Show More
@@ -1,551 +1,557 b'' | |||
|
1 | 1 | # Copyright (C) 2004, 2005 Canonical Ltd |
|
2 | 2 | # |
|
3 | 3 | # This program is free software; you can redistribute it and/or modify |
|
4 | 4 | # it under the terms of the GNU General Public License as published by |
|
5 | 5 | # the Free Software Foundation; either version 2 of the License, or |
|
6 | 6 | # (at your option) any later version. |
|
7 | 7 | # |
|
8 | 8 | # This program is distributed in the hope that it will be useful, |
|
9 | 9 | # but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
10 | 10 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
|
11 | 11 | # GNU General Public License for more details. |
|
12 | 12 | # |
|
13 | 13 | # You should have received a copy of the GNU General Public License |
|
14 | 14 | # along with this program; if not, see <http://www.gnu.org/licenses/>. |
|
15 | 15 | |
|
16 | 16 | # mbp: "you know that thing where cvs gives you conflict markers?" |
|
17 | 17 | # s: "i hate that." |
|
18 | 18 | |
|
19 | 19 | from __future__ import absolute_import |
|
20 | 20 | |
|
21 | 21 | from .i18n import _ |
|
22 | 22 | from . import ( |
|
23 | 23 | error, |
|
24 | 24 | mdiff, |
|
25 | 25 | pycompat, |
|
26 | 26 | ) |
|
27 | 27 | from .utils import stringutil |
|
28 | 28 | |
|
29 | 29 | |
|
30 | 30 | def intersect(ra, rb): |
|
31 | 31 | """Given two ranges return the range where they intersect or None. |
|
32 | 32 | |
|
33 | 33 | >>> intersect((0, 10), (0, 6)) |
|
34 | 34 | (0, 6) |
|
35 | 35 | >>> intersect((0, 10), (5, 15)) |
|
36 | 36 | (5, 10) |
|
37 | 37 | >>> intersect((0, 10), (10, 15)) |
|
38 | 38 | >>> intersect((0, 9), (10, 15)) |
|
39 | 39 | >>> intersect((0, 9), (7, 15)) |
|
40 | 40 | (7, 9) |
|
41 | 41 | """ |
|
42 | 42 | assert ra[0] <= ra[1] |
|
43 | 43 | assert rb[0] <= rb[1] |
|
44 | 44 | |
|
45 | 45 | sa = max(ra[0], rb[0]) |
|
46 | 46 | sb = min(ra[1], rb[1]) |
|
47 | 47 | if sa < sb: |
|
48 | 48 | return sa, sb |
|
49 | 49 | else: |
|
50 | 50 | return None |
|
51 | 51 | |
|
52 | 52 | |
|
53 | 53 | def compare_range(a, astart, aend, b, bstart, bend): |
|
54 | 54 | """Compare a[astart:aend] == b[bstart:bend], without slicing.""" |
|
55 | 55 | if (aend - astart) != (bend - bstart): |
|
56 | 56 | return False |
|
57 | 57 | for ia, ib in zip( |
|
58 | 58 | pycompat.xrange(astart, aend), pycompat.xrange(bstart, bend) |
|
59 | 59 | ): |
|
60 | 60 | if a[ia] != b[ib]: |
|
61 | 61 | return False |
|
62 | 62 | else: |
|
63 | 63 | return True |
|
64 | 64 | |
|
65 | 65 | |
|
66 | 66 | class Merge3Text(object): |
|
67 | 67 | """3-way merge of texts. |
|
68 | 68 | |
|
69 | 69 | Given strings BASE, OTHER, THIS, tries to produce a combined text |
|
70 | 70 | incorporating the changes from both BASE->OTHER and BASE->THIS.""" |
|
71 | 71 | |
|
72 | 72 | def __init__(self, basetext, atext, btext, base=None, a=None, b=None): |
|
73 | 73 | self.basetext = basetext |
|
74 | 74 | self.atext = atext |
|
75 | 75 | self.btext = btext |
|
76 | 76 | if base is None: |
|
77 | 77 | base = mdiff.splitnewlines(basetext) |
|
78 | 78 | if a is None: |
|
79 | 79 | a = mdiff.splitnewlines(atext) |
|
80 | 80 | if b is None: |
|
81 | 81 | b = mdiff.splitnewlines(btext) |
|
82 | 82 | self.base = base |
|
83 | 83 | self.a = a |
|
84 | 84 | self.b = b |
|
85 | 85 | |
|
86 | 86 | def merge_groups(self): |
|
87 | 87 | """Yield sequence of line groups. Each one is a tuple: |
|
88 | 88 | |
|
89 | 89 | 'unchanged', lines |
|
90 | 90 | Lines unchanged from base |
|
91 | 91 | |
|
92 | 92 | 'a', lines |
|
93 | 93 | Lines taken from a |
|
94 | 94 | |
|
95 | 95 | 'same', lines |
|
96 | 96 | Lines taken from a (and equal to b) |
|
97 | 97 | |
|
98 | 98 | 'b', lines |
|
99 | 99 | Lines taken from b |
|
100 | 100 | |
|
101 | 101 | 'conflict', (base_lines, a_lines, b_lines) |
|
102 | 102 | Lines from base were changed to either a or b and conflict. |
|
103 | 103 | """ |
|
104 | 104 | for t in self.merge_regions(): |
|
105 | 105 | what = t[0] |
|
106 | 106 | if what == b'unchanged': |
|
107 | 107 | yield what, self.base[t[1] : t[2]] |
|
108 | 108 | elif what == b'a' or what == b'same': |
|
109 | 109 | yield what, self.a[t[1] : t[2]] |
|
110 | 110 | elif what == b'b': |
|
111 | 111 | yield what, self.b[t[1] : t[2]] |
|
112 | 112 | elif what == b'conflict': |
|
113 | 113 | yield ( |
|
114 | 114 | what, |
|
115 | 115 | ( |
|
116 | 116 | self.base[t[1] : t[2]], |
|
117 | 117 | self.a[t[3] : t[4]], |
|
118 | 118 | self.b[t[5] : t[6]], |
|
119 | 119 | ), |
|
120 | 120 | ) |
|
121 | 121 | else: |
|
122 | 122 | raise ValueError(what) |
|
123 | 123 | |
|
124 | 124 | def merge_regions(self): |
|
125 | 125 | """Return sequences of matching and conflicting regions. |
|
126 | 126 | |
|
127 | 127 | This returns tuples, where the first value says what kind we |
|
128 | 128 | have: |
|
129 | 129 | |
|
130 | 130 | 'unchanged', start, end |
|
131 | 131 | Take a region of base[start:end] |
|
132 | 132 | |
|
133 | 133 | 'same', astart, aend |
|
134 | 134 | b and a are different from base but give the same result |
|
135 | 135 | |
|
136 | 136 | 'a', start, end |
|
137 | 137 | Non-clashing insertion from a[start:end] |
|
138 | 138 | |
|
139 | 139 | 'conflict', zstart, zend, astart, aend, bstart, bend |
|
140 | 140 | Conflict between a and b, with z as common ancestor |
|
141 | 141 | |
|
142 | 142 | Method is as follows: |
|
143 | 143 | |
|
144 | 144 | The two sequences align only on regions which match the base |
|
145 | 145 | and both descendants. These are found by doing a two-way diff |
|
146 | 146 | of each one against the base, and then finding the |
|
147 | 147 | intersections between those regions. These "sync regions" |
|
148 | 148 | are by definition unchanged in both and easily dealt with. |
|
149 | 149 | |
|
150 | 150 | The regions in between can be in any of three cases: |
|
151 | 151 | conflicted, or changed on only one side. |
|
152 | 152 | """ |
|
153 | 153 | |
|
154 | 154 | # section a[0:ia] has been disposed of, etc |
|
155 | 155 | iz = ia = ib = 0 |
|
156 | 156 | |
|
157 | 157 | for region in self.find_sync_regions(): |
|
158 | 158 | zmatch, zend, amatch, aend, bmatch, bend = region |
|
159 | 159 | # print 'match base [%d:%d]' % (zmatch, zend) |
|
160 | 160 | |
|
161 | 161 | matchlen = zend - zmatch |
|
162 | 162 | assert matchlen >= 0 |
|
163 | 163 | assert matchlen == (aend - amatch) |
|
164 | 164 | assert matchlen == (bend - bmatch) |
|
165 | 165 | |
|
166 | 166 | len_a = amatch - ia |
|
167 | 167 | len_b = bmatch - ib |
|
168 | 168 | len_base = zmatch - iz |
|
169 | 169 | assert len_a >= 0 |
|
170 | 170 | assert len_b >= 0 |
|
171 | 171 | assert len_base >= 0 |
|
172 | 172 | |
|
173 | 173 | # print 'unmatched a=%d, b=%d' % (len_a, len_b) |
|
174 | 174 | |
|
175 | 175 | if len_a or len_b: |
|
176 | 176 | # try to avoid actually slicing the lists |
|
177 | 177 | equal_a = compare_range( |
|
178 | 178 | self.a, ia, amatch, self.base, iz, zmatch |
|
179 | 179 | ) |
|
180 | 180 | equal_b = compare_range( |
|
181 | 181 | self.b, ib, bmatch, self.base, iz, zmatch |
|
182 | 182 | ) |
|
183 | 183 | same = compare_range(self.a, ia, amatch, self.b, ib, bmatch) |
|
184 | 184 | |
|
185 | 185 | if same: |
|
186 | 186 | yield b'same', ia, amatch |
|
187 | 187 | elif equal_a and not equal_b: |
|
188 | 188 | yield b'b', ib, bmatch |
|
189 | 189 | elif equal_b and not equal_a: |
|
190 | 190 | yield b'a', ia, amatch |
|
191 | 191 | elif not equal_a and not equal_b: |
|
192 | 192 | yield b'conflict', iz, zmatch, ia, amatch, ib, bmatch |
|
193 | 193 | else: |
|
194 | 194 | raise AssertionError(b"can't handle a=b=base but unmatched") |
|
195 | 195 | |
|
196 | 196 | ia = amatch |
|
197 | 197 | ib = bmatch |
|
198 | 198 | iz = zmatch |
|
199 | 199 | |
|
200 | 200 | # if the same part of the base was deleted on both sides |
|
201 | 201 | # that's OK, we can just skip it. |
|
202 | 202 | |
|
203 | 203 | if matchlen > 0: |
|
204 | 204 | assert ia == amatch |
|
205 | 205 | assert ib == bmatch |
|
206 | 206 | assert iz == zmatch |
|
207 | 207 | |
|
208 | 208 | yield b'unchanged', zmatch, zend |
|
209 | 209 | iz = zend |
|
210 | 210 | ia = aend |
|
211 | 211 | ib = bend |
|
212 | 212 | |
|
213 | 213 | def find_sync_regions(self): |
|
214 | 214 | """Return a list of sync regions, where both descendants match the base. |
|
215 | 215 | |
|
216 | 216 | Generates a list of (base1, base2, a1, a2, b1, b2). There is |
|
217 | 217 | always a zero-length sync region at the end of all the files. |
|
218 | 218 | """ |
|
219 | 219 | |
|
220 | 220 | ia = ib = 0 |
|
221 | 221 | amatches = mdiff.get_matching_blocks(self.basetext, self.atext) |
|
222 | 222 | bmatches = mdiff.get_matching_blocks(self.basetext, self.btext) |
|
223 | 223 | len_a = len(amatches) |
|
224 | 224 | len_b = len(bmatches) |
|
225 | 225 | |
|
226 | 226 | sl = [] |
|
227 | 227 | |
|
228 | 228 | while ia < len_a and ib < len_b: |
|
229 | 229 | abase, amatch, alen = amatches[ia] |
|
230 | 230 | bbase, bmatch, blen = bmatches[ib] |
|
231 | 231 | |
|
232 | 232 | # there is an unconflicted block at i; how long does it |
|
233 | 233 | # extend? until whichever one ends earlier. |
|
234 | 234 | i = intersect((abase, abase + alen), (bbase, bbase + blen)) |
|
235 | 235 | if i: |
|
236 | 236 | intbase = i[0] |
|
237 | 237 | intend = i[1] |
|
238 | 238 | intlen = intend - intbase |
|
239 | 239 | |
|
240 | 240 | # found a match of base[i[0], i[1]]; this may be less than |
|
241 | 241 | # the region that matches in either one |
|
242 | 242 | assert intlen <= alen |
|
243 | 243 | assert intlen <= blen |
|
244 | 244 | assert abase <= intbase |
|
245 | 245 | assert bbase <= intbase |
|
246 | 246 | |
|
247 | 247 | asub = amatch + (intbase - abase) |
|
248 | 248 | bsub = bmatch + (intbase - bbase) |
|
249 | 249 | aend = asub + intlen |
|
250 | 250 | bend = bsub + intlen |
|
251 | 251 | |
|
252 | 252 | assert self.base[intbase:intend] == self.a[asub:aend], ( |
|
253 | 253 | self.base[intbase:intend], |
|
254 | 254 | self.a[asub:aend], |
|
255 | 255 | ) |
|
256 | 256 | |
|
257 | 257 | assert self.base[intbase:intend] == self.b[bsub:bend] |
|
258 | 258 | |
|
259 | 259 | sl.append((intbase, intend, asub, aend, bsub, bend)) |
|
260 | 260 | |
|
261 | 261 | # advance whichever one ends first in the base text |
|
262 | 262 | if (abase + alen) < (bbase + blen): |
|
263 | 263 | ia += 1 |
|
264 | 264 | else: |
|
265 | 265 | ib += 1 |
|
266 | 266 | |
|
267 | 267 | intbase = len(self.base) |
|
268 | 268 | abase = len(self.a) |
|
269 | 269 | bbase = len(self.b) |
|
270 | 270 | sl.append((intbase, intbase, abase, abase, bbase, bbase)) |
|
271 | 271 | |
|
272 | 272 | return sl |
|
273 | 273 | |
|
274 | 274 | |
|
275 | 275 | def _verifytext(text, path, ui, quiet=False, allow_binary=False): |
|
276 | 276 | """verifies that text is non-binary (unless opts[text] is passed, |
|
277 | 277 | then we just warn)""" |
|
278 | 278 | if stringutil.binary(text): |
|
279 | 279 | msg = _(b"%s looks like a binary file.") % path |
|
280 | 280 | if not quiet: |
|
281 | 281 | ui.warn(_(b'warning: %s\n') % msg) |
|
282 | 282 | if not allow_binary: |
|
283 | 283 | raise error.Abort(msg) |
|
284 | 284 | return text |
|
285 | 285 | |
|
286 | 286 | |
|
287 | 287 | def _format_labels(*inputs): |
|
288 | 288 | pad = max(len(input.label) if input.label else 0 for input in inputs) |
|
289 | 289 | labels = [] |
|
290 | 290 | for input in inputs: |
|
291 | 291 | if input.label: |
|
292 | 292 | if input.label_detail: |
|
293 | 293 | label = ( |
|
294 | 294 | (input.label + b':').ljust(pad + 1) |
|
295 | 295 | + b' ' |
|
296 | 296 | + input.label_detail |
|
297 | 297 | ) |
|
298 | 298 | else: |
|
299 | 299 | label = input.label |
|
300 | 300 | # 8 for the prefix of conflict marker lines (e.g. '<<<<<<< ') |
|
301 | 301 | labels.append(stringutil.ellipsis(label, 80 - 8)) |
|
302 | 302 | else: |
|
303 | 303 | labels.append(None) |
|
304 | 304 | return labels |
|
305 | 305 | |
|
306 | 306 | |
|
307 | 307 | def _detect_newline(m3): |
|
308 | 308 | if len(m3.a) > 0: |
|
309 | 309 | if m3.a[0].endswith(b'\r\n'): |
|
310 | 310 | return b'\r\n' |
|
311 | 311 | elif m3.a[0].endswith(b'\r'): |
|
312 | 312 | return b'\r' |
|
313 | 313 | return b'\n' |
|
314 | 314 | |
|
315 | 315 | |
|
316 | 316 | def _minimize(a_lines, b_lines): |
|
317 | 317 | """Trim conflict regions of lines where A and B sides match. |
|
318 | 318 | |
|
319 | 319 | Lines where both A and B have made the same changes at the beginning |
|
320 | 320 | or the end of each merge region are eliminated from the conflict |
|
321 | 321 | region and are instead considered the same. |
|
322 | 322 | """ |
|
323 | 323 | alen = len(a_lines) |
|
324 | 324 | blen = len(b_lines) |
|
325 | 325 | |
|
326 | 326 | # find matches at the front |
|
327 | 327 | ii = 0 |
|
328 | 328 | while ii < alen and ii < blen and a_lines[ii] == b_lines[ii]: |
|
329 | 329 | ii += 1 |
|
330 | 330 | startmatches = ii |
|
331 | 331 | |
|
332 | 332 | # find matches at the end |
|
333 | 333 | ii = 0 |
|
334 | 334 | while ii < alen and ii < blen and a_lines[-ii - 1] == b_lines[-ii - 1]: |
|
335 | 335 | ii += 1 |
|
336 | 336 | endmatches = ii |
|
337 | 337 | |
|
338 | 338 | lines_before = a_lines[:startmatches] |
|
339 | 339 | new_a_lines = a_lines[startmatches : alen - endmatches] |
|
340 | 340 | new_b_lines = b_lines[startmatches : blen - endmatches] |
|
341 | 341 | lines_after = a_lines[alen - endmatches :] |
|
342 | 342 | return lines_before, new_a_lines, new_b_lines, lines_after |
|
343 | 343 | |
|
344 | 344 | |
|
345 | 345 | def render_minimized( |
|
346 | 346 | m3, |
|
347 | 347 | name_a=None, |
|
348 | 348 | name_b=None, |
|
349 | 349 | start_marker=b'<<<<<<<', |
|
350 | 350 | mid_marker=b'=======', |
|
351 | 351 | end_marker=b'>>>>>>>', |
|
352 | 352 | ): |
|
353 | 353 | """Return merge in cvs-like form.""" |
|
354 | 354 | newline = _detect_newline(m3) |
|
355 | 355 | conflicts = False |
|
356 | 356 | if name_a: |
|
357 | 357 | start_marker = start_marker + b' ' + name_a |
|
358 | 358 | if name_b: |
|
359 | 359 | end_marker = end_marker + b' ' + name_b |
|
360 | 360 | merge_groups = m3.merge_groups() |
|
361 | 361 | lines = [] |
|
362 | 362 | for what, group_lines in merge_groups: |
|
363 | 363 | if what == b'conflict': |
|
364 | 364 | conflicts = True |
|
365 | 365 | base_lines, a_lines, b_lines = group_lines |
|
366 | 366 | minimized = _minimize(a_lines, b_lines) |
|
367 | 367 | lines_before, a_lines, b_lines, lines_after = minimized |
|
368 | 368 | lines.extend(lines_before) |
|
369 | 369 | lines.append(start_marker + newline) |
|
370 | 370 | lines.extend(a_lines) |
|
371 | 371 | lines.append(mid_marker + newline) |
|
372 | 372 | lines.extend(b_lines) |
|
373 | 373 | lines.append(end_marker + newline) |
|
374 | 374 | lines.extend(lines_after) |
|
375 | 375 | else: |
|
376 | 376 | lines.extend(group_lines) |
|
377 | 377 | return lines, conflicts |
|
378 | 378 | |
|
379 | 379 | |
|
380 | 380 | def render_merge3(m3, name_a, name_b, name_base): |
|
381 | 381 | """Render conflicts as 3-way conflict markers.""" |
|
382 | 382 | newline = _detect_newline(m3) |
|
383 | 383 | conflicts = False |
|
384 | 384 | lines = [] |
|
385 | 385 | for what, group_lines in m3.merge_groups(): |
|
386 | 386 | if what == b'conflict': |
|
387 | 387 | base_lines, a_lines, b_lines = group_lines |
|
388 | 388 | conflicts = True |
|
389 | 389 | lines.append(b'<<<<<<< ' + name_a + newline) |
|
390 | 390 | lines.extend(a_lines) |
|
391 | 391 | lines.append(b'||||||| ' + name_base + newline) |
|
392 | 392 | lines.extend(base_lines) |
|
393 | 393 | lines.append(b'=======' + newline) |
|
394 | 394 | lines.extend(b_lines) |
|
395 | 395 | lines.append(b'>>>>>>> ' + name_b + newline) |
|
396 | 396 | else: |
|
397 | 397 | lines.extend(group_lines) |
|
398 | 398 | return lines, conflicts |
|
399 | 399 | |
|
400 | 400 | |
|
401 | 401 | def render_mergediff(m3, name_a, name_b, name_base): |
|
402 | 402 | """Render conflicts as conflict markers with one snapshot and one diff.""" |
|
403 | 403 | newline = _detect_newline(m3) |
|
404 | 404 | lines = [] |
|
405 | 405 | conflicts = False |
|
406 | 406 | for what, group_lines in m3.merge_groups(): |
|
407 | 407 | if what == b'conflict': |
|
408 | 408 | base_lines, a_lines, b_lines = group_lines |
|
409 | 409 | base_text = b''.join(base_lines) |
|
410 | 410 | b_blocks = list( |
|
411 | 411 | mdiff.allblocks( |
|
412 | 412 | base_text, |
|
413 | 413 | b''.join(b_lines), |
|
414 | 414 | lines1=base_lines, |
|
415 | 415 | lines2=b_lines, |
|
416 | 416 | ) |
|
417 | 417 | ) |
|
418 | 418 | a_blocks = list( |
|
419 | 419 | mdiff.allblocks( |
|
420 | 420 | base_text, |
|
421 | 421 | b''.join(a_lines), |
|
422 | 422 | lines1=base_lines, |
|
423 | 423 | lines2=b_lines, |
|
424 | 424 | ) |
|
425 | 425 | ) |
|
426 | 426 | |
|
427 | 427 | def matching_lines(blocks): |
|
428 | 428 | return sum( |
|
429 | 429 | block[1] - block[0] |
|
430 | 430 | for block, kind in blocks |
|
431 | 431 | if kind == b'=' |
|
432 | 432 | ) |
|
433 | 433 | |
|
434 | 434 | def diff_lines(blocks, lines1, lines2): |
|
435 | 435 | for block, kind in blocks: |
|
436 | 436 | if kind == b'=': |
|
437 | 437 | for line in lines1[block[0] : block[1]]: |
|
438 | 438 | yield b' ' + line |
|
439 | 439 | else: |
|
440 | 440 | for line in lines1[block[0] : block[1]]: |
|
441 | 441 | yield b'-' + line |
|
442 | 442 | for line in lines2[block[2] : block[3]]: |
|
443 | 443 | yield b'+' + line |
|
444 | 444 | |
|
445 | 445 | lines.append(b"<<<<<<<" + newline) |
|
446 | 446 | if matching_lines(a_blocks) < matching_lines(b_blocks): |
|
447 | 447 | lines.append(b"======= " + name_a + newline) |
|
448 | 448 | lines.extend(a_lines) |
|
449 | 449 | lines.append(b"------- " + name_base + newline) |
|
450 | 450 | lines.append(b"+++++++ " + name_b + newline) |
|
451 | 451 | lines.extend(diff_lines(b_blocks, base_lines, b_lines)) |
|
452 | 452 | else: |
|
453 | 453 | lines.append(b"------- " + name_base + newline) |
|
454 | 454 | lines.append(b"+++++++ " + name_a + newline) |
|
455 | 455 | lines.extend(diff_lines(a_blocks, base_lines, a_lines)) |
|
456 | 456 | lines.append(b"======= " + name_b + newline) |
|
457 | 457 | lines.extend(b_lines) |
|
458 | 458 | lines.append(b">>>>>>>" + newline) |
|
459 | 459 | conflicts = True |
|
460 | 460 | else: |
|
461 | 461 | lines.extend(group_lines) |
|
462 | 462 | return lines, conflicts |
|
463 | 463 | |
|
464 | 464 | |
|
465 | 465 | def _resolve(m3, sides): |
|
466 | 466 | lines = [] |
|
467 | 467 | for what, group_lines in m3.merge_groups(): |
|
468 | 468 | if what == b'conflict': |
|
469 | 469 | for side in sides: |
|
470 | 470 | lines.extend(group_lines[side]) |
|
471 | 471 | else: |
|
472 | 472 | lines.extend(group_lines) |
|
473 | 473 | return lines |
|
474 | 474 | |
|
475 | 475 | |
|
476 | 476 | class MergeInput(object): |
|
477 | 477 | def __init__(self, fctx, label=None, label_detail=None): |
|
478 | 478 | self.fctx = fctx |
|
479 | 479 | self.label = label |
|
480 | 480 | # If the "detail" part is set, then that is rendered after the label and |
|
481 | 481 | # separated by a ':'. The label is padded to make the ':' aligned among |
|
482 | 482 | # all merge inputs. |
|
483 | 483 | self.label_detail = label_detail |
|
484 | self._text = None | |
|
485 | ||
|
486 | def text(self): | |
|
487 | if self._text is None: | |
|
488 | # Merges were always run in the working copy before, which means | |
|
489 | # they used decoded data, if the user defined any repository | |
|
490 | # filters. | |
|
491 | # | |
|
492 | # Maintain that behavior today for BC, though perhaps in the future | |
|
493 | # it'd be worth considering whether merging encoded data (what the | |
|
494 | # repository usually sees) might be more useful. | |
|
495 | self._text = self.fctx.decodeddata() | |
|
496 | return self._text | |
|
484 | 497 | |
|
485 | 498 | |
|
486 | 499 | def simplemerge( |
|
487 | 500 | ui, |
|
488 | 501 | local, |
|
489 | 502 | base, |
|
490 | 503 | other, |
|
491 | 504 | mode=b'merge', |
|
492 | 505 | quiet=False, |
|
493 | 506 | allow_binary=False, |
|
494 | 507 | print_result=False, |
|
495 | 508 | ): |
|
496 | 509 | """Performs the simplemerge algorithm. |
|
497 | 510 | |
|
498 | 511 | The merged result is written into `localctx`. |
|
499 | 512 | """ |
|
500 | 513 | |
|
501 |
def readctx( |
|
|
502 | # Merges were always run in the working copy before, which means | |
|
503 | # they used decoded data, if the user defined any repository | |
|
504 | # filters. | |
|
505 | # | |
|
506 | # Maintain that behavior today for BC, though perhaps in the future | |
|
507 | # it'd be worth considering whether merging encoded data (what the | |
|
508 | # repository usually sees) might be more useful. | |
|
514 | def readctx(input): | |
|
509 | 515 | return _verifytext( |
|
510 |
|
|
|
511 | ctx.path(), | |
|
516 | input.text(), | |
|
517 | input.fctx.path(), | |
|
512 | 518 | ui, |
|
513 | 519 | quiet=quiet, |
|
514 | 520 | allow_binary=allow_binary, |
|
515 | 521 | ) |
|
516 | 522 | |
|
517 | 523 | try: |
|
518 |
localtext = readctx(local |
|
|
519 |
basetext = readctx(base |
|
|
520 |
othertext = readctx(other |
|
|
524 | localtext = readctx(local) | |
|
525 | basetext = readctx(base) | |
|
526 | othertext = readctx(other) | |
|
521 | 527 | except error.Abort: |
|
522 | 528 | return True |
|
523 | 529 | |
|
524 | 530 | m3 = Merge3Text(basetext, localtext, othertext) |
|
525 | 531 | conflicts = False |
|
526 | 532 | if mode == b'union': |
|
527 | 533 | lines = _resolve(m3, (1, 2)) |
|
528 | 534 | elif mode == b'local': |
|
529 | 535 | lines = _resolve(m3, (1,)) |
|
530 | 536 | elif mode == b'other': |
|
531 | 537 | lines = _resolve(m3, (2,)) |
|
532 | 538 | else: |
|
533 | 539 | if mode == b'mergediff': |
|
534 | 540 | labels = _format_labels(local, other, base) |
|
535 | 541 | lines, conflicts = render_mergediff(m3, *labels) |
|
536 | 542 | elif mode == b'merge3': |
|
537 | 543 | labels = _format_labels(local, other, base) |
|
538 | 544 | lines, conflicts = render_merge3(m3, *labels) |
|
539 | 545 | else: |
|
540 | 546 | labels = _format_labels(local, other) |
|
541 | 547 | lines, conflicts = render_minimized(m3, *labels) |
|
542 | 548 | |
|
543 | 549 | mergedtext = b''.join(lines) |
|
544 | 550 | if print_result: |
|
545 | 551 | ui.fout.write(mergedtext) |
|
546 | 552 | else: |
|
547 | 553 | # local.fctx.flags() already has the merged flags (done in |
|
548 | 554 | # mergestate.resolve()) |
|
549 | 555 | local.fctx.write(mergedtext, local.fctx.flags()) |
|
550 | 556 | |
|
551 | 557 | return conflicts |
General Comments 0
You need to be logged in to leave comments.
Login now