]> git.ipfire.org Git - thirdparty/git.git/blob - rerere.c
git-submodule.sh: shorten submodule SHA-1s using rev-parse
[thirdparty/git.git] / rerere.c
1 #include "cache.h"
2 #include "config.h"
3 #include "lockfile.h"
4 #include "string-list.h"
5 #include "rerere.h"
6 #include "xdiff-interface.h"
7 #include "dir.h"
8 #include "resolve-undo.h"
9 #include "ll-merge.h"
10 #include "attr.h"
11 #include "pathspec.h"
12 #include "object-store.h"
13 #include "sha1-lookup.h"
14
15 #define RESOLVED 0
16 #define PUNTED 1
17 #define THREE_STAGED 2
18 void *RERERE_RESOLVED = &RERERE_RESOLVED;
19
20 /* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
21 static int rerere_enabled = -1;
22
23 /* automatically update cleanly resolved paths to the index */
24 static int rerere_autoupdate;
25
26 static int rerere_dir_nr;
27 static int rerere_dir_alloc;
28
29 #define RR_HAS_POSTIMAGE 1
30 #define RR_HAS_PREIMAGE 2
31 static struct rerere_dir {
32 unsigned char hash[GIT_MAX_HEXSZ];
33 int status_alloc, status_nr;
34 unsigned char *status;
35 } **rerere_dir;
36
37 static void free_rerere_dirs(void)
38 {
39 int i;
40 for (i = 0; i < rerere_dir_nr; i++) {
41 free(rerere_dir[i]->status);
42 free(rerere_dir[i]);
43 }
44 FREE_AND_NULL(rerere_dir);
45 rerere_dir_nr = rerere_dir_alloc = 0;
46 }
47
48 static void free_rerere_id(struct string_list_item *item)
49 {
50 free(item->util);
51 }
52
53 static const char *rerere_id_hex(const struct rerere_id *id)
54 {
55 return sha1_to_hex(id->collection->hash);
56 }
57
58 static void fit_variant(struct rerere_dir *rr_dir, int variant)
59 {
60 variant++;
61 ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
62 if (rr_dir->status_nr < variant) {
63 memset(rr_dir->status + rr_dir->status_nr,
64 '\0', variant - rr_dir->status_nr);
65 rr_dir->status_nr = variant;
66 }
67 }
68
69 static void assign_variant(struct rerere_id *id)
70 {
71 int variant;
72 struct rerere_dir *rr_dir = id->collection;
73
74 variant = id->variant;
75 if (variant < 0) {
76 for (variant = 0; variant < rr_dir->status_nr; variant++)
77 if (!rr_dir->status[variant])
78 break;
79 }
80 fit_variant(rr_dir, variant);
81 id->variant = variant;
82 }
83
84 const char *rerere_path(const struct rerere_id *id, const char *file)
85 {
86 if (!file)
87 return git_path("rr-cache/%s", rerere_id_hex(id));
88
89 if (id->variant <= 0)
90 return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
91
92 return git_path("rr-cache/%s/%s.%d",
93 rerere_id_hex(id), file, id->variant);
94 }
95
96 static int is_rr_file(const char *name, const char *filename, int *variant)
97 {
98 const char *suffix;
99 char *ep;
100
101 if (!strcmp(name, filename)) {
102 *variant = 0;
103 return 1;
104 }
105 if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
106 return 0;
107
108 errno = 0;
109 *variant = strtol(suffix + 1, &ep, 10);
110 if (errno || *ep)
111 return 0;
112 return 1;
113 }
114
115 static void scan_rerere_dir(struct rerere_dir *rr_dir)
116 {
117 struct dirent *de;
118 DIR *dir = opendir(git_path("rr-cache/%s", sha1_to_hex(rr_dir->hash)));
119
120 if (!dir)
121 return;
122 while ((de = readdir(dir)) != NULL) {
123 int variant;
124
125 if (is_rr_file(de->d_name, "postimage", &variant)) {
126 fit_variant(rr_dir, variant);
127 rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
128 } else if (is_rr_file(de->d_name, "preimage", &variant)) {
129 fit_variant(rr_dir, variant);
130 rr_dir->status[variant] |= RR_HAS_PREIMAGE;
131 }
132 }
133 closedir(dir);
134 }
135
136 static const unsigned char *rerere_dir_hash(size_t i, void *table)
137 {
138 struct rerere_dir **rr_dir = table;
139 return rr_dir[i]->hash;
140 }
141
142 static struct rerere_dir *find_rerere_dir(const char *hex)
143 {
144 unsigned char hash[GIT_MAX_RAWSZ];
145 struct rerere_dir *rr_dir;
146 int pos;
147
148 if (get_sha1_hex(hex, hash))
149 return NULL; /* BUG */
150 pos = sha1_pos(hash, rerere_dir, rerere_dir_nr, rerere_dir_hash);
151 if (pos < 0) {
152 rr_dir = xmalloc(sizeof(*rr_dir));
153 hashcpy(rr_dir->hash, hash);
154 rr_dir->status = NULL;
155 rr_dir->status_nr = 0;
156 rr_dir->status_alloc = 0;
157 pos = -1 - pos;
158
159 /* Make sure the array is big enough ... */
160 ALLOC_GROW(rerere_dir, rerere_dir_nr + 1, rerere_dir_alloc);
161 /* ... and add it in. */
162 rerere_dir_nr++;
163 MOVE_ARRAY(rerere_dir + pos + 1, rerere_dir + pos,
164 rerere_dir_nr - pos - 1);
165 rerere_dir[pos] = rr_dir;
166 scan_rerere_dir(rr_dir);
167 }
168 return rerere_dir[pos];
169 }
170
171 static int has_rerere_resolution(const struct rerere_id *id)
172 {
173 const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
174 int variant = id->variant;
175
176 if (variant < 0)
177 return 0;
178 return ((id->collection->status[variant] & both) == both);
179 }
180
181 static struct rerere_id *new_rerere_id_hex(char *hex)
182 {
183 struct rerere_id *id = xmalloc(sizeof(*id));
184 id->collection = find_rerere_dir(hex);
185 id->variant = -1; /* not known yet */
186 return id;
187 }
188
189 static struct rerere_id *new_rerere_id(unsigned char *sha1)
190 {
191 return new_rerere_id_hex(sha1_to_hex(sha1));
192 }
193
194 /*
195 * $GIT_DIR/MERGE_RR file is a collection of records, each of which is
196 * "conflict ID", a HT and pathname, terminated with a NUL, and is
197 * used to keep track of the set of paths that "rerere" may need to
198 * work on (i.e. what is left by the previous invocation of "git
199 * rerere" during the current conflict resolution session).
200 */
201 static void read_rr(struct string_list *rr)
202 {
203 struct strbuf buf = STRBUF_INIT;
204 FILE *in = fopen_or_warn(git_path_merge_rr(the_repository), "r");
205
206 if (!in)
207 return;
208 while (!strbuf_getwholeline(&buf, in, '\0')) {
209 char *path;
210 unsigned char hash[GIT_MAX_RAWSZ];
211 struct rerere_id *id;
212 int variant;
213 const unsigned hexsz = the_hash_algo->hexsz;
214
215 /* There has to be the hash, tab, path and then NUL */
216 if (buf.len < hexsz + 2 || get_sha1_hex(buf.buf, hash))
217 die(_("corrupt MERGE_RR"));
218
219 if (buf.buf[hexsz] != '.') {
220 variant = 0;
221 path = buf.buf + hexsz;
222 } else {
223 errno = 0;
224 variant = strtol(buf.buf + hexsz + 1, &path, 10);
225 if (errno)
226 die(_("corrupt MERGE_RR"));
227 }
228 if (*(path++) != '\t')
229 die(_("corrupt MERGE_RR"));
230 buf.buf[hexsz] = '\0';
231 id = new_rerere_id_hex(buf.buf);
232 id->variant = variant;
233 string_list_insert(rr, path)->util = id;
234 }
235 strbuf_release(&buf);
236 fclose(in);
237 }
238
239 static struct lock_file write_lock;
240
241 static int write_rr(struct string_list *rr, int out_fd)
242 {
243 int i;
244 for (i = 0; i < rr->nr; i++) {
245 struct strbuf buf = STRBUF_INIT;
246 struct rerere_id *id;
247
248 assert(rr->items[i].util != RERERE_RESOLVED);
249
250 id = rr->items[i].util;
251 if (!id)
252 continue;
253 assert(id->variant >= 0);
254 if (0 < id->variant)
255 strbuf_addf(&buf, "%s.%d\t%s%c",
256 rerere_id_hex(id), id->variant,
257 rr->items[i].string, 0);
258 else
259 strbuf_addf(&buf, "%s\t%s%c",
260 rerere_id_hex(id),
261 rr->items[i].string, 0);
262
263 if (write_in_full(out_fd, buf.buf, buf.len) < 0)
264 die(_("unable to write rerere record"));
265
266 strbuf_release(&buf);
267 }
268 if (commit_lock_file(&write_lock) != 0)
269 die(_("unable to write rerere record"));
270 return 0;
271 }
272
273 /*
274 * "rerere" interacts with conflicted file contents using this I/O
275 * abstraction. It reads a conflicted contents from one place via
276 * "getline()" method, and optionally can write it out after
277 * normalizing the conflicted hunks to the "output". Subclasses of
278 * rerere_io embed this structure at the beginning of their own
279 * rerere_io object.
280 */
281 struct rerere_io {
282 int (*getline)(struct strbuf *, struct rerere_io *);
283 FILE *output;
284 int wrerror;
285 /* some more stuff */
286 };
287
288 static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
289 {
290 if (!count || *err)
291 return;
292 if (fwrite(p, count, 1, fp) != 1)
293 *err = errno;
294 }
295
296 static inline void ferr_puts(const char *s, FILE *fp, int *err)
297 {
298 ferr_write(s, strlen(s), fp, err);
299 }
300
301 static void rerere_io_putstr(const char *str, struct rerere_io *io)
302 {
303 if (io->output)
304 ferr_puts(str, io->output, &io->wrerror);
305 }
306
307 static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
308 {
309 if (io->output)
310 ferr_write(mem, sz, io->output, &io->wrerror);
311 }
312
313 /*
314 * Subclass of rerere_io that reads from an on-disk file
315 */
316 struct rerere_io_file {
317 struct rerere_io io;
318 FILE *input;
319 };
320
321 /*
322 * ... and its getline() method implementation
323 */
324 static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
325 {
326 struct rerere_io_file *io = (struct rerere_io_file *)io_;
327 return strbuf_getwholeline(sb, io->input, '\n');
328 }
329
330 /*
331 * Require the exact number of conflict marker letters, no more, no
332 * less, followed by SP or any whitespace
333 * (including LF).
334 */
335 static int is_cmarker(char *buf, int marker_char, int marker_size)
336 {
337 int want_sp;
338
339 /*
340 * The beginning of our version and the end of their version
341 * always are labeled like "<<<<< ours" or ">>>>> theirs",
342 * hence we set want_sp for them. Note that the version from
343 * the common ancestor in diff3-style output is not always
344 * labelled (e.g. "||||| common" is often seen but "|||||"
345 * alone is also valid), so we do not set want_sp.
346 */
347 want_sp = (marker_char == '<') || (marker_char == '>');
348
349 while (marker_size--)
350 if (*buf++ != marker_char)
351 return 0;
352 if (want_sp && *buf != ' ')
353 return 0;
354 return isspace(*buf);
355 }
356
357 static void rerere_strbuf_putconflict(struct strbuf *buf, int ch, size_t size)
358 {
359 strbuf_addchars(buf, ch, size);
360 strbuf_addch(buf, '\n');
361 }
362
363 static int handle_conflict(struct strbuf *out, struct rerere_io *io,
364 int marker_size, git_hash_ctx *ctx)
365 {
366 enum {
367 RR_SIDE_1 = 0, RR_SIDE_2, RR_ORIGINAL
368 } hunk = RR_SIDE_1;
369 struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
370 struct strbuf buf = STRBUF_INIT, conflict = STRBUF_INIT;
371 int has_conflicts = -1;
372
373 while (!io->getline(&buf, io)) {
374 if (is_cmarker(buf.buf, '<', marker_size)) {
375 if (handle_conflict(&conflict, io, marker_size, NULL) < 0)
376 break;
377 if (hunk == RR_SIDE_1)
378 strbuf_addbuf(&one, &conflict);
379 else
380 strbuf_addbuf(&two, &conflict);
381 strbuf_release(&conflict);
382 } else if (is_cmarker(buf.buf, '|', marker_size)) {
383 if (hunk != RR_SIDE_1)
384 break;
385 hunk = RR_ORIGINAL;
386 } else if (is_cmarker(buf.buf, '=', marker_size)) {
387 if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
388 break;
389 hunk = RR_SIDE_2;
390 } else if (is_cmarker(buf.buf, '>', marker_size)) {
391 if (hunk != RR_SIDE_2)
392 break;
393 if (strbuf_cmp(&one, &two) > 0)
394 strbuf_swap(&one, &two);
395 has_conflicts = 1;
396 rerere_strbuf_putconflict(out, '<', marker_size);
397 strbuf_addbuf(out, &one);
398 rerere_strbuf_putconflict(out, '=', marker_size);
399 strbuf_addbuf(out, &two);
400 rerere_strbuf_putconflict(out, '>', marker_size);
401 if (ctx) {
402 the_hash_algo->update_fn(ctx, one.buf ?
403 one.buf : "",
404 one.len + 1);
405 the_hash_algo->update_fn(ctx, two.buf ?
406 two.buf : "",
407 two.len + 1);
408 }
409 break;
410 } else if (hunk == RR_SIDE_1)
411 strbuf_addbuf(&one, &buf);
412 else if (hunk == RR_ORIGINAL)
413 ; /* discard */
414 else if (hunk == RR_SIDE_2)
415 strbuf_addbuf(&two, &buf);
416 }
417 strbuf_release(&one);
418 strbuf_release(&two);
419 strbuf_release(&buf);
420
421 return has_conflicts;
422 }
423
424 /*
425 * Read contents a file with conflicts, normalize the conflicts
426 * by (1) discarding the common ancestor version in diff3-style,
427 * (2) reordering our side and their side so that whichever sorts
428 * alphabetically earlier comes before the other one, while
429 * computing the "conflict ID", which is just an SHA-1 hash of
430 * one side of the conflict, NUL, the other side of the conflict,
431 * and NUL concatenated together.
432 *
433 * Return 1 if conflict hunks are found, 0 if there are no conflict
434 * hunks and -1 if an error occured.
435 */
436 static int handle_path(unsigned char *hash, struct rerere_io *io, int marker_size)
437 {
438 git_hash_ctx ctx;
439 struct strbuf buf = STRBUF_INIT, out = STRBUF_INIT;
440 int has_conflicts = 0;
441 if (hash)
442 the_hash_algo->init_fn(&ctx);
443
444 while (!io->getline(&buf, io)) {
445 if (is_cmarker(buf.buf, '<', marker_size)) {
446 has_conflicts = handle_conflict(&out, io, marker_size,
447 hash ? &ctx : NULL);
448 if (has_conflicts < 0)
449 break;
450 rerere_io_putmem(out.buf, out.len, io);
451 strbuf_reset(&out);
452 } else
453 rerere_io_putstr(buf.buf, io);
454 }
455 strbuf_release(&buf);
456 strbuf_release(&out);
457
458 if (hash)
459 the_hash_algo->final_fn(hash, &ctx);
460
461 return has_conflicts;
462 }
463
464 /*
465 * Scan the path for conflicts, do the "handle_path()" thing above, and
466 * return the number of conflict hunks found.
467 */
468 static int handle_file(struct index_state *istate,
469 const char *path, unsigned char *hash, const char *output)
470 {
471 int has_conflicts = 0;
472 struct rerere_io_file io;
473 int marker_size = ll_merge_marker_size(istate, path);
474
475 memset(&io, 0, sizeof(io));
476 io.io.getline = rerere_file_getline;
477 io.input = fopen(path, "r");
478 io.io.wrerror = 0;
479 if (!io.input)
480 return error_errno(_("could not open '%s'"), path);
481
482 if (output) {
483 io.io.output = fopen(output, "w");
484 if (!io.io.output) {
485 error_errno(_("could not write '%s'"), output);
486 fclose(io.input);
487 return -1;
488 }
489 }
490
491 has_conflicts = handle_path(hash, (struct rerere_io *)&io, marker_size);
492
493 fclose(io.input);
494 if (io.io.wrerror)
495 error(_("there were errors while writing '%s' (%s)"),
496 path, strerror(io.io.wrerror));
497 if (io.io.output && fclose(io.io.output))
498 io.io.wrerror = error_errno(_("failed to flush '%s'"), path);
499
500 if (has_conflicts < 0) {
501 if (output)
502 unlink_or_warn(output);
503 return error(_("could not parse conflict hunks in '%s'"), path);
504 }
505 if (io.io.wrerror)
506 return -1;
507 return has_conflicts;
508 }
509
510 /*
511 * Look at a cache entry at "i" and see if it is not conflicting,
512 * conflicting and we are willing to handle, or conflicting and
513 * we are unable to handle, and return the determination in *type.
514 * Return the cache index to be looked at next, by skipping the
515 * stages we have already looked at in this invocation of this
516 * function.
517 */
518 static int check_one_conflict(struct index_state *istate, int i, int *type)
519 {
520 const struct cache_entry *e = istate->cache[i];
521
522 if (!ce_stage(e)) {
523 *type = RESOLVED;
524 return i + 1;
525 }
526
527 *type = PUNTED;
528 while (i < istate->cache_nr && ce_stage(istate->cache[i]) == 1)
529 i++;
530
531 /* Only handle regular files with both stages #2 and #3 */
532 if (i + 1 < istate->cache_nr) {
533 const struct cache_entry *e2 = istate->cache[i];
534 const struct cache_entry *e3 = istate->cache[i + 1];
535 if (ce_stage(e2) == 2 &&
536 ce_stage(e3) == 3 &&
537 ce_same_name(e, e3) &&
538 S_ISREG(e2->ce_mode) &&
539 S_ISREG(e3->ce_mode))
540 *type = THREE_STAGED;
541 }
542
543 /* Skip the entries with the same name */
544 while (i < istate->cache_nr && ce_same_name(e, istate->cache[i]))
545 i++;
546 return i;
547 }
548
549 /*
550 * Scan the index and find paths that have conflicts that rerere can
551 * handle, i.e. the ones that has both stages #2 and #3.
552 *
553 * NEEDSWORK: we do not record or replay a previous "resolve by
554 * deletion" for a delete-modify conflict, as that is inherently risky
555 * without knowing what modification is being discarded. The only
556 * safe case, i.e. both side doing the deletion and modification that
557 * are identical to the previous round, might want to be handled,
558 * though.
559 */
560 static int find_conflict(struct repository *r, struct string_list *conflict)
561 {
562 int i;
563
564 if (read_index(r->index) < 0)
565 return error(_("index file corrupt"));
566
567 for (i = 0; i < r->index->cache_nr;) {
568 int conflict_type;
569 const struct cache_entry *e = r->index->cache[i];
570 i = check_one_conflict(r->index, i, &conflict_type);
571 if (conflict_type == THREE_STAGED)
572 string_list_insert(conflict, (const char *)e->name);
573 }
574 return 0;
575 }
576
577 /*
578 * The merge_rr list is meant to hold outstanding conflicted paths
579 * that rerere could handle. Abuse the list by adding other types of
580 * entries to allow the caller to show "rerere remaining".
581 *
582 * - Conflicted paths that rerere does not handle are added
583 * - Conflicted paths that have been resolved are marked as such
584 * by storing RERERE_RESOLVED to .util field (where conflict ID
585 * is expected to be stored).
586 *
587 * Do *not* write MERGE_RR file out after calling this function.
588 *
589 * NEEDSWORK: we may want to fix the caller that implements "rerere
590 * remaining" to do this without abusing merge_rr.
591 */
592 int rerere_remaining(struct repository *r, struct string_list *merge_rr)
593 {
594 int i;
595
596 if (setup_rerere(merge_rr, RERERE_READONLY))
597 return 0;
598 if (read_index(r->index) < 0)
599 return error(_("index file corrupt"));
600
601 for (i = 0; i < r->index->cache_nr;) {
602 int conflict_type;
603 const struct cache_entry *e = r->index->cache[i];
604 i = check_one_conflict(r->index, i, &conflict_type);
605 if (conflict_type == PUNTED)
606 string_list_insert(merge_rr, (const char *)e->name);
607 else if (conflict_type == RESOLVED) {
608 struct string_list_item *it;
609 it = string_list_lookup(merge_rr, (const char *)e->name);
610 if (it != NULL) {
611 free_rerere_id(it);
612 it->util = RERERE_RESOLVED;
613 }
614 }
615 }
616 return 0;
617 }
618
619 /*
620 * Try using the given conflict resolution "ID" to see
621 * if that recorded conflict resolves cleanly what we
622 * got in the "cur".
623 */
624 static int try_merge(struct index_state *istate,
625 const struct rerere_id *id, const char *path,
626 mmfile_t *cur, mmbuffer_t *result)
627 {
628 int ret;
629 mmfile_t base = {NULL, 0}, other = {NULL, 0};
630
631 if (read_mmfile(&base, rerere_path(id, "preimage")) ||
632 read_mmfile(&other, rerere_path(id, "postimage")))
633 ret = 1;
634 else
635 /*
636 * A three-way merge. Note that this honors user-customizable
637 * low-level merge driver settings.
638 */
639 ret = ll_merge(result, path, &base, NULL, cur, "", &other, "",
640 istate, NULL);
641
642 free(base.ptr);
643 free(other.ptr);
644
645 return ret;
646 }
647
648 /*
649 * Find the conflict identified by "id"; the change between its
650 * "preimage" (i.e. a previous contents with conflict markers) and its
651 * "postimage" (i.e. the corresponding contents with conflicts
652 * resolved) may apply cleanly to the contents stored in "path", i.e.
653 * the conflict this time around.
654 *
655 * Returns 0 for successful replay of recorded resolution, or non-zero
656 * for failure.
657 */
658 static int merge(struct index_state *istate, const struct rerere_id *id, const char *path)
659 {
660 FILE *f;
661 int ret;
662 mmfile_t cur = {NULL, 0};
663 mmbuffer_t result = {NULL, 0};
664
665 /*
666 * Normalize the conflicts in path and write it out to
667 * "thisimage" temporary file.
668 */
669 if ((handle_file(istate, path, NULL, rerere_path(id, "thisimage")) < 0) ||
670 read_mmfile(&cur, rerere_path(id, "thisimage"))) {
671 ret = 1;
672 goto out;
673 }
674
675 ret = try_merge(istate, id, path, &cur, &result);
676 if (ret)
677 goto out;
678
679 /*
680 * A successful replay of recorded resolution.
681 * Mark that "postimage" was used to help gc.
682 */
683 if (utime(rerere_path(id, "postimage"), NULL) < 0)
684 warning_errno(_("failed utime() on '%s'"),
685 rerere_path(id, "postimage"));
686
687 /* Update "path" with the resolution */
688 f = fopen(path, "w");
689 if (!f)
690 return error_errno(_("could not open '%s'"), path);
691 if (fwrite(result.ptr, result.size, 1, f) != 1)
692 error_errno(_("could not write '%s'"), path);
693 if (fclose(f))
694 return error_errno(_("writing '%s' failed"), path);
695
696 out:
697 free(cur.ptr);
698 free(result.ptr);
699
700 return ret;
701 }
702
703 static void update_paths(struct repository *r, struct string_list *update)
704 {
705 struct lock_file index_lock = LOCK_INIT;
706 int i;
707
708 hold_locked_index(&index_lock, LOCK_DIE_ON_ERROR);
709
710 for (i = 0; i < update->nr; i++) {
711 struct string_list_item *item = &update->items[i];
712 if (add_file_to_index(r->index, item->string, 0))
713 exit(128);
714 fprintf_ln(stderr, _("Staged '%s' using previous resolution."),
715 item->string);
716 }
717
718 if (write_locked_index(r->index, &index_lock,
719 COMMIT_LOCK | SKIP_IF_UNCHANGED))
720 die(_("unable to write new index file"));
721 }
722
723 static void remove_variant(struct rerere_id *id)
724 {
725 unlink_or_warn(rerere_path(id, "postimage"));
726 unlink_or_warn(rerere_path(id, "preimage"));
727 id->collection->status[id->variant] = 0;
728 }
729
730 /*
731 * The path indicated by rr_item may still have conflict for which we
732 * have a recorded resolution, in which case replay it and optionally
733 * update it. Or it may have been resolved by the user and we may
734 * only have the preimage for that conflict, in which case the result
735 * needs to be recorded as a resolution in a postimage file.
736 */
737 static void do_rerere_one_path(struct index_state *istate,
738 struct string_list_item *rr_item,
739 struct string_list *update)
740 {
741 const char *path = rr_item->string;
742 struct rerere_id *id = rr_item->util;
743 struct rerere_dir *rr_dir = id->collection;
744 int variant;
745
746 variant = id->variant;
747
748 /* Has the user resolved it already? */
749 if (variant >= 0) {
750 if (!handle_file(istate, path, NULL, NULL)) {
751 copy_file(rerere_path(id, "postimage"), path, 0666);
752 id->collection->status[variant] |= RR_HAS_POSTIMAGE;
753 fprintf_ln(stderr, _("Recorded resolution for '%s'."), path);
754 free_rerere_id(rr_item);
755 rr_item->util = NULL;
756 return;
757 }
758 /*
759 * There may be other variants that can cleanly
760 * replay. Try them and update the variant number for
761 * this one.
762 */
763 }
764
765 /* Does any existing resolution apply cleanly? */
766 for (variant = 0; variant < rr_dir->status_nr; variant++) {
767 const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
768 struct rerere_id vid = *id;
769
770 if ((rr_dir->status[variant] & both) != both)
771 continue;
772
773 vid.variant = variant;
774 if (merge(istate, &vid, path))
775 continue; /* failed to replay */
776
777 /*
778 * If there already is a different variant that applies
779 * cleanly, there is no point maintaining our own variant.
780 */
781 if (0 <= id->variant && id->variant != variant)
782 remove_variant(id);
783
784 if (rerere_autoupdate)
785 string_list_insert(update, path);
786 else
787 fprintf_ln(stderr,
788 _("Resolved '%s' using previous resolution."),
789 path);
790 free_rerere_id(rr_item);
791 rr_item->util = NULL;
792 return;
793 }
794
795 /* None of the existing one applies; we need a new variant */
796 assign_variant(id);
797
798 variant = id->variant;
799 handle_file(istate, path, NULL, rerere_path(id, "preimage"));
800 if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
801 const char *path = rerere_path(id, "postimage");
802 if (unlink(path))
803 die_errno(_("cannot unlink stray '%s'"), path);
804 id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
805 }
806 id->collection->status[variant] |= RR_HAS_PREIMAGE;
807 fprintf_ln(stderr, _("Recorded preimage for '%s'"), path);
808 }
809
810 static int do_plain_rerere(struct repository *r,
811 struct string_list *rr, int fd)
812 {
813 struct string_list conflict = STRING_LIST_INIT_DUP;
814 struct string_list update = STRING_LIST_INIT_DUP;
815 int i;
816
817 find_conflict(r, &conflict);
818
819 /*
820 * MERGE_RR records paths with conflicts immediately after
821 * merge failed. Some of the conflicted paths might have been
822 * hand resolved in the working tree since then, but the
823 * initial run would catch all and register their preimages.
824 */
825 for (i = 0; i < conflict.nr; i++) {
826 struct rerere_id *id;
827 unsigned char hash[GIT_MAX_RAWSZ];
828 const char *path = conflict.items[i].string;
829 int ret;
830
831 /*
832 * Ask handle_file() to scan and assign a
833 * conflict ID. No need to write anything out
834 * yet.
835 */
836 ret = handle_file(r->index, path, hash, NULL);
837 if (ret != 0 && string_list_has_string(rr, path)) {
838 remove_variant(string_list_lookup(rr, path)->util);
839 string_list_remove(rr, path, 1);
840 }
841 if (ret < 1)
842 continue;
843
844 id = new_rerere_id(hash);
845 string_list_insert(rr, path)->util = id;
846
847 /* Ensure that the directory exists. */
848 mkdir_in_gitdir(rerere_path(id, NULL));
849 }
850
851 for (i = 0; i < rr->nr; i++)
852 do_rerere_one_path(r->index, &rr->items[i], &update);
853
854 if (update.nr)
855 update_paths(r, &update);
856
857 return write_rr(rr, fd);
858 }
859
860 static void git_rerere_config(void)
861 {
862 git_config_get_bool("rerere.enabled", &rerere_enabled);
863 git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
864 git_config(git_default_config, NULL);
865 }
866
867 static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
868
869 static int is_rerere_enabled(void)
870 {
871 int rr_cache_exists;
872
873 if (!rerere_enabled)
874 return 0;
875
876 rr_cache_exists = is_directory(git_path_rr_cache());
877 if (rerere_enabled < 0)
878 return rr_cache_exists;
879
880 if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
881 die(_("could not create directory '%s'"), git_path_rr_cache());
882 return 1;
883 }
884
885 int setup_rerere(struct string_list *merge_rr, int flags)
886 {
887 int fd;
888
889 git_rerere_config();
890 if (!is_rerere_enabled())
891 return -1;
892
893 if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
894 rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
895 if (flags & RERERE_READONLY)
896 fd = 0;
897 else
898 fd = hold_lock_file_for_update(&write_lock,
899 git_path_merge_rr(the_repository),
900 LOCK_DIE_ON_ERROR);
901 read_rr(merge_rr);
902 return fd;
903 }
904
905 /*
906 * The main entry point that is called internally from codepaths that
907 * perform mergy operations, possibly leaving conflicted index entries
908 * and working tree files.
909 */
910 int repo_rerere(struct repository *r, int flags)
911 {
912 struct string_list merge_rr = STRING_LIST_INIT_DUP;
913 int fd, status;
914
915 fd = setup_rerere(&merge_rr, flags);
916 if (fd < 0)
917 return 0;
918 status = do_plain_rerere(r, &merge_rr, fd);
919 free_rerere_dirs();
920 return status;
921 }
922
923 /*
924 * Subclass of rerere_io that reads from an in-core buffer that is a
925 * strbuf
926 */
927 struct rerere_io_mem {
928 struct rerere_io io;
929 struct strbuf input;
930 };
931
932 /*
933 * ... and its getline() method implementation
934 */
935 static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
936 {
937 struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
938 char *ep;
939 size_t len;
940
941 strbuf_release(sb);
942 if (!io->input.len)
943 return -1;
944 ep = memchr(io->input.buf, '\n', io->input.len);
945 if (!ep)
946 ep = io->input.buf + io->input.len;
947 else if (*ep == '\n')
948 ep++;
949 len = ep - io->input.buf;
950 strbuf_add(sb, io->input.buf, len);
951 strbuf_remove(&io->input, 0, len);
952 return 0;
953 }
954
955 static int handle_cache(struct index_state *istate,
956 const char *path, unsigned char *hash, const char *output)
957 {
958 mmfile_t mmfile[3] = {{NULL}};
959 mmbuffer_t result = {NULL, 0};
960 const struct cache_entry *ce;
961 int pos, len, i, has_conflicts;
962 struct rerere_io_mem io;
963 int marker_size = ll_merge_marker_size(istate, path);
964
965 /*
966 * Reproduce the conflicted merge in-core
967 */
968 len = strlen(path);
969 pos = index_name_pos(istate, path, len);
970 if (0 <= pos)
971 return -1;
972 pos = -pos - 1;
973
974 while (pos < istate->cache_nr) {
975 enum object_type type;
976 unsigned long size;
977
978 ce = istate->cache[pos++];
979 if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
980 break;
981 i = ce_stage(ce) - 1;
982 if (!mmfile[i].ptr) {
983 mmfile[i].ptr = read_object_file(&ce->oid, &type,
984 &size);
985 mmfile[i].size = size;
986 }
987 }
988 for (i = 0; i < 3; i++)
989 if (!mmfile[i].ptr && !mmfile[i].size)
990 mmfile[i].ptr = xstrdup("");
991
992 /*
993 * NEEDSWORK: handle conflicts from merges with
994 * merge.renormalize set, too?
995 */
996 ll_merge(&result, path, &mmfile[0], NULL,
997 &mmfile[1], "ours",
998 &mmfile[2], "theirs",
999 istate, NULL);
1000 for (i = 0; i < 3; i++)
1001 free(mmfile[i].ptr);
1002
1003 memset(&io, 0, sizeof(io));
1004 io.io.getline = rerere_mem_getline;
1005 if (output)
1006 io.io.output = fopen(output, "w");
1007 else
1008 io.io.output = NULL;
1009 strbuf_init(&io.input, 0);
1010 strbuf_attach(&io.input, result.ptr, result.size, result.size);
1011
1012 /*
1013 * Grab the conflict ID and optionally write the original
1014 * contents with conflict markers out.
1015 */
1016 has_conflicts = handle_path(hash, (struct rerere_io *)&io, marker_size);
1017 strbuf_release(&io.input);
1018 if (io.io.output)
1019 fclose(io.io.output);
1020 return has_conflicts;
1021 }
1022
1023 static int rerere_forget_one_path(struct index_state *istate,
1024 const char *path,
1025 struct string_list *rr)
1026 {
1027 const char *filename;
1028 struct rerere_id *id;
1029 unsigned char hash[GIT_MAX_RAWSZ];
1030 int ret;
1031 struct string_list_item *item;
1032
1033 /*
1034 * Recreate the original conflict from the stages in the
1035 * index and compute the conflict ID
1036 */
1037 ret = handle_cache(istate, path, hash, NULL);
1038 if (ret < 1)
1039 return error(_("could not parse conflict hunks in '%s'"), path);
1040
1041 /* Nuke the recorded resolution for the conflict */
1042 id = new_rerere_id(hash);
1043
1044 for (id->variant = 0;
1045 id->variant < id->collection->status_nr;
1046 id->variant++) {
1047 mmfile_t cur = { NULL, 0 };
1048 mmbuffer_t result = {NULL, 0};
1049 int cleanly_resolved;
1050
1051 if (!has_rerere_resolution(id))
1052 continue;
1053
1054 handle_cache(istate, path, hash, rerere_path(id, "thisimage"));
1055 if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1056 free(cur.ptr);
1057 error(_("failed to update conflicted state in '%s'"), path);
1058 goto fail_exit;
1059 }
1060 cleanly_resolved = !try_merge(istate, id, path, &cur, &result);
1061 free(result.ptr);
1062 free(cur.ptr);
1063 if (cleanly_resolved)
1064 break;
1065 }
1066
1067 if (id->collection->status_nr <= id->variant) {
1068 error(_("no remembered resolution for '%s'"), path);
1069 goto fail_exit;
1070 }
1071
1072 filename = rerere_path(id, "postimage");
1073 if (unlink(filename)) {
1074 if (errno == ENOENT)
1075 error(_("no remembered resolution for '%s'"), path);
1076 else
1077 error_errno(_("cannot unlink '%s'"), filename);
1078 goto fail_exit;
1079 }
1080
1081 /*
1082 * Update the preimage so that the user can resolve the
1083 * conflict in the working tree, run us again to record
1084 * the postimage.
1085 */
1086 handle_cache(istate, path, hash, rerere_path(id, "preimage"));
1087 fprintf_ln(stderr, _("Updated preimage for '%s'"), path);
1088
1089 /*
1090 * And remember that we can record resolution for this
1091 * conflict when the user is done.
1092 */
1093 item = string_list_insert(rr, path);
1094 free_rerere_id(item);
1095 item->util = id;
1096 fprintf(stderr, _("Forgot resolution for '%s'\n"), path);
1097 return 0;
1098
1099 fail_exit:
1100 free(id);
1101 return -1;
1102 }
1103
1104 int rerere_forget(struct repository *r, struct pathspec *pathspec)
1105 {
1106 int i, fd;
1107 struct string_list conflict = STRING_LIST_INIT_DUP;
1108 struct string_list merge_rr = STRING_LIST_INIT_DUP;
1109
1110 if (read_index(r->index) < 0)
1111 return error(_("index file corrupt"));
1112
1113 fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
1114 if (fd < 0)
1115 return 0;
1116
1117 /*
1118 * The paths may have been resolved (incorrectly);
1119 * recover the original conflicted state and then
1120 * find the conflicted paths.
1121 */
1122 unmerge_index(r->index, pathspec);
1123 find_conflict(r, &conflict);
1124 for (i = 0; i < conflict.nr; i++) {
1125 struct string_list_item *it = &conflict.items[i];
1126 if (!match_pathspec(r->index, pathspec, it->string,
1127 strlen(it->string), 0, NULL, 0))
1128 continue;
1129 rerere_forget_one_path(r->index, it->string, &merge_rr);
1130 }
1131 return write_rr(&merge_rr, fd);
1132 }
1133
1134 /*
1135 * Garbage collection support
1136 */
1137
1138 static timestamp_t rerere_created_at(struct rerere_id *id)
1139 {
1140 struct stat st;
1141
1142 return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1143 }
1144
1145 static timestamp_t rerere_last_used_at(struct rerere_id *id)
1146 {
1147 struct stat st;
1148
1149 return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1150 }
1151
1152 /*
1153 * Remove the recorded resolution for a given conflict ID
1154 */
1155 static void unlink_rr_item(struct rerere_id *id)
1156 {
1157 unlink_or_warn(rerere_path(id, "thisimage"));
1158 remove_variant(id);
1159 id->collection->status[id->variant] = 0;
1160 }
1161
1162 static void prune_one(struct rerere_id *id,
1163 timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1164 {
1165 timestamp_t then;
1166 timestamp_t cutoff;
1167
1168 then = rerere_last_used_at(id);
1169 if (then)
1170 cutoff = cutoff_resolve;
1171 else {
1172 then = rerere_created_at(id);
1173 if (!then)
1174 return;
1175 cutoff = cutoff_noresolve;
1176 }
1177 if (then < cutoff)
1178 unlink_rr_item(id);
1179 }
1180
1181 void rerere_gc(struct string_list *rr)
1182 {
1183 struct string_list to_remove = STRING_LIST_INIT_DUP;
1184 DIR *dir;
1185 struct dirent *e;
1186 int i;
1187 timestamp_t now = time(NULL);
1188 timestamp_t cutoff_noresolve = now - 15 * 86400;
1189 timestamp_t cutoff_resolve = now - 60 * 86400;
1190
1191 if (setup_rerere(rr, 0) < 0)
1192 return;
1193
1194 git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
1195 git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
1196 git_config(git_default_config, NULL);
1197 dir = opendir(git_path("rr-cache"));
1198 if (!dir)
1199 die_errno(_("unable to open rr-cache directory"));
1200 /* Collect stale conflict IDs ... */
1201 while ((e = readdir(dir))) {
1202 struct rerere_dir *rr_dir;
1203 struct rerere_id id;
1204 int now_empty;
1205
1206 if (is_dot_or_dotdot(e->d_name))
1207 continue;
1208 rr_dir = find_rerere_dir(e->d_name);
1209 if (!rr_dir)
1210 continue; /* or should we remove e->d_name? */
1211
1212 now_empty = 1;
1213 for (id.variant = 0, id.collection = rr_dir;
1214 id.variant < id.collection->status_nr;
1215 id.variant++) {
1216 prune_one(&id, cutoff_resolve, cutoff_noresolve);
1217 if (id.collection->status[id.variant])
1218 now_empty = 0;
1219 }
1220 if (now_empty)
1221 string_list_append(&to_remove, e->d_name);
1222 }
1223 closedir(dir);
1224
1225 /* ... and then remove the empty directories */
1226 for (i = 0; i < to_remove.nr; i++)
1227 rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
1228 string_list_clear(&to_remove, 0);
1229 rollback_lock_file(&write_lock);
1230 }
1231
1232 /*
1233 * During a conflict resolution, after "rerere" recorded the
1234 * preimages, abandon them if the user did not resolve them or
1235 * record their resolutions. And drop $GIT_DIR/MERGE_RR.
1236 *
1237 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1238 */
1239 void rerere_clear(struct string_list *merge_rr)
1240 {
1241 int i;
1242
1243 if (setup_rerere(merge_rr, 0) < 0)
1244 return;
1245
1246 for (i = 0; i < merge_rr->nr; i++) {
1247 struct rerere_id *id = merge_rr->items[i].util;
1248 if (!has_rerere_resolution(id)) {
1249 unlink_rr_item(id);
1250 rmdir(rerere_path(id, NULL));
1251 }
1252 }
1253 unlink_or_warn(git_path_merge_rr(the_repository));
1254 rollback_lock_file(&write_lock);
1255 }