]>
Commit | Line | Data |
---|---|---|
17e5574b EN |
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 | ||
4296d8f1 | 20 | #include "alloc.h" |
67845745 | 21 | #include "blob.h" |
ef2b3693 | 22 | #include "cache-tree.h" |
4296d8f1 | 23 | #include "commit.h" |
67845745 | 24 | #include "commit-reach.h" |
e4171b1b EN |
25 | #include "diff.h" |
26 | #include "diffcore.h" | |
6681ce5c | 27 | #include "dir.h" |
f591c472 | 28 | #include "ll-merge.h" |
ee4012dc | 29 | #include "object-store.h" |
4204cd59 | 30 | #include "revision.h" |
5b59c3db | 31 | #include "strmap.h" |
c73cda76 | 32 | #include "submodule.h" |
231e2dd4 | 33 | #include "tree.h" |
6681ce5c | 34 | #include "unpack-trees.h" |
c8017176 | 35 | #include "xdiff-interface.h" |
5b59c3db | 36 | |
d2bc1994 EN |
37 | /* |
38 | * We have many arrays of size 3. Whenever we have such an array, the | |
39 | * indices refer to one of the sides of the three-way merge. This is so | |
40 | * pervasive that the constants 0, 1, and 2 are used in many places in the | |
41 | * code (especially in arithmetic operations to find the other side's index | |
42 | * or to compute a relevant mask), but sometimes these enum names are used | |
43 | * to aid code clarity. | |
44 | * | |
45 | * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side" | |
46 | * referred to there is one of these three sides. | |
47 | */ | |
48 | enum merge_side { | |
49 | MERGE_BASE = 0, | |
50 | MERGE_SIDE1 = 1, | |
51 | MERGE_SIDE2 = 2 | |
52 | }; | |
53 | ||
864075ec | 54 | struct rename_info { |
c09376d5 EN |
55 | /* |
56 | * All variables that are arrays of size 3 correspond to data tracked | |
57 | * for the sides in enum merge_side. Index 0 is almost always unused | |
58 | * because we often only need to track information for MERGE_SIDE1 and | |
59 | * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames | |
60 | * are determined relative to what changed since the MERGE_BASE). | |
61 | */ | |
62 | ||
864075ec EN |
63 | /* |
64 | * pairs: pairing of filenames from diffcore_rename() | |
864075ec EN |
65 | */ |
66 | struct diff_queue_struct pairs[3]; | |
67 | ||
c09376d5 EN |
68 | /* |
69 | * dirs_removed: directories removed on a given side of history. | |
70 | */ | |
71 | struct strset dirs_removed[3]; | |
72 | ||
73 | /* | |
74 | * dir_rename_count: tracking where parts of a directory were renamed to | |
75 | * | |
76 | * When files in a directory are renamed, they may not all go to the | |
77 | * same location. Each strmap here tracks: | |
78 | * old_dir => {new_dir => int} | |
79 | * That is, dir_rename_count[side] is a strmap to a strintmap. | |
80 | */ | |
81 | struct strmap dir_rename_count[3]; | |
82 | ||
83 | /* | |
84 | * dir_renames: computed directory renames | |
85 | * | |
86 | * This is a map of old_dir => new_dir and is derived in part from | |
87 | * dir_rename_count. | |
88 | */ | |
89 | struct strmap dir_renames[3]; | |
90 | ||
864075ec EN |
91 | /* |
92 | * needed_limit: value needed for inexact rename detection to run | |
93 | * | |
94 | * If the current rename limit wasn't high enough for inexact | |
95 | * rename detection to run, this records the limit needed. Otherwise, | |
96 | * this value remains 0. | |
97 | */ | |
98 | int needed_limit; | |
99 | }; | |
100 | ||
5b59c3db EN |
101 | struct merge_options_internal { |
102 | /* | |
103 | * paths: primary data structure in all of merge ort. | |
104 | * | |
105 | * The keys of paths: | |
106 | * * are full relative paths from the toplevel of the repository | |
107 | * (e.g. "drivers/firmware/raspberrypi.c"). | |
108 | * * store all relevant paths in the repo, both directories and | |
109 | * files (e.g. drivers, drivers/firmware would also be included) | |
110 | * * these keys serve to intern all the path strings, which allows | |
111 | * us to do pointer comparison on directory names instead of | |
112 | * strcmp; we just have to be careful to use the interned strings. | |
43c1dccb EN |
113 | * (Technically paths_to_free may track some strings that were |
114 | * removed from froms paths.) | |
5b59c3db EN |
115 | * |
116 | * The values of paths: | |
117 | * * either a pointer to a merged_info, or a conflict_info struct | |
118 | * * merged_info contains all relevant information for a | |
119 | * non-conflicted entry. | |
120 | * * conflict_info contains a merged_info, plus any additional | |
121 | * information about a conflict such as the higher orders stages | |
122 | * involved and the names of the paths those came from (handy | |
123 | * once renames get involved). | |
124 | * * a path may start "conflicted" (i.e. point to a conflict_info) | |
125 | * and then a later step (e.g. three-way content merge) determines | |
126 | * it can be cleanly merged, at which point it'll be marked clean | |
127 | * and the algorithm will ignore any data outside the contained | |
128 | * merged_info for that entry | |
129 | * * If an entry remains conflicted, the merged_info portion of a | |
130 | * conflict_info will later be filled with whatever version of | |
131 | * the file should be placed in the working directory (e.g. an | |
132 | * as-merged-as-possible variation that contains conflict markers). | |
133 | */ | |
134 | struct strmap paths; | |
135 | ||
136 | /* | |
137 | * conflicted: a subset of keys->values from "paths" | |
138 | * | |
139 | * conflicted is basically an optimization between process_entries() | |
140 | * and record_conflicted_index_entries(); the latter could loop over | |
141 | * ALL the entries in paths AGAIN and look for the ones that are | |
142 | * still conflicted, but since process_entries() has to loop over | |
143 | * all of them, it saves the ones it couldn't resolve in this strmap | |
144 | * so that record_conflicted_index_entries() can iterate just the | |
145 | * relevant entries. | |
146 | */ | |
147 | struct strmap conflicted; | |
148 | ||
43c1dccb EN |
149 | /* |
150 | * paths_to_free: additional list of strings to free | |
151 | * | |
152 | * If keys are removed from "paths", they are added to paths_to_free | |
153 | * to ensure they are later freed. We avoid free'ing immediately since | |
154 | * other places (e.g. conflict_info.pathnames[]) may still be | |
155 | * referencing these paths. | |
156 | */ | |
157 | struct string_list paths_to_free; | |
158 | ||
c5a6f655 EN |
159 | /* |
160 | * output: special messages and conflict notices for various paths | |
161 | * | |
162 | * This is a map of pathnames (a subset of the keys in "paths" above) | |
163 | * to strbufs. It gathers various warning/conflict/notice messages | |
164 | * for later processing. | |
165 | */ | |
166 | struct strmap output; | |
167 | ||
5b59c3db | 168 | /* |
864075ec EN |
169 | * renames: various data relating to rename detection |
170 | */ | |
171 | struct rename_info renames; | |
172 | ||
5b59c3db | 173 | /* |
05b85c6e | 174 | * current_dir_name, toplevel_dir: temporary vars |
5b59c3db | 175 | * |
05b85c6e EN |
176 | * These are used in collect_merge_info_callback(), and will set the |
177 | * various merged_info.directory_name for the various paths we get; | |
178 | * see documentation for that variable and the requirements placed on | |
179 | * that field. | |
5b59c3db EN |
180 | */ |
181 | const char *current_dir_name; | |
05b85c6e | 182 | const char *toplevel_dir; |
5b59c3db EN |
183 | |
184 | /* call_depth: recursion level counter for merging merge bases */ | |
185 | int call_depth; | |
186 | }; | |
187 | ||
188 | struct version_info { | |
189 | struct object_id oid; | |
190 | unsigned short mode; | |
191 | }; | |
192 | ||
193 | struct merged_info { | |
194 | /* if is_null, ignore result. otherwise result has oid & mode */ | |
195 | struct version_info result; | |
196 | unsigned is_null:1; | |
197 | ||
198 | /* | |
199 | * clean: whether the path in question is cleanly merged. | |
200 | * | |
201 | * see conflict_info.merged for more details. | |
202 | */ | |
203 | unsigned clean:1; | |
204 | ||
205 | /* | |
206 | * basename_offset: offset of basename of path. | |
207 | * | |
208 | * perf optimization to avoid recomputing offset of final '/' | |
209 | * character in pathname (0 if no '/' in pathname). | |
210 | */ | |
211 | size_t basename_offset; | |
212 | ||
213 | /* | |
214 | * directory_name: containing directory name. | |
215 | * | |
216 | * Note that we assume directory_name is constructed such that | |
217 | * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name, | |
218 | * i.e. string equality is equivalent to pointer equality. For this | |
219 | * to hold, we have to be careful setting directory_name. | |
220 | */ | |
221 | const char *directory_name; | |
222 | }; | |
223 | ||
224 | struct conflict_info { | |
225 | /* | |
226 | * merged: the version of the path that will be written to working tree | |
227 | * | |
228 | * WARNING: It is critical to check merged.clean and ensure it is 0 | |
229 | * before reading any conflict_info fields outside of merged. | |
230 | * Allocated merge_info structs will always have clean set to 1. | |
231 | * Allocated conflict_info structs will have merged.clean set to 0 | |
232 | * initially. The merged.clean field is how we know if it is safe | |
233 | * to access other parts of conflict_info besides merged; if a | |
234 | * conflict_info's merged.clean is changed to 1, the rest of the | |
235 | * algorithm is not allowed to look at anything outside of the | |
236 | * merged member anymore. | |
237 | */ | |
238 | struct merged_info merged; | |
239 | ||
240 | /* oids & modes from each of the three trees for this path */ | |
241 | struct version_info stages[3]; | |
242 | ||
243 | /* pathnames for each stage; may differ due to rename detection */ | |
244 | const char *pathnames[3]; | |
245 | ||
246 | /* Whether this path is/was involved in a directory/file conflict */ | |
247 | unsigned df_conflict:1; | |
248 | ||
1c7873cd EN |
249 | /* |
250 | * Whether this path is/was involved in a non-content conflict other | |
251 | * than a directory/file conflict (e.g. rename/rename, rename/delete, | |
252 | * file location based on possible directory rename). | |
253 | */ | |
254 | unsigned path_conflict:1; | |
255 | ||
5b59c3db EN |
256 | /* |
257 | * For filemask and dirmask, the ith bit corresponds to whether the | |
258 | * ith entry is a file (filemask) or a directory (dirmask). Thus, | |
259 | * filemask & dirmask is always zero, and filemask | dirmask is at | |
260 | * most 7 but can be less when a path does not appear as either a | |
261 | * file or a directory on at least one side of history. | |
262 | * | |
263 | * Note that these masks are related to enum merge_side, as the ith | |
264 | * entry corresponds to side i. | |
265 | * | |
266 | * These values come from a traverse_trees() call; more info may be | |
267 | * found looking at tree-walk.h's struct traverse_info, | |
268 | * particularly the documentation above the "fn" member (note that | |
269 | * filemask = mask & ~dirmask from that documentation). | |
270 | */ | |
271 | unsigned filemask:3; | |
272 | unsigned dirmask:3; | |
273 | ||
274 | /* | |
275 | * Optimization to track which stages match, to avoid the need to | |
276 | * recompute it in multiple steps. Either 0 or at least 2 bits are | |
277 | * set; if at least 2 bits are set, their corresponding stages match. | |
278 | */ | |
279 | unsigned match_mask:3; | |
280 | }; | |
281 | ||
04af1879 EN |
282 | /*** Function Grouping: various utility functions ***/ |
283 | ||
98bf9841 EN |
284 | /* |
285 | * For the next three macros, see warning for conflict_info.merged. | |
286 | * | |
287 | * In each of the below, mi is a struct merged_info*, and ci was defined | |
288 | * as a struct conflict_info* (but we need to verify ci isn't actually | |
289 | * pointed at a struct merged_info*). | |
290 | * | |
291 | * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise. | |
292 | * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one. | |
293 | * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first. | |
294 | */ | |
295 | #define INITIALIZE_CI(ci, mi) do { \ | |
296 | (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \ | |
297 | } while (0) | |
298 | #define VERIFY_CI(ci) assert(ci && !ci->merged.clean); | |
299 | #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \ | |
300 | (ci) = (struct conflict_info *)(mi); \ | |
301 | assert((ci) && !(mi)->clean); \ | |
302 | } while (0) | |
303 | ||
89422d29 EN |
304 | static void free_strmap_strings(struct strmap *map) |
305 | { | |
306 | struct hashmap_iter iter; | |
307 | struct strmap_entry *entry; | |
308 | ||
309 | strmap_for_each_entry(map, &iter, entry) { | |
310 | free((char*)entry->key); | |
311 | } | |
312 | } | |
313 | ||
43e9c4ee EN |
314 | static void clear_or_reinit_internal_opts(struct merge_options_internal *opti, |
315 | int reinitialize) | |
101bc5bc | 316 | { |
f5d9fbc2 EN |
317 | struct rename_info *renames = &opti->renames; |
318 | int i; | |
43e9c4ee EN |
319 | void (*strmap_func)(struct strmap *, int) = |
320 | reinitialize ? strmap_partial_clear : strmap_clear; | |
f5d9fbc2 EN |
321 | void (*strset_func)(struct strset *) = |
322 | reinitialize ? strset_partial_clear : strset_clear; | |
101bc5bc EN |
323 | |
324 | /* | |
325 | * We marked opti->paths with strdup_strings = 0, so that we | |
326 | * wouldn't have to make another copy of the fullpath created by | |
327 | * make_traverse_path from setup_path_info(). But, now that we've | |
328 | * used it and have no other references to these strings, it is time | |
329 | * to deallocate them. | |
330 | */ | |
331 | free_strmap_strings(&opti->paths); | |
43e9c4ee | 332 | strmap_func(&opti->paths, 1); |
101bc5bc EN |
333 | |
334 | /* | |
335 | * All keys and values in opti->conflicted are a subset of those in | |
336 | * opti->paths. We don't want to deallocate anything twice, so we | |
337 | * don't free the keys and we pass 0 for free_values. | |
338 | */ | |
43e9c4ee | 339 | strmap_func(&opti->conflicted, 0); |
43c1dccb EN |
340 | |
341 | /* | |
342 | * opti->paths_to_free is similar to opti->paths; we created it with | |
343 | * strdup_strings = 0 to avoid making _another_ copy of the fullpath | |
344 | * but now that we've used it and have no other references to these | |
345 | * strings, it is time to deallocate them. We do so by temporarily | |
346 | * setting strdup_strings to 1. | |
347 | */ | |
348 | opti->paths_to_free.strdup_strings = 1; | |
349 | string_list_clear(&opti->paths_to_free, 0); | |
350 | opti->paths_to_free.strdup_strings = 0; | |
c5a6f655 | 351 | |
f5d9fbc2 EN |
352 | /* Free memory used by various renames maps */ |
353 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; ++i) { | |
354 | struct hashmap_iter iter; | |
355 | struct strmap_entry *entry; | |
356 | ||
357 | strset_func(&renames->dirs_removed[i]); | |
358 | ||
359 | strmap_for_each_entry(&renames->dir_rename_count[i], | |
360 | &iter, entry) { | |
361 | struct strintmap *counts = entry->value; | |
362 | strintmap_clear(counts); | |
363 | } | |
364 | strmap_func(&renames->dir_rename_count[i], 1); | |
365 | ||
366 | strmap_func(&renames->dir_renames[i], 0); | |
367 | } | |
368 | ||
c5a6f655 EN |
369 | if (!reinitialize) { |
370 | struct hashmap_iter iter; | |
371 | struct strmap_entry *e; | |
372 | ||
373 | /* Release and free each strbuf found in output */ | |
374 | strmap_for_each_entry(&opti->output, &iter, e) { | |
375 | struct strbuf *sb = e->value; | |
376 | strbuf_release(sb); | |
377 | /* | |
378 | * While strictly speaking we don't need to free(sb) | |
379 | * here because we could pass free_values=1 when | |
380 | * calling strmap_clear() on opti->output, that would | |
381 | * require strmap_clear to do another | |
382 | * strmap_for_each_entry() loop, so we just free it | |
383 | * while we're iterating anyway. | |
384 | */ | |
385 | free(sb); | |
386 | } | |
387 | strmap_clear(&opti->output, 0); | |
388 | } | |
101bc5bc EN |
389 | } |
390 | ||
0c0d705b EN |
391 | static int err(struct merge_options *opt, const char *err, ...) |
392 | { | |
393 | va_list params; | |
394 | struct strbuf sb = STRBUF_INIT; | |
395 | ||
396 | strbuf_addstr(&sb, "error: "); | |
397 | va_start(params, err); | |
398 | strbuf_vaddf(&sb, err, params); | |
399 | va_end(params); | |
400 | ||
401 | error("%s", sb.buf); | |
402 | strbuf_release(&sb); | |
403 | ||
404 | return -1; | |
405 | } | |
406 | ||
c73cda76 EN |
407 | static void format_commit(struct strbuf *sb, |
408 | int indent, | |
409 | struct commit *commit) | |
410 | { | |
70f19c7f EN |
411 | struct merge_remote_desc *desc; |
412 | struct pretty_print_context ctx = {0}; | |
413 | ctx.abbrev = DEFAULT_ABBREV; | |
414 | ||
415 | strbuf_addchars(sb, ' ', indent); | |
416 | desc = merge_remote_util(commit); | |
417 | if (desc) { | |
418 | strbuf_addf(sb, "virtual %s\n", desc->name); | |
419 | return; | |
420 | } | |
421 | ||
422 | format_commit_message(commit, "%h %s", sb, &ctx); | |
423 | strbuf_addch(sb, '\n'); | |
c73cda76 EN |
424 | } |
425 | ||
c5a6f655 EN |
426 | __attribute__((format (printf, 4, 5))) |
427 | static void path_msg(struct merge_options *opt, | |
428 | const char *path, | |
429 | int omittable_hint, /* skippable under --remerge-diff */ | |
430 | const char *fmt, ...) | |
431 | { | |
432 | va_list ap; | |
433 | struct strbuf *sb = strmap_get(&opt->priv->output, path); | |
434 | if (!sb) { | |
435 | sb = xmalloc(sizeof(*sb)); | |
436 | strbuf_init(sb, 0); | |
437 | strmap_put(&opt->priv->output, path, sb); | |
438 | } | |
439 | ||
440 | va_start(ap, fmt); | |
441 | strbuf_vaddf(sb, fmt, ap); | |
442 | va_end(ap); | |
443 | ||
444 | strbuf_addch(sb, '\n'); | |
445 | } | |
446 | ||
5a1a1e8e EN |
447 | /* add a string to a strbuf, but converting "/" to "_" */ |
448 | static void add_flattened_path(struct strbuf *out, const char *s) | |
449 | { | |
450 | size_t i = out->len; | |
451 | strbuf_addstr(out, s); | |
452 | for (; i < out->len; i++) | |
453 | if (out->buf[i] == '/') | |
454 | out->buf[i] = '_'; | |
455 | } | |
456 | ||
23366d2a EN |
457 | static char *unique_path(struct strmap *existing_paths, |
458 | const char *path, | |
459 | const char *branch) | |
460 | { | |
5a1a1e8e EN |
461 | struct strbuf newpath = STRBUF_INIT; |
462 | int suffix = 0; | |
463 | size_t base_len; | |
464 | ||
465 | strbuf_addf(&newpath, "%s~", path); | |
466 | add_flattened_path(&newpath, branch); | |
467 | ||
468 | base_len = newpath.len; | |
469 | while (strmap_contains(existing_paths, newpath.buf)) { | |
470 | strbuf_setlen(&newpath, base_len); | |
471 | strbuf_addf(&newpath, "_%d", suffix++); | |
472 | } | |
473 | ||
474 | return strbuf_detach(&newpath, NULL); | |
23366d2a EN |
475 | } |
476 | ||
04af1879 EN |
477 | /*** Function Grouping: functions related to collect_merge_info() ***/ |
478 | ||
98bf9841 EN |
479 | static void setup_path_info(struct merge_options *opt, |
480 | struct string_list_item *result, | |
481 | const char *current_dir_name, | |
482 | int current_dir_name_len, | |
483 | char *fullpath, /* we'll take over ownership */ | |
484 | struct name_entry *names, | |
485 | struct name_entry *merged_version, | |
486 | unsigned is_null, /* boolean */ | |
487 | unsigned df_conflict, /* boolean */ | |
488 | unsigned filemask, | |
489 | unsigned dirmask, | |
490 | int resolved /* boolean */) | |
491 | { | |
492 | /* result->util is void*, so mi is a convenience typed variable */ | |
493 | struct merged_info *mi; | |
494 | ||
495 | assert(!is_null || resolved); | |
496 | assert(!df_conflict || !resolved); /* df_conflict implies !resolved */ | |
497 | assert(resolved == (merged_version != NULL)); | |
498 | ||
499 | mi = xcalloc(1, resolved ? sizeof(struct merged_info) : | |
500 | sizeof(struct conflict_info)); | |
501 | mi->directory_name = current_dir_name; | |
502 | mi->basename_offset = current_dir_name_len; | |
503 | mi->clean = !!resolved; | |
504 | if (resolved) { | |
505 | mi->result.mode = merged_version->mode; | |
506 | oidcpy(&mi->result.oid, &merged_version->oid); | |
507 | mi->is_null = !!is_null; | |
508 | } else { | |
509 | int i; | |
510 | struct conflict_info *ci; | |
511 | ||
512 | ASSIGN_AND_VERIFY_CI(ci, mi); | |
513 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
514 | ci->pathnames[i] = fullpath; | |
515 | ci->stages[i].mode = names[i].mode; | |
516 | oidcpy(&ci->stages[i].oid, &names[i].oid); | |
517 | } | |
518 | ci->filemask = filemask; | |
519 | ci->dirmask = dirmask; | |
520 | ci->df_conflict = !!df_conflict; | |
521 | if (dirmask) | |
522 | /* | |
523 | * Assume is_null for now, but if we have entries | |
524 | * under the directory then when it is complete in | |
525 | * write_completed_directory() it'll update this. | |
526 | * Also, for D/F conflicts, we have to handle the | |
527 | * directory first, then clear this bit and process | |
528 | * the file to see how it is handled -- that occurs | |
529 | * near the top of process_entry(). | |
530 | */ | |
531 | mi->is_null = 1; | |
532 | } | |
533 | strmap_put(&opt->priv->paths, fullpath, mi); | |
534 | result->string = fullpath; | |
535 | result->util = mi; | |
536 | } | |
537 | ||
eb3e3e1d EN |
538 | static void collect_rename_info(struct merge_options *opt, |
539 | struct name_entry *names, | |
540 | const char *dirname, | |
541 | const char *fullname, | |
542 | unsigned filemask, | |
543 | unsigned dirmask, | |
544 | unsigned match_mask) | |
545 | { | |
546 | struct rename_info *renames = &opt->priv->renames; | |
547 | ||
548 | /* Update dirs_removed, as needed */ | |
549 | if (dirmask == 1 || dirmask == 3 || dirmask == 5) { | |
550 | /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */ | |
551 | unsigned sides = (0x07 - dirmask)/2; | |
552 | if (sides & 1) | |
553 | strset_add(&renames->dirs_removed[1], fullname); | |
554 | if (sides & 2) | |
555 | strset_add(&renames->dirs_removed[2], fullname); | |
556 | } | |
557 | } | |
558 | ||
d2bc1994 EN |
559 | static int collect_merge_info_callback(int n, |
560 | unsigned long mask, | |
561 | unsigned long dirmask, | |
562 | struct name_entry *names, | |
563 | struct traverse_info *info) | |
564 | { | |
565 | /* | |
566 | * n is 3. Always. | |
567 | * common ancestor (mbase) has mask 1, and stored in index 0 of names | |
568 | * head of side 1 (side1) has mask 2, and stored in index 1 of names | |
569 | * head of side 2 (side2) has mask 4, and stored in index 2 of names | |
570 | */ | |
571 | struct merge_options *opt = info->data; | |
572 | struct merge_options_internal *opti = opt->priv; | |
98bf9841 EN |
573 | struct string_list_item pi; /* Path Info */ |
574 | struct conflict_info *ci; /* typed alias to pi.util (which is void*) */ | |
d2bc1994 EN |
575 | struct name_entry *p; |
576 | size_t len; | |
577 | char *fullpath; | |
98bf9841 | 578 | const char *dirname = opti->current_dir_name; |
d2bc1994 | 579 | unsigned filemask = mask & ~dirmask; |
34e557af | 580 | unsigned match_mask = 0; /* will be updated below */ |
d2bc1994 EN |
581 | unsigned mbase_null = !(mask & 1); |
582 | unsigned side1_null = !(mask & 2); | |
583 | unsigned side2_null = !(mask & 4); | |
885f0063 EN |
584 | unsigned side1_matches_mbase = (!side1_null && !mbase_null && |
585 | names[0].mode == names[1].mode && | |
586 | oideq(&names[0].oid, &names[1].oid)); | |
587 | unsigned side2_matches_mbase = (!side2_null && !mbase_null && | |
588 | names[0].mode == names[2].mode && | |
589 | oideq(&names[0].oid, &names[2].oid)); | |
590 | unsigned sides_match = (!side1_null && !side2_null && | |
591 | names[1].mode == names[2].mode && | |
592 | oideq(&names[1].oid, &names[2].oid)); | |
d2bc1994 | 593 | |
34e557af EN |
594 | /* |
595 | * Note: When a path is a file on one side of history and a directory | |
596 | * in another, we have a directory/file conflict. In such cases, if | |
597 | * the conflict doesn't resolve from renames and deletions, then we | |
598 | * always leave directories where they are and move files out of the | |
599 | * way. Thus, while struct conflict_info has a df_conflict field to | |
600 | * track such conflicts, we ignore that field for any directories at | |
601 | * a path and only pay attention to it for files at the given path. | |
602 | * The fact that we leave directories were they are also means that | |
603 | * we do not need to worry about getting additional df_conflict | |
604 | * information propagated from parent directories down to children | |
605 | * (unlike, say traverse_trees_recursive() in unpack-trees.c, which | |
606 | * sets a newinfo.df_conflicts field specifically to propagate it). | |
607 | */ | |
608 | unsigned df_conflict = (filemask != 0) && (dirmask != 0); | |
609 | ||
d2bc1994 EN |
610 | /* n = 3 is a fundamental assumption. */ |
611 | if (n != 3) | |
612 | BUG("Called collect_merge_info_callback wrong"); | |
613 | ||
614 | /* | |
615 | * A bunch of sanity checks verifying that traverse_trees() calls | |
616 | * us the way I expect. Could just remove these at some point, | |
617 | * though maybe they are helpful to future code readers. | |
618 | */ | |
619 | assert(mbase_null == is_null_oid(&names[0].oid)); | |
620 | assert(side1_null == is_null_oid(&names[1].oid)); | |
621 | assert(side2_null == is_null_oid(&names[2].oid)); | |
622 | assert(!mbase_null || !side1_null || !side2_null); | |
623 | assert(mask > 0 && mask < 8); | |
624 | ||
34e557af EN |
625 | /* Determine match_mask */ |
626 | if (side1_matches_mbase) | |
627 | match_mask = (side2_matches_mbase ? 7 : 3); | |
628 | else if (side2_matches_mbase) | |
629 | match_mask = 5; | |
630 | else if (sides_match) | |
631 | match_mask = 6; | |
632 | ||
d2bc1994 EN |
633 | /* |
634 | * Get the name of the relevant filepath, which we'll pass to | |
635 | * setup_path_info() for tracking. | |
636 | */ | |
637 | p = names; | |
638 | while (!p->mode) | |
639 | p++; | |
640 | len = traverse_path_len(info, p->pathlen); | |
641 | ||
642 | /* +1 in both of the following lines to include the NUL byte */ | |
643 | fullpath = xmalloc(len + 1); | |
644 | make_traverse_path(fullpath, len + 1, info, p->path, p->pathlen); | |
645 | ||
291f29ca EN |
646 | /* |
647 | * If mbase, side1, and side2 all match, we can resolve early. Even | |
648 | * if these are trees, there will be no renames or anything | |
649 | * underneath. | |
650 | */ | |
651 | if (side1_matches_mbase && side2_matches_mbase) { | |
652 | /* mbase, side1, & side2 all match; use mbase as resolution */ | |
653 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
654 | names, names+0, mbase_null, 0, | |
655 | filemask, dirmask, 1); | |
656 | return mask; | |
657 | } | |
658 | ||
eb3e3e1d EN |
659 | /* |
660 | * Gather additional information used in rename detection. | |
661 | */ | |
662 | collect_rename_info(opt, names, dirname, fullpath, | |
663 | filemask, dirmask, match_mask); | |
664 | ||
d2bc1994 | 665 | /* |
98bf9841 EN |
666 | * Record information about the path so we can resolve later in |
667 | * process_entries. | |
d2bc1994 | 668 | */ |
98bf9841 EN |
669 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, |
670 | names, NULL, 0, df_conflict, filemask, dirmask, 0); | |
671 | ||
672 | ci = pi.util; | |
673 | VERIFY_CI(ci); | |
34e557af | 674 | ci->match_mask = match_mask; |
d2bc1994 EN |
675 | |
676 | /* If dirmask, recurse into subdirectories */ | |
677 | if (dirmask) { | |
678 | struct traverse_info newinfo; | |
679 | struct tree_desc t[3]; | |
680 | void *buf[3] = {NULL, NULL, NULL}; | |
681 | const char *original_dir_name; | |
682 | int i, ret; | |
683 | ||
684 | ci->match_mask &= filemask; | |
685 | newinfo = *info; | |
686 | newinfo.prev = info; | |
687 | newinfo.name = p->path; | |
688 | newinfo.namelen = p->pathlen; | |
689 | newinfo.pathlen = st_add3(newinfo.pathlen, p->pathlen, 1); | |
34e557af EN |
690 | /* |
691 | * If this directory we are about to recurse into cared about | |
692 | * its parent directory (the current directory) having a D/F | |
693 | * conflict, then we'd propagate the masks in this way: | |
694 | * newinfo.df_conflicts |= (mask & ~dirmask); | |
695 | * But we don't worry about propagating D/F conflicts. (See | |
696 | * comment near setting of local df_conflict variable near | |
697 | * the beginning of this function). | |
698 | */ | |
d2bc1994 EN |
699 | |
700 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
885f0063 EN |
701 | if (i == 1 && side1_matches_mbase) |
702 | t[1] = t[0]; | |
703 | else if (i == 2 && side2_matches_mbase) | |
704 | t[2] = t[0]; | |
705 | else if (i == 2 && sides_match) | |
706 | t[2] = t[1]; | |
707 | else { | |
708 | const struct object_id *oid = NULL; | |
709 | if (dirmask & 1) | |
710 | oid = &names[i].oid; | |
711 | buf[i] = fill_tree_descriptor(opt->repo, | |
712 | t + i, oid); | |
713 | } | |
d2bc1994 EN |
714 | dirmask >>= 1; |
715 | } | |
716 | ||
717 | original_dir_name = opti->current_dir_name; | |
98bf9841 | 718 | opti->current_dir_name = pi.string; |
d2bc1994 EN |
719 | ret = traverse_trees(NULL, 3, t, &newinfo); |
720 | opti->current_dir_name = original_dir_name; | |
721 | ||
722 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) | |
723 | free(buf[i]); | |
724 | ||
725 | if (ret < 0) | |
726 | return -1; | |
727 | } | |
728 | ||
729 | return mask; | |
730 | } | |
731 | ||
231e2dd4 EN |
732 | static int collect_merge_info(struct merge_options *opt, |
733 | struct tree *merge_base, | |
734 | struct tree *side1, | |
735 | struct tree *side2) | |
736 | { | |
d2bc1994 EN |
737 | int ret; |
738 | struct tree_desc t[3]; | |
739 | struct traverse_info info; | |
d2bc1994 | 740 | |
05b85c6e EN |
741 | opt->priv->toplevel_dir = ""; |
742 | opt->priv->current_dir_name = opt->priv->toplevel_dir; | |
743 | setup_traverse_info(&info, opt->priv->toplevel_dir); | |
d2bc1994 EN |
744 | info.fn = collect_merge_info_callback; |
745 | info.data = opt; | |
746 | info.show_all_errors = 1; | |
747 | ||
748 | parse_tree(merge_base); | |
749 | parse_tree(side1); | |
750 | parse_tree(side2); | |
751 | init_tree_desc(t + 0, merge_base->buffer, merge_base->size); | |
752 | init_tree_desc(t + 1, side1->buffer, side1->size); | |
753 | init_tree_desc(t + 2, side2->buffer, side2->size); | |
754 | ||
557ac035 | 755 | trace2_region_enter("merge", "traverse_trees", opt->repo); |
d2bc1994 | 756 | ret = traverse_trees(NULL, 3, t, &info); |
557ac035 | 757 | trace2_region_leave("merge", "traverse_trees", opt->repo); |
d2bc1994 EN |
758 | |
759 | return ret; | |
231e2dd4 EN |
760 | } |
761 | ||
04af1879 EN |
762 | /*** Function Grouping: functions related to threeway content merges ***/ |
763 | ||
c73cda76 EN |
764 | static int find_first_merges(struct repository *repo, |
765 | const char *path, | |
766 | struct commit *a, | |
767 | struct commit *b, | |
768 | struct object_array *result) | |
769 | { | |
4204cd59 EN |
770 | int i, j; |
771 | struct object_array merges = OBJECT_ARRAY_INIT; | |
772 | struct commit *commit; | |
773 | int contains_another; | |
774 | ||
775 | char merged_revision[GIT_MAX_HEXSZ + 2]; | |
776 | const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path", | |
777 | "--all", merged_revision, NULL }; | |
778 | struct rev_info revs; | |
779 | struct setup_revision_opt rev_opts; | |
780 | ||
781 | memset(result, 0, sizeof(struct object_array)); | |
782 | memset(&rev_opts, 0, sizeof(rev_opts)); | |
783 | ||
784 | /* get all revisions that merge commit a */ | |
785 | xsnprintf(merged_revision, sizeof(merged_revision), "^%s", | |
786 | oid_to_hex(&a->object.oid)); | |
787 | repo_init_revisions(repo, &revs, NULL); | |
788 | rev_opts.submodule = path; | |
789 | /* FIXME: can't handle linked worktrees in submodules yet */ | |
790 | revs.single_worktree = path != NULL; | |
791 | setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts); | |
792 | ||
793 | /* save all revisions from the above list that contain b */ | |
794 | if (prepare_revision_walk(&revs)) | |
795 | die("revision walk setup failed"); | |
796 | while ((commit = get_revision(&revs)) != NULL) { | |
797 | struct object *o = &(commit->object); | |
798 | if (in_merge_bases(b, commit)) | |
799 | add_object_array(o, NULL, &merges); | |
800 | } | |
801 | reset_revision_walk(); | |
802 | ||
803 | /* Now we've got all merges that contain a and b. Prune all | |
804 | * merges that contain another found merge and save them in | |
805 | * result. | |
806 | */ | |
807 | for (i = 0; i < merges.nr; i++) { | |
808 | struct commit *m1 = (struct commit *) merges.objects[i].item; | |
809 | ||
810 | contains_another = 0; | |
811 | for (j = 0; j < merges.nr; j++) { | |
812 | struct commit *m2 = (struct commit *) merges.objects[j].item; | |
813 | if (i != j && in_merge_bases(m2, m1)) { | |
814 | contains_another = 1; | |
815 | break; | |
816 | } | |
817 | } | |
818 | ||
819 | if (!contains_another) | |
820 | add_object_array(merges.objects[i].item, NULL, result); | |
821 | } | |
822 | ||
823 | object_array_clear(&merges); | |
824 | return result->nr; | |
c73cda76 EN |
825 | } |
826 | ||
62fdec17 EN |
827 | static int merge_submodule(struct merge_options *opt, |
828 | const char *path, | |
829 | const struct object_id *o, | |
830 | const struct object_id *a, | |
831 | const struct object_id *b, | |
832 | struct object_id *result) | |
833 | { | |
c73cda76 EN |
834 | struct commit *commit_o, *commit_a, *commit_b; |
835 | int parent_count; | |
836 | struct object_array merges; | |
837 | struct strbuf sb = STRBUF_INIT; | |
838 | ||
839 | int i; | |
840 | int search = !opt->priv->call_depth; | |
841 | ||
842 | /* store fallback answer in result in case we fail */ | |
843 | oidcpy(result, opt->priv->call_depth ? o : a); | |
844 | ||
845 | /* we can not handle deletion conflicts */ | |
846 | if (is_null_oid(o)) | |
847 | return 0; | |
848 | if (is_null_oid(a)) | |
849 | return 0; | |
850 | if (is_null_oid(b)) | |
851 | return 0; | |
852 | ||
853 | if (add_submodule_odb(path)) { | |
854 | path_msg(opt, path, 0, | |
855 | _("Failed to merge submodule %s (not checked out)"), | |
856 | path); | |
857 | return 0; | |
858 | } | |
859 | ||
860 | if (!(commit_o = lookup_commit_reference(opt->repo, o)) || | |
861 | !(commit_a = lookup_commit_reference(opt->repo, a)) || | |
862 | !(commit_b = lookup_commit_reference(opt->repo, b))) { | |
863 | path_msg(opt, path, 0, | |
864 | _("Failed to merge submodule %s (commits not present)"), | |
865 | path); | |
866 | return 0; | |
867 | } | |
868 | ||
869 | /* check whether both changes are forward */ | |
870 | if (!in_merge_bases(commit_o, commit_a) || | |
871 | !in_merge_bases(commit_o, commit_b)) { | |
872 | path_msg(opt, path, 0, | |
873 | _("Failed to merge submodule %s " | |
874 | "(commits don't follow merge-base)"), | |
875 | path); | |
876 | return 0; | |
877 | } | |
878 | ||
879 | /* Case #1: a is contained in b or vice versa */ | |
880 | if (in_merge_bases(commit_a, commit_b)) { | |
881 | oidcpy(result, b); | |
882 | path_msg(opt, path, 1, | |
883 | _("Note: Fast-forwarding submodule %s to %s"), | |
884 | path, oid_to_hex(b)); | |
885 | return 1; | |
886 | } | |
887 | if (in_merge_bases(commit_b, commit_a)) { | |
888 | oidcpy(result, a); | |
889 | path_msg(opt, path, 1, | |
890 | _("Note: Fast-forwarding submodule %s to %s"), | |
891 | path, oid_to_hex(a)); | |
892 | return 1; | |
893 | } | |
894 | ||
895 | /* | |
896 | * Case #2: There are one or more merges that contain a and b in | |
897 | * the submodule. If there is only one, then present it as a | |
898 | * suggestion to the user, but leave it marked unmerged so the | |
899 | * user needs to confirm the resolution. | |
900 | */ | |
901 | ||
902 | /* Skip the search if makes no sense to the calling context. */ | |
903 | if (!search) | |
904 | return 0; | |
905 | ||
906 | /* find commit which merges them */ | |
907 | parent_count = find_first_merges(opt->repo, path, commit_a, commit_b, | |
908 | &merges); | |
909 | switch (parent_count) { | |
910 | case 0: | |
911 | path_msg(opt, path, 0, _("Failed to merge submodule %s"), path); | |
912 | break; | |
913 | ||
914 | case 1: | |
915 | format_commit(&sb, 4, | |
916 | (struct commit *)merges.objects[0].item); | |
917 | path_msg(opt, path, 0, | |
918 | _("Failed to merge submodule %s, but a possible merge " | |
919 | "resolution exists:\n%s\n"), | |
920 | path, sb.buf); | |
921 | path_msg(opt, path, 1, | |
922 | _("If this is correct simply add it to the index " | |
923 | "for example\n" | |
924 | "by using:\n\n" | |
925 | " git update-index --cacheinfo 160000 %s \"%s\"\n\n" | |
926 | "which will accept this suggestion.\n"), | |
927 | oid_to_hex(&merges.objects[0].item->oid), path); | |
928 | strbuf_release(&sb); | |
929 | break; | |
930 | default: | |
931 | for (i = 0; i < merges.nr; i++) | |
932 | format_commit(&sb, 4, | |
933 | (struct commit *)merges.objects[i].item); | |
934 | path_msg(opt, path, 0, | |
935 | _("Failed to merge submodule %s, but multiple " | |
936 | "possible merges exist:\n%s"), path, sb.buf); | |
937 | strbuf_release(&sb); | |
938 | } | |
939 | ||
940 | object_array_clear(&merges); | |
941 | return 0; | |
62fdec17 EN |
942 | } |
943 | ||
944 | static int merge_3way(struct merge_options *opt, | |
945 | const char *path, | |
946 | const struct object_id *o, | |
947 | const struct object_id *a, | |
948 | const struct object_id *b, | |
949 | const char *pathnames[3], | |
950 | const int extra_marker_size, | |
951 | mmbuffer_t *result_buf) | |
952 | { | |
f591c472 EN |
953 | mmfile_t orig, src1, src2; |
954 | struct ll_merge_options ll_opts = {0}; | |
955 | char *base, *name1, *name2; | |
956 | int merge_status; | |
957 | ||
958 | ll_opts.renormalize = opt->renormalize; | |
959 | ll_opts.extra_marker_size = extra_marker_size; | |
960 | ll_opts.xdl_opts = opt->xdl_opts; | |
961 | ||
962 | if (opt->priv->call_depth) { | |
963 | ll_opts.virtual_ancestor = 1; | |
964 | ll_opts.variant = 0; | |
965 | } else { | |
966 | switch (opt->recursive_variant) { | |
967 | case MERGE_VARIANT_OURS: | |
968 | ll_opts.variant = XDL_MERGE_FAVOR_OURS; | |
969 | break; | |
970 | case MERGE_VARIANT_THEIRS: | |
971 | ll_opts.variant = XDL_MERGE_FAVOR_THEIRS; | |
972 | break; | |
973 | default: | |
974 | ll_opts.variant = 0; | |
975 | break; | |
976 | } | |
977 | } | |
978 | ||
979 | assert(pathnames[0] && pathnames[1] && pathnames[2] && opt->ancestor); | |
980 | if (pathnames[0] == pathnames[1] && pathnames[1] == pathnames[2]) { | |
981 | base = mkpathdup("%s", opt->ancestor); | |
982 | name1 = mkpathdup("%s", opt->branch1); | |
983 | name2 = mkpathdup("%s", opt->branch2); | |
984 | } else { | |
985 | base = mkpathdup("%s:%s", opt->ancestor, pathnames[0]); | |
986 | name1 = mkpathdup("%s:%s", opt->branch1, pathnames[1]); | |
987 | name2 = mkpathdup("%s:%s", opt->branch2, pathnames[2]); | |
988 | } | |
989 | ||
990 | read_mmblob(&orig, o); | |
991 | read_mmblob(&src1, a); | |
992 | read_mmblob(&src2, b); | |
993 | ||
994 | merge_status = ll_merge(result_buf, path, &orig, base, | |
995 | &src1, name1, &src2, name2, | |
996 | opt->repo->index, &ll_opts); | |
997 | ||
998 | free(base); | |
999 | free(name1); | |
1000 | free(name2); | |
1001 | free(orig.ptr); | |
1002 | free(src1.ptr); | |
1003 | free(src2.ptr); | |
1004 | return merge_status; | |
62fdec17 EN |
1005 | } |
1006 | ||
e2e9dc03 EN |
1007 | static int handle_content_merge(struct merge_options *opt, |
1008 | const char *path, | |
1009 | const struct version_info *o, | |
1010 | const struct version_info *a, | |
1011 | const struct version_info *b, | |
1012 | const char *pathnames[3], | |
1013 | const int extra_marker_size, | |
1014 | struct version_info *result) | |
1015 | { | |
991bbdca | 1016 | /* |
62fdec17 EN |
1017 | * path is the target location where we want to put the file, and |
1018 | * is used to determine any normalization rules in ll_merge. | |
1019 | * | |
1020 | * The normal case is that path and all entries in pathnames are | |
1021 | * identical, though renames can affect which path we got one of | |
1022 | * the three blobs to merge on various sides of history. | |
1023 | * | |
1024 | * extra_marker_size is the amount to extend conflict markers in | |
1025 | * ll_merge; this is neeed if we have content merges of content | |
1026 | * merges, which happens for example with rename/rename(2to1) and | |
1027 | * rename/add conflicts. | |
1028 | */ | |
1029 | unsigned clean = 1; | |
1030 | ||
1031 | /* | |
1032 | * handle_content_merge() needs both files to be of the same type, i.e. | |
1033 | * both files OR both submodules OR both symlinks. Conflicting types | |
1034 | * needs to be handled elsewhere. | |
991bbdca | 1035 | */ |
62fdec17 EN |
1036 | assert((S_IFMT & a->mode) == (S_IFMT & b->mode)); |
1037 | ||
1038 | /* Merge modes */ | |
1039 | if (a->mode == b->mode || a->mode == o->mode) | |
1040 | result->mode = b->mode; | |
1041 | else { | |
1042 | /* must be the 100644/100755 case */ | |
1043 | assert(S_ISREG(a->mode)); | |
1044 | result->mode = a->mode; | |
1045 | clean = (b->mode == o->mode); | |
1046 | /* | |
1047 | * FIXME: If opt->priv->call_depth && !clean, then we really | |
1048 | * should not make result->mode match either a->mode or | |
1049 | * b->mode; that causes t6036 "check conflicting mode for | |
1050 | * regular file" to fail. It would be best to use some other | |
1051 | * mode, but we'll confuse all kinds of stuff if we use one | |
1052 | * where S_ISREG(result->mode) isn't true, and if we use | |
1053 | * something like 0100666, then tree-walk.c's calls to | |
1054 | * canon_mode() will just normalize that to 100644 for us and | |
1055 | * thus not solve anything. | |
1056 | * | |
1057 | * Figure out if there's some kind of way we can work around | |
1058 | * this... | |
1059 | */ | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * Trivial oid merge. | |
1064 | * | |
1065 | * Note: While one might assume that the next four lines would | |
1066 | * be unnecessary due to the fact that match_mask is often | |
1067 | * setup and already handled, renames don't always take care | |
1068 | * of that. | |
1069 | */ | |
1070 | if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid)) | |
1071 | oidcpy(&result->oid, &b->oid); | |
1072 | else if (oideq(&b->oid, &o->oid)) | |
1073 | oidcpy(&result->oid, &a->oid); | |
1074 | ||
1075 | /* Remaining rules depend on file vs. submodule vs. symlink. */ | |
1076 | else if (S_ISREG(a->mode)) { | |
1077 | mmbuffer_t result_buf; | |
1078 | int ret = 0, merge_status; | |
1079 | int two_way; | |
1080 | ||
1081 | /* | |
1082 | * If 'o' is different type, treat it as null so we do a | |
1083 | * two-way merge. | |
1084 | */ | |
1085 | two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
1086 | ||
1087 | merge_status = merge_3way(opt, path, | |
1088 | two_way ? &null_oid : &o->oid, | |
1089 | &a->oid, &b->oid, | |
1090 | pathnames, extra_marker_size, | |
1091 | &result_buf); | |
1092 | ||
1093 | if ((merge_status < 0) || !result_buf.ptr) | |
1094 | ret = err(opt, _("Failed to execute internal merge")); | |
1095 | ||
1096 | if (!ret && | |
1097 | write_object_file(result_buf.ptr, result_buf.size, | |
1098 | blob_type, &result->oid)) | |
1099 | ret = err(opt, _("Unable to add %s to database"), | |
1100 | path); | |
1101 | ||
1102 | free(result_buf.ptr); | |
1103 | if (ret) | |
1104 | return -1; | |
1105 | clean &= (merge_status == 0); | |
1106 | path_msg(opt, path, 1, _("Auto-merging %s"), path); | |
1107 | } else if (S_ISGITLINK(a->mode)) { | |
1108 | int two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
1109 | clean = merge_submodule(opt, pathnames[0], | |
1110 | two_way ? &null_oid : &o->oid, | |
1111 | &a->oid, &b->oid, &result->oid); | |
1112 | if (opt->priv->call_depth && two_way && !clean) { | |
1113 | result->mode = o->mode; | |
1114 | oidcpy(&result->oid, &o->oid); | |
1115 | } | |
1116 | } else if (S_ISLNK(a->mode)) { | |
1117 | if (opt->priv->call_depth) { | |
1118 | clean = 0; | |
1119 | result->mode = o->mode; | |
1120 | oidcpy(&result->oid, &o->oid); | |
1121 | } else { | |
1122 | switch (opt->recursive_variant) { | |
1123 | case MERGE_VARIANT_NORMAL: | |
1124 | clean = 0; | |
1125 | oidcpy(&result->oid, &a->oid); | |
1126 | break; | |
1127 | case MERGE_VARIANT_OURS: | |
1128 | oidcpy(&result->oid, &a->oid); | |
1129 | break; | |
1130 | case MERGE_VARIANT_THEIRS: | |
1131 | oidcpy(&result->oid, &b->oid); | |
1132 | break; | |
1133 | } | |
1134 | } | |
1135 | } else | |
1136 | BUG("unsupported object type in the tree: %06o for %s", | |
1137 | a->mode, path); | |
1138 | ||
991bbdca | 1139 | return clean; |
e2e9dc03 EN |
1140 | } |
1141 | ||
04af1879 EN |
1142 | /*** Function Grouping: functions related to detect_and_process_renames(), *** |
1143 | *** which are split into directory and regular rename detection sections. ***/ | |
1144 | ||
1145 | /*** Function Grouping: functions related to directory rename detection ***/ | |
1146 | ||
fa5e06d6 EN |
1147 | struct collision_info { |
1148 | struct string_list source_files; | |
1149 | unsigned reported_already:1; | |
1150 | }; | |
1151 | ||
d9d015df EN |
1152 | /* |
1153 | * Return a new string that replaces the beginning portion (which matches | |
1154 | * rename_info->key), with rename_info->util.new_dir. In perl-speak: | |
1155 | * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/); | |
1156 | * NOTE: | |
1157 | * Caller must ensure that old_path starts with rename_info->key + '/'. | |
1158 | */ | |
1159 | static char *apply_dir_rename(struct strmap_entry *rename_info, | |
1160 | const char *old_path) | |
1161 | { | |
fbcfc0cc EN |
1162 | struct strbuf new_path = STRBUF_INIT; |
1163 | const char *old_dir = rename_info->key; | |
1164 | const char *new_dir = rename_info->value; | |
1165 | int oldlen, newlen, new_dir_len; | |
1166 | ||
1167 | oldlen = strlen(old_dir); | |
1168 | if (*new_dir == '\0') | |
1169 | /* | |
1170 | * If someone renamed/merged a subdirectory into the root | |
1171 | * directory (e.g. 'some/subdir' -> ''), then we want to | |
1172 | * avoid returning | |
1173 | * '' + '/filename' | |
1174 | * as the rename; we need to make old_path + oldlen advance | |
1175 | * past the '/' character. | |
1176 | */ | |
1177 | oldlen++; | |
1178 | new_dir_len = strlen(new_dir); | |
1179 | newlen = new_dir_len + (strlen(old_path) - oldlen) + 1; | |
1180 | strbuf_grow(&new_path, newlen); | |
1181 | strbuf_add(&new_path, new_dir, new_dir_len); | |
1182 | strbuf_addstr(&new_path, &old_path[oldlen]); | |
1183 | ||
1184 | return strbuf_detach(&new_path, NULL); | |
d9d015df EN |
1185 | } |
1186 | ||
bea43365 EN |
1187 | static int path_in_way(struct strmap *paths, const char *path, unsigned side_mask) |
1188 | { | |
1189 | struct merged_info *mi = strmap_get(paths, path); | |
1190 | struct conflict_info *ci; | |
1191 | if (!mi) | |
1192 | return 0; | |
1193 | INITIALIZE_CI(ci, mi); | |
1194 | return mi->clean || (side_mask & (ci->filemask | ci->dirmask)); | |
1195 | } | |
1196 | ||
47325e85 EN |
1197 | /* |
1198 | * See if there is a directory rename for path, and if there are any file | |
1199 | * level conflicts on the given side for the renamed location. If there is | |
1200 | * a rename and there are no conflicts, return the new name. Otherwise, | |
1201 | * return NULL. | |
1202 | */ | |
1203 | static char *handle_path_level_conflicts(struct merge_options *opt, | |
1204 | const char *path, | |
1205 | unsigned side_index, | |
1206 | struct strmap_entry *rename_info, | |
1207 | struct strmap *collisions) | |
1208 | { | |
bea43365 EN |
1209 | char *new_path = NULL; |
1210 | struct collision_info *c_info; | |
1211 | int clean = 1; | |
1212 | struct strbuf collision_paths = STRBUF_INIT; | |
1213 | ||
1214 | /* | |
1215 | * entry has the mapping of old directory name to new directory name | |
1216 | * that we want to apply to path. | |
1217 | */ | |
1218 | new_path = apply_dir_rename(rename_info, path); | |
1219 | if (!new_path) | |
1220 | BUG("Failed to apply directory rename!"); | |
1221 | ||
1222 | /* | |
1223 | * The caller needs to have ensured that it has pre-populated | |
1224 | * collisions with all paths that map to new_path. Do a quick check | |
1225 | * to ensure that's the case. | |
1226 | */ | |
1227 | c_info = strmap_get(collisions, new_path); | |
1228 | if (c_info == NULL) | |
1229 | BUG("c_info is NULL"); | |
1230 | ||
1231 | /* | |
1232 | * Check for one-sided add/add/.../add conflicts, i.e. | |
1233 | * where implicit renames from the other side doing | |
1234 | * directory rename(s) can affect this side of history | |
1235 | * to put multiple paths into the same location. Warn | |
1236 | * and bail on directory renames for such paths. | |
1237 | */ | |
1238 | if (c_info->reported_already) { | |
1239 | clean = 0; | |
1240 | } else if (path_in_way(&opt->priv->paths, new_path, 1 << side_index)) { | |
1241 | c_info->reported_already = 1; | |
1242 | strbuf_add_separated_string_list(&collision_paths, ", ", | |
1243 | &c_info->source_files); | |
1244 | path_msg(opt, new_path, 0, | |
1245 | _("CONFLICT (implicit dir rename): Existing file/dir " | |
1246 | "at %s in the way of implicit directory rename(s) " | |
1247 | "putting the following path(s) there: %s."), | |
1248 | new_path, collision_paths.buf); | |
1249 | clean = 0; | |
1250 | } else if (c_info->source_files.nr > 1) { | |
1251 | c_info->reported_already = 1; | |
1252 | strbuf_add_separated_string_list(&collision_paths, ", ", | |
1253 | &c_info->source_files); | |
1254 | path_msg(opt, new_path, 0, | |
1255 | _("CONFLICT (implicit dir rename): Cannot map more " | |
1256 | "than one path to %s; implicit directory renames " | |
1257 | "tried to put these paths there: %s"), | |
1258 | new_path, collision_paths.buf); | |
1259 | clean = 0; | |
1260 | } | |
1261 | ||
1262 | /* Free memory we no longer need */ | |
1263 | strbuf_release(&collision_paths); | |
1264 | if (!clean && new_path) { | |
1265 | free(new_path); | |
1266 | return NULL; | |
1267 | } | |
1268 | ||
1269 | return new_path; | |
47325e85 EN |
1270 | } |
1271 | ||
203c872c EN |
1272 | static void dirname_munge(char *filename) |
1273 | { | |
1274 | char *slash = strrchr(filename, '/'); | |
1275 | if (!slash) | |
1276 | slash = filename; | |
1277 | *slash = '\0'; | |
1278 | } | |
1279 | ||
2f620a4f EN |
1280 | static void increment_count(struct strmap *dir_rename_count, |
1281 | char *old_dir, | |
1282 | char *new_dir) | |
1283 | { | |
1284 | struct strintmap *counts; | |
1285 | struct strmap_entry *e; | |
1286 | ||
1287 | /* Get the {new_dirs -> counts} mapping using old_dir */ | |
1288 | e = strmap_get_entry(dir_rename_count, old_dir); | |
1289 | if (e) { | |
1290 | counts = e->value; | |
1291 | } else { | |
1292 | counts = xmalloc(sizeof(*counts)); | |
1293 | strintmap_init_with_options(counts, 0, NULL, 1); | |
1294 | strmap_put(dir_rename_count, old_dir, counts); | |
1295 | } | |
1296 | ||
1297 | /* Increment the count for new_dir */ | |
1298 | strintmap_incr(counts, new_dir, 1); | |
1299 | } | |
1300 | ||
203c872c EN |
1301 | static void update_dir_rename_counts(struct strmap *dir_rename_count, |
1302 | struct strset *dirs_removed, | |
1303 | const char *oldname, | |
1304 | const char *newname) | |
1305 | { | |
1306 | char *old_dir = xstrdup(oldname); | |
1307 | char *new_dir = xstrdup(newname); | |
1308 | char new_dir_first_char = new_dir[0]; | |
1309 | int first_time_in_loop = 1; | |
1310 | ||
1311 | while (1) { | |
1312 | dirname_munge(old_dir); | |
1313 | dirname_munge(new_dir); | |
1314 | ||
1315 | /* | |
1316 | * When renaming | |
1317 | * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c" | |
1318 | * then this suggests that both | |
1319 | * a/b/c/d/e/ => a/b/some/thing/else/e/ | |
1320 | * a/b/c/d/ => a/b/some/thing/else/ | |
1321 | * so we want to increment counters for both. We do NOT, | |
1322 | * however, also want to suggest that there was the following | |
1323 | * rename: | |
1324 | * a/b/c/ => a/b/some/thing/ | |
1325 | * so we need to quit at that point. | |
1326 | * | |
1327 | * Note the when first_time_in_loop, we only strip off the | |
1328 | * basename, and we don't care if that's different. | |
1329 | */ | |
1330 | if (!first_time_in_loop) { | |
1331 | char *old_sub_dir = strchr(old_dir, '\0')+1; | |
1332 | char *new_sub_dir = strchr(new_dir, '\0')+1; | |
1333 | if (!*new_dir) { | |
1334 | /* | |
1335 | * Special case when renaming to root directory, | |
1336 | * i.e. when new_dir == "". In this case, we had | |
1337 | * something like | |
1338 | * a/b/subdir => subdir | |
1339 | * and so dirname_munge() sets things up so that | |
1340 | * old_dir = "a/b\0subdir\0" | |
1341 | * new_dir = "\0ubdir\0" | |
1342 | * We didn't have a '/' to overwrite a '\0' onto | |
1343 | * in new_dir, so we have to compare differently. | |
1344 | */ | |
1345 | if (new_dir_first_char != old_sub_dir[0] || | |
1346 | strcmp(old_sub_dir+1, new_sub_dir)) | |
1347 | break; | |
1348 | } else { | |
1349 | if (strcmp(old_sub_dir, new_sub_dir)) | |
1350 | break; | |
1351 | } | |
1352 | } | |
1353 | ||
1354 | if (strset_contains(dirs_removed, old_dir)) | |
1355 | increment_count(dir_rename_count, old_dir, new_dir); | |
1356 | else | |
1357 | break; | |
1358 | ||
1359 | /* If we hit toplevel directory ("") for old or new dir, quit */ | |
1360 | if (!*old_dir || !*new_dir) | |
1361 | break; | |
1362 | ||
1363 | first_time_in_loop = 0; | |
1364 | } | |
1365 | ||
1366 | /* Free resources we don't need anymore */ | |
1367 | free(old_dir); | |
1368 | free(new_dir); | |
1369 | } | |
1370 | ||
04264d40 EN |
1371 | static void compute_rename_counts(struct diff_queue_struct *pairs, |
1372 | struct strmap *dir_rename_count, | |
1373 | struct strset *dirs_removed) | |
1374 | { | |
2f620a4f EN |
1375 | int i; |
1376 | ||
1377 | for (i = 0; i < pairs->nr; ++i) { | |
2f620a4f EN |
1378 | struct diff_filepair *pair = pairs->queue[i]; |
1379 | ||
1380 | /* File not part of directory rename if it wasn't renamed */ | |
1381 | if (pair->status != 'R') | |
1382 | continue; | |
1383 | ||
2f620a4f EN |
1384 | /* |
1385 | * Make dir_rename_count contain a map of a map: | |
1386 | * old_directory -> {new_directory -> count} | |
1387 | * In other words, for every pair look at the directories for | |
1388 | * the old filename and the new filename and count how many | |
1389 | * times that pairing occurs. | |
1390 | */ | |
203c872c EN |
1391 | update_dir_rename_counts(dir_rename_count, dirs_removed, |
1392 | pair->one->path, | |
1393 | pair->two->path); | |
2f620a4f | 1394 | } |
04264d40 EN |
1395 | } |
1396 | ||
112e1112 EN |
1397 | static void get_provisional_directory_renames(struct merge_options *opt, |
1398 | unsigned side, | |
1399 | int *clean) | |
1400 | { | |
04264d40 EN |
1401 | struct hashmap_iter iter; |
1402 | struct strmap_entry *entry; | |
1403 | struct rename_info *renames = &opt->priv->renames; | |
1404 | ||
1405 | compute_rename_counts(&renames->pairs[side], | |
1406 | &renames->dir_rename_count[side], | |
1407 | &renames->dirs_removed[side]); | |
1408 | /* | |
1409 | * Collapse | |
1410 | * dir_rename_count: old_directory -> {new_directory -> count} | |
1411 | * down to | |
1412 | * dir_renames: old_directory -> best_new_directory | |
1413 | * where best_new_directory is the one with the unique highest count. | |
1414 | */ | |
1415 | strmap_for_each_entry(&renames->dir_rename_count[side], &iter, entry) { | |
1416 | const char *source_dir = entry->key; | |
1417 | struct strintmap *counts = entry->value; | |
1418 | struct hashmap_iter count_iter; | |
1419 | struct strmap_entry *count_entry; | |
1420 | int max = 0; | |
1421 | int bad_max = 0; | |
1422 | const char *best = NULL; | |
1423 | ||
1424 | strintmap_for_each_entry(counts, &count_iter, count_entry) { | |
1425 | const char *target_dir = count_entry->key; | |
1426 | intptr_t count = (intptr_t)count_entry->value; | |
1427 | ||
1428 | if (count == max) | |
1429 | bad_max = max; | |
1430 | else if (count > max) { | |
1431 | max = count; | |
1432 | best = target_dir; | |
1433 | } | |
1434 | } | |
1435 | ||
1436 | if (bad_max == max) { | |
1437 | path_msg(opt, source_dir, 0, | |
1438 | _("CONFLICT (directory rename split): " | |
1439 | "Unclear where to rename %s to; it was " | |
1440 | "renamed to multiple other directories, with " | |
1441 | "no destination getting a majority of the " | |
1442 | "files."), | |
1443 | source_dir); | |
5ced7c3d EN |
1444 | /* |
1445 | * We should mark this as unclean IF something attempts | |
1446 | * to use this rename. We do not yet have the logic | |
1447 | * in place to detect if this directory rename is being | |
1448 | * used, and optimizations that reduce the number of | |
1449 | * renames cause this to falsely trigger. For now, | |
1450 | * just disable it, causing t6423 testcase 2a to break. | |
1451 | * We'll later fix the detection, and when we do we | |
1452 | * will re-enable setting *clean to 0 (and thereby fix | |
1453 | * t6423 testcase 2a). | |
1454 | */ | |
1455 | /* *clean = 0; */ | |
04264d40 EN |
1456 | } else { |
1457 | strmap_put(&renames->dir_renames[side], | |
1458 | source_dir, (void*)best); | |
1459 | } | |
1460 | } | |
112e1112 EN |
1461 | } |
1462 | ||
1463 | static void handle_directory_level_conflicts(struct merge_options *opt) | |
1464 | { | |
98d0d081 EN |
1465 | struct hashmap_iter iter; |
1466 | struct strmap_entry *entry; | |
1467 | struct string_list duplicated = STRING_LIST_INIT_NODUP; | |
1468 | struct rename_info *renames = &opt->priv->renames; | |
1469 | struct strmap *side1_dir_renames = &renames->dir_renames[MERGE_SIDE1]; | |
1470 | struct strmap *side2_dir_renames = &renames->dir_renames[MERGE_SIDE2]; | |
1471 | int i; | |
1472 | ||
1473 | strmap_for_each_entry(side1_dir_renames, &iter, entry) { | |
1474 | if (strmap_contains(side2_dir_renames, entry->key)) | |
1475 | string_list_append(&duplicated, entry->key); | |
1476 | } | |
1477 | ||
1478 | for (i = 0; i < duplicated.nr; i++) { | |
1479 | strmap_remove(side1_dir_renames, duplicated.items[i].string, 0); | |
1480 | strmap_remove(side2_dir_renames, duplicated.items[i].string, 0); | |
1481 | } | |
1482 | string_list_clear(&duplicated, 0); | |
112e1112 EN |
1483 | } |
1484 | ||
d9d015df EN |
1485 | static struct strmap_entry *check_dir_renamed(const char *path, |
1486 | struct strmap *dir_renames) | |
1487 | { | |
fbcfc0cc EN |
1488 | char *temp = xstrdup(path); |
1489 | char *end; | |
1490 | struct strmap_entry *e = NULL; | |
1491 | ||
1492 | while ((end = strrchr(temp, '/'))) { | |
1493 | *end = '\0'; | |
1494 | e = strmap_get_entry(dir_renames, temp); | |
1495 | if (e) | |
1496 | break; | |
1497 | } | |
1498 | free(temp); | |
1499 | return e; | |
d9d015df EN |
1500 | } |
1501 | ||
fa5e06d6 EN |
1502 | static void compute_collisions(struct strmap *collisions, |
1503 | struct strmap *dir_renames, | |
1504 | struct diff_queue_struct *pairs) | |
1505 | { | |
d9d015df EN |
1506 | int i; |
1507 | ||
1508 | strmap_init_with_options(collisions, NULL, 0); | |
1509 | if (strmap_empty(dir_renames)) | |
1510 | return; | |
1511 | ||
1512 | /* | |
1513 | * Multiple files can be mapped to the same path due to directory | |
1514 | * renames done by the other side of history. Since that other | |
1515 | * side of history could have merged multiple directories into one, | |
1516 | * if our side of history added the same file basename to each of | |
1517 | * those directories, then all N of them would get implicitly | |
1518 | * renamed by the directory rename detection into the same path, | |
1519 | * and we'd get an add/add/.../add conflict, and all those adds | |
1520 | * from *this* side of history. This is not representable in the | |
1521 | * index, and users aren't going to easily be able to make sense of | |
1522 | * it. So we need to provide a good warning about what's | |
1523 | * happening, and fall back to no-directory-rename detection | |
1524 | * behavior for those paths. | |
1525 | * | |
1526 | * See testcases 9e and all of section 5 from t6043 for examples. | |
1527 | */ | |
1528 | for (i = 0; i < pairs->nr; ++i) { | |
1529 | struct strmap_entry *rename_info; | |
1530 | struct collision_info *collision_info; | |
1531 | char *new_path; | |
1532 | struct diff_filepair *pair = pairs->queue[i]; | |
1533 | ||
1534 | if (pair->status != 'A' && pair->status != 'R') | |
1535 | continue; | |
1536 | rename_info = check_dir_renamed(pair->two->path, dir_renames); | |
1537 | if (!rename_info) | |
1538 | continue; | |
1539 | ||
1540 | new_path = apply_dir_rename(rename_info, pair->two->path); | |
1541 | assert(new_path); | |
1542 | collision_info = strmap_get(collisions, new_path); | |
1543 | if (collision_info) { | |
1544 | free(new_path); | |
1545 | } else { | |
1546 | collision_info = xcalloc(1, | |
1547 | sizeof(struct collision_info)); | |
1548 | string_list_init(&collision_info->source_files, 0); | |
1549 | strmap_put(collisions, new_path, collision_info); | |
1550 | } | |
1551 | string_list_insert(&collision_info->source_files, | |
1552 | pair->two->path); | |
1553 | } | |
fa5e06d6 EN |
1554 | } |
1555 | ||
1556 | static char *check_for_directory_rename(struct merge_options *opt, | |
1557 | const char *path, | |
1558 | unsigned side_index, | |
1559 | struct strmap *dir_renames, | |
1560 | struct strmap *dir_rename_exclusions, | |
1561 | struct strmap *collisions, | |
1562 | int *clean_merge) | |
1563 | { | |
47325e85 EN |
1564 | char *new_path = NULL; |
1565 | struct strmap_entry *rename_info; | |
1566 | struct strmap_entry *otherinfo = NULL; | |
1567 | const char *new_dir; | |
1568 | ||
1569 | if (strmap_empty(dir_renames)) | |
1570 | return new_path; | |
1571 | rename_info = check_dir_renamed(path, dir_renames); | |
1572 | if (!rename_info) | |
1573 | return new_path; | |
1574 | /* old_dir = rename_info->key; */ | |
1575 | new_dir = rename_info->value; | |
1576 | ||
1577 | /* | |
1578 | * This next part is a little weird. We do not want to do an | |
1579 | * implicit rename into a directory we renamed on our side, because | |
1580 | * that will result in a spurious rename/rename(1to2) conflict. An | |
1581 | * example: | |
1582 | * Base commit: dumbdir/afile, otherdir/bfile | |
1583 | * Side 1: smrtdir/afile, otherdir/bfile | |
1584 | * Side 2: dumbdir/afile, dumbdir/bfile | |
1585 | * Here, while working on Side 1, we could notice that otherdir was | |
1586 | * renamed/merged to dumbdir, and change the diff_filepair for | |
1587 | * otherdir/bfile into a rename into dumbdir/bfile. However, Side | |
1588 | * 2 will notice the rename from dumbdir to smrtdir, and do the | |
1589 | * transitive rename to move it from dumbdir/bfile to | |
1590 | * smrtdir/bfile. That gives us bfile in dumbdir vs being in | |
1591 | * smrtdir, a rename/rename(1to2) conflict. We really just want | |
1592 | * the file to end up in smrtdir. And the way to achieve that is | |
1593 | * to not let Side1 do the rename to dumbdir, since we know that is | |
1594 | * the source of one of our directory renames. | |
1595 | * | |
1596 | * That's why otherinfo and dir_rename_exclusions is here. | |
1597 | * | |
1598 | * As it turns out, this also prevents N-way transient rename | |
1599 | * confusion; See testcases 9c and 9d of t6043. | |
1600 | */ | |
1601 | otherinfo = strmap_get_entry(dir_rename_exclusions, new_dir); | |
1602 | if (otherinfo) { | |
1603 | path_msg(opt, rename_info->key, 1, | |
1604 | _("WARNING: Avoiding applying %s -> %s rename " | |
1605 | "to %s, because %s itself was renamed."), | |
1606 | rename_info->key, new_dir, path, new_dir); | |
1607 | return NULL; | |
1608 | } | |
1609 | ||
1610 | new_path = handle_path_level_conflicts(opt, path, side_index, | |
1611 | rename_info, collisions); | |
1612 | *clean_merge &= (new_path != NULL); | |
1613 | ||
1614 | return new_path; | |
fa5e06d6 EN |
1615 | } |
1616 | ||
1617 | static void apply_directory_rename_modifications(struct merge_options *opt, | |
1618 | struct diff_filepair *pair, | |
1619 | char *new_path) | |
1620 | { | |
089d82bc EN |
1621 | /* |
1622 | * The basic idea is to get the conflict_info from opt->priv->paths | |
1623 | * at old path, and insert it into new_path; basically just this: | |
1624 | * ci = strmap_get(&opt->priv->paths, old_path); | |
1625 | * strmap_remove(&opt->priv->paths, old_path, 0); | |
1626 | * strmap_put(&opt->priv->paths, new_path, ci); | |
1627 | * However, there are some factors complicating this: | |
1628 | * - opt->priv->paths may already have an entry at new_path | |
1629 | * - Each ci tracks its containing directory, so we need to | |
1630 | * update that | |
1631 | * - If another ci has the same containing directory, then | |
1632 | * the two char*'s MUST point to the same location. See the | |
1633 | * comment in struct merged_info. strcmp equality is not | |
1634 | * enough; we need pointer equality. | |
1635 | * - opt->priv->paths must hold the parent directories of any | |
1636 | * entries that are added. So, if this directory rename | |
1637 | * causes entirely new directories, we must recursively add | |
1638 | * parent directories. | |
1639 | * - For each parent directory added to opt->priv->paths, we | |
1640 | * also need to get its parent directory stored in its | |
1641 | * conflict_info->merged.directory_name with all the same | |
1642 | * requirements about pointer equality. | |
1643 | */ | |
1644 | struct string_list dirs_to_insert = STRING_LIST_INIT_NODUP; | |
1645 | struct conflict_info *ci, *new_ci; | |
1646 | struct strmap_entry *entry; | |
1647 | const char *branch_with_new_path, *branch_with_dir_rename; | |
1648 | const char *old_path = pair->two->path; | |
1649 | const char *parent_name; | |
1650 | const char *cur_path; | |
1651 | int i, len; | |
1652 | ||
1653 | entry = strmap_get_entry(&opt->priv->paths, old_path); | |
1654 | old_path = entry->key; | |
1655 | ci = entry->value; | |
1656 | VERIFY_CI(ci); | |
1657 | ||
1658 | /* Find parent directories missing from opt->priv->paths */ | |
1659 | cur_path = new_path; | |
1660 | while (1) { | |
1661 | /* Find the parent directory of cur_path */ | |
1662 | char *last_slash = strrchr(cur_path, '/'); | |
1663 | if (last_slash) { | |
1664 | parent_name = xstrndup(cur_path, last_slash - cur_path); | |
1665 | } else { | |
1666 | parent_name = opt->priv->toplevel_dir; | |
1667 | break; | |
1668 | } | |
1669 | ||
1670 | /* Look it up in opt->priv->paths */ | |
1671 | entry = strmap_get_entry(&opt->priv->paths, parent_name); | |
1672 | if (entry) { | |
1673 | free((char*)parent_name); | |
1674 | parent_name = entry->key; /* reuse known pointer */ | |
1675 | break; | |
1676 | } | |
1677 | ||
1678 | /* Record this is one of the directories we need to insert */ | |
1679 | string_list_append(&dirs_to_insert, parent_name); | |
1680 | cur_path = parent_name; | |
1681 | } | |
1682 | ||
1683 | /* Traverse dirs_to_insert and insert them into opt->priv->paths */ | |
1684 | for (i = dirs_to_insert.nr-1; i >= 0; --i) { | |
1685 | struct conflict_info *dir_ci; | |
1686 | char *cur_dir = dirs_to_insert.items[i].string; | |
1687 | ||
1688 | dir_ci = xcalloc(1, sizeof(*dir_ci)); | |
1689 | ||
1690 | dir_ci->merged.directory_name = parent_name; | |
1691 | len = strlen(parent_name); | |
1692 | /* len+1 because of trailing '/' character */ | |
1693 | dir_ci->merged.basename_offset = (len > 0 ? len+1 : len); | |
1694 | dir_ci->dirmask = ci->filemask; | |
1695 | strmap_put(&opt->priv->paths, cur_dir, dir_ci); | |
1696 | ||
1697 | parent_name = cur_dir; | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * We are removing old_path from opt->priv->paths. old_path also will | |
1702 | * eventually need to be freed, but it may still be used by e.g. | |
1703 | * ci->pathnames. So, store it in another string-list for now. | |
1704 | */ | |
1705 | string_list_append(&opt->priv->paths_to_free, old_path); | |
1706 | ||
1707 | assert(ci->filemask == 2 || ci->filemask == 4); | |
1708 | assert(ci->dirmask == 0); | |
1709 | strmap_remove(&opt->priv->paths, old_path, 0); | |
1710 | ||
1711 | branch_with_new_path = (ci->filemask == 2) ? opt->branch1 : opt->branch2; | |
1712 | branch_with_dir_rename = (ci->filemask == 2) ? opt->branch2 : opt->branch1; | |
1713 | ||
1714 | /* Now, finally update ci and stick it into opt->priv->paths */ | |
1715 | ci->merged.directory_name = parent_name; | |
1716 | len = strlen(parent_name); | |
1717 | ci->merged.basename_offset = (len > 0 ? len+1 : len); | |
1718 | new_ci = strmap_get(&opt->priv->paths, new_path); | |
1719 | if (!new_ci) { | |
1720 | /* Place ci back into opt->priv->paths, but at new_path */ | |
1721 | strmap_put(&opt->priv->paths, new_path, ci); | |
1722 | } else { | |
1723 | int index; | |
1724 | ||
1725 | /* A few sanity checks */ | |
1726 | VERIFY_CI(new_ci); | |
1727 | assert(ci->filemask == 2 || ci->filemask == 4); | |
1728 | assert((new_ci->filemask & ci->filemask) == 0); | |
1729 | assert(!new_ci->merged.clean); | |
1730 | ||
1731 | /* Copy stuff from ci into new_ci */ | |
1732 | new_ci->filemask |= ci->filemask; | |
1733 | if (new_ci->dirmask) | |
1734 | new_ci->df_conflict = 1; | |
1735 | index = (ci->filemask >> 1); | |
1736 | new_ci->pathnames[index] = ci->pathnames[index]; | |
1737 | new_ci->stages[index].mode = ci->stages[index].mode; | |
1738 | oidcpy(&new_ci->stages[index].oid, &ci->stages[index].oid); | |
1739 | ||
1740 | free(ci); | |
1741 | ci = new_ci; | |
1742 | } | |
1743 | ||
1744 | if (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) { | |
1745 | /* Notify user of updated path */ | |
1746 | if (pair->status == 'A') | |
1747 | path_msg(opt, new_path, 1, | |
1748 | _("Path updated: %s added in %s inside a " | |
1749 | "directory that was renamed in %s; moving " | |
1750 | "it to %s."), | |
1751 | old_path, branch_with_new_path, | |
1752 | branch_with_dir_rename, new_path); | |
1753 | else | |
1754 | path_msg(opt, new_path, 1, | |
1755 | _("Path updated: %s renamed to %s in %s, " | |
1756 | "inside a directory that was renamed in %s; " | |
1757 | "moving it to %s."), | |
1758 | pair->one->path, old_path, branch_with_new_path, | |
1759 | branch_with_dir_rename, new_path); | |
1760 | } else { | |
1761 | /* | |
1762 | * opt->detect_directory_renames has the value | |
1763 | * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts. | |
1764 | */ | |
1765 | ci->path_conflict = 1; | |
1766 | if (pair->status == 'A') | |
1767 | path_msg(opt, new_path, 0, | |
1768 | _("CONFLICT (file location): %s added in %s " | |
1769 | "inside a directory that was renamed in %s, " | |
1770 | "suggesting it should perhaps be moved to " | |
1771 | "%s."), | |
1772 | old_path, branch_with_new_path, | |
1773 | branch_with_dir_rename, new_path); | |
1774 | else | |
1775 | path_msg(opt, new_path, 0, | |
1776 | _("CONFLICT (file location): %s renamed to %s " | |
1777 | "in %s, inside a directory that was renamed " | |
1778 | "in %s, suggesting it should perhaps be " | |
1779 | "moved to %s."), | |
1780 | pair->one->path, old_path, branch_with_new_path, | |
1781 | branch_with_dir_rename, new_path); | |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * Finally, record the new location. | |
1786 | */ | |
1787 | pair->two->path = new_path; | |
fa5e06d6 EN |
1788 | } |
1789 | ||
04af1879 EN |
1790 | /*** Function Grouping: functions related to regular rename detection ***/ |
1791 | ||
e1a124e8 EN |
1792 | static int process_renames(struct merge_options *opt, |
1793 | struct diff_queue_struct *renames) | |
1794 | { | |
c2d267df EN |
1795 | int clean_merge = 1, i; |
1796 | ||
1797 | for (i = 0; i < renames->nr; ++i) { | |
1798 | const char *oldpath = NULL, *newpath; | |
1799 | struct diff_filepair *pair = renames->queue[i]; | |
1800 | struct conflict_info *oldinfo = NULL, *newinfo = NULL; | |
1801 | struct strmap_entry *old_ent, *new_ent; | |
1802 | unsigned int old_sidemask; | |
1803 | int target_index, other_source_index; | |
1804 | int source_deleted, collision, type_changed; | |
2e91ddd2 | 1805 | const char *rename_branch = NULL, *delete_branch = NULL; |
c2d267df EN |
1806 | |
1807 | old_ent = strmap_get_entry(&opt->priv->paths, pair->one->path); | |
c2d267df | 1808 | new_ent = strmap_get_entry(&opt->priv->paths, pair->two->path); |
1b6b902d EN |
1809 | if (old_ent) { |
1810 | oldpath = old_ent->key; | |
1811 | oldinfo = old_ent->value; | |
1812 | } | |
1813 | newpath = pair->two->path; | |
1814 | if (new_ent) { | |
1815 | newpath = new_ent->key; | |
1816 | newinfo = new_ent->value; | |
1817 | } | |
1818 | ||
1819 | /* | |
1820 | * If pair->one->path isn't in opt->priv->paths, that means | |
1821 | * that either directory rename detection removed that | |
1822 | * path, or a parent directory of oldpath was resolved and | |
1823 | * we don't even need the rename; in either case, we can | |
1824 | * skip it. If oldinfo->merged.clean, then the other side | |
1825 | * of history had no changes to oldpath and we don't need | |
1826 | * the rename and can skip it. | |
1827 | */ | |
1828 | if (!oldinfo || oldinfo->merged.clean) | |
1829 | continue; | |
c2d267df EN |
1830 | |
1831 | /* | |
1832 | * diff_filepairs have copies of pathnames, thus we have to | |
1833 | * use standard 'strcmp()' (negated) instead of '=='. | |
1834 | */ | |
1835 | if (i + 1 < renames->nr && | |
1836 | !strcmp(oldpath, renames->queue[i+1]->one->path)) { | |
1837 | /* Handle rename/rename(1to2) or rename/rename(1to1) */ | |
1838 | const char *pathnames[3]; | |
af1e56c4 EN |
1839 | struct version_info merged; |
1840 | struct conflict_info *base, *side1, *side2; | |
53e88a03 | 1841 | unsigned was_binary_blob = 0; |
c2d267df EN |
1842 | |
1843 | pathnames[0] = oldpath; | |
1844 | pathnames[1] = newpath; | |
1845 | pathnames[2] = renames->queue[i+1]->two->path; | |
1846 | ||
af1e56c4 EN |
1847 | base = strmap_get(&opt->priv->paths, pathnames[0]); |
1848 | side1 = strmap_get(&opt->priv->paths, pathnames[1]); | |
1849 | side2 = strmap_get(&opt->priv->paths, pathnames[2]); | |
1850 | ||
1851 | VERIFY_CI(base); | |
1852 | VERIFY_CI(side1); | |
1853 | VERIFY_CI(side2); | |
1854 | ||
c2d267df | 1855 | if (!strcmp(pathnames[1], pathnames[2])) { |
af1e56c4 EN |
1856 | /* Both sides renamed the same way */ |
1857 | assert(side1 == side2); | |
1858 | memcpy(&side1->stages[0], &base->stages[0], | |
1859 | sizeof(merged)); | |
1860 | side1->filemask |= (1 << MERGE_BASE); | |
1861 | /* Mark base as resolved by removal */ | |
1862 | base->merged.is_null = 1; | |
1863 | base->merged.clean = 1; | |
c2d267df EN |
1864 | |
1865 | /* We handled both renames, i.e. i+1 handled */ | |
1866 | i++; | |
1867 | /* Move to next rename */ | |
1868 | continue; | |
1869 | } | |
1870 | ||
1871 | /* This is a rename/rename(1to2) */ | |
53e88a03 EN |
1872 | clean_merge = handle_content_merge(opt, |
1873 | pair->one->path, | |
1874 | &base->stages[0], | |
1875 | &side1->stages[1], | |
1876 | &side2->stages[2], | |
1877 | pathnames, | |
1878 | 1 + 2 * opt->priv->call_depth, | |
1879 | &merged); | |
1880 | if (!clean_merge && | |
1881 | merged.mode == side1->stages[1].mode && | |
1882 | oideq(&merged.oid, &side1->stages[1].oid)) | |
1883 | was_binary_blob = 1; | |
1884 | memcpy(&side1->stages[1], &merged, sizeof(merged)); | |
1885 | if (was_binary_blob) { | |
1886 | /* | |
1887 | * Getting here means we were attempting to | |
1888 | * merge a binary blob. | |
1889 | * | |
1890 | * Since we can't merge binaries, | |
1891 | * handle_content_merge() just takes one | |
1892 | * side. But we don't want to copy the | |
1893 | * contents of one side to both paths. We | |
1894 | * used the contents of side1 above for | |
1895 | * side1->stages, let's use the contents of | |
1896 | * side2 for side2->stages below. | |
1897 | */ | |
1898 | oidcpy(&merged.oid, &side2->stages[2].oid); | |
1899 | merged.mode = side2->stages[2].mode; | |
1900 | } | |
1901 | memcpy(&side2->stages[2], &merged, sizeof(merged)); | |
1902 | ||
1903 | side1->path_conflict = 1; | |
1904 | side2->path_conflict = 1; | |
1905 | /* | |
1906 | * TODO: For renames we normally remove the path at the | |
1907 | * old name. It would thus seem consistent to do the | |
1908 | * same for rename/rename(1to2) cases, but we haven't | |
1909 | * done so traditionally and a number of the regression | |
1910 | * tests now encode an expectation that the file is | |
1911 | * left there at stage 1. If we ever decide to change | |
1912 | * this, add the following two lines here: | |
1913 | * base->merged.is_null = 1; | |
1914 | * base->merged.clean = 1; | |
1915 | * and remove the setting of base->path_conflict to 1. | |
1916 | */ | |
1917 | base->path_conflict = 1; | |
1918 | path_msg(opt, oldpath, 0, | |
1919 | _("CONFLICT (rename/rename): %s renamed to " | |
1920 | "%s in %s and to %s in %s."), | |
1921 | pathnames[0], | |
1922 | pathnames[1], opt->branch1, | |
1923 | pathnames[2], opt->branch2); | |
c2d267df EN |
1924 | |
1925 | i++; /* We handled both renames, i.e. i+1 handled */ | |
1926 | continue; | |
1927 | } | |
1928 | ||
1929 | VERIFY_CI(oldinfo); | |
1930 | VERIFY_CI(newinfo); | |
1931 | target_index = pair->score; /* from collect_renames() */ | |
1932 | assert(target_index == 1 || target_index == 2); | |
1933 | other_source_index = 3 - target_index; | |
1934 | old_sidemask = (1 << other_source_index); /* 2 or 4 */ | |
1935 | source_deleted = (oldinfo->filemask == 1); | |
1936 | collision = ((newinfo->filemask & old_sidemask) != 0); | |
1937 | type_changed = !source_deleted && | |
1938 | (S_ISREG(oldinfo->stages[other_source_index].mode) != | |
1939 | S_ISREG(newinfo->stages[target_index].mode)); | |
1940 | if (type_changed && collision) { | |
6fcccbd7 EN |
1941 | /* |
1942 | * special handling so later blocks can handle this... | |
1943 | * | |
1944 | * if type_changed && collision are both true, then this | |
1945 | * was really a double rename, but one side wasn't | |
1946 | * detected due to lack of break detection. I.e. | |
1947 | * something like | |
1948 | * orig: has normal file 'foo' | |
1949 | * side1: renames 'foo' to 'bar', adds 'foo' symlink | |
1950 | * side2: renames 'foo' to 'bar' | |
1951 | * In this case, the foo->bar rename on side1 won't be | |
1952 | * detected because the new symlink named 'foo' is | |
1953 | * there and we don't do break detection. But we detect | |
1954 | * this here because we don't want to merge the content | |
1955 | * of the foo symlink with the foo->bar file, so we | |
1956 | * have some logic to handle this special case. The | |
1957 | * easiest way to do that is make 'bar' on side1 not | |
1958 | * be considered a colliding file but the other part | |
1959 | * of a normal rename. If the file is very different, | |
1960 | * well we're going to get content merge conflicts | |
1961 | * anyway so it doesn't hurt. And if the colliding | |
1962 | * file also has a different type, that'll be handled | |
1963 | * by the content merge logic in process_entry() too. | |
1964 | * | |
1965 | * See also t6430, 'rename vs. rename/symlink' | |
1966 | */ | |
1967 | collision = 0; | |
c2d267df | 1968 | } |
2e91ddd2 EN |
1969 | if (source_deleted) { |
1970 | if (target_index == 1) { | |
1971 | rename_branch = opt->branch1; | |
1972 | delete_branch = opt->branch2; | |
1973 | } else { | |
1974 | rename_branch = opt->branch2; | |
1975 | delete_branch = opt->branch1; | |
1976 | } | |
1977 | } | |
c2d267df EN |
1978 | |
1979 | assert(source_deleted || oldinfo->filemask & old_sidemask); | |
1980 | ||
1981 | /* Need to check for special types of rename conflicts... */ | |
1982 | if (collision && !source_deleted) { | |
1983 | /* collision: rename/add or rename/rename(2to1) */ | |
35e47e35 EN |
1984 | const char *pathnames[3]; |
1985 | struct version_info merged; | |
1986 | ||
1987 | struct conflict_info *base, *side1, *side2; | |
1988 | unsigned clean; | |
1989 | ||
1990 | pathnames[0] = oldpath; | |
1991 | pathnames[other_source_index] = oldpath; | |
1992 | pathnames[target_index] = newpath; | |
1993 | ||
1994 | base = strmap_get(&opt->priv->paths, pathnames[0]); | |
1995 | side1 = strmap_get(&opt->priv->paths, pathnames[1]); | |
1996 | side2 = strmap_get(&opt->priv->paths, pathnames[2]); | |
1997 | ||
1998 | VERIFY_CI(base); | |
1999 | VERIFY_CI(side1); | |
2000 | VERIFY_CI(side2); | |
2001 | ||
2002 | clean = handle_content_merge(opt, pair->one->path, | |
2003 | &base->stages[0], | |
2004 | &side1->stages[1], | |
2005 | &side2->stages[2], | |
2006 | pathnames, | |
2007 | 1 + 2 * opt->priv->call_depth, | |
2008 | &merged); | |
2009 | ||
2010 | memcpy(&newinfo->stages[target_index], &merged, | |
2011 | sizeof(merged)); | |
2012 | if (!clean) { | |
2013 | path_msg(opt, newpath, 0, | |
2014 | _("CONFLICT (rename involved in " | |
2015 | "collision): rename of %s -> %s has " | |
2016 | "content conflicts AND collides " | |
2017 | "with another path; this may result " | |
2018 | "in nested conflict markers."), | |
2019 | oldpath, newpath); | |
2020 | } | |
c2d267df | 2021 | } else if (collision && source_deleted) { |
35e47e35 EN |
2022 | /* |
2023 | * rename/add/delete or rename/rename(2to1)/delete: | |
2024 | * since oldpath was deleted on the side that didn't | |
2025 | * do the rename, there's not much of a content merge | |
2026 | * we can do for the rename. oldinfo->merged.is_null | |
2027 | * was already set, so we just leave things as-is so | |
2028 | * they look like an add/add conflict. | |
2029 | */ | |
2030 | ||
2031 | newinfo->path_conflict = 1; | |
2032 | path_msg(opt, newpath, 0, | |
2033 | _("CONFLICT (rename/delete): %s renamed " | |
2034 | "to %s in %s, but deleted in %s."), | |
2035 | oldpath, newpath, rename_branch, delete_branch); | |
c2d267df | 2036 | } else { |
2e91ddd2 EN |
2037 | /* |
2038 | * a few different cases...start by copying the | |
2039 | * existing stage(s) from oldinfo over the newinfo | |
2040 | * and update the pathname(s). | |
2041 | */ | |
2042 | memcpy(&newinfo->stages[0], &oldinfo->stages[0], | |
2043 | sizeof(newinfo->stages[0])); | |
2044 | newinfo->filemask |= (1 << MERGE_BASE); | |
2045 | newinfo->pathnames[0] = oldpath; | |
c2d267df EN |
2046 | if (type_changed) { |
2047 | /* rename vs. typechange */ | |
6fcccbd7 EN |
2048 | /* Mark the original as resolved by removal */ |
2049 | memcpy(&oldinfo->stages[0].oid, &null_oid, | |
2050 | sizeof(oldinfo->stages[0].oid)); | |
2051 | oldinfo->stages[0].mode = 0; | |
2052 | oldinfo->filemask &= 0x06; | |
c2d267df EN |
2053 | } else if (source_deleted) { |
2054 | /* rename/delete */ | |
2e91ddd2 EN |
2055 | newinfo->path_conflict = 1; |
2056 | path_msg(opt, newpath, 0, | |
2057 | _("CONFLICT (rename/delete): %s renamed" | |
2058 | " to %s in %s, but deleted in %s."), | |
2059 | oldpath, newpath, | |
2060 | rename_branch, delete_branch); | |
c2d267df EN |
2061 | } else { |
2062 | /* normal rename */ | |
f1665e69 EN |
2063 | memcpy(&newinfo->stages[other_source_index], |
2064 | &oldinfo->stages[other_source_index], | |
2065 | sizeof(newinfo->stages[0])); | |
2066 | newinfo->filemask |= (1 << other_source_index); | |
2067 | newinfo->pathnames[other_source_index] = oldpath; | |
c2d267df EN |
2068 | } |
2069 | } | |
2070 | ||
2071 | if (!type_changed) { | |
2072 | /* Mark the original as resolved by removal */ | |
2073 | oldinfo->merged.is_null = 1; | |
2074 | oldinfo->merged.clean = 1; | |
2075 | } | |
2076 | ||
2077 | } | |
2078 | ||
2079 | return clean_merge; | |
e1a124e8 EN |
2080 | } |
2081 | ||
2082 | static int compare_pairs(const void *a_, const void *b_) | |
2083 | { | |
965a7bc2 EN |
2084 | const struct diff_filepair *a = *((const struct diff_filepair **)a_); |
2085 | const struct diff_filepair *b = *((const struct diff_filepair **)b_); | |
2086 | ||
2087 | return strcmp(a->one->path, b->one->path); | |
e1a124e8 EN |
2088 | } |
2089 | ||
2090 | /* Call diffcore_rename() to compute which files have changed on given side */ | |
2091 | static void detect_regular_renames(struct merge_options *opt, | |
2092 | struct tree *merge_base, | |
2093 | struct tree *side, | |
2094 | unsigned side_index) | |
2095 | { | |
f39d05ca EN |
2096 | struct diff_options diff_opts; |
2097 | struct rename_info *renames = &opt->priv->renames; | |
2098 | ||
2099 | repo_diff_setup(opt->repo, &diff_opts); | |
2100 | diff_opts.flags.recursive = 1; | |
2101 | diff_opts.flags.rename_empty = 0; | |
2102 | diff_opts.detect_rename = DIFF_DETECT_RENAME; | |
2103 | diff_opts.rename_limit = opt->rename_limit; | |
2104 | if (opt->rename_limit <= 0) | |
2105 | diff_opts.rename_limit = 1000; | |
2106 | diff_opts.rename_score = opt->rename_score; | |
2107 | diff_opts.show_rename_progress = opt->show_rename_progress; | |
2108 | diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT; | |
2109 | diff_setup_done(&diff_opts); | |
557ac035 EN |
2110 | |
2111 | trace2_region_enter("diff", "diffcore_rename", opt->repo); | |
f39d05ca EN |
2112 | diff_tree_oid(&merge_base->object.oid, &side->object.oid, "", |
2113 | &diff_opts); | |
2114 | diffcore_std(&diff_opts); | |
557ac035 | 2115 | trace2_region_leave("diff", "diffcore_rename", opt->repo); |
f39d05ca EN |
2116 | |
2117 | if (diff_opts.needed_rename_limit > renames->needed_limit) | |
2118 | renames->needed_limit = diff_opts.needed_rename_limit; | |
2119 | ||
2120 | renames->pairs[side_index] = diff_queued_diff; | |
2121 | ||
2122 | diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT; | |
2123 | diff_queued_diff.nr = 0; | |
2124 | diff_queued_diff.queue = NULL; | |
2125 | diff_flush(&diff_opts); | |
e1a124e8 EN |
2126 | } |
2127 | ||
2128 | /* | |
2129 | * Get information of all renames which occurred in 'side_pairs', discarding | |
2130 | * non-renames. | |
2131 | */ | |
2132 | static int collect_renames(struct merge_options *opt, | |
2133 | struct diff_queue_struct *result, | |
fa5e06d6 EN |
2134 | unsigned side_index, |
2135 | struct strmap *dir_renames_for_side, | |
2136 | struct strmap *rename_exclusions) | |
e1a124e8 | 2137 | { |
965a7bc2 | 2138 | int i, clean = 1; |
fa5e06d6 | 2139 | struct strmap collisions; |
965a7bc2 | 2140 | struct diff_queue_struct *side_pairs; |
fa5e06d6 EN |
2141 | struct hashmap_iter iter; |
2142 | struct strmap_entry *entry; | |
965a7bc2 EN |
2143 | struct rename_info *renames = &opt->priv->renames; |
2144 | ||
2145 | side_pairs = &renames->pairs[side_index]; | |
fa5e06d6 | 2146 | compute_collisions(&collisions, dir_renames_for_side, side_pairs); |
965a7bc2 EN |
2147 | |
2148 | for (i = 0; i < side_pairs->nr; ++i) { | |
2149 | struct diff_filepair *p = side_pairs->queue[i]; | |
fa5e06d6 | 2150 | char *new_path; /* non-NULL only with directory renames */ |
965a7bc2 | 2151 | |
fa5e06d6 | 2152 | if (p->status != 'A' && p->status != 'R') { |
965a7bc2 EN |
2153 | diff_free_filepair(p); |
2154 | continue; | |
2155 | } | |
2156 | ||
fa5e06d6 EN |
2157 | new_path = check_for_directory_rename(opt, p->two->path, |
2158 | side_index, | |
2159 | dir_renames_for_side, | |
2160 | rename_exclusions, | |
2161 | &collisions, | |
2162 | &clean); | |
2163 | ||
2164 | if (p->status != 'R' && !new_path) { | |
2165 | diff_free_filepair(p); | |
2166 | continue; | |
2167 | } | |
2168 | ||
2169 | if (new_path) | |
2170 | apply_directory_rename_modifications(opt, p, new_path); | |
2171 | ||
965a7bc2 EN |
2172 | /* |
2173 | * p->score comes back from diffcore_rename_extended() with | |
2174 | * the similarity of the renamed file. The similarity is | |
2175 | * was used to determine that the two files were related | |
2176 | * and are a rename, which we have already used, but beyond | |
2177 | * that we have no use for the similarity. So p->score is | |
2178 | * now irrelevant. However, process_renames() will need to | |
2179 | * know which side of the merge this rename was associated | |
2180 | * with, so overwrite p->score with that value. | |
2181 | */ | |
2182 | p->score = side_index; | |
2183 | result->queue[result->nr++] = p; | |
2184 | } | |
2185 | ||
fa5e06d6 EN |
2186 | /* Free each value in the collisions map */ |
2187 | strmap_for_each_entry(&collisions, &iter, entry) { | |
2188 | struct collision_info *info = entry->value; | |
2189 | string_list_clear(&info->source_files, 0); | |
2190 | } | |
2191 | /* | |
2192 | * In compute_collisions(), we set collisions.strdup_strings to 0 | |
2193 | * so that we wouldn't have to make another copy of the new_path | |
2194 | * allocated by apply_dir_rename(). But now that we've used them | |
2195 | * and have no other references to these strings, it is time to | |
2196 | * deallocate them. | |
2197 | */ | |
2198 | free_strmap_strings(&collisions); | |
2199 | strmap_clear(&collisions, 1); | |
965a7bc2 | 2200 | return clean; |
e1a124e8 EN |
2201 | } |
2202 | ||
231e2dd4 EN |
2203 | static int detect_and_process_renames(struct merge_options *opt, |
2204 | struct tree *merge_base, | |
2205 | struct tree *side1, | |
2206 | struct tree *side2) | |
2207 | { | |
e1a124e8 EN |
2208 | struct diff_queue_struct combined; |
2209 | struct rename_info *renames = &opt->priv->renames; | |
112e1112 | 2210 | int need_dir_renames, s, clean = 1; |
e1a124e8 EN |
2211 | |
2212 | memset(&combined, 0, sizeof(combined)); | |
2213 | ||
557ac035 | 2214 | trace2_region_enter("merge", "regular renames", opt->repo); |
e1a124e8 EN |
2215 | detect_regular_renames(opt, merge_base, side1, MERGE_SIDE1); |
2216 | detect_regular_renames(opt, merge_base, side2, MERGE_SIDE2); | |
557ac035 | 2217 | trace2_region_leave("merge", "regular renames", opt->repo); |
e1a124e8 | 2218 | |
557ac035 | 2219 | trace2_region_enter("merge", "directory renames", opt->repo); |
112e1112 EN |
2220 | need_dir_renames = |
2221 | !opt->priv->call_depth && | |
2222 | (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE || | |
2223 | opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT); | |
2224 | ||
2225 | if (need_dir_renames) { | |
2226 | get_provisional_directory_renames(opt, MERGE_SIDE1, &clean); | |
2227 | get_provisional_directory_renames(opt, MERGE_SIDE2, &clean); | |
2228 | handle_directory_level_conflicts(opt); | |
2229 | } | |
2230 | ||
e1a124e8 EN |
2231 | ALLOC_GROW(combined.queue, |
2232 | renames->pairs[1].nr + renames->pairs[2].nr, | |
2233 | combined.alloc); | |
fa5e06d6 EN |
2234 | clean &= collect_renames(opt, &combined, MERGE_SIDE1, |
2235 | &renames->dir_renames[2], | |
2236 | &renames->dir_renames[1]); | |
2237 | clean &= collect_renames(opt, &combined, MERGE_SIDE2, | |
2238 | &renames->dir_renames[1], | |
2239 | &renames->dir_renames[2]); | |
e1a124e8 | 2240 | QSORT(combined.queue, combined.nr, compare_pairs); |
557ac035 | 2241 | trace2_region_leave("merge", "directory renames", opt->repo); |
e1a124e8 | 2242 | |
557ac035 | 2243 | trace2_region_enter("merge", "process renames", opt->repo); |
e1a124e8 | 2244 | clean &= process_renames(opt, &combined); |
557ac035 | 2245 | trace2_region_leave("merge", "process renames", opt->repo); |
e1a124e8 EN |
2246 | |
2247 | /* Free memory for renames->pairs[] and combined */ | |
2248 | for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) { | |
2249 | free(renames->pairs[s].queue); | |
2250 | DIFF_QUEUE_CLEAR(&renames->pairs[s]); | |
2251 | } | |
2252 | if (combined.nr) { | |
2253 | int i; | |
2254 | for (i = 0; i < combined.nr; i++) | |
2255 | diff_free_filepair(combined.queue[i]); | |
2256 | free(combined.queue); | |
2257 | } | |
231e2dd4 | 2258 | |
231e2dd4 EN |
2259 | return clean; |
2260 | } | |
2261 | ||
04af1879 EN |
2262 | /*** Function Grouping: functions related to process_entries() ***/ |
2263 | ||
8adffaa8 EN |
2264 | static int string_list_df_name_compare(const char *one, const char *two) |
2265 | { | |
2266 | int onelen = strlen(one); | |
2267 | int twolen = strlen(two); | |
2268 | /* | |
2269 | * Here we only care that entries for D/F conflicts are | |
2270 | * adjacent, in particular with the file of the D/F conflict | |
2271 | * appearing before files below the corresponding directory. | |
2272 | * The order of the rest of the list is irrelevant for us. | |
2273 | * | |
2274 | * To achieve this, we sort with df_name_compare and provide | |
2275 | * the mode S_IFDIR so that D/F conflicts will sort correctly. | |
2276 | * We use the mode S_IFDIR for everything else for simplicity, | |
2277 | * since in other cases any changes in their order due to | |
2278 | * sorting cause no problems for us. | |
2279 | */ | |
2280 | int cmp = df_name_compare(one, onelen, S_IFDIR, | |
2281 | two, twolen, S_IFDIR); | |
2282 | /* | |
2283 | * Now that 'foo' and 'foo/bar' compare equal, we have to make sure | |
2284 | * that 'foo' comes before 'foo/bar'. | |
2285 | */ | |
2286 | if (cmp) | |
2287 | return cmp; | |
2288 | return onelen - twolen; | |
2289 | } | |
2290 | ||
a9945bba | 2291 | struct directory_versions { |
bb470f4e EN |
2292 | /* |
2293 | * versions: list of (basename -> version_info) | |
2294 | * | |
2295 | * The basenames are in reverse lexicographic order of full pathnames, | |
2296 | * as processed in process_entries(). This puts all entries within | |
2297 | * a directory together, and covers the directory itself after | |
2298 | * everything within it, allowing us to write subtrees before needing | |
2299 | * to record information for the tree itself. | |
2300 | */ | |
a9945bba | 2301 | struct string_list versions; |
bb470f4e EN |
2302 | |
2303 | /* | |
2304 | * offsets: list of (full relative path directories -> integer offsets) | |
2305 | * | |
2306 | * Since versions contains basenames from files in multiple different | |
2307 | * directories, we need to know which entries in versions correspond | |
2308 | * to which directories. Values of e.g. | |
2309 | * "" 0 | |
2310 | * src 2 | |
2311 | * src/moduleA 5 | |
2312 | * Would mean that entries 0-1 of versions are files in the toplevel | |
2313 | * directory, entries 2-4 are files under src/, and the remaining | |
2314 | * entries starting at index 5 are files under src/moduleA/. | |
2315 | */ | |
2316 | struct string_list offsets; | |
2317 | ||
2318 | /* | |
2319 | * last_directory: directory that previously processed file found in | |
2320 | * | |
2321 | * last_directory starts NULL, but records the directory in which the | |
2322 | * previous file was found within. As soon as | |
2323 | * directory(current_file) != last_directory | |
2324 | * then we need to start updating accounting in versions & offsets. | |
2325 | * Note that last_directory is always the last path in "offsets" (or | |
2326 | * NULL if "offsets" is empty) so this exists just for quick access. | |
2327 | */ | |
2328 | const char *last_directory; | |
2329 | ||
2330 | /* last_directory_len: cached computation of strlen(last_directory) */ | |
2331 | unsigned last_directory_len; | |
a9945bba EN |
2332 | }; |
2333 | ||
ee4012dc EN |
2334 | static int tree_entry_order(const void *a_, const void *b_) |
2335 | { | |
2336 | const struct string_list_item *a = a_; | |
2337 | const struct string_list_item *b = b_; | |
2338 | ||
2339 | const struct merged_info *ami = a->util; | |
2340 | const struct merged_info *bmi = b->util; | |
2341 | return base_name_compare(a->string, strlen(a->string), ami->result.mode, | |
2342 | b->string, strlen(b->string), bmi->result.mode); | |
2343 | } | |
2344 | ||
2345 | static void write_tree(struct object_id *result_oid, | |
2346 | struct string_list *versions, | |
2347 | unsigned int offset, | |
2348 | size_t hash_size) | |
2349 | { | |
2350 | size_t maxlen = 0, extra; | |
2351 | unsigned int nr = versions->nr - offset; | |
2352 | struct strbuf buf = STRBUF_INIT; | |
2353 | struct string_list relevant_entries = STRING_LIST_INIT_NODUP; | |
2354 | int i; | |
2355 | ||
2356 | /* | |
2357 | * We want to sort the last (versions->nr-offset) entries in versions. | |
2358 | * Do so by abusing the string_list API a bit: make another string_list | |
2359 | * that contains just those entries and then sort them. | |
2360 | * | |
2361 | * We won't use relevant_entries again and will let it just pop off the | |
2362 | * stack, so there won't be allocation worries or anything. | |
2363 | */ | |
2364 | relevant_entries.items = versions->items + offset; | |
2365 | relevant_entries.nr = versions->nr - offset; | |
2366 | QSORT(relevant_entries.items, relevant_entries.nr, tree_entry_order); | |
2367 | ||
2368 | /* Pre-allocate some space in buf */ | |
2369 | extra = hash_size + 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */ | |
2370 | for (i = 0; i < nr; i++) { | |
2371 | maxlen += strlen(versions->items[offset+i].string) + extra; | |
2372 | } | |
2373 | strbuf_grow(&buf, maxlen); | |
2374 | ||
2375 | /* Write each entry out to buf */ | |
2376 | for (i = 0; i < nr; i++) { | |
2377 | struct merged_info *mi = versions->items[offset+i].util; | |
2378 | struct version_info *ri = &mi->result; | |
2379 | strbuf_addf(&buf, "%o %s%c", | |
2380 | ri->mode, | |
2381 | versions->items[offset+i].string, '\0'); | |
2382 | strbuf_add(&buf, ri->oid.hash, hash_size); | |
2383 | } | |
2384 | ||
2385 | /* Write this object file out, and record in result_oid */ | |
2386 | write_object_file(buf.buf, buf.len, tree_type, result_oid); | |
2387 | strbuf_release(&buf); | |
2388 | } | |
2389 | ||
a9945bba EN |
2390 | static void record_entry_for_tree(struct directory_versions *dir_metadata, |
2391 | const char *path, | |
2392 | struct merged_info *mi) | |
2393 | { | |
2394 | const char *basename; | |
2395 | ||
2396 | if (mi->is_null) | |
2397 | /* nothing to record */ | |
2398 | return; | |
2399 | ||
2400 | basename = path + mi->basename_offset; | |
2401 | assert(strchr(basename, '/') == NULL); | |
2402 | string_list_append(&dir_metadata->versions, | |
2403 | basename)->util = &mi->result; | |
2404 | } | |
2405 | ||
bb470f4e EN |
2406 | static void write_completed_directory(struct merge_options *opt, |
2407 | const char *new_directory_name, | |
2408 | struct directory_versions *info) | |
2409 | { | |
2410 | const char *prev_dir; | |
2411 | struct merged_info *dir_info = NULL; | |
2412 | unsigned int offset; | |
2413 | ||
2414 | /* | |
2415 | * Some explanation of info->versions and info->offsets... | |
2416 | * | |
2417 | * process_entries() iterates over all relevant files AND | |
2418 | * directories in reverse lexicographic order, and calls this | |
2419 | * function. Thus, an example of the paths that process_entries() | |
2420 | * could operate on (along with the directories for those paths | |
2421 | * being shown) is: | |
2422 | * | |
2423 | * xtract.c "" | |
2424 | * tokens.txt "" | |
2425 | * src/moduleB/umm.c src/moduleB | |
2426 | * src/moduleB/stuff.h src/moduleB | |
2427 | * src/moduleB/baz.c src/moduleB | |
2428 | * src/moduleB src | |
2429 | * src/moduleA/foo.c src/moduleA | |
2430 | * src/moduleA/bar.c src/moduleA | |
2431 | * src/moduleA src | |
2432 | * src "" | |
2433 | * Makefile "" | |
2434 | * | |
2435 | * info->versions: | |
2436 | * | |
2437 | * always contains the unprocessed entries and their | |
2438 | * version_info information. For example, after the first five | |
2439 | * entries above, info->versions would be: | |
2440 | * | |
2441 | * xtract.c <xtract.c's version_info> | |
2442 | * token.txt <token.txt's version_info> | |
2443 | * umm.c <src/moduleB/umm.c's version_info> | |
2444 | * stuff.h <src/moduleB/stuff.h's version_info> | |
2445 | * baz.c <src/moduleB/baz.c's version_info> | |
2446 | * | |
2447 | * Once a subdirectory is completed we remove the entries in | |
2448 | * that subdirectory from info->versions, writing it as a tree | |
2449 | * (write_tree()). Thus, as soon as we get to src/moduleB, | |
2450 | * info->versions would be updated to | |
2451 | * | |
2452 | * xtract.c <xtract.c's version_info> | |
2453 | * token.txt <token.txt's version_info> | |
2454 | * moduleB <src/moduleB's version_info> | |
2455 | * | |
2456 | * info->offsets: | |
2457 | * | |
2458 | * helps us track which entries in info->versions correspond to | |
2459 | * which directories. When we are N directories deep (e.g. 4 | |
2460 | * for src/modA/submod/subdir/), we have up to N+1 unprocessed | |
2461 | * directories (+1 because of toplevel dir). Corresponding to | |
2462 | * the info->versions example above, after processing five entries | |
2463 | * info->offsets will be: | |
2464 | * | |
2465 | * "" 0 | |
2466 | * src/moduleB 2 | |
2467 | * | |
2468 | * which is used to know that xtract.c & token.txt are from the | |
2469 | * toplevel dirctory, while umm.c & stuff.h & baz.c are from the | |
2470 | * src/moduleB directory. Again, following the example above, | |
2471 | * once we need to process src/moduleB, then info->offsets is | |
2472 | * updated to | |
2473 | * | |
2474 | * "" 0 | |
2475 | * src 2 | |
2476 | * | |
2477 | * which says that moduleB (and only moduleB so far) is in the | |
2478 | * src directory. | |
2479 | * | |
2480 | * One unique thing to note about info->offsets here is that | |
2481 | * "src" was not added to info->offsets until there was a path | |
2482 | * (a file OR directory) immediately below src/ that got | |
2483 | * processed. | |
2484 | * | |
2485 | * Since process_entry() just appends new entries to info->versions, | |
2486 | * write_completed_directory() only needs to do work if the next path | |
2487 | * is in a directory that is different than the last directory found | |
2488 | * in info->offsets. | |
2489 | */ | |
2490 | ||
2491 | /* | |
2492 | * If we are working with the same directory as the last entry, there | |
2493 | * is no work to do. (See comments above the directory_name member of | |
2494 | * struct merged_info for why we can use pointer comparison instead of | |
2495 | * strcmp here.) | |
2496 | */ | |
2497 | if (new_directory_name == info->last_directory) | |
2498 | return; | |
2499 | ||
2500 | /* | |
2501 | * If we are just starting (last_directory is NULL), or last_directory | |
2502 | * is a prefix of the current directory, then we can just update | |
2503 | * info->offsets to record the offset where we started this directory | |
2504 | * and update last_directory to have quick access to it. | |
2505 | */ | |
2506 | if (info->last_directory == NULL || | |
2507 | !strncmp(new_directory_name, info->last_directory, | |
2508 | info->last_directory_len)) { | |
2509 | uintptr_t offset = info->versions.nr; | |
2510 | ||
2511 | info->last_directory = new_directory_name; | |
2512 | info->last_directory_len = strlen(info->last_directory); | |
2513 | /* | |
2514 | * Record the offset into info->versions where we will | |
2515 | * start recording basenames of paths found within | |
2516 | * new_directory_name. | |
2517 | */ | |
2518 | string_list_append(&info->offsets, | |
2519 | info->last_directory)->util = (void*)offset; | |
2520 | return; | |
2521 | } | |
2522 | ||
2523 | /* | |
2524 | * The next entry that will be processed will be within | |
2525 | * new_directory_name. Since at this point we know that | |
2526 | * new_directory_name is within a different directory than | |
2527 | * info->last_directory, we have all entries for info->last_directory | |
2528 | * in info->versions and we need to create a tree object for them. | |
2529 | */ | |
2530 | dir_info = strmap_get(&opt->priv->paths, info->last_directory); | |
2531 | assert(dir_info); | |
2532 | offset = (uintptr_t)info->offsets.items[info->offsets.nr-1].util; | |
2533 | if (offset == info->versions.nr) { | |
2534 | /* | |
2535 | * Actually, we don't need to create a tree object in this | |
2536 | * case. Whenever all files within a directory disappear | |
2537 | * during the merge (e.g. unmodified on one side and | |
2538 | * deleted on the other, or files were renamed elsewhere), | |
2539 | * then we get here and the directory itself needs to be | |
2540 | * omitted from its parent tree as well. | |
2541 | */ | |
2542 | dir_info->is_null = 1; | |
2543 | } else { | |
2544 | /* | |
2545 | * Write out the tree to the git object directory, and also | |
2546 | * record the mode and oid in dir_info->result. | |
2547 | */ | |
2548 | dir_info->is_null = 0; | |
2549 | dir_info->result.mode = S_IFDIR; | |
2550 | write_tree(&dir_info->result.oid, &info->versions, offset, | |
2551 | opt->repo->hash_algo->rawsz); | |
2552 | } | |
2553 | ||
2554 | /* | |
2555 | * We've now used several entries from info->versions and one entry | |
2556 | * from info->offsets, so we get rid of those values. | |
2557 | */ | |
2558 | info->offsets.nr--; | |
2559 | info->versions.nr = offset; | |
2560 | ||
2561 | /* | |
2562 | * Now we've taken care of the completed directory, but we need to | |
2563 | * prepare things since future entries will be in | |
2564 | * new_directory_name. (In particular, process_entry() will be | |
2565 | * appending new entries to info->versions.) So, we need to make | |
2566 | * sure new_directory_name is the last entry in info->offsets. | |
2567 | */ | |
2568 | prev_dir = info->offsets.nr == 0 ? NULL : | |
2569 | info->offsets.items[info->offsets.nr-1].string; | |
2570 | if (new_directory_name != prev_dir) { | |
2571 | uintptr_t c = info->versions.nr; | |
2572 | string_list_append(&info->offsets, | |
2573 | new_directory_name)->util = (void*)c; | |
2574 | } | |
2575 | ||
2576 | /* And, of course, we need to update last_directory to match. */ | |
2577 | info->last_directory = new_directory_name; | |
2578 | info->last_directory_len = strlen(info->last_directory); | |
2579 | } | |
2580 | ||
6a02dd90 EN |
2581 | /* Per entry merge function */ |
2582 | static void process_entry(struct merge_options *opt, | |
2583 | const char *path, | |
a9945bba EN |
2584 | struct conflict_info *ci, |
2585 | struct directory_versions *dir_metadata) | |
6a02dd90 | 2586 | { |
23366d2a EN |
2587 | int df_file_index = 0; |
2588 | ||
6a02dd90 EN |
2589 | VERIFY_CI(ci); |
2590 | assert(ci->filemask >= 0 && ci->filemask <= 7); | |
2591 | /* ci->match_mask == 7 was handled in collect_merge_info_callback() */ | |
2592 | assert(ci->match_mask == 0 || ci->match_mask == 3 || | |
2593 | ci->match_mask == 5 || ci->match_mask == 6); | |
2594 | ||
a9945bba EN |
2595 | if (ci->dirmask) { |
2596 | record_entry_for_tree(dir_metadata, path, &ci->merged); | |
2597 | if (ci->filemask == 0) | |
2598 | /* nothing else to handle */ | |
2599 | return; | |
2600 | assert(ci->df_conflict); | |
2601 | } | |
2602 | ||
0ccfa4e5 EN |
2603 | if (ci->df_conflict && ci->merged.result.mode == 0) { |
2604 | int i; | |
2605 | ||
2606 | /* | |
2607 | * directory no longer in the way, but we do have a file we | |
2608 | * need to place here so we need to clean away the "directory | |
2609 | * merges to nothing" result. | |
2610 | */ | |
2611 | ci->df_conflict = 0; | |
2612 | assert(ci->filemask != 0); | |
2613 | ci->merged.clean = 0; | |
2614 | ci->merged.is_null = 0; | |
2615 | /* and we want to zero out any directory-related entries */ | |
2616 | ci->match_mask = (ci->match_mask & ~ci->dirmask); | |
2617 | ci->dirmask = 0; | |
2618 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
2619 | if (ci->filemask & (1 << i)) | |
2620 | continue; | |
2621 | ci->stages[i].mode = 0; | |
2622 | oidcpy(&ci->stages[i].oid, &null_oid); | |
2623 | } | |
2624 | } else if (ci->df_conflict && ci->merged.result.mode != 0) { | |
23366d2a EN |
2625 | /* |
2626 | * This started out as a D/F conflict, and the entries in | |
2627 | * the competing directory were not removed by the merge as | |
2628 | * evidenced by write_completed_directory() writing a value | |
2629 | * to ci->merged.result.mode. | |
2630 | */ | |
2631 | struct conflict_info *new_ci; | |
2632 | const char *branch; | |
2633 | const char *old_path = path; | |
2634 | int i; | |
2635 | ||
2636 | assert(ci->merged.result.mode == S_IFDIR); | |
2637 | ||
2638 | /* | |
2639 | * If filemask is 1, we can just ignore the file as having | |
2640 | * been deleted on both sides. We do not want to overwrite | |
2641 | * ci->merged.result, since it stores the tree for all the | |
2642 | * files under it. | |
2643 | */ | |
2644 | if (ci->filemask == 1) { | |
2645 | ci->filemask = 0; | |
2646 | return; | |
2647 | } | |
2648 | ||
2649 | /* | |
2650 | * This file still exists on at least one side, and we want | |
2651 | * the directory to remain here, so we need to move this | |
2652 | * path to some new location. | |
2653 | */ | |
2654 | new_ci = xcalloc(1, sizeof(*new_ci)); | |
2655 | /* We don't really want new_ci->merged.result copied, but it'll | |
2656 | * be overwritten below so it doesn't matter. We also don't | |
2657 | * want any directory mode/oid values copied, but we'll zero | |
2658 | * those out immediately. We do want the rest of ci copied. | |
2659 | */ | |
2660 | memcpy(new_ci, ci, sizeof(*ci)); | |
2661 | new_ci->match_mask = (new_ci->match_mask & ~new_ci->dirmask); | |
2662 | new_ci->dirmask = 0; | |
2663 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
2664 | if (new_ci->filemask & (1 << i)) | |
2665 | continue; | |
2666 | /* zero out any entries related to directories */ | |
2667 | new_ci->stages[i].mode = 0; | |
2668 | oidcpy(&new_ci->stages[i].oid, &null_oid); | |
2669 | } | |
2670 | ||
2671 | /* | |
2672 | * Find out which side this file came from; note that we | |
2673 | * cannot just use ci->filemask, because renames could cause | |
2674 | * the filemask to go back to 7. So we use dirmask, then | |
2675 | * pick the opposite side's index. | |
2676 | */ | |
2677 | df_file_index = (ci->dirmask & (1 << 1)) ? 2 : 1; | |
2678 | branch = (df_file_index == 1) ? opt->branch1 : opt->branch2; | |
2679 | path = unique_path(&opt->priv->paths, path, branch); | |
2680 | strmap_put(&opt->priv->paths, path, new_ci); | |
2681 | ||
2682 | path_msg(opt, path, 0, | |
2683 | _("CONFLICT (file/directory): directory in the way " | |
2684 | "of %s from %s; moving it to %s instead."), | |
2685 | old_path, branch, path); | |
2686 | ||
2687 | /* | |
2688 | * Zero out the filemask for the old ci. At this point, ci | |
2689 | * was just an entry for a directory, so we don't need to | |
2690 | * do anything more with it. | |
2691 | */ | |
2692 | ci->filemask = 0; | |
2693 | ||
2694 | /* | |
2695 | * Now note that we're working on the new entry (path was | |
2696 | * updated above. | |
2697 | */ | |
2698 | ci = new_ci; | |
6a02dd90 EN |
2699 | } |
2700 | ||
2701 | /* | |
2702 | * NOTE: Below there is a long switch-like if-elseif-elseif... block | |
2703 | * which the code goes through even for the df_conflict cases | |
23366d2a | 2704 | * above. |
6a02dd90 EN |
2705 | */ |
2706 | if (ci->match_mask) { | |
2707 | ci->merged.clean = 1; | |
2708 | if (ci->match_mask == 6) { | |
2709 | /* stages[1] == stages[2] */ | |
2710 | ci->merged.result.mode = ci->stages[1].mode; | |
2711 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
2712 | } else { | |
2713 | /* determine the mask of the side that didn't match */ | |
2714 | unsigned int othermask = 7 & ~ci->match_mask; | |
2715 | int side = (othermask == 4) ? 2 : 1; | |
2716 | ||
2717 | ci->merged.result.mode = ci->stages[side].mode; | |
2718 | ci->merged.is_null = !ci->merged.result.mode; | |
2719 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
2720 | ||
2721 | assert(othermask == 2 || othermask == 4); | |
2722 | assert(ci->merged.is_null == | |
2723 | (ci->filemask == ci->match_mask)); | |
2724 | } | |
2725 | } else if (ci->filemask >= 6 && | |
2726 | (S_IFMT & ci->stages[1].mode) != | |
2727 | (S_IFMT & ci->stages[2].mode)) { | |
4ef88fc3 EN |
2728 | /* Two different items from (file/submodule/symlink) */ |
2729 | if (opt->priv->call_depth) { | |
2730 | /* Just use the version from the merge base */ | |
2731 | ci->merged.clean = 0; | |
2732 | oidcpy(&ci->merged.result.oid, &ci->stages[0].oid); | |
2733 | ci->merged.result.mode = ci->stages[0].mode; | |
2734 | ci->merged.is_null = (ci->merged.result.mode == 0); | |
2735 | } else { | |
2736 | /* Handle by renaming one or both to separate paths. */ | |
2737 | unsigned o_mode = ci->stages[0].mode; | |
2738 | unsigned a_mode = ci->stages[1].mode; | |
2739 | unsigned b_mode = ci->stages[2].mode; | |
2740 | struct conflict_info *new_ci; | |
2741 | const char *a_path = NULL, *b_path = NULL; | |
2742 | int rename_a = 0, rename_b = 0; | |
2743 | ||
2744 | new_ci = xmalloc(sizeof(*new_ci)); | |
2745 | ||
2746 | if (S_ISREG(a_mode)) | |
2747 | rename_a = 1; | |
2748 | else if (S_ISREG(b_mode)) | |
2749 | rename_b = 1; | |
2750 | else { | |
2751 | rename_a = 1; | |
2752 | rename_b = 1; | |
2753 | } | |
2754 | ||
2755 | path_msg(opt, path, 0, | |
2756 | _("CONFLICT (distinct types): %s had different " | |
2757 | "types on each side; renamed %s of them so " | |
2758 | "each can be recorded somewhere."), | |
2759 | path, | |
2760 | (rename_a && rename_b) ? _("both") : _("one")); | |
2761 | ||
2762 | ci->merged.clean = 0; | |
2763 | memcpy(new_ci, ci, sizeof(*new_ci)); | |
2764 | ||
2765 | /* Put b into new_ci, removing a from stages */ | |
2766 | new_ci->merged.result.mode = ci->stages[2].mode; | |
2767 | oidcpy(&new_ci->merged.result.oid, &ci->stages[2].oid); | |
2768 | new_ci->stages[1].mode = 0; | |
2769 | oidcpy(&new_ci->stages[1].oid, &null_oid); | |
2770 | new_ci->filemask = 5; | |
2771 | if ((S_IFMT & b_mode) != (S_IFMT & o_mode)) { | |
2772 | new_ci->stages[0].mode = 0; | |
2773 | oidcpy(&new_ci->stages[0].oid, &null_oid); | |
2774 | new_ci->filemask = 4; | |
2775 | } | |
2776 | ||
2777 | /* Leave only a in ci, fixing stages. */ | |
2778 | ci->merged.result.mode = ci->stages[1].mode; | |
2779 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
2780 | ci->stages[2].mode = 0; | |
2781 | oidcpy(&ci->stages[2].oid, &null_oid); | |
2782 | ci->filemask = 3; | |
2783 | if ((S_IFMT & a_mode) != (S_IFMT & o_mode)) { | |
2784 | ci->stages[0].mode = 0; | |
2785 | oidcpy(&ci->stages[0].oid, &null_oid); | |
2786 | ci->filemask = 2; | |
2787 | } | |
2788 | ||
2789 | /* Insert entries into opt->priv_paths */ | |
2790 | assert(rename_a || rename_b); | |
2791 | if (rename_a) { | |
2792 | a_path = unique_path(&opt->priv->paths, | |
2793 | path, opt->branch1); | |
2794 | strmap_put(&opt->priv->paths, a_path, ci); | |
2795 | } | |
2796 | ||
2797 | if (rename_b) | |
2798 | b_path = unique_path(&opt->priv->paths, | |
2799 | path, opt->branch2); | |
2800 | else | |
2801 | b_path = path; | |
2802 | strmap_put(&opt->priv->paths, b_path, new_ci); | |
2803 | ||
2804 | if (rename_a && rename_b) { | |
2805 | strmap_remove(&opt->priv->paths, path, 0); | |
2806 | /* | |
2807 | * We removed path from opt->priv->paths. path | |
2808 | * will also eventually need to be freed, but | |
2809 | * it may still be used by e.g. ci->pathnames. | |
2810 | * So, store it in another string-list for now. | |
2811 | */ | |
2812 | string_list_append(&opt->priv->paths_to_free, | |
2813 | path); | |
2814 | } | |
2815 | ||
2816 | /* | |
2817 | * Do special handling for b_path since process_entry() | |
2818 | * won't be called on it specially. | |
2819 | */ | |
2820 | strmap_put(&opt->priv->conflicted, b_path, new_ci); | |
2821 | record_entry_for_tree(dir_metadata, b_path, | |
2822 | &new_ci->merged); | |
2823 | ||
2824 | /* | |
2825 | * Remaining code for processing this entry should | |
2826 | * think in terms of processing a_path. | |
2827 | */ | |
2828 | if (a_path) | |
2829 | path = a_path; | |
2830 | } | |
6a02dd90 | 2831 | } else if (ci->filemask >= 6) { |
991bbdca EN |
2832 | /* Need a two-way or three-way content merge */ |
2833 | struct version_info merged_file; | |
2834 | unsigned clean_merge; | |
2835 | struct version_info *o = &ci->stages[0]; | |
2836 | struct version_info *a = &ci->stages[1]; | |
2837 | struct version_info *b = &ci->stages[2]; | |
2838 | ||
2839 | clean_merge = handle_content_merge(opt, path, o, a, b, | |
2840 | ci->pathnames, | |
2841 | opt->priv->call_depth * 2, | |
2842 | &merged_file); | |
2843 | ci->merged.clean = clean_merge && | |
2844 | !ci->df_conflict && !ci->path_conflict; | |
2845 | ci->merged.result.mode = merged_file.mode; | |
2846 | ci->merged.is_null = (merged_file.mode == 0); | |
2847 | oidcpy(&ci->merged.result.oid, &merged_file.oid); | |
2848 | if (clean_merge && ci->df_conflict) { | |
2849 | assert(df_file_index == 1 || df_file_index == 2); | |
2850 | ci->filemask = 1 << df_file_index; | |
2851 | ci->stages[df_file_index].mode = merged_file.mode; | |
2852 | oidcpy(&ci->stages[df_file_index].oid, &merged_file.oid); | |
2853 | } | |
2854 | if (!clean_merge) { | |
2855 | const char *reason = _("content"); | |
2856 | if (ci->filemask == 6) | |
2857 | reason = _("add/add"); | |
2858 | if (S_ISGITLINK(merged_file.mode)) | |
2859 | reason = _("submodule"); | |
2860 | path_msg(opt, path, 0, | |
2861 | _("CONFLICT (%s): Merge conflict in %s"), | |
2862 | reason, path); | |
2863 | } | |
6a02dd90 EN |
2864 | } else if (ci->filemask == 3 || ci->filemask == 5) { |
2865 | /* Modify/delete */ | |
c5a6f655 EN |
2866 | const char *modify_branch, *delete_branch; |
2867 | int side = (ci->filemask == 5) ? 2 : 1; | |
2868 | int index = opt->priv->call_depth ? 0 : side; | |
2869 | ||
2870 | ci->merged.result.mode = ci->stages[index].mode; | |
2871 | oidcpy(&ci->merged.result.oid, &ci->stages[index].oid); | |
2872 | ci->merged.clean = 0; | |
2873 | ||
2874 | modify_branch = (side == 1) ? opt->branch1 : opt->branch2; | |
2875 | delete_branch = (side == 1) ? opt->branch2 : opt->branch1; | |
2876 | ||
2e91ddd2 EN |
2877 | if (ci->path_conflict && |
2878 | oideq(&ci->stages[0].oid, &ci->stages[side].oid)) { | |
2879 | /* | |
2880 | * This came from a rename/delete; no action to take, | |
2881 | * but avoid printing "modify/delete" conflict notice | |
2882 | * since the contents were not modified. | |
2883 | */ | |
2884 | } else { | |
2885 | path_msg(opt, path, 0, | |
2886 | _("CONFLICT (modify/delete): %s deleted in %s " | |
2887 | "and modified in %s. Version %s of %s left " | |
2888 | "in tree."), | |
2889 | path, delete_branch, modify_branch, | |
2890 | modify_branch, path); | |
2891 | } | |
6a02dd90 EN |
2892 | } else if (ci->filemask == 2 || ci->filemask == 4) { |
2893 | /* Added on one side */ | |
2894 | int side = (ci->filemask == 4) ? 2 : 1; | |
2895 | ci->merged.result.mode = ci->stages[side].mode; | |
2896 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
53e88a03 | 2897 | ci->merged.clean = !ci->df_conflict && !ci->path_conflict; |
6a02dd90 EN |
2898 | } else if (ci->filemask == 1) { |
2899 | /* Deleted on both sides */ | |
2900 | ci->merged.is_null = 1; | |
2901 | ci->merged.result.mode = 0; | |
2902 | oidcpy(&ci->merged.result.oid, &null_oid); | |
53e88a03 | 2903 | ci->merged.clean = !ci->path_conflict; |
6a02dd90 EN |
2904 | } |
2905 | ||
2906 | /* | |
2907 | * If still conflicted, record it separately. This allows us to later | |
2908 | * iterate over just conflicted entries when updating the index instead | |
2909 | * of iterating over all entries. | |
2910 | */ | |
2911 | if (!ci->merged.clean) | |
2912 | strmap_put(&opt->priv->conflicted, path, ci); | |
a9945bba | 2913 | record_entry_for_tree(dir_metadata, path, &ci->merged); |
6a02dd90 EN |
2914 | } |
2915 | ||
231e2dd4 EN |
2916 | static void process_entries(struct merge_options *opt, |
2917 | struct object_id *result_oid) | |
2918 | { | |
6a02dd90 EN |
2919 | struct hashmap_iter iter; |
2920 | struct strmap_entry *e; | |
8adffaa8 EN |
2921 | struct string_list plist = STRING_LIST_INIT_NODUP; |
2922 | struct string_list_item *entry; | |
bb470f4e EN |
2923 | struct directory_versions dir_metadata = { STRING_LIST_INIT_NODUP, |
2924 | STRING_LIST_INIT_NODUP, | |
2925 | NULL, 0 }; | |
6a02dd90 | 2926 | |
557ac035 | 2927 | trace2_region_enter("merge", "process_entries setup", opt->repo); |
6a02dd90 EN |
2928 | if (strmap_empty(&opt->priv->paths)) { |
2929 | oidcpy(result_oid, opt->repo->hash_algo->empty_tree); | |
2930 | return; | |
2931 | } | |
2932 | ||
8adffaa8 | 2933 | /* Hack to pre-allocate plist to the desired size */ |
557ac035 | 2934 | trace2_region_enter("merge", "plist grow", opt->repo); |
8adffaa8 | 2935 | ALLOC_GROW(plist.items, strmap_get_size(&opt->priv->paths), plist.alloc); |
557ac035 | 2936 | trace2_region_leave("merge", "plist grow", opt->repo); |
8adffaa8 EN |
2937 | |
2938 | /* Put every entry from paths into plist, then sort */ | |
557ac035 | 2939 | trace2_region_enter("merge", "plist copy", opt->repo); |
6a02dd90 | 2940 | strmap_for_each_entry(&opt->priv->paths, &iter, e) { |
8adffaa8 EN |
2941 | string_list_append(&plist, e->key)->util = e->value; |
2942 | } | |
557ac035 EN |
2943 | trace2_region_leave("merge", "plist copy", opt->repo); |
2944 | ||
2945 | trace2_region_enter("merge", "plist special sort", opt->repo); | |
8adffaa8 EN |
2946 | plist.cmp = string_list_df_name_compare; |
2947 | string_list_sort(&plist); | |
557ac035 EN |
2948 | trace2_region_leave("merge", "plist special sort", opt->repo); |
2949 | ||
2950 | trace2_region_leave("merge", "process_entries setup", opt->repo); | |
8adffaa8 EN |
2951 | |
2952 | /* | |
2953 | * Iterate over the items in reverse order, so we can handle paths | |
2954 | * below a directory before needing to handle the directory itself. | |
bb470f4e EN |
2955 | * |
2956 | * This allows us to write subtrees before we need to write trees, | |
2957 | * and it also enables sane handling of directory/file conflicts | |
2958 | * (because it allows us to know whether the directory is still in | |
2959 | * the way when it is time to process the file at the same path). | |
8adffaa8 | 2960 | */ |
557ac035 | 2961 | trace2_region_enter("merge", "processing", opt->repo); |
8adffaa8 EN |
2962 | for (entry = &plist.items[plist.nr-1]; entry >= plist.items; --entry) { |
2963 | char *path = entry->string; | |
6a02dd90 EN |
2964 | /* |
2965 | * NOTE: mi may actually be a pointer to a conflict_info, but | |
2966 | * we have to check mi->clean first to see if it's safe to | |
2967 | * reassign to such a pointer type. | |
2968 | */ | |
8adffaa8 | 2969 | struct merged_info *mi = entry->util; |
6a02dd90 | 2970 | |
bb470f4e EN |
2971 | write_completed_directory(opt, mi->directory_name, |
2972 | &dir_metadata); | |
a9945bba EN |
2973 | if (mi->clean) |
2974 | record_entry_for_tree(&dir_metadata, path, mi); | |
2975 | else { | |
8adffaa8 | 2976 | struct conflict_info *ci = (struct conflict_info *)mi; |
a9945bba | 2977 | process_entry(opt, path, ci, &dir_metadata); |
8adffaa8 | 2978 | } |
6a02dd90 | 2979 | } |
557ac035 | 2980 | trace2_region_leave("merge", "processing", opt->repo); |
6a02dd90 | 2981 | |
557ac035 | 2982 | trace2_region_enter("merge", "process_entries cleanup", opt->repo); |
bb470f4e EN |
2983 | if (dir_metadata.offsets.nr != 1 || |
2984 | (uintptr_t)dir_metadata.offsets.items[0].util != 0) { | |
2985 | printf("dir_metadata.offsets.nr = %d (should be 1)\n", | |
2986 | dir_metadata.offsets.nr); | |
2987 | printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n", | |
2988 | (unsigned)(uintptr_t)dir_metadata.offsets.items[0].util); | |
2989 | fflush(stdout); | |
2990 | BUG("dir_metadata accounting completely off; shouldn't happen"); | |
2991 | } | |
ee4012dc EN |
2992 | write_tree(result_oid, &dir_metadata.versions, 0, |
2993 | opt->repo->hash_algo->rawsz); | |
8adffaa8 | 2994 | string_list_clear(&plist, 0); |
a9945bba | 2995 | string_list_clear(&dir_metadata.versions, 0); |
bb470f4e | 2996 | string_list_clear(&dir_metadata.offsets, 0); |
557ac035 | 2997 | trace2_region_leave("merge", "process_entries cleanup", opt->repo); |
231e2dd4 EN |
2998 | } |
2999 | ||
04af1879 EN |
3000 | /*** Function Grouping: functions related to merge_switch_to_result() ***/ |
3001 | ||
9fefce68 EN |
3002 | static int checkout(struct merge_options *opt, |
3003 | struct tree *prev, | |
3004 | struct tree *next) | |
3005 | { | |
6681ce5c EN |
3006 | /* Switch the index/working copy from old to new */ |
3007 | int ret; | |
3008 | struct tree_desc trees[2]; | |
3009 | struct unpack_trees_options unpack_opts; | |
3010 | ||
3011 | memset(&unpack_opts, 0, sizeof(unpack_opts)); | |
3012 | unpack_opts.head_idx = -1; | |
3013 | unpack_opts.src_index = opt->repo->index; | |
3014 | unpack_opts.dst_index = opt->repo->index; | |
3015 | ||
3016 | setup_unpack_trees_porcelain(&unpack_opts, "merge"); | |
3017 | ||
3018 | /* | |
3019 | * NOTE: if this were just "git checkout" code, we would probably | |
3020 | * read or refresh the cache and check for a conflicted index, but | |
3021 | * builtin/merge.c or sequencer.c really needs to read the index | |
3022 | * and check for conflicted entries before starting merging for a | |
3023 | * good user experience (no sense waiting for merges/rebases before | |
3024 | * erroring out), so there's no reason to duplicate that work here. | |
3025 | */ | |
3026 | ||
3027 | /* 2-way merge to the new branch */ | |
3028 | unpack_opts.update = 1; | |
3029 | unpack_opts.merge = 1; | |
3030 | unpack_opts.quiet = 0; /* FIXME: sequencer might want quiet? */ | |
3031 | unpack_opts.verbose_update = (opt->verbosity > 2); | |
3032 | unpack_opts.fn = twoway_merge; | |
3033 | if (1/* FIXME: opts->overwrite_ignore*/) { | |
3034 | unpack_opts.dir = xcalloc(1, sizeof(*unpack_opts.dir)); | |
3035 | unpack_opts.dir->flags |= DIR_SHOW_IGNORED; | |
3036 | setup_standard_excludes(unpack_opts.dir); | |
3037 | } | |
3038 | parse_tree(prev); | |
3039 | init_tree_desc(&trees[0], prev->buffer, prev->size); | |
3040 | parse_tree(next); | |
3041 | init_tree_desc(&trees[1], next->buffer, next->size); | |
3042 | ||
3043 | ret = unpack_trees(2, trees, &unpack_opts); | |
3044 | clear_unpack_trees_porcelain(&unpack_opts); | |
3045 | dir_clear(unpack_opts.dir); | |
3046 | FREE_AND_NULL(unpack_opts.dir); | |
3047 | return ret; | |
9fefce68 EN |
3048 | } |
3049 | ||
3050 | static int record_conflicted_index_entries(struct merge_options *opt, | |
3051 | struct index_state *index, | |
3052 | struct strmap *paths, | |
3053 | struct strmap *conflicted) | |
3054 | { | |
ef2b3693 EN |
3055 | struct hashmap_iter iter; |
3056 | struct strmap_entry *e; | |
3057 | int errs = 0; | |
3058 | int original_cache_nr; | |
3059 | ||
9fefce68 EN |
3060 | if (strmap_empty(conflicted)) |
3061 | return 0; | |
3062 | ||
ef2b3693 EN |
3063 | original_cache_nr = index->cache_nr; |
3064 | ||
3065 | /* Put every entry from paths into plist, then sort */ | |
3066 | strmap_for_each_entry(conflicted, &iter, e) { | |
3067 | const char *path = e->key; | |
3068 | struct conflict_info *ci = e->value; | |
3069 | int pos; | |
3070 | struct cache_entry *ce; | |
3071 | int i; | |
3072 | ||
3073 | VERIFY_CI(ci); | |
3074 | ||
3075 | /* | |
3076 | * The index will already have a stage=0 entry for this path, | |
3077 | * because we created an as-merged-as-possible version of the | |
3078 | * file and checkout() moved the working copy and index over | |
3079 | * to that version. | |
3080 | * | |
3081 | * However, previous iterations through this loop will have | |
3082 | * added unstaged entries to the end of the cache which | |
3083 | * ignore the standard alphabetical ordering of cache | |
3084 | * entries and break invariants needed for index_name_pos() | |
3085 | * to work. However, we know the entry we want is before | |
3086 | * those appended cache entries, so do a temporary swap on | |
3087 | * cache_nr to only look through entries of interest. | |
3088 | */ | |
3089 | SWAP(index->cache_nr, original_cache_nr); | |
3090 | pos = index_name_pos(index, path, strlen(path)); | |
3091 | SWAP(index->cache_nr, original_cache_nr); | |
3092 | if (pos < 0) { | |
3093 | if (ci->filemask != 1) | |
3094 | BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path); | |
3095 | cache_tree_invalidate_path(index, path); | |
3096 | } else { | |
3097 | ce = index->cache[pos]; | |
3098 | ||
3099 | /* | |
3100 | * Clean paths with CE_SKIP_WORKTREE set will not be | |
3101 | * written to the working tree by the unpack_trees() | |
3102 | * call in checkout(). Our conflicted entries would | |
3103 | * have appeared clean to that code since we ignored | |
3104 | * the higher order stages. Thus, we need override | |
3105 | * the CE_SKIP_WORKTREE bit and manually write those | |
3106 | * files to the working disk here. | |
3107 | * | |
3108 | * TODO: Implement this CE_SKIP_WORKTREE fixup. | |
3109 | */ | |
3110 | ||
3111 | /* | |
3112 | * Mark this cache entry for removal and instead add | |
3113 | * new stage>0 entries corresponding to the | |
3114 | * conflicts. If there are many conflicted entries, we | |
3115 | * want to avoid memmove'ing O(NM) entries by | |
3116 | * inserting the new entries one at a time. So, | |
3117 | * instead, we just add the new cache entries to the | |
3118 | * end (ignoring normal index requirements on sort | |
3119 | * order) and sort the index once we're all done. | |
3120 | */ | |
3121 | ce->ce_flags |= CE_REMOVE; | |
3122 | } | |
3123 | ||
3124 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
3125 | struct version_info *vi; | |
3126 | if (!(ci->filemask & (1ul << i))) | |
3127 | continue; | |
3128 | vi = &ci->stages[i]; | |
3129 | ce = make_cache_entry(index, vi->mode, &vi->oid, | |
3130 | path, i+1, 0); | |
3131 | add_index_entry(index, ce, ADD_CACHE_JUST_APPEND); | |
3132 | } | |
3133 | } | |
3134 | ||
3135 | /* | |
3136 | * Remove the unused cache entries (and invalidate the relevant | |
3137 | * cache-trees), then sort the index entries to get the conflicted | |
3138 | * entries we added to the end into their right locations. | |
3139 | */ | |
3140 | remove_marked_cache_entries(index, 1); | |
3141 | QSORT(index->cache, index->cache_nr, cmp_cache_name_compare); | |
3142 | ||
3143 | return errs; | |
9fefce68 EN |
3144 | } |
3145 | ||
17e5574b EN |
3146 | void merge_switch_to_result(struct merge_options *opt, |
3147 | struct tree *head, | |
3148 | struct merge_result *result, | |
3149 | int update_worktree_and_index, | |
3150 | int display_update_msgs) | |
3151 | { | |
9fefce68 EN |
3152 | assert(opt->priv == NULL); |
3153 | if (result->clean >= 0 && update_worktree_and_index) { | |
3154 | struct merge_options_internal *opti = result->priv; | |
3155 | ||
557ac035 | 3156 | trace2_region_enter("merge", "checkout", opt->repo); |
9fefce68 EN |
3157 | if (checkout(opt, head, result->tree)) { |
3158 | /* failure to function */ | |
3159 | result->clean = -1; | |
3160 | return; | |
3161 | } | |
557ac035 | 3162 | trace2_region_leave("merge", "checkout", opt->repo); |
9fefce68 | 3163 | |
557ac035 | 3164 | trace2_region_enter("merge", "record_conflicted", opt->repo); |
9fefce68 EN |
3165 | if (record_conflicted_index_entries(opt, opt->repo->index, |
3166 | &opti->paths, | |
3167 | &opti->conflicted)) { | |
3168 | /* failure to function */ | |
3169 | result->clean = -1; | |
3170 | return; | |
3171 | } | |
557ac035 | 3172 | trace2_region_leave("merge", "record_conflicted", opt->repo); |
9fefce68 EN |
3173 | } |
3174 | ||
3175 | if (display_update_msgs) { | |
c5a6f655 EN |
3176 | struct merge_options_internal *opti = result->priv; |
3177 | struct hashmap_iter iter; | |
3178 | struct strmap_entry *e; | |
3179 | struct string_list olist = STRING_LIST_INIT_NODUP; | |
3180 | int i; | |
3181 | ||
557ac035 EN |
3182 | trace2_region_enter("merge", "display messages", opt->repo); |
3183 | ||
c5a6f655 EN |
3184 | /* Hack to pre-allocate olist to the desired size */ |
3185 | ALLOC_GROW(olist.items, strmap_get_size(&opti->output), | |
3186 | olist.alloc); | |
3187 | ||
3188 | /* Put every entry from output into olist, then sort */ | |
3189 | strmap_for_each_entry(&opti->output, &iter, e) { | |
3190 | string_list_append(&olist, e->key)->util = e->value; | |
3191 | } | |
3192 | string_list_sort(&olist); | |
3193 | ||
3194 | /* Iterate over the items, printing them */ | |
3195 | for (i = 0; i < olist.nr; ++i) { | |
3196 | struct strbuf *sb = olist.items[i].util; | |
3197 | ||
3198 | printf("%s", sb->buf); | |
3199 | } | |
3200 | string_list_clear(&olist, 0); | |
f39d05ca EN |
3201 | |
3202 | /* Also include needed rename limit adjustment now */ | |
3203 | diff_warn_rename_limit("merge.renamelimit", | |
3204 | opti->renames.needed_limit, 0); | |
557ac035 EN |
3205 | |
3206 | trace2_region_leave("merge", "display messages", opt->repo); | |
9fefce68 EN |
3207 | } |
3208 | ||
17e5574b EN |
3209 | merge_finalize(opt, result); |
3210 | } | |
3211 | ||
3212 | void merge_finalize(struct merge_options *opt, | |
3213 | struct merge_result *result) | |
3214 | { | |
89422d29 EN |
3215 | struct merge_options_internal *opti = result->priv; |
3216 | ||
3217 | assert(opt->priv == NULL); | |
3218 | ||
43e9c4ee | 3219 | clear_or_reinit_internal_opts(opti, 0); |
89422d29 | 3220 | FREE_AND_NULL(opti); |
17e5574b EN |
3221 | } |
3222 | ||
04af1879 EN |
3223 | /*** Function Grouping: helper functions for merge_incore_*() ***/ |
3224 | ||
4296d8f1 EN |
3225 | static inline void set_commit_tree(struct commit *c, struct tree *t) |
3226 | { | |
3227 | c->maybe_tree = t; | |
3228 | } | |
3229 | ||
4296d8f1 EN |
3230 | static struct commit *make_virtual_commit(struct repository *repo, |
3231 | struct tree *tree, | |
3232 | const char *comment) | |
3233 | { | |
3234 | struct commit *commit = alloc_commit_node(repo); | |
3235 | ||
3236 | set_merge_remote_desc(commit, comment, (struct object *)commit); | |
3237 | set_commit_tree(commit, tree); | |
3238 | commit->object.parsed = 1; | |
3239 | return commit; | |
3240 | } | |
3241 | ||
231e2dd4 EN |
3242 | static void merge_start(struct merge_options *opt, struct merge_result *result) |
3243 | { | |
f5d9fbc2 EN |
3244 | struct rename_info *renames; |
3245 | int i; | |
3246 | ||
e4171b1b | 3247 | /* Sanity checks on opt */ |
557ac035 | 3248 | trace2_region_enter("merge", "sanity checks", opt->repo); |
e4171b1b EN |
3249 | assert(opt->repo); |
3250 | ||
3251 | assert(opt->branch1 && opt->branch2); | |
3252 | ||
3253 | assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE && | |
3254 | opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE); | |
3255 | assert(opt->rename_limit >= -1); | |
3256 | assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE); | |
3257 | assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1); | |
3258 | ||
3259 | assert(opt->xdl_opts >= 0); | |
3260 | assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL && | |
3261 | opt->recursive_variant <= MERGE_VARIANT_THEIRS); | |
3262 | ||
3263 | /* | |
3264 | * detect_renames, verbosity, buffer_output, and obuf are ignored | |
3265 | * fields that were used by "recursive" rather than "ort" -- but | |
3266 | * sanity check them anyway. | |
3267 | */ | |
3268 | assert(opt->detect_renames >= -1 && | |
3269 | opt->detect_renames <= DIFF_DETECT_COPY); | |
3270 | assert(opt->verbosity >= 0 && opt->verbosity <= 5); | |
3271 | assert(opt->buffer_output <= 2); | |
3272 | assert(opt->obuf.len == 0); | |
3273 | ||
3274 | assert(opt->priv == NULL); | |
cf8937ac EN |
3275 | if (result->priv) { |
3276 | opt->priv = result->priv; | |
3277 | result->priv = NULL; | |
3278 | /* | |
3279 | * opt->priv non-NULL means we had results from a previous | |
3280 | * run; do a few sanity checks that user didn't mess with | |
3281 | * it in an obvious fashion. | |
3282 | */ | |
3283 | assert(opt->priv->call_depth == 0); | |
3284 | assert(!opt->priv->toplevel_dir || | |
3285 | 0 == strlen(opt->priv->toplevel_dir)); | |
3286 | } | |
557ac035 | 3287 | trace2_region_leave("merge", "sanity checks", opt->repo); |
e4171b1b | 3288 | |
c8017176 EN |
3289 | /* Default to histogram diff. Actually, just hardcode it...for now. */ |
3290 | opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF); | |
3291 | ||
e4171b1b | 3292 | /* Initialization of opt->priv, our internal merge data */ |
557ac035 | 3293 | trace2_region_enter("merge", "allocate/init", opt->repo); |
cf8937ac EN |
3294 | if (opt->priv) { |
3295 | clear_or_reinit_internal_opts(opt->priv, 1); | |
3296 | trace2_region_leave("merge", "allocate/init", opt->repo); | |
3297 | return; | |
3298 | } | |
e4171b1b EN |
3299 | opt->priv = xcalloc(1, sizeof(*opt->priv)); |
3300 | ||
f5d9fbc2 EN |
3301 | /* Initialization of various renames fields */ |
3302 | renames = &opt->priv->renames; | |
3303 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) { | |
3304 | strset_init_with_options(&renames->dirs_removed[i], | |
3305 | NULL, 0); | |
3306 | strmap_init_with_options(&renames->dir_rename_count[i], | |
3307 | NULL, 1); | |
3308 | strmap_init_with_options(&renames->dir_renames[i], | |
3309 | NULL, 0); | |
3310 | } | |
3311 | ||
e4171b1b EN |
3312 | /* |
3313 | * Although we initialize opt->priv->paths with strdup_strings=0, | |
3314 | * that's just to avoid making yet another copy of an allocated | |
3315 | * string. Putting the entry into paths means we are taking | |
43c1dccb | 3316 | * ownership, so we will later free it. paths_to_free is similar. |
e4171b1b EN |
3317 | * |
3318 | * In contrast, conflicted just has a subset of keys from paths, so | |
3319 | * we don't want to free those (it'd be a duplicate free). | |
3320 | */ | |
3321 | strmap_init_with_options(&opt->priv->paths, NULL, 0); | |
3322 | strmap_init_with_options(&opt->priv->conflicted, NULL, 0); | |
43c1dccb | 3323 | string_list_init(&opt->priv->paths_to_free, 0); |
c5a6f655 EN |
3324 | |
3325 | /* | |
3326 | * keys & strbufs in output will sometimes need to outlive "paths", | |
3327 | * so it will have a copy of relevant keys. It's probably a small | |
3328 | * subset of the overall paths that have special output. | |
3329 | */ | |
3330 | strmap_init(&opt->priv->output); | |
557ac035 EN |
3331 | |
3332 | trace2_region_leave("merge", "allocate/init", opt->repo); | |
231e2dd4 EN |
3333 | } |
3334 | ||
04af1879 EN |
3335 | /*** Function Grouping: merge_incore_*() and their internal variants ***/ |
3336 | ||
231e2dd4 EN |
3337 | /* |
3338 | * Originally from merge_trees_internal(); heavily adapted, though. | |
3339 | */ | |
3340 | static void merge_ort_nonrecursive_internal(struct merge_options *opt, | |
3341 | struct tree *merge_base, | |
3342 | struct tree *side1, | |
3343 | struct tree *side2, | |
3344 | struct merge_result *result) | |
3345 | { | |
3346 | struct object_id working_tree_oid; | |
3347 | ||
557ac035 | 3348 | trace2_region_enter("merge", "collect_merge_info", opt->repo); |
0c0d705b EN |
3349 | if (collect_merge_info(opt, merge_base, side1, side2) != 0) { |
3350 | /* | |
3351 | * TRANSLATORS: The %s arguments are: 1) tree hash of a merge | |
3352 | * base, and 2-3) the trees for the two trees we're merging. | |
3353 | */ | |
3354 | err(opt, _("collecting merge info failed for trees %s, %s, %s"), | |
3355 | oid_to_hex(&merge_base->object.oid), | |
3356 | oid_to_hex(&side1->object.oid), | |
3357 | oid_to_hex(&side2->object.oid)); | |
3358 | result->clean = -1; | |
3359 | return; | |
3360 | } | |
557ac035 | 3361 | trace2_region_leave("merge", "collect_merge_info", opt->repo); |
0c0d705b | 3362 | |
557ac035 | 3363 | trace2_region_enter("merge", "renames", opt->repo); |
231e2dd4 EN |
3364 | result->clean = detect_and_process_renames(opt, merge_base, |
3365 | side1, side2); | |
557ac035 EN |
3366 | trace2_region_leave("merge", "renames", opt->repo); |
3367 | ||
3368 | trace2_region_enter("merge", "process_entries", opt->repo); | |
231e2dd4 | 3369 | process_entries(opt, &working_tree_oid); |
557ac035 | 3370 | trace2_region_leave("merge", "process_entries", opt->repo); |
231e2dd4 EN |
3371 | |
3372 | /* Set return values */ | |
3373 | result->tree = parse_tree_indirect(&working_tree_oid); | |
3374 | /* existence of conflicted entries implies unclean */ | |
3375 | result->clean &= strmap_empty(&opt->priv->conflicted); | |
3376 | if (!opt->priv->call_depth) { | |
3377 | result->priv = opt->priv; | |
3378 | opt->priv = NULL; | |
3379 | } | |
3380 | } | |
3381 | ||
8119214f EN |
3382 | /* |
3383 | * Originally from merge_recursive_internal(); somewhat adapted, though. | |
3384 | */ | |
3385 | static void merge_ort_internal(struct merge_options *opt, | |
3386 | struct commit_list *merge_bases, | |
3387 | struct commit *h1, | |
3388 | struct commit *h2, | |
3389 | struct merge_result *result) | |
3390 | { | |
3391 | struct commit_list *iter; | |
3392 | struct commit *merged_merge_bases; | |
3393 | const char *ancestor_name; | |
3394 | struct strbuf merge_base_abbrev = STRBUF_INIT; | |
3395 | ||
3396 | if (!merge_bases) { | |
3397 | merge_bases = get_merge_bases(h1, h2); | |
3398 | /* See merge-ort.h:merge_incore_recursive() declaration NOTE */ | |
3399 | merge_bases = reverse_commit_list(merge_bases); | |
3400 | } | |
3401 | ||
3402 | merged_merge_bases = pop_commit(&merge_bases); | |
3403 | if (merged_merge_bases == NULL) { | |
3404 | /* if there is no common ancestor, use an empty tree */ | |
3405 | struct tree *tree; | |
3406 | ||
3407 | tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree); | |
3408 | merged_merge_bases = make_virtual_commit(opt->repo, tree, | |
3409 | "ancestor"); | |
3410 | ancestor_name = "empty tree"; | |
3411 | } else if (merge_bases) { | |
3412 | ancestor_name = "merged common ancestors"; | |
3413 | } else { | |
3414 | strbuf_add_unique_abbrev(&merge_base_abbrev, | |
3415 | &merged_merge_bases->object.oid, | |
3416 | DEFAULT_ABBREV); | |
3417 | ancestor_name = merge_base_abbrev.buf; | |
3418 | } | |
3419 | ||
3420 | for (iter = merge_bases; iter; iter = iter->next) { | |
3421 | const char *saved_b1, *saved_b2; | |
3422 | struct commit *prev = merged_merge_bases; | |
3423 | ||
3424 | opt->priv->call_depth++; | |
3425 | /* | |
3426 | * When the merge fails, the result contains files | |
3427 | * with conflict markers. The cleanness flag is | |
3428 | * ignored (unless indicating an error), it was never | |
3429 | * actually used, as result of merge_trees has always | |
3430 | * overwritten it: the committed "conflicts" were | |
3431 | * already resolved. | |
3432 | */ | |
3433 | saved_b1 = opt->branch1; | |
3434 | saved_b2 = opt->branch2; | |
3435 | opt->branch1 = "Temporary merge branch 1"; | |
3436 | opt->branch2 = "Temporary merge branch 2"; | |
3437 | merge_ort_internal(opt, NULL, prev, iter->item, result); | |
3438 | if (result->clean < 0) | |
3439 | return; | |
3440 | opt->branch1 = saved_b1; | |
3441 | opt->branch2 = saved_b2; | |
3442 | opt->priv->call_depth--; | |
3443 | ||
3444 | merged_merge_bases = make_virtual_commit(opt->repo, | |
3445 | result->tree, | |
3446 | "merged tree"); | |
3447 | commit_list_insert(prev, &merged_merge_bases->parents); | |
3448 | commit_list_insert(iter->item, | |
3449 | &merged_merge_bases->parents->next); | |
3450 | ||
3451 | clear_or_reinit_internal_opts(opt->priv, 1); | |
3452 | } | |
3453 | ||
3454 | opt->ancestor = ancestor_name; | |
3455 | merge_ort_nonrecursive_internal(opt, | |
3456 | repo_get_commit_tree(opt->repo, | |
3457 | merged_merge_bases), | |
3458 | repo_get_commit_tree(opt->repo, h1), | |
3459 | repo_get_commit_tree(opt->repo, h2), | |
3460 | result); | |
3461 | strbuf_release(&merge_base_abbrev); | |
3462 | opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */ | |
3463 | } | |
3464 | ||
17e5574b EN |
3465 | void merge_incore_nonrecursive(struct merge_options *opt, |
3466 | struct tree *merge_base, | |
3467 | struct tree *side1, | |
3468 | struct tree *side2, | |
3469 | struct merge_result *result) | |
3470 | { | |
557ac035 EN |
3471 | trace2_region_enter("merge", "incore_nonrecursive", opt->repo); |
3472 | ||
3473 | trace2_region_enter("merge", "merge_start", opt->repo); | |
231e2dd4 EN |
3474 | assert(opt->ancestor != NULL); |
3475 | merge_start(opt, result); | |
557ac035 EN |
3476 | trace2_region_leave("merge", "merge_start", opt->repo); |
3477 | ||
231e2dd4 | 3478 | merge_ort_nonrecursive_internal(opt, merge_base, side1, side2, result); |
557ac035 | 3479 | trace2_region_leave("merge", "incore_nonrecursive", opt->repo); |
17e5574b EN |
3480 | } |
3481 | ||
3482 | void merge_incore_recursive(struct merge_options *opt, | |
3483 | struct commit_list *merge_bases, | |
3484 | struct commit *side1, | |
3485 | struct commit *side2, | |
3486 | struct merge_result *result) | |
3487 | { | |
557ac035 EN |
3488 | trace2_region_enter("merge", "incore_recursive", opt->repo); |
3489 | ||
8119214f EN |
3490 | /* We set the ancestor label based on the merge_bases */ |
3491 | assert(opt->ancestor == NULL); | |
3492 | ||
557ac035 | 3493 | trace2_region_enter("merge", "merge_start", opt->repo); |
8119214f | 3494 | merge_start(opt, result); |
557ac035 EN |
3495 | trace2_region_leave("merge", "merge_start", opt->repo); |
3496 | ||
8119214f | 3497 | merge_ort_internal(opt, merge_bases, side1, side2, result); |
557ac035 | 3498 | trace2_region_leave("merge", "incore_recursive", opt->repo); |
17e5574b | 3499 | } |