Show More
@@ -1,379 +1,379 b'' | |||
|
1 | 1 | # synthrepo.py - repo synthesis |
|
2 | 2 | # |
|
3 | 3 | # Copyright 2012 Facebook |
|
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 | '''synthesize structurally interesting change history |
|
9 | 9 | |
|
10 | 10 | This extension is useful for creating a repository with properties |
|
11 | 11 | that are statistically similar to an existing repository. During |
|
12 | 12 | analysis, a simple probability table is constructed from the history |
|
13 | 13 | of an existing repository. During synthesis, these properties are |
|
14 | 14 | reconstructed. |
|
15 | 15 | |
|
16 | 16 | Properties that are analyzed and synthesized include the following: |
|
17 | 17 | |
|
18 | 18 | - Lines added or removed when an existing file is modified |
|
19 | 19 | - Number and sizes of files added |
|
20 | 20 | - Number of files removed |
|
21 | 21 | - Line lengths |
|
22 | 22 | - Topological distance to parent changeset(s) |
|
23 | 23 | - Probability of a commit being a merge |
|
24 | 24 | - Probability of a newly added file being added to a new directory |
|
25 | 25 | - Interarrival time, and time zone, of commits |
|
26 | 26 | |
|
27 | 27 | A few obvious properties that are not currently handled realistically: |
|
28 | 28 | |
|
29 | 29 | - Merges are treated as regular commits with two parents, which is not |
|
30 | 30 | realistic |
|
31 | 31 | - Modifications are not treated as operations on hunks of lines, but |
|
32 | 32 | as insertions and deletions of randomly chosen single lines |
|
33 | 33 | - Committer ID (always random) |
|
34 | 34 | - Executability of files |
|
35 | 35 | - Symlinks and binary files are ignored |
|
36 | 36 | ''' |
|
37 | 37 | |
|
38 | import bisect, collections, json, os, random, time | |
|
38 | import bisect, collections, json, os, random, time, sys | |
|
39 | 39 | from mercurial import cmdutil, context, patch, scmutil, url, util, hg |
|
40 | 40 | from mercurial.i18n import _ |
|
41 | 41 | from mercurial.node import nullrev, nullid |
|
42 | 42 | |
|
43 | 43 | testedwith = 'internal' |
|
44 | 44 | |
|
45 | 45 | cmdtable = {} |
|
46 | 46 | command = cmdutil.command(cmdtable) |
|
47 | 47 | |
|
48 | 48 | newfile = set(('new fi', 'rename', 'copy f', 'copy t')) |
|
49 | 49 | |
|
50 | 50 | def zerodict(): |
|
51 | 51 | return collections.defaultdict(lambda: 0) |
|
52 | 52 | |
|
53 | 53 | def roundto(x, k): |
|
54 | 54 | if x > k * 2: |
|
55 | 55 | return int(round(x / float(k)) * k) |
|
56 | 56 | return int(round(x)) |
|
57 | 57 | |
|
58 | 58 | def parsegitdiff(lines): |
|
59 | 59 | filename, mar, lineadd, lineremove = None, None, zerodict(), 0 |
|
60 | 60 | binary = False |
|
61 | 61 | for line in lines: |
|
62 | 62 | start = line[:6] |
|
63 | 63 | if start == 'diff -': |
|
64 | 64 | if filename: |
|
65 | 65 | yield filename, mar, lineadd, lineremove, binary |
|
66 | 66 | mar, lineadd, lineremove, binary = 'm', zerodict(), 0, False |
|
67 | 67 | filename = patch.gitre.match(line).group(1) |
|
68 | 68 | elif start in newfile: |
|
69 | 69 | mar = 'a' |
|
70 | 70 | elif start == 'GIT bi': |
|
71 | 71 | binary = True |
|
72 | 72 | elif start == 'delete': |
|
73 | 73 | mar = 'r' |
|
74 | 74 | elif start: |
|
75 | 75 | s = start[0] |
|
76 | 76 | if s == '-' and not line.startswith('--- '): |
|
77 | 77 | lineremove += 1 |
|
78 | 78 | elif s == '+' and not line.startswith('+++ '): |
|
79 | 79 | lineadd[roundto(len(line) - 1, 5)] += 1 |
|
80 | 80 | if filename: |
|
81 | 81 | yield filename, mar, lineadd, lineremove, binary |
|
82 | 82 | |
|
83 | 83 | @command('analyze', |
|
84 | 84 | [('o', 'output', [], _('write output to given file'), _('FILE')), |
|
85 | 85 | ('r', 'rev', [], _('analyze specified revisions'), _('REV'))], |
|
86 | 86 | _('hg analyze')) |
|
87 | 87 | def analyze(ui, repo, *revs, **opts): |
|
88 | 88 | '''create a simple model of a repository to use for later synthesis |
|
89 | 89 | |
|
90 | 90 | This command examines every changeset in the given range (or all |
|
91 | 91 | of history if none are specified) and creates a simple statistical |
|
92 | 92 | model of the history of the repository. |
|
93 | 93 | |
|
94 | 94 | The model is written out to a JSON file, and can be used by |
|
95 | 95 | :hg:`synthesize` to create or augment a repository with synthetic |
|
96 | 96 | commits that have a structure that is statistically similar to the |
|
97 | 97 | analyzed repository. |
|
98 | 98 | ''' |
|
99 | 99 | |
|
100 | 100 | revs = list(revs) |
|
101 | 101 | revs.extend(opts['rev']) |
|
102 | 102 | if not revs: |
|
103 | 103 | revs = [':'] |
|
104 | 104 | |
|
105 | 105 | output = opts['output'] |
|
106 | 106 | if not output: |
|
107 | 107 | output = os.path.basename(repo.root) + '.json' |
|
108 | 108 | |
|
109 | 109 | if output == '-': |
|
110 | 110 | fp = sys.stdout |
|
111 | 111 | else: |
|
112 | 112 | fp = open(output, 'w') |
|
113 | 113 | |
|
114 | 114 | revs = scmutil.revrange(repo, revs) |
|
115 | 115 | revs.sort() |
|
116 | 116 | |
|
117 | 117 | lineschanged = zerodict() |
|
118 | 118 | children = zerodict() |
|
119 | 119 | p1distance = zerodict() |
|
120 | 120 | p2distance = zerodict() |
|
121 | 121 | linesinfilesadded = zerodict() |
|
122 | 122 | fileschanged = zerodict() |
|
123 | 123 | filesadded = zerodict() |
|
124 | 124 | filesremoved = zerodict() |
|
125 | 125 | linelengths = zerodict() |
|
126 | 126 | interarrival = zerodict() |
|
127 | 127 | parents = zerodict() |
|
128 | 128 | dirsadded = zerodict() |
|
129 | 129 | tzoffset = zerodict() |
|
130 | 130 | |
|
131 | 131 | progress = ui.progress |
|
132 | 132 | _analyzing = _('analyzing') |
|
133 | 133 | _changesets = _('changesets') |
|
134 | 134 | _total = len(revs) |
|
135 | 135 | |
|
136 | 136 | for i, rev in enumerate(revs): |
|
137 | 137 | progress(_analyzing, i, unit=_changesets, total=_total) |
|
138 | 138 | ctx = repo[rev] |
|
139 | 139 | pl = ctx.parents() |
|
140 | 140 | pctx = pl[0] |
|
141 | 141 | prev = pctx.rev() |
|
142 | 142 | children[prev] += 1 |
|
143 | 143 | p1distance[rev - prev] += 1 |
|
144 | 144 | parents[len(pl)] += 1 |
|
145 | 145 | tzoffset[ctx.date()[1]] += 1 |
|
146 | 146 | if len(pl) > 1: |
|
147 | 147 | p2distance[rev - pl[1].rev()] += 1 |
|
148 | 148 | if prev == rev - 1: |
|
149 | 149 | lastctx = pctx |
|
150 | 150 | else: |
|
151 | 151 | lastctx = repo[rev - 1] |
|
152 | 152 | if lastctx.rev() != nullrev: |
|
153 | 153 | interarrival[roundto(ctx.date()[0] - lastctx.date()[0], 300)] += 1 |
|
154 | 154 | diff = sum((d.splitlines() |
|
155 | 155 | for d in ctx.diff(pctx, opts=dict(git=True))), []) |
|
156 | 156 | fileadds, diradds, fileremoves, filechanges = 0, 0, 0, 0 |
|
157 | 157 | for filename, mar, lineadd, lineremove, binary in parsegitdiff(diff): |
|
158 | 158 | if binary: |
|
159 | 159 | continue |
|
160 | 160 | added = sum(lineadd.itervalues(), 0) |
|
161 | 161 | if mar == 'm': |
|
162 | 162 | if added and lineremove: |
|
163 | 163 | lineschanged[roundto(added, 5), roundto(lineremove, 5)] += 1 |
|
164 | 164 | filechanges += 1 |
|
165 | 165 | elif mar == 'a': |
|
166 | 166 | fileadds += 1 |
|
167 | 167 | if '/' in filename: |
|
168 | 168 | filedir = filename.rsplit('/', 1)[0] |
|
169 | 169 | if filedir not in pctx.dirs(): |
|
170 | 170 | diradds += 1 |
|
171 | 171 | linesinfilesadded[roundto(added, 5)] += 1 |
|
172 | 172 | elif mar == 'r': |
|
173 | 173 | fileremoves += 1 |
|
174 | 174 | for length, count in lineadd.iteritems(): |
|
175 | 175 | linelengths[length] += count |
|
176 | 176 | fileschanged[filechanges] += 1 |
|
177 | 177 | filesadded[fileadds] += 1 |
|
178 | 178 | dirsadded[diradds] += 1 |
|
179 | 179 | filesremoved[fileremoves] += 1 |
|
180 | 180 | |
|
181 | 181 | invchildren = zerodict() |
|
182 | 182 | |
|
183 | 183 | for rev, count in children.iteritems(): |
|
184 | 184 | invchildren[count] += 1 |
|
185 | 185 | |
|
186 | 186 | if output != '-': |
|
187 | 187 | ui.status(_('writing output to %s\n') % output) |
|
188 | 188 | |
|
189 | 189 | def pronk(d): |
|
190 | 190 | return sorted(d.iteritems(), key=lambda x: x[1], reverse=True) |
|
191 | 191 | |
|
192 | 192 | json.dump(dict(revs=len(revs), |
|
193 | 193 | lineschanged=pronk(lineschanged), |
|
194 | 194 | children=pronk(invchildren), |
|
195 | 195 | fileschanged=pronk(fileschanged), |
|
196 | 196 | filesadded=pronk(filesadded), |
|
197 | 197 | linesinfilesadded=pronk(linesinfilesadded), |
|
198 | 198 | dirsadded=pronk(dirsadded), |
|
199 | 199 | filesremoved=pronk(filesremoved), |
|
200 | 200 | linelengths=pronk(linelengths), |
|
201 | 201 | parents=pronk(parents), |
|
202 | 202 | p1distance=pronk(p1distance), |
|
203 | 203 | p2distance=pronk(p2distance), |
|
204 | 204 | interarrival=pronk(interarrival), |
|
205 | 205 | tzoffset=pronk(tzoffset), |
|
206 | 206 | ), |
|
207 | 207 | fp) |
|
208 | 208 | fp.close() |
|
209 | 209 | |
|
210 | 210 | @command('synthesize', |
|
211 | 211 | [('c', 'count', 0, _('create given number of commits'), _('COUNT')), |
|
212 | 212 | ('', 'dict', '', _('path to a dictionary of words'), _('FILE'))], |
|
213 | 213 | _('hg synthesize [OPTION].. DESCFILE')) |
|
214 | 214 | def synthesize(ui, repo, descpath, **opts): |
|
215 | 215 | '''synthesize commits based on a model of an existing repository |
|
216 | 216 | |
|
217 | 217 | The model must have been generated by :hg:`analyze`. Commits will |
|
218 | 218 | be generated randomly according to the probabilities described in |
|
219 | 219 | the model. |
|
220 | 220 | |
|
221 | 221 | When synthesizing new content, commit descriptions, and user |
|
222 | 222 | names, words will be chosen randomly from a dictionary that is |
|
223 | 223 | presumed to contain one word per line. Use --dict to specify the |
|
224 | 224 | path to an alternate dictionary to use. |
|
225 | 225 | ''' |
|
226 | 226 | try: |
|
227 | 227 | fp = hg.openpath(ui, descpath) |
|
228 | 228 | except Exception, err: |
|
229 | 229 | raise util.Abort('%s: %s' % (descpath, err[0].strerror)) |
|
230 | 230 | desc = json.load(fp) |
|
231 | 231 | fp.close() |
|
232 | 232 | |
|
233 | 233 | def cdf(l): |
|
234 | 234 | if not l: |
|
235 | 235 | return [], [] |
|
236 | 236 | vals, probs = zip(*sorted(l, key=lambda x: x[1], reverse=True)) |
|
237 | 237 | t = float(sum(probs, 0)) |
|
238 | 238 | s, cdfs = 0, [] |
|
239 | 239 | for v in probs: |
|
240 | 240 | s += v |
|
241 | 241 | cdfs.append(s / t) |
|
242 | 242 | return vals, cdfs |
|
243 | 243 | |
|
244 | 244 | lineschanged = cdf(desc['lineschanged']) |
|
245 | 245 | fileschanged = cdf(desc['fileschanged']) |
|
246 | 246 | filesadded = cdf(desc['filesadded']) |
|
247 | 247 | dirsadded = cdf(desc['dirsadded']) |
|
248 | 248 | filesremoved = cdf(desc['filesremoved']) |
|
249 | 249 | linelengths = cdf(desc['linelengths']) |
|
250 | 250 | parents = cdf(desc['parents']) |
|
251 | 251 | p1distance = cdf(desc['p1distance']) |
|
252 | 252 | p2distance = cdf(desc['p2distance']) |
|
253 | 253 | interarrival = cdf(desc['interarrival']) |
|
254 | 254 | linesinfilesadded = cdf(desc['linesinfilesadded']) |
|
255 | 255 | tzoffset = cdf(desc['tzoffset']) |
|
256 | 256 | |
|
257 | 257 | dictfile = opts.get('dict') or '/usr/share/dict/words' |
|
258 | 258 | try: |
|
259 | 259 | fp = open(dictfile, 'rU') |
|
260 | 260 | except IOError, err: |
|
261 | 261 | raise util.Abort('%s: %s' % (dictfile, err.strerror)) |
|
262 | 262 | words = fp.read().splitlines() |
|
263 | 263 | fp.close() |
|
264 | 264 | |
|
265 | 265 | def pick(cdf): |
|
266 | 266 | return cdf[0][bisect.bisect_left(cdf[1], random.random())] |
|
267 | 267 | |
|
268 | 268 | def makeline(minimum=0): |
|
269 | 269 | total = max(minimum, pick(linelengths)) |
|
270 | 270 | c, l = 0, [] |
|
271 | 271 | while c < total: |
|
272 | 272 | w = random.choice(words) |
|
273 | 273 | c += len(w) + 1 |
|
274 | 274 | l.append(w) |
|
275 | 275 | return ' '.join(l) |
|
276 | 276 | |
|
277 | 277 | wlock = repo.wlock() |
|
278 | 278 | lock = repo.lock() |
|
279 | 279 | |
|
280 | 280 | nevertouch = set(('.hgsub', '.hgignore', '.hgtags')) |
|
281 | 281 | |
|
282 | 282 | progress = ui.progress |
|
283 | 283 | _synthesizing = _('synthesizing') |
|
284 | 284 | _changesets = _('changesets') |
|
285 | 285 | |
|
286 | 286 | count = int(opts['count']) |
|
287 | 287 | heads = set(map(repo.changelog.rev, repo.heads())) |
|
288 | 288 | for i in xrange(count): |
|
289 | 289 | progress(_synthesizing, i, unit=_changesets, total=count) |
|
290 | 290 | |
|
291 | 291 | node = repo.changelog.node |
|
292 | 292 | revs = len(repo) |
|
293 | 293 | |
|
294 | 294 | def pickhead(heads, distance): |
|
295 | 295 | if heads: |
|
296 | 296 | lheads = sorted(heads) |
|
297 | 297 | rev = revs - min(pick(distance), revs) |
|
298 | 298 | if rev < lheads[-1]: |
|
299 | 299 | rev = lheads[bisect.bisect_left(lheads, rev)] |
|
300 | 300 | else: |
|
301 | 301 | rev = lheads[-1] |
|
302 | 302 | return rev, node(rev) |
|
303 | 303 | return nullrev, nullid |
|
304 | 304 | |
|
305 | 305 | r1 = revs - min(pick(p1distance), revs) |
|
306 | 306 | p1 = node(r1) |
|
307 | 307 | |
|
308 | 308 | # the number of heads will grow without bound if we use a pure |
|
309 | 309 | # model, so artificially constrain their proliferation |
|
310 | 310 | if pick(parents) == 2 or len(heads) > random.randint(1, 20): |
|
311 | 311 | r2, p2 = pickhead(heads.difference([r1]), p2distance) |
|
312 | 312 | else: |
|
313 | 313 | r2, p2 = nullrev, nullid |
|
314 | 314 | |
|
315 | 315 | pl = [p1, p2] |
|
316 | 316 | pctx = repo[r1] |
|
317 | 317 | mf = pctx.manifest() |
|
318 | 318 | mfk = mf.keys() |
|
319 | 319 | changes = {} |
|
320 | 320 | if mfk: |
|
321 | 321 | for __ in xrange(pick(fileschanged)): |
|
322 | 322 | for __ in xrange(10): |
|
323 | 323 | fctx = pctx.filectx(random.choice(mfk)) |
|
324 | 324 | path = fctx.path() |
|
325 | 325 | if not (path in nevertouch or fctx.isbinary() or |
|
326 | 326 | 'l' in fctx.flags()): |
|
327 | 327 | break |
|
328 | 328 | lines = fctx.data().splitlines() |
|
329 | 329 | add, remove = pick(lineschanged) |
|
330 | 330 | for __ in xrange(remove): |
|
331 | 331 | if not lines: |
|
332 | 332 | break |
|
333 | 333 | del lines[random.randrange(0, len(lines))] |
|
334 | 334 | for __ in xrange(add): |
|
335 | 335 | lines.insert(random.randint(0, len(lines)), makeline()) |
|
336 | 336 | path = fctx.path() |
|
337 | 337 | changes[path] = context.memfilectx(path, |
|
338 | 338 | '\n'.join(lines) + '\n') |
|
339 | 339 | for __ in xrange(pick(filesremoved)): |
|
340 | 340 | path = random.choice(mfk) |
|
341 | 341 | for __ in xrange(10): |
|
342 | 342 | path = random.choice(mfk) |
|
343 | 343 | if path not in changes: |
|
344 | 344 | changes[path] = None |
|
345 | 345 | break |
|
346 | 346 | if filesadded: |
|
347 | 347 | dirs = list(pctx.dirs()) |
|
348 | 348 | dirs.append('') |
|
349 | 349 | for __ in xrange(pick(filesadded)): |
|
350 | 350 | path = [random.choice(dirs)] |
|
351 | 351 | if pick(dirsadded): |
|
352 | 352 | path.append(random.choice(words)) |
|
353 | 353 | path.append(random.choice(words)) |
|
354 | 354 | path = '/'.join(filter(None, path)) |
|
355 | 355 | data = '\n'.join(makeline() |
|
356 | 356 | for __ in xrange(pick(linesinfilesadded))) + '\n' |
|
357 | 357 | changes[path] = context.memfilectx(path, data) |
|
358 | 358 | def filectxfn(repo, memctx, path): |
|
359 | 359 | data = changes[path] |
|
360 | 360 | if data is None: |
|
361 | 361 | raise IOError |
|
362 | 362 | return data |
|
363 | 363 | if not changes: |
|
364 | 364 | continue |
|
365 | 365 | if revs: |
|
366 | 366 | date = repo['tip'].date()[0] + pick(interarrival) |
|
367 | 367 | else: |
|
368 | 368 | date = time.time() - (86400 * count) |
|
369 | 369 | user = random.choice(words) + '@' + random.choice(words) |
|
370 | 370 | mc = context.memctx(repo, pl, makeline(minimum=2), |
|
371 | 371 | sorted(changes.iterkeys()), |
|
372 | 372 | filectxfn, user, '%d %d' % (date, pick(tzoffset))) |
|
373 | 373 | newnode = mc.commit() |
|
374 | 374 | heads.add(repo.changelog.rev(newnode)) |
|
375 | 375 | heads.discard(r1) |
|
376 | 376 | heads.discard(r2) |
|
377 | 377 | |
|
378 | 378 | lock.release() |
|
379 | 379 | wlock.release() |
General Comments 0
You need to be logged in to leave comments.
Login now