?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
merge: fix out-of-bounds memory access
[git/mjg.git] / builtin-merge-recursive.c
blobb9738655adc66386e55eccafa9ab891bdcf30960
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "cache-tree.h"
8 #include "commit.h"
9 #include "blob.h"
10 #include "builtin.h"
11 #include "tree-walk.h"
12 #include "diff.h"
13 #include "diffcore.h"
14 #include "tag.h"
15 #include "unpack-trees.h"
16 #include "string-list.h"
17 #include "xdiff-interface.h"
18 #include "ll-merge.h"
19 #include "interpolate.h"
20 #include "attr.h"
21 #include "dir.h"
22 #include "merge-recursive.h"
23
24 static int subtree_merge;
25
26 static struct tree *shift_tree_object(struct tree *one, struct tree *two)
27 {
28 unsigned char shifted[20];
29
30 /*
31 * NEEDSWORK: this limits the recursion depth to hardcoded
32 * value '2' to avoid excessive overhead.
33 */
34 shift_tree(one->object.sha1, two->object.sha1, shifted, 2);
35 if (!hashcmp(two->object.sha1, shifted))
36 return two;
37 return lookup_tree(shifted);
38 }
39
40 /*
41 * A virtual commit has
42 * - (const char *)commit->util set to the name, and
43 * - *(int *)commit->object.sha1 set to the virtual id.
44 */
45
46 static struct commit *make_virtual_commit(struct tree *tree, const char *comment)
47 {
48 struct commit *commit = xcalloc(1, sizeof(struct commit));
49 static unsigned virtual_id = 1;
50 commit->tree = tree;
51 commit->util = (void*)comment;
52 *(int*)commit->object.sha1 = virtual_id++;
53 /* avoid warnings */
54 commit->object.parsed = 1;
55 return commit;
56 }
57
58 /*
59 * Since we use get_tree_entry(), which does not put the read object into
60 * the object pool, we cannot rely on a == b.
61 */
62 static int sha_eq(const unsigned char *a, const unsigned char *b)
63 {
64 if (!a && !b)
65 return 2;
66 return a && b && hashcmp(a, b) == 0;
67 }
68
69 /*
70 * Since we want to write the index eventually, we cannot reuse the index
71 * for these (temporary) data.
72 */
73 struct stage_data
74 {
75 struct
76 {
77 unsigned mode;
78 unsigned char sha[20];
79 } stages[4];
80 unsigned processed:1;
81 };
82
83 static struct string_list current_file_set = {NULL, 0, 0, 1};
84 static struct string_list current_directory_set = {NULL, 0, 0, 1};
85
86 static int call_depth = 0;
87 static int verbosity = 2;
88 static int diff_rename_limit = -1;
89 static int merge_rename_limit = -1;
90 static int buffer_output = 1;
91 static struct strbuf obuf = STRBUF_INIT;
92
93 static int show(int v)
94 {
95 return (!call_depth && verbosity >= v) || verbosity >= 5;
96 }
97
98 static void flush_output(void)
99 {
100 if (obuf.len) {
101 fputs(obuf.buf, stdout);
102 strbuf_reset(&obuf);
103 }
104 }
105
106 static void output(int v, const char *fmt, ...)
107 {
108 int len;
109 va_list ap;
110
111 if (!show(v))
112 return;
113
114 strbuf_grow(&obuf, call_depth * 2 + 2);
115 memset(obuf.buf + obuf.len, ' ', call_depth * 2);
116 strbuf_setlen(&obuf, obuf.len + call_depth * 2);
117
118 va_start(ap, fmt);
119 len = vsnprintf(obuf.buf + obuf.len, strbuf_avail(&obuf), fmt, ap);
120 va_end(ap);
121
122 if (len < 0)
123 len = 0;
124 if (len >= strbuf_avail(&obuf)) {
125 strbuf_grow(&obuf, len + 2);
126 va_start(ap, fmt);
127 len = vsnprintf(obuf.buf + obuf.len, strbuf_avail(&obuf), fmt, ap);
128 va_end(ap);
129 if (len >= strbuf_avail(&obuf)) {
130 die("this should not happen, your snprintf is broken");
131 }
132 }
133 strbuf_setlen(&obuf, obuf.len + len);
134 strbuf_add(&obuf, "\n", 1);
135 if (!buffer_output)
136 flush_output();
137 }
138
139 static void output_commit_title(struct commit *commit)
140 {
141 int i;
142 flush_output();
143 for (i = call_depth; i--;)
144 fputs(" ", stdout);
145 if (commit->util)
146 printf("virtual %s\n", (char *)commit->util);
147 else {
148 printf("%s ", find_unique_abbrev(commit->object.sha1, DEFAULT_ABBREV));
149 if (parse_commit(commit) != 0)
150 printf("(bad commit)\n");
151 else {
152 const char *s;
153 int len;
154 for (s = commit->buffer; *s; s++)
155 if (*s == '\n' && s[1] == '\n') {
156 s += 2;
157 break;
158 }
159 for (len = 0; s[len] && '\n' != s[len]; len++)
160 ; /* do nothing */
161 printf("%.*s\n", len, s);
162 }
163 }
164 }
165
166 static int add_cacheinfo(unsigned int mode, const unsigned char *sha1,
167 const char *path, int stage, int refresh, int options)
168 {
169 struct cache_entry *ce;
170 ce = make_cache_entry(mode, sha1 ? sha1 : null_sha1, path, stage, refresh);
171 if (!ce)
172 return error("addinfo_cache failed for path '%s'", path);
173 return add_cache_entry(ce, options);
174 }
175
176 /*
177 * This is a global variable which is used in a number of places but
178 * only written to in the 'merge' function.
179 *
180 * index_only == 1 => Don't leave any non-stage 0 entries in the cache and
181 * don't update the working directory.
182 * 0 => Leave unmerged entries in the cache and update
183 * the working directory.
184 */
185 static int index_only = 0;
186
187 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
188 {
189 parse_tree(tree);
190 init_tree_desc(desc, tree->buffer, tree->size);
191 }
192
193 static int git_merge_trees(int index_only,
194 struct tree *common,
195 struct tree *head,
196 struct tree *merge)
197 {
198 int rc;
199 struct tree_desc t[3];
200 struct unpack_trees_options opts;
201
202 memset(&opts, 0, sizeof(opts));
203 if (index_only)
204 opts.index_only = 1;
205 else
206 opts.update = 1;
207 opts.merge = 1;
208 opts.head_idx = 2;
209 opts.fn = threeway_merge;
210 opts.src_index = &the_index;
211 opts.dst_index = &the_index;
212
213 init_tree_desc_from_tree(t+0, common);
214 init_tree_desc_from_tree(t+1, head);
215 init_tree_desc_from_tree(t+2, merge);
216
217 rc = unpack_trees(3, t, &opts);
218 cache_tree_free(&active_cache_tree);
219 return rc;
220 }
221
222 struct tree *write_tree_from_memory(void)
223 {
224 struct tree *result = NULL;
225
226 if (unmerged_cache()) {
227 int i;
228 output(0, "There are unmerged index entries:");
229 for (i = 0; i < active_nr; i++) {
230 struct cache_entry *ce = active_cache[i];
231 if (ce_stage(ce))
232 output(0, "%d %.*s", ce_stage(ce), ce_namelen(ce), ce->name);
233 }
234 return NULL;
235 }
236
237 if (!active_cache_tree)
238 active_cache_tree = cache_tree();
239
240 if (!cache_tree_fully_valid(active_cache_tree) &&
241 cache_tree_update(active_cache_tree,
242 active_cache, active_nr, 0, 0) < 0)
243 die("error building trees");
244
245 result = lookup_tree(active_cache_tree->sha1);
246
247 return result;
248 }
249
250 static int save_files_dirs(const unsigned char *sha1,
251 const char *base, int baselen, const char *path,
252 unsigned int mode, int stage, void *context)
253 {
254 int len = strlen(path);
255 char *newpath = xmalloc(baselen + len + 1);
256 memcpy(newpath, base, baselen);
257 memcpy(newpath + baselen, path, len);
258 newpath[baselen + len] = '\0';
259
260 if (S_ISDIR(mode))
261 string_list_insert(newpath, &current_directory_set);
262 else
263 string_list_insert(newpath, &current_file_set);
264 free(newpath);
265
266 return READ_TREE_RECURSIVE;
267 }
268
269 static int get_files_dirs(struct tree *tree)
270 {
271 int n;
272 if (read_tree_recursive(tree, "", 0, 0, NULL, save_files_dirs, NULL))
273 return 0;
274 n = current_file_set.nr + current_directory_set.nr;
275 return n;
276 }
277
278 /*
279 * Returns an index_entry instance which doesn't have to correspond to
280 * a real cache entry in Git's index.
281 */
282 static struct stage_data *insert_stage_data(const char *path,
283 struct tree *o, struct tree *a, struct tree *b,
284 struct string_list *entries)
285 {
286 struct string_list_item *item;
287 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
288 get_tree_entry(o->object.sha1, path,
289 e->stages[1].sha, &e->stages[1].mode);
290 get_tree_entry(a->object.sha1, path,
291 e->stages[2].sha, &e->stages[2].mode);
292 get_tree_entry(b->object.sha1, path,
293 e->stages[3].sha, &e->stages[3].mode);
294 item = string_list_insert(path, entries);
295 item->util = e;
296 return e;
297 }
298
299 /*
300 * Create a dictionary mapping file names to stage_data objects. The
301 * dictionary contains one entry for every path with a non-zero stage entry.
302 */
303 static struct string_list *get_unmerged(void)
304 {
305 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
306 int i;
307
308 unmerged->strdup_strings = 1;
309
310 for (i = 0; i < active_nr; i++) {
311 struct string_list_item *item;
312 struct stage_data *e;
313 struct cache_entry *ce = active_cache[i];
314 if (!ce_stage(ce))
315 continue;
316
317 item = string_list_lookup(ce->name, unmerged);
318 if (!item) {
319 item = string_list_insert(ce->name, unmerged);
320 item->util = xcalloc(1, sizeof(struct stage_data));
321 }
322 e = item->util;
323 e->stages[ce_stage(ce)].mode = ce->ce_mode;
324 hashcpy(e->stages[ce_stage(ce)].sha, ce->sha1);
325 }
326
327 return unmerged;
328 }
329
330 struct rename
331 {
332 struct diff_filepair *pair;
333 struct stage_data *src_entry;
334 struct stage_data *dst_entry;
335 unsigned processed:1;
336 };
337
338 /*
339 * Get information of all renames which occurred between 'o_tree' and
340 * 'tree'. We need the three trees in the merge ('o_tree', 'a_tree' and
341 * 'b_tree') to be able to associate the correct cache entries with
342 * the rename information. 'tree' is always equal to either a_tree or b_tree.
343 */
344 static struct string_list *get_renames(struct tree *tree,
345 struct tree *o_tree,
346 struct tree *a_tree,
347 struct tree *b_tree,
348 struct string_list *entries)
349 {
350 int i;
351 struct string_list *renames;
352 struct diff_options opts;
353
354 renames = xcalloc(1, sizeof(struct string_list));
355 diff_setup(&opts);
356 DIFF_OPT_SET(&opts, RECURSIVE);
357 opts.detect_rename = DIFF_DETECT_RENAME;
358 opts.rename_limit = merge_rename_limit >= 0 ? merge_rename_limit :
359 diff_rename_limit >= 0 ? diff_rename_limit :
360 500;
361 opts.warn_on_too_large_rename = 1;
362 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
363 if (diff_setup_done(&opts) < 0)
364 die("diff setup failed");
365 diff_tree_sha1(o_tree->object.sha1, tree->object.sha1, "", &opts);
366 diffcore_std(&opts);
367 for (i = 0; i < diff_queued_diff.nr; ++i) {
368 struct string_list_item *item;
369 struct rename *re;
370 struct diff_filepair *pair = diff_queued_diff.queue[i];
371 if (pair->status != 'R') {
372 diff_free_filepair(pair);
373 continue;
374 }
375 re = xmalloc(sizeof(*re));
376 re->processed = 0;
377 re->pair = pair;
378 item = string_list_lookup(re->pair->one->path, entries);
379 if (!item)
380 re->src_entry = insert_stage_data(re->pair->one->path,
381 o_tree, a_tree, b_tree, entries);
382 else
383 re->src_entry = item->util;
384
385 item = string_list_lookup(re->pair->two->path, entries);
386 if (!item)
387 re->dst_entry = insert_stage_data(re->pair->two->path,
388 o_tree, a_tree, b_tree, entries);
389 else
390 re->dst_entry = item->util;
391 item = string_list_insert(pair->one->path, renames);
392 item->util = re;
393 }
394 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
395 diff_queued_diff.nr = 0;
396 diff_flush(&opts);
397 return renames;
398 }
399
400 static int update_stages(const char *path, struct diff_filespec *o,
401 struct diff_filespec *a, struct diff_filespec *b,
402 int clear)
403 {
404 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE;
405 if (clear)
406 if (remove_file_from_cache(path))
407 return -1;
408 if (o)
409 if (add_cacheinfo(o->mode, o->sha1, path, 1, 0, options))
410 return -1;
411 if (a)
412 if (add_cacheinfo(a->mode, a->sha1, path, 2, 0, options))
413 return -1;
414 if (b)
415 if (add_cacheinfo(b->mode, b->sha1, path, 3, 0, options))
416 return -1;
417 return 0;
418 }
419
420 static int remove_file(int clean, const char *path, int no_wd)
421 {
422 int update_cache = index_only || clean;
423 int update_working_directory = !index_only && !no_wd;
424
425 if (update_cache) {
426 if (remove_file_from_cache(path))
427 return -1;
428 }
429 if (update_working_directory) {
430 if (remove_path(path))
431 return -1;
432 }
433 return 0;
434 }
435
436 static char *unique_path(const char *path, const char *branch)
437 {
438 char *newpath = xmalloc(strlen(path) + 1 + strlen(branch) + 8 + 1);
439 int suffix = 0;
440 struct stat st;
441 char *p = newpath + strlen(path);
442 strcpy(newpath, path);
443 *(p++) = '~';
444 strcpy(p, branch);
445 for (; *p; ++p)
446 if ('/' == *p)
447 *p = '_';
448 while (string_list_has_string(&current_file_set, newpath) ||
449 string_list_has_string(&current_directory_set, newpath) ||
450 lstat(newpath, &st) == 0)
451 sprintf(p, "_%d", suffix++);
452
453 string_list_insert(newpath, &current_file_set);
454 return newpath;
455 }
456
457 static void flush_buffer(int fd, const char *buf, unsigned long size)
458 {
459 while (size > 0) {
460 long ret = write_in_full(fd, buf, size);
461 if (ret < 0) {
462 /* Ignore epipe */
463 if (errno == EPIPE)
464 break;
465 die("merge-recursive: %s", strerror(errno));
466 } else if (!ret) {
467 die("merge-recursive: disk full?");
468 }
469 size -= ret;
470 buf += ret;
471 }
472 }
473
474 static int make_room_for_path(const char *path)
475 {
476 int status;
477 const char *msg = "failed to create path '%s'%s";
478
479 status = safe_create_leading_directories_const(path);
480 if (status) {
481 if (status == -3) {
482 /* something else exists */
483 error(msg, path, ": perhaps a D/F conflict?");
484 return -1;
485 }
486 die(msg, path, "");
487 }
488
489 /* Successful unlink is good.. */
490 if (!unlink(path))
491 return 0;
492 /* .. and so is no existing file */
493 if (errno == ENOENT)
494 return 0;
495 /* .. but not some other error (who really cares what?) */
496 return error(msg, path, ": perhaps a D/F conflict?");
497 }
498
499 static void update_file_flags(const unsigned char *sha,
500 unsigned mode,
501 const char *path,
502 int update_cache,
503 int update_wd)
504 {
505 if (index_only)
506 update_wd = 0;
507
508 if (update_wd) {
509 enum object_type type;
510 void *buf;
511 unsigned long size;
512
513 if (S_ISGITLINK(mode))
514 die("cannot read object %s '%s': It is a submodule!",
515 sha1_to_hex(sha), path);
516
517 buf = read_sha1_file(sha, &type, &size);
518 if (!buf)
519 die("cannot read object %s '%s'", sha1_to_hex(sha), path);
520 if (type != OBJ_BLOB)
521 die("blob expected for %s '%s'", sha1_to_hex(sha), path);
522 if (S_ISREG(mode)) {
523 struct strbuf strbuf;
524 strbuf_init(&strbuf, 0);
525 if (convert_to_working_tree(path, buf, size, &strbuf)) {
526 free(buf);
527 size = strbuf.len;
528 buf = strbuf_detach(&strbuf, NULL);
529 }
530 }
531
532 if (make_room_for_path(path) < 0) {
533 update_wd = 0;
534 free(buf);
535 goto update_index;
536 }
537 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
538 int fd;
539 if (mode & 0100)
540 mode = 0777;
541 else
542 mode = 0666;
543 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
544 if (fd < 0)
545 die("failed to open %s: %s", path, strerror(errno));
546 flush_buffer(fd, buf, size);
547 close(fd);
548 } else if (S_ISLNK(mode)) {
549 char *lnk = xmemdupz(buf, size);
550 safe_create_leading_directories_const(path);
551 unlink(path);
552 symlink(lnk, path);
553 free(lnk);
554 } else
555 die("do not know what to do with %06o %s '%s'",
556 mode, sha1_to_hex(sha), path);
557 free(buf);
558 }
559 update_index:
560 if (update_cache)
561 add_cacheinfo(mode, sha, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
562 }
563
564 static void update_file(int clean,
565 const unsigned char *sha,
566 unsigned mode,
567 const char *path)
568 {
569 update_file_flags(sha, mode, path, index_only || clean, !index_only);
570 }
571
572 /* Low level file merging, update and removal */
573
574 struct merge_file_info
575 {
576 unsigned char sha[20];
577 unsigned mode;
578 unsigned clean:1,
579 merge:1;
580 };
581
582 static void fill_mm(const unsigned char *sha1, mmfile_t *mm)
583 {
584 unsigned long size;
585 enum object_type type;
586
587 if (!hashcmp(sha1, null_sha1)) {
588 mm->ptr = xstrdup("");
589 mm->size = 0;
590 return;
591 }
592
593 mm->ptr = read_sha1_file(sha1, &type, &size);
594 if (!mm->ptr || type != OBJ_BLOB)
595 die("unable to read blob object %s", sha1_to_hex(sha1));
596 mm->size = size;
597 }
598
599 static int merge_3way(mmbuffer_t *result_buf,
600 struct diff_filespec *o,
601 struct diff_filespec *a,
602 struct diff_filespec *b,
603 const char *branch1,
604 const char *branch2)
605 {
606 mmfile_t orig, src1, src2;
607 char *name1, *name2;
608 int merge_status;
609
610 name1 = xstrdup(mkpath("%s:%s", branch1, a->path));
611 name2 = xstrdup(mkpath("%s:%s", branch2, b->path));
612
613 fill_mm(o->sha1, &orig);
614 fill_mm(a->sha1, &src1);
615 fill_mm(b->sha1, &src2);
616
617 merge_status = ll_merge(result_buf, a->path, &orig,
618 &src1, name1, &src2, name2,
619 index_only);
620
621 free(name1);
622 free(name2);
623 free(orig.ptr);
624 free(src1.ptr);
625 free(src2.ptr);
626 return merge_status;
627 }
628
629 static struct merge_file_info merge_file(struct diff_filespec *o,
630 struct diff_filespec *a, struct diff_filespec *b,
631 const char *branch1, const char *branch2)
632 {
633 struct merge_file_info result;
634 result.merge = 0;
635 result.clean = 1;
636
637 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
638 result.clean = 0;
639 if (S_ISREG(a->mode)) {
640 result.mode = a->mode;
641 hashcpy(result.sha, a->sha1);
642 } else {
643 result.mode = b->mode;
644 hashcpy(result.sha, b->sha1);
645 }
646 } else {
647 if (!sha_eq(a->sha1, o->sha1) && !sha_eq(b->sha1, o->sha1))
648 result.merge = 1;
649
650 /*
651 * Merge modes
652 */
653 if (a->mode == b->mode || a->mode == o->mode)
654 result.mode = b->mode;
655 else {
656 result.mode = a->mode;
657 if (b->mode != o->mode) {
658 result.clean = 0;
659 result.merge = 1;
660 }
661 }
662
663 if (sha_eq(a->sha1, b->sha1) || sha_eq(a->sha1, o->sha1))
664 hashcpy(result.sha, b->sha1);
665 else if (sha_eq(b->sha1, o->sha1))
666 hashcpy(result.sha, a->sha1);
667 else if (S_ISREG(a->mode)) {
668 mmbuffer_t result_buf;
669 int merge_status;
670
671 merge_status = merge_3way(&result_buf, o, a, b,
672 branch1, branch2);
673
674 if ((merge_status < 0) || !result_buf.ptr)
675 die("Failed to execute internal merge");
676
677 if (write_sha1_file(result_buf.ptr, result_buf.size,
678 blob_type, result.sha))
679 die("Unable to add %s to database",
680 a->path);
681
682 free(result_buf.ptr);
683 result.clean = (merge_status == 0);
684 } else if (S_ISGITLINK(a->mode)) {
685 result.clean = 0;
686 hashcpy(result.sha, a->sha1);
687 } else if (S_ISLNK(a->mode)) {
688 hashcpy(result.sha, a->sha1);
689
690 if (!sha_eq(a->sha1, b->sha1))
691 result.clean = 0;
692 } else {
693 die("unsupported object type in the tree");
694 }
695 }
696
697 return result;
698 }
699
700 static void conflict_rename_rename(struct rename *ren1,
701 const char *branch1,
702 struct rename *ren2,
703 const char *branch2)
704 {
705 char *del[2];
706 int delp = 0;
707 const char *ren1_dst = ren1->pair->two->path;
708 const char *ren2_dst = ren2->pair->two->path;
709 const char *dst_name1 = ren1_dst;
710 const char *dst_name2 = ren2_dst;
711 if (string_list_has_string(&current_directory_set, ren1_dst)) {
712 dst_name1 = del[delp++] = unique_path(ren1_dst, branch1);
713 output(1, "%s is a directory in %s added as %s instead",
714 ren1_dst, branch2, dst_name1);
715 remove_file(0, ren1_dst, 0);
716 }
717 if (string_list_has_string(&current_directory_set, ren2_dst)) {
718 dst_name2 = del[delp++] = unique_path(ren2_dst, branch2);
719 output(1, "%s is a directory in %s added as %s instead",
720 ren2_dst, branch1, dst_name2);
721 remove_file(0, ren2_dst, 0);
722 }
723 if (index_only) {
724 remove_file_from_cache(dst_name1);
725 remove_file_from_cache(dst_name2);
726 /*
727 * Uncomment to leave the conflicting names in the resulting tree
728 *
729 * update_file(0, ren1->pair->two->sha1, ren1->pair->two->mode, dst_name1);
730 * update_file(0, ren2->pair->two->sha1, ren2->pair->two->mode, dst_name2);
731 */
732 } else {
733 update_stages(dst_name1, NULL, ren1->pair->two, NULL, 1);
734 update_stages(dst_name2, NULL, NULL, ren2->pair->two, 1);
735 }
736 while (delp--)
737 free(del[delp]);
738 }
739
740 static void conflict_rename_dir(struct rename *ren1,
741 const char *branch1)
742 {
743 char *new_path = unique_path(ren1->pair->two->path, branch1);
744 output(1, "Renamed %s to %s instead", ren1->pair->one->path, new_path);
745 remove_file(0, ren1->pair->two->path, 0);
746 update_file(0, ren1->pair->two->sha1, ren1->pair->two->mode, new_path);
747 free(new_path);
748 }
749
750 static void conflict_rename_rename_2(struct rename *ren1,
751 const char *branch1,
752 struct rename *ren2,
753 const char *branch2)
754 {
755 char *new_path1 = unique_path(ren1->pair->two->path, branch1);
756 char *new_path2 = unique_path(ren2->pair->two->path, branch2);
757 output(1, "Renamed %s to %s and %s to %s instead",
758 ren1->pair->one->path, new_path1,
759 ren2->pair->one->path, new_path2);
760 remove_file(0, ren1->pair->two->path, 0);
761 update_file(0, ren1->pair->two->sha1, ren1->pair->two->mode, new_path1);
762 update_file(0, ren2->pair->two->sha1, ren2->pair->two->mode, new_path2);
763 free(new_path2);
764 free(new_path1);
765 }
766
767 static int process_renames(struct string_list *a_renames,
768 struct string_list *b_renames,
769 const char *a_branch,
770 const char *b_branch)
771 {
772 int clean_merge = 1, i, j;
773 struct string_list a_by_dst = {NULL, 0, 0, 0}, b_by_dst = {NULL, 0, 0, 0};
774 const struct rename *sre;
775
776 for (i = 0; i < a_renames->nr; i++) {
777 sre = a_renames->items[i].util;
778 string_list_insert(sre->pair->two->path, &a_by_dst)->util
779 = sre->dst_entry;
780 }
781 for (i = 0; i < b_renames->nr; i++) {
782 sre = b_renames->items[i].util;
783 string_list_insert(sre->pair->two->path, &b_by_dst)->util
784 = sre->dst_entry;
785 }
786
787 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
788 int compare;
789 char *src;
790 struct string_list *renames1, *renames2, *renames2Dst;
791 struct rename *ren1 = NULL, *ren2 = NULL;
792 const char *branch1, *branch2;
793 const char *ren1_src, *ren1_dst;
794
795 if (i >= a_renames->nr) {
796 compare = 1;
797 ren2 = b_renames->items[j++].util;
798 } else if (j >= b_renames->nr) {
799 compare = -1;
800 ren1 = a_renames->items[i++].util;
801 } else {
802 compare = strcmp(a_renames->items[i].string,
803 b_renames->items[j].string);
804 if (compare <= 0)
805 ren1 = a_renames->items[i++].util;
806 if (compare >= 0)
807 ren2 = b_renames->items[j++].util;
808 }
809
810 /* TODO: refactor, so that 1/2 are not needed */
811 if (ren1) {
812 renames1 = a_renames;
813 renames2 = b_renames;
814 renames2Dst = &b_by_dst;
815 branch1 = a_branch;
816 branch2 = b_branch;
817 } else {
818 struct rename *tmp;
819 renames1 = b_renames;
820 renames2 = a_renames;
821 renames2Dst = &a_by_dst;
822 branch1 = b_branch;
823 branch2 = a_branch;
824 tmp = ren2;
825 ren2 = ren1;
826 ren1 = tmp;
827 }
828 src = ren1->pair->one->path;
829
830 ren1->dst_entry->processed = 1;
831 ren1->src_entry->processed = 1;
832
833 if (ren1->processed)
834 continue;
835 ren1->processed = 1;
836
837 ren1_src = ren1->pair->one->path;
838 ren1_dst = ren1->pair->two->path;
839
840 if (ren2) {
841 const char *ren2_src = ren2->pair->one->path;
842 const char *ren2_dst = ren2->pair->two->path;
843 /* Renamed in 1 and renamed in 2 */
844 if (strcmp(ren1_src, ren2_src) != 0)
845 die("ren1.src != ren2.src");
846 ren2->dst_entry->processed = 1;
847 ren2->processed = 1;
848 if (strcmp(ren1_dst, ren2_dst) != 0) {
849 clean_merge = 0;
850 output(1, "CONFLICT (rename/rename): "
851 "Rename \"%s\"->\"%s\" in branch \"%s\" "
852 "rename \"%s\"->\"%s\" in \"%s\"%s",
853 src, ren1_dst, branch1,
854 src, ren2_dst, branch2,
855 index_only ? " (left unresolved)": "");
856 if (index_only) {
857 remove_file_from_cache(src);
858 update_file(0, ren1->pair->one->sha1,
859 ren1->pair->one->mode, src);
860 }
861 conflict_rename_rename(ren1, branch1, ren2, branch2);
862 } else {
863 struct merge_file_info mfi;
864 remove_file(1, ren1_src, 1);
865 mfi = merge_file(ren1->pair->one,
866 ren1->pair->two,
867 ren2->pair->two,
868 branch1,
869 branch2);
870 if (mfi.merge || !mfi.clean)
871 output(1, "Renamed %s->%s", src, ren1_dst);
872
873 if (mfi.merge)
874 output(2, "Auto-merged %s", ren1_dst);
875
876 if (!mfi.clean) {
877 output(1, "CONFLICT (content): merge conflict in %s",
878 ren1_dst);
879 clean_merge = 0;
880
881 if (!index_only)
882 update_stages(ren1_dst,
883 ren1->pair->one,
884 ren1->pair->two,
885 ren2->pair->two,
886 1 /* clear */);
887 }
888 update_file(mfi.clean, mfi.sha, mfi.mode, ren1_dst);
889 }
890 } else {
891 /* Renamed in 1, maybe changed in 2 */
892 struct string_list_item *item;
893 /* we only use sha1 and mode of these */
894 struct diff_filespec src_other, dst_other;
895 int try_merge, stage = a_renames == renames1 ? 3: 2;
896
897 remove_file(1, ren1_src, index_only || stage == 3);
898
899 hashcpy(src_other.sha1, ren1->src_entry->stages[stage].sha);
900 src_other.mode = ren1->src_entry->stages[stage].mode;
901 hashcpy(dst_other.sha1, ren1->dst_entry->stages[stage].sha);
902 dst_other.mode = ren1->dst_entry->stages[stage].mode;
903
904 try_merge = 0;
905
906 if (string_list_has_string(&current_directory_set, ren1_dst)) {
907 clean_merge = 0;
908 output(1, "CONFLICT (rename/directory): Renamed %s->%s in %s "
909 " directory %s added in %s",
910 ren1_src, ren1_dst, branch1,
911 ren1_dst, branch2);
912 conflict_rename_dir(ren1, branch1);
913 } else if (sha_eq(src_other.sha1, null_sha1)) {
914 clean_merge = 0;
915 output(1, "CONFLICT (rename/delete): Renamed %s->%s in %s "
916 "and deleted in %s",
917 ren1_src, ren1_dst, branch1,
918 branch2);
919 update_file(0, ren1->pair->two->sha1, ren1->pair->two->mode, ren1_dst);
920 } else if (!sha_eq(dst_other.sha1, null_sha1)) {
921 const char *new_path;
922 clean_merge = 0;
923 try_merge = 1;
924 output(1, "CONFLICT (rename/add): Renamed %s->%s in %s. "
925 "%s added in %s",
926 ren1_src, ren1_dst, branch1,
927 ren1_dst, branch2);
928 new_path = unique_path(ren1_dst, branch2);
929 output(1, "Added as %s instead", new_path);
930 update_file(0, dst_other.sha1, dst_other.mode, new_path);
931 } else if ((item = string_list_lookup(ren1_dst, renames2Dst))) {
932 ren2 = item->util;
933 clean_merge = 0;
934 ren2->processed = 1;
935 output(1, "CONFLICT (rename/rename): Renamed %s->%s in %s. "
936 "Renamed %s->%s in %s",
937 ren1_src, ren1_dst, branch1,
938 ren2->pair->one->path, ren2->pair->two->path, branch2);
939 conflict_rename_rename_2(ren1, branch1, ren2, branch2);
940 } else
941 try_merge = 1;
942
943 if (try_merge) {
944 struct diff_filespec *o, *a, *b;
945 struct merge_file_info mfi;
946 src_other.path = (char *)ren1_src;
947
948 o = ren1->pair->one;
949 if (a_renames == renames1) {
950 a = ren1->pair->two;
951 b = &src_other;
952 } else {
953 b = ren1->pair->two;
954 a = &src_other;
955 }
956 mfi = merge_file(o, a, b,
957 a_branch, b_branch);
958
959 if (mfi.clean &&
960 sha_eq(mfi.sha, ren1->pair->two->sha1) &&
961 mfi.mode == ren1->pair->two->mode)
962 /*
963 * This messaged is part of
964 * t6022 test. If you change
965 * it update the test too.
966 */
967 output(3, "Skipped %s (merged same as existing)", ren1_dst);
968 else {
969 if (mfi.merge || !mfi.clean)
970 output(1, "Renamed %s => %s", ren1_src, ren1_dst);
971 if (mfi.merge)
972 output(2, "Auto-merged %s", ren1_dst);
973 if (!mfi.clean) {
974 output(1, "CONFLICT (rename/modify): Merge conflict in %s",
975 ren1_dst);
976 clean_merge = 0;
977
978 if (!index_only)
979 update_stages(ren1_dst,
980 o, a, b, 1);
981 }
982 update_file(mfi.clean, mfi.sha, mfi.mode, ren1_dst);
983 }
984 }
985 }
986 }
987 string_list_clear(&a_by_dst, 0);
988 string_list_clear(&b_by_dst, 0);
989
990 return clean_merge;
991 }
992
993 static unsigned char *stage_sha(const unsigned char *sha, unsigned mode)
994 {
995 return (is_null_sha1(sha) || mode == 0) ? NULL: (unsigned char *)sha;
996 }
997
998 /* Per entry merge function */
999 static int process_entry(const char *path, struct stage_data *entry,
1000 const char *branch1,
1001 const char *branch2)
1002 {
1003 /*
1004 printf("processing entry, clean cache: %s\n", index_only ? "yes": "no");
1005 print_index_entry("\tpath: ", entry);
1006 */
1007 int clean_merge = 1;
1008 unsigned o_mode = entry->stages[1].mode;
1009 unsigned a_mode = entry->stages[2].mode;
1010 unsigned b_mode = entry->stages[3].mode;
1011 unsigned char *o_sha = stage_sha(entry->stages[1].sha, o_mode);
1012 unsigned char *a_sha = stage_sha(entry->stages[2].sha, a_mode);
1013 unsigned char *b_sha = stage_sha(entry->stages[3].sha, b_mode);
1014
1015 if (o_sha && (!a_sha || !b_sha)) {
1016 /* Case A: Deleted in one */
1017 if ((!a_sha && !b_sha) ||
1018 (sha_eq(a_sha, o_sha) && !b_sha) ||
1019 (!a_sha && sha_eq(b_sha, o_sha))) {
1020 /* Deleted in both or deleted in one and
1021 * unchanged in the other */
1022 if (a_sha)
1023 output(2, "Removed %s", path);
1024 /* do not touch working file if it did not exist */
1025 remove_file(1, path, !a_sha);
1026 } else {
1027 /* Deleted in one and changed in the other */
1028 clean_merge = 0;
1029 if (!a_sha) {
1030 output(1, "CONFLICT (delete/modify): %s deleted in %s "
1031 "and modified in %s. Version %s of %s left in tree.",
1032 path, branch1,
1033 branch2, branch2, path);
1034 update_file(0, b_sha, b_mode, path);
1035 } else {
1036 output(1, "CONFLICT (delete/modify): %s deleted in %s "
1037 "and modified in %s. Version %s of %s left in tree.",
1038 path, branch2,
1039 branch1, branch1, path);
1040 update_file(0, a_sha, a_mode, path);
1041 }
1042 }
1043
1044 } else if ((!o_sha && a_sha && !b_sha) ||
1045 (!o_sha && !a_sha && b_sha)) {
1046 /* Case B: Added in one. */
1047 const char *add_branch;
1048 const char *other_branch;
1049 unsigned mode;
1050 const unsigned char *sha;
1051 const char *conf;
1052
1053 if (a_sha) {
1054 add_branch = branch1;
1055 other_branch = branch2;
1056 mode = a_mode;
1057 sha = a_sha;
1058 conf = "file/directory";
1059 } else {
1060 add_branch = branch2;
1061 other_branch = branch1;
1062 mode = b_mode;
1063 sha = b_sha;
1064 conf = "directory/file";
1065 }
1066 if (string_list_has_string(&current_directory_set, path)) {
1067 const char *new_path = unique_path(path, add_branch);
1068 clean_merge = 0;
1069 output(1, "CONFLICT (%s): There is a directory with name %s in %s. "
1070 "Added %s as %s",
1071 conf, path, other_branch, path, new_path);
1072 remove_file(0, path, 0);
1073 update_file(0, sha, mode, new_path);
1074 } else {
1075 output(2, "Added %s", path);
1076 update_file(1, sha, mode, path);
1077 }
1078 } else if (a_sha && b_sha) {
1079 /* Case C: Added in both (check for same permissions) and */
1080 /* case D: Modified in both, but differently. */
1081 const char *reason = "content";
1082 struct merge_file_info mfi;
1083 struct diff_filespec o, a, b;
1084
1085 if (!o_sha) {
1086 reason = "add/add";
1087 o_sha = (unsigned char *)null_sha1;
1088 }
1089 output(2, "Auto-merged %s", path);
1090 o.path = a.path = b.path = (char *)path;
1091 hashcpy(o.sha1, o_sha);
1092 o.mode = o_mode;
1093 hashcpy(a.sha1, a_sha);
1094 a.mode = a_mode;
1095 hashcpy(b.sha1, b_sha);
1096 b.mode = b_mode;
1097
1098 mfi = merge_file(&o, &a, &b,
1099 branch1, branch2);
1100
1101 clean_merge = mfi.clean;
1102 if (mfi.clean)
1103 update_file(1, mfi.sha, mfi.mode, path);
1104 else if (S_ISGITLINK(mfi.mode))
1105 output(1, "CONFLICT (submodule): Merge conflict in %s "
1106 "- needs %s", path, sha1_to_hex(b.sha1));
1107 else {
1108 output(1, "CONFLICT (%s): Merge conflict in %s",
1109 reason, path);
1110
1111 if (index_only)
1112 update_file(0, mfi.sha, mfi.mode, path);
1113 else
1114 update_file_flags(mfi.sha, mfi.mode, path,
1115 0 /* update_cache */, 1 /* update_working_directory */);
1116 }
1117 } else if (!o_sha && !a_sha && !b_sha) {
1118 /*
1119 * this entry was deleted altogether. a_mode == 0 means
1120 * we had that path and want to actively remove it.
1121 */
1122 remove_file(1, path, !a_mode);
1123 } else
1124 die("Fatal merge failure, shouldn't happen.");
1125
1126 return clean_merge;
1127 }
1128
1129 int merge_trees(struct tree *head,
1130 struct tree *merge,
1131 struct tree *common,
1132 const char *branch1,
1133 const char *branch2,
1134 struct tree **result)
1135 {
1136 int code, clean;
1137
1138 if (subtree_merge) {
1139 merge = shift_tree_object(head, merge);
1140 common = shift_tree_object(head, common);
1141 }
1142
1143 if (sha_eq(common->object.sha1, merge->object.sha1)) {
1144 output(0, "Already uptodate!");
1145 *result = head;
1146 return 1;
1147 }
1148
1149 code = git_merge_trees(index_only, common, head, merge);
1150
1151 if (code != 0)
1152 die("merging of trees %s and %s failed",
1153 sha1_to_hex(head->object.sha1),
1154 sha1_to_hex(merge->object.sha1));
1155
1156 if (unmerged_cache()) {
1157 struct string_list *entries, *re_head, *re_merge;
1158 int i;
1159 string_list_clear(&current_file_set, 1);
1160 string_list_clear(&current_directory_set, 1);
1161 get_files_dirs(head);
1162 get_files_dirs(merge);
1163
1164 entries = get_unmerged();
1165 re_head = get_renames(head, common, head, merge, entries);
1166 re_merge = get_renames(merge, common, head, merge, entries);
1167 clean = process_renames(re_head, re_merge,
1168 branch1, branch2);
1169 for (i = 0; i < entries->nr; i++) {
1170 const char *path = entries->items[i].string;
1171 struct stage_data *e = entries->items[i].util;
1172 if (!e->processed
1173 && !process_entry(path, e, branch1, branch2))
1174 clean = 0;
1175 }
1176
1177 string_list_clear(re_merge, 0);
1178 string_list_clear(re_head, 0);
1179 string_list_clear(entries, 1);
1180
1181 }
1182 else
1183 clean = 1;
1184
1185 if (index_only)
1186 *result = write_tree_from_memory();
1187
1188 return clean;
1189 }
1190
1191 static struct commit_list *reverse_commit_list(struct commit_list *list)
1192 {
1193 struct commit_list *next = NULL, *current, *backup;
1194 for (current = list; current; current = backup) {
1195 backup = current->next;
1196 current->next = next;
1197 next = current;
1198 }
1199 return next;
1200 }
1201
1202 /*
1203 * Merge the commits h1 and h2, return the resulting virtual
1204 * commit object and a flag indicating the cleanness of the merge.
1205 */
1206 int merge_recursive(struct commit *h1,
1207 struct commit *h2,
1208 const char *branch1,
1209 const char *branch2,
1210 struct commit_list *ca,
1211 struct commit **result)
1212 {
1213 struct commit_list *iter;
1214 struct commit *merged_common_ancestors;
1215 struct tree *mrtree = mrtree;
1216 int clean;
1217
1218 if (show(4)) {
1219 output(4, "Merging:");
1220 output_commit_title(h1);
1221 output_commit_title(h2);
1222 }
1223
1224 if (!ca) {
1225 ca = get_merge_bases(h1, h2, 1);
1226 ca = reverse_commit_list(ca);
1227 }
1228
1229 if (show(5)) {
1230 output(5, "found %u common ancestor(s):", commit_list_count(ca));
1231 for (iter = ca; iter; iter = iter->next)
1232 output_commit_title(iter->item);
1233 }
1234
1235 merged_common_ancestors = pop_commit(&ca);
1236 if (merged_common_ancestors == NULL) {
1237 /* if there is no common ancestor, make an empty tree */
1238 struct tree *tree = xcalloc(1, sizeof(struct tree));
1239
1240 tree->object.parsed = 1;
1241 tree->object.type = OBJ_TREE;
1242 pretend_sha1_file(NULL, 0, OBJ_TREE, tree->object.sha1);
1243 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
1244 }
1245
1246 for (iter = ca; iter; iter = iter->next) {
1247 call_depth++;
1248 /*
1249 * When the merge fails, the result contains files
1250 * with conflict markers. The cleanness flag is
1251 * ignored, it was never actually used, as result of
1252 * merge_trees has always overwritten it: the committed
1253 * "conflicts" were already resolved.
1254 */
1255 discard_cache();
1256 merge_recursive(merged_common_ancestors, iter->item,
1257 "Temporary merge branch 1",
1258 "Temporary merge branch 2",
1259 NULL,
1260 &merged_common_ancestors);
1261 call_depth--;
1262
1263 if (!merged_common_ancestors)
1264 die("merge returned no commit");
1265 }
1266
1267 discard_cache();
1268 if (!call_depth) {
1269 read_cache();
1270 index_only = 0;
1271 } else
1272 index_only = 1;
1273
1274 clean = merge_trees(h1->tree, h2->tree, merged_common_ancestors->tree,
1275 branch1, branch2, &mrtree);
1276
1277 if (index_only) {
1278 *result = make_virtual_commit(mrtree, "merged tree");
1279 commit_list_insert(h1, &(*result)->parents);
1280 commit_list_insert(h2, &(*result)->parents->next);
1281 }
1282 flush_output();
1283 return clean;
1284 }
1285
1286 static const char *better_branch_name(const char *branch)
1287 {
1288 static char githead_env[8 + 40 + 1];
1289 char *name;
1290
1291 if (strlen(branch) != 40)
1292 return branch;
1293 sprintf(githead_env, "GITHEAD_%s", branch);
1294 name = getenv(githead_env);
1295 return name ? name : branch;
1296 }
1297
1298 static struct commit *get_ref(const char *ref)
1299 {
1300 unsigned char sha1[20];
1301 struct object *object;
1302
1303 if (get_sha1(ref, sha1))
1304 die("Could not resolve ref '%s'", ref);
1305 object = deref_tag(parse_object(sha1), ref, strlen(ref));
1306 if (!object)
1307 return NULL;
1308 if (object->type == OBJ_TREE)
1309 return make_virtual_commit((struct tree*)object,
1310 better_branch_name(ref));
1311 if (object->type != OBJ_COMMIT)
1312 return NULL;
1313 if (parse_commit((struct commit *)object))
1314 die("Could not parse commit '%s'", sha1_to_hex(object->sha1));
1315 return (struct commit *)object;
1316 }
1317
1318 static int merge_config(const char *var, const char *value, void *cb)
1319 {
1320 if (!strcasecmp(var, "merge.verbosity")) {
1321 verbosity = git_config_int(var, value);
1322 return 0;
1323 }
1324 if (!strcasecmp(var, "diff.renamelimit")) {
1325 diff_rename_limit = git_config_int(var, value);
1326 return 0;
1327 }
1328 if (!strcasecmp(var, "merge.renamelimit")) {
1329 merge_rename_limit = git_config_int(var, value);
1330 return 0;
1331 }
1332 return git_default_config(var, value, cb);
1333 }
1334
1335 int cmd_merge_recursive(int argc, const char **argv, const char *prefix)
1336 {
1337 static const char *bases[20];
1338 static unsigned bases_count = 0;
1339 int i, clean;
1340 const char *branch1, *branch2;
1341 struct commit *result, *h1, *h2;
1342 struct commit_list *ca = NULL;
1343 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
1344 int index_fd;
1345
1346 if (argv[0]) {
1347 int namelen = strlen(argv[0]);
1348 if (8 < namelen &&
1349 !strcmp(argv[0] + namelen - 8, "-subtree"))
1350 subtree_merge = 1;
1351 }
1352
1353 git_config(merge_config, NULL);
1354 if (getenv("GIT_MERGE_VERBOSITY"))
1355 verbosity = strtol(getenv("GIT_MERGE_VERBOSITY"), NULL, 10);
1356
1357 if (argc < 4)
1358 die("Usage: %s <base>... -- <head> <remote> ...\n", argv[0]);
1359
1360 for (i = 1; i < argc; ++i) {
1361 if (!strcmp(argv[i], "--"))
1362 break;
1363 if (bases_count < sizeof(bases)/sizeof(*bases))
1364 bases[bases_count++] = argv[i];
1365 }
1366 if (argc - i != 3) /* "--" "<head>" "<remote>" */
1367 die("Not handling anything other than two heads merge.");
1368 if (verbosity >= 5)
1369 buffer_output = 0;
1370
1371 branch1 = argv[++i];
1372 branch2 = argv[++i];
1373
1374 h1 = get_ref(branch1);
1375 h2 = get_ref(branch2);
1376
1377 branch1 = better_branch_name(branch1);
1378 branch2 = better_branch_name(branch2);
1379
1380 if (show(3))
1381 printf("Merging %s with %s\n", branch1, branch2);
1382
1383 index_fd = hold_locked_index(lock, 1);
1384
1385 for (i = 0; i < bases_count; i++) {
1386 struct commit *ancestor = get_ref(bases[i]);
1387 ca = commit_list_insert(ancestor, &ca);
1388 }
1389 clean = merge_recursive(h1, h2, branch1, branch2, ca, &result);
1390
1391 if (active_cache_changed &&
1392 (write_cache(index_fd, active_cache, active_nr) ||
1393 commit_locked_index(lock)))
1394 die ("unable to write %s", get_index_file());
1395
1396 return clean ? 0: 1;
1397 }
Michael J Gruber's git tree