]>
Commit | Line | Data |
---|---|---|
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 "git-compat-util.h" | |
18 | #include "merge-ort.h" | |
19 | ||
20 | #include "alloc.h" | |
21 | #include "advice.h" | |
22 | #include "attr.h" | |
23 | #include "cache-tree.h" | |
24 | #include "commit.h" | |
25 | #include "commit-reach.h" | |
26 | #include "diff.h" | |
27 | #include "diffcore.h" | |
28 | #include "dir.h" | |
29 | #include "environment.h" | |
30 | #include "gettext.h" | |
31 | #include "hex.h" | |
32 | #include "entry.h" | |
33 | #include "merge-ll.h" | |
34 | #include "match-trees.h" | |
35 | #include "mem-pool.h" | |
36 | #include "object-name.h" | |
37 | #include "object-store-ll.h" | |
38 | #include "oid-array.h" | |
39 | #include "path.h" | |
40 | #include "promisor-remote.h" | |
41 | #include "read-cache-ll.h" | |
42 | #include "refs.h" | |
43 | #include "revision.h" | |
44 | #include "sparse-index.h" | |
45 | #include "strmap.h" | |
46 | #include "trace2.h" | |
47 | #include "tree.h" | |
48 | #include "unpack-trees.h" | |
49 | #include "xdiff-interface.h" | |
50 | ||
51 | /* | |
52 | * We have many arrays of size 3. Whenever we have such an array, the | |
53 | * indices refer to one of the sides of the three-way merge. This is so | |
54 | * pervasive that the constants 0, 1, and 2 are used in many places in the | |
55 | * code (especially in arithmetic operations to find the other side's index | |
56 | * or to compute a relevant mask), but sometimes these enum names are used | |
57 | * to aid code clarity. | |
58 | * | |
59 | * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side" | |
60 | * referred to there is one of these three sides. | |
61 | */ | |
62 | enum merge_side { | |
63 | MERGE_BASE = 0, | |
64 | MERGE_SIDE1 = 1, | |
65 | MERGE_SIDE2 = 2 | |
66 | }; | |
67 | ||
68 | static unsigned RESULT_INITIALIZED = 0x1abe11ed; /* unlikely accidental value */ | |
69 | ||
70 | struct traversal_callback_data { | |
71 | unsigned long mask; | |
72 | unsigned long dirmask; | |
73 | struct name_entry names[3]; | |
74 | }; | |
75 | ||
76 | struct deferred_traversal_data { | |
77 | /* | |
78 | * possible_trivial_merges: directories to be explored only when needed | |
79 | * | |
80 | * possible_trivial_merges is a map of directory names to | |
81 | * dir_rename_mask. When we detect that a directory is unchanged on | |
82 | * one side, we can sometimes resolve the directory without recursing | |
83 | * into it. Renames are the only things that can prevent such an | |
84 | * optimization. However, for rename sources: | |
85 | * - If no parent directory needed directory rename detection, then | |
86 | * no path under such a directory can be a relevant_source. | |
87 | * and for rename destinations: | |
88 | * - If no cached rename has a target path under the directory AND | |
89 | * - If there are no unpaired relevant_sources elsewhere in the | |
90 | * repository | |
91 | * then we don't need any path under this directory for a rename | |
92 | * destination. The only way to know the last item above is to defer | |
93 | * handling such directories until the end of collect_merge_info(), | |
94 | * in handle_deferred_entries(). | |
95 | * | |
96 | * For each we store dir_rename_mask, since that's the only bit of | |
97 | * information we need, other than the path, to resume the recursive | |
98 | * traversal. | |
99 | */ | |
100 | struct strintmap possible_trivial_merges; | |
101 | ||
102 | /* | |
103 | * trivial_merges_okay: if trivial directory merges are okay | |
104 | * | |
105 | * See possible_trivial_merges above. The "no unpaired | |
106 | * relevant_sources elsewhere in the repository" is a single boolean | |
107 | * per merge side, which we store here. Note that while 0 means no, | |
108 | * 1 only means "maybe" rather than "yes"; we optimistically set it | |
109 | * to 1 initially and only clear when we determine it is unsafe to | |
110 | * do trivial directory merges. | |
111 | */ | |
112 | unsigned trivial_merges_okay; | |
113 | ||
114 | /* | |
115 | * target_dirs: ancestor directories of rename targets | |
116 | * | |
117 | * target_dirs contains all directory names that are an ancestor of | |
118 | * any rename destination. | |
119 | */ | |
120 | struct strset target_dirs; | |
121 | }; | |
122 | ||
123 | struct rename_info { | |
124 | /* | |
125 | * All variables that are arrays of size 3 correspond to data tracked | |
126 | * for the sides in enum merge_side. Index 0 is almost always unused | |
127 | * because we often only need to track information for MERGE_SIDE1 and | |
128 | * MERGE_SIDE2 (MERGE_BASE can't have rename information since renames | |
129 | * are determined relative to what changed since the MERGE_BASE). | |
130 | */ | |
131 | ||
132 | /* | |
133 | * pairs: pairing of filenames from diffcore_rename() | |
134 | */ | |
135 | struct diff_queue_struct pairs[3]; | |
136 | ||
137 | /* | |
138 | * dirs_removed: directories removed on a given side of history. | |
139 | * | |
140 | * The keys of dirs_removed[side] are the directories that were removed | |
141 | * on the given side of history. The value of the strintmap for each | |
142 | * directory is a value from enum dir_rename_relevance. | |
143 | */ | |
144 | struct strintmap dirs_removed[3]; | |
145 | ||
146 | /* | |
147 | * dir_rename_count: tracking where parts of a directory were renamed to | |
148 | * | |
149 | * When files in a directory are renamed, they may not all go to the | |
150 | * same location. Each strmap here tracks: | |
151 | * old_dir => {new_dir => int} | |
152 | * That is, dir_rename_count[side] is a strmap to a strintmap. | |
153 | */ | |
154 | struct strmap dir_rename_count[3]; | |
155 | ||
156 | /* | |
157 | * dir_renames: computed directory renames | |
158 | * | |
159 | * This is a map of old_dir => new_dir and is derived in part from | |
160 | * dir_rename_count. | |
161 | */ | |
162 | struct strmap dir_renames[3]; | |
163 | ||
164 | /* | |
165 | * relevant_sources: deleted paths wanted in rename detection, and why | |
166 | * | |
167 | * relevant_sources is a set of deleted paths on each side of | |
168 | * history for which we need rename detection. If a path is deleted | |
169 | * on one side of history, we need to detect if it is part of a | |
170 | * rename if either | |
171 | * * the file is modified/deleted on the other side of history | |
172 | * * we need to detect renames for an ancestor directory | |
173 | * If neither of those are true, we can skip rename detection for | |
174 | * that path. The reason is stored as a value from enum | |
175 | * file_rename_relevance, as the reason can inform the algorithm in | |
176 | * diffcore_rename_extended(). | |
177 | */ | |
178 | struct strintmap relevant_sources[3]; | |
179 | ||
180 | struct deferred_traversal_data deferred[3]; | |
181 | ||
182 | /* | |
183 | * dir_rename_mask: | |
184 | * 0: optimization removing unmodified potential rename source okay | |
185 | * 2 or 4: optimization okay, but must check for files added to dir | |
186 | * 7: optimization forbidden; need rename source in case of dir rename | |
187 | */ | |
188 | unsigned dir_rename_mask:3; | |
189 | ||
190 | /* | |
191 | * callback_data_*: supporting data structures for alternate traversal | |
192 | * | |
193 | * We sometimes need to be able to traverse through all the files | |
194 | * in a given tree before all immediate subdirectories within that | |
195 | * tree. Since traverse_trees() doesn't do that naturally, we have | |
196 | * a traverse_trees_wrapper() that stores any immediate | |
197 | * subdirectories while traversing files, then traverses the | |
198 | * immediate subdirectories later. These callback_data* variables | |
199 | * store the information for the subdirectories so that we can do | |
200 | * that traversal order. | |
201 | */ | |
202 | struct traversal_callback_data *callback_data; | |
203 | int callback_data_nr, callback_data_alloc; | |
204 | char *callback_data_traverse_path; | |
205 | ||
206 | /* | |
207 | * merge_trees: trees passed to the merge algorithm for the merge | |
208 | * | |
209 | * merge_trees records the trees passed to the merge algorithm. But, | |
210 | * this data also is stored in merge_result->priv. If a sequence of | |
211 | * merges are being done (such as when cherry-picking or rebasing), | |
212 | * the next merge can look at this and re-use information from | |
213 | * previous merges under certain circumstances. | |
214 | * | |
215 | * See also all the cached_* variables. | |
216 | */ | |
217 | struct tree *merge_trees[3]; | |
218 | ||
219 | /* | |
220 | * cached_pairs_valid_side: which side's cached info can be reused | |
221 | * | |
222 | * See the description for merge_trees. For repeated merges, at most | |
223 | * only one side's cached information can be used. Valid values: | |
224 | * MERGE_SIDE2: cached data from side2 can be reused | |
225 | * MERGE_SIDE1: cached data from side1 can be reused | |
226 | * 0: no cached data can be reused | |
227 | * -1: See redo_after_renames; both sides can be reused. | |
228 | */ | |
229 | int cached_pairs_valid_side; | |
230 | ||
231 | /* | |
232 | * cached_pairs: Caching of renames and deletions. | |
233 | * | |
234 | * These are mappings recording renames and deletions of individual | |
235 | * files (not directories). They are thus a map from an old | |
236 | * filename to either NULL (for deletions) or a new filename (for | |
237 | * renames). | |
238 | */ | |
239 | struct strmap cached_pairs[3]; | |
240 | ||
241 | /* | |
242 | * cached_target_names: just the destinations from cached_pairs | |
243 | * | |
244 | * We sometimes want a fast lookup to determine if a given filename | |
245 | * is one of the destinations in cached_pairs. cached_target_names | |
246 | * is thus duplicative information, but it provides a fast lookup. | |
247 | */ | |
248 | struct strset cached_target_names[3]; | |
249 | ||
250 | /* | |
251 | * cached_irrelevant: Caching of rename_sources that aren't relevant. | |
252 | * | |
253 | * If we try to detect a rename for a source path and succeed, it's | |
254 | * part of a rename. If we try to detect a rename for a source path | |
255 | * and fail, then it's a delete. If we do not try to detect a rename | |
256 | * for a path, then we don't know if it's a rename or a delete. If | |
257 | * merge-ort doesn't think the path is relevant, then we just won't | |
258 | * cache anything for that path. But there's a slight problem in | |
259 | * that merge-ort can think a path is RELEVANT_LOCATION, but due to | |
260 | * commit 9bd342137e ("diffcore-rename: determine which | |
261 | * relevant_sources are no longer relevant", 2021-03-13), | |
262 | * diffcore-rename can downgrade the path to RELEVANT_NO_MORE. To | |
263 | * avoid excessive calls to diffcore_rename_extended() we still need | |
264 | * to cache such paths, though we cannot record them as either | |
265 | * renames or deletes. So we cache them here as a "turned out to be | |
266 | * irrelevant *for this commit*" as they are often also irrelevant | |
267 | * for subsequent commits, though we will have to do some extra | |
268 | * checking to see whether such paths become relevant for rename | |
269 | * detection when cherry-picking/rebasing subsequent commits. | |
270 | */ | |
271 | struct strset cached_irrelevant[3]; | |
272 | ||
273 | /* | |
274 | * redo_after_renames: optimization flag for "restarting" the merge | |
275 | * | |
276 | * Sometimes it pays to detect renames, cache them, and then | |
277 | * restart the merge operation from the beginning. The reason for | |
278 | * this is that when we know where all the renames are, we know | |
279 | * whether a certain directory has any paths under it affected -- | |
280 | * and if a directory is not affected then it permits us to do | |
281 | * trivial tree merging in more cases. Doing trivial tree merging | |
282 | * prevents the need to run process_entry() on every path | |
283 | * underneath trees that can be trivially merged, and | |
284 | * process_entry() is more expensive than collect_merge_info() -- | |
285 | * plus, the second collect_merge_info() will be much faster since | |
286 | * it doesn't have to recurse into the relevant trees. | |
287 | * | |
288 | * Values for this flag: | |
289 | * 0 = don't bother, not worth it (or conditions not yet checked) | |
290 | * 1 = conditions for optimization met, optimization worthwhile | |
291 | * 2 = we already did it (don't restart merge yet again) | |
292 | */ | |
293 | unsigned redo_after_renames; | |
294 | ||
295 | /* | |
296 | * needed_limit: value needed for inexact rename detection to run | |
297 | * | |
298 | * If the current rename limit wasn't high enough for inexact | |
299 | * rename detection to run, this records the limit needed. Otherwise, | |
300 | * this value remains 0. | |
301 | */ | |
302 | int needed_limit; | |
303 | }; | |
304 | ||
305 | struct merge_options_internal { | |
306 | /* | |
307 | * paths: primary data structure in all of merge ort. | |
308 | * | |
309 | * The keys of paths: | |
310 | * * are full relative paths from the toplevel of the repository | |
311 | * (e.g. "drivers/firmware/raspberrypi.c"). | |
312 | * * store all relevant paths in the repo, both directories and | |
313 | * files (e.g. drivers, drivers/firmware would also be included) | |
314 | * * these keys serve to intern all the path strings, which allows | |
315 | * us to do pointer comparison on directory names instead of | |
316 | * strcmp; we just have to be careful to use the interned strings. | |
317 | * | |
318 | * The values of paths: | |
319 | * * either a pointer to a merged_info, or a conflict_info struct | |
320 | * * merged_info contains all relevant information for a | |
321 | * non-conflicted entry. | |
322 | * * conflict_info contains a merged_info, plus any additional | |
323 | * information about a conflict such as the higher orders stages | |
324 | * involved and the names of the paths those came from (handy | |
325 | * once renames get involved). | |
326 | * * a path may start "conflicted" (i.e. point to a conflict_info) | |
327 | * and then a later step (e.g. three-way content merge) determines | |
328 | * it can be cleanly merged, at which point it'll be marked clean | |
329 | * and the algorithm will ignore any data outside the contained | |
330 | * merged_info for that entry | |
331 | * * If an entry remains conflicted, the merged_info portion of a | |
332 | * conflict_info will later be filled with whatever version of | |
333 | * the file should be placed in the working directory (e.g. an | |
334 | * as-merged-as-possible variation that contains conflict markers). | |
335 | */ | |
336 | struct strmap paths; | |
337 | ||
338 | /* | |
339 | * conflicted: a subset of keys->values from "paths" | |
340 | * | |
341 | * conflicted is basically an optimization between process_entries() | |
342 | * and record_conflicted_index_entries(); the latter could loop over | |
343 | * ALL the entries in paths AGAIN and look for the ones that are | |
344 | * still conflicted, but since process_entries() has to loop over | |
345 | * all of them, it saves the ones it couldn't resolve in this strmap | |
346 | * so that record_conflicted_index_entries() can iterate just the | |
347 | * relevant entries. | |
348 | */ | |
349 | struct strmap conflicted; | |
350 | ||
351 | /* | |
352 | * pool: memory pool for fast allocation/deallocation | |
353 | * | |
354 | * We allocate room for lots of filenames and auxiliary data | |
355 | * structures in merge_options_internal, and it tends to all be | |
356 | * freed together too. Using a memory pool for these provides a | |
357 | * nice speedup. | |
358 | */ | |
359 | struct mem_pool pool; | |
360 | ||
361 | /* | |
362 | * conflicts: logical conflicts and messages stored by _primary_ path | |
363 | * | |
364 | * This is a map of pathnames (a subset of the keys in "paths" above) | |
365 | * to struct string_list, with each item's `util` containing a | |
366 | * `struct logical_conflict_info`. Note, though, that for each path, | |
367 | * it only stores the logical conflicts for which that path is the | |
368 | * primary path; the path might be part of additional conflicts. | |
369 | */ | |
370 | struct strmap conflicts; | |
371 | ||
372 | /* | |
373 | * renames: various data relating to rename detection | |
374 | */ | |
375 | struct rename_info renames; | |
376 | ||
377 | /* | |
378 | * attr_index: hacky minimal index used for renormalization | |
379 | * | |
380 | * renormalization code _requires_ an index, though it only needs to | |
381 | * find a .gitattributes file within the index. So, when | |
382 | * renormalization is important, we create a special index with just | |
383 | * that one file. | |
384 | */ | |
385 | struct index_state attr_index; | |
386 | ||
387 | /* | |
388 | * current_dir_name, toplevel_dir: temporary vars | |
389 | * | |
390 | * These are used in collect_merge_info_callback(), and will set the | |
391 | * various merged_info.directory_name for the various paths we get; | |
392 | * see documentation for that variable and the requirements placed on | |
393 | * that field. | |
394 | */ | |
395 | const char *current_dir_name; | |
396 | const char *toplevel_dir; | |
397 | ||
398 | /* call_depth: recursion level counter for merging merge bases */ | |
399 | int call_depth; | |
400 | ||
401 | /* field that holds submodule conflict information */ | |
402 | struct string_list conflicted_submodules; | |
403 | }; | |
404 | ||
405 | struct conflicted_submodule_item { | |
406 | char *abbrev; | |
407 | int flag; | |
408 | }; | |
409 | ||
410 | static void conflicted_submodule_item_free(void *util, const char *str UNUSED) | |
411 | { | |
412 | struct conflicted_submodule_item *item = util; | |
413 | ||
414 | free(item->abbrev); | |
415 | free(item); | |
416 | } | |
417 | ||
418 | struct version_info { | |
419 | struct object_id oid; | |
420 | unsigned short mode; | |
421 | }; | |
422 | ||
423 | struct merged_info { | |
424 | /* if is_null, ignore result. otherwise result has oid & mode */ | |
425 | struct version_info result; | |
426 | unsigned is_null:1; | |
427 | ||
428 | /* | |
429 | * clean: whether the path in question is cleanly merged. | |
430 | * | |
431 | * see conflict_info.merged for more details. | |
432 | */ | |
433 | unsigned clean:1; | |
434 | ||
435 | /* | |
436 | * basename_offset: offset of basename of path. | |
437 | * | |
438 | * perf optimization to avoid recomputing offset of final '/' | |
439 | * character in pathname (0 if no '/' in pathname). | |
440 | */ | |
441 | size_t basename_offset; | |
442 | ||
443 | /* | |
444 | * directory_name: containing directory name. | |
445 | * | |
446 | * Note that we assume directory_name is constructed such that | |
447 | * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name, | |
448 | * i.e. string equality is equivalent to pointer equality. For this | |
449 | * to hold, we have to be careful setting directory_name. | |
450 | */ | |
451 | const char *directory_name; | |
452 | }; | |
453 | ||
454 | struct conflict_info { | |
455 | /* | |
456 | * merged: the version of the path that will be written to working tree | |
457 | * | |
458 | * WARNING: It is critical to check merged.clean and ensure it is 0 | |
459 | * before reading any conflict_info fields outside of merged. | |
460 | * Allocated merge_info structs will always have clean set to 1. | |
461 | * Allocated conflict_info structs will have merged.clean set to 0 | |
462 | * initially. The merged.clean field is how we know if it is safe | |
463 | * to access other parts of conflict_info besides merged; if a | |
464 | * conflict_info's merged.clean is changed to 1, the rest of the | |
465 | * algorithm is not allowed to look at anything outside of the | |
466 | * merged member anymore. | |
467 | */ | |
468 | struct merged_info merged; | |
469 | ||
470 | /* oids & modes from each of the three trees for this path */ | |
471 | struct version_info stages[3]; | |
472 | ||
473 | /* pathnames for each stage; may differ due to rename detection */ | |
474 | const char *pathnames[3]; | |
475 | ||
476 | /* Whether this path is/was involved in a directory/file conflict */ | |
477 | unsigned df_conflict:1; | |
478 | ||
479 | /* | |
480 | * Whether this path is/was involved in a non-content conflict other | |
481 | * than a directory/file conflict (e.g. rename/rename, rename/delete, | |
482 | * file location based on possible directory rename). | |
483 | */ | |
484 | unsigned path_conflict:1; | |
485 | ||
486 | /* | |
487 | * For filemask and dirmask, the ith bit corresponds to whether the | |
488 | * ith entry is a file (filemask) or a directory (dirmask). Thus, | |
489 | * filemask & dirmask is always zero, and filemask | dirmask is at | |
490 | * most 7 but can be less when a path does not appear as either a | |
491 | * file or a directory on at least one side of history. | |
492 | * | |
493 | * Note that these masks are related to enum merge_side, as the ith | |
494 | * entry corresponds to side i. | |
495 | * | |
496 | * These values come from a traverse_trees() call; more info may be | |
497 | * found looking at tree-walk.h's struct traverse_info, | |
498 | * particularly the documentation above the "fn" member (note that | |
499 | * filemask = mask & ~dirmask from that documentation). | |
500 | */ | |
501 | unsigned filemask:3; | |
502 | unsigned dirmask:3; | |
503 | ||
504 | /* | |
505 | * Optimization to track which stages match, to avoid the need to | |
506 | * recompute it in multiple steps. Either 0 or at least 2 bits are | |
507 | * set; if at least 2 bits are set, their corresponding stages match. | |
508 | */ | |
509 | unsigned match_mask:3; | |
510 | }; | |
511 | ||
512 | enum conflict_and_info_types { | |
513 | /* "Simple" conflicts and informational messages */ | |
514 | INFO_AUTO_MERGING = 0, | |
515 | CONFLICT_CONTENTS, /* text file that failed to merge */ | |
516 | CONFLICT_BINARY, | |
517 | CONFLICT_FILE_DIRECTORY, | |
518 | CONFLICT_DISTINCT_MODES, | |
519 | CONFLICT_MODIFY_DELETE, | |
520 | ||
521 | /* Regular rename */ | |
522 | CONFLICT_RENAME_RENAME, /* same file renamed differently */ | |
523 | CONFLICT_RENAME_COLLIDES, /* rename/add or two files renamed to 1 */ | |
524 | CONFLICT_RENAME_DELETE, | |
525 | ||
526 | /* Basic directory rename */ | |
527 | CONFLICT_DIR_RENAME_SUGGESTED, | |
528 | INFO_DIR_RENAME_APPLIED, | |
529 | ||
530 | /* Special directory rename cases */ | |
531 | INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME, | |
532 | CONFLICT_DIR_RENAME_FILE_IN_WAY, | |
533 | CONFLICT_DIR_RENAME_COLLISION, | |
534 | CONFLICT_DIR_RENAME_SPLIT, | |
535 | ||
536 | /* Basic submodule */ | |
537 | INFO_SUBMODULE_FAST_FORWARDING, | |
538 | CONFLICT_SUBMODULE_FAILED_TO_MERGE, | |
539 | ||
540 | /* Special submodule cases broken out from FAILED_TO_MERGE */ | |
541 | CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION, | |
542 | CONFLICT_SUBMODULE_NOT_INITIALIZED, | |
543 | CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE, | |
544 | CONFLICT_SUBMODULE_MAY_HAVE_REWINDS, | |
545 | CONFLICT_SUBMODULE_NULL_MERGE_BASE, | |
546 | CONFLICT_SUBMODULE_CORRUPT, | |
547 | ||
548 | /* Keep this entry _last_ in the list */ | |
549 | NB_CONFLICT_TYPES, | |
550 | }; | |
551 | ||
552 | /* | |
553 | * Short description of conflict type, relied upon by external tools. | |
554 | * | |
555 | * We can add more entries, but DO NOT change any of these strings. Also, | |
556 | * Order MUST match conflict_info_and_types. | |
557 | */ | |
558 | static const char *type_short_descriptions[] = { | |
559 | /*** "Simple" conflicts and informational messages ***/ | |
560 | [INFO_AUTO_MERGING] = "Auto-merging", | |
561 | [CONFLICT_CONTENTS] = "CONFLICT (contents)", | |
562 | [CONFLICT_BINARY] = "CONFLICT (binary)", | |
563 | [CONFLICT_FILE_DIRECTORY] = "CONFLICT (file/directory)", | |
564 | [CONFLICT_DISTINCT_MODES] = "CONFLICT (distinct modes)", | |
565 | [CONFLICT_MODIFY_DELETE] = "CONFLICT (modify/delete)", | |
566 | ||
567 | /*** Regular rename ***/ | |
568 | [CONFLICT_RENAME_RENAME] = "CONFLICT (rename/rename)", | |
569 | [CONFLICT_RENAME_COLLIDES] = "CONFLICT (rename involved in collision)", | |
570 | [CONFLICT_RENAME_DELETE] = "CONFLICT (rename/delete)", | |
571 | ||
572 | /*** Basic directory rename ***/ | |
573 | [CONFLICT_DIR_RENAME_SUGGESTED] = | |
574 | "CONFLICT (directory rename suggested)", | |
575 | [INFO_DIR_RENAME_APPLIED] = "Path updated due to directory rename", | |
576 | ||
577 | /*** Special directory rename cases ***/ | |
578 | [INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME] = | |
579 | "Directory rename skipped since directory was renamed on both sides", | |
580 | [CONFLICT_DIR_RENAME_FILE_IN_WAY] = | |
581 | "CONFLICT (file in way of directory rename)", | |
582 | [CONFLICT_DIR_RENAME_COLLISION] = "CONFLICT(directory rename collision)", | |
583 | [CONFLICT_DIR_RENAME_SPLIT] = "CONFLICT(directory rename unclear split)", | |
584 | ||
585 | /*** Basic submodule ***/ | |
586 | [INFO_SUBMODULE_FAST_FORWARDING] = "Fast forwarding submodule", | |
587 | [CONFLICT_SUBMODULE_FAILED_TO_MERGE] = "CONFLICT (submodule)", | |
588 | ||
589 | /*** Special submodule cases broken out from FAILED_TO_MERGE ***/ | |
590 | [CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION] = | |
591 | "CONFLICT (submodule with possible resolution)", | |
592 | [CONFLICT_SUBMODULE_NOT_INITIALIZED] = | |
593 | "CONFLICT (submodule not initialized)", | |
594 | [CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE] = | |
595 | "CONFLICT (submodule history not available)", | |
596 | [CONFLICT_SUBMODULE_MAY_HAVE_REWINDS] = | |
597 | "CONFLICT (submodule may have rewinds)", | |
598 | [CONFLICT_SUBMODULE_NULL_MERGE_BASE] = | |
599 | "CONFLICT (submodule lacks merge base)", | |
600 | [CONFLICT_SUBMODULE_CORRUPT] = | |
601 | "CONFLICT (submodule corrupt)" | |
602 | }; | |
603 | ||
604 | struct logical_conflict_info { | |
605 | enum conflict_and_info_types type; | |
606 | struct strvec paths; | |
607 | }; | |
608 | ||
609 | /*** Function Grouping: various utility functions ***/ | |
610 | ||
611 | /* | |
612 | * For the next three macros, see warning for conflict_info.merged. | |
613 | * | |
614 | * In each of the below, mi is a struct merged_info*, and ci was defined | |
615 | * as a struct conflict_info* (but we need to verify ci isn't actually | |
616 | * pointed at a struct merged_info*). | |
617 | * | |
618 | * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise. | |
619 | * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one. | |
620 | * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first. | |
621 | */ | |
622 | #define INITIALIZE_CI(ci, mi) do { \ | |
623 | (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \ | |
624 | } while (0) | |
625 | #define VERIFY_CI(ci) assert(ci && !ci->merged.clean); | |
626 | #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \ | |
627 | (ci) = (struct conflict_info *)(mi); \ | |
628 | assert((ci) && !(mi)->clean); \ | |
629 | } while (0) | |
630 | ||
631 | static void free_strmap_strings(struct strmap *map) | |
632 | { | |
633 | struct hashmap_iter iter; | |
634 | struct strmap_entry *entry; | |
635 | ||
636 | strmap_for_each_entry(map, &iter, entry) { | |
637 | free((char*)entry->key); | |
638 | } | |
639 | } | |
640 | ||
641 | static void clear_or_reinit_internal_opts(struct merge_options_internal *opti, | |
642 | int reinitialize) | |
643 | { | |
644 | struct rename_info *renames = &opti->renames; | |
645 | int i; | |
646 | void (*strmap_clear_func)(struct strmap *, int) = | |
647 | reinitialize ? strmap_partial_clear : strmap_clear; | |
648 | void (*strintmap_clear_func)(struct strintmap *) = | |
649 | reinitialize ? strintmap_partial_clear : strintmap_clear; | |
650 | void (*strset_clear_func)(struct strset *) = | |
651 | reinitialize ? strset_partial_clear : strset_clear; | |
652 | ||
653 | strmap_clear_func(&opti->paths, 0); | |
654 | ||
655 | /* | |
656 | * All keys and values in opti->conflicted are a subset of those in | |
657 | * opti->paths. We don't want to deallocate anything twice, so we | |
658 | * don't free the keys and we pass 0 for free_values. | |
659 | */ | |
660 | strmap_clear_func(&opti->conflicted, 0); | |
661 | ||
662 | if (opti->attr_index.cache_nr) /* true iff opt->renormalize */ | |
663 | discard_index(&opti->attr_index); | |
664 | ||
665 | /* Free memory used by various renames maps */ | |
666 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; ++i) { | |
667 | strintmap_clear_func(&renames->dirs_removed[i]); | |
668 | strmap_clear_func(&renames->dir_renames[i], 0); | |
669 | strintmap_clear_func(&renames->relevant_sources[i]); | |
670 | if (!reinitialize) | |
671 | assert(renames->cached_pairs_valid_side == 0); | |
672 | if (i != renames->cached_pairs_valid_side && | |
673 | -1 != renames->cached_pairs_valid_side) { | |
674 | strset_clear_func(&renames->cached_target_names[i]); | |
675 | strmap_clear_func(&renames->cached_pairs[i], 1); | |
676 | strset_clear_func(&renames->cached_irrelevant[i]); | |
677 | partial_clear_dir_rename_count(&renames->dir_rename_count[i]); | |
678 | if (!reinitialize) | |
679 | strmap_clear(&renames->dir_rename_count[i], 1); | |
680 | } | |
681 | } | |
682 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; ++i) { | |
683 | strintmap_clear_func(&renames->deferred[i].possible_trivial_merges); | |
684 | strset_clear_func(&renames->deferred[i].target_dirs); | |
685 | renames->deferred[i].trivial_merges_okay = 1; /* 1 == maybe */ | |
686 | } | |
687 | renames->cached_pairs_valid_side = 0; | |
688 | renames->dir_rename_mask = 0; | |
689 | ||
690 | if (!reinitialize) { | |
691 | struct hashmap_iter iter; | |
692 | struct strmap_entry *e; | |
693 | ||
694 | /* Release and free each strbuf found in output */ | |
695 | strmap_for_each_entry(&opti->conflicts, &iter, e) { | |
696 | struct string_list *list = e->value; | |
697 | for (int i = 0; i < list->nr; i++) { | |
698 | struct logical_conflict_info *info = | |
699 | list->items[i].util; | |
700 | strvec_clear(&info->paths); | |
701 | } | |
702 | /* | |
703 | * While strictly speaking we don't need to | |
704 | * free(conflicts) here because we could pass | |
705 | * free_values=1 when calling strmap_clear() on | |
706 | * opti->conflicts, that would require strmap_clear | |
707 | * to do another strmap_for_each_entry() loop, so we | |
708 | * just free it while we're iterating anyway. | |
709 | */ | |
710 | string_list_clear(list, 1); | |
711 | free(list); | |
712 | } | |
713 | strmap_clear(&opti->conflicts, 0); | |
714 | } | |
715 | ||
716 | mem_pool_discard(&opti->pool, 0); | |
717 | ||
718 | string_list_clear_func(&opti->conflicted_submodules, | |
719 | conflicted_submodule_item_free); | |
720 | ||
721 | /* Clean out callback_data as well. */ | |
722 | FREE_AND_NULL(renames->callback_data); | |
723 | renames->callback_data_nr = renames->callback_data_alloc = 0; | |
724 | } | |
725 | ||
726 | static void format_commit(struct strbuf *sb, | |
727 | int indent, | |
728 | struct repository *repo, | |
729 | struct commit *commit) | |
730 | { | |
731 | struct merge_remote_desc *desc; | |
732 | struct pretty_print_context ctx = {0}; | |
733 | ctx.abbrev = DEFAULT_ABBREV; | |
734 | ||
735 | strbuf_addchars(sb, ' ', indent); | |
736 | desc = merge_remote_util(commit); | |
737 | if (desc) { | |
738 | strbuf_addf(sb, "virtual %s\n", desc->name); | |
739 | return; | |
740 | } | |
741 | ||
742 | repo_format_commit_message(repo, commit, "%h %s", sb, &ctx); | |
743 | strbuf_addch(sb, '\n'); | |
744 | } | |
745 | ||
746 | __attribute__((format (printf, 8, 9))) | |
747 | static void path_msg(struct merge_options *opt, | |
748 | enum conflict_and_info_types type, | |
749 | int omittable_hint, /* skippable under --remerge-diff */ | |
750 | const char *primary_path, | |
751 | const char *other_path_1, /* may be NULL */ | |
752 | const char *other_path_2, /* may be NULL */ | |
753 | struct string_list *other_paths, /* may be NULL */ | |
754 | const char *fmt, ...) | |
755 | { | |
756 | va_list ap; | |
757 | struct string_list *path_conflicts; | |
758 | struct logical_conflict_info *info; | |
759 | struct strbuf buf = STRBUF_INIT; | |
760 | struct strbuf *dest; | |
761 | struct strbuf tmp = STRBUF_INIT; | |
762 | ||
763 | /* Sanity checks */ | |
764 | assert(omittable_hint == | |
765 | !starts_with(type_short_descriptions[type], "CONFLICT") || | |
766 | type == CONFLICT_DIR_RENAME_SUGGESTED); | |
767 | if (opt->record_conflict_msgs_as_headers && omittable_hint) | |
768 | return; /* Do not record mere hints in headers */ | |
769 | if (opt->priv->call_depth && opt->verbosity < 5) | |
770 | return; /* Ignore messages from inner merges */ | |
771 | ||
772 | /* Ensure path_conflicts (ptr to array of logical_conflict) allocated */ | |
773 | path_conflicts = strmap_get(&opt->priv->conflicts, primary_path); | |
774 | if (!path_conflicts) { | |
775 | path_conflicts = xmalloc(sizeof(*path_conflicts)); | |
776 | string_list_init_dup(path_conflicts); | |
777 | strmap_put(&opt->priv->conflicts, primary_path, path_conflicts); | |
778 | } | |
779 | ||
780 | /* Add a logical_conflict at the end to store info from this call */ | |
781 | info = xcalloc(1, sizeof(*info)); | |
782 | info->type = type; | |
783 | strvec_init(&info->paths); | |
784 | ||
785 | /* Handle the list of paths */ | |
786 | strvec_push(&info->paths, primary_path); | |
787 | if (other_path_1) | |
788 | strvec_push(&info->paths, other_path_1); | |
789 | if (other_path_2) | |
790 | strvec_push(&info->paths, other_path_2); | |
791 | if (other_paths) | |
792 | for (int i = 0; i < other_paths->nr; i++) | |
793 | strvec_push(&info->paths, other_paths->items[i].string); | |
794 | ||
795 | /* Handle message and its format, in normal case */ | |
796 | dest = (opt->record_conflict_msgs_as_headers ? &tmp : &buf); | |
797 | ||
798 | va_start(ap, fmt); | |
799 | if (opt->priv->call_depth) { | |
800 | strbuf_addchars(dest, ' ', 2); | |
801 | strbuf_addstr(dest, "From inner merge:"); | |
802 | strbuf_addchars(dest, ' ', opt->priv->call_depth * 2); | |
803 | } | |
804 | strbuf_vaddf(dest, fmt, ap); | |
805 | va_end(ap); | |
806 | ||
807 | /* Handle specialized formatting of message under --remerge-diff */ | |
808 | if (opt->record_conflict_msgs_as_headers) { | |
809 | int i_sb = 0, i_tmp = 0; | |
810 | ||
811 | /* Start with the specified prefix */ | |
812 | if (opt->msg_header_prefix) | |
813 | strbuf_addf(&buf, "%s ", opt->msg_header_prefix); | |
814 | ||
815 | /* Copy tmp to sb, adding spaces after newlines */ | |
816 | strbuf_grow(&buf, buf.len + 2*tmp.len); /* more than sufficient */ | |
817 | for (; i_tmp < tmp.len; i_tmp++, i_sb++) { | |
818 | /* Copy next character from tmp to sb */ | |
819 | buf.buf[buf.len + i_sb] = tmp.buf[i_tmp]; | |
820 | ||
821 | /* If we copied a newline, add a space */ | |
822 | if (tmp.buf[i_tmp] == '\n') | |
823 | buf.buf[++i_sb] = ' '; | |
824 | } | |
825 | /* Update length and ensure it's NUL-terminated */ | |
826 | buf.len += i_sb; | |
827 | buf.buf[buf.len] = '\0'; | |
828 | ||
829 | strbuf_release(&tmp); | |
830 | } | |
831 | string_list_append_nodup(path_conflicts, strbuf_detach(&buf, NULL)) | |
832 | ->util = info; | |
833 | } | |
834 | ||
835 | static struct diff_filespec *pool_alloc_filespec(struct mem_pool *pool, | |
836 | const char *path) | |
837 | { | |
838 | /* Similar to alloc_filespec(), but allocate from pool and reuse path */ | |
839 | struct diff_filespec *spec; | |
840 | ||
841 | spec = mem_pool_calloc(pool, 1, sizeof(*spec)); | |
842 | spec->path = (char*)path; /* spec won't modify it */ | |
843 | ||
844 | spec->count = 1; | |
845 | spec->is_binary = -1; | |
846 | return spec; | |
847 | } | |
848 | ||
849 | static struct diff_filepair *pool_diff_queue(struct mem_pool *pool, | |
850 | struct diff_queue_struct *queue, | |
851 | struct diff_filespec *one, | |
852 | struct diff_filespec *two) | |
853 | { | |
854 | /* Same code as diff_queue(), except allocate from pool */ | |
855 | struct diff_filepair *dp; | |
856 | ||
857 | dp = mem_pool_calloc(pool, 1, sizeof(*dp)); | |
858 | dp->one = one; | |
859 | dp->two = two; | |
860 | if (queue) | |
861 | diff_q(queue, dp); | |
862 | return dp; | |
863 | } | |
864 | ||
865 | /* add a string to a strbuf, but converting "/" to "_" */ | |
866 | static void add_flattened_path(struct strbuf *out, const char *s) | |
867 | { | |
868 | size_t i = out->len; | |
869 | strbuf_addstr(out, s); | |
870 | for (; i < out->len; i++) | |
871 | if (out->buf[i] == '/') | |
872 | out->buf[i] = '_'; | |
873 | } | |
874 | ||
875 | static char *unique_path(struct merge_options *opt, | |
876 | const char *path, | |
877 | const char *branch) | |
878 | { | |
879 | char *ret = NULL; | |
880 | struct strbuf newpath = STRBUF_INIT; | |
881 | int suffix = 0; | |
882 | size_t base_len; | |
883 | struct strmap *existing_paths = &opt->priv->paths; | |
884 | ||
885 | strbuf_addf(&newpath, "%s~", path); | |
886 | add_flattened_path(&newpath, branch); | |
887 | ||
888 | base_len = newpath.len; | |
889 | while (strmap_contains(existing_paths, newpath.buf)) { | |
890 | strbuf_setlen(&newpath, base_len); | |
891 | strbuf_addf(&newpath, "_%d", suffix++); | |
892 | } | |
893 | ||
894 | /* Track the new path in our memory pool */ | |
895 | ret = mem_pool_alloc(&opt->priv->pool, newpath.len + 1); | |
896 | memcpy(ret, newpath.buf, newpath.len + 1); | |
897 | strbuf_release(&newpath); | |
898 | return ret; | |
899 | } | |
900 | ||
901 | /*** Function Grouping: functions related to collect_merge_info() ***/ | |
902 | ||
903 | static int traverse_trees_wrapper_callback(int n, | |
904 | unsigned long mask, | |
905 | unsigned long dirmask, | |
906 | struct name_entry *names, | |
907 | struct traverse_info *info) | |
908 | { | |
909 | struct merge_options *opt = info->data; | |
910 | struct rename_info *renames = &opt->priv->renames; | |
911 | unsigned filemask = mask & ~dirmask; | |
912 | ||
913 | assert(n==3); | |
914 | ||
915 | if (!renames->callback_data_traverse_path) | |
916 | renames->callback_data_traverse_path = xstrdup(info->traverse_path); | |
917 | ||
918 | if (filemask && filemask == renames->dir_rename_mask) | |
919 | renames->dir_rename_mask = 0x07; | |
920 | ||
921 | ALLOC_GROW(renames->callback_data, renames->callback_data_nr + 1, | |
922 | renames->callback_data_alloc); | |
923 | renames->callback_data[renames->callback_data_nr].mask = mask; | |
924 | renames->callback_data[renames->callback_data_nr].dirmask = dirmask; | |
925 | COPY_ARRAY(renames->callback_data[renames->callback_data_nr].names, | |
926 | names, 3); | |
927 | renames->callback_data_nr++; | |
928 | ||
929 | return mask; | |
930 | } | |
931 | ||
932 | /* | |
933 | * Much like traverse_trees(), BUT: | |
934 | * - read all the tree entries FIRST, saving them | |
935 | * - note that the above step provides an opportunity to compute necessary | |
936 | * additional details before the "real" traversal | |
937 | * - loop through the saved entries and call the original callback on them | |
938 | */ | |
939 | static int traverse_trees_wrapper(struct index_state *istate, | |
940 | int n, | |
941 | struct tree_desc *t, | |
942 | struct traverse_info *info) | |
943 | { | |
944 | int ret, i, old_offset; | |
945 | traverse_callback_t old_fn; | |
946 | char *old_callback_data_traverse_path; | |
947 | struct merge_options *opt = info->data; | |
948 | struct rename_info *renames = &opt->priv->renames; | |
949 | ||
950 | assert(renames->dir_rename_mask == 2 || renames->dir_rename_mask == 4); | |
951 | ||
952 | old_callback_data_traverse_path = renames->callback_data_traverse_path; | |
953 | old_fn = info->fn; | |
954 | old_offset = renames->callback_data_nr; | |
955 | ||
956 | renames->callback_data_traverse_path = NULL; | |
957 | info->fn = traverse_trees_wrapper_callback; | |
958 | ret = traverse_trees(istate, n, t, info); | |
959 | if (ret < 0) | |
960 | return ret; | |
961 | ||
962 | info->traverse_path = renames->callback_data_traverse_path; | |
963 | info->fn = old_fn; | |
964 | for (i = old_offset; i < renames->callback_data_nr; ++i) { | |
965 | info->fn(n, | |
966 | renames->callback_data[i].mask, | |
967 | renames->callback_data[i].dirmask, | |
968 | renames->callback_data[i].names, | |
969 | info); | |
970 | } | |
971 | ||
972 | renames->callback_data_nr = old_offset; | |
973 | free(renames->callback_data_traverse_path); | |
974 | renames->callback_data_traverse_path = old_callback_data_traverse_path; | |
975 | info->traverse_path = NULL; | |
976 | return 0; | |
977 | } | |
978 | ||
979 | static void setup_path_info(struct merge_options *opt, | |
980 | struct string_list_item *result, | |
981 | const char *current_dir_name, | |
982 | int current_dir_name_len, | |
983 | char *fullpath, /* we'll take over ownership */ | |
984 | struct name_entry *names, | |
985 | struct name_entry *merged_version, | |
986 | unsigned is_null, /* boolean */ | |
987 | unsigned df_conflict, /* boolean */ | |
988 | unsigned filemask, | |
989 | unsigned dirmask, | |
990 | int resolved /* boolean */) | |
991 | { | |
992 | /* result->util is void*, so mi is a convenience typed variable */ | |
993 | struct merged_info *mi; | |
994 | ||
995 | assert(!is_null || resolved); | |
996 | assert(!df_conflict || !resolved); /* df_conflict implies !resolved */ | |
997 | assert(resolved == (merged_version != NULL)); | |
998 | ||
999 | mi = mem_pool_calloc(&opt->priv->pool, 1, | |
1000 | resolved ? sizeof(struct merged_info) : | |
1001 | sizeof(struct conflict_info)); | |
1002 | mi->directory_name = current_dir_name; | |
1003 | mi->basename_offset = current_dir_name_len; | |
1004 | mi->clean = !!resolved; | |
1005 | if (resolved) { | |
1006 | mi->result.mode = merged_version->mode; | |
1007 | oidcpy(&mi->result.oid, &merged_version->oid); | |
1008 | mi->is_null = !!is_null; | |
1009 | } else { | |
1010 | int i; | |
1011 | struct conflict_info *ci; | |
1012 | ||
1013 | ASSIGN_AND_VERIFY_CI(ci, mi); | |
1014 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
1015 | ci->pathnames[i] = fullpath; | |
1016 | ci->stages[i].mode = names[i].mode; | |
1017 | oidcpy(&ci->stages[i].oid, &names[i].oid); | |
1018 | } | |
1019 | ci->filemask = filemask; | |
1020 | ci->dirmask = dirmask; | |
1021 | ci->df_conflict = !!df_conflict; | |
1022 | if (dirmask) | |
1023 | /* | |
1024 | * Assume is_null for now, but if we have entries | |
1025 | * under the directory then when it is complete in | |
1026 | * write_completed_directory() it'll update this. | |
1027 | * Also, for D/F conflicts, we have to handle the | |
1028 | * directory first, then clear this bit and process | |
1029 | * the file to see how it is handled -- that occurs | |
1030 | * near the top of process_entry(). | |
1031 | */ | |
1032 | mi->is_null = 1; | |
1033 | } | |
1034 | strmap_put(&opt->priv->paths, fullpath, mi); | |
1035 | result->string = fullpath; | |
1036 | result->util = mi; | |
1037 | } | |
1038 | ||
1039 | static void add_pair(struct merge_options *opt, | |
1040 | struct name_entry *names, | |
1041 | const char *pathname, | |
1042 | unsigned side, | |
1043 | unsigned is_add /* if false, is_delete */, | |
1044 | unsigned match_mask, | |
1045 | unsigned dir_rename_mask) | |
1046 | { | |
1047 | struct diff_filespec *one, *two; | |
1048 | struct rename_info *renames = &opt->priv->renames; | |
1049 | int names_idx = is_add ? side : 0; | |
1050 | ||
1051 | if (is_add) { | |
1052 | assert(match_mask == 0 || match_mask == 6); | |
1053 | if (strset_contains(&renames->cached_target_names[side], | |
1054 | pathname)) | |
1055 | return; | |
1056 | } else { | |
1057 | unsigned content_relevant = (match_mask == 0); | |
1058 | unsigned location_relevant = (dir_rename_mask == 0x07); | |
1059 | ||
1060 | assert(match_mask == 0 || match_mask == 3 || match_mask == 5); | |
1061 | ||
1062 | /* | |
1063 | * If pathname is found in cached_irrelevant[side] due to | |
1064 | * previous pick but for this commit content is relevant, | |
1065 | * then we need to remove it from cached_irrelevant. | |
1066 | */ | |
1067 | if (content_relevant) | |
1068 | /* strset_remove is no-op if strset doesn't have key */ | |
1069 | strset_remove(&renames->cached_irrelevant[side], | |
1070 | pathname); | |
1071 | ||
1072 | /* | |
1073 | * We do not need to re-detect renames for paths that we already | |
1074 | * know the pairing, i.e. for cached_pairs (or | |
1075 | * cached_irrelevant). However, handle_deferred_entries() needs | |
1076 | * to loop over the union of keys from relevant_sources[side] and | |
1077 | * cached_pairs[side], so for simplicity we set relevant_sources | |
1078 | * for all the cached_pairs too and then strip them back out in | |
1079 | * prune_cached_from_relevant() at the beginning of | |
1080 | * detect_regular_renames(). | |
1081 | */ | |
1082 | if (content_relevant || location_relevant) { | |
1083 | /* content_relevant trumps location_relevant */ | |
1084 | strintmap_set(&renames->relevant_sources[side], pathname, | |
1085 | content_relevant ? RELEVANT_CONTENT : RELEVANT_LOCATION); | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * Avoid creating pair if we've already cached rename results. | |
1090 | * Note that we do this after setting relevant_sources[side] | |
1091 | * as noted in the comment above. | |
1092 | */ | |
1093 | if (strmap_contains(&renames->cached_pairs[side], pathname) || | |
1094 | strset_contains(&renames->cached_irrelevant[side], pathname)) | |
1095 | return; | |
1096 | } | |
1097 | ||
1098 | one = pool_alloc_filespec(&opt->priv->pool, pathname); | |
1099 | two = pool_alloc_filespec(&opt->priv->pool, pathname); | |
1100 | fill_filespec(is_add ? two : one, | |
1101 | &names[names_idx].oid, 1, names[names_idx].mode); | |
1102 | pool_diff_queue(&opt->priv->pool, &renames->pairs[side], one, two); | |
1103 | } | |
1104 | ||
1105 | static void collect_rename_info(struct merge_options *opt, | |
1106 | struct name_entry *names, | |
1107 | const char *dirname, | |
1108 | const char *fullname, | |
1109 | unsigned filemask, | |
1110 | unsigned dirmask, | |
1111 | unsigned match_mask) | |
1112 | { | |
1113 | struct rename_info *renames = &opt->priv->renames; | |
1114 | unsigned side; | |
1115 | ||
1116 | /* | |
1117 | * Update dir_rename_mask (determines ignore-rename-source validity) | |
1118 | * | |
1119 | * dir_rename_mask helps us keep track of when directory rename | |
1120 | * detection may be relevant. Basically, whenver a directory is | |
1121 | * removed on one side of history, and a file is added to that | |
1122 | * directory on the other side of history, directory rename | |
1123 | * detection is relevant (meaning we have to detect renames for all | |
1124 | * files within that directory to deduce where the directory | |
1125 | * moved). Also, whenever a directory needs directory rename | |
1126 | * detection, due to the "majority rules" choice for where to move | |
1127 | * it (see t6423 testcase 1f), we also need to detect renames for | |
1128 | * all files within subdirectories of that directory as well. | |
1129 | * | |
1130 | * Here we haven't looked at files within the directory yet, we are | |
1131 | * just looking at the directory itself. So, if we aren't yet in | |
1132 | * a case where a parent directory needed directory rename detection | |
1133 | * (i.e. dir_rename_mask != 0x07), and if the directory was removed | |
1134 | * on one side of history, record the mask of the other side of | |
1135 | * history in dir_rename_mask. | |
1136 | */ | |
1137 | if (renames->dir_rename_mask != 0x07 && | |
1138 | (dirmask == 3 || dirmask == 5)) { | |
1139 | /* simple sanity check */ | |
1140 | assert(renames->dir_rename_mask == 0 || | |
1141 | renames->dir_rename_mask == (dirmask & ~1)); | |
1142 | /* update dir_rename_mask; have it record mask of new side */ | |
1143 | renames->dir_rename_mask = (dirmask & ~1); | |
1144 | } | |
1145 | ||
1146 | /* Update dirs_removed, as needed */ | |
1147 | if (dirmask == 1 || dirmask == 3 || dirmask == 5) { | |
1148 | /* absent_mask = 0x07 - dirmask; sides = absent_mask/2 */ | |
1149 | unsigned sides = (0x07 - dirmask)/2; | |
1150 | unsigned relevance = (renames->dir_rename_mask == 0x07) ? | |
1151 | RELEVANT_FOR_ANCESTOR : NOT_RELEVANT; | |
1152 | /* | |
1153 | * Record relevance of this directory. However, note that | |
1154 | * when collect_merge_info_callback() recurses into this | |
1155 | * directory and calls collect_rename_info() on paths | |
1156 | * within that directory, if we find a path that was added | |
1157 | * to this directory on the other side of history, we will | |
1158 | * upgrade this value to RELEVANT_FOR_SELF; see below. | |
1159 | */ | |
1160 | if (sides & 1) | |
1161 | strintmap_set(&renames->dirs_removed[1], fullname, | |
1162 | relevance); | |
1163 | if (sides & 2) | |
1164 | strintmap_set(&renames->dirs_removed[2], fullname, | |
1165 | relevance); | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * Here's the block that potentially upgrades to RELEVANT_FOR_SELF. | |
1170 | * When we run across a file added to a directory. In such a case, | |
1171 | * find the directory of the file and upgrade its relevance. | |
1172 | */ | |
1173 | if (renames->dir_rename_mask == 0x07 && | |
1174 | (filemask == 2 || filemask == 4)) { | |
1175 | /* | |
1176 | * Need directory rename for parent directory on other side | |
1177 | * of history from added file. Thus | |
1178 | * side = (~filemask & 0x06) >> 1 | |
1179 | * or | |
1180 | * side = 3 - (filemask/2). | |
1181 | */ | |
1182 | unsigned side = 3 - (filemask >> 1); | |
1183 | strintmap_set(&renames->dirs_removed[side], dirname, | |
1184 | RELEVANT_FOR_SELF); | |
1185 | } | |
1186 | ||
1187 | if (filemask == 0 || filemask == 7) | |
1188 | return; | |
1189 | ||
1190 | for (side = MERGE_SIDE1; side <= MERGE_SIDE2; ++side) { | |
1191 | unsigned side_mask = (1 << side); | |
1192 | ||
1193 | /* Check for deletion on side */ | |
1194 | if ((filemask & 1) && !(filemask & side_mask)) | |
1195 | add_pair(opt, names, fullname, side, 0 /* delete */, | |
1196 | match_mask & filemask, | |
1197 | renames->dir_rename_mask); | |
1198 | ||
1199 | /* Check for addition on side */ | |
1200 | if (!(filemask & 1) && (filemask & side_mask)) | |
1201 | add_pair(opt, names, fullname, side, 1 /* add */, | |
1202 | match_mask & filemask, | |
1203 | renames->dir_rename_mask); | |
1204 | } | |
1205 | } | |
1206 | ||
1207 | static int collect_merge_info_callback(int n, | |
1208 | unsigned long mask, | |
1209 | unsigned long dirmask, | |
1210 | struct name_entry *names, | |
1211 | struct traverse_info *info) | |
1212 | { | |
1213 | /* | |
1214 | * n is 3. Always. | |
1215 | * common ancestor (mbase) has mask 1, and stored in index 0 of names | |
1216 | * head of side 1 (side1) has mask 2, and stored in index 1 of names | |
1217 | * head of side 2 (side2) has mask 4, and stored in index 2 of names | |
1218 | */ | |
1219 | struct merge_options *opt = info->data; | |
1220 | struct merge_options_internal *opti = opt->priv; | |
1221 | struct rename_info *renames = &opt->priv->renames; | |
1222 | struct string_list_item pi; /* Path Info */ | |
1223 | struct conflict_info *ci; /* typed alias to pi.util (which is void*) */ | |
1224 | struct name_entry *p; | |
1225 | size_t len; | |
1226 | char *fullpath; | |
1227 | const char *dirname = opti->current_dir_name; | |
1228 | unsigned prev_dir_rename_mask = renames->dir_rename_mask; | |
1229 | unsigned filemask = mask & ~dirmask; | |
1230 | unsigned match_mask = 0; /* will be updated below */ | |
1231 | unsigned mbase_null = !(mask & 1); | |
1232 | unsigned side1_null = !(mask & 2); | |
1233 | unsigned side2_null = !(mask & 4); | |
1234 | unsigned side1_matches_mbase = (!side1_null && !mbase_null && | |
1235 | names[0].mode == names[1].mode && | |
1236 | oideq(&names[0].oid, &names[1].oid)); | |
1237 | unsigned side2_matches_mbase = (!side2_null && !mbase_null && | |
1238 | names[0].mode == names[2].mode && | |
1239 | oideq(&names[0].oid, &names[2].oid)); | |
1240 | unsigned sides_match = (!side1_null && !side2_null && | |
1241 | names[1].mode == names[2].mode && | |
1242 | oideq(&names[1].oid, &names[2].oid)); | |
1243 | ||
1244 | /* | |
1245 | * Note: When a path is a file on one side of history and a directory | |
1246 | * in another, we have a directory/file conflict. In such cases, if | |
1247 | * the conflict doesn't resolve from renames and deletions, then we | |
1248 | * always leave directories where they are and move files out of the | |
1249 | * way. Thus, while struct conflict_info has a df_conflict field to | |
1250 | * track such conflicts, we ignore that field for any directories at | |
1251 | * a path and only pay attention to it for files at the given path. | |
1252 | * The fact that we leave directories were they are also means that | |
1253 | * we do not need to worry about getting additional df_conflict | |
1254 | * information propagated from parent directories down to children | |
1255 | * (unlike, say traverse_trees_recursive() in unpack-trees.c, which | |
1256 | * sets a newinfo.df_conflicts field specifically to propagate it). | |
1257 | */ | |
1258 | unsigned df_conflict = (filemask != 0) && (dirmask != 0); | |
1259 | ||
1260 | /* n = 3 is a fundamental assumption. */ | |
1261 | if (n != 3) | |
1262 | BUG("Called collect_merge_info_callback wrong"); | |
1263 | ||
1264 | /* | |
1265 | * A bunch of sanity checks verifying that traverse_trees() calls | |
1266 | * us the way I expect. Could just remove these at some point, | |
1267 | * though maybe they are helpful to future code readers. | |
1268 | */ | |
1269 | assert(mbase_null == is_null_oid(&names[0].oid)); | |
1270 | assert(side1_null == is_null_oid(&names[1].oid)); | |
1271 | assert(side2_null == is_null_oid(&names[2].oid)); | |
1272 | assert(!mbase_null || !side1_null || !side2_null); | |
1273 | assert(mask > 0 && mask < 8); | |
1274 | ||
1275 | /* Determine match_mask */ | |
1276 | if (side1_matches_mbase) | |
1277 | match_mask = (side2_matches_mbase ? 7 : 3); | |
1278 | else if (side2_matches_mbase) | |
1279 | match_mask = 5; | |
1280 | else if (sides_match) | |
1281 | match_mask = 6; | |
1282 | ||
1283 | /* | |
1284 | * Get the name of the relevant filepath, which we'll pass to | |
1285 | * setup_path_info() for tracking. | |
1286 | */ | |
1287 | p = names; | |
1288 | while (!p->mode) | |
1289 | p++; | |
1290 | len = traverse_path_len(info, p->pathlen); | |
1291 | ||
1292 | /* +1 in both of the following lines to include the NUL byte */ | |
1293 | fullpath = mem_pool_alloc(&opt->priv->pool, len + 1); | |
1294 | make_traverse_path(fullpath, len + 1, info, p->path, p->pathlen); | |
1295 | ||
1296 | /* | |
1297 | * If mbase, side1, and side2 all match, we can resolve early. Even | |
1298 | * if these are trees, there will be no renames or anything | |
1299 | * underneath. | |
1300 | */ | |
1301 | if (side1_matches_mbase && side2_matches_mbase) { | |
1302 | /* mbase, side1, & side2 all match; use mbase as resolution */ | |
1303 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
1304 | names, names+0, mbase_null, 0 /* df_conflict */, | |
1305 | filemask, dirmask, 1 /* resolved */); | |
1306 | return mask; | |
1307 | } | |
1308 | ||
1309 | /* | |
1310 | * If the sides match, and all three paths are present and are | |
1311 | * files, then we can take either as the resolution. We can't do | |
1312 | * this with trees, because there may be rename sources from the | |
1313 | * merge_base. | |
1314 | */ | |
1315 | if (sides_match && filemask == 0x07) { | |
1316 | /* use side1 (== side2) version as resolution */ | |
1317 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
1318 | names, names+1, side1_null, 0, | |
1319 | filemask, dirmask, 1); | |
1320 | return mask; | |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * If side1 matches mbase and all three paths are present and are | |
1325 | * files, then we can use side2 as the resolution. We cannot | |
1326 | * necessarily do so this for trees, because there may be rename | |
1327 | * destinations within side2. | |
1328 | */ | |
1329 | if (side1_matches_mbase && filemask == 0x07) { | |
1330 | /* use side2 version as resolution */ | |
1331 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
1332 | names, names+2, side2_null, 0, | |
1333 | filemask, dirmask, 1); | |
1334 | return mask; | |
1335 | } | |
1336 | ||
1337 | /* Similar to above but swapping sides 1 and 2 */ | |
1338 | if (side2_matches_mbase && filemask == 0x07) { | |
1339 | /* use side1 version as resolution */ | |
1340 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
1341 | names, names+1, side1_null, 0, | |
1342 | filemask, dirmask, 1); | |
1343 | return mask; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Sometimes we can tell that a source path need not be included in | |
1348 | * rename detection -- namely, whenever either | |
1349 | * side1_matches_mbase && side2_null | |
1350 | * or | |
1351 | * side2_matches_mbase && side1_null | |
1352 | * However, we call collect_rename_info() even in those cases, | |
1353 | * because exact renames are cheap and would let us remove both a | |
1354 | * source and destination path. We'll cull the unneeded sources | |
1355 | * later. | |
1356 | */ | |
1357 | collect_rename_info(opt, names, dirname, fullpath, | |
1358 | filemask, dirmask, match_mask); | |
1359 | ||
1360 | /* | |
1361 | * None of the special cases above matched, so we have a | |
1362 | * provisional conflict. (Rename detection might allow us to | |
1363 | * unconflict some more cases, but that comes later so all we can | |
1364 | * do now is record the different non-null file hashes.) | |
1365 | */ | |
1366 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
1367 | names, NULL, 0, df_conflict, filemask, dirmask, 0); | |
1368 | ||
1369 | ci = pi.util; | |
1370 | VERIFY_CI(ci); | |
1371 | ci->match_mask = match_mask; | |
1372 | ||
1373 | /* If dirmask, recurse into subdirectories */ | |
1374 | if (dirmask) { | |
1375 | struct traverse_info newinfo; | |
1376 | struct tree_desc t[3]; | |
1377 | void *buf[3] = {NULL, NULL, NULL}; | |
1378 | const char *original_dir_name; | |
1379 | int i, ret, side; | |
1380 | ||
1381 | /* | |
1382 | * Check for whether we can avoid recursing due to one side | |
1383 | * matching the merge base. The side that does NOT match is | |
1384 | * the one that might have a rename destination we need. | |
1385 | */ | |
1386 | assert(!side1_matches_mbase || !side2_matches_mbase); | |
1387 | side = side1_matches_mbase ? MERGE_SIDE2 : | |
1388 | side2_matches_mbase ? MERGE_SIDE1 : MERGE_BASE; | |
1389 | if (filemask == 0 && (dirmask == 2 || dirmask == 4)) { | |
1390 | /* | |
1391 | * Also defer recursing into new directories; set up a | |
1392 | * few variables to let us do so. | |
1393 | */ | |
1394 | ci->match_mask = (7 - dirmask); | |
1395 | side = dirmask / 2; | |
1396 | } | |
1397 | if (renames->dir_rename_mask != 0x07 && | |
1398 | side != MERGE_BASE && | |
1399 | renames->deferred[side].trivial_merges_okay && | |
1400 | !strset_contains(&renames->deferred[side].target_dirs, | |
1401 | pi.string)) { | |
1402 | strintmap_set(&renames->deferred[side].possible_trivial_merges, | |
1403 | pi.string, renames->dir_rename_mask); | |
1404 | renames->dir_rename_mask = prev_dir_rename_mask; | |
1405 | return mask; | |
1406 | } | |
1407 | ||
1408 | /* We need to recurse */ | |
1409 | ci->match_mask &= filemask; | |
1410 | newinfo = *info; | |
1411 | newinfo.prev = info; | |
1412 | newinfo.name = p->path; | |
1413 | newinfo.namelen = p->pathlen; | |
1414 | newinfo.pathlen = st_add3(newinfo.pathlen, p->pathlen, 1); | |
1415 | /* | |
1416 | * If this directory we are about to recurse into cared about | |
1417 | * its parent directory (the current directory) having a D/F | |
1418 | * conflict, then we'd propagate the masks in this way: | |
1419 | * newinfo.df_conflicts |= (mask & ~dirmask); | |
1420 | * But we don't worry about propagating D/F conflicts. (See | |
1421 | * comment near setting of local df_conflict variable near | |
1422 | * the beginning of this function). | |
1423 | */ | |
1424 | ||
1425 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
1426 | if (i == 1 && side1_matches_mbase) | |
1427 | t[1] = t[0]; | |
1428 | else if (i == 2 && side2_matches_mbase) | |
1429 | t[2] = t[0]; | |
1430 | else if (i == 2 && sides_match) | |
1431 | t[2] = t[1]; | |
1432 | else { | |
1433 | const struct object_id *oid = NULL; | |
1434 | if (dirmask & 1) | |
1435 | oid = &names[i].oid; | |
1436 | buf[i] = fill_tree_descriptor(opt->repo, | |
1437 | t + i, oid); | |
1438 | } | |
1439 | dirmask >>= 1; | |
1440 | } | |
1441 | ||
1442 | original_dir_name = opti->current_dir_name; | |
1443 | opti->current_dir_name = pi.string; | |
1444 | if (renames->dir_rename_mask == 0 || | |
1445 | renames->dir_rename_mask == 0x07) | |
1446 | ret = traverse_trees(NULL, 3, t, &newinfo); | |
1447 | else | |
1448 | ret = traverse_trees_wrapper(NULL, 3, t, &newinfo); | |
1449 | opti->current_dir_name = original_dir_name; | |
1450 | renames->dir_rename_mask = prev_dir_rename_mask; | |
1451 | ||
1452 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) | |
1453 | free(buf[i]); | |
1454 | ||
1455 | if (ret < 0) | |
1456 | return -1; | |
1457 | } | |
1458 | ||
1459 | return mask; | |
1460 | } | |
1461 | ||
1462 | static void resolve_trivial_directory_merge(struct conflict_info *ci, int side) | |
1463 | { | |
1464 | VERIFY_CI(ci); | |
1465 | assert((side == 1 && ci->match_mask == 5) || | |
1466 | (side == 2 && ci->match_mask == 3)); | |
1467 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
1468 | ci->merged.result.mode = ci->stages[side].mode; | |
1469 | ci->merged.is_null = is_null_oid(&ci->stages[side].oid); | |
1470 | ci->match_mask = 0; | |
1471 | ci->merged.clean = 1; /* (ci->filemask == 0); */ | |
1472 | } | |
1473 | ||
1474 | static int handle_deferred_entries(struct merge_options *opt, | |
1475 | struct traverse_info *info) | |
1476 | { | |
1477 | struct rename_info *renames = &opt->priv->renames; | |
1478 | struct hashmap_iter iter; | |
1479 | struct strmap_entry *entry; | |
1480 | int side, ret = 0; | |
1481 | int path_count_before, path_count_after = 0; | |
1482 | ||
1483 | path_count_before = strmap_get_size(&opt->priv->paths); | |
1484 | for (side = MERGE_SIDE1; side <= MERGE_SIDE2; side++) { | |
1485 | unsigned optimization_okay = 1; | |
1486 | struct strintmap copy; | |
1487 | ||
1488 | /* Loop over the set of paths we need to know rename info for */ | |
1489 | strset_for_each_entry(&renames->relevant_sources[side], | |
1490 | &iter, entry) { | |
1491 | char *rename_target, *dir, *dir_marker; | |
1492 | struct strmap_entry *e; | |
1493 | ||
1494 | /* | |
1495 | * If we don't know delete/rename info for this path, | |
1496 | * then we need to recurse into all trees to get all | |
1497 | * adds to make sure we have it. | |
1498 | */ | |
1499 | if (strset_contains(&renames->cached_irrelevant[side], | |
1500 | entry->key)) | |
1501 | continue; | |
1502 | e = strmap_get_entry(&renames->cached_pairs[side], | |
1503 | entry->key); | |
1504 | if (!e) { | |
1505 | optimization_okay = 0; | |
1506 | break; | |
1507 | } | |
1508 | ||
1509 | /* If this is a delete, we have enough info already */ | |
1510 | rename_target = e->value; | |
1511 | if (!rename_target) | |
1512 | continue; | |
1513 | ||
1514 | /* If we already walked the rename target, we're good */ | |
1515 | if (strmap_contains(&opt->priv->paths, rename_target)) | |
1516 | continue; | |
1517 | ||
1518 | /* | |
1519 | * Otherwise, we need to get a list of directories that | |
1520 | * will need to be recursed into to get this | |
1521 | * rename_target. | |
1522 | */ | |
1523 | dir = xstrdup(rename_target); | |
1524 | while ((dir_marker = strrchr(dir, '/'))) { | |
1525 | *dir_marker = '\0'; | |
1526 | if (strset_contains(&renames->deferred[side].target_dirs, | |
1527 | dir)) | |
1528 | break; | |
1529 | strset_add(&renames->deferred[side].target_dirs, | |
1530 | dir); | |
1531 | } | |
1532 | free(dir); | |
1533 | } | |
1534 | renames->deferred[side].trivial_merges_okay = optimization_okay; | |
1535 | /* | |
1536 | * We need to recurse into any directories in | |
1537 | * possible_trivial_merges[side] found in target_dirs[side]. | |
1538 | * But when we recurse, we may need to queue up some of the | |
1539 | * subdirectories for possible_trivial_merges[side]. Since | |
1540 | * we can't safely iterate through a hashmap while also adding | |
1541 | * entries, move the entries into 'copy', iterate over 'copy', | |
1542 | * and then we'll also iterate anything added into | |
1543 | * possible_trivial_merges[side] once this loop is done. | |
1544 | */ | |
1545 | copy = renames->deferred[side].possible_trivial_merges; | |
1546 | strintmap_init_with_options(&renames->deferred[side].possible_trivial_merges, | |
1547 | 0, | |
1548 | &opt->priv->pool, | |
1549 | 0); | |
1550 | strintmap_for_each_entry(©, &iter, entry) { | |
1551 | const char *path = entry->key; | |
1552 | unsigned dir_rename_mask = (intptr_t)entry->value; | |
1553 | struct conflict_info *ci; | |
1554 | unsigned dirmask; | |
1555 | struct tree_desc t[3]; | |
1556 | void *buf[3] = {NULL,}; | |
1557 | int i; | |
1558 | ||
1559 | ci = strmap_get(&opt->priv->paths, path); | |
1560 | VERIFY_CI(ci); | |
1561 | dirmask = ci->dirmask; | |
1562 | ||
1563 | if (optimization_okay && | |
1564 | !strset_contains(&renames->deferred[side].target_dirs, | |
1565 | path)) { | |
1566 | resolve_trivial_directory_merge(ci, side); | |
1567 | continue; | |
1568 | } | |
1569 | ||
1570 | info->name = path; | |
1571 | info->namelen = strlen(path); | |
1572 | info->pathlen = info->namelen + 1; | |
1573 | ||
1574 | for (i = 0; i < 3; i++, dirmask >>= 1) { | |
1575 | if (i == 1 && ci->match_mask == 3) | |
1576 | t[1] = t[0]; | |
1577 | else if (i == 2 && ci->match_mask == 5) | |
1578 | t[2] = t[0]; | |
1579 | else if (i == 2 && ci->match_mask == 6) | |
1580 | t[2] = t[1]; | |
1581 | else { | |
1582 | const struct object_id *oid = NULL; | |
1583 | if (dirmask & 1) | |
1584 | oid = &ci->stages[i].oid; | |
1585 | buf[i] = fill_tree_descriptor(opt->repo, | |
1586 | t+i, oid); | |
1587 | } | |
1588 | } | |
1589 | ||
1590 | ci->match_mask &= ci->filemask; | |
1591 | opt->priv->current_dir_name = path; | |
1592 | renames->dir_rename_mask = dir_rename_mask; | |
1593 | if (renames->dir_rename_mask == 0 || | |
1594 | renames->dir_rename_mask == 0x07) | |
1595 | ret = traverse_trees(NULL, 3, t, info); | |
1596 | else | |
1597 | ret = traverse_trees_wrapper(NULL, 3, t, info); | |
1598 | ||
1599 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) | |
1600 | free(buf[i]); | |
1601 | ||
1602 | if (ret < 0) | |
1603 | return ret; | |
1604 | } | |
1605 | strintmap_clear(©); | |
1606 | strintmap_for_each_entry(&renames->deferred[side].possible_trivial_merges, | |
1607 | &iter, entry) { | |
1608 | const char *path = entry->key; | |
1609 | struct conflict_info *ci; | |
1610 | ||
1611 | ci = strmap_get(&opt->priv->paths, path); | |
1612 | VERIFY_CI(ci); | |
1613 | ||
1614 | assert(renames->deferred[side].trivial_merges_okay && | |
1615 | !strset_contains(&renames->deferred[side].target_dirs, | |
1616 | path)); | |
1617 | resolve_trivial_directory_merge(ci, side); | |
1618 | } | |
1619 | if (!optimization_okay || path_count_after) | |
1620 | path_count_after = strmap_get_size(&opt->priv->paths); | |
1621 | } | |
1622 | if (path_count_after) { | |
1623 | /* | |
1624 | * The choice of wanted_factor here does not affect | |
1625 | * correctness, only performance. When the | |
1626 | * path_count_after / path_count_before | |
1627 | * ratio is high, redoing after renames is a big | |
1628 | * performance boost. I suspect that redoing is a wash | |
1629 | * somewhere near a value of 2, and below that redoing will | |
1630 | * slow things down. I applied a fudge factor and picked | |
1631 | * 3; see the commit message when this was introduced for | |
1632 | * back of the envelope calculations for this ratio. | |
1633 | */ | |
1634 | const int wanted_factor = 3; | |
1635 | ||
1636 | /* We should only redo collect_merge_info one time */ | |
1637 | assert(renames->redo_after_renames == 0); | |
1638 | ||
1639 | if (path_count_after / path_count_before >= wanted_factor) { | |
1640 | renames->redo_after_renames = 1; | |
1641 | renames->cached_pairs_valid_side = -1; | |
1642 | } | |
1643 | } else if (renames->redo_after_renames == 2) | |
1644 | renames->redo_after_renames = 0; | |
1645 | return ret; | |
1646 | } | |
1647 | ||
1648 | static int collect_merge_info(struct merge_options *opt, | |
1649 | struct tree *merge_base, | |
1650 | struct tree *side1, | |
1651 | struct tree *side2) | |
1652 | { | |
1653 | int ret; | |
1654 | struct tree_desc t[3]; | |
1655 | struct traverse_info info; | |
1656 | ||
1657 | opt->priv->toplevel_dir = ""; | |
1658 | opt->priv->current_dir_name = opt->priv->toplevel_dir; | |
1659 | setup_traverse_info(&info, opt->priv->toplevel_dir); | |
1660 | info.fn = collect_merge_info_callback; | |
1661 | info.data = opt; | |
1662 | info.show_all_errors = 1; | |
1663 | ||
1664 | if (parse_tree(merge_base) < 0 || | |
1665 | parse_tree(side1) < 0 || | |
1666 | parse_tree(side2) < 0) | |
1667 | return -1; | |
1668 | init_tree_desc(t + 0, &merge_base->object.oid, | |
1669 | merge_base->buffer, merge_base->size); | |
1670 | init_tree_desc(t + 1, &side1->object.oid, side1->buffer, side1->size); | |
1671 | init_tree_desc(t + 2, &side2->object.oid, side2->buffer, side2->size); | |
1672 | ||
1673 | trace2_region_enter("merge", "traverse_trees", opt->repo); | |
1674 | ret = traverse_trees(NULL, 3, t, &info); | |
1675 | if (ret == 0) | |
1676 | ret = handle_deferred_entries(opt, &info); | |
1677 | trace2_region_leave("merge", "traverse_trees", opt->repo); | |
1678 | ||
1679 | return ret; | |
1680 | } | |
1681 | ||
1682 | /*** Function Grouping: functions related to threeway content merges ***/ | |
1683 | ||
1684 | static int find_first_merges(struct repository *repo, | |
1685 | const char *path, | |
1686 | struct commit *a, | |
1687 | struct commit *b, | |
1688 | struct object_array *result) | |
1689 | { | |
1690 | int i, j; | |
1691 | struct object_array merges = OBJECT_ARRAY_INIT; | |
1692 | struct commit *commit; | |
1693 | int contains_another; | |
1694 | ||
1695 | char merged_revision[GIT_MAX_HEXSZ + 2]; | |
1696 | const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path", | |
1697 | "--all", merged_revision, NULL }; | |
1698 | struct rev_info revs; | |
1699 | struct setup_revision_opt rev_opts; | |
1700 | ||
1701 | memset(result, 0, sizeof(struct object_array)); | |
1702 | memset(&rev_opts, 0, sizeof(rev_opts)); | |
1703 | ||
1704 | /* get all revisions that merge commit a */ | |
1705 | xsnprintf(merged_revision, sizeof(merged_revision), "^%s", | |
1706 | oid_to_hex(&a->object.oid)); | |
1707 | repo_init_revisions(repo, &revs, NULL); | |
1708 | /* FIXME: can't handle linked worktrees in submodules yet */ | |
1709 | revs.single_worktree = path != NULL; | |
1710 | setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts); | |
1711 | ||
1712 | /* save all revisions from the above list that contain b */ | |
1713 | if (prepare_revision_walk(&revs)) | |
1714 | die("revision walk setup failed"); | |
1715 | while ((commit = get_revision(&revs)) != NULL) { | |
1716 | struct object *o = &(commit->object); | |
1717 | int ret = repo_in_merge_bases(repo, b, commit); | |
1718 | ||
1719 | if (ret < 0) { | |
1720 | object_array_clear(&merges); | |
1721 | release_revisions(&revs); | |
1722 | return ret; | |
1723 | } | |
1724 | if (ret > 0) | |
1725 | add_object_array(o, NULL, &merges); | |
1726 | } | |
1727 | reset_revision_walk(); | |
1728 | ||
1729 | /* Now we've got all merges that contain a and b. Prune all | |
1730 | * merges that contain another found merge and save them in | |
1731 | * result. | |
1732 | */ | |
1733 | for (i = 0; i < merges.nr; i++) { | |
1734 | struct commit *m1 = (struct commit *) merges.objects[i].item; | |
1735 | ||
1736 | contains_another = 0; | |
1737 | for (j = 0; j < merges.nr; j++) { | |
1738 | struct commit *m2 = (struct commit *) merges.objects[j].item; | |
1739 | if (i != j) { | |
1740 | int ret = repo_in_merge_bases(repo, m2, m1); | |
1741 | if (ret < 0) { | |
1742 | object_array_clear(&merges); | |
1743 | release_revisions(&revs); | |
1744 | return ret; | |
1745 | } | |
1746 | if (ret > 0) { | |
1747 | contains_another = 1; | |
1748 | break; | |
1749 | } | |
1750 | } | |
1751 | } | |
1752 | ||
1753 | if (!contains_another) | |
1754 | add_object_array(merges.objects[i].item, NULL, result); | |
1755 | } | |
1756 | ||
1757 | object_array_clear(&merges); | |
1758 | release_revisions(&revs); | |
1759 | return result->nr; | |
1760 | } | |
1761 | ||
1762 | static int merge_submodule(struct merge_options *opt, | |
1763 | const char *path, | |
1764 | const struct object_id *o, | |
1765 | const struct object_id *a, | |
1766 | const struct object_id *b, | |
1767 | struct object_id *result) | |
1768 | { | |
1769 | struct repository subrepo; | |
1770 | struct strbuf sb = STRBUF_INIT; | |
1771 | int ret = 0, ret2; | |
1772 | struct commit *commit_o, *commit_a, *commit_b; | |
1773 | int parent_count; | |
1774 | struct object_array merges; | |
1775 | ||
1776 | int i; | |
1777 | int search = !opt->priv->call_depth; | |
1778 | int sub_not_initialized = 1; | |
1779 | int sub_flag = CONFLICT_SUBMODULE_FAILED_TO_MERGE; | |
1780 | ||
1781 | /* store fallback answer in result in case we fail */ | |
1782 | oidcpy(result, opt->priv->call_depth ? o : a); | |
1783 | ||
1784 | /* we can not handle deletion conflicts */ | |
1785 | if (is_null_oid(a) || is_null_oid(b)) | |
1786 | BUG("submodule deleted on one side; this should be handled outside of merge_submodule()"); | |
1787 | ||
1788 | if ((sub_not_initialized = repo_submodule_init(&subrepo, | |
1789 | opt->repo, path, null_oid()))) { | |
1790 | path_msg(opt, CONFLICT_SUBMODULE_NOT_INITIALIZED, 0, | |
1791 | path, NULL, NULL, NULL, | |
1792 | _("Failed to merge submodule %s (not checked out)"), | |
1793 | path); | |
1794 | sub_flag = CONFLICT_SUBMODULE_NOT_INITIALIZED; | |
1795 | goto cleanup; | |
1796 | } | |
1797 | ||
1798 | if (is_null_oid(o)) { | |
1799 | path_msg(opt, CONFLICT_SUBMODULE_NULL_MERGE_BASE, 0, | |
1800 | path, NULL, NULL, NULL, | |
1801 | _("Failed to merge submodule %s (no merge base)"), | |
1802 | path); | |
1803 | goto cleanup; | |
1804 | } | |
1805 | ||
1806 | if (!(commit_o = lookup_commit_reference(&subrepo, o)) || | |
1807 | !(commit_a = lookup_commit_reference(&subrepo, a)) || | |
1808 | !(commit_b = lookup_commit_reference(&subrepo, b))) { | |
1809 | path_msg(opt, CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE, 0, | |
1810 | path, NULL, NULL, NULL, | |
1811 | _("Failed to merge submodule %s (commits not present)"), | |
1812 | path); | |
1813 | sub_flag = CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE; | |
1814 | goto cleanup; | |
1815 | } | |
1816 | ||
1817 | /* check whether both changes are forward */ | |
1818 | ret2 = repo_in_merge_bases(&subrepo, commit_o, commit_a); | |
1819 | if (ret2 < 0) { | |
1820 | path_msg(opt, CONFLICT_SUBMODULE_CORRUPT, 0, | |
1821 | path, NULL, NULL, NULL, | |
1822 | _("Failed to merge submodule %s " | |
1823 | "(repository corrupt)"), | |
1824 | path); | |
1825 | ret = -1; | |
1826 | goto cleanup; | |
1827 | } | |
1828 | if (ret2 > 0) | |
1829 | ret2 = repo_in_merge_bases(&subrepo, commit_o, commit_b); | |
1830 | if (ret2 < 0) { | |
1831 | path_msg(opt, CONFLICT_SUBMODULE_CORRUPT, 0, | |
1832 | path, NULL, NULL, NULL, | |
1833 | _("Failed to merge submodule %s " | |
1834 | "(repository corrupt)"), | |
1835 | path); | |
1836 | ret = -1; | |
1837 | goto cleanup; | |
1838 | } | |
1839 | if (!ret2) { | |
1840 | path_msg(opt, CONFLICT_SUBMODULE_MAY_HAVE_REWINDS, 0, | |
1841 | path, NULL, NULL, NULL, | |
1842 | _("Failed to merge submodule %s " | |
1843 | "(commits don't follow merge-base)"), | |
1844 | path); | |
1845 | goto cleanup; | |
1846 | } | |
1847 | ||
1848 | /* Case #1: a is contained in b or vice versa */ | |
1849 | ret2 = repo_in_merge_bases(&subrepo, commit_a, commit_b); | |
1850 | if (ret2 < 0) { | |
1851 | path_msg(opt, CONFLICT_SUBMODULE_CORRUPT, 0, | |
1852 | path, NULL, NULL, NULL, | |
1853 | _("Failed to merge submodule %s " | |
1854 | "(repository corrupt)"), | |
1855 | path); | |
1856 | ret = -1; | |
1857 | goto cleanup; | |
1858 | } | |
1859 | if (ret2 > 0) { | |
1860 | oidcpy(result, b); | |
1861 | path_msg(opt, INFO_SUBMODULE_FAST_FORWARDING, 1, | |
1862 | path, NULL, NULL, NULL, | |
1863 | _("Note: Fast-forwarding submodule %s to %s"), | |
1864 | path, oid_to_hex(b)); | |
1865 | ret = 1; | |
1866 | goto cleanup; | |
1867 | } | |
1868 | ret2 = repo_in_merge_bases(&subrepo, commit_b, commit_a); | |
1869 | if (ret2 < 0) { | |
1870 | path_msg(opt, CONFLICT_SUBMODULE_CORRUPT, 0, | |
1871 | path, NULL, NULL, NULL, | |
1872 | _("Failed to merge submodule %s " | |
1873 | "(repository corrupt)"), | |
1874 | path); | |
1875 | ret = -1; | |
1876 | goto cleanup; | |
1877 | } | |
1878 | if (ret2 > 0) { | |
1879 | oidcpy(result, a); | |
1880 | path_msg(opt, INFO_SUBMODULE_FAST_FORWARDING, 1, | |
1881 | path, NULL, NULL, NULL, | |
1882 | _("Note: Fast-forwarding submodule %s to %s"), | |
1883 | path, oid_to_hex(a)); | |
1884 | ret = 1; | |
1885 | goto cleanup; | |
1886 | } | |
1887 | ||
1888 | /* | |
1889 | * Case #2: There are one or more merges that contain a and b in | |
1890 | * the submodule. If there is only one, then present it as a | |
1891 | * suggestion to the user, but leave it marked unmerged so the | |
1892 | * user needs to confirm the resolution. | |
1893 | */ | |
1894 | ||
1895 | /* Skip the search if makes no sense to the calling context. */ | |
1896 | if (!search) | |
1897 | goto cleanup; | |
1898 | ||
1899 | /* find commit which merges them */ | |
1900 | parent_count = find_first_merges(&subrepo, path, commit_a, commit_b, | |
1901 | &merges); | |
1902 | switch (parent_count) { | |
1903 | case -1: | |
1904 | path_msg(opt, CONFLICT_SUBMODULE_CORRUPT, 0, | |
1905 | path, NULL, NULL, NULL, | |
1906 | _("Failed to merge submodule %s " | |
1907 | "(repository corrupt)"), | |
1908 | path); | |
1909 | ret = -1; | |
1910 | break; | |
1911 | case 0: | |
1912 | path_msg(opt, CONFLICT_SUBMODULE_FAILED_TO_MERGE, 0, | |
1913 | path, NULL, NULL, NULL, | |
1914 | _("Failed to merge submodule %s"), path); | |
1915 | break; | |
1916 | ||
1917 | case 1: | |
1918 | format_commit(&sb, 4, &subrepo, | |
1919 | (struct commit *)merges.objects[0].item); | |
1920 | path_msg(opt, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION, 0, | |
1921 | path, NULL, NULL, NULL, | |
1922 | _("Failed to merge submodule %s, but a possible merge " | |
1923 | "resolution exists: %s"), | |
1924 | path, sb.buf); | |
1925 | strbuf_release(&sb); | |
1926 | break; | |
1927 | default: | |
1928 | for (i = 0; i < merges.nr; i++) | |
1929 | format_commit(&sb, 4, &subrepo, | |
1930 | (struct commit *)merges.objects[i].item); | |
1931 | path_msg(opt, CONFLICT_SUBMODULE_FAILED_TO_MERGE_BUT_POSSIBLE_RESOLUTION, 0, | |
1932 | path, NULL, NULL, NULL, | |
1933 | _("Failed to merge submodule %s, but multiple " | |
1934 | "possible merges exist:\n%s"), path, sb.buf); | |
1935 | strbuf_release(&sb); | |
1936 | } | |
1937 | ||
1938 | object_array_clear(&merges); | |
1939 | cleanup: | |
1940 | if (!opt->priv->call_depth && !ret) { | |
1941 | struct string_list *csub = &opt->priv->conflicted_submodules; | |
1942 | struct conflicted_submodule_item *util; | |
1943 | const char *abbrev; | |
1944 | ||
1945 | util = xmalloc(sizeof(*util)); | |
1946 | util->flag = sub_flag; | |
1947 | util->abbrev = NULL; | |
1948 | if (!sub_not_initialized) { | |
1949 | abbrev = repo_find_unique_abbrev(&subrepo, b, DEFAULT_ABBREV); | |
1950 | util->abbrev = xstrdup(abbrev); | |
1951 | } | |
1952 | string_list_append(csub, path)->util = util; | |
1953 | } | |
1954 | ||
1955 | if (!sub_not_initialized) | |
1956 | repo_clear(&subrepo); | |
1957 | return ret; | |
1958 | } | |
1959 | ||
1960 | static void initialize_attr_index(struct merge_options *opt) | |
1961 | { | |
1962 | /* | |
1963 | * The renormalize_buffer() functions require attributes, and | |
1964 | * annoyingly those can only be read from the working tree or from | |
1965 | * an index_state. merge-ort doesn't have an index_state, so we | |
1966 | * generate a fake one containing only attribute information. | |
1967 | */ | |
1968 | struct merged_info *mi; | |
1969 | struct index_state *attr_index = &opt->priv->attr_index; | |
1970 | struct cache_entry *ce; | |
1971 | ||
1972 | attr_index->repo = opt->repo; | |
1973 | attr_index->initialized = 1; | |
1974 | ||
1975 | if (!opt->renormalize) | |
1976 | return; | |
1977 | ||
1978 | mi = strmap_get(&opt->priv->paths, GITATTRIBUTES_FILE); | |
1979 | if (!mi) | |
1980 | return; | |
1981 | ||
1982 | if (mi->clean) { | |
1983 | int len = strlen(GITATTRIBUTES_FILE); | |
1984 | ce = make_empty_cache_entry(attr_index, len); | |
1985 | ce->ce_mode = create_ce_mode(mi->result.mode); | |
1986 | ce->ce_flags = create_ce_flags(0); | |
1987 | ce->ce_namelen = len; | |
1988 | oidcpy(&ce->oid, &mi->result.oid); | |
1989 | memcpy(ce->name, GITATTRIBUTES_FILE, len); | |
1990 | add_index_entry(attr_index, ce, | |
1991 | ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE); | |
1992 | get_stream_filter(attr_index, GITATTRIBUTES_FILE, &ce->oid); | |
1993 | } else { | |
1994 | int stage, len; | |
1995 | struct conflict_info *ci; | |
1996 | ||
1997 | ASSIGN_AND_VERIFY_CI(ci, mi); | |
1998 | for (stage = 0; stage < 3; stage++) { | |
1999 | unsigned stage_mask = (1 << stage); | |
2000 | ||
2001 | if (!(ci->filemask & stage_mask)) | |
2002 | continue; | |
2003 | len = strlen(GITATTRIBUTES_FILE); | |
2004 | ce = make_empty_cache_entry(attr_index, len); | |
2005 | ce->ce_mode = create_ce_mode(ci->stages[stage].mode); | |
2006 | ce->ce_flags = create_ce_flags(stage); | |
2007 | ce->ce_namelen = len; | |
2008 | oidcpy(&ce->oid, &ci->stages[stage].oid); | |
2009 | memcpy(ce->name, GITATTRIBUTES_FILE, len); | |
2010 | add_index_entry(attr_index, ce, | |
2011 | ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE); | |
2012 | get_stream_filter(attr_index, GITATTRIBUTES_FILE, | |
2013 | &ce->oid); | |
2014 | } | |
2015 | } | |
2016 | } | |
2017 | ||
2018 | static int merge_3way(struct merge_options *opt, | |
2019 | const char *path, | |
2020 | const struct object_id *o, | |
2021 | const struct object_id *a, | |
2022 | const struct object_id *b, | |
2023 | const char *pathnames[3], | |
2024 | const int extra_marker_size, | |
2025 | mmbuffer_t *result_buf) | |
2026 | { | |
2027 | mmfile_t orig, src1, src2; | |
2028 | struct ll_merge_options ll_opts = LL_MERGE_OPTIONS_INIT; | |
2029 | char *base, *name1, *name2; | |
2030 | enum ll_merge_result merge_status; | |
2031 | ||
2032 | if (!opt->priv->attr_index.initialized) | |
2033 | initialize_attr_index(opt); | |
2034 | ||
2035 | ll_opts.renormalize = opt->renormalize; | |
2036 | ll_opts.extra_marker_size = extra_marker_size; | |
2037 | ll_opts.xdl_opts = opt->xdl_opts; | |
2038 | ll_opts.conflict_style = opt->conflict_style; | |
2039 | ||
2040 | if (opt->priv->call_depth) { | |
2041 | ll_opts.virtual_ancestor = 1; | |
2042 | ll_opts.variant = 0; | |
2043 | } else { | |
2044 | switch (opt->recursive_variant) { | |
2045 | case MERGE_VARIANT_OURS: | |
2046 | ll_opts.variant = XDL_MERGE_FAVOR_OURS; | |
2047 | break; | |
2048 | case MERGE_VARIANT_THEIRS: | |
2049 | ll_opts.variant = XDL_MERGE_FAVOR_THEIRS; | |
2050 | break; | |
2051 | default: | |
2052 | ll_opts.variant = 0; | |
2053 | break; | |
2054 | } | |
2055 | } | |
2056 | ||
2057 | assert(pathnames[0] && pathnames[1] && pathnames[2] && opt->ancestor); | |
2058 | if (pathnames[0] == pathnames[1] && pathnames[1] == pathnames[2]) { | |
2059 | base = mkpathdup("%s", opt->ancestor); | |
2060 | name1 = mkpathdup("%s", opt->branch1); | |
2061 | name2 = mkpathdup("%s", opt->branch2); | |
2062 | } else { | |
2063 | base = mkpathdup("%s:%s", opt->ancestor, pathnames[0]); | |
2064 | name1 = mkpathdup("%s:%s", opt->branch1, pathnames[1]); | |
2065 | name2 = mkpathdup("%s:%s", opt->branch2, pathnames[2]); | |
2066 | } | |
2067 | ||
2068 | read_mmblob(&orig, o); | |
2069 | read_mmblob(&src1, a); | |
2070 | read_mmblob(&src2, b); | |
2071 | ||
2072 | merge_status = ll_merge(result_buf, path, &orig, base, | |
2073 | &src1, name1, &src2, name2, | |
2074 | &opt->priv->attr_index, &ll_opts); | |
2075 | if (merge_status == LL_MERGE_BINARY_CONFLICT) | |
2076 | path_msg(opt, CONFLICT_BINARY, 0, | |
2077 | path, NULL, NULL, NULL, | |
2078 | "warning: Cannot merge binary files: %s (%s vs. %s)", | |
2079 | path, name1, name2); | |
2080 | ||
2081 | free(base); | |
2082 | free(name1); | |
2083 | free(name2); | |
2084 | free(orig.ptr); | |
2085 | free(src1.ptr); | |
2086 | free(src2.ptr); | |
2087 | return merge_status; | |
2088 | } | |
2089 | ||
2090 | static int handle_content_merge(struct merge_options *opt, | |
2091 | const char *path, | |
2092 | const struct version_info *o, | |
2093 | const struct version_info *a, | |
2094 | const struct version_info *b, | |
2095 | const char *pathnames[3], | |
2096 | const int extra_marker_size, | |
2097 | struct version_info *result) | |
2098 | { | |
2099 | /* | |
2100 | * path is the target location where we want to put the file, and | |
2101 | * is used to determine any normalization rules in ll_merge. | |
2102 | * | |
2103 | * The normal case is that path and all entries in pathnames are | |
2104 | * identical, though renames can affect which path we got one of | |
2105 | * the three blobs to merge on various sides of history. | |
2106 | * | |
2107 | * extra_marker_size is the amount to extend conflict markers in | |
2108 | * ll_merge; this is needed if we have content merges of content | |
2109 | * merges, which happens for example with rename/rename(2to1) and | |
2110 | * rename/add conflicts. | |
2111 | */ | |
2112 | unsigned clean = 1; | |
2113 | ||
2114 | /* | |
2115 | * handle_content_merge() needs both files to be of the same type, i.e. | |
2116 | * both files OR both submodules OR both symlinks. Conflicting types | |
2117 | * needs to be handled elsewhere. | |
2118 | */ | |
2119 | assert((S_IFMT & a->mode) == (S_IFMT & b->mode)); | |
2120 | ||
2121 | /* Merge modes */ | |
2122 | if (a->mode == b->mode || a->mode == o->mode) | |
2123 | result->mode = b->mode; | |
2124 | else { | |
2125 | /* must be the 100644/100755 case */ | |
2126 | assert(S_ISREG(a->mode)); | |
2127 | result->mode = a->mode; | |
2128 | clean = (b->mode == o->mode); | |
2129 | /* | |
2130 | * FIXME: If opt->priv->call_depth && !clean, then we really | |
2131 | * should not make result->mode match either a->mode or | |
2132 | * b->mode; that causes t6036 "check conflicting mode for | |
2133 | * regular file" to fail. It would be best to use some other | |
2134 | * mode, but we'll confuse all kinds of stuff if we use one | |
2135 | * where S_ISREG(result->mode) isn't true, and if we use | |
2136 | * something like 0100666, then tree-walk.c's calls to | |
2137 | * canon_mode() will just normalize that to 100644 for us and | |
2138 | * thus not solve anything. | |
2139 | * | |
2140 | * Figure out if there's some kind of way we can work around | |
2141 | * this... | |
2142 | */ | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * Trivial oid merge. | |
2147 | * | |
2148 | * Note: While one might assume that the next four lines would | |
2149 | * be unnecessary due to the fact that match_mask is often | |
2150 | * setup and already handled, renames don't always take care | |
2151 | * of that. | |
2152 | */ | |
2153 | if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid)) | |
2154 | oidcpy(&result->oid, &b->oid); | |
2155 | else if (oideq(&b->oid, &o->oid)) | |
2156 | oidcpy(&result->oid, &a->oid); | |
2157 | ||
2158 | /* Remaining rules depend on file vs. submodule vs. symlink. */ | |
2159 | else if (S_ISREG(a->mode)) { | |
2160 | mmbuffer_t result_buf; | |
2161 | int ret = 0, merge_status; | |
2162 | int two_way; | |
2163 | ||
2164 | /* | |
2165 | * If 'o' is different type, treat it as null so we do a | |
2166 | * two-way merge. | |
2167 | */ | |
2168 | two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
2169 | ||
2170 | merge_status = merge_3way(opt, path, | |
2171 | two_way ? null_oid() : &o->oid, | |
2172 | &a->oid, &b->oid, | |
2173 | pathnames, extra_marker_size, | |
2174 | &result_buf); | |
2175 | ||
2176 | if ((merge_status < 0) || !result_buf.ptr) | |
2177 | ret = error(_("failed to execute internal merge")); | |
2178 | ||
2179 | if (!ret && | |
2180 | write_object_file(result_buf.ptr, result_buf.size, | |
2181 | OBJ_BLOB, &result->oid)) | |
2182 | ret = error(_("unable to add %s to database"), path); | |
2183 | ||
2184 | free(result_buf.ptr); | |
2185 | if (ret) | |
2186 | return -1; | |
2187 | clean &= (merge_status == 0); | |
2188 | path_msg(opt, INFO_AUTO_MERGING, 1, path, NULL, NULL, NULL, | |
2189 | _("Auto-merging %s"), path); | |
2190 | } else if (S_ISGITLINK(a->mode)) { | |
2191 | int two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
2192 | clean = merge_submodule(opt, pathnames[0], | |
2193 | two_way ? null_oid() : &o->oid, | |
2194 | &a->oid, &b->oid, &result->oid); | |
2195 | if (opt->priv->call_depth && two_way && !clean) { | |
2196 | result->mode = o->mode; | |
2197 | oidcpy(&result->oid, &o->oid); | |
2198 | } | |
2199 | } else if (S_ISLNK(a->mode)) { | |
2200 | if (opt->priv->call_depth) { | |
2201 | clean = 0; | |
2202 | result->mode = o->mode; | |
2203 | oidcpy(&result->oid, &o->oid); | |
2204 | } else { | |
2205 | switch (opt->recursive_variant) { | |
2206 | case MERGE_VARIANT_NORMAL: | |
2207 | clean = 0; | |
2208 | oidcpy(&result->oid, &a->oid); | |
2209 | break; | |
2210 | case MERGE_VARIANT_OURS: | |
2211 | oidcpy(&result->oid, &a->oid); | |
2212 | break; | |
2213 | case MERGE_VARIANT_THEIRS: | |
2214 | oidcpy(&result->oid, &b->oid); | |
2215 | break; | |
2216 | } | |
2217 | } | |
2218 | } else | |
2219 | BUG("unsupported object type in the tree: %06o for %s", | |
2220 | a->mode, path); | |
2221 | ||
2222 | return clean; | |
2223 | } | |
2224 | ||
2225 | /*** Function Grouping: functions related to detect_and_process_renames(), *** | |
2226 | *** which are split into directory and regular rename detection sections. ***/ | |
2227 | ||
2228 | /*** Function Grouping: functions related to directory rename detection ***/ | |
2229 | ||
2230 | struct collision_info { | |
2231 | struct string_list source_files; | |
2232 | unsigned reported_already:1; | |
2233 | }; | |
2234 | ||
2235 | /* | |
2236 | * Return a new string that replaces the beginning portion (which matches | |
2237 | * rename_info->key), with rename_info->util.new_dir. In perl-speak: | |
2238 | * new_path_name = (old_path =~ s/rename_info->key/rename_info->value/); | |
2239 | * NOTE: | |
2240 | * Caller must ensure that old_path starts with rename_info->key + '/'. | |
2241 | */ | |
2242 | static char *apply_dir_rename(struct strmap_entry *rename_info, | |
2243 | const char *old_path) | |
2244 | { | |
2245 | struct strbuf new_path = STRBUF_INIT; | |
2246 | const char *old_dir = rename_info->key; | |
2247 | const char *new_dir = rename_info->value; | |
2248 | int oldlen, newlen, new_dir_len; | |
2249 | ||
2250 | oldlen = strlen(old_dir); | |
2251 | if (*new_dir == '\0') | |
2252 | /* | |
2253 | * If someone renamed/merged a subdirectory into the root | |
2254 | * directory (e.g. 'some/subdir' -> ''), then we want to | |
2255 | * avoid returning | |
2256 | * '' + '/filename' | |
2257 | * as the rename; we need to make old_path + oldlen advance | |
2258 | * past the '/' character. | |
2259 | */ | |
2260 | oldlen++; | |
2261 | new_dir_len = strlen(new_dir); | |
2262 | newlen = new_dir_len + (strlen(old_path) - oldlen) + 1; | |
2263 | strbuf_grow(&new_path, newlen); | |
2264 | strbuf_add(&new_path, new_dir, new_dir_len); | |
2265 | strbuf_addstr(&new_path, &old_path[oldlen]); | |
2266 | ||
2267 | return strbuf_detach(&new_path, NULL); | |
2268 | } | |
2269 | ||
2270 | static int path_in_way(struct strmap *paths, const char *path, unsigned side_mask) | |
2271 | { | |
2272 | struct merged_info *mi = strmap_get(paths, path); | |
2273 | struct conflict_info *ci; | |
2274 | if (!mi) | |
2275 | return 0; | |
2276 | INITIALIZE_CI(ci, mi); | |
2277 | return mi->clean || (side_mask & (ci->filemask | ci->dirmask)); | |
2278 | } | |
2279 | ||
2280 | /* | |
2281 | * See if there is a directory rename for path, and if there are any file | |
2282 | * level conflicts on the given side for the renamed location. If there is | |
2283 | * a rename and there are no conflicts, return the new name. Otherwise, | |
2284 | * return NULL. | |
2285 | */ | |
2286 | static char *handle_path_level_conflicts(struct merge_options *opt, | |
2287 | const char *path, | |
2288 | unsigned side_index, | |
2289 | struct strmap_entry *rename_info, | |
2290 | struct strmap *collisions) | |
2291 | { | |
2292 | char *new_path = NULL; | |
2293 | struct collision_info *c_info; | |
2294 | int clean = 1; | |
2295 | struct strbuf collision_paths = STRBUF_INIT; | |
2296 | ||
2297 | /* | |
2298 | * entry has the mapping of old directory name to new directory name | |
2299 | * that we want to apply to path. | |
2300 | */ | |
2301 | new_path = apply_dir_rename(rename_info, path); | |
2302 | if (!new_path) | |
2303 | BUG("Failed to apply directory rename!"); | |
2304 | ||
2305 | /* | |
2306 | * The caller needs to have ensured that it has pre-populated | |
2307 | * collisions with all paths that map to new_path. Do a quick check | |
2308 | * to ensure that's the case. | |
2309 | */ | |
2310 | c_info = strmap_get(collisions, new_path); | |
2311 | if (!c_info) | |
2312 | BUG("c_info is NULL"); | |
2313 | ||
2314 | /* | |
2315 | * Check for one-sided add/add/.../add conflicts, i.e. | |
2316 | * where implicit renames from the other side doing | |
2317 | * directory rename(s) can affect this side of history | |
2318 | * to put multiple paths into the same location. Warn | |
2319 | * and bail on directory renames for such paths. | |
2320 | */ | |
2321 | if (c_info->reported_already) { | |
2322 | clean = 0; | |
2323 | } else if (path_in_way(&opt->priv->paths, new_path, 1 << side_index)) { | |
2324 | c_info->reported_already = 1; | |
2325 | strbuf_add_separated_string_list(&collision_paths, ", ", | |
2326 | &c_info->source_files); | |
2327 | path_msg(opt, CONFLICT_DIR_RENAME_FILE_IN_WAY, 0, | |
2328 | new_path, NULL, NULL, &c_info->source_files, | |
2329 | _("CONFLICT (implicit dir rename): Existing " | |
2330 | "file/dir at %s in the way of implicit " | |
2331 | "directory rename(s) putting the following " | |
2332 | "path(s) there: %s."), | |
2333 | new_path, collision_paths.buf); | |
2334 | clean = 0; | |
2335 | } else if (c_info->source_files.nr > 1) { | |
2336 | c_info->reported_already = 1; | |
2337 | strbuf_add_separated_string_list(&collision_paths, ", ", | |
2338 | &c_info->source_files); | |
2339 | path_msg(opt, CONFLICT_DIR_RENAME_COLLISION, 0, | |
2340 | new_path, NULL, NULL, &c_info->source_files, | |
2341 | _("CONFLICT (implicit dir rename): Cannot map " | |
2342 | "more than one path to %s; implicit directory " | |
2343 | "renames tried to put these paths there: %s"), | |
2344 | new_path, collision_paths.buf); | |
2345 | clean = 0; | |
2346 | } | |
2347 | ||
2348 | /* Free memory we no longer need */ | |
2349 | strbuf_release(&collision_paths); | |
2350 | if (!clean && new_path) { | |
2351 | free(new_path); | |
2352 | return NULL; | |
2353 | } | |
2354 | ||
2355 | return new_path; | |
2356 | } | |
2357 | ||
2358 | static void get_provisional_directory_renames(struct merge_options *opt, | |
2359 | unsigned side, | |
2360 | int *clean) | |
2361 | { | |
2362 | struct hashmap_iter iter; | |
2363 | struct strmap_entry *entry; | |
2364 | struct rename_info *renames = &opt->priv->renames; | |
2365 | ||
2366 | /* | |
2367 | * Collapse | |
2368 | * dir_rename_count: old_directory -> {new_directory -> count} | |
2369 | * down to | |
2370 | * dir_renames: old_directory -> best_new_directory | |
2371 | * where best_new_directory is the one with the unique highest count. | |
2372 | */ | |
2373 | strmap_for_each_entry(&renames->dir_rename_count[side], &iter, entry) { | |
2374 | const char *source_dir = entry->key; | |
2375 | struct strintmap *counts = entry->value; | |
2376 | struct hashmap_iter count_iter; | |
2377 | struct strmap_entry *count_entry; | |
2378 | int max = 0; | |
2379 | int bad_max = 0; | |
2380 | const char *best = NULL; | |
2381 | ||
2382 | strintmap_for_each_entry(counts, &count_iter, count_entry) { | |
2383 | const char *target_dir = count_entry->key; | |
2384 | intptr_t count = (intptr_t)count_entry->value; | |
2385 | ||
2386 | if (count == max) | |
2387 | bad_max = max; | |
2388 | else if (count > max) { | |
2389 | max = count; | |
2390 | best = target_dir; | |
2391 | } | |
2392 | } | |
2393 | ||
2394 | if (max == 0) | |
2395 | continue; | |
2396 | ||
2397 | if (bad_max == max) { | |
2398 | path_msg(opt, CONFLICT_DIR_RENAME_SPLIT, 0, | |
2399 | source_dir, NULL, NULL, NULL, | |
2400 | _("CONFLICT (directory rename split): " | |
2401 | "Unclear where to rename %s to; it was " | |
2402 | "renamed to multiple other directories, " | |
2403 | "with no destination getting a majority of " | |
2404 | "the files."), | |
2405 | source_dir); | |
2406 | *clean = 0; | |
2407 | } else { | |
2408 | strmap_put(&renames->dir_renames[side], | |
2409 | source_dir, (void*)best); | |
2410 | } | |
2411 | } | |
2412 | } | |
2413 | ||
2414 | static void handle_directory_level_conflicts(struct merge_options *opt) | |
2415 | { | |
2416 | struct hashmap_iter iter; | |
2417 | struct strmap_entry *entry; | |
2418 | struct string_list duplicated = STRING_LIST_INIT_NODUP; | |
2419 | struct rename_info *renames = &opt->priv->renames; | |
2420 | struct strmap *side1_dir_renames = &renames->dir_renames[MERGE_SIDE1]; | |
2421 | struct strmap *side2_dir_renames = &renames->dir_renames[MERGE_SIDE2]; | |
2422 | int i; | |
2423 | ||
2424 | strmap_for_each_entry(side1_dir_renames, &iter, entry) { | |
2425 | if (strmap_contains(side2_dir_renames, entry->key)) | |
2426 | string_list_append(&duplicated, entry->key); | |
2427 | } | |
2428 | ||
2429 | for (i = 0; i < duplicated.nr; i++) { | |
2430 | strmap_remove(side1_dir_renames, duplicated.items[i].string, 0); | |
2431 | strmap_remove(side2_dir_renames, duplicated.items[i].string, 0); | |
2432 | } | |
2433 | string_list_clear(&duplicated, 0); | |
2434 | } | |
2435 | ||
2436 | static struct strmap_entry *check_dir_renamed(const char *path, | |
2437 | struct strmap *dir_renames) | |
2438 | { | |
2439 | char *temp = xstrdup(path); | |
2440 | char *end; | |
2441 | struct strmap_entry *e = NULL; | |
2442 | ||
2443 | while ((end = strrchr(temp, '/'))) { | |
2444 | *end = '\0'; | |
2445 | e = strmap_get_entry(dir_renames, temp); | |
2446 | if (e) | |
2447 | break; | |
2448 | } | |
2449 | free(temp); | |
2450 | return e; | |
2451 | } | |
2452 | ||
2453 | static void compute_collisions(struct strmap *collisions, | |
2454 | struct strmap *dir_renames, | |
2455 | struct diff_queue_struct *pairs) | |
2456 | { | |
2457 | int i; | |
2458 | ||
2459 | strmap_init_with_options(collisions, NULL, 0); | |
2460 | if (strmap_empty(dir_renames)) | |
2461 | return; | |
2462 | ||
2463 | /* | |
2464 | * Multiple files can be mapped to the same path due to directory | |
2465 | * renames done by the other side of history. Since that other | |
2466 | * side of history could have merged multiple directories into one, | |
2467 | * if our side of history added the same file basename to each of | |
2468 | * those directories, then all N of them would get implicitly | |
2469 | * renamed by the directory rename detection into the same path, | |
2470 | * and we'd get an add/add/.../add conflict, and all those adds | |
2471 | * from *this* side of history. This is not representable in the | |
2472 | * index, and users aren't going to easily be able to make sense of | |
2473 | * it. So we need to provide a good warning about what's | |
2474 | * happening, and fall back to no-directory-rename detection | |
2475 | * behavior for those paths. | |
2476 | * | |
2477 | * See testcases 9e and all of section 5 from t6043 for examples. | |
2478 | */ | |
2479 | for (i = 0; i < pairs->nr; ++i) { | |
2480 | struct strmap_entry *rename_info; | |
2481 | struct collision_info *collision_info; | |
2482 | char *new_path; | |
2483 | struct diff_filepair *pair = pairs->queue[i]; | |
2484 | ||
2485 | if (pair->status != 'A' && pair->status != 'R') | |
2486 | continue; | |
2487 | rename_info = check_dir_renamed(pair->two->path, dir_renames); | |
2488 | if (!rename_info) | |
2489 | continue; | |
2490 | ||
2491 | new_path = apply_dir_rename(rename_info, pair->two->path); | |
2492 | assert(new_path); | |
2493 | collision_info = strmap_get(collisions, new_path); | |
2494 | if (collision_info) { | |
2495 | free(new_path); | |
2496 | } else { | |
2497 | CALLOC_ARRAY(collision_info, 1); | |
2498 | string_list_init_nodup(&collision_info->source_files); | |
2499 | strmap_put(collisions, new_path, collision_info); | |
2500 | } | |
2501 | string_list_insert(&collision_info->source_files, | |
2502 | pair->two->path); | |
2503 | } | |
2504 | } | |
2505 | ||
2506 | static void free_collisions(struct strmap *collisions) | |
2507 | { | |
2508 | struct hashmap_iter iter; | |
2509 | struct strmap_entry *entry; | |
2510 | ||
2511 | /* Free each value in the collisions map */ | |
2512 | strmap_for_each_entry(collisions, &iter, entry) { | |
2513 | struct collision_info *info = entry->value; | |
2514 | string_list_clear(&info->source_files, 0); | |
2515 | } | |
2516 | /* | |
2517 | * In compute_collisions(), we set collisions.strdup_strings to 0 | |
2518 | * so that we wouldn't have to make another copy of the new_path | |
2519 | * allocated by apply_dir_rename(). But now that we've used them | |
2520 | * and have no other references to these strings, it is time to | |
2521 | * deallocate them. | |
2522 | */ | |
2523 | free_strmap_strings(collisions); | |
2524 | strmap_clear(collisions, 1); | |
2525 | } | |
2526 | ||
2527 | static char *check_for_directory_rename(struct merge_options *opt, | |
2528 | const char *path, | |
2529 | unsigned side_index, | |
2530 | struct strmap *dir_renames, | |
2531 | struct strmap *dir_rename_exclusions, | |
2532 | struct strmap *collisions, | |
2533 | int *clean_merge) | |
2534 | { | |
2535 | char *new_path; | |
2536 | struct strmap_entry *rename_info; | |
2537 | struct strmap_entry *otherinfo; | |
2538 | const char *new_dir; | |
2539 | int other_side = 3 - side_index; | |
2540 | ||
2541 | /* | |
2542 | * Cases where we don't have or don't want a directory rename for | |
2543 | * this path. | |
2544 | */ | |
2545 | if (strmap_empty(dir_renames)) | |
2546 | return NULL; | |
2547 | if (strmap_get(&collisions[other_side], path)) | |
2548 | return NULL; | |
2549 | rename_info = check_dir_renamed(path, dir_renames); | |
2550 | if (!rename_info) | |
2551 | return NULL; | |
2552 | ||
2553 | /* | |
2554 | * This next part is a little weird. We do not want to do an | |
2555 | * implicit rename into a directory we renamed on our side, because | |
2556 | * that will result in a spurious rename/rename(1to2) conflict. An | |
2557 | * example: | |
2558 | * Base commit: dumbdir/afile, otherdir/bfile | |
2559 | * Side 1: smrtdir/afile, otherdir/bfile | |
2560 | * Side 2: dumbdir/afile, dumbdir/bfile | |
2561 | * Here, while working on Side 1, we could notice that otherdir was | |
2562 | * renamed/merged to dumbdir, and change the diff_filepair for | |
2563 | * otherdir/bfile into a rename into dumbdir/bfile. However, Side | |
2564 | * 2 will notice the rename from dumbdir to smrtdir, and do the | |
2565 | * transitive rename to move it from dumbdir/bfile to | |
2566 | * smrtdir/bfile. That gives us bfile in dumbdir vs being in | |
2567 | * smrtdir, a rename/rename(1to2) conflict. We really just want | |
2568 | * the file to end up in smrtdir. And the way to achieve that is | |
2569 | * to not let Side1 do the rename to dumbdir, since we know that is | |
2570 | * the source of one of our directory renames. | |
2571 | * | |
2572 | * That's why otherinfo and dir_rename_exclusions is here. | |
2573 | * | |
2574 | * As it turns out, this also prevents N-way transient rename | |
2575 | * confusion; See testcases 9c and 9d of t6043. | |
2576 | */ | |
2577 | new_dir = rename_info->value; /* old_dir = rename_info->key; */ | |
2578 | otherinfo = strmap_get_entry(dir_rename_exclusions, new_dir); | |
2579 | if (otherinfo) { | |
2580 | path_msg(opt, INFO_DIR_RENAME_SKIPPED_DUE_TO_RERENAME, 1, | |
2581 | rename_info->key, path, new_dir, NULL, | |
2582 | _("WARNING: Avoiding applying %s -> %s rename " | |
2583 | "to %s, because %s itself was renamed."), | |
2584 | rename_info->key, new_dir, path, new_dir); | |
2585 | return NULL; | |
2586 | } | |
2587 | ||
2588 | new_path = handle_path_level_conflicts(opt, path, side_index, | |
2589 | rename_info, | |
2590 | &collisions[side_index]); | |
2591 | *clean_merge &= (new_path != NULL); | |
2592 | ||
2593 | return new_path; | |
2594 | } | |
2595 | ||
2596 | static void apply_directory_rename_modifications(struct merge_options *opt, | |
2597 | struct diff_filepair *pair, | |
2598 | char *new_path) | |
2599 | { | |
2600 | /* | |
2601 | * The basic idea is to get the conflict_info from opt->priv->paths | |
2602 | * at old path, and insert it into new_path; basically just this: | |
2603 | * ci = strmap_get(&opt->priv->paths, old_path); | |
2604 | * strmap_remove(&opt->priv->paths, old_path, 0); | |
2605 | * strmap_put(&opt->priv->paths, new_path, ci); | |
2606 | * However, there are some factors complicating this: | |
2607 | * - opt->priv->paths may already have an entry at new_path | |
2608 | * - Each ci tracks its containing directory, so we need to | |
2609 | * update that | |
2610 | * - If another ci has the same containing directory, then | |
2611 | * the two char*'s MUST point to the same location. See the | |
2612 | * comment in struct merged_info. strcmp equality is not | |
2613 | * enough; we need pointer equality. | |
2614 | * - opt->priv->paths must hold the parent directories of any | |
2615 | * entries that are added. So, if this directory rename | |
2616 | * causes entirely new directories, we must recursively add | |
2617 | * parent directories. | |
2618 | * - For each parent directory added to opt->priv->paths, we | |
2619 | * also need to get its parent directory stored in its | |
2620 | * conflict_info->merged.directory_name with all the same | |
2621 | * requirements about pointer equality. | |
2622 | */ | |
2623 | struct string_list dirs_to_insert = STRING_LIST_INIT_NODUP; | |
2624 | struct conflict_info *ci, *new_ci; | |
2625 | struct strmap_entry *entry; | |
2626 | const char *branch_with_new_path, *branch_with_dir_rename; | |
2627 | const char *old_path = pair->two->path; | |
2628 | const char *parent_name; | |
2629 | const char *cur_path; | |
2630 | int i, len; | |
2631 | ||
2632 | entry = strmap_get_entry(&opt->priv->paths, old_path); | |
2633 | old_path = entry->key; | |
2634 | ci = entry->value; | |
2635 | VERIFY_CI(ci); | |
2636 | ||
2637 | /* Find parent directories missing from opt->priv->paths */ | |
2638 | cur_path = mem_pool_strdup(&opt->priv->pool, new_path); | |
2639 | free((char*)new_path); | |
2640 | new_path = (char *)cur_path; | |
2641 | ||
2642 | while (1) { | |
2643 | /* Find the parent directory of cur_path */ | |
2644 | char *last_slash = strrchr(cur_path, '/'); | |
2645 | if (last_slash) { | |
2646 | parent_name = mem_pool_strndup(&opt->priv->pool, | |
2647 | cur_path, | |
2648 | last_slash - cur_path); | |
2649 | } else { | |
2650 | parent_name = opt->priv->toplevel_dir; | |
2651 | break; | |
2652 | } | |
2653 | ||
2654 | /* Look it up in opt->priv->paths */ | |
2655 | entry = strmap_get_entry(&opt->priv->paths, parent_name); | |
2656 | if (entry) { | |
2657 | parent_name = entry->key; /* reuse known pointer */ | |
2658 | break; | |
2659 | } | |
2660 | ||
2661 | /* Record this is one of the directories we need to insert */ | |
2662 | string_list_append(&dirs_to_insert, parent_name); | |
2663 | cur_path = parent_name; | |
2664 | } | |
2665 | ||
2666 | /* Traverse dirs_to_insert and insert them into opt->priv->paths */ | |
2667 | for (i = dirs_to_insert.nr-1; i >= 0; --i) { | |
2668 | struct conflict_info *dir_ci; | |
2669 | char *cur_dir = dirs_to_insert.items[i].string; | |
2670 | ||
2671 | CALLOC_ARRAY(dir_ci, 1); | |
2672 | ||
2673 | dir_ci->merged.directory_name = parent_name; | |
2674 | len = strlen(parent_name); | |
2675 | /* len+1 because of trailing '/' character */ | |
2676 | dir_ci->merged.basename_offset = (len > 0 ? len+1 : len); | |
2677 | dir_ci->dirmask = ci->filemask; | |
2678 | strmap_put(&opt->priv->paths, cur_dir, dir_ci); | |
2679 | ||
2680 | parent_name = cur_dir; | |
2681 | } | |
2682 | ||
2683 | assert(ci->filemask == 2 || ci->filemask == 4); | |
2684 | assert(ci->dirmask == 0 || ci->dirmask == 1); | |
2685 | if (ci->dirmask == 0) | |
2686 | strmap_remove(&opt->priv->paths, old_path, 0); | |
2687 | else { | |
2688 | /* | |
2689 | * This file exists on one side, but we still had a directory | |
2690 | * at the old location that we can't remove until after | |
2691 | * processing all paths below it. So, make a copy of ci in | |
2692 | * new_ci and only put the file information into it. | |
2693 | */ | |
2694 | new_ci = mem_pool_calloc(&opt->priv->pool, 1, sizeof(*new_ci)); | |
2695 | memcpy(new_ci, ci, sizeof(*ci)); | |
2696 | assert(!new_ci->match_mask); | |
2697 | new_ci->dirmask = 0; | |
2698 | new_ci->stages[1].mode = 0; | |
2699 | oidcpy(&new_ci->stages[1].oid, null_oid()); | |
2700 | ||
2701 | /* | |
2702 | * Now that we have the file information in new_ci, make sure | |
2703 | * ci only has the directory information. | |
2704 | */ | |
2705 | ci->filemask = 0; | |
2706 | ci->merged.clean = 1; | |
2707 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
2708 | if (ci->dirmask & (1 << i)) | |
2709 | continue; | |
2710 | /* zero out any entries related to files */ | |
2711 | ci->stages[i].mode = 0; | |
2712 | oidcpy(&ci->stages[i].oid, null_oid()); | |
2713 | } | |
2714 | ||
2715 | /* Now we want to focus on new_ci, so reassign ci to it. */ | |
2716 | ci = new_ci; | |
2717 | } | |
2718 | ||
2719 | branch_with_new_path = (ci->filemask == 2) ? opt->branch1 : opt->branch2; | |
2720 | branch_with_dir_rename = (ci->filemask == 2) ? opt->branch2 : opt->branch1; | |
2721 | ||
2722 | /* Now, finally update ci and stick it into opt->priv->paths */ | |
2723 | ci->merged.directory_name = parent_name; | |
2724 | len = strlen(parent_name); | |
2725 | ci->merged.basename_offset = (len > 0 ? len+1 : len); | |
2726 | new_ci = strmap_get(&opt->priv->paths, new_path); | |
2727 | if (!new_ci) { | |
2728 | /* Place ci back into opt->priv->paths, but at new_path */ | |
2729 | strmap_put(&opt->priv->paths, new_path, ci); | |
2730 | } else { | |
2731 | int index; | |
2732 | ||
2733 | /* A few sanity checks */ | |
2734 | VERIFY_CI(new_ci); | |
2735 | assert(ci->filemask == 2 || ci->filemask == 4); | |
2736 | assert((new_ci->filemask & ci->filemask) == 0); | |
2737 | assert(!new_ci->merged.clean); | |
2738 | ||
2739 | /* Copy stuff from ci into new_ci */ | |
2740 | new_ci->filemask |= ci->filemask; | |
2741 | if (new_ci->dirmask) | |
2742 | new_ci->df_conflict = 1; | |
2743 | index = (ci->filemask >> 1); | |
2744 | new_ci->pathnames[index] = ci->pathnames[index]; | |
2745 | new_ci->stages[index].mode = ci->stages[index].mode; | |
2746 | oidcpy(&new_ci->stages[index].oid, &ci->stages[index].oid); | |
2747 | ||
2748 | ci = new_ci; | |
2749 | } | |
2750 | ||
2751 | if (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) { | |
2752 | /* Notify user of updated path */ | |
2753 | if (pair->status == 'A') | |
2754 | path_msg(opt, INFO_DIR_RENAME_APPLIED, 1, | |
2755 | new_path, old_path, NULL, NULL, | |
2756 | _("Path updated: %s added in %s inside a " | |
2757 | "directory that was renamed in %s; moving " | |
2758 | "it to %s."), | |
2759 | old_path, branch_with_new_path, | |
2760 | branch_with_dir_rename, new_path); | |
2761 | else | |
2762 | path_msg(opt, INFO_DIR_RENAME_APPLIED, 1, | |
2763 | new_path, old_path, NULL, NULL, | |
2764 | _("Path updated: %s renamed to %s in %s, " | |
2765 | "inside a directory that was renamed in %s; " | |
2766 | "moving it to %s."), | |
2767 | pair->one->path, old_path, branch_with_new_path, | |
2768 | branch_with_dir_rename, new_path); | |
2769 | } else { | |
2770 | /* | |
2771 | * opt->detect_directory_renames has the value | |
2772 | * MERGE_DIRECTORY_RENAMES_CONFLICT, so mark these as conflicts. | |
2773 | */ | |
2774 | ci->path_conflict = 1; | |
2775 | if (pair->status == 'A') | |
2776 | path_msg(opt, CONFLICT_DIR_RENAME_SUGGESTED, 1, | |
2777 | new_path, old_path, NULL, NULL, | |
2778 | _("CONFLICT (file location): %s added in %s " | |
2779 | "inside a directory that was renamed in %s, " | |
2780 | "suggesting it should perhaps be moved to " | |
2781 | "%s."), | |
2782 | old_path, branch_with_new_path, | |
2783 | branch_with_dir_rename, new_path); | |
2784 | else | |
2785 | path_msg(opt, CONFLICT_DIR_RENAME_SUGGESTED, 1, | |
2786 | new_path, old_path, NULL, NULL, | |
2787 | _("CONFLICT (file location): %s renamed to %s " | |
2788 | "in %s, inside a directory that was renamed " | |
2789 | "in %s, suggesting it should perhaps be " | |
2790 | "moved to %s."), | |
2791 | pair->one->path, old_path, branch_with_new_path, | |
2792 | branch_with_dir_rename, new_path); | |
2793 | } | |
2794 | ||
2795 | /* | |
2796 | * Finally, record the new location. | |
2797 | */ | |
2798 | pair->two->path = new_path; | |
2799 | } | |
2800 | ||
2801 | /*** Function Grouping: functions related to regular rename detection ***/ | |
2802 | ||
2803 | static int process_renames(struct merge_options *opt, | |
2804 | struct diff_queue_struct *renames) | |
2805 | { | |
2806 | int clean_merge = 1, i; | |
2807 | ||
2808 | for (i = 0; i < renames->nr; ++i) { | |
2809 | const char *oldpath = NULL, *newpath; | |
2810 | struct diff_filepair *pair = renames->queue[i]; | |
2811 | struct conflict_info *oldinfo = NULL, *newinfo = NULL; | |
2812 | struct strmap_entry *old_ent, *new_ent; | |
2813 | unsigned int old_sidemask; | |
2814 | int target_index, other_source_index; | |
2815 | int source_deleted, collision, type_changed; | |
2816 | const char *rename_branch = NULL, *delete_branch = NULL; | |
2817 | ||
2818 | old_ent = strmap_get_entry(&opt->priv->paths, pair->one->path); | |
2819 | new_ent = strmap_get_entry(&opt->priv->paths, pair->two->path); | |
2820 | if (old_ent) { | |
2821 | oldpath = old_ent->key; | |
2822 | oldinfo = old_ent->value; | |
2823 | } | |
2824 | newpath = pair->two->path; | |
2825 | if (new_ent) { | |
2826 | newpath = new_ent->key; | |
2827 | newinfo = new_ent->value; | |
2828 | } | |
2829 | ||
2830 | /* | |
2831 | * If pair->one->path isn't in opt->priv->paths, that means | |
2832 | * that either directory rename detection removed that | |
2833 | * path, or a parent directory of oldpath was resolved and | |
2834 | * we don't even need the rename; in either case, we can | |
2835 | * skip it. If oldinfo->merged.clean, then the other side | |
2836 | * of history had no changes to oldpath and we don't need | |
2837 | * the rename and can skip it. | |
2838 | */ | |
2839 | if (!oldinfo || oldinfo->merged.clean) | |
2840 | continue; | |
2841 | ||
2842 | /* | |
2843 | * diff_filepairs have copies of pathnames, thus we have to | |
2844 | * use standard 'strcmp()' (negated) instead of '=='. | |
2845 | */ | |
2846 | if (i + 1 < renames->nr && | |
2847 | !strcmp(oldpath, renames->queue[i+1]->one->path)) { | |
2848 | /* Handle rename/rename(1to2) or rename/rename(1to1) */ | |
2849 | const char *pathnames[3]; | |
2850 | struct version_info merged; | |
2851 | struct conflict_info *base, *side1, *side2; | |
2852 | unsigned was_binary_blob = 0; | |
2853 | ||
2854 | pathnames[0] = oldpath; | |
2855 | pathnames[1] = newpath; | |
2856 | pathnames[2] = renames->queue[i+1]->two->path; | |
2857 | ||
2858 | base = strmap_get(&opt->priv->paths, pathnames[0]); | |
2859 | side1 = strmap_get(&opt->priv->paths, pathnames[1]); | |
2860 | side2 = strmap_get(&opt->priv->paths, pathnames[2]); | |
2861 | ||
2862 | VERIFY_CI(base); | |
2863 | VERIFY_CI(side1); | |
2864 | VERIFY_CI(side2); | |
2865 | ||
2866 | if (!strcmp(pathnames[1], pathnames[2])) { | |
2867 | struct rename_info *ri = &opt->priv->renames; | |
2868 | int j; | |
2869 | ||
2870 | /* Both sides renamed the same way */ | |
2871 | assert(side1 == side2); | |
2872 | memcpy(&side1->stages[0], &base->stages[0], | |
2873 | sizeof(merged)); | |
2874 | side1->filemask |= (1 << MERGE_BASE); | |
2875 | /* Mark base as resolved by removal */ | |
2876 | base->merged.is_null = 1; | |
2877 | base->merged.clean = 1; | |
2878 | ||
2879 | /* | |
2880 | * Disable remembering renames optimization; | |
2881 | * rename/rename(1to1) is incredibly rare, and | |
2882 | * just disabling the optimization is easier | |
2883 | * than purging cached_pairs, | |
2884 | * cached_target_names, and dir_rename_counts. | |
2885 | */ | |
2886 | for (j = 0; j < 3; j++) | |
2887 | ri->merge_trees[j] = NULL; | |
2888 | ||
2889 | /* We handled both renames, i.e. i+1 handled */ | |
2890 | i++; | |
2891 | /* Move to next rename */ | |
2892 | continue; | |
2893 | } | |
2894 | ||
2895 | /* This is a rename/rename(1to2) */ | |
2896 | clean_merge = handle_content_merge(opt, | |
2897 | pair->one->path, | |
2898 | &base->stages[0], | |
2899 | &side1->stages[1], | |
2900 | &side2->stages[2], | |
2901 | pathnames, | |
2902 | 1 + 2 * opt->priv->call_depth, | |
2903 | &merged); | |
2904 | if (clean_merge < 0) | |
2905 | return -1; | |
2906 | if (!clean_merge && | |
2907 | merged.mode == side1->stages[1].mode && | |
2908 | oideq(&merged.oid, &side1->stages[1].oid)) | |
2909 | was_binary_blob = 1; | |
2910 | memcpy(&side1->stages[1], &merged, sizeof(merged)); | |
2911 | if (was_binary_blob) { | |
2912 | /* | |
2913 | * Getting here means we were attempting to | |
2914 | * merge a binary blob. | |
2915 | * | |
2916 | * Since we can't merge binaries, | |
2917 | * handle_content_merge() just takes one | |
2918 | * side. But we don't want to copy the | |
2919 | * contents of one side to both paths. We | |
2920 | * used the contents of side1 above for | |
2921 | * side1->stages, let's use the contents of | |
2922 | * side2 for side2->stages below. | |
2923 | */ | |
2924 | oidcpy(&merged.oid, &side2->stages[2].oid); | |
2925 | merged.mode = side2->stages[2].mode; | |
2926 | } | |
2927 | memcpy(&side2->stages[2], &merged, sizeof(merged)); | |
2928 | ||
2929 | side1->path_conflict = 1; | |
2930 | side2->path_conflict = 1; | |
2931 | /* | |
2932 | * TODO: For renames we normally remove the path at the | |
2933 | * old name. It would thus seem consistent to do the | |
2934 | * same for rename/rename(1to2) cases, but we haven't | |
2935 | * done so traditionally and a number of the regression | |
2936 | * tests now encode an expectation that the file is | |
2937 | * left there at stage 1. If we ever decide to change | |
2938 | * this, add the following two lines here: | |
2939 | * base->merged.is_null = 1; | |
2940 | * base->merged.clean = 1; | |
2941 | * and remove the setting of base->path_conflict to 1. | |
2942 | */ | |
2943 | base->path_conflict = 1; | |
2944 | path_msg(opt, CONFLICT_RENAME_RENAME, 0, | |
2945 | pathnames[0], pathnames[1], pathnames[2], NULL, | |
2946 | _("CONFLICT (rename/rename): %s renamed to " | |
2947 | "%s in %s and to %s in %s."), | |
2948 | pathnames[0], | |
2949 | pathnames[1], opt->branch1, | |
2950 | pathnames[2], opt->branch2); | |
2951 | ||
2952 | i++; /* We handled both renames, i.e. i+1 handled */ | |
2953 | continue; | |
2954 | } | |
2955 | ||
2956 | VERIFY_CI(oldinfo); | |
2957 | VERIFY_CI(newinfo); | |
2958 | target_index = pair->score; /* from collect_renames() */ | |
2959 | assert(target_index == 1 || target_index == 2); | |
2960 | other_source_index = 3 - target_index; | |
2961 | old_sidemask = (1 << other_source_index); /* 2 or 4 */ | |
2962 | source_deleted = (oldinfo->filemask == 1); | |
2963 | collision = ((newinfo->filemask & old_sidemask) != 0); | |
2964 | type_changed = !source_deleted && | |
2965 | (S_ISREG(oldinfo->stages[other_source_index].mode) != | |
2966 | S_ISREG(newinfo->stages[target_index].mode)); | |
2967 | if (type_changed && collision) { | |
2968 | /* | |
2969 | * special handling so later blocks can handle this... | |
2970 | * | |
2971 | * if type_changed && collision are both true, then this | |
2972 | * was really a double rename, but one side wasn't | |
2973 | * detected due to lack of break detection. I.e. | |
2974 | * something like | |
2975 | * orig: has normal file 'foo' | |
2976 | * side1: renames 'foo' to 'bar', adds 'foo' symlink | |
2977 | * side2: renames 'foo' to 'bar' | |
2978 | * In this case, the foo->bar rename on side1 won't be | |
2979 | * detected because the new symlink named 'foo' is | |
2980 | * there and we don't do break detection. But we detect | |
2981 | * this here because we don't want to merge the content | |
2982 | * of the foo symlink with the foo->bar file, so we | |
2983 | * have some logic to handle this special case. The | |
2984 | * easiest way to do that is make 'bar' on side1 not | |
2985 | * be considered a colliding file but the other part | |
2986 | * of a normal rename. If the file is very different, | |
2987 | * well we're going to get content merge conflicts | |
2988 | * anyway so it doesn't hurt. And if the colliding | |
2989 | * file also has a different type, that'll be handled | |
2990 | * by the content merge logic in process_entry() too. | |
2991 | * | |
2992 | * See also t6430, 'rename vs. rename/symlink' | |
2993 | */ | |
2994 | collision = 0; | |
2995 | } | |
2996 | if (source_deleted) { | |
2997 | if (target_index == 1) { | |
2998 | rename_branch = opt->branch1; | |
2999 | delete_branch = opt->branch2; | |
3000 | } else { | |
3001 | rename_branch = opt->branch2; | |
3002 | delete_branch = opt->branch1; | |
3003 | } | |
3004 | } | |
3005 | ||
3006 | assert(source_deleted || oldinfo->filemask & old_sidemask); | |
3007 | ||
3008 | /* Need to check for special types of rename conflicts... */ | |
3009 | if (collision && !source_deleted) { | |
3010 | /* collision: rename/add or rename/rename(2to1) */ | |
3011 | const char *pathnames[3]; | |
3012 | struct version_info merged; | |
3013 | ||
3014 | struct conflict_info *base, *side1, *side2; | |
3015 | int clean; | |
3016 | ||
3017 | pathnames[0] = oldpath; | |
3018 | pathnames[other_source_index] = oldpath; | |
3019 | pathnames[target_index] = newpath; | |
3020 | ||
3021 | base = strmap_get(&opt->priv->paths, pathnames[0]); | |
3022 | side1 = strmap_get(&opt->priv->paths, pathnames[1]); | |
3023 | side2 = strmap_get(&opt->priv->paths, pathnames[2]); | |
3024 | ||
3025 | VERIFY_CI(base); | |
3026 | VERIFY_CI(side1); | |
3027 | VERIFY_CI(side2); | |
3028 | ||
3029 | clean = handle_content_merge(opt, pair->one->path, | |
3030 | &base->stages[0], | |
3031 | &side1->stages[1], | |
3032 | &side2->stages[2], | |
3033 | pathnames, | |
3034 | 1 + 2 * opt->priv->call_depth, | |
3035 | &merged); | |
3036 | if (clean < 0) | |
3037 | return -1; | |
3038 | ||
3039 | memcpy(&newinfo->stages[target_index], &merged, | |
3040 | sizeof(merged)); | |
3041 | if (!clean) { | |
3042 | path_msg(opt, CONFLICT_RENAME_COLLIDES, 0, | |
3043 | newpath, oldpath, NULL, NULL, | |
3044 | _("CONFLICT (rename involved in " | |
3045 | "collision): rename of %s -> %s has " | |
3046 | "content conflicts AND collides " | |
3047 | "with another path; this may result " | |
3048 | "in nested conflict markers."), | |
3049 | oldpath, newpath); | |
3050 | } | |
3051 | } else if (collision && source_deleted) { | |
3052 | /* | |
3053 | * rename/add/delete or rename/rename(2to1)/delete: | |
3054 | * since oldpath was deleted on the side that didn't | |
3055 | * do the rename, there's not much of a content merge | |
3056 | * we can do for the rename. oldinfo->merged.is_null | |
3057 | * was already set, so we just leave things as-is so | |
3058 | * they look like an add/add conflict. | |
3059 | */ | |
3060 | ||
3061 | newinfo->path_conflict = 1; | |
3062 | path_msg(opt, CONFLICT_RENAME_DELETE, 0, | |
3063 | newpath, oldpath, NULL, NULL, | |
3064 | _("CONFLICT (rename/delete): %s renamed " | |
3065 | "to %s in %s, but deleted in %s."), | |
3066 | oldpath, newpath, rename_branch, delete_branch); | |
3067 | } else { | |
3068 | /* | |
3069 | * a few different cases...start by copying the | |
3070 | * existing stage(s) from oldinfo over the newinfo | |
3071 | * and update the pathname(s). | |
3072 | */ | |
3073 | memcpy(&newinfo->stages[0], &oldinfo->stages[0], | |
3074 | sizeof(newinfo->stages[0])); | |
3075 | newinfo->filemask |= (1 << MERGE_BASE); | |
3076 | newinfo->pathnames[0] = oldpath; | |
3077 | if (type_changed) { | |
3078 | /* rename vs. typechange */ | |
3079 | /* Mark the original as resolved by removal */ | |
3080 | memcpy(&oldinfo->stages[0].oid, null_oid(), | |
3081 | sizeof(oldinfo->stages[0].oid)); | |
3082 | oldinfo->stages[0].mode = 0; | |
3083 | oldinfo->filemask &= 0x06; | |
3084 | } else if (source_deleted) { | |
3085 | /* rename/delete */ | |
3086 | newinfo->path_conflict = 1; | |
3087 | path_msg(opt, CONFLICT_RENAME_DELETE, 0, | |
3088 | newpath, oldpath, NULL, NULL, | |
3089 | _("CONFLICT (rename/delete): %s renamed" | |
3090 | " to %s in %s, but deleted in %s."), | |
3091 | oldpath, newpath, | |
3092 | rename_branch, delete_branch); | |
3093 | } else { | |
3094 | /* normal rename */ | |
3095 | memcpy(&newinfo->stages[other_source_index], | |
3096 | &oldinfo->stages[other_source_index], | |
3097 | sizeof(newinfo->stages[0])); | |
3098 | newinfo->filemask |= (1 << other_source_index); | |
3099 | newinfo->pathnames[other_source_index] = oldpath; | |
3100 | } | |
3101 | } | |
3102 | ||
3103 | if (!type_changed) { | |
3104 | /* Mark the original as resolved by removal */ | |
3105 | oldinfo->merged.is_null = 1; | |
3106 | oldinfo->merged.clean = 1; | |
3107 | } | |
3108 | ||
3109 | } | |
3110 | ||
3111 | return clean_merge; | |
3112 | } | |
3113 | ||
3114 | static inline int possible_side_renames(struct rename_info *renames, | |
3115 | unsigned side_index) | |
3116 | { | |
3117 | return renames->pairs[side_index].nr > 0 && | |
3118 | !strintmap_empty(&renames->relevant_sources[side_index]); | |
3119 | } | |
3120 | ||
3121 | static inline int possible_renames(struct rename_info *renames) | |
3122 | { | |
3123 | return possible_side_renames(renames, 1) || | |
3124 | possible_side_renames(renames, 2) || | |
3125 | !strmap_empty(&renames->cached_pairs[1]) || | |
3126 | !strmap_empty(&renames->cached_pairs[2]); | |
3127 | } | |
3128 | ||
3129 | static void resolve_diffpair_statuses(struct diff_queue_struct *q) | |
3130 | { | |
3131 | /* | |
3132 | * A simplified version of diff_resolve_rename_copy(); would probably | |
3133 | * just use that function but it's static... | |
3134 | */ | |
3135 | int i; | |
3136 | struct diff_filepair *p; | |
3137 | ||
3138 | for (i = 0; i < q->nr; ++i) { | |
3139 | p = q->queue[i]; | |
3140 | p->status = 0; /* undecided */ | |
3141 | if (!DIFF_FILE_VALID(p->one)) | |
3142 | p->status = DIFF_STATUS_ADDED; | |
3143 | else if (!DIFF_FILE_VALID(p->two)) | |
3144 | p->status = DIFF_STATUS_DELETED; | |
3145 | else if (DIFF_PAIR_RENAME(p)) | |
3146 | p->status = DIFF_STATUS_RENAMED; | |
3147 | } | |
3148 | } | |
3149 | ||
3150 | static void prune_cached_from_relevant(struct rename_info *renames, | |
3151 | unsigned side) | |
3152 | { | |
3153 | /* Reason for this function described in add_pair() */ | |
3154 | struct hashmap_iter iter; | |
3155 | struct strmap_entry *entry; | |
3156 | ||
3157 | /* Remove from relevant_sources all entries in cached_pairs[side] */ | |
3158 | strmap_for_each_entry(&renames->cached_pairs[side], &iter, entry) { | |
3159 | strintmap_remove(&renames->relevant_sources[side], | |
3160 | entry->key); | |
3161 | } | |
3162 | /* Remove from relevant_sources all entries in cached_irrelevant[side] */ | |
3163 | strset_for_each_entry(&renames->cached_irrelevant[side], &iter, entry) { | |
3164 | strintmap_remove(&renames->relevant_sources[side], | |
3165 | entry->key); | |
3166 | } | |
3167 | } | |
3168 | ||
3169 | static void use_cached_pairs(struct merge_options *opt, | |
3170 | struct strmap *cached_pairs, | |
3171 | struct diff_queue_struct *pairs) | |
3172 | { | |
3173 | struct hashmap_iter iter; | |
3174 | struct strmap_entry *entry; | |
3175 | ||
3176 | /* | |
3177 | * Add to side_pairs all entries from renames->cached_pairs[side_index]. | |
3178 | * (Info in cached_irrelevant[side_index] is not relevant here.) | |
3179 | */ | |
3180 | strmap_for_each_entry(cached_pairs, &iter, entry) { | |
3181 | struct diff_filespec *one, *two; | |
3182 | const char *old_name = entry->key; | |
3183 | const char *new_name = entry->value; | |
3184 | if (!new_name) | |
3185 | new_name = old_name; | |
3186 | ||
3187 | /* | |
3188 | * cached_pairs has *copies* of old_name and new_name, | |
3189 | * because it has to persist across merges. Since | |
3190 | * pool_alloc_filespec() will just re-use the existing | |
3191 | * filenames, which will also get re-used by | |
3192 | * opt->priv->paths if they become renames, and then | |
3193 | * get freed at the end of the merge, that would leave | |
3194 | * the copy in cached_pairs dangling. Avoid this by | |
3195 | * making a copy here. | |
3196 | */ | |
3197 | old_name = mem_pool_strdup(&opt->priv->pool, old_name); | |
3198 | new_name = mem_pool_strdup(&opt->priv->pool, new_name); | |
3199 | ||
3200 | /* We don't care about oid/mode, only filenames and status */ | |
3201 | one = pool_alloc_filespec(&opt->priv->pool, old_name); | |
3202 | two = pool_alloc_filespec(&opt->priv->pool, new_name); | |
3203 | pool_diff_queue(&opt->priv->pool, pairs, one, two); | |
3204 | pairs->queue[pairs->nr-1]->status = entry->value ? 'R' : 'D'; | |
3205 | } | |
3206 | } | |
3207 | ||
3208 | static void cache_new_pair(struct rename_info *renames, | |
3209 | int side, | |
3210 | char *old_path, | |
3211 | char *new_path, | |
3212 | int free_old_value) | |
3213 | { | |
3214 | char *old_value; | |
3215 | new_path = xstrdup(new_path); | |
3216 | old_value = strmap_put(&renames->cached_pairs[side], | |
3217 | old_path, new_path); | |
3218 | strset_add(&renames->cached_target_names[side], new_path); | |
3219 | if (free_old_value) | |
3220 | free(old_value); | |
3221 | else | |
3222 | assert(!old_value); | |
3223 | } | |
3224 | ||
3225 | static void possibly_cache_new_pair(struct rename_info *renames, | |
3226 | struct diff_filepair *p, | |
3227 | unsigned side, | |
3228 | char *new_path) | |
3229 | { | |
3230 | int dir_renamed_side = 0; | |
3231 | ||
3232 | if (new_path) { | |
3233 | /* | |
3234 | * Directory renames happen on the other side of history from | |
3235 | * the side that adds new files to the old directory. | |
3236 | */ | |
3237 | dir_renamed_side = 3 - side; | |
3238 | } else { | |
3239 | int val = strintmap_get(&renames->relevant_sources[side], | |
3240 | p->one->path); | |
3241 | if (val == RELEVANT_NO_MORE) { | |
3242 | assert(p->status == 'D'); | |
3243 | strset_add(&renames->cached_irrelevant[side], | |
3244 | p->one->path); | |
3245 | } | |
3246 | if (val <= 0) | |
3247 | return; | |
3248 | } | |
3249 | ||
3250 | if (p->status == 'D') { | |
3251 | /* | |
3252 | * If we already had this delete, we'll just set it's value | |
3253 | * to NULL again, so no harm. | |
3254 | */ | |
3255 | strmap_put(&renames->cached_pairs[side], p->one->path, NULL); | |
3256 | } else if (p->status == 'R') { | |
3257 | if (!new_path) | |
3258 | new_path = p->two->path; | |
3259 | else | |
3260 | cache_new_pair(renames, dir_renamed_side, | |
3261 | p->two->path, new_path, 0); | |
3262 | cache_new_pair(renames, side, p->one->path, new_path, 1); | |
3263 | } else if (p->status == 'A' && new_path) { | |
3264 | cache_new_pair(renames, dir_renamed_side, | |
3265 | p->two->path, new_path, 0); | |
3266 | } | |
3267 | } | |
3268 | ||
3269 | static int compare_pairs(const void *a_, const void *b_) | |
3270 | { | |
3271 | const struct diff_filepair *a = *((const struct diff_filepair **)a_); | |
3272 | const struct diff_filepair *b = *((const struct diff_filepair **)b_); | |
3273 | ||
3274 | return strcmp(a->one->path, b->one->path); | |
3275 | } | |
3276 | ||
3277 | /* Call diffcore_rename() to update deleted/added pairs into rename pairs */ | |
3278 | static int detect_regular_renames(struct merge_options *opt, | |
3279 | unsigned side_index) | |
3280 | { | |
3281 | struct diff_options diff_opts; | |
3282 | struct rename_info *renames = &opt->priv->renames; | |
3283 | ||
3284 | prune_cached_from_relevant(renames, side_index); | |
3285 | if (!possible_side_renames(renames, side_index)) { | |
3286 | /* | |
3287 | * No rename detection needed for this side, but we still need | |
3288 | * to make sure 'adds' are marked correctly in case the other | |
3289 | * side had directory renames. | |
3290 | */ | |
3291 | resolve_diffpair_statuses(&renames->pairs[side_index]); | |
3292 | return 0; | |
3293 | } | |
3294 | ||
3295 | partial_clear_dir_rename_count(&renames->dir_rename_count[side_index]); | |
3296 | repo_diff_setup(opt->repo, &diff_opts); | |
3297 | diff_opts.flags.recursive = 1; | |
3298 | diff_opts.flags.rename_empty = 0; | |
3299 | diff_opts.detect_rename = DIFF_DETECT_RENAME; | |
3300 | diff_opts.rename_limit = opt->rename_limit; | |
3301 | if (opt->rename_limit <= 0) | |
3302 | diff_opts.rename_limit = 7000; | |
3303 | diff_opts.rename_score = opt->rename_score; | |
3304 | diff_opts.show_rename_progress = opt->show_rename_progress; | |
3305 | diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT; | |
3306 | diff_setup_done(&diff_opts); | |
3307 | ||
3308 | diff_queued_diff = renames->pairs[side_index]; | |
3309 | trace2_region_enter("diff", "diffcore_rename", opt->repo); | |
3310 | diffcore_rename_extended(&diff_opts, | |
3311 | &opt->priv->pool, | |
3312 | &renames->relevant_sources[side_index], | |
3313 | &renames->dirs_removed[side_index], | |
3314 | &renames->dir_rename_count[side_index], | |
3315 | &renames->cached_pairs[side_index]); | |
3316 | trace2_region_leave("diff", "diffcore_rename", opt->repo); | |
3317 | resolve_diffpair_statuses(&diff_queued_diff); | |
3318 | ||
3319 | if (diff_opts.needed_rename_limit > 0) | |
3320 | renames->redo_after_renames = 0; | |
3321 | if (diff_opts.needed_rename_limit > renames->needed_limit) | |
3322 | renames->needed_limit = diff_opts.needed_rename_limit; | |
3323 | ||
3324 | renames->pairs[side_index] = diff_queued_diff; | |
3325 | ||
3326 | diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT; | |
3327 | diff_queued_diff.nr = 0; | |
3328 | diff_queued_diff.queue = NULL; | |
3329 | diff_flush(&diff_opts); | |
3330 | ||
3331 | return 1; | |
3332 | } | |
3333 | ||
3334 | /* | |
3335 | * Get information of all renames which occurred in 'side_pairs', making use | |
3336 | * of any implicit directory renames in side_dir_renames (also making use of | |
3337 | * implicit directory renames rename_exclusions as needed by | |
3338 | * check_for_directory_rename()). Add all (updated) renames into result. | |
3339 | */ | |
3340 | static int collect_renames(struct merge_options *opt, | |
3341 | struct diff_queue_struct *result, | |
3342 | unsigned side_index, | |
3343 | struct strmap *collisions, | |
3344 | struct strmap *dir_renames_for_side, | |
3345 | struct strmap *rename_exclusions) | |
3346 | { | |
3347 | int i, clean = 1; | |
3348 | struct diff_queue_struct *side_pairs; | |
3349 | struct rename_info *renames = &opt->priv->renames; | |
3350 | ||
3351 | side_pairs = &renames->pairs[side_index]; | |
3352 | ||
3353 | for (i = 0; i < side_pairs->nr; ++i) { | |
3354 | struct diff_filepair *p = side_pairs->queue[i]; | |
3355 | char *new_path; /* non-NULL only with directory renames */ | |
3356 | ||
3357 | if (p->status != 'A' && p->status != 'R') { | |
3358 | possibly_cache_new_pair(renames, p, side_index, NULL); | |
3359 | pool_diff_free_filepair(&opt->priv->pool, p); | |
3360 | continue; | |
3361 | } | |
3362 | ||
3363 | new_path = check_for_directory_rename(opt, p->two->path, | |
3364 | side_index, | |
3365 | dir_renames_for_side, | |
3366 | rename_exclusions, | |
3367 | collisions, | |
3368 | &clean); | |
3369 | ||
3370 | possibly_cache_new_pair(renames, p, side_index, new_path); | |
3371 | if (p->status != 'R' && !new_path) { | |
3372 | pool_diff_free_filepair(&opt->priv->pool, p); | |
3373 | continue; | |
3374 | } | |
3375 | ||
3376 | if (new_path) | |
3377 | apply_directory_rename_modifications(opt, p, new_path); | |
3378 | ||
3379 | /* | |
3380 | * p->score comes back from diffcore_rename_extended() with | |
3381 | * the similarity of the renamed file. The similarity is | |
3382 | * was used to determine that the two files were related | |
3383 | * and are a rename, which we have already used, but beyond | |
3384 | * that we have no use for the similarity. So p->score is | |
3385 | * now irrelevant. However, process_renames() will need to | |
3386 | * know which side of the merge this rename was associated | |
3387 | * with, so overwrite p->score with that value. | |
3388 | */ | |
3389 | p->score = side_index; | |
3390 | result->queue[result->nr++] = p; | |
3391 | } | |
3392 | ||
3393 | return clean; | |
3394 | } | |
3395 | ||
3396 | static int detect_and_process_renames(struct merge_options *opt) | |
3397 | { | |
3398 | struct diff_queue_struct combined = { 0 }; | |
3399 | struct rename_info *renames = &opt->priv->renames; | |
3400 | struct strmap collisions[3]; | |
3401 | int need_dir_renames, s, i, clean = 1; | |
3402 | unsigned detection_run = 0; | |
3403 | ||
3404 | if (!possible_renames(renames)) | |
3405 | goto cleanup; | |
3406 | ||
3407 | trace2_region_enter("merge", "regular renames", opt->repo); | |
3408 | detection_run |= detect_regular_renames(opt, MERGE_SIDE1); | |
3409 | detection_run |= detect_regular_renames(opt, MERGE_SIDE2); | |
3410 | if (renames->needed_limit) { | |
3411 | renames->cached_pairs_valid_side = 0; | |
3412 | renames->redo_after_renames = 0; | |
3413 | } | |
3414 | if (renames->redo_after_renames && detection_run) { | |
3415 | int i, side; | |
3416 | struct diff_filepair *p; | |
3417 | ||
3418 | /* Cache the renames, we found */ | |
3419 | for (side = MERGE_SIDE1; side <= MERGE_SIDE2; side++) { | |
3420 | for (i = 0; i < renames->pairs[side].nr; ++i) { | |
3421 | p = renames->pairs[side].queue[i]; | |
3422 | possibly_cache_new_pair(renames, p, side, NULL); | |
3423 | } | |
3424 | } | |
3425 | ||
3426 | /* Restart the merge with the cached renames */ | |
3427 | renames->redo_after_renames = 2; | |
3428 | trace2_region_leave("merge", "regular renames", opt->repo); | |
3429 | goto cleanup; | |
3430 | } | |
3431 | use_cached_pairs(opt, &renames->cached_pairs[1], &renames->pairs[1]); | |
3432 | use_cached_pairs(opt, &renames->cached_pairs[2], &renames->pairs[2]); | |
3433 | trace2_region_leave("merge", "regular renames", opt->repo); | |
3434 | ||
3435 | trace2_region_enter("merge", "directory renames", opt->repo); | |
3436 | need_dir_renames = | |
3437 | !opt->priv->call_depth && | |
3438 | (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE || | |
3439 | opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT); | |
3440 | ||
3441 | if (need_dir_renames) { | |
3442 | get_provisional_directory_renames(opt, MERGE_SIDE1, &clean); | |
3443 | get_provisional_directory_renames(opt, MERGE_SIDE2, &clean); | |
3444 | handle_directory_level_conflicts(opt); | |
3445 | } | |
3446 | ||
3447 | ALLOC_GROW(combined.queue, | |
3448 | renames->pairs[1].nr + renames->pairs[2].nr, | |
3449 | combined.alloc); | |
3450 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) { | |
3451 | int other_side = 3 - i; | |
3452 | compute_collisions(&collisions[i], | |
3453 | &renames->dir_renames[other_side], | |
3454 | &renames->pairs[i]); | |
3455 | } | |
3456 | clean &= collect_renames(opt, &combined, MERGE_SIDE1, | |
3457 | collisions, | |
3458 | &renames->dir_renames[2], | |
3459 | &renames->dir_renames[1]); | |
3460 | clean &= collect_renames(opt, &combined, MERGE_SIDE2, | |
3461 | collisions, | |
3462 | &renames->dir_renames[1], | |
3463 | &renames->dir_renames[2]); | |
3464 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) | |
3465 | free_collisions(&collisions[i]); | |
3466 | STABLE_QSORT(combined.queue, combined.nr, compare_pairs); | |
3467 | trace2_region_leave("merge", "directory renames", opt->repo); | |
3468 | ||
3469 | trace2_region_enter("merge", "process renames", opt->repo); | |
3470 | clean &= process_renames(opt, &combined); | |
3471 | trace2_region_leave("merge", "process renames", opt->repo); | |
3472 | ||
3473 | goto simple_cleanup; /* collect_renames() handles some of cleanup */ | |
3474 | ||
3475 | cleanup: | |
3476 | /* | |
3477 | * Free now unneeded filepairs, which would have been handled | |
3478 | * in collect_renames() normally but we skipped that code. | |
3479 | */ | |
3480 | for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) { | |
3481 | struct diff_queue_struct *side_pairs; | |
3482 | int i; | |
3483 | ||
3484 | side_pairs = &renames->pairs[s]; | |
3485 | for (i = 0; i < side_pairs->nr; ++i) { | |
3486 | struct diff_filepair *p = side_pairs->queue[i]; | |
3487 | pool_diff_free_filepair(&opt->priv->pool, p); | |
3488 | } | |
3489 | } | |
3490 | ||
3491 | simple_cleanup: | |
3492 | /* Free memory for renames->pairs[] and combined */ | |
3493 | for (s = MERGE_SIDE1; s <= MERGE_SIDE2; s++) { | |
3494 | free(renames->pairs[s].queue); | |
3495 | DIFF_QUEUE_CLEAR(&renames->pairs[s]); | |
3496 | } | |
3497 | for (i = 0; i < combined.nr; i++) | |
3498 | pool_diff_free_filepair(&opt->priv->pool, combined.queue[i]); | |
3499 | free(combined.queue); | |
3500 | ||
3501 | return clean; | |
3502 | } | |
3503 | ||
3504 | /*** Function Grouping: functions related to process_entries() ***/ | |
3505 | ||
3506 | static int sort_dirs_next_to_their_children(const char *one, const char *two) | |
3507 | { | |
3508 | unsigned char c1, c2; | |
3509 | ||
3510 | /* | |
3511 | * Here we only care that entries for directories appear adjacent | |
3512 | * to and before files underneath the directory. We can achieve | |
3513 | * that by pretending to add a trailing slash to every file and | |
3514 | * then sorting. In other words, we do not want the natural | |
3515 | * sorting of | |
3516 | * foo | |
3517 | * foo.txt | |
3518 | * foo/bar | |
3519 | * Instead, we want "foo" to sort as though it were "foo/", so that | |
3520 | * we instead get | |
3521 | * foo.txt | |
3522 | * foo | |
3523 | * foo/bar | |
3524 | * To achieve this, we basically implement our own strcmp, except that | |
3525 | * if we get to the end of either string instead of comparing NUL to | |
3526 | * another character, we compare '/' to it. | |
3527 | * | |
3528 | * If this unusual "sort as though '/' were appended" perplexes | |
3529 | * you, perhaps it will help to note that this is not the final | |
3530 | * sort. write_tree() will sort again without the trailing slash | |
3531 | * magic, but just on paths immediately under a given tree. | |
3532 | * | |
3533 | * The reason to not use df_name_compare directly was that it was | |
3534 | * just too expensive (we don't have the string lengths handy), so | |
3535 | * it was reimplemented. | |
3536 | */ | |
3537 | ||
3538 | /* | |
3539 | * NOTE: This function will never be called with two equal strings, | |
3540 | * because it is used to sort the keys of a strmap, and strmaps have | |
3541 | * unique keys by construction. That simplifies our c1==c2 handling | |
3542 | * below. | |
3543 | */ | |
3544 | ||
3545 | while (*one && (*one == *two)) { | |
3546 | one++; | |
3547 | two++; | |
3548 | } | |
3549 | ||
3550 | c1 = *one ? *one : '/'; | |
3551 | c2 = *two ? *two : '/'; | |
3552 | ||
3553 | if (c1 == c2) { | |
3554 | /* Getting here means one is a leading directory of the other */ | |
3555 | return (*one) ? 1 : -1; | |
3556 | } else | |
3557 | return c1 - c2; | |
3558 | } | |
3559 | ||
3560 | static int read_oid_strbuf(const struct object_id *oid, | |
3561 | struct strbuf *dst) | |
3562 | { | |
3563 | void *buf; | |
3564 | enum object_type type; | |
3565 | unsigned long size; | |
3566 | buf = repo_read_object_file(the_repository, oid, &type, &size); | |
3567 | if (!buf) | |
3568 | return error(_("cannot read object %s"), oid_to_hex(oid)); | |
3569 | if (type != OBJ_BLOB) { | |
3570 | free(buf); | |
3571 | return error(_("object %s is not a blob"), oid_to_hex(oid)); | |
3572 | } | |
3573 | strbuf_attach(dst, buf, size, size + 1); | |
3574 | return 0; | |
3575 | } | |
3576 | ||
3577 | static int blob_unchanged(struct merge_options *opt, | |
3578 | const struct version_info *base, | |
3579 | const struct version_info *side, | |
3580 | const char *path) | |
3581 | { | |
3582 | struct strbuf basebuf = STRBUF_INIT; | |
3583 | struct strbuf sidebuf = STRBUF_INIT; | |
3584 | int ret = 0; /* assume changed for safety */ | |
3585 | struct index_state *idx = &opt->priv->attr_index; | |
3586 | ||
3587 | if (!idx->initialized) | |
3588 | initialize_attr_index(opt); | |
3589 | ||
3590 | if (base->mode != side->mode) | |
3591 | return 0; | |
3592 | if (oideq(&base->oid, &side->oid)) | |
3593 | return 1; | |
3594 | ||
3595 | if (read_oid_strbuf(&base->oid, &basebuf) || | |
3596 | read_oid_strbuf(&side->oid, &sidebuf)) | |
3597 | goto error_return; | |
3598 | /* | |
3599 | * Note: binary | is used so that both renormalizations are | |
3600 | * performed. Comparison can be skipped if both files are | |
3601 | * unchanged since their sha1s have already been compared. | |
3602 | */ | |
3603 | if (renormalize_buffer(idx, path, basebuf.buf, basebuf.len, &basebuf) | | |
3604 | renormalize_buffer(idx, path, sidebuf.buf, sidebuf.len, &sidebuf)) | |
3605 | ret = (basebuf.len == sidebuf.len && | |
3606 | !memcmp(basebuf.buf, sidebuf.buf, basebuf.len)); | |
3607 | ||
3608 | error_return: | |
3609 | strbuf_release(&basebuf); | |
3610 | strbuf_release(&sidebuf); | |
3611 | return ret; | |
3612 | } | |
3613 | ||
3614 | struct directory_versions { | |
3615 | /* | |
3616 | * versions: list of (basename -> version_info) | |
3617 | * | |
3618 | * The basenames are in reverse lexicographic order of full pathnames, | |
3619 | * as processed in process_entries(). This puts all entries within | |
3620 | * a directory together, and covers the directory itself after | |
3621 | * everything within it, allowing us to write subtrees before needing | |
3622 | * to record information for the tree itself. | |
3623 | */ | |
3624 | struct string_list versions; | |
3625 | ||
3626 | /* | |
3627 | * offsets: list of (full relative path directories -> integer offsets) | |
3628 | * | |
3629 | * Since versions contains basenames from files in multiple different | |
3630 | * directories, we need to know which entries in versions correspond | |
3631 | * to which directories. Values of e.g. | |
3632 | * "" 0 | |
3633 | * src 2 | |
3634 | * src/moduleA 5 | |
3635 | * Would mean that entries 0-1 of versions are files in the toplevel | |
3636 | * directory, entries 2-4 are files under src/, and the remaining | |
3637 | * entries starting at index 5 are files under src/moduleA/. | |
3638 | */ | |
3639 | struct string_list offsets; | |
3640 | ||
3641 | /* | |
3642 | * last_directory: directory that previously processed file found in | |
3643 | * | |
3644 | * last_directory starts NULL, but records the directory in which the | |
3645 | * previous file was found within. As soon as | |
3646 | * directory(current_file) != last_directory | |
3647 | * then we need to start updating accounting in versions & offsets. | |
3648 | * Note that last_directory is always the last path in "offsets" (or | |
3649 | * NULL if "offsets" is empty) so this exists just for quick access. | |
3650 | */ | |
3651 | const char *last_directory; | |
3652 | ||
3653 | /* last_directory_len: cached computation of strlen(last_directory) */ | |
3654 | unsigned last_directory_len; | |
3655 | }; | |
3656 | ||
3657 | static int tree_entry_order(const void *a_, const void *b_) | |
3658 | { | |
3659 | const struct string_list_item *a = a_; | |
3660 | const struct string_list_item *b = b_; | |
3661 | ||
3662 | const struct merged_info *ami = a->util; | |
3663 | const struct merged_info *bmi = b->util; | |
3664 | return base_name_compare(a->string, strlen(a->string), ami->result.mode, | |
3665 | b->string, strlen(b->string), bmi->result.mode); | |
3666 | } | |
3667 | ||
3668 | static int write_tree(struct object_id *result_oid, | |
3669 | struct string_list *versions, | |
3670 | unsigned int offset, | |
3671 | size_t hash_size) | |
3672 | { | |
3673 | size_t maxlen = 0, extra; | |
3674 | unsigned int nr; | |
3675 | struct strbuf buf = STRBUF_INIT; | |
3676 | int i, ret = 0; | |
3677 | ||
3678 | assert(offset <= versions->nr); | |
3679 | nr = versions->nr - offset; | |
3680 | if (versions->nr) | |
3681 | /* No need for STABLE_QSORT -- filenames must be unique */ | |
3682 | QSORT(versions->items + offset, nr, tree_entry_order); | |
3683 | ||
3684 | /* Pre-allocate some space in buf */ | |
3685 | extra = hash_size + 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */ | |
3686 | for (i = 0; i < nr; i++) { | |
3687 | maxlen += strlen(versions->items[offset+i].string) + extra; | |
3688 | } | |
3689 | strbuf_grow(&buf, maxlen); | |
3690 | ||
3691 | /* Write each entry out to buf */ | |
3692 | for (i = 0; i < nr; i++) { | |
3693 | struct merged_info *mi = versions->items[offset+i].util; | |
3694 | struct version_info *ri = &mi->result; | |
3695 | strbuf_addf(&buf, "%o %s%c", | |
3696 | ri->mode, | |
3697 | versions->items[offset+i].string, '\0'); | |
3698 | strbuf_add(&buf, ri->oid.hash, hash_size); | |
3699 | } | |
3700 | ||
3701 | /* Write this object file out, and record in result_oid */ | |
3702 | if (write_object_file(buf.buf, buf.len, OBJ_TREE, result_oid)) | |
3703 | ret = -1; | |
3704 | strbuf_release(&buf); | |
3705 | return ret; | |
3706 | } | |
3707 | ||
3708 | static void record_entry_for_tree(struct directory_versions *dir_metadata, | |
3709 | const char *path, | |
3710 | struct merged_info *mi) | |
3711 | { | |
3712 | const char *basename; | |
3713 | ||
3714 | if (mi->is_null) | |
3715 | /* nothing to record */ | |
3716 | return; | |
3717 | ||
3718 | basename = path + mi->basename_offset; | |
3719 | assert(strchr(basename, '/') == NULL); | |
3720 | string_list_append(&dir_metadata->versions, | |
3721 | basename)->util = &mi->result; | |
3722 | } | |
3723 | ||
3724 | static int write_completed_directory(struct merge_options *opt, | |
3725 | const char *new_directory_name, | |
3726 | struct directory_versions *info) | |
3727 | { | |
3728 | const char *prev_dir; | |
3729 | struct merged_info *dir_info = NULL; | |
3730 | unsigned int offset, ret = 0; | |
3731 | ||
3732 | /* | |
3733 | * Some explanation of info->versions and info->offsets... | |
3734 | * | |
3735 | * process_entries() iterates over all relevant files AND | |
3736 | * directories in reverse lexicographic order, and calls this | |
3737 | * function. Thus, an example of the paths that process_entries() | |
3738 | * could operate on (along with the directories for those paths | |
3739 | * being shown) is: | |
3740 | * | |
3741 | * xtract.c "" | |
3742 | * tokens.txt "" | |
3743 | * src/moduleB/umm.c src/moduleB | |
3744 | * src/moduleB/stuff.h src/moduleB | |
3745 | * src/moduleB/baz.c src/moduleB | |
3746 | * src/moduleB src | |
3747 | * src/moduleA/foo.c src/moduleA | |
3748 | * src/moduleA/bar.c src/moduleA | |
3749 | * src/moduleA src | |
3750 | * src "" | |
3751 | * Makefile "" | |
3752 | * | |
3753 | * info->versions: | |
3754 | * | |
3755 | * always contains the unprocessed entries and their | |
3756 | * version_info information. For example, after the first five | |
3757 | * entries above, info->versions would be: | |
3758 | * | |
3759 | * xtract.c <xtract.c's version_info> | |
3760 | * token.txt <token.txt's version_info> | |
3761 | * umm.c <src/moduleB/umm.c's version_info> | |
3762 | * stuff.h <src/moduleB/stuff.h's version_info> | |
3763 | * baz.c <src/moduleB/baz.c's version_info> | |
3764 | * | |
3765 | * Once a subdirectory is completed we remove the entries in | |
3766 | * that subdirectory from info->versions, writing it as a tree | |
3767 | * (write_tree()). Thus, as soon as we get to src/moduleB, | |
3768 | * info->versions would be updated to | |
3769 | * | |
3770 | * xtract.c <xtract.c's version_info> | |
3771 | * token.txt <token.txt's version_info> | |
3772 | * moduleB <src/moduleB's version_info> | |
3773 | * | |
3774 | * info->offsets: | |
3775 | * | |
3776 | * helps us track which entries in info->versions correspond to | |
3777 | * which directories. When we are N directories deep (e.g. 4 | |
3778 | * for src/modA/submod/subdir/), we have up to N+1 unprocessed | |
3779 | * directories (+1 because of toplevel dir). Corresponding to | |
3780 | * the info->versions example above, after processing five entries | |
3781 | * info->offsets will be: | |
3782 | * | |
3783 | * "" 0 | |
3784 | * src/moduleB 2 | |
3785 | * | |
3786 | * which is used to know that xtract.c & token.txt are from the | |
3787 | * toplevel dirctory, while umm.c & stuff.h & baz.c are from the | |
3788 | * src/moduleB directory. Again, following the example above, | |
3789 | * once we need to process src/moduleB, then info->offsets is | |
3790 | * updated to | |
3791 | * | |
3792 | * "" 0 | |
3793 | * src 2 | |
3794 | * | |
3795 | * which says that moduleB (and only moduleB so far) is in the | |
3796 | * src directory. | |
3797 | * | |
3798 | * One unique thing to note about info->offsets here is that | |
3799 | * "src" was not added to info->offsets until there was a path | |
3800 | * (a file OR directory) immediately below src/ that got | |
3801 | * processed. | |
3802 | * | |
3803 | * Since process_entry() just appends new entries to info->versions, | |
3804 | * write_completed_directory() only needs to do work if the next path | |
3805 | * is in a directory that is different than the last directory found | |
3806 | * in info->offsets. | |
3807 | */ | |
3808 | ||
3809 | /* | |
3810 | * If we are working with the same directory as the last entry, there | |
3811 | * is no work to do. (See comments above the directory_name member of | |
3812 | * struct merged_info for why we can use pointer comparison instead of | |
3813 | * strcmp here.) | |
3814 | */ | |
3815 | if (new_directory_name == info->last_directory) | |
3816 | return 0; | |
3817 | ||
3818 | /* | |
3819 | * If we are just starting (last_directory is NULL), or last_directory | |
3820 | * is a prefix of the current directory, then we can just update | |
3821 | * info->offsets to record the offset where we started this directory | |
3822 | * and update last_directory to have quick access to it. | |
3823 | */ | |
3824 | if (info->last_directory == NULL || | |
3825 | !strncmp(new_directory_name, info->last_directory, | |
3826 | info->last_directory_len)) { | |
3827 | uintptr_t offset = info->versions.nr; | |
3828 | ||
3829 | info->last_directory = new_directory_name; | |
3830 | info->last_directory_len = strlen(info->last_directory); | |
3831 | /* | |
3832 | * Record the offset into info->versions where we will | |
3833 | * start recording basenames of paths found within | |
3834 | * new_directory_name. | |
3835 | */ | |
3836 | string_list_append(&info->offsets, | |
3837 | info->last_directory)->util = (void*)offset; | |
3838 | return 0; | |
3839 | } | |
3840 | ||
3841 | /* | |
3842 | * The next entry that will be processed will be within | |
3843 | * new_directory_name. Since at this point we know that | |
3844 | * new_directory_name is within a different directory than | |
3845 | * info->last_directory, we have all entries for info->last_directory | |
3846 | * in info->versions and we need to create a tree object for them. | |
3847 | */ | |
3848 | dir_info = strmap_get(&opt->priv->paths, info->last_directory); | |
3849 | assert(dir_info); | |
3850 | offset = (uintptr_t)info->offsets.items[info->offsets.nr-1].util; | |
3851 | if (offset == info->versions.nr) { | |
3852 | /* | |
3853 | * Actually, we don't need to create a tree object in this | |
3854 | * case. Whenever all files within a directory disappear | |
3855 | * during the merge (e.g. unmodified on one side and | |
3856 | * deleted on the other, or files were renamed elsewhere), | |
3857 | * then we get here and the directory itself needs to be | |
3858 | * omitted from its parent tree as well. | |
3859 | */ | |
3860 | dir_info->is_null = 1; | |
3861 | } else { | |
3862 | /* | |
3863 | * Write out the tree to the git object directory, and also | |
3864 | * record the mode and oid in dir_info->result. | |
3865 | */ | |
3866 | dir_info->is_null = 0; | |
3867 | dir_info->result.mode = S_IFDIR; | |
3868 | if (write_tree(&dir_info->result.oid, &info->versions, offset, | |
3869 | opt->repo->hash_algo->rawsz) < 0) | |
3870 | ret = -1; | |
3871 | } | |
3872 | ||
3873 | /* | |
3874 | * We've now used several entries from info->versions and one entry | |
3875 | * from info->offsets, so we get rid of those values. | |
3876 | */ | |
3877 | info->offsets.nr--; | |
3878 | info->versions.nr = offset; | |
3879 | ||
3880 | /* | |
3881 | * Now we've taken care of the completed directory, but we need to | |
3882 | * prepare things since future entries will be in | |
3883 | * new_directory_name. (In particular, process_entry() will be | |
3884 | * appending new entries to info->versions.) So, we need to make | |
3885 | * sure new_directory_name is the last entry in info->offsets. | |
3886 | */ | |
3887 | prev_dir = info->offsets.nr == 0 ? NULL : | |
3888 | info->offsets.items[info->offsets.nr-1].string; | |
3889 | if (new_directory_name != prev_dir) { | |
3890 | uintptr_t c = info->versions.nr; | |
3891 | string_list_append(&info->offsets, | |
3892 | new_directory_name)->util = (void*)c; | |
3893 | } | |
3894 | ||
3895 | /* And, of course, we need to update last_directory to match. */ | |
3896 | info->last_directory = new_directory_name; | |
3897 | info->last_directory_len = strlen(info->last_directory); | |
3898 | ||
3899 | return ret; | |
3900 | } | |
3901 | ||
3902 | /* Per entry merge function */ | |
3903 | static int process_entry(struct merge_options *opt, | |
3904 | const char *path, | |
3905 | struct conflict_info *ci, | |
3906 | struct directory_versions *dir_metadata) | |
3907 | { | |
3908 | int df_file_index = 0; | |
3909 | ||
3910 | VERIFY_CI(ci); | |
3911 | assert(ci->filemask >= 0 && ci->filemask <= 7); | |
3912 | /* ci->match_mask == 7 was handled in collect_merge_info_callback() */ | |
3913 | assert(ci->match_mask == 0 || ci->match_mask == 3 || | |
3914 | ci->match_mask == 5 || ci->match_mask == 6); | |
3915 | ||
3916 | if (ci->dirmask) { | |
3917 | record_entry_for_tree(dir_metadata, path, &ci->merged); | |
3918 | if (ci->filemask == 0) | |
3919 | /* nothing else to handle */ | |
3920 | return 0; | |
3921 | assert(ci->df_conflict); | |
3922 | } | |
3923 | ||
3924 | if (ci->df_conflict && ci->merged.result.mode == 0) { | |
3925 | int i; | |
3926 | ||
3927 | /* | |
3928 | * directory no longer in the way, but we do have a file we | |
3929 | * need to place here so we need to clean away the "directory | |
3930 | * merges to nothing" result. | |
3931 | */ | |
3932 | ci->df_conflict = 0; | |
3933 | assert(ci->filemask != 0); | |
3934 | ci->merged.clean = 0; | |
3935 | ci->merged.is_null = 0; | |
3936 | /* and we want to zero out any directory-related entries */ | |
3937 | ci->match_mask = (ci->match_mask & ~ci->dirmask); | |
3938 | ci->dirmask = 0; | |
3939 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
3940 | if (ci->filemask & (1 << i)) | |
3941 | continue; | |
3942 | ci->stages[i].mode = 0; | |
3943 | oidcpy(&ci->stages[i].oid, null_oid()); | |
3944 | } | |
3945 | } else if (ci->df_conflict && ci->merged.result.mode != 0) { | |
3946 | /* | |
3947 | * This started out as a D/F conflict, and the entries in | |
3948 | * the competing directory were not removed by the merge as | |
3949 | * evidenced by write_completed_directory() writing a value | |
3950 | * to ci->merged.result.mode. | |
3951 | */ | |
3952 | struct conflict_info *new_ci; | |
3953 | const char *branch; | |
3954 | const char *old_path = path; | |
3955 | int i; | |
3956 | ||
3957 | assert(ci->merged.result.mode == S_IFDIR); | |
3958 | ||
3959 | /* | |
3960 | * If filemask is 1, we can just ignore the file as having | |
3961 | * been deleted on both sides. We do not want to overwrite | |
3962 | * ci->merged.result, since it stores the tree for all the | |
3963 | * files under it. | |
3964 | */ | |
3965 | if (ci->filemask == 1) { | |
3966 | ci->filemask = 0; | |
3967 | return 0; | |
3968 | } | |
3969 | ||
3970 | /* | |
3971 | * This file still exists on at least one side, and we want | |
3972 | * the directory to remain here, so we need to move this | |
3973 | * path to some new location. | |
3974 | */ | |
3975 | new_ci = mem_pool_calloc(&opt->priv->pool, 1, sizeof(*new_ci)); | |
3976 | ||
3977 | /* We don't really want new_ci->merged.result copied, but it'll | |
3978 | * be overwritten below so it doesn't matter. We also don't | |
3979 | * want any directory mode/oid values copied, but we'll zero | |
3980 | * those out immediately. We do want the rest of ci copied. | |
3981 | */ | |
3982 | memcpy(new_ci, ci, sizeof(*ci)); | |
3983 | new_ci->match_mask = (new_ci->match_mask & ~new_ci->dirmask); | |
3984 | new_ci->dirmask = 0; | |
3985 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
3986 | if (new_ci->filemask & (1 << i)) | |
3987 | continue; | |
3988 | /* zero out any entries related to directories */ | |
3989 | new_ci->stages[i].mode = 0; | |
3990 | oidcpy(&new_ci->stages[i].oid, null_oid()); | |
3991 | } | |
3992 | ||
3993 | /* | |
3994 | * Find out which side this file came from; note that we | |
3995 | * cannot just use ci->filemask, because renames could cause | |
3996 | * the filemask to go back to 7. So we use dirmask, then | |
3997 | * pick the opposite side's index. | |
3998 | */ | |
3999 | df_file_index = (ci->dirmask & (1 << 1)) ? 2 : 1; | |
4000 | branch = (df_file_index == 1) ? opt->branch1 : opt->branch2; | |
4001 | path = unique_path(opt, path, branch); | |
4002 | strmap_put(&opt->priv->paths, path, new_ci); | |
4003 | ||
4004 | path_msg(opt, CONFLICT_FILE_DIRECTORY, 0, | |
4005 | path, old_path, NULL, NULL, | |
4006 | _("CONFLICT (file/directory): directory in the way " | |
4007 | "of %s from %s; moving it to %s instead."), | |
4008 | old_path, branch, path); | |
4009 | ||
4010 | /* | |
4011 | * Zero out the filemask for the old ci. At this point, ci | |
4012 | * was just an entry for a directory, so we don't need to | |
4013 | * do anything more with it. | |
4014 | */ | |
4015 | ci->filemask = 0; | |
4016 | ||
4017 | /* | |
4018 | * Now note that we're working on the new entry (path was | |
4019 | * updated above. | |
4020 | */ | |
4021 | ci = new_ci; | |
4022 | } | |
4023 | ||
4024 | /* | |
4025 | * NOTE: Below there is a long switch-like if-elseif-elseif... block | |
4026 | * which the code goes through even for the df_conflict cases | |
4027 | * above. | |
4028 | */ | |
4029 | if (ci->match_mask) { | |
4030 | ci->merged.clean = !ci->df_conflict && !ci->path_conflict; | |
4031 | if (ci->match_mask == 6) { | |
4032 | /* stages[1] == stages[2] */ | |
4033 | ci->merged.result.mode = ci->stages[1].mode; | |
4034 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
4035 | } else { | |
4036 | /* determine the mask of the side that didn't match */ | |
4037 | unsigned int othermask = 7 & ~ci->match_mask; | |
4038 | int side = (othermask == 4) ? 2 : 1; | |
4039 | ||
4040 | ci->merged.result.mode = ci->stages[side].mode; | |
4041 | ci->merged.is_null = !ci->merged.result.mode; | |
4042 | if (ci->merged.is_null) | |
4043 | ci->merged.clean = 1; | |
4044 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
4045 | ||
4046 | assert(othermask == 2 || othermask == 4); | |
4047 | assert(ci->merged.is_null == | |
4048 | (ci->filemask == ci->match_mask)); | |
4049 | } | |
4050 | } else if (ci->filemask >= 6 && | |
4051 | (S_IFMT & ci->stages[1].mode) != | |
4052 | (S_IFMT & ci->stages[2].mode)) { | |
4053 | /* Two different items from (file/submodule/symlink) */ | |
4054 | if (opt->priv->call_depth) { | |
4055 | /* Just use the version from the merge base */ | |
4056 | ci->merged.clean = 0; | |
4057 | oidcpy(&ci->merged.result.oid, &ci->stages[0].oid); | |
4058 | ci->merged.result.mode = ci->stages[0].mode; | |
4059 | ci->merged.is_null = (ci->merged.result.mode == 0); | |
4060 | } else { | |
4061 | /* Handle by renaming one or both to separate paths. */ | |
4062 | unsigned o_mode = ci->stages[0].mode; | |
4063 | unsigned a_mode = ci->stages[1].mode; | |
4064 | unsigned b_mode = ci->stages[2].mode; | |
4065 | struct conflict_info *new_ci; | |
4066 | const char *a_path = NULL, *b_path = NULL; | |
4067 | int rename_a = 0, rename_b = 0; | |
4068 | ||
4069 | new_ci = mem_pool_alloc(&opt->priv->pool, | |
4070 | sizeof(*new_ci)); | |
4071 | ||
4072 | if (S_ISREG(a_mode)) | |
4073 | rename_a = 1; | |
4074 | else if (S_ISREG(b_mode)) | |
4075 | rename_b = 1; | |
4076 | else { | |
4077 | rename_a = 1; | |
4078 | rename_b = 1; | |
4079 | } | |
4080 | ||
4081 | if (rename_a) | |
4082 | a_path = unique_path(opt, path, opt->branch1); | |
4083 | if (rename_b) | |
4084 | b_path = unique_path(opt, path, opt->branch2); | |
4085 | ||
4086 | if (rename_a && rename_b) { | |
4087 | path_msg(opt, CONFLICT_DISTINCT_MODES, 0, | |
4088 | path, a_path, b_path, NULL, | |
4089 | _("CONFLICT (distinct types): %s had " | |
4090 | "different types on each side; " | |
4091 | "renamed both of them so each can " | |
4092 | "be recorded somewhere."), | |
4093 | path); | |
4094 | } else { | |
4095 | path_msg(opt, CONFLICT_DISTINCT_MODES, 0, | |
4096 | path, rename_a ? a_path : b_path, | |
4097 | NULL, NULL, | |
4098 | _("CONFLICT (distinct types): %s had " | |
4099 | "different types on each side; " | |
4100 | "renamed one of them so each can be " | |
4101 | "recorded somewhere."), | |
4102 | path); | |
4103 | } | |
4104 | ||
4105 | ci->merged.clean = 0; | |
4106 | memcpy(new_ci, ci, sizeof(*new_ci)); | |
4107 | ||
4108 | /* Put b into new_ci, removing a from stages */ | |
4109 | new_ci->merged.result.mode = ci->stages[2].mode; | |
4110 | oidcpy(&new_ci->merged.result.oid, &ci->stages[2].oid); | |
4111 | new_ci->stages[1].mode = 0; | |
4112 | oidcpy(&new_ci->stages[1].oid, null_oid()); | |
4113 | new_ci->filemask = 5; | |
4114 | if ((S_IFMT & b_mode) != (S_IFMT & o_mode)) { | |
4115 | new_ci->stages[0].mode = 0; | |
4116 | oidcpy(&new_ci->stages[0].oid, null_oid()); | |
4117 | new_ci->filemask = 4; | |
4118 | } | |
4119 | ||
4120 | /* Leave only a in ci, fixing stages. */ | |
4121 | ci->merged.result.mode = ci->stages[1].mode; | |
4122 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
4123 | ci->stages[2].mode = 0; | |
4124 | oidcpy(&ci->stages[2].oid, null_oid()); | |
4125 | ci->filemask = 3; | |
4126 | if ((S_IFMT & a_mode) != (S_IFMT & o_mode)) { | |
4127 | ci->stages[0].mode = 0; | |
4128 | oidcpy(&ci->stages[0].oid, null_oid()); | |
4129 | ci->filemask = 2; | |
4130 | } | |
4131 | ||
4132 | /* Insert entries into opt->priv_paths */ | |
4133 | assert(rename_a || rename_b); | |
4134 | if (rename_a) | |
4135 | strmap_put(&opt->priv->paths, a_path, ci); | |
4136 | ||
4137 | if (!rename_b) | |
4138 | b_path = path; | |
4139 | strmap_put(&opt->priv->paths, b_path, new_ci); | |
4140 | ||
4141 | if (rename_a && rename_b) | |
4142 | strmap_remove(&opt->priv->paths, path, 0); | |
4143 | ||
4144 | /* | |
4145 | * Do special handling for b_path since process_entry() | |
4146 | * won't be called on it specially. | |
4147 | */ | |
4148 | strmap_put(&opt->priv->conflicted, b_path, new_ci); | |
4149 | record_entry_for_tree(dir_metadata, b_path, | |
4150 | &new_ci->merged); | |
4151 | ||
4152 | /* | |
4153 | * Remaining code for processing this entry should | |
4154 | * think in terms of processing a_path. | |
4155 | */ | |
4156 | if (a_path) | |
4157 | path = a_path; | |
4158 | } | |
4159 | } else if (ci->filemask >= 6) { | |
4160 | /* Need a two-way or three-way content merge */ | |
4161 | struct version_info merged_file; | |
4162 | int clean_merge; | |
4163 | struct version_info *o = &ci->stages[0]; | |
4164 | struct version_info *a = &ci->stages[1]; | |
4165 | struct version_info *b = &ci->stages[2]; | |
4166 | ||
4167 | clean_merge = handle_content_merge(opt, path, o, a, b, | |
4168 | ci->pathnames, | |
4169 | opt->priv->call_depth * 2, | |
4170 | &merged_file); | |
4171 | if (clean_merge < 0) | |
4172 | return -1; | |
4173 | ci->merged.clean = clean_merge && | |
4174 | !ci->df_conflict && !ci->path_conflict; | |
4175 | ci->merged.result.mode = merged_file.mode; | |
4176 | ci->merged.is_null = (merged_file.mode == 0); | |
4177 | oidcpy(&ci->merged.result.oid, &merged_file.oid); | |
4178 | if (clean_merge && ci->df_conflict) { | |
4179 | assert(df_file_index == 1 || df_file_index == 2); | |
4180 | ci->filemask = 1 << df_file_index; | |
4181 | ci->stages[df_file_index].mode = merged_file.mode; | |
4182 | oidcpy(&ci->stages[df_file_index].oid, &merged_file.oid); | |
4183 | } | |
4184 | if (!clean_merge) { | |
4185 | const char *reason = _("content"); | |
4186 | if (ci->filemask == 6) | |
4187 | reason = _("add/add"); | |
4188 | if (S_ISGITLINK(merged_file.mode)) | |
4189 | reason = _("submodule"); | |
4190 | path_msg(opt, CONFLICT_CONTENTS, 0, | |
4191 | path, NULL, NULL, NULL, | |
4192 | _("CONFLICT (%s): Merge conflict in %s"), | |
4193 | reason, path); | |
4194 | } | |
4195 | } else if (ci->filemask == 3 || ci->filemask == 5) { | |
4196 | /* Modify/delete */ | |
4197 | const char *modify_branch, *delete_branch; | |
4198 | int side = (ci->filemask == 5) ? 2 : 1; | |
4199 | int index = opt->priv->call_depth ? 0 : side; | |
4200 | ||
4201 | ci->merged.result.mode = ci->stages[index].mode; | |
4202 | oidcpy(&ci->merged.result.oid, &ci->stages[index].oid); | |
4203 | ci->merged.clean = 0; | |
4204 | ||
4205 | modify_branch = (side == 1) ? opt->branch1 : opt->branch2; | |
4206 | delete_branch = (side == 1) ? opt->branch2 : opt->branch1; | |
4207 | ||
4208 | if (opt->renormalize && | |
4209 | blob_unchanged(opt, &ci->stages[0], &ci->stages[side], | |
4210 | path)) { | |
4211 | if (!ci->path_conflict) { | |
4212 | /* | |
4213 | * Blob unchanged after renormalization, so | |
4214 | * there's no modify/delete conflict after all; | |
4215 | * we can just remove the file. | |
4216 | */ | |
4217 | ci->merged.is_null = 1; | |
4218 | ci->merged.clean = 1; | |
4219 | /* | |
4220 | * file goes away => even if there was a | |
4221 | * directory/file conflict there isn't one now. | |
4222 | */ | |
4223 | ci->df_conflict = 0; | |
4224 | } else { | |
4225 | /* rename/delete, so conflict remains */ | |
4226 | } | |
4227 | } else if (ci->path_conflict && | |
4228 | oideq(&ci->stages[0].oid, &ci->stages[side].oid)) { | |
4229 | /* | |
4230 | * This came from a rename/delete; no action to take, | |
4231 | * but avoid printing "modify/delete" conflict notice | |
4232 | * since the contents were not modified. | |
4233 | */ | |
4234 | } else { | |
4235 | path_msg(opt, CONFLICT_MODIFY_DELETE, 0, | |
4236 | path, NULL, NULL, NULL, | |
4237 | _("CONFLICT (modify/delete): %s deleted in %s " | |
4238 | "and modified in %s. Version %s of %s left " | |
4239 | "in tree."), | |
4240 | path, delete_branch, modify_branch, | |
4241 | modify_branch, path); | |
4242 | } | |
4243 | } else if (ci->filemask == 2 || ci->filemask == 4) { | |
4244 | /* Added on one side */ | |
4245 | int side = (ci->filemask == 4) ? 2 : 1; | |
4246 | ci->merged.result.mode = ci->stages[side].mode; | |
4247 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
4248 | ci->merged.clean = !ci->df_conflict && !ci->path_conflict; | |
4249 | } else if (ci->filemask == 1) { | |
4250 | /* Deleted on both sides */ | |
4251 | ci->merged.is_null = 1; | |
4252 | ci->merged.result.mode = 0; | |
4253 | oidcpy(&ci->merged.result.oid, null_oid()); | |
4254 | assert(!ci->df_conflict); | |
4255 | ci->merged.clean = !ci->path_conflict; | |
4256 | } | |
4257 | ||
4258 | /* | |
4259 | * If still conflicted, record it separately. This allows us to later | |
4260 | * iterate over just conflicted entries when updating the index instead | |
4261 | * of iterating over all entries. | |
4262 | */ | |
4263 | if (!ci->merged.clean) | |
4264 | strmap_put(&opt->priv->conflicted, path, ci); | |
4265 | ||
4266 | /* Record metadata for ci->merged in dir_metadata */ | |
4267 | record_entry_for_tree(dir_metadata, path, &ci->merged); | |
4268 | return 0; | |
4269 | } | |
4270 | ||
4271 | static void prefetch_for_content_merges(struct merge_options *opt, | |
4272 | struct string_list *plist) | |
4273 | { | |
4274 | struct string_list_item *e; | |
4275 | struct oid_array to_fetch = OID_ARRAY_INIT; | |
4276 | ||
4277 | if (opt->repo != the_repository || !repo_has_promisor_remote(the_repository)) | |
4278 | return; | |
4279 | ||
4280 | for (e = &plist->items[plist->nr-1]; e >= plist->items; --e) { | |
4281 | /* char *path = e->string; */ | |
4282 | struct conflict_info *ci = e->util; | |
4283 | int i; | |
4284 | ||
4285 | /* Ignore clean entries */ | |
4286 | if (ci->merged.clean) | |
4287 | continue; | |
4288 | ||
4289 | /* Ignore entries that don't need a content merge */ | |
4290 | if (ci->match_mask || ci->filemask < 6 || | |
4291 | !S_ISREG(ci->stages[1].mode) || | |
4292 | !S_ISREG(ci->stages[2].mode) || | |
4293 | oideq(&ci->stages[1].oid, &ci->stages[2].oid)) | |
4294 | continue; | |
4295 | ||
4296 | /* Also don't need content merge if base matches either side */ | |
4297 | if (ci->filemask == 7 && | |
4298 | S_ISREG(ci->stages[0].mode) && | |
4299 | (oideq(&ci->stages[0].oid, &ci->stages[1].oid) || | |
4300 | oideq(&ci->stages[0].oid, &ci->stages[2].oid))) | |
4301 | continue; | |
4302 | ||
4303 | for (i = 0; i < 3; i++) { | |
4304 | unsigned side_mask = (1 << i); | |
4305 | struct version_info *vi = &ci->stages[i]; | |
4306 | ||
4307 | if ((ci->filemask & side_mask) && | |
4308 | S_ISREG(vi->mode) && | |
4309 | oid_object_info_extended(opt->repo, &vi->oid, NULL, | |
4310 | OBJECT_INFO_FOR_PREFETCH)) | |
4311 | oid_array_append(&to_fetch, &vi->oid); | |
4312 | } | |
4313 | } | |
4314 | ||
4315 | promisor_remote_get_direct(opt->repo, to_fetch.oid, to_fetch.nr); | |
4316 | oid_array_clear(&to_fetch); | |
4317 | } | |
4318 | ||
4319 | static int process_entries(struct merge_options *opt, | |
4320 | struct object_id *result_oid) | |
4321 | { | |
4322 | struct hashmap_iter iter; | |
4323 | struct strmap_entry *e; | |
4324 | struct string_list plist = STRING_LIST_INIT_NODUP; | |
4325 | struct string_list_item *entry; | |
4326 | struct directory_versions dir_metadata = { STRING_LIST_INIT_NODUP, | |
4327 | STRING_LIST_INIT_NODUP, | |
4328 | NULL, 0 }; | |
4329 | int ret = 0; | |
4330 | ||
4331 | trace2_region_enter("merge", "process_entries setup", opt->repo); | |
4332 | if (strmap_empty(&opt->priv->paths)) { | |
4333 | oidcpy(result_oid, opt->repo->hash_algo->empty_tree); | |
4334 | return 0; | |
4335 | } | |
4336 | ||
4337 | /* Hack to pre-allocate plist to the desired size */ | |
4338 | trace2_region_enter("merge", "plist grow", opt->repo); | |
4339 | ALLOC_GROW(plist.items, strmap_get_size(&opt->priv->paths), plist.alloc); | |
4340 | trace2_region_leave("merge", "plist grow", opt->repo); | |
4341 | ||
4342 | /* Put every entry from paths into plist, then sort */ | |
4343 | trace2_region_enter("merge", "plist copy", opt->repo); | |
4344 | strmap_for_each_entry(&opt->priv->paths, &iter, e) { | |
4345 | string_list_append(&plist, e->key)->util = e->value; | |
4346 | } | |
4347 | trace2_region_leave("merge", "plist copy", opt->repo); | |
4348 | ||
4349 | trace2_region_enter("merge", "plist special sort", opt->repo); | |
4350 | plist.cmp = sort_dirs_next_to_their_children; | |
4351 | string_list_sort(&plist); | |
4352 | trace2_region_leave("merge", "plist special sort", opt->repo); | |
4353 | ||
4354 | trace2_region_leave("merge", "process_entries setup", opt->repo); | |
4355 | ||
4356 | /* | |
4357 | * Iterate over the items in reverse order, so we can handle paths | |
4358 | * below a directory before needing to handle the directory itself. | |
4359 | * | |
4360 | * This allows us to write subtrees before we need to write trees, | |
4361 | * and it also enables sane handling of directory/file conflicts | |
4362 | * (because it allows us to know whether the directory is still in | |
4363 | * the way when it is time to process the file at the same path). | |
4364 | */ | |
4365 | trace2_region_enter("merge", "processing", opt->repo); | |
4366 | prefetch_for_content_merges(opt, &plist); | |
4367 | for (entry = &plist.items[plist.nr-1]; entry >= plist.items; --entry) { | |
4368 | char *path = entry->string; | |
4369 | /* | |
4370 | * NOTE: mi may actually be a pointer to a conflict_info, but | |
4371 | * we have to check mi->clean first to see if it's safe to | |
4372 | * reassign to such a pointer type. | |
4373 | */ | |
4374 | struct merged_info *mi = entry->util; | |
4375 | ||
4376 | if (write_completed_directory(opt, mi->directory_name, | |
4377 | &dir_metadata) < 0) { | |
4378 | ret = -1; | |
4379 | goto cleanup; | |
4380 | } | |
4381 | if (mi->clean) | |
4382 | record_entry_for_tree(&dir_metadata, path, mi); | |
4383 | else { | |
4384 | struct conflict_info *ci = (struct conflict_info *)mi; | |
4385 | if (process_entry(opt, path, ci, &dir_metadata) < 0) { | |
4386 | ret = -1; | |
4387 | goto cleanup; | |
4388 | }; | |
4389 | } | |
4390 | } | |
4391 | trace2_region_leave("merge", "processing", opt->repo); | |
4392 | ||
4393 | trace2_region_enter("merge", "process_entries cleanup", opt->repo); | |
4394 | if (dir_metadata.offsets.nr != 1 || | |
4395 | (uintptr_t)dir_metadata.offsets.items[0].util != 0) { | |
4396 | printf("dir_metadata.offsets.nr = %"PRIuMAX" (should be 1)\n", | |
4397 | (uintmax_t)dir_metadata.offsets.nr); | |
4398 | printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n", | |
4399 | (unsigned)(uintptr_t)dir_metadata.offsets.items[0].util); | |
4400 | fflush(stdout); | |
4401 | BUG("dir_metadata accounting completely off; shouldn't happen"); | |
4402 | } | |
4403 | if (write_tree(result_oid, &dir_metadata.versions, 0, | |
4404 | opt->repo->hash_algo->rawsz) < 0) | |
4405 | ret = -1; | |
4406 | cleanup: | |
4407 | string_list_clear(&plist, 0); | |
4408 | string_list_clear(&dir_metadata.versions, 0); | |
4409 | string_list_clear(&dir_metadata.offsets, 0); | |
4410 | trace2_region_leave("merge", "process_entries cleanup", opt->repo); | |
4411 | ||
4412 | return ret; | |
4413 | } | |
4414 | ||
4415 | /*** Function Grouping: functions related to merge_switch_to_result() ***/ | |
4416 | ||
4417 | static int checkout(struct merge_options *opt, | |
4418 | struct tree *prev, | |
4419 | struct tree *next) | |
4420 | { | |
4421 | /* Switch the index/working copy from old to new */ | |
4422 | int ret; | |
4423 | struct tree_desc trees[2]; | |
4424 | struct unpack_trees_options unpack_opts; | |
4425 | ||
4426 | memset(&unpack_opts, 0, sizeof(unpack_opts)); | |
4427 | unpack_opts.head_idx = -1; | |
4428 | unpack_opts.src_index = opt->repo->index; | |
4429 | unpack_opts.dst_index = opt->repo->index; | |
4430 | ||
4431 | setup_unpack_trees_porcelain(&unpack_opts, "merge"); | |
4432 | ||
4433 | /* | |
4434 | * NOTE: if this were just "git checkout" code, we would probably | |
4435 | * read or refresh the cache and check for a conflicted index, but | |
4436 | * builtin/merge.c or sequencer.c really needs to read the index | |
4437 | * and check for conflicted entries before starting merging for a | |
4438 | * good user experience (no sense waiting for merges/rebases before | |
4439 | * erroring out), so there's no reason to duplicate that work here. | |
4440 | */ | |
4441 | ||
4442 | /* 2-way merge to the new branch */ | |
4443 | unpack_opts.update = 1; | |
4444 | unpack_opts.merge = 1; | |
4445 | unpack_opts.quiet = 0; /* FIXME: sequencer might want quiet? */ | |
4446 | unpack_opts.verbose_update = (opt->verbosity > 2); | |
4447 | unpack_opts.fn = twoway_merge; | |
4448 | unpack_opts.preserve_ignored = 0; /* FIXME: !opts->overwrite_ignore */ | |
4449 | if (parse_tree(prev) < 0) | |
4450 | return -1; | |
4451 | init_tree_desc(&trees[0], &prev->object.oid, prev->buffer, prev->size); | |
4452 | if (parse_tree(next) < 0) | |
4453 | return -1; | |
4454 | init_tree_desc(&trees[1], &next->object.oid, next->buffer, next->size); | |
4455 | ||
4456 | ret = unpack_trees(2, trees, &unpack_opts); | |
4457 | clear_unpack_trees_porcelain(&unpack_opts); | |
4458 | return ret; | |
4459 | } | |
4460 | ||
4461 | static int record_conflicted_index_entries(struct merge_options *opt) | |
4462 | { | |
4463 | struct hashmap_iter iter; | |
4464 | struct strmap_entry *e; | |
4465 | struct index_state *index = opt->repo->index; | |
4466 | struct checkout state = CHECKOUT_INIT; | |
4467 | int errs = 0; | |
4468 | int original_cache_nr; | |
4469 | ||
4470 | if (strmap_empty(&opt->priv->conflicted)) | |
4471 | return 0; | |
4472 | ||
4473 | /* | |
4474 | * We are in a conflicted state. These conflicts might be inside | |
4475 | * sparse-directory entries, so check if any entries are outside | |
4476 | * of the sparse-checkout cone preemptively. | |
4477 | * | |
4478 | * We set original_cache_nr below, but that might change if | |
4479 | * index_name_pos() calls ask for paths within sparse directories. | |
4480 | */ | |
4481 | strmap_for_each_entry(&opt->priv->conflicted, &iter, e) { | |
4482 | if (!path_in_sparse_checkout(e->key, index)) { | |
4483 | ensure_full_index(index); | |
4484 | break; | |
4485 | } | |
4486 | } | |
4487 | ||
4488 | /* If any entries have skip_worktree set, we'll have to check 'em out */ | |
4489 | state.force = 1; | |
4490 | state.quiet = 1; | |
4491 | state.refresh_cache = 1; | |
4492 | state.istate = index; | |
4493 | original_cache_nr = index->cache_nr; | |
4494 | ||
4495 | /* Append every entry from conflicted into index, then sort */ | |
4496 | strmap_for_each_entry(&opt->priv->conflicted, &iter, e) { | |
4497 | const char *path = e->key; | |
4498 | struct conflict_info *ci = e->value; | |
4499 | int pos; | |
4500 | struct cache_entry *ce; | |
4501 | int i; | |
4502 | ||
4503 | VERIFY_CI(ci); | |
4504 | ||
4505 | /* | |
4506 | * The index will already have a stage=0 entry for this path, | |
4507 | * because we created an as-merged-as-possible version of the | |
4508 | * file and checkout() moved the working copy and index over | |
4509 | * to that version. | |
4510 | * | |
4511 | * However, previous iterations through this loop will have | |
4512 | * added unstaged entries to the end of the cache which | |
4513 | * ignore the standard alphabetical ordering of cache | |
4514 | * entries and break invariants needed for index_name_pos() | |
4515 | * to work. However, we know the entry we want is before | |
4516 | * those appended cache entries, so do a temporary swap on | |
4517 | * cache_nr to only look through entries of interest. | |
4518 | */ | |
4519 | SWAP(index->cache_nr, original_cache_nr); | |
4520 | pos = index_name_pos(index, path, strlen(path)); | |
4521 | SWAP(index->cache_nr, original_cache_nr); | |
4522 | if (pos < 0) { | |
4523 | if (ci->filemask != 1) | |
4524 | BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path); | |
4525 | cache_tree_invalidate_path(index, path); | |
4526 | } else { | |
4527 | ce = index->cache[pos]; | |
4528 | ||
4529 | /* | |
4530 | * Clean paths with CE_SKIP_WORKTREE set will not be | |
4531 | * written to the working tree by the unpack_trees() | |
4532 | * call in checkout(). Our conflicted entries would | |
4533 | * have appeared clean to that code since we ignored | |
4534 | * the higher order stages. Thus, we need override | |
4535 | * the CE_SKIP_WORKTREE bit and manually write those | |
4536 | * files to the working disk here. | |
4537 | */ | |
4538 | if (ce_skip_worktree(ce)) | |
4539 | errs |= checkout_entry(ce, &state, NULL, NULL); | |
4540 | ||
4541 | /* | |
4542 | * Mark this cache entry for removal and instead add | |
4543 | * new stage>0 entries corresponding to the | |
4544 | * conflicts. If there are many conflicted entries, we | |
4545 | * want to avoid memmove'ing O(NM) entries by | |
4546 | * inserting the new entries one at a time. So, | |
4547 | * instead, we just add the new cache entries to the | |
4548 | * end (ignoring normal index requirements on sort | |
4549 | * order) and sort the index once we're all done. | |
4550 | */ | |
4551 | ce->ce_flags |= CE_REMOVE; | |
4552 | } | |
4553 | ||
4554 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
4555 | struct version_info *vi; | |
4556 | if (!(ci->filemask & (1ul << i))) | |
4557 | continue; | |
4558 | vi = &ci->stages[i]; | |
4559 | ce = make_cache_entry(index, vi->mode, &vi->oid, | |
4560 | path, i+1, 0); | |
4561 | add_index_entry(index, ce, ADD_CACHE_JUST_APPEND); | |
4562 | } | |
4563 | } | |
4564 | ||
4565 | /* | |
4566 | * Remove the unused cache entries (and invalidate the relevant | |
4567 | * cache-trees), then sort the index entries to get the conflicted | |
4568 | * entries we added to the end into their right locations. | |
4569 | */ | |
4570 | remove_marked_cache_entries(index, 1); | |
4571 | /* | |
4572 | * No need for STABLE_QSORT -- cmp_cache_name_compare sorts primarily | |
4573 | * on filename and secondarily on stage, and (name, stage #) are a | |
4574 | * unique tuple. | |
4575 | */ | |
4576 | QSORT(index->cache, index->cache_nr, cmp_cache_name_compare); | |
4577 | ||
4578 | return errs; | |
4579 | } | |
4580 | ||
4581 | static void print_submodule_conflict_suggestion(struct string_list *csub) { | |
4582 | struct string_list_item *item; | |
4583 | struct strbuf msg = STRBUF_INIT; | |
4584 | struct strbuf tmp = STRBUF_INIT; | |
4585 | struct strbuf subs = STRBUF_INIT; | |
4586 | ||
4587 | if (!csub->nr) | |
4588 | return; | |
4589 | ||
4590 | strbuf_add_separated_string_list(&subs, " ", csub); | |
4591 | for_each_string_list_item(item, csub) { | |
4592 | struct conflicted_submodule_item *util = item->util; | |
4593 | ||
4594 | /* | |
4595 | * NEEDSWORK: The steps to resolve these errors deserve a more | |
4596 | * detailed explanation than what is currently printed below. | |
4597 | */ | |
4598 | if (util->flag == CONFLICT_SUBMODULE_NOT_INITIALIZED || | |
4599 | util->flag == CONFLICT_SUBMODULE_HISTORY_NOT_AVAILABLE) | |
4600 | continue; | |
4601 | ||
4602 | /* | |
4603 | * TRANSLATORS: This is a line of advice to resolve a merge | |
4604 | * conflict in a submodule. The first argument is the submodule | |
4605 | * name, and the second argument is the abbreviated id of the | |
4606 | * commit that needs to be merged. For example: | |
4607 | * - go to submodule (mysubmodule), and either merge commit abc1234" | |
4608 | */ | |
4609 | strbuf_addf(&tmp, _(" - go to submodule (%s), and either merge commit %s\n" | |
4610 | " or update to an existing commit which has merged those changes\n"), | |
4611 | item->string, util->abbrev); | |
4612 | } | |
4613 | ||
4614 | /* | |
4615 | * TRANSLATORS: This is a detailed message for resolving submodule | |
4616 | * conflicts. The first argument is string containing one step per | |
4617 | * submodule. The second is a space-separated list of submodule names. | |
4618 | */ | |
4619 | strbuf_addf(&msg, | |
4620 | _("Recursive merging with submodules currently only supports trivial cases.\n" | |
4621 | "Please manually handle the merging of each conflicted submodule.\n" | |
4622 | "This can be accomplished with the following steps:\n" | |
4623 | "%s" | |
4624 | " - come back to superproject and run:\n\n" | |
4625 | " git add %s\n\n" | |
4626 | " to record the above merge or update\n" | |
4627 | " - resolve any other conflicts in the superproject\n" | |
4628 | " - commit the resulting index in the superproject\n"), | |
4629 | tmp.buf, subs.buf); | |
4630 | ||
4631 | advise_if_enabled(ADVICE_SUBMODULE_MERGE_CONFLICT, "%s", msg.buf); | |
4632 | ||
4633 | strbuf_release(&subs); | |
4634 | strbuf_release(&tmp); | |
4635 | strbuf_release(&msg); | |
4636 | } | |
4637 | ||
4638 | void merge_display_update_messages(struct merge_options *opt, | |
4639 | int detailed, | |
4640 | struct merge_result *result) | |
4641 | { | |
4642 | struct merge_options_internal *opti = result->priv; | |
4643 | struct hashmap_iter iter; | |
4644 | struct strmap_entry *e; | |
4645 | struct string_list olist = STRING_LIST_INIT_NODUP; | |
4646 | ||
4647 | if (opt->record_conflict_msgs_as_headers) | |
4648 | BUG("Either display conflict messages or record them as headers, not both"); | |
4649 | ||
4650 | trace2_region_enter("merge", "display messages", opt->repo); | |
4651 | ||
4652 | /* Hack to pre-allocate olist to the desired size */ | |
4653 | ALLOC_GROW(olist.items, strmap_get_size(&opti->conflicts), | |
4654 | olist.alloc); | |
4655 | ||
4656 | /* Put every entry from output into olist, then sort */ | |
4657 | strmap_for_each_entry(&opti->conflicts, &iter, e) { | |
4658 | string_list_append(&olist, e->key)->util = e->value; | |
4659 | } | |
4660 | string_list_sort(&olist); | |
4661 | ||
4662 | /* Iterate over the items, printing them */ | |
4663 | for (int path_nr = 0; path_nr < olist.nr; ++path_nr) { | |
4664 | struct string_list *conflicts = olist.items[path_nr].util; | |
4665 | for (int i = 0; i < conflicts->nr; i++) { | |
4666 | struct logical_conflict_info *info = | |
4667 | conflicts->items[i].util; | |
4668 | ||
4669 | if (detailed) { | |
4670 | printf("%lu", (unsigned long)info->paths.nr); | |
4671 | putchar('\0'); | |
4672 | for (int n = 0; n < info->paths.nr; n++) { | |
4673 | fputs(info->paths.v[n], stdout); | |
4674 | putchar('\0'); | |
4675 | } | |
4676 | fputs(type_short_descriptions[info->type], | |
4677 | stdout); | |
4678 | putchar('\0'); | |
4679 | } | |
4680 | puts(conflicts->items[i].string); | |
4681 | if (detailed) | |
4682 | putchar('\0'); | |
4683 | } | |
4684 | } | |
4685 | string_list_clear(&olist, 0); | |
4686 | ||
4687 | print_submodule_conflict_suggestion(&opti->conflicted_submodules); | |
4688 | ||
4689 | /* Also include needed rename limit adjustment now */ | |
4690 | diff_warn_rename_limit("merge.renamelimit", | |
4691 | opti->renames.needed_limit, 0); | |
4692 | ||
4693 | trace2_region_leave("merge", "display messages", opt->repo); | |
4694 | } | |
4695 | ||
4696 | void merge_get_conflicted_files(struct merge_result *result, | |
4697 | struct string_list *conflicted_files) | |
4698 | { | |
4699 | struct hashmap_iter iter; | |
4700 | struct strmap_entry *e; | |
4701 | struct merge_options_internal *opti = result->priv; | |
4702 | ||
4703 | strmap_for_each_entry(&opti->conflicted, &iter, e) { | |
4704 | const char *path = e->key; | |
4705 | struct conflict_info *ci = e->value; | |
4706 | int i; | |
4707 | ||
4708 | VERIFY_CI(ci); | |
4709 | ||
4710 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
4711 | struct stage_info *si; | |
4712 | ||
4713 | if (!(ci->filemask & (1ul << i))) | |
4714 | continue; | |
4715 | ||
4716 | si = xmalloc(sizeof(*si)); | |
4717 | si->stage = i+1; | |
4718 | si->mode = ci->stages[i].mode; | |
4719 | oidcpy(&si->oid, &ci->stages[i].oid); | |
4720 | string_list_append(conflicted_files, path)->util = si; | |
4721 | } | |
4722 | } | |
4723 | /* string_list_sort() uses a stable sort, so we're good */ | |
4724 | string_list_sort(conflicted_files); | |
4725 | } | |
4726 | ||
4727 | void merge_switch_to_result(struct merge_options *opt, | |
4728 | struct tree *head, | |
4729 | struct merge_result *result, | |
4730 | int update_worktree_and_index, | |
4731 | int display_update_msgs) | |
4732 | { | |
4733 | assert(opt->priv == NULL); | |
4734 | if (result->clean >= 0 && update_worktree_and_index) { | |
4735 | trace2_region_enter("merge", "checkout", opt->repo); | |
4736 | if (checkout(opt, head, result->tree)) { | |
4737 | /* failure to function */ | |
4738 | result->clean = -1; | |
4739 | merge_finalize(opt, result); | |
4740 | trace2_region_leave("merge", "checkout", opt->repo); | |
4741 | return; | |
4742 | } | |
4743 | trace2_region_leave("merge", "checkout", opt->repo); | |
4744 | ||
4745 | trace2_region_enter("merge", "record_conflicted", opt->repo); | |
4746 | opt->priv = result->priv; | |
4747 | if (record_conflicted_index_entries(opt)) { | |
4748 | /* failure to function */ | |
4749 | opt->priv = NULL; | |
4750 | result->clean = -1; | |
4751 | merge_finalize(opt, result); | |
4752 | trace2_region_leave("merge", "record_conflicted", | |
4753 | opt->repo); | |
4754 | return; | |
4755 | } | |
4756 | opt->priv = NULL; | |
4757 | trace2_region_leave("merge", "record_conflicted", opt->repo); | |
4758 | ||
4759 | trace2_region_enter("merge", "write_auto_merge", opt->repo); | |
4760 | if (refs_update_ref(get_main_ref_store(opt->repo), "", "AUTO_MERGE", | |
4761 | &result->tree->object.oid, NULL, REF_NO_DEREF, | |
4762 | UPDATE_REFS_MSG_ON_ERR)) { | |
4763 | /* failure to function */ | |
4764 | opt->priv = NULL; | |
4765 | result->clean = -1; | |
4766 | merge_finalize(opt, result); | |
4767 | trace2_region_leave("merge", "write_auto_merge", | |
4768 | opt->repo); | |
4769 | return; | |
4770 | } | |
4771 | trace2_region_leave("merge", "write_auto_merge", opt->repo); | |
4772 | } | |
4773 | if (display_update_msgs) | |
4774 | merge_display_update_messages(opt, /* detailed */ 0, result); | |
4775 | ||
4776 | merge_finalize(opt, result); | |
4777 | } | |
4778 | ||
4779 | void merge_finalize(struct merge_options *opt, | |
4780 | struct merge_result *result) | |
4781 | { | |
4782 | if (opt->renormalize) | |
4783 | git_attr_set_direction(GIT_ATTR_CHECKIN); | |
4784 | assert(opt->priv == NULL); | |
4785 | ||
4786 | if (result->priv) { | |
4787 | clear_or_reinit_internal_opts(result->priv, 0); | |
4788 | FREE_AND_NULL(result->priv); | |
4789 | } | |
4790 | } | |
4791 | ||
4792 | /*** Function Grouping: helper functions for merge_incore_*() ***/ | |
4793 | ||
4794 | static struct tree *shift_tree_object(struct repository *repo, | |
4795 | struct tree *one, struct tree *two, | |
4796 | const char *subtree_shift) | |
4797 | { | |
4798 | struct object_id shifted; | |
4799 | ||
4800 | if (!*subtree_shift) { | |
4801 | shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0); | |
4802 | } else { | |
4803 | shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted, | |
4804 | subtree_shift); | |
4805 | } | |
4806 | if (oideq(&two->object.oid, &shifted)) | |
4807 | return two; | |
4808 | return lookup_tree(repo, &shifted); | |
4809 | } | |
4810 | ||
4811 | static inline void set_commit_tree(struct commit *c, struct tree *t) | |
4812 | { | |
4813 | c->maybe_tree = t; | |
4814 | } | |
4815 | ||
4816 | static struct commit *make_virtual_commit(struct repository *repo, | |
4817 | struct tree *tree, | |
4818 | const char *comment) | |
4819 | { | |
4820 | struct commit *commit = alloc_commit_node(repo); | |
4821 | ||
4822 | set_merge_remote_desc(commit, comment, (struct object *)commit); | |
4823 | set_commit_tree(commit, tree); | |
4824 | commit->object.parsed = 1; | |
4825 | return commit; | |
4826 | } | |
4827 | ||
4828 | static void merge_start(struct merge_options *opt, struct merge_result *result) | |
4829 | { | |
4830 | struct rename_info *renames; | |
4831 | int i; | |
4832 | struct mem_pool *pool = NULL; | |
4833 | ||
4834 | /* Sanity checks on opt */ | |
4835 | trace2_region_enter("merge", "sanity checks", opt->repo); | |
4836 | assert(opt->repo); | |
4837 | ||
4838 | assert(opt->branch1 && opt->branch2); | |
4839 | ||
4840 | assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE && | |
4841 | opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE); | |
4842 | assert(opt->rename_limit >= -1); | |
4843 | assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE); | |
4844 | assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1); | |
4845 | ||
4846 | assert(opt->xdl_opts >= 0); | |
4847 | assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL && | |
4848 | opt->recursive_variant <= MERGE_VARIANT_THEIRS); | |
4849 | ||
4850 | if (opt->msg_header_prefix) | |
4851 | assert(opt->record_conflict_msgs_as_headers); | |
4852 | ||
4853 | /* | |
4854 | * detect_renames, verbosity, buffer_output, and obuf are ignored | |
4855 | * fields that were used by "recursive" rather than "ort" -- but | |
4856 | * sanity check them anyway. | |
4857 | */ | |
4858 | assert(opt->detect_renames >= -1 && | |
4859 | opt->detect_renames <= DIFF_DETECT_COPY); | |
4860 | assert(opt->verbosity >= 0 && opt->verbosity <= 5); | |
4861 | assert(opt->buffer_output <= 2); | |
4862 | assert(opt->obuf.len == 0); | |
4863 | ||
4864 | assert(opt->priv == NULL); | |
4865 | if (result->_properly_initialized != 0 && | |
4866 | result->_properly_initialized != RESULT_INITIALIZED) | |
4867 | BUG("struct merge_result passed to merge_incore_*recursive() must be zeroed or filled with values from a previous run"); | |
4868 | assert(!!result->priv == !!result->_properly_initialized); | |
4869 | if (result->priv) { | |
4870 | opt->priv = result->priv; | |
4871 | result->priv = NULL; | |
4872 | /* | |
4873 | * opt->priv non-NULL means we had results from a previous | |
4874 | * run; do a few sanity checks that user didn't mess with | |
4875 | * it in an obvious fashion. | |
4876 | */ | |
4877 | assert(opt->priv->call_depth == 0); | |
4878 | assert(!opt->priv->toplevel_dir || | |
4879 | 0 == strlen(opt->priv->toplevel_dir)); | |
4880 | } | |
4881 | trace2_region_leave("merge", "sanity checks", opt->repo); | |
4882 | ||
4883 | /* Default to histogram diff. Actually, just hardcode it...for now. */ | |
4884 | opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF); | |
4885 | ||
4886 | /* Handle attr direction stuff for renormalization */ | |
4887 | if (opt->renormalize) | |
4888 | git_attr_set_direction(GIT_ATTR_CHECKOUT); | |
4889 | ||
4890 | /* Initialization of opt->priv, our internal merge data */ | |
4891 | trace2_region_enter("merge", "allocate/init", opt->repo); | |
4892 | if (opt->priv) { | |
4893 | clear_or_reinit_internal_opts(opt->priv, 1); | |
4894 | string_list_init_nodup(&opt->priv->conflicted_submodules); | |
4895 | trace2_region_leave("merge", "allocate/init", opt->repo); | |
4896 | return; | |
4897 | } | |
4898 | opt->priv = xcalloc(1, sizeof(*opt->priv)); | |
4899 | ||
4900 | /* Initialization of various renames fields */ | |
4901 | renames = &opt->priv->renames; | |
4902 | mem_pool_init(&opt->priv->pool, 0); | |
4903 | pool = &opt->priv->pool; | |
4904 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) { | |
4905 | strintmap_init_with_options(&renames->dirs_removed[i], | |
4906 | NOT_RELEVANT, pool, 0); | |
4907 | strmap_init_with_options(&renames->dir_rename_count[i], | |
4908 | NULL, 1); | |
4909 | strmap_init_with_options(&renames->dir_renames[i], | |
4910 | NULL, 0); | |
4911 | /* | |
4912 | * relevant_sources uses -1 for the default, because we need | |
4913 | * to be able to distinguish not-in-strintmap from valid | |
4914 | * relevant_source values from enum file_rename_relevance. | |
4915 | * In particular, possibly_cache_new_pair() expects a negative | |
4916 | * value for not-found entries. | |
4917 | */ | |
4918 | strintmap_init_with_options(&renames->relevant_sources[i], | |
4919 | -1 /* explicitly invalid */, | |
4920 | pool, 0); | |
4921 | strmap_init_with_options(&renames->cached_pairs[i], | |
4922 | NULL, 1); | |
4923 | strset_init_with_options(&renames->cached_irrelevant[i], | |
4924 | NULL, 1); | |
4925 | strset_init_with_options(&renames->cached_target_names[i], | |
4926 | NULL, 0); | |
4927 | } | |
4928 | for (i = MERGE_SIDE1; i <= MERGE_SIDE2; i++) { | |
4929 | strintmap_init_with_options(&renames->deferred[i].possible_trivial_merges, | |
4930 | 0, pool, 0); | |
4931 | strset_init_with_options(&renames->deferred[i].target_dirs, | |
4932 | pool, 1); | |
4933 | renames->deferred[i].trivial_merges_okay = 1; /* 1 == maybe */ | |
4934 | } | |
4935 | ||
4936 | /* | |
4937 | * Although we initialize opt->priv->paths with strdup_strings=0, | |
4938 | * that's just to avoid making yet another copy of an allocated | |
4939 | * string. Putting the entry into paths means we are taking | |
4940 | * ownership, so we will later free it. | |
4941 | * | |
4942 | * In contrast, conflicted just has a subset of keys from paths, so | |
4943 | * we don't want to free those (it'd be a duplicate free). | |
4944 | */ | |
4945 | strmap_init_with_options(&opt->priv->paths, pool, 0); | |
4946 | strmap_init_with_options(&opt->priv->conflicted, pool, 0); | |
4947 | ||
4948 | /* | |
4949 | * keys & string_lists in conflicts will sometimes need to outlive | |
4950 | * "paths", so it will have a copy of relevant keys. It's probably | |
4951 | * a small subset of the overall paths that have special output. | |
4952 | */ | |
4953 | strmap_init(&opt->priv->conflicts); | |
4954 | ||
4955 | trace2_region_leave("merge", "allocate/init", opt->repo); | |
4956 | } | |
4957 | ||
4958 | static void merge_check_renames_reusable(struct merge_result *result, | |
4959 | struct tree *merge_base, | |
4960 | struct tree *side1, | |
4961 | struct tree *side2) | |
4962 | { | |
4963 | struct rename_info *renames; | |
4964 | struct tree **merge_trees; | |
4965 | struct merge_options_internal *opti = result->priv; | |
4966 | ||
4967 | if (!opti) | |
4968 | return; | |
4969 | ||
4970 | renames = &opti->renames; | |
4971 | merge_trees = renames->merge_trees; | |
4972 | ||
4973 | /* | |
4974 | * Handle case where previous merge operation did not want cache to | |
4975 | * take effect, e.g. because rename/rename(1to1) makes it invalid. | |
4976 | */ | |
4977 | if (!merge_trees[0]) { | |
4978 | assert(!merge_trees[0] && !merge_trees[1] && !merge_trees[2]); | |
4979 | renames->cached_pairs_valid_side = 0; /* neither side valid */ | |
4980 | return; | |
4981 | } | |
4982 | ||
4983 | /* | |
4984 | * Handle other cases; note that merge_trees[0..2] will only | |
4985 | * be NULL if opti is, or if all three were manually set to | |
4986 | * NULL by e.g. rename/rename(1to1) handling. | |
4987 | */ | |
4988 | assert(merge_trees[0] && merge_trees[1] && merge_trees[2]); | |
4989 | ||
4990 | /* Check if we meet a condition for re-using cached_pairs */ | |
4991 | if (oideq(&merge_base->object.oid, &merge_trees[2]->object.oid) && | |
4992 | oideq(&side1->object.oid, &result->tree->object.oid)) | |
4993 | renames->cached_pairs_valid_side = MERGE_SIDE1; | |
4994 | else if (oideq(&merge_base->object.oid, &merge_trees[1]->object.oid) && | |
4995 | oideq(&side2->object.oid, &result->tree->object.oid)) | |
4996 | renames->cached_pairs_valid_side = MERGE_SIDE2; | |
4997 | else | |
4998 | renames->cached_pairs_valid_side = 0; /* neither side valid */ | |
4999 | } | |
5000 | ||
5001 | /*** Function Grouping: merge_incore_*() and their internal variants ***/ | |
5002 | ||
5003 | /* | |
5004 | * Originally from merge_trees_internal(); heavily adapted, though. | |
5005 | */ | |
5006 | static void merge_ort_nonrecursive_internal(struct merge_options *opt, | |
5007 | struct tree *merge_base, | |
5008 | struct tree *side1, | |
5009 | struct tree *side2, | |
5010 | struct merge_result *result) | |
5011 | { | |
5012 | struct object_id working_tree_oid; | |
5013 | ||
5014 | if (opt->subtree_shift) { | |
5015 | side2 = shift_tree_object(opt->repo, side1, side2, | |
5016 | opt->subtree_shift); | |
5017 | merge_base = shift_tree_object(opt->repo, side1, merge_base, | |
5018 | opt->subtree_shift); | |
5019 | } | |
5020 | ||
5021 | redo: | |
5022 | trace2_region_enter("merge", "collect_merge_info", opt->repo); | |
5023 | if (collect_merge_info(opt, merge_base, side1, side2) != 0) { | |
5024 | /* | |
5025 | * TRANSLATORS: The %s arguments are: 1) tree hash of a merge | |
5026 | * base, and 2-3) the trees for the two trees we're merging. | |
5027 | */ | |
5028 | error(_("collecting merge info failed for trees %s, %s, %s"), | |
5029 | oid_to_hex(&merge_base->object.oid), | |
5030 | oid_to_hex(&side1->object.oid), | |
5031 | oid_to_hex(&side2->object.oid)); | |
5032 | result->clean = -1; | |
5033 | return; | |
5034 | } | |
5035 | trace2_region_leave("merge", "collect_merge_info", opt->repo); | |
5036 | ||
5037 | trace2_region_enter("merge", "renames", opt->repo); | |
5038 | result->clean = detect_and_process_renames(opt); | |
5039 | trace2_region_leave("merge", "renames", opt->repo); | |
5040 | if (opt->priv->renames.redo_after_renames == 2) { | |
5041 | trace2_region_enter("merge", "reset_maps", opt->repo); | |
5042 | clear_or_reinit_internal_opts(opt->priv, 1); | |
5043 | trace2_region_leave("merge", "reset_maps", opt->repo); | |
5044 | goto redo; | |
5045 | } | |
5046 | ||
5047 | trace2_region_enter("merge", "process_entries", opt->repo); | |
5048 | if (process_entries(opt, &working_tree_oid) < 0) | |
5049 | result->clean = -1; | |
5050 | trace2_region_leave("merge", "process_entries", opt->repo); | |
5051 | ||
5052 | /* Set return values */ | |
5053 | result->path_messages = &opt->priv->conflicts; | |
5054 | ||
5055 | if (result->clean >= 0) { | |
5056 | result->tree = parse_tree_indirect(&working_tree_oid); | |
5057 | if (!result->tree) | |
5058 | die(_("unable to read tree (%s)"), | |
5059 | oid_to_hex(&working_tree_oid)); | |
5060 | /* existence of conflicted entries implies unclean */ | |
5061 | result->clean &= strmap_empty(&opt->priv->conflicted); | |
5062 | } | |
5063 | if (!opt->priv->call_depth) { | |
5064 | result->priv = opt->priv; | |
5065 | result->_properly_initialized = RESULT_INITIALIZED; | |
5066 | opt->priv = NULL; | |
5067 | } | |
5068 | } | |
5069 | ||
5070 | /* | |
5071 | * Originally from merge_recursive_internal(); somewhat adapted, though. | |
5072 | */ | |
5073 | static void merge_ort_internal(struct merge_options *opt, | |
5074 | struct commit_list *merge_bases, | |
5075 | struct commit *h1, | |
5076 | struct commit *h2, | |
5077 | struct merge_result *result) | |
5078 | { | |
5079 | struct commit *next; | |
5080 | struct commit *merged_merge_bases; | |
5081 | const char *ancestor_name; | |
5082 | struct strbuf merge_base_abbrev = STRBUF_INIT; | |
5083 | ||
5084 | if (!merge_bases) { | |
5085 | if (repo_get_merge_bases(the_repository, h1, h2, | |
5086 | &merge_bases) < 0) { | |
5087 | result->clean = -1; | |
5088 | return; | |
5089 | } | |
5090 | /* See merge-ort.h:merge_incore_recursive() declaration NOTE */ | |
5091 | merge_bases = reverse_commit_list(merge_bases); | |
5092 | } | |
5093 | ||
5094 | merged_merge_bases = pop_commit(&merge_bases); | |
5095 | if (!merged_merge_bases) { | |
5096 | /* if there is no common ancestor, use an empty tree */ | |
5097 | struct tree *tree; | |
5098 | ||
5099 | tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree); | |
5100 | merged_merge_bases = make_virtual_commit(opt->repo, tree, | |
5101 | "ancestor"); | |
5102 | ancestor_name = "empty tree"; | |
5103 | } else if (merge_bases) { | |
5104 | ancestor_name = "merged common ancestors"; | |
5105 | } else { | |
5106 | strbuf_add_unique_abbrev(&merge_base_abbrev, | |
5107 | &merged_merge_bases->object.oid, | |
5108 | DEFAULT_ABBREV); | |
5109 | ancestor_name = merge_base_abbrev.buf; | |
5110 | } | |
5111 | ||
5112 | for (next = pop_commit(&merge_bases); next; | |
5113 | next = pop_commit(&merge_bases)) { | |
5114 | const char *saved_b1, *saved_b2; | |
5115 | struct commit *prev = merged_merge_bases; | |
5116 | ||
5117 | opt->priv->call_depth++; | |
5118 | /* | |
5119 | * When the merge fails, the result contains files | |
5120 | * with conflict markers. The cleanness flag is | |
5121 | * ignored (unless indicating an error), it was never | |
5122 | * actually used, as result of merge_trees has always | |
5123 | * overwritten it: the committed "conflicts" were | |
5124 | * already resolved. | |
5125 | */ | |
5126 | saved_b1 = opt->branch1; | |
5127 | saved_b2 = opt->branch2; | |
5128 | opt->branch1 = "Temporary merge branch 1"; | |
5129 | opt->branch2 = "Temporary merge branch 2"; | |
5130 | merge_ort_internal(opt, NULL, prev, next, result); | |
5131 | if (result->clean < 0) | |
5132 | return; | |
5133 | opt->branch1 = saved_b1; | |
5134 | opt->branch2 = saved_b2; | |
5135 | opt->priv->call_depth--; | |
5136 | ||
5137 | merged_merge_bases = make_virtual_commit(opt->repo, | |
5138 | result->tree, | |
5139 | "merged tree"); | |
5140 | commit_list_insert(prev, &merged_merge_bases->parents); | |
5141 | commit_list_insert(next, &merged_merge_bases->parents->next); | |
5142 | ||
5143 | clear_or_reinit_internal_opts(opt->priv, 1); | |
5144 | } | |
5145 | ||
5146 | opt->ancestor = ancestor_name; | |
5147 | merge_ort_nonrecursive_internal(opt, | |
5148 | repo_get_commit_tree(opt->repo, | |
5149 | merged_merge_bases), | |
5150 | repo_get_commit_tree(opt->repo, h1), | |
5151 | repo_get_commit_tree(opt->repo, h2), | |
5152 | result); | |
5153 | strbuf_release(&merge_base_abbrev); | |
5154 | opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */ | |
5155 | } | |
5156 | ||
5157 | void merge_incore_nonrecursive(struct merge_options *opt, | |
5158 | struct tree *merge_base, | |
5159 | struct tree *side1, | |
5160 | struct tree *side2, | |
5161 | struct merge_result *result) | |
5162 | { | |
5163 | trace2_region_enter("merge", "incore_nonrecursive", opt->repo); | |
5164 | ||
5165 | trace2_region_enter("merge", "merge_start", opt->repo); | |
5166 | assert(opt->ancestor != NULL); | |
5167 | merge_check_renames_reusable(result, merge_base, side1, side2); | |
5168 | merge_start(opt, result); | |
5169 | /* | |
5170 | * Record the trees used in this merge, so if there's a next merge in | |
5171 | * a cherry-pick or rebase sequence it might be able to take advantage | |
5172 | * of the cached_pairs in that next merge. | |
5173 | */ | |
5174 | opt->priv->renames.merge_trees[0] = merge_base; | |
5175 | opt->priv->renames.merge_trees[1] = side1; | |
5176 | opt->priv->renames.merge_trees[2] = side2; | |
5177 | trace2_region_leave("merge", "merge_start", opt->repo); | |
5178 | ||
5179 | merge_ort_nonrecursive_internal(opt, merge_base, side1, side2, result); | |
5180 | trace2_region_leave("merge", "incore_nonrecursive", opt->repo); | |
5181 | } | |
5182 | ||
5183 | void merge_incore_recursive(struct merge_options *opt, | |
5184 | struct commit_list *merge_bases, | |
5185 | struct commit *side1, | |
5186 | struct commit *side2, | |
5187 | struct merge_result *result) | |
5188 | { | |
5189 | trace2_region_enter("merge", "incore_recursive", opt->repo); | |
5190 | ||
5191 | /* We set the ancestor label based on the merge_bases */ | |
5192 | assert(opt->ancestor == NULL); | |
5193 | ||
5194 | trace2_region_enter("merge", "merge_start", opt->repo); | |
5195 | merge_start(opt, result); | |
5196 | trace2_region_leave("merge", "merge_start", opt->repo); | |
5197 | ||
5198 | merge_ort_internal(opt, merge_bases, side1, side2, result); | |
5199 | trace2_region_leave("merge", "incore_recursive", opt->repo); | |
5200 | } |