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