]>
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" |
ea305a68 | 21 | #include "attr.h" |
67845745 | 22 | #include "blob.h" |
ef2b3693 | 23 | #include "cache-tree.h" |
4296d8f1 | 24 | #include "commit.h" |
67845745 | 25 | #include "commit-reach.h" |
e4171b1b EN |
26 | #include "diff.h" |
27 | #include "diffcore.h" | |
6681ce5c | 28 | #include "dir.h" |
7bec8e7f | 29 | #include "entry.h" |
f591c472 | 30 | #include "ll-merge.h" |
ee4012dc | 31 | #include "object-store.h" |
2bff554b | 32 | #include "promisor-remote.h" |
4204cd59 | 33 | #include "revision.h" |
5b59c3db | 34 | #include "strmap.h" |
c73cda76 | 35 | #include "submodule.h" |
231e2dd4 | 36 | #include "tree.h" |
6681ce5c | 37 | #include "unpack-trees.h" |
c8017176 | 38 | #include "xdiff-interface.h" |
5b59c3db | 39 | |
d2bc1994 EN |
40 | /* |
41 | * We have many arrays of size 3. Whenever we have such an array, the | |
42 | * indices refer to one of the sides of the three-way merge. This is so | |
43 | * pervasive that the constants 0, 1, and 2 are used in many places in the | |
44 | * code (especially in arithmetic operations to find the other side's index | |
45 | * or to compute a relevant mask), but sometimes these enum names are used | |
46 | * to aid code clarity. | |
47 | * | |
48 | * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side" | |
49 | * referred to there is one of these three sides. | |
50 | */ | |
51 | enum merge_side { | |
52 | MERGE_BASE = 0, | |
53 | MERGE_SIDE1 = 1, | |
54 | MERGE_SIDE2 = 2 | |
55 | }; | |
56 | ||
19ceb486 EN |
57 | static unsigned RESULT_INITIALIZED = 0x1abe11ed; /* unlikely accidental value */ |
58 | ||
beb06145 EN |
59 | struct traversal_callback_data { |
60 | unsigned long mask; | |
61 | unsigned long dirmask; | |
62 | struct name_entry names[3]; | |
63 | }; | |
64 | ||
864075ec | 65 | struct rename_info { |
c09376d5 EN |
66 | /* |
67 | * All variables that are arrays of size 3 correspond to data tracked | |
68 | * for the sides in enum merge_side. Index 0 is almost always unused | |
69 | * because we often only need to track information for MERGE_SIDE1 and | |
70 | * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames | |
71 | * are determined relative to what changed since the MERGE_BASE). | |
72 | */ | |
73 | ||
864075ec EN |
74 | /* |
75 | * pairs: pairing of filenames from diffcore_rename() | |
864075ec EN |
76 | */ |
77 | struct diff_queue_struct pairs[3]; | |
78 | ||
c09376d5 EN |
79 | /* |
80 | * dirs_removed: directories removed on a given side of history. | |
fb52938e EN |
81 | * |
82 | * The keys of dirs_removed[side] are the directories that were removed | |
83 | * on the given side of history. The value of the strintmap for each | |
84 | * directory is a value from enum dir_rename_relevance. | |
c09376d5 | 85 | */ |
a49b55d5 | 86 | struct strintmap dirs_removed[3]; |
c09376d5 EN |
87 | |
88 | /* | |
89 | * dir_rename_count: tracking where parts of a directory were renamed to | |
90 | * | |
91 | * When files in a directory are renamed, they may not all go to the | |
92 | * same location. Each strmap here tracks: | |
93 | * old_dir => {new_dir => int} | |
94 | * That is, dir_rename_count[side] is a strmap to a strintmap. | |
95 | */ | |
96 | struct strmap dir_rename_count[3]; | |
97 | ||
98 | /* | |
99 | * dir_renames: computed directory renames | |
100 | * | |
101 | * This is a map of old_dir => new_dir and is derived in part from | |
102 | * dir_rename_count. | |
103 | */ | |
104 | struct strmap dir_renames[3]; | |
105 | ||
32a56dfb | 106 | /* |
ec59da60 | 107 | * relevant_sources: deleted paths wanted in rename detection, and why |
32a56dfb EN |
108 | * |
109 | * relevant_sources is a set of deleted paths on each side of | |
110 | * history for which we need rename detection. If a path is deleted | |
111 | * on one side of history, we need to detect if it is part of a | |
112 | * rename if either | |
32a56dfb | 113 | * * the file is modified/deleted on the other side of history |
ec59da60 | 114 | * * we need to detect renames for an ancestor directory |
32a56dfb | 115 | * If neither of those are true, we can skip rename detection for |
ec59da60 EN |
116 | * that path. The reason is stored as a value from enum |
117 | * file_rename_relevance, as the reason can inform the algorithm in | |
118 | * diffcore_rename_extended(). | |
32a56dfb | 119 | */ |
a49b55d5 | 120 | struct strintmap relevant_sources[3]; |
32a56dfb | 121 | |
2fd9eda4 EN |
122 | /* |
123 | * dir_rename_mask: | |
124 | * 0: optimization removing unmodified potential rename source okay | |
125 | * 2 or 4: optimization okay, but must check for files added to dir | |
126 | * 7: optimization forbidden; need rename source in case of dir rename | |
127 | */ | |
128 | unsigned dir_rename_mask:3; | |
129 | ||
beb06145 EN |
130 | /* |
131 | * callback_data_*: supporting data structures for alternate traversal | |
132 | * | |
133 | * We sometimes need to be able to traverse through all the files | |
134 | * in a given tree before all immediate subdirectories within that | |
135 | * tree. Since traverse_trees() doesn't do that naturally, we have | |
136 | * a traverse_trees_wrapper() that stores any immediate | |
137 | * subdirectories while traversing files, then traverses the | |
138 | * immediate subdirectories later. These callback_data* variables | |
139 | * store the information for the subdirectories so that we can do | |
140 | * that traversal order. | |
141 | */ | |
142 | struct traversal_callback_data *callback_data; | |
143 | int callback_data_nr, callback_data_alloc; | |
144 | char *callback_data_traverse_path; | |
145 | ||
64aceb6d EN |
146 | /* |
147 | * merge_trees: trees passed to the merge algorithm for the merge | |
148 | * | |
149 | * merge_trees records the trees passed to the merge algorithm. But, | |
150 | * this data also is stored in merge_result->priv. If a sequence of | |
151 | * merges are being done (such as when cherry-picking or rebasing), | |
152 | * the next merge can look at this and re-use information from | |
153 | * previous merges under certain circumstances. | |
154 | * | |
155 | * See also all the cached_* variables. | |
156 | */ | |
157 | struct tree *merge_trees[3]; | |
158 | ||
159 | /* | |
160 | * cached_pairs_valid_side: which side's cached info can be reused | |
161 | * | |
162 | * See the description for merge_trees. For repeated merges, at most | |
163 | * only one side's cached information can be used. Valid values: | |
164 | * MERGE_SIDE2: cached data from side2 can be reused | |
165 | * MERGE_SIDE1: cached data from side1 can be reused | |
166 | * 0: no cached data can be reused | |
167 | */ | |
168 | int cached_pairs_valid_side; | |
169 | ||
d29bd6d7 EN |
170 | /* |
171 | * cached_pairs: Caching of renames and deletions. | |
172 | * | |
173 | * These are mappings recording renames and deletions of individual | |
174 | * files (not directories). They are thus a map from an old | |
175 | * filename to either NULL (for deletions) or a new filename (for | |
176 | * renames). | |
177 | */ | |
178 | struct strmap cached_pairs[3]; | |
179 | ||
180 | /* | |
181 | * cached_target_names: just the destinations from cached_pairs | |
182 | * | |
183 | * We sometimes want a fast lookup to determine if a given filename | |
184 | * is one of the destinations in cached_pairs. cached_target_names | |
185 | * is thus duplicative information, but it provides a fast lookup. | |
186 | */ | |
187 | struct strset cached_target_names[3]; | |
188 | ||
189 | /* | |
190 | * cached_irrelevant: Caching of rename_sources that aren't relevant. | |
191 | * | |
192 | * If we try to detect a rename for a source path and succeed, it's | |
193 | * part of a rename. If we try to detect a rename for a source path | |
194 | * and fail, then it's a delete. If we do not try to detect a rename | |
195 | * for a path, then we don't know if it's a rename or a delete. If | |
196 | * merge-ort doesn't think the path is relevant, then we just won't | |
197 | * cache anything for that path. But there's a slight problem in | |
198 | * that merge-ort can think a path is RELEVANT_LOCATION, but due to | |
199 | * commit 9bd342137e ("diffcore-rename: determine which | |
200 | * relevant_sources are no longer relevant", 2021-03-13), | |
201 | * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To | |
202 | * avoid excessive calls to diffcore_rename_extended() we still need | |
203 | * to cache such paths, though we cannot record them as either | |
204 | * renames or deletes. So we cache them here as a "turned out to be | |
205 | * irrelevant *for this commit*" as they are often also irrelevant | |
206 | * for subsequent commits, though we will have to do some extra | |
207 | * checking to see whether such paths become relevant for rename | |
208 | * detection when cherry-picking/rebasing subsequent commits. | |
209 | */ | |
210 | struct strset cached_irrelevant[3]; | |
211 | ||
864075ec EN |
212 | /* |
213 | * needed_limit: value needed for inexact rename detection to run | |
214 | * | |
215 | * If the current rename limit wasn't high enough for inexact | |
216 | * rename detection to run, this records the limit needed. Otherwise, | |
217 | * this value remains 0. | |
218 | */ | |
219 | int needed_limit; | |
220 | }; | |
221 | ||
5b59c3db EN |
222 | struct merge_options_internal { |
223 | /* | |
224 | * paths: primary data structure in all of merge ort. | |
225 | * | |
226 | * The keys of paths: | |
227 | * * are full relative paths from the toplevel of the repository | |
228 | * (e.g. "drivers/firmware/raspberrypi.c"). | |
229 | * * store all relevant paths in the repo, both directories and | |
230 | * files (e.g. drivers, drivers/firmware would also be included) | |
231 | * * these keys serve to intern all the path strings, which allows | |
232 | * us to do pointer comparison on directory names instead of | |
233 | * strcmp; we just have to be careful to use the interned strings. | |
43c1dccb EN |
234 | * (Technically paths_to_free may track some strings that were |
235 | * removed from froms paths.) | |
5b59c3db EN |
236 | * |
237 | * The values of paths: | |
238 | * * either a pointer to a merged_info, or a conflict_info struct | |
239 | * * merged_info contains all relevant information for a | |
240 | * non-conflicted entry. | |
241 | * * conflict_info contains a merged_info, plus any additional | |
242 | * information about a conflict such as the higher orders stages | |
243 | * involved and the names of the paths those came from (handy | |
244 | * once renames get involved). | |
245 | * * a path may start "conflicted" (i.e. point to a conflict_info) | |
246 | * and then a later step (e.g. three-way content merge) determines | |
247 | * it can be cleanly merged, at which point it'll be marked clean | |
248 | * and the algorithm will ignore any data outside the contained | |
249 | * merged_info for that entry | |
250 | * * If an entry remains conflicted, the merged_info portion of a | |
251 | * conflict_info will later be filled with whatever version of | |
252 | * the file should be placed in the working directory (e.g. an | |
253 | * as-merged-as-possible variation that contains conflict markers). | |
254 | */ | |
255 | struct strmap paths; | |
256 | ||
257 | /* | |
258 | * conflicted: a subset of keys->values from "paths" | |
259 | * | |
260 | * conflicted is basically an optimization between process_entries() | |
261 | * and record_conflicted_index_entries(); the latter could loop over | |
262 | * ALL the entries in paths AGAIN and look for the ones that are | |
263 | * still conflicted, but since process_entries() has to loop over | |
264 | * all of them, it saves the ones it couldn't resolve in this strmap | |
265 | * so that record_conflicted_index_entries() can iterate just the | |
266 | * relevant entries. | |
267 | */ | |
268 | struct strmap conflicted; | |
269 | ||
43c1dccb EN |
270 | /* |
271 | * paths_to_free: additional list of strings to free | |
272 | * | |
273 | * If keys are removed from "paths", they are added to paths_to_free | |
274 | * to ensure they are later freed. We avoid free'ing immediately since | |
275 | * other places (e.g. conflict_info.pathnames[]) may still be | |
276 | * referencing these paths. | |
277 | */ | |
278 | struct string_list paths_to_free; | |
279 | ||
c5a6f655 EN |
280 | /* |
281 | * output: special messages and conflict notices for various paths | |
282 | * | |
283 | * This is a map of pathnames (a subset of the keys in "paths" above) | |
284 | * to strbufs. It gathers various warning/conflict/notice messages | |
285 | * for later processing. | |
286 | */ | |
287 | struct strmap output; | |
288 | ||
5b59c3db | 289 | /* |
864075ec EN |
290 | * renames: various data relating to rename detection |
291 | */ | |
292 | struct rename_info renames; | |
293 | ||
ea305a68 EN |
294 | /* |
295 | * attr_index: hacky minimal index used for renormalization | |
296 | * | |
297 | * renormalization code _requires_ an index, though it only needs to | |
298 | * find a .gitattributes file within the index. So, when | |
299 | * renormalization is important, we create a special index with just | |
300 | * that one file. | |
301 | */ | |
302 | struct index_state attr_index; | |
303 | ||
5b59c3db | 304 | /* |
05b85c6e | 305 | * current_dir_name, toplevel_dir: temporary vars |
5b59c3db | 306 | * |
05b85c6e EN |
307 | * These are used in collect_merge_info_callback(), and will set the |
308 | * various merged_info.directory_name for the various paths we get; | |
309 | * see documentation for that variable and the requirements placed on | |
310 | * that field. | |
5b59c3db EN |
311 | */ |
312 | const char *current_dir_name; | |
05b85c6e | 313 | const char *toplevel_dir; |
5b59c3db EN |
314 | |
315 | /* call_depth: recursion level counter for merging merge bases */ | |
316 | int call_depth; | |
317 | }; | |
318 | ||
319 | struct version_info { | |
320 | struct object_id oid; | |
321 | unsigned short mode; | |
322 | }; | |
323 | ||
324 | struct merged_info { | |
325 | /* if is_null, ignore result. otherwise result has oid & mode */ | |
326 | struct version_info result; | |
327 | unsigned is_null:1; | |
328 | ||
329 | /* | |
330 | * clean: whether the path in question is cleanly merged. | |
331 | * | |
332 | * see conflict_info.merged for more details. | |
333 | */ | |
334 | unsigned clean:1; | |
335 | ||
336 | /* | |
337 | * basename_offset: offset of basename of path. | |
338 | * | |
339 | * perf optimization to avoid recomputing offset of final '/' | |
340 | * character in pathname (0 if no '/' in pathname). | |
341 | */ | |
342 | size_t basename_offset; | |
343 | ||
344 | /* | |
345 | * directory_name: containing directory name. | |
346 | * | |
347 | * Note that we assume directory_name is constructed such that | |
348 | * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name, | |
349 | * i.e. string equality is equivalent to pointer equality. For this | |
350 | * to hold, we have to be careful setting directory_name. | |
351 | */ | |
352 | const char *directory_name; | |
353 | }; | |
354 | ||
355 | struct conflict_info { | |
356 | /* | |
357 | * merged: the version of the path that will be written to working tree | |
358 | * | |
359 | * WARNING: It is critical to check merged.clean and ensure it is 0 | |
360 | * before reading any conflict_info fields outside of merged. | |
361 | * Allocated merge_info structs will always have clean set to 1. | |
362 | * Allocated conflict_info structs will have merged.clean set to 0 | |
363 | * initially. The merged.clean field is how we know if it is safe | |
364 | * to access other parts of conflict_info besides merged; if a | |
365 | * conflict_info's merged.clean is changed to 1, the rest of the | |
366 | * algorithm is not allowed to look at anything outside of the | |
367 | * merged member anymore. | |
368 | */ | |
369 | struct merged_info merged; | |
370 | ||
371 | /* oids & modes from each of the three trees for this path */ | |
372 | struct version_info stages[3]; | |
373 | ||
374 | /* pathnames for each stage; may differ due to rename detection */ | |
375 | const char *pathnames[3]; | |
376 | ||
377 | /* Whether this path is/was involved in a directory/file conflict */ | |
378 | unsigned df_conflict:1; | |
379 | ||
1c7873cd EN |
380 | /* |
381 | * Whether this path is/was involved in a non-content conflict other | |
382 | * than a directory/file conflict (e.g. rename/rename, rename/delete, | |
383 | * file location based on possible directory rename). | |
384 | */ | |
385 | unsigned path_conflict:1; | |
386 | ||
5b59c3db EN |
387 | /* |
388 | * For filemask and dirmask, the ith bit corresponds to whether the | |
389 | * ith entry is a file (filemask) or a directory (dirmask). Thus, | |
390 | * filemask & dirmask is always zero, and filemask | dirmask is at | |
391 | * most 7 but can be less when a path does not appear as either a | |
392 | * file or a directory on at least one side of history. | |
393 | * | |
394 | * Note that these masks are related to enum merge_side, as the ith | |
395 | * entry corresponds to side i. | |
396 | * | |
397 | * These values come from a traverse_trees() call; more info may be | |
398 | * found looking at tree-walk.h's struct traverse_info, | |
399 | * particularly the documentation above the "fn" member (note that | |
400 | * filemask = mask & ~dirmask from that documentation). | |
401 | */ | |
402 | unsigned filemask:3; | |
403 | unsigned dirmask:3; | |
404 | ||
405 | /* | |
406 | * Optimization to track which stages match, to avoid the need to | |
407 | * recompute it in multiple steps. Either 0 or at least 2 bits are | |
408 | * set; if at least 2 bits are set, their corresponding stages match. | |
409 | */ | |
410 | unsigned match_mask:3; | |
411 | }; | |
412 | ||
04af1879 EN |
413 | /*** Function Grouping: various utility functions ***/ |
414 | ||
98bf9841 EN |
415 | /* |
416 | * For the next three macros, see warning for conflict_info.merged. | |
417 | * | |
418 | * In each of the below, mi is a struct merged_info*, and ci was defined | |
419 | * as a struct conflict_info* (but we need to verify ci isn't actually | |
420 | * pointed at a struct merged_info*). | |
421 | * | |
422 | * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise. | |
423 | * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one. | |
424 | * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first. | |
425 | */ | |
426 | #define INITIALIZE_CI(ci, mi) do { \ | |
427 | (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \ | |
428 | } while (0) | |
429 | #define VERIFY_CI(ci) assert(ci && !ci->merged.clean); | |
430 | #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \ | |
431 | (ci) = (struct conflict_info *)(mi); \ | |
432 | assert((ci) && !(mi)->clean); \ | |
433 | } while (0) | |
434 | ||
89422d29 EN |
435 | static void free_strmap_strings(struct strmap *map) |
436 | { | |
437 | struct hashmap_iter iter; | |
438 | struct strmap_entry *entry; | |
439 | ||
440 | strmap_for_each_entry(map, &iter, entry) { | |
441 | free((char*)entry->key); | |
442 | } | |
443 | } | |
444 | ||
43e9c4ee EN |
445 | static void clear_or_reinit_internal_opts(struct merge_options_internal *opti, |
446 | int reinitialize) | |
101bc5bc | 447 | { |
f5d9fbc2 EN |
448 | struct rename_info *renames = &opti->renames; |
449 | int i; | |
43e9c4ee EN |
450 | void (*strmap_func)(struct strmap *, int) = |
451 | reinitialize ? strmap_partial_clear : strmap_clear; | |
a49b55d5 EN |
452 | void (*strintmap_func)(struct strintmap *) = |
453 | reinitialize ? strintmap_partial_clear : strintmap_clear; | |
d29bd6d7 EN |
454 | void (*strset_func)(struct strset *) = |
455 | reinitialize ? strset_partial_clear : strset_clear; | |
101bc5bc EN |
456 | |
457 | /* | |
458 | * We marked opti->paths with strdup_strings = 0, so that we | |
459 | * wouldn't have to make another copy of the fullpath created by | |
460 | * make_traverse_path from setup_path_info(). But, now that we've | |
461 | * used it and have no other references to these strings, it is time | |
462 | * to deallocate them. | |
463 | */ | |
464 | free_strmap_strings(&opti->paths); | |
43e9c4ee | 465 | strmap_func(&opti->paths, 1); |
101bc5bc EN |
466 | |
467 | /* | |
468 | * All keys and values in opti->conflicted are a subset of those in | |
469 | * opti->paths. We don't want to deallocate anything twice, so we | |
470 | * don't free the keys and we pass 0 for free_values. | |
471 | */ | |
43e9c4ee | 472 | strmap_func(&opti->conflicted, 0); |
43c1dccb EN |
473 | |
474 | /* | |
475 | * opti->paths_to_free is similar to opti->paths; we created it with | |
476 | * strdup_strings = 0 to avoid making _another_ copy of the fullpath | |
477 | * but now that we've used it and have no other references to these | |
478 | * strings, it is time to deallocate them. We do so by temporarily | |
479 | * setting strdup_strings to 1. | |
480 | */ | |
481 | opti->paths_to_free.strdup_strings = 1; | |
482 | string_list_clear(&opti->paths_to_free, 0); | |
483 | opti->paths_to_free.strdup_strings = 0; | |
c5a6f655 | 484 | |
1218b3ab | 485 | if (opti->attr_index.cache_nr) /* true iff opt->renormalize */ |
ea305a68 EN |
486 | discard_index(&opti->attr_index); |
487 | ||
f5d9fbc2 EN |
488 | /* Free memory used by various renames maps */ |
489 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; ++i) { | |
a49b55d5 | 490 | strintmap_func(&renames->dirs_removed[i]); |
f5d9fbc2 | 491 | strmap_func(&renames->dir_renames[i], 0); |
a49b55d5 | 492 | strintmap_func(&renames->relevant_sources[i]); |
d5098029 EN |
493 | if (!reinitialize) |
494 | assert(renames->cached_pairs_valid_side == 0); | |
495 | if (i != renames->cached_pairs_valid_side) { | |
496 | strset_func(&renames->cached_target_names[i]); | |
497 | strmap_func(&renames->cached_pairs[i], 1); | |
498 | strset_func(&renames->cached_irrelevant[i]); | |
499 | partial_clear_dir_rename_count(&renames->dir_rename_count[i]); | |
500 | if (!reinitialize) | |
501 | strmap_clear(&renames->dir_rename_count[i], 1); | |
502 | } | |
f5d9fbc2 | 503 | } |
64aceb6d EN |
504 | renames->cached_pairs_valid_side = 0; |
505 | renames->dir_rename_mask = 0; | |
f5d9fbc2 | 506 | |
c5a6f655 EN |
507 | if (!reinitialize) { |
508 | struct hashmap_iter iter; | |
509 | struct strmap_entry *e; | |
510 | ||
511 | /* Release and free each strbuf found in output */ | |
512 | strmap_for_each_entry(&opti->output, &iter, e) { | |
513 | struct strbuf *sb = e->value; | |
514 | strbuf_release(sb); | |
515 | /* | |
516 | * While strictly speaking we don't need to free(sb) | |
517 | * here because we could pass free_values=1 when | |
518 | * calling strmap_clear() on opti->output, that would | |
519 | * require strmap_clear to do another | |
520 | * strmap_for_each_entry() loop, so we just free it | |
521 | * while we're iterating anyway. | |
522 | */ | |
523 | free(sb); | |
524 | } | |
525 | strmap_clear(&opti->output, 0); | |
526 | } | |
beb06145 EN |
527 | |
528 | /* Clean out callback_data as well. */ | |
529 | FREE_AND_NULL(renames->callback_data); | |
530 | renames->callback_data_nr = renames->callback_data_alloc = 0; | |
101bc5bc EN |
531 | } |
532 | ||
0c0d705b EN |
533 | static int err(struct merge_options *opt, const char *err, ...) |
534 | { | |
535 | va_list params; | |
536 | struct strbuf sb = STRBUF_INIT; | |
537 | ||
538 | strbuf_addstr(&sb, "error: "); | |
539 | va_start(params, err); | |
540 | strbuf_vaddf(&sb, err, params); | |
541 | va_end(params); | |
542 | ||
543 | error("%s", sb.buf); | |
544 | strbuf_release(&sb); | |
545 | ||
546 | return -1; | |
547 | } | |
548 | ||
c73cda76 EN |
549 | static void format_commit(struct strbuf *sb, |
550 | int indent, | |
551 | struct commit *commit) | |
552 | { | |
70f19c7f EN |
553 | struct merge_remote_desc *desc; |
554 | struct pretty_print_context ctx = {0}; | |
555 | ctx.abbrev = DEFAULT_ABBREV; | |
556 | ||
557 | strbuf_addchars(sb, ' ', indent); | |
558 | desc = merge_remote_util(commit); | |
559 | if (desc) { | |
560 | strbuf_addf(sb, "virtual %s\n", desc->name); | |
561 | return; | |
562 | } | |
563 | ||
564 | format_commit_message(commit, "%h %s", sb, &ctx); | |
565 | strbuf_addch(sb, '\n'); | |
c73cda76 EN |
566 | } |
567 | ||
c5a6f655 EN |
568 | __attribute__((format (printf, 4, 5))) |
569 | static void path_msg(struct merge_options *opt, | |
570 | const char *path, | |
571 | int omittable_hint, /* skippable under --remerge-diff */ | |
572 | const char *fmt, ...) | |
573 | { | |
574 | va_list ap; | |
575 | struct strbuf *sb = strmap_get(&opt->priv->output, path); | |
576 | if (!sb) { | |
577 | sb = xmalloc(sizeof(*sb)); | |
578 | strbuf_init(sb, 0); | |
579 | strmap_put(&opt->priv->output, path, sb); | |
580 | } | |
581 | ||
582 | va_start(ap, fmt); | |
583 | strbuf_vaddf(sb, fmt, ap); | |
584 | va_end(ap); | |
585 | ||
586 | strbuf_addch(sb, '\n'); | |
587 | } | |
588 | ||
5a1a1e8e EN |
589 | /* add a string to a strbuf, but converting "/" to "_" */ |
590 | static void add_flattened_path(struct strbuf *out, const char *s) | |
591 | { | |
592 | size_t i = out->len; | |
593 | strbuf_addstr(out, s); | |
594 | for (; i < out->len; i++) | |
595 | if (out->buf[i] == '/') | |
596 | out->buf[i] = '_'; | |
597 | } | |
598 | ||
23366d2a EN |
599 | static char *unique_path(struct strmap *existing_paths, |
600 | const char *path, | |
601 | const char *branch) | |
602 | { | |
5a1a1e8e EN |
603 | struct strbuf newpath = STRBUF_INIT; |
604 | int suffix = 0; | |
605 | size_t base_len; | |
606 | ||
607 | strbuf_addf(&newpath, "%s~", path); | |
608 | add_flattened_path(&newpath, branch); | |
609 | ||
610 | base_len = newpath.len; | |
611 | while (strmap_contains(existing_paths, newpath.buf)) { | |
612 | strbuf_setlen(&newpath, base_len); | |
613 | strbuf_addf(&newpath, "_%d", suffix++); | |
614 | } | |
615 | ||
616 | return strbuf_detach(&newpath, NULL); | |
23366d2a EN |
617 | } |
618 | ||
04af1879 EN |
619 | /*** Function Grouping: functions related to collect_merge_info() ***/ |
620 | ||
a68e6cea EN |
621 | static int traverse_trees_wrapper_callback(int n, |
622 | unsigned long mask, | |
623 | unsigned long dirmask, | |
624 | struct name_entry *names, | |
625 | struct traverse_info *info) | |
626 | { | |
627 | struct merge_options *opt = info->data; | |
628 | struct rename_info *renames = &opt->priv->renames; | |
2fd9eda4 | 629 | unsigned filemask = mask & ~dirmask; |
a68e6cea EN |
630 | |
631 | assert(n==3); | |
632 | ||
633 | if (!renames->callback_data_traverse_path) | |
634 | renames->callback_data_traverse_path = xstrdup(info->traverse_path); | |
635 | ||
2fd9eda4 EN |
636 | if (filemask && filemask == renames->dir_rename_mask) |
637 | renames->dir_rename_mask = 0x07; | |
638 | ||
a68e6cea EN |
639 | ALLOC_GROW(renames->callback_data, renames->callback_data_nr + 1, |
640 | renames->callback_data_alloc); | |
641 | renames->callback_data[renames->callback_data_nr].mask = mask; | |
642 | renames->callback_data[renames->callback_data_nr].dirmask = dirmask; | |
643 | COPY_ARRAY(renames->callback_data[renames->callback_data_nr].names, | |
644 | names, 3); | |
645 | renames->callback_data_nr++; | |
646 | ||
647 | return mask; | |
648 | } | |
649 | ||
650 | /* | |
651 | * Much like traverse_trees(), BUT: | |
652 | * - read all the tree entries FIRST, saving them | |
653 | * - note that the above step provides an opportunity to compute necessary | |
654 | * additional details before the "real" traversal | |
655 | * - loop through the saved entries and call the original callback on them | |
656 | */ | |
a68e6cea EN |
657 | static int traverse_trees_wrapper(struct index_state *istate, |
658 | int n, | |
659 | struct tree_desc *t, | |
660 | struct traverse_info *info) | |
661 | { | |
662 | int ret, i, old_offset; | |
663 | traverse_callback_t old_fn; | |
664 | char *old_callback_data_traverse_path; | |
665 | struct merge_options *opt = info->data; | |
666 | struct rename_info *renames = &opt->priv->renames; | |
667 | ||
2fd9eda4 EN |
668 | assert(renames->dir_rename_mask == 2 || renames->dir_rename_mask == 4); |
669 | ||
a68e6cea EN |
670 | old_callback_data_traverse_path = renames->callback_data_traverse_path; |
671 | old_fn = info->fn; | |
672 | old_offset = renames->callback_data_nr; | |
673 | ||
674 | renames->callback_data_traverse_path = NULL; | |
675 | info->fn = traverse_trees_wrapper_callback; | |
676 | ret = traverse_trees(istate, n, t, info); | |
677 | if (ret < 0) | |
678 | return ret; | |
679 | ||
680 | info->traverse_path = renames->callback_data_traverse_path; | |
681 | info->fn = old_fn; | |
682 | for (i = old_offset; i < renames->callback_data_nr; ++i) { | |
683 | info->fn(n, | |
684 | renames->callback_data[i].mask, | |
685 | renames->callback_data[i].dirmask, | |
686 | renames->callback_data[i].names, | |
687 | info); | |
688 | } | |
689 | ||
690 | renames->callback_data_nr = old_offset; | |
691 | free(renames->callback_data_traverse_path); | |
692 | renames->callback_data_traverse_path = old_callback_data_traverse_path; | |
693 | info->traverse_path = NULL; | |
694 | return 0; | |
695 | } | |
696 | ||
98bf9841 EN |
697 | static void setup_path_info(struct merge_options *opt, |
698 | struct string_list_item *result, | |
699 | const char *current_dir_name, | |
700 | int current_dir_name_len, | |
701 | char *fullpath, /* we'll take over ownership */ | |
702 | struct name_entry *names, | |
703 | struct name_entry *merged_version, | |
704 | unsigned is_null, /* boolean */ | |
705 | unsigned df_conflict, /* boolean */ | |
706 | unsigned filemask, | |
707 | unsigned dirmask, | |
708 | int resolved /* boolean */) | |
709 | { | |
710 | /* result->util is void*, so mi is a convenience typed variable */ | |
711 | struct merged_info *mi; | |
712 | ||
713 | assert(!is_null || resolved); | |
714 | assert(!df_conflict || !resolved); /* df_conflict implies !resolved */ | |
715 | assert(resolved == (merged_version != NULL)); | |
716 | ||
717 | mi = xcalloc(1, resolved ? sizeof(struct merged_info) : | |
718 | sizeof(struct conflict_info)); | |
719 | mi->directory_name = current_dir_name; | |
720 | mi->basename_offset = current_dir_name_len; | |
721 | mi->clean = !!resolved; | |
722 | if (resolved) { | |
723 | mi->result.mode = merged_version->mode; | |
724 | oidcpy(&mi->result.oid, &merged_version->oid); | |
725 | mi->is_null = !!is_null; | |
726 | } else { | |
727 | int i; | |
728 | struct conflict_info *ci; | |
729 | ||
730 | ASSIGN_AND_VERIFY_CI(ci, mi); | |
731 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
732 | ci->pathnames[i] = fullpath; | |
733 | ci->stages[i].mode = names[i].mode; | |
734 | oidcpy(&ci->stages[i].oid, &names[i].oid); | |
735 | } | |
736 | ci->filemask = filemask; | |
737 | ci->dirmask = dirmask; | |
738 | ci->df_conflict = !!df_conflict; | |
739 | if (dirmask) | |
740 | /* | |
741 | * Assume is_null for now, but if we have entries | |
742 | * under the directory then when it is complete in | |
743 | * write_completed_directory() it'll update this. | |
744 | * Also, for D/F conflicts, we have to handle the | |
745 | * directory first, then clear this bit and process | |
746 | * the file to see how it is handled -- that occurs | |
747 | * near the top of process_entry(). | |
748 | */ | |
749 | mi->is_null = 1; | |
750 | } | |
751 | strmap_put(&opt->priv->paths, fullpath, mi); | |
752 | result->string = fullpath; | |
753 | result->util = mi; | |
754 | } | |
755 | ||
f78cf976 EN |
756 | static void add_pair(struct merge_options *opt, |
757 | struct name_entry *names, | |
758 | const char *pathname, | |
759 | unsigned side, | |
32a56dfb | 760 | unsigned is_add /* if false, is_delete */, |
2fd9eda4 EN |
761 | unsigned match_mask, |
762 | unsigned dir_rename_mask) | |
f78cf976 EN |
763 | { |
764 | struct diff_filespec *one, *two; | |
765 | struct rename_info *renames = &opt->priv->renames; | |
766 | int names_idx = is_add ? side : 0; | |
767 | ||
25e65b6d | 768 | if (is_add) { |
ef68c3d8 | 769 | assert(match_mask == 0 || match_mask == 6); |
25e65b6d EN |
770 | if (strset_contains(&renames->cached_target_names[side], |
771 | pathname)) | |
772 | return; | |
773 | } else { | |
32a56dfb | 774 | unsigned content_relevant = (match_mask == 0); |
2fd9eda4 | 775 | unsigned location_relevant = (dir_rename_mask == 0x07); |
32a56dfb | 776 | |
ef68c3d8 EN |
777 | assert(match_mask == 0 || match_mask == 3 || match_mask == 5); |
778 | ||
25e65b6d EN |
779 | /* |
780 | * If pathname is found in cached_irrelevant[side] due to | |
781 | * previous pick but for this commit content is relevant, | |
782 | * then we need to remove it from cached_irrelevant. | |
783 | */ | |
784 | if (content_relevant) | |
785 | /* strset_remove is no-op if strset doesn't have key */ | |
786 | strset_remove(&renames->cached_irrelevant[side], | |
787 | pathname); | |
788 | ||
789 | /* | |
790 | * We do not need to re-detect renames for paths that we already | |
791 | * know the pairing, i.e. for cached_pairs (or | |
792 | * cached_irrelevant). However, handle_deferred_entries() needs | |
793 | * to loop over the union of keys from relevant_sources[side] and | |
794 | * cached_pairs[side], so for simplicity we set relevant_sources | |
795 | * for all the cached_pairs too and then strip them back out in | |
796 | * prune_cached_from_relevant() at the beginning of | |
797 | * detect_regular_renames(). | |
798 | */ | |
ec59da60 EN |
799 | if (content_relevant || location_relevant) { |
800 | /* content_relevant trumps location_relevant */ | |
801 | strintmap_set(&renames->relevant_sources[side], pathname, | |
802 | content_relevant ? RELEVANT_CONTENT : RELEVANT_LOCATION); | |
803 | } | |
25e65b6d EN |
804 | |
805 | /* | |
806 | * Avoid creating pair if we've already cached rename results. | |
807 | * Note that we do this after setting relevant_sources[side] | |
808 | * as noted in the comment above. | |
809 | */ | |
810 | if (strmap_contains(&renames->cached_pairs[side], pathname) || | |
811 | strset_contains(&renames->cached_irrelevant[side], pathname)) | |
812 | return; | |
32a56dfb EN |
813 | } |
814 | ||
f78cf976 EN |
815 | one = alloc_filespec(pathname); |
816 | two = alloc_filespec(pathname); | |
817 | fill_filespec(is_add ? two : one, | |
818 | &names[names_idx].oid, 1, names[names_idx].mode); | |
819 | diff_queue(&renames->pairs[side], one, two); | |
820 | } | |
821 | ||
eb3e3e1d EN |
822 | static void collect_rename_info(struct merge_options *opt, |
823 | struct name_entry *names, | |
824 | const char *dirname, | |
825 | const char *fullname, | |
826 | unsigned filemask, | |
827 | unsigned dirmask, | |
828 | unsigned match_mask) | |
829 | { | |
830 | struct rename_info *renames = &opt->priv->renames; | |
f78cf976 | 831 | unsigned side; |
eb3e3e1d | 832 | |
2fd9eda4 EN |
833 | /* |
834 | * Update dir_rename_mask (determines ignore-rename-source validity) | |
835 | * | |
836 | * dir_rename_mask helps us keep track of when directory rename | |
837 | * detection may be relevant. Basically, whenver a directory is | |
838 | * removed on one side of history, and a file is added to that | |
839 | * directory on the other side of history, directory rename | |
840 | * detection is relevant (meaning we have to detect renames for all | |
841 | * files within that directory to deduce where the directory | |
842 | * moved). Also, whenever a directory needs directory rename | |
843 | * detection, due to the "majority rules" choice for where to move | |
844 | * it (see t6423 testcase 1f), we also need to detect renames for | |
845 | * all files within subdirectories of that directory as well. | |
846 | * | |
847 | * Here we haven't looked at files within the directory yet, we are | |
848 | * just looking at the directory itself. So, if we aren't yet in | |
849 | * a case where a parent directory needed directory rename detection | |
850 | * (i.e. dir_rename_mask != 0x07), and if the directory was removed | |
851 | * on one side of history, record the mask of the other side of | |
852 | * history in dir_rename_mask. | |
853 | */ | |
854 | if (renames->dir_rename_mask != 0x07 && | |
855 | (dirmask == 3 || dirmask == 5)) { | |
856 | /* simple sanity check */ | |
857 | assert(renames->dir_rename_mask == 0 || | |
858 | renames->dir_rename_mask == (dirmask & ~1)); | |
859 | /* update dir_rename_mask; have it record mask of new side */ | |
860 | renames->dir_rename_mask = (dirmask & ~1); | |
861 | } | |
862 | ||
eb3e3e1d EN |
863 | /* Update dirs_removed, as needed */ |
864 | if (dirmask == 1 || dirmask == 3 || dirmask == 5) { | |
865 | /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */ | |
866 | unsigned sides = (0x07 - dirmask)/2; | |
fb52938e EN |
867 | unsigned relevance = (renames->dir_rename_mask == 0x07) ? |
868 | RELEVANT_FOR_ANCESTOR : NOT_RELEVANT; | |
869 | /* | |
870 | * Record relevance of this directory. However, note that | |
871 | * when collect_merge_info_callback() recurses into this | |
872 | * directory and calls collect_rename_info() on paths | |
873 | * within that directory, if we find a path that was added | |
874 | * to this directory on the other side of history, we will | |
875 | * upgrade this value to RELEVANT_FOR_SELF; see below. | |
876 | */ | |
eb3e3e1d | 877 | if (sides & 1) |
fb52938e EN |
878 | strintmap_set(&renames->dirs_removed[1], fullname, |
879 | relevance); | |
eb3e3e1d | 880 | if (sides & 2) |
fb52938e EN |
881 | strintmap_set(&renames->dirs_removed[2], fullname, |
882 | relevance); | |
883 | } | |
884 | ||
885 | /* | |
886 | * Here's the block that potentially upgrades to RELEVANT_FOR_SELF. | |
887 | * When we run across a file added to a directory. In such a case, | |
888 | * find the directory of the file and upgrade its relevance. | |
889 | */ | |
890 | if (renames->dir_rename_mask == 0x07 && | |
891 | (filemask == 2 || filemask == 4)) { | |
892 | /* | |
893 | * Need directory rename for parent directory on other side | |
894 | * of history from added file. Thus | |
895 | * side = (~filemask & 0x06) >> 1 | |
896 | * or | |
897 | * side = 3 - (filemask/2). | |
898 | */ | |
899 | unsigned side = 3 - (filemask >> 1); | |
900 | strintmap_set(&renames->dirs_removed[side], dirname, | |
901 | RELEVANT_FOR_SELF); | |
eb3e3e1d | 902 | } |
f78cf976 EN |
903 | |
904 | if (filemask == 0 || filemask == 7) | |
905 | return; | |
906 | ||
907 | for (side = MERGE_SIDE1; side <= MERGE_SIDE2; ++side) { | |
908 | unsigned side_mask = (1 << side); | |
909 | ||
910 | /* Check for deletion on side */ | |
911 | if ((filemask & 1) && !(filemask & side_mask)) | |
32a56dfb | 912 | add_pair(opt, names, fullname, side, 0 /* delete */, |
2fd9eda4 EN |
913 | match_mask & filemask, |
914 | renames->dir_rename_mask); | |
f78cf976 EN |
915 | |
916 | /* Check for addition on side */ | |
917 | if (!(filemask & 1) && (filemask & side_mask)) | |
32a56dfb | 918 | add_pair(opt, names, fullname, side, 1 /* add */, |
2fd9eda4 EN |
919 | match_mask & filemask, |
920 | renames->dir_rename_mask); | |
f78cf976 | 921 | } |
eb3e3e1d EN |
922 | } |
923 | ||
d2bc1994 EN |
924 | static int collect_merge_info_callback(int n, |
925 | unsigned long mask, | |
926 | unsigned long dirmask, | |
927 | struct name_entry *names, | |
928 | struct traverse_info *info) | |
929 | { | |
930 | /* | |
931 | * n is 3. Always. | |
932 | * common ancestor (mbase) has mask 1, and stored in index 0 of names | |
933 | * head of side 1 (side1) has mask 2, and stored in index 1 of names | |
934 | * head of side 2 (side2) has mask 4, and stored in index 2 of names | |
935 | */ | |
936 | struct merge_options *opt = info->data; | |
937 | struct merge_options_internal *opti = opt->priv; | |
2fd9eda4 | 938 | struct rename_info *renames = &opt->priv->renames; |
98bf9841 EN |
939 | struct string_list_item pi; /* Path Info */ |
940 | struct conflict_info *ci; /* typed alias to pi.util (which is void*) */ | |
d2bc1994 EN |
941 | struct name_entry *p; |
942 | size_t len; | |
943 | char *fullpath; | |
98bf9841 | 944 | const char *dirname = opti->current_dir_name; |
2fd9eda4 | 945 | unsigned prev_dir_rename_mask = renames->dir_rename_mask; |
d2bc1994 | 946 | unsigned filemask = mask & ~dirmask; |
34e557af | 947 | unsigned match_mask = 0; /* will be updated below */ |
d2bc1994 EN |
948 | unsigned mbase_null = !(mask & 1); |
949 | unsigned side1_null = !(mask & 2); | |
950 | unsigned side2_null = !(mask & 4); | |
885f0063 EN |
951 | unsigned side1_matches_mbase = (!side1_null && !mbase_null && |
952 | names[0].mode == names[1].mode && | |
953 | oideq(&names[0].oid, &names[1].oid)); | |
954 | unsigned side2_matches_mbase = (!side2_null && !mbase_null && | |
955 | names[0].mode == names[2].mode && | |
956 | oideq(&names[0].oid, &names[2].oid)); | |
957 | unsigned sides_match = (!side1_null && !side2_null && | |
958 | names[1].mode == names[2].mode && | |
959 | oideq(&names[1].oid, &names[2].oid)); | |
d2bc1994 | 960 | |
34e557af EN |
961 | /* |
962 | * Note: When a path is a file on one side of history and a directory | |
963 | * in another, we have a directory/file conflict. In such cases, if | |
964 | * the conflict doesn't resolve from renames and deletions, then we | |
965 | * always leave directories where they are and move files out of the | |
966 | * way. Thus, while struct conflict_info has a df_conflict field to | |
967 | * track such conflicts, we ignore that field for any directories at | |
968 | * a path and only pay attention to it for files at the given path. | |
969 | * The fact that we leave directories were they are also means that | |
970 | * we do not need to worry about getting additional df_conflict | |
971 | * information propagated from parent directories down to children | |
972 | * (unlike, say traverse_trees_recursive() in unpack-trees.c, which | |
973 | * sets a newinfo.df_conflicts field specifically to propagate it). | |
974 | */ | |
975 | unsigned df_conflict = (filemask != 0) && (dirmask != 0); | |
976 | ||
d2bc1994 EN |
977 | /* n = 3 is a fundamental assumption. */ |
978 | if (n != 3) | |
979 | BUG("Called collect_merge_info_callback wrong"); | |
980 | ||
981 | /* | |
982 | * A bunch of sanity checks verifying that traverse_trees() calls | |
983 | * us the way I expect. Could just remove these at some point, | |
984 | * though maybe they are helpful to future code readers. | |
985 | */ | |
986 | assert(mbase_null == is_null_oid(&names[0].oid)); | |
987 | assert(side1_null == is_null_oid(&names[1].oid)); | |
988 | assert(side2_null == is_null_oid(&names[2].oid)); | |
989 | assert(!mbase_null || !side1_null || !side2_null); | |
990 | assert(mask > 0 && mask < 8); | |
991 | ||
34e557af EN |
992 | /* Determine match_mask */ |
993 | if (side1_matches_mbase) | |
994 | match_mask = (side2_matches_mbase ? 7 : 3); | |
995 | else if (side2_matches_mbase) | |
996 | match_mask = 5; | |
997 | else if (sides_match) | |
998 | match_mask = 6; | |
999 | ||
d2bc1994 EN |
1000 | /* |
1001 | * Get the name of the relevant filepath, which we'll pass to | |
1002 | * setup_path_info() for tracking. | |
1003 | */ | |
1004 | p = names; | |
1005 | while (!p->mode) | |
1006 | p++; | |
1007 | len = traverse_path_len(info, p->pathlen); | |
1008 | ||
1009 | /* +1 in both of the following lines to include the NUL byte */ | |
1010 | fullpath = xmalloc(len + 1); | |
1011 | make_traverse_path(fullpath, len + 1, info, p->path, p->pathlen); | |
1012 | ||
291f29ca EN |
1013 | /* |
1014 | * If mbase, side1, and side2 all match, we can resolve early. Even | |
1015 | * if these are trees, there will be no renames or anything | |
1016 | * underneath. | |
1017 | */ | |
1018 | if (side1_matches_mbase && side2_matches_mbase) { | |
1019 | /* mbase, side1, & side2 all match; use mbase as resolution */ | |
1020 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
1021 | names, names+0, mbase_null, 0, | |
1022 | filemask, dirmask, 1); | |
1023 | return mask; | |
1024 | } | |
1025 | ||
eb3e3e1d EN |
1026 | /* |
1027 | * Gather additional information used in rename detection. | |
1028 | */ | |
1029 | collect_rename_info(opt, names, dirname, fullpath, | |
1030 | filemask, dirmask, match_mask); | |
1031 | ||
d2bc1994 | 1032 | /* |
98bf9841 EN |
1033 | * Record information about the path so we can resolve later in |
1034 | * process_entries. | |
d2bc1994 | 1035 | */ |
98bf9841 EN |
1036 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, |
1037 | names, NULL, 0, df_conflict, filemask, dirmask, 0); | |
1038 | ||
1039 | ci = pi.util; | |
1040 | VERIFY_CI(ci); | |
34e557af | 1041 | ci->match_mask = match_mask; |
d2bc1994 EN |
1042 | |
1043 | /* If dirmask, recurse into subdirectories */ | |
1044 | if (dirmask) { | |
1045 | struct traverse_info newinfo; | |
1046 | struct tree_desc t[3]; | |
1047 | void *buf[3] = {NULL, NULL, NULL}; | |
1048 | const char *original_dir_name; | |
1049 | int i, ret; | |
1050 | ||
1051 | ci->match_mask &= filemask; | |
1052 | newinfo = *info; | |
1053 | newinfo.prev = info; | |
1054 | newinfo.name = p->path; | |
1055 | newinfo.namelen = p->pathlen; | |
1056 | newinfo.pathlen = st_add3(newinfo.pathlen, p->pathlen, 1); | |
34e557af EN |
1057 | /* |
1058 | * If this directory we are about to recurse into cared about | |
1059 | * its parent directory (the current directory) having a D/F | |
1060 | * conflict, then we'd propagate the masks in this way: | |
1061 | * newinfo.df_conflicts |= (mask & ~dirmask); | |
1062 | * But we don't worry about propagating D/F conflicts. (See | |
1063 | * comment near setting of local df_conflict variable near | |
1064 | * the beginning of this function). | |
1065 | */ | |
d2bc1994 EN |
1066 | |
1067 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
885f0063 EN |
1068 | if (i == 1 && side1_matches_mbase) |
1069 | t[1] = t[0]; | |
1070 | else if (i == 2 && side2_matches_mbase) | |
1071 | t[2] = t[0]; | |
1072 | else if (i == 2 && sides_match) | |
1073 | t[2] = t[1]; | |
1074 | else { | |
1075 | const struct object_id *oid = NULL; | |
1076 | if (dirmask & 1) | |
1077 | oid = &names[i].oid; | |
1078 | buf[i] = fill_tree_descriptor(opt->repo, | |
1079 | t + i, oid); | |
1080 | } | |
d2bc1994 EN |
1081 | dirmask >>= 1; |
1082 | } | |
1083 | ||
1084 | original_dir_name = opti->current_dir_name; | |
98bf9841 | 1085 | opti->current_dir_name = pi.string; |
2fd9eda4 EN |
1086 | if (renames->dir_rename_mask == 0 || |
1087 | renames->dir_rename_mask == 0x07) | |
1088 | ret = traverse_trees(NULL, 3, t, &newinfo); | |
1089 | else | |
1090 | ret = traverse_trees_wrapper(NULL, 3, t, &newinfo); | |
d2bc1994 | 1091 | opti->current_dir_name = original_dir_name; |
2fd9eda4 | 1092 | renames->dir_rename_mask = prev_dir_rename_mask; |
d2bc1994 EN |
1093 | |
1094 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) | |
1095 | free(buf[i]); | |
1096 | ||
1097 | if (ret < 0) | |
1098 | return -1; | |
1099 | } | |
1100 | ||
1101 | return mask; | |
1102 | } | |
1103 | ||
231e2dd4 EN |
1104 | static int collect_merge_info(struct merge_options *opt, |
1105 | struct tree *merge_base, | |
1106 | struct tree *side1, | |
1107 | struct tree *side2) | |
1108 | { | |
d2bc1994 EN |
1109 | int ret; |
1110 | struct tree_desc t[3]; | |
1111 | struct traverse_info info; | |
d2bc1994 | 1112 | |
05b85c6e EN |
1113 | opt->priv->toplevel_dir = ""; |
1114 | opt->priv->current_dir_name = opt->priv->toplevel_dir; | |
1115 | setup_traverse_info(&info, opt->priv->toplevel_dir); | |
d2bc1994 EN |
1116 | info.fn = collect_merge_info_callback; |
1117 | info.data = opt; | |
1118 | info.show_all_errors = 1; | |
1119 | ||
1120 | parse_tree(merge_base); | |
1121 | parse_tree(side1); | |
1122 | parse_tree(side2); | |
1123 | init_tree_desc(t + 0, merge_base->buffer, merge_base->size); | |
1124 | init_tree_desc(t + 1, side1->buffer, side1->size); | |
1125 | init_tree_desc(t + 2, side2->buffer, side2->size); | |
1126 | ||
557ac035 | 1127 | trace2_region_enter("merge", "traverse_trees", opt->repo); |
d2bc1994 | 1128 | ret = traverse_trees(NULL, 3, t, &info); |
557ac035 | 1129 | trace2_region_leave("merge", "traverse_trees", opt->repo); |
d2bc1994 EN |
1130 | |
1131 | return ret; | |
231e2dd4 EN |
1132 | } |
1133 | ||
04af1879 EN |
1134 | /*** Function Grouping: functions related to threeway content merges ***/ |
1135 | ||
c73cda76 EN |
1136 | static int find_first_merges(struct repository *repo, |
1137 | const char *path, | |
1138 | struct commit *a, | |
1139 | struct commit *b, | |
1140 | struct object_array *result) | |
1141 | { | |
4204cd59 EN |
1142 | int i, j; |
1143 | struct object_array merges = OBJECT_ARRAY_INIT; | |
1144 | struct commit *commit; | |
1145 | int contains_another; | |
1146 | ||
1147 | char merged_revision[GIT_MAX_HEXSZ + 2]; | |
1148 | const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path", | |
1149 | "--all", merged_revision, NULL }; | |
1150 | struct rev_info revs; | |
1151 | struct setup_revision_opt rev_opts; | |
1152 | ||
1153 | memset(result, 0, sizeof(struct object_array)); | |
1154 | memset(&rev_opts, 0, sizeof(rev_opts)); | |
1155 | ||
1156 | /* get all revisions that merge commit a */ | |
1157 | xsnprintf(merged_revision, sizeof(merged_revision), "^%s", | |
1158 | oid_to_hex(&a->object.oid)); | |
1159 | repo_init_revisions(repo, &revs, NULL); | |
1160 | rev_opts.submodule = path; | |
1161 | /* FIXME: can't handle linked worktrees in submodules yet */ | |
1162 | revs.single_worktree = path != NULL; | |
1163 | setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts); | |
1164 | ||
1165 | /* save all revisions from the above list that contain b */ | |
1166 | if (prepare_revision_walk(&revs)) | |
1167 | die("revision walk setup failed"); | |
1168 | while ((commit = get_revision(&revs)) != NULL) { | |
1169 | struct object *o = &(commit->object); | |
1170 | if (in_merge_bases(b, commit)) | |
1171 | add_object_array(o, NULL, &merges); | |
1172 | } | |
1173 | reset_revision_walk(); | |
1174 | ||
1175 | /* Now we've got all merges that contain a and b. Prune all | |
1176 | * merges that contain another found merge and save them in | |
1177 | * result. | |
1178 | */ | |
1179 | for (i = 0; i < merges.nr; i++) { | |
1180 | struct commit *m1 = (struct commit *) merges.objects[i].item; | |
1181 | ||
1182 | contains_another = 0; | |
1183 | for (j = 0; j < merges.nr; j++) { | |
1184 | struct commit *m2 = (struct commit *) merges.objects[j].item; | |
1185 | if (i != j && in_merge_bases(m2, m1)) { | |
1186 | contains_another = 1; | |
1187 | break; | |
1188 | } | |
1189 | } | |
1190 | ||
1191 | if (!contains_another) | |
1192 | add_object_array(merges.objects[i].item, NULL, result); | |
1193 | } | |
1194 | ||
1195 | object_array_clear(&merges); | |
1196 | return result->nr; | |
c73cda76 EN |
1197 | } |
1198 | ||
62fdec17 EN |
1199 | static int merge_submodule(struct merge_options *opt, |
1200 | const char *path, | |
1201 | const struct object_id *o, | |
1202 | const struct object_id *a, | |
1203 | const struct object_id *b, | |
1204 | struct object_id *result) | |
1205 | { | |
c73cda76 EN |
1206 | struct commit *commit_o, *commit_a, *commit_b; |
1207 | int parent_count; | |
1208 | struct object_array merges; | |
1209 | struct strbuf sb = STRBUF_INIT; | |
1210 | ||
1211 | int i; | |
1212 | int search = !opt->priv->call_depth; | |
1213 | ||
1214 | /* store fallback answer in result in case we fail */ | |
1215 | oidcpy(result, opt->priv->call_depth ? o : a); | |
1216 | ||
1217 | /* we can not handle deletion conflicts */ | |
1218 | if (is_null_oid(o)) | |
1219 | return 0; | |
1220 | if (is_null_oid(a)) | |
1221 | return 0; | |
1222 | if (is_null_oid(b)) | |
1223 | return 0; | |
1224 | ||
1225 | if (add_submodule_odb(path)) { | |
1226 | path_msg(opt, path, 0, | |
1227 | _("Failed to merge submodule %s (not checked out)"), | |
1228 | path); | |
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | if (!(commit_o = lookup_commit_reference(opt->repo, o)) || | |
1233 | !(commit_a = lookup_commit_reference(opt->repo, a)) || | |
1234 | !(commit_b = lookup_commit_reference(opt->repo, b))) { | |
1235 | path_msg(opt, path, 0, | |
1236 | _("Failed to merge submodule %s (commits not present)"), | |
1237 | path); | |
1238 | return 0; | |
1239 | } | |
1240 | ||
1241 | /* check whether both changes are forward */ | |
1242 | if (!in_merge_bases(commit_o, commit_a) || | |
1243 | !in_merge_bases(commit_o, commit_b)) { | |
1244 | path_msg(opt, path, 0, | |
1245 | _("Failed to merge submodule %s " | |
1246 | "(commits don't follow merge-base)"), | |
1247 | path); | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | /* Case #1: a is contained in b or vice versa */ | |
1252 | if (in_merge_bases(commit_a, commit_b)) { | |
1253 | oidcpy(result, b); | |
1254 | path_msg(opt, path, 1, | |
1255 | _("Note: Fast-forwarding submodule %s to %s"), | |
1256 | path, oid_to_hex(b)); | |
1257 | return 1; | |
1258 | } | |
1259 | if (in_merge_bases(commit_b, commit_a)) { | |
1260 | oidcpy(result, a); | |
1261 | path_msg(opt, path, 1, | |
1262 | _("Note: Fast-forwarding submodule %s to %s"), | |
1263 | path, oid_to_hex(a)); | |
1264 | return 1; | |
1265 | } | |
1266 | ||
1267 | /* | |
1268 | * Case #2: There are one or more merges that contain a and b in | |
1269 | * the submodule. If there is only one, then present it as a | |
1270 | * suggestion to the user, but leave it marked unmerged so the | |
1271 | * user needs to confirm the resolution. | |
1272 | */ | |
1273 | ||
1274 | /* Skip the search if makes no sense to the calling context. */ | |
1275 | if (!search) | |
1276 | return 0; | |
1277 | ||
1278 | /* find commit which merges them */ | |
1279 | parent_count = find_first_merges(opt->repo, path, commit_a, commit_b, | |
1280 | &merges); | |
1281 | switch (parent_count) { | |
1282 | case 0: | |
1283 | path_msg(opt, path, 0, _("Failed to merge submodule %s"), path); | |
1284 | break; | |
1285 | ||
1286 | case 1: | |
1287 | format_commit(&sb, 4, | |
1288 | (struct commit *)merges.objects[0].item); | |
1289 | path_msg(opt, path, 0, | |
1290 | _("Failed to merge submodule %s, but a possible merge " | |
1291 | "resolution exists:\n%s\n"), | |
1292 | path, sb.buf); | |
1293 | path_msg(opt, path, 1, | |
1294 | _("If this is correct simply add it to the index " | |
1295 | "for example\n" | |
1296 | "by using:\n\n" | |
1297 | " git update-index --cacheinfo 160000 %s \"%s\"\n\n" | |
1298 | "which will accept this suggestion.\n"), | |
1299 | oid_to_hex(&merges.objects[0].item->oid), path); | |
1300 | strbuf_release(&sb); | |
1301 | break; | |
1302 | default: | |
1303 | for (i = 0; i < merges.nr; i++) | |
1304 | format_commit(&sb, 4, | |
1305 | (struct commit *)merges.objects[i].item); | |
1306 | path_msg(opt, path, 0, | |
1307 | _("Failed to merge submodule %s, but multiple " | |
1308 | "possible merges exist:\n%s"), path, sb.buf); | |
1309 | strbuf_release(&sb); | |
1310 | } | |
1311 | ||
1312 | object_array_clear(&merges); | |
1313 | return 0; | |
62fdec17 EN |
1314 | } |
1315 | ||
1218b3ab EN |
1316 | static void initialize_attr_index(struct merge_options *opt) |
1317 | { | |
1318 | /* | |
1319 | * The renormalize_buffer() functions require attributes, and | |
1320 | * annoyingly those can only be read from the working tree or from | |
1321 | * an index_state. merge-ort doesn't have an index_state, so we | |
1322 | * generate a fake one containing only attribute information. | |
1323 | */ | |
1324 | struct merged_info *mi; | |
1325 | struct index_state *attr_index = &opt->priv->attr_index; | |
1326 | struct cache_entry *ce; | |
1327 | ||
1328 | attr_index->initialized = 1; | |
1329 | ||
1330 | if (!opt->renormalize) | |
1331 | return; | |
1332 | ||
1333 | mi = strmap_get(&opt->priv->paths, GITATTRIBUTES_FILE); | |
1334 | if (!mi) | |
1335 | return; | |
1336 | ||
1337 | if (mi->clean) { | |
1338 | int len = strlen(GITATTRIBUTES_FILE); | |
1339 | ce = make_empty_cache_entry(attr_index, len); | |
1340 | ce->ce_mode = create_ce_mode(mi->result.mode); | |
1341 | ce->ce_flags = create_ce_flags(0); | |
1342 | ce->ce_namelen = len; | |
1343 | oidcpy(&ce->oid, &mi->result.oid); | |
1344 | memcpy(ce->name, GITATTRIBUTES_FILE, len); | |
1345 | add_index_entry(attr_index, ce, | |
1346 | ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE); | |
1347 | get_stream_filter(attr_index, GITATTRIBUTES_FILE, &ce->oid); | |
1348 | } else { | |
1349 | int stage, len; | |
1350 | struct conflict_info *ci; | |
1351 | ||
1352 | ASSIGN_AND_VERIFY_CI(ci, mi); | |
1353 | for (stage = 0; stage < 3; stage++) { | |
1354 | unsigned stage_mask = (1 << stage); | |
1355 | ||
1356 | if (!(ci->filemask & stage_mask)) | |
1357 | continue; | |
1358 | len = strlen(GITATTRIBUTES_FILE); | |
1359 | ce = make_empty_cache_entry(attr_index, len); | |
1360 | ce->ce_mode = create_ce_mode(ci->stages[stage].mode); | |
1361 | ce->ce_flags = create_ce_flags(stage); | |
1362 | ce->ce_namelen = len; | |
1363 | oidcpy(&ce->oid, &ci->stages[stage].oid); | |
1364 | memcpy(ce->name, GITATTRIBUTES_FILE, len); | |
1365 | add_index_entry(attr_index, ce, | |
1366 | ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE); | |
1367 | get_stream_filter(attr_index, GITATTRIBUTES_FILE, | |
1368 | &ce->oid); | |
1369 | } | |
1370 | } | |
1371 | } | |
1372 | ||
62fdec17 EN |
1373 | static int merge_3way(struct merge_options *opt, |
1374 | const char *path, | |
1375 | const struct object_id *o, | |
1376 | const struct object_id *a, | |
1377 | const struct object_id *b, | |
1378 | const char *pathnames[3], | |
1379 | const int extra_marker_size, | |
1380 | mmbuffer_t *result_buf) | |
1381 | { | |
f591c472 EN |
1382 | mmfile_t orig, src1, src2; |
1383 | struct ll_merge_options ll_opts = {0}; | |
1384 | char *base, *name1, *name2; | |
1385 | int merge_status; | |
1386 | ||
1218b3ab EN |
1387 | if (!opt->priv->attr_index.initialized) |
1388 | initialize_attr_index(opt); | |
1389 | ||
f591c472 EN |
1390 | ll_opts.renormalize = opt->renormalize; |
1391 | ll_opts.extra_marker_size = extra_marker_size; | |
1392 | ll_opts.xdl_opts = opt->xdl_opts; | |
1393 | ||
1394 | if (opt->priv->call_depth) { | |
1395 | ll_opts.virtual_ancestor = 1; | |
1396 | ll_opts.variant = 0; | |
1397 | } else { | |
1398 | switch (opt->recursive_variant) { | |
1399 | case MERGE_VARIANT_OURS: | |
1400 | ll_opts.variant = XDL_MERGE_FAVOR_OURS; | |
1401 | break; | |
1402 | case MERGE_VARIANT_THEIRS: | |
1403 | ll_opts.variant = XDL_MERGE_FAVOR_THEIRS; | |
1404 | break; | |
1405 | default: | |
1406 | ll_opts.variant = 0; | |
1407 | break; | |
1408 | } | |
1409 | } | |
1410 | ||
1411 | assert(pathnames[0] && pathnames[1] && pathnames[2] && opt->ancestor); | |
1412 | if (pathnames[0] == pathnames[1] && pathnames[1] == pathnames[2]) { | |
1413 | base = mkpathdup("%s", opt->ancestor); | |
1414 | name1 = mkpathdup("%s", opt->branch1); | |
1415 | name2 = mkpathdup("%s", opt->branch2); | |
1416 | } else { | |
1417 | base = mkpathdup("%s:%s", opt->ancestor, pathnames[0]); | |
1418 | name1 = mkpathdup("%s:%s", opt->branch1, pathnames[1]); | |
1419 | name2 = mkpathdup("%s:%s", opt->branch2, pathnames[2]); | |
1420 | } | |
1421 | ||
1422 | read_mmblob(&orig, o); | |
1423 | read_mmblob(&src1, a); | |
1424 | read_mmblob(&src2, b); | |
1425 | ||
1426 | merge_status = ll_merge(result_buf, path, &orig, base, | |
1427 | &src1, name1, &src2, name2, | |
1218b3ab | 1428 | &opt->priv->attr_index, &ll_opts); |
f591c472 EN |
1429 | |
1430 | free(base); | |
1431 | free(name1); | |
1432 | free(name2); | |
1433 | free(orig.ptr); | |
1434 | free(src1.ptr); | |
1435 | free(src2.ptr); | |
1436 | return merge_status; | |
62fdec17 EN |
1437 | } |
1438 | ||
e2e9dc03 EN |
1439 | static int handle_content_merge(struct merge_options *opt, |
1440 | const char *path, | |
1441 | const struct version_info *o, | |
1442 | const struct version_info *a, | |
1443 | const struct version_info *b, | |
1444 | const char *pathnames[3], | |
1445 | const int extra_marker_size, | |
1446 | struct version_info *result) | |
1447 | { | |
991bbdca | 1448 | /* |
62fdec17 EN |
1449 | * path is the target location where we want to put the file, and |
1450 | * is used to determine any normalization rules in ll_merge. | |
1451 | * | |
1452 | * The normal case is that path and all entries in pathnames are | |
1453 | * identical, though renames can affect which path we got one of | |
1454 | * the three blobs to merge on various sides of history. | |
1455 | * | |
1456 | * extra_marker_size is the amount to extend conflict markers in | |
1457 | * ll_merge; this is neeed if we have content merges of content | |
1458 | * merges, which happens for example with rename/rename(2to1) and | |
1459 | * rename/add conflicts. | |
1460 | */ | |
1461 | unsigned clean = 1; | |
1462 | ||
1463 | /* | |
1464 | * handle_content_merge() needs both files to be of the same type, i.e. | |
1465 | * both files OR both submodules OR both symlinks. Conflicting types | |
1466 | * needs to be handled elsewhere. | |
991bbdca | 1467 | */ |
62fdec17 EN |
1468 | assert((S_IFMT & a->mode) == (S_IFMT & b->mode)); |
1469 | ||
1470 | /* Merge modes */ | |
1471 | if (a->mode == b->mode || a->mode == o->mode) | |
1472 | result->mode = b->mode; | |
1473 | else { | |
1474 | /* must be the 100644/100755 case */ | |
1475 | assert(S_ISREG(a->mode)); | |
1476 | result->mode = a->mode; | |
1477 | clean = (b->mode == o->mode); | |
1478 | /* | |
1479 | * FIXME: If opt->priv->call_depth && !clean, then we really | |
1480 | * should not make result->mode match either a->mode or | |
1481 | * b->mode; that causes t6036 "check conflicting mode for | |
1482 | * regular file" to fail. It would be best to use some other | |
1483 | * mode, but we'll confuse all kinds of stuff if we use one | |
1484 | * where S_ISREG(result->mode) isn't true, and if we use | |
1485 | * something like 0100666, then tree-walk.c's calls to | |
1486 | * canon_mode() will just normalize that to 100644 for us and | |
1487 | * thus not solve anything. | |
1488 | * | |
1489 | * Figure out if there's some kind of way we can work around | |
1490 | * this... | |
1491 | */ | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Trivial oid merge. | |
1496 | * | |
1497 | * Note: While one might assume that the next four lines would | |
1498 | * be unnecessary due to the fact that match_mask is often | |
1499 | * setup and already handled, renames don't always take care | |
1500 | * of that. | |
1501 | */ | |
1502 | if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid)) | |
1503 | oidcpy(&result->oid, &b->oid); | |
1504 | else if (oideq(&b->oid, &o->oid)) | |
1505 | oidcpy(&result->oid, &a->oid); | |
1506 | ||
1507 | /* Remaining rules depend on file vs. submodule vs. symlink. */ | |
1508 | else if (S_ISREG(a->mode)) { | |
1509 | mmbuffer_t result_buf; | |
1510 | int ret = 0, merge_status; | |
1511 | int two_way; | |
1512 | ||
1513 | /* | |
1514 | * If 'o' is different type, treat it as null so we do a | |
1515 | * two-way merge. | |
1516 | */ | |
1517 | two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
1518 | ||
1519 | merge_status = merge_3way(opt, path, | |
14228447 | 1520 | two_way ? null_oid() : &o->oid, |
62fdec17 EN |
1521 | &a->oid, &b->oid, |
1522 | pathnames, extra_marker_size, | |
1523 | &result_buf); | |
1524 | ||
1525 | if ((merge_status < 0) || !result_buf.ptr) | |
1526 | ret = err(opt, _("Failed to execute internal merge")); | |
1527 | ||
1528 | if (!ret && | |
1529 | write_object_file(result_buf.ptr, result_buf.size, | |
1530 | blob_type, &result->oid)) | |
1531 | ret = err(opt, _("Unable to add %s to database"), | |
1532 | path); | |
1533 | ||
1534 | free(result_buf.ptr); | |
1535 | if (ret) | |
1536 | return -1; | |
1537 | clean &= (merge_status == 0); | |
1538 | path_msg(opt, path, 1, _("Auto-merging %s"), path); | |
1539 | } else if (S_ISGITLINK(a->mode)) { | |
1540 | int two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
1541 | clean = merge_submodule(opt, pathnames[0], | |
14228447 | 1542 | two_way ? null_oid() : &o->oid, |
62fdec17 EN |
1543 | &a->oid, &b->oid, &result->oid); |
1544 | if (opt->priv->call_depth && two_way && !clean) { | |
1545 | result->mode = o->mode; | |
1546 | oidcpy(&result->oid, &o->oid); | |
1547 | } | |
1548 | } else if (S_ISLNK(a->mode)) { | |
1549 | if (opt->priv->call_depth) { | |
1550 | clean = 0; | |
1551 | result->mode = o->mode; | |
1552 | oidcpy(&result->oid, &o->oid); | |
1553 | } else { | |
1554 | switch (opt->recursive_variant) { | |
1555 | case MERGE_VARIANT_NORMAL: | |
1556 | clean = 0; | |
1557 | oidcpy(&result->oid, &a->oid); | |
1558 | break; | |
1559 | case MERGE_VARIANT_OURS: | |
1560 | oidcpy(&result->oid, &a->oid); | |
1561 | break; | |
1562 | case MERGE_VARIANT_THEIRS: | |
1563 | oidcpy(&result->oid, &b->oid); | |
1564 | break; | |
1565 | } | |
1566 | } | |
1567 | } else | |
1568 | BUG("unsupported object type in the tree: %06o for %s", | |
1569 | a->mode, path); | |
1570 | ||
991bbdca | 1571 | return clean; |
e2e9dc03 EN |
1572 | } |
1573 | ||
04af1879 EN |
1574 | /*** Function Grouping: functions related to detect_and_process_renames(), *** |
1575 | *** which are split into directory and regular rename detection sections. ***/ | |
1576 | ||
1577 | /*** Function Grouping: functions related to directory rename detection ***/ | |
1578 | ||
fa5e06d6 EN |
1579 | struct collision_info { |
1580 | struct string_list source_files; | |
1581 | unsigned reported_already:1; | |
1582 | }; | |
1583 | ||
d9d015df EN |
1584 | /* |
1585 | * Return a new string that replaces the beginning portion (which matches | |
1586 | * rename_info->key), with rename_info->util.new_dir. In perl-speak: | |
1587 | * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/); | |
1588 | * NOTE: | |
1589 | * Caller must ensure that old_path starts with rename_info->key + '/'. | |
1590 | */ | |
1591 | static char *apply_dir_rename(struct strmap_entry *rename_info, | |
1592 | const char *old_path) | |
1593 | { | |
fbcfc0cc EN |
1594 | struct strbuf new_path = STRBUF_INIT; |
1595 | const char *old_dir = rename_info->key; | |
1596 | const char *new_dir = rename_info->value; | |
1597 | int oldlen, newlen, new_dir_len; | |
1598 | ||
1599 | oldlen = strlen(old_dir); | |
1600 | if (*new_dir == '\0') | |
1601 | /* | |
1602 | * If someone renamed/merged a subdirectory into the root | |
1603 | * directory (e.g. 'some/subdir' -> ''), then we want to | |
1604 | * avoid returning | |
1605 | * '' + '/filename' | |
1606 | * as the rename; we need to make old_path + oldlen advance | |
1607 | * past the '/' character. | |
1608 | */ | |
1609 | oldlen++; | |
1610 | new_dir_len = strlen(new_dir); | |
1611 | newlen = new_dir_len + (strlen(old_path) - oldlen) + 1; | |
1612 | strbuf_grow(&new_path, newlen); | |
1613 | strbuf_add(&new_path, new_dir, new_dir_len); | |
1614 | strbuf_addstr(&new_path, &old_path[oldlen]); | |
1615 | ||
1616 | return strbuf_detach(&new_path, NULL); | |
d9d015df EN |
1617 | } |
1618 | ||
bea43365 EN |
1619 | static int path_in_way(struct strmap *paths, const char *path, unsigned side_mask) |
1620 | { | |
1621 | struct merged_info *mi = strmap_get(paths, path); | |
1622 | struct conflict_info *ci; | |
1623 | if (!mi) | |
1624 | return 0; | |
1625 | INITIALIZE_CI(ci, mi); | |
1626 | return mi->clean || (side_mask & (ci->filemask | ci->dirmask)); | |
1627 | } | |
1628 | ||
47325e85 EN |
1629 | /* |
1630 | * See if there is a directory rename for path, and if there are any file | |
1631 | * level conflicts on the given side for the renamed location. If there is | |
1632 | * a rename and there are no conflicts, return the new name. Otherwise, | |
1633 | * return NULL. | |
1634 | */ | |
1635 | static char *handle_path_level_conflicts(struct merge_options *opt, | |
1636 | const char *path, | |
1637 | unsigned side_index, | |
1638 | struct strmap_entry *rename_info, | |
1639 | struct strmap *collisions) | |
1640 | { | |
bea43365 EN |
1641 | char *new_path = NULL; |
1642 | struct collision_info *c_info; | |
1643 | int clean = 1; | |
1644 | struct strbuf collision_paths = STRBUF_INIT; | |
1645 | ||
1646 | /* | |
1647 | * entry has the mapping of old directory name to new directory name | |
1648 | * that we want to apply to path. | |
1649 | */ | |
1650 | new_path = apply_dir_rename(rename_info, path); | |
1651 | if (!new_path) | |
1652 | BUG("Failed to apply directory rename!"); | |
1653 | ||
1654 | /* | |
1655 | * The caller needs to have ensured that it has pre-populated | |
1656 | * collisions with all paths that map to new_path. Do a quick check | |
1657 | * to ensure that's the case. | |
1658 | */ | |
1659 | c_info = strmap_get(collisions, new_path); | |
1660 | if (c_info == NULL) | |
1661 | BUG("c_info is NULL"); | |
1662 | ||
1663 | /* | |
1664 | * Check for one-sided add/add/.../add conflicts, i.e. | |
1665 | * where implicit renames from the other side doing | |
1666 | * directory rename(s) can affect this side of history | |
1667 | * to put multiple paths into the same location. Warn | |
1668 | * and bail on directory renames for such paths. | |
1669 | */ | |
1670 | if (c_info->reported_already) { | |
1671 | clean = 0; | |
1672 | } else if (path_in_way(&opt->priv->paths, new_path, 1 << side_index)) { | |
1673 | c_info->reported_already = 1; | |
1674 | strbuf_add_separated_string_list(&collision_paths, ", ", | |
1675 | &c_info->source_files); | |
1676 | path_msg(opt, new_path, 0, | |
1677 | _("CONFLICT (implicit dir rename): Existing file/dir " | |
1678 | "at %s in the way of implicit directory rename(s) " | |
1679 | "putting the following path(s) there: %s."), | |
1680 | new_path, collision_paths.buf); | |
1681 | clean = 0; | |
1682 | } else if (c_info->source_files.nr > 1) { | |
1683 | c_info->reported_already = 1; | |
1684 | strbuf_add_separated_string_list(&collision_paths, ", ", | |
1685 | &c_info->source_files); | |
1686 | path_msg(opt, new_path, 0, | |
1687 | _("CONFLICT (implicit dir rename): Cannot map more " | |
1688 | "than one path to %s; implicit directory renames " | |
1689 | "tried to put these paths there: %s"), | |
1690 | new_path, collision_paths.buf); | |
1691 | clean = 0; | |
1692 | } | |
1693 | ||
1694 | /* Free memory we no longer need */ | |
1695 | strbuf_release(&collision_paths); | |
1696 | if (!clean && new_path) { | |
1697 | free(new_path); | |
1698 | return NULL; | |
1699 | } | |
1700 | ||
1701 | return new_path; | |
47325e85 EN |
1702 | } |
1703 | ||
112e1112 EN |
1704 | static void get_provisional_directory_renames(struct merge_options *opt, |
1705 | unsigned side, | |
1706 | int *clean) | |
1707 | { | |
04264d40 EN |
1708 | struct hashmap_iter iter; |
1709 | struct strmap_entry *entry; | |
1710 | struct rename_info *renames = &opt->priv->renames; | |
1711 | ||
04264d40 EN |
1712 | /* |
1713 | * Collapse | |
1714 | * dir_rename_count: old_directory -> {new_directory -> count} | |
1715 | * down to | |
1716 | * dir_renames: old_directory -> best_new_directory | |
1717 | * where best_new_directory is the one with the unique highest count. | |
1718 | */ | |
1719 | strmap_for_each_entry(&renames->dir_rename_count[side], &iter, entry) { | |
1720 | const char *source_dir = entry->key; | |
1721 | struct strintmap *counts = entry->value; | |
1722 | struct hashmap_iter count_iter; | |
1723 | struct strmap_entry *count_entry; | |
1724 | int max = 0; | |
1725 | int bad_max = 0; | |
1726 | const char *best = NULL; | |
1727 | ||
1728 | strintmap_for_each_entry(counts, &count_iter, count_entry) { | |
1729 | const char *target_dir = count_entry->key; | |
1730 | intptr_t count = (intptr_t)count_entry->value; | |
1731 | ||
1732 | if (count == max) | |
1733 | bad_max = max; | |
1734 | else if (count > max) { | |
1735 | max = count; | |
1736 | best = target_dir; | |
1737 | } | |
1738 | } | |
1739 | ||
bf238b71 EN |
1740 | if (max == 0) |
1741 | continue; | |
1742 | ||
04264d40 EN |
1743 | if (bad_max == max) { |
1744 | path_msg(opt, source_dir, 0, | |
1745 | _("CONFLICT (directory rename split): " | |
1746 | "Unclear where to rename %s to; it was " | |
1747 | "renamed to multiple other directories, with " | |
1748 | "no destination getting a majority of the " | |
1749 | "files."), | |
1750 | source_dir); | |
41376b58 | 1751 | *clean = 0; |
04264d40 EN |
1752 | } else { |
1753 | strmap_put(&renames->dir_renames[side], | |
1754 | source_dir, (void*)best); | |
1755 | } | |
1756 | } | |
112e1112 EN |
1757 | } |
1758 | ||
1759 | static void handle_directory_level_conflicts(struct merge_options *opt) | |
1760 | { | |
98d0d081 EN |
1761 | struct hashmap_iter iter; |
1762 | struct strmap_entry *entry; | |
1763 | struct string_list duplicated = STRING_LIST_INIT_NODUP; | |
1764 | struct rename_info *renames = &opt->priv->renames; | |
1765 | struct strmap *side1_dir_renames = &renames->dir_renames[MERGE_SIDE1]; | |
1766 | struct strmap *side2_dir_renames = &renames->dir_renames[MERGE_SIDE2]; | |
1767 | int i; | |
1768 | ||
1769 | strmap_for_each_entry(side1_dir_renames, &iter, entry) { | |
1770 | if (strmap_contains(side2_dir_renames, entry->key)) | |
1771 | string_list_append(&duplicated, entry->key); | |
1772 | } | |
1773 | ||
1774 | for (i = 0; i < duplicated.nr; i++) { | |
1775 | strmap_remove(side1_dir_renames, duplicated.items[i].string, 0); | |
1776 | strmap_remove(side2_dir_renames, duplicated.items[i].string, 0); | |
1777 | } | |
1778 | string_list_clear(&duplicated, 0); | |
112e1112 EN |
1779 | } |
1780 | ||
d9d015df EN |
1781 | static struct strmap_entry *check_dir_renamed(const char *path, |
1782 | struct strmap *dir_renames) | |
1783 | { | |
fbcfc0cc EN |
1784 | char *temp = xstrdup(path); |
1785 | char *end; | |
1786 | struct strmap_entry *e = NULL; | |
1787 | ||
1788 | while ((end = strrchr(temp, '/'))) { | |
1789 | *end = '\0'; | |
1790 | e = strmap_get_entry(dir_renames, temp); | |
1791 | if (e) | |
1792 | break; | |
1793 | } | |
1794 | free(temp); | |
1795 | return e; | |
d9d015df EN |
1796 | } |
1797 | ||
fa5e06d6 EN |
1798 | static void compute_collisions(struct strmap *collisions, |
1799 | struct strmap *dir_renames, | |
1800 | struct diff_queue_struct *pairs) | |
1801 | { | |
d9d015df EN |
1802 | int i; |
1803 | ||
1804 | strmap_init_with_options(collisions, NULL, 0); | |
1805 | if (strmap_empty(dir_renames)) | |
1806 | return; | |
1807 | ||
1808 | /* | |
1809 | * Multiple files can be mapped to the same path due to directory | |
1810 | * renames done by the other side of history. Since that other | |
1811 | * side of history could have merged multiple directories into one, | |
1812 | * if our side of history added the same file basename to each of | |
1813 | * those directories, then all N of them would get implicitly | |
1814 | * renamed by the directory rename detection into the same path, | |
1815 | * and we'd get an add/add/.../add conflict, and all those adds | |
1816 | * from *this* side of history. This is not representable in the | |
1817 | * index, and users aren't going to easily be able to make sense of | |
1818 | * it. So we need to provide a good warning about what's | |
1819 | * happening, and fall back to no-directory-rename detection | |
1820 | * behavior for those paths. | |
1821 | * | |
1822 | * See testcases 9e and all of section 5 from t6043 for examples. | |
1823 | */ | |
1824 | for (i = 0; i < pairs->nr; ++i) { | |
1825 | struct strmap_entry *rename_info; | |
1826 | struct collision_info *collision_info; | |
1827 | char *new_path; | |
1828 | struct diff_filepair *pair = pairs->queue[i]; | |
1829 | ||
1830 | if (pair->status != 'A' && pair->status != 'R') | |
1831 | continue; | |
1832 | rename_info = check_dir_renamed(pair->two->path, dir_renames); | |
1833 | if (!rename_info) | |
1834 | continue; | |
1835 | ||
1836 | new_path = apply_dir_rename(rename_info, pair->two->path); | |
1837 | assert(new_path); | |
1838 | collision_info = strmap_get(collisions, new_path); | |
1839 | if (collision_info) { | |
1840 | free(new_path); | |
1841 | } else { | |
ca56dadb | 1842 | CALLOC_ARRAY(collision_info, 1); |
d9d015df EN |
1843 | string_list_init(&collision_info->source_files, 0); |
1844 | strmap_put(collisions, new_path, collision_info); | |
1845 | } | |
1846 | string_list_insert(&collision_info->source_files, | |
1847 | pair->two->path); | |
1848 | } | |
fa5e06d6 EN |
1849 | } |
1850 | ||
1851 | static char *check_for_directory_rename(struct merge_options *opt, | |
1852 | const char *path, | |
1853 | unsigned side_index, | |
1854 | struct strmap *dir_renames, | |
1855 | struct strmap *dir_rename_exclusions, | |
1856 | struct strmap *collisions, | |
1857 | int *clean_merge) | |
1858 | { | |
47325e85 EN |
1859 | char *new_path = NULL; |
1860 | struct strmap_entry *rename_info; | |
1861 | struct strmap_entry *otherinfo = NULL; | |
1862 | const char *new_dir; | |
1863 | ||
1864 | if (strmap_empty(dir_renames)) | |
1865 | return new_path; | |
1866 | rename_info = check_dir_renamed(path, dir_renames); | |
1867 | if (!rename_info) | |
1868 | return new_path; | |
1869 | /* old_dir = rename_info->key; */ | |
1870 | new_dir = rename_info->value; | |
1871 | ||
1872 | /* | |
1873 | * This next part is a little weird. We do not want to do an | |
1874 | * implicit rename into a directory we renamed on our side, because | |
1875 | * that will result in a spurious rename/rename(1to2) conflict. An | |
1876 | * example: | |
1877 | * Base commit: dumbdir/afile, otherdir/bfile | |
1878 | * Side 1: smrtdir/afile, otherdir/bfile | |
1879 | * Side 2: dumbdir/afile, dumbdir/bfile | |
1880 | * Here, while working on Side 1, we could notice that otherdir was | |
1881 | * renamed/merged to dumbdir, and change the diff_filepair for | |
1882 | * otherdir/bfile into a rename into dumbdir/bfile. However, Side | |
1883 | * 2 will notice the rename from dumbdir to smrtdir, and do the | |
1884 | * transitive rename to move it from dumbdir/bfile to | |
1885 | * smrtdir/bfile. That gives us bfile in dumbdir vs being in | |
1886 | * smrtdir, a rename/rename(1to2) conflict. We really just want | |
1887 | * the file to end up in smrtdir. And the way to achieve that is | |
1888 | * to not let Side1 do the rename to dumbdir, since we know that is | |
1889 | * the source of one of our directory renames. | |
1890 | * | |
1891 | * That's why otherinfo and dir_rename_exclusions is here. | |
1892 | * | |
1893 | * As it turns out, this also prevents N-way transient rename | |
1894 | * confusion; See testcases 9c and 9d of t6043. | |
1895 | */ | |
1896 | otherinfo = strmap_get_entry(dir_rename_exclusions, new_dir); | |
1897 | if (otherinfo) { | |
1898 | path_msg(opt, rename_info->key, 1, | |
1899 | _("WARNING: Avoiding applying %s -> %s rename " | |
1900 | "to %s, because %s itself was renamed."), | |
1901 | rename_info->key, new_dir, path, new_dir); | |
1902 | return NULL; | |
1903 | } | |
1904 | ||
1905 | new_path = handle_path_level_conflicts(opt, path, side_index, | |
1906 | rename_info, collisions); | |
1907 | *clean_merge &= (new_path != NULL); | |
1908 | ||
1909 | return new_path; | |
fa5e06d6 EN |
1910 | } |
1911 | ||
1912 | static void apply_directory_rename_modifications(struct merge_options *opt, | |
1913 | struct diff_filepair *pair, | |
1914 | char *new_path) | |
1915 | { | |
089d82bc EN |
1916 | /* |
1917 | * The basic idea is to get the conflict_info from opt->priv->paths | |
1918 | * at old path, and insert it into new_path; basically just this: | |
1919 | * ci = strmap_get(&opt->priv->paths, old_path); | |
1920 | * strmap_remove(&opt->priv->paths, old_path, 0); | |
1921 | * strmap_put(&opt->priv->paths, new_path, ci); | |
1922 | * However, there are some factors complicating this: | |
1923 | * - opt->priv->paths may already have an entry at new_path | |
1924 | * - Each ci tracks its containing directory, so we need to | |
1925 | * update that | |
1926 | * - If another ci has the same containing directory, then | |
1927 | * the two char*'s MUST point to the same location. See the | |
1928 | * comment in struct merged_info. strcmp equality is not | |
1929 | * enough; we need pointer equality. | |
1930 | * - opt->priv->paths must hold the parent directories of any | |
1931 | * entries that are added. So, if this directory rename | |
1932 | * causes entirely new directories, we must recursively add | |
1933 | * parent directories. | |
1934 | * - For each parent directory added to opt->priv->paths, we | |
1935 | * also need to get its parent directory stored in its | |
1936 | * conflict_info->merged.directory_name with all the same | |
1937 | * requirements about pointer equality. | |
1938 | */ | |
1939 | struct string_list dirs_to_insert = STRING_LIST_INIT_NODUP; | |
1940 | struct conflict_info *ci, *new_ci; | |
1941 | struct strmap_entry *entry; | |
1942 | const char *branch_with_new_path, *branch_with_dir_rename; | |
1943 | const char *old_path = pair->two->path; | |
1944 | const char *parent_name; | |
1945 | const char *cur_path; | |
1946 | int i, len; | |
1947 | ||
1948 | entry = strmap_get_entry(&opt->priv->paths, old_path); | |
1949 | old_path = entry->key; | |
1950 | ci = entry->value; | |
1951 | VERIFY_CI(ci); | |
1952 | ||
1953 | /* Find parent directories missing from opt->priv->paths */ | |
1954 | cur_path = new_path; | |
1955 | while (1) { | |
1956 | /* Find the parent directory of cur_path */ | |
1957 | char *last_slash = strrchr(cur_path, '/'); | |
1958 | if (last_slash) { | |
1959 | parent_name = xstrndup(cur_path, last_slash - cur_path); | |
1960 | } else { | |
1961 | parent_name = opt->priv->toplevel_dir; | |
1962 | break; | |
1963 | } | |
1964 | ||
1965 | /* Look it up in opt->priv->paths */ | |
1966 | entry = strmap_get_entry(&opt->priv->paths, parent_name); | |
1967 | if (entry) { | |
1968 | free((char*)parent_name); | |
1969 | parent_name = entry->key; /* reuse known pointer */ | |
1970 | break; | |
1971 | } | |
1972 | ||
1973 | /* Record this is one of the directories we need to insert */ | |
1974 | string_list_append(&dirs_to_insert, parent_name); | |
1975 | cur_path = parent_name; | |
1976 | } | |
1977 | ||
1978 | /* Traverse dirs_to_insert and insert them into opt->priv->paths */ | |
1979 | for (i = dirs_to_insert.nr-1; i >= 0; --i) { | |
1980 | struct conflict_info *dir_ci; | |
1981 | char *cur_dir = dirs_to_insert.items[i].string; | |
1982 | ||
ca56dadb | 1983 | CALLOC_ARRAY(dir_ci, 1); |
089d82bc EN |
1984 | |
1985 | dir_ci->merged.directory_name = parent_name; | |
1986 | len = strlen(parent_name); | |
1987 | /* len+1 because of trailing '/' character */ | |
1988 | dir_ci->merged.basename_offset = (len > 0 ? len+1 : len); | |
1989 | dir_ci->dirmask = ci->filemask; | |
1990 | strmap_put(&opt->priv->paths, cur_dir, dir_ci); | |
1991 | ||
1992 | parent_name = cur_dir; | |
1993 | } | |
1994 | ||
1995 | /* | |
1996 | * We are removing old_path from opt->priv->paths. old_path also will | |
1997 | * eventually need to be freed, but it may still be used by e.g. | |
1998 | * ci->pathnames. So, store it in another string-list for now. | |
1999 | */ | |
2000 | string_list_append(&opt->priv->paths_to_free, old_path); | |
2001 | ||
2002 | assert(ci->filemask == 2 || ci->filemask == 4); | |
2003 | assert(ci->dirmask == 0); | |
2004 | strmap_remove(&opt->priv->paths, old_path, 0); | |
2005 | ||
2006 | branch_with_new_path = (ci->filemask == 2) ? opt->branch1 : opt->branch2; | |
2007 | branch_with_dir_rename = (ci->filemask == 2) ? opt->branch2 : opt->branch1; | |
2008 | ||
2009 | /* Now, finally update ci and stick it into opt->priv->paths */ | |
2010 | ci->merged.directory_name = parent_name; | |
2011 | len = strlen(parent_name); | |
2012 | ci->merged.basename_offset = (len > 0 ? len+1 : len); | |
2013 | new_ci = strmap_get(&opt->priv->paths, new_path); | |
2014 | if (!new_ci) { | |
2015 | /* Place ci back into opt->priv->paths, but at new_path */ | |
2016 | strmap_put(&opt->priv->paths, new_path, ci); | |
2017 | } else { | |
2018 | int index; | |
2019 | ||
2020 | /* A few sanity checks */ | |
2021 | VERIFY_CI(new_ci); | |
2022 | assert(ci->filemask == 2 || ci->filemask == 4); | |
2023 | assert((new_ci->filemask & ci->filemask) == 0); | |
2024 | assert(!new_ci->merged.clean); | |
2025 | ||
2026 | /* Copy stuff from ci into new_ci */ | |
2027 | new_ci->filemask |= ci->filemask; | |
2028 | if (new_ci->dirmask) | |
2029 | new_ci->df_conflict = 1; | |
2030 | index = (ci->filemask >> 1); | |
2031 | new_ci->pathnames[index] = ci->pathnames[index]; | |
2032 | new_ci->stages[index].mode = ci->stages[index].mode; | |
2033 | oidcpy(&new_ci->stages[index].oid, &ci->stages[index].oid); | |
2034 | ||
2035 | free(ci); | |
2036 | ci = new_ci; | |
2037 | } | |
2038 | ||
2039 | if (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) { | |
2040 | /* Notify user of updated path */ | |
2041 | if (pair->status == 'A') | |
2042 | path_msg(opt, new_path, 1, | |
2043 | _("Path updated: %s added in %s inside a " | |
2044 | "directory that was renamed in %s; moving " | |
2045 | "it to %s."), | |
2046 | old_path, branch_with_new_path, | |
2047 | branch_with_dir_rename, new_path); | |
2048 | else | |
2049 | path_msg(opt, new_path, 1, | |
2050 | _("Path updated: %s renamed to %s in %s, " | |
2051 | "inside a directory that was renamed in %s; " | |
2052 | "moving it to %s."), | |
2053 | pair->one->path, old_path, branch_with_new_path, | |
2054 | branch_with_dir_rename, new_path); | |
2055 | } else { | |
2056 | /* | |
2057 | * opt->detect_directory_renames has the value | |
2058 | * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts. | |
2059 | */ | |
2060 | ci->path_conflict = 1; | |
2061 | if (pair->status == 'A') | |
2062 | path_msg(opt, new_path, 0, | |
2063 | _("CONFLICT (file location): %s added in %s " | |
2064 | "inside a directory that was renamed in %s, " | |
2065 | "suggesting it should perhaps be moved to " | |
2066 | "%s."), | |
2067 | old_path, branch_with_new_path, | |
2068 | branch_with_dir_rename, new_path); | |
2069 | else | |
2070 | path_msg(opt, new_path, 0, | |
2071 | _("CONFLICT (file location): %s renamed to %s " | |
2072 | "in %s, inside a directory that was renamed " | |
2073 | "in %s, suggesting it should perhaps be " | |
2074 | "moved to %s."), | |
2075 | pair->one->path, old_path, branch_with_new_path, | |
2076 | branch_with_dir_rename, new_path); | |
2077 | } | |
2078 | ||
2079 | /* | |
2080 | * Finally, record the new location. | |
2081 | */ | |
2082 | pair->two->path = new_path; | |
fa5e06d6 EN |
2083 | } |
2084 | ||
04af1879 EN |
2085 | /*** Function Grouping: functions related to regular rename detection ***/ |
2086 | ||
e1a124e8 EN |
2087 | static int process_renames(struct merge_options *opt, |
2088 | struct diff_queue_struct *renames) | |
2089 | { | |
c2d267df EN |
2090 | int clean_merge = 1, i; |
2091 | ||
2092 | for (i = 0; i < renames->nr; ++i) { | |
2093 | const char *oldpath = NULL, *newpath; | |
2094 | struct diff_filepair *pair = renames->queue[i]; | |
2095 | struct conflict_info *oldinfo = NULL, *newinfo = NULL; | |
2096 | struct strmap_entry *old_ent, *new_ent; | |
2097 | unsigned int old_sidemask; | |
2098 | int target_index, other_source_index; | |
2099 | int source_deleted, collision, type_changed; | |
2e91ddd2 | 2100 | const char *rename_branch = NULL, *delete_branch = NULL; |
c2d267df EN |
2101 | |
2102 | old_ent = strmap_get_entry(&opt->priv->paths, pair->one->path); | |
c2d267df | 2103 | new_ent = strmap_get_entry(&opt->priv->paths, pair->two->path); |
1b6b902d EN |
2104 | if (old_ent) { |
2105 | oldpath = old_ent->key; | |
2106 | oldinfo = old_ent->value; | |
2107 | } | |
2108 | newpath = pair->two->path; | |
2109 | if (new_ent) { | |
2110 | newpath = new_ent->key; | |
2111 | newinfo = new_ent->value; | |
2112 | } | |
2113 | ||
2114 | /* | |
2115 | * If pair->one->path isn't in opt->priv->paths, that means | |
2116 | * that either directory rename detection removed that | |
2117 | * path, or a parent directory of oldpath was resolved and | |
2118 | * we don't even need the rename; in either case, we can | |
2119 | * skip it. If oldinfo->merged.clean, then the other side | |
2120 | * of history had no changes to oldpath and we don't need | |
2121 | * the rename and can skip it. | |
2122 | */ | |
2123 | if (!oldinfo || oldinfo->merged.clean) | |
2124 | continue; | |
c2d267df EN |
2125 | |
2126 | /* | |
2127 | * diff_filepairs have copies of pathnames, thus we have to | |
2128 | * use standard 'strcmp()' (negated) instead of '=='. | |
2129 | */ | |
2130 | if (i + 1 < renames->nr && | |
2131 | !strcmp(oldpath, renames->queue[i+1]->one->path)) { | |
2132 | /* Handle rename/rename(1to2) or rename/rename(1to1) */ | |
2133 | const char *pathnames[3]; | |
af1e56c4 EN |
2134 | struct version_info merged; |
2135 | struct conflict_info *base, *side1, *side2; | |
53e88a03 | 2136 | unsigned was_binary_blob = 0; |
c2d267df EN |
2137 | |
2138 | pathnames[0] = oldpath; | |
2139 | pathnames[1] = newpath; | |
2140 | pathnames[2] = renames->queue[i+1]->two->path; | |
2141 | ||
af1e56c4 EN |
2142 | base = strmap_get(&opt->priv->paths, pathnames[0]); |
2143 | side1 = strmap_get(&opt->priv->paths, pathnames[1]); | |
2144 | side2 = strmap_get(&opt->priv->paths, pathnames[2]); | |
2145 | ||
2146 | VERIFY_CI(base); | |
2147 | VERIFY_CI(side1); | |
2148 | VERIFY_CI(side2); | |
2149 | ||
c2d267df | 2150 | if (!strcmp(pathnames[1], pathnames[2])) { |
cbdca289 EN |
2151 | struct rename_info *ri = &opt->priv->renames; |
2152 | int j; | |
2153 | ||
af1e56c4 EN |
2154 | /* Both sides renamed the same way */ |
2155 | assert(side1 == side2); | |
2156 | memcpy(&side1->stages[0], &base->stages[0], | |
2157 | sizeof(merged)); | |
2158 | side1->filemask |= (1 << MERGE_BASE); | |
2159 | /* Mark base as resolved by removal */ | |
2160 | base->merged.is_null = 1; | |
2161 | base->merged.clean = 1; | |
c2d267df | 2162 | |
cbdca289 EN |
2163 | /* |
2164 | * Disable remembering renames optimization; | |
2165 | * rename/rename(1to1) is incredibly rare, and | |
2166 | * just disabling the optimization is easier | |
2167 | * than purging cached_pairs, | |
2168 | * cached_target_names, and dir_rename_counts. | |
2169 | */ | |
2170 | for (j = 0; j < 3; j++) | |
2171 | ri->merge_trees[j] = NULL; | |
2172 | ||
c2d267df EN |
2173 | /* We handled both renames, i.e. i+1 handled */ |
2174 | i++; | |
2175 | /* Move to next rename */ | |
2176 | continue; | |
2177 | } | |
2178 | ||
2179 | /* This is a rename/rename(1to2) */ | |
53e88a03 EN |
2180 | clean_merge = handle_content_merge(opt, |
2181 | pair->one->path, | |
2182 | &base->stages[0], | |
2183 | &side1->stages[1], | |
2184 | &side2->stages[2], | |
2185 | pathnames, | |
2186 | 1 + 2 * opt->priv->call_depth, | |
2187 | &merged); | |
2188 | if (!clean_merge && | |
2189 | merged.mode == side1->stages[1].mode && | |
2190 | oideq(&merged.oid, &side1->stages[1].oid)) | |
2191 | was_binary_blob = 1; | |
2192 | memcpy(&side1->stages[1], &merged, sizeof(merged)); | |
2193 | if (was_binary_blob) { | |
2194 | /* | |
2195 | * Getting here means we were attempting to | |
2196 | * merge a binary blob. | |
2197 | * | |
2198 | * Since we can't merge binaries, | |
2199 | * handle_content_merge() just takes one | |
2200 | * side. But we don't want to copy the | |
2201 | * contents of one side to both paths. We | |
2202 | * used the contents of side1 above for | |
2203 | * side1->stages, let's use the contents of | |
2204 | * side2 for side2->stages below. | |
2205 | */ | |
2206 | oidcpy(&merged.oid, &side2->stages[2].oid); | |
2207 | merged.mode = side2->stages[2].mode; | |
2208 | } | |
2209 | memcpy(&side2->stages[2], &merged, sizeof(merged)); | |
2210 | ||
2211 | side1->path_conflict = 1; | |
2212 | side2->path_conflict = 1; | |
2213 | /* | |
2214 | * TODO: For renames we normally remove the path at the | |
2215 | * old name. It would thus seem consistent to do the | |
2216 | * same for rename/rename(1to2) cases, but we haven't | |
2217 | * done so traditionally and a number of the regression | |
2218 | * tests now encode an expectation that the file is | |
2219 | * left there at stage 1. If we ever decide to change | |
2220 | * this, add the following two lines here: | |
2221 | * base->merged.is_null = 1; | |
2222 | * base->merged.clean = 1; | |
2223 | * and remove the setting of base->path_conflict to 1. | |
2224 | */ | |
2225 | base->path_conflict = 1; | |
2226 | path_msg(opt, oldpath, 0, | |
2227 | _("CONFLICT (rename/rename): %s renamed to " | |
2228 | "%s in %s and to %s in %s."), | |
2229 | pathnames[0], | |
2230 | pathnames[1], opt->branch1, | |
2231 | pathnames[2], opt->branch2); | |
c2d267df EN |
2232 | |
2233 | i++; /* We handled both renames, i.e. i+1 handled */ | |
2234 | continue; | |
2235 | } | |
2236 | ||
2237 | VERIFY_CI(oldinfo); | |
2238 | VERIFY_CI(newinfo); | |
2239 | target_index = pair->score; /* from collect_renames() */ | |
2240 | assert(target_index == 1 || target_index == 2); | |
2241 | other_source_index = 3 - target_index; | |
2242 | old_sidemask = (1 << other_source_index); /* 2 or 4 */ | |
2243 | source_deleted = (oldinfo->filemask == 1); | |
2244 | collision = ((newinfo->filemask & old_sidemask) != 0); | |
2245 | type_changed = !source_deleted && | |
2246 | (S_ISREG(oldinfo->stages[other_source_index].mode) != | |
2247 | S_ISREG(newinfo->stages[target_index].mode)); | |
2248 | if (type_changed && collision) { | |
6fcccbd7 EN |
2249 | /* |
2250 | * special handling so later blocks can handle this... | |
2251 | * | |
2252 | * if type_changed && collision are both true, then this | |
2253 | * was really a double rename, but one side wasn't | |
2254 | * detected due to lack of break detection. I.e. | |
2255 | * something like | |
2256 | * orig: has normal file 'foo' | |
2257 | * side1: renames 'foo' to 'bar', adds 'foo' symlink | |
2258 | * side2: renames 'foo' to 'bar' | |
2259 | * In this case, the foo->bar rename on side1 won't be | |
2260 | * detected because the new symlink named 'foo' is | |
2261 | * there and we don't do break detection. But we detect | |
2262 | * this here because we don't want to merge the content | |
2263 | * of the foo symlink with the foo->bar file, so we | |
2264 | * have some logic to handle this special case. The | |
2265 | * easiest way to do that is make 'bar' on side1 not | |
2266 | * be considered a colliding file but the other part | |
2267 | * of a normal rename. If the file is very different, | |
2268 | * well we're going to get content merge conflicts | |
2269 | * anyway so it doesn't hurt. And if the colliding | |
2270 | * file also has a different type, that'll be handled | |
2271 | * by the content merge logic in process_entry() too. | |
2272 | * | |
2273 | * See also t6430, 'rename vs. rename/symlink' | |
2274 | */ | |
2275 | collision = 0; | |
c2d267df | 2276 | } |
2e91ddd2 EN |
2277 | if (source_deleted) { |
2278 | if (target_index == 1) { | |
2279 | rename_branch = opt->branch1; | |
2280 | delete_branch = opt->branch2; | |
2281 | } else { | |
2282 | rename_branch = opt->branch2; | |
2283 | delete_branch = opt->branch1; | |
2284 | } | |
2285 | } | |
c2d267df EN |
2286 | |
2287 | assert(source_deleted || oldinfo->filemask & old_sidemask); | |
2288 | ||
2289 | /* Need to check for special types of rename conflicts... */ | |
2290 | if (collision && !source_deleted) { | |
2291 | /* collision: rename/add or rename/rename(2to1) */ | |
35e47e35 EN |
2292 | const char *pathnames[3]; |
2293 | struct version_info merged; | |
2294 | ||
2295 | struct conflict_info *base, *side1, *side2; | |
2296 | unsigned clean; | |
2297 | ||
2298 | pathnames[0] = oldpath; | |
2299 | pathnames[other_source_index] = oldpath; | |
2300 | pathnames[target_index] = newpath; | |
2301 | ||
2302 | base = strmap_get(&opt->priv->paths, pathnames[0]); | |
2303 | side1 = strmap_get(&opt->priv->paths, pathnames[1]); | |
2304 | side2 = strmap_get(&opt->priv->paths, pathnames[2]); | |
2305 | ||
2306 | VERIFY_CI(base); | |
2307 | VERIFY_CI(side1); | |
2308 | VERIFY_CI(side2); | |
2309 | ||
2310 | clean = handle_content_merge(opt, pair->one->path, | |
2311 | &base->stages[0], | |
2312 | &side1->stages[1], | |
2313 | &side2->stages[2], | |
2314 | pathnames, | |
2315 | 1 + 2 * opt->priv->call_depth, | |
2316 | &merged); | |
2317 | ||
2318 | memcpy(&newinfo->stages[target_index], &merged, | |
2319 | sizeof(merged)); | |
2320 | if (!clean) { | |
2321 | path_msg(opt, newpath, 0, | |
2322 | _("CONFLICT (rename involved in " | |
2323 | "collision): rename of %s -> %s has " | |
2324 | "content conflicts AND collides " | |
2325 | "with another path; this may result " | |
2326 | "in nested conflict markers."), | |
2327 | oldpath, newpath); | |
2328 | } | |
c2d267df | 2329 | } else if (collision && source_deleted) { |
35e47e35 EN |
2330 | /* |
2331 | * rename/add/delete or rename/rename(2to1)/delete: | |
2332 | * since oldpath was deleted on the side that didn't | |
2333 | * do the rename, there's not much of a content merge | |
2334 | * we can do for the rename. oldinfo->merged.is_null | |
2335 | * was already set, so we just leave things as-is so | |
2336 | * they look like an add/add conflict. | |
2337 | */ | |
2338 | ||
2339 | newinfo->path_conflict = 1; | |
2340 | path_msg(opt, newpath, 0, | |
2341 | _("CONFLICT (rename/delete): %s renamed " | |
2342 | "to %s in %s, but deleted in %s."), | |
2343 | oldpath, newpath, rename_branch, delete_branch); | |
c2d267df | 2344 | } else { |
2e91ddd2 EN |
2345 | /* |
2346 | * a few different cases...start by copying the | |
2347 | * existing stage(s) from oldinfo over the newinfo | |
2348 | * and update the pathname(s). | |
2349 | */ | |
2350 | memcpy(&newinfo->stages[0], &oldinfo->stages[0], | |
2351 | sizeof(newinfo->stages[0])); | |
2352 | newinfo->filemask |= (1 << MERGE_BASE); | |
2353 | newinfo->pathnames[0] = oldpath; | |
c2d267df EN |
2354 | if (type_changed) { |
2355 | /* rename vs. typechange */ | |
6fcccbd7 | 2356 | /* Mark the original as resolved by removal */ |
14228447 | 2357 | memcpy(&oldinfo->stages[0].oid, null_oid(), |
6fcccbd7 EN |
2358 | sizeof(oldinfo->stages[0].oid)); |
2359 | oldinfo->stages[0].mode = 0; | |
2360 | oldinfo->filemask &= 0x06; | |
c2d267df EN |
2361 | } else if (source_deleted) { |
2362 | /* rename/delete */ | |
2e91ddd2 EN |
2363 | newinfo->path_conflict = 1; |
2364 | path_msg(opt, newpath, 0, | |
2365 | _("CONFLICT (rename/delete): %s renamed" | |
2366 | " to %s in %s, but deleted in %s."), | |
2367 | oldpath, newpath, | |
2368 | rename_branch, delete_branch); | |
c2d267df EN |
2369 | } else { |
2370 | /* normal rename */ | |
f1665e69 EN |
2371 | memcpy(&newinfo->stages[other_source_index], |
2372 | &oldinfo->stages[other_source_index], | |
2373 | sizeof(newinfo->stages[0])); | |
2374 | newinfo->filemask |= (1 << other_source_index); | |
2375 | newinfo->pathnames[other_source_index] = oldpath; | |
c2d267df EN |
2376 | } |
2377 | } | |
2378 | ||
2379 | if (!type_changed) { | |
2380 | /* Mark the original as resolved by removal */ | |
2381 | oldinfo->merged.is_null = 1; | |
2382 | oldinfo->merged.clean = 1; | |
2383 | } | |
2384 | ||
2385 | } | |
2386 | ||
2387 | return clean_merge; | |
e1a124e8 EN |
2388 | } |
2389 | ||
f89b4f2b EN |
2390 | static inline int possible_side_renames(struct rename_info *renames, |
2391 | unsigned side_index) | |
2392 | { | |
2393 | return renames->pairs[side_index].nr > 0 && | |
a49b55d5 | 2394 | !strintmap_empty(&renames->relevant_sources[side_index]); |
f89b4f2b EN |
2395 | } |
2396 | ||
2397 | static inline int possible_renames(struct rename_info *renames) | |
2398 | { | |
2399 | return possible_side_renames(renames, 1) || | |
25e65b6d EN |
2400 | possible_side_renames(renames, 2) || |
2401 | !strmap_empty(&renames->cached_pairs[1]) || | |
2402 | !strmap_empty(&renames->cached_pairs[2]); | |
f89b4f2b EN |
2403 | } |
2404 | ||
f78cf976 EN |
2405 | static void resolve_diffpair_statuses(struct diff_queue_struct *q) |
2406 | { | |
2407 | /* | |
2408 | * A simplified version of diff_resolve_rename_copy(); would probably | |
2409 | * just use that function but it's static... | |
2410 | */ | |
2411 | int i; | |
2412 | struct diff_filepair *p; | |
2413 | ||
2414 | for (i = 0; i < q->nr; ++i) { | |
2415 | p = q->queue[i]; | |
2416 | p->status = 0; /* undecided */ | |
2417 | if (!DIFF_FILE_VALID(p->one)) | |
2418 | p->status = DIFF_STATUS_ADDED; | |
2419 | else if (!DIFF_FILE_VALID(p->two)) | |
2420 | p->status = DIFF_STATUS_DELETED; | |
2421 | else if (DIFF_PAIR_RENAME(p)) | |
2422 | p->status = DIFF_STATUS_RENAMED; | |
2423 | } | |
2424 | } | |
2425 | ||
86b41b38 EN |
2426 | static void prune_cached_from_relevant(struct rename_info *renames, |
2427 | unsigned side) | |
2428 | { | |
2429 | /* Reason for this function described in add_pair() */ | |
2430 | struct hashmap_iter iter; | |
2431 | struct strmap_entry *entry; | |
2432 | ||
2433 | /* Remove from relevant_sources all entries in cached_pairs[side] */ | |
2434 | strmap_for_each_entry(&renames->cached_pairs[side], &iter, entry) { | |
2435 | strintmap_remove(&renames->relevant_sources[side], | |
2436 | entry->key); | |
2437 | } | |
2438 | /* Remove from relevant_sources all entries in cached_irrelevant[side] */ | |
2439 | strset_for_each_entry(&renames->cached_irrelevant[side], &iter, entry) { | |
2440 | strintmap_remove(&renames->relevant_sources[side], | |
2441 | entry->key); | |
2442 | } | |
2443 | } | |
2444 | ||
86b41b38 EN |
2445 | static void use_cached_pairs(struct merge_options *opt, |
2446 | struct strmap *cached_pairs, | |
2447 | struct diff_queue_struct *pairs) | |
2448 | { | |
2449 | struct hashmap_iter iter; | |
2450 | struct strmap_entry *entry; | |
2451 | ||
2452 | /* | |
2453 | * Add to side_pairs all entries from renames->cached_pairs[side_index]. | |
2454 | * (Info in cached_irrelevant[side_index] is not relevant here.) | |
2455 | */ | |
2456 | strmap_for_each_entry(cached_pairs, &iter, entry) { | |
2457 | struct diff_filespec *one, *two; | |
2458 | const char *old_name = entry->key; | |
2459 | const char *new_name = entry->value; | |
2460 | if (!new_name) | |
2461 | new_name = old_name; | |
2462 | ||
2463 | /* We don't care about oid/mode, only filenames and status */ | |
2464 | one = alloc_filespec(old_name); | |
2465 | two = alloc_filespec(new_name); | |
2466 | diff_queue(pairs, one, two); | |
2467 | pairs->queue[pairs->nr-1]->status = entry->value ? 'R' : 'D'; | |
2468 | } | |
2469 | } | |
2470 | ||
2734f2e3 EN |
2471 | static void cache_new_pair(struct rename_info *renames, |
2472 | int side, | |
2473 | char *old_path, | |
2474 | char *new_path, | |
2475 | int free_old_value) | |
2476 | { | |
2477 | char *old_value; | |
2478 | new_path = xstrdup(new_path); | |
2479 | old_value = strmap_put(&renames->cached_pairs[side], | |
2480 | old_path, new_path); | |
2481 | strset_add(&renames->cached_target_names[side], new_path); | |
2482 | if (free_old_value) | |
2483 | free(old_value); | |
2484 | else | |
2485 | assert(!old_value); | |
2486 | } | |
2487 | ||
2488 | static void possibly_cache_new_pair(struct rename_info *renames, | |
2489 | struct diff_filepair *p, | |
2490 | unsigned side, | |
2491 | char *new_path) | |
2492 | { | |
2493 | int dir_renamed_side = 0; | |
2494 | ||
2495 | if (new_path) { | |
2496 | /* | |
2497 | * Directory renames happen on the other side of history from | |
2498 | * the side that adds new files to the old directory. | |
2499 | */ | |
2500 | dir_renamed_side = 3 - side; | |
2501 | } else { | |
2502 | int val = strintmap_get(&renames->relevant_sources[side], | |
2503 | p->one->path); | |
2504 | if (val == RELEVANT_NO_MORE) { | |
2505 | assert(p->status == 'D'); | |
2506 | strset_add(&renames->cached_irrelevant[side], | |
2507 | p->one->path); | |
2508 | } | |
2509 | if (val <= 0) | |
2510 | return; | |
2511 | } | |
2512 | ||
2513 | if (p->status == 'D') { | |
2514 | /* | |
2515 | * If we already had this delete, we'll just set it's value | |
2516 | * to NULL again, so no harm. | |
2517 | */ | |
2518 | strmap_put(&renames->cached_pairs[side], p->one->path, NULL); | |
2519 | } else if (p->status == 'R') { | |
2520 | if (!new_path) | |
2521 | new_path = p->two->path; | |
2522 | else | |
2523 | cache_new_pair(renames, dir_renamed_side, | |
2524 | p->two->path, new_path, 0); | |
2525 | cache_new_pair(renames, side, p->one->path, new_path, 1); | |
2526 | } else if (p->status == 'A' && new_path) { | |
2527 | cache_new_pair(renames, dir_renamed_side, | |
2528 | p->two->path, new_path, 0); | |
2529 | } | |
2530 | } | |
2531 | ||
e1a124e8 EN |
2532 | static int compare_pairs(const void *a_, const void *b_) |
2533 | { | |
965a7bc2 EN |
2534 | const struct diff_filepair *a = *((const struct diff_filepair **)a_); |
2535 | const struct diff_filepair *b = *((const struct diff_filepair **)b_); | |
2536 | ||
2537 | return strcmp(a->one->path, b->one->path); | |
e1a124e8 EN |
2538 | } |
2539 | ||
356da0f9 | 2540 | /* Call diffcore_rename() to update deleted/added pairs into rename pairs */ |
e1a124e8 | 2541 | static void detect_regular_renames(struct merge_options *opt, |
e1a124e8 EN |
2542 | unsigned side_index) |
2543 | { | |
f39d05ca EN |
2544 | struct diff_options diff_opts; |
2545 | struct rename_info *renames = &opt->priv->renames; | |
2546 | ||
25e65b6d | 2547 | prune_cached_from_relevant(renames, side_index); |
f89b4f2b EN |
2548 | if (!possible_side_renames(renames, side_index)) { |
2549 | /* | |
2550 | * No rename detection needed for this side, but we still need | |
2551 | * to make sure 'adds' are marked correctly in case the other | |
2552 | * side had directory renames. | |
2553 | */ | |
2554 | resolve_diffpair_statuses(&renames->pairs[side_index]); | |
2555 | return; | |
2556 | } | |
2557 | ||
d5098029 | 2558 | partial_clear_dir_rename_count(&renames->dir_rename_count[side_index]); |
f39d05ca EN |
2559 | repo_diff_setup(opt->repo, &diff_opts); |
2560 | diff_opts.flags.recursive = 1; | |
2561 | diff_opts.flags.rename_empty = 0; | |
2562 | diff_opts.detect_rename = DIFF_DETECT_RENAME; | |
2563 | diff_opts.rename_limit = opt->rename_limit; | |
2564 | if (opt->rename_limit <= 0) | |
2565 | diff_opts.rename_limit = 1000; | |
2566 | diff_opts.rename_score = opt->rename_score; | |
2567 | diff_opts.show_rename_progress = opt->show_rename_progress; | |
2568 | diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT; | |
2569 | diff_setup_done(&diff_opts); | |
557ac035 | 2570 | |
f78cf976 | 2571 | diff_queued_diff = renames->pairs[side_index]; |
557ac035 | 2572 | trace2_region_enter("diff", "diffcore_rename", opt->repo); |
0c4fd732 | 2573 | diffcore_rename_extended(&diff_opts, |
174791f0 | 2574 | &renames->relevant_sources[side_index], |
0c4fd732 | 2575 | &renames->dirs_removed[side_index], |
25e65b6d EN |
2576 | &renames->dir_rename_count[side_index], |
2577 | &renames->cached_pairs[side_index]); | |
557ac035 | 2578 | trace2_region_leave("diff", "diffcore_rename", opt->repo); |
f78cf976 | 2579 | resolve_diffpair_statuses(&diff_queued_diff); |
f39d05ca EN |
2580 | |
2581 | if (diff_opts.needed_rename_limit > renames->needed_limit) | |
2582 | renames->needed_limit = diff_opts.needed_rename_limit; | |
2583 | ||
2584 | renames->pairs[side_index] = diff_queued_diff; | |
2585 | ||
2586 | diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT; | |
2587 | diff_queued_diff.nr = 0; | |
2588 | diff_queued_diff.queue = NULL; | |
2589 | diff_flush(&diff_opts); | |
e1a124e8 EN |
2590 | } |
2591 | ||
2592 | /* | |
356da0f9 EN |
2593 | * Get information of all renames which occurred in 'side_pairs', making use |
2594 | * of any implicit directory renames in side_dir_renames (also making use of | |
2595 | * implicit directory renames rename_exclusions as needed by | |
2596 | * check_for_directory_rename()). Add all (updated) renames into result. | |
e1a124e8 EN |
2597 | */ |
2598 | static int collect_renames(struct merge_options *opt, | |
2599 | struct diff_queue_struct *result, | |
fa5e06d6 EN |
2600 | unsigned side_index, |
2601 | struct strmap *dir_renames_for_side, | |
2602 | struct strmap *rename_exclusions) | |
e1a124e8 | 2603 | { |
965a7bc2 | 2604 | int i, clean = 1; |
fa5e06d6 | 2605 | struct strmap collisions; |
965a7bc2 | 2606 | struct diff_queue_struct *side_pairs; |
fa5e06d6 EN |
2607 | struct hashmap_iter iter; |
2608 | struct strmap_entry *entry; | |
965a7bc2 EN |
2609 | struct rename_info *renames = &opt->priv->renames; |
2610 | ||
2611 | side_pairs = &renames->pairs[side_index]; | |
fa5e06d6 | 2612 | compute_collisions(&collisions, dir_renames_for_side, side_pairs); |
965a7bc2 EN |
2613 | |
2614 | for (i = 0; i < side_pairs->nr; ++i) { | |
2615 | struct diff_filepair *p = side_pairs->queue[i]; | |
fa5e06d6 | 2616 | char *new_path; /* non-NULL only with directory renames */ |
965a7bc2 | 2617 | |
fa5e06d6 | 2618 | if (p->status != 'A' && p->status != 'R') { |
2734f2e3 | 2619 | possibly_cache_new_pair(renames, p, side_index, NULL); |
965a7bc2 EN |
2620 | diff_free_filepair(p); |
2621 | continue; | |
2622 | } | |
2623 | ||
fa5e06d6 EN |
2624 | new_path = check_for_directory_rename(opt, p->two->path, |
2625 | side_index, | |
2626 | dir_renames_for_side, | |
2627 | rename_exclusions, | |
2628 | &collisions, | |
2629 | &clean); | |
2630 | ||
2734f2e3 | 2631 | possibly_cache_new_pair(renames, p, side_index, new_path); |
fa5e06d6 EN |
2632 | if (p->status != 'R' && !new_path) { |
2633 | diff_free_filepair(p); | |
2634 | continue; | |
2635 | } | |
2636 | ||
2637 | if (new_path) | |
2638 | apply_directory_rename_modifications(opt, p, new_path); | |
2639 | ||
965a7bc2 EN |
2640 | /* |
2641 | * p->score comes back from diffcore_rename_extended() with | |
2642 | * the similarity of the renamed file. The similarity is | |
2643 | * was used to determine that the two files were related | |
2644 | * and are a rename, which we have already used, but beyond | |
2645 | * that we have no use for the similarity. So p->score is | |
2646 | * now irrelevant. However, process_renames() will need to | |
2647 | * know which side of the merge this rename was associated | |
2648 | * with, so overwrite p->score with that value. | |
2649 | */ | |
2650 | p->score = side_index; | |
2651 | result->queue[result->nr++] = p; | |
2652 | } | |
2653 | ||
fa5e06d6 EN |
2654 | /* Free each value in the collisions map */ |
2655 | strmap_for_each_entry(&collisions, &iter, entry) { | |
2656 | struct collision_info *info = entry->value; | |
2657 | string_list_clear(&info->source_files, 0); | |
2658 | } | |
2659 | /* | |
2660 | * In compute_collisions(), we set collisions.strdup_strings to 0 | |
2661 | * so that we wouldn't have to make another copy of the new_path | |
2662 | * allocated by apply_dir_rename(). But now that we've used them | |
2663 | * and have no other references to these strings, it is time to | |
2664 | * deallocate them. | |
2665 | */ | |
2666 | free_strmap_strings(&collisions); | |
2667 | strmap_clear(&collisions, 1); | |
965a7bc2 | 2668 | return clean; |
e1a124e8 EN |
2669 | } |
2670 | ||
231e2dd4 EN |
2671 | static int detect_and_process_renames(struct merge_options *opt, |
2672 | struct tree *merge_base, | |
2673 | struct tree *side1, | |
2674 | struct tree *side2) | |
2675 | { | |
e1a124e8 EN |
2676 | struct diff_queue_struct combined; |
2677 | struct rename_info *renames = &opt->priv->renames; | |
112e1112 | 2678 | int need_dir_renames, s, clean = 1; |
e1a124e8 EN |
2679 | |
2680 | memset(&combined, 0, sizeof(combined)); | |
f89b4f2b EN |
2681 | if (!possible_renames(renames)) |
2682 | goto cleanup; | |
e1a124e8 | 2683 | |
557ac035 | 2684 | trace2_region_enter("merge", "regular renames", opt->repo); |
f78cf976 EN |
2685 | detect_regular_renames(opt, MERGE_SIDE1); |
2686 | detect_regular_renames(opt, MERGE_SIDE2); | |
25e65b6d EN |
2687 | use_cached_pairs(opt, &renames->cached_pairs[1], &renames->pairs[1]); |
2688 | use_cached_pairs(opt, &renames->cached_pairs[2], &renames->pairs[2]); | |
557ac035 | 2689 | trace2_region_leave("merge", "regular renames", opt->repo); |
e1a124e8 | 2690 | |
557ac035 | 2691 | trace2_region_enter("merge", "directory renames", opt->repo); |
112e1112 EN |
2692 | need_dir_renames = |
2693 | !opt->priv->call_depth && | |
2694 | (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE || | |
2695 | opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT); | |
2696 | ||
2697 | if (need_dir_renames) { | |
2698 | get_provisional_directory_renames(opt, MERGE_SIDE1, &clean); | |
2699 | get_provisional_directory_renames(opt, MERGE_SIDE2, &clean); | |
2700 | handle_directory_level_conflicts(opt); | |
2701 | } | |
2702 | ||
e1a124e8 EN |
2703 | ALLOC_GROW(combined.queue, |
2704 | renames->pairs[1].nr + renames->pairs[2].nr, | |
2705 | combined.alloc); | |
fa5e06d6 EN |
2706 | clean &= collect_renames(opt, &combined, MERGE_SIDE1, |
2707 | &renames->dir_renames[2], | |
2708 | &renames->dir_renames[1]); | |
2709 | clean &= collect_renames(opt, &combined, MERGE_SIDE2, | |
2710 | &renames->dir_renames[1], | |
2711 | &renames->dir_renames[2]); | |
72b30910 | 2712 | STABLE_QSORT(combined.queue, combined.nr, compare_pairs); |
557ac035 | 2713 | trace2_region_leave("merge", "directory renames", opt->repo); |
e1a124e8 | 2714 | |
557ac035 | 2715 | trace2_region_enter("merge", "process renames", opt->repo); |
e1a124e8 | 2716 | clean &= process_renames(opt, &combined); |
557ac035 | 2717 | trace2_region_leave("merge", "process renames", opt->repo); |
e1a124e8 | 2718 | |
f89b4f2b EN |
2719 | goto simple_cleanup; /* collect_renames() handles some of cleanup */ |
2720 | ||
2721 | cleanup: | |
2722 | /* | |
2723 | * Free now unneeded filepairs, which would have been handled | |
2724 | * in collect_renames() normally but we skipped that code. | |
2725 | */ | |
2726 | for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) { | |
2727 | struct diff_queue_struct *side_pairs; | |
2728 | int i; | |
2729 | ||
2730 | side_pairs = &renames->pairs[s]; | |
2731 | for (i = 0; i < side_pairs->nr; ++i) { | |
2732 | struct diff_filepair *p = side_pairs->queue[i]; | |
2733 | diff_free_filepair(p); | |
2734 | } | |
2735 | } | |
2736 | ||
2737 | simple_cleanup: | |
e1a124e8 EN |
2738 | /* Free memory for renames->pairs[] and combined */ |
2739 | for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) { | |
2740 | free(renames->pairs[s].queue); | |
2741 | DIFF_QUEUE_CLEAR(&renames->pairs[s]); | |
2742 | } | |
2743 | if (combined.nr) { | |
2744 | int i; | |
2745 | for (i = 0; i < combined.nr; i++) | |
2746 | diff_free_filepair(combined.queue[i]); | |
2747 | free(combined.queue); | |
2748 | } | |
231e2dd4 | 2749 | |
231e2dd4 EN |
2750 | return clean; |
2751 | } | |
2752 | ||
04af1879 EN |
2753 | /*** Function Grouping: functions related to process_entries() ***/ |
2754 | ||
5a3743da | 2755 | static int sort_dirs_next_to_their_children(const char *one, const char *two) |
8adffaa8 | 2756 | { |
5a3743da EN |
2757 | unsigned char c1, c2; |
2758 | ||
2759 | /* | |
2760 | * Here we only care that entries for directories appear adjacent | |
2761 | * to and before files underneath the directory. We can achieve | |
2762 | * that by pretending to add a trailing slash to every file and | |
2763 | * then sorting. In other words, we do not want the natural | |
2764 | * sorting of | |
2765 | * foo | |
2766 | * foo.txt | |
2767 | * foo/bar | |
2768 | * Instead, we want "foo" to sort as though it were "foo/", so that | |
2769 | * we instead get | |
2770 | * foo.txt | |
2771 | * foo | |
2772 | * foo/bar | |
2773 | * To achieve this, we basically implement our own strcmp, except that | |
2774 | * if we get to the end of either string instead of comparing NUL to | |
2775 | * another character, we compare '/' to it. | |
2776 | * | |
2777 | * If this unusual "sort as though '/' were appended" perplexes | |
2778 | * you, perhaps it will help to note that this is not the final | |
2779 | * sort. write_tree() will sort again without the trailing slash | |
2780 | * magic, but just on paths immediately under a given tree. | |
8adffaa8 | 2781 | * |
5a3743da EN |
2782 | * The reason to not use df_name_compare directly was that it was |
2783 | * just too expensive (we don't have the string lengths handy), so | |
2784 | * it was reimplemented. | |
8adffaa8 | 2785 | */ |
5a3743da | 2786 | |
8adffaa8 | 2787 | /* |
5a3743da EN |
2788 | * NOTE: This function will never be called with two equal strings, |
2789 | * because it is used to sort the keys of a strmap, and strmaps have | |
2790 | * unique keys by construction. That simplifies our c1==c2 handling | |
2791 | * below. | |
8adffaa8 | 2792 | */ |
5a3743da EN |
2793 | |
2794 | while (*one && (*one == *two)) { | |
2795 | one++; | |
2796 | two++; | |
2797 | } | |
2798 | ||
2799 | c1 = *one ? *one : '/'; | |
2800 | c2 = *two ? *two : '/'; | |
2801 | ||
2802 | if (c1 == c2) { | |
2803 | /* Getting here means one is a leading directory of the other */ | |
2804 | return (*one) ? 1 : -1; | |
2805 | } else | |
2806 | return c1 - c2; | |
8adffaa8 EN |
2807 | } |
2808 | ||
3860220b EN |
2809 | static int read_oid_strbuf(struct merge_options *opt, |
2810 | const struct object_id *oid, | |
2811 | struct strbuf *dst) | |
2812 | { | |
2813 | void *buf; | |
2814 | enum object_type type; | |
2815 | unsigned long size; | |
2816 | buf = read_object_file(oid, &type, &size); | |
2817 | if (!buf) | |
2818 | return err(opt, _("cannot read object %s"), oid_to_hex(oid)); | |
2819 | if (type != OBJ_BLOB) { | |
2820 | free(buf); | |
2821 | return err(opt, _("object %s is not a blob"), oid_to_hex(oid)); | |
2822 | } | |
2823 | strbuf_attach(dst, buf, size, size + 1); | |
2824 | return 0; | |
2825 | } | |
2826 | ||
2827 | static int blob_unchanged(struct merge_options *opt, | |
2828 | const struct version_info *base, | |
2829 | const struct version_info *side, | |
2830 | const char *path) | |
2831 | { | |
2832 | struct strbuf basebuf = STRBUF_INIT; | |
2833 | struct strbuf sidebuf = STRBUF_INIT; | |
2834 | int ret = 0; /* assume changed for safety */ | |
8e978529 | 2835 | struct index_state *idx = &opt->priv->attr_index; |
3860220b EN |
2836 | |
2837 | if (!idx->initialized) | |
2838 | initialize_attr_index(opt); | |
2839 | ||
2840 | if (base->mode != side->mode) | |
2841 | return 0; | |
2842 | if (oideq(&base->oid, &side->oid)) | |
2843 | return 1; | |
2844 | ||
2845 | if (read_oid_strbuf(opt, &base->oid, &basebuf) || | |
2846 | read_oid_strbuf(opt, &side->oid, &sidebuf)) | |
2847 | goto error_return; | |
2848 | /* | |
2849 | * Note: binary | is used so that both renormalizations are | |
2850 | * performed. Comparison can be skipped if both files are | |
2851 | * unchanged since their sha1s have already been compared. | |
2852 | */ | |
2853 | if (renormalize_buffer(idx, path, basebuf.buf, basebuf.len, &basebuf) | | |
2854 | renormalize_buffer(idx, path, sidebuf.buf, sidebuf.len, &sidebuf)) | |
2855 | ret = (basebuf.len == sidebuf.len && | |
2856 | !memcmp(basebuf.buf, sidebuf.buf, basebuf.len)); | |
2857 | ||
2858 | error_return: | |
2859 | strbuf_release(&basebuf); | |
2860 | strbuf_release(&sidebuf); | |
2861 | return ret; | |
2862 | } | |
2863 | ||
a9945bba | 2864 | struct directory_versions { |
bb470f4e EN |
2865 | /* |
2866 | * versions: list of (basename -> version_info) | |
2867 | * | |
2868 | * The basenames are in reverse lexicographic order of full pathnames, | |
2869 | * as processed in process_entries(). This puts all entries within | |
2870 | * a directory together, and covers the directory itself after | |
2871 | * everything within it, allowing us to write subtrees before needing | |
2872 | * to record information for the tree itself. | |
2873 | */ | |
a9945bba | 2874 | struct string_list versions; |
bb470f4e EN |
2875 | |
2876 | /* | |
2877 | * offsets: list of (full relative path directories -> integer offsets) | |
2878 | * | |
2879 | * Since versions contains basenames from files in multiple different | |
2880 | * directories, we need to know which entries in versions correspond | |
2881 | * to which directories. Values of e.g. | |
2882 | * "" 0 | |
2883 | * src 2 | |
2884 | * src/moduleA 5 | |
2885 | * Would mean that entries 0-1 of versions are files in the toplevel | |
2886 | * directory, entries 2-4 are files under src/, and the remaining | |
2887 | * entries starting at index 5 are files under src/moduleA/. | |
2888 | */ | |
2889 | struct string_list offsets; | |
2890 | ||
2891 | /* | |
2892 | * last_directory: directory that previously processed file found in | |
2893 | * | |
2894 | * last_directory starts NULL, but records the directory in which the | |
2895 | * previous file was found within. As soon as | |
2896 | * directory(current_file) != last_directory | |
2897 | * then we need to start updating accounting in versions & offsets. | |
2898 | * Note that last_directory is always the last path in "offsets" (or | |
2899 | * NULL if "offsets" is empty) so this exists just for quick access. | |
2900 | */ | |
2901 | const char *last_directory; | |
2902 | ||
2903 | /* last_directory_len: cached computation of strlen(last_directory) */ | |
2904 | unsigned last_directory_len; | |
a9945bba EN |
2905 | }; |
2906 | ||
ee4012dc EN |
2907 | static int tree_entry_order(const void *a_, const void *b_) |
2908 | { | |
2909 | const struct string_list_item *a = a_; | |
2910 | const struct string_list_item *b = b_; | |
2911 | ||
2912 | const struct merged_info *ami = a->util; | |
2913 | const struct merged_info *bmi = b->util; | |
2914 | return base_name_compare(a->string, strlen(a->string), ami->result.mode, | |
2915 | b->string, strlen(b->string), bmi->result.mode); | |
2916 | } | |
2917 | ||
2918 | static void write_tree(struct object_id *result_oid, | |
2919 | struct string_list *versions, | |
2920 | unsigned int offset, | |
2921 | size_t hash_size) | |
2922 | { | |
2923 | size_t maxlen = 0, extra; | |
c1ea48a8 | 2924 | unsigned int nr; |
ee4012dc | 2925 | struct strbuf buf = STRBUF_INIT; |
ee4012dc EN |
2926 | int i; |
2927 | ||
c1ea48a8 AH |
2928 | assert(offset <= versions->nr); |
2929 | nr = versions->nr - offset; | |
2930 | if (versions->nr) | |
257ae76b | 2931 | /* No need for STABLE_QSORT -- filenames must be unique */ |
c1ea48a8 | 2932 | QSORT(versions->items + offset, nr, tree_entry_order); |
ee4012dc EN |
2933 | |
2934 | /* Pre-allocate some space in buf */ | |
2935 | extra = hash_size + 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */ | |
2936 | for (i = 0; i < nr; i++) { | |
2937 | maxlen += strlen(versions->items[offset+i].string) + extra; | |
2938 | } | |
2939 | strbuf_grow(&buf, maxlen); | |
2940 | ||
2941 | /* Write each entry out to buf */ | |
2942 | for (i = 0; i < nr; i++) { | |
2943 | struct merged_info *mi = versions->items[offset+i].util; | |
2944 | struct version_info *ri = &mi->result; | |
2945 | strbuf_addf(&buf, "%o %s%c", | |
2946 | ri->mode, | |
2947 | versions->items[offset+i].string, '\0'); | |
2948 | strbuf_add(&buf, ri->oid.hash, hash_size); | |
2949 | } | |
2950 | ||
2951 | /* Write this object file out, and record in result_oid */ | |
2952 | write_object_file(buf.buf, buf.len, tree_type, result_oid); | |
2953 | strbuf_release(&buf); | |
2954 | } | |
2955 | ||
a9945bba EN |
2956 | static void record_entry_for_tree(struct directory_versions *dir_metadata, |
2957 | const char *path, | |
2958 | struct merged_info *mi) | |
2959 | { | |
2960 | const char *basename; | |
2961 | ||
2962 | if (mi->is_null) | |
2963 | /* nothing to record */ | |
2964 | return; | |
2965 | ||
2966 | basename = path + mi->basename_offset; | |
2967 | assert(strchr(basename, '/') == NULL); | |
2968 | string_list_append(&dir_metadata->versions, | |
2969 | basename)->util = &mi->result; | |
2970 | } | |
2971 | ||
bb470f4e EN |
2972 | static void write_completed_directory(struct merge_options *opt, |
2973 | const char *new_directory_name, | |
2974 | struct directory_versions *info) | |
2975 | { | |
2976 | const char *prev_dir; | |
2977 | struct merged_info *dir_info = NULL; | |
2978 | unsigned int offset; | |
2979 | ||
2980 | /* | |
2981 | * Some explanation of info->versions and info->offsets... | |
2982 | * | |
2983 | * process_entries() iterates over all relevant files AND | |
2984 | * directories in reverse lexicographic order, and calls this | |
2985 | * function. Thus, an example of the paths that process_entries() | |
2986 | * could operate on (along with the directories for those paths | |
2987 | * being shown) is: | |
2988 | * | |
2989 | * xtract.c "" | |
2990 | * tokens.txt "" | |
2991 | * src/moduleB/umm.c src/moduleB | |
2992 | * src/moduleB/stuff.h src/moduleB | |
2993 | * src/moduleB/baz.c src/moduleB | |
2994 | * src/moduleB src | |
2995 | * src/moduleA/foo.c src/moduleA | |
2996 | * src/moduleA/bar.c src/moduleA | |
2997 | * src/moduleA src | |
2998 | * src "" | |
2999 | * Makefile "" | |
3000 | * | |
3001 | * info->versions: | |
3002 | * | |
3003 | * always contains the unprocessed entries and their | |
3004 | * version_info information. For example, after the first five | |
3005 | * entries above, info->versions would be: | |
3006 | * | |
3007 | * xtract.c <xtract.c's version_info> | |
3008 | * token.txt <token.txt's version_info> | |
3009 | * umm.c <src/moduleB/umm.c's version_info> | |
3010 | * stuff.h <src/moduleB/stuff.h's version_info> | |
3011 | * baz.c <src/moduleB/baz.c's version_info> | |
3012 | * | |
3013 | * Once a subdirectory is completed we remove the entries in | |
3014 | * that subdirectory from info->versions, writing it as a tree | |
3015 | * (write_tree()). Thus, as soon as we get to src/moduleB, | |
3016 | * info->versions would be updated to | |
3017 | * | |
3018 | * xtract.c <xtract.c's version_info> | |
3019 | * token.txt <token.txt's version_info> | |
3020 | * moduleB <src/moduleB's version_info> | |
3021 | * | |
3022 | * info->offsets: | |
3023 | * | |
3024 | * helps us track which entries in info->versions correspond to | |
3025 | * which directories. When we are N directories deep (e.g. 4 | |
3026 | * for src/modA/submod/subdir/), we have up to N+1 unprocessed | |
3027 | * directories (+1 because of toplevel dir). Corresponding to | |
3028 | * the info->versions example above, after processing five entries | |
3029 | * info->offsets will be: | |
3030 | * | |
3031 | * "" 0 | |
3032 | * src/moduleB 2 | |
3033 | * | |
3034 | * which is used to know that xtract.c & token.txt are from the | |
3035 | * toplevel dirctory, while umm.c & stuff.h & baz.c are from the | |
3036 | * src/moduleB directory. Again, following the example above, | |
3037 | * once we need to process src/moduleB, then info->offsets is | |
3038 | * updated to | |
3039 | * | |
3040 | * "" 0 | |
3041 | * src 2 | |
3042 | * | |
3043 | * which says that moduleB (and only moduleB so far) is in the | |
3044 | * src directory. | |
3045 | * | |
3046 | * One unique thing to note about info->offsets here is that | |
3047 | * "src" was not added to info->offsets until there was a path | |
3048 | * (a file OR directory) immediately below src/ that got | |
3049 | * processed. | |
3050 | * | |
3051 | * Since process_entry() just appends new entries to info->versions, | |
3052 | * write_completed_directory() only needs to do work if the next path | |
3053 | * is in a directory that is different than the last directory found | |
3054 | * in info->offsets. | |
3055 | */ | |
3056 | ||
3057 | /* | |
3058 | * If we are working with the same directory as the last entry, there | |
3059 | * is no work to do. (See comments above the directory_name member of | |
3060 | * struct merged_info for why we can use pointer comparison instead of | |
3061 | * strcmp here.) | |
3062 | */ | |
3063 | if (new_directory_name == info->last_directory) | |
3064 | return; | |
3065 | ||
3066 | /* | |
3067 | * If we are just starting (last_directory is NULL), or last_directory | |
3068 | * is a prefix of the current directory, then we can just update | |
3069 | * info->offsets to record the offset where we started this directory | |
3070 | * and update last_directory to have quick access to it. | |
3071 | */ | |
3072 | if (info->last_directory == NULL || | |
3073 | !strncmp(new_directory_name, info->last_directory, | |
3074 | info->last_directory_len)) { | |
3075 | uintptr_t offset = info->versions.nr; | |
3076 | ||
3077 | info->last_directory = new_directory_name; | |
3078 | info->last_directory_len = strlen(info->last_directory); | |
3079 | /* | |
3080 | * Record the offset into info->versions where we will | |
3081 | * start recording basenames of paths found within | |
3082 | * new_directory_name. | |
3083 | */ | |
3084 | string_list_append(&info->offsets, | |
3085 | info->last_directory)->util = (void*)offset; | |
3086 | return; | |
3087 | } | |
3088 | ||
3089 | /* | |
3090 | * The next entry that will be processed will be within | |
3091 | * new_directory_name. Since at this point we know that | |
3092 | * new_directory_name is within a different directory than | |
3093 | * info->last_directory, we have all entries for info->last_directory | |
3094 | * in info->versions and we need to create a tree object for them. | |
3095 | */ | |
3096 | dir_info = strmap_get(&opt->priv->paths, info->last_directory); | |
3097 | assert(dir_info); | |
3098 | offset = (uintptr_t)info->offsets.items[info->offsets.nr-1].util; | |
3099 | if (offset == info->versions.nr) { | |
3100 | /* | |
3101 | * Actually, we don't need to create a tree object in this | |
3102 | * case. Whenever all files within a directory disappear | |
3103 | * during the merge (e.g. unmodified on one side and | |
3104 | * deleted on the other, or files were renamed elsewhere), | |
3105 | * then we get here and the directory itself needs to be | |
3106 | * omitted from its parent tree as well. | |
3107 | */ | |
3108 | dir_info->is_null = 1; | |
3109 | } else { | |
3110 | /* | |
3111 | * Write out the tree to the git object directory, and also | |
3112 | * record the mode and oid in dir_info->result. | |
3113 | */ | |
3114 | dir_info->is_null = 0; | |
3115 | dir_info->result.mode = S_IFDIR; | |
3116 | write_tree(&dir_info->result.oid, &info->versions, offset, | |
3117 | opt->repo->hash_algo->rawsz); | |
3118 | } | |
3119 | ||
3120 | /* | |
3121 | * We've now used several entries from info->versions and one entry | |
3122 | * from info->offsets, so we get rid of those values. | |
3123 | */ | |
3124 | info->offsets.nr--; | |
3125 | info->versions.nr = offset; | |
3126 | ||
3127 | /* | |
3128 | * Now we've taken care of the completed directory, but we need to | |
3129 | * prepare things since future entries will be in | |
3130 | * new_directory_name. (In particular, process_entry() will be | |
3131 | * appending new entries to info->versions.) So, we need to make | |
3132 | * sure new_directory_name is the last entry in info->offsets. | |
3133 | */ | |
3134 | prev_dir = info->offsets.nr == 0 ? NULL : | |
3135 | info->offsets.items[info->offsets.nr-1].string; | |
3136 | if (new_directory_name != prev_dir) { | |
3137 | uintptr_t c = info->versions.nr; | |
3138 | string_list_append(&info->offsets, | |
3139 | new_directory_name)->util = (void*)c; | |
3140 | } | |
3141 | ||
3142 | /* And, of course, we need to update last_directory to match. */ | |
3143 | info->last_directory = new_directory_name; | |
3144 | info->last_directory_len = strlen(info->last_directory); | |
3145 | } | |
3146 | ||
6a02dd90 EN |
3147 | /* Per entry merge function */ |
3148 | static void process_entry(struct merge_options *opt, | |
3149 | const char *path, | |
a9945bba EN |
3150 | struct conflict_info *ci, |
3151 | struct directory_versions *dir_metadata) | |
6a02dd90 | 3152 | { |
23366d2a EN |
3153 | int df_file_index = 0; |
3154 | ||
6a02dd90 EN |
3155 | VERIFY_CI(ci); |
3156 | assert(ci->filemask >= 0 && ci->filemask <= 7); | |
3157 | /* ci->match_mask == 7 was handled in collect_merge_info_callback() */ | |
3158 | assert(ci->match_mask == 0 || ci->match_mask == 3 || | |
3159 | ci->match_mask == 5 || ci->match_mask == 6); | |
3160 | ||
a9945bba EN |
3161 | if (ci->dirmask) { |
3162 | record_entry_for_tree(dir_metadata, path, &ci->merged); | |
3163 | if (ci->filemask == 0) | |
3164 | /* nothing else to handle */ | |
3165 | return; | |
3166 | assert(ci->df_conflict); | |
3167 | } | |
3168 | ||
0ccfa4e5 EN |
3169 | if (ci->df_conflict && ci->merged.result.mode == 0) { |
3170 | int i; | |
3171 | ||
3172 | /* | |
3173 | * directory no longer in the way, but we do have a file we | |
3174 | * need to place here so we need to clean away the "directory | |
3175 | * merges to nothing" result. | |
3176 | */ | |
3177 | ci->df_conflict = 0; | |
3178 | assert(ci->filemask != 0); | |
3179 | ci->merged.clean = 0; | |
3180 | ci->merged.is_null = 0; | |
3181 | /* and we want to zero out any directory-related entries */ | |
3182 | ci->match_mask = (ci->match_mask & ~ci->dirmask); | |
3183 | ci->dirmask = 0; | |
3184 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
3185 | if (ci->filemask & (1 << i)) | |
3186 | continue; | |
3187 | ci->stages[i].mode = 0; | |
14228447 | 3188 | oidcpy(&ci->stages[i].oid, null_oid()); |
0ccfa4e5 EN |
3189 | } |
3190 | } else if (ci->df_conflict && ci->merged.result.mode != 0) { | |
23366d2a EN |
3191 | /* |
3192 | * This started out as a D/F conflict, and the entries in | |
3193 | * the competing directory were not removed by the merge as | |
3194 | * evidenced by write_completed_directory() writing a value | |
3195 | * to ci->merged.result.mode. | |
3196 | */ | |
3197 | struct conflict_info *new_ci; | |
3198 | const char *branch; | |
3199 | const char *old_path = path; | |
3200 | int i; | |
3201 | ||
3202 | assert(ci->merged.result.mode == S_IFDIR); | |
3203 | ||
3204 | /* | |
3205 | * If filemask is 1, we can just ignore the file as having | |
3206 | * been deleted on both sides. We do not want to overwrite | |
3207 | * ci->merged.result, since it stores the tree for all the | |
3208 | * files under it. | |
3209 | */ | |
3210 | if (ci->filemask == 1) { | |
3211 | ci->filemask = 0; | |
3212 | return; | |
3213 | } | |
3214 | ||
3215 | /* | |
3216 | * This file still exists on at least one side, and we want | |
3217 | * the directory to remain here, so we need to move this | |
3218 | * path to some new location. | |
3219 | */ | |
ca56dadb | 3220 | CALLOC_ARRAY(new_ci, 1); |
23366d2a EN |
3221 | /* We don't really want new_ci->merged.result copied, but it'll |
3222 | * be overwritten below so it doesn't matter. We also don't | |
3223 | * want any directory mode/oid values copied, but we'll zero | |
3224 | * those out immediately. We do want the rest of ci copied. | |
3225 | */ | |
3226 | memcpy(new_ci, ci, sizeof(*ci)); | |
3227 | new_ci->match_mask = (new_ci->match_mask & ~new_ci->dirmask); | |
3228 | new_ci->dirmask = 0; | |
3229 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
3230 | if (new_ci->filemask & (1 << i)) | |
3231 | continue; | |
3232 | /* zero out any entries related to directories */ | |
3233 | new_ci->stages[i].mode = 0; | |
14228447 | 3234 | oidcpy(&new_ci->stages[i].oid, null_oid()); |
23366d2a EN |
3235 | } |
3236 | ||
3237 | /* | |
3238 | * Find out which side this file came from; note that we | |
3239 | * cannot just use ci->filemask, because renames could cause | |
3240 | * the filemask to go back to 7. So we use dirmask, then | |
3241 | * pick the opposite side's index. | |
3242 | */ | |
3243 | df_file_index = (ci->dirmask & (1 << 1)) ? 2 : 1; | |
3244 | branch = (df_file_index == 1) ? opt->branch1 : opt->branch2; | |
3245 | path = unique_path(&opt->priv->paths, path, branch); | |
3246 | strmap_put(&opt->priv->paths, path, new_ci); | |
3247 | ||
3248 | path_msg(opt, path, 0, | |
3249 | _("CONFLICT (file/directory): directory in the way " | |
3250 | "of %s from %s; moving it to %s instead."), | |
3251 | old_path, branch, path); | |
3252 | ||
3253 | /* | |
3254 | * Zero out the filemask for the old ci. At this point, ci | |
3255 | * was just an entry for a directory, so we don't need to | |
3256 | * do anything more with it. | |
3257 | */ | |
3258 | ci->filemask = 0; | |
3259 | ||
3260 | /* | |
3261 | * Now note that we're working on the new entry (path was | |
3262 | * updated above. | |
3263 | */ | |
3264 | ci = new_ci; | |
6a02dd90 EN |
3265 | } |
3266 | ||
3267 | /* | |
3268 | * NOTE: Below there is a long switch-like if-elseif-elseif... block | |
3269 | * which the code goes through even for the df_conflict cases | |
23366d2a | 3270 | * above. |
6a02dd90 EN |
3271 | */ |
3272 | if (ci->match_mask) { | |
a492d533 | 3273 | ci->merged.clean = !ci->df_conflict && !ci->path_conflict; |
6a02dd90 EN |
3274 | if (ci->match_mask == 6) { |
3275 | /* stages[1] == stages[2] */ | |
3276 | ci->merged.result.mode = ci->stages[1].mode; | |
3277 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
3278 | } else { | |
3279 | /* determine the mask of the side that didn't match */ | |
3280 | unsigned int othermask = 7 & ~ci->match_mask; | |
3281 | int side = (othermask == 4) ? 2 : 1; | |
3282 | ||
3283 | ci->merged.result.mode = ci->stages[side].mode; | |
3284 | ci->merged.is_null = !ci->merged.result.mode; | |
a492d533 EN |
3285 | if (ci->merged.is_null) |
3286 | ci->merged.clean = 1; | |
6a02dd90 EN |
3287 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); |
3288 | ||
3289 | assert(othermask == 2 || othermask == 4); | |
3290 | assert(ci->merged.is_null == | |
3291 | (ci->filemask == ci->match_mask)); | |
3292 | } | |
3293 | } else if (ci->filemask >= 6 && | |
3294 | (S_IFMT & ci->stages[1].mode) != | |
3295 | (S_IFMT & ci->stages[2].mode)) { | |
4ef88fc3 EN |
3296 | /* Two different items from (file/submodule/symlink) */ |
3297 | if (opt->priv->call_depth) { | |
3298 | /* Just use the version from the merge base */ | |
3299 | ci->merged.clean = 0; | |
3300 | oidcpy(&ci->merged.result.oid, &ci->stages[0].oid); | |
3301 | ci->merged.result.mode = ci->stages[0].mode; | |
3302 | ci->merged.is_null = (ci->merged.result.mode == 0); | |
3303 | } else { | |
3304 | /* Handle by renaming one or both to separate paths. */ | |
3305 | unsigned o_mode = ci->stages[0].mode; | |
3306 | unsigned a_mode = ci->stages[1].mode; | |
3307 | unsigned b_mode = ci->stages[2].mode; | |
3308 | struct conflict_info *new_ci; | |
3309 | const char *a_path = NULL, *b_path = NULL; | |
3310 | int rename_a = 0, rename_b = 0; | |
3311 | ||
3312 | new_ci = xmalloc(sizeof(*new_ci)); | |
3313 | ||
3314 | if (S_ISREG(a_mode)) | |
3315 | rename_a = 1; | |
3316 | else if (S_ISREG(b_mode)) | |
3317 | rename_b = 1; | |
3318 | else { | |
3319 | rename_a = 1; | |
3320 | rename_b = 1; | |
3321 | } | |
3322 | ||
0e59f7ad AH |
3323 | if (rename_a && rename_b) { |
3324 | path_msg(opt, path, 0, | |
3325 | _("CONFLICT (distinct types): %s had " | |
3326 | "different types on each side; " | |
3327 | "renamed both of them so each can " | |
3328 | "be recorded somewhere."), | |
3329 | path); | |
3330 | } else { | |
3331 | path_msg(opt, path, 0, | |
3332 | _("CONFLICT (distinct types): %s had " | |
3333 | "different types on each side; " | |
3334 | "renamed one of them so each can be " | |
3335 | "recorded somewhere."), | |
3336 | path); | |
3337 | } | |
4ef88fc3 EN |
3338 | |
3339 | ci->merged.clean = 0; | |
3340 | memcpy(new_ci, ci, sizeof(*new_ci)); | |
3341 | ||
3342 | /* Put b into new_ci, removing a from stages */ | |
3343 | new_ci->merged.result.mode = ci->stages[2].mode; | |
3344 | oidcpy(&new_ci->merged.result.oid, &ci->stages[2].oid); | |
3345 | new_ci->stages[1].mode = 0; | |
14228447 | 3346 | oidcpy(&new_ci->stages[1].oid, null_oid()); |
4ef88fc3 EN |
3347 | new_ci->filemask = 5; |
3348 | if ((S_IFMT & b_mode) != (S_IFMT & o_mode)) { | |
3349 | new_ci->stages[0].mode = 0; | |
14228447 | 3350 | oidcpy(&new_ci->stages[0].oid, null_oid()); |
4ef88fc3 EN |
3351 | new_ci->filemask = 4; |
3352 | } | |
3353 | ||
3354 | /* Leave only a in ci, fixing stages. */ | |
3355 | ci->merged.result.mode = ci->stages[1].mode; | |
3356 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
3357 | ci->stages[2].mode = 0; | |
14228447 | 3358 | oidcpy(&ci->stages[2].oid, null_oid()); |
4ef88fc3 EN |
3359 | ci->filemask = 3; |
3360 | if ((S_IFMT & a_mode) != (S_IFMT & o_mode)) { | |
3361 | ci->stages[0].mode = 0; | |
14228447 | 3362 | oidcpy(&ci->stages[0].oid, null_oid()); |
4ef88fc3 EN |
3363 | ci->filemask = 2; |
3364 | } | |
3365 | ||
3366 | /* Insert entries into opt->priv_paths */ | |
3367 | assert(rename_a || rename_b); | |
3368 | if (rename_a) { | |
3369 | a_path = unique_path(&opt->priv->paths, | |
3370 | path, opt->branch1); | |
3371 | strmap_put(&opt->priv->paths, a_path, ci); | |
3372 | } | |
3373 | ||
3374 | if (rename_b) | |
3375 | b_path = unique_path(&opt->priv->paths, | |
3376 | path, opt->branch2); | |
3377 | else | |
3378 | b_path = path; | |
3379 | strmap_put(&opt->priv->paths, b_path, new_ci); | |
3380 | ||
3381 | if (rename_a && rename_b) { | |
3382 | strmap_remove(&opt->priv->paths, path, 0); | |
3383 | /* | |
3384 | * We removed path from opt->priv->paths. path | |
3385 | * will also eventually need to be freed, but | |
3386 | * it may still be used by e.g. ci->pathnames. | |
3387 | * So, store it in another string-list for now. | |
3388 | */ | |
3389 | string_list_append(&opt->priv->paths_to_free, | |
3390 | path); | |
3391 | } | |
3392 | ||
3393 | /* | |
3394 | * Do special handling for b_path since process_entry() | |
3395 | * won't be called on it specially. | |
3396 | */ | |
3397 | strmap_put(&opt->priv->conflicted, b_path, new_ci); | |
3398 | record_entry_for_tree(dir_metadata, b_path, | |
3399 | &new_ci->merged); | |
3400 | ||
3401 | /* | |
3402 | * Remaining code for processing this entry should | |
3403 | * think in terms of processing a_path. | |
3404 | */ | |
3405 | if (a_path) | |
3406 | path = a_path; | |
3407 | } | |
6a02dd90 | 3408 | } else if (ci->filemask >= 6) { |
991bbdca EN |
3409 | /* Need a two-way or three-way content merge */ |
3410 | struct version_info merged_file; | |
3411 | unsigned clean_merge; | |
3412 | struct version_info *o = &ci->stages[0]; | |
3413 | struct version_info *a = &ci->stages[1]; | |
3414 | struct version_info *b = &ci->stages[2]; | |
3415 | ||
3416 | clean_merge = handle_content_merge(opt, path, o, a, b, | |
3417 | ci->pathnames, | |
3418 | opt->priv->call_depth * 2, | |
3419 | &merged_file); | |
3420 | ci->merged.clean = clean_merge && | |
3421 | !ci->df_conflict && !ci->path_conflict; | |
3422 | ci->merged.result.mode = merged_file.mode; | |
3423 | ci->merged.is_null = (merged_file.mode == 0); | |
3424 | oidcpy(&ci->merged.result.oid, &merged_file.oid); | |
3425 | if (clean_merge && ci->df_conflict) { | |
3426 | assert(df_file_index == 1 || df_file_index == 2); | |
3427 | ci->filemask = 1 << df_file_index; | |
3428 | ci->stages[df_file_index].mode = merged_file.mode; | |
3429 | oidcpy(&ci->stages[df_file_index].oid, &merged_file.oid); | |
3430 | } | |
3431 | if (!clean_merge) { | |
3432 | const char *reason = _("content"); | |
3433 | if (ci->filemask == 6) | |
3434 | reason = _("add/add"); | |
3435 | if (S_ISGITLINK(merged_file.mode)) | |
3436 | reason = _("submodule"); | |
3437 | path_msg(opt, path, 0, | |
3438 | _("CONFLICT (%s): Merge conflict in %s"), | |
3439 | reason, path); | |
3440 | } | |
6a02dd90 EN |
3441 | } else if (ci->filemask == 3 || ci->filemask == 5) { |
3442 | /* Modify/delete */ | |
c5a6f655 EN |
3443 | const char *modify_branch, *delete_branch; |
3444 | int side = (ci->filemask == 5) ? 2 : 1; | |
3445 | int index = opt->priv->call_depth ? 0 : side; | |
3446 | ||
3447 | ci->merged.result.mode = ci->stages[index].mode; | |
3448 | oidcpy(&ci->merged.result.oid, &ci->stages[index].oid); | |
3449 | ci->merged.clean = 0; | |
3450 | ||
3451 | modify_branch = (side == 1) ? opt->branch1 : opt->branch2; | |
3452 | delete_branch = (side == 1) ? opt->branch2 : opt->branch1; | |
3453 | ||
3860220b EN |
3454 | if (opt->renormalize && |
3455 | blob_unchanged(opt, &ci->stages[0], &ci->stages[side], | |
3456 | path)) { | |
3457 | ci->merged.is_null = 1; | |
3458 | ci->merged.clean = 1; | |
a492d533 | 3459 | assert(!ci->df_conflict && !ci->path_conflict); |
3860220b EN |
3460 | } else if (ci->path_conflict && |
3461 | oideq(&ci->stages[0].oid, &ci->stages[side].oid)) { | |
2e91ddd2 EN |
3462 | /* |
3463 | * This came from a rename/delete; no action to take, | |
3464 | * but avoid printing "modify/delete" conflict notice | |
3465 | * since the contents were not modified. | |
3466 | */ | |
3467 | } else { | |
3468 | path_msg(opt, path, 0, | |
3469 | _("CONFLICT (modify/delete): %s deleted in %s " | |
3470 | "and modified in %s. Version %s of %s left " | |
3471 | "in tree."), | |
3472 | path, delete_branch, modify_branch, | |
3473 | modify_branch, path); | |
3474 | } | |
6a02dd90 EN |
3475 | } else if (ci->filemask == 2 || ci->filemask == 4) { |
3476 | /* Added on one side */ | |
3477 | int side = (ci->filemask == 4) ? 2 : 1; | |
3478 | ci->merged.result.mode = ci->stages[side].mode; | |
3479 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
53e88a03 | 3480 | ci->merged.clean = !ci->df_conflict && !ci->path_conflict; |
6a02dd90 EN |
3481 | } else if (ci->filemask == 1) { |
3482 | /* Deleted on both sides */ | |
3483 | ci->merged.is_null = 1; | |
3484 | ci->merged.result.mode = 0; | |
14228447 | 3485 | oidcpy(&ci->merged.result.oid, null_oid()); |
a492d533 | 3486 | assert(!ci->df_conflict); |
53e88a03 | 3487 | ci->merged.clean = !ci->path_conflict; |
6a02dd90 EN |
3488 | } |
3489 | ||
3490 | /* | |
3491 | * If still conflicted, record it separately. This allows us to later | |
3492 | * iterate over just conflicted entries when updating the index instead | |
3493 | * of iterating over all entries. | |
3494 | */ | |
3495 | if (!ci->merged.clean) | |
3496 | strmap_put(&opt->priv->conflicted, path, ci); | |
ef68c3d8 EN |
3497 | |
3498 | /* Record metadata for ci->merged in dir_metadata */ | |
a9945bba | 3499 | record_entry_for_tree(dir_metadata, path, &ci->merged); |
6a02dd90 EN |
3500 | } |
3501 | ||
2bff554b EN |
3502 | static void prefetch_for_content_merges(struct merge_options *opt, |
3503 | struct string_list *plist) | |
3504 | { | |
3505 | struct string_list_item *e; | |
3506 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
3507 | ||
3508 | if (opt->repo != the_repository || !has_promisor_remote()) | |
3509 | return; | |
3510 | ||
3511 | for (e = &plist->items[plist->nr-1]; e >= plist->items; --e) { | |
3512 | /* char *path = e->string; */ | |
3513 | struct conflict_info *ci = e->util; | |
3514 | int i; | |
3515 | ||
3516 | /* Ignore clean entries */ | |
3517 | if (ci->merged.clean) | |
3518 | continue; | |
3519 | ||
3520 | /* Ignore entries that don't need a content merge */ | |
3521 | if (ci->match_mask || ci->filemask < 6 || | |
3522 | !S_ISREG(ci->stages[1].mode) || | |
3523 | !S_ISREG(ci->stages[2].mode) || | |
3524 | oideq(&ci->stages[1].oid, &ci->stages[2].oid)) | |
3525 | continue; | |
3526 | ||
3527 | /* Also don't need content merge if base matches either side */ | |
3528 | if (ci->filemask == 7 && | |
3529 | S_ISREG(ci->stages[0].mode) && | |
3530 | (oideq(&ci->stages[0].oid, &ci->stages[1].oid) || | |
3531 | oideq(&ci->stages[0].oid, &ci->stages[2].oid))) | |
3532 | continue; | |
3533 | ||
3534 | for (i = 0; i < 3; i++) { | |
3535 | unsigned side_mask = (1 << i); | |
3536 | struct version_info *vi = &ci->stages[i]; | |
3537 | ||
3538 | if ((ci->filemask & side_mask) && | |
3539 | S_ISREG(vi->mode) && | |
3540 | oid_object_info_extended(opt->repo, &vi->oid, NULL, | |
3541 | OBJECT_INFO_FOR_PREFETCH)) | |
3542 | oid_array_append(&to_fetch, &vi->oid); | |
3543 | } | |
3544 | } | |
3545 | ||
3546 | promisor_remote_get_direct(opt->repo, to_fetch.oid, to_fetch.nr); | |
3547 | oid_array_clear(&to_fetch); | |
3548 | } | |
3549 | ||
231e2dd4 EN |
3550 | static void process_entries(struct merge_options *opt, |
3551 | struct object_id *result_oid) | |
3552 | { | |
6a02dd90 EN |
3553 | struct hashmap_iter iter; |
3554 | struct strmap_entry *e; | |
8adffaa8 EN |
3555 | struct string_list plist = STRING_LIST_INIT_NODUP; |
3556 | struct string_list_item *entry; | |
bb470f4e EN |
3557 | struct directory_versions dir_metadata = { STRING_LIST_INIT_NODUP, |
3558 | STRING_LIST_INIT_NODUP, | |
3559 | NULL, 0 }; | |
6a02dd90 | 3560 | |
557ac035 | 3561 | trace2_region_enter("merge", "process_entries setup", opt->repo); |
6a02dd90 EN |
3562 | if (strmap_empty(&opt->priv->paths)) { |
3563 | oidcpy(result_oid, opt->repo->hash_algo->empty_tree); | |
3564 | return; | |
3565 | } | |
3566 | ||
8adffaa8 | 3567 | /* Hack to pre-allocate plist to the desired size */ |
557ac035 | 3568 | trace2_region_enter("merge", "plist grow", opt->repo); |
8adffaa8 | 3569 | ALLOC_GROW(plist.items, strmap_get_size(&opt->priv->paths), plist.alloc); |
557ac035 | 3570 | trace2_region_leave("merge", "plist grow", opt->repo); |
8adffaa8 EN |
3571 | |
3572 | /* Put every entry from paths into plist, then sort */ | |
557ac035 | 3573 | trace2_region_enter("merge", "plist copy", opt->repo); |
6a02dd90 | 3574 | strmap_for_each_entry(&opt->priv->paths, &iter, e) { |
8adffaa8 EN |
3575 | string_list_append(&plist, e->key)->util = e->value; |
3576 | } | |
557ac035 EN |
3577 | trace2_region_leave("merge", "plist copy", opt->repo); |
3578 | ||
3579 | trace2_region_enter("merge", "plist special sort", opt->repo); | |
5a3743da | 3580 | plist.cmp = sort_dirs_next_to_their_children; |
8adffaa8 | 3581 | string_list_sort(&plist); |
557ac035 EN |
3582 | trace2_region_leave("merge", "plist special sort", opt->repo); |
3583 | ||
3584 | trace2_region_leave("merge", "process_entries setup", opt->repo); | |
8adffaa8 EN |
3585 | |
3586 | /* | |
3587 | * Iterate over the items in reverse order, so we can handle paths | |
3588 | * below a directory before needing to handle the directory itself. | |
bb470f4e EN |
3589 | * |
3590 | * This allows us to write subtrees before we need to write trees, | |
3591 | * and it also enables sane handling of directory/file conflicts | |
3592 | * (because it allows us to know whether the directory is still in | |
3593 | * the way when it is time to process the file at the same path). | |
8adffaa8 | 3594 | */ |
557ac035 | 3595 | trace2_region_enter("merge", "processing", opt->repo); |
2bff554b | 3596 | prefetch_for_content_merges(opt, &plist); |
8adffaa8 EN |
3597 | for (entry = &plist.items[plist.nr-1]; entry >= plist.items; --entry) { |
3598 | char *path = entry->string; | |
6a02dd90 EN |
3599 | /* |
3600 | * NOTE: mi may actually be a pointer to a conflict_info, but | |
3601 | * we have to check mi->clean first to see if it's safe to | |
3602 | * reassign to such a pointer type. | |
3603 | */ | |
8adffaa8 | 3604 | struct merged_info *mi = entry->util; |
6a02dd90 | 3605 | |
bb470f4e EN |
3606 | write_completed_directory(opt, mi->directory_name, |
3607 | &dir_metadata); | |
a9945bba EN |
3608 | if (mi->clean) |
3609 | record_entry_for_tree(&dir_metadata, path, mi); | |
3610 | else { | |
8adffaa8 | 3611 | struct conflict_info *ci = (struct conflict_info *)mi; |
a9945bba | 3612 | process_entry(opt, path, ci, &dir_metadata); |
8adffaa8 | 3613 | } |
6a02dd90 | 3614 | } |
557ac035 | 3615 | trace2_region_leave("merge", "processing", opt->repo); |
6a02dd90 | 3616 | |
557ac035 | 3617 | trace2_region_enter("merge", "process_entries cleanup", opt->repo); |
bb470f4e EN |
3618 | if (dir_metadata.offsets.nr != 1 || |
3619 | (uintptr_t)dir_metadata.offsets.items[0].util != 0) { | |
3620 | printf("dir_metadata.offsets.nr = %d (should be 1)\n", | |
3621 | dir_metadata.offsets.nr); | |
3622 | printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n", | |
3623 | (unsigned)(uintptr_t)dir_metadata.offsets.items[0].util); | |
3624 | fflush(stdout); | |
3625 | BUG("dir_metadata accounting completely off; shouldn't happen"); | |
3626 | } | |
ee4012dc EN |
3627 | write_tree(result_oid, &dir_metadata.versions, 0, |
3628 | opt->repo->hash_algo->rawsz); | |
8adffaa8 | 3629 | string_list_clear(&plist, 0); |
a9945bba | 3630 | string_list_clear(&dir_metadata.versions, 0); |
bb470f4e | 3631 | string_list_clear(&dir_metadata.offsets, 0); |
557ac035 | 3632 | trace2_region_leave("merge", "process_entries cleanup", opt->repo); |
231e2dd4 EN |
3633 | } |
3634 | ||
04af1879 EN |
3635 | /*** Function Grouping: functions related to merge_switch_to_result() ***/ |
3636 | ||
9fefce68 EN |
3637 | static int checkout(struct merge_options *opt, |
3638 | struct tree *prev, | |
3639 | struct tree *next) | |
3640 | { | |
6681ce5c EN |
3641 | /* Switch the index/working copy from old to new */ |
3642 | int ret; | |
3643 | struct tree_desc trees[2]; | |
3644 | struct unpack_trees_options unpack_opts; | |
3645 | ||
3646 | memset(&unpack_opts, 0, sizeof(unpack_opts)); | |
3647 | unpack_opts.head_idx = -1; | |
3648 | unpack_opts.src_index = opt->repo->index; | |
3649 | unpack_opts.dst_index = opt->repo->index; | |
3650 | ||
3651 | setup_unpack_trees_porcelain(&unpack_opts, "merge"); | |
3652 | ||
3653 | /* | |
3654 | * NOTE: if this were just "git checkout" code, we would probably | |
3655 | * read or refresh the cache and check for a conflicted index, but | |
3656 | * builtin/merge.c or sequencer.c really needs to read the index | |
3657 | * and check for conflicted entries before starting merging for a | |
3658 | * good user experience (no sense waiting for merges/rebases before | |
3659 | * erroring out), so there's no reason to duplicate that work here. | |
3660 | */ | |
3661 | ||
3662 | /* 2-way merge to the new branch */ | |
3663 | unpack_opts.update = 1; | |
3664 | unpack_opts.merge = 1; | |
3665 | unpack_opts.quiet = 0; /* FIXME: sequencer might want quiet? */ | |
3666 | unpack_opts.verbose_update = (opt->verbosity > 2); | |
3667 | unpack_opts.fn = twoway_merge; | |
3668 | if (1/* FIXME: opts->overwrite_ignore*/) { | |
ca56dadb | 3669 | CALLOC_ARRAY(unpack_opts.dir, 1); |
6681ce5c EN |
3670 | unpack_opts.dir->flags |= DIR_SHOW_IGNORED; |
3671 | setup_standard_excludes(unpack_opts.dir); | |
3672 | } | |
3673 | parse_tree(prev); | |
3674 | init_tree_desc(&trees[0], prev->buffer, prev->size); | |
3675 | parse_tree(next); | |
3676 | init_tree_desc(&trees[1], next->buffer, next->size); | |
3677 | ||
3678 | ret = unpack_trees(2, trees, &unpack_opts); | |
3679 | clear_unpack_trees_porcelain(&unpack_opts); | |
3680 | dir_clear(unpack_opts.dir); | |
3681 | FREE_AND_NULL(unpack_opts.dir); | |
3682 | return ret; | |
9fefce68 EN |
3683 | } |
3684 | ||
66b209b8 | 3685 | static int record_conflicted_index_entries(struct merge_options *opt) |
9fefce68 | 3686 | { |
ef2b3693 EN |
3687 | struct hashmap_iter iter; |
3688 | struct strmap_entry *e; | |
66b209b8 EN |
3689 | struct index_state *index = opt->repo->index; |
3690 | struct checkout state = CHECKOUT_INIT; | |
ef2b3693 EN |
3691 | int errs = 0; |
3692 | int original_cache_nr; | |
3693 | ||
66b209b8 | 3694 | if (strmap_empty(&opt->priv->conflicted)) |
9fefce68 EN |
3695 | return 0; |
3696 | ||
66b209b8 EN |
3697 | /* If any entries have skip_worktree set, we'll have to check 'em out */ |
3698 | state.force = 1; | |
3699 | state.quiet = 1; | |
3700 | state.refresh_cache = 1; | |
3701 | state.istate = index; | |
ef2b3693 EN |
3702 | original_cache_nr = index->cache_nr; |
3703 | ||
3704 | /* Put every entry from paths into plist, then sort */ | |
66b209b8 | 3705 | strmap_for_each_entry(&opt->priv->conflicted, &iter, e) { |
ef2b3693 EN |
3706 | const char *path = e->key; |
3707 | struct conflict_info *ci = e->value; | |
3708 | int pos; | |
3709 | struct cache_entry *ce; | |
3710 | int i; | |
3711 | ||
3712 | VERIFY_CI(ci); | |
3713 | ||
3714 | /* | |
3715 | * The index will already have a stage=0 entry for this path, | |
3716 | * because we created an as-merged-as-possible version of the | |
3717 | * file and checkout() moved the working copy and index over | |
3718 | * to that version. | |
3719 | * | |
3720 | * However, previous iterations through this loop will have | |
3721 | * added unstaged entries to the end of the cache which | |
3722 | * ignore the standard alphabetical ordering of cache | |
3723 | * entries and break invariants needed for index_name_pos() | |
3724 | * to work. However, we know the entry we want is before | |
3725 | * those appended cache entries, so do a temporary swap on | |
3726 | * cache_nr to only look through entries of interest. | |
3727 | */ | |
3728 | SWAP(index->cache_nr, original_cache_nr); | |
3729 | pos = index_name_pos(index, path, strlen(path)); | |
3730 | SWAP(index->cache_nr, original_cache_nr); | |
3731 | if (pos < 0) { | |
3732 | if (ci->filemask != 1) | |
3733 | BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path); | |
3734 | cache_tree_invalidate_path(index, path); | |
3735 | } else { | |
3736 | ce = index->cache[pos]; | |
3737 | ||
3738 | /* | |
3739 | * Clean paths with CE_SKIP_WORKTREE set will not be | |
3740 | * written to the working tree by the unpack_trees() | |
3741 | * call in checkout(). Our conflicted entries would | |
3742 | * have appeared clean to that code since we ignored | |
3743 | * the higher order stages. Thus, we need override | |
3744 | * the CE_SKIP_WORKTREE bit and manually write those | |
3745 | * files to the working disk here. | |
ef2b3693 | 3746 | */ |
66b209b8 EN |
3747 | if (ce_skip_worktree(ce)) { |
3748 | struct stat st; | |
3749 | ||
3750 | if (!lstat(path, &st)) { | |
3751 | char *new_name = unique_path(&opt->priv->paths, | |
3752 | path, | |
3753 | "cruft"); | |
3754 | ||
3755 | path_msg(opt, path, 1, | |
3756 | _("Note: %s not up to date and in way of checking out conflicted version; old copy renamed to %s"), | |
3757 | path, new_name); | |
3758 | errs |= rename(path, new_name); | |
3759 | free(new_name); | |
3760 | } | |
3761 | errs |= checkout_entry(ce, &state, NULL, NULL); | |
3762 | } | |
ef2b3693 EN |
3763 | |
3764 | /* | |
3765 | * Mark this cache entry for removal and instead add | |
3766 | * new stage>0 entries corresponding to the | |
3767 | * conflicts. If there are many conflicted entries, we | |
3768 | * want to avoid memmove'ing O(NM) entries by | |
3769 | * inserting the new entries one at a time. So, | |
3770 | * instead, we just add the new cache entries to the | |
3771 | * end (ignoring normal index requirements on sort | |
3772 | * order) and sort the index once we're all done. | |
3773 | */ | |
3774 | ce->ce_flags |= CE_REMOVE; | |
3775 | } | |
3776 | ||
3777 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
3778 | struct version_info *vi; | |
3779 | if (!(ci->filemask & (1ul << i))) | |
3780 | continue; | |
3781 | vi = &ci->stages[i]; | |
3782 | ce = make_cache_entry(index, vi->mode, &vi->oid, | |
3783 | path, i+1, 0); | |
3784 | add_index_entry(index, ce, ADD_CACHE_JUST_APPEND); | |
3785 | } | |
3786 | } | |
3787 | ||
3788 | /* | |
3789 | * Remove the unused cache entries (and invalidate the relevant | |
3790 | * cache-trees), then sort the index entries to get the conflicted | |
3791 | * entries we added to the end into their right locations. | |
3792 | */ | |
3793 | remove_marked_cache_entries(index, 1); | |
72b30910 EN |
3794 | /* |
3795 | * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily | |
3796 | * on filename and secondarily on stage, and (name, stage #) are a | |
3797 | * unique tuple. | |
3798 | */ | |
ef2b3693 EN |
3799 | QSORT(index->cache, index->cache_nr, cmp_cache_name_compare); |
3800 | ||
3801 | return errs; | |
9fefce68 EN |
3802 | } |
3803 | ||
17e5574b EN |
3804 | void merge_switch_to_result(struct merge_options *opt, |
3805 | struct tree *head, | |
3806 | struct merge_result *result, | |
3807 | int update_worktree_and_index, | |
3808 | int display_update_msgs) | |
3809 | { | |
9fefce68 EN |
3810 | assert(opt->priv == NULL); |
3811 | if (result->clean >= 0 && update_worktree_and_index) { | |
5291828d EN |
3812 | const char *filename; |
3813 | FILE *fp; | |
9fefce68 | 3814 | |
557ac035 | 3815 | trace2_region_enter("merge", "checkout", opt->repo); |
9fefce68 EN |
3816 | if (checkout(opt, head, result->tree)) { |
3817 | /* failure to function */ | |
3818 | result->clean = -1; | |
3819 | return; | |
3820 | } | |
557ac035 | 3821 | trace2_region_leave("merge", "checkout", opt->repo); |
9fefce68 | 3822 | |
557ac035 | 3823 | trace2_region_enter("merge", "record_conflicted", opt->repo); |
66b209b8 EN |
3824 | opt->priv = result->priv; |
3825 | if (record_conflicted_index_entries(opt)) { | |
9fefce68 | 3826 | /* failure to function */ |
66b209b8 | 3827 | opt->priv = NULL; |
9fefce68 EN |
3828 | result->clean = -1; |
3829 | return; | |
3830 | } | |
66b209b8 | 3831 | opt->priv = NULL; |
557ac035 | 3832 | trace2_region_leave("merge", "record_conflicted", opt->repo); |
5291828d EN |
3833 | |
3834 | trace2_region_enter("merge", "write_auto_merge", opt->repo); | |
3835 | filename = git_path_auto_merge(opt->repo); | |
3836 | fp = xfopen(filename, "w"); | |
3837 | fprintf(fp, "%s\n", oid_to_hex(&result->tree->object.oid)); | |
3838 | fclose(fp); | |
3839 | trace2_region_leave("merge", "write_auto_merge", opt->repo); | |
9fefce68 EN |
3840 | } |
3841 | ||
3842 | if (display_update_msgs) { | |
c5a6f655 EN |
3843 | struct merge_options_internal *opti = result->priv; |
3844 | struct hashmap_iter iter; | |
3845 | struct strmap_entry *e; | |
3846 | struct string_list olist = STRING_LIST_INIT_NODUP; | |
3847 | int i; | |
3848 | ||
557ac035 EN |
3849 | trace2_region_enter("merge", "display messages", opt->repo); |
3850 | ||
c5a6f655 EN |
3851 | /* Hack to pre-allocate olist to the desired size */ |
3852 | ALLOC_GROW(olist.items, strmap_get_size(&opti->output), | |
3853 | olist.alloc); | |
3854 | ||
3855 | /* Put every entry from output into olist, then sort */ | |
3856 | strmap_for_each_entry(&opti->output, &iter, e) { | |
3857 | string_list_append(&olist, e->key)->util = e->value; | |
3858 | } | |
3859 | string_list_sort(&olist); | |
3860 | ||
3861 | /* Iterate over the items, printing them */ | |
3862 | for (i = 0; i < olist.nr; ++i) { | |
3863 | struct strbuf *sb = olist.items[i].util; | |
3864 | ||
3865 | printf("%s", sb->buf); | |
3866 | } | |
3867 | string_list_clear(&olist, 0); | |
f39d05ca EN |
3868 | |
3869 | /* Also include needed rename limit adjustment now */ | |
3870 | diff_warn_rename_limit("merge.renamelimit", | |
3871 | opti->renames.needed_limit, 0); | |
557ac035 EN |
3872 | |
3873 | trace2_region_leave("merge", "display messages", opt->repo); | |
9fefce68 EN |
3874 | } |
3875 | ||
17e5574b EN |
3876 | merge_finalize(opt, result); |
3877 | } | |
3878 | ||
3879 | void merge_finalize(struct merge_options *opt, | |
3880 | struct merge_result *result) | |
3881 | { | |
89422d29 EN |
3882 | struct merge_options_internal *opti = result->priv; |
3883 | ||
ea305a68 EN |
3884 | if (opt->renormalize) |
3885 | git_attr_set_direction(GIT_ATTR_CHECKIN); | |
89422d29 EN |
3886 | assert(opt->priv == NULL); |
3887 | ||
43e9c4ee | 3888 | clear_or_reinit_internal_opts(opti, 0); |
89422d29 | 3889 | FREE_AND_NULL(opti); |
17e5574b EN |
3890 | } |
3891 | ||
04af1879 EN |
3892 | /*** Function Grouping: helper functions for merge_incore_*() ***/ |
3893 | ||
3639dfb3 EN |
3894 | static struct tree *shift_tree_object(struct repository *repo, |
3895 | struct tree *one, struct tree *two, | |
3896 | const char *subtree_shift) | |
3897 | { | |
3898 | struct object_id shifted; | |
3899 | ||
3900 | if (!*subtree_shift) { | |
3901 | shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0); | |
3902 | } else { | |
3903 | shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted, | |
3904 | subtree_shift); | |
3905 | } | |
3906 | if (oideq(&two->object.oid, &shifted)) | |
3907 | return two; | |
3908 | return lookup_tree(repo, &shifted); | |
3909 | } | |
3910 | ||
4296d8f1 EN |
3911 | static inline void set_commit_tree(struct commit *c, struct tree *t) |
3912 | { | |
3913 | c->maybe_tree = t; | |
3914 | } | |
3915 | ||
4296d8f1 EN |
3916 | static struct commit *make_virtual_commit(struct repository *repo, |
3917 | struct tree *tree, | |
3918 | const char *comment) | |
3919 | { | |
3920 | struct commit *commit = alloc_commit_node(repo); | |
3921 | ||
3922 | set_merge_remote_desc(commit, comment, (struct object *)commit); | |
3923 | set_commit_tree(commit, tree); | |
3924 | commit->object.parsed = 1; | |
3925 | return commit; | |
3926 | } | |
3927 | ||
231e2dd4 EN |
3928 | static void merge_start(struct merge_options *opt, struct merge_result *result) |
3929 | { | |
f5d9fbc2 EN |
3930 | struct rename_info *renames; |
3931 | int i; | |
3932 | ||
e4171b1b | 3933 | /* Sanity checks on opt */ |
557ac035 | 3934 | trace2_region_enter("merge", "sanity checks", opt->repo); |
e4171b1b EN |
3935 | assert(opt->repo); |
3936 | ||
3937 | assert(opt->branch1 && opt->branch2); | |
3938 | ||
3939 | assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE && | |
3940 | opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE); | |
3941 | assert(opt->rename_limit >= -1); | |
3942 | assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE); | |
3943 | assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1); | |
3944 | ||
3945 | assert(opt->xdl_opts >= 0); | |
3946 | assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL && | |
3947 | opt->recursive_variant <= MERGE_VARIANT_THEIRS); | |
3948 | ||
3949 | /* | |
3950 | * detect_renames, verbosity, buffer_output, and obuf are ignored | |
3951 | * fields that were used by "recursive" rather than "ort" -- but | |
3952 | * sanity check them anyway. | |
3953 | */ | |
3954 | assert(opt->detect_renames >= -1 && | |
3955 | opt->detect_renames <= DIFF_DETECT_COPY); | |
3956 | assert(opt->verbosity >= 0 && opt->verbosity <= 5); | |
3957 | assert(opt->buffer_output <= 2); | |
3958 | assert(opt->obuf.len == 0); | |
3959 | ||
3960 | assert(opt->priv == NULL); | |
19ceb486 EN |
3961 | if (result->_properly_initialized != 0 && |
3962 | result->_properly_initialized != RESULT_INITIALIZED) | |
3963 | BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run"); | |
3964 | assert(!!result->priv == !!result->_properly_initialized); | |
cf8937ac EN |
3965 | if (result->priv) { |
3966 | opt->priv = result->priv; | |
3967 | result->priv = NULL; | |
3968 | /* | |
3969 | * opt->priv non-NULL means we had results from a previous | |
3970 | * run; do a few sanity checks that user didn't mess with | |
3971 | * it in an obvious fashion. | |
3972 | */ | |
3973 | assert(opt->priv->call_depth == 0); | |
3974 | assert(!opt->priv->toplevel_dir || | |
3975 | 0 == strlen(opt->priv->toplevel_dir)); | |
3976 | } | |
557ac035 | 3977 | trace2_region_leave("merge", "sanity checks", opt->repo); |
e4171b1b | 3978 | |
c8017176 EN |
3979 | /* Default to histogram diff. Actually, just hardcode it...for now. */ |
3980 | opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF); | |
3981 | ||
ea305a68 EN |
3982 | /* Handle attr direction stuff for renormalization */ |
3983 | if (opt->renormalize) | |
3984 | git_attr_set_direction(GIT_ATTR_CHECKOUT); | |
3985 | ||
e4171b1b | 3986 | /* Initialization of opt->priv, our internal merge data */ |
557ac035 | 3987 | trace2_region_enter("merge", "allocate/init", opt->repo); |
cf8937ac EN |
3988 | if (opt->priv) { |
3989 | clear_or_reinit_internal_opts(opt->priv, 1); | |
3990 | trace2_region_leave("merge", "allocate/init", opt->repo); | |
3991 | return; | |
3992 | } | |
e4171b1b EN |
3993 | opt->priv = xcalloc(1, sizeof(*opt->priv)); |
3994 | ||
f5d9fbc2 EN |
3995 | /* Initialization of various renames fields */ |
3996 | renames = &opt->priv->renames; | |
3997 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) { | |
a49b55d5 | 3998 | strintmap_init_with_options(&renames->dirs_removed[i], |
fb52938e | 3999 | NOT_RELEVANT, NULL, 0); |
f5d9fbc2 EN |
4000 | strmap_init_with_options(&renames->dir_rename_count[i], |
4001 | NULL, 1); | |
4002 | strmap_init_with_options(&renames->dir_renames[i], | |
4003 | NULL, 0); | |
2734f2e3 EN |
4004 | /* |
4005 | * relevant_sources uses -1 for the default, because we need | |
4006 | * to be able to distinguish not-in-strintmap from valid | |
4007 | * relevant_source values from enum file_rename_relevance. | |
4008 | * In particular, possibly_cache_new_pair() expects a negative | |
4009 | * value for not-found entries. | |
4010 | */ | |
a49b55d5 | 4011 | strintmap_init_with_options(&renames->relevant_sources[i], |
2734f2e3 EN |
4012 | -1 /* explicitly invalid */, |
4013 | NULL, 0); | |
d29bd6d7 EN |
4014 | strmap_init_with_options(&renames->cached_pairs[i], |
4015 | NULL, 1); | |
4016 | strset_init_with_options(&renames->cached_irrelevant[i], | |
4017 | NULL, 1); | |
4018 | strset_init_with_options(&renames->cached_target_names[i], | |
4019 | NULL, 0); | |
f5d9fbc2 EN |
4020 | } |
4021 | ||
e4171b1b EN |
4022 | /* |
4023 | * Although we initialize opt->priv->paths with strdup_strings=0, | |
4024 | * that's just to avoid making yet another copy of an allocated | |
4025 | * string. Putting the entry into paths means we are taking | |
43c1dccb | 4026 | * ownership, so we will later free it. paths_to_free is similar. |
e4171b1b EN |
4027 | * |
4028 | * In contrast, conflicted just has a subset of keys from paths, so | |
4029 | * we don't want to free those (it'd be a duplicate free). | |
4030 | */ | |
4031 | strmap_init_with_options(&opt->priv->paths, NULL, 0); | |
4032 | strmap_init_with_options(&opt->priv->conflicted, NULL, 0); | |
43c1dccb | 4033 | string_list_init(&opt->priv->paths_to_free, 0); |
c5a6f655 EN |
4034 | |
4035 | /* | |
4036 | * keys & strbufs in output will sometimes need to outlive "paths", | |
4037 | * so it will have a copy of relevant keys. It's probably a small | |
4038 | * subset of the overall paths that have special output. | |
4039 | */ | |
4040 | strmap_init(&opt->priv->output); | |
557ac035 EN |
4041 | |
4042 | trace2_region_leave("merge", "allocate/init", opt->repo); | |
231e2dd4 EN |
4043 | } |
4044 | ||
64aceb6d EN |
4045 | static void merge_check_renames_reusable(struct merge_options *opt, |
4046 | struct merge_result *result, | |
4047 | struct tree *merge_base, | |
4048 | struct tree *side1, | |
4049 | struct tree *side2) | |
4050 | { | |
4051 | struct rename_info *renames; | |
4052 | struct tree **merge_trees; | |
4053 | struct merge_options_internal *opti = result->priv; | |
4054 | ||
4055 | if (!opti) | |
4056 | return; | |
4057 | ||
4058 | renames = &opti->renames; | |
4059 | merge_trees = renames->merge_trees; | |
cbdca289 EN |
4060 | |
4061 | /* | |
4062 | * Handle case where previous merge operation did not want cache to | |
4063 | * take effect, e.g. because rename/rename(1to1) makes it invalid. | |
4064 | */ | |
4065 | if (!merge_trees[0]) { | |
4066 | assert(!merge_trees[0] && !merge_trees[1] && !merge_trees[2]); | |
4067 | renames->cached_pairs_valid_side = 0; /* neither side valid */ | |
4068 | return; | |
4069 | } | |
4070 | ||
4071 | /* | |
4072 | * Handle other cases; note that merge_trees[0..2] will only | |
4073 | * be NULL if opti is, or if all three were manually set to | |
4074 | * NULL by e.g. rename/rename(1to1) handling. | |
4075 | */ | |
64aceb6d EN |
4076 | assert(merge_trees[0] && merge_trees[1] && merge_trees[2]); |
4077 | ||
4078 | /* Check if we meet a condition for re-using cached_pairs */ | |
4079 | if (oideq(&merge_base->object.oid, &merge_trees[2]->object.oid) && | |
4080 | oideq(&side1->object.oid, &result->tree->object.oid)) | |
4081 | renames->cached_pairs_valid_side = MERGE_SIDE1; | |
4082 | else if (oideq(&merge_base->object.oid, &merge_trees[1]->object.oid) && | |
4083 | oideq(&side2->object.oid, &result->tree->object.oid)) | |
4084 | renames->cached_pairs_valid_side = MERGE_SIDE2; | |
4085 | else | |
4086 | renames->cached_pairs_valid_side = 0; /* neither side valid */ | |
4087 | } | |
4088 | ||
04af1879 EN |
4089 | /*** Function Grouping: merge_incore_*() and their internal variants ***/ |
4090 | ||
231e2dd4 EN |
4091 | /* |
4092 | * Originally from merge_trees_internal(); heavily adapted, though. | |
4093 | */ | |
4094 | static void merge_ort_nonrecursive_internal(struct merge_options *opt, | |
4095 | struct tree *merge_base, | |
4096 | struct tree *side1, | |
4097 | struct tree *side2, | |
4098 | struct merge_result *result) | |
4099 | { | |
4100 | struct object_id working_tree_oid; | |
4101 | ||
3639dfb3 EN |
4102 | if (opt->subtree_shift) { |
4103 | side2 = shift_tree_object(opt->repo, side1, side2, | |
4104 | opt->subtree_shift); | |
4105 | merge_base = shift_tree_object(opt->repo, side1, merge_base, | |
4106 | opt->subtree_shift); | |
4107 | } | |
4108 | ||
557ac035 | 4109 | trace2_region_enter("merge", "collect_merge_info", opt->repo); |
0c0d705b EN |
4110 | if (collect_merge_info(opt, merge_base, side1, side2) != 0) { |
4111 | /* | |
4112 | * TRANSLATORS: The %s arguments are: 1) tree hash of a merge | |
4113 | * base, and 2-3) the trees for the two trees we're merging. | |
4114 | */ | |
4115 | err(opt, _("collecting merge info failed for trees %s, %s, %s"), | |
4116 | oid_to_hex(&merge_base->object.oid), | |
4117 | oid_to_hex(&side1->object.oid), | |
4118 | oid_to_hex(&side2->object.oid)); | |
4119 | result->clean = -1; | |
4120 | return; | |
4121 | } | |
557ac035 | 4122 | trace2_region_leave("merge", "collect_merge_info", opt->repo); |
0c0d705b | 4123 | |
557ac035 | 4124 | trace2_region_enter("merge", "renames", opt->repo); |
231e2dd4 EN |
4125 | result->clean = detect_and_process_renames(opt, merge_base, |
4126 | side1, side2); | |
557ac035 EN |
4127 | trace2_region_leave("merge", "renames", opt->repo); |
4128 | ||
4129 | trace2_region_enter("merge", "process_entries", opt->repo); | |
231e2dd4 | 4130 | process_entries(opt, &working_tree_oid); |
557ac035 | 4131 | trace2_region_leave("merge", "process_entries", opt->repo); |
231e2dd4 EN |
4132 | |
4133 | /* Set return values */ | |
4134 | result->tree = parse_tree_indirect(&working_tree_oid); | |
4135 | /* existence of conflicted entries implies unclean */ | |
4136 | result->clean &= strmap_empty(&opt->priv->conflicted); | |
4137 | if (!opt->priv->call_depth) { | |
4138 | result->priv = opt->priv; | |
19ceb486 | 4139 | result->_properly_initialized = RESULT_INITIALIZED; |
231e2dd4 EN |
4140 | opt->priv = NULL; |
4141 | } | |
4142 | } | |
4143 | ||
8119214f EN |
4144 | /* |
4145 | * Originally from merge_recursive_internal(); somewhat adapted, though. | |
4146 | */ | |
4147 | static void merge_ort_internal(struct merge_options *opt, | |
4148 | struct commit_list *merge_bases, | |
4149 | struct commit *h1, | |
4150 | struct commit *h2, | |
4151 | struct merge_result *result) | |
4152 | { | |
4153 | struct commit_list *iter; | |
4154 | struct commit *merged_merge_bases; | |
4155 | const char *ancestor_name; | |
4156 | struct strbuf merge_base_abbrev = STRBUF_INIT; | |
4157 | ||
4158 | if (!merge_bases) { | |
4159 | merge_bases = get_merge_bases(h1, h2); | |
4160 | /* See merge-ort.h:merge_incore_recursive() declaration NOTE */ | |
4161 | merge_bases = reverse_commit_list(merge_bases); | |
4162 | } | |
4163 | ||
4164 | merged_merge_bases = pop_commit(&merge_bases); | |
4165 | if (merged_merge_bases == NULL) { | |
4166 | /* if there is no common ancestor, use an empty tree */ | |
4167 | struct tree *tree; | |
4168 | ||
4169 | tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree); | |
4170 | merged_merge_bases = make_virtual_commit(opt->repo, tree, | |
4171 | "ancestor"); | |
4172 | ancestor_name = "empty tree"; | |
4173 | } else if (merge_bases) { | |
4174 | ancestor_name = "merged common ancestors"; | |
4175 | } else { | |
4176 | strbuf_add_unique_abbrev(&merge_base_abbrev, | |
4177 | &merged_merge_bases->object.oid, | |
4178 | DEFAULT_ABBREV); | |
4179 | ancestor_name = merge_base_abbrev.buf; | |
4180 | } | |
4181 | ||
4182 | for (iter = merge_bases; iter; iter = iter->next) { | |
4183 | const char *saved_b1, *saved_b2; | |
4184 | struct commit *prev = merged_merge_bases; | |
4185 | ||
4186 | opt->priv->call_depth++; | |
4187 | /* | |
4188 | * When the merge fails, the result contains files | |
4189 | * with conflict markers. The cleanness flag is | |
4190 | * ignored (unless indicating an error), it was never | |
4191 | * actually used, as result of merge_trees has always | |
4192 | * overwritten it: the committed "conflicts" were | |
4193 | * already resolved. | |
4194 | */ | |
4195 | saved_b1 = opt->branch1; | |
4196 | saved_b2 = opt->branch2; | |
4197 | opt->branch1 = "Temporary merge branch 1"; | |
4198 | opt->branch2 = "Temporary merge branch 2"; | |
4199 | merge_ort_internal(opt, NULL, prev, iter->item, result); | |
4200 | if (result->clean < 0) | |
4201 | return; | |
4202 | opt->branch1 = saved_b1; | |
4203 | opt->branch2 = saved_b2; | |
4204 | opt->priv->call_depth--; | |
4205 | ||
4206 | merged_merge_bases = make_virtual_commit(opt->repo, | |
4207 | result->tree, | |
4208 | "merged tree"); | |
4209 | commit_list_insert(prev, &merged_merge_bases->parents); | |
4210 | commit_list_insert(iter->item, | |
4211 | &merged_merge_bases->parents->next); | |
4212 | ||
4213 | clear_or_reinit_internal_opts(opt->priv, 1); | |
4214 | } | |
4215 | ||
4216 | opt->ancestor = ancestor_name; | |
4217 | merge_ort_nonrecursive_internal(opt, | |
4218 | repo_get_commit_tree(opt->repo, | |
4219 | merged_merge_bases), | |
4220 | repo_get_commit_tree(opt->repo, h1), | |
4221 | repo_get_commit_tree(opt->repo, h2), | |
4222 | result); | |
4223 | strbuf_release(&merge_base_abbrev); | |
4224 | opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */ | |
4225 | } | |
4226 | ||
17e5574b EN |
4227 | void merge_incore_nonrecursive(struct merge_options *opt, |
4228 | struct tree *merge_base, | |
4229 | struct tree *side1, | |
4230 | struct tree *side2, | |
4231 | struct merge_result *result) | |
4232 | { | |
557ac035 EN |
4233 | trace2_region_enter("merge", "incore_nonrecursive", opt->repo); |
4234 | ||
4235 | trace2_region_enter("merge", "merge_start", opt->repo); | |
231e2dd4 | 4236 | assert(opt->ancestor != NULL); |
64aceb6d | 4237 | merge_check_renames_reusable(opt, result, merge_base, side1, side2); |
231e2dd4 | 4238 | merge_start(opt, result); |
64aceb6d EN |
4239 | /* |
4240 | * Record the trees used in this merge, so if there's a next merge in | |
4241 | * a cherry-pick or rebase sequence it might be able to take advantage | |
4242 | * of the cached_pairs in that next merge. | |
4243 | */ | |
4244 | opt->priv->renames.merge_trees[0] = merge_base; | |
4245 | opt->priv->renames.merge_trees[1] = side1; | |
4246 | opt->priv->renames.merge_trees[2] = side2; | |
557ac035 EN |
4247 | trace2_region_leave("merge", "merge_start", opt->repo); |
4248 | ||
231e2dd4 | 4249 | merge_ort_nonrecursive_internal(opt, merge_base, side1, side2, result); |
557ac035 | 4250 | trace2_region_leave("merge", "incore_nonrecursive", opt->repo); |
17e5574b EN |
4251 | } |
4252 | ||
4253 | void merge_incore_recursive(struct merge_options *opt, | |
4254 | struct commit_list *merge_bases, | |
4255 | struct commit *side1, | |
4256 | struct commit *side2, | |
4257 | struct merge_result *result) | |
4258 | { | |
557ac035 EN |
4259 | trace2_region_enter("merge", "incore_recursive", opt->repo); |
4260 | ||
8119214f EN |
4261 | /* We set the ancestor label based on the merge_bases */ |
4262 | assert(opt->ancestor == NULL); | |
4263 | ||
557ac035 | 4264 | trace2_region_enter("merge", "merge_start", opt->repo); |
8119214f | 4265 | merge_start(opt, result); |
557ac035 EN |
4266 | trace2_region_leave("merge", "merge_start", opt->repo); |
4267 | ||
8119214f | 4268 | merge_ort_internal(opt, merge_bases, side1, side2, result); |
557ac035 | 4269 | trace2_region_leave("merge", "incore_recursive", opt->repo); |
17e5574b | 4270 | } |