rerere: avoid buffer overrun
[git/git.git] / rerere.c
CommitLineData
5b2fd956 1#include "cache.h"
b2141fc1 2#include "config.h"
697cc8ef 3#include "lockfile.h"
c455c87c 4#include "string-list.h"
5b2fd956 5#include "rerere.h"
5b2fd956 6#include "xdiff-interface.h"
dea4562b
JH
7#include "dir.h"
8#include "resolve-undo.h"
9#include "ll-merge.h"
8588567c 10#include "attr.h"
01a10b0a 11#include "pathspec.h"
1869bbe1 12#include "sha1-lookup.h"
5b2fd956 13
ac49f5ca
MZ
14#define RESOLVED 0
15#define PUNTED 1
16#define THREE_STAGED 2
17void *RERERE_RESOLVED = &RERERE_RESOLVED;
18
5b2fd956
SB
19/* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
20static int rerere_enabled = -1;
21
22/* automatically update cleanly resolved paths to the index */
23static int rerere_autoupdate;
24
1869bbe1
JH
25static int rerere_dir_nr;
26static int rerere_dir_alloc;
27
2c7929b1
JH
28#define RR_HAS_POSTIMAGE 1
29#define RR_HAS_PREIMAGE 2
1869bbe1
JH
30static struct rerere_dir {
31 unsigned char sha1[20];
a13d1370
JH
32 int status_alloc, status_nr;
33 unsigned char *status;
1869bbe1
JH
34} **rerere_dir;
35
36static void free_rerere_dirs(void)
37{
38 int i;
a13d1370
JH
39 for (i = 0; i < rerere_dir_nr; i++) {
40 free(rerere_dir[i]->status);
1869bbe1 41 free(rerere_dir[i]);
a13d1370 42 }
88ce3ef6 43 FREE_AND_NULL(rerere_dir);
1869bbe1 44 rerere_dir_nr = rerere_dir_alloc = 0;
1869bbe1
JH
45}
46
1d51eced 47static void free_rerere_id(struct string_list_item *item)
5b2fd956 48{
1d51eced 49 free(item->util);
5b2fd956
SB
50}
51
1d51eced
JH
52static const char *rerere_id_hex(const struct rerere_id *id)
53{
1869bbe1 54 return sha1_to_hex(id->collection->sha1);
1d51eced
JH
55}
56
a13d1370
JH
57static void fit_variant(struct rerere_dir *rr_dir, int variant)
58{
59 variant++;
60 ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
61 if (rr_dir->status_nr < variant) {
62 memset(rr_dir->status + rr_dir->status_nr,
63 '\0', variant - rr_dir->status_nr);
64 rr_dir->status_nr = variant;
65 }
66}
67
68static void assign_variant(struct rerere_id *id)
69{
70 int variant;
71 struct rerere_dir *rr_dir = id->collection;
72
73 variant = id->variant;
74 if (variant < 0) {
629716d2
JH
75 for (variant = 0; variant < rr_dir->status_nr; variant++)
76 if (!rr_dir->status[variant])
77 break;
a13d1370
JH
78 }
79 fit_variant(rr_dir, variant);
80 id->variant = variant;
1d51eced
JH
81}
82
83const char *rerere_path(const struct rerere_id *id, const char *file)
84{
85 if (!file)
86 return git_path("rr-cache/%s", rerere_id_hex(id));
87
a13d1370
JH
88 if (id->variant <= 0)
89 return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
90
91 return git_path("rr-cache/%s/%s.%d",
92 rerere_id_hex(id), file, id->variant);
5b2fd956
SB
93}
94
a13d1370 95static int is_rr_file(const char *name, const char *filename, int *variant)
2c7929b1 96{
a13d1370
JH
97 const char *suffix;
98 char *ep;
99
100 if (!strcmp(name, filename)) {
101 *variant = 0;
102 return 1;
103 }
104 if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
105 return 0;
106
107 errno = 0;
108 *variant = strtol(suffix + 1, &ep, 10);
109 if (errno || *ep)
110 return 0;
111 return 1;
2c7929b1
JH
112}
113
114static void scan_rerere_dir(struct rerere_dir *rr_dir)
115{
116 struct dirent *de;
117 DIR *dir = opendir(git_path("rr-cache/%s", sha1_to_hex(rr_dir->sha1)));
118
119 if (!dir)
120 return;
121 while ((de = readdir(dir)) != NULL) {
a13d1370
JH
122 int variant;
123
124 if (is_rr_file(de->d_name, "postimage", &variant)) {
125 fit_variant(rr_dir, variant);
126 rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
127 } else if (is_rr_file(de->d_name, "preimage", &variant)) {
128 fit_variant(rr_dir, variant);
129 rr_dir->status[variant] |= RR_HAS_PREIMAGE;
130 }
2c7929b1
JH
131 }
132 closedir(dir);
133}
134
1869bbe1
JH
135static const unsigned char *rerere_dir_sha1(size_t i, void *table)
136{
137 struct rerere_dir **rr_dir = table;
138 return rr_dir[i]->sha1;
139}
140
141static struct rerere_dir *find_rerere_dir(const char *hex)
142{
143 unsigned char sha1[20];
144 struct rerere_dir *rr_dir;
145 int pos;
146
147 if (get_sha1_hex(hex, sha1))
148 return NULL; /* BUG */
149 pos = sha1_pos(sha1, rerere_dir, rerere_dir_nr, rerere_dir_sha1);
150 if (pos < 0) {
151 rr_dir = xmalloc(sizeof(*rr_dir));
152 hashcpy(rr_dir->sha1, sha1);
a13d1370
JH
153 rr_dir->status = NULL;
154 rr_dir->status_nr = 0;
155 rr_dir->status_alloc = 0;
1869bbe1
JH
156 pos = -1 - pos;
157
158 /* Make sure the array is big enough ... */
159 ALLOC_GROW(rerere_dir, rerere_dir_nr + 1, rerere_dir_alloc);
160 /* ... and add it in. */
161 rerere_dir_nr++;
f919ffeb
SG
162 MOVE_ARRAY(rerere_dir + pos + 1, rerere_dir + pos,
163 rerere_dir_nr - pos - 1);
1869bbe1 164 rerere_dir[pos] = rr_dir;
2c7929b1 165 scan_rerere_dir(rr_dir);
1869bbe1
JH
166 }
167 return rerere_dir[pos];
5b2fd956
SB
168}
169
1d51eced 170static int has_rerere_resolution(const struct rerere_id *id)
5b2fd956 171{
05dd9f13 172 const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
a13d1370 173 int variant = id->variant;
1d51eced 174
a13d1370
JH
175 if (variant < 0)
176 return 0;
177 return ((id->collection->status[variant] & both) == both);
5b2fd956
SB
178}
179
1d51eced
JH
180static struct rerere_id *new_rerere_id_hex(char *hex)
181{
182 struct rerere_id *id = xmalloc(sizeof(*id));
1869bbe1 183 id->collection = find_rerere_dir(hex);
a13d1370 184 id->variant = -1; /* not known yet */
1d51eced
JH
185 return id;
186}
187
188static struct rerere_id *new_rerere_id(unsigned char *sha1)
189{
190 return new_rerere_id_hex(sha1_to_hex(sha1));
5b2fd956
SB
191}
192
4b68c2a0
JH
193/*
194 * $GIT_DIR/MERGE_RR file is a collection of records, each of which is
195 * "conflict ID", a HT and pathname, terminated with a NUL, and is
196 * used to keep track of the set of paths that "rerere" may need to
197 * work on (i.e. what is left by the previous invocation of "git
198 * rerere" during the current conflict resolution session).
199 */
c455c87c 200static void read_rr(struct string_list *rr)
5b2fd956 201{
f5800f6a 202 struct strbuf buf = STRBUF_INIT;
e9d983f1 203 FILE *in = fopen_or_warn(git_path_merge_rr(), "r");
f5800f6a 204
5b2fd956
SB
205 if (!in)
206 return;
f5800f6a
JH
207 while (!strbuf_getwholeline(&buf, in, '\0')) {
208 char *path;
209 unsigned char sha1[20];
1d51eced 210 struct rerere_id *id;
a13d1370 211 int variant;
f5800f6a
JH
212
213 /* There has to be the hash, tab, path and then NUL */
214 if (buf.len < 42 || get_sha1_hex(buf.buf, sha1))
5b2fd956 215 die("corrupt MERGE_RR");
f5800f6a 216
a13d1370
JH
217 if (buf.buf[40] != '.') {
218 variant = 0;
219 path = buf.buf + 40;
220 } else {
221 errno = 0;
222 variant = strtol(buf.buf + 41, &path, 10);
223 if (errno)
224 die("corrupt MERGE_RR");
225 }
226 if (*(path++) != '\t')
5b2fd956 227 die("corrupt MERGE_RR");
f5800f6a 228 buf.buf[40] = '\0';
1d51eced 229 id = new_rerere_id_hex(buf.buf);
a13d1370 230 id->variant = variant;
1d51eced 231 string_list_insert(rr, path)->util = id;
5b2fd956 232 }
f5800f6a 233 strbuf_release(&buf);
5b2fd956
SB
234 fclose(in);
235}
236
237static struct lock_file write_lock;
238
c455c87c 239static int write_rr(struct string_list *rr, int out_fd)
5b2fd956
SB
240{
241 int i;
242 for (i = 0; i < rr->nr; i++) {
e2cb6a95 243 struct strbuf buf = STRBUF_INIT;
1d51eced 244 struct rerere_id *id;
e2cb6a95
JH
245
246 assert(rr->items[i].util != RERERE_RESOLVED);
1d51eced
JH
247
248 id = rr->items[i].util;
249 if (!id)
5b2fd956 250 continue;
a13d1370
JH
251 assert(id->variant >= 0);
252 if (0 < id->variant)
253 strbuf_addf(&buf, "%s.%d\t%s%c",
254 rerere_id_hex(id), id->variant,
255 rr->items[i].string, 0);
256 else
257 strbuf_addf(&buf, "%s\t%s%c",
258 rerere_id_hex(id),
259 rr->items[i].string, 0);
260
06f46f23 261 if (write_in_full(out_fd, buf.buf, buf.len) < 0)
5b2fd956 262 die("unable to write rerere record");
e2cb6a95
JH
263
264 strbuf_release(&buf);
5b2fd956
SB
265 }
266 if (commit_lock_file(&write_lock) != 0)
267 die("unable to write rerere record");
268 return 0;
269}
270
a96847cc
JH
271/*
272 * "rerere" interacts with conflicted file contents using this I/O
273 * abstraction. It reads a conflicted contents from one place via
274 * "getline()" method, and optionally can write it out after
275 * normalizing the conflicted hunks to the "output". Subclasses of
276 * rerere_io embed this structure at the beginning of their own
277 * rerere_io object.
278 */
279struct rerere_io {
280 int (*getline)(struct strbuf *, struct rerere_io *);
281 FILE *output;
282 int wrerror;
283 /* some more stuff */
284};
285
47d32af2
AR
286static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
287{
288 if (!count || *err)
289 return;
290 if (fwrite(p, count, 1, fp) != 1)
291 *err = errno;
292}
293
294static inline void ferr_puts(const char *s, FILE *fp, int *err)
295{
296 ferr_write(s, strlen(s), fp, err);
297}
298
27d6b085
JH
299static void rerere_io_putstr(const char *str, struct rerere_io *io)
300{
301 if (io->output)
302 ferr_puts(str, io->output, &io->wrerror);
303}
304
a96847cc
JH
305/*
306 * Write a conflict marker to io->output (if defined).
307 */
191f2417
JH
308static void rerere_io_putconflict(int ch, int size, struct rerere_io *io)
309{
310 char buf[64];
311
312 while (size) {
d3c2749d 313 if (size <= sizeof(buf) - 2) {
191f2417
JH
314 memset(buf, ch, size);
315 buf[size] = '\n';
316 buf[size + 1] = '\0';
317 size = 0;
318 } else {
319 int sz = sizeof(buf) - 1;
d3c2749d
JH
320
321 /*
322 * Make sure we will not write everything out
323 * in this round by leaving at least 1 byte
324 * for the next round, giving the next round
325 * a chance to add the terminating LF. Yuck.
326 */
191f2417
JH
327 if (size <= sz)
328 sz -= (sz - size) + 1;
329 memset(buf, ch, sz);
330 buf[sz] = '\0';
331 size -= sz;
332 }
333 rerere_io_putstr(buf, io);
334 }
335}
336
27d6b085
JH
337static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
338{
339 if (io->output)
340 ferr_write(mem, sz, io->output, &io->wrerror);
341}
342
a96847cc
JH
343/*
344 * Subclass of rerere_io that reads from an on-disk file
345 */
27d6b085
JH
346struct rerere_io_file {
347 struct rerere_io io;
348 FILE *input;
349};
350
a96847cc
JH
351/*
352 * ... and its getline() method implementation
353 */
27d6b085
JH
354static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
355{
356 struct rerere_io_file *io = (struct rerere_io_file *)io_;
357 return strbuf_getwholeline(sb, io->input, '\n');
358}
359
67711cdc
JH
360/*
361 * Require the exact number of conflict marker letters, no more, no
362 * less, followed by SP or any whitespace
363 * (including LF).
364 */
365static int is_cmarker(char *buf, int marker_char, int marker_size)
191f2417 366{
67711cdc
JH
367 int want_sp;
368
369 /*
370 * The beginning of our version and the end of their version
371 * always are labeled like "<<<<< ours" or ">>>>> theirs",
372 * hence we set want_sp for them. Note that the version from
373 * the common ancestor in diff3-style output is not always
374 * labelled (e.g. "||||| common" is often seen but "|||||"
375 * alone is also valid), so we do not set want_sp.
376 */
377 want_sp = (marker_char == '<') || (marker_char == '>');
378
191f2417
JH
379 while (marker_size--)
380 if (*buf++ != marker_char)
381 return 0;
382 if (want_sp && *buf != ' ')
383 return 0;
384 return isspace(*buf);
385}
386
cc899eca
JH
387/*
388 * Read contents a file with conflicts, normalize the conflicts
389 * by (1) discarding the common ancestor version in diff3-style,
390 * (2) reordering our side and their side so that whichever sorts
391 * alphabetically earlier comes before the other one, while
392 * computing the "conflict ID", which is just an SHA-1 hash of
393 * one side of the conflict, NUL, the other side of the conflict,
394 * and NUL concatenated together.
395 *
396 * Return the number of conflict hunks found.
397 *
398 * NEEDSWORK: the logic and theory of operation behind this conflict
399 * normalization may deserve to be documented somewhere, perhaps in
400 * Documentation/technical/rerere.txt.
401 */
191f2417 402static int handle_path(unsigned char *sha1, struct rerere_io *io, int marker_size)
5b2fd956 403{
9126f009 404 git_SHA_CTX ctx;
cc58d7df
JH
405 int hunk_no = 0;
406 enum {
4b05548f 407 RR_CONTEXT = 0, RR_SIDE_1, RR_SIDE_2, RR_ORIGINAL
cc58d7df 408 } hunk = RR_CONTEXT;
f285a2d7 409 struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
d58ee6db 410 struct strbuf buf = STRBUF_INIT;
5b2fd956
SB
411
412 if (sha1)
9126f009 413 git_SHA1_Init(&ctx);
5b2fd956 414
27d6b085 415 while (!io->getline(&buf, io)) {
67711cdc 416 if (is_cmarker(buf.buf, '<', marker_size)) {
cc58d7df 417 if (hunk != RR_CONTEXT)
5b2fd956 418 goto bad;
cc58d7df 419 hunk = RR_SIDE_1;
67711cdc 420 } else if (is_cmarker(buf.buf, '|', marker_size)) {
cc58d7df 421 if (hunk != RR_SIDE_1)
5b2fd956 422 goto bad;
387c9d49 423 hunk = RR_ORIGINAL;
67711cdc 424 } else if (is_cmarker(buf.buf, '=', marker_size)) {
387c9d49 425 if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
5b2fd956 426 goto bad;
cc58d7df 427 hunk = RR_SIDE_2;
67711cdc 428 } else if (is_cmarker(buf.buf, '>', marker_size)) {
cc58d7df 429 if (hunk != RR_SIDE_2)
5b2fd956
SB
430 goto bad;
431 if (strbuf_cmp(&one, &two) > 0)
432 strbuf_swap(&one, &two);
433 hunk_no++;
cc58d7df 434 hunk = RR_CONTEXT;
191f2417 435 rerere_io_putconflict('<', marker_size, io);
27d6b085 436 rerere_io_putmem(one.buf, one.len, io);
191f2417 437 rerere_io_putconflict('=', marker_size, io);
27d6b085 438 rerere_io_putmem(two.buf, two.len, io);
191f2417 439 rerere_io_putconflict('>', marker_size, io);
5b2fd956 440 if (sha1) {
9126f009 441 git_SHA1_Update(&ctx, one.buf ? one.buf : "",
5b2fd956 442 one.len + 1);
9126f009 443 git_SHA1_Update(&ctx, two.buf ? two.buf : "",
5b2fd956
SB
444 two.len + 1);
445 }
446 strbuf_reset(&one);
447 strbuf_reset(&two);
cc58d7df 448 } else if (hunk == RR_SIDE_1)
e992d1eb 449 strbuf_addbuf(&one, &buf);
387c9d49
JH
450 else if (hunk == RR_ORIGINAL)
451 ; /* discard */
cc58d7df 452 else if (hunk == RR_SIDE_2)
e992d1eb 453 strbuf_addbuf(&two, &buf);
27d6b085
JH
454 else
455 rerere_io_putstr(buf.buf, io);
5b2fd956
SB
456 continue;
457 bad:
458 hunk = 99; /* force error exit */
459 break;
460 }
461 strbuf_release(&one);
462 strbuf_release(&two);
d58ee6db 463 strbuf_release(&buf);
5b2fd956 464
5b2fd956 465 if (sha1)
9126f009 466 git_SHA1_Final(sha1, &ctx);
27d6b085
JH
467 if (hunk != RR_CONTEXT)
468 return -1;
469 return hunk_no;
470}
471
cc899eca
JH
472/*
473 * Scan the path for conflicts, do the "handle_path()" thing above, and
474 * return the number of conflict hunks found.
475 */
27d6b085
JH
476static int handle_file(const char *path, unsigned char *sha1, const char *output)
477{
478 int hunk_no = 0;
479 struct rerere_io_file io;
8588567c 480 int marker_size = ll_merge_marker_size(path);
27d6b085
JH
481
482 memset(&io, 0, sizeof(io));
483 io.io.getline = rerere_file_getline;
484 io.input = fopen(path, "r");
485 io.io.wrerror = 0;
486 if (!io.input)
5118d7f4 487 return error_errno("Could not open %s", path);
27d6b085
JH
488
489 if (output) {
490 io.io.output = fopen(output, "w");
491 if (!io.io.output) {
f7566f07 492 error_errno("Could not write %s", output);
27d6b085 493 fclose(io.input);
f7566f07 494 return -1;
27d6b085
JH
495 }
496 }
497
191f2417 498 hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
27d6b085
JH
499
500 fclose(io.input);
501 if (io.io.wrerror)
502 error("There were errors while writing %s (%s)",
503 path, strerror(io.io.wrerror));
504 if (io.io.output && fclose(io.io.output))
033e011e 505 io.io.wrerror = error_errno("Failed to flush %s", path);
27d6b085
JH
506
507 if (hunk_no < 0) {
5b2fd956 508 if (output)
691f1a28 509 unlink_or_warn(output);
5b2fd956
SB
510 return error("Could not parse conflict hunks in %s", path);
511 }
27d6b085 512 if (io.io.wrerror)
47d32af2 513 return -1;
5b2fd956
SB
514 return hunk_no;
515}
516
4b68c2a0
JH
517/*
518 * Look at a cache entry at "i" and see if it is not conflicting,
519 * conflicting and we are willing to handle, or conflicting and
520 * we are unable to handle, and return the determination in *type.
521 * Return the cache index to be looked at next, by skipping the
522 * stages we have already looked at in this invocation of this
523 * function.
524 */
ac49f5ca 525static int check_one_conflict(int i, int *type)
5b2fd956 526{
9c5e6c80 527 const struct cache_entry *e = active_cache[i];
ac49f5ca
MZ
528
529 if (!ce_stage(e)) {
530 *type = RESOLVED;
531 return i + 1;
532 }
533
534 *type = PUNTED;
ad2bf0d9 535 while (i < active_nr && ce_stage(active_cache[i]) == 1)
fb70a06d 536 i++;
ac49f5ca
MZ
537
538 /* Only handle regular files with both stages #2 and #3 */
539 if (i + 1 < active_nr) {
9c5e6c80
NTND
540 const struct cache_entry *e2 = active_cache[i];
541 const struct cache_entry *e3 = active_cache[i + 1];
5b2fd956
SB
542 if (ce_stage(e2) == 2 &&
543 ce_stage(e3) == 3 &&
ac49f5ca 544 ce_same_name(e, e3) &&
5b2fd956 545 S_ISREG(e2->ce_mode) &&
ac49f5ca
MZ
546 S_ISREG(e3->ce_mode))
547 *type = THREE_STAGED;
548 }
549
550 /* Skip the entries with the same name */
551 while (i < active_nr && ce_same_name(e, active_cache[i]))
552 i++;
553 return i;
554}
555
4b68c2a0
JH
556/*
557 * Scan the index and find paths that have conflicts that rerere can
558 * handle, i.e. the ones that has both stages #2 and #3.
559 *
560 * NEEDSWORK: we do not record or replay a previous "resolve by
561 * deletion" for a delete-modify conflict, as that is inherently risky
562 * without knowing what modification is being discarded. The only
563 * safe case, i.e. both side doing the deletion and modification that
564 * are identical to the previous round, might want to be handled,
565 * though.
566 */
ac49f5ca
MZ
567static int find_conflict(struct string_list *conflict)
568{
569 int i;
570 if (read_cache() < 0)
571 return error("Could not read index");
572
573 for (i = 0; i < active_nr;) {
574 int conflict_type;
9c5e6c80 575 const struct cache_entry *e = active_cache[i];
ac49f5ca
MZ
576 i = check_one_conflict(i, &conflict_type);
577 if (conflict_type == THREE_STAGED)
578 string_list_insert(conflict, (const char *)e->name);
579 }
580 return 0;
581}
582
4b68c2a0
JH
583/*
584 * The merge_rr list is meant to hold outstanding conflicted paths
585 * that rerere could handle. Abuse the list by adding other types of
586 * entries to allow the caller to show "rerere remaining".
587 *
588 * - Conflicted paths that rerere does not handle are added
589 * - Conflicted paths that have been resolved are marked as such
590 * by storing RERERE_RESOLVED to .util field (where conflict ID
591 * is expected to be stored).
592 *
593 * Do *not* write MERGE_RR file out after calling this function.
594 *
595 * NEEDSWORK: we may want to fix the caller that implements "rerere
596 * remaining" to do this without abusing merge_rr.
597 */
ac49f5ca
MZ
598int rerere_remaining(struct string_list *merge_rr)
599{
600 int i;
9dd330e6
JK
601 if (setup_rerere(merge_rr, RERERE_READONLY))
602 return 0;
ac49f5ca
MZ
603 if (read_cache() < 0)
604 return error("Could not read index");
605
606 for (i = 0; i < active_nr;) {
607 int conflict_type;
9c5e6c80 608 const struct cache_entry *e = active_cache[i];
ac49f5ca
MZ
609 i = check_one_conflict(i, &conflict_type);
610 if (conflict_type == PUNTED)
611 string_list_insert(merge_rr, (const char *)e->name);
612 else if (conflict_type == RESOLVED) {
613 struct string_list_item *it;
614 it = string_list_lookup(merge_rr, (const char *)e->name);
615 if (it != NULL) {
1d51eced 616 free_rerere_id(it);
ac49f5ca
MZ
617 it->util = RERERE_RESOLVED;
618 }
5b2fd956
SB
619 }
620 }
621 return 0;
622}
623
0ce02b36
JH
624/*
625 * Try using the given conflict resolution "ID" to see
626 * if that recorded conflict resolves cleanly what we
627 * got in the "cur".
628 */
629static int try_merge(const struct rerere_id *id, const char *path,
630 mmfile_t *cur, mmbuffer_t *result)
631{
632 int ret;
633 mmfile_t base = {NULL, 0}, other = {NULL, 0};
634
635 if (read_mmfile(&base, rerere_path(id, "preimage")) ||
636 read_mmfile(&other, rerere_path(id, "postimage")))
637 ret = 1;
638 else
639 /*
640 * A three-way merge. Note that this honors user-customizable
641 * low-level merge driver settings.
642 */
643 ret = ll_merge(result, path, &base, NULL, cur, "", &other, "", NULL);
644
645 free(base.ptr);
646 free(other.ptr);
647
648 return ret;
649}
650
cc899eca 651/*
18bb9934 652 * Find the conflict identified by "id"; the change between its
cc899eca
JH
653 * "preimage" (i.e. a previous contents with conflict markers) and its
654 * "postimage" (i.e. the corresponding contents with conflicts
655 * resolved) may apply cleanly to the contents stored in "path", i.e.
656 * the conflict this time around.
657 *
658 * Returns 0 for successful replay of recorded resolution, or non-zero
659 * for failure.
660 */
1d51eced 661static int merge(const struct rerere_id *id, const char *path)
5b2fd956 662{
15ed07d5 663 FILE *f;
5b2fd956 664 int ret;
0ce02b36 665 mmfile_t cur = {NULL, 0};
5b2fd956 666 mmbuffer_t result = {NULL, 0};
5b2fd956 667
cc899eca
JH
668 /*
669 * Normalize the conflicts in path and write it out to
670 * "thisimage" temporary file.
671 */
0ce02b36
JH
672 if ((handle_file(path, NULL, rerere_path(id, "thisimage")) < 0) ||
673 read_mmfile(&cur, rerere_path(id, "thisimage"))) {
689b8c29
BW
674 ret = 1;
675 goto out;
676 }
cc899eca 677
0ce02b36 678 ret = try_merge(id, path, &cur, &result);
15ed07d5
JH
679 if (ret)
680 goto out;
681
682 /*
683 * A successful replay of recorded resolution.
684 * Mark that "postimage" was used to help gc.
685 */
686 if (utime(rerere_path(id, "postimage"), NULL) < 0)
033e011e
NTND
687 warning_errno("failed utime() on %s",
688 rerere_path(id, "postimage"));
15ed07d5
JH
689
690 /* Update "path" with the resolution */
691 f = fopen(path, "w");
692 if (!f)
033e011e 693 return error_errno("Could not open %s", path);
15ed07d5 694 if (fwrite(result.ptr, result.size, 1, f) != 1)
033e011e 695 error_errno("Could not write %s", path);
15ed07d5 696 if (fclose(f))
033e011e 697 return error_errno("Writing %s failed", path);
5b2fd956 698
689b8c29 699out:
5b2fd956 700 free(cur.ptr);
5b2fd956
SB
701 free(result.ptr);
702
703 return ret;
704}
705
706static struct lock_file index_lock;
707
89ea9035 708static void update_paths(struct string_list *update)
5b2fd956
SB
709{
710 int i;
5b2fd956 711
b3e83cc7 712 hold_locked_index(&index_lock, LOCK_DIE_ON_ERROR);
5b2fd956
SB
713
714 for (i = 0; i < update->nr; i++) {
c455c87c 715 struct string_list_item *item = &update->items[i];
89ea9035
JN
716 if (add_file_to_cache(item->string, 0))
717 exit(128);
a14c7ab8
JH
718 fprintf(stderr, "Staged '%s' using previous resolution.\n",
719 item->string);
5b2fd956
SB
720 }
721
61000814
722 if (write_locked_index(&the_index, &index_lock,
723 COMMIT_LOCK | SKIP_IF_UNCHANGED))
724 die("Unable to write new index file");
5b2fd956
SB
725}
726
629716d2
JH
727static void remove_variant(struct rerere_id *id)
728{
729 unlink_or_warn(rerere_path(id, "postimage"));
730 unlink_or_warn(rerere_path(id, "preimage"));
731 id->collection->status[id->variant] = 0;
732}
733
8e7768b2
JH
734/*
735 * The path indicated by rr_item may still have conflict for which we
736 * have a recorded resolution, in which case replay it and optionally
737 * update it. Or it may have been resolved by the user and we may
738 * only have the preimage for that conflict, in which case the result
739 * needs to be recorded as a resolution in a postimage file.
740 */
741static void do_rerere_one_path(struct string_list_item *rr_item,
742 struct string_list *update)
743{
744 const char *path = rr_item->string;
a13d1370 745 struct rerere_id *id = rr_item->util;
629716d2 746 struct rerere_dir *rr_dir = id->collection;
a13d1370
JH
747 int variant;
748
a13d1370 749 variant = id->variant;
8e7768b2 750
629716d2
JH
751 /* Has the user resolved it already? */
752 if (variant >= 0) {
753 if (!handle_file(path, NULL, NULL)) {
754 copy_file(rerere_path(id, "postimage"), path, 0666);
755 id->collection->status[variant] |= RR_HAS_POSTIMAGE;
756 fprintf(stderr, "Recorded resolution for '%s'.\n", path);
757 free_rerere_id(rr_item);
758 rr_item->util = NULL;
759 return;
760 }
c0a5423b 761 /*
629716d2
JH
762 * There may be other variants that can cleanly
763 * replay. Try them and update the variant number for
764 * this one.
c0a5423b 765 */
629716d2
JH
766 }
767
768 /* Does any existing resolution apply cleanly? */
769 for (variant = 0; variant < rr_dir->status_nr; variant++) {
770 const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
771 struct rerere_id vid = *id;
772
773 if ((rr_dir->status[variant] & both) != both)
774 continue;
8e7768b2 775
629716d2
JH
776 vid.variant = variant;
777 if (merge(&vid, path))
778 continue; /* failed to replay */
779
780 /*
781 * If there already is a different variant that applies
782 * cleanly, there is no point maintaining our own variant.
783 */
784 if (0 <= id->variant && id->variant != variant)
785 remove_variant(id);
8e7768b2
JH
786
787 if (rerere_autoupdate)
788 string_list_insert(update, path);
789 else
790 fprintf(stderr,
791 "Resolved '%s' using previous resolution.\n",
792 path);
629716d2
JH
793 free_rerere_id(rr_item);
794 rr_item->util = NULL;
925d73c4 795 return;
8e7768b2 796 }
629716d2
JH
797
798 /* None of the existing one applies; we need a new variant */
799 assign_variant(id);
800
801 variant = id->variant;
802 handle_file(path, NULL, rerere_path(id, "preimage"));
803 if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
804 const char *path = rerere_path(id, "postimage");
805 if (unlink(path))
806 die_errno("cannot unlink stray '%s'", path);
807 id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
808 }
809 id->collection->status[variant] |= RR_HAS_PREIMAGE;
810 fprintf(stderr, "Recorded preimage for '%s'\n", path);
8e7768b2
JH
811}
812
c455c87c 813static int do_plain_rerere(struct string_list *rr, int fd)
5b2fd956 814{
183113a5
TF
815 struct string_list conflict = STRING_LIST_INIT_DUP;
816 struct string_list update = STRING_LIST_INIT_DUP;
5b2fd956
SB
817 int i;
818
819 find_conflict(&conflict);
820
821 /*
cc899eca
JH
822 * MERGE_RR records paths with conflicts immediately after
823 * merge failed. Some of the conflicted paths might have been
824 * hand resolved in the working tree since then, but the
825 * initial run would catch all and register their preimages.
5b2fd956 826 */
5b2fd956 827 for (i = 0; i < conflict.nr; i++) {
1d51eced 828 struct rerere_id *id;
c7a25d37 829 unsigned char sha1[20];
c455c87c 830 const char *path = conflict.items[i].string;
1d51eced 831 int ret;
5b2fd956 832
c7a25d37
JH
833 if (string_list_has_string(rr, path))
834 continue;
5b2fd956 835
c7a25d37
JH
836 /*
837 * Ask handle_file() to scan and assign a
838 * conflict ID. No need to write anything out
839 * yet.
840 */
841 ret = handle_file(path, sha1, NULL);
842 if (ret < 1)
843 continue;
5b2fd956 844
1d51eced 845 id = new_rerere_id(sha1);
18bb9934 846 string_list_insert(rr, path)->util = id;
cc899eca 847
c0a5423b
JH
848 /* Ensure that the directory exists. */
849 mkdir_in_gitdir(rerere_path(id, NULL));
5b2fd956
SB
850 }
851
8e7768b2
JH
852 for (i = 0; i < rr->nr; i++)
853 do_rerere_one_path(&rr->items[i], &update);
5b2fd956
SB
854
855 if (update.nr)
856 update_paths(&update);
857
858 return write_rr(rr, fd);
859}
860
633e5ad3 861static void git_rerere_config(void)
5b2fd956 862{
633e5ad3
TA
863 git_config_get_bool("rerere.enabled", &rerere_enabled);
864 git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
865 git_config(git_default_config, NULL);
5b2fd956
SB
866}
867
f932729c
JK
868static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
869
5b2fd956
SB
870static int is_rerere_enabled(void)
871{
5b2fd956
SB
872 int rr_cache_exists;
873
874 if (!rerere_enabled)
875 return 0;
876
f932729c 877 rr_cache_exists = is_directory(git_path_rr_cache());
5b2fd956
SB
878 if (rerere_enabled < 0)
879 return rr_cache_exists;
880
f932729c
JK
881 if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
882 die("Could not create directory %s", git_path_rr_cache());
5b2fd956
SB
883 return 1;
884}
885
cb6020bb 886int setup_rerere(struct string_list *merge_rr, int flags)
5b2fd956
SB
887{
888 int fd;
889
633e5ad3 890 git_rerere_config();
5b2fd956
SB
891 if (!is_rerere_enabled())
892 return -1;
893
cb6020bb
JH
894 if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
895 rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
9dd330e6
JK
896 if (flags & RERERE_READONLY)
897 fd = 0;
898 else
899 fd = hold_lock_file_for_update(&write_lock, git_path_merge_rr(),
900 LOCK_DIE_ON_ERROR);
5b2fd956
SB
901 read_rr(merge_rr);
902 return fd;
903}
904
cc899eca
JH
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 */
cb6020bb 910int rerere(int flags)
5b2fd956 911{
183113a5 912 struct string_list merge_rr = STRING_LIST_INIT_DUP;
1869bbe1 913 int fd, status;
5b2fd956 914
cb6020bb 915 fd = setup_rerere(&merge_rr, flags);
5b2fd956
SB
916 if (fd < 0)
917 return 0;
1869bbe1
JH
918 status = do_plain_rerere(&merge_rr, fd);
919 free_rerere_dirs();
920 return status;
5b2fd956 921}
dea4562b 922
3d730ed9
JH
923/*
924 * Subclass of rerere_io that reads from an in-core buffer that is a
925 * strbuf
926 */
927struct rerere_io_mem {
928 struct rerere_io io;
929 struct strbuf input;
930};
931
932/*
933 * ... and its getline() method implementation
934 */
935static 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
955static int handle_cache(const char *path, unsigned char *sha1, const char *output)
956{
957 mmfile_t mmfile[3] = {{NULL}};
958 mmbuffer_t result = {NULL, 0};
959 const struct cache_entry *ce;
960 int pos, len, i, hunk_no;
961 struct rerere_io_mem io;
962 int marker_size = ll_merge_marker_size(path);
963
964 /*
965 * Reproduce the conflicted merge in-core
966 */
967 len = strlen(path);
968 pos = cache_name_pos(path, len);
969 if (0 <= pos)
970 return -1;
971 pos = -pos - 1;
972
973 while (pos < active_nr) {
974 enum object_type type;
975 unsigned long size;
976
977 ce = active_cache[pos++];
978 if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
979 break;
980 i = ce_stage(ce) - 1;
981 if (!mmfile[i].ptr) {
99d1a986 982 mmfile[i].ptr = read_sha1_file(ce->oid.hash, &type,
983 &size);
3d730ed9
JH
984 mmfile[i].size = size;
985 }
986 }
987 for (i = 0; i < 3; i++)
988 if (!mmfile[i].ptr && !mmfile[i].size)
989 mmfile[i].ptr = xstrdup("");
990
991 /*
992 * NEEDSWORK: handle conflicts from merges with
993 * merge.renormalize set, too?
994 */
995 ll_merge(&result, path, &mmfile[0], NULL,
996 &mmfile[1], "ours",
997 &mmfile[2], "theirs", NULL);
998 for (i = 0; i < 3; i++)
999 free(mmfile[i].ptr);
1000
1001 memset(&io, 0, sizeof(io));
1002 io.io.getline = rerere_mem_getline;
1003 if (output)
1004 io.io.output = fopen(output, "w");
1005 else
1006 io.io.output = NULL;
1007 strbuf_init(&io.input, 0);
1008 strbuf_attach(&io.input, result.ptr, result.size, result.size);
1009
1010 /*
1011 * Grab the conflict ID and optionally write the original
1012 * contents with conflict markers out.
1013 */
1014 hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
1015 strbuf_release(&io.input);
1016 if (io.io.output)
1017 fclose(io.io.output);
1018 return hunk_no;
5b2fd956 1019}
dea4562b
JH
1020
1021static int rerere_forget_one_path(const char *path, struct string_list *rr)
1022{
1023 const char *filename;
1d51eced 1024 struct rerere_id *id;
dea4562b
JH
1025 unsigned char sha1[20];
1026 int ret;
8d9b5a4a 1027 struct string_list_item *item;
dea4562b 1028
963ec003
JH
1029 /*
1030 * Recreate the original conflict from the stages in the
1031 * index and compute the conflict ID
1032 */
dea4562b
JH
1033 ret = handle_cache(path, sha1, NULL);
1034 if (ret < 1)
1035 return error("Could not parse conflict hunks in '%s'", path);
963ec003
JH
1036
1037 /* Nuke the recorded resolution for the conflict */
1d51eced 1038 id = new_rerere_id(sha1);
890fca84
JH
1039
1040 for (id->variant = 0;
1041 id->variant < id->collection->status_nr;
1042 id->variant++) {
1043 mmfile_t cur = { NULL, 0 };
1044 mmbuffer_t result = {NULL, 0};
1045 int cleanly_resolved;
1046
1047 if (!has_rerere_resolution(id))
1048 continue;
1049
1050 handle_cache(path, sha1, rerere_path(id, "thisimage"));
1051 if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1052 free(cur.ptr);
8f449614
JH
1053 error("Failed to update conflicted state in '%s'", path);
1054 goto fail_exit;
890fca84
JH
1055 }
1056 cleanly_resolved = !try_merge(id, path, &cur, &result);
1057 free(result.ptr);
1058 free(cur.ptr);
1059 if (cleanly_resolved)
1060 break;
1061 }
1062
8f449614
JH
1063 if (id->collection->status_nr <= id->variant) {
1064 error("no remembered resolution for '%s'", path);
1065 goto fail_exit;
1066 }
890fca84 1067
18bb9934 1068 filename = rerere_path(id, "postimage");
8f449614
JH
1069 if (unlink(filename)) {
1070 if (errno == ENOENT)
1071 error("no remembered resolution for %s", path);
1072 else
ec34a8b1 1073 error_errno("cannot unlink %s", filename);
8f449614 1074 goto fail_exit;
d9d501b0 1075 }
dea4562b 1076
963ec003
JH
1077 /*
1078 * Update the preimage so that the user can resolve the
1079 * conflict in the working tree, run us again to record
1080 * the postimage.
1081 */
18bb9934 1082 handle_cache(path, sha1, rerere_path(id, "preimage"));
dea4562b
JH
1083 fprintf(stderr, "Updated preimage for '%s'\n", path);
1084
963ec003
JH
1085 /*
1086 * And remember that we can record resolution for this
1087 * conflict when the user is done.
1088 */
8d9b5a4a 1089 item = string_list_insert(rr, path);
1d51eced 1090 free_rerere_id(item);
18bb9934 1091 item->util = id;
dea4562b
JH
1092 fprintf(stderr, "Forgot resolution for %s\n", path);
1093 return 0;
8f449614
JH
1094
1095fail_exit:
1096 free(id);
1097 return -1;
dea4562b
JH
1098}
1099
01a10b0a 1100int rerere_forget(struct pathspec *pathspec)
dea4562b
JH
1101{
1102 int i, fd;
183113a5
TF
1103 struct string_list conflict = STRING_LIST_INIT_DUP;
1104 struct string_list merge_rr = STRING_LIST_INIT_DUP;
dea4562b
JH
1105
1106 if (read_cache() < 0)
1107 return error("Could not read index");
1108
6751e047 1109 fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
0544574c
JK
1110 if (fd < 0)
1111 return 0;
dea4562b 1112
963ec003
JH
1113 /*
1114 * The paths may have been resolved (incorrectly);
1115 * recover the original conflicted state and then
1116 * find the conflicted paths.
1117 */
dea4562b
JH
1118 unmerge_cache(pathspec);
1119 find_conflict(&conflict);
1120 for (i = 0; i < conflict.nr; i++) {
1121 struct string_list_item *it = &conflict.items[i];
854b0959 1122 if (!match_pathspec(pathspec, it->string,
ae8d0824 1123 strlen(it->string), 0, NULL, 0))
dea4562b
JH
1124 continue;
1125 rerere_forget_one_path(it->string, &merge_rr);
1126 }
1127 return write_rr(&merge_rr, fd);
1128}
0f891e7d 1129
e828de82
JH
1130/*
1131 * Garbage collection support
1132 */
1d51eced 1133
5ea82279 1134static timestamp_t rerere_created_at(struct rerere_id *id)
0f891e7d
JH
1135{
1136 struct stat st;
1d51eced 1137
18bb9934 1138 return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
0f891e7d
JH
1139}
1140
5ea82279 1141static timestamp_t rerere_last_used_at(struct rerere_id *id)
0f891e7d
JH
1142{
1143 struct stat st;
1d51eced 1144
18bb9934 1145 return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
0f891e7d
JH
1146}
1147
e828de82
JH
1148/*
1149 * Remove the recorded resolution for a given conflict ID
1150 */
1d51eced 1151static void unlink_rr_item(struct rerere_id *id)
0f891e7d 1152{
1be1e851
JH
1153 unlink_or_warn(rerere_path(id, "thisimage"));
1154 remove_variant(id);
1155 id->collection->status[id->variant] = 0;
1156}
1157
5ea82279
JH
1158static void prune_one(struct rerere_id *id,
1159 timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1be1e851 1160{
5ea82279
JH
1161 timestamp_t then;
1162 timestamp_t cutoff;
1be1e851
JH
1163
1164 then = rerere_last_used_at(id);
1165 if (then)
1166 cutoff = cutoff_resolve;
1167 else {
1168 then = rerere_created_at(id);
1169 if (!then)
1170 return;
1171 cutoff = cutoff_noresolve;
1172 }
5ea82279 1173 if (then < cutoff)
1be1e851 1174 unlink_rr_item(id);
0f891e7d
JH
1175}
1176
0f891e7d
JH
1177void rerere_gc(struct string_list *rr)
1178{
1179 struct string_list to_remove = STRING_LIST_INIT_DUP;
1180 DIR *dir;
1181 struct dirent *e;
1be1e851 1182 int i;
5ea82279
JH
1183 timestamp_t now = time(NULL);
1184 timestamp_t cutoff_noresolve = now - 15 * 86400;
1185 timestamp_t cutoff_resolve = now - 60 * 86400;
0f891e7d 1186
9dd330e6
JK
1187 if (setup_rerere(rr, 0) < 0)
1188 return;
1189
6e96cb52
JH
1190 git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
1191 git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
633e5ad3 1192 git_config(git_default_config, NULL);
0f891e7d
JH
1193 dir = opendir(git_path("rr-cache"));
1194 if (!dir)
1195 die_errno("unable to open rr-cache directory");
e828de82 1196 /* Collect stale conflict IDs ... */
0f891e7d 1197 while ((e = readdir(dir))) {
1be1e851
JH
1198 struct rerere_dir *rr_dir;
1199 struct rerere_id id;
1200 int now_empty;
1201
0f891e7d
JH
1202 if (is_dot_or_dotdot(e->d_name))
1203 continue;
1be1e851
JH
1204 rr_dir = find_rerere_dir(e->d_name);
1205 if (!rr_dir)
1206 continue; /* or should we remove e->d_name? */
1207
1208 now_empty = 1;
1209 for (id.variant = 0, id.collection = rr_dir;
1210 id.variant < id.collection->status_nr;
1211 id.variant++) {
5ea82279 1212 prune_one(&id, cutoff_resolve, cutoff_noresolve);
1be1e851
JH
1213 if (id.collection->status[id.variant])
1214 now_empty = 0;
0f891e7d 1215 }
1be1e851 1216 if (now_empty)
0f891e7d
JH
1217 string_list_append(&to_remove, e->d_name);
1218 }
a9930e35 1219 closedir(dir);
1be1e851
JH
1220
1221 /* ... and then remove the empty directories */
0f891e7d 1222 for (i = 0; i < to_remove.nr; i++)
1be1e851 1223 rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
0f891e7d 1224 string_list_clear(&to_remove, 0);
9dd330e6 1225 rollback_lock_file(&write_lock);
0f891e7d
JH
1226}
1227
e828de82
JH
1228/*
1229 * During a conflict resolution, after "rerere" recorded the
1230 * preimages, abandon them if the user did not resolve them or
1231 * record their resolutions. And drop $GIT_DIR/MERGE_RR.
1232 *
1233 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1234 */
0f891e7d
JH
1235void rerere_clear(struct string_list *merge_rr)
1236{
1237 int i;
1238
9dd330e6
JK
1239 if (setup_rerere(merge_rr, 0) < 0)
1240 return;
1241
0f891e7d 1242 for (i = 0; i < merge_rr->nr; i++) {
1d51eced 1243 struct rerere_id *id = merge_rr->items[i].util;
1be1e851 1244 if (!has_rerere_resolution(id)) {
18bb9934 1245 unlink_rr_item(id);
1be1e851
JH
1246 rmdir(rerere_path(id, NULL));
1247 }
0f891e7d 1248 }
f932729c 1249 unlink_or_warn(git_path_merge_rr());
9dd330e6 1250 rollback_lock_file(&write_lock);
0f891e7d 1251}