rerere.con commit multi-pack-index: load into memory (4d80560)
   1#include "cache.h"
   2#include "config.h"
   3#include "lockfile.h"
   4#include "string-list.h"
   5#include "rerere.h"
   6#include "xdiff-interface.h"
   7#include "dir.h"
   8#include "resolve-undo.h"
   9#include "ll-merge.h"
  10#include "attr.h"
  11#include "pathspec.h"
  12#include "sha1-lookup.h"
  13
  14#define RESOLVED 0
  15#define PUNTED 1
  16#define THREE_STAGED 2
  17void *RERERE_RESOLVED = &RERERE_RESOLVED;
  18
  19/* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
  20static int rerere_enabled = -1;
  21
  22/* automatically update cleanly resolved paths to the index */
  23static int rerere_autoupdate;
  24
  25static int rerere_dir_nr;
  26static int rerere_dir_alloc;
  27
  28#define RR_HAS_POSTIMAGE 1
  29#define RR_HAS_PREIMAGE 2
  30static struct rerere_dir {
  31        unsigned char sha1[20];
  32        int status_alloc, status_nr;
  33        unsigned char *status;
  34} **rerere_dir;
  35
  36static void free_rerere_dirs(void)
  37{
  38        int i;
  39        for (i = 0; i < rerere_dir_nr; i++) {
  40                free(rerere_dir[i]->status);
  41                free(rerere_dir[i]);
  42        }
  43        FREE_AND_NULL(rerere_dir);
  44        rerere_dir_nr = rerere_dir_alloc = 0;
  45}
  46
  47static void free_rerere_id(struct string_list_item *item)
  48{
  49        free(item->util);
  50}
  51
  52static const char *rerere_id_hex(const struct rerere_id *id)
  53{
  54        return sha1_to_hex(id->collection->sha1);
  55}
  56
  57static void fit_variant(struct rerere_dir *rr_dir, int variant)
  58{
  59        variant++;
  60        ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
  61        if (rr_dir->status_nr < variant) {
  62                memset(rr_dir->status + rr_dir->status_nr,
  63                       '\0', variant - rr_dir->status_nr);
  64                rr_dir->status_nr = variant;
  65        }
  66}
  67
  68static void assign_variant(struct rerere_id *id)
  69{
  70        int variant;
  71        struct rerere_dir *rr_dir = id->collection;
  72
  73        variant = id->variant;
  74        if (variant < 0) {
  75                for (variant = 0; variant < rr_dir->status_nr; variant++)
  76                        if (!rr_dir->status[variant])
  77                                break;
  78        }
  79        fit_variant(rr_dir, variant);
  80        id->variant = variant;
  81}
  82
  83const char *rerere_path(const struct rerere_id *id, const char *file)
  84{
  85        if (!file)
  86                return git_path("rr-cache/%s", rerere_id_hex(id));
  87
  88        if (id->variant <= 0)
  89                return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
  90
  91        return git_path("rr-cache/%s/%s.%d",
  92                        rerere_id_hex(id), file, id->variant);
  93}
  94
  95static int is_rr_file(const char *name, const char *filename, int *variant)
  96{
  97        const char *suffix;
  98        char *ep;
  99
 100        if (!strcmp(name, filename)) {
 101                *variant = 0;
 102                return 1;
 103        }
 104        if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
 105                return 0;
 106
 107        errno = 0;
 108        *variant = strtol(suffix + 1, &ep, 10);
 109        if (errno || *ep)
 110                return 0;
 111        return 1;
 112}
 113
 114static void scan_rerere_dir(struct rerere_dir *rr_dir)
 115{
 116        struct dirent *de;
 117        DIR *dir = opendir(git_path("rr-cache/%s", sha1_to_hex(rr_dir->sha1)));
 118
 119        if (!dir)
 120                return;
 121        while ((de = readdir(dir)) != NULL) {
 122                int variant;
 123
 124                if (is_rr_file(de->d_name, "postimage", &variant)) {
 125                        fit_variant(rr_dir, variant);
 126                        rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
 127                } else if (is_rr_file(de->d_name, "preimage", &variant)) {
 128                        fit_variant(rr_dir, variant);
 129                        rr_dir->status[variant] |= RR_HAS_PREIMAGE;
 130                }
 131        }
 132        closedir(dir);
 133}
 134
 135static const unsigned char *rerere_dir_sha1(size_t i, void *table)
 136{
 137        struct rerere_dir **rr_dir = table;
 138        return rr_dir[i]->sha1;
 139}
 140
 141static struct rerere_dir *find_rerere_dir(const char *hex)
 142{
 143        unsigned char sha1[20];
 144        struct rerere_dir *rr_dir;
 145        int pos;
 146
 147        if (get_sha1_hex(hex, sha1))
 148                return NULL; /* BUG */
 149        pos = sha1_pos(sha1, rerere_dir, rerere_dir_nr, rerere_dir_sha1);
 150        if (pos < 0) {
 151                rr_dir = xmalloc(sizeof(*rr_dir));
 152                hashcpy(rr_dir->sha1, sha1);
 153                rr_dir->status = NULL;
 154                rr_dir->status_nr = 0;
 155                rr_dir->status_alloc = 0;
 156                pos = -1 - pos;
 157
 158                /* Make sure the array is big enough ... */
 159                ALLOC_GROW(rerere_dir, rerere_dir_nr + 1, rerere_dir_alloc);
 160                /* ... and add it in. */
 161                rerere_dir_nr++;
 162                MOVE_ARRAY(rerere_dir + pos + 1, rerere_dir + pos,
 163                           rerere_dir_nr - pos - 1);
 164                rerere_dir[pos] = rr_dir;
 165                scan_rerere_dir(rr_dir);
 166        }
 167        return rerere_dir[pos];
 168}
 169
 170static int has_rerere_resolution(const struct rerere_id *id)
 171{
 172        const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
 173        int variant = id->variant;
 174
 175        if (variant < 0)
 176                return 0;
 177        return ((id->collection->status[variant] & both) == both);
 178}
 179
 180static struct rerere_id *new_rerere_id_hex(char *hex)
 181{
 182        struct rerere_id *id = xmalloc(sizeof(*id));
 183        id->collection = find_rerere_dir(hex);
 184        id->variant = -1; /* not known yet */
 185        return id;
 186}
 187
 188static struct rerere_id *new_rerere_id(unsigned char *sha1)
 189{
 190        return new_rerere_id_hex(sha1_to_hex(sha1));
 191}
 192
 193/*
 194 * $GIT_DIR/MERGE_RR file is a collection of records, each of which is
 195 * "conflict ID", a HT and pathname, terminated with a NUL, and is
 196 * used to keep track of the set of paths that "rerere" may need to
 197 * work on (i.e. what is left by the previous invocation of "git
 198 * rerere" during the current conflict resolution session).
 199 */
 200static void read_rr(struct string_list *rr)
 201{
 202        struct strbuf buf = STRBUF_INIT;
 203        FILE *in = fopen_or_warn(git_path_merge_rr(), "r");
 204
 205        if (!in)
 206                return;
 207        while (!strbuf_getwholeline(&buf, in, '\0')) {
 208                char *path;
 209                unsigned char sha1[20];
 210                struct rerere_id *id;
 211                int variant;
 212
 213                /* There has to be the hash, tab, path and then NUL */
 214                if (buf.len < 42 || get_sha1_hex(buf.buf, sha1))
 215                        die("corrupt MERGE_RR");
 216
 217                if (buf.buf[40] != '.') {
 218                        variant = 0;
 219                        path = buf.buf + 40;
 220                } else {
 221                        errno = 0;
 222                        variant = strtol(buf.buf + 41, &path, 10);
 223                        if (errno)
 224                                die("corrupt MERGE_RR");
 225                }
 226                if (*(path++) != '\t')
 227                        die("corrupt MERGE_RR");
 228                buf.buf[40] = '\0';
 229                id = new_rerere_id_hex(buf.buf);
 230                id->variant = variant;
 231                string_list_insert(rr, path)->util = id;
 232        }
 233        strbuf_release(&buf);
 234        fclose(in);
 235}
 236
 237static struct lock_file write_lock;
 238
 239static int write_rr(struct string_list *rr, int out_fd)
 240{
 241        int i;
 242        for (i = 0; i < rr->nr; i++) {
 243                struct strbuf buf = STRBUF_INIT;
 244                struct rerere_id *id;
 245
 246                assert(rr->items[i].util != RERERE_RESOLVED);
 247
 248                id = rr->items[i].util;
 249                if (!id)
 250                        continue;
 251                assert(id->variant >= 0);
 252                if (0 < id->variant)
 253                        strbuf_addf(&buf, "%s.%d\t%s%c",
 254                                    rerere_id_hex(id), id->variant,
 255                                    rr->items[i].string, 0);
 256                else
 257                        strbuf_addf(&buf, "%s\t%s%c",
 258                                    rerere_id_hex(id),
 259                                    rr->items[i].string, 0);
 260
 261                if (write_in_full(out_fd, buf.buf, buf.len) < 0)
 262                        die("unable to write rerere record");
 263
 264                strbuf_release(&buf);
 265        }
 266        if (commit_lock_file(&write_lock) != 0)
 267                die("unable to write rerere record");
 268        return 0;
 269}
 270
 271/*
 272 * "rerere" interacts with conflicted file contents using this I/O
 273 * abstraction.  It reads a conflicted contents from one place via
 274 * "getline()" method, and optionally can write it out after
 275 * normalizing the conflicted hunks to the "output".  Subclasses of
 276 * rerere_io embed this structure at the beginning of their own
 277 * rerere_io object.
 278 */
 279struct rerere_io {
 280        int (*getline)(struct strbuf *, struct rerere_io *);
 281        FILE *output;
 282        int wrerror;
 283        /* some more stuff */
 284};
 285
 286static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
 287{
 288        if (!count || *err)
 289                return;
 290        if (fwrite(p, count, 1, fp) != 1)
 291                *err = errno;
 292}
 293
 294static inline void ferr_puts(const char *s, FILE *fp, int *err)
 295{
 296        ferr_write(s, strlen(s), fp, err);
 297}
 298
 299static void rerere_io_putstr(const char *str, struct rerere_io *io)
 300{
 301        if (io->output)
 302                ferr_puts(str, io->output, &io->wrerror);
 303}
 304
 305/*
 306 * Write a conflict marker to io->output (if defined).
 307 */
 308static void rerere_io_putconflict(int ch, int size, struct rerere_io *io)
 309{
 310        char buf[64];
 311
 312        while (size) {
 313                if (size <= sizeof(buf) - 2) {
 314                        memset(buf, ch, size);
 315                        buf[size] = '\n';
 316                        buf[size + 1] = '\0';
 317                        size = 0;
 318                } else {
 319                        int sz = sizeof(buf) - 1;
 320
 321                        /*
 322                         * Make sure we will not write everything out
 323                         * in this round by leaving at least 1 byte
 324                         * for the next round, giving the next round
 325                         * a chance to add the terminating LF.  Yuck.
 326                         */
 327                        if (size <= sz)
 328                                sz -= (sz - size) + 1;
 329                        memset(buf, ch, sz);
 330                        buf[sz] = '\0';
 331                        size -= sz;
 332                }
 333                rerere_io_putstr(buf, io);
 334        }
 335}
 336
 337static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
 338{
 339        if (io->output)
 340                ferr_write(mem, sz, io->output, &io->wrerror);
 341}
 342
 343/*
 344 * Subclass of rerere_io that reads from an on-disk file
 345 */
 346struct rerere_io_file {
 347        struct rerere_io io;
 348        FILE *input;
 349};
 350
 351/*
 352 * ... and its getline() method implementation
 353 */
 354static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
 355{
 356        struct rerere_io_file *io = (struct rerere_io_file *)io_;
 357        return strbuf_getwholeline(sb, io->input, '\n');
 358}
 359
 360/*
 361 * Require the exact number of conflict marker letters, no more, no
 362 * less, followed by SP or any whitespace
 363 * (including LF).
 364 */
 365static int is_cmarker(char *buf, int marker_char, int marker_size)
 366{
 367        int want_sp;
 368
 369        /*
 370         * The beginning of our version and the end of their version
 371         * always are labeled like "<<<<< ours" or ">>>>> theirs",
 372         * hence we set want_sp for them.  Note that the version from
 373         * the common ancestor in diff3-style output is not always
 374         * labelled (e.g. "||||| common" is often seen but "|||||"
 375         * alone is also valid), so we do not set want_sp.
 376         */
 377        want_sp = (marker_char == '<') || (marker_char == '>');
 378
 379        while (marker_size--)
 380                if (*buf++ != marker_char)
 381                        return 0;
 382        if (want_sp && *buf != ' ')
 383                return 0;
 384        return isspace(*buf);
 385}
 386
 387/*
 388 * Read contents a file with conflicts, normalize the conflicts
 389 * by (1) discarding the common ancestor version in diff3-style,
 390 * (2) reordering our side and their side so that whichever sorts
 391 * alphabetically earlier comes before the other one, while
 392 * computing the "conflict ID", which is just an SHA-1 hash of
 393 * one side of the conflict, NUL, the other side of the conflict,
 394 * and NUL concatenated together.
 395 *
 396 * Return the number of conflict hunks found.
 397 *
 398 * NEEDSWORK: the logic and theory of operation behind this conflict
 399 * normalization may deserve to be documented somewhere, perhaps in
 400 * Documentation/technical/rerere.txt.
 401 */
 402static int handle_path(unsigned char *sha1, struct rerere_io *io, int marker_size)
 403{
 404        git_SHA_CTX ctx;
 405        int hunk_no = 0;
 406        enum {
 407                RR_CONTEXT = 0, RR_SIDE_1, RR_SIDE_2, RR_ORIGINAL
 408        } hunk = RR_CONTEXT;
 409        struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
 410        struct strbuf buf = STRBUF_INIT;
 411
 412        if (sha1)
 413                git_SHA1_Init(&ctx);
 414
 415        while (!io->getline(&buf, io)) {
 416                if (is_cmarker(buf.buf, '<', marker_size)) {
 417                        if (hunk != RR_CONTEXT)
 418                                goto bad;
 419                        hunk = RR_SIDE_1;
 420                } else if (is_cmarker(buf.buf, '|', marker_size)) {
 421                        if (hunk != RR_SIDE_1)
 422                                goto bad;
 423                        hunk = RR_ORIGINAL;
 424                } else if (is_cmarker(buf.buf, '=', marker_size)) {
 425                        if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
 426                                goto bad;
 427                        hunk = RR_SIDE_2;
 428                } else if (is_cmarker(buf.buf, '>', marker_size)) {
 429                        if (hunk != RR_SIDE_2)
 430                                goto bad;
 431                        if (strbuf_cmp(&one, &two) > 0)
 432                                strbuf_swap(&one, &two);
 433                        hunk_no++;
 434                        hunk = RR_CONTEXT;
 435                        rerere_io_putconflict('<', marker_size, io);
 436                        rerere_io_putmem(one.buf, one.len, io);
 437                        rerere_io_putconflict('=', marker_size, io);
 438                        rerere_io_putmem(two.buf, two.len, io);
 439                        rerere_io_putconflict('>', marker_size, io);
 440                        if (sha1) {
 441                                git_SHA1_Update(&ctx, one.buf ? one.buf : "",
 442                                            one.len + 1);
 443                                git_SHA1_Update(&ctx, two.buf ? two.buf : "",
 444                                            two.len + 1);
 445                        }
 446                        strbuf_reset(&one);
 447                        strbuf_reset(&two);
 448                } else if (hunk == RR_SIDE_1)
 449                        strbuf_addbuf(&one, &buf);
 450                else if (hunk == RR_ORIGINAL)
 451                        ; /* discard */
 452                else if (hunk == RR_SIDE_2)
 453                        strbuf_addbuf(&two, &buf);
 454                else
 455                        rerere_io_putstr(buf.buf, io);
 456                continue;
 457        bad:
 458                hunk = 99; /* force error exit */
 459                break;
 460        }
 461        strbuf_release(&one);
 462        strbuf_release(&two);
 463        strbuf_release(&buf);
 464
 465        if (sha1)
 466                git_SHA1_Final(sha1, &ctx);
 467        if (hunk != RR_CONTEXT)
 468                return -1;
 469        return hunk_no;
 470}
 471
 472/*
 473 * Scan the path for conflicts, do the "handle_path()" thing above, and
 474 * return the number of conflict hunks found.
 475 */
 476static int handle_file(const char *path, unsigned char *sha1, const char *output)
 477{
 478        int hunk_no = 0;
 479        struct rerere_io_file io;
 480        int marker_size = ll_merge_marker_size(path);
 481
 482        memset(&io, 0, sizeof(io));
 483        io.io.getline = rerere_file_getline;
 484        io.input = fopen(path, "r");
 485        io.io.wrerror = 0;
 486        if (!io.input)
 487                return error_errno("Could not open %s", path);
 488
 489        if (output) {
 490                io.io.output = fopen(output, "w");
 491                if (!io.io.output) {
 492                        error_errno("Could not write %s", output);
 493                        fclose(io.input);
 494                        return -1;
 495                }
 496        }
 497
 498        hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
 499
 500        fclose(io.input);
 501        if (io.io.wrerror)
 502                error("There were errors while writing %s (%s)",
 503                      path, strerror(io.io.wrerror));
 504        if (io.io.output && fclose(io.io.output))
 505                io.io.wrerror = error_errno("Failed to flush %s", path);
 506
 507        if (hunk_no < 0) {
 508                if (output)
 509                        unlink_or_warn(output);
 510                return error("Could not parse conflict hunks in %s", path);
 511        }
 512        if (io.io.wrerror)
 513                return -1;
 514        return hunk_no;
 515}
 516
 517/*
 518 * Look at a cache entry at "i" and see if it is not conflicting,
 519 * conflicting and we are willing to handle, or conflicting and
 520 * we are unable to handle, and return the determination in *type.
 521 * Return the cache index to be looked at next, by skipping the
 522 * stages we have already looked at in this invocation of this
 523 * function.
 524 */
 525static int check_one_conflict(int i, int *type)
 526{
 527        const struct cache_entry *e = active_cache[i];
 528
 529        if (!ce_stage(e)) {
 530                *type = RESOLVED;
 531                return i + 1;
 532        }
 533
 534        *type = PUNTED;
 535        while (ce_stage(active_cache[i]) == 1)
 536                i++;
 537
 538        /* Only handle regular files with both stages #2 and #3 */
 539        if (i + 1 < active_nr) {
 540                const struct cache_entry *e2 = active_cache[i];
 541                const struct cache_entry *e3 = active_cache[i + 1];
 542                if (ce_stage(e2) == 2 &&
 543                    ce_stage(e3) == 3 &&
 544                    ce_same_name(e, e3) &&
 545                    S_ISREG(e2->ce_mode) &&
 546                    S_ISREG(e3->ce_mode))
 547                        *type = THREE_STAGED;
 548        }
 549
 550        /* Skip the entries with the same name */
 551        while (i < active_nr && ce_same_name(e, active_cache[i]))
 552                i++;
 553        return i;
 554}
 555
 556/*
 557 * Scan the index and find paths that have conflicts that rerere can
 558 * handle, i.e. the ones that has both stages #2 and #3.
 559 *
 560 * NEEDSWORK: we do not record or replay a previous "resolve by
 561 * deletion" for a delete-modify conflict, as that is inherently risky
 562 * without knowing what modification is being discarded.  The only
 563 * safe case, i.e. both side doing the deletion and modification that
 564 * are identical to the previous round, might want to be handled,
 565 * though.
 566 */
 567static int find_conflict(struct string_list *conflict)
 568{
 569        int i;
 570        if (read_cache() < 0)
 571                return error("Could not read index");
 572
 573        for (i = 0; i < active_nr;) {
 574                int conflict_type;
 575                const struct cache_entry *e = active_cache[i];
 576                i = check_one_conflict(i, &conflict_type);
 577                if (conflict_type == THREE_STAGED)
 578                        string_list_insert(conflict, (const char *)e->name);
 579        }
 580        return 0;
 581}
 582
 583/*
 584 * The merge_rr list is meant to hold outstanding conflicted paths
 585 * that rerere could handle.  Abuse the list by adding other types of
 586 * entries to allow the caller to show "rerere remaining".
 587 *
 588 * - Conflicted paths that rerere does not handle are added
 589 * - Conflicted paths that have been resolved are marked as such
 590 *   by storing RERERE_RESOLVED to .util field (where conflict ID
 591 *   is expected to be stored).
 592 *
 593 * Do *not* write MERGE_RR file out after calling this function.
 594 *
 595 * NEEDSWORK: we may want to fix the caller that implements "rerere
 596 * remaining" to do this without abusing merge_rr.
 597 */
 598int rerere_remaining(struct string_list *merge_rr)
 599{
 600        int i;
 601        if (setup_rerere(merge_rr, RERERE_READONLY))
 602                return 0;
 603        if (read_cache() < 0)
 604                return error("Could not read index");
 605
 606        for (i = 0; i < active_nr;) {
 607                int conflict_type;
 608                const struct cache_entry *e = active_cache[i];
 609                i = check_one_conflict(i, &conflict_type);
 610                if (conflict_type == PUNTED)
 611                        string_list_insert(merge_rr, (const char *)e->name);
 612                else if (conflict_type == RESOLVED) {
 613                        struct string_list_item *it;
 614                        it = string_list_lookup(merge_rr, (const char *)e->name);
 615                        if (it != NULL) {
 616                                free_rerere_id(it);
 617                                it->util = RERERE_RESOLVED;
 618                        }
 619                }
 620        }
 621        return 0;
 622}
 623
 624/*
 625 * Try using the given conflict resolution "ID" to see
 626 * if that recorded conflict resolves cleanly what we
 627 * got in the "cur".
 628 */
 629static int try_merge(const struct rerere_id *id, const char *path,
 630                     mmfile_t *cur, mmbuffer_t *result)
 631{
 632        int ret;
 633        mmfile_t base = {NULL, 0}, other = {NULL, 0};
 634
 635        if (read_mmfile(&base, rerere_path(id, "preimage")) ||
 636            read_mmfile(&other, rerere_path(id, "postimage")))
 637                ret = 1;
 638        else
 639                /*
 640                 * A three-way merge. Note that this honors user-customizable
 641                 * low-level merge driver settings.
 642                 */
 643                ret = ll_merge(result, path, &base, NULL, cur, "", &other, "", NULL);
 644
 645        free(base.ptr);
 646        free(other.ptr);
 647
 648        return ret;
 649}
 650
 651/*
 652 * Find the conflict identified by "id"; the change between its
 653 * "preimage" (i.e. a previous contents with conflict markers) and its
 654 * "postimage" (i.e. the corresponding contents with conflicts
 655 * resolved) may apply cleanly to the contents stored in "path", i.e.
 656 * the conflict this time around.
 657 *
 658 * Returns 0 for successful replay of recorded resolution, or non-zero
 659 * for failure.
 660 */
 661static int merge(const struct rerere_id *id, const char *path)
 662{
 663        FILE *f;
 664        int ret;
 665        mmfile_t cur = {NULL, 0};
 666        mmbuffer_t result = {NULL, 0};
 667
 668        /*
 669         * Normalize the conflicts in path and write it out to
 670         * "thisimage" temporary file.
 671         */
 672        if ((handle_file(path, NULL, rerere_path(id, "thisimage")) < 0) ||
 673            read_mmfile(&cur, rerere_path(id, "thisimage"))) {
 674                ret = 1;
 675                goto out;
 676        }
 677
 678        ret = try_merge(id, path, &cur, &result);
 679        if (ret)
 680                goto out;
 681
 682        /*
 683         * A successful replay of recorded resolution.
 684         * Mark that "postimage" was used to help gc.
 685         */
 686        if (utime(rerere_path(id, "postimage"), NULL) < 0)
 687                warning_errno("failed utime() on %s",
 688                              rerere_path(id, "postimage"));
 689
 690        /* Update "path" with the resolution */
 691        f = fopen(path, "w");
 692        if (!f)
 693                return error_errno("Could not open %s", path);
 694        if (fwrite(result.ptr, result.size, 1, f) != 1)
 695                error_errno("Could not write %s", path);
 696        if (fclose(f))
 697                return error_errno("Writing %s failed", path);
 698
 699out:
 700        free(cur.ptr);
 701        free(result.ptr);
 702
 703        return ret;
 704}
 705
 706static void update_paths(struct string_list *update)
 707{
 708        struct lock_file index_lock = LOCK_INIT;
 709        int i;
 710
 711        hold_locked_index(&index_lock, LOCK_DIE_ON_ERROR);
 712
 713        for (i = 0; i < update->nr; i++) {
 714                struct string_list_item *item = &update->items[i];
 715                if (add_file_to_cache(item->string, 0))
 716                        exit(128);
 717                fprintf(stderr, "Staged '%s' using previous resolution.\n",
 718                        item->string);
 719        }
 720
 721        if (write_locked_index(&the_index, &index_lock,
 722                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
 723                die("Unable to write new index file");
 724}
 725
 726static void remove_variant(struct rerere_id *id)
 727{
 728        unlink_or_warn(rerere_path(id, "postimage"));
 729        unlink_or_warn(rerere_path(id, "preimage"));
 730        id->collection->status[id->variant] = 0;
 731}
 732
 733/*
 734 * The path indicated by rr_item may still have conflict for which we
 735 * have a recorded resolution, in which case replay it and optionally
 736 * update it.  Or it may have been resolved by the user and we may
 737 * only have the preimage for that conflict, in which case the result
 738 * needs to be recorded as a resolution in a postimage file.
 739 */
 740static void do_rerere_one_path(struct string_list_item *rr_item,
 741                               struct string_list *update)
 742{
 743        const char *path = rr_item->string;
 744        struct rerere_id *id = rr_item->util;
 745        struct rerere_dir *rr_dir = id->collection;
 746        int variant;
 747
 748        variant = id->variant;
 749
 750        /* Has the user resolved it already? */
 751        if (variant >= 0) {
 752                if (!handle_file(path, NULL, NULL)) {
 753                        copy_file(rerere_path(id, "postimage"), path, 0666);
 754                        id->collection->status[variant] |= RR_HAS_POSTIMAGE;
 755                        fprintf(stderr, "Recorded resolution for '%s'.\n", path);
 756                        free_rerere_id(rr_item);
 757                        rr_item->util = NULL;
 758                        return;
 759                }
 760                /*
 761                 * There may be other variants that can cleanly
 762                 * replay.  Try them and update the variant number for
 763                 * this one.
 764                 */
 765        }
 766
 767        /* Does any existing resolution apply cleanly? */
 768        for (variant = 0; variant < rr_dir->status_nr; variant++) {
 769                const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
 770                struct rerere_id vid = *id;
 771
 772                if ((rr_dir->status[variant] & both) != both)
 773                        continue;
 774
 775                vid.variant = variant;
 776                if (merge(&vid, path))
 777                        continue; /* failed to replay */
 778
 779                /*
 780                 * If there already is a different variant that applies
 781                 * cleanly, there is no point maintaining our own variant.
 782                 */
 783                if (0 <= id->variant && id->variant != variant)
 784                        remove_variant(id);
 785
 786                if (rerere_autoupdate)
 787                        string_list_insert(update, path);
 788                else
 789                        fprintf(stderr,
 790                                "Resolved '%s' using previous resolution.\n",
 791                                path);
 792                free_rerere_id(rr_item);
 793                rr_item->util = NULL;
 794                return;
 795        }
 796
 797        /* None of the existing one applies; we need a new variant */
 798        assign_variant(id);
 799
 800        variant = id->variant;
 801        handle_file(path, NULL, rerere_path(id, "preimage"));
 802        if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
 803                const char *path = rerere_path(id, "postimage");
 804                if (unlink(path))
 805                        die_errno("cannot unlink stray '%s'", path);
 806                id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
 807        }
 808        id->collection->status[variant] |= RR_HAS_PREIMAGE;
 809        fprintf(stderr, "Recorded preimage for '%s'\n", path);
 810}
 811
 812static int do_plain_rerere(struct string_list *rr, int fd)
 813{
 814        struct string_list conflict = STRING_LIST_INIT_DUP;
 815        struct string_list update = STRING_LIST_INIT_DUP;
 816        int i;
 817
 818        find_conflict(&conflict);
 819
 820        /*
 821         * MERGE_RR records paths with conflicts immediately after
 822         * merge failed.  Some of the conflicted paths might have been
 823         * hand resolved in the working tree since then, but the
 824         * initial run would catch all and register their preimages.
 825         */
 826        for (i = 0; i < conflict.nr; i++) {
 827                struct rerere_id *id;
 828                unsigned char sha1[20];
 829                const char *path = conflict.items[i].string;
 830                int ret;
 831
 832                if (string_list_has_string(rr, path))
 833                        continue;
 834
 835                /*
 836                 * Ask handle_file() to scan and assign a
 837                 * conflict ID.  No need to write anything out
 838                 * yet.
 839                 */
 840                ret = handle_file(path, sha1, NULL);
 841                if (ret < 1)
 842                        continue;
 843
 844                id = new_rerere_id(sha1);
 845                string_list_insert(rr, path)->util = id;
 846
 847                /* Ensure that the directory exists. */
 848                mkdir_in_gitdir(rerere_path(id, NULL));
 849        }
 850
 851        for (i = 0; i < rr->nr; i++)
 852                do_rerere_one_path(&rr->items[i], &update);
 853
 854        if (update.nr)
 855                update_paths(&update);
 856
 857        return write_rr(rr, fd);
 858}
 859
 860static void git_rerere_config(void)
 861{
 862        git_config_get_bool("rerere.enabled", &rerere_enabled);
 863        git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
 864        git_config(git_default_config, NULL);
 865}
 866
 867static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
 868
 869static int is_rerere_enabled(void)
 870{
 871        int rr_cache_exists;
 872
 873        if (!rerere_enabled)
 874                return 0;
 875
 876        rr_cache_exists = is_directory(git_path_rr_cache());
 877        if (rerere_enabled < 0)
 878                return rr_cache_exists;
 879
 880        if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
 881                die("Could not create directory %s", git_path_rr_cache());
 882        return 1;
 883}
 884
 885int setup_rerere(struct string_list *merge_rr, int flags)
 886{
 887        int fd;
 888
 889        git_rerere_config();
 890        if (!is_rerere_enabled())
 891                return -1;
 892
 893        if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
 894                rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
 895        if (flags & RERERE_READONLY)
 896                fd = 0;
 897        else
 898                fd = hold_lock_file_for_update(&write_lock, git_path_merge_rr(),
 899                                               LOCK_DIE_ON_ERROR);
 900        read_rr(merge_rr);
 901        return fd;
 902}
 903
 904/*
 905 * The main entry point that is called internally from codepaths that
 906 * perform mergy operations, possibly leaving conflicted index entries
 907 * and working tree files.
 908 */
 909int rerere(int flags)
 910{
 911        struct string_list merge_rr = STRING_LIST_INIT_DUP;
 912        int fd, status;
 913
 914        fd = setup_rerere(&merge_rr, flags);
 915        if (fd < 0)
 916                return 0;
 917        status = do_plain_rerere(&merge_rr, fd);
 918        free_rerere_dirs();
 919        return status;
 920}
 921
 922/*
 923 * Subclass of rerere_io that reads from an in-core buffer that is a
 924 * strbuf
 925 */
 926struct rerere_io_mem {
 927        struct rerere_io io;
 928        struct strbuf input;
 929};
 930
 931/*
 932 * ... and its getline() method implementation
 933 */
 934static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
 935{
 936        struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
 937        char *ep;
 938        size_t len;
 939
 940        strbuf_release(sb);
 941        if (!io->input.len)
 942                return -1;
 943        ep = memchr(io->input.buf, '\n', io->input.len);
 944        if (!ep)
 945                ep = io->input.buf + io->input.len;
 946        else if (*ep == '\n')
 947                ep++;
 948        len = ep - io->input.buf;
 949        strbuf_add(sb, io->input.buf, len);
 950        strbuf_remove(&io->input, 0, len);
 951        return 0;
 952}
 953
 954static int handle_cache(const char *path, unsigned char *sha1, const char *output)
 955{
 956        mmfile_t mmfile[3] = {{NULL}};
 957        mmbuffer_t result = {NULL, 0};
 958        const struct cache_entry *ce;
 959        int pos, len, i, hunk_no;
 960        struct rerere_io_mem io;
 961        int marker_size = ll_merge_marker_size(path);
 962
 963        /*
 964         * Reproduce the conflicted merge in-core
 965         */
 966        len = strlen(path);
 967        pos = cache_name_pos(path, len);
 968        if (0 <= pos)
 969                return -1;
 970        pos = -pos - 1;
 971
 972        while (pos < active_nr) {
 973                enum object_type type;
 974                unsigned long size;
 975
 976                ce = active_cache[pos++];
 977                if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
 978                        break;
 979                i = ce_stage(ce) - 1;
 980                if (!mmfile[i].ptr) {
 981                        mmfile[i].ptr = read_object_file(&ce->oid, &type,
 982                                                         &size);
 983                        mmfile[i].size = size;
 984                }
 985        }
 986        for (i = 0; i < 3; i++)
 987                if (!mmfile[i].ptr && !mmfile[i].size)
 988                        mmfile[i].ptr = xstrdup("");
 989
 990        /*
 991         * NEEDSWORK: handle conflicts from merges with
 992         * merge.renormalize set, too?
 993         */
 994        ll_merge(&result, path, &mmfile[0], NULL,
 995                 &mmfile[1], "ours",
 996                 &mmfile[2], "theirs", NULL);
 997        for (i = 0; i < 3; i++)
 998                free(mmfile[i].ptr);
 999
1000        memset(&io, 0, sizeof(io));
1001        io.io.getline = rerere_mem_getline;
1002        if (output)
1003                io.io.output = fopen(output, "w");
1004        else
1005                io.io.output = NULL;
1006        strbuf_init(&io.input, 0);
1007        strbuf_attach(&io.input, result.ptr, result.size, result.size);
1008
1009        /*
1010         * Grab the conflict ID and optionally write the original
1011         * contents with conflict markers out.
1012         */
1013        hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
1014        strbuf_release(&io.input);
1015        if (io.io.output)
1016                fclose(io.io.output);
1017        return hunk_no;
1018}
1019
1020static int rerere_forget_one_path(const char *path, struct string_list *rr)
1021{
1022        const char *filename;
1023        struct rerere_id *id;
1024        unsigned char sha1[20];
1025        int ret;
1026        struct string_list_item *item;
1027
1028        /*
1029         * Recreate the original conflict from the stages in the
1030         * index and compute the conflict ID
1031         */
1032        ret = handle_cache(path, sha1, NULL);
1033        if (ret < 1)
1034                return error("Could not parse conflict hunks in '%s'", path);
1035
1036        /* Nuke the recorded resolution for the conflict */
1037        id = new_rerere_id(sha1);
1038
1039        for (id->variant = 0;
1040             id->variant < id->collection->status_nr;
1041             id->variant++) {
1042                mmfile_t cur = { NULL, 0 };
1043                mmbuffer_t result = {NULL, 0};
1044                int cleanly_resolved;
1045
1046                if (!has_rerere_resolution(id))
1047                        continue;
1048
1049                handle_cache(path, sha1, rerere_path(id, "thisimage"));
1050                if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
1051                        free(cur.ptr);
1052                        error("Failed to update conflicted state in '%s'", path);
1053                        goto fail_exit;
1054                }
1055                cleanly_resolved = !try_merge(id, path, &cur, &result);
1056                free(result.ptr);
1057                free(cur.ptr);
1058                if (cleanly_resolved)
1059                        break;
1060        }
1061
1062        if (id->collection->status_nr <= id->variant) {
1063                error("no remembered resolution for '%s'", path);
1064                goto fail_exit;
1065        }
1066
1067        filename = rerere_path(id, "postimage");
1068        if (unlink(filename)) {
1069                if (errno == ENOENT)
1070                        error("no remembered resolution for %s", path);
1071                else
1072                        error_errno("cannot unlink %s", filename);
1073                goto fail_exit;
1074        }
1075
1076        /*
1077         * Update the preimage so that the user can resolve the
1078         * conflict in the working tree, run us again to record
1079         * the postimage.
1080         */
1081        handle_cache(path, sha1, rerere_path(id, "preimage"));
1082        fprintf(stderr, "Updated preimage for '%s'\n", path);
1083
1084        /*
1085         * And remember that we can record resolution for this
1086         * conflict when the user is done.
1087         */
1088        item = string_list_insert(rr, path);
1089        free_rerere_id(item);
1090        item->util = id;
1091        fprintf(stderr, "Forgot resolution for %s\n", path);
1092        return 0;
1093
1094fail_exit:
1095        free(id);
1096        return -1;
1097}
1098
1099int rerere_forget(struct pathspec *pathspec)
1100{
1101        int i, fd;
1102        struct string_list conflict = STRING_LIST_INIT_DUP;
1103        struct string_list merge_rr = STRING_LIST_INIT_DUP;
1104
1105        if (read_cache() < 0)
1106                return error("Could not read index");
1107
1108        fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
1109        if (fd < 0)
1110                return 0;
1111
1112        /*
1113         * The paths may have been resolved (incorrectly);
1114         * recover the original conflicted state and then
1115         * find the conflicted paths.
1116         */
1117        unmerge_cache(pathspec);
1118        find_conflict(&conflict);
1119        for (i = 0; i < conflict.nr; i++) {
1120                struct string_list_item *it = &conflict.items[i];
1121                if (!match_pathspec(pathspec, it->string,
1122                                    strlen(it->string), 0, NULL, 0))
1123                        continue;
1124                rerere_forget_one_path(it->string, &merge_rr);
1125        }
1126        return write_rr(&merge_rr, fd);
1127}
1128
1129/*
1130 * Garbage collection support
1131 */
1132
1133static timestamp_t rerere_created_at(struct rerere_id *id)
1134{
1135        struct stat st;
1136
1137        return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1138}
1139
1140static timestamp_t rerere_last_used_at(struct rerere_id *id)
1141{
1142        struct stat st;
1143
1144        return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1145}
1146
1147/*
1148 * Remove the recorded resolution for a given conflict ID
1149 */
1150static void unlink_rr_item(struct rerere_id *id)
1151{
1152        unlink_or_warn(rerere_path(id, "thisimage"));
1153        remove_variant(id);
1154        id->collection->status[id->variant] = 0;
1155}
1156
1157static void prune_one(struct rerere_id *id,
1158                      timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
1159{
1160        timestamp_t then;
1161        timestamp_t cutoff;
1162
1163        then = rerere_last_used_at(id);
1164        if (then)
1165                cutoff = cutoff_resolve;
1166        else {
1167                then = rerere_created_at(id);
1168                if (!then)
1169                        return;
1170                cutoff = cutoff_noresolve;
1171        }
1172        if (then < cutoff)
1173                unlink_rr_item(id);
1174}
1175
1176void rerere_gc(struct string_list *rr)
1177{
1178        struct string_list to_remove = STRING_LIST_INIT_DUP;
1179        DIR *dir;
1180        struct dirent *e;
1181        int i;
1182        timestamp_t now = time(NULL);
1183        timestamp_t cutoff_noresolve = now - 15 * 86400;
1184        timestamp_t cutoff_resolve = now - 60 * 86400;
1185
1186        if (setup_rerere(rr, 0) < 0)
1187                return;
1188
1189        git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
1190        git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
1191        git_config(git_default_config, NULL);
1192        dir = opendir(git_path("rr-cache"));
1193        if (!dir)
1194                die_errno("unable to open rr-cache directory");
1195        /* Collect stale conflict IDs ... */
1196        while ((e = readdir(dir))) {
1197                struct rerere_dir *rr_dir;
1198                struct rerere_id id;
1199                int now_empty;
1200
1201                if (is_dot_or_dotdot(e->d_name))
1202                        continue;
1203                rr_dir = find_rerere_dir(e->d_name);
1204                if (!rr_dir)
1205                        continue; /* or should we remove e->d_name? */
1206
1207                now_empty = 1;
1208                for (id.variant = 0, id.collection = rr_dir;
1209                     id.variant < id.collection->status_nr;
1210                     id.variant++) {
1211                        prune_one(&id, cutoff_resolve, cutoff_noresolve);
1212                        if (id.collection->status[id.variant])
1213                                now_empty = 0;
1214                }
1215                if (now_empty)
1216                        string_list_append(&to_remove, e->d_name);
1217        }
1218        closedir(dir);
1219
1220        /* ... and then remove the empty directories */
1221        for (i = 0; i < to_remove.nr; i++)
1222                rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
1223        string_list_clear(&to_remove, 0);
1224        rollback_lock_file(&write_lock);
1225}
1226
1227/*
1228 * During a conflict resolution, after "rerere" recorded the
1229 * preimages, abandon them if the user did not resolve them or
1230 * record their resolutions.  And drop $GIT_DIR/MERGE_RR.
1231 *
1232 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1233 */
1234void rerere_clear(struct string_list *merge_rr)
1235{
1236        int i;
1237
1238        if (setup_rerere(merge_rr, 0) < 0)
1239                return;
1240
1241        for (i = 0; i < merge_rr->nr; i++) {
1242                struct rerere_id *id = merge_rr->items[i].util;
1243                if (!has_rerere_resolution(id)) {
1244                        unlink_rr_item(id);
1245                        rmdir(rerere_path(id, NULL));
1246                }
1247        }
1248        unlink_or_warn(git_path_merge_rr());
1249        rollback_lock_file(&write_lock);
1250}