353227cdaf9214d48c0d656698c7e2a94847d843
   1#include "cache.h"
   2#include "lockfile.h"
   3#include "string-list.h"
   4#include "rerere.h"
   5#include "xdiff-interface.h"
   6#include "dir.h"
   7#include "resolve-undo.h"
   8#include "ll-merge.h"
   9#include "attr.h"
  10#include "pathspec.h"
  11#include "sha1-lookup.h"
  12
  13#define RESOLVED 0
  14#define PUNTED 1
  15#define THREE_STAGED 2
  16void *RERERE_RESOLVED = &RERERE_RESOLVED;
  17
  18/* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
  19static int rerere_enabled = -1;
  20
  21/* automatically update cleanly resolved paths to the index */
  22static int rerere_autoupdate;
  23
  24static char *merge_rr_path;
  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(rerere_dir);
  45        rerere_dir_nr = rerere_dir_alloc = 0;
  46        rerere_dir = NULL;
  47}
  48
  49static void free_rerere_id(struct string_list_item *item)
  50{
  51        free(item->util);
  52}
  53
  54static const char *rerere_id_hex(const struct rerere_id *id)
  55{
  56        return sha1_to_hex(id->collection->sha1);
  57}
  58
  59static void fit_variant(struct rerere_dir *rr_dir, int variant)
  60{
  61        variant++;
  62        ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
  63        if (rr_dir->status_nr < variant) {
  64                memset(rr_dir->status + rr_dir->status_nr,
  65                       '\0', variant - rr_dir->status_nr);
  66                rr_dir->status_nr = variant;
  67        }
  68}
  69
  70static void assign_variant(struct rerere_id *id)
  71{
  72        int variant;
  73        struct rerere_dir *rr_dir = id->collection;
  74
  75        variant = id->variant;
  76        if (variant < 0) {
  77                for (variant = 0; variant < rr_dir->status_nr; variant++)
  78                        if (!rr_dir->status[variant])
  79                                break;
  80        }
  81        fit_variant(rr_dir, variant);
  82        id->variant = variant;
  83}
  84
  85const char *rerere_path(const struct rerere_id *id, const char *file)
  86{
  87        if (!file)
  88                return git_path("rr-cache/%s", rerere_id_hex(id));
  89
  90        if (id->variant <= 0)
  91                return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
  92
  93        return git_path("rr-cache/%s/%s.%d",
  94                        rerere_id_hex(id), file, id->variant);
  95}
  96
  97static int is_rr_file(const char *name, const char *filename, int *variant)
  98{
  99        const char *suffix;
 100        char *ep;
 101
 102        if (!strcmp(name, filename)) {
 103                *variant = 0;
 104                return 1;
 105        }
 106        if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
 107                return 0;
 108
 109        errno = 0;
 110        *variant = strtol(suffix + 1, &ep, 10);
 111        if (errno || *ep)
 112                return 0;
 113        return 1;
 114}
 115
 116static void scan_rerere_dir(struct rerere_dir *rr_dir)
 117{
 118        struct dirent *de;
 119        DIR *dir = opendir(git_path("rr-cache/%s", sha1_to_hex(rr_dir->sha1)));
 120
 121        if (!dir)
 122                return;
 123        while ((de = readdir(dir)) != NULL) {
 124                int variant;
 125
 126                if (is_rr_file(de->d_name, "postimage", &variant)) {
 127                        fit_variant(rr_dir, variant);
 128                        rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
 129                } else if (is_rr_file(de->d_name, "preimage", &variant)) {
 130                        fit_variant(rr_dir, variant);
 131                        rr_dir->status[variant] |= RR_HAS_PREIMAGE;
 132                }
 133        }
 134        closedir(dir);
 135}
 136
 137static const unsigned char *rerere_dir_sha1(size_t i, void *table)
 138{
 139        struct rerere_dir **rr_dir = table;
 140        return rr_dir[i]->sha1;
 141}
 142
 143static struct rerere_dir *find_rerere_dir(const char *hex)
 144{
 145        unsigned char sha1[20];
 146        struct rerere_dir *rr_dir;
 147        int pos;
 148
 149        if (get_sha1_hex(hex, sha1))
 150                return NULL; /* BUG */
 151        pos = sha1_pos(sha1, rerere_dir, rerere_dir_nr, rerere_dir_sha1);
 152        if (pos < 0) {
 153                rr_dir = xmalloc(sizeof(*rr_dir));
 154                hashcpy(rr_dir->sha1, sha1);
 155                rr_dir->status = NULL;
 156                rr_dir->status_nr = 0;
 157                rr_dir->status_alloc = 0;
 158                pos = -1 - pos;
 159
 160                /* Make sure the array is big enough ... */
 161                ALLOC_GROW(rerere_dir, rerere_dir_nr + 1, rerere_dir_alloc);
 162                /* ... and add it in. */
 163                rerere_dir_nr++;
 164                memmove(rerere_dir + pos + 1, rerere_dir + pos,
 165                        (rerere_dir_nr - pos - 1) * sizeof(*rerere_dir));
 166                rerere_dir[pos] = rr_dir;
 167                scan_rerere_dir(rr_dir);
 168        }
 169        return rerere_dir[pos];
 170}
 171
 172static int has_rerere_resolution(const struct rerere_id *id)
 173{
 174        const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
 175        int variant = id->variant;
 176
 177        if (variant < 0)
 178                return 0;
 179        return ((id->collection->status[variant] & both) == both);
 180}
 181
 182static struct rerere_id *new_rerere_id_hex(char *hex)
 183{
 184        struct rerere_id *id = xmalloc(sizeof(*id));
 185        id->collection = find_rerere_dir(hex);
 186        id->variant = -1; /* not known yet */
 187        return id;
 188}
 189
 190static struct rerere_id *new_rerere_id(unsigned char *sha1)
 191{
 192        return new_rerere_id_hex(sha1_to_hex(sha1));
 193}
 194
 195/*
 196 * $GIT_DIR/MERGE_RR file is a collection of records, each of which is
 197 * "conflict ID", a HT and pathname, terminated with a NUL, and is
 198 * used to keep track of the set of paths that "rerere" may need to
 199 * work on (i.e. what is left by the previous invocation of "git
 200 * rerere" during the current conflict resolution session).
 201 */
 202static void read_rr(struct string_list *rr)
 203{
 204        struct strbuf buf = STRBUF_INIT;
 205        FILE *in = fopen(merge_rr_path, "r");
 206
 207        if (!in)
 208                return;
 209        while (!strbuf_getwholeline(&buf, in, '\0')) {
 210                char *path;
 211                unsigned char sha1[20];
 212                struct rerere_id *id;
 213                int variant;
 214
 215                /* There has to be the hash, tab, path and then NUL */
 216                if (buf.len < 42 || get_sha1_hex(buf.buf, sha1))
 217                        die("corrupt MERGE_RR");
 218
 219                if (buf.buf[40] != '.') {
 220                        variant = 0;
 221                        path = buf.buf + 40;
 222                } else {
 223                        errno = 0;
 224                        variant = strtol(buf.buf + 41, &path, 10);
 225                        if (errno)
 226                                die("corrupt MERGE_RR");
 227                }
 228                if (*(path++) != '\t')
 229                        die("corrupt MERGE_RR");
 230                buf.buf[40] = '\0';
 231                id = new_rerere_id_hex(buf.buf);
 232                id->variant = variant;
 233                string_list_insert(rr, path)->util = id;
 234        }
 235        strbuf_release(&buf);
 236        fclose(in);
 237}
 238
 239static struct lock_file write_lock;
 240
 241static int write_rr(struct string_list *rr, int out_fd)
 242{
 243        int i;
 244        for (i = 0; i < rr->nr; i++) {
 245                struct strbuf buf = STRBUF_INIT;
 246                struct rerere_id *id;
 247
 248                assert(rr->items[i].util != RERERE_RESOLVED);
 249
 250                id = rr->items[i].util;
 251                if (!id)
 252                        continue;
 253                assert(id->variant >= 0);
 254                if (0 < id->variant)
 255                        strbuf_addf(&buf, "%s.%d\t%s%c",
 256                                    rerere_id_hex(id), id->variant,
 257                                    rr->items[i].string, 0);
 258                else
 259                        strbuf_addf(&buf, "%s\t%s%c",
 260                                    rerere_id_hex(id),
 261                                    rr->items[i].string, 0);
 262
 263                if (write_in_full(out_fd, buf.buf, buf.len) != buf.len)
 264                        die("unable to write rerere record");
 265
 266                strbuf_release(&buf);
 267        }
 268        if (commit_lock_file(&write_lock) != 0)
 269                die("unable to write rerere record");
 270        return 0;
 271}
 272
 273/*
 274 * "rerere" interacts with conflicted file contents using this I/O
 275 * abstraction.  It reads a conflicted contents from one place via
 276 * "getline()" method, and optionally can write it out after
 277 * normalizing the conflicted hunks to the "output".  Subclasses of
 278 * rerere_io embed this structure at the beginning of their own
 279 * rerere_io object.
 280 */
 281struct rerere_io {
 282        int (*getline)(struct strbuf *, struct rerere_io *);
 283        FILE *output;
 284        int wrerror;
 285        /* some more stuff */
 286};
 287
 288static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
 289{
 290        if (!count || *err)
 291                return;
 292        if (fwrite(p, count, 1, fp) != 1)
 293                *err = errno;
 294}
 295
 296static inline void ferr_puts(const char *s, FILE *fp, int *err)
 297{
 298        ferr_write(s, strlen(s), fp, err);
 299}
 300
 301static void rerere_io_putstr(const char *str, struct rerere_io *io)
 302{
 303        if (io->output)
 304                ferr_puts(str, io->output, &io->wrerror);
 305}
 306
 307/*
 308 * Write a conflict marker to io->output (if defined).
 309 */
 310static void rerere_io_putconflict(int ch, int size, struct rerere_io *io)
 311{
 312        char buf[64];
 313
 314        while (size) {
 315                if (size <= sizeof(buf) - 2) {
 316                        memset(buf, ch, size);
 317                        buf[size] = '\n';
 318                        buf[size + 1] = '\0';
 319                        size = 0;
 320                } else {
 321                        int sz = sizeof(buf) - 1;
 322
 323                        /*
 324                         * Make sure we will not write everything out
 325                         * in this round by leaving at least 1 byte
 326                         * for the next round, giving the next round
 327                         * a chance to add the terminating LF.  Yuck.
 328                         */
 329                        if (size <= sz)
 330                                sz -= (sz - size) + 1;
 331                        memset(buf, ch, sz);
 332                        buf[sz] = '\0';
 333                        size -= sz;
 334                }
 335                rerere_io_putstr(buf, io);
 336        }
 337}
 338
 339static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
 340{
 341        if (io->output)
 342                ferr_write(mem, sz, io->output, &io->wrerror);
 343}
 344
 345/*
 346 * Subclass of rerere_io that reads from an on-disk file
 347 */
 348struct rerere_io_file {
 349        struct rerere_io io;
 350        FILE *input;
 351};
 352
 353/*
 354 * ... and its getline() method implementation
 355 */
 356static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
 357{
 358        struct rerere_io_file *io = (struct rerere_io_file *)io_;
 359        return strbuf_getwholeline(sb, io->input, '\n');
 360}
 361
 362/*
 363 * Require the exact number of conflict marker letters, no more, no
 364 * less, followed by SP or any whitespace
 365 * (including LF).
 366 */
 367static int is_cmarker(char *buf, int marker_char, int marker_size)
 368{
 369        int want_sp;
 370
 371        /*
 372         * The beginning of our version and the end of their version
 373         * always are labeled like "<<<<< ours" or ">>>>> theirs",
 374         * hence we set want_sp for them.  Note that the version from
 375         * the common ancestor in diff3-style output is not always
 376         * labelled (e.g. "||||| common" is often seen but "|||||"
 377         * alone is also valid), so we do not set want_sp.
 378         */
 379        want_sp = (marker_char == '<') || (marker_char == '>');
 380
 381        while (marker_size--)
 382                if (*buf++ != marker_char)
 383                        return 0;
 384        if (want_sp && *buf != ' ')
 385                return 0;
 386        return isspace(*buf);
 387}
 388
 389/*
 390 * Read contents a file with conflicts, normalize the conflicts
 391 * by (1) discarding the common ancestor version in diff3-style,
 392 * (2) reordering our side and their side so that whichever sorts
 393 * alphabetically earlier comes before the other one, while
 394 * computing the "conflict ID", which is just an SHA-1 hash of
 395 * one side of the conflict, NUL, the other side of the conflict,
 396 * and NUL concatenated together.
 397 *
 398 * Return the number of conflict hunks found.
 399 *
 400 * NEEDSWORK: the logic and theory of operation behind this conflict
 401 * normalization may deserve to be documented somewhere, perhaps in
 402 * Documentation/technical/rerere.txt.
 403 */
 404static int handle_path(unsigned char *sha1, struct rerere_io *io, int marker_size)
 405{
 406        git_SHA_CTX ctx;
 407        int hunk_no = 0;
 408        enum {
 409                RR_CONTEXT = 0, RR_SIDE_1, RR_SIDE_2, RR_ORIGINAL
 410        } hunk = RR_CONTEXT;
 411        struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
 412        struct strbuf buf = STRBUF_INIT;
 413
 414        if (sha1)
 415                git_SHA1_Init(&ctx);
 416
 417        while (!io->getline(&buf, io)) {
 418                if (is_cmarker(buf.buf, '<', marker_size)) {
 419                        if (hunk != RR_CONTEXT)
 420                                goto bad;
 421                        hunk = RR_SIDE_1;
 422                } else if (is_cmarker(buf.buf, '|', marker_size)) {
 423                        if (hunk != RR_SIDE_1)
 424                                goto bad;
 425                        hunk = RR_ORIGINAL;
 426                } else if (is_cmarker(buf.buf, '=', marker_size)) {
 427                        if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
 428                                goto bad;
 429                        hunk = RR_SIDE_2;
 430                } else if (is_cmarker(buf.buf, '>', marker_size)) {
 431                        if (hunk != RR_SIDE_2)
 432                                goto bad;
 433                        if (strbuf_cmp(&one, &two) > 0)
 434                                strbuf_swap(&one, &two);
 435                        hunk_no++;
 436                        hunk = RR_CONTEXT;
 437                        rerere_io_putconflict('<', marker_size, io);
 438                        rerere_io_putmem(one.buf, one.len, io);
 439                        rerere_io_putconflict('=', marker_size, io);
 440                        rerere_io_putmem(two.buf, two.len, io);
 441                        rerere_io_putconflict('>', marker_size, io);
 442                        if (sha1) {
 443                                git_SHA1_Update(&ctx, one.buf ? one.buf : "",
 444                                            one.len + 1);
 445                                git_SHA1_Update(&ctx, two.buf ? two.buf : "",
 446                                            two.len + 1);
 447                        }
 448                        strbuf_reset(&one);
 449                        strbuf_reset(&two);
 450                } else if (hunk == RR_SIDE_1)
 451                        strbuf_addbuf(&one, &buf);
 452                else if (hunk == RR_ORIGINAL)
 453                        ; /* discard */
 454                else if (hunk == RR_SIDE_2)
 455                        strbuf_addbuf(&two, &buf);
 456                else
 457                        rerere_io_putstr(buf.buf, io);
 458                continue;
 459        bad:
 460                hunk = 99; /* force error exit */
 461                break;
 462        }
 463        strbuf_release(&one);
 464        strbuf_release(&two);
 465        strbuf_release(&buf);
 466
 467        if (sha1)
 468                git_SHA1_Final(sha1, &ctx);
 469        if (hunk != RR_CONTEXT)
 470                return -1;
 471        return hunk_no;
 472}
 473
 474/*
 475 * Scan the path for conflicts, do the "handle_path()" thing above, and
 476 * return the number of conflict hunks found.
 477 */
 478static int handle_file(const char *path, unsigned char *sha1, const char *output)
 479{
 480        int hunk_no = 0;
 481        struct rerere_io_file io;
 482        int marker_size = ll_merge_marker_size(path);
 483
 484        memset(&io, 0, sizeof(io));
 485        io.io.getline = rerere_file_getline;
 486        io.input = fopen(path, "r");
 487        io.io.wrerror = 0;
 488        if (!io.input)
 489                return error("Could not open %s", path);
 490
 491        if (output) {
 492                io.io.output = fopen(output, "w");
 493                if (!io.io.output) {
 494                        fclose(io.input);
 495                        return error("Could not write %s", output);
 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("Failed to flush %s: %s",
 507                                      path, strerror(errno));
 508
 509        if (hunk_no < 0) {
 510                if (output)
 511                        unlink_or_warn(output);
 512                return error("Could not parse conflict hunks in %s", path);
 513        }
 514        if (io.io.wrerror)
 515                return -1;
 516        return hunk_no;
 517}
 518
 519/*
 520 * Subclass of rerere_io that reads from an in-core buffer that is a
 521 * strbuf
 522 */
 523struct rerere_io_mem {
 524        struct rerere_io io;
 525        struct strbuf input;
 526};
 527
 528/*
 529 * ... and its getline() method implementation
 530 */
 531static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
 532{
 533        struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
 534        char *ep;
 535        size_t len;
 536
 537        strbuf_release(sb);
 538        if (!io->input.len)
 539                return -1;
 540        ep = memchr(io->input.buf, '\n', io->input.len);
 541        if (!ep)
 542                ep = io->input.buf + io->input.len;
 543        else if (*ep == '\n')
 544                ep++;
 545        len = ep - io->input.buf;
 546        strbuf_add(sb, io->input.buf, len);
 547        strbuf_remove(&io->input, 0, len);
 548        return 0;
 549}
 550
 551static int handle_cache(const char *path, unsigned char *sha1, const char *output)
 552{
 553        mmfile_t mmfile[3] = {{NULL}};
 554        mmbuffer_t result = {NULL, 0};
 555        const struct cache_entry *ce;
 556        int pos, len, i, hunk_no;
 557        struct rerere_io_mem io;
 558        int marker_size = ll_merge_marker_size(path);
 559
 560        /*
 561         * Reproduce the conflicted merge in-core
 562         */
 563        len = strlen(path);
 564        pos = cache_name_pos(path, len);
 565        if (0 <= pos)
 566                return -1;
 567        pos = -pos - 1;
 568
 569        while (pos < active_nr) {
 570                enum object_type type;
 571                unsigned long size;
 572
 573                ce = active_cache[pos++];
 574                if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
 575                        break;
 576                i = ce_stage(ce) - 1;
 577                if (!mmfile[i].ptr) {
 578                        mmfile[i].ptr = read_sha1_file(ce->sha1, &type, &size);
 579                        mmfile[i].size = size;
 580                }
 581        }
 582        for (i = 0; i < 3; i++)
 583                if (!mmfile[i].ptr && !mmfile[i].size)
 584                        mmfile[i].ptr = xstrdup("");
 585
 586        /*
 587         * NEEDSWORK: handle conflicts from merges with
 588         * merge.renormalize set, too
 589         */
 590        ll_merge(&result, path, &mmfile[0], NULL,
 591                 &mmfile[1], "ours",
 592                 &mmfile[2], "theirs", NULL);
 593        for (i = 0; i < 3; i++)
 594                free(mmfile[i].ptr);
 595
 596        memset(&io, 0, sizeof(io));
 597        io.io.getline = rerere_mem_getline;
 598        if (output)
 599                io.io.output = fopen(output, "w");
 600        else
 601                io.io.output = NULL;
 602        strbuf_init(&io.input, 0);
 603        strbuf_attach(&io.input, result.ptr, result.size, result.size);
 604
 605        /*
 606         * Grab the conflict ID and optionally write the original
 607         * contents with conflict markers out.
 608         */
 609        hunk_no = handle_path(sha1, (struct rerere_io *)&io, marker_size);
 610        strbuf_release(&io.input);
 611        if (io.io.output)
 612                fclose(io.io.output);
 613        return hunk_no;
 614}
 615
 616/*
 617 * Look at a cache entry at "i" and see if it is not conflicting,
 618 * conflicting and we are willing to handle, or conflicting and
 619 * we are unable to handle, and return the determination in *type.
 620 * Return the cache index to be looked at next, by skipping the
 621 * stages we have already looked at in this invocation of this
 622 * function.
 623 */
 624static int check_one_conflict(int i, int *type)
 625{
 626        const struct cache_entry *e = active_cache[i];
 627
 628        if (!ce_stage(e)) {
 629                *type = RESOLVED;
 630                return i + 1;
 631        }
 632
 633        *type = PUNTED;
 634        while (ce_stage(active_cache[i]) == 1)
 635                i++;
 636
 637        /* Only handle regular files with both stages #2 and #3 */
 638        if (i + 1 < active_nr) {
 639                const struct cache_entry *e2 = active_cache[i];
 640                const struct cache_entry *e3 = active_cache[i + 1];
 641                if (ce_stage(e2) == 2 &&
 642                    ce_stage(e3) == 3 &&
 643                    ce_same_name(e, e3) &&
 644                    S_ISREG(e2->ce_mode) &&
 645                    S_ISREG(e3->ce_mode))
 646                        *type = THREE_STAGED;
 647        }
 648
 649        /* Skip the entries with the same name */
 650        while (i < active_nr && ce_same_name(e, active_cache[i]))
 651                i++;
 652        return i;
 653}
 654
 655/*
 656 * Scan the index and find paths that have conflicts that rerere can
 657 * handle, i.e. the ones that has both stages #2 and #3.
 658 *
 659 * NEEDSWORK: we do not record or replay a previous "resolve by
 660 * deletion" for a delete-modify conflict, as that is inherently risky
 661 * without knowing what modification is being discarded.  The only
 662 * safe case, i.e. both side doing the deletion and modification that
 663 * are identical to the previous round, might want to be handled,
 664 * though.
 665 */
 666static int find_conflict(struct string_list *conflict)
 667{
 668        int i;
 669        if (read_cache() < 0)
 670                return error("Could not read index");
 671
 672        for (i = 0; i < active_nr;) {
 673                int conflict_type;
 674                const struct cache_entry *e = active_cache[i];
 675                i = check_one_conflict(i, &conflict_type);
 676                if (conflict_type == THREE_STAGED)
 677                        string_list_insert(conflict, (const char *)e->name);
 678        }
 679        return 0;
 680}
 681
 682/*
 683 * The merge_rr list is meant to hold outstanding conflicted paths
 684 * that rerere could handle.  Abuse the list by adding other types of
 685 * entries to allow the caller to show "rerere remaining".
 686 *
 687 * - Conflicted paths that rerere does not handle are added
 688 * - Conflicted paths that have been resolved are marked as such
 689 *   by storing RERERE_RESOLVED to .util field (where conflict ID
 690 *   is expected to be stored).
 691 *
 692 * Do *not* write MERGE_RR file out after calling this function.
 693 *
 694 * NEEDSWORK: we may want to fix the caller that implements "rerere
 695 * remaining" to do this without abusing merge_rr.
 696 */
 697int rerere_remaining(struct string_list *merge_rr)
 698{
 699        int i;
 700        if (read_cache() < 0)
 701                return error("Could not read index");
 702
 703        for (i = 0; i < active_nr;) {
 704                int conflict_type;
 705                const struct cache_entry *e = active_cache[i];
 706                i = check_one_conflict(i, &conflict_type);
 707                if (conflict_type == PUNTED)
 708                        string_list_insert(merge_rr, (const char *)e->name);
 709                else if (conflict_type == RESOLVED) {
 710                        struct string_list_item *it;
 711                        it = string_list_lookup(merge_rr, (const char *)e->name);
 712                        if (it != NULL) {
 713                                free_rerere_id(it);
 714                                it->util = RERERE_RESOLVED;
 715                        }
 716                }
 717        }
 718        return 0;
 719}
 720
 721/*
 722 * Find the conflict identified by "id"; the change between its
 723 * "preimage" (i.e. a previous contents with conflict markers) and its
 724 * "postimage" (i.e. the corresponding contents with conflicts
 725 * resolved) may apply cleanly to the contents stored in "path", i.e.
 726 * the conflict this time around.
 727 *
 728 * Returns 0 for successful replay of recorded resolution, or non-zero
 729 * for failure.
 730 */
 731static int merge(const struct rerere_id *id, const char *path)
 732{
 733        FILE *f;
 734        int ret;
 735        mmfile_t cur = {NULL, 0}, base = {NULL, 0}, other = {NULL, 0};
 736        mmbuffer_t result = {NULL, 0};
 737
 738        /*
 739         * Normalize the conflicts in path and write it out to
 740         * "thisimage" temporary file.
 741         */
 742        if (handle_file(path, NULL, rerere_path(id, "thisimage")) < 0) {
 743                ret = 1;
 744                goto out;
 745        }
 746
 747        if (read_mmfile(&cur, rerere_path(id, "thisimage")) ||
 748            read_mmfile(&base, rerere_path(id, "preimage")) ||
 749            read_mmfile(&other, rerere_path(id, "postimage"))) {
 750                ret = 1;
 751                goto out;
 752        }
 753
 754        /*
 755         * A three-way merge. Note that this honors user-customizable
 756         * low-level merge driver settings.
 757         */
 758        ret = ll_merge(&result, path, &base, NULL, &cur, "", &other, "", NULL);
 759        if (ret)
 760                goto out;
 761
 762        /*
 763         * A successful replay of recorded resolution.
 764         * Mark that "postimage" was used to help gc.
 765         */
 766        if (utime(rerere_path(id, "postimage"), NULL) < 0)
 767                warning("failed utime() on %s: %s",
 768                        rerere_path(id, "postimage"),
 769                        strerror(errno));
 770
 771        /* Update "path" with the resolution */
 772        f = fopen(path, "w");
 773        if (!f)
 774                return error("Could not open %s: %s", path,
 775                             strerror(errno));
 776        if (fwrite(result.ptr, result.size, 1, f) != 1)
 777                error("Could not write %s: %s", path, strerror(errno));
 778        if (fclose(f))
 779                return error("Writing %s failed: %s", path,
 780                             strerror(errno));
 781
 782out:
 783        free(cur.ptr);
 784        free(base.ptr);
 785        free(other.ptr);
 786        free(result.ptr);
 787
 788        return ret;
 789}
 790
 791static struct lock_file index_lock;
 792
 793static void update_paths(struct string_list *update)
 794{
 795        int i;
 796
 797        hold_locked_index(&index_lock, 1);
 798
 799        for (i = 0; i < update->nr; i++) {
 800                struct string_list_item *item = &update->items[i];
 801                if (add_file_to_cache(item->string, 0))
 802                        exit(128);
 803                fprintf(stderr, "Staged '%s' using previous resolution.\n",
 804                        item->string);
 805        }
 806
 807        if (active_cache_changed) {
 808                if (write_locked_index(&the_index, &index_lock, COMMIT_LOCK))
 809                        die("Unable to write new index file");
 810        } else
 811                rollback_lock_file(&index_lock);
 812}
 813
 814static void remove_variant(struct rerere_id *id)
 815{
 816        unlink_or_warn(rerere_path(id, "postimage"));
 817        unlink_or_warn(rerere_path(id, "preimage"));
 818        id->collection->status[id->variant] = 0;
 819}
 820
 821/*
 822 * The path indicated by rr_item may still have conflict for which we
 823 * have a recorded resolution, in which case replay it and optionally
 824 * update it.  Or it may have been resolved by the user and we may
 825 * only have the preimage for that conflict, in which case the result
 826 * needs to be recorded as a resolution in a postimage file.
 827 */
 828static void do_rerere_one_path(struct string_list_item *rr_item,
 829                               struct string_list *update)
 830{
 831        const char *path = rr_item->string;
 832        struct rerere_id *id = rr_item->util;
 833        struct rerere_dir *rr_dir = id->collection;
 834        int variant;
 835
 836        variant = id->variant;
 837
 838        /* Has the user resolved it already? */
 839        if (variant >= 0) {
 840                if (!handle_file(path, NULL, NULL)) {
 841                        copy_file(rerere_path(id, "postimage"), path, 0666);
 842                        id->collection->status[variant] |= RR_HAS_POSTIMAGE;
 843                        fprintf(stderr, "Recorded resolution for '%s'.\n", path);
 844                        free_rerere_id(rr_item);
 845                        rr_item->util = NULL;
 846                        return;
 847                }
 848                /*
 849                 * There may be other variants that can cleanly
 850                 * replay.  Try them and update the variant number for
 851                 * this one.
 852                 */
 853        }
 854
 855        /* Does any existing resolution apply cleanly? */
 856        for (variant = 0; variant < rr_dir->status_nr; variant++) {
 857                const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
 858                struct rerere_id vid = *id;
 859
 860                if ((rr_dir->status[variant] & both) != both)
 861                        continue;
 862
 863                vid.variant = variant;
 864                if (merge(&vid, path))
 865                        continue; /* failed to replay */
 866
 867                /*
 868                 * If there already is a different variant that applies
 869                 * cleanly, there is no point maintaining our own variant.
 870                 */
 871                if (0 <= id->variant && id->variant != variant)
 872                        remove_variant(id);
 873
 874                if (rerere_autoupdate)
 875                        string_list_insert(update, path);
 876                else
 877                        fprintf(stderr,
 878                                "Resolved '%s' using previous resolution.\n",
 879                                path);
 880                free_rerere_id(rr_item);
 881                rr_item->util = NULL;
 882                return;
 883        }
 884
 885        /* None of the existing one applies; we need a new variant */
 886        assign_variant(id);
 887
 888        variant = id->variant;
 889        handle_file(path, NULL, rerere_path(id, "preimage"));
 890        if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
 891                const char *path = rerere_path(id, "postimage");
 892                if (unlink(path))
 893                        die_errno("cannot unlink stray '%s'", path);
 894                id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
 895        }
 896        id->collection->status[variant] |= RR_HAS_PREIMAGE;
 897        fprintf(stderr, "Recorded preimage for '%s'\n", path);
 898}
 899
 900static int do_plain_rerere(struct string_list *rr, int fd)
 901{
 902        struct string_list conflict = STRING_LIST_INIT_DUP;
 903        struct string_list update = STRING_LIST_INIT_DUP;
 904        int i;
 905
 906        find_conflict(&conflict);
 907
 908        /*
 909         * MERGE_RR records paths with conflicts immediately after
 910         * merge failed.  Some of the conflicted paths might have been
 911         * hand resolved in the working tree since then, but the
 912         * initial run would catch all and register their preimages.
 913         */
 914        for (i = 0; i < conflict.nr; i++) {
 915                struct rerere_id *id;
 916                unsigned char sha1[20];
 917                const char *path = conflict.items[i].string;
 918                int ret;
 919
 920                if (string_list_has_string(rr, path))
 921                        continue;
 922
 923                /*
 924                 * Ask handle_file() to scan and assign a
 925                 * conflict ID.  No need to write anything out
 926                 * yet.
 927                 */
 928                ret = handle_file(path, sha1, NULL);
 929                if (ret < 1)
 930                        continue;
 931
 932                id = new_rerere_id(sha1);
 933                string_list_insert(rr, path)->util = id;
 934
 935                /* Ensure that the directory exists. */
 936                mkdir_in_gitdir(rerere_path(id, NULL));
 937        }
 938
 939        for (i = 0; i < rr->nr; i++)
 940                do_rerere_one_path(&rr->items[i], &update);
 941
 942        if (update.nr)
 943                update_paths(&update);
 944
 945        return write_rr(rr, fd);
 946}
 947
 948static void git_rerere_config(void)
 949{
 950        git_config_get_bool("rerere.enabled", &rerere_enabled);
 951        git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
 952        git_config(git_default_config, NULL);
 953}
 954
 955static int is_rerere_enabled(void)
 956{
 957        const char *rr_cache;
 958        int rr_cache_exists;
 959
 960        if (!rerere_enabled)
 961                return 0;
 962
 963        rr_cache = git_path("rr-cache");
 964        rr_cache_exists = is_directory(rr_cache);
 965        if (rerere_enabled < 0)
 966                return rr_cache_exists;
 967
 968        if (!rr_cache_exists && mkdir_in_gitdir(rr_cache))
 969                die("Could not create directory %s", rr_cache);
 970        return 1;
 971}
 972
 973int setup_rerere(struct string_list *merge_rr, int flags)
 974{
 975        int fd;
 976
 977        git_rerere_config();
 978        if (!is_rerere_enabled())
 979                return -1;
 980
 981        if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
 982                rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
 983        merge_rr_path = git_pathdup("MERGE_RR");
 984        fd = hold_lock_file_for_update(&write_lock, merge_rr_path,
 985                                       LOCK_DIE_ON_ERROR);
 986        read_rr(merge_rr);
 987        return fd;
 988}
 989
 990/*
 991 * The main entry point that is called internally from codepaths that
 992 * perform mergy operations, possibly leaving conflicted index entries
 993 * and working tree files.
 994 */
 995int rerere(int flags)
 996{
 997        struct string_list merge_rr = STRING_LIST_INIT_DUP;
 998        int fd, status;
 999
1000        fd = setup_rerere(&merge_rr, flags);
1001        if (fd < 0)
1002                return 0;
1003        status = do_plain_rerere(&merge_rr, fd);
1004        free_rerere_dirs();
1005        return status;
1006}
1007
1008static int rerere_forget_one_path(const char *path, struct string_list *rr)
1009{
1010        const char *filename;
1011        struct rerere_id *id;
1012        unsigned char sha1[20];
1013        int ret;
1014        struct string_list_item *item;
1015
1016        /*
1017         * Recreate the original conflict from the stages in the
1018         * index and compute the conflict ID
1019         */
1020        ret = handle_cache(path, sha1, NULL);
1021        if (ret < 1)
1022                return error("Could not parse conflict hunks in '%s'", path);
1023
1024        /* Nuke the recorded resolution for the conflict */
1025        id = new_rerere_id(sha1);
1026        id->variant = 0; /* for now */
1027        filename = rerere_path(id, "postimage");
1028        if (unlink(filename))
1029                return (errno == ENOENT
1030                        ? error("no remembered resolution for %s", path)
1031                        : error("cannot unlink %s: %s", filename, strerror(errno)));
1032
1033        /*
1034         * Update the preimage so that the user can resolve the
1035         * conflict in the working tree, run us again to record
1036         * the postimage.
1037         */
1038        handle_cache(path, sha1, rerere_path(id, "preimage"));
1039        fprintf(stderr, "Updated preimage for '%s'\n", path);
1040
1041        /*
1042         * And remember that we can record resolution for this
1043         * conflict when the user is done.
1044         */
1045        item = string_list_insert(rr, path);
1046        free_rerere_id(item);
1047        item->util = id;
1048        fprintf(stderr, "Forgot resolution for %s\n", path);
1049        return 0;
1050}
1051
1052int rerere_forget(struct pathspec *pathspec)
1053{
1054        int i, fd;
1055        struct string_list conflict = STRING_LIST_INIT_DUP;
1056        struct string_list merge_rr = STRING_LIST_INIT_DUP;
1057
1058        if (read_cache() < 0)
1059                return error("Could not read index");
1060
1061        fd = setup_rerere(&merge_rr, RERERE_NOAUTOUPDATE);
1062
1063        /*
1064         * The paths may have been resolved (incorrectly);
1065         * recover the original conflicted state and then
1066         * find the conflicted paths.
1067         */
1068        unmerge_cache(pathspec);
1069        find_conflict(&conflict);
1070        for (i = 0; i < conflict.nr; i++) {
1071                struct string_list_item *it = &conflict.items[i];
1072                if (!match_pathspec(pathspec, it->string,
1073                                    strlen(it->string), 0, NULL, 0))
1074                        continue;
1075                rerere_forget_one_path(it->string, &merge_rr);
1076        }
1077        return write_rr(&merge_rr, fd);
1078}
1079
1080/*
1081 * Garbage collection support
1082 */
1083
1084/*
1085 * Note that this is not reentrant but is used only one-at-a-time
1086 * so it does not matter right now.
1087 */
1088static struct rerere_id *dirname_to_id(const char *name)
1089{
1090        static struct rerere_id id;
1091        id.collection = find_rerere_dir(name);
1092        return &id;
1093}
1094
1095static time_t rerere_created_at(const char *dir_name)
1096{
1097        struct stat st;
1098        struct rerere_id *id = dirname_to_id(dir_name);
1099
1100        return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
1101}
1102
1103static time_t rerere_last_used_at(const char *dir_name)
1104{
1105        struct stat st;
1106        struct rerere_id *id = dirname_to_id(dir_name);
1107
1108        return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
1109}
1110
1111/*
1112 * Remove the recorded resolution for a given conflict ID
1113 */
1114static void unlink_rr_item(struct rerere_id *id)
1115{
1116        unlink(rerere_path(id, "thisimage"));
1117        unlink(rerere_path(id, "preimage"));
1118        unlink(rerere_path(id, "postimage"));
1119        /*
1120         * NEEDSWORK: what if this rmdir() fails?  Wouldn't we then
1121         * assume that we already have preimage recorded in
1122         * do_plain_rerere()?
1123         */
1124        rmdir(rerere_path(id, NULL));
1125}
1126
1127void rerere_gc(struct string_list *rr)
1128{
1129        struct string_list to_remove = STRING_LIST_INIT_DUP;
1130        DIR *dir;
1131        struct dirent *e;
1132        int i, cutoff;
1133        time_t now = time(NULL), then;
1134        int cutoff_noresolve = 15;
1135        int cutoff_resolve = 60;
1136
1137        git_config_get_int("gc.rerereresolved", &cutoff_resolve);
1138        git_config_get_int("gc.rerereunresolved", &cutoff_noresolve);
1139        git_config(git_default_config, NULL);
1140        dir = opendir(git_path("rr-cache"));
1141        if (!dir)
1142                die_errno("unable to open rr-cache directory");
1143        /* Collect stale conflict IDs ... */
1144        while ((e = readdir(dir))) {
1145                if (is_dot_or_dotdot(e->d_name))
1146                        continue;
1147
1148                then = rerere_last_used_at(e->d_name);
1149                if (then) {
1150                        cutoff = cutoff_resolve;
1151                } else {
1152                        then = rerere_created_at(e->d_name);
1153                        if (!then)
1154                                continue;
1155                        cutoff = cutoff_noresolve;
1156                }
1157                if (then < now - cutoff * 86400)
1158                        string_list_append(&to_remove, e->d_name);
1159        }
1160        closedir(dir);
1161        /* ... and then remove them one-by-one */
1162        for (i = 0; i < to_remove.nr; i++)
1163                unlink_rr_item(dirname_to_id(to_remove.items[i].string));
1164        string_list_clear(&to_remove, 0);
1165}
1166
1167/*
1168 * During a conflict resolution, after "rerere" recorded the
1169 * preimages, abandon them if the user did not resolve them or
1170 * record their resolutions.  And drop $GIT_DIR/MERGE_RR.
1171 *
1172 * NEEDSWORK: shouldn't we be calling this from "reset --hard"?
1173 */
1174void rerere_clear(struct string_list *merge_rr)
1175{
1176        int i;
1177
1178        for (i = 0; i < merge_rr->nr; i++) {
1179                struct rerere_id *id = merge_rr->items[i].util;
1180                if (!has_rerere_resolution(id))
1181                        unlink_rr_item(id);
1182        }
1183        unlink_or_warn(git_path("MERGE_RR"));
1184}