bf803043e28676ccde5a0d75f955872b2c3a2022
   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 "sha1-lookup.h"
  13
  14#define RESOLVED 0
  15#define PUNTED 1
  16#define THREE_STAGED 2
  17void *RERERE_RESOLVED = &RERERE_RESOLVED;
  18
  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
  25static int rerere_dir_nr;
  26static int rerere_dir_alloc;
  27
  28#define RR_HAS_POSTIMAGE 1
  29#define RR_HAS_PREIMAGE 2
  30static struct rerere_dir {
  31        unsigned char sha1[20];
  32        int status_alloc, status_nr;
  33        unsigned char *status;
  34} **rerere_dir;
  35
  36static void free_rerere_dirs(void)
  37{
  38        int i;
  39        for (i = 0; i < rerere_dir_nr; i++) {
  40                free(rerere_dir[i]->status);
  41                free(rerere_dir[i]);
  42        }
  43        FREE_AND_NULL(rerere_dir);
  44        rerere_dir_nr = rerere_dir_alloc = 0;
  45}
  46
  47static void free_rerere_id(struct string_list_item *item)
  48{
  49        free(item->util);
  50}
  51
  52static const char *rerere_id_hex(const struct rerere_id *id)
  53{
  54        return sha1_to_hex(id->collection->sha1);
  55}
  56
  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) {
  75                for (variant = 0; variant < rr_dir->status_nr; variant++)
  76                        if (!rr_dir->status[variant])
  77                                break;
  78        }
  79        fit_variant(rr_dir, variant);
  80        id->variant = variant;
  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
  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);
  93}
  94
  95static int is_rr_file(const char *name, const char *filename, int *variant)
  96{
  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;
 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) {
 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                }
 131        }
 132        closedir(dir);
 133}
 134
 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);
 153                rr_dir->status = NULL;
 154                rr_dir->status_nr = 0;
 155                rr_dir->status_alloc = 0;
 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++;
 162                MOVE_ARRAY(rerere_dir + pos + 1, rerere_dir + pos,
 163                           rerere_dir_nr - pos - 1);
 164                rerere_dir[pos] = rr_dir;
 165                scan_rerere_dir(rr_dir);
 166        }
 167        return rerere_dir[pos];
 168}
 169
 170static int has_rerere_resolution(const struct rerere_id *id)
 171{
 172        const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
 173        int variant = id->variant;
 174
 175        if (variant < 0)
 176                return 0;
 177        return ((id->collection->status[variant] & both) == both);
 178}
 179
 180static struct rerere_id *new_rerere_id_hex(char *hex)
 181{
 182        struct rerere_id *id = xmalloc(sizeof(*id));
 183        id->collection = find_rerere_dir(hex);
 184        id->variant = -1; /* not known yet */
 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));
 191}
 192
 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 */
 200static void read_rr(struct string_list *rr)
 201{
 202        struct strbuf buf = STRBUF_INIT;
 203        FILE *in = fopen_or_warn(git_path_merge_rr(), "r");
 204
 205        if (!in)
 206                return;
 207        while (!strbuf_getwholeline(&buf, in, '\0')) {
 208                char *path;
 209                unsigned char sha1[20];
 210                struct rerere_id *id;
 211                int variant;
 212
 213                /* There has to be the hash, tab, path and then NUL */
 214                if (buf.len < 42 || get_sha1_hex(buf.buf, sha1))
 215                        die(_("corrupt MERGE_RR"));
 216
 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')
 227                        die(_("corrupt MERGE_RR"));
 228                buf.buf[40] = '\0';
 229                id = new_rerere_id_hex(buf.buf);
 230                id->variant = variant;
 231                string_list_insert(rr, path)->util = id;
 232        }
 233        strbuf_release(&buf);
 234        fclose(in);
 235}
 236
 237static struct lock_file write_lock;
 238
 239static int write_rr(struct string_list *rr, int out_fd)
 240{
 241        int i;
 242        for (i = 0; i < rr->nr; i++) {
 243                struct strbuf buf = STRBUF_INIT;
 244                struct rerere_id *id;
 245
 246                assert(rr->items[i].util != RERERE_RESOLVED);
 247
 248                id = rr->items[i].util;
 249                if (!id)
 250                        continue;
 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
 261                if (write_in_full(out_fd, buf.buf, buf.len) < 0)
 262                        die(_("unable to write rerere record"));
 263
 264                strbuf_release(&buf);
 265        }
 266        if (commit_lock_file(&write_lock) != 0)
 267                die(_("unable to write rerere record"));
 268        return 0;
 269}
 270
 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
 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
 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
 305/*
 306 * Write a conflict marker to io->output (if defined).
 307 */
 308static void rerere_io_putconflict(int ch, int size, struct rerere_io *io)
 309{
 310        char buf[64];
 311
 312        while (size) {
 313                if (size <= sizeof(buf) - 2) {
 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;
 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                         */
 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
 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
 343/*
 344 * Subclass of rerere_io that reads from an on-disk file
 345 */
 346struct rerere_io_file {
 347        struct rerere_io io;
 348        FILE *input;
 349};
 350
 351/*
 352 * ... and its getline() method implementation
 353 */
 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
 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)
 366{
 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
 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
 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 1 if conflict hunks are found, 0 if there are no conflict
 397 * hunks and -1 if an error occured.
 398 */
 399static int handle_path(unsigned char *sha1, struct rerere_io *io, int marker_size)
 400{
 401        git_SHA_CTX ctx;
 402        int has_conflicts = 0;
 403        enum {
 404                RR_CONTEXT = 0, RR_SIDE_1, RR_SIDE_2, RR_ORIGINAL
 405        } hunk = RR_CONTEXT;
 406        struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
 407        struct strbuf buf = STRBUF_INIT;
 408
 409        if (sha1)
 410                git_SHA1_Init(&ctx);
 411
 412        while (!io->getline(&buf, io)) {
 413                if (is_cmarker(buf.buf, '<', marker_size)) {
 414                        if (hunk != RR_CONTEXT)
 415                                goto bad;
 416                        hunk = RR_SIDE_1;
 417                } else if (is_cmarker(buf.buf, '|', marker_size)) {
 418                        if (hunk != RR_SIDE_1)
 419                                goto bad;
 420                        hunk = RR_ORIGINAL;
 421                } else if (is_cmarker(buf.buf, '=', marker_size)) {
 422                        if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
 423                                goto bad;
 424                        hunk = RR_SIDE_2;
 425                } else if (is_cmarker(buf.buf, '>', marker_size)) {
 426                        if (hunk != RR_SIDE_2)
 427                                goto bad;
 428                        if (strbuf_cmp(&one, &two) > 0)
 429                                strbuf_swap(&one, &two);
 430                        has_conflicts = 1;
 431                        hunk = RR_CONTEXT;
 432                        rerere_io_putconflict('<', marker_size, io);
 433                        rerere_io_putmem(one.buf, one.len, io);
 434                        rerere_io_putconflict('=', marker_size, io);
 435                        rerere_io_putmem(two.buf, two.len, io);
 436                        rerere_io_putconflict('>', marker_size, io);
 437                        if (sha1) {
 438                                git_SHA1_Update(&ctx, one.buf ? one.buf : "",
 439                                            one.len + 1);
 440                                git_SHA1_Update(&ctx, two.buf ? two.buf : "",
 441                                            two.len + 1);
 442                        }
 443                        strbuf_reset(&one);
 444                        strbuf_reset(&two);
 445                } else if (hunk == RR_SIDE_1)
 446                        strbuf_addbuf(&one, &buf);
 447                else if (hunk == RR_ORIGINAL)
 448                        ; /* discard */
 449                else if (hunk == RR_SIDE_2)
 450                        strbuf_addbuf(&two, &buf);
 451                else
 452                        rerere_io_putstr(buf.buf, io);
 453                continue;
 454        bad:
 455                hunk = 99; /* force error exit */
 456                break;
 457        }
 458        strbuf_release(&one);
 459        strbuf_release(&two);
 460        strbuf_release(&buf);
 461
 462        if (sha1)
 463                git_SHA1_Final(sha1, &ctx);
 464        if (hunk != RR_CONTEXT)
 465                return -1;
 466        return has_conflicts;
 467}
 468
 469/*
 470 * Scan the path for conflicts, do the "handle_path()" thing above, and
 471 * return the number of conflict hunks found.
 472 */
 473static int handle_file(const char *path, unsigned char *sha1, const char *output)
 474{
 475        int has_conflicts = 0;
 476        struct rerere_io_file io;
 477        int marker_size = ll_merge_marker_size(path);
 478
 479        memset(&io, 0, sizeof(io));
 480        io.io.getline = rerere_file_getline;
 481        io.input = fopen(path, "r");
 482        io.io.wrerror = 0;
 483        if (!io.input)
 484                return error_errno(_("could not open '%s'"), path);
 485
 486        if (output) {
 487                io.io.output = fopen(output, "w");
 488                if (!io.io.output) {
 489                        error_errno(_("could not write '%s'"), output);
 490                        fclose(io.input);
 491                        return -1;
 492                }
 493        }
 494
 495        has_conflicts = handle_path(sha1, (struct rerere_io *)&io, marker_size);
 496
 497        fclose(io.input);
 498        if (io.io.wrerror)
 499                error(_("there were errors while writing '%s' (%s)"),
 500                      path, strerror(io.io.wrerror));
 501        if (io.io.output && fclose(io.io.output))
 502                io.io.wrerror = error_errno(_("failed to flush '%s'"), path);
 503
 504        if (has_conflicts < 0) {
 505                if (output)
 506                        unlink_or_warn(output);
 507                return error(_("could not parse conflict hunks in '%s'"), path);
 508        }
 509        if (io.io.wrerror)
 510                return -1;
 511        return has_conflicts;
 512}
 513
 514/*
 515 * Look at a cache entry at "i" and see if it is not conflicting,
 516 * conflicting and we are willing to handle, or conflicting and
 517 * we are unable to handle, and return the determination in *type.
 518 * Return the cache index to be looked at next, by skipping the
 519 * stages we have already looked at in this invocation of this
 520 * function.
 521 */
 522static int check_one_conflict(int i, int *type)
 523{
 524        const struct cache_entry *e = active_cache[i];
 525
 526        if (!ce_stage(e)) {
 527                *type = RESOLVED;
 528                return i + 1;
 529        }
 530
 531        *type = PUNTED;
 532        while (ce_stage(active_cache[i]) == 1)
 533                i++;
 534
 535        /* Only handle regular files with both stages #2 and #3 */
 536        if (i + 1 < active_nr) {
 537                const struct cache_entry *e2 = active_cache[i];
 538                const struct cache_entry *e3 = active_cache[i + 1];
 539                if (ce_stage(e2) == 2 &&
 540                    ce_stage(e3) == 3 &&
 541                    ce_same_name(e, e3) &&
 542                    S_ISREG(e2->ce_mode) &&
 543                    S_ISREG(e3->ce_mode))
 544                        *type = THREE_STAGED;
 545        }
 546
 547        /* Skip the entries with the same name */
 548        while (i < active_nr && ce_same_name(e, active_cache[i]))
 549                i++;
 550        return i;
 551}
 552
 553/*
 554 * Scan the index and find paths that have conflicts that rerere can
 555 * handle, i.e. the ones that has both stages #2 and #3.
 556 *
 557 * NEEDSWORK: we do not record or replay a previous "resolve by
 558 * deletion" for a delete-modify conflict, as that is inherently risky
 559 * without knowing what modification is being discarded.  The only
 560 * safe case, i.e. both side doing the deletion and modification that
 561 * are identical to the previous round, might want to be handled,
 562 * though.
 563 */
 564static int find_conflict(struct string_list *conflict)
 565{
 566        int i;
 567        if (read_cache() < 0)
 568                return error(_("index file corrupt"));
 569
 570        for (i = 0; i < active_nr;) {
 571                int conflict_type;
 572                const struct cache_entry *e = active_cache[i];
 573                i = check_one_conflict(i, &conflict_type);
 574                if (conflict_type == THREE_STAGED)
 575                        string_list_insert(conflict, (const char *)e->name);
 576        }
 577        return 0;
 578}
 579
 580/*
 581 * The merge_rr list is meant to hold outstanding conflicted paths
 582 * that rerere could handle.  Abuse the list by adding other types of
 583 * entries to allow the caller to show "rerere remaining".
 584 *
 585 * - Conflicted paths that rerere does not handle are added
 586 * - Conflicted paths that have been resolved are marked as such
 587 *   by storing RERERE_RESOLVED to .util field (where conflict ID
 588 *   is expected to be stored).
 589 *
 590 * Do *not* write MERGE_RR file out after calling this function.
 591 *
 592 * NEEDSWORK: we may want to fix the caller that implements "rerere
 593 * remaining" to do this without abusing merge_rr.
 594 */
 595int rerere_remaining(struct string_list *merge_rr)
 596{
 597        int i;
 598        if (setup_rerere(merge_rr, RERERE_READONLY))
 599                return 0;
 600        if (read_cache() < 0)
 601                return error(_("index file corrupt"));
 602
 603        for (i = 0; i < active_nr;) {
 604                int conflict_type;
 605                const struct cache_entry *e = active_cache[i];
 606                i = check_one_conflict(i, &conflict_type);
 607                if (conflict_type == PUNTED)
 608                        string_list_insert(merge_rr, (const char *)e->name);
 609                else if (conflict_type == RESOLVED) {
 610                        struct string_list_item *it;
 611                        it = string_list_lookup(merge_rr, (const char *)e->name);
 612                        if (it != NULL) {
 613                                free_rerere_id(it);
 614                                it->util = RERERE_RESOLVED;
 615                        }
 616                }
 617        }
 618        return 0;
 619}
 620
 621/*
 622 * Try using the given conflict resolution "ID" to see
 623 * if that recorded conflict resolves cleanly what we
 624 * got in the "cur".
 625 */
 626static int try_merge(const struct rerere_id *id, const char *path,
 627                     mmfile_t *cur, mmbuffer_t *result)
 628{
 629        int ret;
 630        mmfile_t base = {NULL, 0}, other = {NULL, 0};
 631
 632        if (read_mmfile(&base, rerere_path(id, "preimage")) ||
 633            read_mmfile(&other, rerere_path(id, "postimage")))
 634                ret = 1;
 635        else
 636                /*
 637                 * A three-way merge. Note that this honors user-customizable
 638                 * low-level merge driver settings.
 639                 */
 640                ret = ll_merge(result, path, &base, NULL, cur, "", &other, "", 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 */
 658static int merge(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(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(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
 696out:
 697        free(cur.ptr);
 698        free(result.ptr);
 699
 700        return ret;
 701}
 702
 703static void update_paths(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_cache(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(&the_index, &index_lock,
 719                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
 720                die(_("unable to write new index file"));
 721}
 722
 723static 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 */
 737static void do_rerere_one_path(struct string_list_item *rr_item,
 738                               struct string_list *update)
 739{
 740        const char *path = rr_item->string;
 741        struct rerere_id *id = rr_item->util;
 742        struct rerere_dir *rr_dir = id->collection;
 743        int variant;
 744
 745        variant = id->variant;
 746
 747        /* Has the user resolved it already? */
 748        if (variant >= 0) {
 749                if (!handle_file(path, NULL, NULL)) {
 750                        copy_file(rerere_path(id, "postimage"), path, 0666);
 751                        id->collection->status[variant] |= RR_HAS_POSTIMAGE;
 752                        fprintf_ln(stderr, _("Recorded resolution for '%s'."), path);
 753                        free_rerere_id(rr_item);
 754                        rr_item->util = NULL;
 755                        return;
 756                }
 757                /*
 758                 * There may be other variants that can cleanly
 759                 * replay.  Try them and update the variant number for
 760                 * this one.
 761                 */
 762        }
 763
 764        /* Does any existing resolution apply cleanly? */
 765        for (variant = 0; variant < rr_dir->status_nr; variant++) {
 766                const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
 767                struct rerere_id vid = *id;
 768
 769                if ((rr_dir->status[variant] & both) != both)
 770                        continue;
 771
 772                vid.variant = variant;
 773                if (merge(&vid, path))
 774                        continue; /* failed to replay */
 775
 776                /*
 777                 * If there already is a different variant that applies
 778                 * cleanly, there is no point maintaining our own variant.
 779                 */
 780                if (0 <= id->variant && id->variant != variant)
 781                        remove_variant(id);
 782
 783                if (rerere_autoupdate)
 784                        string_list_insert(update, path);
 785                else
 786                        fprintf_ln(stderr,
 787                                   _("Resolved '%s' using previous resolution."),
 788                                   path);
 789                free_rerere_id(rr_item);
 790                rr_item->util = NULL;
 791                return;
 792        }
 793
 794        /* None of the existing one applies; we need a new variant */
 795        assign_variant(id);
 796
 797        variant = id->variant;
 798        handle_file(path, NULL, rerere_path(id, "preimage"));
 799        if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
 800                const char *path = rerere_path(id, "postimage");
 801                if (unlink(path))
 802                        die_errno(_("cannot unlink stray '%s'"), path);
 803                id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
 804        }
 805        id->collection->status[variant] |= RR_HAS_PREIMAGE;
 806        fprintf_ln(stderr, _("Recorded preimage for '%s'"), path);
 807}
 808
 809static int do_plain_rerere(struct string_list *rr, int fd)
 810{
 811        struct string_list conflict = STRING_LIST_INIT_DUP;
 812        struct string_list update = STRING_LIST_INIT_DUP;
 813        int i;
 814
 815        find_conflict(&conflict);
 816
 817        /*
 818         * MERGE_RR records paths with conflicts immediately after
 819         * merge failed.  Some of the conflicted paths might have been
 820         * hand resolved in the working tree since then, but the
 821         * initial run would catch all and register their preimages.
 822         */
 823        for (i = 0; i < conflict.nr; i++) {
 824                struct rerere_id *id;
 825                unsigned char sha1[20];
 826                const char *path = conflict.items[i].string;
 827                int ret, has_string;
 828
 829                /*
 830                 * Ask handle_file() to scan and assign a
 831                 * conflict ID.  No need to write anything out
 832                 * yet.
 833                 */
 834                ret = handle_file(path, sha1, NULL);
 835                has_string = string_list_has_string(rr, path);
 836                if (ret < 0 && has_string) {
 837                        remove_variant(string_list_lookup(rr, path)->util);
 838                        string_list_remove(rr, path, 1);
 839                }
 840                if (ret < 1 || has_string)
 841                        continue;
 842
 843                id = new_rerere_id(sha1);
 844                string_list_insert(rr, path)->util = id;
 845
 846                /* Ensure that the directory exists. */
 847                mkdir_in_gitdir(rerere_path(id, NULL));
 848        }
 849
 850        for (i = 0; i < rr->nr; i++)
 851                do_rerere_one_path(&rr->items[i], &update);
 852
 853        if (update.nr)
 854                update_paths(&update);
 855
 856        return write_rr(rr, fd);
 857}
 858
 859static 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
 866static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
 867
 868static 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
 884int setup_rerere(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, git_path_merge_rr(),
 898                                               LOCK_DIE_ON_ERROR);
 899        read_rr(merge_rr);
 900        return fd;
 901}
 902
 903/*
 904 * The main entry point that is called internally from codepaths that
 905 * perform mergy operations, possibly leaving conflicted index entries
 906 * and working tree files.
 907 */
 908int rerere(int flags)
 909{
 910        struct string_list merge_rr = STRING_LIST_INIT_DUP;
 911        int fd, status;
 912
 913        fd = setup_rerere(&merge_rr, flags);
 914        if (fd < 0)
 915                return 0;
 916        status = do_plain_rerere(&merge_rr, fd);
 917        free_rerere_dirs();
 918        return status;
 919}
 920
 921/*
 922 * Subclass of rerere_io that reads from an in-core buffer that is a
 923 * strbuf
 924 */
 925struct rerere_io_mem {
 926        struct rerere_io io;
 927        struct strbuf input;
 928};
 929
 930/*
 931 * ... and its getline() method implementation
 932 */
 933static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
 934{
 935        struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
 936        char *ep;
 937        size_t len;
 938
 939        strbuf_release(sb);
 940        if (!io->input.len)
 941                return -1;
 942        ep = memchr(io->input.buf, '\n', io->input.len);
 943        if (!ep)
 944                ep = io->input.buf + io->input.len;
 945        else if (*ep == '\n')
 946                ep++;
 947        len = ep - io->input.buf;
 948        strbuf_add(sb, io->input.buf, len);
 949        strbuf_remove(&io->input, 0, len);
 950        return 0;
 951}
 952
 953static int handle_cache(const char *path, unsigned char *sha1, const char *output)
 954{
 955        mmfile_t mmfile[3] = {{NULL}};
 956        mmbuffer_t result = {NULL, 0};
 957        const struct cache_entry *ce;
 958        int pos, len, i, has_conflicts;
 959        struct rerere_io_mem io;
 960        int marker_size = ll_merge_marker_size(path);
 961
 962        /*
 963         * Reproduce the conflicted merge in-core
 964         */
 965        len = strlen(path);
 966        pos = cache_name_pos(path, len);
 967        if (0 <= pos)
 968                return -1;
 969        pos = -pos - 1;
 970
 971        while (pos < active_nr) {
 972                enum object_type type;
 973                unsigned long size;
 974
 975                ce = active_cache[pos++];
 976                if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
 977                        break;
 978                i = ce_stage(ce) - 1;
 979                if (!mmfile[i].ptr) {
 980                        mmfile[i].ptr = read_object_file(&ce->oid, &type,
 981                                                         &size);
 982                        mmfile[i].size = size;
 983                }
 984        }
 985        for (i = 0; i < 3; i++)
 986                if (!mmfile[i].ptr && !mmfile[i].size)
 987                        mmfile[i].ptr = xstrdup("");
 988
 989        /*
 990         * NEEDSWORK: handle conflicts from merges with
 991         * merge.renormalize set, too?
 992         */
 993        ll_merge(&result, path, &mmfile[0], NULL,
 994                 &mmfile[1], "ours",
 995                 &mmfile[2], "theirs", NULL);
 996        for (i = 0; i < 3; i++)
 997                free(mmfile[i].ptr);
 998
 999        memset(&io, 0, sizeof(io));
1000        io.io.getline = rerere_mem_getline;
1001        if (output)
1002                io.io.output = fopen(output, "w");
1003        else
1004                io.io.output = NULL;
1005        strbuf_init(&io.input, 0);
1006        strbuf_attach(&io.input, result.ptr, result.size, result.size);
1007
1008        /*
1009         * Grab the conflict ID and optionally write the original
1010         * contents with conflict markers out.
1011         */
1012        has_conflicts = handle_path(sha1, (struct rerere_io *)&io, marker_size);
1013        strbuf_release(&io.input);
1014        if (io.io.output)
1015                fclose(io.io.output);
1016        return has_conflicts;
1017}
1018
1019static int rerere_forget_one_path(const char *path, struct string_list *rr)
1020{
1021        const char *filename;
1022        struct rerere_id *id;
1023        unsigned char sha1[20];
1024        int ret;
1025        struct string_list_item *item;
1026
1027        /*
1028         * Recreate the original conflict from the stages in the
1029         * index and compute the conflict ID
1030         */
1031        ret = handle_cache(path, sha1, NULL);
1032        if (ret < 1)
1033                return error(_("could not parse conflict hunks in '%s'"), path);
1034
1035        /* Nuke the recorded resolution for the conflict */
1036        id = new_rerere_id(sha1);
1037
1038        for (id->variant = 0;
1039             id->variant < id->collection->status_nr;
1040             id->variant++) {
1041                mmfile_t cur = { NULL, 0 };
1042                mmbuffer_t result = {NULL, 0};
1043                int cleanly_resolved;
1044
1045                if (!has_rerere_resolution(id))
1046                        continue;
1047
1048                handle_cache(path, sha1, rerere_path(id, "thisimage"));
1049                if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1050                        free(cur.ptr);
1051                        error(_("failed to update conflicted state in '%s'"), path);
1052                        goto fail_exit;
1053                }
1054                cleanly_resolved = !try_merge(id, path, &cur, &result);
1055                free(result.ptr);
1056                free(cur.ptr);
1057                if (cleanly_resolved)
1058                        break;
1059        }
1060
1061        if (id->collection->status_nr <= id->variant) {
1062                error(_("no remembered resolution for '%s'"), path);
1063                goto fail_exit;
1064        }
1065
1066        filename = rerere_path(id, "postimage");
1067        if (unlink(filename)) {
1068                if (errno == ENOENT)
1069                        error(_("no remembered resolution for '%s'"), path);
1070                else
1071                        error_errno(_("cannot unlink '%s'"), filename);
1072                goto fail_exit;
1073        }
1074
1075        /*
1076         * Update the preimage so that the user can resolve the
1077         * conflict in the working tree, run us again to record
1078         * the postimage.
1079         */
1080        handle_cache(path, sha1, rerere_path(id, "preimage"));
1081        fprintf_ln(stderr, _("Updated preimage for '%s'"), path);
1082
1083        /*
1084         * And remember that we can record resolution for this
1085         * conflict when the user is done.
1086         */
1087        item = string_list_insert(rr, path);
1088        free_rerere_id(item);
1089        item->util = id;
1090        fprintf(stderr, _("Forgot resolution for '%s'\n"), path);
1091        return 0;
1092
1093fail_exit:
1094        free(id);
1095        return -1;
1096}
1097
1098int rerere_forget(struct pathspec *pathspec)
1099{
1100        int i, fd;
1101        struct string_list conflict = STRING_LIST_INIT_DUP;
1102        struct string_list merge_rr = STRING_LIST_INIT_DUP;
1103
1104        if (read_cache() < 0)
1105                return error(_("index file corrupt"));
1106
1107        fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
1108        if (fd < 0)
1109                return 0;
1110
1111        /*
1112         * The paths may have been resolved (incorrectly);
1113         * recover the original conflicted state and then
1114         * find the conflicted paths.
1115         */
1116        unmerge_cache(pathspec);
1117        find_conflict(&conflict);
1118        for (i = 0; i < conflict.nr; i++) {
1119                struct string_list_item *it = &conflict.items[i];
1120                if (!match_pathspec(pathspec, it->string,
1121                                    strlen(it->string), 0, NULL, 0))
1122                        continue;
1123                rerere_forget_one_path(it->string, &merge_rr);
1124        }
1125        return write_rr(&merge_rr, fd);
1126}
1127
1128/*
1129 * Garbage collection support
1130 */
1131
1132static timestamp_t rerere_created_at(struct rerere_id *id)
1133{
1134        struct stat st;
1135
1136        return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1137}
1138
1139static timestamp_t rerere_last_used_at(struct rerere_id *id)
1140{
1141        struct stat st;
1142
1143        return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1144}
1145
1146/*
1147 * Remove the recorded resolution for a given conflict ID
1148 */
1149static void unlink_rr_item(struct rerere_id *id)
1150{
1151        unlink_or_warn(rerere_path(id, "thisimage"));
1152        remove_variant(id);
1153        id->collection->status[id->variant] = 0;
1154}
1155
1156static void prune_one(struct rerere_id *id,
1157                      timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1158{
1159        timestamp_t then;
1160        timestamp_t cutoff;
1161
1162        then = rerere_last_used_at(id);
1163        if (then)
1164                cutoff = cutoff_resolve;
1165        else {
1166                then = rerere_created_at(id);
1167                if (!then)
1168                        return;
1169                cutoff = cutoff_noresolve;
1170        }
1171        if (then < cutoff)
1172                unlink_rr_item(id);
1173}
1174
1175void rerere_gc(struct string_list *rr)
1176{
1177        struct string_list to_remove = STRING_LIST_INIT_DUP;
1178        DIR *dir;
1179        struct dirent *e;
1180        int i;
1181        timestamp_t now = time(NULL);
1182        timestamp_t cutoff_noresolve = now - 15 * 86400;
1183        timestamp_t cutoff_resolve = now - 60 * 86400;
1184
1185        if (setup_rerere(rr, 0) < 0)
1186                return;
1187
1188        git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
1189        git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
1190        git_config(git_default_config, NULL);
1191        dir = opendir(git_path("rr-cache"));
1192        if (!dir)
1193                die_errno(_("unable to open rr-cache directory"));
1194        /* Collect stale conflict IDs ... */
1195        while ((e = readdir(dir))) {
1196                struct rerere_dir *rr_dir;
1197                struct rerere_id id;
1198                int now_empty;
1199
1200                if (is_dot_or_dotdot(e->d_name))
1201                        continue;
1202                rr_dir = find_rerere_dir(e->d_name);
1203                if (!rr_dir)
1204                        continue; /* or should we remove e->d_name? */
1205
1206                now_empty = 1;
1207                for (id.variant = 0, id.collection = rr_dir;
1208                     id.variant < id.collection->status_nr;
1209                     id.variant++) {
1210                        prune_one(&id, cutoff_resolve, cutoff_noresolve);
1211                        if (id.collection->status[id.variant])
1212                                now_empty = 0;
1213                }
1214                if (now_empty)
1215                        string_list_append(&to_remove, e->d_name);
1216        }
1217        closedir(dir);
1218
1219        /* ... and then remove the empty directories */
1220        for (i = 0; i < to_remove.nr; i++)
1221                rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
1222        string_list_clear(&to_remove, 0);
1223        rollback_lock_file(&write_lock);
1224}
1225
1226/*
1227 * During a conflict resolution, after "rerere" recorded the
1228 * preimages, abandon them if the user did not resolve them or
1229 * record their resolutions.  And drop $GIT_DIR/MERGE_RR.
1230 *
1231 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1232 */
1233void rerere_clear(struct string_list *merge_rr)
1234{
1235        int i;
1236
1237        if (setup_rerere(merge_rr, 0) < 0)
1238                return;
1239
1240        for (i = 0; i < merge_rr->nr; i++) {
1241                struct rerere_id *id = merge_rr->items[i].util;
1242                if (!has_rerere_resolution(id)) {
1243                        unlink_rr_item(id);
1244                        rmdir(rerere_path(id, NULL));
1245                }
1246        }
1247        unlink_or_warn(git_path_merge_rr());
1248        rollback_lock_file(&write_lock);
1249}