git

1
#define USE_THE_REPOSITORY_VARIABLE
2

3
#include "git-compat-util.h"
4
#include "abspath.h"
5
#include "config.h"
6
#include "copy.h"
7
#include "gettext.h"
8
#include "hex.h"
9
#include "lockfile.h"
10
#include "string-list.h"
11
#include "read-cache-ll.h"
12
#include "rerere.h"
13
#include "xdiff-interface.h"
14
#include "dir.h"
15
#include "resolve-undo.h"
16
#include "merge-ll.h"
17
#include "path.h"
18
#include "pathspec.h"
19
#include "object-file.h"
20
#include "object-store-ll.h"
21
#include "strmap.h"
22

23
#define RESOLVED 0
24
#define PUNTED 1
25
#define THREE_STAGED 2
26
void *RERERE_RESOLVED = &RERERE_RESOLVED;
27

28
/* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
29
static int rerere_enabled = -1;
30

31
/* automatically update cleanly resolved paths to the index */
32
static int rerere_autoupdate;
33

34
#define RR_HAS_POSTIMAGE 1
35
#define RR_HAS_PREIMAGE 2
36
struct rerere_dir {
37
	int status_alloc, status_nr;
38
	unsigned char *status;
39
	char name[FLEX_ARRAY];
40
};
41

42
static struct strmap rerere_dirs = STRMAP_INIT;
43

44
static void free_rerere_dirs(void)
45
{
46
	struct hashmap_iter iter;
47
	struct strmap_entry *ent;
48

49
	strmap_for_each_entry(&rerere_dirs, &iter, ent) {
50
		struct rerere_dir *rr_dir = ent->value;
51
		free(rr_dir->status);
52
		free(rr_dir);
53
	}
54
	strmap_clear(&rerere_dirs, 0);
55
}
56

57
static void free_rerere_id(struct string_list_item *item)
58
{
59
	free(item->util);
60
}
61

62
static const char *rerere_id_hex(const struct rerere_id *id)
63
{
64
	return id->collection->name;
65
}
66

67
static void fit_variant(struct rerere_dir *rr_dir, int variant)
68
{
69
	variant++;
70
	ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
71
	if (rr_dir->status_nr < variant) {
72
		memset(rr_dir->status + rr_dir->status_nr,
73
		       '\0', variant - rr_dir->status_nr);
74
		rr_dir->status_nr = variant;
75
	}
76
}
77

78
static void assign_variant(struct rerere_id *id)
79
{
80
	int variant;
81
	struct rerere_dir *rr_dir = id->collection;
82

83
	variant = id->variant;
84
	if (variant < 0) {
85
		for (variant = 0; variant < rr_dir->status_nr; variant++)
86
			if (!rr_dir->status[variant])
87
				break;
88
	}
89
	fit_variant(rr_dir, variant);
90
	id->variant = variant;
91
}
92

93
const char *rerere_path(const struct rerere_id *id, const char *file)
94
{
95
	if (!file)
96
		return git_path("rr-cache/%s", rerere_id_hex(id));
97

98
	if (id->variant <= 0)
99
		return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
100

101
	return git_path("rr-cache/%s/%s.%d",
102
			rerere_id_hex(id), file, id->variant);
103
}
104

105
static int is_rr_file(const char *name, const char *filename, int *variant)
106
{
107
	const char *suffix;
108
	char *ep;
109

110
	if (!strcmp(name, filename)) {
111
		*variant = 0;
112
		return 1;
113
	}
114
	if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
115
		return 0;
116

117
	errno = 0;
118
	*variant = strtol(suffix + 1, &ep, 10);
119
	if (errno || *ep)
120
		return 0;
121
	return 1;
122
}
123

124
static void scan_rerere_dir(struct rerere_dir *rr_dir)
125
{
126
	struct dirent *de;
127
	DIR *dir = opendir(git_path("rr-cache/%s", rr_dir->name));
128

129
	if (!dir)
130
		return;
131
	while ((de = readdir(dir)) != NULL) {
132
		int variant;
133

134
		if (is_rr_file(de->d_name, "postimage", &variant)) {
135
			fit_variant(rr_dir, variant);
136
			rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
137
		} else if (is_rr_file(de->d_name, "preimage", &variant)) {
138
			fit_variant(rr_dir, variant);
139
			rr_dir->status[variant] |= RR_HAS_PREIMAGE;
140
		}
141
	}
142
	closedir(dir);
143
}
144

145
static struct rerere_dir *find_rerere_dir(const char *hex)
146
{
147
	struct rerere_dir *rr_dir;
148

149
	rr_dir = strmap_get(&rerere_dirs, hex);
150
	if (!rr_dir) {
151
		FLEX_ALLOC_STR(rr_dir, name, hex);
152
		rr_dir->status = NULL;
153
		rr_dir->status_nr = 0;
154
		rr_dir->status_alloc = 0;
155
		strmap_put(&rerere_dirs, hex, rr_dir);
156

157
		scan_rerere_dir(rr_dir);
158
	}
159
	return rr_dir;
160
}
161

162
static int has_rerere_resolution(const struct rerere_id *id)
163
{
164
	const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
165
	int variant = id->variant;
166

167
	if (variant < 0)
168
		return 0;
169
	return ((id->collection->status[variant] & both) == both);
170
}
171

172
static struct rerere_id *new_rerere_id_hex(char *hex)
173
{
174
	struct rerere_id *id = xmalloc(sizeof(*id));
175
	id->collection = find_rerere_dir(hex);
176
	id->variant = -1; /* not known yet */
177
	return id;
178
}
179

180
static struct rerere_id *new_rerere_id(unsigned char *hash)
181
{
182
	return new_rerere_id_hex(hash_to_hex(hash));
183
}
184

185
/*
186
 * $GIT_DIR/MERGE_RR file is a collection of records, each of which is
187
 * "conflict ID", a HT and pathname, terminated with a NUL, and is
188
 * used to keep track of the set of paths that "rerere" may need to
189
 * work on (i.e. what is left by the previous invocation of "git
190
 * rerere" during the current conflict resolution session).
191
 */
192
static void read_rr(struct repository *r, struct string_list *rr)
193
{
194
	struct strbuf buf = STRBUF_INIT;
195
	FILE *in = fopen_or_warn(git_path_merge_rr(r), "r");
196

197
	if (!in)
198
		return;
199
	while (!strbuf_getwholeline(&buf, in, '\0')) {
200
		char *path;
201
		unsigned char hash[GIT_MAX_RAWSZ];
202
		struct rerere_id *id;
203
		int variant;
204
		const unsigned hexsz = the_hash_algo->hexsz;
205

206
		/* There has to be the hash, tab, path and then NUL */
207
		if (buf.len < hexsz + 2 || get_hash_hex(buf.buf, hash))
208
			die(_("corrupt MERGE_RR"));
209

210
		if (buf.buf[hexsz] != '.') {
211
			variant = 0;
212
			path = buf.buf + hexsz;
213
		} else {
214
			errno = 0;
215
			variant = strtol(buf.buf + hexsz + 1, &path, 10);
216
			if (errno)
217
				die(_("corrupt MERGE_RR"));
218
		}
219
		if (*(path++) != '\t')
220
			die(_("corrupt MERGE_RR"));
221
		buf.buf[hexsz] = '\0';
222
		id = new_rerere_id_hex(buf.buf);
223
		id->variant = variant;
224
		/*
225
		 * make sure id->collection->status has enough space
226
		 * for the variant we are interested in
227
		 */
228
		fit_variant(id->collection, variant);
229
		string_list_insert(rr, path)->util = id;
230
	}
231
	strbuf_release(&buf);
232
	fclose(in);
233
}
234

235
static struct lock_file write_lock;
236

237
static int write_rr(struct string_list *rr, int out_fd)
238
{
239
	int i;
240
	for (i = 0; i < rr->nr; i++) {
241
		struct strbuf buf = STRBUF_INIT;
242
		struct rerere_id *id;
243

244
		assert(rr->items[i].util != RERERE_RESOLVED);
245

246
		id = rr->items[i].util;
247
		if (!id)
248
			continue;
249
		assert(id->variant >= 0);
250
		if (0 < id->variant)
251
			strbuf_addf(&buf, "%s.%d\t%s%c",
252
				    rerere_id_hex(id), id->variant,
253
				    rr->items[i].string, 0);
254
		else
255
			strbuf_addf(&buf, "%s\t%s%c",
256
				    rerere_id_hex(id),
257
				    rr->items[i].string, 0);
258

259
		if (write_in_full(out_fd, buf.buf, buf.len) < 0)
260
			die(_("unable to write rerere record"));
261

262
		strbuf_release(&buf);
263
	}
264
	if (commit_lock_file(&write_lock) != 0)
265
		die(_("unable to write rerere record"));
266
	return 0;
267
}
268

269
/*
270
 * "rerere" interacts with conflicted file contents using this I/O
271
 * abstraction.  It reads a conflicted contents from one place via
272
 * "getline()" method, and optionally can write it out after
273
 * normalizing the conflicted hunks to the "output".  Subclasses of
274
 * rerere_io embed this structure at the beginning of their own
275
 * rerere_io object.
276
 */
277
struct rerere_io {
278
	int (*getline)(struct strbuf *, struct rerere_io *);
279
	FILE *output;
280
	int wrerror;
281
	/* some more stuff */
282
};
283

284
static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
285
{
286
	if (!count || *err)
287
		return;
288
	if (fwrite(p, count, 1, fp) != 1)
289
		*err = errno;
290
}
291

292
static inline void ferr_puts(const char *s, FILE *fp, int *err)
293
{
294
	ferr_write(s, strlen(s), fp, err);
295
}
296

297
static void rerere_io_putstr(const char *str, struct rerere_io *io)
298
{
299
	if (io->output)
300
		ferr_puts(str, io->output, &io->wrerror);
301
}
302

303
static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
304
{
305
	if (io->output)
306
		ferr_write(mem, sz, io->output, &io->wrerror);
307
}
308

309
/*
310
 * Subclass of rerere_io that reads from an on-disk file
311
 */
312
struct rerere_io_file {
313
	struct rerere_io io;
314
	FILE *input;
315
};
316

317
/*
318
 * ... and its getline() method implementation
319
 */
320
static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
321
{
322
	struct rerere_io_file *io = (struct rerere_io_file *)io_;
323
	return strbuf_getwholeline(sb, io->input, '\n');
324
}
325

326
/*
327
 * Require the exact number of conflict marker letters, no more, no
328
 * less, followed by SP or any whitespace
329
 * (including LF).
330
 */
331
static int is_cmarker(char *buf, int marker_char, int marker_size)
332
{
333
	int want_sp;
334

335
	/*
336
	 * The beginning of our version and the end of their version
337
	 * always are labeled like "<<<<< ours" or ">>>>> theirs",
338
	 * hence we set want_sp for them.  Note that the version from
339
	 * the common ancestor in diff3-style output is not always
340
	 * labelled (e.g. "||||| common" is often seen but "|||||"
341
	 * alone is also valid), so we do not set want_sp.
342
	 */
343
	want_sp = (marker_char == '<') || (marker_char == '>');
344

345
	while (marker_size--)
346
		if (*buf++ != marker_char)
347
			return 0;
348
	if (want_sp && *buf != ' ')
349
		return 0;
350
	return isspace(*buf);
351
}
352

353
static void rerere_strbuf_putconflict(struct strbuf *buf, int ch, size_t size)
354
{
355
	strbuf_addchars(buf, ch, size);
356
	strbuf_addch(buf, '\n');
357
}
358

359
static int handle_conflict(struct strbuf *out, struct rerere_io *io,
360
			   int marker_size, git_hash_ctx *ctx)
361
{
362
	enum {
363
		RR_SIDE_1 = 0, RR_SIDE_2, RR_ORIGINAL
364
	} hunk = RR_SIDE_1;
365
	struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
366
	struct strbuf buf = STRBUF_INIT, conflict = STRBUF_INIT;
367
	int has_conflicts = -1;
368

369
	while (!io->getline(&buf, io)) {
370
		if (is_cmarker(buf.buf, '<', marker_size)) {
371
			if (handle_conflict(&conflict, io, marker_size, NULL) < 0)
372
				break;
373
			if (hunk == RR_SIDE_1)
374
				strbuf_addbuf(&one, &conflict);
375
			else
376
				strbuf_addbuf(&two, &conflict);
377
			strbuf_release(&conflict);
378
		} else if (is_cmarker(buf.buf, '|', marker_size)) {
379
			if (hunk != RR_SIDE_1)
380
				break;
381
			hunk = RR_ORIGINAL;
382
		} else if (is_cmarker(buf.buf, '=', marker_size)) {
383
			if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
384
				break;
385
			hunk = RR_SIDE_2;
386
		} else if (is_cmarker(buf.buf, '>', marker_size)) {
387
			if (hunk != RR_SIDE_2)
388
				break;
389
			if (strbuf_cmp(&one, &two) > 0)
390
				strbuf_swap(&one, &two);
391
			has_conflicts = 1;
392
			rerere_strbuf_putconflict(out, '<', marker_size);
393
			strbuf_addbuf(out, &one);
394
			rerere_strbuf_putconflict(out, '=', marker_size);
395
			strbuf_addbuf(out, &two);
396
			rerere_strbuf_putconflict(out, '>', marker_size);
397
			if (ctx) {
398
				the_hash_algo->update_fn(ctx, one.buf ?
399
							 one.buf : "",
400
							 one.len + 1);
401
				the_hash_algo->update_fn(ctx, two.buf ?
402
							 two.buf : "",
403
							 two.len + 1);
404
			}
405
			break;
406
		} else if (hunk == RR_SIDE_1)
407
			strbuf_addbuf(&one, &buf);
408
		else if (hunk == RR_ORIGINAL)
409
			; /* discard */
410
		else if (hunk == RR_SIDE_2)
411
			strbuf_addbuf(&two, &buf);
412
	}
413
	strbuf_release(&one);
414
	strbuf_release(&two);
415
	strbuf_release(&buf);
416

417
	return has_conflicts;
418
}
419

420
/*
421
 * Read contents a file with conflicts, normalize the conflicts
422
 * by (1) discarding the common ancestor version in diff3-style,
423
 * (2) reordering our side and their side so that whichever sorts
424
 * alphabetically earlier comes before the other one, while
425
 * computing the "conflict ID", which is just an SHA-1 hash of
426
 * one side of the conflict, NUL, the other side of the conflict,
427
 * and NUL concatenated together.
428
 *
429
 * Return 1 if conflict hunks are found, 0 if there are no conflict
430
 * hunks and -1 if an error occurred.
431
 */
432
static int handle_path(unsigned char *hash, struct rerere_io *io, int marker_size)
433
{
434
	git_hash_ctx ctx;
435
	struct strbuf buf = STRBUF_INIT, out = STRBUF_INIT;
436
	int has_conflicts = 0;
437
	if (hash)
438
		the_hash_algo->init_fn(&ctx);
439

440
	while (!io->getline(&buf, io)) {
441
		if (is_cmarker(buf.buf, '<', marker_size)) {
442
			has_conflicts = handle_conflict(&out, io, marker_size,
443
							hash ? &ctx : NULL);
444
			if (has_conflicts < 0)
445
				break;
446
			rerere_io_putmem(out.buf, out.len, io);
447
			strbuf_reset(&out);
448
		} else
449
			rerere_io_putstr(buf.buf, io);
450
	}
451
	strbuf_release(&buf);
452
	strbuf_release(&out);
453

454
	if (hash)
455
		the_hash_algo->final_fn(hash, &ctx);
456

457
	return has_conflicts;
458
}
459

460
/*
461
 * Scan the path for conflicts, do the "handle_path()" thing above, and
462
 * return the number of conflict hunks found.
463
 */
464
static int handle_file(struct index_state *istate,
465
		       const char *path, unsigned char *hash, const char *output)
466
{
467
	int has_conflicts = 0;
468
	struct rerere_io_file io;
469
	int marker_size = ll_merge_marker_size(istate, path);
470

471
	memset(&io, 0, sizeof(io));
472
	io.io.getline = rerere_file_getline;
473
	io.input = fopen(path, "r");
474
	io.io.wrerror = 0;
475
	if (!io.input)
476
		return error_errno(_("could not open '%s'"), path);
477

478
	if (output) {
479
		io.io.output = fopen(output, "w");
480
		if (!io.io.output) {
481
			error_errno(_("could not write '%s'"), output);
482
			fclose(io.input);
483
			return -1;
484
		}
485
	}
486

487
	has_conflicts = handle_path(hash, (struct rerere_io *)&io, marker_size);
488

489
	fclose(io.input);
490
	if (io.io.wrerror)
491
		error(_("there were errors while writing '%s' (%s)"),
492
		      path, strerror(io.io.wrerror));
493
	if (io.io.output && fclose(io.io.output))
494
		io.io.wrerror = error_errno(_("failed to flush '%s'"), path);
495

496
	if (has_conflicts < 0) {
497
		if (output)
498
			unlink_or_warn(output);
499
		return error(_("could not parse conflict hunks in '%s'"), path);
500
	}
501
	if (io.io.wrerror)
502
		return -1;
503
	return has_conflicts;
504
}
505

506
/*
507
 * Look at a cache entry at "i" and see if it is not conflicting,
508
 * conflicting and we are willing to handle, or conflicting and
509
 * we are unable to handle, and return the determination in *type.
510
 * Return the cache index to be looked at next, by skipping the
511
 * stages we have already looked at in this invocation of this
512
 * function.
513
 */
514
static int check_one_conflict(struct index_state *istate, int i, int *type)
515
{
516
	const struct cache_entry *e = istate->cache[i];
517

518
	if (!ce_stage(e)) {
519
		*type = RESOLVED;
520
		return i + 1;
521
	}
522

523
	*type = PUNTED;
524
	while (i < istate->cache_nr && ce_stage(istate->cache[i]) == 1)
525
		i++;
526

527
	/* Only handle regular files with both stages #2 and #3 */
528
	if (i + 1 < istate->cache_nr) {
529
		const struct cache_entry *e2 = istate->cache[i];
530
		const struct cache_entry *e3 = istate->cache[i + 1];
531
		if (ce_stage(e2) == 2 &&
532
		    ce_stage(e3) == 3 &&
533
		    ce_same_name(e, e3) &&
534
		    S_ISREG(e2->ce_mode) &&
535
		    S_ISREG(e3->ce_mode))
536
			*type = THREE_STAGED;
537
	}
538

539
	/* Skip the entries with the same name */
540
	while (i < istate->cache_nr && ce_same_name(e, istate->cache[i]))
541
		i++;
542
	return i;
543
}
544

545
/*
546
 * Scan the index and find paths that have conflicts that rerere can
547
 * handle, i.e. the ones that has both stages #2 and #3.
548
 *
549
 * NEEDSWORK: we do not record or replay a previous "resolve by
550
 * deletion" for a delete-modify conflict, as that is inherently risky
551
 * without knowing what modification is being discarded.  The only
552
 * safe case, i.e. both side doing the deletion and modification that
553
 * are identical to the previous round, might want to be handled,
554
 * though.
555
 */
556
static int find_conflict(struct repository *r, struct string_list *conflict)
557
{
558
	int i;
559

560
	if (repo_read_index(r) < 0)
561
		return error(_("index file corrupt"));
562

563
	for (i = 0; i < r->index->cache_nr;) {
564
		int conflict_type;
565
		const struct cache_entry *e = r->index->cache[i];
566
		i = check_one_conflict(r->index, i, &conflict_type);
567
		if (conflict_type == THREE_STAGED)
568
			string_list_insert(conflict, (const char *)e->name);
569
	}
570
	return 0;
571
}
572

573
/*
574
 * The merge_rr list is meant to hold outstanding conflicted paths
575
 * that rerere could handle.  Abuse the list by adding other types of
576
 * entries to allow the caller to show "rerere remaining".
577
 *
578
 * - Conflicted paths that rerere does not handle are added
579
 * - Conflicted paths that have been resolved are marked as such
580
 *   by storing RERERE_RESOLVED to .util field (where conflict ID
581
 *   is expected to be stored).
582
 *
583
 * Do *not* write MERGE_RR file out after calling this function.
584
 *
585
 * NEEDSWORK: we may want to fix the caller that implements "rerere
586
 * remaining" to do this without abusing merge_rr.
587
 */
588
int rerere_remaining(struct repository *r, struct string_list *merge_rr)
589
{
590
	int i;
591

592
	if (setup_rerere(r, merge_rr, RERERE_READONLY))
593
		return 0;
594
	if (repo_read_index(r) < 0)
595
		return error(_("index file corrupt"));
596

597
	for (i = 0; i < r->index->cache_nr;) {
598
		int conflict_type;
599
		const struct cache_entry *e = r->index->cache[i];
600
		i = check_one_conflict(r->index, i, &conflict_type);
601
		if (conflict_type == PUNTED)
602
			string_list_insert(merge_rr, (const char *)e->name);
603
		else if (conflict_type == RESOLVED) {
604
			struct string_list_item *it;
605
			it = string_list_lookup(merge_rr, (const char *)e->name);
606
			if (it) {
607
				free_rerere_id(it);
608
				it->util = RERERE_RESOLVED;
609
			}
610
		}
611
	}
612
	return 0;
613
}
614

615
/*
616
 * Try using the given conflict resolution "ID" to see
617
 * if that recorded conflict resolves cleanly what we
618
 * got in the "cur".
619
 */
620
static int try_merge(struct index_state *istate,
621
		     const struct rerere_id *id, const char *path,
622
		     mmfile_t *cur, mmbuffer_t *result)
623
{
624
	enum ll_merge_result ret;
625
	mmfile_t base = {NULL, 0}, other = {NULL, 0};
626

627
	if (read_mmfile(&base, rerere_path(id, "preimage")) ||
628
	    read_mmfile(&other, rerere_path(id, "postimage"))) {
629
		ret = LL_MERGE_CONFLICT;
630
	} else {
631
		/*
632
		 * A three-way merge. Note that this honors user-customizable
633
		 * low-level merge driver settings.
634
		 */
635
		ret = ll_merge(result, path, &base, NULL, cur, "", &other, "",
636
			       istate, NULL);
637
	}
638

639
	free(base.ptr);
640
	free(other.ptr);
641

642
	return ret;
643
}
644

645
/*
646
 * Find the conflict identified by "id"; the change between its
647
 * "preimage" (i.e. a previous contents with conflict markers) and its
648
 * "postimage" (i.e. the corresponding contents with conflicts
649
 * resolved) may apply cleanly to the contents stored in "path", i.e.
650
 * the conflict this time around.
651
 *
652
 * Returns 0 for successful replay of recorded resolution, or non-zero
653
 * for failure.
654
 */
655
static int merge(struct index_state *istate, const struct rerere_id *id, const char *path)
656
{
657
	FILE *f;
658
	int ret;
659
	mmfile_t cur = {NULL, 0};
660
	mmbuffer_t result = {NULL, 0};
661

662
	/*
663
	 * Normalize the conflicts in path and write it out to
664
	 * "thisimage" temporary file.
665
	 */
666
	if ((handle_file(istate, path, NULL, rerere_path(id, "thisimage")) < 0) ||
667
	    read_mmfile(&cur, rerere_path(id, "thisimage"))) {
668
		ret = 1;
669
		goto out;
670
	}
671

672
	ret = try_merge(istate, id, path, &cur, &result);
673
	if (ret)
674
		goto out;
675

676
	/*
677
	 * A successful replay of recorded resolution.
678
	 * Mark that "postimage" was used to help gc.
679
	 */
680
	if (utime(rerere_path(id, "postimage"), NULL) < 0)
681
		warning_errno(_("failed utime() on '%s'"),
682
			      rerere_path(id, "postimage"));
683

684
	/* Update "path" with the resolution */
685
	f = fopen(path, "w");
686
	if (!f)
687
		return error_errno(_("could not open '%s'"), path);
688
	if (fwrite(result.ptr, result.size, 1, f) != 1)
689
		error_errno(_("could not write '%s'"), path);
690
	if (fclose(f))
691
		return error_errno(_("writing '%s' failed"), path);
692

693
out:
694
	free(cur.ptr);
695
	free(result.ptr);
696

697
	return ret;
698
}
699

700
static void update_paths(struct repository *r, struct string_list *update)
701
{
702
	struct lock_file index_lock = LOCK_INIT;
703
	int i;
704

705
	repo_hold_locked_index(r, &index_lock, LOCK_DIE_ON_ERROR);
706

707
	for (i = 0; i < update->nr; i++) {
708
		struct string_list_item *item = &update->items[i];
709
		if (add_file_to_index(r->index, item->string, 0))
710
			exit(128);
711
		fprintf_ln(stderr, _("Staged '%s' using previous resolution."),
712
			item->string);
713
	}
714

715
	if (write_locked_index(r->index, &index_lock,
716
			       COMMIT_LOCK | SKIP_IF_UNCHANGED))
717
		die(_("unable to write new index file"));
718
}
719

720
static void remove_variant(struct rerere_id *id)
721
{
722
	unlink_or_warn(rerere_path(id, "postimage"));
723
	unlink_or_warn(rerere_path(id, "preimage"));
724
	id->collection->status[id->variant] = 0;
725
}
726

727
/*
728
 * The path indicated by rr_item may still have conflict for which we
729
 * have a recorded resolution, in which case replay it and optionally
730
 * update it.  Or it may have been resolved by the user and we may
731
 * only have the preimage for that conflict, in which case the result
732
 * needs to be recorded as a resolution in a postimage file.
733
 */
734
static void do_rerere_one_path(struct index_state *istate,
735
			       struct string_list_item *rr_item,
736
			       struct string_list *update)
737
{
738
	const char *path = rr_item->string;
739
	struct rerere_id *id = rr_item->util;
740
	struct rerere_dir *rr_dir = id->collection;
741
	int variant;
742

743
	variant = id->variant;
744

745
	/* Has the user resolved it already? */
746
	if (variant >= 0) {
747
		if (!handle_file(istate, path, NULL, NULL)) {
748
			copy_file(rerere_path(id, "postimage"), path, 0666);
749
			id->collection->status[variant] |= RR_HAS_POSTIMAGE;
750
			fprintf_ln(stderr, _("Recorded resolution for '%s'."), path);
751
			free_rerere_id(rr_item);
752
			rr_item->util = NULL;
753
			return;
754
		}
755
		/*
756
		 * There may be other variants that can cleanly
757
		 * replay.  Try them and update the variant number for
758
		 * this one.
759
		 */
760
	}
761

762
	/* Does any existing resolution apply cleanly? */
763
	for (variant = 0; variant < rr_dir->status_nr; variant++) {
764
		const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
765
		struct rerere_id vid = *id;
766

767
		if ((rr_dir->status[variant] & both) != both)
768
			continue;
769

770
		vid.variant = variant;
771
		if (merge(istate, &vid, path))
772
			continue; /* failed to replay */
773

774
		/*
775
		 * If there already is a different variant that applies
776
		 * cleanly, there is no point maintaining our own variant.
777
		 */
778
		if (0 <= id->variant && id->variant != variant)
779
			remove_variant(id);
780

781
		if (rerere_autoupdate)
782
			string_list_insert(update, path);
783
		else
784
			fprintf_ln(stderr,
785
				   _("Resolved '%s' using previous resolution."),
786
				   path);
787
		free_rerere_id(rr_item);
788
		rr_item->util = NULL;
789
		return;
790
	}
791

792
	/* None of the existing one applies; we need a new variant */
793
	assign_variant(id);
794

795
	variant = id->variant;
796
	handle_file(istate, path, NULL, rerere_path(id, "preimage"));
797
	if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
798
		const char *path = rerere_path(id, "postimage");
799
		if (unlink(path))
800
			die_errno(_("cannot unlink stray '%s'"), path);
801
		id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
802
	}
803
	id->collection->status[variant] |= RR_HAS_PREIMAGE;
804
	fprintf_ln(stderr, _("Recorded preimage for '%s'"), path);
805
}
806

807
static int do_plain_rerere(struct repository *r,
808
			   struct string_list *rr, int fd)
809
{
810
	struct string_list conflict = STRING_LIST_INIT_DUP;
811
	struct string_list update = STRING_LIST_INIT_DUP;
812
	int i;
813

814
	find_conflict(r, &conflict);
815

816
	/*
817
	 * MERGE_RR records paths with conflicts immediately after
818
	 * merge failed.  Some of the conflicted paths might have been
819
	 * hand resolved in the working tree since then, but the
820
	 * initial run would catch all and register their preimages.
821
	 */
822
	for (i = 0; i < conflict.nr; i++) {
823
		struct rerere_id *id;
824
		unsigned char hash[GIT_MAX_RAWSZ];
825
		const char *path = conflict.items[i].string;
826
		int ret;
827

828
		/*
829
		 * Ask handle_file() to scan and assign a
830
		 * conflict ID.  No need to write anything out
831
		 * yet.
832
		 */
833
		ret = handle_file(r->index, path, hash, NULL);
834
		if (ret != 0 && string_list_has_string(rr, path)) {
835
			remove_variant(string_list_lookup(rr, path)->util);
836
			string_list_remove(rr, path, 1);
837
		}
838
		if (ret < 1)
839
			continue;
840

841
		id = new_rerere_id(hash);
842
		string_list_insert(rr, path)->util = id;
843

844
		/* Ensure that the directory exists. */
845
		mkdir_in_gitdir(rerere_path(id, NULL));
846
	}
847

848
	for (i = 0; i < rr->nr; i++)
849
		do_rerere_one_path(r->index, &rr->items[i], &update);
850

851
	if (update.nr)
852
		update_paths(r, &update);
853

854
	string_list_clear(&conflict, 0);
855
	string_list_clear(&update, 0);
856
	return write_rr(rr, fd);
857
}
858

859
static void git_rerere_config(void)
860
{
861
	git_config_get_bool("rerere.enabled", &rerere_enabled);
862
	git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
863
	git_config(git_default_config, NULL);
864
}
865

866
static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
867

868
static int is_rerere_enabled(void)
869
{
870
	int rr_cache_exists;
871

872
	if (!rerere_enabled)
873
		return 0;
874

875
	rr_cache_exists = is_directory(git_path_rr_cache());
876
	if (rerere_enabled < 0)
877
		return rr_cache_exists;
878

879
	if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
880
		die(_("could not create directory '%s'"), git_path_rr_cache());
881
	return 1;
882
}
883

884
int setup_rerere(struct repository *r, struct string_list *merge_rr, int flags)
885
{
886
	int fd;
887

888
	git_rerere_config();
889
	if (!is_rerere_enabled())
890
		return -1;
891

892
	if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
893
		rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
894
	if (flags & RERERE_READONLY)
895
		fd = 0;
896
	else
897
		fd = hold_lock_file_for_update(&write_lock,
898
					       git_path_merge_rr(r),
899
					       LOCK_DIE_ON_ERROR);
900
	read_rr(r, merge_rr);
901
	return fd;
902
}
903

904
/*
905
 * The main entry point that is called internally from codepaths that
906
 * perform mergy operations, possibly leaving conflicted index entries
907
 * and working tree files.
908
 */
909
int repo_rerere(struct repository *r, int flags)
910
{
911
	struct string_list merge_rr = STRING_LIST_INIT_DUP;
912
	int fd, status;
913

914
	fd = setup_rerere(r, &merge_rr, flags);
915
	if (fd < 0)
916
		return 0;
917
	status = do_plain_rerere(r, &merge_rr, fd);
918
	free_rerere_dirs();
919
	string_list_clear(&merge_rr, 1);
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 = repo_read_object_file(the_repository,
984
							      &ce->oid, &type,
985
							      &size);
986
			if (!mmfile[i].ptr)
987
				die(_("unable to read %s"),
988
				    oid_to_hex(&ce->oid));
989
			mmfile[i].size = size;
990
		}
991
	}
992
	for (i = 0; i < 3; i++)
993
		if (!mmfile[i].ptr && !mmfile[i].size)
994
			mmfile[i].ptr = xstrdup("");
995

996
	/*
997
	 * NEEDSWORK: handle conflicts from merges with
998
	 * merge.renormalize set, too?
999
	 */
1000
	ll_merge(&result, path, &mmfile[0], NULL,
1001
		 &mmfile[1], "ours",
1002
		 &mmfile[2], "theirs",
1003
		 istate, NULL);
1004
	for (i = 0; i < 3; i++)
1005
		free(mmfile[i].ptr);
1006

1007
	memset(&io, 0, sizeof(io));
1008
	io.io.getline = rerere_mem_getline;
1009
	if (output)
1010
		io.io.output = fopen(output, "w");
1011
	else
1012
		io.io.output = NULL;
1013
	strbuf_init(&io.input, 0);
1014
	strbuf_attach(&io.input, result.ptr, result.size, result.size);
1015

1016
	/*
1017
	 * Grab the conflict ID and optionally write the original
1018
	 * contents with conflict markers out.
1019
	 */
1020
	has_conflicts = handle_path(hash, (struct rerere_io *)&io, marker_size);
1021
	strbuf_release(&io.input);
1022
	if (io.io.output)
1023
		fclose(io.io.output);
1024
	return has_conflicts;
1025
}
1026

1027
static int rerere_forget_one_path(struct index_state *istate,
1028
				  const char *path,
1029
				  struct string_list *rr)
1030
{
1031
	const char *filename;
1032
	struct rerere_id *id;
1033
	unsigned char hash[GIT_MAX_RAWSZ];
1034
	int ret;
1035
	struct string_list_item *item;
1036

1037
	/*
1038
	 * Recreate the original conflict from the stages in the
1039
	 * index and compute the conflict ID
1040
	 */
1041
	ret = handle_cache(istate, path, hash, NULL);
1042
	if (ret < 1)
1043
		return error(_("could not parse conflict hunks in '%s'"), path);
1044

1045
	/* Nuke the recorded resolution for the conflict */
1046
	id = new_rerere_id(hash);
1047

1048
	for (id->variant = 0;
1049
	     id->variant < id->collection->status_nr;
1050
	     id->variant++) {
1051
		mmfile_t cur = { NULL, 0 };
1052
		mmbuffer_t result = {NULL, 0};
1053
		int cleanly_resolved;
1054

1055
		if (!has_rerere_resolution(id))
1056
			continue;
1057

1058
		handle_cache(istate, path, hash, rerere_path(id, "thisimage"));
1059
		if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1060
			free(cur.ptr);
1061
			error(_("failed to update conflicted state in '%s'"), path);
1062
			goto fail_exit;
1063
		}
1064
		cleanly_resolved = !try_merge(istate, id, path, &cur, &result);
1065
		free(result.ptr);
1066
		free(cur.ptr);
1067
		if (cleanly_resolved)
1068
			break;
1069
	}
1070

1071
	if (id->collection->status_nr <= id->variant) {
1072
		error(_("no remembered resolution for '%s'"), path);
1073
		goto fail_exit;
1074
	}
1075

1076
	filename = rerere_path(id, "postimage");
1077
	if (unlink(filename)) {
1078
		if (errno == ENOENT)
1079
			error(_("no remembered resolution for '%s'"), path);
1080
		else
1081
			error_errno(_("cannot unlink '%s'"), filename);
1082
		goto fail_exit;
1083
	}
1084

1085
	/*
1086
	 * Update the preimage so that the user can resolve the
1087
	 * conflict in the working tree, run us again to record
1088
	 * the postimage.
1089
	 */
1090
	handle_cache(istate, path, hash, rerere_path(id, "preimage"));
1091
	fprintf_ln(stderr, _("Updated preimage for '%s'"), path);
1092

1093
	/*
1094
	 * And remember that we can record resolution for this
1095
	 * conflict when the user is done.
1096
	 */
1097
	item = string_list_insert(rr, path);
1098
	free_rerere_id(item);
1099
	item->util = id;
1100
	fprintf(stderr, _("Forgot resolution for '%s'\n"), path);
1101
	return 0;
1102

1103
fail_exit:
1104
	free(id);
1105
	return -1;
1106
}
1107

1108
int rerere_forget(struct repository *r, struct pathspec *pathspec)
1109
{
1110
	int i, fd, ret;
1111
	struct string_list conflict = STRING_LIST_INIT_DUP;
1112
	struct string_list merge_rr = STRING_LIST_INIT_DUP;
1113

1114
	if (repo_read_index(r) < 0)
1115
		return error(_("index file corrupt"));
1116

1117
	fd = setup_rerere(r, &merge_rr, RERERE_NOAUTOUPDATE);
1118
	if (fd < 0)
1119
		return 0;
1120

1121
	/*
1122
	 * The paths may have been resolved (incorrectly);
1123
	 * recover the original conflicted state and then
1124
	 * find the conflicted paths.
1125
	 */
1126
	unmerge_index(r->index, pathspec, 0);
1127
	find_conflict(r, &conflict);
1128
	for (i = 0; i < conflict.nr; i++) {
1129
		struct string_list_item *it = &conflict.items[i];
1130
		if (!match_pathspec(r->index, pathspec, it->string,
1131
				    strlen(it->string), 0, NULL, 0))
1132
			continue;
1133
		rerere_forget_one_path(r->index, it->string, &merge_rr);
1134
	}
1135

1136
	ret = write_rr(&merge_rr, fd);
1137

1138
	string_list_clear(&conflict, 0);
1139
	string_list_clear(&merge_rr, 1);
1140
	return ret;
1141
}
1142

1143
/*
1144
 * Garbage collection support
1145
 */
1146

1147
static timestamp_t rerere_created_at(struct rerere_id *id)
1148
{
1149
	struct stat st;
1150

1151
	return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1152
}
1153

1154
static timestamp_t rerere_last_used_at(struct rerere_id *id)
1155
{
1156
	struct stat st;
1157

1158
	return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1159
}
1160

1161
/*
1162
 * Remove the recorded resolution for a given conflict ID
1163
 */
1164
static void unlink_rr_item(struct rerere_id *id)
1165
{
1166
	unlink_or_warn(rerere_path(id, "thisimage"));
1167
	remove_variant(id);
1168
	id->collection->status[id->variant] = 0;
1169
}
1170

1171
static void prune_one(struct rerere_id *id,
1172
		      timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1173
{
1174
	timestamp_t then;
1175
	timestamp_t cutoff;
1176

1177
	then = rerere_last_used_at(id);
1178
	if (then)
1179
		cutoff = cutoff_resolve;
1180
	else {
1181
		then = rerere_created_at(id);
1182
		if (!then)
1183
			return;
1184
		cutoff = cutoff_noresolve;
1185
	}
1186
	if (then < cutoff)
1187
		unlink_rr_item(id);
1188
}
1189

1190
/* Does the basename in "path" look plausibly like an rr-cache entry? */
1191
static int is_rr_cache_dirname(const char *path)
1192
{
1193
	struct object_id oid;
1194
	const char *end;
1195
	return !parse_oid_hex(path, &oid, &end) && !*end;
1196
}
1197

1198
void rerere_gc(struct repository *r, struct string_list *rr)
1199
{
1200
	struct string_list to_remove = STRING_LIST_INIT_DUP;
1201
	DIR *dir;
1202
	struct dirent *e;
1203
	int i;
1204
	timestamp_t now = time(NULL);
1205
	timestamp_t cutoff_noresolve = now - 15 * 86400;
1206
	timestamp_t cutoff_resolve = now - 60 * 86400;
1207

1208
	if (setup_rerere(r, rr, 0) < 0)
1209
		return;
1210

1211
	repo_config_get_expiry_in_days(the_repository, "gc.rerereresolved",
1212
				       &cutoff_resolve, now);
1213
	repo_config_get_expiry_in_days(the_repository, "gc.rerereunresolved",
1214
				       &cutoff_noresolve, now);
1215
	git_config(git_default_config, NULL);
1216
	dir = opendir(git_path("rr-cache"));
1217
	if (!dir)
1218
		die_errno(_("unable to open rr-cache directory"));
1219
	/* Collect stale conflict IDs ... */
1220
	while ((e = readdir_skip_dot_and_dotdot(dir))) {
1221
		struct rerere_dir *rr_dir;
1222
		struct rerere_id id;
1223
		int now_empty;
1224

1225
		if (!is_rr_cache_dirname(e->d_name))
1226
			continue; /* or should we remove e->d_name? */
1227

1228
		rr_dir = find_rerere_dir(e->d_name);
1229

1230
		now_empty = 1;
1231
		for (id.variant = 0, id.collection = rr_dir;
1232
		     id.variant < id.collection->status_nr;
1233
		     id.variant++) {
1234
			prune_one(&id, cutoff_resolve, cutoff_noresolve);
1235
			if (id.collection->status[id.variant])
1236
				now_empty = 0;
1237
		}
1238
		if (now_empty)
1239
			string_list_append(&to_remove, e->d_name);
1240
	}
1241
	closedir(dir);
1242

1243
	/* ... and then remove the empty directories */
1244
	for (i = 0; i < to_remove.nr; i++)
1245
		rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
1246
	string_list_clear(&to_remove, 0);
1247
	rollback_lock_file(&write_lock);
1248
}
1249

1250
/*
1251
 * During a conflict resolution, after "rerere" recorded the
1252
 * preimages, abandon them if the user did not resolve them or
1253
 * record their resolutions.  And drop $GIT_DIR/MERGE_RR.
1254
 *
1255
 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1256
 */
1257
void rerere_clear(struct repository *r, struct string_list *merge_rr)
1258
{
1259
	int i;
1260

1261
	if (setup_rerere(r, merge_rr, 0) < 0)
1262
		return;
1263

1264
	for (i = 0; i < merge_rr->nr; i++) {
1265
		struct rerere_id *id = merge_rr->items[i].util;
1266
		if (!has_rerere_resolution(id)) {
1267
			unlink_rr_item(id);
1268
			rmdir(rerere_path(id, NULL));
1269
		}
1270
	}
1271
	unlink_or_warn(git_path_merge_rr(r));
1272
	rollback_lock_file(&write_lock);
1273
}
1274