builtin / blame.con commit Merge branch 'maint' (1ba5c53)
   1/*
   2 * Blame
   3 *
   4 * Copyright (c) 2006, Junio C Hamano
   5 */
   6
   7#include "cache.h"
   8#include "builtin.h"
   9#include "blob.h"
  10#include "commit.h"
  11#include "tag.h"
  12#include "tree-walk.h"
  13#include "diff.h"
  14#include "diffcore.h"
  15#include "revision.h"
  16#include "quote.h"
  17#include "xdiff-interface.h"
  18#include "cache-tree.h"
  19#include "string-list.h"
  20#include "mailmap.h"
  21#include "parse-options.h"
  22#include "utf8.h"
  23#include "userdiff.h"
  24
  25static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
  26
  27static const char *blame_opt_usage[] = {
  28        blame_usage,
  29        "",
  30        "[rev-opts] are documented in git-rev-list(1)",
  31        NULL
  32};
  33
  34static int longest_file;
  35static int longest_author;
  36static int max_orig_digits;
  37static int max_digits;
  38static int max_score_digits;
  39static int show_root;
  40static int reverse;
  41static int blank_boundary;
  42static int incremental;
  43static int xdl_opts;
  44
  45static enum date_mode blame_date_mode = DATE_ISO8601;
  46static size_t blame_date_width;
  47
  48static struct string_list mailmap;
  49
  50#ifndef DEBUG
  51#define DEBUG 0
  52#endif
  53
  54/* stats */
  55static int num_read_blob;
  56static int num_get_patch;
  57static int num_commits;
  58
  59#define PICKAXE_BLAME_MOVE              01
  60#define PICKAXE_BLAME_COPY              02
  61#define PICKAXE_BLAME_COPY_HARDER       04
  62#define PICKAXE_BLAME_COPY_HARDEST      010
  63
  64/*
  65 * blame for a blame_entry with score lower than these thresholds
  66 * is not passed to the parent using move/copy logic.
  67 */
  68static unsigned blame_move_score;
  69static unsigned blame_copy_score;
  70#define BLAME_DEFAULT_MOVE_SCORE        20
  71#define BLAME_DEFAULT_COPY_SCORE        40
  72
  73/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
  74#define METAINFO_SHOWN          (1u<<12)
  75#define MORE_THAN_ONE_PATH      (1u<<13)
  76
  77/*
  78 * One blob in a commit that is being suspected
  79 */
  80struct origin {
  81        int refcnt;
  82        struct origin *previous;
  83        struct commit *commit;
  84        mmfile_t file;
  85        unsigned char blob_sha1[20];
  86        char path[FLEX_ARRAY];
  87};
  88
  89/*
  90 * Prepare diff_filespec and convert it using diff textconv API
  91 * if the textconv driver exists.
  92 * Return 1 if the conversion succeeds, 0 otherwise.
  93 */
  94int textconv_object(const char *path,
  95                    const unsigned char *sha1,
  96                    char **buf,
  97                    unsigned long *buf_size)
  98{
  99        struct diff_filespec *df;
 100        struct userdiff_driver *textconv;
 101
 102        df = alloc_filespec(path);
 103        fill_filespec(df, sha1, S_IFREG | 0664);
 104        textconv = get_textconv(df);
 105        if (!textconv) {
 106                free_filespec(df);
 107                return 0;
 108        }
 109
 110        *buf_size = fill_textconv(textconv, df, buf);
 111        free_filespec(df);
 112        return 1;
 113}
 114
 115/*
 116 * Given an origin, prepare mmfile_t structure to be used by the
 117 * diff machinery
 118 */
 119static void fill_origin_blob(struct diff_options *opt,
 120                             struct origin *o, mmfile_t *file)
 121{
 122        if (!o->file.ptr) {
 123                enum object_type type;
 124                unsigned long file_size;
 125
 126                num_read_blob++;
 127                if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
 128                    textconv_object(o->path, o->blob_sha1, &file->ptr, &file_size))
 129                        ;
 130                else
 131                        file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
 132                file->size = file_size;
 133
 134                if (!file->ptr)
 135                        die("Cannot read blob %s for path %s",
 136                            sha1_to_hex(o->blob_sha1),
 137                            o->path);
 138                o->file = *file;
 139        }
 140        else
 141                *file = o->file;
 142}
 143
 144/*
 145 * Origin is refcounted and usually we keep the blob contents to be
 146 * reused.
 147 */
 148static inline struct origin *origin_incref(struct origin *o)
 149{
 150        if (o)
 151                o->refcnt++;
 152        return o;
 153}
 154
 155static void origin_decref(struct origin *o)
 156{
 157        if (o && --o->refcnt <= 0) {
 158                if (o->previous)
 159                        origin_decref(o->previous);
 160                free(o->file.ptr);
 161                free(o);
 162        }
 163}
 164
 165static void drop_origin_blob(struct origin *o)
 166{
 167        if (o->file.ptr) {
 168                free(o->file.ptr);
 169                o->file.ptr = NULL;
 170        }
 171}
 172
 173/*
 174 * Each group of lines is described by a blame_entry; it can be split
 175 * as we pass blame to the parents.  They form a linked list in the
 176 * scoreboard structure, sorted by the target line number.
 177 */
 178struct blame_entry {
 179        struct blame_entry *prev;
 180        struct blame_entry *next;
 181
 182        /* the first line of this group in the final image;
 183         * internally all line numbers are 0 based.
 184         */
 185        int lno;
 186
 187        /* how many lines this group has */
 188        int num_lines;
 189
 190        /* the commit that introduced this group into the final image */
 191        struct origin *suspect;
 192
 193        /* true if the suspect is truly guilty; false while we have not
 194         * checked if the group came from one of its parents.
 195         */
 196        char guilty;
 197
 198        /* true if the entry has been scanned for copies in the current parent
 199         */
 200        char scanned;
 201
 202        /* the line number of the first line of this group in the
 203         * suspect's file; internally all line numbers are 0 based.
 204         */
 205        int s_lno;
 206
 207        /* how significant this entry is -- cached to avoid
 208         * scanning the lines over and over.
 209         */
 210        unsigned score;
 211};
 212
 213/*
 214 * The current state of the blame assignment.
 215 */
 216struct scoreboard {
 217        /* the final commit (i.e. where we started digging from) */
 218        struct commit *final;
 219        struct rev_info *revs;
 220        const char *path;
 221
 222        /*
 223         * The contents in the final image.
 224         * Used by many functions to obtain contents of the nth line,
 225         * indexed with scoreboard.lineno[blame_entry.lno].
 226         */
 227        const char *final_buf;
 228        unsigned long final_buf_size;
 229
 230        /* linked list of blames */
 231        struct blame_entry *ent;
 232
 233        /* look-up a line in the final buffer */
 234        int num_lines;
 235        int *lineno;
 236};
 237
 238static inline int same_suspect(struct origin *a, struct origin *b)
 239{
 240        if (a == b)
 241                return 1;
 242        if (a->commit != b->commit)
 243                return 0;
 244        return !strcmp(a->path, b->path);
 245}
 246
 247static void sanity_check_refcnt(struct scoreboard *);
 248
 249/*
 250 * If two blame entries that are next to each other came from
 251 * contiguous lines in the same origin (i.e. <commit, path> pair),
 252 * merge them together.
 253 */
 254static void coalesce(struct scoreboard *sb)
 255{
 256        struct blame_entry *ent, *next;
 257
 258        for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 259                if (same_suspect(ent->suspect, next->suspect) &&
 260                    ent->guilty == next->guilty &&
 261                    ent->s_lno + ent->num_lines == next->s_lno) {
 262                        ent->num_lines += next->num_lines;
 263                        ent->next = next->next;
 264                        if (ent->next)
 265                                ent->next->prev = ent;
 266                        origin_decref(next->suspect);
 267                        free(next);
 268                        ent->score = 0;
 269                        next = ent; /* again */
 270                }
 271        }
 272
 273        if (DEBUG) /* sanity */
 274                sanity_check_refcnt(sb);
 275}
 276
 277/*
 278 * Given a commit and a path in it, create a new origin structure.
 279 * The callers that add blame to the scoreboard should use
 280 * get_origin() to obtain shared, refcounted copy instead of calling
 281 * this function directly.
 282 */
 283static struct origin *make_origin(struct commit *commit, const char *path)
 284{
 285        struct origin *o;
 286        o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 287        o->commit = commit;
 288        o->refcnt = 1;
 289        strcpy(o->path, path);
 290        return o;
 291}
 292
 293/*
 294 * Locate an existing origin or create a new one.
 295 */
 296static struct origin *get_origin(struct scoreboard *sb,
 297                                 struct commit *commit,
 298                                 const char *path)
 299{
 300        struct blame_entry *e;
 301
 302        for (e = sb->ent; e; e = e->next) {
 303                if (e->suspect->commit == commit &&
 304                    !strcmp(e->suspect->path, path))
 305                        return origin_incref(e->suspect);
 306        }
 307        return make_origin(commit, path);
 308}
 309
 310/*
 311 * Fill the blob_sha1 field of an origin if it hasn't, so that later
 312 * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 313 * for an origin is also used to pass the blame for the entire file to
 314 * the parent to detect the case where a child's blob is identical to
 315 * that of its parent's.
 316 */
 317static int fill_blob_sha1(struct origin *origin)
 318{
 319        unsigned mode;
 320        if (!is_null_sha1(origin->blob_sha1))
 321                return 0;
 322        if (get_tree_entry(origin->commit->object.sha1,
 323                           origin->path,
 324                           origin->blob_sha1, &mode))
 325                goto error_out;
 326        if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 327                goto error_out;
 328        return 0;
 329 error_out:
 330        hashclr(origin->blob_sha1);
 331        return -1;
 332}
 333
 334/*
 335 * We have an origin -- check if the same path exists in the
 336 * parent and return an origin structure to represent it.
 337 */
 338static struct origin *find_origin(struct scoreboard *sb,
 339                                  struct commit *parent,
 340                                  struct origin *origin)
 341{
 342        struct origin *porigin = NULL;
 343        struct diff_options diff_opts;
 344        const char *paths[2];
 345
 346        if (parent->util) {
 347                /*
 348                 * Each commit object can cache one origin in that
 349                 * commit.  This is a freestanding copy of origin and
 350                 * not refcounted.
 351                 */
 352                struct origin *cached = parent->util;
 353                if (!strcmp(cached->path, origin->path)) {
 354                        /*
 355                         * The same path between origin and its parent
 356                         * without renaming -- the most common case.
 357                         */
 358                        porigin = get_origin(sb, parent, cached->path);
 359
 360                        /*
 361                         * If the origin was newly created (i.e. get_origin
 362                         * would call make_origin if none is found in the
 363                         * scoreboard), it does not know the blob_sha1,
 364                         * so copy it.  Otherwise porigin was in the
 365                         * scoreboard and already knows blob_sha1.
 366                         */
 367                        if (porigin->refcnt == 1)
 368                                hashcpy(porigin->blob_sha1, cached->blob_sha1);
 369                        return porigin;
 370                }
 371                /* otherwise it was not very useful; free it */
 372                free(parent->util);
 373                parent->util = NULL;
 374        }
 375
 376        /* See if the origin->path is different between parent
 377         * and origin first.  Most of the time they are the
 378         * same and diff-tree is fairly efficient about this.
 379         */
 380        diff_setup(&diff_opts);
 381        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 382        diff_opts.detect_rename = 0;
 383        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 384        paths[0] = origin->path;
 385        paths[1] = NULL;
 386
 387        diff_tree_setup_paths(paths, &diff_opts);
 388        if (diff_setup_done(&diff_opts) < 0)
 389                die("diff-setup");
 390
 391        if (is_null_sha1(origin->commit->object.sha1))
 392                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 393        else
 394                diff_tree_sha1(parent->tree->object.sha1,
 395                               origin->commit->tree->object.sha1,
 396                               "", &diff_opts);
 397        diffcore_std(&diff_opts);
 398
 399        if (!diff_queued_diff.nr) {
 400                /* The path is the same as parent */
 401                porigin = get_origin(sb, parent, origin->path);
 402                hashcpy(porigin->blob_sha1, origin->blob_sha1);
 403        } else {
 404                /*
 405                 * Since origin->path is a pathspec, if the parent
 406                 * commit had it as a directory, we will see a whole
 407                 * bunch of deletion of files in the directory that we
 408                 * do not care about.
 409                 */
 410                int i;
 411                struct diff_filepair *p = NULL;
 412                for (i = 0; i < diff_queued_diff.nr; i++) {
 413                        const char *name;
 414                        p = diff_queued_diff.queue[i];
 415                        name = p->one->path ? p->one->path : p->two->path;
 416                        if (!strcmp(name, origin->path))
 417                                break;
 418                }
 419                if (!p)
 420                        die("internal error in blame::find_origin");
 421                switch (p->status) {
 422                default:
 423                        die("internal error in blame::find_origin (%c)",
 424                            p->status);
 425                case 'M':
 426                        porigin = get_origin(sb, parent, origin->path);
 427                        hashcpy(porigin->blob_sha1, p->one->sha1);
 428                        break;
 429                case 'A':
 430                case 'T':
 431                        /* Did not exist in parent, or type changed */
 432                        break;
 433                }
 434        }
 435        diff_flush(&diff_opts);
 436        diff_tree_release_paths(&diff_opts);
 437        if (porigin) {
 438                /*
 439                 * Create a freestanding copy that is not part of
 440                 * the refcounted origin found in the scoreboard, and
 441                 * cache it in the commit.
 442                 */
 443                struct origin *cached;
 444
 445                cached = make_origin(porigin->commit, porigin->path);
 446                hashcpy(cached->blob_sha1, porigin->blob_sha1);
 447                parent->util = cached;
 448        }
 449        return porigin;
 450}
 451
 452/*
 453 * We have an origin -- find the path that corresponds to it in its
 454 * parent and return an origin structure to represent it.
 455 */
 456static struct origin *find_rename(struct scoreboard *sb,
 457                                  struct commit *parent,
 458                                  struct origin *origin)
 459{
 460        struct origin *porigin = NULL;
 461        struct diff_options diff_opts;
 462        int i;
 463        const char *paths[2];
 464
 465        diff_setup(&diff_opts);
 466        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 467        diff_opts.detect_rename = DIFF_DETECT_RENAME;
 468        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 469        diff_opts.single_follow = origin->path;
 470        paths[0] = NULL;
 471        diff_tree_setup_paths(paths, &diff_opts);
 472        if (diff_setup_done(&diff_opts) < 0)
 473                die("diff-setup");
 474
 475        if (is_null_sha1(origin->commit->object.sha1))
 476                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 477        else
 478                diff_tree_sha1(parent->tree->object.sha1,
 479                               origin->commit->tree->object.sha1,
 480                               "", &diff_opts);
 481        diffcore_std(&diff_opts);
 482
 483        for (i = 0; i < diff_queued_diff.nr; i++) {
 484                struct diff_filepair *p = diff_queued_diff.queue[i];
 485                if ((p->status == 'R' || p->status == 'C') &&
 486                    !strcmp(p->two->path, origin->path)) {
 487                        porigin = get_origin(sb, parent, p->one->path);
 488                        hashcpy(porigin->blob_sha1, p->one->sha1);
 489                        break;
 490                }
 491        }
 492        diff_flush(&diff_opts);
 493        diff_tree_release_paths(&diff_opts);
 494        return porigin;
 495}
 496
 497/*
 498 * Link in a new blame entry to the scoreboard.  Entries that cover the
 499 * same line range have been removed from the scoreboard previously.
 500 */
 501static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 502{
 503        struct blame_entry *ent, *prev = NULL;
 504
 505        origin_incref(e->suspect);
 506
 507        for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 508                prev = ent;
 509
 510        /* prev, if not NULL, is the last one that is below e */
 511        e->prev = prev;
 512        if (prev) {
 513                e->next = prev->next;
 514                prev->next = e;
 515        }
 516        else {
 517                e->next = sb->ent;
 518                sb->ent = e;
 519        }
 520        if (e->next)
 521                e->next->prev = e;
 522}
 523
 524/*
 525 * src typically is on-stack; we want to copy the information in it to
 526 * a malloced blame_entry that is already on the linked list of the
 527 * scoreboard.  The origin of dst loses a refcnt while the origin of src
 528 * gains one.
 529 */
 530static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 531{
 532        struct blame_entry *p, *n;
 533
 534        p = dst->prev;
 535        n = dst->next;
 536        origin_incref(src->suspect);
 537        origin_decref(dst->suspect);
 538        memcpy(dst, src, sizeof(*src));
 539        dst->prev = p;
 540        dst->next = n;
 541        dst->score = 0;
 542}
 543
 544static const char *nth_line(struct scoreboard *sb, int lno)
 545{
 546        return sb->final_buf + sb->lineno[lno];
 547}
 548
 549/*
 550 * It is known that lines between tlno to same came from parent, and e
 551 * has an overlap with that range.  it also is known that parent's
 552 * line plno corresponds to e's line tlno.
 553 *
 554 *                <---- e ----->
 555 *                   <------>
 556 *                   <------------>
 557 *             <------------>
 558 *             <------------------>
 559 *
 560 * Split e into potentially three parts; before this chunk, the chunk
 561 * to be blamed for the parent, and after that portion.
 562 */
 563static void split_overlap(struct blame_entry *split,
 564                          struct blame_entry *e,
 565                          int tlno, int plno, int same,
 566                          struct origin *parent)
 567{
 568        int chunk_end_lno;
 569        memset(split, 0, sizeof(struct blame_entry [3]));
 570
 571        if (e->s_lno < tlno) {
 572                /* there is a pre-chunk part not blamed on parent */
 573                split[0].suspect = origin_incref(e->suspect);
 574                split[0].lno = e->lno;
 575                split[0].s_lno = e->s_lno;
 576                split[0].num_lines = tlno - e->s_lno;
 577                split[1].lno = e->lno + tlno - e->s_lno;
 578                split[1].s_lno = plno;
 579        }
 580        else {
 581                split[1].lno = e->lno;
 582                split[1].s_lno = plno + (e->s_lno - tlno);
 583        }
 584
 585        if (same < e->s_lno + e->num_lines) {
 586                /* there is a post-chunk part not blamed on parent */
 587                split[2].suspect = origin_incref(e->suspect);
 588                split[2].lno = e->lno + (same - e->s_lno);
 589                split[2].s_lno = e->s_lno + (same - e->s_lno);
 590                split[2].num_lines = e->s_lno + e->num_lines - same;
 591                chunk_end_lno = split[2].lno;
 592        }
 593        else
 594                chunk_end_lno = e->lno + e->num_lines;
 595        split[1].num_lines = chunk_end_lno - split[1].lno;
 596
 597        /*
 598         * if it turns out there is nothing to blame the parent for,
 599         * forget about the splitting.  !split[1].suspect signals this.
 600         */
 601        if (split[1].num_lines < 1)
 602                return;
 603        split[1].suspect = origin_incref(parent);
 604}
 605
 606/*
 607 * split_overlap() divided an existing blame e into up to three parts
 608 * in split.  Adjust the linked list of blames in the scoreboard to
 609 * reflect the split.
 610 */
 611static void split_blame(struct scoreboard *sb,
 612                        struct blame_entry *split,
 613                        struct blame_entry *e)
 614{
 615        struct blame_entry *new_entry;
 616
 617        if (split[0].suspect && split[2].suspect) {
 618                /* The first part (reuse storage for the existing entry e) */
 619                dup_entry(e, &split[0]);
 620
 621                /* The last part -- me */
 622                new_entry = xmalloc(sizeof(*new_entry));
 623                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 624                add_blame_entry(sb, new_entry);
 625
 626                /* ... and the middle part -- parent */
 627                new_entry = xmalloc(sizeof(*new_entry));
 628                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 629                add_blame_entry(sb, new_entry);
 630        }
 631        else if (!split[0].suspect && !split[2].suspect)
 632                /*
 633                 * The parent covers the entire area; reuse storage for
 634                 * e and replace it with the parent.
 635                 */
 636                dup_entry(e, &split[1]);
 637        else if (split[0].suspect) {
 638                /* me and then parent */
 639                dup_entry(e, &split[0]);
 640
 641                new_entry = xmalloc(sizeof(*new_entry));
 642                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 643                add_blame_entry(sb, new_entry);
 644        }
 645        else {
 646                /* parent and then me */
 647                dup_entry(e, &split[1]);
 648
 649                new_entry = xmalloc(sizeof(*new_entry));
 650                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 651                add_blame_entry(sb, new_entry);
 652        }
 653
 654        if (DEBUG) { /* sanity */
 655                struct blame_entry *ent;
 656                int lno = sb->ent->lno, corrupt = 0;
 657
 658                for (ent = sb->ent; ent; ent = ent->next) {
 659                        if (lno != ent->lno)
 660                                corrupt = 1;
 661                        if (ent->s_lno < 0)
 662                                corrupt = 1;
 663                        lno += ent->num_lines;
 664                }
 665                if (corrupt) {
 666                        lno = sb->ent->lno;
 667                        for (ent = sb->ent; ent; ent = ent->next) {
 668                                printf("L %8d l %8d n %8d\n",
 669                                       lno, ent->lno, ent->num_lines);
 670                                lno = ent->lno + ent->num_lines;
 671                        }
 672                        die("oops");
 673                }
 674        }
 675}
 676
 677/*
 678 * After splitting the blame, the origins used by the
 679 * on-stack blame_entry should lose one refcnt each.
 680 */
 681static void decref_split(struct blame_entry *split)
 682{
 683        int i;
 684
 685        for (i = 0; i < 3; i++)
 686                origin_decref(split[i].suspect);
 687}
 688
 689/*
 690 * Helper for blame_chunk().  blame_entry e is known to overlap with
 691 * the patch hunk; split it and pass blame to the parent.
 692 */
 693static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 694                          int tlno, int plno, int same,
 695                          struct origin *parent)
 696{
 697        struct blame_entry split[3];
 698
 699        split_overlap(split, e, tlno, plno, same, parent);
 700        if (split[1].suspect)
 701                split_blame(sb, split, e);
 702        decref_split(split);
 703}
 704
 705/*
 706 * Find the line number of the last line the target is suspected for.
 707 */
 708static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 709{
 710        struct blame_entry *e;
 711        int last_in_target = -1;
 712
 713        for (e = sb->ent; e; e = e->next) {
 714                if (e->guilty || !same_suspect(e->suspect, target))
 715                        continue;
 716                if (last_in_target < e->s_lno + e->num_lines)
 717                        last_in_target = e->s_lno + e->num_lines;
 718        }
 719        return last_in_target;
 720}
 721
 722/*
 723 * Process one hunk from the patch between the current suspect for
 724 * blame_entry e and its parent.  Find and split the overlap, and
 725 * pass blame to the overlapping part to the parent.
 726 */
 727static void blame_chunk(struct scoreboard *sb,
 728                        int tlno, int plno, int same,
 729                        struct origin *target, struct origin *parent)
 730{
 731        struct blame_entry *e;
 732
 733        for (e = sb->ent; e; e = e->next) {
 734                if (e->guilty || !same_suspect(e->suspect, target))
 735                        continue;
 736                if (same <= e->s_lno)
 737                        continue;
 738                if (tlno < e->s_lno + e->num_lines)
 739                        blame_overlap(sb, e, tlno, plno, same, parent);
 740        }
 741}
 742
 743struct blame_chunk_cb_data {
 744        struct scoreboard *sb;
 745        struct origin *target;
 746        struct origin *parent;
 747        long plno;
 748        long tlno;
 749};
 750
 751static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
 752{
 753        struct blame_chunk_cb_data *d = data;
 754        blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
 755        d->plno = p_next;
 756        d->tlno = t_next;
 757}
 758
 759/*
 760 * We are looking at the origin 'target' and aiming to pass blame
 761 * for the lines it is suspected to its parent.  Run diff to find
 762 * which lines came from parent and pass blame for them.
 763 */
 764static int pass_blame_to_parent(struct scoreboard *sb,
 765                                struct origin *target,
 766                                struct origin *parent)
 767{
 768        int last_in_target;
 769        mmfile_t file_p, file_o;
 770        struct blame_chunk_cb_data d;
 771        xpparam_t xpp;
 772        xdemitconf_t xecfg;
 773        memset(&d, 0, sizeof(d));
 774        d.sb = sb; d.target = target; d.parent = parent;
 775        last_in_target = find_last_in_target(sb, target);
 776        if (last_in_target < 0)
 777                return 1; /* nothing remains for this target */
 778
 779        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 780        fill_origin_blob(&sb->revs->diffopt, target, &file_o);
 781        num_get_patch++;
 782
 783        memset(&xpp, 0, sizeof(xpp));
 784        xpp.flags = xdl_opts;
 785        memset(&xecfg, 0, sizeof(xecfg));
 786        xecfg.ctxlen = 0;
 787        xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
 788        /* The rest (i.e. anything after tlno) are the same as the parent */
 789        blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
 790
 791        return 0;
 792}
 793
 794/*
 795 * The lines in blame_entry after splitting blames many times can become
 796 * very small and trivial, and at some point it becomes pointless to
 797 * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 798 * ordinary C program, and it is not worth to say it was copied from
 799 * totally unrelated file in the parent.
 800 *
 801 * Compute how trivial the lines in the blame_entry are.
 802 */
 803static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 804{
 805        unsigned score;
 806        const char *cp, *ep;
 807
 808        if (e->score)
 809                return e->score;
 810
 811        score = 1;
 812        cp = nth_line(sb, e->lno);
 813        ep = nth_line(sb, e->lno + e->num_lines);
 814        while (cp < ep) {
 815                unsigned ch = *((unsigned char *)cp);
 816                if (isalnum(ch))
 817                        score++;
 818                cp++;
 819        }
 820        e->score = score;
 821        return score;
 822}
 823
 824/*
 825 * best_so_far[] and this[] are both a split of an existing blame_entry
 826 * that passes blame to the parent.  Maintain best_so_far the best split
 827 * so far, by comparing this and best_so_far and copying this into
 828 * bst_so_far as needed.
 829 */
 830static void copy_split_if_better(struct scoreboard *sb,
 831                                 struct blame_entry *best_so_far,
 832                                 struct blame_entry *this)
 833{
 834        int i;
 835
 836        if (!this[1].suspect)
 837                return;
 838        if (best_so_far[1].suspect) {
 839                if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 840                        return;
 841        }
 842
 843        for (i = 0; i < 3; i++)
 844                origin_incref(this[i].suspect);
 845        decref_split(best_so_far);
 846        memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 847}
 848
 849/*
 850 * We are looking at a part of the final image represented by
 851 * ent (tlno and same are offset by ent->s_lno).
 852 * tlno is where we are looking at in the final image.
 853 * up to (but not including) same match preimage.
 854 * plno is where we are looking at in the preimage.
 855 *
 856 * <-------------- final image ---------------------->
 857 *       <------ent------>
 858 *         ^tlno ^same
 859 *    <---------preimage----->
 860 *         ^plno
 861 *
 862 * All line numbers are 0-based.
 863 */
 864static void handle_split(struct scoreboard *sb,
 865                         struct blame_entry *ent,
 866                         int tlno, int plno, int same,
 867                         struct origin *parent,
 868                         struct blame_entry *split)
 869{
 870        if (ent->num_lines <= tlno)
 871                return;
 872        if (tlno < same) {
 873                struct blame_entry this[3];
 874                tlno += ent->s_lno;
 875                same += ent->s_lno;
 876                split_overlap(this, ent, tlno, plno, same, parent);
 877                copy_split_if_better(sb, split, this);
 878                decref_split(this);
 879        }
 880}
 881
 882struct handle_split_cb_data {
 883        struct scoreboard *sb;
 884        struct blame_entry *ent;
 885        struct origin *parent;
 886        struct blame_entry *split;
 887        long plno;
 888        long tlno;
 889};
 890
 891static void handle_split_cb(void *data, long same, long p_next, long t_next)
 892{
 893        struct handle_split_cb_data *d = data;
 894        handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
 895        d->plno = p_next;
 896        d->tlno = t_next;
 897}
 898
 899/*
 900 * Find the lines from parent that are the same as ent so that
 901 * we can pass blames to it.  file_p has the blob contents for
 902 * the parent.
 903 */
 904static void find_copy_in_blob(struct scoreboard *sb,
 905                              struct blame_entry *ent,
 906                              struct origin *parent,
 907                              struct blame_entry *split,
 908                              mmfile_t *file_p)
 909{
 910        const char *cp;
 911        int cnt;
 912        mmfile_t file_o;
 913        struct handle_split_cb_data d;
 914        xpparam_t xpp;
 915        xdemitconf_t xecfg;
 916        memset(&d, 0, sizeof(d));
 917        d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
 918        /*
 919         * Prepare mmfile that contains only the lines in ent.
 920         */
 921        cp = nth_line(sb, ent->lno);
 922        file_o.ptr = (char *) cp;
 923        cnt = ent->num_lines;
 924
 925        while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 926                if (*cp++ == '\n')
 927                        cnt--;
 928        }
 929        file_o.size = cp - file_o.ptr;
 930
 931        /*
 932         * file_o is a part of final image we are annotating.
 933         * file_p partially may match that image.
 934         */
 935        memset(&xpp, 0, sizeof(xpp));
 936        xpp.flags = xdl_opts;
 937        memset(&xecfg, 0, sizeof(xecfg));
 938        xecfg.ctxlen = 1;
 939        memset(split, 0, sizeof(struct blame_entry [3]));
 940        xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
 941        /* remainder, if any, all match the preimage */
 942        handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
 943}
 944
 945/*
 946 * See if lines currently target is suspected for can be attributed to
 947 * parent.
 948 */
 949static int find_move_in_parent(struct scoreboard *sb,
 950                               struct origin *target,
 951                               struct origin *parent)
 952{
 953        int last_in_target, made_progress;
 954        struct blame_entry *e, split[3];
 955        mmfile_t file_p;
 956
 957        last_in_target = find_last_in_target(sb, target);
 958        if (last_in_target < 0)
 959                return 1; /* nothing remains for this target */
 960
 961        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 962        if (!file_p.ptr)
 963                return 0;
 964
 965        made_progress = 1;
 966        while (made_progress) {
 967                made_progress = 0;
 968                for (e = sb->ent; e; e = e->next) {
 969                        if (e->guilty || !same_suspect(e->suspect, target) ||
 970                            ent_score(sb, e) < blame_move_score)
 971                                continue;
 972                        find_copy_in_blob(sb, e, parent, split, &file_p);
 973                        if (split[1].suspect &&
 974                            blame_move_score < ent_score(sb, &split[1])) {
 975                                split_blame(sb, split, e);
 976                                made_progress = 1;
 977                        }
 978                        decref_split(split);
 979                }
 980        }
 981        return 0;
 982}
 983
 984struct blame_list {
 985        struct blame_entry *ent;
 986        struct blame_entry split[3];
 987};
 988
 989/*
 990 * Count the number of entries the target is suspected for,
 991 * and prepare a list of entry and the best split.
 992 */
 993static struct blame_list *setup_blame_list(struct scoreboard *sb,
 994                                           struct origin *target,
 995                                           int min_score,
 996                                           int *num_ents_p)
 997{
 998        struct blame_entry *e;
 999        int num_ents, i;
1000        struct blame_list *blame_list = NULL;
1001
1002        for (e = sb->ent, num_ents = 0; e; e = e->next)
1003                if (!e->scanned && !e->guilty &&
1004                    same_suspect(e->suspect, target) &&
1005                    min_score < ent_score(sb, e))
1006                        num_ents++;
1007        if (num_ents) {
1008                blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1009                for (e = sb->ent, i = 0; e; e = e->next)
1010                        if (!e->scanned && !e->guilty &&
1011                            same_suspect(e->suspect, target) &&
1012                            min_score < ent_score(sb, e))
1013                                blame_list[i++].ent = e;
1014        }
1015        *num_ents_p = num_ents;
1016        return blame_list;
1017}
1018
1019/*
1020 * Reset the scanned status on all entries.
1021 */
1022static void reset_scanned_flag(struct scoreboard *sb)
1023{
1024        struct blame_entry *e;
1025        for (e = sb->ent; e; e = e->next)
1026                e->scanned = 0;
1027}
1028
1029/*
1030 * For lines target is suspected for, see if we can find code movement
1031 * across file boundary from the parent commit.  porigin is the path
1032 * in the parent we already tried.
1033 */
1034static int find_copy_in_parent(struct scoreboard *sb,
1035                               struct origin *target,
1036                               struct commit *parent,
1037                               struct origin *porigin,
1038                               int opt)
1039{
1040        struct diff_options diff_opts;
1041        const char *paths[1];
1042        int i, j;
1043        int retval;
1044        struct blame_list *blame_list;
1045        int num_ents;
1046
1047        blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1048        if (!blame_list)
1049                return 1; /* nothing remains for this target */
1050
1051        diff_setup(&diff_opts);
1052        DIFF_OPT_SET(&diff_opts, RECURSIVE);
1053        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1054
1055        paths[0] = NULL;
1056        diff_tree_setup_paths(paths, &diff_opts);
1057        if (diff_setup_done(&diff_opts) < 0)
1058                die("diff-setup");
1059
1060        /* Try "find copies harder" on new path if requested;
1061         * we do not want to use diffcore_rename() actually to
1062         * match things up; find_copies_harder is set only to
1063         * force diff_tree_sha1() to feed all filepairs to diff_queue,
1064         * and this code needs to be after diff_setup_done(), which
1065         * usually makes find-copies-harder imply copy detection.
1066         */
1067        if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1068            || ((opt & PICKAXE_BLAME_COPY_HARDER)
1069                && (!porigin || strcmp(target->path, porigin->path))))
1070                DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1071
1072        if (is_null_sha1(target->commit->object.sha1))
1073                do_diff_cache(parent->tree->object.sha1, &diff_opts);
1074        else
1075                diff_tree_sha1(parent->tree->object.sha1,
1076                               target->commit->tree->object.sha1,
1077                               "", &diff_opts);
1078
1079        if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1080                diffcore_std(&diff_opts);
1081
1082        retval = 0;
1083        while (1) {
1084                int made_progress = 0;
1085
1086                for (i = 0; i < diff_queued_diff.nr; i++) {
1087                        struct diff_filepair *p = diff_queued_diff.queue[i];
1088                        struct origin *norigin;
1089                        mmfile_t file_p;
1090                        struct blame_entry this[3];
1091
1092                        if (!DIFF_FILE_VALID(p->one))
1093                                continue; /* does not exist in parent */
1094                        if (S_ISGITLINK(p->one->mode))
1095                                continue; /* ignore git links */
1096                        if (porigin && !strcmp(p->one->path, porigin->path))
1097                                /* find_move already dealt with this path */
1098                                continue;
1099
1100                        norigin = get_origin(sb, parent, p->one->path);
1101                        hashcpy(norigin->blob_sha1, p->one->sha1);
1102                        fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1103                        if (!file_p.ptr)
1104                                continue;
1105
1106                        for (j = 0; j < num_ents; j++) {
1107                                find_copy_in_blob(sb, blame_list[j].ent,
1108                                                  norigin, this, &file_p);
1109                                copy_split_if_better(sb, blame_list[j].split,
1110                                                     this);
1111                                decref_split(this);
1112                        }
1113                        origin_decref(norigin);
1114                }
1115
1116                for (j = 0; j < num_ents; j++) {
1117                        struct blame_entry *split = blame_list[j].split;
1118                        if (split[1].suspect &&
1119                            blame_copy_score < ent_score(sb, &split[1])) {
1120                                split_blame(sb, split, blame_list[j].ent);
1121                                made_progress = 1;
1122                        }
1123                        else
1124                                blame_list[j].ent->scanned = 1;
1125                        decref_split(split);
1126                }
1127                free(blame_list);
1128
1129                if (!made_progress)
1130                        break;
1131                blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1132                if (!blame_list) {
1133                        retval = 1;
1134                        break;
1135                }
1136        }
1137        reset_scanned_flag(sb);
1138        diff_flush(&diff_opts);
1139        diff_tree_release_paths(&diff_opts);
1140        return retval;
1141}
1142
1143/*
1144 * The blobs of origin and porigin exactly match, so everything
1145 * origin is suspected for can be blamed on the parent.
1146 */
1147static void pass_whole_blame(struct scoreboard *sb,
1148                             struct origin *origin, struct origin *porigin)
1149{
1150        struct blame_entry *e;
1151
1152        if (!porigin->file.ptr && origin->file.ptr) {
1153                /* Steal its file */
1154                porigin->file = origin->file;
1155                origin->file.ptr = NULL;
1156        }
1157        for (e = sb->ent; e; e = e->next) {
1158                if (!same_suspect(e->suspect, origin))
1159                        continue;
1160                origin_incref(porigin);
1161                origin_decref(e->suspect);
1162                e->suspect = porigin;
1163        }
1164}
1165
1166/*
1167 * We pass blame from the current commit to its parents.  We keep saying
1168 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1169 * exonerate ourselves.
1170 */
1171static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1172{
1173        if (!reverse)
1174                return commit->parents;
1175        return lookup_decoration(&revs->children, &commit->object);
1176}
1177
1178static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1179{
1180        int cnt;
1181        struct commit_list *l = first_scapegoat(revs, commit);
1182        for (cnt = 0; l; l = l->next)
1183                cnt++;
1184        return cnt;
1185}
1186
1187#define MAXSG 16
1188
1189static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1190{
1191        struct rev_info *revs = sb->revs;
1192        int i, pass, num_sg;
1193        struct commit *commit = origin->commit;
1194        struct commit_list *sg;
1195        struct origin *sg_buf[MAXSG];
1196        struct origin *porigin, **sg_origin = sg_buf;
1197
1198        num_sg = num_scapegoats(revs, commit);
1199        if (!num_sg)
1200                goto finish;
1201        else if (num_sg < ARRAY_SIZE(sg_buf))
1202                memset(sg_buf, 0, sizeof(sg_buf));
1203        else
1204                sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1205
1206        /*
1207         * The first pass looks for unrenamed path to optimize for
1208         * common cases, then we look for renames in the second pass.
1209         */
1210        for (pass = 0; pass < 2; pass++) {
1211                struct origin *(*find)(struct scoreboard *,
1212                                       struct commit *, struct origin *);
1213                find = pass ? find_rename : find_origin;
1214
1215                for (i = 0, sg = first_scapegoat(revs, commit);
1216                     i < num_sg && sg;
1217                     sg = sg->next, i++) {
1218                        struct commit *p = sg->item;
1219                        int j, same;
1220
1221                        if (sg_origin[i])
1222                                continue;
1223                        if (parse_commit(p))
1224                                continue;
1225                        porigin = find(sb, p, origin);
1226                        if (!porigin)
1227                                continue;
1228                        if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1229                                pass_whole_blame(sb, origin, porigin);
1230                                origin_decref(porigin);
1231                                goto finish;
1232                        }
1233                        for (j = same = 0; j < i; j++)
1234                                if (sg_origin[j] &&
1235                                    !hashcmp(sg_origin[j]->blob_sha1,
1236                                             porigin->blob_sha1)) {
1237                                        same = 1;
1238                                        break;
1239                                }
1240                        if (!same)
1241                                sg_origin[i] = porigin;
1242                        else
1243                                origin_decref(porigin);
1244                }
1245        }
1246
1247        num_commits++;
1248        for (i = 0, sg = first_scapegoat(revs, commit);
1249             i < num_sg && sg;
1250             sg = sg->next, i++) {
1251                struct origin *porigin = sg_origin[i];
1252                if (!porigin)
1253                        continue;
1254                if (!origin->previous) {
1255                        origin_incref(porigin);
1256                        origin->previous = porigin;
1257                }
1258                if (pass_blame_to_parent(sb, origin, porigin))
1259                        goto finish;
1260        }
1261
1262        /*
1263         * Optionally find moves in parents' files.
1264         */
1265        if (opt & PICKAXE_BLAME_MOVE)
1266                for (i = 0, sg = first_scapegoat(revs, commit);
1267                     i < num_sg && sg;
1268                     sg = sg->next, i++) {
1269                        struct origin *porigin = sg_origin[i];
1270                        if (!porigin)
1271                                continue;
1272                        if (find_move_in_parent(sb, origin, porigin))
1273                                goto finish;
1274                }
1275
1276        /*
1277         * Optionally find copies from parents' files.
1278         */
1279        if (opt & PICKAXE_BLAME_COPY)
1280                for (i = 0, sg = first_scapegoat(revs, commit);
1281                     i < num_sg && sg;
1282                     sg = sg->next, i++) {
1283                        struct origin *porigin = sg_origin[i];
1284                        if (find_copy_in_parent(sb, origin, sg->item,
1285                                                porigin, opt))
1286                                goto finish;
1287                }
1288
1289 finish:
1290        for (i = 0; i < num_sg; i++) {
1291                if (sg_origin[i]) {
1292                        drop_origin_blob(sg_origin[i]);
1293                        origin_decref(sg_origin[i]);
1294                }
1295        }
1296        drop_origin_blob(origin);
1297        if (sg_buf != sg_origin)
1298                free(sg_origin);
1299}
1300
1301/*
1302 * Information on commits, used for output.
1303 */
1304struct commit_info
1305{
1306        const char *author;
1307        const char *author_mail;
1308        unsigned long author_time;
1309        const char *author_tz;
1310
1311        /* filled only when asked for details */
1312        const char *committer;
1313        const char *committer_mail;
1314        unsigned long committer_time;
1315        const char *committer_tz;
1316
1317        const char *summary;
1318};
1319
1320/*
1321 * Parse author/committer line in the commit object buffer
1322 */
1323static void get_ac_line(const char *inbuf, const char *what,
1324                        int person_len, char *person,
1325                        int mail_len, char *mail,
1326                        unsigned long *time, const char **tz)
1327{
1328        int len, tzlen, maillen;
1329        char *tmp, *endp, *timepos, *mailpos;
1330
1331        tmp = strstr(inbuf, what);
1332        if (!tmp)
1333                goto error_out;
1334        tmp += strlen(what);
1335        endp = strchr(tmp, '\n');
1336        if (!endp)
1337                len = strlen(tmp);
1338        else
1339                len = endp - tmp;
1340        if (person_len <= len) {
1341        error_out:
1342                /* Ugh */
1343                *tz = "(unknown)";
1344                strcpy(person, *tz);
1345                strcpy(mail, *tz);
1346                *time = 0;
1347                return;
1348        }
1349        memcpy(person, tmp, len);
1350
1351        tmp = person;
1352        tmp += len;
1353        *tmp = 0;
1354        while (person < tmp && *tmp != ' ')
1355                tmp--;
1356        if (tmp <= person)
1357                goto error_out;
1358        *tz = tmp+1;
1359        tzlen = (person+len)-(tmp+1);
1360
1361        *tmp = 0;
1362        while (person < tmp && *tmp != ' ')
1363                tmp--;
1364        if (tmp <= person)
1365                goto error_out;
1366        *time = strtoul(tmp, NULL, 10);
1367        timepos = tmp;
1368
1369        *tmp = 0;
1370        while (person < tmp && *tmp != ' ')
1371                tmp--;
1372        if (tmp <= person)
1373                return;
1374        mailpos = tmp + 1;
1375        *tmp = 0;
1376        maillen = timepos - tmp;
1377        memcpy(mail, mailpos, maillen);
1378
1379        if (!mailmap.nr)
1380                return;
1381
1382        /*
1383         * mailmap expansion may make the name longer.
1384         * make room by pushing stuff down.
1385         */
1386        tmp = person + person_len - (tzlen + 1);
1387        memmove(tmp, *tz, tzlen);
1388        tmp[tzlen] = 0;
1389        *tz = tmp;
1390
1391        /*
1392         * Now, convert both name and e-mail using mailmap
1393         */
1394        if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1395                /* Add a trailing '>' to email, since map_user returns plain emails
1396                   Note: It already has '<', since we replace from mail+1 */
1397                mailpos = memchr(mail, '\0', mail_len);
1398                if (mailpos && mailpos-mail < mail_len - 1) {
1399                        *mailpos = '>';
1400                        *(mailpos+1) = '\0';
1401                }
1402        }
1403}
1404
1405static void get_commit_info(struct commit *commit,
1406                            struct commit_info *ret,
1407                            int detailed)
1408{
1409        int len;
1410        char *tmp, *endp, *reencoded, *message;
1411        static char author_name[1024];
1412        static char author_mail[1024];
1413        static char committer_name[1024];
1414        static char committer_mail[1024];
1415        static char summary_buf[1024];
1416
1417        /*
1418         * We've operated without save_commit_buffer, so
1419         * we now need to populate them for output.
1420         */
1421        if (!commit->buffer) {
1422                enum object_type type;
1423                unsigned long size;
1424                commit->buffer =
1425                        read_sha1_file(commit->object.sha1, &type, &size);
1426                if (!commit->buffer)
1427                        die("Cannot read commit %s",
1428                            sha1_to_hex(commit->object.sha1));
1429        }
1430        reencoded = reencode_commit_message(commit, NULL);
1431        message   = reencoded ? reencoded : commit->buffer;
1432        ret->author = author_name;
1433        ret->author_mail = author_mail;
1434        get_ac_line(message, "\nauthor ",
1435                    sizeof(author_name), author_name,
1436                    sizeof(author_mail), author_mail,
1437                    &ret->author_time, &ret->author_tz);
1438
1439        if (!detailed) {
1440                free(reencoded);
1441                return;
1442        }
1443
1444        ret->committer = committer_name;
1445        ret->committer_mail = committer_mail;
1446        get_ac_line(message, "\ncommitter ",
1447                    sizeof(committer_name), committer_name,
1448                    sizeof(committer_mail), committer_mail,
1449                    &ret->committer_time, &ret->committer_tz);
1450
1451        ret->summary = summary_buf;
1452        tmp = strstr(message, "\n\n");
1453        if (!tmp) {
1454        error_out:
1455                sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1456                free(reencoded);
1457                return;
1458        }
1459        tmp += 2;
1460        endp = strchr(tmp, '\n');
1461        if (!endp)
1462                endp = tmp + strlen(tmp);
1463        len = endp - tmp;
1464        if (len >= sizeof(summary_buf) || len == 0)
1465                goto error_out;
1466        memcpy(summary_buf, tmp, len);
1467        summary_buf[len] = 0;
1468        free(reencoded);
1469}
1470
1471/*
1472 * To allow LF and other nonportable characters in pathnames,
1473 * they are c-style quoted as needed.
1474 */
1475static void write_filename_info(const char *path)
1476{
1477        printf("filename ");
1478        write_name_quoted(path, stdout, '\n');
1479}
1480
1481/*
1482 * Porcelain/Incremental format wants to show a lot of details per
1483 * commit.  Instead of repeating this every line, emit it only once,
1484 * the first time each commit appears in the output.
1485 */
1486static int emit_one_suspect_detail(struct origin *suspect)
1487{
1488        struct commit_info ci;
1489
1490        if (suspect->commit->object.flags & METAINFO_SHOWN)
1491                return 0;
1492
1493        suspect->commit->object.flags |= METAINFO_SHOWN;
1494        get_commit_info(suspect->commit, &ci, 1);
1495        printf("author %s\n", ci.author);
1496        printf("author-mail %s\n", ci.author_mail);
1497        printf("author-time %lu\n", ci.author_time);
1498        printf("author-tz %s\n", ci.author_tz);
1499        printf("committer %s\n", ci.committer);
1500        printf("committer-mail %s\n", ci.committer_mail);
1501        printf("committer-time %lu\n", ci.committer_time);
1502        printf("committer-tz %s\n", ci.committer_tz);
1503        printf("summary %s\n", ci.summary);
1504        if (suspect->commit->object.flags & UNINTERESTING)
1505                printf("boundary\n");
1506        if (suspect->previous) {
1507                struct origin *prev = suspect->previous;
1508                printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1509                write_name_quoted(prev->path, stdout, '\n');
1510        }
1511        return 1;
1512}
1513
1514/*
1515 * The blame_entry is found to be guilty for the range.  Mark it
1516 * as such, and show it in incremental output.
1517 */
1518static void found_guilty_entry(struct blame_entry *ent)
1519{
1520        if (ent->guilty)
1521                return;
1522        ent->guilty = 1;
1523        if (incremental) {
1524                struct origin *suspect = ent->suspect;
1525
1526                printf("%s %d %d %d\n",
1527                       sha1_to_hex(suspect->commit->object.sha1),
1528                       ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1529                emit_one_suspect_detail(suspect);
1530                write_filename_info(suspect->path);
1531                maybe_flush_or_die(stdout, "stdout");
1532        }
1533}
1534
1535/*
1536 * The main loop -- while the scoreboard has lines whose true origin
1537 * is still unknown, pick one blame_entry, and allow its current
1538 * suspect to pass blames to its parents.
1539 */
1540static void assign_blame(struct scoreboard *sb, int opt)
1541{
1542        struct rev_info *revs = sb->revs;
1543
1544        while (1) {
1545                struct blame_entry *ent;
1546                struct commit *commit;
1547                struct origin *suspect = NULL;
1548
1549                /* find one suspect to break down */
1550                for (ent = sb->ent; !suspect && ent; ent = ent->next)
1551                        if (!ent->guilty)
1552                                suspect = ent->suspect;
1553                if (!suspect)
1554                        return; /* all done */
1555
1556                /*
1557                 * We will use this suspect later in the loop,
1558                 * so hold onto it in the meantime.
1559                 */
1560                origin_incref(suspect);
1561                commit = suspect->commit;
1562                if (!commit->object.parsed)
1563                        parse_commit(commit);
1564                if (reverse ||
1565                    (!(commit->object.flags & UNINTERESTING) &&
1566                     !(revs->max_age != -1 && commit->date < revs->max_age)))
1567                        pass_blame(sb, suspect, opt);
1568                else {
1569                        commit->object.flags |= UNINTERESTING;
1570                        if (commit->object.parsed)
1571                                mark_parents_uninteresting(commit);
1572                }
1573                /* treat root commit as boundary */
1574                if (!commit->parents && !show_root)
1575                        commit->object.flags |= UNINTERESTING;
1576
1577                /* Take responsibility for the remaining entries */
1578                for (ent = sb->ent; ent; ent = ent->next)
1579                        if (same_suspect(ent->suspect, suspect))
1580                                found_guilty_entry(ent);
1581                origin_decref(suspect);
1582
1583                if (DEBUG) /* sanity */
1584                        sanity_check_refcnt(sb);
1585        }
1586}
1587
1588static const char *format_time(unsigned long time, const char *tz_str,
1589                               int show_raw_time)
1590{
1591        static char time_buf[128];
1592        const char *time_str;
1593        int time_len;
1594        int tz;
1595
1596        if (show_raw_time) {
1597                sprintf(time_buf, "%lu %s", time, tz_str);
1598        }
1599        else {
1600                tz = atoi(tz_str);
1601                time_str = show_date(time, tz, blame_date_mode);
1602                time_len = strlen(time_str);
1603                memcpy(time_buf, time_str, time_len);
1604                memset(time_buf + time_len, ' ', blame_date_width - time_len);
1605        }
1606        return time_buf;
1607}
1608
1609#define OUTPUT_ANNOTATE_COMPAT  001
1610#define OUTPUT_LONG_OBJECT_NAME 002
1611#define OUTPUT_RAW_TIMESTAMP    004
1612#define OUTPUT_PORCELAIN        010
1613#define OUTPUT_SHOW_NAME        020
1614#define OUTPUT_SHOW_NUMBER      040
1615#define OUTPUT_SHOW_SCORE      0100
1616#define OUTPUT_NO_AUTHOR       0200
1617
1618static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1619{
1620        int cnt;
1621        const char *cp;
1622        struct origin *suspect = ent->suspect;
1623        char hex[41];
1624
1625        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1626        printf("%s%c%d %d %d\n",
1627               hex,
1628               ent->guilty ? ' ' : '*', /* purely for debugging */
1629               ent->s_lno + 1,
1630               ent->lno + 1,
1631               ent->num_lines);
1632        if (emit_one_suspect_detail(suspect) ||
1633            (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1634                write_filename_info(suspect->path);
1635
1636        cp = nth_line(sb, ent->lno);
1637        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1638                char ch;
1639                if (cnt)
1640                        printf("%s %d %d\n", hex,
1641                               ent->s_lno + 1 + cnt,
1642                               ent->lno + 1 + cnt);
1643                putchar('\t');
1644                do {
1645                        ch = *cp++;
1646                        putchar(ch);
1647                } while (ch != '\n' &&
1648                         cp < sb->final_buf + sb->final_buf_size);
1649        }
1650
1651        if (sb->final_buf_size && cp[-1] != '\n')
1652                putchar('\n');
1653}
1654
1655static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1656{
1657        int cnt;
1658        const char *cp;
1659        struct origin *suspect = ent->suspect;
1660        struct commit_info ci;
1661        char hex[41];
1662        int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1663
1664        get_commit_info(suspect->commit, &ci, 1);
1665        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1666
1667        cp = nth_line(sb, ent->lno);
1668        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1669                char ch;
1670                int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1671
1672                if (suspect->commit->object.flags & UNINTERESTING) {
1673                        if (blank_boundary)
1674                                memset(hex, ' ', length);
1675                        else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1676                                length--;
1677                                putchar('^');
1678                        }
1679                }
1680
1681                printf("%.*s", length, hex);
1682                if (opt & OUTPUT_ANNOTATE_COMPAT)
1683                        printf("\t(%10s\t%10s\t%d)", ci.author,
1684                               format_time(ci.author_time, ci.author_tz,
1685                                           show_raw_time),
1686                               ent->lno + 1 + cnt);
1687                else {
1688                        if (opt & OUTPUT_SHOW_SCORE)
1689                                printf(" %*d %02d",
1690                                       max_score_digits, ent->score,
1691                                       ent->suspect->refcnt);
1692                        if (opt & OUTPUT_SHOW_NAME)
1693                                printf(" %-*.*s", longest_file, longest_file,
1694                                       suspect->path);
1695                        if (opt & OUTPUT_SHOW_NUMBER)
1696                                printf(" %*d", max_orig_digits,
1697                                       ent->s_lno + 1 + cnt);
1698
1699                        if (!(opt & OUTPUT_NO_AUTHOR)) {
1700                                int pad = longest_author - utf8_strwidth(ci.author);
1701                                printf(" (%s%*s %10s",
1702                                       ci.author, pad, "",
1703                                       format_time(ci.author_time,
1704                                                   ci.author_tz,
1705                                                   show_raw_time));
1706                        }
1707                        printf(" %*d) ",
1708                               max_digits, ent->lno + 1 + cnt);
1709                }
1710                do {
1711                        ch = *cp++;
1712                        putchar(ch);
1713                } while (ch != '\n' &&
1714                         cp < sb->final_buf + sb->final_buf_size);
1715        }
1716
1717        if (sb->final_buf_size && cp[-1] != '\n')
1718                putchar('\n');
1719}
1720
1721static void output(struct scoreboard *sb, int option)
1722{
1723        struct blame_entry *ent;
1724
1725        if (option & OUTPUT_PORCELAIN) {
1726                for (ent = sb->ent; ent; ent = ent->next) {
1727                        struct blame_entry *oth;
1728                        struct origin *suspect = ent->suspect;
1729                        struct commit *commit = suspect->commit;
1730                        if (commit->object.flags & MORE_THAN_ONE_PATH)
1731                                continue;
1732                        for (oth = ent->next; oth; oth = oth->next) {
1733                                if ((oth->suspect->commit != commit) ||
1734                                    !strcmp(oth->suspect->path, suspect->path))
1735                                        continue;
1736                                commit->object.flags |= MORE_THAN_ONE_PATH;
1737                                break;
1738                        }
1739                }
1740        }
1741
1742        for (ent = sb->ent; ent; ent = ent->next) {
1743                if (option & OUTPUT_PORCELAIN)
1744                        emit_porcelain(sb, ent);
1745                else {
1746                        emit_other(sb, ent, option);
1747                }
1748        }
1749}
1750
1751/*
1752 * To allow quick access to the contents of nth line in the
1753 * final image, prepare an index in the scoreboard.
1754 */
1755static int prepare_lines(struct scoreboard *sb)
1756{
1757        const char *buf = sb->final_buf;
1758        unsigned long len = sb->final_buf_size;
1759        int num = 0, incomplete = 0, bol = 1;
1760
1761        if (len && buf[len-1] != '\n')
1762                incomplete++; /* incomplete line at the end */
1763        while (len--) {
1764                if (bol) {
1765                        sb->lineno = xrealloc(sb->lineno,
1766                                              sizeof(int *) * (num + 1));
1767                        sb->lineno[num] = buf - sb->final_buf;
1768                        bol = 0;
1769                }
1770                if (*buf++ == '\n') {
1771                        num++;
1772                        bol = 1;
1773                }
1774        }
1775        sb->lineno = xrealloc(sb->lineno,
1776                              sizeof(int *) * (num + incomplete + 1));
1777        sb->lineno[num + incomplete] = buf - sb->final_buf;
1778        sb->num_lines = num + incomplete;
1779        return sb->num_lines;
1780}
1781
1782/*
1783 * Add phony grafts for use with -S; this is primarily to
1784 * support git's cvsserver that wants to give a linear history
1785 * to its clients.
1786 */
1787static int read_ancestry(const char *graft_file)
1788{
1789        FILE *fp = fopen(graft_file, "r");
1790        char buf[1024];
1791        if (!fp)
1792                return -1;
1793        while (fgets(buf, sizeof(buf), fp)) {
1794                /* The format is just "Commit Parent1 Parent2 ...\n" */
1795                int len = strlen(buf);
1796                struct commit_graft *graft = read_graft_line(buf, len);
1797                if (graft)
1798                        register_commit_graft(graft, 0);
1799        }
1800        fclose(fp);
1801        return 0;
1802}
1803
1804/*
1805 * How many columns do we need to show line numbers in decimal?
1806 */
1807static int lineno_width(int lines)
1808{
1809        int i, width;
1810
1811        for (width = 1, i = 10; i <= lines; width++)
1812                i *= 10;
1813        return width;
1814}
1815
1816/*
1817 * How many columns do we need to show line numbers, authors,
1818 * and filenames?
1819 */
1820static void find_alignment(struct scoreboard *sb, int *option)
1821{
1822        int longest_src_lines = 0;
1823        int longest_dst_lines = 0;
1824        unsigned largest_score = 0;
1825        struct blame_entry *e;
1826
1827        for (e = sb->ent; e; e = e->next) {
1828                struct origin *suspect = e->suspect;
1829                struct commit_info ci;
1830                int num;
1831
1832                if (strcmp(suspect->path, sb->path))
1833                        *option |= OUTPUT_SHOW_NAME;
1834                num = strlen(suspect->path);
1835                if (longest_file < num)
1836                        longest_file = num;
1837                if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1838                        suspect->commit->object.flags |= METAINFO_SHOWN;
1839                        get_commit_info(suspect->commit, &ci, 1);
1840                        num = utf8_strwidth(ci.author);
1841                        if (longest_author < num)
1842                                longest_author = num;
1843                }
1844                num = e->s_lno + e->num_lines;
1845                if (longest_src_lines < num)
1846                        longest_src_lines = num;
1847                num = e->lno + e->num_lines;
1848                if (longest_dst_lines < num)
1849                        longest_dst_lines = num;
1850                if (largest_score < ent_score(sb, e))
1851                        largest_score = ent_score(sb, e);
1852        }
1853        max_orig_digits = lineno_width(longest_src_lines);
1854        max_digits = lineno_width(longest_dst_lines);
1855        max_score_digits = lineno_width(largest_score);
1856}
1857
1858/*
1859 * For debugging -- origin is refcounted, and this asserts that
1860 * we do not underflow.
1861 */
1862static void sanity_check_refcnt(struct scoreboard *sb)
1863{
1864        int baa = 0;
1865        struct blame_entry *ent;
1866
1867        for (ent = sb->ent; ent; ent = ent->next) {
1868                /* Nobody should have zero or negative refcnt */
1869                if (ent->suspect->refcnt <= 0) {
1870                        fprintf(stderr, "%s in %s has negative refcnt %d\n",
1871                                ent->suspect->path,
1872                                sha1_to_hex(ent->suspect->commit->object.sha1),
1873                                ent->suspect->refcnt);
1874                        baa = 1;
1875                }
1876        }
1877        if (baa) {
1878                int opt = 0160;
1879                find_alignment(sb, &opt);
1880                output(sb, opt);
1881                die("Baa %d!", baa);
1882        }
1883}
1884
1885/*
1886 * Used for the command line parsing; check if the path exists
1887 * in the working tree.
1888 */
1889static int has_string_in_work_tree(const char *path)
1890{
1891        struct stat st;
1892        return !lstat(path, &st);
1893}
1894
1895static unsigned parse_score(const char *arg)
1896{
1897        char *end;
1898        unsigned long score = strtoul(arg, &end, 10);
1899        if (*end)
1900                return 0;
1901        return score;
1902}
1903
1904static const char *add_prefix(const char *prefix, const char *path)
1905{
1906        return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1907}
1908
1909/*
1910 * Parsing of (comma separated) one item in the -L option
1911 */
1912static const char *parse_loc(const char *spec,
1913                             struct scoreboard *sb, long lno,
1914                             long begin, long *ret)
1915{
1916        char *term;
1917        const char *line;
1918        long num;
1919        int reg_error;
1920        regex_t regexp;
1921        regmatch_t match[1];
1922
1923        /* Allow "-L <something>,+20" to mean starting at <something>
1924         * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1925         * <something>.
1926         */
1927        if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1928                num = strtol(spec + 1, &term, 10);
1929                if (term != spec + 1) {
1930                        if (spec[0] == '-')
1931                                num = 0 - num;
1932                        if (0 < num)
1933                                *ret = begin + num - 2;
1934                        else if (!num)
1935                                *ret = begin;
1936                        else
1937                                *ret = begin + num;
1938                        return term;
1939                }
1940                return spec;
1941        }
1942        num = strtol(spec, &term, 10);
1943        if (term != spec) {
1944                *ret = num;
1945                return term;
1946        }
1947        if (spec[0] != '/')
1948                return spec;
1949
1950        /* it could be a regexp of form /.../ */
1951        for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1952                if (*term == '\\')
1953                        term++;
1954        }
1955        if (*term != '/')
1956                return spec;
1957
1958        /* try [spec+1 .. term-1] as regexp */
1959        *term = 0;
1960        begin--; /* input is in human terms */
1961        line = nth_line(sb, begin);
1962
1963        if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1964            !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1965                const char *cp = line + match[0].rm_so;
1966                const char *nline;
1967
1968                while (begin++ < lno) {
1969                        nline = nth_line(sb, begin);
1970                        if (line <= cp && cp < nline)
1971                                break;
1972                        line = nline;
1973                }
1974                *ret = begin;
1975                regfree(&regexp);
1976                *term++ = '/';
1977                return term;
1978        }
1979        else {
1980                char errbuf[1024];
1981                regerror(reg_error, &regexp, errbuf, 1024);
1982                die("-L parameter '%s': %s", spec + 1, errbuf);
1983        }
1984}
1985
1986/*
1987 * Parsing of -L option
1988 */
1989static void prepare_blame_range(struct scoreboard *sb,
1990                                const char *bottomtop,
1991                                long lno,
1992                                long *bottom, long *top)
1993{
1994        const char *term;
1995
1996        term = parse_loc(bottomtop, sb, lno, 1, bottom);
1997        if (*term == ',') {
1998                term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1999                if (*term)
2000                        usage(blame_usage);
2001        }
2002        if (*term)
2003                usage(blame_usage);
2004}
2005
2006static int git_blame_config(const char *var, const char *value, void *cb)
2007{
2008        if (!strcmp(var, "blame.showroot")) {
2009                show_root = git_config_bool(var, value);
2010                return 0;
2011        }
2012        if (!strcmp(var, "blame.blankboundary")) {
2013                blank_boundary = git_config_bool(var, value);
2014                return 0;
2015        }
2016        if (!strcmp(var, "blame.date")) {
2017                if (!value)
2018                        return config_error_nonbool(var);
2019                blame_date_mode = parse_date_format(value);
2020                return 0;
2021        }
2022
2023        switch (userdiff_config(var, value)) {
2024        case 0:
2025                break;
2026        case -1:
2027                return -1;
2028        default:
2029                return 0;
2030        }
2031
2032        return git_default_config(var, value, cb);
2033}
2034
2035/*
2036 * Prepare a dummy commit that represents the work tree (or staged) item.
2037 * Note that annotating work tree item never works in the reverse.
2038 */
2039static struct commit *fake_working_tree_commit(struct diff_options *opt,
2040                                               const char *path,
2041                                               const char *contents_from)
2042{
2043        struct commit *commit;
2044        struct origin *origin;
2045        unsigned char head_sha1[20];
2046        struct strbuf buf = STRBUF_INIT;
2047        const char *ident;
2048        time_t now;
2049        int size, len;
2050        struct cache_entry *ce;
2051        unsigned mode;
2052
2053        if (get_sha1("HEAD", head_sha1))
2054                die("No such ref: HEAD");
2055
2056        time(&now);
2057        commit = xcalloc(1, sizeof(*commit));
2058        commit->parents = xcalloc(1, sizeof(*commit->parents));
2059        commit->parents->item = lookup_commit_reference(head_sha1);
2060        commit->object.parsed = 1;
2061        commit->date = now;
2062        commit->object.type = OBJ_COMMIT;
2063
2064        origin = make_origin(commit, path);
2065
2066        if (!contents_from || strcmp("-", contents_from)) {
2067                struct stat st;
2068                const char *read_from;
2069                unsigned long buf_len;
2070
2071                if (contents_from) {
2072                        if (stat(contents_from, &st) < 0)
2073                                die_errno("Cannot stat '%s'", contents_from);
2074                        read_from = contents_from;
2075                }
2076                else {
2077                        if (lstat(path, &st) < 0)
2078                                die_errno("Cannot lstat '%s'", path);
2079                        read_from = path;
2080                }
2081                mode = canon_mode(st.st_mode);
2082
2083                switch (st.st_mode & S_IFMT) {
2084                case S_IFREG:
2085                        if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2086                            textconv_object(read_from, null_sha1, &buf.buf, &buf_len))
2087                                buf.len = buf_len;
2088                        else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2089                                die_errno("cannot open or read '%s'", read_from);
2090                        break;
2091                case S_IFLNK:
2092                        if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2093                                die_errno("cannot readlink '%s'", read_from);
2094                        break;
2095                default:
2096                        die("unsupported file type %s", read_from);
2097                }
2098        }
2099        else {
2100                /* Reading from stdin */
2101                contents_from = "standard input";
2102                mode = 0;
2103                if (strbuf_read(&buf, 0, 0) < 0)
2104                        die_errno("failed to read from stdin");
2105        }
2106        convert_to_git(path, buf.buf, buf.len, &buf, 0);
2107        origin->file.ptr = buf.buf;
2108        origin->file.size = buf.len;
2109        pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2110        commit->util = origin;
2111
2112        /*
2113         * Read the current index, replace the path entry with
2114         * origin->blob_sha1 without mucking with its mode or type
2115         * bits; we are not going to write this index out -- we just
2116         * want to run "diff-index --cached".
2117         */
2118        discard_cache();
2119        read_cache();
2120
2121        len = strlen(path);
2122        if (!mode) {
2123                int pos = cache_name_pos(path, len);
2124                if (0 <= pos)
2125                        mode = active_cache[pos]->ce_mode;
2126                else
2127                        /* Let's not bother reading from HEAD tree */
2128                        mode = S_IFREG | 0644;
2129        }
2130        size = cache_entry_size(len);
2131        ce = xcalloc(1, size);
2132        hashcpy(ce->sha1, origin->blob_sha1);
2133        memcpy(ce->name, path, len);
2134        ce->ce_flags = create_ce_flags(len, 0);
2135        ce->ce_mode = create_ce_mode(mode);
2136        add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2137
2138        /*
2139         * We are not going to write this out, so this does not matter
2140         * right now, but someday we might optimize diff-index --cached
2141         * with cache-tree information.
2142         */
2143        cache_tree_invalidate_path(active_cache_tree, path);
2144
2145        commit->buffer = xmalloc(400);
2146        ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2147        snprintf(commit->buffer, 400,
2148                "tree 0000000000000000000000000000000000000000\n"
2149                "parent %s\n"
2150                "author %s\n"
2151                "committer %s\n\n"
2152                "Version of %s from %s\n",
2153                sha1_to_hex(head_sha1),
2154                ident, ident, path, contents_from ? contents_from : path);
2155        return commit;
2156}
2157
2158static const char *prepare_final(struct scoreboard *sb)
2159{
2160        int i;
2161        const char *final_commit_name = NULL;
2162        struct rev_info *revs = sb->revs;
2163
2164        /*
2165         * There must be one and only one positive commit in the
2166         * revs->pending array.
2167         */
2168        for (i = 0; i < revs->pending.nr; i++) {
2169                struct object *obj = revs->pending.objects[i].item;
2170                if (obj->flags & UNINTERESTING)
2171                        continue;
2172                while (obj->type == OBJ_TAG)
2173                        obj = deref_tag(obj, NULL, 0);
2174                if (obj->type != OBJ_COMMIT)
2175                        die("Non commit %s?", revs->pending.objects[i].name);
2176                if (sb->final)
2177                        die("More than one commit to dig from %s and %s?",
2178                            revs->pending.objects[i].name,
2179                            final_commit_name);
2180                sb->final = (struct commit *) obj;
2181                final_commit_name = revs->pending.objects[i].name;
2182        }
2183        return final_commit_name;
2184}
2185
2186static const char *prepare_initial(struct scoreboard *sb)
2187{
2188        int i;
2189        const char *final_commit_name = NULL;
2190        struct rev_info *revs = sb->revs;
2191
2192        /*
2193         * There must be one and only one negative commit, and it must be
2194         * the boundary.
2195         */
2196        for (i = 0; i < revs->pending.nr; i++) {
2197                struct object *obj = revs->pending.objects[i].item;
2198                if (!(obj->flags & UNINTERESTING))
2199                        continue;
2200                while (obj->type == OBJ_TAG)
2201                        obj = deref_tag(obj, NULL, 0);
2202                if (obj->type != OBJ_COMMIT)
2203                        die("Non commit %s?", revs->pending.objects[i].name);
2204                if (sb->final)
2205                        die("More than one commit to dig down to %s and %s?",
2206                            revs->pending.objects[i].name,
2207                            final_commit_name);
2208                sb->final = (struct commit *) obj;
2209                final_commit_name = revs->pending.objects[i].name;
2210        }
2211        if (!final_commit_name)
2212                die("No commit to dig down to?");
2213        return final_commit_name;
2214}
2215
2216static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2217{
2218        int *opt = option->value;
2219
2220        /*
2221         * -C enables copy from removed files;
2222         * -C -C enables copy from existing files, but only
2223         *       when blaming a new file;
2224         * -C -C -C enables copy from existing files for
2225         *          everybody
2226         */
2227        if (*opt & PICKAXE_BLAME_COPY_HARDER)
2228                *opt |= PICKAXE_BLAME_COPY_HARDEST;
2229        if (*opt & PICKAXE_BLAME_COPY)
2230                *opt |= PICKAXE_BLAME_COPY_HARDER;
2231        *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2232
2233        if (arg)
2234                blame_copy_score = parse_score(arg);
2235        return 0;
2236}
2237
2238static int blame_move_callback(const struct option *option, const char *arg, int unset)
2239{
2240        int *opt = option->value;
2241
2242        *opt |= PICKAXE_BLAME_MOVE;
2243
2244        if (arg)
2245                blame_move_score = parse_score(arg);
2246        return 0;
2247}
2248
2249static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2250{
2251        const char **bottomtop = option->value;
2252        if (!arg)
2253                return -1;
2254        if (*bottomtop)
2255                die("More than one '-L n,m' option given");
2256        *bottomtop = arg;
2257        return 0;
2258}
2259
2260int cmd_blame(int argc, const char **argv, const char *prefix)
2261{
2262        struct rev_info revs;
2263        const char *path;
2264        struct scoreboard sb;
2265        struct origin *o;
2266        struct blame_entry *ent;
2267        long dashdash_pos, bottom, top, lno;
2268        const char *final_commit_name = NULL;
2269        enum object_type type;
2270
2271        static const char *bottomtop = NULL;
2272        static int output_option = 0, opt = 0;
2273        static int show_stats = 0;
2274        static const char *revs_file = NULL;
2275        static const char *contents_from = NULL;
2276        static const struct option options[] = {
2277                OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2278                OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2279                OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2280                OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2281                OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2282                OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2283                OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2284                OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2285                OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2286                OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2287                OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2288                OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2289                OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2290                OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2291                OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2292                { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2293                { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2294                OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2295                OPT_END()
2296        };
2297
2298        struct parse_opt_ctx_t ctx;
2299        int cmd_is_annotate = !strcmp(argv[0], "annotate");
2300
2301        git_config(git_blame_config, NULL);
2302        init_revisions(&revs, NULL);
2303        revs.date_mode = blame_date_mode;
2304        DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2305
2306        save_commit_buffer = 0;
2307        dashdash_pos = 0;
2308
2309        parse_options_start(&ctx, argc, argv, prefix, PARSE_OPT_KEEP_DASHDASH |
2310                            PARSE_OPT_KEEP_ARGV0);
2311        for (;;) {
2312                switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2313                case PARSE_OPT_HELP:
2314                        exit(129);
2315                case PARSE_OPT_DONE:
2316                        if (ctx.argv[0])
2317                                dashdash_pos = ctx.cpidx;
2318                        goto parse_done;
2319                }
2320
2321                if (!strcmp(ctx.argv[0], "--reverse")) {
2322                        ctx.argv[0] = "--children";
2323                        reverse = 1;
2324                }
2325                parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2326        }
2327parse_done:
2328        argc = parse_options_end(&ctx);
2329
2330        if (revs_file && read_ancestry(revs_file))
2331                die_errno("reading graft file '%s' failed", revs_file);
2332
2333        if (cmd_is_annotate) {
2334                output_option |= OUTPUT_ANNOTATE_COMPAT;
2335                blame_date_mode = DATE_ISO8601;
2336        } else {
2337                blame_date_mode = revs.date_mode;
2338        }
2339
2340        /* The maximum width used to show the dates */
2341        switch (blame_date_mode) {
2342        case DATE_RFC2822:
2343                blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2344                break;
2345        case DATE_ISO8601:
2346                blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2347                break;
2348        case DATE_RAW:
2349                blame_date_width = sizeof("1161298804 -0700");
2350                break;
2351        case DATE_SHORT:
2352                blame_date_width = sizeof("2006-10-19");
2353                break;
2354        case DATE_RELATIVE:
2355                /* "normal" is used as the fallback for "relative" */
2356        case DATE_LOCAL:
2357        case DATE_NORMAL:
2358                blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2359                break;
2360        }
2361        blame_date_width -= 1; /* strip the null */
2362
2363        if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2364                opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2365                        PICKAXE_BLAME_COPY_HARDER);
2366
2367        if (!blame_move_score)
2368                blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2369        if (!blame_copy_score)
2370                blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2371
2372        /*
2373         * We have collected options unknown to us in argv[1..unk]
2374         * which are to be passed to revision machinery if we are
2375         * going to do the "bottom" processing.
2376         *
2377         * The remaining are:
2378         *
2379         * (1) if dashdash_pos != 0, its either
2380         *     "blame [revisions] -- <path>" or
2381         *     "blame -- <path> <rev>"
2382         *
2383         * (2) otherwise, its one of the two:
2384         *     "blame [revisions] <path>"
2385         *     "blame <path> <rev>"
2386         *
2387         * Note that we must strip out <path> from the arguments: we do not
2388         * want the path pruning but we may want "bottom" processing.
2389         */
2390        if (dashdash_pos) {
2391                switch (argc - dashdash_pos - 1) {
2392                case 2: /* (1b) */
2393                        if (argc != 4)
2394                                usage_with_options(blame_opt_usage, options);
2395                        /* reorder for the new way: <rev> -- <path> */
2396                        argv[1] = argv[3];
2397                        argv[3] = argv[2];
2398                        argv[2] = "--";
2399                        /* FALLTHROUGH */
2400                case 1: /* (1a) */
2401                        path = add_prefix(prefix, argv[--argc]);
2402                        argv[argc] = NULL;
2403                        break;
2404                default:
2405                        usage_with_options(blame_opt_usage, options);
2406                }
2407        } else {
2408                if (argc < 2)
2409                        usage_with_options(blame_opt_usage, options);
2410                path = add_prefix(prefix, argv[argc - 1]);
2411                if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2412                        path = add_prefix(prefix, argv[1]);
2413                        argv[1] = argv[2];
2414                }
2415                argv[argc - 1] = "--";
2416
2417                setup_work_tree();
2418                if (!has_string_in_work_tree(path))
2419                        die_errno("cannot stat path '%s'", path);
2420        }
2421
2422        revs.disable_stdin = 1;
2423        setup_revisions(argc, argv, &revs, NULL);
2424        memset(&sb, 0, sizeof(sb));
2425
2426        sb.revs = &revs;
2427        if (!reverse)
2428                final_commit_name = prepare_final(&sb);
2429        else if (contents_from)
2430                die("--contents and --children do not blend well.");
2431        else
2432                final_commit_name = prepare_initial(&sb);
2433
2434        if (!sb.final) {
2435                /*
2436                 * "--not A B -- path" without anything positive;
2437                 * do not default to HEAD, but use the working tree
2438                 * or "--contents".
2439                 */
2440                setup_work_tree();
2441                sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2442                                                    path, contents_from);
2443                add_pending_object(&revs, &(sb.final->object), ":");
2444        }
2445        else if (contents_from)
2446                die("Cannot use --contents with final commit object name");
2447
2448        /*
2449         * If we have bottom, this will mark the ancestors of the
2450         * bottom commits we would reach while traversing as
2451         * uninteresting.
2452         */
2453        if (prepare_revision_walk(&revs))
2454                die("revision walk setup failed");
2455
2456        if (is_null_sha1(sb.final->object.sha1)) {
2457                char *buf;
2458                o = sb.final->util;
2459                buf = xmalloc(o->file.size + 1);
2460                memcpy(buf, o->file.ptr, o->file.size + 1);
2461                sb.final_buf = buf;
2462                sb.final_buf_size = o->file.size;
2463        }
2464        else {
2465                o = get_origin(&sb, sb.final, path);
2466                if (fill_blob_sha1(o))
2467                        die("no such path %s in %s", path, final_commit_name);
2468
2469                if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2470                    textconv_object(path, o->blob_sha1, (char **) &sb.final_buf,
2471                                    &sb.final_buf_size))
2472                        ;
2473                else
2474                        sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2475                                                      &sb.final_buf_size);
2476
2477                if (!sb.final_buf)
2478                        die("Cannot read blob %s for path %s",
2479                            sha1_to_hex(o->blob_sha1),
2480                            path);
2481        }
2482        num_read_blob++;
2483        lno = prepare_lines(&sb);
2484
2485        bottom = top = 0;
2486        if (bottomtop)
2487                prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2488        if (bottom && top && top < bottom) {
2489                long tmp;
2490                tmp = top; top = bottom; bottom = tmp;
2491        }
2492        if (bottom < 1)
2493                bottom = 1;
2494        if (top < 1)
2495                top = lno;
2496        bottom--;
2497        if (lno < top || lno < bottom)
2498                die("file %s has only %lu lines", path, lno);
2499
2500        ent = xcalloc(1, sizeof(*ent));
2501        ent->lno = bottom;
2502        ent->num_lines = top - bottom;
2503        ent->suspect = o;
2504        ent->s_lno = bottom;
2505
2506        sb.ent = ent;
2507        sb.path = path;
2508
2509        read_mailmap(&mailmap, NULL);
2510
2511        if (!incremental)
2512                setup_pager();
2513
2514        assign_blame(&sb, opt);
2515
2516        if (incremental)
2517                return 0;
2518
2519        coalesce(&sb);
2520
2521        if (!(output_option & OUTPUT_PORCELAIN))
2522                find_alignment(&sb, &output_option);
2523
2524        output(&sb, output_option);
2525        free((void *)sb.final_buf);
2526        for (ent = sb.ent; ent; ) {
2527                struct blame_entry *e = ent->next;
2528                free(ent);
2529                ent = e;
2530        }
2531
2532        if (show_stats) {
2533                printf("num read blob: %d\n", num_read_blob);
2534                printf("num get patch: %d\n", num_get_patch);
2535                printf("num commits: %d\n", num_commits);
2536        }
2537        return 0;
2538}