rerere.con commit rebase -i: implement the main part of interactive rebase as a builtin (53bbcfb)
   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
 306/*
 307 * Write a conflict marker to io->output (if defined).
 308 */
 309static void rerere_io_putconflict(int ch, int size, struct rerere_io *io)
 310{
 311        char buf[64];
 312
 313        while (size) {
 314                if (size <= sizeof(buf) - 2) {
 315                        memset(buf, ch, size);
 316                        buf[size] = '\n';
 317                        buf[size + 1] = '\0';
 318                        size = 0;
 319                } else {
 320                        int sz = sizeof(buf) - 1;
 321
 322                        /*
 323                         * Make sure we will not write everything out
 324                         * in this round by leaving at least 1 byte
 325                         * for the next round, giving the next round
 326                         * a chance to add the terminating LF.  Yuck.
 327                         */
 328                        if (size <= sz)
 329                                sz -= (sz - size) + 1;
 330                        memset(buf, ch, sz);
 331                        buf[sz] = '\0';
 332                        size -= sz;
 333                }
 334                rerere_io_putstr(buf, io);
 335        }
 336}
 337
 338static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
 339{
 340        if (io->output)
 341                ferr_write(mem, sz, io->output, &io->wrerror);
 342}
 343
 344/*
 345 * Subclass of rerere_io that reads from an on-disk file
 346 */
 347struct rerere_io_file {
 348        struct rerere_io io;
 349        FILE *input;
 350};
 351
 352/*
 353 * ... and its getline() method implementation
 354 */
 355static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
 356{
 357        struct rerere_io_file *io = (struct rerere_io_file *)io_;
 358        return strbuf_getwholeline(sb, io->input, '\n');
 359}
 360
 361/*
 362 * Require the exact number of conflict marker letters, no more, no
 363 * less, followed by SP or any whitespace
 364 * (including LF).
 365 */
 366static int is_cmarker(char *buf, int marker_char, int marker_size)
 367{
 368        int want_sp;
 369
 370        /*
 371         * The beginning of our version and the end of their version
 372         * always are labeled like "<<<<< ours" or ">>>>> theirs",
 373         * hence we set want_sp for them.  Note that the version from
 374         * the common ancestor in diff3-style output is not always
 375         * labelled (e.g. "||||| common" is often seen but "|||||"
 376         * alone is also valid), so we do not set want_sp.
 377         */
 378        want_sp = (marker_char == '<') || (marker_char == '>');
 379
 380        while (marker_size--)
 381                if (*buf++ != marker_char)
 382                        return 0;
 383        if (want_sp && *buf != ' ')
 384                return 0;
 385        return isspace(*buf);
 386}
 387
 388/*
 389 * Read contents a file with conflicts, normalize the conflicts
 390 * by (1) discarding the common ancestor version in diff3-style,
 391 * (2) reordering our side and their side so that whichever sorts
 392 * alphabetically earlier comes before the other one, while
 393 * computing the "conflict ID", which is just an SHA-1 hash of
 394 * one side of the conflict, NUL, the other side of the conflict,
 395 * and NUL concatenated together.
 396 *
 397 * Return the number of conflict hunks found.
 398 *
 399 * NEEDSWORK: the logic and theory of operation behind this conflict
 400 * normalization may deserve to be documented somewhere, perhaps in
 401 * Documentation/technical/rerere.txt.
 402 */
 403static int handle_path(unsigned char *sha1, struct rerere_io *io, int marker_size)
 404{
 405        git_SHA_CTX ctx;
 406        int hunk_no = 0;
 407        enum {
 408                RR_CONTEXT = 0, RR_SIDE_1, RR_SIDE_2, RR_ORIGINAL
 409        } hunk = RR_CONTEXT;
 410        struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
 411        struct strbuf buf = STRBUF_INIT;
 412
 413        if (sha1)
 414                git_SHA1_Init(&ctx);
 415
 416        while (!io->getline(&buf, io)) {
 417                if (is_cmarker(buf.buf, '<', marker_size)) {
 418                        if (hunk != RR_CONTEXT)
 419                                goto bad;
 420                        hunk = RR_SIDE_1;
 421                } else if (is_cmarker(buf.buf, '|', marker_size)) {
 422                        if (hunk != RR_SIDE_1)
 423                                goto bad;
 424                        hunk = RR_ORIGINAL;
 425                } else if (is_cmarker(buf.buf, '=', marker_size)) {
 426                        if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
 427                                goto bad;
 428                        hunk = RR_SIDE_2;
 429                } else if (is_cmarker(buf.buf, '>', marker_size)) {
 430                        if (hunk != RR_SIDE_2)
 431                                goto bad;
 432                        if (strbuf_cmp(&one, &two) > 0)
 433                                strbuf_swap(&one, &two);
 434                        hunk_no++;
 435                        hunk = RR_CONTEXT;
 436                        rerere_io_putconflict('<', marker_size, io);
 437                        rerere_io_putmem(one.buf, one.len, io);
 438                        rerere_io_putconflict('=', marker_size, io);
 439                        rerere_io_putmem(two.buf, two.len, io);
 440                        rerere_io_putconflict('>', marker_size, io);
 441                        if (sha1) {
 442                                git_SHA1_Update(&ctx, one.buf ? one.buf : "",
 443                                            one.len + 1);
 444                                git_SHA1_Update(&ctx, two.buf ? two.buf : "",
 445                                            two.len + 1);
 446                        }
 447                        strbuf_reset(&one);
 448                        strbuf_reset(&two);
 449                } else if (hunk == RR_SIDE_1)
 450                        strbuf_addbuf(&one, &buf);
 451                else if (hunk == RR_ORIGINAL)
 452                        ; /* discard */
 453                else if (hunk == RR_SIDE_2)
 454                        strbuf_addbuf(&two, &buf);
 455                else
 456                        rerere_io_putstr(buf.buf, io);
 457                continue;
 458        bad:
 459                hunk = 99; /* force error exit */
 460                break;
 461        }
 462        strbuf_release(&one);
 463        strbuf_release(&two);
 464        strbuf_release(&buf);
 465
 466        if (sha1)
 467                git_SHA1_Final(sha1, &ctx);
 468        if (hunk != RR_CONTEXT)
 469                return -1;
 470        return hunk_no;
 471}
 472
 473/*
 474 * Scan the path for conflicts, do the "handle_path()" thing above, and
 475 * return the number of conflict hunks found.
 476 */
 477static int handle_file(const char *path, unsigned char *sha1, const char *output)
 478{
 479        int hunk_no = 0;
 480        struct rerere_io_file io;
 481        int marker_size = ll_merge_marker_size(path);
 482
 483        memset(&io, 0, sizeof(io));
 484        io.io.getline = rerere_file_getline;
 485        io.input = fopen(path, "r");
 486        io.io.wrerror = 0;
 487        if (!io.input)
 488                return error_errno("Could not open %s", path);
 489
 490        if (output) {
 491                io.io.output = fopen(output, "w");
 492                if (!io.io.output) {
 493                        error_errno("Could not write %s", output);
 494                        fclose(io.input);
 495                        return -1;
 496                }
 497        }
 498
 499        hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
 500
 501        fclose(io.input);
 502        if (io.io.wrerror)
 503                error("There were errors while writing %s (%s)",
 504                      path, strerror(io.io.wrerror));
 505        if (io.io.output && fclose(io.io.output))
 506                io.io.wrerror = error_errno("Failed to flush %s", path);
 507
 508        if (hunk_no < 0) {
 509                if (output)
 510                        unlink_or_warn(output);
 511                return error("Could not parse conflict hunks in %s", path);
 512        }
 513        if (io.io.wrerror)
 514                return -1;
 515        return hunk_no;
 516}
 517
 518/*
 519 * Look at a cache entry at "i" and see if it is not conflicting,
 520 * conflicting and we are willing to handle, or conflicting and
 521 * we are unable to handle, and return the determination in *type.
 522 * Return the cache index to be looked at next, by skipping the
 523 * stages we have already looked at in this invocation of this
 524 * function.
 525 */
 526static int check_one_conflict(int i, int *type)
 527{
 528        const struct cache_entry *e = active_cache[i];
 529
 530        if (!ce_stage(e)) {
 531                *type = RESOLVED;
 532                return i + 1;
 533        }
 534
 535        *type = PUNTED;
 536        while (ce_stage(active_cache[i]) == 1)
 537                i++;
 538
 539        /* Only handle regular files with both stages #2 and #3 */
 540        if (i + 1 < active_nr) {
 541                const struct cache_entry *e2 = active_cache[i];
 542                const struct cache_entry *e3 = active_cache[i + 1];
 543                if (ce_stage(e2) == 2 &&
 544                    ce_stage(e3) == 3 &&
 545                    ce_same_name(e, e3) &&
 546                    S_ISREG(e2->ce_mode) &&
 547                    S_ISREG(e3->ce_mode))
 548                        *type = THREE_STAGED;
 549        }
 550
 551        /* Skip the entries with the same name */
 552        while (i < active_nr && ce_same_name(e, active_cache[i]))
 553                i++;
 554        return i;
 555}
 556
 557/*
 558 * Scan the index and find paths that have conflicts that rerere can
 559 * handle, i.e. the ones that has both stages #2 and #3.
 560 *
 561 * NEEDSWORK: we do not record or replay a previous "resolve by
 562 * deletion" for a delete-modify conflict, as that is inherently risky
 563 * without knowing what modification is being discarded.  The only
 564 * safe case, i.e. both side doing the deletion and modification that
 565 * are identical to the previous round, might want to be handled,
 566 * though.
 567 */
 568static int find_conflict(struct string_list *conflict)
 569{
 570        int i;
 571        if (read_cache() < 0)
 572                return error("Could not read index");
 573
 574        for (i = 0; i < active_nr;) {
 575                int conflict_type;
 576                const struct cache_entry *e = active_cache[i];
 577                i = check_one_conflict(i, &conflict_type);
 578                if (conflict_type == THREE_STAGED)
 579                        string_list_insert(conflict, (const char *)e->name);
 580        }
 581        return 0;
 582}
 583
 584/*
 585 * The merge_rr list is meant to hold outstanding conflicted paths
 586 * that rerere could handle.  Abuse the list by adding other types of
 587 * entries to allow the caller to show "rerere remaining".
 588 *
 589 * - Conflicted paths that rerere does not handle are added
 590 * - Conflicted paths that have been resolved are marked as such
 591 *   by storing RERERE_RESOLVED to .util field (where conflict ID
 592 *   is expected to be stored).
 593 *
 594 * Do *not* write MERGE_RR file out after calling this function.
 595 *
 596 * NEEDSWORK: we may want to fix the caller that implements "rerere
 597 * remaining" to do this without abusing merge_rr.
 598 */
 599int rerere_remaining(struct string_list *merge_rr)
 600{
 601        int i;
 602        if (setup_rerere(merge_rr, RERERE_READONLY))
 603                return 0;
 604        if (read_cache() < 0)
 605                return error("Could not read index");
 606
 607        for (i = 0; i < active_nr;) {
 608                int conflict_type;
 609                const struct cache_entry *e = active_cache[i];
 610                i = check_one_conflict(i, &conflict_type);
 611                if (conflict_type == PUNTED)
 612                        string_list_insert(merge_rr, (const char *)e->name);
 613                else if (conflict_type == RESOLVED) {
 614                        struct string_list_item *it;
 615                        it = string_list_lookup(merge_rr, (const char *)e->name);
 616                        if (it != NULL) {
 617                                free_rerere_id(it);
 618                                it->util = RERERE_RESOLVED;
 619                        }
 620                }
 621        }
 622        return 0;
 623}
 624
 625/*
 626 * Try using the given conflict resolution "ID" to see
 627 * if that recorded conflict resolves cleanly what we
 628 * got in the "cur".
 629 */
 630static int try_merge(const struct rerere_id *id, const char *path,
 631                     mmfile_t *cur, mmbuffer_t *result)
 632{
 633        int ret;
 634        mmfile_t base = {NULL, 0}, other = {NULL, 0};
 635
 636        if (read_mmfile(&base, rerere_path(id, "preimage")) ||
 637            read_mmfile(&other, rerere_path(id, "postimage")))
 638                ret = 1;
 639        else
 640                /*
 641                 * A three-way merge. Note that this honors user-customizable
 642                 * low-level merge driver settings.
 643                 */
 644                ret = ll_merge(result, path, &base, NULL, cur, "", &other, "", NULL);
 645
 646        free(base.ptr);
 647        free(other.ptr);
 648
 649        return ret;
 650}
 651
 652/*
 653 * Find the conflict identified by "id"; the change between its
 654 * "preimage" (i.e. a previous contents with conflict markers) and its
 655 * "postimage" (i.e. the corresponding contents with conflicts
 656 * resolved) may apply cleanly to the contents stored in "path", i.e.
 657 * the conflict this time around.
 658 *
 659 * Returns 0 for successful replay of recorded resolution, or non-zero
 660 * for failure.
 661 */
 662static int merge(const struct rerere_id *id, const char *path)
 663{
 664        FILE *f;
 665        int ret;
 666        mmfile_t cur = {NULL, 0};
 667        mmbuffer_t result = {NULL, 0};
 668
 669        /*
 670         * Normalize the conflicts in path and write it out to
 671         * "thisimage" temporary file.
 672         */
 673        if ((handle_file(path, NULL, rerere_path(id, "thisimage")) < 0) ||
 674            read_mmfile(&cur, rerere_path(id, "thisimage"))) {
 675                ret = 1;
 676                goto out;
 677        }
 678
 679        ret = try_merge(id, path, &cur, &result);
 680        if (ret)
 681                goto out;
 682
 683        /*
 684         * A successful replay of recorded resolution.
 685         * Mark that "postimage" was used to help gc.
 686         */
 687        if (utime(rerere_path(id, "postimage"), NULL) < 0)
 688                warning_errno("failed utime() on %s",
 689                              rerere_path(id, "postimage"));
 690
 691        /* Update "path" with the resolution */
 692        f = fopen(path, "w");
 693        if (!f)
 694                return error_errno("Could not open %s", path);
 695        if (fwrite(result.ptr, result.size, 1, f) != 1)
 696                error_errno("Could not write %s", path);
 697        if (fclose(f))
 698                return error_errno("Writing %s failed", path);
 699
 700out:
 701        free(cur.ptr);
 702        free(result.ptr);
 703
 704        return ret;
 705}
 706
 707static void update_paths(struct string_list *update)
 708{
 709        struct lock_file index_lock = LOCK_INIT;
 710        int i;
 711
 712        hold_locked_index(&index_lock, LOCK_DIE_ON_ERROR);
 713
 714        for (i = 0; i < update->nr; i++) {
 715                struct string_list_item *item = &update->items[i];
 716                if (add_file_to_cache(item->string, 0))
 717                        exit(128);
 718                fprintf(stderr, "Staged '%s' using previous resolution.\n",
 719                        item->string);
 720        }
 721
 722        if (write_locked_index(&the_index, &index_lock,
 723                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
 724                die("Unable to write new index file");
 725}
 726
 727static void remove_variant(struct rerere_id *id)
 728{
 729        unlink_or_warn(rerere_path(id, "postimage"));
 730        unlink_or_warn(rerere_path(id, "preimage"));
 731        id->collection->status[id->variant] = 0;
 732}
 733
 734/*
 735 * The path indicated by rr_item may still have conflict for which we
 736 * have a recorded resolution, in which case replay it and optionally
 737 * update it.  Or it may have been resolved by the user and we may
 738 * only have the preimage for that conflict, in which case the result
 739 * needs to be recorded as a resolution in a postimage file.
 740 */
 741static void do_rerere_one_path(struct string_list_item *rr_item,
 742                               struct string_list *update)
 743{
 744        const char *path = rr_item->string;
 745        struct rerere_id *id = rr_item->util;
 746        struct rerere_dir *rr_dir = id->collection;
 747        int variant;
 748
 749        variant = id->variant;
 750
 751        /* Has the user resolved it already? */
 752        if (variant >= 0) {
 753                if (!handle_file(path, NULL, NULL)) {
 754                        copy_file(rerere_path(id, "postimage"), path, 0666);
 755                        id->collection->status[variant] |= RR_HAS_POSTIMAGE;
 756                        fprintf(stderr, "Recorded resolution for '%s'.\n", path);
 757                        free_rerere_id(rr_item);
 758                        rr_item->util = NULL;
 759                        return;
 760                }
 761                /*
 762                 * There may be other variants that can cleanly
 763                 * replay.  Try them and update the variant number for
 764                 * this one.
 765                 */
 766        }
 767
 768        /* Does any existing resolution apply cleanly? */
 769        for (variant = 0; variant < rr_dir->status_nr; variant++) {
 770                const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
 771                struct rerere_id vid = *id;
 772
 773                if ((rr_dir->status[variant] & both) != both)
 774                        continue;
 775
 776                vid.variant = variant;
 777                if (merge(&vid, path))
 778                        continue; /* failed to replay */
 779
 780                /*
 781                 * If there already is a different variant that applies
 782                 * cleanly, there is no point maintaining our own variant.
 783                 */
 784                if (0 <= id->variant && id->variant != variant)
 785                        remove_variant(id);
 786
 787                if (rerere_autoupdate)
 788                        string_list_insert(update, path);
 789                else
 790                        fprintf(stderr,
 791                                "Resolved '%s' using previous resolution.\n",
 792                                path);
 793                free_rerere_id(rr_item);
 794                rr_item->util = NULL;
 795                return;
 796        }
 797
 798        /* None of the existing one applies; we need a new variant */
 799        assign_variant(id);
 800
 801        variant = id->variant;
 802        handle_file(path, NULL, rerere_path(id, "preimage"));
 803        if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
 804                const char *path = rerere_path(id, "postimage");
 805                if (unlink(path))
 806                        die_errno("cannot unlink stray '%s'", path);
 807                id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
 808        }
 809        id->collection->status[variant] |= RR_HAS_PREIMAGE;
 810        fprintf(stderr, "Recorded preimage for '%s'\n", path);
 811}
 812
 813static int do_plain_rerere(struct string_list *rr, int fd)
 814{
 815        struct string_list conflict = STRING_LIST_INIT_DUP;
 816        struct string_list update = STRING_LIST_INIT_DUP;
 817        int i;
 818
 819        find_conflict(&conflict);
 820
 821        /*
 822         * MERGE_RR records paths with conflicts immediately after
 823         * merge failed.  Some of the conflicted paths might have been
 824         * hand resolved in the working tree since then, but the
 825         * initial run would catch all and register their preimages.
 826         */
 827        for (i = 0; i < conflict.nr; i++) {
 828                struct rerere_id *id;
 829                unsigned char sha1[20];
 830                const char *path = conflict.items[i].string;
 831                int ret;
 832
 833                if (string_list_has_string(rr, path))
 834                        continue;
 835
 836                /*
 837                 * Ask handle_file() to scan and assign a
 838                 * conflict ID.  No need to write anything out
 839                 * yet.
 840                 */
 841                ret = handle_file(path, sha1, NULL);
 842                if (ret < 1)
 843                        continue;
 844
 845                id = new_rerere_id(sha1);
 846                string_list_insert(rr, path)->util = id;
 847
 848                /* Ensure that the directory exists. */
 849                mkdir_in_gitdir(rerere_path(id, NULL));
 850        }
 851
 852        for (i = 0; i < rr->nr; i++)
 853                do_rerere_one_path(&rr->items[i], &update);
 854
 855        if (update.nr)
 856                update_paths(&update);
 857
 858        return write_rr(rr, fd);
 859}
 860
 861static void git_rerere_config(void)
 862{
 863        git_config_get_bool("rerere.enabled", &rerere_enabled);
 864        git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
 865        git_config(git_default_config, NULL);
 866}
 867
 868static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
 869
 870static int is_rerere_enabled(void)
 871{
 872        int rr_cache_exists;
 873
 874        if (!rerere_enabled)
 875                return 0;
 876
 877        rr_cache_exists = is_directory(git_path_rr_cache());
 878        if (rerere_enabled < 0)
 879                return rr_cache_exists;
 880
 881        if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
 882                die("Could not create directory %s", git_path_rr_cache());
 883        return 1;
 884}
 885
 886int setup_rerere(struct string_list *merge_rr, int flags)
 887{
 888        int fd;
 889
 890        git_rerere_config();
 891        if (!is_rerere_enabled())
 892                return -1;
 893
 894        if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
 895                rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
 896        if (flags & RERERE_READONLY)
 897                fd = 0;
 898        else
 899                fd = hold_lock_file_for_update(&write_lock,
 900                                               git_path_merge_rr(the_repository),
 901                                               LOCK_DIE_ON_ERROR);
 902        read_rr(merge_rr);
 903        return fd;
 904}
 905
 906/*
 907 * The main entry point that is called internally from codepaths that
 908 * perform mergy operations, possibly leaving conflicted index entries
 909 * and working tree files.
 910 */
 911int rerere(int flags)
 912{
 913        struct string_list merge_rr = STRING_LIST_INIT_DUP;
 914        int fd, status;
 915
 916        fd = setup_rerere(&merge_rr, flags);
 917        if (fd < 0)
 918                return 0;
 919        status = do_plain_rerere(&merge_rr, fd);
 920        free_rerere_dirs();
 921        return status;
 922}
 923
 924/*
 925 * Subclass of rerere_io that reads from an in-core buffer that is a
 926 * strbuf
 927 */
 928struct rerere_io_mem {
 929        struct rerere_io io;
 930        struct strbuf input;
 931};
 932
 933/*
 934 * ... and its getline() method implementation
 935 */
 936static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
 937{
 938        struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
 939        char *ep;
 940        size_t len;
 941
 942        strbuf_release(sb);
 943        if (!io->input.len)
 944                return -1;
 945        ep = memchr(io->input.buf, '\n', io->input.len);
 946        if (!ep)
 947                ep = io->input.buf + io->input.len;
 948        else if (*ep == '\n')
 949                ep++;
 950        len = ep - io->input.buf;
 951        strbuf_add(sb, io->input.buf, len);
 952        strbuf_remove(&io->input, 0, len);
 953        return 0;
 954}
 955
 956static int handle_cache(const char *path, unsigned char *sha1, const char *output)
 957{
 958        mmfile_t mmfile[3] = {{NULL}};
 959        mmbuffer_t result = {NULL, 0};
 960        const struct cache_entry *ce;
 961        int pos, len, i, hunk_no;
 962        struct rerere_io_mem io;
 963        int marker_size = ll_merge_marker_size(path);
 964
 965        /*
 966         * Reproduce the conflicted merge in-core
 967         */
 968        len = strlen(path);
 969        pos = cache_name_pos(path, len);
 970        if (0 <= pos)
 971                return -1;
 972        pos = -pos - 1;
 973
 974        while (pos < active_nr) {
 975                enum object_type type;
 976                unsigned long size;
 977
 978                ce = active_cache[pos++];
 979                if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
 980                        break;
 981                i = ce_stage(ce) - 1;
 982                if (!mmfile[i].ptr) {
 983                        mmfile[i].ptr = read_object_file(&ce->oid, &type,
 984                                                         &size);
 985                        mmfile[i].size = size;
 986                }
 987        }
 988        for (i = 0; i < 3; i++)
 989                if (!mmfile[i].ptr && !mmfile[i].size)
 990                        mmfile[i].ptr = xstrdup("");
 991
 992        /*
 993         * NEEDSWORK: handle conflicts from merges with
 994         * merge.renormalize set, too?
 995         */
 996        ll_merge(&result, path, &mmfile[0], NULL,
 997                 &mmfile[1], "ours",
 998                 &mmfile[2], "theirs", NULL);
 999        for (i = 0; i < 3; i++)
1000                free(mmfile[i].ptr);
1001
1002        memset(&io, 0, sizeof(io));
1003        io.io.getline = rerere_mem_getline;
1004        if (output)
1005                io.io.output = fopen(output, "w");
1006        else
1007                io.io.output = NULL;
1008        strbuf_init(&io.input, 0);
1009        strbuf_attach(&io.input, result.ptr, result.size, result.size);
1010
1011        /*
1012         * Grab the conflict ID and optionally write the original
1013         * contents with conflict markers out.
1014         */
1015        hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
1016        strbuf_release(&io.input);
1017        if (io.io.output)
1018                fclose(io.io.output);
1019        return hunk_no;
1020}
1021
1022static int rerere_forget_one_path(const char *path, struct string_list *rr)
1023{
1024        const char *filename;
1025        struct rerere_id *id;
1026        unsigned char sha1[20];
1027        int ret;
1028        struct string_list_item *item;
1029
1030        /*
1031         * Recreate the original conflict from the stages in the
1032         * index and compute the conflict ID
1033         */
1034        ret = handle_cache(path, sha1, NULL);
1035        if (ret < 1)
1036                return error("Could not parse conflict hunks in '%s'", path);
1037
1038        /* Nuke the recorded resolution for the conflict */
1039        id = new_rerere_id(sha1);
1040
1041        for (id->variant = 0;
1042             id->variant < id->collection->status_nr;
1043             id->variant++) {
1044                mmfile_t cur = { NULL, 0 };
1045                mmbuffer_t result = {NULL, 0};
1046                int cleanly_resolved;
1047
1048                if (!has_rerere_resolution(id))
1049                        continue;
1050
1051                handle_cache(path, sha1, rerere_path(id, "thisimage"));
1052                if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1053                        free(cur.ptr);
1054                        error("Failed to update conflicted state in '%s'", path);
1055                        goto fail_exit;
1056                }
1057                cleanly_resolved = !try_merge(id, path, &cur, &result);
1058                free(result.ptr);
1059                free(cur.ptr);
1060                if (cleanly_resolved)
1061                        break;
1062        }
1063
1064        if (id->collection->status_nr <= id->variant) {
1065                error("no remembered resolution for '%s'", path);
1066                goto fail_exit;
1067        }
1068
1069        filename = rerere_path(id, "postimage");
1070        if (unlink(filename)) {
1071                if (errno == ENOENT)
1072                        error("no remembered resolution for %s", path);
1073                else
1074                        error_errno("cannot unlink %s", filename);
1075                goto fail_exit;
1076        }
1077
1078        /*
1079         * Update the preimage so that the user can resolve the
1080         * conflict in the working tree, run us again to record
1081         * the postimage.
1082         */
1083        handle_cache(path, sha1, rerere_path(id, "preimage"));
1084        fprintf(stderr, "Updated preimage for '%s'\n", path);
1085
1086        /*
1087         * And remember that we can record resolution for this
1088         * conflict when the user is done.
1089         */
1090        item = string_list_insert(rr, path);
1091        free_rerere_id(item);
1092        item->util = id;
1093        fprintf(stderr, "Forgot resolution for %s\n", path);
1094        return 0;
1095
1096fail_exit:
1097        free(id);
1098        return -1;
1099}
1100
1101int rerere_forget(struct pathspec *pathspec)
1102{
1103        int i, fd;
1104        struct string_list conflict = STRING_LIST_INIT_DUP;
1105        struct string_list merge_rr = STRING_LIST_INIT_DUP;
1106
1107        if (read_cache() < 0)
1108                return error("Could not read index");
1109
1110        fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
1111        if (fd < 0)
1112                return 0;
1113
1114        /*
1115         * The paths may have been resolved (incorrectly);
1116         * recover the original conflicted state and then
1117         * find the conflicted paths.
1118         */
1119        unmerge_cache(pathspec);
1120        find_conflict(&conflict);
1121        for (i = 0; i < conflict.nr; i++) {
1122                struct string_list_item *it = &conflict.items[i];
1123                if (!match_pathspec(pathspec, it->string,
1124                                    strlen(it->string), 0, NULL, 0))
1125                        continue;
1126                rerere_forget_one_path(it->string, &merge_rr);
1127        }
1128        return write_rr(&merge_rr, fd);
1129}
1130
1131/*
1132 * Garbage collection support
1133 */
1134
1135static timestamp_t rerere_created_at(struct rerere_id *id)
1136{
1137        struct stat st;
1138
1139        return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1140}
1141
1142static timestamp_t rerere_last_used_at(struct rerere_id *id)
1143{
1144        struct stat st;
1145
1146        return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1147}
1148
1149/*
1150 * Remove the recorded resolution for a given conflict ID
1151 */
1152static void unlink_rr_item(struct rerere_id *id)
1153{
1154        unlink_or_warn(rerere_path(id, "thisimage"));
1155        remove_variant(id);
1156        id->collection->status[id->variant] = 0;
1157}
1158
1159static void prune_one(struct rerere_id *id,
1160                      timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1161{
1162        timestamp_t then;
1163        timestamp_t cutoff;
1164
1165        then = rerere_last_used_at(id);
1166        if (then)
1167                cutoff = cutoff_resolve;
1168        else {
1169                then = rerere_created_at(id);
1170                if (!then)
1171                        return;
1172                cutoff = cutoff_noresolve;
1173        }
1174        if (then < cutoff)
1175                unlink_rr_item(id);
1176}
1177
1178void rerere_gc(struct string_list *rr)
1179{
1180        struct string_list to_remove = STRING_LIST_INIT_DUP;
1181        DIR *dir;
1182        struct dirent *e;
1183        int i;
1184        timestamp_t now = time(NULL);
1185        timestamp_t cutoff_noresolve = now - 15 * 86400;
1186        timestamp_t cutoff_resolve = now - 60 * 86400;
1187
1188        if (setup_rerere(rr, 0) < 0)
1189                return;
1190
1191        git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
1192        git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
1193        git_config(git_default_config, NULL);
1194        dir = opendir(git_path("rr-cache"));
1195        if (!dir)
1196                die_errno("unable to open rr-cache directory");
1197        /* Collect stale conflict IDs ... */
1198        while ((e = readdir(dir))) {
1199                struct rerere_dir *rr_dir;
1200                struct rerere_id id;
1201                int now_empty;
1202
1203                if (is_dot_or_dotdot(e->d_name))
1204                        continue;
1205                rr_dir = find_rerere_dir(e->d_name);
1206                if (!rr_dir)
1207                        continue; /* or should we remove e->d_name? */
1208
1209                now_empty = 1;
1210                for (id.variant = 0, id.collection = rr_dir;
1211                     id.variant < id.collection->status_nr;
1212                     id.variant++) {
1213                        prune_one(&id, cutoff_resolve, cutoff_noresolve);
1214                        if (id.collection->status[id.variant])
1215                                now_empty = 0;
1216                }
1217                if (now_empty)
1218                        string_list_append(&to_remove, e->d_name);
1219        }
1220        closedir(dir);
1221
1222        /* ... and then remove the empty directories */
1223        for (i = 0; i < to_remove.nr; i++)
1224                rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
1225        string_list_clear(&to_remove, 0);
1226        rollback_lock_file(&write_lock);
1227}
1228
1229/*
1230 * During a conflict resolution, after "rerere" recorded the
1231 * preimages, abandon them if the user did not resolve them or
1232 * record their resolutions.  And drop $GIT_DIR/MERGE_RR.
1233 *
1234 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1235 */
1236void rerere_clear(struct string_list *merge_rr)
1237{
1238        int i;
1239
1240        if (setup_rerere(merge_rr, 0) < 0)
1241                return;
1242
1243        for (i = 0; i < merge_rr->nr; i++) {
1244                struct rerere_id *id = merge_rr->items[i].util;
1245                if (!has_rerere_resolution(id)) {
1246                        unlink_rr_item(id);
1247                        rmdir(rerere_path(id, NULL));
1248                }
1249        }
1250        unlink_or_warn(git_path_merge_rr(the_repository));
1251        rollback_lock_file(&write_lock);
1252}