | ) | |
50 |
|
50 | |||
51 | # Note for extension authors: ONLY specify testedwith = 'ships-with-hg-core' for |
|
51 | # Note for extension authors: ONLY specify testedwith = 'ships-with-hg-core' for | |
52 | # extensions which SHIP WITH MERCURIAL. Non-mainline extensions should |
|
52 | # extensions which SHIP WITH MERCURIAL. Non-mainline extensions should | |
53 | # be specifying the version(s) of Mercurial they are tested with, or |
|
53 | # be specifying the version(s) of Mercurial they are tested with, or | |
54 | # leave the attribute unspecified. |
|
54 | # leave the attribute unspecified. | |
55 | testedwith = 'ships-with-hg-core' |
|
55 | testedwith = 'ships-with-hg-core' | |
56 |
|
56 | |||
57 | REQUIREMENT = 'testonly-simplestore' |
|
57 | REQUIREMENT = 'testonly-simplestore' | |
58 |
|
58 | |||
59 | def validatenode(node): |
|
59 | def validatenode(node): | |
60 | if isinstance(node, int): |
|
60 | if isinstance(node, int): | |
61 | raise ValueError('expected node; got int') |
|
61 | raise ValueError('expected node; got int') | |
62 |
|
62 | |||
63 | if len(node) != 20: |
|
63 | if len(node) != 20: | |
64 | raise ValueError('expected 20 byte node') |
|
64 | raise ValueError('expected 20 byte node') | |
65 |
|
65 | |||
66 | def validaterev(rev): |
|
66 | def validaterev(rev): | |
67 | if not isinstance(rev, int): |
|
67 | if not isinstance(rev, int): | |
68 | raise ValueError('expected int') |
|
68 | raise ValueError('expected int') | |
69 |
|
69 | |||
70 | class simplestoreerror(error.StorageError): |
|
70 | class simplestoreerror(error.StorageError): | |
71 | pass |
|
71 | pass | |
72 |
|
72 | |||
73 | @interfaceutil.implementer(repository.irevisiondelta) |
|
73 | @interfaceutil.implementer(repository.irevisiondelta) | |
74 | @attr.s(slots=True) |
|
74 | @attr.s(slots=True) | |
75 | class simplestorerevisiondelta(object): |
|
75 | class simplestorerevisiondelta(object): | |
76 | node = attr.ib() |
|
76 | node = attr.ib() | |
77 | p1node = attr.ib() |
|
77 | p1node = attr.ib() | |
78 | p2node = attr.ib() |
|
78 | p2node = attr.ib() | |
79 | basenode = attr.ib() |
|
79 | basenode = attr.ib() | |
80 | flags = attr.ib() |
|
80 | flags = attr.ib() | |
81 | baserevisionsize = attr.ib() |
|
81 | baserevisionsize = attr.ib() | |
82 | revision = attr.ib() |
|
82 | revision = attr.ib() | |
83 | delta = attr.ib() |
|
83 | delta = attr.ib() | |
84 | linknode = attr.ib(default=None) |
|
84 | linknode = attr.ib(default=None) | |
85 |
|
85 | |||
86 | @interfaceutil.implementer(repository.iverifyproblem) |
|
86 | @interfaceutil.implementer(repository.iverifyproblem) | |
87 | @attr.s(frozen=True) |
|
87 | @attr.s(frozen=True) | |
88 | class simplefilestoreproblem(object): |
|
88 | class simplefilestoreproblem(object): | |
89 | warning = attr.ib(default=None) |
|
89 | warning = attr.ib(default=None) | |
90 | error = attr.ib(default=None) |
|
90 | error = attr.ib(default=None) | |
91 | node = attr.ib(default=None) |
|
91 | node = attr.ib(default=None) | |
92 |
|
92 | |||
93 | @interfaceutil.implementer(repository.ifilestorage) |
|
93 | @interfaceutil.implementer(repository.ifilestorage) | |
94 | class filestorage(flagutil.flagprocessorsmixin): |
|
94 | class filestorage(object): | |
95 | """Implements storage for a tracked path. |
|
95 | """Implements storage for a tracked path. | |
96 |
|
96 | |||
97 | Data is stored in the VFS in a directory corresponding to the tracked |
|
97 | Data is stored in the VFS in a directory corresponding to the tracked | |
98 | path. |
|
98 | path. | |
99 |
|
99 | |||
100 | Index data is stored in an ``index`` file using CBOR. |
|
100 | Index data is stored in an ``index`` file using CBOR. | |
101 |
|
101 | |||
102 | Fulltext data is stored in files having names of the node. |
|
102 | Fulltext data is stored in files having names of the node. | |
103 | """ |
|
103 | """ | |
104 |
|
104 | |||
105 | _flagserrorclass = simplestoreerror |
|
105 | _flagserrorclass = simplestoreerror | |
106 |
|
106 | |||
107 | def __init__(self, svfs, path): |
|
107 | def __init__(self, svfs, path): | |
108 | self._svfs = svfs |
|
108 | self._svfs = svfs | |
109 | self._path = path |
|
109 | self._path = path | |
110 |
|
110 | |||
111 | self._storepath = b'/'.join([b'data', path]) |
|
111 | self._storepath = b'/'.join([b'data', path]) | |
112 | self._indexpath = b'/'.join([self._storepath, b'index']) |
|
112 | self._indexpath = b'/'.join([self._storepath, b'index']) | |
113 |
|
113 | |||
114 | indexdata = self._svfs.tryread(self._indexpath) |
|
114 | indexdata = self._svfs.tryread(self._indexpath) | |
115 | if indexdata: |
|
115 | if indexdata: | |
116 | indexdata = cborutil.decodeall(indexdata) |
|
116 | indexdata = cborutil.decodeall(indexdata) | |
117 |
|
117 | |||
118 | self._indexdata = indexdata or [] |
|
118 | self._indexdata = indexdata or [] | |
119 | self._indexbynode = {} |
|
119 | self._indexbynode = {} | |
120 | self._indexbyrev = {} |
|
120 | self._indexbyrev = {} | |
121 | self._index = [] |
|
121 | self._index = [] | |
122 | self._refreshindex() |
|
122 | self._refreshindex() | |
123 |
|
123 | |||
124 | self._flagprocessors = dict(flagutil.flagprocessors) |
|
124 | self._flagprocessors = dict(flagutil.flagprocessors) | |
125 |
|
125 | |||
126 | def _refreshindex(self): |
|
126 | def _refreshindex(self): | |
127 | self._indexbynode.clear() |
|
127 | self._indexbynode.clear() | |
128 | self._indexbyrev.clear() |
|
128 | self._indexbyrev.clear() | |
129 | self._index = [] |
|
129 | self._index = [] | |
130 |
|
130 | |||
131 | for i, entry in enumerate(self._indexdata): |
|
131 | for i, entry in enumerate(self._indexdata): | |
132 | self._indexbynode[entry[b'node']] = entry |
|
132 | self._indexbynode[entry[b'node']] = entry | |
133 | self._indexbyrev[i] = entry |
|
133 | self._indexbyrev[i] = entry | |
134 |
|
134 | |||
135 | self._indexbynode[nullid] = { |
|
135 | self._indexbynode[nullid] = { | |
136 | b'node': nullid, |
|
136 | b'node': nullid, | |
137 | b'p1': nullid, |
|
137 | b'p1': nullid, | |
138 | b'p2': nullid, |
|
138 | b'p2': nullid, | |
139 | b'linkrev': nullrev, |
|
139 | b'linkrev': nullrev, | |
140 | b'flags': 0, |
|
140 | b'flags': 0, | |
141 | } |
|
141 | } | |
142 |
|
142 | |||
143 | self._indexbyrev[nullrev] = { |
|
143 | self._indexbyrev[nullrev] = { | |
144 | b'node': nullid, |
|
144 | b'node': nullid, | |
145 | b'p1': nullid, |
|
145 | b'p1': nullid, | |
146 | b'p2': nullid, |
|
146 | b'p2': nullid, | |
147 | b'linkrev': nullrev, |
|
147 | b'linkrev': nullrev, | |
148 | b'flags': 0, |
|
148 | b'flags': 0, | |
149 | } |
|
149 | } | |
150 |
|
150 | |||
151 | for i, entry in enumerate(self._indexdata): |
|
151 | for i, entry in enumerate(self._indexdata): | |
152 | p1rev, p2rev = self.parentrevs(self.rev(entry[b'node'])) |
|
152 | p1rev, p2rev = self.parentrevs(self.rev(entry[b'node'])) | |
153 |
|
153 | |||
154 | # start, length, rawsize, chainbase, linkrev, p1, p2, node |
|
154 | # start, length, rawsize, chainbase, linkrev, p1, p2, node | |
155 | self._index.append((0, 0, 0, -1, entry[b'linkrev'], p1rev, p2rev, |
|
155 | self._index.append((0, 0, 0, -1, entry[b'linkrev'], p1rev, p2rev, | |
156 | entry[b'node'])) |
|
156 | entry[b'node'])) | |
157 |
|
157 | |||
158 | self._index.append((0, 0, 0, -1, -1, -1, -1, nullid)) |
|
158 | self._index.append((0, 0, 0, -1, -1, -1, -1, nullid)) | |
159 |
|
159 | |||
160 | def __len__(self): |
|
160 | def __len__(self): | |
161 | return len(self._indexdata) |
|
161 | return len(self._indexdata) | |
162 |
|
162 | |||
163 | def __iter__(self): |
|
163 | def __iter__(self): | |
164 | return iter(range(len(self))) |
|
164 | return iter(range(len(self))) | |
165 |
|
165 | |||
166 | def revs(self, start=0, stop=None): |
|
166 | def revs(self, start=0, stop=None): | |
167 | step = 1 |
|
167 | step = 1 | |
168 | if stop is not None: |
|
168 | if stop is not None: | |
169 | if start > stop: |
|
169 | if start > stop: | |
170 | step = -1 |
|
170 | step = -1 | |
171 |
|
171 | |||
172 | stop += step |
|
172 | stop += step | |
173 | else: |
|
173 | else: | |
174 | stop = len(self) |
|
174 | stop = len(self) | |
175 |
|
175 | |||
176 | return range(start, stop, step) |
|
176 | return range(start, stop, step) | |
177 |
|
177 | |||
178 | def parents(self, node): |
|
178 | def parents(self, node): | |
179 | validatenode(node) |
|
179 | validatenode(node) | |
180 |
|
180 | |||
181 | if node not in self._indexbynode: |
|
181 | if node not in self._indexbynode: | |
182 | raise KeyError('unknown node') |
|
182 | raise KeyError('unknown node') | |
183 |
|
183 | |||
184 | entry = self._indexbynode[node] |
|
184 | entry = self._indexbynode[node] | |
185 |
|
185 | |||
186 | return entry[b'p1'], entry[b'p2'] |
|
186 | return entry[b'p1'], entry[b'p2'] | |
187 |
|
187 | |||
188 | def parentrevs(self, rev): |
|
188 | def parentrevs(self, rev): | |
189 | p1, p2 = self.parents(self._indexbyrev[rev][b'node']) |
|
189 | p1, p2 = self.parents(self._indexbyrev[rev][b'node']) | |
190 | return self.rev(p1), self.rev(p2) |
|
190 | return self.rev(p1), self.rev(p2) | |
191 |
|
191 | |||
192 | def rev(self, node): |
|
192 | def rev(self, node): | |
193 | validatenode(node) |
|
193 | validatenode(node) | |
194 |
|
194 | |||
195 | try: |
|
195 | try: | |
196 | self._indexbynode[node] |
|
196 | self._indexbynode[node] | |
197 | except KeyError: |
|
197 | except KeyError: | |
198 | raise error.LookupError(node, self._indexpath, _('no node')) |
|
198 | raise error.LookupError(node, self._indexpath, _('no node')) | |
199 |
|
199 | |||
200 | for rev, entry in self._indexbyrev.items(): |
|
200 | for rev, entry in self._indexbyrev.items(): | |
201 | if entry[b'node'] == node: |
|
201 | if entry[b'node'] == node: | |
202 | return rev |
|
202 | return rev | |
203 |
|
203 | |||
204 | raise error.ProgrammingError('this should not occur') |
|
204 | raise error.ProgrammingError('this should not occur') | |
205 |
|
205 | |||
206 | def node(self, rev): |
|
206 | def node(self, rev): | |
207 | validaterev(rev) |
|
207 | validaterev(rev) | |
208 |
|
208 | |||
209 | return self._indexbyrev[rev][b'node'] |
|
209 | return self._indexbyrev[rev][b'node'] | |
210 |
|
210 | |||
211 | def hasnode(self, node): |
|
211 | def hasnode(self, node): | |
212 | validatenode(node) |
|
212 | validatenode(node) | |
213 | return node in self._indexbynode |
|
213 | return node in self._indexbynode | |
214 |
|
214 | |||
215 | def censorrevision(self, tr, censornode, tombstone=b''): |
|
215 | def censorrevision(self, tr, censornode, tombstone=b''): | |
216 | raise NotImplementedError('TODO') |
|
216 | raise NotImplementedError('TODO') | |
217 |
|
217 | |||
218 | def lookup(self, node): |
|
218 | def lookup(self, node): | |
219 | if isinstance(node, int): |
|
219 | if isinstance(node, int): | |
220 | return self.node(node) |
|
220 | return self.node(node) | |
221 |
|
221 | |||
222 | if len(node) == 20: |
|
222 | if len(node) == 20: | |
223 | self.rev(node) |
|
223 | self.rev(node) | |
224 | return node |
|
224 | return node | |
225 |
|
225 | |||
226 | try: |
|
226 | try: | |
227 | rev = int(node) |
|
227 | rev = int(node) | |
228 | if '%d' % rev != node: |
|
228 | if '%d' % rev != node: | |
229 | raise ValueError |
|
229 | raise ValueError | |
230 |
|
230 | |||
231 | if rev < 0: |
|
231 | if rev < 0: | |
232 | rev = len(self) + rev |
|
232 | rev = len(self) + rev | |
233 | if rev < 0 or rev >= len(self): |
|
233 | if rev < 0 or rev >= len(self): | |
234 | raise ValueError |
|
234 | raise ValueError | |
235 |
|
235 | |||
236 | return self.node(rev) |
|
236 | return self.node(rev) | |
237 | except (ValueError, OverflowError): |
|
237 | except (ValueError, OverflowError): | |
238 | pass |
|
238 | pass | |
239 |
|
239 | |||
240 | if len(node) == 40: |
|
240 | if len(node) == 40: | |
241 | try: |
|
241 | try: | |
242 | rawnode = bin(node) |
|
242 | rawnode = bin(node) | |
243 | self.rev(rawnode) |
|
243 | self.rev(rawnode) | |
244 | return rawnode |
|
244 | return rawnode | |
245 | except TypeError: |
|
245 | except TypeError: | |
246 | pass |
|
246 | pass | |
247 |
|
247 | |||
248 | raise error.LookupError(node, self._path, _('invalid lookup input')) |
|
248 | raise error.LookupError(node, self._path, _('invalid lookup input')) | |
249 |
|
249 | |||
250 | def linkrev(self, rev): |
|
250 | def linkrev(self, rev): | |
251 | validaterev(rev) |
|
251 | validaterev(rev) | |
252 |
|
252 | |||
253 | return self._indexbyrev[rev][b'linkrev'] |
|
253 | return self._indexbyrev[rev][b'linkrev'] | |
254 |
|
254 | |||
255 | def _flags(self, rev): |
|
255 | def _flags(self, rev): | |
256 | validaterev(rev) |
|
256 | validaterev(rev) | |
257 |
|
257 | |||
258 | return self._indexbyrev[rev][b'flags'] |
|
258 | return self._indexbyrev[rev][b'flags'] | |
259 |
|
259 | |||
260 | def _candelta(self, baserev, rev): |
|
260 | def _candelta(self, baserev, rev): | |
261 | validaterev(baserev) |
|
261 | validaterev(baserev) | |
262 | validaterev(rev) |
|
262 | validaterev(rev) | |
263 |
|
263 | |||
264 | if ((self._flags(baserev) & revlog.REVIDX_RAWTEXT_CHANGING_FLAGS) |
|
264 | if ((self._flags(baserev) & revlog.REVIDX_RAWTEXT_CHANGING_FLAGS) | |
265 | or (self._flags(rev) & revlog.REVIDX_RAWTEXT_CHANGING_FLAGS)): |
|
265 | or (self._flags(rev) & revlog.REVIDX_RAWTEXT_CHANGING_FLAGS)): | |
266 | return False |
|
266 | return False | |
267 |
|
267 | |||
268 | return True |
|
268 | return True | |
269 |
|
269 | |||
270 | def checkhash(self, text, node, p1=None, p2=None, rev=None): |
|
270 | def checkhash(self, text, node, p1=None, p2=None, rev=None): | |
271 | if p1 is None and p2 is None: |
|
271 | if p1 is None and p2 is None: | |
272 | p1, p2 = self.parents(node) |
|
272 | p1, p2 = self.parents(node) | |
273 | if node != storageutil.hashrevisionsha1(text, p1, p2): |
|
273 | if node != storageutil.hashrevisionsha1(text, p1, p2): | |
274 | raise simplestoreerror(_("integrity check failed on %s") % |
|
274 | raise simplestoreerror(_("integrity check failed on %s") % | |
275 | self._path) |
|
275 | self._path) | |
276 |
|
276 | |||
277 | def revision(self, nodeorrev, raw=False): |
|
277 | def revision(self, nodeorrev, raw=False): | |
278 | if isinstance(nodeorrev, int): |
|
278 | if isinstance(nodeorrev, int): | |
279 | node = self.node(nodeorrev) |
|
279 | node = self.node(nodeorrev) | |
280 | else: |
|
280 | else: | |
281 | node = nodeorrev |
|
281 | node = nodeorrev | |
282 | validatenode(node) |
|
282 | validatenode(node) | |
283 |
|
283 | |||
284 | if node == nullid: |
|
284 | if node == nullid: | |
285 | return b'' |
|
285 | return b'' | |
286 |
|
286 | |||
287 | rev = self.rev(node) |
|
287 | rev = self.rev(node) | |
288 | flags = self._flags(rev) |
|
288 | flags = self._flags(rev) | |
289 |
|
289 | |||
290 | path = b'/'.join([self._storepath, hex(node)]) |
|
290 | path = b'/'.join([self._storepath, hex(node)]) | |
291 | rawtext = self._svfs.read(path) |
|
291 | rawtext = self._svfs.read(path) | |
292 |
|
292 | |||
293 | if raw: |
|
293 | if raw: | |
294 | validatehash = flagutil.processflagsraw(self, rawtext, flags) |
|
294 | validatehash = flagutil.processflagsraw(self, rawtext, flags) | |
295 | text = rawtext |
|
295 | text = rawtext | |
296 | else: |
|
296 | else: | |
297 | r = flagutil.processflagsread(self, rawtext, flags) |
|
297 | r = flagutil.processflagsread(self, rawtext, flags) | |
298 | text, validatehash, sidedata = r |
|
298 | text, validatehash, sidedata = r | |
299 | if validatehash: |
|
299 | if validatehash: | |
300 | self.checkhash(text, node, rev=rev) |
|
300 | self.checkhash(text, node, rev=rev) | |
301 |
|
301 | |||
302 | return text |
|
302 | return text | |
303 |
|
303 | |||
304 | def rawdata(self, nodeorrev): |
|
304 | def rawdata(self, nodeorrev): | |
305 | return self.revision(raw=True) |
|
305 | return self.revision(raw=True) | |
306 |
|
306 | |||
307 | def read(self, node): |
|
307 | def read(self, node): | |
308 | validatenode(node) |
|
308 | validatenode(node) | |
309 |
|
309 | |||
310 | revision = self.revision(node) |
|
310 | revision = self.revision(node) | |
311 |
|
311 | |||
312 | if not revision.startswith(b'\1\n'): |
|
312 | if not revision.startswith(b'\1\n'): | |
313 | return revision |
|
313 | return revision | |
314 |
|
314 | |||
315 | start = revision.index(b'\1\n', 2) |
|
315 | start = revision.index(b'\1\n', 2) | |
316 | return revision[start + 2:] |
|
316 | return revision[start + 2:] | |
317 |
|
317 | |||
318 | def renamed(self, node): |
|
318 | def renamed(self, node): | |
319 | validatenode(node) |
|
319 | validatenode(node) | |
320 |
|
320 | |||
321 | if self.parents(node)[0] != nullid: |
|
321 | if self.parents(node)[0] != nullid: | |
322 | return False |
|
322 | return False | |
323 |
|
323 | |||
324 | fulltext = self.revision(node) |
|
324 | fulltext = self.revision(node) | |
325 | m = storageutil.parsemeta(fulltext)[0] |
|
325 | m = storageutil.parsemeta(fulltext)[0] | |
326 |
|
326 | |||
327 | if m and 'copy' in m: |
|
327 | if m and 'copy' in m: | |
328 | return m['copy'], bin(m['copyrev']) |
|
328 | return m['copy'], bin(m['copyrev']) | |
329 |
|
329 | |||
330 | return False |
|
330 | return False | |
331 |
|
331 | |||
332 | def cmp(self, node, text): |
|
332 | def cmp(self, node, text): | |
333 | validatenode(node) |
|
333 | validatenode(node) | |
334 |
|
334 | |||
335 | t = text |
|
335 | t = text | |
336 |
|
336 | |||
337 | if text.startswith(b'\1\n'): |
|
337 | if text.startswith(b'\1\n'): | |
338 | t = b'\1\n\1\n' + text |
|
338 | t = b'\1\n\1\n' + text | |
339 |
|
339 | |||
340 | p1, p2 = self.parents(node) |
|
340 | p1, p2 = self.parents(node) | |
341 |
|
341 | |||
342 | if storageutil.hashrevisionsha1(t, p1, p2) == node: |
|
342 | if storageutil.hashrevisionsha1(t, p1, p2) == node: | |
343 | return False |
|
343 | return False | |
344 |
|
344 | |||
345 | if self.iscensored(self.rev(node)): |
|
345 | if self.iscensored(self.rev(node)): | |
346 | return text != b'' |
|
346 | return text != b'' | |
347 |
|
347 | |||
348 | if self.renamed(node): |
|
348 | if self.renamed(node): | |
349 | t2 = self.read(node) |
|
349 | t2 = self.read(node) | |
350 | return t2 != text |
|
350 | return t2 != text | |
351 |
|
351 | |||
352 | return True |
|
352 | return True | |
353 |
|
353 | |||
354 | def size(self, rev): |
|
354 | def size(self, rev): | |
355 | validaterev(rev) |
|
355 | validaterev(rev) | |
356 |
|
356 | |||
357 | node = self._indexbyrev[rev][b'node'] |
|
357 | node = self._indexbyrev[rev][b'node'] | |
358 |
|
358 | |||
359 | if self.renamed(node): |
|
359 | if self.renamed(node): | |
360 | return len(self.read(node)) |
|
360 | return len(self.read(node)) | |
361 |
|
361 | |||
362 | if self.iscensored(rev): |
|
362 | if self.iscensored(rev): | |
363 | return 0 |
|
363 | return 0 | |
364 |
|
364 | |||
365 | return len(self.revision(node)) |
|
365 | return len(self.revision(node)) | |
366 |
|
366 | |||
367 | def iscensored(self, rev): |
|
367 | def iscensored(self, rev): | |
368 | validaterev(rev) |
|
368 | validaterev(rev) | |
369 |
|
369 | |||
370 | return self._flags(rev) & repository.REVISION_FLAG_CENSORED |
|
370 | return self._flags(rev) & repository.REVISION_FLAG_CENSORED | |
371 |
|
371 | |||
372 | def commonancestorsheads(self, a, b): |
|
372 | def commonancestorsheads(self, a, b): | |
373 | validatenode(a) |
|
373 | validatenode(a) | |
374 | validatenode(b) |
|
374 | validatenode(b) | |
375 |
|
375 | |||
376 | a = self.rev(a) |
|
376 | a = self.rev(a) | |
377 | b = self.rev(b) |
|
377 | b = self.rev(b) | |
378 |
|
378 | |||
379 | ancestors = ancestor.commonancestorsheads(self.parentrevs, a, b) |
|
379 | ancestors = ancestor.commonancestorsheads(self.parentrevs, a, b) | |
380 | return pycompat.maplist(self.node, ancestors) |
|
380 | return pycompat.maplist(self.node, ancestors) | |
381 |
|
381 | |||
382 | def descendants(self, revs): |
|
382 | def descendants(self, revs): | |
383 | # This is a copy of revlog.descendants() |
|
383 | # This is a copy of revlog.descendants() | |
384 | first = min(revs) |
|
384 | first = min(revs) | |
385 | if first == nullrev: |
|
385 | if first == nullrev: | |
386 | for i in self: |
|
386 | for i in self: | |
387 | yield i |
|
387 | yield i | |
388 | return |
|
388 | return | |
389 |
|
389 | |||
390 | seen = set(revs) |
|
390 | seen = set(revs) | |
391 | for i in self.revs(start=first + 1): |
|
391 | for i in self.revs(start=first + 1): | |
392 | for x in self.parentrevs(i): |
|
392 | for x in self.parentrevs(i): | |
393 | if x != nullrev and x in seen: |
|
393 | if x != nullrev and x in seen: | |
394 | seen.add(i) |
|
394 | seen.add(i) | |
395 | yield i |
|
395 | yield i | |
396 | break |
|
396 | break | |
397 |
|
397 | |||
398 | # Required by verify. |
|
398 | # Required by verify. | |
399 | def files(self): |
|
399 | def files(self): | |
400 | entries = self._svfs.listdir(self._storepath) |
|
400 | entries = self._svfs.listdir(self._storepath) | |
401 |
|
401 | |||
402 | # Strip out undo.backup.* files created as part of transaction |
|
402 | # Strip out undo.backup.* files created as part of transaction | |
403 | # recording. |
|
403 | # recording. | |
404 | entries = [f for f in entries if not f.startswith('undo.backup.')] |
|
404 | entries = [f for f in entries if not f.startswith('undo.backup.')] | |
405 |
|
405 | |||
406 | return [b'/'.join((self._storepath, f)) for f in entries] |
|
406 | return [b'/'.join((self._storepath, f)) for f in entries] | |
407 |
|
407 | |||
408 | def storageinfo(self, exclusivefiles=False, sharedfiles=False, |
|
408 | def storageinfo(self, exclusivefiles=False, sharedfiles=False, | |
409 | revisionscount=False, trackedsize=False, |
|
409 | revisionscount=False, trackedsize=False, | |
410 | storedsize=False): |
|
410 | storedsize=False): | |
411 | # TODO do a real implementation of this |
|
411 | # TODO do a real implementation of this | |
412 | return { |
|
412 | return { | |
413 | 'exclusivefiles': [], |
|
413 | 'exclusivefiles': [], | |
414 | 'sharedfiles': [], |
|
414 | 'sharedfiles': [], | |
415 | 'revisionscount': len(self), |
|
415 | 'revisionscount': len(self), | |
416 | 'trackedsize': 0, |
|
416 | 'trackedsize': 0, | |
417 | 'storedsize': None, |
|
417 | 'storedsize': None, | |
418 | } |
|
418 | } | |
419 |
|
419 | |||
420 | def verifyintegrity(self, state): |
|
420 | def verifyintegrity(self, state): | |
421 | state['skipread'] = set() |
|
421 | state['skipread'] = set() | |
422 | for rev in self: |
|
422 | for rev in self: | |
423 | node = self.node(rev) |
|
423 | node = self.node(rev) | |
424 | try: |
|
424 | try: | |
425 | self.revision(node) |
|
425 | self.revision(node) | |
426 | except Exception as e: |
|
426 | except Exception as e: | |
427 | yield simplefilestoreproblem( |
|
427 | yield simplefilestoreproblem( | |
428 | error='unpacking %s: %s' % (node, e), |
|
428 | error='unpacking %s: %s' % (node, e), | |
429 | node=node) |
|
429 | node=node) | |
430 | state['skipread'].add(node) |
|
430 | state['skipread'].add(node) | |
431 |
|
431 | |||
432 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, |
|
432 | def emitrevisions(self, nodes, nodesorder=None, revisiondata=False, | |
433 | assumehaveparentrevisions=False, |
|
433 | assumehaveparentrevisions=False, | |
434 | deltamode=repository.CG_DELTAMODE_STD): |
|
434 | deltamode=repository.CG_DELTAMODE_STD): | |
435 | # TODO this will probably break on some ordering options. |
|
435 | # TODO this will probably break on some ordering options. | |
436 | nodes = [n for n in nodes if n != nullid] |
|
436 | nodes = [n for n in nodes if n != nullid] | |
437 | if not nodes: |
|
437 | if not nodes: | |
438 | return |
|
438 | return | |
439 | for delta in storageutil.emitrevisions( |
|
439 | for delta in storageutil.emitrevisions( | |
440 | self, nodes, nodesorder, simplestorerevisiondelta, |
|
440 | self, nodes, nodesorder, simplestorerevisiondelta, | |
441 | revisiondata=revisiondata, |
|
441 | revisiondata=revisiondata, | |
442 | assumehaveparentrevisions=assumehaveparentrevisions, |
|
442 | assumehaveparentrevisions=assumehaveparentrevisions, | |
443 | deltamode=deltamode): |
|
443 | deltamode=deltamode): | |
444 | yield delta |
|
444 | yield delta | |
445 |
|
445 | |||
446 | def add(self, text, meta, transaction, linkrev, p1, p2): |
|
446 | def add(self, text, meta, transaction, linkrev, p1, p2): | |
447 | if meta or text.startswith(b'\1\n'): |
|
447 | if meta or text.startswith(b'\1\n'): | |
448 | text = storageutil.packmeta(meta, text) |
|
448 | text = storageutil.packmeta(meta, text) | |
449 |
|
449 | |||
450 | return self.addrevision(text, transaction, linkrev, p1, p2) |
|
450 | return self.addrevision(text, transaction, linkrev, p1, p2) | |
451 |
|
451 | |||
452 | def addrevision(self, text, transaction, linkrev, p1, p2, node=None, |
|
452 | def addrevision(self, text, transaction, linkrev, p1, p2, node=None, | |
453 | flags=revlog.REVIDX_DEFAULT_FLAGS, cachedelta=None): |
|
453 | flags=revlog.REVIDX_DEFAULT_FLAGS, cachedelta=None): | |
454 | validatenode(p1) |
|
454 | validatenode(p1) | |
455 | validatenode(p2) |
|
455 | validatenode(p2) | |
456 |
|
456 | |||
457 | if flags: |
|
457 | if flags: | |
458 | node = node or storageutil.hashrevisionsha1(text, p1, p2) |
|
458 | node = node or storageutil.hashrevisionsha1(text, p1, p2) | |
459 |
|
459 | |||
460 | rawtext, validatehash = flagutil.processflagswrite(self, text, flags) |
|
460 | rawtext, validatehash = flagutil.processflagswrite(self, text, flags) | |
461 |
|
461 | |||
462 | node = node or storageutil.hashrevisionsha1(text, p1, p2) |
|
462 | node = node or storageutil.hashrevisionsha1(text, p1, p2) | |
463 |
|
463 | |||
464 | if node in self._indexbynode: |
|
464 | if node in self._indexbynode: | |
465 | return node |
|
465 | return node | |
466 |
|
466 | |||
467 | if validatehash: |
|
467 | if validatehash: | |
468 | self.checkhash(rawtext, node, p1=p1, p2=p2) |
|
468 | self.checkhash(rawtext, node, p1=p1, p2=p2) | |
469 |
|
469 | |||
470 | return self._addrawrevision(node, rawtext, transaction, linkrev, p1, p2, |
|
470 | return self._addrawrevision(node, rawtext, transaction, linkrev, p1, p2, | |
471 | flags) |
|
471 | flags) | |
472 |
|
472 | |||
473 | def _addrawrevision(self, node, rawtext, transaction, link, p1, p2, flags): |
|
473 | def _addrawrevision(self, node, rawtext, transaction, link, p1, p2, flags): | |
474 | transaction.addbackup(self._indexpath) |
|
474 | transaction.addbackup(self._indexpath) | |
475 |
|
475 | |||
476 | path = b'/'.join([self._storepath, hex(node)]) |
|
476 | path = b'/'.join([self._storepath, hex(node)]) | |
477 |
|
477 | |||
478 | self._svfs.write(path, rawtext) |
|
478 | self._svfs.write(path, rawtext) | |
479 |
|
479 | |||
480 | self._indexdata.append({ |
|
480 | self._indexdata.append({ | |
481 | b'node': node, |
|
481 | b'node': node, | |
482 | b'p1': p1, |
|
482 | b'p1': p1, | |
483 | b'p2': p2, |
|
483 | b'p2': p2, | |
484 | b'linkrev': link, |
|
484 | b'linkrev': link, | |
485 | b'flags': flags, |
|
485 | b'flags': flags, | |
486 | }) |
|
486 | }) | |
487 |
|
487 | |||
488 | self._reflectindexupdate() |
|
488 | self._reflectindexupdate() | |
489 |
|
489 | |||
490 | return node |
|
490 | return node | |
491 |
|
491 | |||
492 | def _reflectindexupdate(self): |
|
492 | def _reflectindexupdate(self): | |
493 | self._refreshindex() |
|
493 | self._refreshindex() | |
494 | self._svfs.write(self._indexpath, |
|
494 | self._svfs.write(self._indexpath, | |
495 | ''.join(cborutil.streamencode(self._indexdata))) |
|
495 | ''.join(cborutil.streamencode(self._indexdata))) | |
496 |
|
496 | |||
497 | def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None, |
|
497 | def addgroup(self, deltas, linkmapper, transaction, addrevisioncb=None, | |
498 | maybemissingparents=False): |
|
498 | maybemissingparents=False): | |
499 | if maybemissingparents: |
|
499 | if maybemissingparents: | |
500 | raise error.Abort(_('simple store does not support missing parents ' |
|
500 | raise error.Abort(_('simple store does not support missing parents ' | |
501 | 'write mode')) |
|
501 | 'write mode')) | |
502 |
|
502 | |||
503 | nodes = [] |
|
503 | nodes = [] | |
504 |
|
504 | |||
505 | transaction.addbackup(self._indexpath) |
|
505 | transaction.addbackup(self._indexpath) | |
506 |
|
506 | |||
507 | for node, p1, p2, linknode, deltabase, delta, flags in deltas: |
|
507 | for node, p1, p2, linknode, deltabase, delta, flags in deltas: | |
508 | linkrev = linkmapper(linknode) |
|
508 | linkrev = linkmapper(linknode) | |
509 | flags = flags or revlog.REVIDX_DEFAULT_FLAGS |
|
509 | flags = flags or revlog.REVIDX_DEFAULT_FLAGS | |
510 |
|
510 | |||
511 | nodes.append(node) |
|
511 | nodes.append(node) | |
512 |
|
512 | |||
513 | if node in self._indexbynode: |
|
513 | if node in self._indexbynode: | |
514 | continue |
|
514 | continue | |
515 |
|
515 | |||
516 | # Need to resolve the fulltext from the delta base. |
|
516 | # Need to resolve the fulltext from the delta base. | |
517 | if deltabase == nullid: |
|
517 | if deltabase == nullid: | |
518 | text = mdiff.patch(b'', delta) |
|
518 | text = mdiff.patch(b'', delta) | |
519 | else: |
|
519 | else: | |
520 | text = mdiff.patch(self.revision(deltabase), delta) |
|
520 | text = mdiff.patch(self.revision(deltabase), delta) | |
521 |
|
521 | |||
522 | self._addrawrevision(node, text, transaction, linkrev, p1, p2, |
|
522 | self._addrawrevision(node, text, transaction, linkrev, p1, p2, | |
523 | flags) |
|
523 | flags) | |
524 |
|
524 | |||
525 | if addrevisioncb: |
|
525 | if addrevisioncb: | |
526 | addrevisioncb(self, node) |
|
526 | addrevisioncb(self, node) | |
527 | return nodes |
|
527 | return nodes | |
528 |
|
528 | |||
529 | def _headrevs(self): |
|
529 | def _headrevs(self): | |
530 | # Assume all revisions are heads by default. |
|
530 | # Assume all revisions are heads by default. | |
531 | revishead = {rev: True for rev in self._indexbyrev} |
|
531 | revishead = {rev: True for rev in self._indexbyrev} | |
532 |
|
532 | |||
533 | for rev, entry in self._indexbyrev.items(): |
|
533 | for rev, entry in self._indexbyrev.items(): | |
534 | # Unset head flag for all seen parents. |
|
534 | # Unset head flag for all seen parents. | |
535 | revishead[self.rev(entry[b'p1'])] = False |
|
535 | revishead[self.rev(entry[b'p1'])] = False | |
536 | revishead[self.rev(entry[b'p2'])] = False |
|
536 | revishead[self.rev(entry[b'p2'])] = False | |
537 |
|
537 | |||
538 | return [rev for rev, ishead in sorted(revishead.items()) |
|
538 | return [rev for rev, ishead in sorted(revishead.items()) | |
539 | if ishead] |
|
539 | if ishead] | |
540 |
|
540 | |||
541 | def heads(self, start=None, stop=None): |
|
541 | def heads(self, start=None, stop=None): | |
542 | # This is copied from revlog.py. |
|
542 | # This is copied from revlog.py. | |
543 | if start is None and stop is None: |
|
543 | if start is None and stop is None: | |
544 | if not len(self): |
|
544 | if not len(self): | |
545 | return [nullid] |
|
545 | return [nullid] | |
546 | return [self.node(r) for r in self._headrevs()] |
|
546 | return [self.node(r) for r in self._headrevs()] | |
547 |
|
547 | |||
548 | if start is None: |
|
548 | if start is None: | |
549 | start = nullid |
|
549 | start = nullid | |
550 | if stop is None: |
|
550 | if stop is None: | |
551 | stop = [] |
|
551 | stop = [] | |
552 | stoprevs = set([self.rev(n) for n in stop]) |
|
552 | stoprevs = set([self.rev(n) for n in stop]) | |
553 | startrev = self.rev(start) |
|
553 | startrev = self.rev(start) | |
554 | reachable = {startrev} |
|
554 | reachable = {startrev} | |
555 | heads = {startrev} |
|
555 | heads = {startrev} | |
556 |
|
556 | |||
557 | parentrevs = self.parentrevs |
|
557 | parentrevs = self.parentrevs | |
558 | for r in self.revs(start=startrev + 1): |
|
558 | for r in self.revs(start=startrev + 1): | |
559 | for p in parentrevs(r): |
|
559 | for p in parentrevs(r): | |
560 | if p in reachable: |
|
560 | if p in reachable: | |
561 | if r not in stoprevs: |
|
561 | if r not in stoprevs: | |
562 | reachable.add(r) |
|
562 | reachable.add(r) | |
563 | heads.add(r) |
|
563 | heads.add(r) | |
564 | if p in heads and p not in stoprevs: |
|
564 | if p in heads and p not in stoprevs: | |
565 | heads.remove(p) |
|
565 | heads.remove(p) | |
566 |
|
566 | |||
567 | return [self.node(r) for r in heads] |
|
567 | return [self.node(r) for r in heads] | |
568 |
|
568 | |||
569 | def children(self, node): |
|
569 | def children(self, node): | |
570 | validatenode(node) |
|
570 | validatenode(node) | |
571 |
|
571 | |||
572 | # This is a copy of revlog.children(). |
|
572 | # This is a copy of revlog.children(). | |
573 | c = [] |
|
573 | c = [] | |
574 | p = self.rev(node) |
|
574 | p = self.rev(node) | |
575 | for r in self.revs(start=p + 1): |
|
575 | for r in self.revs(start=p + 1): | |
576 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] |
|
576 | prevs = [pr for pr in self.parentrevs(r) if pr != nullrev] | |
577 | if prevs: |
|
577 | if prevs: | |
578 | for pr in prevs: |
|
578 | for pr in prevs: | |
579 | if pr == p: |
|
579 | if pr == p: | |
580 | c.append(self.node(r)) |
|
580 | c.append(self.node(r)) | |
581 | elif p == nullrev: |
|
581 | elif p == nullrev: | |
582 | c.append(self.node(r)) |
|
582 | c.append(self.node(r)) | |
583 | return c |
|
583 | return c | |
584 |
|
584 | |||
585 | def getstrippoint(self, minlink): |
|
585 | def getstrippoint(self, minlink): | |
586 | return storageutil.resolvestripinfo( |
|
586 | return storageutil.resolvestripinfo( | |
587 | minlink, len(self) - 1, self._headrevs(), self.linkrev, |
|
587 | minlink, len(self) - 1, self._headrevs(), self.linkrev, | |
588 | self.parentrevs) |
|
588 | self.parentrevs) | |
589 |
|
589 | |||
590 | def strip(self, minlink, transaction): |
|
590 | def strip(self, minlink, transaction): | |
591 | if not len(self): |
|
591 | if not len(self): | |
592 | return |
|
592 | return | |
593 |
|
593 | |||
594 | rev, _ignored = self.getstrippoint(minlink) |
|
594 | rev, _ignored = self.getstrippoint(minlink) | |
595 | if rev == len(self): |
|
595 | if rev == len(self): | |
596 | return |
|
596 | return | |
597 |
|
597 | |||
598 | # Purge index data starting at the requested revision. |
|
598 | # Purge index data starting at the requested revision. | |
599 | self._indexdata[rev:] = [] |
|
599 | self._indexdata[rev:] = [] | |
600 | self._reflectindexupdate() |
|
600 | self._reflectindexupdate() | |
601 |
|
601 | |||
602 | def issimplestorefile(f, kind, st): |
|
602 | def issimplestorefile(f, kind, st): | |
603 | if kind != stat.S_IFREG: |
|
603 | if kind != stat.S_IFREG: | |
604 | return False |
|
604 | return False | |
605 |
|
605 | |||
606 | if store.isrevlog(f, kind, st): |
|
606 | if store.isrevlog(f, kind, st): | |
607 | return False |
|
607 | return False | |
608 |
|
608 | |||
609 | # Ignore transaction undo files. |
|
609 | # Ignore transaction undo files. | |
610 | if f.startswith('undo.'): |
|
610 | if f.startswith('undo.'): | |
611 | return False |
|
611 | return False | |
612 |
|
612 | |||
613 | # Otherwise assume it belongs to the simple store. |
|
613 | # Otherwise assume it belongs to the simple store. | |
614 | return True |
|
614 | return True | |
615 |
|
615 | |||
616 | class simplestore(store.encodedstore): |
|
616 | class simplestore(store.encodedstore): | |
617 | def datafiles(self): |
|
617 | def datafiles(self): | |
618 | for x in super(simplestore, self).datafiles(): |
|
618 | for x in super(simplestore, self).datafiles(): | |
619 | yield x |
|
619 | yield x | |
620 |
|
620 | |||
621 | # Supplement with non-revlog files. |
|
621 | # Supplement with non-revlog files. | |
622 | extrafiles = self._walk('data', True, filefilter=issimplestorefile) |
|
622 | extrafiles = self._walk('data', True, filefilter=issimplestorefile) | |
623 |
|
623 | |||
624 | for unencoded, encoded, size in extrafiles: |
|
624 | for unencoded, encoded, size in extrafiles: | |
625 | try: |
|
625 | try: | |
626 | unencoded = store.decodefilename(unencoded) |
|
626 | unencoded = store.decodefilename(unencoded) | |
627 | except KeyError: |
|
627 | except KeyError: | |
628 | unencoded = None |
|
628 | unencoded = None | |
629 |
|
629 | |||
630 | yield unencoded, encoded, size |
|
630 | yield unencoded, encoded, size | |
631 |
|
631 | |||
632 | def reposetup(ui, repo): |
|
632 | def reposetup(ui, repo): | |
633 | if not repo.local(): |
|
633 | if not repo.local(): | |
634 | return |
|
634 | return | |
635 |
|
635 | |||
636 | if isinstance(repo, bundlerepo.bundlerepository): |
|
636 | if isinstance(repo, bundlerepo.bundlerepository): | |
637 | raise error.Abort(_('cannot use simple store with bundlerepo')) |
|
637 | raise error.Abort(_('cannot use simple store with bundlerepo')) | |
638 |
|
638 | |||
639 | class simplestorerepo(repo.__class__): |
|
639 | class simplestorerepo(repo.__class__): | |
640 | def file(self, f): |
|
640 | def file(self, f): | |
641 | return filestorage(self.svfs, f) |
|
641 | return filestorage(self.svfs, f) | |
642 |
|
642 | |||
643 | repo.__class__ = simplestorerepo |
|
643 | repo.__class__ = simplestorerepo | |
644 |
|
644 | |||
645 | def featuresetup(ui, supported): |
|
645 | def featuresetup(ui, supported): | |
646 | supported.add(REQUIREMENT) |
|
646 | supported.add(REQUIREMENT) | |
647 |
|
647 | |||
648 | def newreporequirements(orig, ui, createopts): |
|
648 | def newreporequirements(orig, ui, createopts): | |
649 | """Modifies default requirements for new repos to use the simple store.""" |
|
649 | """Modifies default requirements for new repos to use the simple store.""" | |
650 | requirements = orig(ui, createopts) |
|
650 | requirements = orig(ui, createopts) | |
651 |
|
651 | |||
652 | # These requirements are only used to affect creation of the store |
|
652 | # These requirements are only used to affect creation of the store | |
653 | # object. We have our own store. So we can remove them. |
|
653 | # object. We have our own store. So we can remove them. | |
654 | # TODO do this once we feel like taking the test hit. |
|
654 | # TODO do this once we feel like taking the test hit. | |
655 | #if 'fncache' in requirements: |
|
655 | #if 'fncache' in requirements: | |
656 | # requirements.remove('fncache') |
|
656 | # requirements.remove('fncache') | |
657 | #if 'dotencode' in requirements: |
|
657 | #if 'dotencode' in requirements: | |
658 | # requirements.remove('dotencode') |
|
658 | # requirements.remove('dotencode') | |
659 |
|
659 | |||
660 | requirements.add(REQUIREMENT) |
|
660 | requirements.add(REQUIREMENT) | |
661 |
|
661 | |||
662 | return requirements |
|
662 | return requirements | |
663 |
|
663 | |||
664 | def makestore(orig, requirements, path, vfstype): |
|
664 | def makestore(orig, requirements, path, vfstype): | |
665 | if REQUIREMENT not in requirements: |
|
665 | if REQUIREMENT not in requirements: | |
666 | return orig(requirements, path, vfstype) |
|
666 | return orig(requirements, path, vfstype) | |
667 |
|
667 | |||
668 | return simplestore(path, vfstype) |
|
668 | return simplestore(path, vfstype) | |
669 |
|
669 | |||
670 | def verifierinit(orig, self, *args, **kwargs): |
|
670 | def verifierinit(orig, self, *args, **kwargs): | |
671 | orig(self, *args, **kwargs) |
|
671 | orig(self, *args, **kwargs) | |
672 |
|
672 | |||
673 | # We don't care that files in the store don't align with what is |
|
673 | # We don't care that files in the store don't align with what is | |
674 | # advertised. So suppress these warnings. |
|
674 | # advertised. So suppress these warnings. | |
675 | self.warnorphanstorefiles = False |
|
675 | self.warnorphanstorefiles = False | |
676 |
|
676 | |||
677 | def extsetup(ui): |
|
677 | def extsetup(ui): | |
678 | localrepo.featuresetupfuncs.add(featuresetup) |
|
678 | localrepo.featuresetupfuncs.add(featuresetup) | |
679 |
|
679 | |||
680 | extensions.wrapfunction(localrepo, 'newreporequirements', |
|
680 | extensions.wrapfunction(localrepo, 'newreporequirements', | |
681 | newreporequirements) |
|
681 | newreporequirements) | |
682 | extensions.wrapfunction(localrepo, 'makestore', makestore) |
|
682 | extensions.wrapfunction(localrepo, 'makestore', makestore) | |
683 | extensions.wrapfunction(verify.verifier, '__init__', verifierinit) |
|
683 | extensions.wrapfunction(verify.verifier, '__init__', verifierinit) |
General Comments 0
You need to be logged in to leave comments.
Login now