rerere.con commit fetch: make the code more understandable (9528b80)
   1#include "cache.h"
   2#include "config.h"
   3#include "lockfile.h"
   4#include "string-list.h"
   5#include "rerere.h"
   6#include "xdiff-interface.h"
   7#include "dir.h"
   8#include "resolve-undo.h"
   9#include "ll-merge.h"
  10#include "attr.h"
  11#include "pathspec.h"
  12#include "object-store.h"
  13#include "sha1-lookup.h"
  14
  15#define RESOLVED 0
  16#define PUNTED 1
  17#define THREE_STAGED 2
  18void *RERERE_RESOLVED = &RERERE_RESOLVED;
  19
  20/* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
  21static int rerere_enabled = -1;
  22
  23/* automatically update cleanly resolved paths to the index */
  24static int rerere_autoupdate;
  25
  26static int rerere_dir_nr;
  27static int rerere_dir_alloc;
  28
  29#define RR_HAS_POSTIMAGE 1
  30#define RR_HAS_PREIMAGE 2
  31static struct rerere_dir {
  32        unsigned char sha1[20];
  33        int status_alloc, status_nr;
  34        unsigned char *status;
  35} **rerere_dir;
  36
  37static void free_rerere_dirs(void)
  38{
  39        int i;
  40        for (i = 0; i < rerere_dir_nr; i++) {
  41                free(rerere_dir[i]->status);
  42                free(rerere_dir[i]);
  43        }
  44        FREE_AND_NULL(rerere_dir);
  45        rerere_dir_nr = rerere_dir_alloc = 0;
  46}
  47
  48static void free_rerere_id(struct string_list_item *item)
  49{
  50        free(item->util);
  51}
  52
  53static const char *rerere_id_hex(const struct rerere_id *id)
  54{
  55        return sha1_to_hex(id->collection->sha1);
  56}
  57
  58static void fit_variant(struct rerere_dir *rr_dir, int variant)
  59{
  60        variant++;
  61        ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
  62        if (rr_dir->status_nr < variant) {
  63                memset(rr_dir->status + rr_dir->status_nr,
  64                       '\0', variant - rr_dir->status_nr);
  65                rr_dir->status_nr = variant;
  66        }
  67}
  68
  69static void assign_variant(struct rerere_id *id)
  70{
  71        int variant;
  72        struct rerere_dir *rr_dir = id->collection;
  73
  74        variant = id->variant;
  75        if (variant < 0) {
  76                for (variant = 0; variant < rr_dir->status_nr; variant++)
  77                        if (!rr_dir->status[variant])
  78                                break;
  79        }
  80        fit_variant(rr_dir, variant);
  81        id->variant = variant;
  82}
  83
  84const char *rerere_path(const struct rerere_id *id, const char *file)
  85{
  86        if (!file)
  87                return git_path("rr-cache/%s", rerere_id_hex(id));
  88
  89        if (id->variant <= 0)
  90                return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
  91
  92        return git_path("rr-cache/%s/%s.%d",
  93                        rerere_id_hex(id), file, id->variant);
  94}
  95
  96static int is_rr_file(const char *name, const char *filename, int *variant)
  97{
  98        const char *suffix;
  99        char *ep;
 100
 101        if (!strcmp(name, filename)) {
 102                *variant = 0;
 103                return 1;
 104        }
 105        if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
 106                return 0;
 107
 108        errno = 0;
 109        *variant = strtol(suffix + 1, &ep, 10);
 110        if (errno || *ep)
 111                return 0;
 112        return 1;
 113}
 114
 115static void scan_rerere_dir(struct rerere_dir *rr_dir)
 116{
 117        struct dirent *de;
 118        DIR *dir = opendir(git_path("rr-cache/%s", sha1_to_hex(rr_dir->sha1)));
 119
 120        if (!dir)
 121                return;
 122        while ((de = readdir(dir)) != NULL) {
 123                int variant;
 124
 125                if (is_rr_file(de->d_name, "postimage", &variant)) {
 126                        fit_variant(rr_dir, variant);
 127                        rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
 128                } else if (is_rr_file(de->d_name, "preimage", &variant)) {
 129                        fit_variant(rr_dir, variant);
 130                        rr_dir->status[variant] |= RR_HAS_PREIMAGE;
 131                }
 132        }
 133        closedir(dir);
 134}
 135
 136static const unsigned char *rerere_dir_sha1(size_t i, void *table)
 137{
 138        struct rerere_dir **rr_dir = table;
 139        return rr_dir[i]->sha1;
 140}
 141
 142static struct rerere_dir *find_rerere_dir(const char *hex)
 143{
 144        unsigned char sha1[20];
 145        struct rerere_dir *rr_dir;
 146        int pos;
 147
 148        if (get_sha1_hex(hex, sha1))
 149                return NULL; /* BUG */
 150        pos = sha1_pos(sha1, rerere_dir, rerere_dir_nr, rerere_dir_sha1);
 151        if (pos < 0) {
 152                rr_dir = xmalloc(sizeof(*rr_dir));
 153                hashcpy(rr_dir->sha1, sha1);
 154                rr_dir->status = NULL;
 155                rr_dir->status_nr = 0;
 156                rr_dir->status_alloc = 0;
 157                pos = -1 - pos;
 158
 159                /* Make sure the array is big enough ... */
 160                ALLOC_GROW(rerere_dir, rerere_dir_nr + 1, rerere_dir_alloc);
 161                /* ... and add it in. */
 162                rerere_dir_nr++;
 163                MOVE_ARRAY(rerere_dir + pos + 1, rerere_dir + pos,
 164                           rerere_dir_nr - pos - 1);
 165                rerere_dir[pos] = rr_dir;
 166                scan_rerere_dir(rr_dir);
 167        }
 168        return rerere_dir[pos];
 169}
 170
 171static int has_rerere_resolution(const struct rerere_id *id)
 172{
 173        const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
 174        int variant = id->variant;
 175
 176        if (variant < 0)
 177                return 0;
 178        return ((id->collection->status[variant] & both) == both);
 179}
 180
 181static struct rerere_id *new_rerere_id_hex(char *hex)
 182{
 183        struct rerere_id *id = xmalloc(sizeof(*id));
 184        id->collection = find_rerere_dir(hex);
 185        id->variant = -1; /* not known yet */
 186        return id;
 187}
 188
 189static struct rerere_id *new_rerere_id(unsigned char *sha1)
 190{
 191        return new_rerere_id_hex(sha1_to_hex(sha1));
 192}
 193
 194/*
 195 * $GIT_DIR/MERGE_RR file is a collection of records, each of which is
 196 * "conflict ID", a HT and pathname, terminated with a NUL, and is
 197 * used to keep track of the set of paths that "rerere" may need to
 198 * work on (i.e. what is left by the previous invocation of "git
 199 * rerere" during the current conflict resolution session).
 200 */
 201static void read_rr(struct string_list *rr)
 202{
 203        struct strbuf buf = STRBUF_INIT;
 204        FILE *in = fopen_or_warn(git_path_merge_rr(the_repository), "r");
 205
 206        if (!in)
 207                return;
 208        while (!strbuf_getwholeline(&buf, in, '\0')) {
 209                char *path;
 210                unsigned char sha1[20];
 211                struct rerere_id *id;
 212                int variant;
 213
 214                /* There has to be the hash, tab, path and then NUL */
 215                if (buf.len < 42 || get_sha1_hex(buf.buf, sha1))
 216                        die(_("corrupt MERGE_RR"));
 217
 218                if (buf.buf[40] != '.') {
 219                        variant = 0;
 220                        path = buf.buf + 40;
 221                } else {
 222                        errno = 0;
 223                        variant = strtol(buf.buf + 41, &path, 10);
 224                        if (errno)
 225                                die(_("corrupt MERGE_RR"));
 226                }
 227                if (*(path++) != '\t')
 228                        die(_("corrupt MERGE_RR"));
 229                buf.buf[40] = '\0';
 230                id = new_rerere_id_hex(buf.buf);
 231                id->variant = variant;
 232                string_list_insert(rr, path)->util = id;
 233        }
 234        strbuf_release(&buf);
 235        fclose(in);
 236}
 237
 238static struct lock_file write_lock;
 239
 240static int write_rr(struct string_list *rr, int out_fd)
 241{
 242        int i;
 243        for (i = 0; i < rr->nr; i++) {
 244                struct strbuf buf = STRBUF_INIT;
 245                struct rerere_id *id;
 246
 247                assert(rr->items[i].util != RERERE_RESOLVED);
 248
 249                id = rr->items[i].util;
 250                if (!id)
 251                        continue;
 252                assert(id->variant >= 0);
 253                if (0 < id->variant)
 254                        strbuf_addf(&buf, "%s.%d\t%s%c",
 255                                    rerere_id_hex(id), id->variant,
 256                                    rr->items[i].string, 0);
 257                else
 258                        strbuf_addf(&buf, "%s\t%s%c",
 259                                    rerere_id_hex(id),
 260                                    rr->items[i].string, 0);
 261
 262                if (write_in_full(out_fd, buf.buf, buf.len) < 0)
 263                        die(_("unable to write rerere record"));
 264
 265                strbuf_release(&buf);
 266        }
 267        if (commit_lock_file(&write_lock) != 0)
 268                die(_("unable to write rerere record"));
 269        return 0;
 270}
 271
 272/*
 273 * "rerere" interacts with conflicted file contents using this I/O
 274 * abstraction.  It reads a conflicted contents from one place via
 275 * "getline()" method, and optionally can write it out after
 276 * normalizing the conflicted hunks to the "output".  Subclasses of
 277 * rerere_io embed this structure at the beginning of their own
 278 * rerere_io object.
 279 */
 280struct rerere_io {
 281        int (*getline)(struct strbuf *, struct rerere_io *);
 282        FILE *output;
 283        int wrerror;
 284        /* some more stuff */
 285};
 286
 287static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
 288{
 289        if (!count || *err)
 290                return;
 291        if (fwrite(p, count, 1, fp) != 1)
 292                *err = errno;
 293}
 294
 295static inline void ferr_puts(const char *s, FILE *fp, int *err)
 296{
 297        ferr_write(s, strlen(s), fp, err);
 298}
 299
 300static void rerere_io_putstr(const char *str, struct rerere_io *io)
 301{
 302        if (io->output)
 303                ferr_puts(str, io->output, &io->wrerror);
 304}
 305
 306static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
 307{
 308        if (io->output)
 309                ferr_write(mem, sz, io->output, &io->wrerror);
 310}
 311
 312/*
 313 * Subclass of rerere_io that reads from an on-disk file
 314 */
 315struct rerere_io_file {
 316        struct rerere_io io;
 317        FILE *input;
 318};
 319
 320/*
 321 * ... and its getline() method implementation
 322 */
 323static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
 324{
 325        struct rerere_io_file *io = (struct rerere_io_file *)io_;
 326        return strbuf_getwholeline(sb, io->input, '\n');
 327}
 328
 329/*
 330 * Require the exact number of conflict marker letters, no more, no
 331 * less, followed by SP or any whitespace
 332 * (including LF).
 333 */
 334static int is_cmarker(char *buf, int marker_char, int marker_size)
 335{
 336        int want_sp;
 337
 338        /*
 339         * The beginning of our version and the end of their version
 340         * always are labeled like "<<<<< ours" or ">>>>> theirs",
 341         * hence we set want_sp for them.  Note that the version from
 342         * the common ancestor in diff3-style output is not always
 343         * labelled (e.g. "||||| common" is often seen but "|||||"
 344         * alone is also valid), so we do not set want_sp.
 345         */
 346        want_sp = (marker_char == '<') || (marker_char == '>');
 347
 348        while (marker_size--)
 349                if (*buf++ != marker_char)
 350                        return 0;
 351        if (want_sp && *buf != ' ')
 352                return 0;
 353        return isspace(*buf);
 354}
 355
 356static void rerere_strbuf_putconflict(struct strbuf *buf, int ch, size_t size)
 357{
 358        strbuf_addchars(buf, ch, size);
 359        strbuf_addch(buf, '\n');
 360}
 361
 362static int handle_conflict(struct strbuf *out, struct rerere_io *io,
 363                           int marker_size, git_SHA_CTX *ctx)
 364{
 365        enum {
 366                RR_SIDE_1 = 0, RR_SIDE_2, RR_ORIGINAL
 367        } hunk = RR_SIDE_1;
 368        struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
 369        struct strbuf buf = STRBUF_INIT, conflict = STRBUF_INIT;
 370        int has_conflicts = -1;
 371
 372        while (!io->getline(&buf, io)) {
 373                if (is_cmarker(buf.buf, '<', marker_size)) {
 374                        if (handle_conflict(&conflict, io, marker_size, NULL) < 0)
 375                                break;
 376                        if (hunk == RR_SIDE_1)
 377                                strbuf_addbuf(&one, &conflict);
 378                        else
 379                                strbuf_addbuf(&two, &conflict);
 380                        strbuf_release(&conflict);
 381                } else if (is_cmarker(buf.buf, '|', marker_size)) {
 382                        if (hunk != RR_SIDE_1)
 383                                break;
 384                        hunk = RR_ORIGINAL;
 385                } else if (is_cmarker(buf.buf, '=', marker_size)) {
 386                        if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
 387                                break;
 388                        hunk = RR_SIDE_2;
 389                } else if (is_cmarker(buf.buf, '>', marker_size)) {
 390                        if (hunk != RR_SIDE_2)
 391                                break;
 392                        if (strbuf_cmp(&one, &two) > 0)
 393                                strbuf_swap(&one, &two);
 394                        has_conflicts = 1;
 395                        rerere_strbuf_putconflict(out, '<', marker_size);
 396                        strbuf_addbuf(out, &one);
 397                        rerere_strbuf_putconflict(out, '=', marker_size);
 398                        strbuf_addbuf(out, &two);
 399                        rerere_strbuf_putconflict(out, '>', marker_size);
 400                        if (ctx) {
 401                                git_SHA1_Update(ctx, one.buf ? one.buf : "",
 402                                            one.len + 1);
 403                                git_SHA1_Update(ctx, two.buf ? two.buf : "",
 404                                            two.len + 1);
 405                        }
 406                        break;
 407                } else if (hunk == RR_SIDE_1)
 408                        strbuf_addbuf(&one, &buf);
 409                else if (hunk == RR_ORIGINAL)
 410                        ; /* discard */
 411                else if (hunk == RR_SIDE_2)
 412                        strbuf_addbuf(&two, &buf);
 413        }
 414        strbuf_release(&one);
 415        strbuf_release(&two);
 416        strbuf_release(&buf);
 417
 418        return has_conflicts;
 419}
 420
 421/*
 422 * Read contents a file with conflicts, normalize the conflicts
 423 * by (1) discarding the common ancestor version in diff3-style,
 424 * (2) reordering our side and their side so that whichever sorts
 425 * alphabetically earlier comes before the other one, while
 426 * computing the "conflict ID", which is just an SHA-1 hash of
 427 * one side of the conflict, NUL, the other side of the conflict,
 428 * and NUL concatenated together.
 429 *
 430 * Return 1 if conflict hunks are found, 0 if there are no conflict
 431 * hunks and -1 if an error occured.
 432 */
 433static int handle_path(unsigned char *sha1, struct rerere_io *io, int marker_size)
 434{
 435        git_SHA_CTX ctx;
 436        struct strbuf buf = STRBUF_INIT, out = STRBUF_INIT;
 437        int has_conflicts = 0;
 438        if (sha1)
 439                git_SHA1_Init(&ctx);
 440
 441        while (!io->getline(&buf, io)) {
 442                if (is_cmarker(buf.buf, '<', marker_size)) {
 443                        has_conflicts = handle_conflict(&out, io, marker_size,
 444                                                        sha1 ? &ctx : NULL);
 445                        if (has_conflicts < 0)
 446                                break;
 447                        rerere_io_putmem(out.buf, out.len, io);
 448                        strbuf_reset(&out);
 449                } else
 450                        rerere_io_putstr(buf.buf, io);
 451        }
 452        strbuf_release(&buf);
 453        strbuf_release(&out);
 454
 455        if (sha1)
 456                git_SHA1_Final(sha1, &ctx);
 457
 458        return has_conflicts;
 459}
 460
 461/*
 462 * Scan the path for conflicts, do the "handle_path()" thing above, and
 463 * return the number of conflict hunks found.
 464 */
 465static int handle_file(const char *path, unsigned char *sha1, const char *output)
 466{
 467        int has_conflicts = 0;
 468        struct rerere_io_file io;
 469        int marker_size = ll_merge_marker_size(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(sha1, (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 */
 514static int check_one_conflict(int i, int *type)
 515{
 516        const struct cache_entry *e = active_cache[i];
 517
 518        if (!ce_stage(e)) {
 519                *type = RESOLVED;
 520                return i + 1;
 521        }
 522
 523        *type = PUNTED;
 524        while (i < active_nr && ce_stage(active_cache[i]) == 1)
 525                i++;
 526
 527        /* Only handle regular files with both stages #2 and #3 */
 528        if (i + 1 < active_nr) {
 529                const struct cache_entry *e2 = active_cache[i];
 530                const struct cache_entry *e3 = active_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 < active_nr && ce_same_name(e, active_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 */
 556static int find_conflict(struct string_list *conflict)
 557{
 558        int i;
 559        if (read_cache() < 0)
 560                return error(_("index file corrupt"));
 561
 562        for (i = 0; i < active_nr;) {
 563                int conflict_type;
 564                const struct cache_entry *e = active_cache[i];
 565                i = check_one_conflict(i, &conflict_type);
 566                if (conflict_type == THREE_STAGED)
 567                        string_list_insert(conflict, (const char *)e->name);
 568        }
 569        return 0;
 570}
 571
 572/*
 573 * The merge_rr list is meant to hold outstanding conflicted paths
 574 * that rerere could handle.  Abuse the list by adding other types of
 575 * entries to allow the caller to show "rerere remaining".
 576 *
 577 * - Conflicted paths that rerere does not handle are added
 578 * - Conflicted paths that have been resolved are marked as such
 579 *   by storing RERERE_RESOLVED to .util field (where conflict ID
 580 *   is expected to be stored).
 581 *
 582 * Do *not* write MERGE_RR file out after calling this function.
 583 *
 584 * NEEDSWORK: we may want to fix the caller that implements "rerere
 585 * remaining" to do this without abusing merge_rr.
 586 */
 587int rerere_remaining(struct string_list *merge_rr)
 588{
 589        int i;
 590        if (setup_rerere(merge_rr, RERERE_READONLY))
 591                return 0;
 592        if (read_cache() < 0)
 593                return error(_("index file corrupt"));
 594
 595        for (i = 0; i < active_nr;) {
 596                int conflict_type;
 597                const struct cache_entry *e = active_cache[i];
 598                i = check_one_conflict(i, &conflict_type);
 599                if (conflict_type == PUNTED)
 600                        string_list_insert(merge_rr, (const char *)e->name);
 601                else if (conflict_type == RESOLVED) {
 602                        struct string_list_item *it;
 603                        it = string_list_lookup(merge_rr, (const char *)e->name);
 604                        if (it != NULL) {
 605                                free_rerere_id(it);
 606                                it->util = RERERE_RESOLVED;
 607                        }
 608                }
 609        }
 610        return 0;
 611}
 612
 613/*
 614 * Try using the given conflict resolution "ID" to see
 615 * if that recorded conflict resolves cleanly what we
 616 * got in the "cur".
 617 */
 618static int try_merge(const struct rerere_id *id, const char *path,
 619                     mmfile_t *cur, mmbuffer_t *result)
 620{
 621        int ret;
 622        mmfile_t base = {NULL, 0}, other = {NULL, 0};
 623
 624        if (read_mmfile(&base, rerere_path(id, "preimage")) ||
 625            read_mmfile(&other, rerere_path(id, "postimage")))
 626                ret = 1;
 627        else
 628                /*
 629                 * A three-way merge. Note that this honors user-customizable
 630                 * low-level merge driver settings.
 631                 */
 632                ret = ll_merge(result, path, &base, NULL, cur, "", &other, "", NULL);
 633
 634        free(base.ptr);
 635        free(other.ptr);
 636
 637        return ret;
 638}
 639
 640/*
 641 * Find the conflict identified by "id"; the change between its
 642 * "preimage" (i.e. a previous contents with conflict markers) and its
 643 * "postimage" (i.e. the corresponding contents with conflicts
 644 * resolved) may apply cleanly to the contents stored in "path", i.e.
 645 * the conflict this time around.
 646 *
 647 * Returns 0 for successful replay of recorded resolution, or non-zero
 648 * for failure.
 649 */
 650static int merge(const struct rerere_id *id, const char *path)
 651{
 652        FILE *f;
 653        int ret;
 654        mmfile_t cur = {NULL, 0};
 655        mmbuffer_t result = {NULL, 0};
 656
 657        /*
 658         * Normalize the conflicts in path and write it out to
 659         * "thisimage" temporary file.
 660         */
 661        if ((handle_file(path, NULL, rerere_path(id, "thisimage")) < 0) ||
 662            read_mmfile(&cur, rerere_path(id, "thisimage"))) {
 663                ret = 1;
 664                goto out;
 665        }
 666
 667        ret = try_merge(id, path, &cur, &result);
 668        if (ret)
 669                goto out;
 670
 671        /*
 672         * A successful replay of recorded resolution.
 673         * Mark that "postimage" was used to help gc.
 674         */
 675        if (utime(rerere_path(id, "postimage"), NULL) < 0)
 676                warning_errno(_("failed utime() on '%s'"),
 677                              rerere_path(id, "postimage"));
 678
 679        /* Update "path" with the resolution */
 680        f = fopen(path, "w");
 681        if (!f)
 682                return error_errno(_("could not open '%s'"), path);
 683        if (fwrite(result.ptr, result.size, 1, f) != 1)
 684                error_errno(_("could not write '%s'"), path);
 685        if (fclose(f))
 686                return error_errno(_("writing '%s' failed"), path);
 687
 688out:
 689        free(cur.ptr);
 690        free(result.ptr);
 691
 692        return ret;
 693}
 694
 695static void update_paths(struct string_list *update)
 696{
 697        struct lock_file index_lock = LOCK_INIT;
 698        int i;
 699
 700        hold_locked_index(&index_lock, LOCK_DIE_ON_ERROR);
 701
 702        for (i = 0; i < update->nr; i++) {
 703                struct string_list_item *item = &update->items[i];
 704                if (add_file_to_cache(item->string, 0))
 705                        exit(128);
 706                fprintf_ln(stderr, _("Staged '%s' using previous resolution."),
 707                        item->string);
 708        }
 709
 710        if (write_locked_index(&the_index, &index_lock,
 711                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
 712                die(_("unable to write new index file"));
 713}
 714
 715static void remove_variant(struct rerere_id *id)
 716{
 717        unlink_or_warn(rerere_path(id, "postimage"));
 718        unlink_or_warn(rerere_path(id, "preimage"));
 719        id->collection->status[id->variant] = 0;
 720}
 721
 722/*
 723 * The path indicated by rr_item may still have conflict for which we
 724 * have a recorded resolution, in which case replay it and optionally
 725 * update it.  Or it may have been resolved by the user and we may
 726 * only have the preimage for that conflict, in which case the result
 727 * needs to be recorded as a resolution in a postimage file.
 728 */
 729static void do_rerere_one_path(struct string_list_item *rr_item,
 730                               struct string_list *update)
 731{
 732        const char *path = rr_item->string;
 733        struct rerere_id *id = rr_item->util;
 734        struct rerere_dir *rr_dir = id->collection;
 735        int variant;
 736
 737        variant = id->variant;
 738
 739        /* Has the user resolved it already? */
 740        if (variant >= 0) {
 741                if (!handle_file(path, NULL, NULL)) {
 742                        copy_file(rerere_path(id, "postimage"), path, 0666);
 743                        id->collection->status[variant] |= RR_HAS_POSTIMAGE;
 744                        fprintf_ln(stderr, _("Recorded resolution for '%s'."), path);
 745                        free_rerere_id(rr_item);
 746                        rr_item->util = NULL;
 747                        return;
 748                }
 749                /*
 750                 * There may be other variants that can cleanly
 751                 * replay.  Try them and update the variant number for
 752                 * this one.
 753                 */
 754        }
 755
 756        /* Does any existing resolution apply cleanly? */
 757        for (variant = 0; variant < rr_dir->status_nr; variant++) {
 758                const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
 759                struct rerere_id vid = *id;
 760
 761                if ((rr_dir->status[variant] & both) != both)
 762                        continue;
 763
 764                vid.variant = variant;
 765                if (merge(&vid, path))
 766                        continue; /* failed to replay */
 767
 768                /*
 769                 * If there already is a different variant that applies
 770                 * cleanly, there is no point maintaining our own variant.
 771                 */
 772                if (0 <= id->variant && id->variant != variant)
 773                        remove_variant(id);
 774
 775                if (rerere_autoupdate)
 776                        string_list_insert(update, path);
 777                else
 778                        fprintf_ln(stderr,
 779                                   _("Resolved '%s' using previous resolution."),
 780                                   path);
 781                free_rerere_id(rr_item);
 782                rr_item->util = NULL;
 783                return;
 784        }
 785
 786        /* None of the existing one applies; we need a new variant */
 787        assign_variant(id);
 788
 789        variant = id->variant;
 790        handle_file(path, NULL, rerere_path(id, "preimage"));
 791        if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
 792                const char *path = rerere_path(id, "postimage");
 793                if (unlink(path))
 794                        die_errno(_("cannot unlink stray '%s'"), path);
 795                id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
 796        }
 797        id->collection->status[variant] |= RR_HAS_PREIMAGE;
 798        fprintf_ln(stderr, _("Recorded preimage for '%s'"), path);
 799}
 800
 801static int do_plain_rerere(struct string_list *rr, int fd)
 802{
 803        struct string_list conflict = STRING_LIST_INIT_DUP;
 804        struct string_list update = STRING_LIST_INIT_DUP;
 805        int i;
 806
 807        find_conflict(&conflict);
 808
 809        /*
 810         * MERGE_RR records paths with conflicts immediately after
 811         * merge failed.  Some of the conflicted paths might have been
 812         * hand resolved in the working tree since then, but the
 813         * initial run would catch all and register their preimages.
 814         */
 815        for (i = 0; i < conflict.nr; i++) {
 816                struct rerere_id *id;
 817                unsigned char sha1[20];
 818                const char *path = conflict.items[i].string;
 819                int ret;
 820
 821                /*
 822                 * Ask handle_file() to scan and assign a
 823                 * conflict ID.  No need to write anything out
 824                 * yet.
 825                 */
 826                ret = handle_file(path, sha1, NULL);
 827                if (ret != 0 && string_list_has_string(rr, path)) {
 828                        remove_variant(string_list_lookup(rr, path)->util);
 829                        string_list_remove(rr, path, 1);
 830                }
 831                if (ret < 1)
 832                        continue;
 833
 834                id = new_rerere_id(sha1);
 835                string_list_insert(rr, path)->util = id;
 836
 837                /* Ensure that the directory exists. */
 838                mkdir_in_gitdir(rerere_path(id, NULL));
 839        }
 840
 841        for (i = 0; i < rr->nr; i++)
 842                do_rerere_one_path(&rr->items[i], &update);
 843
 844        if (update.nr)
 845                update_paths(&update);
 846
 847        return write_rr(rr, fd);
 848}
 849
 850static void git_rerere_config(void)
 851{
 852        git_config_get_bool("rerere.enabled", &rerere_enabled);
 853        git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
 854        git_config(git_default_config, NULL);
 855}
 856
 857static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
 858
 859static int is_rerere_enabled(void)
 860{
 861        int rr_cache_exists;
 862
 863        if (!rerere_enabled)
 864                return 0;
 865
 866        rr_cache_exists = is_directory(git_path_rr_cache());
 867        if (rerere_enabled < 0)
 868                return rr_cache_exists;
 869
 870        if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
 871                die(_("could not create directory '%s'"), git_path_rr_cache());
 872        return 1;
 873}
 874
 875int setup_rerere(struct string_list *merge_rr, int flags)
 876{
 877        int fd;
 878
 879        git_rerere_config();
 880        if (!is_rerere_enabled())
 881                return -1;
 882
 883        if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
 884                rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
 885        if (flags & RERERE_READONLY)
 886                fd = 0;
 887        else
 888                fd = hold_lock_file_for_update(&write_lock,
 889                                               git_path_merge_rr(the_repository),
 890                                               LOCK_DIE_ON_ERROR);
 891        read_rr(merge_rr);
 892        return fd;
 893}
 894
 895/*
 896 * The main entry point that is called internally from codepaths that
 897 * perform mergy operations, possibly leaving conflicted index entries
 898 * and working tree files.
 899 */
 900int rerere(int flags)
 901{
 902        struct string_list merge_rr = STRING_LIST_INIT_DUP;
 903        int fd, status;
 904
 905        fd = setup_rerere(&merge_rr, flags);
 906        if (fd < 0)
 907                return 0;
 908        status = do_plain_rerere(&merge_rr, fd);
 909        free_rerere_dirs();
 910        return status;
 911}
 912
 913/*
 914 * Subclass of rerere_io that reads from an in-core buffer that is a
 915 * strbuf
 916 */
 917struct rerere_io_mem {
 918        struct rerere_io io;
 919        struct strbuf input;
 920};
 921
 922/*
 923 * ... and its getline() method implementation
 924 */
 925static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
 926{
 927        struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
 928        char *ep;
 929        size_t len;
 930
 931        strbuf_release(sb);
 932        if (!io->input.len)
 933                return -1;
 934        ep = memchr(io->input.buf, '\n', io->input.len);
 935        if (!ep)
 936                ep = io->input.buf + io->input.len;
 937        else if (*ep == '\n')
 938                ep++;
 939        len = ep - io->input.buf;
 940        strbuf_add(sb, io->input.buf, len);
 941        strbuf_remove(&io->input, 0, len);
 942        return 0;
 943}
 944
 945static int handle_cache(const char *path, unsigned char *sha1, const char *output)
 946{
 947        mmfile_t mmfile[3] = {{NULL}};
 948        mmbuffer_t result = {NULL, 0};
 949        const struct cache_entry *ce;
 950        int pos, len, i, has_conflicts;
 951        struct rerere_io_mem io;
 952        int marker_size = ll_merge_marker_size(path);
 953
 954        /*
 955         * Reproduce the conflicted merge in-core
 956         */
 957        len = strlen(path);
 958        pos = cache_name_pos(path, len);
 959        if (0 <= pos)
 960                return -1;
 961        pos = -pos - 1;
 962
 963        while (pos < active_nr) {
 964                enum object_type type;
 965                unsigned long size;
 966
 967                ce = active_cache[pos++];
 968                if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
 969                        break;
 970                i = ce_stage(ce) - 1;
 971                if (!mmfile[i].ptr) {
 972                        mmfile[i].ptr = read_object_file(&ce->oid, &type,
 973                                                         &size);
 974                        mmfile[i].size = size;
 975                }
 976        }
 977        for (i = 0; i < 3; i++)
 978                if (!mmfile[i].ptr && !mmfile[i].size)
 979                        mmfile[i].ptr = xstrdup("");
 980
 981        /*
 982         * NEEDSWORK: handle conflicts from merges with
 983         * merge.renormalize set, too?
 984         */
 985        ll_merge(&result, path, &mmfile[0], NULL,
 986                 &mmfile[1], "ours",
 987                 &mmfile[2], "theirs", NULL);
 988        for (i = 0; i < 3; i++)
 989                free(mmfile[i].ptr);
 990
 991        memset(&io, 0, sizeof(io));
 992        io.io.getline = rerere_mem_getline;
 993        if (output)
 994                io.io.output = fopen(output, "w");
 995        else
 996                io.io.output = NULL;
 997        strbuf_init(&io.input, 0);
 998        strbuf_attach(&io.input, result.ptr, result.size, result.size);
 999
1000        /*
1001         * Grab the conflict ID and optionally write the original
1002         * contents with conflict markers out.
1003         */
1004        has_conflicts = handle_path(sha1, (struct rerere_io *)&io, marker_size);
1005        strbuf_release(&io.input);
1006        if (io.io.output)
1007                fclose(io.io.output);
1008        return has_conflicts;
1009}
1010
1011static int rerere_forget_one_path(const char *path, struct string_list *rr)
1012{
1013        const char *filename;
1014        struct rerere_id *id;
1015        unsigned char sha1[20];
1016        int ret;
1017        struct string_list_item *item;
1018
1019        /*
1020         * Recreate the original conflict from the stages in the
1021         * index and compute the conflict ID
1022         */
1023        ret = handle_cache(path, sha1, NULL);
1024        if (ret < 1)
1025                return error(_("could not parse conflict hunks in '%s'"), path);
1026
1027        /* Nuke the recorded resolution for the conflict */
1028        id = new_rerere_id(sha1);
1029
1030        for (id->variant = 0;
1031             id->variant < id->collection->status_nr;
1032             id->variant++) {
1033                mmfile_t cur = { NULL, 0 };
1034                mmbuffer_t result = {NULL, 0};
1035                int cleanly_resolved;
1036
1037                if (!has_rerere_resolution(id))
1038                        continue;
1039
1040                handle_cache(path, sha1, rerere_path(id, "thisimage"));
1041                if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1042                        free(cur.ptr);
1043                        error(_("failed to update conflicted state in '%s'"), path);
1044                        goto fail_exit;
1045                }
1046                cleanly_resolved = !try_merge(id, path, &cur, &result);
1047                free(result.ptr);
1048                free(cur.ptr);
1049                if (cleanly_resolved)
1050                        break;
1051        }
1052
1053        if (id->collection->status_nr <= id->variant) {
1054                error(_("no remembered resolution for '%s'"), path);
1055                goto fail_exit;
1056        }
1057
1058        filename = rerere_path(id, "postimage");
1059        if (unlink(filename)) {
1060                if (errno == ENOENT)
1061                        error(_("no remembered resolution for '%s'"), path);
1062                else
1063                        error_errno(_("cannot unlink '%s'"), filename);
1064                goto fail_exit;
1065        }
1066
1067        /*
1068         * Update the preimage so that the user can resolve the
1069         * conflict in the working tree, run us again to record
1070         * the postimage.
1071         */
1072        handle_cache(path, sha1, rerere_path(id, "preimage"));
1073        fprintf_ln(stderr, _("Updated preimage for '%s'"), path);
1074
1075        /*
1076         * And remember that we can record resolution for this
1077         * conflict when the user is done.
1078         */
1079        item = string_list_insert(rr, path);
1080        free_rerere_id(item);
1081        item->util = id;
1082        fprintf(stderr, _("Forgot resolution for '%s'\n"), path);
1083        return 0;
1084
1085fail_exit:
1086        free(id);
1087        return -1;
1088}
1089
1090int rerere_forget(struct pathspec *pathspec)
1091{
1092        int i, fd;
1093        struct string_list conflict = STRING_LIST_INIT_DUP;
1094        struct string_list merge_rr = STRING_LIST_INIT_DUP;
1095
1096        if (read_cache() < 0)
1097                return error(_("index file corrupt"));
1098
1099        fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
1100        if (fd < 0)
1101                return 0;
1102
1103        /*
1104         * The paths may have been resolved (incorrectly);
1105         * recover the original conflicted state and then
1106         * find the conflicted paths.
1107         */
1108        unmerge_cache(pathspec);
1109        find_conflict(&conflict);
1110        for (i = 0; i < conflict.nr; i++) {
1111                struct string_list_item *it = &conflict.items[i];
1112                if (!match_pathspec(&the_index, pathspec, it->string,
1113                                    strlen(it->string), 0, NULL, 0))
1114                        continue;
1115                rerere_forget_one_path(it->string, &merge_rr);
1116        }
1117        return write_rr(&merge_rr, fd);
1118}
1119
1120/*
1121 * Garbage collection support
1122 */
1123
1124static timestamp_t rerere_created_at(struct rerere_id *id)
1125{
1126        struct stat st;
1127
1128        return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1129}
1130
1131static timestamp_t rerere_last_used_at(struct rerere_id *id)
1132{
1133        struct stat st;
1134
1135        return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1136}
1137
1138/*
1139 * Remove the recorded resolution for a given conflict ID
1140 */
1141static void unlink_rr_item(struct rerere_id *id)
1142{
1143        unlink_or_warn(rerere_path(id, "thisimage"));
1144        remove_variant(id);
1145        id->collection->status[id->variant] = 0;
1146}
1147
1148static void prune_one(struct rerere_id *id,
1149                      timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1150{
1151        timestamp_t then;
1152        timestamp_t cutoff;
1153
1154        then = rerere_last_used_at(id);
1155        if (then)
1156                cutoff = cutoff_resolve;
1157        else {
1158                then = rerere_created_at(id);
1159                if (!then)
1160                        return;
1161                cutoff = cutoff_noresolve;
1162        }
1163        if (then < cutoff)
1164                unlink_rr_item(id);
1165}
1166
1167void rerere_gc(struct string_list *rr)
1168{
1169        struct string_list to_remove = STRING_LIST_INIT_DUP;
1170        DIR *dir;
1171        struct dirent *e;
1172        int i;
1173        timestamp_t now = time(NULL);
1174        timestamp_t cutoff_noresolve = now - 15 * 86400;
1175        timestamp_t cutoff_resolve = now - 60 * 86400;
1176
1177        if (setup_rerere(rr, 0) < 0)
1178                return;
1179
1180        git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
1181        git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
1182        git_config(git_default_config, NULL);
1183        dir = opendir(git_path("rr-cache"));
1184        if (!dir)
1185                die_errno(_("unable to open rr-cache directory"));
1186        /* Collect stale conflict IDs ... */
1187        while ((e = readdir(dir))) {
1188                struct rerere_dir *rr_dir;
1189                struct rerere_id id;
1190                int now_empty;
1191
1192                if (is_dot_or_dotdot(e->d_name))
1193                        continue;
1194                rr_dir = find_rerere_dir(e->d_name);
1195                if (!rr_dir)
1196                        continue; /* or should we remove e->d_name? */
1197
1198                now_empty = 1;
1199                for (id.variant = 0, id.collection = rr_dir;
1200                     id.variant < id.collection->status_nr;
1201                     id.variant++) {
1202                        prune_one(&id, cutoff_resolve, cutoff_noresolve);
1203                        if (id.collection->status[id.variant])
1204                                now_empty = 0;
1205                }
1206                if (now_empty)
1207                        string_list_append(&to_remove, e->d_name);
1208        }
1209        closedir(dir);
1210
1211        /* ... and then remove the empty directories */
1212        for (i = 0; i < to_remove.nr; i++)
1213                rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
1214        string_list_clear(&to_remove, 0);
1215        rollback_lock_file(&write_lock);
1216}
1217
1218/*
1219 * During a conflict resolution, after "rerere" recorded the
1220 * preimages, abandon them if the user did not resolve them or
1221 * record their resolutions.  And drop $GIT_DIR/MERGE_RR.
1222 *
1223 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1224 */
1225void rerere_clear(struct string_list *merge_rr)
1226{
1227        int i;
1228
1229        if (setup_rerere(merge_rr, 0) < 0)
1230                return;
1231
1232        for (i = 0; i < merge_rr->nr; i++) {
1233                struct rerere_id *id = merge_rr->items[i].util;
1234                if (!has_rerere_resolution(id)) {
1235                        unlink_rr_item(id);
1236                        rmdir(rerere_path(id, NULL));
1237                }
1238        }
1239        unlink_or_warn(git_path_merge_rr(the_repository));
1240        rollback_lock_file(&write_lock);
1241}