]> git.ipfire.org Git - thirdparty/git.git/blob - merge-ort.c
Merge branch 'zh/arg-help-format'
[thirdparty/git.git] / merge-ort.c
1 /*
2 * "Ostensibly Recursive's Twin" merge strategy, or "ort" for short. Meant
3 * as a drop-in replacement for the "recursive" merge strategy, allowing one
4 * to replace
5 *
6 * git merge [-s recursive]
7 *
8 * with
9 *
10 * git merge -s ort
11 *
12 * Note: git's parser allows the space between '-s' and its argument to be
13 * missing. (Should I have backronymed "ham", "alsa", "kip", "nap, "alvo",
14 * "cale", "peedy", or "ins" instead of "ort"?)
15 */
16
17 #include "cache.h"
18 #include "merge-ort.h"
19
20 #include "alloc.h"
21 #include "blob.h"
22 #include "cache-tree.h"
23 #include "commit.h"
24 #include "commit-reach.h"
25 #include "diff.h"
26 #include "diffcore.h"
27 #include "dir.h"
28 #include "object-store.h"
29 #include "strmap.h"
30 #include "tree.h"
31 #include "unpack-trees.h"
32 #include "xdiff-interface.h"
33
34 /*
35 * We have many arrays of size 3. Whenever we have such an array, the
36 * indices refer to one of the sides of the three-way merge. This is so
37 * pervasive that the constants 0, 1, and 2 are used in many places in the
38 * code (especially in arithmetic operations to find the other side's index
39 * or to compute a relevant mask), but sometimes these enum names are used
40 * to aid code clarity.
41 *
42 * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side"
43 * referred to there is one of these three sides.
44 */
45 enum merge_side {
46 MERGE_BASE = 0,
47 MERGE_SIDE1 = 1,
48 MERGE_SIDE2 = 2
49 };
50
51 struct merge_options_internal {
52 /*
53 * paths: primary data structure in all of merge ort.
54 *
55 * The keys of paths:
56 * * are full relative paths from the toplevel of the repository
57 * (e.g. "drivers/firmware/raspberrypi.c").
58 * * store all relevant paths in the repo, both directories and
59 * files (e.g. drivers, drivers/firmware would also be included)
60 * * these keys serve to intern all the path strings, which allows
61 * us to do pointer comparison on directory names instead of
62 * strcmp; we just have to be careful to use the interned strings.
63 * (Technically paths_to_free may track some strings that were
64 * removed from froms paths.)
65 *
66 * The values of paths:
67 * * either a pointer to a merged_info, or a conflict_info struct
68 * * merged_info contains all relevant information for a
69 * non-conflicted entry.
70 * * conflict_info contains a merged_info, plus any additional
71 * information about a conflict such as the higher orders stages
72 * involved and the names of the paths those came from (handy
73 * once renames get involved).
74 * * a path may start "conflicted" (i.e. point to a conflict_info)
75 * and then a later step (e.g. three-way content merge) determines
76 * it can be cleanly merged, at which point it'll be marked clean
77 * and the algorithm will ignore any data outside the contained
78 * merged_info for that entry
79 * * If an entry remains conflicted, the merged_info portion of a
80 * conflict_info will later be filled with whatever version of
81 * the file should be placed in the working directory (e.g. an
82 * as-merged-as-possible variation that contains conflict markers).
83 */
84 struct strmap paths;
85
86 /*
87 * conflicted: a subset of keys->values from "paths"
88 *
89 * conflicted is basically an optimization between process_entries()
90 * and record_conflicted_index_entries(); the latter could loop over
91 * ALL the entries in paths AGAIN and look for the ones that are
92 * still conflicted, but since process_entries() has to loop over
93 * all of them, it saves the ones it couldn't resolve in this strmap
94 * so that record_conflicted_index_entries() can iterate just the
95 * relevant entries.
96 */
97 struct strmap conflicted;
98
99 /*
100 * paths_to_free: additional list of strings to free
101 *
102 * If keys are removed from "paths", they are added to paths_to_free
103 * to ensure they are later freed. We avoid free'ing immediately since
104 * other places (e.g. conflict_info.pathnames[]) may still be
105 * referencing these paths.
106 */
107 struct string_list paths_to_free;
108
109 /*
110 * output: special messages and conflict notices for various paths
111 *
112 * This is a map of pathnames (a subset of the keys in "paths" above)
113 * to strbufs. It gathers various warning/conflict/notice messages
114 * for later processing.
115 */
116 struct strmap output;
117
118 /*
119 * current_dir_name: temporary var used in collect_merge_info_callback()
120 *
121 * Used to set merged_info.directory_name; see documentation for that
122 * variable and the requirements placed on that field.
123 */
124 const char *current_dir_name;
125
126 /* call_depth: recursion level counter for merging merge bases */
127 int call_depth;
128 };
129
130 struct version_info {
131 struct object_id oid;
132 unsigned short mode;
133 };
134
135 struct merged_info {
136 /* if is_null, ignore result. otherwise result has oid & mode */
137 struct version_info result;
138 unsigned is_null:1;
139
140 /*
141 * clean: whether the path in question is cleanly merged.
142 *
143 * see conflict_info.merged for more details.
144 */
145 unsigned clean:1;
146
147 /*
148 * basename_offset: offset of basename of path.
149 *
150 * perf optimization to avoid recomputing offset of final '/'
151 * character in pathname (0 if no '/' in pathname).
152 */
153 size_t basename_offset;
154
155 /*
156 * directory_name: containing directory name.
157 *
158 * Note that we assume directory_name is constructed such that
159 * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name,
160 * i.e. string equality is equivalent to pointer equality. For this
161 * to hold, we have to be careful setting directory_name.
162 */
163 const char *directory_name;
164 };
165
166 struct conflict_info {
167 /*
168 * merged: the version of the path that will be written to working tree
169 *
170 * WARNING: It is critical to check merged.clean and ensure it is 0
171 * before reading any conflict_info fields outside of merged.
172 * Allocated merge_info structs will always have clean set to 1.
173 * Allocated conflict_info structs will have merged.clean set to 0
174 * initially. The merged.clean field is how we know if it is safe
175 * to access other parts of conflict_info besides merged; if a
176 * conflict_info's merged.clean is changed to 1, the rest of the
177 * algorithm is not allowed to look at anything outside of the
178 * merged member anymore.
179 */
180 struct merged_info merged;
181
182 /* oids & modes from each of the three trees for this path */
183 struct version_info stages[3];
184
185 /* pathnames for each stage; may differ due to rename detection */
186 const char *pathnames[3];
187
188 /* Whether this path is/was involved in a directory/file conflict */
189 unsigned df_conflict:1;
190
191 /*
192 * Whether this path is/was involved in a non-content conflict other
193 * than a directory/file conflict (e.g. rename/rename, rename/delete,
194 * file location based on possible directory rename).
195 */
196 unsigned path_conflict:1;
197
198 /*
199 * For filemask and dirmask, the ith bit corresponds to whether the
200 * ith entry is a file (filemask) or a directory (dirmask). Thus,
201 * filemask & dirmask is always zero, and filemask | dirmask is at
202 * most 7 but can be less when a path does not appear as either a
203 * file or a directory on at least one side of history.
204 *
205 * Note that these masks are related to enum merge_side, as the ith
206 * entry corresponds to side i.
207 *
208 * These values come from a traverse_trees() call; more info may be
209 * found looking at tree-walk.h's struct traverse_info,
210 * particularly the documentation above the "fn" member (note that
211 * filemask = mask & ~dirmask from that documentation).
212 */
213 unsigned filemask:3;
214 unsigned dirmask:3;
215
216 /*
217 * Optimization to track which stages match, to avoid the need to
218 * recompute it in multiple steps. Either 0 or at least 2 bits are
219 * set; if at least 2 bits are set, their corresponding stages match.
220 */
221 unsigned match_mask:3;
222 };
223
224 /*** Function Grouping: various utility functions ***/
225
226 /*
227 * For the next three macros, see warning for conflict_info.merged.
228 *
229 * In each of the below, mi is a struct merged_info*, and ci was defined
230 * as a struct conflict_info* (but we need to verify ci isn't actually
231 * pointed at a struct merged_info*).
232 *
233 * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise.
234 * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one.
235 * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first.
236 */
237 #define INITIALIZE_CI(ci, mi) do { \
238 (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \
239 } while (0)
240 #define VERIFY_CI(ci) assert(ci && !ci->merged.clean);
241 #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \
242 (ci) = (struct conflict_info *)(mi); \
243 assert((ci) && !(mi)->clean); \
244 } while (0)
245
246 static void free_strmap_strings(struct strmap *map)
247 {
248 struct hashmap_iter iter;
249 struct strmap_entry *entry;
250
251 strmap_for_each_entry(map, &iter, entry) {
252 free((char*)entry->key);
253 }
254 }
255
256 static void clear_or_reinit_internal_opts(struct merge_options_internal *opti,
257 int reinitialize)
258 {
259 void (*strmap_func)(struct strmap *, int) =
260 reinitialize ? strmap_partial_clear : strmap_clear;
261
262 /*
263 * We marked opti->paths with strdup_strings = 0, so that we
264 * wouldn't have to make another copy of the fullpath created by
265 * make_traverse_path from setup_path_info(). But, now that we've
266 * used it and have no other references to these strings, it is time
267 * to deallocate them.
268 */
269 free_strmap_strings(&opti->paths);
270 strmap_func(&opti->paths, 1);
271
272 /*
273 * All keys and values in opti->conflicted are a subset of those in
274 * opti->paths. We don't want to deallocate anything twice, so we
275 * don't free the keys and we pass 0 for free_values.
276 */
277 strmap_func(&opti->conflicted, 0);
278
279 /*
280 * opti->paths_to_free is similar to opti->paths; we created it with
281 * strdup_strings = 0 to avoid making _another_ copy of the fullpath
282 * but now that we've used it and have no other references to these
283 * strings, it is time to deallocate them. We do so by temporarily
284 * setting strdup_strings to 1.
285 */
286 opti->paths_to_free.strdup_strings = 1;
287 string_list_clear(&opti->paths_to_free, 0);
288 opti->paths_to_free.strdup_strings = 0;
289
290 if (!reinitialize) {
291 struct hashmap_iter iter;
292 struct strmap_entry *e;
293
294 /* Release and free each strbuf found in output */
295 strmap_for_each_entry(&opti->output, &iter, e) {
296 struct strbuf *sb = e->value;
297 strbuf_release(sb);
298 /*
299 * While strictly speaking we don't need to free(sb)
300 * here because we could pass free_values=1 when
301 * calling strmap_clear() on opti->output, that would
302 * require strmap_clear to do another
303 * strmap_for_each_entry() loop, so we just free it
304 * while we're iterating anyway.
305 */
306 free(sb);
307 }
308 strmap_clear(&opti->output, 0);
309 }
310 }
311
312 static int err(struct merge_options *opt, const char *err, ...)
313 {
314 va_list params;
315 struct strbuf sb = STRBUF_INIT;
316
317 strbuf_addstr(&sb, "error: ");
318 va_start(params, err);
319 strbuf_vaddf(&sb, err, params);
320 va_end(params);
321
322 error("%s", sb.buf);
323 strbuf_release(&sb);
324
325 return -1;
326 }
327
328 __attribute__((format (printf, 4, 5)))
329 static void path_msg(struct merge_options *opt,
330 const char *path,
331 int omittable_hint, /* skippable under --remerge-diff */
332 const char *fmt, ...)
333 {
334 va_list ap;
335 struct strbuf *sb = strmap_get(&opt->priv->output, path);
336 if (!sb) {
337 sb = xmalloc(sizeof(*sb));
338 strbuf_init(sb, 0);
339 strmap_put(&opt->priv->output, path, sb);
340 }
341
342 va_start(ap, fmt);
343 strbuf_vaddf(sb, fmt, ap);
344 va_end(ap);
345
346 strbuf_addch(sb, '\n');
347 }
348
349 /*** Function Grouping: functions related to collect_merge_info() ***/
350
351 static void setup_path_info(struct merge_options *opt,
352 struct string_list_item *result,
353 const char *current_dir_name,
354 int current_dir_name_len,
355 char *fullpath, /* we'll take over ownership */
356 struct name_entry *names,
357 struct name_entry *merged_version,
358 unsigned is_null, /* boolean */
359 unsigned df_conflict, /* boolean */
360 unsigned filemask,
361 unsigned dirmask,
362 int resolved /* boolean */)
363 {
364 /* result->util is void*, so mi is a convenience typed variable */
365 struct merged_info *mi;
366
367 assert(!is_null || resolved);
368 assert(!df_conflict || !resolved); /* df_conflict implies !resolved */
369 assert(resolved == (merged_version != NULL));
370
371 mi = xcalloc(1, resolved ? sizeof(struct merged_info) :
372 sizeof(struct conflict_info));
373 mi->directory_name = current_dir_name;
374 mi->basename_offset = current_dir_name_len;
375 mi->clean = !!resolved;
376 if (resolved) {
377 mi->result.mode = merged_version->mode;
378 oidcpy(&mi->result.oid, &merged_version->oid);
379 mi->is_null = !!is_null;
380 } else {
381 int i;
382 struct conflict_info *ci;
383
384 ASSIGN_AND_VERIFY_CI(ci, mi);
385 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
386 ci->pathnames[i] = fullpath;
387 ci->stages[i].mode = names[i].mode;
388 oidcpy(&ci->stages[i].oid, &names[i].oid);
389 }
390 ci->filemask = filemask;
391 ci->dirmask = dirmask;
392 ci->df_conflict = !!df_conflict;
393 if (dirmask)
394 /*
395 * Assume is_null for now, but if we have entries
396 * under the directory then when it is complete in
397 * write_completed_directory() it'll update this.
398 * Also, for D/F conflicts, we have to handle the
399 * directory first, then clear this bit and process
400 * the file to see how it is handled -- that occurs
401 * near the top of process_entry().
402 */
403 mi->is_null = 1;
404 }
405 strmap_put(&opt->priv->paths, fullpath, mi);
406 result->string = fullpath;
407 result->util = mi;
408 }
409
410 static int collect_merge_info_callback(int n,
411 unsigned long mask,
412 unsigned long dirmask,
413 struct name_entry *names,
414 struct traverse_info *info)
415 {
416 /*
417 * n is 3. Always.
418 * common ancestor (mbase) has mask 1, and stored in index 0 of names
419 * head of side 1 (side1) has mask 2, and stored in index 1 of names
420 * head of side 2 (side2) has mask 4, and stored in index 2 of names
421 */
422 struct merge_options *opt = info->data;
423 struct merge_options_internal *opti = opt->priv;
424 struct string_list_item pi; /* Path Info */
425 struct conflict_info *ci; /* typed alias to pi.util (which is void*) */
426 struct name_entry *p;
427 size_t len;
428 char *fullpath;
429 const char *dirname = opti->current_dir_name;
430 unsigned filemask = mask & ~dirmask;
431 unsigned match_mask = 0; /* will be updated below */
432 unsigned mbase_null = !(mask & 1);
433 unsigned side1_null = !(mask & 2);
434 unsigned side2_null = !(mask & 4);
435 unsigned side1_matches_mbase = (!side1_null && !mbase_null &&
436 names[0].mode == names[1].mode &&
437 oideq(&names[0].oid, &names[1].oid));
438 unsigned side2_matches_mbase = (!side2_null && !mbase_null &&
439 names[0].mode == names[2].mode &&
440 oideq(&names[0].oid, &names[2].oid));
441 unsigned sides_match = (!side1_null && !side2_null &&
442 names[1].mode == names[2].mode &&
443 oideq(&names[1].oid, &names[2].oid));
444
445 /*
446 * Note: When a path is a file on one side of history and a directory
447 * in another, we have a directory/file conflict. In such cases, if
448 * the conflict doesn't resolve from renames and deletions, then we
449 * always leave directories where they are and move files out of the
450 * way. Thus, while struct conflict_info has a df_conflict field to
451 * track such conflicts, we ignore that field for any directories at
452 * a path and only pay attention to it for files at the given path.
453 * The fact that we leave directories were they are also means that
454 * we do not need to worry about getting additional df_conflict
455 * information propagated from parent directories down to children
456 * (unlike, say traverse_trees_recursive() in unpack-trees.c, which
457 * sets a newinfo.df_conflicts field specifically to propagate it).
458 */
459 unsigned df_conflict = (filemask != 0) && (dirmask != 0);
460
461 /* n = 3 is a fundamental assumption. */
462 if (n != 3)
463 BUG("Called collect_merge_info_callback wrong");
464
465 /*
466 * A bunch of sanity checks verifying that traverse_trees() calls
467 * us the way I expect. Could just remove these at some point,
468 * though maybe they are helpful to future code readers.
469 */
470 assert(mbase_null == is_null_oid(&names[0].oid));
471 assert(side1_null == is_null_oid(&names[1].oid));
472 assert(side2_null == is_null_oid(&names[2].oid));
473 assert(!mbase_null || !side1_null || !side2_null);
474 assert(mask > 0 && mask < 8);
475
476 /* Determine match_mask */
477 if (side1_matches_mbase)
478 match_mask = (side2_matches_mbase ? 7 : 3);
479 else if (side2_matches_mbase)
480 match_mask = 5;
481 else if (sides_match)
482 match_mask = 6;
483
484 /*
485 * Get the name of the relevant filepath, which we'll pass to
486 * setup_path_info() for tracking.
487 */
488 p = names;
489 while (!p->mode)
490 p++;
491 len = traverse_path_len(info, p->pathlen);
492
493 /* +1 in both of the following lines to include the NUL byte */
494 fullpath = xmalloc(len + 1);
495 make_traverse_path(fullpath, len + 1, info, p->path, p->pathlen);
496
497 /*
498 * If mbase, side1, and side2 all match, we can resolve early. Even
499 * if these are trees, there will be no renames or anything
500 * underneath.
501 */
502 if (side1_matches_mbase && side2_matches_mbase) {
503 /* mbase, side1, & side2 all match; use mbase as resolution */
504 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
505 names, names+0, mbase_null, 0,
506 filemask, dirmask, 1);
507 return mask;
508 }
509
510 /*
511 * Record information about the path so we can resolve later in
512 * process_entries.
513 */
514 setup_path_info(opt, &pi, dirname, info->pathlen, fullpath,
515 names, NULL, 0, df_conflict, filemask, dirmask, 0);
516
517 ci = pi.util;
518 VERIFY_CI(ci);
519 ci->match_mask = match_mask;
520
521 /* If dirmask, recurse into subdirectories */
522 if (dirmask) {
523 struct traverse_info newinfo;
524 struct tree_desc t[3];
525 void *buf[3] = {NULL, NULL, NULL};
526 const char *original_dir_name;
527 int i, ret;
528
529 ci->match_mask &= filemask;
530 newinfo = *info;
531 newinfo.prev = info;
532 newinfo.name = p->path;
533 newinfo.namelen = p->pathlen;
534 newinfo.pathlen = st_add3(newinfo.pathlen, p->pathlen, 1);
535 /*
536 * If this directory we are about to recurse into cared about
537 * its parent directory (the current directory) having a D/F
538 * conflict, then we'd propagate the masks in this way:
539 * newinfo.df_conflicts |= (mask & ~dirmask);
540 * But we don't worry about propagating D/F conflicts. (See
541 * comment near setting of local df_conflict variable near
542 * the beginning of this function).
543 */
544
545 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
546 if (i == 1 && side1_matches_mbase)
547 t[1] = t[0];
548 else if (i == 2 && side2_matches_mbase)
549 t[2] = t[0];
550 else if (i == 2 && sides_match)
551 t[2] = t[1];
552 else {
553 const struct object_id *oid = NULL;
554 if (dirmask & 1)
555 oid = &names[i].oid;
556 buf[i] = fill_tree_descriptor(opt->repo,
557 t + i, oid);
558 }
559 dirmask >>= 1;
560 }
561
562 original_dir_name = opti->current_dir_name;
563 opti->current_dir_name = pi.string;
564 ret = traverse_trees(NULL, 3, t, &newinfo);
565 opti->current_dir_name = original_dir_name;
566
567 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++)
568 free(buf[i]);
569
570 if (ret < 0)
571 return -1;
572 }
573
574 return mask;
575 }
576
577 static int collect_merge_info(struct merge_options *opt,
578 struct tree *merge_base,
579 struct tree *side1,
580 struct tree *side2)
581 {
582 int ret;
583 struct tree_desc t[3];
584 struct traverse_info info;
585 const char *toplevel_dir_placeholder = "";
586
587 opt->priv->current_dir_name = toplevel_dir_placeholder;
588 setup_traverse_info(&info, toplevel_dir_placeholder);
589 info.fn = collect_merge_info_callback;
590 info.data = opt;
591 info.show_all_errors = 1;
592
593 parse_tree(merge_base);
594 parse_tree(side1);
595 parse_tree(side2);
596 init_tree_desc(t + 0, merge_base->buffer, merge_base->size);
597 init_tree_desc(t + 1, side1->buffer, side1->size);
598 init_tree_desc(t + 2, side2->buffer, side2->size);
599
600 ret = traverse_trees(NULL, 3, t, &info);
601
602 return ret;
603 }
604
605 /*** Function Grouping: functions related to threeway content merges ***/
606
607 static int handle_content_merge(struct merge_options *opt,
608 const char *path,
609 const struct version_info *o,
610 const struct version_info *a,
611 const struct version_info *b,
612 const char *pathnames[3],
613 const int extra_marker_size,
614 struct version_info *result)
615 {
616 die("Not yet implemented");
617 }
618
619 /*** Function Grouping: functions related to detect_and_process_renames(), ***
620 *** which are split into directory and regular rename detection sections. ***/
621
622 /*** Function Grouping: functions related to directory rename detection ***/
623
624 /*** Function Grouping: functions related to regular rename detection ***/
625
626 static int detect_and_process_renames(struct merge_options *opt,
627 struct tree *merge_base,
628 struct tree *side1,
629 struct tree *side2)
630 {
631 int clean = 1;
632
633 /*
634 * Rename detection works by detecting file similarity. Here we use
635 * a really easy-to-implement scheme: files are similar IFF they have
636 * the same filename. Therefore, by this scheme, there are no renames.
637 *
638 * TODO: Actually implement a real rename detection scheme.
639 */
640 return clean;
641 }
642
643 /*** Function Grouping: functions related to process_entries() ***/
644
645 static int string_list_df_name_compare(const char *one, const char *two)
646 {
647 int onelen = strlen(one);
648 int twolen = strlen(two);
649 /*
650 * Here we only care that entries for D/F conflicts are
651 * adjacent, in particular with the file of the D/F conflict
652 * appearing before files below the corresponding directory.
653 * The order of the rest of the list is irrelevant for us.
654 *
655 * To achieve this, we sort with df_name_compare and provide
656 * the mode S_IFDIR so that D/F conflicts will sort correctly.
657 * We use the mode S_IFDIR for everything else for simplicity,
658 * since in other cases any changes in their order due to
659 * sorting cause no problems for us.
660 */
661 int cmp = df_name_compare(one, onelen, S_IFDIR,
662 two, twolen, S_IFDIR);
663 /*
664 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
665 * that 'foo' comes before 'foo/bar'.
666 */
667 if (cmp)
668 return cmp;
669 return onelen - twolen;
670 }
671
672 struct directory_versions {
673 /*
674 * versions: list of (basename -> version_info)
675 *
676 * The basenames are in reverse lexicographic order of full pathnames,
677 * as processed in process_entries(). This puts all entries within
678 * a directory together, and covers the directory itself after
679 * everything within it, allowing us to write subtrees before needing
680 * to record information for the tree itself.
681 */
682 struct string_list versions;
683
684 /*
685 * offsets: list of (full relative path directories -> integer offsets)
686 *
687 * Since versions contains basenames from files in multiple different
688 * directories, we need to know which entries in versions correspond
689 * to which directories. Values of e.g.
690 * "" 0
691 * src 2
692 * src/moduleA 5
693 * Would mean that entries 0-1 of versions are files in the toplevel
694 * directory, entries 2-4 are files under src/, and the remaining
695 * entries starting at index 5 are files under src/moduleA/.
696 */
697 struct string_list offsets;
698
699 /*
700 * last_directory: directory that previously processed file found in
701 *
702 * last_directory starts NULL, but records the directory in which the
703 * previous file was found within. As soon as
704 * directory(current_file) != last_directory
705 * then we need to start updating accounting in versions & offsets.
706 * Note that last_directory is always the last path in "offsets" (or
707 * NULL if "offsets" is empty) so this exists just for quick access.
708 */
709 const char *last_directory;
710
711 /* last_directory_len: cached computation of strlen(last_directory) */
712 unsigned last_directory_len;
713 };
714
715 static int tree_entry_order(const void *a_, const void *b_)
716 {
717 const struct string_list_item *a = a_;
718 const struct string_list_item *b = b_;
719
720 const struct merged_info *ami = a->util;
721 const struct merged_info *bmi = b->util;
722 return base_name_compare(a->string, strlen(a->string), ami->result.mode,
723 b->string, strlen(b->string), bmi->result.mode);
724 }
725
726 static void write_tree(struct object_id *result_oid,
727 struct string_list *versions,
728 unsigned int offset,
729 size_t hash_size)
730 {
731 size_t maxlen = 0, extra;
732 unsigned int nr = versions->nr - offset;
733 struct strbuf buf = STRBUF_INIT;
734 struct string_list relevant_entries = STRING_LIST_INIT_NODUP;
735 int i;
736
737 /*
738 * We want to sort the last (versions->nr-offset) entries in versions.
739 * Do so by abusing the string_list API a bit: make another string_list
740 * that contains just those entries and then sort them.
741 *
742 * We won't use relevant_entries again and will let it just pop off the
743 * stack, so there won't be allocation worries or anything.
744 */
745 relevant_entries.items = versions->items + offset;
746 relevant_entries.nr = versions->nr - offset;
747 QSORT(relevant_entries.items, relevant_entries.nr, tree_entry_order);
748
749 /* Pre-allocate some space in buf */
750 extra = hash_size + 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */
751 for (i = 0; i < nr; i++) {
752 maxlen += strlen(versions->items[offset+i].string) + extra;
753 }
754 strbuf_grow(&buf, maxlen);
755
756 /* Write each entry out to buf */
757 for (i = 0; i < nr; i++) {
758 struct merged_info *mi = versions->items[offset+i].util;
759 struct version_info *ri = &mi->result;
760 strbuf_addf(&buf, "%o %s%c",
761 ri->mode,
762 versions->items[offset+i].string, '\0');
763 strbuf_add(&buf, ri->oid.hash, hash_size);
764 }
765
766 /* Write this object file out, and record in result_oid */
767 write_object_file(buf.buf, buf.len, tree_type, result_oid);
768 strbuf_release(&buf);
769 }
770
771 static void record_entry_for_tree(struct directory_versions *dir_metadata,
772 const char *path,
773 struct merged_info *mi)
774 {
775 const char *basename;
776
777 if (mi->is_null)
778 /* nothing to record */
779 return;
780
781 basename = path + mi->basename_offset;
782 assert(strchr(basename, '/') == NULL);
783 string_list_append(&dir_metadata->versions,
784 basename)->util = &mi->result;
785 }
786
787 static void write_completed_directory(struct merge_options *opt,
788 const char *new_directory_name,
789 struct directory_versions *info)
790 {
791 const char *prev_dir;
792 struct merged_info *dir_info = NULL;
793 unsigned int offset;
794
795 /*
796 * Some explanation of info->versions and info->offsets...
797 *
798 * process_entries() iterates over all relevant files AND
799 * directories in reverse lexicographic order, and calls this
800 * function. Thus, an example of the paths that process_entries()
801 * could operate on (along with the directories for those paths
802 * being shown) is:
803 *
804 * xtract.c ""
805 * tokens.txt ""
806 * src/moduleB/umm.c src/moduleB
807 * src/moduleB/stuff.h src/moduleB
808 * src/moduleB/baz.c src/moduleB
809 * src/moduleB src
810 * src/moduleA/foo.c src/moduleA
811 * src/moduleA/bar.c src/moduleA
812 * src/moduleA src
813 * src ""
814 * Makefile ""
815 *
816 * info->versions:
817 *
818 * always contains the unprocessed entries and their
819 * version_info information. For example, after the first five
820 * entries above, info->versions would be:
821 *
822 * xtract.c <xtract.c's version_info>
823 * token.txt <token.txt's version_info>
824 * umm.c <src/moduleB/umm.c's version_info>
825 * stuff.h <src/moduleB/stuff.h's version_info>
826 * baz.c <src/moduleB/baz.c's version_info>
827 *
828 * Once a subdirectory is completed we remove the entries in
829 * that subdirectory from info->versions, writing it as a tree
830 * (write_tree()). Thus, as soon as we get to src/moduleB,
831 * info->versions would be updated to
832 *
833 * xtract.c <xtract.c's version_info>
834 * token.txt <token.txt's version_info>
835 * moduleB <src/moduleB's version_info>
836 *
837 * info->offsets:
838 *
839 * helps us track which entries in info->versions correspond to
840 * which directories. When we are N directories deep (e.g. 4
841 * for src/modA/submod/subdir/), we have up to N+1 unprocessed
842 * directories (+1 because of toplevel dir). Corresponding to
843 * the info->versions example above, after processing five entries
844 * info->offsets will be:
845 *
846 * "" 0
847 * src/moduleB 2
848 *
849 * which is used to know that xtract.c & token.txt are from the
850 * toplevel dirctory, while umm.c & stuff.h & baz.c are from the
851 * src/moduleB directory. Again, following the example above,
852 * once we need to process src/moduleB, then info->offsets is
853 * updated to
854 *
855 * "" 0
856 * src 2
857 *
858 * which says that moduleB (and only moduleB so far) is in the
859 * src directory.
860 *
861 * One unique thing to note about info->offsets here is that
862 * "src" was not added to info->offsets until there was a path
863 * (a file OR directory) immediately below src/ that got
864 * processed.
865 *
866 * Since process_entry() just appends new entries to info->versions,
867 * write_completed_directory() only needs to do work if the next path
868 * is in a directory that is different than the last directory found
869 * in info->offsets.
870 */
871
872 /*
873 * If we are working with the same directory as the last entry, there
874 * is no work to do. (See comments above the directory_name member of
875 * struct merged_info for why we can use pointer comparison instead of
876 * strcmp here.)
877 */
878 if (new_directory_name == info->last_directory)
879 return;
880
881 /*
882 * If we are just starting (last_directory is NULL), or last_directory
883 * is a prefix of the current directory, then we can just update
884 * info->offsets to record the offset where we started this directory
885 * and update last_directory to have quick access to it.
886 */
887 if (info->last_directory == NULL ||
888 !strncmp(new_directory_name, info->last_directory,
889 info->last_directory_len)) {
890 uintptr_t offset = info->versions.nr;
891
892 info->last_directory = new_directory_name;
893 info->last_directory_len = strlen(info->last_directory);
894 /*
895 * Record the offset into info->versions where we will
896 * start recording basenames of paths found within
897 * new_directory_name.
898 */
899 string_list_append(&info->offsets,
900 info->last_directory)->util = (void*)offset;
901 return;
902 }
903
904 /*
905 * The next entry that will be processed will be within
906 * new_directory_name. Since at this point we know that
907 * new_directory_name is within a different directory than
908 * info->last_directory, we have all entries for info->last_directory
909 * in info->versions and we need to create a tree object for them.
910 */
911 dir_info = strmap_get(&opt->priv->paths, info->last_directory);
912 assert(dir_info);
913 offset = (uintptr_t)info->offsets.items[info->offsets.nr-1].util;
914 if (offset == info->versions.nr) {
915 /*
916 * Actually, we don't need to create a tree object in this
917 * case. Whenever all files within a directory disappear
918 * during the merge (e.g. unmodified on one side and
919 * deleted on the other, or files were renamed elsewhere),
920 * then we get here and the directory itself needs to be
921 * omitted from its parent tree as well.
922 */
923 dir_info->is_null = 1;
924 } else {
925 /*
926 * Write out the tree to the git object directory, and also
927 * record the mode and oid in dir_info->result.
928 */
929 dir_info->is_null = 0;
930 dir_info->result.mode = S_IFDIR;
931 write_tree(&dir_info->result.oid, &info->versions, offset,
932 opt->repo->hash_algo->rawsz);
933 }
934
935 /*
936 * We've now used several entries from info->versions and one entry
937 * from info->offsets, so we get rid of those values.
938 */
939 info->offsets.nr--;
940 info->versions.nr = offset;
941
942 /*
943 * Now we've taken care of the completed directory, but we need to
944 * prepare things since future entries will be in
945 * new_directory_name. (In particular, process_entry() will be
946 * appending new entries to info->versions.) So, we need to make
947 * sure new_directory_name is the last entry in info->offsets.
948 */
949 prev_dir = info->offsets.nr == 0 ? NULL :
950 info->offsets.items[info->offsets.nr-1].string;
951 if (new_directory_name != prev_dir) {
952 uintptr_t c = info->versions.nr;
953 string_list_append(&info->offsets,
954 new_directory_name)->util = (void*)c;
955 }
956
957 /* And, of course, we need to update last_directory to match. */
958 info->last_directory = new_directory_name;
959 info->last_directory_len = strlen(info->last_directory);
960 }
961
962 /* Per entry merge function */
963 static void process_entry(struct merge_options *opt,
964 const char *path,
965 struct conflict_info *ci,
966 struct directory_versions *dir_metadata)
967 {
968 VERIFY_CI(ci);
969 assert(ci->filemask >= 0 && ci->filemask <= 7);
970 /* ci->match_mask == 7 was handled in collect_merge_info_callback() */
971 assert(ci->match_mask == 0 || ci->match_mask == 3 ||
972 ci->match_mask == 5 || ci->match_mask == 6);
973
974 if (ci->dirmask) {
975 record_entry_for_tree(dir_metadata, path, &ci->merged);
976 if (ci->filemask == 0)
977 /* nothing else to handle */
978 return;
979 assert(ci->df_conflict);
980 }
981
982 if (ci->df_conflict) {
983 die("Not yet implemented.");
984 }
985
986 /*
987 * NOTE: Below there is a long switch-like if-elseif-elseif... block
988 * which the code goes through even for the df_conflict cases
989 * above. Well, it will once we don't die-not-implemented above.
990 */
991 if (ci->match_mask) {
992 ci->merged.clean = 1;
993 if (ci->match_mask == 6) {
994 /* stages[1] == stages[2] */
995 ci->merged.result.mode = ci->stages[1].mode;
996 oidcpy(&ci->merged.result.oid, &ci->stages[1].oid);
997 } else {
998 /* determine the mask of the side that didn't match */
999 unsigned int othermask = 7 & ~ci->match_mask;
1000 int side = (othermask == 4) ? 2 : 1;
1001
1002 ci->merged.result.mode = ci->stages[side].mode;
1003 ci->merged.is_null = !ci->merged.result.mode;
1004 oidcpy(&ci->merged.result.oid, &ci->stages[side].oid);
1005
1006 assert(othermask == 2 || othermask == 4);
1007 assert(ci->merged.is_null ==
1008 (ci->filemask == ci->match_mask));
1009 }
1010 } else if (ci->filemask >= 6 &&
1011 (S_IFMT & ci->stages[1].mode) !=
1012 (S_IFMT & ci->stages[2].mode)) {
1013 /*
1014 * Two different items from (file/submodule/symlink)
1015 */
1016 die("Not yet implemented.");
1017 } else if (ci->filemask >= 6) {
1018 /*
1019 * TODO: Needs a two-way or three-way content merge, but we're
1020 * just being lazy and copying the version from HEAD and
1021 * leaving it as conflicted.
1022 */
1023 ci->merged.clean = 0;
1024 ci->merged.result.mode = ci->stages[1].mode;
1025 oidcpy(&ci->merged.result.oid, &ci->stages[1].oid);
1026 /* When we fix above, we'll call handle_content_merge() */
1027 (void)handle_content_merge;
1028 } else if (ci->filemask == 3 || ci->filemask == 5) {
1029 /* Modify/delete */
1030 const char *modify_branch, *delete_branch;
1031 int side = (ci->filemask == 5) ? 2 : 1;
1032 int index = opt->priv->call_depth ? 0 : side;
1033
1034 ci->merged.result.mode = ci->stages[index].mode;
1035 oidcpy(&ci->merged.result.oid, &ci->stages[index].oid);
1036 ci->merged.clean = 0;
1037
1038 modify_branch = (side == 1) ? opt->branch1 : opt->branch2;
1039 delete_branch = (side == 1) ? opt->branch2 : opt->branch1;
1040
1041 path_msg(opt, path, 0,
1042 _("CONFLICT (modify/delete): %s deleted in %s "
1043 "and modified in %s. Version %s of %s left "
1044 "in tree."),
1045 path, delete_branch, modify_branch,
1046 modify_branch, path);
1047 } else if (ci->filemask == 2 || ci->filemask == 4) {
1048 /* Added on one side */
1049 int side = (ci->filemask == 4) ? 2 : 1;
1050 ci->merged.result.mode = ci->stages[side].mode;
1051 oidcpy(&ci->merged.result.oid, &ci->stages[side].oid);
1052 ci->merged.clean = !ci->df_conflict;
1053 } else if (ci->filemask == 1) {
1054 /* Deleted on both sides */
1055 ci->merged.is_null = 1;
1056 ci->merged.result.mode = 0;
1057 oidcpy(&ci->merged.result.oid, &null_oid);
1058 ci->merged.clean = 1;
1059 }
1060
1061 /*
1062 * If still conflicted, record it separately. This allows us to later
1063 * iterate over just conflicted entries when updating the index instead
1064 * of iterating over all entries.
1065 */
1066 if (!ci->merged.clean)
1067 strmap_put(&opt->priv->conflicted, path, ci);
1068 record_entry_for_tree(dir_metadata, path, &ci->merged);
1069 }
1070
1071 static void process_entries(struct merge_options *opt,
1072 struct object_id *result_oid)
1073 {
1074 struct hashmap_iter iter;
1075 struct strmap_entry *e;
1076 struct string_list plist = STRING_LIST_INIT_NODUP;
1077 struct string_list_item *entry;
1078 struct directory_versions dir_metadata = { STRING_LIST_INIT_NODUP,
1079 STRING_LIST_INIT_NODUP,
1080 NULL, 0 };
1081
1082 if (strmap_empty(&opt->priv->paths)) {
1083 oidcpy(result_oid, opt->repo->hash_algo->empty_tree);
1084 return;
1085 }
1086
1087 /* Hack to pre-allocate plist to the desired size */
1088 ALLOC_GROW(plist.items, strmap_get_size(&opt->priv->paths), plist.alloc);
1089
1090 /* Put every entry from paths into plist, then sort */
1091 strmap_for_each_entry(&opt->priv->paths, &iter, e) {
1092 string_list_append(&plist, e->key)->util = e->value;
1093 }
1094 plist.cmp = string_list_df_name_compare;
1095 string_list_sort(&plist);
1096
1097 /*
1098 * Iterate over the items in reverse order, so we can handle paths
1099 * below a directory before needing to handle the directory itself.
1100 *
1101 * This allows us to write subtrees before we need to write trees,
1102 * and it also enables sane handling of directory/file conflicts
1103 * (because it allows us to know whether the directory is still in
1104 * the way when it is time to process the file at the same path).
1105 */
1106 for (entry = &plist.items[plist.nr-1]; entry >= plist.items; --entry) {
1107 char *path = entry->string;
1108 /*
1109 * NOTE: mi may actually be a pointer to a conflict_info, but
1110 * we have to check mi->clean first to see if it's safe to
1111 * reassign to such a pointer type.
1112 */
1113 struct merged_info *mi = entry->util;
1114
1115 write_completed_directory(opt, mi->directory_name,
1116 &dir_metadata);
1117 if (mi->clean)
1118 record_entry_for_tree(&dir_metadata, path, mi);
1119 else {
1120 struct conflict_info *ci = (struct conflict_info *)mi;
1121 process_entry(opt, path, ci, &dir_metadata);
1122 }
1123 }
1124
1125 if (dir_metadata.offsets.nr != 1 ||
1126 (uintptr_t)dir_metadata.offsets.items[0].util != 0) {
1127 printf("dir_metadata.offsets.nr = %d (should be 1)\n",
1128 dir_metadata.offsets.nr);
1129 printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n",
1130 (unsigned)(uintptr_t)dir_metadata.offsets.items[0].util);
1131 fflush(stdout);
1132 BUG("dir_metadata accounting completely off; shouldn't happen");
1133 }
1134 write_tree(result_oid, &dir_metadata.versions, 0,
1135 opt->repo->hash_algo->rawsz);
1136 string_list_clear(&plist, 0);
1137 string_list_clear(&dir_metadata.versions, 0);
1138 string_list_clear(&dir_metadata.offsets, 0);
1139 }
1140
1141 /*** Function Grouping: functions related to merge_switch_to_result() ***/
1142
1143 static int checkout(struct merge_options *opt,
1144 struct tree *prev,
1145 struct tree *next)
1146 {
1147 /* Switch the index/working copy from old to new */
1148 int ret;
1149 struct tree_desc trees[2];
1150 struct unpack_trees_options unpack_opts;
1151
1152 memset(&unpack_opts, 0, sizeof(unpack_opts));
1153 unpack_opts.head_idx = -1;
1154 unpack_opts.src_index = opt->repo->index;
1155 unpack_opts.dst_index = opt->repo->index;
1156
1157 setup_unpack_trees_porcelain(&unpack_opts, "merge");
1158
1159 /*
1160 * NOTE: if this were just "git checkout" code, we would probably
1161 * read or refresh the cache and check for a conflicted index, but
1162 * builtin/merge.c or sequencer.c really needs to read the index
1163 * and check for conflicted entries before starting merging for a
1164 * good user experience (no sense waiting for merges/rebases before
1165 * erroring out), so there's no reason to duplicate that work here.
1166 */
1167
1168 /* 2-way merge to the new branch */
1169 unpack_opts.update = 1;
1170 unpack_opts.merge = 1;
1171 unpack_opts.quiet = 0; /* FIXME: sequencer might want quiet? */
1172 unpack_opts.verbose_update = (opt->verbosity > 2);
1173 unpack_opts.fn = twoway_merge;
1174 if (1/* FIXME: opts->overwrite_ignore*/) {
1175 unpack_opts.dir = xcalloc(1, sizeof(*unpack_opts.dir));
1176 unpack_opts.dir->flags |= DIR_SHOW_IGNORED;
1177 setup_standard_excludes(unpack_opts.dir);
1178 }
1179 parse_tree(prev);
1180 init_tree_desc(&trees[0], prev->buffer, prev->size);
1181 parse_tree(next);
1182 init_tree_desc(&trees[1], next->buffer, next->size);
1183
1184 ret = unpack_trees(2, trees, &unpack_opts);
1185 clear_unpack_trees_porcelain(&unpack_opts);
1186 dir_clear(unpack_opts.dir);
1187 FREE_AND_NULL(unpack_opts.dir);
1188 return ret;
1189 }
1190
1191 static int record_conflicted_index_entries(struct merge_options *opt,
1192 struct index_state *index,
1193 struct strmap *paths,
1194 struct strmap *conflicted)
1195 {
1196 struct hashmap_iter iter;
1197 struct strmap_entry *e;
1198 int errs = 0;
1199 int original_cache_nr;
1200
1201 if (strmap_empty(conflicted))
1202 return 0;
1203
1204 original_cache_nr = index->cache_nr;
1205
1206 /* Put every entry from paths into plist, then sort */
1207 strmap_for_each_entry(conflicted, &iter, e) {
1208 const char *path = e->key;
1209 struct conflict_info *ci = e->value;
1210 int pos;
1211 struct cache_entry *ce;
1212 int i;
1213
1214 VERIFY_CI(ci);
1215
1216 /*
1217 * The index will already have a stage=0 entry for this path,
1218 * because we created an as-merged-as-possible version of the
1219 * file and checkout() moved the working copy and index over
1220 * to that version.
1221 *
1222 * However, previous iterations through this loop will have
1223 * added unstaged entries to the end of the cache which
1224 * ignore the standard alphabetical ordering of cache
1225 * entries and break invariants needed for index_name_pos()
1226 * to work. However, we know the entry we want is before
1227 * those appended cache entries, so do a temporary swap on
1228 * cache_nr to only look through entries of interest.
1229 */
1230 SWAP(index->cache_nr, original_cache_nr);
1231 pos = index_name_pos(index, path, strlen(path));
1232 SWAP(index->cache_nr, original_cache_nr);
1233 if (pos < 0) {
1234 if (ci->filemask != 1)
1235 BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path);
1236 cache_tree_invalidate_path(index, path);
1237 } else {
1238 ce = index->cache[pos];
1239
1240 /*
1241 * Clean paths with CE_SKIP_WORKTREE set will not be
1242 * written to the working tree by the unpack_trees()
1243 * call in checkout(). Our conflicted entries would
1244 * have appeared clean to that code since we ignored
1245 * the higher order stages. Thus, we need override
1246 * the CE_SKIP_WORKTREE bit and manually write those
1247 * files to the working disk here.
1248 *
1249 * TODO: Implement this CE_SKIP_WORKTREE fixup.
1250 */
1251
1252 /*
1253 * Mark this cache entry for removal and instead add
1254 * new stage>0 entries corresponding to the
1255 * conflicts. If there are many conflicted entries, we
1256 * want to avoid memmove'ing O(NM) entries by
1257 * inserting the new entries one at a time. So,
1258 * instead, we just add the new cache entries to the
1259 * end (ignoring normal index requirements on sort
1260 * order) and sort the index once we're all done.
1261 */
1262 ce->ce_flags |= CE_REMOVE;
1263 }
1264
1265 for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) {
1266 struct version_info *vi;
1267 if (!(ci->filemask & (1ul << i)))
1268 continue;
1269 vi = &ci->stages[i];
1270 ce = make_cache_entry(index, vi->mode, &vi->oid,
1271 path, i+1, 0);
1272 add_index_entry(index, ce, ADD_CACHE_JUST_APPEND);
1273 }
1274 }
1275
1276 /*
1277 * Remove the unused cache entries (and invalidate the relevant
1278 * cache-trees), then sort the index entries to get the conflicted
1279 * entries we added to the end into their right locations.
1280 */
1281 remove_marked_cache_entries(index, 1);
1282 QSORT(index->cache, index->cache_nr, cmp_cache_name_compare);
1283
1284 return errs;
1285 }
1286
1287 void merge_switch_to_result(struct merge_options *opt,
1288 struct tree *head,
1289 struct merge_result *result,
1290 int update_worktree_and_index,
1291 int display_update_msgs)
1292 {
1293 assert(opt->priv == NULL);
1294 if (result->clean >= 0 && update_worktree_and_index) {
1295 struct merge_options_internal *opti = result->priv;
1296
1297 if (checkout(opt, head, result->tree)) {
1298 /* failure to function */
1299 result->clean = -1;
1300 return;
1301 }
1302
1303 if (record_conflicted_index_entries(opt, opt->repo->index,
1304 &opti->paths,
1305 &opti->conflicted)) {
1306 /* failure to function */
1307 result->clean = -1;
1308 return;
1309 }
1310 }
1311
1312 if (display_update_msgs) {
1313 struct merge_options_internal *opti = result->priv;
1314 struct hashmap_iter iter;
1315 struct strmap_entry *e;
1316 struct string_list olist = STRING_LIST_INIT_NODUP;
1317 int i;
1318
1319 /* Hack to pre-allocate olist to the desired size */
1320 ALLOC_GROW(olist.items, strmap_get_size(&opti->output),
1321 olist.alloc);
1322
1323 /* Put every entry from output into olist, then sort */
1324 strmap_for_each_entry(&opti->output, &iter, e) {
1325 string_list_append(&olist, e->key)->util = e->value;
1326 }
1327 string_list_sort(&olist);
1328
1329 /* Iterate over the items, printing them */
1330 for (i = 0; i < olist.nr; ++i) {
1331 struct strbuf *sb = olist.items[i].util;
1332
1333 printf("%s", sb->buf);
1334 }
1335 string_list_clear(&olist, 0);
1336 }
1337
1338 merge_finalize(opt, result);
1339 }
1340
1341 void merge_finalize(struct merge_options *opt,
1342 struct merge_result *result)
1343 {
1344 struct merge_options_internal *opti = result->priv;
1345
1346 assert(opt->priv == NULL);
1347
1348 clear_or_reinit_internal_opts(opti, 0);
1349 FREE_AND_NULL(opti);
1350 }
1351
1352 /*** Function Grouping: helper functions for merge_incore_*() ***/
1353
1354 static inline void set_commit_tree(struct commit *c, struct tree *t)
1355 {
1356 c->maybe_tree = t;
1357 }
1358
1359 static struct commit *make_virtual_commit(struct repository *repo,
1360 struct tree *tree,
1361 const char *comment)
1362 {
1363 struct commit *commit = alloc_commit_node(repo);
1364
1365 set_merge_remote_desc(commit, comment, (struct object *)commit);
1366 set_commit_tree(commit, tree);
1367 commit->object.parsed = 1;
1368 return commit;
1369 }
1370
1371 static void merge_start(struct merge_options *opt, struct merge_result *result)
1372 {
1373 /* Sanity checks on opt */
1374 assert(opt->repo);
1375
1376 assert(opt->branch1 && opt->branch2);
1377
1378 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
1379 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
1380 assert(opt->rename_limit >= -1);
1381 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
1382 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
1383
1384 assert(opt->xdl_opts >= 0);
1385 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
1386 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
1387
1388 /*
1389 * detect_renames, verbosity, buffer_output, and obuf are ignored
1390 * fields that were used by "recursive" rather than "ort" -- but
1391 * sanity check them anyway.
1392 */
1393 assert(opt->detect_renames >= -1 &&
1394 opt->detect_renames <= DIFF_DETECT_COPY);
1395 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
1396 assert(opt->buffer_output <= 2);
1397 assert(opt->obuf.len == 0);
1398
1399 assert(opt->priv == NULL);
1400
1401 /* Default to histogram diff. Actually, just hardcode it...for now. */
1402 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
1403
1404 /* Initialization of opt->priv, our internal merge data */
1405 opt->priv = xcalloc(1, sizeof(*opt->priv));
1406
1407 /*
1408 * Although we initialize opt->priv->paths with strdup_strings=0,
1409 * that's just to avoid making yet another copy of an allocated
1410 * string. Putting the entry into paths means we are taking
1411 * ownership, so we will later free it. paths_to_free is similar.
1412 *
1413 * In contrast, conflicted just has a subset of keys from paths, so
1414 * we don't want to free those (it'd be a duplicate free).
1415 */
1416 strmap_init_with_options(&opt->priv->paths, NULL, 0);
1417 strmap_init_with_options(&opt->priv->conflicted, NULL, 0);
1418 string_list_init(&opt->priv->paths_to_free, 0);
1419
1420 /*
1421 * keys & strbufs in output will sometimes need to outlive "paths",
1422 * so it will have a copy of relevant keys. It's probably a small
1423 * subset of the overall paths that have special output.
1424 */
1425 strmap_init(&opt->priv->output);
1426 }
1427
1428 /*** Function Grouping: merge_incore_*() and their internal variants ***/
1429
1430 /*
1431 * Originally from merge_trees_internal(); heavily adapted, though.
1432 */
1433 static void merge_ort_nonrecursive_internal(struct merge_options *opt,
1434 struct tree *merge_base,
1435 struct tree *side1,
1436 struct tree *side2,
1437 struct merge_result *result)
1438 {
1439 struct object_id working_tree_oid;
1440
1441 if (collect_merge_info(opt, merge_base, side1, side2) != 0) {
1442 /*
1443 * TRANSLATORS: The %s arguments are: 1) tree hash of a merge
1444 * base, and 2-3) the trees for the two trees we're merging.
1445 */
1446 err(opt, _("collecting merge info failed for trees %s, %s, %s"),
1447 oid_to_hex(&merge_base->object.oid),
1448 oid_to_hex(&side1->object.oid),
1449 oid_to_hex(&side2->object.oid));
1450 result->clean = -1;
1451 return;
1452 }
1453
1454 result->clean = detect_and_process_renames(opt, merge_base,
1455 side1, side2);
1456 process_entries(opt, &working_tree_oid);
1457
1458 /* Set return values */
1459 result->tree = parse_tree_indirect(&working_tree_oid);
1460 /* existence of conflicted entries implies unclean */
1461 result->clean &= strmap_empty(&opt->priv->conflicted);
1462 if (!opt->priv->call_depth) {
1463 result->priv = opt->priv;
1464 opt->priv = NULL;
1465 }
1466 }
1467
1468 /*
1469 * Originally from merge_recursive_internal(); somewhat adapted, though.
1470 */
1471 static void merge_ort_internal(struct merge_options *opt,
1472 struct commit_list *merge_bases,
1473 struct commit *h1,
1474 struct commit *h2,
1475 struct merge_result *result)
1476 {
1477 struct commit_list *iter;
1478 struct commit *merged_merge_bases;
1479 const char *ancestor_name;
1480 struct strbuf merge_base_abbrev = STRBUF_INIT;
1481
1482 if (!merge_bases) {
1483 merge_bases = get_merge_bases(h1, h2);
1484 /* See merge-ort.h:merge_incore_recursive() declaration NOTE */
1485 merge_bases = reverse_commit_list(merge_bases);
1486 }
1487
1488 merged_merge_bases = pop_commit(&merge_bases);
1489 if (merged_merge_bases == NULL) {
1490 /* if there is no common ancestor, use an empty tree */
1491 struct tree *tree;
1492
1493 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
1494 merged_merge_bases = make_virtual_commit(opt->repo, tree,
1495 "ancestor");
1496 ancestor_name = "empty tree";
1497 } else if (merge_bases) {
1498 ancestor_name = "merged common ancestors";
1499 } else {
1500 strbuf_add_unique_abbrev(&merge_base_abbrev,
1501 &merged_merge_bases->object.oid,
1502 DEFAULT_ABBREV);
1503 ancestor_name = merge_base_abbrev.buf;
1504 }
1505
1506 for (iter = merge_bases; iter; iter = iter->next) {
1507 const char *saved_b1, *saved_b2;
1508 struct commit *prev = merged_merge_bases;
1509
1510 opt->priv->call_depth++;
1511 /*
1512 * When the merge fails, the result contains files
1513 * with conflict markers. The cleanness flag is
1514 * ignored (unless indicating an error), it was never
1515 * actually used, as result of merge_trees has always
1516 * overwritten it: the committed "conflicts" were
1517 * already resolved.
1518 */
1519 saved_b1 = opt->branch1;
1520 saved_b2 = opt->branch2;
1521 opt->branch1 = "Temporary merge branch 1";
1522 opt->branch2 = "Temporary merge branch 2";
1523 merge_ort_internal(opt, NULL, prev, iter->item, result);
1524 if (result->clean < 0)
1525 return;
1526 opt->branch1 = saved_b1;
1527 opt->branch2 = saved_b2;
1528 opt->priv->call_depth--;
1529
1530 merged_merge_bases = make_virtual_commit(opt->repo,
1531 result->tree,
1532 "merged tree");
1533 commit_list_insert(prev, &merged_merge_bases->parents);
1534 commit_list_insert(iter->item,
1535 &merged_merge_bases->parents->next);
1536
1537 clear_or_reinit_internal_opts(opt->priv, 1);
1538 }
1539
1540 opt->ancestor = ancestor_name;
1541 merge_ort_nonrecursive_internal(opt,
1542 repo_get_commit_tree(opt->repo,
1543 merged_merge_bases),
1544 repo_get_commit_tree(opt->repo, h1),
1545 repo_get_commit_tree(opt->repo, h2),
1546 result);
1547 strbuf_release(&merge_base_abbrev);
1548 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
1549 }
1550
1551 void merge_incore_nonrecursive(struct merge_options *opt,
1552 struct tree *merge_base,
1553 struct tree *side1,
1554 struct tree *side2,
1555 struct merge_result *result)
1556 {
1557 assert(opt->ancestor != NULL);
1558 merge_start(opt, result);
1559 merge_ort_nonrecursive_internal(opt, merge_base, side1, side2, result);
1560 }
1561
1562 void merge_incore_recursive(struct merge_options *opt,
1563 struct commit_list *merge_bases,
1564 struct commit *side1,
1565 struct commit *side2,
1566 struct merge_result *result)
1567 {
1568 /* We set the ancestor label based on the merge_bases */
1569 assert(opt->ancestor == NULL);
1570
1571 merge_start(opt, result);
1572 merge_ort_internal(opt, merge_bases, side1, side2, result);
1573 }