Show More
@@ -1,318 +1,211 b'' | |||
|
1 | 1 | # narrowchangegroup.py - narrow clone changegroup creation and consumption |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2017 Google, Inc. |
|
4 | 4 | # |
|
5 | 5 | # This software may be used and distributed according to the terms of the |
|
6 | 6 | # GNU General Public License version 2 or any later version. |
|
7 | 7 | |
|
8 | 8 | from __future__ import absolute_import |
|
9 | 9 | |
|
10 | 10 | from mercurial.i18n import _ |
|
11 | 11 | from mercurial import ( |
|
12 | 12 | changegroup, |
|
13 | 13 | error, |
|
14 | 14 | extensions, |
|
15 | 15 | node, |
|
16 | pycompat, | |
|
17 | 16 | util, |
|
18 | 17 | ) |
|
19 | 18 | |
|
20 | 19 | def setup(): |
|
21 | 20 | def generatefiles(orig, self, changedfiles, linknodes, commonrevs, |
|
22 | 21 | source): |
|
23 | 22 | changedfiles = list(filter(self._filematcher, changedfiles)) |
|
24 | 23 | |
|
25 | 24 | if getattr(self, 'is_shallow', False): |
|
26 | 25 | # See comment in generate() for why this sadness is a thing. |
|
27 | 26 | mfdicts = self._mfdicts |
|
28 | 27 | del self._mfdicts |
|
29 | 28 | # In a shallow clone, the linknodes callback needs to also include |
|
30 | 29 | # those file nodes that are in the manifests we sent but weren't |
|
31 | 30 | # introduced by those manifests. |
|
32 | 31 | commonctxs = [self._repo[c] for c in commonrevs] |
|
33 | 32 | oldlinknodes = linknodes |
|
34 | 33 | clrev = self._repo.changelog.rev |
|
35 | 34 | def linknodes(flog, fname): |
|
36 | 35 | for c in commonctxs: |
|
37 | 36 | try: |
|
38 | 37 | fnode = c.filenode(fname) |
|
39 | 38 | self.clrev_to_localrev[c.rev()] = flog.rev(fnode) |
|
40 | 39 | except error.ManifestLookupError: |
|
41 | 40 | pass |
|
42 | 41 | links = oldlinknodes(flog, fname) |
|
43 | 42 | if len(links) != len(mfdicts): |
|
44 | 43 | for mf, lr in mfdicts: |
|
45 | 44 | fnode = mf.get(fname, None) |
|
46 | 45 | if fnode in links: |
|
47 | 46 | links[fnode] = min(links[fnode], lr, key=clrev) |
|
48 | 47 | elif fnode: |
|
49 | 48 | links[fnode] = lr |
|
50 | 49 | return links |
|
51 | 50 | return orig(self, changedfiles, linknodes, commonrevs, source) |
|
52 | 51 | extensions.wrapfunction( |
|
53 | 52 | changegroup.cg1packer, 'generatefiles', generatefiles) |
|
54 | 53 | |
|
55 | 54 | def close(orig, self): |
|
56 | 55 | getattr(self, 'clrev_to_localrev', {}).clear() |
|
57 | 56 | if getattr(self, 'next_clrev_to_localrev', {}): |
|
58 | 57 | self.clrev_to_localrev = self.next_clrev_to_localrev |
|
59 | 58 | del self.next_clrev_to_localrev |
|
60 | 59 | self.changelog_done = True |
|
61 | 60 | return orig(self) |
|
62 | 61 | extensions.wrapfunction(changegroup.cg1packer, 'close', close) |
|
63 | 62 | |
|
64 | 63 | # In a perfect world, we'd generate better ellipsis-ified graphs |
|
65 | 64 | # for non-changelog revlogs. In practice, we haven't started doing |
|
66 | 65 | # that yet, so the resulting DAGs for the manifestlog and filelogs |
|
67 | 66 | # are actually full of bogus parentage on all the ellipsis |
|
68 | 67 | # nodes. This has the side effect that, while the contents are |
|
69 | 68 | # correct, the individual DAGs might be completely out of whack in |
|
70 | 69 | # a case like 882681bc3166 and its ancestors (back about 10 |
|
71 | 70 | # revisions or so) in the main hg repo. |
|
72 | 71 | # |
|
73 | 72 | # The one invariant we *know* holds is that the new (potentially |
|
74 | 73 | # bogus) DAG shape will be valid if we order the nodes in the |
|
75 | 74 | # order that they're introduced in dramatis personae by the |
|
76 | 75 | # changelog, so what we do is we sort the non-changelog histories |
|
77 | 76 | # by the order in which they are used by the changelog. |
|
78 | 77 | def _sortgroup(orig, self, revlog, nodelist, lookup): |
|
79 | 78 | if not util.safehasattr(self, 'full_nodes') or not self.clnode_to_rev: |
|
80 | 79 | return orig(self, revlog, nodelist, lookup) |
|
81 | 80 | key = lambda n: self.clnode_to_rev[lookup(n)] |
|
82 | 81 | return [revlog.rev(n) for n in sorted(nodelist, key=key)] |
|
83 | 82 | |
|
84 | 83 | extensions.wrapfunction(changegroup.cg1packer, '_sortgroup', _sortgroup) |
|
85 | 84 | |
|
86 | 85 | def generate(orig, self, commonrevs, clnodes, fastpathlinkrev, source): |
|
87 | 86 | '''yield a sequence of changegroup chunks (strings)''' |
|
88 | 87 | # Note: other than delegating to orig, the only deviation in |
|
89 | 88 | # logic from normal hg's generate is marked with BEGIN/END |
|
90 | 89 | # NARROW HACK. |
|
91 | 90 | if not util.safehasattr(self, 'full_nodes'): |
|
92 | 91 | # not sending a narrow bundle |
|
93 | 92 | for x in orig(self, commonrevs, clnodes, fastpathlinkrev, source): |
|
94 | 93 | yield x |
|
95 | 94 | return |
|
96 | 95 | |
|
97 | 96 | repo = self._repo |
|
98 | 97 | cl = repo.changelog |
|
99 | 98 | mfl = repo.manifestlog |
|
100 | 99 | mfrevlog = mfl._revlog |
|
101 | 100 | |
|
102 | 101 | clrevorder = {} |
|
103 | 102 | mfs = {} # needed manifests |
|
104 | 103 | fnodes = {} # needed file nodes |
|
105 | 104 | changedfiles = set() |
|
106 | 105 | |
|
107 | 106 | # Callback for the changelog, used to collect changed files and manifest |
|
108 | 107 | # nodes. |
|
109 | 108 | # Returns the linkrev node (identity in the changelog case). |
|
110 | 109 | def lookupcl(x): |
|
111 | 110 | c = cl.read(x) |
|
112 | 111 | clrevorder[x] = len(clrevorder) |
|
113 | 112 | # BEGIN NARROW HACK |
|
114 | 113 | # |
|
115 | 114 | # Only update mfs if x is going to be sent. Otherwise we |
|
116 | 115 | # end up with bogus linkrevs specified for manifests and |
|
117 | 116 | # we skip some manifest nodes that we should otherwise |
|
118 | 117 | # have sent. |
|
119 | 118 | if x in self.full_nodes or cl.rev(x) in self.precomputed_ellipsis: |
|
120 | 119 | n = c[0] |
|
121 | 120 | # record the first changeset introducing this manifest version |
|
122 | 121 | mfs.setdefault(n, x) |
|
123 | 122 | # Set this narrow-specific dict so we have the lowest manifest |
|
124 | 123 | # revnum to look up for this cl revnum. (Part of mapping |
|
125 | 124 | # changelog ellipsis parents to manifest ellipsis parents) |
|
126 | 125 | self.next_clrev_to_localrev.setdefault(cl.rev(x), |
|
127 | 126 | mfrevlog.rev(n)) |
|
128 | 127 | # We can't trust the changed files list in the changeset if the |
|
129 | 128 | # client requested a shallow clone. |
|
130 | 129 | if self.is_shallow: |
|
131 | 130 | changedfiles.update(mfl[c[0]].read().keys()) |
|
132 | 131 | else: |
|
133 | 132 | changedfiles.update(c[3]) |
|
134 | 133 | # END NARROW HACK |
|
135 | 134 | # Record a complete list of potentially-changed files in |
|
136 | 135 | # this manifest. |
|
137 | 136 | return x |
|
138 | 137 | |
|
139 | 138 | self._verbosenote(_('uncompressed size of bundle content:\n')) |
|
140 | 139 | size = 0 |
|
141 | 140 | for chunk in self.group(clnodes, cl, lookupcl, units=_('changesets')): |
|
142 | 141 | size += len(chunk) |
|
143 | 142 | yield chunk |
|
144 | 143 | self._verbosenote(_('%8.i (changelog)\n') % size) |
|
145 | 144 | |
|
146 | 145 | # We need to make sure that the linkrev in the changegroup refers to |
|
147 | 146 | # the first changeset that introduced the manifest or file revision. |
|
148 | 147 | # The fastpath is usually safer than the slowpath, because the filelogs |
|
149 | 148 | # are walked in revlog order. |
|
150 | 149 | # |
|
151 | 150 | # When taking the slowpath with reorder=None and the manifest revlog |
|
152 | 151 | # uses generaldelta, the manifest may be walked in the "wrong" order. |
|
153 | 152 | # Without 'clrevorder', we would get an incorrect linkrev (see fix in |
|
154 | 153 | # cc0ff93d0c0c). |
|
155 | 154 | # |
|
156 | 155 | # When taking the fastpath, we are only vulnerable to reordering |
|
157 | 156 | # of the changelog itself. The changelog never uses generaldelta, so |
|
158 | 157 | # it is only reordered when reorder=True. To handle this case, we |
|
159 | 158 | # simply take the slowpath, which already has the 'clrevorder' logic. |
|
160 | 159 | # This was also fixed in cc0ff93d0c0c. |
|
161 | 160 | fastpathlinkrev = fastpathlinkrev and not self._reorder |
|
162 | 161 | # Treemanifests don't work correctly with fastpathlinkrev |
|
163 | 162 | # either, because we don't discover which directory nodes to |
|
164 | 163 | # send along with files. This could probably be fixed. |
|
165 | 164 | fastpathlinkrev = fastpathlinkrev and ( |
|
166 | 165 | 'treemanifest' not in repo.requirements) |
|
167 | 166 | # Shallow clones also don't work correctly with fastpathlinkrev |
|
168 | 167 | # because file nodes may need to be sent for a manifest even if they |
|
169 | 168 | # weren't introduced by that manifest. |
|
170 | 169 | fastpathlinkrev = fastpathlinkrev and not self.is_shallow |
|
171 | 170 | |
|
172 | 171 | for chunk in self.generatemanifests(commonrevs, clrevorder, |
|
173 | 172 | fastpathlinkrev, mfs, fnodes, source): |
|
174 | 173 | yield chunk |
|
175 | 174 | # BEGIN NARROW HACK |
|
176 | 175 | mfdicts = None |
|
177 | 176 | if self.is_shallow: |
|
178 | 177 | mfdicts = [(self._repo.manifestlog[n].read(), lr) |
|
179 | 178 | for (n, lr) in mfs.iteritems()] |
|
180 | 179 | # END NARROW HACK |
|
181 | 180 | mfs.clear() |
|
182 | 181 | clrevs = set(cl.rev(x) for x in clnodes) |
|
183 | 182 | |
|
184 | 183 | if not fastpathlinkrev: |
|
185 | 184 | def linknodes(unused, fname): |
|
186 | 185 | return fnodes.get(fname, {}) |
|
187 | 186 | else: |
|
188 | 187 | cln = cl.node |
|
189 | 188 | def linknodes(filerevlog, fname): |
|
190 | 189 | llr = filerevlog.linkrev |
|
191 | 190 | fln = filerevlog.node |
|
192 | 191 | revs = ((r, llr(r)) for r in filerevlog) |
|
193 | 192 | return dict((fln(r), cln(lr)) for r, lr in revs if lr in clrevs) |
|
194 | 193 | |
|
195 | 194 | # BEGIN NARROW HACK |
|
196 | 195 | # |
|
197 | 196 | # We need to pass the mfdicts variable down into |
|
198 | 197 | # generatefiles(), but more than one command might have |
|
199 | 198 | # wrapped generatefiles so we can't modify the function |
|
200 | 199 | # signature. Instead, we pass the data to ourselves using an |
|
201 | 200 | # instance attribute. I'm sorry. |
|
202 | 201 | self._mfdicts = mfdicts |
|
203 | 202 | # END NARROW HACK |
|
204 | 203 | for chunk in self.generatefiles(changedfiles, linknodes, commonrevs, |
|
205 | 204 | source): |
|
206 | 205 | yield chunk |
|
207 | 206 | |
|
208 | 207 | yield self.close() |
|
209 | 208 | |
|
210 | 209 | if clnodes: |
|
211 | 210 | repo.hook('outgoing', node=node.hex(clnodes[0]), source=source) |
|
212 | 211 | extensions.wrapfunction(changegroup.cg1packer, 'generate', generate) |
|
213 | ||
|
214 | def revchunk(orig, self, revlog, rev, prev, linknode): | |
|
215 | if not util.safehasattr(self, 'full_nodes'): | |
|
216 | # not sending a narrow changegroup | |
|
217 | for x in orig(self, revlog, rev, prev, linknode): | |
|
218 | yield x | |
|
219 | return | |
|
220 | # build up some mapping information that's useful later. See | |
|
221 | # the local() nested function below. | |
|
222 | if not self.changelog_done: | |
|
223 | self.clnode_to_rev[linknode] = rev | |
|
224 | linkrev = rev | |
|
225 | self.clrev_to_localrev[linkrev] = rev | |
|
226 | else: | |
|
227 | linkrev = self.clnode_to_rev[linknode] | |
|
228 | self.clrev_to_localrev[linkrev] = rev | |
|
229 | # This is a node to send in full, because the changeset it | |
|
230 | # corresponds to was a full changeset. | |
|
231 | if linknode in self.full_nodes: | |
|
232 | for x in orig(self, revlog, rev, prev, linknode): | |
|
233 | yield x | |
|
234 | return | |
|
235 | # At this point, a node can either be one we should skip or an | |
|
236 | # ellipsis. If it's not an ellipsis, bail immediately. | |
|
237 | if linkrev not in self.precomputed_ellipsis: | |
|
238 | return | |
|
239 | linkparents = self.precomputed_ellipsis[linkrev] | |
|
240 | def local(clrev): | |
|
241 | """Turn a changelog revnum into a local revnum. | |
|
242 | ||
|
243 | The ellipsis dag is stored as revnums on the changelog, | |
|
244 | but when we're producing ellipsis entries for | |
|
245 | non-changelog revlogs, we need to turn those numbers into | |
|
246 | something local. This does that for us, and during the | |
|
247 | changelog sending phase will also expand the stored | |
|
248 | mappings as needed. | |
|
249 | """ | |
|
250 | if clrev == node.nullrev: | |
|
251 | return node.nullrev | |
|
252 | if not self.changelog_done: | |
|
253 | # If we're doing the changelog, it's possible that we | |
|
254 | # have a parent that is already on the client, and we | |
|
255 | # need to store some extra mapping information so that | |
|
256 | # our contained ellipsis nodes will be able to resolve | |
|
257 | # their parents. | |
|
258 | if clrev not in self.clrev_to_localrev: | |
|
259 | clnode = revlog.node(clrev) | |
|
260 | self.clnode_to_rev[clnode] = clrev | |
|
261 | return clrev | |
|
262 | # Walk the ellipsis-ized changelog breadth-first looking for a | |
|
263 | # change that has been linked from the current revlog. | |
|
264 | # | |
|
265 | # For a flat manifest revlog only a single step should be necessary | |
|
266 | # as all relevant changelog entries are relevant to the flat | |
|
267 | # manifest. | |
|
268 | # | |
|
269 | # For a filelog or tree manifest dirlog however not every changelog | |
|
270 | # entry will have been relevant, so we need to skip some changelog | |
|
271 | # nodes even after ellipsis-izing. | |
|
272 | walk = [clrev] | |
|
273 | while walk: | |
|
274 | p = walk[0] | |
|
275 | walk = walk[1:] | |
|
276 | if p in self.clrev_to_localrev: | |
|
277 | return self.clrev_to_localrev[p] | |
|
278 | elif p in self.full_nodes: | |
|
279 | walk.extend([pp for pp in self._repo.changelog.parentrevs(p) | |
|
280 | if pp != node.nullrev]) | |
|
281 | elif p in self.precomputed_ellipsis: | |
|
282 | walk.extend([pp for pp in self.precomputed_ellipsis[p] | |
|
283 | if pp != node.nullrev]) | |
|
284 | else: | |
|
285 | # In this case, we've got an ellipsis with parents | |
|
286 | # outside the current bundle (likely an | |
|
287 | # incremental pull). We "know" that we can use the | |
|
288 | # value of this same revlog at whatever revision | |
|
289 | # is pointed to by linknode. "Know" is in scare | |
|
290 | # quotes because I haven't done enough examination | |
|
291 | # of edge cases to convince myself this is really | |
|
292 | # a fact - it works for all the (admittedly | |
|
293 | # thorough) cases in our testsuite, but I would be | |
|
294 | # somewhat unsurprised to find a case in the wild | |
|
295 | # where this breaks down a bit. That said, I don't | |
|
296 | # know if it would hurt anything. | |
|
297 | for i in pycompat.xrange(rev, 0, -1): | |
|
298 | if revlog.linkrev(i) == clrev: | |
|
299 | return i | |
|
300 | # We failed to resolve a parent for this node, so | |
|
301 | # we crash the changegroup construction. | |
|
302 | raise error.Abort( | |
|
303 | 'unable to resolve parent while packing %r %r' | |
|
304 | ' for changeset %r' % (revlog.indexfile, rev, clrev)) | |
|
305 | return node.nullrev | |
|
306 | ||
|
307 | if not linkparents or ( | |
|
308 | revlog.parentrevs(rev) == (node.nullrev, node.nullrev)): | |
|
309 | p1, p2 = node.nullrev, node.nullrev | |
|
310 | elif len(linkparents) == 1: | |
|
311 | p1, = sorted(local(p) for p in linkparents) | |
|
312 | p2 = node.nullrev | |
|
313 | else: | |
|
314 | p1, p2 = sorted(local(p) for p in linkparents) | |
|
315 | n = revlog.node(rev) | |
|
316 | yield changegroup.ellipsisdata( | |
|
317 | self, rev, revlog, p1, p2, revlog.revision(n), linknode) | |
|
318 | extensions.wrapfunction(changegroup.cg1packer, 'revchunk', revchunk) |
@@ -1,1105 +1,1221 b'' | |||
|
1 | 1 | # changegroup.py - Mercurial changegroup manipulation functions |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2006 Matt Mackall <mpm@selenic.com> |
|
4 | 4 | # |
|
5 | 5 | # This software may be used and distributed according to the terms of the |
|
6 | 6 | # GNU General Public License version 2 or any later version. |
|
7 | 7 | |
|
8 | 8 | from __future__ import absolute_import |
|
9 | 9 | |
|
10 | 10 | import os |
|
11 | 11 | import struct |
|
12 | 12 | import weakref |
|
13 | 13 | |
|
14 | 14 | from .i18n import _ |
|
15 | 15 | from .node import ( |
|
16 | 16 | hex, |
|
17 | 17 | nullid, |
|
18 | 18 | nullrev, |
|
19 | 19 | short, |
|
20 | 20 | ) |
|
21 | 21 | |
|
22 | 22 | from . import ( |
|
23 | 23 | dagutil, |
|
24 | 24 | error, |
|
25 | 25 | manifest, |
|
26 | 26 | match as matchmod, |
|
27 | 27 | mdiff, |
|
28 | 28 | phases, |
|
29 | 29 | pycompat, |
|
30 | 30 | repository, |
|
31 | 31 | revlog, |
|
32 | 32 | util, |
|
33 | 33 | ) |
|
34 | 34 | |
|
35 | 35 | from .utils import ( |
|
36 | 36 | stringutil, |
|
37 | 37 | ) |
|
38 | 38 | |
|
39 | 39 | _CHANGEGROUPV1_DELTA_HEADER = "20s20s20s20s" |
|
40 | 40 | _CHANGEGROUPV2_DELTA_HEADER = "20s20s20s20s20s" |
|
41 | 41 | _CHANGEGROUPV3_DELTA_HEADER = ">20s20s20s20s20sH" |
|
42 | 42 | |
|
43 | 43 | LFS_REQUIREMENT = 'lfs' |
|
44 | 44 | |
|
45 | 45 | readexactly = util.readexactly |
|
46 | 46 | |
|
47 | 47 | def getchunk(stream): |
|
48 | 48 | """return the next chunk from stream as a string""" |
|
49 | 49 | d = readexactly(stream, 4) |
|
50 | 50 | l = struct.unpack(">l", d)[0] |
|
51 | 51 | if l <= 4: |
|
52 | 52 | if l: |
|
53 | 53 | raise error.Abort(_("invalid chunk length %d") % l) |
|
54 | 54 | return "" |
|
55 | 55 | return readexactly(stream, l - 4) |
|
56 | 56 | |
|
57 | 57 | def chunkheader(length): |
|
58 | 58 | """return a changegroup chunk header (string)""" |
|
59 | 59 | return struct.pack(">l", length + 4) |
|
60 | 60 | |
|
61 | 61 | def closechunk(): |
|
62 | 62 | """return a changegroup chunk header (string) for a zero-length chunk""" |
|
63 | 63 | return struct.pack(">l", 0) |
|
64 | 64 | |
|
65 | 65 | def writechunks(ui, chunks, filename, vfs=None): |
|
66 | 66 | """Write chunks to a file and return its filename. |
|
67 | 67 | |
|
68 | 68 | The stream is assumed to be a bundle file. |
|
69 | 69 | Existing files will not be overwritten. |
|
70 | 70 | If no filename is specified, a temporary file is created. |
|
71 | 71 | """ |
|
72 | 72 | fh = None |
|
73 | 73 | cleanup = None |
|
74 | 74 | try: |
|
75 | 75 | if filename: |
|
76 | 76 | if vfs: |
|
77 | 77 | fh = vfs.open(filename, "wb") |
|
78 | 78 | else: |
|
79 | 79 | # Increase default buffer size because default is usually |
|
80 | 80 | # small (4k is common on Linux). |
|
81 | 81 | fh = open(filename, "wb", 131072) |
|
82 | 82 | else: |
|
83 | 83 | fd, filename = pycompat.mkstemp(prefix="hg-bundle-", suffix=".hg") |
|
84 | 84 | fh = os.fdopen(fd, r"wb") |
|
85 | 85 | cleanup = filename |
|
86 | 86 | for c in chunks: |
|
87 | 87 | fh.write(c) |
|
88 | 88 | cleanup = None |
|
89 | 89 | return filename |
|
90 | 90 | finally: |
|
91 | 91 | if fh is not None: |
|
92 | 92 | fh.close() |
|
93 | 93 | if cleanup is not None: |
|
94 | 94 | if filename and vfs: |
|
95 | 95 | vfs.unlink(cleanup) |
|
96 | 96 | else: |
|
97 | 97 | os.unlink(cleanup) |
|
98 | 98 | |
|
99 | 99 | class cg1unpacker(object): |
|
100 | 100 | """Unpacker for cg1 changegroup streams. |
|
101 | 101 | |
|
102 | 102 | A changegroup unpacker handles the framing of the revision data in |
|
103 | 103 | the wire format. Most consumers will want to use the apply() |
|
104 | 104 | method to add the changes from the changegroup to a repository. |
|
105 | 105 | |
|
106 | 106 | If you're forwarding a changegroup unmodified to another consumer, |
|
107 | 107 | use getchunks(), which returns an iterator of changegroup |
|
108 | 108 | chunks. This is mostly useful for cases where you need to know the |
|
109 | 109 | data stream has ended by observing the end of the changegroup. |
|
110 | 110 | |
|
111 | 111 | deltachunk() is useful only if you're applying delta data. Most |
|
112 | 112 | consumers should prefer apply() instead. |
|
113 | 113 | |
|
114 | 114 | A few other public methods exist. Those are used only for |
|
115 | 115 | bundlerepo and some debug commands - their use is discouraged. |
|
116 | 116 | """ |
|
117 | 117 | deltaheader = _CHANGEGROUPV1_DELTA_HEADER |
|
118 | 118 | deltaheadersize = struct.calcsize(deltaheader) |
|
119 | 119 | version = '01' |
|
120 | 120 | _grouplistcount = 1 # One list of files after the manifests |
|
121 | 121 | |
|
122 | 122 | def __init__(self, fh, alg, extras=None): |
|
123 | 123 | if alg is None: |
|
124 | 124 | alg = 'UN' |
|
125 | 125 | if alg not in util.compengines.supportedbundletypes: |
|
126 | 126 | raise error.Abort(_('unknown stream compression type: %s') |
|
127 | 127 | % alg) |
|
128 | 128 | if alg == 'BZ': |
|
129 | 129 | alg = '_truncatedBZ' |
|
130 | 130 | |
|
131 | 131 | compengine = util.compengines.forbundletype(alg) |
|
132 | 132 | self._stream = compengine.decompressorreader(fh) |
|
133 | 133 | self._type = alg |
|
134 | 134 | self.extras = extras or {} |
|
135 | 135 | self.callback = None |
|
136 | 136 | |
|
137 | 137 | # These methods (compressed, read, seek, tell) all appear to only |
|
138 | 138 | # be used by bundlerepo, but it's a little hard to tell. |
|
139 | 139 | def compressed(self): |
|
140 | 140 | return self._type is not None and self._type != 'UN' |
|
141 | 141 | def read(self, l): |
|
142 | 142 | return self._stream.read(l) |
|
143 | 143 | def seek(self, pos): |
|
144 | 144 | return self._stream.seek(pos) |
|
145 | 145 | def tell(self): |
|
146 | 146 | return self._stream.tell() |
|
147 | 147 | def close(self): |
|
148 | 148 | return self._stream.close() |
|
149 | 149 | |
|
150 | 150 | def _chunklength(self): |
|
151 | 151 | d = readexactly(self._stream, 4) |
|
152 | 152 | l = struct.unpack(">l", d)[0] |
|
153 | 153 | if l <= 4: |
|
154 | 154 | if l: |
|
155 | 155 | raise error.Abort(_("invalid chunk length %d") % l) |
|
156 | 156 | return 0 |
|
157 | 157 | if self.callback: |
|
158 | 158 | self.callback() |
|
159 | 159 | return l - 4 |
|
160 | 160 | |
|
161 | 161 | def changelogheader(self): |
|
162 | 162 | """v10 does not have a changelog header chunk""" |
|
163 | 163 | return {} |
|
164 | 164 | |
|
165 | 165 | def manifestheader(self): |
|
166 | 166 | """v10 does not have a manifest header chunk""" |
|
167 | 167 | return {} |
|
168 | 168 | |
|
169 | 169 | def filelogheader(self): |
|
170 | 170 | """return the header of the filelogs chunk, v10 only has the filename""" |
|
171 | 171 | l = self._chunklength() |
|
172 | 172 | if not l: |
|
173 | 173 | return {} |
|
174 | 174 | fname = readexactly(self._stream, l) |
|
175 | 175 | return {'filename': fname} |
|
176 | 176 | |
|
177 | 177 | def _deltaheader(self, headertuple, prevnode): |
|
178 | 178 | node, p1, p2, cs = headertuple |
|
179 | 179 | if prevnode is None: |
|
180 | 180 | deltabase = p1 |
|
181 | 181 | else: |
|
182 | 182 | deltabase = prevnode |
|
183 | 183 | flags = 0 |
|
184 | 184 | return node, p1, p2, deltabase, cs, flags |
|
185 | 185 | |
|
186 | 186 | def deltachunk(self, prevnode): |
|
187 | 187 | l = self._chunklength() |
|
188 | 188 | if not l: |
|
189 | 189 | return {} |
|
190 | 190 | headerdata = readexactly(self._stream, self.deltaheadersize) |
|
191 | 191 | header = struct.unpack(self.deltaheader, headerdata) |
|
192 | 192 | delta = readexactly(self._stream, l - self.deltaheadersize) |
|
193 | 193 | node, p1, p2, deltabase, cs, flags = self._deltaheader(header, prevnode) |
|
194 | 194 | return (node, p1, p2, cs, deltabase, delta, flags) |
|
195 | 195 | |
|
196 | 196 | def getchunks(self): |
|
197 | 197 | """returns all the chunks contains in the bundle |
|
198 | 198 | |
|
199 | 199 | Used when you need to forward the binary stream to a file or another |
|
200 | 200 | network API. To do so, it parse the changegroup data, otherwise it will |
|
201 | 201 | block in case of sshrepo because it don't know the end of the stream. |
|
202 | 202 | """ |
|
203 | 203 | # For changegroup 1 and 2, we expect 3 parts: changelog, manifestlog, |
|
204 | 204 | # and a list of filelogs. For changegroup 3, we expect 4 parts: |
|
205 | 205 | # changelog, manifestlog, a list of tree manifestlogs, and a list of |
|
206 | 206 | # filelogs. |
|
207 | 207 | # |
|
208 | 208 | # Changelog and manifestlog parts are terminated with empty chunks. The |
|
209 | 209 | # tree and file parts are a list of entry sections. Each entry section |
|
210 | 210 | # is a series of chunks terminating in an empty chunk. The list of these |
|
211 | 211 | # entry sections is terminated in yet another empty chunk, so we know |
|
212 | 212 | # we've reached the end of the tree/file list when we reach an empty |
|
213 | 213 | # chunk that was proceeded by no non-empty chunks. |
|
214 | 214 | |
|
215 | 215 | parts = 0 |
|
216 | 216 | while parts < 2 + self._grouplistcount: |
|
217 | 217 | noentries = True |
|
218 | 218 | while True: |
|
219 | 219 | chunk = getchunk(self) |
|
220 | 220 | if not chunk: |
|
221 | 221 | # The first two empty chunks represent the end of the |
|
222 | 222 | # changelog and the manifestlog portions. The remaining |
|
223 | 223 | # empty chunks represent either A) the end of individual |
|
224 | 224 | # tree or file entries in the file list, or B) the end of |
|
225 | 225 | # the entire list. It's the end of the entire list if there |
|
226 | 226 | # were no entries (i.e. noentries is True). |
|
227 | 227 | if parts < 2: |
|
228 | 228 | parts += 1 |
|
229 | 229 | elif noentries: |
|
230 | 230 | parts += 1 |
|
231 | 231 | break |
|
232 | 232 | noentries = False |
|
233 | 233 | yield chunkheader(len(chunk)) |
|
234 | 234 | pos = 0 |
|
235 | 235 | while pos < len(chunk): |
|
236 | 236 | next = pos + 2**20 |
|
237 | 237 | yield chunk[pos:next] |
|
238 | 238 | pos = next |
|
239 | 239 | yield closechunk() |
|
240 | 240 | |
|
241 | 241 | def _unpackmanifests(self, repo, revmap, trp, prog): |
|
242 | 242 | self.callback = prog.increment |
|
243 | 243 | # no need to check for empty manifest group here: |
|
244 | 244 | # if the result of the merge of 1 and 2 is the same in 3 and 4, |
|
245 | 245 | # no new manifest will be created and the manifest group will |
|
246 | 246 | # be empty during the pull |
|
247 | 247 | self.manifestheader() |
|
248 | 248 | deltas = self.deltaiter() |
|
249 | 249 | repo.manifestlog.addgroup(deltas, revmap, trp) |
|
250 | 250 | prog.complete() |
|
251 | 251 | self.callback = None |
|
252 | 252 | |
|
253 | 253 | def apply(self, repo, tr, srctype, url, targetphase=phases.draft, |
|
254 | 254 | expectedtotal=None): |
|
255 | 255 | """Add the changegroup returned by source.read() to this repo. |
|
256 | 256 | srctype is a string like 'push', 'pull', or 'unbundle'. url is |
|
257 | 257 | the URL of the repo where this changegroup is coming from. |
|
258 | 258 | |
|
259 | 259 | Return an integer summarizing the change to this repo: |
|
260 | 260 | - nothing changed or no source: 0 |
|
261 | 261 | - more heads than before: 1+added heads (2..n) |
|
262 | 262 | - fewer heads than before: -1-removed heads (-2..-n) |
|
263 | 263 | - number of heads stays the same: 1 |
|
264 | 264 | """ |
|
265 | 265 | repo = repo.unfiltered() |
|
266 | 266 | def csmap(x): |
|
267 | 267 | repo.ui.debug("add changeset %s\n" % short(x)) |
|
268 | 268 | return len(cl) |
|
269 | 269 | |
|
270 | 270 | def revmap(x): |
|
271 | 271 | return cl.rev(x) |
|
272 | 272 | |
|
273 | 273 | changesets = files = revisions = 0 |
|
274 | 274 | |
|
275 | 275 | try: |
|
276 | 276 | # The transaction may already carry source information. In this |
|
277 | 277 | # case we use the top level data. We overwrite the argument |
|
278 | 278 | # because we need to use the top level value (if they exist) |
|
279 | 279 | # in this function. |
|
280 | 280 | srctype = tr.hookargs.setdefault('source', srctype) |
|
281 | 281 | url = tr.hookargs.setdefault('url', url) |
|
282 | 282 | repo.hook('prechangegroup', |
|
283 | 283 | throw=True, **pycompat.strkwargs(tr.hookargs)) |
|
284 | 284 | |
|
285 | 285 | # write changelog data to temp files so concurrent readers |
|
286 | 286 | # will not see an inconsistent view |
|
287 | 287 | cl = repo.changelog |
|
288 | 288 | cl.delayupdate(tr) |
|
289 | 289 | oldheads = set(cl.heads()) |
|
290 | 290 | |
|
291 | 291 | trp = weakref.proxy(tr) |
|
292 | 292 | # pull off the changeset group |
|
293 | 293 | repo.ui.status(_("adding changesets\n")) |
|
294 | 294 | clstart = len(cl) |
|
295 | 295 | progress = repo.ui.makeprogress(_('changesets'), unit=_('chunks'), |
|
296 | 296 | total=expectedtotal) |
|
297 | 297 | self.callback = progress.increment |
|
298 | 298 | |
|
299 | 299 | efiles = set() |
|
300 | 300 | def onchangelog(cl, node): |
|
301 | 301 | efiles.update(cl.readfiles(node)) |
|
302 | 302 | |
|
303 | 303 | self.changelogheader() |
|
304 | 304 | deltas = self.deltaiter() |
|
305 | 305 | cgnodes = cl.addgroup(deltas, csmap, trp, addrevisioncb=onchangelog) |
|
306 | 306 | efiles = len(efiles) |
|
307 | 307 | |
|
308 | 308 | if not cgnodes: |
|
309 | 309 | repo.ui.develwarn('applied empty changegroup', |
|
310 | 310 | config='warn-empty-changegroup') |
|
311 | 311 | clend = len(cl) |
|
312 | 312 | changesets = clend - clstart |
|
313 | 313 | progress.complete() |
|
314 | 314 | self.callback = None |
|
315 | 315 | |
|
316 | 316 | # pull off the manifest group |
|
317 | 317 | repo.ui.status(_("adding manifests\n")) |
|
318 | 318 | # We know that we'll never have more manifests than we had |
|
319 | 319 | # changesets. |
|
320 | 320 | progress = repo.ui.makeprogress(_('manifests'), unit=_('chunks'), |
|
321 | 321 | total=changesets) |
|
322 | 322 | self._unpackmanifests(repo, revmap, trp, progress) |
|
323 | 323 | |
|
324 | 324 | needfiles = {} |
|
325 | 325 | if repo.ui.configbool('server', 'validate'): |
|
326 | 326 | cl = repo.changelog |
|
327 | 327 | ml = repo.manifestlog |
|
328 | 328 | # validate incoming csets have their manifests |
|
329 | 329 | for cset in pycompat.xrange(clstart, clend): |
|
330 | 330 | mfnode = cl.changelogrevision(cset).manifest |
|
331 | 331 | mfest = ml[mfnode].readdelta() |
|
332 | 332 | # store file cgnodes we must see |
|
333 | 333 | for f, n in mfest.iteritems(): |
|
334 | 334 | needfiles.setdefault(f, set()).add(n) |
|
335 | 335 | |
|
336 | 336 | # process the files |
|
337 | 337 | repo.ui.status(_("adding file changes\n")) |
|
338 | 338 | newrevs, newfiles = _addchangegroupfiles( |
|
339 | 339 | repo, self, revmap, trp, efiles, needfiles) |
|
340 | 340 | revisions += newrevs |
|
341 | 341 | files += newfiles |
|
342 | 342 | |
|
343 | 343 | deltaheads = 0 |
|
344 | 344 | if oldheads: |
|
345 | 345 | heads = cl.heads() |
|
346 | 346 | deltaheads = len(heads) - len(oldheads) |
|
347 | 347 | for h in heads: |
|
348 | 348 | if h not in oldheads and repo[h].closesbranch(): |
|
349 | 349 | deltaheads -= 1 |
|
350 | 350 | htext = "" |
|
351 | 351 | if deltaheads: |
|
352 | 352 | htext = _(" (%+d heads)") % deltaheads |
|
353 | 353 | |
|
354 | 354 | repo.ui.status(_("added %d changesets" |
|
355 | 355 | " with %d changes to %d files%s\n") |
|
356 | 356 | % (changesets, revisions, files, htext)) |
|
357 | 357 | repo.invalidatevolatilesets() |
|
358 | 358 | |
|
359 | 359 | if changesets > 0: |
|
360 | 360 | if 'node' not in tr.hookargs: |
|
361 | 361 | tr.hookargs['node'] = hex(cl.node(clstart)) |
|
362 | 362 | tr.hookargs['node_last'] = hex(cl.node(clend - 1)) |
|
363 | 363 | hookargs = dict(tr.hookargs) |
|
364 | 364 | else: |
|
365 | 365 | hookargs = dict(tr.hookargs) |
|
366 | 366 | hookargs['node'] = hex(cl.node(clstart)) |
|
367 | 367 | hookargs['node_last'] = hex(cl.node(clend - 1)) |
|
368 | 368 | repo.hook('pretxnchangegroup', |
|
369 | 369 | throw=True, **pycompat.strkwargs(hookargs)) |
|
370 | 370 | |
|
371 | 371 | added = [cl.node(r) for r in pycompat.xrange(clstart, clend)] |
|
372 | 372 | phaseall = None |
|
373 | 373 | if srctype in ('push', 'serve'): |
|
374 | 374 | # Old servers can not push the boundary themselves. |
|
375 | 375 | # New servers won't push the boundary if changeset already |
|
376 | 376 | # exists locally as secret |
|
377 | 377 | # |
|
378 | 378 | # We should not use added here but the list of all change in |
|
379 | 379 | # the bundle |
|
380 | 380 | if repo.publishing(): |
|
381 | 381 | targetphase = phaseall = phases.public |
|
382 | 382 | else: |
|
383 | 383 | # closer target phase computation |
|
384 | 384 | |
|
385 | 385 | # Those changesets have been pushed from the |
|
386 | 386 | # outside, their phases are going to be pushed |
|
387 | 387 | # alongside. Therefor `targetphase` is |
|
388 | 388 | # ignored. |
|
389 | 389 | targetphase = phaseall = phases.draft |
|
390 | 390 | if added: |
|
391 | 391 | phases.registernew(repo, tr, targetphase, added) |
|
392 | 392 | if phaseall is not None: |
|
393 | 393 | phases.advanceboundary(repo, tr, phaseall, cgnodes) |
|
394 | 394 | |
|
395 | 395 | if changesets > 0: |
|
396 | 396 | |
|
397 | 397 | def runhooks(): |
|
398 | 398 | # These hooks run when the lock releases, not when the |
|
399 | 399 | # transaction closes. So it's possible for the changelog |
|
400 | 400 | # to have changed since we last saw it. |
|
401 | 401 | if clstart >= len(repo): |
|
402 | 402 | return |
|
403 | 403 | |
|
404 | 404 | repo.hook("changegroup", **pycompat.strkwargs(hookargs)) |
|
405 | 405 | |
|
406 | 406 | for n in added: |
|
407 | 407 | args = hookargs.copy() |
|
408 | 408 | args['node'] = hex(n) |
|
409 | 409 | del args['node_last'] |
|
410 | 410 | repo.hook("incoming", **pycompat.strkwargs(args)) |
|
411 | 411 | |
|
412 | 412 | newheads = [h for h in repo.heads() |
|
413 | 413 | if h not in oldheads] |
|
414 | 414 | repo.ui.log("incoming", |
|
415 | 415 | "%d incoming changes - new heads: %s\n", |
|
416 | 416 | len(added), |
|
417 | 417 | ', '.join([hex(c[:6]) for c in newheads])) |
|
418 | 418 | |
|
419 | 419 | tr.addpostclose('changegroup-runhooks-%020i' % clstart, |
|
420 | 420 | lambda tr: repo._afterlock(runhooks)) |
|
421 | 421 | finally: |
|
422 | 422 | repo.ui.flush() |
|
423 | 423 | # never return 0 here: |
|
424 | 424 | if deltaheads < 0: |
|
425 | 425 | ret = deltaheads - 1 |
|
426 | 426 | else: |
|
427 | 427 | ret = deltaheads + 1 |
|
428 | 428 | return ret |
|
429 | 429 | |
|
430 | 430 | def deltaiter(self): |
|
431 | 431 | """ |
|
432 | 432 | returns an iterator of the deltas in this changegroup |
|
433 | 433 | |
|
434 | 434 | Useful for passing to the underlying storage system to be stored. |
|
435 | 435 | """ |
|
436 | 436 | chain = None |
|
437 | 437 | for chunkdata in iter(lambda: self.deltachunk(chain), {}): |
|
438 | 438 | # Chunkdata: (node, p1, p2, cs, deltabase, delta, flags) |
|
439 | 439 | yield chunkdata |
|
440 | 440 | chain = chunkdata[0] |
|
441 | 441 | |
|
442 | 442 | class cg2unpacker(cg1unpacker): |
|
443 | 443 | """Unpacker for cg2 streams. |
|
444 | 444 | |
|
445 | 445 | cg2 streams add support for generaldelta, so the delta header |
|
446 | 446 | format is slightly different. All other features about the data |
|
447 | 447 | remain the same. |
|
448 | 448 | """ |
|
449 | 449 | deltaheader = _CHANGEGROUPV2_DELTA_HEADER |
|
450 | 450 | deltaheadersize = struct.calcsize(deltaheader) |
|
451 | 451 | version = '02' |
|
452 | 452 | |
|
453 | 453 | def _deltaheader(self, headertuple, prevnode): |
|
454 | 454 | node, p1, p2, deltabase, cs = headertuple |
|
455 | 455 | flags = 0 |
|
456 | 456 | return node, p1, p2, deltabase, cs, flags |
|
457 | 457 | |
|
458 | 458 | class cg3unpacker(cg2unpacker): |
|
459 | 459 | """Unpacker for cg3 streams. |
|
460 | 460 | |
|
461 | 461 | cg3 streams add support for exchanging treemanifests and revlog |
|
462 | 462 | flags. It adds the revlog flags to the delta header and an empty chunk |
|
463 | 463 | separating manifests and files. |
|
464 | 464 | """ |
|
465 | 465 | deltaheader = _CHANGEGROUPV3_DELTA_HEADER |
|
466 | 466 | deltaheadersize = struct.calcsize(deltaheader) |
|
467 | 467 | version = '03' |
|
468 | 468 | _grouplistcount = 2 # One list of manifests and one list of files |
|
469 | 469 | |
|
470 | 470 | def _deltaheader(self, headertuple, prevnode): |
|
471 | 471 | node, p1, p2, deltabase, cs, flags = headertuple |
|
472 | 472 | return node, p1, p2, deltabase, cs, flags |
|
473 | 473 | |
|
474 | 474 | def _unpackmanifests(self, repo, revmap, trp, prog): |
|
475 | 475 | super(cg3unpacker, self)._unpackmanifests(repo, revmap, trp, prog) |
|
476 | 476 | for chunkdata in iter(self.filelogheader, {}): |
|
477 | 477 | # If we get here, there are directory manifests in the changegroup |
|
478 | 478 | d = chunkdata["filename"] |
|
479 | 479 | repo.ui.debug("adding %s revisions\n" % d) |
|
480 | 480 | dirlog = repo.manifestlog._revlog.dirlog(d) |
|
481 | 481 | deltas = self.deltaiter() |
|
482 | 482 | if not dirlog.addgroup(deltas, revmap, trp): |
|
483 | 483 | raise error.Abort(_("received dir revlog group is empty")) |
|
484 | 484 | |
|
485 | 485 | class headerlessfixup(object): |
|
486 | 486 | def __init__(self, fh, h): |
|
487 | 487 | self._h = h |
|
488 | 488 | self._fh = fh |
|
489 | 489 | def read(self, n): |
|
490 | 490 | if self._h: |
|
491 | 491 | d, self._h = self._h[:n], self._h[n:] |
|
492 | 492 | if len(d) < n: |
|
493 | 493 | d += readexactly(self._fh, n - len(d)) |
|
494 | 494 | return d |
|
495 | 495 | return readexactly(self._fh, n) |
|
496 | 496 | |
|
497 | 497 | def ellipsisdata(packer, rev, revlog_, p1, p2, data, linknode): |
|
498 | 498 | n = revlog_.node(rev) |
|
499 | 499 | p1n, p2n = revlog_.node(p1), revlog_.node(p2) |
|
500 | 500 | flags = revlog_.flags(rev) |
|
501 | 501 | flags |= revlog.REVIDX_ELLIPSIS |
|
502 | 502 | meta = packer.builddeltaheader( |
|
503 | 503 | n, p1n, p2n, nullid, linknode, flags) |
|
504 | 504 | # TODO: try and actually send deltas for ellipsis data blocks |
|
505 | 505 | diffheader = mdiff.trivialdiffheader(len(data)) |
|
506 | 506 | l = len(meta) + len(diffheader) + len(data) |
|
507 | 507 | return ''.join((chunkheader(l), |
|
508 | 508 | meta, |
|
509 | 509 | diffheader, |
|
510 | 510 | data)) |
|
511 | 511 | |
|
512 | 512 | class cg1packer(object): |
|
513 | 513 | deltaheader = _CHANGEGROUPV1_DELTA_HEADER |
|
514 | 514 | version = '01' |
|
515 | 515 | def __init__(self, repo, filematcher, bundlecaps=None): |
|
516 | 516 | """Given a source repo, construct a bundler. |
|
517 | 517 | |
|
518 | 518 | filematcher is a matcher that matches on files to include in the |
|
519 | 519 | changegroup. Used to facilitate sparse changegroups. |
|
520 | 520 | |
|
521 | 521 | bundlecaps is optional and can be used to specify the set of |
|
522 | 522 | capabilities which can be used to build the bundle. While bundlecaps is |
|
523 | 523 | unused in core Mercurial, extensions rely on this feature to communicate |
|
524 | 524 | capabilities to customize the changegroup packer. |
|
525 | 525 | """ |
|
526 | 526 | assert filematcher |
|
527 | 527 | self._filematcher = filematcher |
|
528 | 528 | |
|
529 | 529 | # Set of capabilities we can use to build the bundle. |
|
530 | 530 | if bundlecaps is None: |
|
531 | 531 | bundlecaps = set() |
|
532 | 532 | self._bundlecaps = bundlecaps |
|
533 | 533 | # experimental config: bundle.reorder |
|
534 | 534 | reorder = repo.ui.config('bundle', 'reorder') |
|
535 | 535 | if reorder == 'auto': |
|
536 | 536 | reorder = None |
|
537 | 537 | else: |
|
538 | 538 | reorder = stringutil.parsebool(reorder) |
|
539 | 539 | self._repo = repo |
|
540 | 540 | self._reorder = reorder |
|
541 | 541 | if self._repo.ui.verbose and not self._repo.ui.debugflag: |
|
542 | 542 | self._verbosenote = self._repo.ui.note |
|
543 | 543 | else: |
|
544 | 544 | self._verbosenote = lambda s: None |
|
545 | 545 | |
|
546 | 546 | def close(self): |
|
547 | 547 | return closechunk() |
|
548 | 548 | |
|
549 | 549 | def fileheader(self, fname): |
|
550 | 550 | return chunkheader(len(fname)) + fname |
|
551 | 551 | |
|
552 | 552 | # Extracted both for clarity and for overriding in extensions. |
|
553 | 553 | def _sortgroup(self, revlog, nodelist, lookup): |
|
554 | 554 | """Sort nodes for change group and turn them into revnums.""" |
|
555 | 555 | # for generaldelta revlogs, we linearize the revs; this will both be |
|
556 | 556 | # much quicker and generate a much smaller bundle |
|
557 | 557 | if (revlog._generaldelta and self._reorder is None) or self._reorder: |
|
558 | 558 | dag = dagutil.revlogdag(revlog) |
|
559 | 559 | return dag.linearize(set(revlog.rev(n) for n in nodelist)) |
|
560 | 560 | else: |
|
561 | 561 | return sorted([revlog.rev(n) for n in nodelist]) |
|
562 | 562 | |
|
563 | 563 | def group(self, nodelist, revlog, lookup, units=None): |
|
564 | 564 | """Calculate a delta group, yielding a sequence of changegroup chunks |
|
565 | 565 | (strings). |
|
566 | 566 | |
|
567 | 567 | Given a list of changeset revs, return a set of deltas and |
|
568 | 568 | metadata corresponding to nodes. The first delta is |
|
569 | 569 | first parent(nodelist[0]) -> nodelist[0], the receiver is |
|
570 | 570 | guaranteed to have this parent as it has all history before |
|
571 | 571 | these changesets. In the case firstparent is nullrev the |
|
572 | 572 | changegroup starts with a full revision. |
|
573 | 573 | |
|
574 | 574 | If units is not None, progress detail will be generated, units specifies |
|
575 | 575 | the type of revlog that is touched (changelog, manifest, etc.). |
|
576 | 576 | """ |
|
577 | 577 | # if we don't have any revisions touched by these changesets, bail |
|
578 | 578 | if len(nodelist) == 0: |
|
579 | 579 | yield self.close() |
|
580 | 580 | return |
|
581 | 581 | |
|
582 | 582 | revs = self._sortgroup(revlog, nodelist, lookup) |
|
583 | 583 | |
|
584 | 584 | # add the parent of the first rev |
|
585 | 585 | p = revlog.parentrevs(revs[0])[0] |
|
586 | 586 | revs.insert(0, p) |
|
587 | 587 | |
|
588 | 588 | # build deltas |
|
589 | 589 | progress = None |
|
590 | 590 | if units is not None: |
|
591 | 591 | progress = self._repo.ui.makeprogress(_('bundling'), unit=units, |
|
592 | 592 | total=(len(revs) - 1)) |
|
593 | 593 | for r in pycompat.xrange(len(revs) - 1): |
|
594 | 594 | if progress: |
|
595 | 595 | progress.update(r + 1) |
|
596 | 596 | prev, curr = revs[r], revs[r + 1] |
|
597 | 597 | linknode = lookup(revlog.node(curr)) |
|
598 | 598 | for c in self.revchunk(revlog, curr, prev, linknode): |
|
599 | 599 | yield c |
|
600 | 600 | |
|
601 | 601 | if progress: |
|
602 | 602 | progress.complete() |
|
603 | 603 | yield self.close() |
|
604 | 604 | |
|
605 | 605 | # filter any nodes that claim to be part of the known set |
|
606 | 606 | def prune(self, revlog, missing, commonrevs): |
|
607 | 607 | # TODO this violates storage abstraction for manifests. |
|
608 | 608 | if isinstance(revlog, manifest.manifestrevlog): |
|
609 | 609 | if not self._filematcher.visitdir(revlog._dir[:-1] or '.'): |
|
610 | 610 | return [] |
|
611 | 611 | |
|
612 | 612 | rr, rl = revlog.rev, revlog.linkrev |
|
613 | 613 | return [n for n in missing if rl(rr(n)) not in commonrevs] |
|
614 | 614 | |
|
615 | 615 | def _packmanifests(self, dir, mfnodes, lookuplinknode): |
|
616 | 616 | """Pack flat manifests into a changegroup stream.""" |
|
617 | 617 | assert not dir |
|
618 | 618 | for chunk in self.group(mfnodes, self._repo.manifestlog._revlog, |
|
619 | 619 | lookuplinknode, units=_('manifests')): |
|
620 | 620 | yield chunk |
|
621 | 621 | |
|
622 | 622 | def _manifestsdone(self): |
|
623 | 623 | return '' |
|
624 | 624 | |
|
625 | 625 | def generate(self, commonrevs, clnodes, fastpathlinkrev, source): |
|
626 | 626 | '''yield a sequence of changegroup chunks (strings)''' |
|
627 | 627 | repo = self._repo |
|
628 | 628 | cl = repo.changelog |
|
629 | 629 | |
|
630 | 630 | clrevorder = {} |
|
631 | 631 | mfs = {} # needed manifests |
|
632 | 632 | fnodes = {} # needed file nodes |
|
633 | 633 | changedfiles = set() |
|
634 | 634 | |
|
635 | 635 | # Callback for the changelog, used to collect changed files and manifest |
|
636 | 636 | # nodes. |
|
637 | 637 | # Returns the linkrev node (identity in the changelog case). |
|
638 | 638 | def lookupcl(x): |
|
639 | 639 | c = cl.read(x) |
|
640 | 640 | clrevorder[x] = len(clrevorder) |
|
641 | 641 | n = c[0] |
|
642 | 642 | # record the first changeset introducing this manifest version |
|
643 | 643 | mfs.setdefault(n, x) |
|
644 | 644 | # Record a complete list of potentially-changed files in |
|
645 | 645 | # this manifest. |
|
646 | 646 | changedfiles.update(c[3]) |
|
647 | 647 | return x |
|
648 | 648 | |
|
649 | 649 | self._verbosenote(_('uncompressed size of bundle content:\n')) |
|
650 | 650 | size = 0 |
|
651 | 651 | for chunk in self.group(clnodes, cl, lookupcl, units=_('changesets')): |
|
652 | 652 | size += len(chunk) |
|
653 | 653 | yield chunk |
|
654 | 654 | self._verbosenote(_('%8.i (changelog)\n') % size) |
|
655 | 655 | |
|
656 | 656 | # We need to make sure that the linkrev in the changegroup refers to |
|
657 | 657 | # the first changeset that introduced the manifest or file revision. |
|
658 | 658 | # The fastpath is usually safer than the slowpath, because the filelogs |
|
659 | 659 | # are walked in revlog order. |
|
660 | 660 | # |
|
661 | 661 | # When taking the slowpath with reorder=None and the manifest revlog |
|
662 | 662 | # uses generaldelta, the manifest may be walked in the "wrong" order. |
|
663 | 663 | # Without 'clrevorder', we would get an incorrect linkrev (see fix in |
|
664 | 664 | # cc0ff93d0c0c). |
|
665 | 665 | # |
|
666 | 666 | # When taking the fastpath, we are only vulnerable to reordering |
|
667 | 667 | # of the changelog itself. The changelog never uses generaldelta, so |
|
668 | 668 | # it is only reordered when reorder=True. To handle this case, we |
|
669 | 669 | # simply take the slowpath, which already has the 'clrevorder' logic. |
|
670 | 670 | # This was also fixed in cc0ff93d0c0c. |
|
671 | 671 | fastpathlinkrev = fastpathlinkrev and not self._reorder |
|
672 | 672 | # Treemanifests don't work correctly with fastpathlinkrev |
|
673 | 673 | # either, because we don't discover which directory nodes to |
|
674 | 674 | # send along with files. This could probably be fixed. |
|
675 | 675 | fastpathlinkrev = fastpathlinkrev and ( |
|
676 | 676 | 'treemanifest' not in repo.requirements) |
|
677 | 677 | |
|
678 | 678 | for chunk in self.generatemanifests(commonrevs, clrevorder, |
|
679 | 679 | fastpathlinkrev, mfs, fnodes, source): |
|
680 | 680 | yield chunk |
|
681 | 681 | mfs.clear() |
|
682 | 682 | clrevs = set(cl.rev(x) for x in clnodes) |
|
683 | 683 | |
|
684 | 684 | if not fastpathlinkrev: |
|
685 | 685 | def linknodes(unused, fname): |
|
686 | 686 | return fnodes.get(fname, {}) |
|
687 | 687 | else: |
|
688 | 688 | cln = cl.node |
|
689 | 689 | def linknodes(filerevlog, fname): |
|
690 | 690 | llr = filerevlog.linkrev |
|
691 | 691 | fln = filerevlog.node |
|
692 | 692 | revs = ((r, llr(r)) for r in filerevlog) |
|
693 | 693 | return dict((fln(r), cln(lr)) for r, lr in revs if lr in clrevs) |
|
694 | 694 | |
|
695 | 695 | for chunk in self.generatefiles(changedfiles, linknodes, commonrevs, |
|
696 | 696 | source): |
|
697 | 697 | yield chunk |
|
698 | 698 | |
|
699 | 699 | yield self.close() |
|
700 | 700 | |
|
701 | 701 | if clnodes: |
|
702 | 702 | repo.hook('outgoing', node=hex(clnodes[0]), source=source) |
|
703 | 703 | |
|
704 | 704 | def generatemanifests(self, commonrevs, clrevorder, fastpathlinkrev, mfs, |
|
705 | 705 | fnodes, source): |
|
706 | 706 | """Returns an iterator of changegroup chunks containing manifests. |
|
707 | 707 | |
|
708 | 708 | `source` is unused here, but is used by extensions like remotefilelog to |
|
709 | 709 | change what is sent based in pulls vs pushes, etc. |
|
710 | 710 | """ |
|
711 | 711 | repo = self._repo |
|
712 | 712 | mfl = repo.manifestlog |
|
713 | 713 | dirlog = mfl._revlog.dirlog |
|
714 | 714 | tmfnodes = {'': mfs} |
|
715 | 715 | |
|
716 | 716 | # Callback for the manifest, used to collect linkrevs for filelog |
|
717 | 717 | # revisions. |
|
718 | 718 | # Returns the linkrev node (collected in lookupcl). |
|
719 | 719 | def makelookupmflinknode(dir, nodes): |
|
720 | 720 | if fastpathlinkrev: |
|
721 | 721 | assert not dir |
|
722 | 722 | return mfs.__getitem__ |
|
723 | 723 | |
|
724 | 724 | def lookupmflinknode(x): |
|
725 | 725 | """Callback for looking up the linknode for manifests. |
|
726 | 726 | |
|
727 | 727 | Returns the linkrev node for the specified manifest. |
|
728 | 728 | |
|
729 | 729 | SIDE EFFECT: |
|
730 | 730 | |
|
731 | 731 | 1) fclnodes gets populated with the list of relevant |
|
732 | 732 | file nodes if we're not using fastpathlinkrev |
|
733 | 733 | 2) When treemanifests are in use, collects treemanifest nodes |
|
734 | 734 | to send |
|
735 | 735 | |
|
736 | 736 | Note that this means manifests must be completely sent to |
|
737 | 737 | the client before you can trust the list of files and |
|
738 | 738 | treemanifests to send. |
|
739 | 739 | """ |
|
740 | 740 | clnode = nodes[x] |
|
741 | 741 | mdata = mfl.get(dir, x).readfast(shallow=True) |
|
742 | 742 | for p, n, fl in mdata.iterentries(): |
|
743 | 743 | if fl == 't': # subdirectory manifest |
|
744 | 744 | subdir = dir + p + '/' |
|
745 | 745 | tmfclnodes = tmfnodes.setdefault(subdir, {}) |
|
746 | 746 | tmfclnode = tmfclnodes.setdefault(n, clnode) |
|
747 | 747 | if clrevorder[clnode] < clrevorder[tmfclnode]: |
|
748 | 748 | tmfclnodes[n] = clnode |
|
749 | 749 | else: |
|
750 | 750 | f = dir + p |
|
751 | 751 | fclnodes = fnodes.setdefault(f, {}) |
|
752 | 752 | fclnode = fclnodes.setdefault(n, clnode) |
|
753 | 753 | if clrevorder[clnode] < clrevorder[fclnode]: |
|
754 | 754 | fclnodes[n] = clnode |
|
755 | 755 | return clnode |
|
756 | 756 | return lookupmflinknode |
|
757 | 757 | |
|
758 | 758 | size = 0 |
|
759 | 759 | while tmfnodes: |
|
760 | 760 | dir, nodes = tmfnodes.popitem() |
|
761 | 761 | prunednodes = self.prune(dirlog(dir), nodes, commonrevs) |
|
762 | 762 | if not dir or prunednodes: |
|
763 | 763 | for x in self._packmanifests(dir, prunednodes, |
|
764 | 764 | makelookupmflinknode(dir, nodes)): |
|
765 | 765 | size += len(x) |
|
766 | 766 | yield x |
|
767 | 767 | self._verbosenote(_('%8.i (manifests)\n') % size) |
|
768 | 768 | yield self._manifestsdone() |
|
769 | 769 | |
|
770 | 770 | # The 'source' parameter is useful for extensions |
|
771 | 771 | def generatefiles(self, changedfiles, linknodes, commonrevs, source): |
|
772 | 772 | repo = self._repo |
|
773 | 773 | progress = repo.ui.makeprogress(_('bundling'), unit=_('files'), |
|
774 | 774 | total=len(changedfiles)) |
|
775 | 775 | for i, fname in enumerate(sorted(changedfiles)): |
|
776 | 776 | filerevlog = repo.file(fname) |
|
777 | 777 | if not filerevlog: |
|
778 | 778 | raise error.Abort(_("empty or missing file data for %s") % |
|
779 | 779 | fname) |
|
780 | 780 | |
|
781 | 781 | linkrevnodes = linknodes(filerevlog, fname) |
|
782 | 782 | # Lookup for filenodes, we collected the linkrev nodes above in the |
|
783 | 783 | # fastpath case and with lookupmf in the slowpath case. |
|
784 | 784 | def lookupfilelog(x): |
|
785 | 785 | return linkrevnodes[x] |
|
786 | 786 | |
|
787 | 787 | filenodes = self.prune(filerevlog, linkrevnodes, commonrevs) |
|
788 | 788 | if filenodes: |
|
789 | 789 | progress.update(i + 1, item=fname) |
|
790 | 790 | h = self.fileheader(fname) |
|
791 | 791 | size = len(h) |
|
792 | 792 | yield h |
|
793 | 793 | for chunk in self.group(filenodes, filerevlog, lookupfilelog): |
|
794 | 794 | size += len(chunk) |
|
795 | 795 | yield chunk |
|
796 | 796 | self._verbosenote(_('%8.i %s\n') % (size, fname)) |
|
797 | 797 | progress.complete() |
|
798 | 798 | |
|
799 | 799 | def deltaparent(self, revlog, rev, p1, p2, prev): |
|
800 | 800 | if not revlog.candelta(prev, rev): |
|
801 | 801 | raise error.ProgrammingError('cg1 should not be used in this case') |
|
802 | 802 | return prev |
|
803 | 803 | |
|
804 | 804 | def revchunk(self, revlog, rev, prev, linknode): |
|
805 | if util.safehasattr(self, 'full_nodes'): | |
|
806 | fn = self._revchunknarrow | |
|
807 | else: | |
|
808 | fn = self._revchunknormal | |
|
809 | ||
|
810 | return fn(revlog, rev, prev, linknode) | |
|
811 | ||
|
812 | def _revchunknormal(self, revlog, rev, prev, linknode): | |
|
805 | 813 | node = revlog.node(rev) |
|
806 | 814 | p1, p2 = revlog.parentrevs(rev) |
|
807 | 815 | base = self.deltaparent(revlog, rev, p1, p2, prev) |
|
808 | 816 | |
|
809 | 817 | prefix = '' |
|
810 | 818 | if revlog.iscensored(base) or revlog.iscensored(rev): |
|
811 | 819 | try: |
|
812 | 820 | delta = revlog.revision(node, raw=True) |
|
813 | 821 | except error.CensoredNodeError as e: |
|
814 | 822 | delta = e.tombstone |
|
815 | 823 | if base == nullrev: |
|
816 | 824 | prefix = mdiff.trivialdiffheader(len(delta)) |
|
817 | 825 | else: |
|
818 | 826 | baselen = revlog.rawsize(base) |
|
819 | 827 | prefix = mdiff.replacediffheader(baselen, len(delta)) |
|
820 | 828 | elif base == nullrev: |
|
821 | 829 | delta = revlog.revision(node, raw=True) |
|
822 | 830 | prefix = mdiff.trivialdiffheader(len(delta)) |
|
823 | 831 | else: |
|
824 | 832 | delta = revlog.revdiff(base, rev) |
|
825 | 833 | p1n, p2n = revlog.parents(node) |
|
826 | 834 | basenode = revlog.node(base) |
|
827 | 835 | flags = revlog.flags(rev) |
|
828 | 836 | meta = self.builddeltaheader(node, p1n, p2n, basenode, linknode, flags) |
|
829 | 837 | meta += prefix |
|
830 | 838 | l = len(meta) + len(delta) |
|
831 | 839 | yield chunkheader(l) |
|
832 | 840 | yield meta |
|
833 | 841 | yield delta |
|
842 | ||
|
843 | def _revchunknarrow(self, revlog, rev, prev, linknode): | |
|
844 | # build up some mapping information that's useful later. See | |
|
845 | # the local() nested function below. | |
|
846 | if not self.changelog_done: | |
|
847 | self.clnode_to_rev[linknode] = rev | |
|
848 | linkrev = rev | |
|
849 | self.clrev_to_localrev[linkrev] = rev | |
|
850 | else: | |
|
851 | linkrev = self.clnode_to_rev[linknode] | |
|
852 | self.clrev_to_localrev[linkrev] = rev | |
|
853 | ||
|
854 | # This is a node to send in full, because the changeset it | |
|
855 | # corresponds to was a full changeset. | |
|
856 | if linknode in self.full_nodes: | |
|
857 | for x in self._revchunknormal(revlog, rev, prev, linknode): | |
|
858 | yield x | |
|
859 | return | |
|
860 | ||
|
861 | # At this point, a node can either be one we should skip or an | |
|
862 | # ellipsis. If it's not an ellipsis, bail immediately. | |
|
863 | if linkrev not in self.precomputed_ellipsis: | |
|
864 | return | |
|
865 | ||
|
866 | linkparents = self.precomputed_ellipsis[linkrev] | |
|
867 | def local(clrev): | |
|
868 | """Turn a changelog revnum into a local revnum. | |
|
869 | ||
|
870 | The ellipsis dag is stored as revnums on the changelog, | |
|
871 | but when we're producing ellipsis entries for | |
|
872 | non-changelog revlogs, we need to turn those numbers into | |
|
873 | something local. This does that for us, and during the | |
|
874 | changelog sending phase will also expand the stored | |
|
875 | mappings as needed. | |
|
876 | """ | |
|
877 | if clrev == nullrev: | |
|
878 | return nullrev | |
|
879 | ||
|
880 | if not self.changelog_done: | |
|
881 | # If we're doing the changelog, it's possible that we | |
|
882 | # have a parent that is already on the client, and we | |
|
883 | # need to store some extra mapping information so that | |
|
884 | # our contained ellipsis nodes will be able to resolve | |
|
885 | # their parents. | |
|
886 | if clrev not in self.clrev_to_localrev: | |
|
887 | clnode = revlog.node(clrev) | |
|
888 | self.clnode_to_rev[clnode] = clrev | |
|
889 | return clrev | |
|
890 | ||
|
891 | # Walk the ellipsis-ized changelog breadth-first looking for a | |
|
892 | # change that has been linked from the current revlog. | |
|
893 | # | |
|
894 | # For a flat manifest revlog only a single step should be necessary | |
|
895 | # as all relevant changelog entries are relevant to the flat | |
|
896 | # manifest. | |
|
897 | # | |
|
898 | # For a filelog or tree manifest dirlog however not every changelog | |
|
899 | # entry will have been relevant, so we need to skip some changelog | |
|
900 | # nodes even after ellipsis-izing. | |
|
901 | walk = [clrev] | |
|
902 | while walk: | |
|
903 | p = walk[0] | |
|
904 | walk = walk[1:] | |
|
905 | if p in self.clrev_to_localrev: | |
|
906 | return self.clrev_to_localrev[p] | |
|
907 | elif p in self.full_nodes: | |
|
908 | walk.extend([pp for pp in self._repo.changelog.parentrevs(p) | |
|
909 | if pp != nullrev]) | |
|
910 | elif p in self.precomputed_ellipsis: | |
|
911 | walk.extend([pp for pp in self.precomputed_ellipsis[p] | |
|
912 | if pp != nullrev]) | |
|
913 | else: | |
|
914 | # In this case, we've got an ellipsis with parents | |
|
915 | # outside the current bundle (likely an | |
|
916 | # incremental pull). We "know" that we can use the | |
|
917 | # value of this same revlog at whatever revision | |
|
918 | # is pointed to by linknode. "Know" is in scare | |
|
919 | # quotes because I haven't done enough examination | |
|
920 | # of edge cases to convince myself this is really | |
|
921 | # a fact - it works for all the (admittedly | |
|
922 | # thorough) cases in our testsuite, but I would be | |
|
923 | # somewhat unsurprised to find a case in the wild | |
|
924 | # where this breaks down a bit. That said, I don't | |
|
925 | # know if it would hurt anything. | |
|
926 | for i in pycompat.xrange(rev, 0, -1): | |
|
927 | if revlog.linkrev(i) == clrev: | |
|
928 | return i | |
|
929 | # We failed to resolve a parent for this node, so | |
|
930 | # we crash the changegroup construction. | |
|
931 | raise error.Abort( | |
|
932 | 'unable to resolve parent while packing %r %r' | |
|
933 | ' for changeset %r' % (revlog.indexfile, rev, clrev)) | |
|
934 | ||
|
935 | return nullrev | |
|
936 | ||
|
937 | if not linkparents or ( | |
|
938 | revlog.parentrevs(rev) == (nullrev, nullrev)): | |
|
939 | p1, p2 = nullrev, nullrev | |
|
940 | elif len(linkparents) == 1: | |
|
941 | p1, = sorted(local(p) for p in linkparents) | |
|
942 | p2 = nullrev | |
|
943 | else: | |
|
944 | p1, p2 = sorted(local(p) for p in linkparents) | |
|
945 | n = revlog.node(rev) | |
|
946 | ||
|
947 | yield ellipsisdata( | |
|
948 | self, rev, revlog, p1, p2, revlog.revision(n), linknode) | |
|
949 | ||
|
834 | 950 | def builddeltaheader(self, node, p1n, p2n, basenode, linknode, flags): |
|
835 | 951 | # do nothing with basenode, it is implicitly the previous one in HG10 |
|
836 | 952 | # do nothing with flags, it is implicitly 0 for cg1 and cg2 |
|
837 | 953 | return struct.pack(self.deltaheader, node, p1n, p2n, linknode) |
|
838 | 954 | |
|
839 | 955 | class cg2packer(cg1packer): |
|
840 | 956 | version = '02' |
|
841 | 957 | deltaheader = _CHANGEGROUPV2_DELTA_HEADER |
|
842 | 958 | |
|
843 | 959 | def __init__(self, repo, filematcher, bundlecaps=None): |
|
844 | 960 | super(cg2packer, self).__init__(repo, filematcher, |
|
845 | 961 | bundlecaps=bundlecaps) |
|
846 | 962 | |
|
847 | 963 | if self._reorder is None: |
|
848 | 964 | # Since generaldelta is directly supported by cg2, reordering |
|
849 | 965 | # generally doesn't help, so we disable it by default (treating |
|
850 | 966 | # bundle.reorder=auto just like bundle.reorder=False). |
|
851 | 967 | self._reorder = False |
|
852 | 968 | |
|
853 | 969 | def deltaparent(self, revlog, rev, p1, p2, prev): |
|
854 | 970 | # Narrow ellipses mode. |
|
855 | 971 | if util.safehasattr(self, 'full_nodes'): |
|
856 | 972 | # TODO: send better deltas when in narrow mode. |
|
857 | 973 | # |
|
858 | 974 | # changegroup.group() loops over revisions to send, |
|
859 | 975 | # including revisions we'll skip. What this means is that |
|
860 | 976 | # `prev` will be a potentially useless delta base for all |
|
861 | 977 | # ellipsis nodes, as the client likely won't have it. In |
|
862 | 978 | # the future we should do bookkeeping about which nodes |
|
863 | 979 | # have been sent to the client, and try to be |
|
864 | 980 | # significantly smarter about delta bases. This is |
|
865 | 981 | # slightly tricky because this same code has to work for |
|
866 | 982 | # all revlogs, and we don't have the linkrev/linknode here. |
|
867 | 983 | return p1 |
|
868 | 984 | |
|
869 | 985 | dp = revlog.deltaparent(rev) |
|
870 | 986 | if dp == nullrev and revlog.storedeltachains: |
|
871 | 987 | # Avoid sending full revisions when delta parent is null. Pick prev |
|
872 | 988 | # in that case. It's tempting to pick p1 in this case, as p1 will |
|
873 | 989 | # be smaller in the common case. However, computing a delta against |
|
874 | 990 | # p1 may require resolving the raw text of p1, which could be |
|
875 | 991 | # expensive. The revlog caches should have prev cached, meaning |
|
876 | 992 | # less CPU for changegroup generation. There is likely room to add |
|
877 | 993 | # a flag and/or config option to control this behavior. |
|
878 | 994 | base = prev |
|
879 | 995 | elif dp == nullrev: |
|
880 | 996 | # revlog is configured to use full snapshot for a reason, |
|
881 | 997 | # stick to full snapshot. |
|
882 | 998 | base = nullrev |
|
883 | 999 | elif dp not in (p1, p2, prev): |
|
884 | 1000 | # Pick prev when we can't be sure remote has the base revision. |
|
885 | 1001 | return prev |
|
886 | 1002 | else: |
|
887 | 1003 | base = dp |
|
888 | 1004 | if base != nullrev and not revlog.candelta(base, rev): |
|
889 | 1005 | base = nullrev |
|
890 | 1006 | return base |
|
891 | 1007 | |
|
892 | 1008 | def builddeltaheader(self, node, p1n, p2n, basenode, linknode, flags): |
|
893 | 1009 | # Do nothing with flags, it is implicitly 0 in cg1 and cg2 |
|
894 | 1010 | return struct.pack(self.deltaheader, node, p1n, p2n, basenode, linknode) |
|
895 | 1011 | |
|
896 | 1012 | class cg3packer(cg2packer): |
|
897 | 1013 | version = '03' |
|
898 | 1014 | deltaheader = _CHANGEGROUPV3_DELTA_HEADER |
|
899 | 1015 | |
|
900 | 1016 | def _packmanifests(self, dir, mfnodes, lookuplinknode): |
|
901 | 1017 | if dir: |
|
902 | 1018 | yield self.fileheader(dir) |
|
903 | 1019 | |
|
904 | 1020 | dirlog = self._repo.manifestlog._revlog.dirlog(dir) |
|
905 | 1021 | for chunk in self.group(mfnodes, dirlog, lookuplinknode, |
|
906 | 1022 | units=_('manifests')): |
|
907 | 1023 | yield chunk |
|
908 | 1024 | |
|
909 | 1025 | def _manifestsdone(self): |
|
910 | 1026 | return self.close() |
|
911 | 1027 | |
|
912 | 1028 | def builddeltaheader(self, node, p1n, p2n, basenode, linknode, flags): |
|
913 | 1029 | return struct.pack( |
|
914 | 1030 | self.deltaheader, node, p1n, p2n, basenode, linknode, flags) |
|
915 | 1031 | |
|
916 | 1032 | _packermap = {'01': (cg1packer, cg1unpacker), |
|
917 | 1033 | # cg2 adds support for exchanging generaldelta |
|
918 | 1034 | '02': (cg2packer, cg2unpacker), |
|
919 | 1035 | # cg3 adds support for exchanging revlog flags and treemanifests |
|
920 | 1036 | '03': (cg3packer, cg3unpacker), |
|
921 | 1037 | } |
|
922 | 1038 | |
|
923 | 1039 | def allsupportedversions(repo): |
|
924 | 1040 | versions = set(_packermap.keys()) |
|
925 | 1041 | if not (repo.ui.configbool('experimental', 'changegroup3') or |
|
926 | 1042 | repo.ui.configbool('experimental', 'treemanifest') or |
|
927 | 1043 | 'treemanifest' in repo.requirements): |
|
928 | 1044 | versions.discard('03') |
|
929 | 1045 | return versions |
|
930 | 1046 | |
|
931 | 1047 | # Changegroup versions that can be applied to the repo |
|
932 | 1048 | def supportedincomingversions(repo): |
|
933 | 1049 | return allsupportedversions(repo) |
|
934 | 1050 | |
|
935 | 1051 | # Changegroup versions that can be created from the repo |
|
936 | 1052 | def supportedoutgoingversions(repo): |
|
937 | 1053 | versions = allsupportedversions(repo) |
|
938 | 1054 | if 'treemanifest' in repo.requirements: |
|
939 | 1055 | # Versions 01 and 02 support only flat manifests and it's just too |
|
940 | 1056 | # expensive to convert between the flat manifest and tree manifest on |
|
941 | 1057 | # the fly. Since tree manifests are hashed differently, all of history |
|
942 | 1058 | # would have to be converted. Instead, we simply don't even pretend to |
|
943 | 1059 | # support versions 01 and 02. |
|
944 | 1060 | versions.discard('01') |
|
945 | 1061 | versions.discard('02') |
|
946 | 1062 | if repository.NARROW_REQUIREMENT in repo.requirements: |
|
947 | 1063 | # Versions 01 and 02 don't support revlog flags, and we need to |
|
948 | 1064 | # support that for stripping and unbundling to work. |
|
949 | 1065 | versions.discard('01') |
|
950 | 1066 | versions.discard('02') |
|
951 | 1067 | if LFS_REQUIREMENT in repo.requirements: |
|
952 | 1068 | # Versions 01 and 02 don't support revlog flags, and we need to |
|
953 | 1069 | # mark LFS entries with REVIDX_EXTSTORED. |
|
954 | 1070 | versions.discard('01') |
|
955 | 1071 | versions.discard('02') |
|
956 | 1072 | |
|
957 | 1073 | return versions |
|
958 | 1074 | |
|
959 | 1075 | def localversion(repo): |
|
960 | 1076 | # Finds the best version to use for bundles that are meant to be used |
|
961 | 1077 | # locally, such as those from strip and shelve, and temporary bundles. |
|
962 | 1078 | return max(supportedoutgoingversions(repo)) |
|
963 | 1079 | |
|
964 | 1080 | def safeversion(repo): |
|
965 | 1081 | # Finds the smallest version that it's safe to assume clients of the repo |
|
966 | 1082 | # will support. For example, all hg versions that support generaldelta also |
|
967 | 1083 | # support changegroup 02. |
|
968 | 1084 | versions = supportedoutgoingversions(repo) |
|
969 | 1085 | if 'generaldelta' in repo.requirements: |
|
970 | 1086 | versions.discard('01') |
|
971 | 1087 | assert versions |
|
972 | 1088 | return min(versions) |
|
973 | 1089 | |
|
974 | 1090 | def getbundler(version, repo, bundlecaps=None, filematcher=None): |
|
975 | 1091 | assert version in supportedoutgoingversions(repo) |
|
976 | 1092 | |
|
977 | 1093 | if filematcher is None: |
|
978 | 1094 | filematcher = matchmod.alwaysmatcher(repo.root, '') |
|
979 | 1095 | |
|
980 | 1096 | if version == '01' and not filematcher.always(): |
|
981 | 1097 | raise error.ProgrammingError('version 01 changegroups do not support ' |
|
982 | 1098 | 'sparse file matchers') |
|
983 | 1099 | |
|
984 | 1100 | # Requested files could include files not in the local store. So |
|
985 | 1101 | # filter those out. |
|
986 | 1102 | filematcher = matchmod.intersectmatchers(repo.narrowmatch(), |
|
987 | 1103 | filematcher) |
|
988 | 1104 | |
|
989 | 1105 | return _packermap[version][0](repo, filematcher=filematcher, |
|
990 | 1106 | bundlecaps=bundlecaps) |
|
991 | 1107 | |
|
992 | 1108 | def getunbundler(version, fh, alg, extras=None): |
|
993 | 1109 | return _packermap[version][1](fh, alg, extras=extras) |
|
994 | 1110 | |
|
995 | 1111 | def _changegroupinfo(repo, nodes, source): |
|
996 | 1112 | if repo.ui.verbose or source == 'bundle': |
|
997 | 1113 | repo.ui.status(_("%d changesets found\n") % len(nodes)) |
|
998 | 1114 | if repo.ui.debugflag: |
|
999 | 1115 | repo.ui.debug("list of changesets:\n") |
|
1000 | 1116 | for node in nodes: |
|
1001 | 1117 | repo.ui.debug("%s\n" % hex(node)) |
|
1002 | 1118 | |
|
1003 | 1119 | def makechangegroup(repo, outgoing, version, source, fastpath=False, |
|
1004 | 1120 | bundlecaps=None): |
|
1005 | 1121 | cgstream = makestream(repo, outgoing, version, source, |
|
1006 | 1122 | fastpath=fastpath, bundlecaps=bundlecaps) |
|
1007 | 1123 | return getunbundler(version, util.chunkbuffer(cgstream), None, |
|
1008 | 1124 | {'clcount': len(outgoing.missing) }) |
|
1009 | 1125 | |
|
1010 | 1126 | def makestream(repo, outgoing, version, source, fastpath=False, |
|
1011 | 1127 | bundlecaps=None, filematcher=None): |
|
1012 | 1128 | bundler = getbundler(version, repo, bundlecaps=bundlecaps, |
|
1013 | 1129 | filematcher=filematcher) |
|
1014 | 1130 | |
|
1015 | 1131 | repo = repo.unfiltered() |
|
1016 | 1132 | commonrevs = outgoing.common |
|
1017 | 1133 | csets = outgoing.missing |
|
1018 | 1134 | heads = outgoing.missingheads |
|
1019 | 1135 | # We go through the fast path if we get told to, or if all (unfiltered |
|
1020 | 1136 | # heads have been requested (since we then know there all linkrevs will |
|
1021 | 1137 | # be pulled by the client). |
|
1022 | 1138 | heads.sort() |
|
1023 | 1139 | fastpathlinkrev = fastpath or ( |
|
1024 | 1140 | repo.filtername is None and heads == sorted(repo.heads())) |
|
1025 | 1141 | |
|
1026 | 1142 | repo.hook('preoutgoing', throw=True, source=source) |
|
1027 | 1143 | _changegroupinfo(repo, csets, source) |
|
1028 | 1144 | return bundler.generate(commonrevs, csets, fastpathlinkrev, source) |
|
1029 | 1145 | |
|
1030 | 1146 | def _addchangegroupfiles(repo, source, revmap, trp, expectedfiles, needfiles): |
|
1031 | 1147 | revisions = 0 |
|
1032 | 1148 | files = 0 |
|
1033 | 1149 | progress = repo.ui.makeprogress(_('files'), unit=_('files'), |
|
1034 | 1150 | total=expectedfiles) |
|
1035 | 1151 | for chunkdata in iter(source.filelogheader, {}): |
|
1036 | 1152 | files += 1 |
|
1037 | 1153 | f = chunkdata["filename"] |
|
1038 | 1154 | repo.ui.debug("adding %s revisions\n" % f) |
|
1039 | 1155 | progress.increment() |
|
1040 | 1156 | fl = repo.file(f) |
|
1041 | 1157 | o = len(fl) |
|
1042 | 1158 | try: |
|
1043 | 1159 | deltas = source.deltaiter() |
|
1044 | 1160 | if not fl.addgroup(deltas, revmap, trp): |
|
1045 | 1161 | raise error.Abort(_("received file revlog group is empty")) |
|
1046 | 1162 | except error.CensoredBaseError as e: |
|
1047 | 1163 | raise error.Abort(_("received delta base is censored: %s") % e) |
|
1048 | 1164 | revisions += len(fl) - o |
|
1049 | 1165 | if f in needfiles: |
|
1050 | 1166 | needs = needfiles[f] |
|
1051 | 1167 | for new in pycompat.xrange(o, len(fl)): |
|
1052 | 1168 | n = fl.node(new) |
|
1053 | 1169 | if n in needs: |
|
1054 | 1170 | needs.remove(n) |
|
1055 | 1171 | else: |
|
1056 | 1172 | raise error.Abort( |
|
1057 | 1173 | _("received spurious file revlog entry")) |
|
1058 | 1174 | if not needs: |
|
1059 | 1175 | del needfiles[f] |
|
1060 | 1176 | progress.complete() |
|
1061 | 1177 | |
|
1062 | 1178 | for f, needs in needfiles.iteritems(): |
|
1063 | 1179 | fl = repo.file(f) |
|
1064 | 1180 | for n in needs: |
|
1065 | 1181 | try: |
|
1066 | 1182 | fl.rev(n) |
|
1067 | 1183 | except error.LookupError: |
|
1068 | 1184 | raise error.Abort( |
|
1069 | 1185 | _('missing file data for %s:%s - run hg verify') % |
|
1070 | 1186 | (f, hex(n))) |
|
1071 | 1187 | |
|
1072 | 1188 | return revisions, files |
|
1073 | 1189 | |
|
1074 | 1190 | def _packellipsischangegroup(repo, common, match, relevant_nodes, |
|
1075 | 1191 | ellipsisroots, visitnodes, depth, source, version): |
|
1076 | 1192 | if version in ('01', '02'): |
|
1077 | 1193 | raise error.Abort( |
|
1078 | 1194 | 'ellipsis nodes require at least cg3 on client and server, ' |
|
1079 | 1195 | 'but negotiated version %s' % version) |
|
1080 | 1196 | # We wrap cg1packer.revchunk, using a side channel to pass |
|
1081 | 1197 | # relevant_nodes into that area. Then if linknode isn't in the |
|
1082 | 1198 | # set, we know we have an ellipsis node and we should defer |
|
1083 | 1199 | # sending that node's data. We override close() to detect |
|
1084 | 1200 | # pending ellipsis nodes and flush them. |
|
1085 | 1201 | packer = getbundler(version, repo, filematcher=match) |
|
1086 | 1202 | # Give the packer the list of nodes which should not be |
|
1087 | 1203 | # ellipsis nodes. We store this rather than the set of nodes |
|
1088 | 1204 | # that should be an ellipsis because for very large histories |
|
1089 | 1205 | # we expect this to be significantly smaller. |
|
1090 | 1206 | packer.full_nodes = relevant_nodes |
|
1091 | 1207 | # Maps ellipsis revs to their roots at the changelog level. |
|
1092 | 1208 | packer.precomputed_ellipsis = ellipsisroots |
|
1093 | 1209 | # Maps CL revs to per-revlog revisions. Cleared in close() at |
|
1094 | 1210 | # the end of each group. |
|
1095 | 1211 | packer.clrev_to_localrev = {} |
|
1096 | 1212 | packer.next_clrev_to_localrev = {} |
|
1097 | 1213 | # Maps changelog nodes to changelog revs. Filled in once |
|
1098 | 1214 | # during changelog stage and then left unmodified. |
|
1099 | 1215 | packer.clnode_to_rev = {} |
|
1100 | 1216 | packer.changelog_done = False |
|
1101 | 1217 | # If true, informs the packer that it is serving shallow content and might |
|
1102 | 1218 | # need to pack file contents not introduced by the changes being packed. |
|
1103 | 1219 | packer.is_shallow = depth is not None |
|
1104 | 1220 | |
|
1105 | 1221 | return packer.generate(common, visitnodes, False, source) |
General Comments 0
You need to be logged in to leave comments.
Login now