]>
Commit | Line | Data |
---|---|---|
17e5574b EN |
1 | /* |
2 | * "Ostensibly Recursive's Twin" merge strategy, or "ort" for short. Meant | |
3 | * as a drop-in replacement for the "recursive" merge strategy, allowing one | |
4 | * to replace | |
5 | * | |
6 | * git merge [-s recursive] | |
7 | * | |
8 | * with | |
9 | * | |
10 | * git merge -s ort | |
11 | * | |
12 | * Note: git's parser allows the space between '-s' and its argument to be | |
13 | * missing. (Should I have backronymed "ham", "alsa", "kip", "nap, "alvo", | |
14 | * "cale", "peedy", or "ins" instead of "ort"?) | |
15 | */ | |
16 | ||
17 | #include "cache.h" | |
18 | #include "merge-ort.h" | |
19 | ||
67845745 | 20 | #include "blob.h" |
ef2b3693 | 21 | #include "cache-tree.h" |
67845745 | 22 | #include "commit-reach.h" |
e4171b1b EN |
23 | #include "diff.h" |
24 | #include "diffcore.h" | |
6681ce5c | 25 | #include "dir.h" |
f591c472 | 26 | #include "ll-merge.h" |
ee4012dc | 27 | #include "object-store.h" |
5b59c3db | 28 | #include "strmap.h" |
c73cda76 | 29 | #include "submodule.h" |
231e2dd4 | 30 | #include "tree.h" |
6681ce5c | 31 | #include "unpack-trees.h" |
c8017176 | 32 | #include "xdiff-interface.h" |
5b59c3db | 33 | |
d2bc1994 EN |
34 | /* |
35 | * We have many arrays of size 3. Whenever we have such an array, the | |
36 | * indices refer to one of the sides of the three-way merge. This is so | |
37 | * pervasive that the constants 0, 1, and 2 are used in many places in the | |
38 | * code (especially in arithmetic operations to find the other side's index | |
39 | * or to compute a relevant mask), but sometimes these enum names are used | |
40 | * to aid code clarity. | |
41 | * | |
42 | * See also 'filemask' and 'dirmask' in struct conflict_info; the "ith side" | |
43 | * referred to there is one of these three sides. | |
44 | */ | |
45 | enum merge_side { | |
46 | MERGE_BASE = 0, | |
47 | MERGE_SIDE1 = 1, | |
48 | MERGE_SIDE2 = 2 | |
49 | }; | |
50 | ||
5b59c3db EN |
51 | struct merge_options_internal { |
52 | /* | |
53 | * paths: primary data structure in all of merge ort. | |
54 | * | |
55 | * The keys of paths: | |
56 | * * are full relative paths from the toplevel of the repository | |
57 | * (e.g. "drivers/firmware/raspberrypi.c"). | |
58 | * * store all relevant paths in the repo, both directories and | |
59 | * files (e.g. drivers, drivers/firmware would also be included) | |
60 | * * these keys serve to intern all the path strings, which allows | |
61 | * us to do pointer comparison on directory names instead of | |
62 | * strcmp; we just have to be careful to use the interned strings. | |
43c1dccb EN |
63 | * (Technically paths_to_free may track some strings that were |
64 | * removed from froms paths.) | |
5b59c3db EN |
65 | * |
66 | * The values of paths: | |
67 | * * either a pointer to a merged_info, or a conflict_info struct | |
68 | * * merged_info contains all relevant information for a | |
69 | * non-conflicted entry. | |
70 | * * conflict_info contains a merged_info, plus any additional | |
71 | * information about a conflict such as the higher orders stages | |
72 | * involved and the names of the paths those came from (handy | |
73 | * once renames get involved). | |
74 | * * a path may start "conflicted" (i.e. point to a conflict_info) | |
75 | * and then a later step (e.g. three-way content merge) determines | |
76 | * it can be cleanly merged, at which point it'll be marked clean | |
77 | * and the algorithm will ignore any data outside the contained | |
78 | * merged_info for that entry | |
79 | * * If an entry remains conflicted, the merged_info portion of a | |
80 | * conflict_info will later be filled with whatever version of | |
81 | * the file should be placed in the working directory (e.g. an | |
82 | * as-merged-as-possible variation that contains conflict markers). | |
83 | */ | |
84 | struct strmap paths; | |
85 | ||
86 | /* | |
87 | * conflicted: a subset of keys->values from "paths" | |
88 | * | |
89 | * conflicted is basically an optimization between process_entries() | |
90 | * and record_conflicted_index_entries(); the latter could loop over | |
91 | * ALL the entries in paths AGAIN and look for the ones that are | |
92 | * still conflicted, but since process_entries() has to loop over | |
93 | * all of them, it saves the ones it couldn't resolve in this strmap | |
94 | * so that record_conflicted_index_entries() can iterate just the | |
95 | * relevant entries. | |
96 | */ | |
97 | struct strmap conflicted; | |
98 | ||
43c1dccb EN |
99 | /* |
100 | * paths_to_free: additional list of strings to free | |
101 | * | |
102 | * If keys are removed from "paths", they are added to paths_to_free | |
103 | * to ensure they are later freed. We avoid free'ing immediately since | |
104 | * other places (e.g. conflict_info.pathnames[]) may still be | |
105 | * referencing these paths. | |
106 | */ | |
107 | struct string_list paths_to_free; | |
108 | ||
c5a6f655 EN |
109 | /* |
110 | * output: special messages and conflict notices for various paths | |
111 | * | |
112 | * This is a map of pathnames (a subset of the keys in "paths" above) | |
113 | * to strbufs. It gathers various warning/conflict/notice messages | |
114 | * for later processing. | |
115 | */ | |
116 | struct strmap output; | |
117 | ||
5b59c3db EN |
118 | /* |
119 | * current_dir_name: temporary var used in collect_merge_info_callback() | |
120 | * | |
121 | * Used to set merged_info.directory_name; see documentation for that | |
122 | * variable and the requirements placed on that field. | |
123 | */ | |
124 | const char *current_dir_name; | |
125 | ||
126 | /* call_depth: recursion level counter for merging merge bases */ | |
127 | int call_depth; | |
128 | }; | |
129 | ||
130 | struct version_info { | |
131 | struct object_id oid; | |
132 | unsigned short mode; | |
133 | }; | |
134 | ||
135 | struct merged_info { | |
136 | /* if is_null, ignore result. otherwise result has oid & mode */ | |
137 | struct version_info result; | |
138 | unsigned is_null:1; | |
139 | ||
140 | /* | |
141 | * clean: whether the path in question is cleanly merged. | |
142 | * | |
143 | * see conflict_info.merged for more details. | |
144 | */ | |
145 | unsigned clean:1; | |
146 | ||
147 | /* | |
148 | * basename_offset: offset of basename of path. | |
149 | * | |
150 | * perf optimization to avoid recomputing offset of final '/' | |
151 | * character in pathname (0 if no '/' in pathname). | |
152 | */ | |
153 | size_t basename_offset; | |
154 | ||
155 | /* | |
156 | * directory_name: containing directory name. | |
157 | * | |
158 | * Note that we assume directory_name is constructed such that | |
159 | * strcmp(dir1_name, dir2_name) == 0 iff dir1_name == dir2_name, | |
160 | * i.e. string equality is equivalent to pointer equality. For this | |
161 | * to hold, we have to be careful setting directory_name. | |
162 | */ | |
163 | const char *directory_name; | |
164 | }; | |
165 | ||
166 | struct conflict_info { | |
167 | /* | |
168 | * merged: the version of the path that will be written to working tree | |
169 | * | |
170 | * WARNING: It is critical to check merged.clean and ensure it is 0 | |
171 | * before reading any conflict_info fields outside of merged. | |
172 | * Allocated merge_info structs will always have clean set to 1. | |
173 | * Allocated conflict_info structs will have merged.clean set to 0 | |
174 | * initially. The merged.clean field is how we know if it is safe | |
175 | * to access other parts of conflict_info besides merged; if a | |
176 | * conflict_info's merged.clean is changed to 1, the rest of the | |
177 | * algorithm is not allowed to look at anything outside of the | |
178 | * merged member anymore. | |
179 | */ | |
180 | struct merged_info merged; | |
181 | ||
182 | /* oids & modes from each of the three trees for this path */ | |
183 | struct version_info stages[3]; | |
184 | ||
185 | /* pathnames for each stage; may differ due to rename detection */ | |
186 | const char *pathnames[3]; | |
187 | ||
188 | /* Whether this path is/was involved in a directory/file conflict */ | |
189 | unsigned df_conflict:1; | |
190 | ||
1c7873cd EN |
191 | /* |
192 | * Whether this path is/was involved in a non-content conflict other | |
193 | * than a directory/file conflict (e.g. rename/rename, rename/delete, | |
194 | * file location based on possible directory rename). | |
195 | */ | |
196 | unsigned path_conflict:1; | |
197 | ||
5b59c3db EN |
198 | /* |
199 | * For filemask and dirmask, the ith bit corresponds to whether the | |
200 | * ith entry is a file (filemask) or a directory (dirmask). Thus, | |
201 | * filemask & dirmask is always zero, and filemask | dirmask is at | |
202 | * most 7 but can be less when a path does not appear as either a | |
203 | * file or a directory on at least one side of history. | |
204 | * | |
205 | * Note that these masks are related to enum merge_side, as the ith | |
206 | * entry corresponds to side i. | |
207 | * | |
208 | * These values come from a traverse_trees() call; more info may be | |
209 | * found looking at tree-walk.h's struct traverse_info, | |
210 | * particularly the documentation above the "fn" member (note that | |
211 | * filemask = mask & ~dirmask from that documentation). | |
212 | */ | |
213 | unsigned filemask:3; | |
214 | unsigned dirmask:3; | |
215 | ||
216 | /* | |
217 | * Optimization to track which stages match, to avoid the need to | |
218 | * recompute it in multiple steps. Either 0 or at least 2 bits are | |
219 | * set; if at least 2 bits are set, their corresponding stages match. | |
220 | */ | |
221 | unsigned match_mask:3; | |
222 | }; | |
223 | ||
04af1879 EN |
224 | /*** Function Grouping: various utility functions ***/ |
225 | ||
98bf9841 EN |
226 | /* |
227 | * For the next three macros, see warning for conflict_info.merged. | |
228 | * | |
229 | * In each of the below, mi is a struct merged_info*, and ci was defined | |
230 | * as a struct conflict_info* (but we need to verify ci isn't actually | |
231 | * pointed at a struct merged_info*). | |
232 | * | |
233 | * INITIALIZE_CI: Assign ci to mi but only if it's safe; set to NULL otherwise. | |
234 | * VERIFY_CI: Ensure that something we assigned to a conflict_info* is one. | |
235 | * ASSIGN_AND_VERIFY_CI: Similar to VERIFY_CI but do assignment first. | |
236 | */ | |
237 | #define INITIALIZE_CI(ci, mi) do { \ | |
238 | (ci) = (!(mi) || (mi)->clean) ? NULL : (struct conflict_info *)(mi); \ | |
239 | } while (0) | |
240 | #define VERIFY_CI(ci) assert(ci && !ci->merged.clean); | |
241 | #define ASSIGN_AND_VERIFY_CI(ci, mi) do { \ | |
242 | (ci) = (struct conflict_info *)(mi); \ | |
243 | assert((ci) && !(mi)->clean); \ | |
244 | } while (0) | |
245 | ||
89422d29 EN |
246 | static void free_strmap_strings(struct strmap *map) |
247 | { | |
248 | struct hashmap_iter iter; | |
249 | struct strmap_entry *entry; | |
250 | ||
251 | strmap_for_each_entry(map, &iter, entry) { | |
252 | free((char*)entry->key); | |
253 | } | |
254 | } | |
255 | ||
101bc5bc EN |
256 | static void clear_internal_opts(struct merge_options_internal *opti, |
257 | int reinitialize) | |
258 | { | |
259 | assert(!reinitialize); | |
260 | ||
261 | /* | |
262 | * We marked opti->paths with strdup_strings = 0, so that we | |
263 | * wouldn't have to make another copy of the fullpath created by | |
264 | * make_traverse_path from setup_path_info(). But, now that we've | |
265 | * used it and have no other references to these strings, it is time | |
266 | * to deallocate them. | |
267 | */ | |
268 | free_strmap_strings(&opti->paths); | |
269 | strmap_clear(&opti->paths, 1); | |
270 | ||
271 | /* | |
272 | * All keys and values in opti->conflicted are a subset of those in | |
273 | * opti->paths. We don't want to deallocate anything twice, so we | |
274 | * don't free the keys and we pass 0 for free_values. | |
275 | */ | |
276 | strmap_clear(&opti->conflicted, 0); | |
43c1dccb EN |
277 | |
278 | /* | |
279 | * opti->paths_to_free is similar to opti->paths; we created it with | |
280 | * strdup_strings = 0 to avoid making _another_ copy of the fullpath | |
281 | * but now that we've used it and have no other references to these | |
282 | * strings, it is time to deallocate them. We do so by temporarily | |
283 | * setting strdup_strings to 1. | |
284 | */ | |
285 | opti->paths_to_free.strdup_strings = 1; | |
286 | string_list_clear(&opti->paths_to_free, 0); | |
287 | opti->paths_to_free.strdup_strings = 0; | |
c5a6f655 EN |
288 | |
289 | if (!reinitialize) { | |
290 | struct hashmap_iter iter; | |
291 | struct strmap_entry *e; | |
292 | ||
293 | /* Release and free each strbuf found in output */ | |
294 | strmap_for_each_entry(&opti->output, &iter, e) { | |
295 | struct strbuf *sb = e->value; | |
296 | strbuf_release(sb); | |
297 | /* | |
298 | * While strictly speaking we don't need to free(sb) | |
299 | * here because we could pass free_values=1 when | |
300 | * calling strmap_clear() on opti->output, that would | |
301 | * require strmap_clear to do another | |
302 | * strmap_for_each_entry() loop, so we just free it | |
303 | * while we're iterating anyway. | |
304 | */ | |
305 | free(sb); | |
306 | } | |
307 | strmap_clear(&opti->output, 0); | |
308 | } | |
101bc5bc EN |
309 | } |
310 | ||
0c0d705b EN |
311 | static int err(struct merge_options *opt, const char *err, ...) |
312 | { | |
313 | va_list params; | |
314 | struct strbuf sb = STRBUF_INIT; | |
315 | ||
316 | strbuf_addstr(&sb, "error: "); | |
317 | va_start(params, err); | |
318 | strbuf_vaddf(&sb, err, params); | |
319 | va_end(params); | |
320 | ||
321 | error("%s", sb.buf); | |
322 | strbuf_release(&sb); | |
323 | ||
324 | return -1; | |
325 | } | |
326 | ||
c73cda76 EN |
327 | static void format_commit(struct strbuf *sb, |
328 | int indent, | |
329 | struct commit *commit) | |
330 | { | |
70f19c7f EN |
331 | struct merge_remote_desc *desc; |
332 | struct pretty_print_context ctx = {0}; | |
333 | ctx.abbrev = DEFAULT_ABBREV; | |
334 | ||
335 | strbuf_addchars(sb, ' ', indent); | |
336 | desc = merge_remote_util(commit); | |
337 | if (desc) { | |
338 | strbuf_addf(sb, "virtual %s\n", desc->name); | |
339 | return; | |
340 | } | |
341 | ||
342 | format_commit_message(commit, "%h %s", sb, &ctx); | |
343 | strbuf_addch(sb, '\n'); | |
c73cda76 EN |
344 | } |
345 | ||
c5a6f655 EN |
346 | __attribute__((format (printf, 4, 5))) |
347 | static void path_msg(struct merge_options *opt, | |
348 | const char *path, | |
349 | int omittable_hint, /* skippable under --remerge-diff */ | |
350 | const char *fmt, ...) | |
351 | { | |
352 | va_list ap; | |
353 | struct strbuf *sb = strmap_get(&opt->priv->output, path); | |
354 | if (!sb) { | |
355 | sb = xmalloc(sizeof(*sb)); | |
356 | strbuf_init(sb, 0); | |
357 | strmap_put(&opt->priv->output, path, sb); | |
358 | } | |
359 | ||
360 | va_start(ap, fmt); | |
361 | strbuf_vaddf(sb, fmt, ap); | |
362 | va_end(ap); | |
363 | ||
364 | strbuf_addch(sb, '\n'); | |
365 | } | |
366 | ||
5a1a1e8e EN |
367 | /* add a string to a strbuf, but converting "/" to "_" */ |
368 | static void add_flattened_path(struct strbuf *out, const char *s) | |
369 | { | |
370 | size_t i = out->len; | |
371 | strbuf_addstr(out, s); | |
372 | for (; i < out->len; i++) | |
373 | if (out->buf[i] == '/') | |
374 | out->buf[i] = '_'; | |
375 | } | |
376 | ||
23366d2a EN |
377 | static char *unique_path(struct strmap *existing_paths, |
378 | const char *path, | |
379 | const char *branch) | |
380 | { | |
5a1a1e8e EN |
381 | struct strbuf newpath = STRBUF_INIT; |
382 | int suffix = 0; | |
383 | size_t base_len; | |
384 | ||
385 | strbuf_addf(&newpath, "%s~", path); | |
386 | add_flattened_path(&newpath, branch); | |
387 | ||
388 | base_len = newpath.len; | |
389 | while (strmap_contains(existing_paths, newpath.buf)) { | |
390 | strbuf_setlen(&newpath, base_len); | |
391 | strbuf_addf(&newpath, "_%d", suffix++); | |
392 | } | |
393 | ||
394 | return strbuf_detach(&newpath, NULL); | |
23366d2a EN |
395 | } |
396 | ||
04af1879 EN |
397 | /*** Function Grouping: functions related to collect_merge_info() ***/ |
398 | ||
98bf9841 EN |
399 | static void setup_path_info(struct merge_options *opt, |
400 | struct string_list_item *result, | |
401 | const char *current_dir_name, | |
402 | int current_dir_name_len, | |
403 | char *fullpath, /* we'll take over ownership */ | |
404 | struct name_entry *names, | |
405 | struct name_entry *merged_version, | |
406 | unsigned is_null, /* boolean */ | |
407 | unsigned df_conflict, /* boolean */ | |
408 | unsigned filemask, | |
409 | unsigned dirmask, | |
410 | int resolved /* boolean */) | |
411 | { | |
412 | /* result->util is void*, so mi is a convenience typed variable */ | |
413 | struct merged_info *mi; | |
414 | ||
415 | assert(!is_null || resolved); | |
416 | assert(!df_conflict || !resolved); /* df_conflict implies !resolved */ | |
417 | assert(resolved == (merged_version != NULL)); | |
418 | ||
419 | mi = xcalloc(1, resolved ? sizeof(struct merged_info) : | |
420 | sizeof(struct conflict_info)); | |
421 | mi->directory_name = current_dir_name; | |
422 | mi->basename_offset = current_dir_name_len; | |
423 | mi->clean = !!resolved; | |
424 | if (resolved) { | |
425 | mi->result.mode = merged_version->mode; | |
426 | oidcpy(&mi->result.oid, &merged_version->oid); | |
427 | mi->is_null = !!is_null; | |
428 | } else { | |
429 | int i; | |
430 | struct conflict_info *ci; | |
431 | ||
432 | ASSIGN_AND_VERIFY_CI(ci, mi); | |
433 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
434 | ci->pathnames[i] = fullpath; | |
435 | ci->stages[i].mode = names[i].mode; | |
436 | oidcpy(&ci->stages[i].oid, &names[i].oid); | |
437 | } | |
438 | ci->filemask = filemask; | |
439 | ci->dirmask = dirmask; | |
440 | ci->df_conflict = !!df_conflict; | |
441 | if (dirmask) | |
442 | /* | |
443 | * Assume is_null for now, but if we have entries | |
444 | * under the directory then when it is complete in | |
445 | * write_completed_directory() it'll update this. | |
446 | * Also, for D/F conflicts, we have to handle the | |
447 | * directory first, then clear this bit and process | |
448 | * the file to see how it is handled -- that occurs | |
449 | * near the top of process_entry(). | |
450 | */ | |
451 | mi->is_null = 1; | |
452 | } | |
453 | strmap_put(&opt->priv->paths, fullpath, mi); | |
454 | result->string = fullpath; | |
455 | result->util = mi; | |
456 | } | |
457 | ||
d2bc1994 EN |
458 | static int collect_merge_info_callback(int n, |
459 | unsigned long mask, | |
460 | unsigned long dirmask, | |
461 | struct name_entry *names, | |
462 | struct traverse_info *info) | |
463 | { | |
464 | /* | |
465 | * n is 3. Always. | |
466 | * common ancestor (mbase) has mask 1, and stored in index 0 of names | |
467 | * head of side 1 (side1) has mask 2, and stored in index 1 of names | |
468 | * head of side 2 (side2) has mask 4, and stored in index 2 of names | |
469 | */ | |
470 | struct merge_options *opt = info->data; | |
471 | struct merge_options_internal *opti = opt->priv; | |
98bf9841 EN |
472 | struct string_list_item pi; /* Path Info */ |
473 | struct conflict_info *ci; /* typed alias to pi.util (which is void*) */ | |
d2bc1994 EN |
474 | struct name_entry *p; |
475 | size_t len; | |
476 | char *fullpath; | |
98bf9841 | 477 | const char *dirname = opti->current_dir_name; |
d2bc1994 | 478 | unsigned filemask = mask & ~dirmask; |
34e557af | 479 | unsigned match_mask = 0; /* will be updated below */ |
d2bc1994 EN |
480 | unsigned mbase_null = !(mask & 1); |
481 | unsigned side1_null = !(mask & 2); | |
482 | unsigned side2_null = !(mask & 4); | |
885f0063 EN |
483 | unsigned side1_matches_mbase = (!side1_null && !mbase_null && |
484 | names[0].mode == names[1].mode && | |
485 | oideq(&names[0].oid, &names[1].oid)); | |
486 | unsigned side2_matches_mbase = (!side2_null && !mbase_null && | |
487 | names[0].mode == names[2].mode && | |
488 | oideq(&names[0].oid, &names[2].oid)); | |
489 | unsigned sides_match = (!side1_null && !side2_null && | |
490 | names[1].mode == names[2].mode && | |
491 | oideq(&names[1].oid, &names[2].oid)); | |
d2bc1994 | 492 | |
34e557af EN |
493 | /* |
494 | * Note: When a path is a file on one side of history and a directory | |
495 | * in another, we have a directory/file conflict. In such cases, if | |
496 | * the conflict doesn't resolve from renames and deletions, then we | |
497 | * always leave directories where they are and move files out of the | |
498 | * way. Thus, while struct conflict_info has a df_conflict field to | |
499 | * track such conflicts, we ignore that field for any directories at | |
500 | * a path and only pay attention to it for files at the given path. | |
501 | * The fact that we leave directories were they are also means that | |
502 | * we do not need to worry about getting additional df_conflict | |
503 | * information propagated from parent directories down to children | |
504 | * (unlike, say traverse_trees_recursive() in unpack-trees.c, which | |
505 | * sets a newinfo.df_conflicts field specifically to propagate it). | |
506 | */ | |
507 | unsigned df_conflict = (filemask != 0) && (dirmask != 0); | |
508 | ||
d2bc1994 EN |
509 | /* n = 3 is a fundamental assumption. */ |
510 | if (n != 3) | |
511 | BUG("Called collect_merge_info_callback wrong"); | |
512 | ||
513 | /* | |
514 | * A bunch of sanity checks verifying that traverse_trees() calls | |
515 | * us the way I expect. Could just remove these at some point, | |
516 | * though maybe they are helpful to future code readers. | |
517 | */ | |
518 | assert(mbase_null == is_null_oid(&names[0].oid)); | |
519 | assert(side1_null == is_null_oid(&names[1].oid)); | |
520 | assert(side2_null == is_null_oid(&names[2].oid)); | |
521 | assert(!mbase_null || !side1_null || !side2_null); | |
522 | assert(mask > 0 && mask < 8); | |
523 | ||
34e557af EN |
524 | /* Determine match_mask */ |
525 | if (side1_matches_mbase) | |
526 | match_mask = (side2_matches_mbase ? 7 : 3); | |
527 | else if (side2_matches_mbase) | |
528 | match_mask = 5; | |
529 | else if (sides_match) | |
530 | match_mask = 6; | |
531 | ||
d2bc1994 EN |
532 | /* |
533 | * Get the name of the relevant filepath, which we'll pass to | |
534 | * setup_path_info() for tracking. | |
535 | */ | |
536 | p = names; | |
537 | while (!p->mode) | |
538 | p++; | |
539 | len = traverse_path_len(info, p->pathlen); | |
540 | ||
541 | /* +1 in both of the following lines to include the NUL byte */ | |
542 | fullpath = xmalloc(len + 1); | |
543 | make_traverse_path(fullpath, len + 1, info, p->path, p->pathlen); | |
544 | ||
291f29ca EN |
545 | /* |
546 | * If mbase, side1, and side2 all match, we can resolve early. Even | |
547 | * if these are trees, there will be no renames or anything | |
548 | * underneath. | |
549 | */ | |
550 | if (side1_matches_mbase && side2_matches_mbase) { | |
551 | /* mbase, side1, & side2 all match; use mbase as resolution */ | |
552 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, | |
553 | names, names+0, mbase_null, 0, | |
554 | filemask, dirmask, 1); | |
555 | return mask; | |
556 | } | |
557 | ||
d2bc1994 | 558 | /* |
98bf9841 EN |
559 | * Record information about the path so we can resolve later in |
560 | * process_entries. | |
d2bc1994 | 561 | */ |
98bf9841 EN |
562 | setup_path_info(opt, &pi, dirname, info->pathlen, fullpath, |
563 | names, NULL, 0, df_conflict, filemask, dirmask, 0); | |
564 | ||
565 | ci = pi.util; | |
566 | VERIFY_CI(ci); | |
34e557af | 567 | ci->match_mask = match_mask; |
d2bc1994 EN |
568 | |
569 | /* If dirmask, recurse into subdirectories */ | |
570 | if (dirmask) { | |
571 | struct traverse_info newinfo; | |
572 | struct tree_desc t[3]; | |
573 | void *buf[3] = {NULL, NULL, NULL}; | |
574 | const char *original_dir_name; | |
575 | int i, ret; | |
576 | ||
577 | ci->match_mask &= filemask; | |
578 | newinfo = *info; | |
579 | newinfo.prev = info; | |
580 | newinfo.name = p->path; | |
581 | newinfo.namelen = p->pathlen; | |
582 | newinfo.pathlen = st_add3(newinfo.pathlen, p->pathlen, 1); | |
34e557af EN |
583 | /* |
584 | * If this directory we are about to recurse into cared about | |
585 | * its parent directory (the current directory) having a D/F | |
586 | * conflict, then we'd propagate the masks in this way: | |
587 | * newinfo.df_conflicts |= (mask & ~dirmask); | |
588 | * But we don't worry about propagating D/F conflicts. (See | |
589 | * comment near setting of local df_conflict variable near | |
590 | * the beginning of this function). | |
591 | */ | |
d2bc1994 EN |
592 | |
593 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
885f0063 EN |
594 | if (i == 1 && side1_matches_mbase) |
595 | t[1] = t[0]; | |
596 | else if (i == 2 && side2_matches_mbase) | |
597 | t[2] = t[0]; | |
598 | else if (i == 2 && sides_match) | |
599 | t[2] = t[1]; | |
600 | else { | |
601 | const struct object_id *oid = NULL; | |
602 | if (dirmask & 1) | |
603 | oid = &names[i].oid; | |
604 | buf[i] = fill_tree_descriptor(opt->repo, | |
605 | t + i, oid); | |
606 | } | |
d2bc1994 EN |
607 | dirmask >>= 1; |
608 | } | |
609 | ||
610 | original_dir_name = opti->current_dir_name; | |
98bf9841 | 611 | opti->current_dir_name = pi.string; |
d2bc1994 EN |
612 | ret = traverse_trees(NULL, 3, t, &newinfo); |
613 | opti->current_dir_name = original_dir_name; | |
614 | ||
615 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) | |
616 | free(buf[i]); | |
617 | ||
618 | if (ret < 0) | |
619 | return -1; | |
620 | } | |
621 | ||
622 | return mask; | |
623 | } | |
624 | ||
231e2dd4 EN |
625 | static int collect_merge_info(struct merge_options *opt, |
626 | struct tree *merge_base, | |
627 | struct tree *side1, | |
628 | struct tree *side2) | |
629 | { | |
d2bc1994 EN |
630 | int ret; |
631 | struct tree_desc t[3]; | |
632 | struct traverse_info info; | |
633 | const char *toplevel_dir_placeholder = ""; | |
634 | ||
635 | opt->priv->current_dir_name = toplevel_dir_placeholder; | |
636 | setup_traverse_info(&info, toplevel_dir_placeholder); | |
637 | info.fn = collect_merge_info_callback; | |
638 | info.data = opt; | |
639 | info.show_all_errors = 1; | |
640 | ||
641 | parse_tree(merge_base); | |
642 | parse_tree(side1); | |
643 | parse_tree(side2); | |
644 | init_tree_desc(t + 0, merge_base->buffer, merge_base->size); | |
645 | init_tree_desc(t + 1, side1->buffer, side1->size); | |
646 | init_tree_desc(t + 2, side2->buffer, side2->size); | |
647 | ||
648 | ret = traverse_trees(NULL, 3, t, &info); | |
649 | ||
650 | return ret; | |
231e2dd4 EN |
651 | } |
652 | ||
04af1879 EN |
653 | /*** Function Grouping: functions related to threeway content merges ***/ |
654 | ||
c73cda76 EN |
655 | static int find_first_merges(struct repository *repo, |
656 | const char *path, | |
657 | struct commit *a, | |
658 | struct commit *b, | |
659 | struct object_array *result) | |
660 | { | |
661 | die("Not yet implemented."); | |
662 | } | |
663 | ||
62fdec17 EN |
664 | static int merge_submodule(struct merge_options *opt, |
665 | const char *path, | |
666 | const struct object_id *o, | |
667 | const struct object_id *a, | |
668 | const struct object_id *b, | |
669 | struct object_id *result) | |
670 | { | |
c73cda76 EN |
671 | struct commit *commit_o, *commit_a, *commit_b; |
672 | int parent_count; | |
673 | struct object_array merges; | |
674 | struct strbuf sb = STRBUF_INIT; | |
675 | ||
676 | int i; | |
677 | int search = !opt->priv->call_depth; | |
678 | ||
679 | /* store fallback answer in result in case we fail */ | |
680 | oidcpy(result, opt->priv->call_depth ? o : a); | |
681 | ||
682 | /* we can not handle deletion conflicts */ | |
683 | if (is_null_oid(o)) | |
684 | return 0; | |
685 | if (is_null_oid(a)) | |
686 | return 0; | |
687 | if (is_null_oid(b)) | |
688 | return 0; | |
689 | ||
690 | if (add_submodule_odb(path)) { | |
691 | path_msg(opt, path, 0, | |
692 | _("Failed to merge submodule %s (not checked out)"), | |
693 | path); | |
694 | return 0; | |
695 | } | |
696 | ||
697 | if (!(commit_o = lookup_commit_reference(opt->repo, o)) || | |
698 | !(commit_a = lookup_commit_reference(opt->repo, a)) || | |
699 | !(commit_b = lookup_commit_reference(opt->repo, b))) { | |
700 | path_msg(opt, path, 0, | |
701 | _("Failed to merge submodule %s (commits not present)"), | |
702 | path); | |
703 | return 0; | |
704 | } | |
705 | ||
706 | /* check whether both changes are forward */ | |
707 | if (!in_merge_bases(commit_o, commit_a) || | |
708 | !in_merge_bases(commit_o, commit_b)) { | |
709 | path_msg(opt, path, 0, | |
710 | _("Failed to merge submodule %s " | |
711 | "(commits don't follow merge-base)"), | |
712 | path); | |
713 | return 0; | |
714 | } | |
715 | ||
716 | /* Case #1: a is contained in b or vice versa */ | |
717 | if (in_merge_bases(commit_a, commit_b)) { | |
718 | oidcpy(result, b); | |
719 | path_msg(opt, path, 1, | |
720 | _("Note: Fast-forwarding submodule %s to %s"), | |
721 | path, oid_to_hex(b)); | |
722 | return 1; | |
723 | } | |
724 | if (in_merge_bases(commit_b, commit_a)) { | |
725 | oidcpy(result, a); | |
726 | path_msg(opt, path, 1, | |
727 | _("Note: Fast-forwarding submodule %s to %s"), | |
728 | path, oid_to_hex(a)); | |
729 | return 1; | |
730 | } | |
731 | ||
732 | /* | |
733 | * Case #2: There are one or more merges that contain a and b in | |
734 | * the submodule. If there is only one, then present it as a | |
735 | * suggestion to the user, but leave it marked unmerged so the | |
736 | * user needs to confirm the resolution. | |
737 | */ | |
738 | ||
739 | /* Skip the search if makes no sense to the calling context. */ | |
740 | if (!search) | |
741 | return 0; | |
742 | ||
743 | /* find commit which merges them */ | |
744 | parent_count = find_first_merges(opt->repo, path, commit_a, commit_b, | |
745 | &merges); | |
746 | switch (parent_count) { | |
747 | case 0: | |
748 | path_msg(opt, path, 0, _("Failed to merge submodule %s"), path); | |
749 | break; | |
750 | ||
751 | case 1: | |
752 | format_commit(&sb, 4, | |
753 | (struct commit *)merges.objects[0].item); | |
754 | path_msg(opt, path, 0, | |
755 | _("Failed to merge submodule %s, but a possible merge " | |
756 | "resolution exists:\n%s\n"), | |
757 | path, sb.buf); | |
758 | path_msg(opt, path, 1, | |
759 | _("If this is correct simply add it to the index " | |
760 | "for example\n" | |
761 | "by using:\n\n" | |
762 | " git update-index --cacheinfo 160000 %s \"%s\"\n\n" | |
763 | "which will accept this suggestion.\n"), | |
764 | oid_to_hex(&merges.objects[0].item->oid), path); | |
765 | strbuf_release(&sb); | |
766 | break; | |
767 | default: | |
768 | for (i = 0; i < merges.nr; i++) | |
769 | format_commit(&sb, 4, | |
770 | (struct commit *)merges.objects[i].item); | |
771 | path_msg(opt, path, 0, | |
772 | _("Failed to merge submodule %s, but multiple " | |
773 | "possible merges exist:\n%s"), path, sb.buf); | |
774 | strbuf_release(&sb); | |
775 | } | |
776 | ||
777 | object_array_clear(&merges); | |
778 | return 0; | |
62fdec17 EN |
779 | } |
780 | ||
781 | static int merge_3way(struct merge_options *opt, | |
782 | const char *path, | |
783 | const struct object_id *o, | |
784 | const struct object_id *a, | |
785 | const struct object_id *b, | |
786 | const char *pathnames[3], | |
787 | const int extra_marker_size, | |
788 | mmbuffer_t *result_buf) | |
789 | { | |
f591c472 EN |
790 | mmfile_t orig, src1, src2; |
791 | struct ll_merge_options ll_opts = {0}; | |
792 | char *base, *name1, *name2; | |
793 | int merge_status; | |
794 | ||
795 | ll_opts.renormalize = opt->renormalize; | |
796 | ll_opts.extra_marker_size = extra_marker_size; | |
797 | ll_opts.xdl_opts = opt->xdl_opts; | |
798 | ||
799 | if (opt->priv->call_depth) { | |
800 | ll_opts.virtual_ancestor = 1; | |
801 | ll_opts.variant = 0; | |
802 | } else { | |
803 | switch (opt->recursive_variant) { | |
804 | case MERGE_VARIANT_OURS: | |
805 | ll_opts.variant = XDL_MERGE_FAVOR_OURS; | |
806 | break; | |
807 | case MERGE_VARIANT_THEIRS: | |
808 | ll_opts.variant = XDL_MERGE_FAVOR_THEIRS; | |
809 | break; | |
810 | default: | |
811 | ll_opts.variant = 0; | |
812 | break; | |
813 | } | |
814 | } | |
815 | ||
816 | assert(pathnames[0] && pathnames[1] && pathnames[2] && opt->ancestor); | |
817 | if (pathnames[0] == pathnames[1] && pathnames[1] == pathnames[2]) { | |
818 | base = mkpathdup("%s", opt->ancestor); | |
819 | name1 = mkpathdup("%s", opt->branch1); | |
820 | name2 = mkpathdup("%s", opt->branch2); | |
821 | } else { | |
822 | base = mkpathdup("%s:%s", opt->ancestor, pathnames[0]); | |
823 | name1 = mkpathdup("%s:%s", opt->branch1, pathnames[1]); | |
824 | name2 = mkpathdup("%s:%s", opt->branch2, pathnames[2]); | |
825 | } | |
826 | ||
827 | read_mmblob(&orig, o); | |
828 | read_mmblob(&src1, a); | |
829 | read_mmblob(&src2, b); | |
830 | ||
831 | merge_status = ll_merge(result_buf, path, &orig, base, | |
832 | &src1, name1, &src2, name2, | |
833 | opt->repo->index, &ll_opts); | |
834 | ||
835 | free(base); | |
836 | free(name1); | |
837 | free(name2); | |
838 | free(orig.ptr); | |
839 | free(src1.ptr); | |
840 | free(src2.ptr); | |
841 | return merge_status; | |
62fdec17 EN |
842 | } |
843 | ||
e2e9dc03 EN |
844 | static int handle_content_merge(struct merge_options *opt, |
845 | const char *path, | |
846 | const struct version_info *o, | |
847 | const struct version_info *a, | |
848 | const struct version_info *b, | |
849 | const char *pathnames[3], | |
850 | const int extra_marker_size, | |
851 | struct version_info *result) | |
852 | { | |
991bbdca | 853 | /* |
62fdec17 EN |
854 | * path is the target location where we want to put the file, and |
855 | * is used to determine any normalization rules in ll_merge. | |
856 | * | |
857 | * The normal case is that path and all entries in pathnames are | |
858 | * identical, though renames can affect which path we got one of | |
859 | * the three blobs to merge on various sides of history. | |
860 | * | |
861 | * extra_marker_size is the amount to extend conflict markers in | |
862 | * ll_merge; this is neeed if we have content merges of content | |
863 | * merges, which happens for example with rename/rename(2to1) and | |
864 | * rename/add conflicts. | |
865 | */ | |
866 | unsigned clean = 1; | |
867 | ||
868 | /* | |
869 | * handle_content_merge() needs both files to be of the same type, i.e. | |
870 | * both files OR both submodules OR both symlinks. Conflicting types | |
871 | * needs to be handled elsewhere. | |
991bbdca | 872 | */ |
62fdec17 EN |
873 | assert((S_IFMT & a->mode) == (S_IFMT & b->mode)); |
874 | ||
875 | /* Merge modes */ | |
876 | if (a->mode == b->mode || a->mode == o->mode) | |
877 | result->mode = b->mode; | |
878 | else { | |
879 | /* must be the 100644/100755 case */ | |
880 | assert(S_ISREG(a->mode)); | |
881 | result->mode = a->mode; | |
882 | clean = (b->mode == o->mode); | |
883 | /* | |
884 | * FIXME: If opt->priv->call_depth && !clean, then we really | |
885 | * should not make result->mode match either a->mode or | |
886 | * b->mode; that causes t6036 "check conflicting mode for | |
887 | * regular file" to fail. It would be best to use some other | |
888 | * mode, but we'll confuse all kinds of stuff if we use one | |
889 | * where S_ISREG(result->mode) isn't true, and if we use | |
890 | * something like 0100666, then tree-walk.c's calls to | |
891 | * canon_mode() will just normalize that to 100644 for us and | |
892 | * thus not solve anything. | |
893 | * | |
894 | * Figure out if there's some kind of way we can work around | |
895 | * this... | |
896 | */ | |
897 | } | |
898 | ||
899 | /* | |
900 | * Trivial oid merge. | |
901 | * | |
902 | * Note: While one might assume that the next four lines would | |
903 | * be unnecessary due to the fact that match_mask is often | |
904 | * setup and already handled, renames don't always take care | |
905 | * of that. | |
906 | */ | |
907 | if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid)) | |
908 | oidcpy(&result->oid, &b->oid); | |
909 | else if (oideq(&b->oid, &o->oid)) | |
910 | oidcpy(&result->oid, &a->oid); | |
911 | ||
912 | /* Remaining rules depend on file vs. submodule vs. symlink. */ | |
913 | else if (S_ISREG(a->mode)) { | |
914 | mmbuffer_t result_buf; | |
915 | int ret = 0, merge_status; | |
916 | int two_way; | |
917 | ||
918 | /* | |
919 | * If 'o' is different type, treat it as null so we do a | |
920 | * two-way merge. | |
921 | */ | |
922 | two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
923 | ||
924 | merge_status = merge_3way(opt, path, | |
925 | two_way ? &null_oid : &o->oid, | |
926 | &a->oid, &b->oid, | |
927 | pathnames, extra_marker_size, | |
928 | &result_buf); | |
929 | ||
930 | if ((merge_status < 0) || !result_buf.ptr) | |
931 | ret = err(opt, _("Failed to execute internal merge")); | |
932 | ||
933 | if (!ret && | |
934 | write_object_file(result_buf.ptr, result_buf.size, | |
935 | blob_type, &result->oid)) | |
936 | ret = err(opt, _("Unable to add %s to database"), | |
937 | path); | |
938 | ||
939 | free(result_buf.ptr); | |
940 | if (ret) | |
941 | return -1; | |
942 | clean &= (merge_status == 0); | |
943 | path_msg(opt, path, 1, _("Auto-merging %s"), path); | |
944 | } else if (S_ISGITLINK(a->mode)) { | |
945 | int two_way = ((S_IFMT & o->mode) != (S_IFMT & a->mode)); | |
946 | clean = merge_submodule(opt, pathnames[0], | |
947 | two_way ? &null_oid : &o->oid, | |
948 | &a->oid, &b->oid, &result->oid); | |
949 | if (opt->priv->call_depth && two_way && !clean) { | |
950 | result->mode = o->mode; | |
951 | oidcpy(&result->oid, &o->oid); | |
952 | } | |
953 | } else if (S_ISLNK(a->mode)) { | |
954 | if (opt->priv->call_depth) { | |
955 | clean = 0; | |
956 | result->mode = o->mode; | |
957 | oidcpy(&result->oid, &o->oid); | |
958 | } else { | |
959 | switch (opt->recursive_variant) { | |
960 | case MERGE_VARIANT_NORMAL: | |
961 | clean = 0; | |
962 | oidcpy(&result->oid, &a->oid); | |
963 | break; | |
964 | case MERGE_VARIANT_OURS: | |
965 | oidcpy(&result->oid, &a->oid); | |
966 | break; | |
967 | case MERGE_VARIANT_THEIRS: | |
968 | oidcpy(&result->oid, &b->oid); | |
969 | break; | |
970 | } | |
971 | } | |
972 | } else | |
973 | BUG("unsupported object type in the tree: %06o for %s", | |
974 | a->mode, path); | |
975 | ||
991bbdca | 976 | return clean; |
e2e9dc03 EN |
977 | } |
978 | ||
04af1879 EN |
979 | /*** Function Grouping: functions related to detect_and_process_renames(), *** |
980 | *** which are split into directory and regular rename detection sections. ***/ | |
981 | ||
982 | /*** Function Grouping: functions related to directory rename detection ***/ | |
983 | ||
984 | /*** Function Grouping: functions related to regular rename detection ***/ | |
985 | ||
231e2dd4 EN |
986 | static int detect_and_process_renames(struct merge_options *opt, |
987 | struct tree *merge_base, | |
988 | struct tree *side1, | |
989 | struct tree *side2) | |
990 | { | |
991 | int clean = 1; | |
992 | ||
993 | /* | |
994 | * Rename detection works by detecting file similarity. Here we use | |
995 | * a really easy-to-implement scheme: files are similar IFF they have | |
996 | * the same filename. Therefore, by this scheme, there are no renames. | |
997 | * | |
998 | * TODO: Actually implement a real rename detection scheme. | |
999 | */ | |
1000 | return clean; | |
1001 | } | |
1002 | ||
04af1879 EN |
1003 | /*** Function Grouping: functions related to process_entries() ***/ |
1004 | ||
8adffaa8 EN |
1005 | static int string_list_df_name_compare(const char *one, const char *two) |
1006 | { | |
1007 | int onelen = strlen(one); | |
1008 | int twolen = strlen(two); | |
1009 | /* | |
1010 | * Here we only care that entries for D/F conflicts are | |
1011 | * adjacent, in particular with the file of the D/F conflict | |
1012 | * appearing before files below the corresponding directory. | |
1013 | * The order of the rest of the list is irrelevant for us. | |
1014 | * | |
1015 | * To achieve this, we sort with df_name_compare and provide | |
1016 | * the mode S_IFDIR so that D/F conflicts will sort correctly. | |
1017 | * We use the mode S_IFDIR for everything else for simplicity, | |
1018 | * since in other cases any changes in their order due to | |
1019 | * sorting cause no problems for us. | |
1020 | */ | |
1021 | int cmp = df_name_compare(one, onelen, S_IFDIR, | |
1022 | two, twolen, S_IFDIR); | |
1023 | /* | |
1024 | * Now that 'foo' and 'foo/bar' compare equal, we have to make sure | |
1025 | * that 'foo' comes before 'foo/bar'. | |
1026 | */ | |
1027 | if (cmp) | |
1028 | return cmp; | |
1029 | return onelen - twolen; | |
1030 | } | |
1031 | ||
a9945bba | 1032 | struct directory_versions { |
bb470f4e EN |
1033 | /* |
1034 | * versions: list of (basename -> version_info) | |
1035 | * | |
1036 | * The basenames are in reverse lexicographic order of full pathnames, | |
1037 | * as processed in process_entries(). This puts all entries within | |
1038 | * a directory together, and covers the directory itself after | |
1039 | * everything within it, allowing us to write subtrees before needing | |
1040 | * to record information for the tree itself. | |
1041 | */ | |
a9945bba | 1042 | struct string_list versions; |
bb470f4e EN |
1043 | |
1044 | /* | |
1045 | * offsets: list of (full relative path directories -> integer offsets) | |
1046 | * | |
1047 | * Since versions contains basenames from files in multiple different | |
1048 | * directories, we need to know which entries in versions correspond | |
1049 | * to which directories. Values of e.g. | |
1050 | * "" 0 | |
1051 | * src 2 | |
1052 | * src/moduleA 5 | |
1053 | * Would mean that entries 0-1 of versions are files in the toplevel | |
1054 | * directory, entries 2-4 are files under src/, and the remaining | |
1055 | * entries starting at index 5 are files under src/moduleA/. | |
1056 | */ | |
1057 | struct string_list offsets; | |
1058 | ||
1059 | /* | |
1060 | * last_directory: directory that previously processed file found in | |
1061 | * | |
1062 | * last_directory starts NULL, but records the directory in which the | |
1063 | * previous file was found within. As soon as | |
1064 | * directory(current_file) != last_directory | |
1065 | * then we need to start updating accounting in versions & offsets. | |
1066 | * Note that last_directory is always the last path in "offsets" (or | |
1067 | * NULL if "offsets" is empty) so this exists just for quick access. | |
1068 | */ | |
1069 | const char *last_directory; | |
1070 | ||
1071 | /* last_directory_len: cached computation of strlen(last_directory) */ | |
1072 | unsigned last_directory_len; | |
a9945bba EN |
1073 | }; |
1074 | ||
ee4012dc EN |
1075 | static int tree_entry_order(const void *a_, const void *b_) |
1076 | { | |
1077 | const struct string_list_item *a = a_; | |
1078 | const struct string_list_item *b = b_; | |
1079 | ||
1080 | const struct merged_info *ami = a->util; | |
1081 | const struct merged_info *bmi = b->util; | |
1082 | return base_name_compare(a->string, strlen(a->string), ami->result.mode, | |
1083 | b->string, strlen(b->string), bmi->result.mode); | |
1084 | } | |
1085 | ||
1086 | static void write_tree(struct object_id *result_oid, | |
1087 | struct string_list *versions, | |
1088 | unsigned int offset, | |
1089 | size_t hash_size) | |
1090 | { | |
1091 | size_t maxlen = 0, extra; | |
1092 | unsigned int nr = versions->nr - offset; | |
1093 | struct strbuf buf = STRBUF_INIT; | |
1094 | struct string_list relevant_entries = STRING_LIST_INIT_NODUP; | |
1095 | int i; | |
1096 | ||
1097 | /* | |
1098 | * We want to sort the last (versions->nr-offset) entries in versions. | |
1099 | * Do so by abusing the string_list API a bit: make another string_list | |
1100 | * that contains just those entries and then sort them. | |
1101 | * | |
1102 | * We won't use relevant_entries again and will let it just pop off the | |
1103 | * stack, so there won't be allocation worries or anything. | |
1104 | */ | |
1105 | relevant_entries.items = versions->items + offset; | |
1106 | relevant_entries.nr = versions->nr - offset; | |
1107 | QSORT(relevant_entries.items, relevant_entries.nr, tree_entry_order); | |
1108 | ||
1109 | /* Pre-allocate some space in buf */ | |
1110 | extra = hash_size + 8; /* 8: 6 for mode, 1 for space, 1 for NUL char */ | |
1111 | for (i = 0; i < nr; i++) { | |
1112 | maxlen += strlen(versions->items[offset+i].string) + extra; | |
1113 | } | |
1114 | strbuf_grow(&buf, maxlen); | |
1115 | ||
1116 | /* Write each entry out to buf */ | |
1117 | for (i = 0; i < nr; i++) { | |
1118 | struct merged_info *mi = versions->items[offset+i].util; | |
1119 | struct version_info *ri = &mi->result; | |
1120 | strbuf_addf(&buf, "%o %s%c", | |
1121 | ri->mode, | |
1122 | versions->items[offset+i].string, '\0'); | |
1123 | strbuf_add(&buf, ri->oid.hash, hash_size); | |
1124 | } | |
1125 | ||
1126 | /* Write this object file out, and record in result_oid */ | |
1127 | write_object_file(buf.buf, buf.len, tree_type, result_oid); | |
1128 | strbuf_release(&buf); | |
1129 | } | |
1130 | ||
a9945bba EN |
1131 | static void record_entry_for_tree(struct directory_versions *dir_metadata, |
1132 | const char *path, | |
1133 | struct merged_info *mi) | |
1134 | { | |
1135 | const char *basename; | |
1136 | ||
1137 | if (mi->is_null) | |
1138 | /* nothing to record */ | |
1139 | return; | |
1140 | ||
1141 | basename = path + mi->basename_offset; | |
1142 | assert(strchr(basename, '/') == NULL); | |
1143 | string_list_append(&dir_metadata->versions, | |
1144 | basename)->util = &mi->result; | |
1145 | } | |
1146 | ||
bb470f4e EN |
1147 | static void write_completed_directory(struct merge_options *opt, |
1148 | const char *new_directory_name, | |
1149 | struct directory_versions *info) | |
1150 | { | |
1151 | const char *prev_dir; | |
1152 | struct merged_info *dir_info = NULL; | |
1153 | unsigned int offset; | |
1154 | ||
1155 | /* | |
1156 | * Some explanation of info->versions and info->offsets... | |
1157 | * | |
1158 | * process_entries() iterates over all relevant files AND | |
1159 | * directories in reverse lexicographic order, and calls this | |
1160 | * function. Thus, an example of the paths that process_entries() | |
1161 | * could operate on (along with the directories for those paths | |
1162 | * being shown) is: | |
1163 | * | |
1164 | * xtract.c "" | |
1165 | * tokens.txt "" | |
1166 | * src/moduleB/umm.c src/moduleB | |
1167 | * src/moduleB/stuff.h src/moduleB | |
1168 | * src/moduleB/baz.c src/moduleB | |
1169 | * src/moduleB src | |
1170 | * src/moduleA/foo.c src/moduleA | |
1171 | * src/moduleA/bar.c src/moduleA | |
1172 | * src/moduleA src | |
1173 | * src "" | |
1174 | * Makefile "" | |
1175 | * | |
1176 | * info->versions: | |
1177 | * | |
1178 | * always contains the unprocessed entries and their | |
1179 | * version_info information. For example, after the first five | |
1180 | * entries above, info->versions would be: | |
1181 | * | |
1182 | * xtract.c <xtract.c's version_info> | |
1183 | * token.txt <token.txt's version_info> | |
1184 | * umm.c <src/moduleB/umm.c's version_info> | |
1185 | * stuff.h <src/moduleB/stuff.h's version_info> | |
1186 | * baz.c <src/moduleB/baz.c's version_info> | |
1187 | * | |
1188 | * Once a subdirectory is completed we remove the entries in | |
1189 | * that subdirectory from info->versions, writing it as a tree | |
1190 | * (write_tree()). Thus, as soon as we get to src/moduleB, | |
1191 | * info->versions would be updated to | |
1192 | * | |
1193 | * xtract.c <xtract.c's version_info> | |
1194 | * token.txt <token.txt's version_info> | |
1195 | * moduleB <src/moduleB's version_info> | |
1196 | * | |
1197 | * info->offsets: | |
1198 | * | |
1199 | * helps us track which entries in info->versions correspond to | |
1200 | * which directories. When we are N directories deep (e.g. 4 | |
1201 | * for src/modA/submod/subdir/), we have up to N+1 unprocessed | |
1202 | * directories (+1 because of toplevel dir). Corresponding to | |
1203 | * the info->versions example above, after processing five entries | |
1204 | * info->offsets will be: | |
1205 | * | |
1206 | * "" 0 | |
1207 | * src/moduleB 2 | |
1208 | * | |
1209 | * which is used to know that xtract.c & token.txt are from the | |
1210 | * toplevel dirctory, while umm.c & stuff.h & baz.c are from the | |
1211 | * src/moduleB directory. Again, following the example above, | |
1212 | * once we need to process src/moduleB, then info->offsets is | |
1213 | * updated to | |
1214 | * | |
1215 | * "" 0 | |
1216 | * src 2 | |
1217 | * | |
1218 | * which says that moduleB (and only moduleB so far) is in the | |
1219 | * src directory. | |
1220 | * | |
1221 | * One unique thing to note about info->offsets here is that | |
1222 | * "src" was not added to info->offsets until there was a path | |
1223 | * (a file OR directory) immediately below src/ that got | |
1224 | * processed. | |
1225 | * | |
1226 | * Since process_entry() just appends new entries to info->versions, | |
1227 | * write_completed_directory() only needs to do work if the next path | |
1228 | * is in a directory that is different than the last directory found | |
1229 | * in info->offsets. | |
1230 | */ | |
1231 | ||
1232 | /* | |
1233 | * If we are working with the same directory as the last entry, there | |
1234 | * is no work to do. (See comments above the directory_name member of | |
1235 | * struct merged_info for why we can use pointer comparison instead of | |
1236 | * strcmp here.) | |
1237 | */ | |
1238 | if (new_directory_name == info->last_directory) | |
1239 | return; | |
1240 | ||
1241 | /* | |
1242 | * If we are just starting (last_directory is NULL), or last_directory | |
1243 | * is a prefix of the current directory, then we can just update | |
1244 | * info->offsets to record the offset where we started this directory | |
1245 | * and update last_directory to have quick access to it. | |
1246 | */ | |
1247 | if (info->last_directory == NULL || | |
1248 | !strncmp(new_directory_name, info->last_directory, | |
1249 | info->last_directory_len)) { | |
1250 | uintptr_t offset = info->versions.nr; | |
1251 | ||
1252 | info->last_directory = new_directory_name; | |
1253 | info->last_directory_len = strlen(info->last_directory); | |
1254 | /* | |
1255 | * Record the offset into info->versions where we will | |
1256 | * start recording basenames of paths found within | |
1257 | * new_directory_name. | |
1258 | */ | |
1259 | string_list_append(&info->offsets, | |
1260 | info->last_directory)->util = (void*)offset; | |
1261 | return; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * The next entry that will be processed will be within | |
1266 | * new_directory_name. Since at this point we know that | |
1267 | * new_directory_name is within a different directory than | |
1268 | * info->last_directory, we have all entries for info->last_directory | |
1269 | * in info->versions and we need to create a tree object for them. | |
1270 | */ | |
1271 | dir_info = strmap_get(&opt->priv->paths, info->last_directory); | |
1272 | assert(dir_info); | |
1273 | offset = (uintptr_t)info->offsets.items[info->offsets.nr-1].util; | |
1274 | if (offset == info->versions.nr) { | |
1275 | /* | |
1276 | * Actually, we don't need to create a tree object in this | |
1277 | * case. Whenever all files within a directory disappear | |
1278 | * during the merge (e.g. unmodified on one side and | |
1279 | * deleted on the other, or files were renamed elsewhere), | |
1280 | * then we get here and the directory itself needs to be | |
1281 | * omitted from its parent tree as well. | |
1282 | */ | |
1283 | dir_info->is_null = 1; | |
1284 | } else { | |
1285 | /* | |
1286 | * Write out the tree to the git object directory, and also | |
1287 | * record the mode and oid in dir_info->result. | |
1288 | */ | |
1289 | dir_info->is_null = 0; | |
1290 | dir_info->result.mode = S_IFDIR; | |
1291 | write_tree(&dir_info->result.oid, &info->versions, offset, | |
1292 | opt->repo->hash_algo->rawsz); | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * We've now used several entries from info->versions and one entry | |
1297 | * from info->offsets, so we get rid of those values. | |
1298 | */ | |
1299 | info->offsets.nr--; | |
1300 | info->versions.nr = offset; | |
1301 | ||
1302 | /* | |
1303 | * Now we've taken care of the completed directory, but we need to | |
1304 | * prepare things since future entries will be in | |
1305 | * new_directory_name. (In particular, process_entry() will be | |
1306 | * appending new entries to info->versions.) So, we need to make | |
1307 | * sure new_directory_name is the last entry in info->offsets. | |
1308 | */ | |
1309 | prev_dir = info->offsets.nr == 0 ? NULL : | |
1310 | info->offsets.items[info->offsets.nr-1].string; | |
1311 | if (new_directory_name != prev_dir) { | |
1312 | uintptr_t c = info->versions.nr; | |
1313 | string_list_append(&info->offsets, | |
1314 | new_directory_name)->util = (void*)c; | |
1315 | } | |
1316 | ||
1317 | /* And, of course, we need to update last_directory to match. */ | |
1318 | info->last_directory = new_directory_name; | |
1319 | info->last_directory_len = strlen(info->last_directory); | |
1320 | } | |
1321 | ||
6a02dd90 EN |
1322 | /* Per entry merge function */ |
1323 | static void process_entry(struct merge_options *opt, | |
1324 | const char *path, | |
a9945bba EN |
1325 | struct conflict_info *ci, |
1326 | struct directory_versions *dir_metadata) | |
6a02dd90 | 1327 | { |
23366d2a EN |
1328 | int df_file_index = 0; |
1329 | ||
6a02dd90 EN |
1330 | VERIFY_CI(ci); |
1331 | assert(ci->filemask >= 0 && ci->filemask <= 7); | |
1332 | /* ci->match_mask == 7 was handled in collect_merge_info_callback() */ | |
1333 | assert(ci->match_mask == 0 || ci->match_mask == 3 || | |
1334 | ci->match_mask == 5 || ci->match_mask == 6); | |
1335 | ||
a9945bba EN |
1336 | if (ci->dirmask) { |
1337 | record_entry_for_tree(dir_metadata, path, &ci->merged); | |
1338 | if (ci->filemask == 0) | |
1339 | /* nothing else to handle */ | |
1340 | return; | |
1341 | assert(ci->df_conflict); | |
1342 | } | |
1343 | ||
0ccfa4e5 EN |
1344 | if (ci->df_conflict && ci->merged.result.mode == 0) { |
1345 | int i; | |
1346 | ||
1347 | /* | |
1348 | * directory no longer in the way, but we do have a file we | |
1349 | * need to place here so we need to clean away the "directory | |
1350 | * merges to nothing" result. | |
1351 | */ | |
1352 | ci->df_conflict = 0; | |
1353 | assert(ci->filemask != 0); | |
1354 | ci->merged.clean = 0; | |
1355 | ci->merged.is_null = 0; | |
1356 | /* and we want to zero out any directory-related entries */ | |
1357 | ci->match_mask = (ci->match_mask & ~ci->dirmask); | |
1358 | ci->dirmask = 0; | |
1359 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
1360 | if (ci->filemask & (1 << i)) | |
1361 | continue; | |
1362 | ci->stages[i].mode = 0; | |
1363 | oidcpy(&ci->stages[i].oid, &null_oid); | |
1364 | } | |
1365 | } else if (ci->df_conflict && ci->merged.result.mode != 0) { | |
23366d2a EN |
1366 | /* |
1367 | * This started out as a D/F conflict, and the entries in | |
1368 | * the competing directory were not removed by the merge as | |
1369 | * evidenced by write_completed_directory() writing a value | |
1370 | * to ci->merged.result.mode. | |
1371 | */ | |
1372 | struct conflict_info *new_ci; | |
1373 | const char *branch; | |
1374 | const char *old_path = path; | |
1375 | int i; | |
1376 | ||
1377 | assert(ci->merged.result.mode == S_IFDIR); | |
1378 | ||
1379 | /* | |
1380 | * If filemask is 1, we can just ignore the file as having | |
1381 | * been deleted on both sides. We do not want to overwrite | |
1382 | * ci->merged.result, since it stores the tree for all the | |
1383 | * files under it. | |
1384 | */ | |
1385 | if (ci->filemask == 1) { | |
1386 | ci->filemask = 0; | |
1387 | return; | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * This file still exists on at least one side, and we want | |
1392 | * the directory to remain here, so we need to move this | |
1393 | * path to some new location. | |
1394 | */ | |
1395 | new_ci = xcalloc(1, sizeof(*new_ci)); | |
1396 | /* We don't really want new_ci->merged.result copied, but it'll | |
1397 | * be overwritten below so it doesn't matter. We also don't | |
1398 | * want any directory mode/oid values copied, but we'll zero | |
1399 | * those out immediately. We do want the rest of ci copied. | |
1400 | */ | |
1401 | memcpy(new_ci, ci, sizeof(*ci)); | |
1402 | new_ci->match_mask = (new_ci->match_mask & ~new_ci->dirmask); | |
1403 | new_ci->dirmask = 0; | |
1404 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
1405 | if (new_ci->filemask & (1 << i)) | |
1406 | continue; | |
1407 | /* zero out any entries related to directories */ | |
1408 | new_ci->stages[i].mode = 0; | |
1409 | oidcpy(&new_ci->stages[i].oid, &null_oid); | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * Find out which side this file came from; note that we | |
1414 | * cannot just use ci->filemask, because renames could cause | |
1415 | * the filemask to go back to 7. So we use dirmask, then | |
1416 | * pick the opposite side's index. | |
1417 | */ | |
1418 | df_file_index = (ci->dirmask & (1 << 1)) ? 2 : 1; | |
1419 | branch = (df_file_index == 1) ? opt->branch1 : opt->branch2; | |
1420 | path = unique_path(&opt->priv->paths, path, branch); | |
1421 | strmap_put(&opt->priv->paths, path, new_ci); | |
1422 | ||
1423 | path_msg(opt, path, 0, | |
1424 | _("CONFLICT (file/directory): directory in the way " | |
1425 | "of %s from %s; moving it to %s instead."), | |
1426 | old_path, branch, path); | |
1427 | ||
1428 | /* | |
1429 | * Zero out the filemask for the old ci. At this point, ci | |
1430 | * was just an entry for a directory, so we don't need to | |
1431 | * do anything more with it. | |
1432 | */ | |
1433 | ci->filemask = 0; | |
1434 | ||
1435 | /* | |
1436 | * Now note that we're working on the new entry (path was | |
1437 | * updated above. | |
1438 | */ | |
1439 | ci = new_ci; | |
6a02dd90 EN |
1440 | } |
1441 | ||
1442 | /* | |
1443 | * NOTE: Below there is a long switch-like if-elseif-elseif... block | |
1444 | * which the code goes through even for the df_conflict cases | |
23366d2a | 1445 | * above. |
6a02dd90 EN |
1446 | */ |
1447 | if (ci->match_mask) { | |
1448 | ci->merged.clean = 1; | |
1449 | if (ci->match_mask == 6) { | |
1450 | /* stages[1] == stages[2] */ | |
1451 | ci->merged.result.mode = ci->stages[1].mode; | |
1452 | oidcpy(&ci->merged.result.oid, &ci->stages[1].oid); | |
1453 | } else { | |
1454 | /* determine the mask of the side that didn't match */ | |
1455 | unsigned int othermask = 7 & ~ci->match_mask; | |
1456 | int side = (othermask == 4) ? 2 : 1; | |
1457 | ||
1458 | ci->merged.result.mode = ci->stages[side].mode; | |
1459 | ci->merged.is_null = !ci->merged.result.mode; | |
1460 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
1461 | ||
1462 | assert(othermask == 2 || othermask == 4); | |
1463 | assert(ci->merged.is_null == | |
1464 | (ci->filemask == ci->match_mask)); | |
1465 | } | |
1466 | } else if (ci->filemask >= 6 && | |
1467 | (S_IFMT & ci->stages[1].mode) != | |
1468 | (S_IFMT & ci->stages[2].mode)) { | |
1469 | /* | |
1470 | * Two different items from (file/submodule/symlink) | |
1471 | */ | |
1472 | die("Not yet implemented."); | |
1473 | } else if (ci->filemask >= 6) { | |
991bbdca EN |
1474 | /* Need a two-way or three-way content merge */ |
1475 | struct version_info merged_file; | |
1476 | unsigned clean_merge; | |
1477 | struct version_info *o = &ci->stages[0]; | |
1478 | struct version_info *a = &ci->stages[1]; | |
1479 | struct version_info *b = &ci->stages[2]; | |
1480 | ||
1481 | clean_merge = handle_content_merge(opt, path, o, a, b, | |
1482 | ci->pathnames, | |
1483 | opt->priv->call_depth * 2, | |
1484 | &merged_file); | |
1485 | ci->merged.clean = clean_merge && | |
1486 | !ci->df_conflict && !ci->path_conflict; | |
1487 | ci->merged.result.mode = merged_file.mode; | |
1488 | ci->merged.is_null = (merged_file.mode == 0); | |
1489 | oidcpy(&ci->merged.result.oid, &merged_file.oid); | |
1490 | if (clean_merge && ci->df_conflict) { | |
1491 | assert(df_file_index == 1 || df_file_index == 2); | |
1492 | ci->filemask = 1 << df_file_index; | |
1493 | ci->stages[df_file_index].mode = merged_file.mode; | |
1494 | oidcpy(&ci->stages[df_file_index].oid, &merged_file.oid); | |
1495 | } | |
1496 | if (!clean_merge) { | |
1497 | const char *reason = _("content"); | |
1498 | if (ci->filemask == 6) | |
1499 | reason = _("add/add"); | |
1500 | if (S_ISGITLINK(merged_file.mode)) | |
1501 | reason = _("submodule"); | |
1502 | path_msg(opt, path, 0, | |
1503 | _("CONFLICT (%s): Merge conflict in %s"), | |
1504 | reason, path); | |
1505 | } | |
6a02dd90 EN |
1506 | } else if (ci->filemask == 3 || ci->filemask == 5) { |
1507 | /* Modify/delete */ | |
c5a6f655 EN |
1508 | const char *modify_branch, *delete_branch; |
1509 | int side = (ci->filemask == 5) ? 2 : 1; | |
1510 | int index = opt->priv->call_depth ? 0 : side; | |
1511 | ||
1512 | ci->merged.result.mode = ci->stages[index].mode; | |
1513 | oidcpy(&ci->merged.result.oid, &ci->stages[index].oid); | |
1514 | ci->merged.clean = 0; | |
1515 | ||
1516 | modify_branch = (side == 1) ? opt->branch1 : opt->branch2; | |
1517 | delete_branch = (side == 1) ? opt->branch2 : opt->branch1; | |
1518 | ||
1519 | path_msg(opt, path, 0, | |
1520 | _("CONFLICT (modify/delete): %s deleted in %s " | |
1521 | "and modified in %s. Version %s of %s left " | |
1522 | "in tree."), | |
1523 | path, delete_branch, modify_branch, | |
1524 | modify_branch, path); | |
6a02dd90 EN |
1525 | } else if (ci->filemask == 2 || ci->filemask == 4) { |
1526 | /* Added on one side */ | |
1527 | int side = (ci->filemask == 4) ? 2 : 1; | |
1528 | ci->merged.result.mode = ci->stages[side].mode; | |
1529 | oidcpy(&ci->merged.result.oid, &ci->stages[side].oid); | |
1530 | ci->merged.clean = !ci->df_conflict; | |
1531 | } else if (ci->filemask == 1) { | |
1532 | /* Deleted on both sides */ | |
1533 | ci->merged.is_null = 1; | |
1534 | ci->merged.result.mode = 0; | |
1535 | oidcpy(&ci->merged.result.oid, &null_oid); | |
1536 | ci->merged.clean = 1; | |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * If still conflicted, record it separately. This allows us to later | |
1541 | * iterate over just conflicted entries when updating the index instead | |
1542 | * of iterating over all entries. | |
1543 | */ | |
1544 | if (!ci->merged.clean) | |
1545 | strmap_put(&opt->priv->conflicted, path, ci); | |
a9945bba | 1546 | record_entry_for_tree(dir_metadata, path, &ci->merged); |
6a02dd90 EN |
1547 | } |
1548 | ||
231e2dd4 EN |
1549 | static void process_entries(struct merge_options *opt, |
1550 | struct object_id *result_oid) | |
1551 | { | |
6a02dd90 EN |
1552 | struct hashmap_iter iter; |
1553 | struct strmap_entry *e; | |
8adffaa8 EN |
1554 | struct string_list plist = STRING_LIST_INIT_NODUP; |
1555 | struct string_list_item *entry; | |
bb470f4e EN |
1556 | struct directory_versions dir_metadata = { STRING_LIST_INIT_NODUP, |
1557 | STRING_LIST_INIT_NODUP, | |
1558 | NULL, 0 }; | |
6a02dd90 EN |
1559 | |
1560 | if (strmap_empty(&opt->priv->paths)) { | |
1561 | oidcpy(result_oid, opt->repo->hash_algo->empty_tree); | |
1562 | return; | |
1563 | } | |
1564 | ||
8adffaa8 EN |
1565 | /* Hack to pre-allocate plist to the desired size */ |
1566 | ALLOC_GROW(plist.items, strmap_get_size(&opt->priv->paths), plist.alloc); | |
1567 | ||
1568 | /* Put every entry from paths into plist, then sort */ | |
6a02dd90 | 1569 | strmap_for_each_entry(&opt->priv->paths, &iter, e) { |
8adffaa8 EN |
1570 | string_list_append(&plist, e->key)->util = e->value; |
1571 | } | |
1572 | plist.cmp = string_list_df_name_compare; | |
1573 | string_list_sort(&plist); | |
1574 | ||
1575 | /* | |
1576 | * Iterate over the items in reverse order, so we can handle paths | |
1577 | * below a directory before needing to handle the directory itself. | |
bb470f4e EN |
1578 | * |
1579 | * This allows us to write subtrees before we need to write trees, | |
1580 | * and it also enables sane handling of directory/file conflicts | |
1581 | * (because it allows us to know whether the directory is still in | |
1582 | * the way when it is time to process the file at the same path). | |
8adffaa8 EN |
1583 | */ |
1584 | for (entry = &plist.items[plist.nr-1]; entry >= plist.items; --entry) { | |
1585 | char *path = entry->string; | |
6a02dd90 EN |
1586 | /* |
1587 | * NOTE: mi may actually be a pointer to a conflict_info, but | |
1588 | * we have to check mi->clean first to see if it's safe to | |
1589 | * reassign to such a pointer type. | |
1590 | */ | |
8adffaa8 | 1591 | struct merged_info *mi = entry->util; |
6a02dd90 | 1592 | |
bb470f4e EN |
1593 | write_completed_directory(opt, mi->directory_name, |
1594 | &dir_metadata); | |
a9945bba EN |
1595 | if (mi->clean) |
1596 | record_entry_for_tree(&dir_metadata, path, mi); | |
1597 | else { | |
8adffaa8 | 1598 | struct conflict_info *ci = (struct conflict_info *)mi; |
a9945bba | 1599 | process_entry(opt, path, ci, &dir_metadata); |
8adffaa8 | 1600 | } |
6a02dd90 EN |
1601 | } |
1602 | ||
bb470f4e EN |
1603 | if (dir_metadata.offsets.nr != 1 || |
1604 | (uintptr_t)dir_metadata.offsets.items[0].util != 0) { | |
1605 | printf("dir_metadata.offsets.nr = %d (should be 1)\n", | |
1606 | dir_metadata.offsets.nr); | |
1607 | printf("dir_metadata.offsets.items[0].util = %u (should be 0)\n", | |
1608 | (unsigned)(uintptr_t)dir_metadata.offsets.items[0].util); | |
1609 | fflush(stdout); | |
1610 | BUG("dir_metadata accounting completely off; shouldn't happen"); | |
1611 | } | |
ee4012dc EN |
1612 | write_tree(result_oid, &dir_metadata.versions, 0, |
1613 | opt->repo->hash_algo->rawsz); | |
8adffaa8 | 1614 | string_list_clear(&plist, 0); |
a9945bba | 1615 | string_list_clear(&dir_metadata.versions, 0); |
bb470f4e | 1616 | string_list_clear(&dir_metadata.offsets, 0); |
231e2dd4 EN |
1617 | } |
1618 | ||
04af1879 EN |
1619 | /*** Function Grouping: functions related to merge_switch_to_result() ***/ |
1620 | ||
9fefce68 EN |
1621 | static int checkout(struct merge_options *opt, |
1622 | struct tree *prev, | |
1623 | struct tree *next) | |
1624 | { | |
6681ce5c EN |
1625 | /* Switch the index/working copy from old to new */ |
1626 | int ret; | |
1627 | struct tree_desc trees[2]; | |
1628 | struct unpack_trees_options unpack_opts; | |
1629 | ||
1630 | memset(&unpack_opts, 0, sizeof(unpack_opts)); | |
1631 | unpack_opts.head_idx = -1; | |
1632 | unpack_opts.src_index = opt->repo->index; | |
1633 | unpack_opts.dst_index = opt->repo->index; | |
1634 | ||
1635 | setup_unpack_trees_porcelain(&unpack_opts, "merge"); | |
1636 | ||
1637 | /* | |
1638 | * NOTE: if this were just "git checkout" code, we would probably | |
1639 | * read or refresh the cache and check for a conflicted index, but | |
1640 | * builtin/merge.c or sequencer.c really needs to read the index | |
1641 | * and check for conflicted entries before starting merging for a | |
1642 | * good user experience (no sense waiting for merges/rebases before | |
1643 | * erroring out), so there's no reason to duplicate that work here. | |
1644 | */ | |
1645 | ||
1646 | /* 2-way merge to the new branch */ | |
1647 | unpack_opts.update = 1; | |
1648 | unpack_opts.merge = 1; | |
1649 | unpack_opts.quiet = 0; /* FIXME: sequencer might want quiet? */ | |
1650 | unpack_opts.verbose_update = (opt->verbosity > 2); | |
1651 | unpack_opts.fn = twoway_merge; | |
1652 | if (1/* FIXME: opts->overwrite_ignore*/) { | |
1653 | unpack_opts.dir = xcalloc(1, sizeof(*unpack_opts.dir)); | |
1654 | unpack_opts.dir->flags |= DIR_SHOW_IGNORED; | |
1655 | setup_standard_excludes(unpack_opts.dir); | |
1656 | } | |
1657 | parse_tree(prev); | |
1658 | init_tree_desc(&trees[0], prev->buffer, prev->size); | |
1659 | parse_tree(next); | |
1660 | init_tree_desc(&trees[1], next->buffer, next->size); | |
1661 | ||
1662 | ret = unpack_trees(2, trees, &unpack_opts); | |
1663 | clear_unpack_trees_porcelain(&unpack_opts); | |
1664 | dir_clear(unpack_opts.dir); | |
1665 | FREE_AND_NULL(unpack_opts.dir); | |
1666 | return ret; | |
9fefce68 EN |
1667 | } |
1668 | ||
1669 | static int record_conflicted_index_entries(struct merge_options *opt, | |
1670 | struct index_state *index, | |
1671 | struct strmap *paths, | |
1672 | struct strmap *conflicted) | |
1673 | { | |
ef2b3693 EN |
1674 | struct hashmap_iter iter; |
1675 | struct strmap_entry *e; | |
1676 | int errs = 0; | |
1677 | int original_cache_nr; | |
1678 | ||
9fefce68 EN |
1679 | if (strmap_empty(conflicted)) |
1680 | return 0; | |
1681 | ||
ef2b3693 EN |
1682 | original_cache_nr = index->cache_nr; |
1683 | ||
1684 | /* Put every entry from paths into plist, then sort */ | |
1685 | strmap_for_each_entry(conflicted, &iter, e) { | |
1686 | const char *path = e->key; | |
1687 | struct conflict_info *ci = e->value; | |
1688 | int pos; | |
1689 | struct cache_entry *ce; | |
1690 | int i; | |
1691 | ||
1692 | VERIFY_CI(ci); | |
1693 | ||
1694 | /* | |
1695 | * The index will already have a stage=0 entry for this path, | |
1696 | * because we created an as-merged-as-possible version of the | |
1697 | * file and checkout() moved the working copy and index over | |
1698 | * to that version. | |
1699 | * | |
1700 | * However, previous iterations through this loop will have | |
1701 | * added unstaged entries to the end of the cache which | |
1702 | * ignore the standard alphabetical ordering of cache | |
1703 | * entries and break invariants needed for index_name_pos() | |
1704 | * to work. However, we know the entry we want is before | |
1705 | * those appended cache entries, so do a temporary swap on | |
1706 | * cache_nr to only look through entries of interest. | |
1707 | */ | |
1708 | SWAP(index->cache_nr, original_cache_nr); | |
1709 | pos = index_name_pos(index, path, strlen(path)); | |
1710 | SWAP(index->cache_nr, original_cache_nr); | |
1711 | if (pos < 0) { | |
1712 | if (ci->filemask != 1) | |
1713 | BUG("Conflicted %s but nothing in basic working tree or index; this shouldn't happen", path); | |
1714 | cache_tree_invalidate_path(index, path); | |
1715 | } else { | |
1716 | ce = index->cache[pos]; | |
1717 | ||
1718 | /* | |
1719 | * Clean paths with CE_SKIP_WORKTREE set will not be | |
1720 | * written to the working tree by the unpack_trees() | |
1721 | * call in checkout(). Our conflicted entries would | |
1722 | * have appeared clean to that code since we ignored | |
1723 | * the higher order stages. Thus, we need override | |
1724 | * the CE_SKIP_WORKTREE bit and manually write those | |
1725 | * files to the working disk here. | |
1726 | * | |
1727 | * TODO: Implement this CE_SKIP_WORKTREE fixup. | |
1728 | */ | |
1729 | ||
1730 | /* | |
1731 | * Mark this cache entry for removal and instead add | |
1732 | * new stage>0 entries corresponding to the | |
1733 | * conflicts. If there are many conflicted entries, we | |
1734 | * want to avoid memmove'ing O(NM) entries by | |
1735 | * inserting the new entries one at a time. So, | |
1736 | * instead, we just add the new cache entries to the | |
1737 | * end (ignoring normal index requirements on sort | |
1738 | * order) and sort the index once we're all done. | |
1739 | */ | |
1740 | ce->ce_flags |= CE_REMOVE; | |
1741 | } | |
1742 | ||
1743 | for (i = MERGE_BASE; i <= MERGE_SIDE2; i++) { | |
1744 | struct version_info *vi; | |
1745 | if (!(ci->filemask & (1ul << i))) | |
1746 | continue; | |
1747 | vi = &ci->stages[i]; | |
1748 | ce = make_cache_entry(index, vi->mode, &vi->oid, | |
1749 | path, i+1, 0); | |
1750 | add_index_entry(index, ce, ADD_CACHE_JUST_APPEND); | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | /* | |
1755 | * Remove the unused cache entries (and invalidate the relevant | |
1756 | * cache-trees), then sort the index entries to get the conflicted | |
1757 | * entries we added to the end into their right locations. | |
1758 | */ | |
1759 | remove_marked_cache_entries(index, 1); | |
1760 | QSORT(index->cache, index->cache_nr, cmp_cache_name_compare); | |
1761 | ||
1762 | return errs; | |
9fefce68 EN |
1763 | } |
1764 | ||
17e5574b EN |
1765 | void merge_switch_to_result(struct merge_options *opt, |
1766 | struct tree *head, | |
1767 | struct merge_result *result, | |
1768 | int update_worktree_and_index, | |
1769 | int display_update_msgs) | |
1770 | { | |
9fefce68 EN |
1771 | assert(opt->priv == NULL); |
1772 | if (result->clean >= 0 && update_worktree_and_index) { | |
1773 | struct merge_options_internal *opti = result->priv; | |
1774 | ||
1775 | if (checkout(opt, head, result->tree)) { | |
1776 | /* failure to function */ | |
1777 | result->clean = -1; | |
1778 | return; | |
1779 | } | |
1780 | ||
1781 | if (record_conflicted_index_entries(opt, opt->repo->index, | |
1782 | &opti->paths, | |
1783 | &opti->conflicted)) { | |
1784 | /* failure to function */ | |
1785 | result->clean = -1; | |
1786 | return; | |
1787 | } | |
1788 | } | |
1789 | ||
1790 | if (display_update_msgs) { | |
c5a6f655 EN |
1791 | struct merge_options_internal *opti = result->priv; |
1792 | struct hashmap_iter iter; | |
1793 | struct strmap_entry *e; | |
1794 | struct string_list olist = STRING_LIST_INIT_NODUP; | |
1795 | int i; | |
1796 | ||
1797 | /* Hack to pre-allocate olist to the desired size */ | |
1798 | ALLOC_GROW(olist.items, strmap_get_size(&opti->output), | |
1799 | olist.alloc); | |
1800 | ||
1801 | /* Put every entry from output into olist, then sort */ | |
1802 | strmap_for_each_entry(&opti->output, &iter, e) { | |
1803 | string_list_append(&olist, e->key)->util = e->value; | |
1804 | } | |
1805 | string_list_sort(&olist); | |
1806 | ||
1807 | /* Iterate over the items, printing them */ | |
1808 | for (i = 0; i < olist.nr; ++i) { | |
1809 | struct strbuf *sb = olist.items[i].util; | |
1810 | ||
1811 | printf("%s", sb->buf); | |
1812 | } | |
1813 | string_list_clear(&olist, 0); | |
9fefce68 EN |
1814 | } |
1815 | ||
17e5574b EN |
1816 | merge_finalize(opt, result); |
1817 | } | |
1818 | ||
1819 | void merge_finalize(struct merge_options *opt, | |
1820 | struct merge_result *result) | |
1821 | { | |
89422d29 EN |
1822 | struct merge_options_internal *opti = result->priv; |
1823 | ||
1824 | assert(opt->priv == NULL); | |
1825 | ||
101bc5bc | 1826 | clear_internal_opts(opti, 0); |
89422d29 | 1827 | FREE_AND_NULL(opti); |
17e5574b EN |
1828 | } |
1829 | ||
04af1879 EN |
1830 | /*** Function Grouping: helper functions for merge_incore_*() ***/ |
1831 | ||
231e2dd4 EN |
1832 | static void merge_start(struct merge_options *opt, struct merge_result *result) |
1833 | { | |
e4171b1b EN |
1834 | /* Sanity checks on opt */ |
1835 | assert(opt->repo); | |
1836 | ||
1837 | assert(opt->branch1 && opt->branch2); | |
1838 | ||
1839 | assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE && | |
1840 | opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE); | |
1841 | assert(opt->rename_limit >= -1); | |
1842 | assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE); | |
1843 | assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1); | |
1844 | ||
1845 | assert(opt->xdl_opts >= 0); | |
1846 | assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL && | |
1847 | opt->recursive_variant <= MERGE_VARIANT_THEIRS); | |
1848 | ||
1849 | /* | |
1850 | * detect_renames, verbosity, buffer_output, and obuf are ignored | |
1851 | * fields that were used by "recursive" rather than "ort" -- but | |
1852 | * sanity check them anyway. | |
1853 | */ | |
1854 | assert(opt->detect_renames >= -1 && | |
1855 | opt->detect_renames <= DIFF_DETECT_COPY); | |
1856 | assert(opt->verbosity >= 0 && opt->verbosity <= 5); | |
1857 | assert(opt->buffer_output <= 2); | |
1858 | assert(opt->obuf.len == 0); | |
1859 | ||
1860 | assert(opt->priv == NULL); | |
1861 | ||
c8017176 EN |
1862 | /* Default to histogram diff. Actually, just hardcode it...for now. */ |
1863 | opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF); | |
1864 | ||
e4171b1b EN |
1865 | /* Initialization of opt->priv, our internal merge data */ |
1866 | opt->priv = xcalloc(1, sizeof(*opt->priv)); | |
1867 | ||
1868 | /* | |
1869 | * Although we initialize opt->priv->paths with strdup_strings=0, | |
1870 | * that's just to avoid making yet another copy of an allocated | |
1871 | * string. Putting the entry into paths means we are taking | |
43c1dccb | 1872 | * ownership, so we will later free it. paths_to_free is similar. |
e4171b1b EN |
1873 | * |
1874 | * In contrast, conflicted just has a subset of keys from paths, so | |
1875 | * we don't want to free those (it'd be a duplicate free). | |
1876 | */ | |
1877 | strmap_init_with_options(&opt->priv->paths, NULL, 0); | |
1878 | strmap_init_with_options(&opt->priv->conflicted, NULL, 0); | |
43c1dccb | 1879 | string_list_init(&opt->priv->paths_to_free, 0); |
c5a6f655 EN |
1880 | |
1881 | /* | |
1882 | * keys & strbufs in output will sometimes need to outlive "paths", | |
1883 | * so it will have a copy of relevant keys. It's probably a small | |
1884 | * subset of the overall paths that have special output. | |
1885 | */ | |
1886 | strmap_init(&opt->priv->output); | |
231e2dd4 EN |
1887 | } |
1888 | ||
04af1879 EN |
1889 | /*** Function Grouping: merge_incore_*() and their internal variants ***/ |
1890 | ||
231e2dd4 EN |
1891 | /* |
1892 | * Originally from merge_trees_internal(); heavily adapted, though. | |
1893 | */ | |
1894 | static void merge_ort_nonrecursive_internal(struct merge_options *opt, | |
1895 | struct tree *merge_base, | |
1896 | struct tree *side1, | |
1897 | struct tree *side2, | |
1898 | struct merge_result *result) | |
1899 | { | |
1900 | struct object_id working_tree_oid; | |
1901 | ||
0c0d705b EN |
1902 | if (collect_merge_info(opt, merge_base, side1, side2) != 0) { |
1903 | /* | |
1904 | * TRANSLATORS: The %s arguments are: 1) tree hash of a merge | |
1905 | * base, and 2-3) the trees for the two trees we're merging. | |
1906 | */ | |
1907 | err(opt, _("collecting merge info failed for trees %s, %s, %s"), | |
1908 | oid_to_hex(&merge_base->object.oid), | |
1909 | oid_to_hex(&side1->object.oid), | |
1910 | oid_to_hex(&side2->object.oid)); | |
1911 | result->clean = -1; | |
1912 | return; | |
1913 | } | |
1914 | ||
231e2dd4 EN |
1915 | result->clean = detect_and_process_renames(opt, merge_base, |
1916 | side1, side2); | |
1917 | process_entries(opt, &working_tree_oid); | |
1918 | ||
1919 | /* Set return values */ | |
1920 | result->tree = parse_tree_indirect(&working_tree_oid); | |
1921 | /* existence of conflicted entries implies unclean */ | |
1922 | result->clean &= strmap_empty(&opt->priv->conflicted); | |
1923 | if (!opt->priv->call_depth) { | |
1924 | result->priv = opt->priv; | |
1925 | opt->priv = NULL; | |
1926 | } | |
1927 | } | |
1928 | ||
17e5574b EN |
1929 | void merge_incore_nonrecursive(struct merge_options *opt, |
1930 | struct tree *merge_base, | |
1931 | struct tree *side1, | |
1932 | struct tree *side2, | |
1933 | struct merge_result *result) | |
1934 | { | |
231e2dd4 EN |
1935 | assert(opt->ancestor != NULL); |
1936 | merge_start(opt, result); | |
1937 | merge_ort_nonrecursive_internal(opt, merge_base, side1, side2, result); | |
17e5574b EN |
1938 | } |
1939 | ||
1940 | void merge_incore_recursive(struct merge_options *opt, | |
1941 | struct commit_list *merge_bases, | |
1942 | struct commit *side1, | |
1943 | struct commit *side2, | |
1944 | struct merge_result *result) | |
1945 | { | |
1946 | die("Not yet implemented"); | |
1947 | } |