builtin / blame.con commit Merge branch 'np/lookup-object-hashing' (3fb9d68)
   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#include "line-range.h"
  25#include "line-log.h"
  26
  27static char blame_usage[] = N_("git blame [options] [rev-opts] [rev] [--] file");
  28
  29static const char *blame_opt_usage[] = {
  30        blame_usage,
  31        "",
  32        N_("[rev-opts] are documented in git-rev-list(1)"),
  33        NULL
  34};
  35
  36static int longest_file;
  37static int longest_author;
  38static int max_orig_digits;
  39static int max_digits;
  40static int max_score_digits;
  41static int show_root;
  42static int reverse;
  43static int blank_boundary;
  44static int incremental;
  45static int xdl_opts;
  46static int abbrev = -1;
  47static int no_whole_file_rename;
  48
  49static enum date_mode blame_date_mode = DATE_ISO8601;
  50static size_t blame_date_width;
  51
  52static struct string_list mailmap;
  53
  54#ifndef DEBUG
  55#define DEBUG 0
  56#endif
  57
  58/* stats */
  59static int num_read_blob;
  60static int num_get_patch;
  61static int num_commits;
  62
  63#define PICKAXE_BLAME_MOVE              01
  64#define PICKAXE_BLAME_COPY              02
  65#define PICKAXE_BLAME_COPY_HARDER       04
  66#define PICKAXE_BLAME_COPY_HARDEST      010
  67
  68/*
  69 * blame for a blame_entry with score lower than these thresholds
  70 * is not passed to the parent using move/copy logic.
  71 */
  72static unsigned blame_move_score;
  73static unsigned blame_copy_score;
  74#define BLAME_DEFAULT_MOVE_SCORE        20
  75#define BLAME_DEFAULT_COPY_SCORE        40
  76
  77/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
  78#define METAINFO_SHOWN          (1u<<12)
  79#define MORE_THAN_ONE_PATH      (1u<<13)
  80
  81/*
  82 * One blob in a commit that is being suspected
  83 */
  84struct origin {
  85        int refcnt;
  86        struct origin *previous;
  87        struct commit *commit;
  88        mmfile_t file;
  89        unsigned char blob_sha1[20];
  90        unsigned mode;
  91        char path[FLEX_ARRAY];
  92};
  93
  94static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
  95                      xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
  96{
  97        xpparam_t xpp = {0};
  98        xdemitconf_t xecfg = {0};
  99        xdemitcb_t ecb = {NULL};
 100
 101        xpp.flags = xdl_opts;
 102        xecfg.ctxlen = ctxlen;
 103        xecfg.hunk_func = hunk_func;
 104        ecb.priv = cb_data;
 105        return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
 106}
 107
 108/*
 109 * Prepare diff_filespec and convert it using diff textconv API
 110 * if the textconv driver exists.
 111 * Return 1 if the conversion succeeds, 0 otherwise.
 112 */
 113int textconv_object(const char *path,
 114                    unsigned mode,
 115                    const unsigned char *sha1,
 116                    int sha1_valid,
 117                    char **buf,
 118                    unsigned long *buf_size)
 119{
 120        struct diff_filespec *df;
 121        struct userdiff_driver *textconv;
 122
 123        df = alloc_filespec(path);
 124        fill_filespec(df, sha1, sha1_valid, mode);
 125        textconv = get_textconv(df);
 126        if (!textconv) {
 127                free_filespec(df);
 128                return 0;
 129        }
 130
 131        *buf_size = fill_textconv(textconv, df, buf);
 132        free_filespec(df);
 133        return 1;
 134}
 135
 136/*
 137 * Given an origin, prepare mmfile_t structure to be used by the
 138 * diff machinery
 139 */
 140static void fill_origin_blob(struct diff_options *opt,
 141                             struct origin *o, mmfile_t *file)
 142{
 143        if (!o->file.ptr) {
 144                enum object_type type;
 145                unsigned long file_size;
 146
 147                num_read_blob++;
 148                if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
 149                    textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
 150                        ;
 151                else
 152                        file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
 153                file->size = file_size;
 154
 155                if (!file->ptr)
 156                        die("Cannot read blob %s for path %s",
 157                            sha1_to_hex(o->blob_sha1),
 158                            o->path);
 159                o->file = *file;
 160        }
 161        else
 162                *file = o->file;
 163}
 164
 165/*
 166 * Origin is refcounted and usually we keep the blob contents to be
 167 * reused.
 168 */
 169static inline struct origin *origin_incref(struct origin *o)
 170{
 171        if (o)
 172                o->refcnt++;
 173        return o;
 174}
 175
 176static void origin_decref(struct origin *o)
 177{
 178        if (o && --o->refcnt <= 0) {
 179                if (o->previous)
 180                        origin_decref(o->previous);
 181                free(o->file.ptr);
 182                free(o);
 183        }
 184}
 185
 186static void drop_origin_blob(struct origin *o)
 187{
 188        if (o->file.ptr) {
 189                free(o->file.ptr);
 190                o->file.ptr = NULL;
 191        }
 192}
 193
 194/*
 195 * Each group of lines is described by a blame_entry; it can be split
 196 * as we pass blame to the parents.  They form a linked list in the
 197 * scoreboard structure, sorted by the target line number.
 198 */
 199struct blame_entry {
 200        struct blame_entry *prev;
 201        struct blame_entry *next;
 202
 203        /* the first line of this group in the final image;
 204         * internally all line numbers are 0 based.
 205         */
 206        int lno;
 207
 208        /* how many lines this group has */
 209        int num_lines;
 210
 211        /* the commit that introduced this group into the final image */
 212        struct origin *suspect;
 213
 214        /* true if the suspect is truly guilty; false while we have not
 215         * checked if the group came from one of its parents.
 216         */
 217        char guilty;
 218
 219        /* true if the entry has been scanned for copies in the current parent
 220         */
 221        char scanned;
 222
 223        /* the line number of the first line of this group in the
 224         * suspect's file; internally all line numbers are 0 based.
 225         */
 226        int s_lno;
 227
 228        /* how significant this entry is -- cached to avoid
 229         * scanning the lines over and over.
 230         */
 231        unsigned score;
 232};
 233
 234/*
 235 * The current state of the blame assignment.
 236 */
 237struct scoreboard {
 238        /* the final commit (i.e. where we started digging from) */
 239        struct commit *final;
 240        struct rev_info *revs;
 241        const char *path;
 242
 243        /*
 244         * The contents in the final image.
 245         * Used by many functions to obtain contents of the nth line,
 246         * indexed with scoreboard.lineno[blame_entry.lno].
 247         */
 248        const char *final_buf;
 249        unsigned long final_buf_size;
 250
 251        /* linked list of blames */
 252        struct blame_entry *ent;
 253
 254        /* look-up a line in the final buffer */
 255        int num_lines;
 256        int *lineno;
 257};
 258
 259static inline int same_suspect(struct origin *a, struct origin *b)
 260{
 261        if (a == b)
 262                return 1;
 263        if (a->commit != b->commit)
 264                return 0;
 265        return !strcmp(a->path, b->path);
 266}
 267
 268static void sanity_check_refcnt(struct scoreboard *);
 269
 270/*
 271 * If two blame entries that are next to each other came from
 272 * contiguous lines in the same origin (i.e. <commit, path> pair),
 273 * merge them together.
 274 */
 275static void coalesce(struct scoreboard *sb)
 276{
 277        struct blame_entry *ent, *next;
 278
 279        for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 280                if (same_suspect(ent->suspect, next->suspect) &&
 281                    ent->guilty == next->guilty &&
 282                    ent->s_lno + ent->num_lines == next->s_lno) {
 283                        ent->num_lines += next->num_lines;
 284                        ent->next = next->next;
 285                        if (ent->next)
 286                                ent->next->prev = ent;
 287                        origin_decref(next->suspect);
 288                        free(next);
 289                        ent->score = 0;
 290                        next = ent; /* again */
 291                }
 292        }
 293
 294        if (DEBUG) /* sanity */
 295                sanity_check_refcnt(sb);
 296}
 297
 298/*
 299 * Given a commit and a path in it, create a new origin structure.
 300 * The callers that add blame to the scoreboard should use
 301 * get_origin() to obtain shared, refcounted copy instead of calling
 302 * this function directly.
 303 */
 304static struct origin *make_origin(struct commit *commit, const char *path)
 305{
 306        struct origin *o;
 307        o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 308        o->commit = commit;
 309        o->refcnt = 1;
 310        strcpy(o->path, path);
 311        return o;
 312}
 313
 314/*
 315 * Locate an existing origin or create a new one.
 316 */
 317static struct origin *get_origin(struct scoreboard *sb,
 318                                 struct commit *commit,
 319                                 const char *path)
 320{
 321        struct blame_entry *e;
 322
 323        for (e = sb->ent; e; e = e->next) {
 324                if (e->suspect->commit == commit &&
 325                    !strcmp(e->suspect->path, path))
 326                        return origin_incref(e->suspect);
 327        }
 328        return make_origin(commit, path);
 329}
 330
 331/*
 332 * Fill the blob_sha1 field of an origin if it hasn't, so that later
 333 * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 334 * for an origin is also used to pass the blame for the entire file to
 335 * the parent to detect the case where a child's blob is identical to
 336 * that of its parent's.
 337 *
 338 * This also fills origin->mode for corresponding tree path.
 339 */
 340static int fill_blob_sha1_and_mode(struct origin *origin)
 341{
 342        if (!is_null_sha1(origin->blob_sha1))
 343                return 0;
 344        if (get_tree_entry(origin->commit->object.sha1,
 345                           origin->path,
 346                           origin->blob_sha1, &origin->mode))
 347                goto error_out;
 348        if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 349                goto error_out;
 350        return 0;
 351 error_out:
 352        hashclr(origin->blob_sha1);
 353        origin->mode = S_IFINVALID;
 354        return -1;
 355}
 356
 357/*
 358 * We have an origin -- check if the same path exists in the
 359 * parent and return an origin structure to represent it.
 360 */
 361static struct origin *find_origin(struct scoreboard *sb,
 362                                  struct commit *parent,
 363                                  struct origin *origin)
 364{
 365        struct origin *porigin = NULL;
 366        struct diff_options diff_opts;
 367        const char *paths[2];
 368
 369        if (parent->util) {
 370                /*
 371                 * Each commit object can cache one origin in that
 372                 * commit.  This is a freestanding copy of origin and
 373                 * not refcounted.
 374                 */
 375                struct origin *cached = parent->util;
 376                if (!strcmp(cached->path, origin->path)) {
 377                        /*
 378                         * The same path between origin and its parent
 379                         * without renaming -- the most common case.
 380                         */
 381                        porigin = get_origin(sb, parent, cached->path);
 382
 383                        /*
 384                         * If the origin was newly created (i.e. get_origin
 385                         * would call make_origin if none is found in the
 386                         * scoreboard), it does not know the blob_sha1/mode,
 387                         * so copy it.  Otherwise porigin was in the
 388                         * scoreboard and already knows blob_sha1/mode.
 389                         */
 390                        if (porigin->refcnt == 1) {
 391                                hashcpy(porigin->blob_sha1, cached->blob_sha1);
 392                                porigin->mode = cached->mode;
 393                        }
 394                        return porigin;
 395                }
 396                /* otherwise it was not very useful; free it */
 397                free(parent->util);
 398                parent->util = NULL;
 399        }
 400
 401        /* See if the origin->path is different between parent
 402         * and origin first.  Most of the time they are the
 403         * same and diff-tree is fairly efficient about this.
 404         */
 405        diff_setup(&diff_opts);
 406        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 407        diff_opts.detect_rename = 0;
 408        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 409        paths[0] = origin->path;
 410        paths[1] = NULL;
 411
 412        parse_pathspec(&diff_opts.pathspec, PATHSPEC_ALL_MAGIC, 0, "", paths);
 413        diff_setup_done(&diff_opts);
 414
 415        if (is_null_sha1(origin->commit->object.sha1))
 416                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 417        else
 418                diff_tree_sha1(parent->tree->object.sha1,
 419                               origin->commit->tree->object.sha1,
 420                               "", &diff_opts);
 421        diffcore_std(&diff_opts);
 422
 423        if (!diff_queued_diff.nr) {
 424                /* The path is the same as parent */
 425                porigin = get_origin(sb, parent, origin->path);
 426                hashcpy(porigin->blob_sha1, origin->blob_sha1);
 427                porigin->mode = origin->mode;
 428        } else {
 429                /*
 430                 * Since origin->path is a pathspec, if the parent
 431                 * commit had it as a directory, we will see a whole
 432                 * bunch of deletion of files in the directory that we
 433                 * do not care about.
 434                 */
 435                int i;
 436                struct diff_filepair *p = NULL;
 437                for (i = 0; i < diff_queued_diff.nr; i++) {
 438                        const char *name;
 439                        p = diff_queued_diff.queue[i];
 440                        name = p->one->path ? p->one->path : p->two->path;
 441                        if (!strcmp(name, origin->path))
 442                                break;
 443                }
 444                if (!p)
 445                        die("internal error in blame::find_origin");
 446                switch (p->status) {
 447                default:
 448                        die("internal error in blame::find_origin (%c)",
 449                            p->status);
 450                case 'M':
 451                        porigin = get_origin(sb, parent, origin->path);
 452                        hashcpy(porigin->blob_sha1, p->one->sha1);
 453                        porigin->mode = p->one->mode;
 454                        break;
 455                case 'A':
 456                case 'T':
 457                        /* Did not exist in parent, or type changed */
 458                        break;
 459                }
 460        }
 461        diff_flush(&diff_opts);
 462        free_pathspec(&diff_opts.pathspec);
 463        if (porigin) {
 464                /*
 465                 * Create a freestanding copy that is not part of
 466                 * the refcounted origin found in the scoreboard, and
 467                 * cache it in the commit.
 468                 */
 469                struct origin *cached;
 470
 471                cached = make_origin(porigin->commit, porigin->path);
 472                hashcpy(cached->blob_sha1, porigin->blob_sha1);
 473                cached->mode = porigin->mode;
 474                parent->util = cached;
 475        }
 476        return porigin;
 477}
 478
 479/*
 480 * We have an origin -- find the path that corresponds to it in its
 481 * parent and return an origin structure to represent it.
 482 */
 483static struct origin *find_rename(struct scoreboard *sb,
 484                                  struct commit *parent,
 485                                  struct origin *origin)
 486{
 487        struct origin *porigin = NULL;
 488        struct diff_options diff_opts;
 489        int i;
 490
 491        diff_setup(&diff_opts);
 492        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 493        diff_opts.detect_rename = DIFF_DETECT_RENAME;
 494        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 495        diff_opts.single_follow = origin->path;
 496        diff_setup_done(&diff_opts);
 497
 498        if (is_null_sha1(origin->commit->object.sha1))
 499                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 500        else
 501                diff_tree_sha1(parent->tree->object.sha1,
 502                               origin->commit->tree->object.sha1,
 503                               "", &diff_opts);
 504        diffcore_std(&diff_opts);
 505
 506        for (i = 0; i < diff_queued_diff.nr; i++) {
 507                struct diff_filepair *p = diff_queued_diff.queue[i];
 508                if ((p->status == 'R' || p->status == 'C') &&
 509                    !strcmp(p->two->path, origin->path)) {
 510                        porigin = get_origin(sb, parent, p->one->path);
 511                        hashcpy(porigin->blob_sha1, p->one->sha1);
 512                        porigin->mode = p->one->mode;
 513                        break;
 514                }
 515        }
 516        diff_flush(&diff_opts);
 517        free_pathspec(&diff_opts.pathspec);
 518        return porigin;
 519}
 520
 521/*
 522 * Link in a new blame entry to the scoreboard.  Entries that cover the
 523 * same line range have been removed from the scoreboard previously.
 524 */
 525static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 526{
 527        struct blame_entry *ent, *prev = NULL;
 528
 529        origin_incref(e->suspect);
 530
 531        for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 532                prev = ent;
 533
 534        /* prev, if not NULL, is the last one that is below e */
 535        e->prev = prev;
 536        if (prev) {
 537                e->next = prev->next;
 538                prev->next = e;
 539        }
 540        else {
 541                e->next = sb->ent;
 542                sb->ent = e;
 543        }
 544        if (e->next)
 545                e->next->prev = e;
 546}
 547
 548/*
 549 * src typically is on-stack; we want to copy the information in it to
 550 * a malloced blame_entry that is already on the linked list of the
 551 * scoreboard.  The origin of dst loses a refcnt while the origin of src
 552 * gains one.
 553 */
 554static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 555{
 556        struct blame_entry *p, *n;
 557
 558        p = dst->prev;
 559        n = dst->next;
 560        origin_incref(src->suspect);
 561        origin_decref(dst->suspect);
 562        memcpy(dst, src, sizeof(*src));
 563        dst->prev = p;
 564        dst->next = n;
 565        dst->score = 0;
 566}
 567
 568static const char *nth_line(struct scoreboard *sb, long lno)
 569{
 570        return sb->final_buf + sb->lineno[lno];
 571}
 572
 573static const char *nth_line_cb(void *data, long lno)
 574{
 575        return nth_line((struct scoreboard *)data, lno);
 576}
 577
 578/*
 579 * It is known that lines between tlno to same came from parent, and e
 580 * has an overlap with that range.  it also is known that parent's
 581 * line plno corresponds to e's line tlno.
 582 *
 583 *                <---- e ----->
 584 *                   <------>
 585 *                   <------------>
 586 *             <------------>
 587 *             <------------------>
 588 *
 589 * Split e into potentially three parts; before this chunk, the chunk
 590 * to be blamed for the parent, and after that portion.
 591 */
 592static void split_overlap(struct blame_entry *split,
 593                          struct blame_entry *e,
 594                          int tlno, int plno, int same,
 595                          struct origin *parent)
 596{
 597        int chunk_end_lno;
 598        memset(split, 0, sizeof(struct blame_entry [3]));
 599
 600        if (e->s_lno < tlno) {
 601                /* there is a pre-chunk part not blamed on parent */
 602                split[0].suspect = origin_incref(e->suspect);
 603                split[0].lno = e->lno;
 604                split[0].s_lno = e->s_lno;
 605                split[0].num_lines = tlno - e->s_lno;
 606                split[1].lno = e->lno + tlno - e->s_lno;
 607                split[1].s_lno = plno;
 608        }
 609        else {
 610                split[1].lno = e->lno;
 611                split[1].s_lno = plno + (e->s_lno - tlno);
 612        }
 613
 614        if (same < e->s_lno + e->num_lines) {
 615                /* there is a post-chunk part not blamed on parent */
 616                split[2].suspect = origin_incref(e->suspect);
 617                split[2].lno = e->lno + (same - e->s_lno);
 618                split[2].s_lno = e->s_lno + (same - e->s_lno);
 619                split[2].num_lines = e->s_lno + e->num_lines - same;
 620                chunk_end_lno = split[2].lno;
 621        }
 622        else
 623                chunk_end_lno = e->lno + e->num_lines;
 624        split[1].num_lines = chunk_end_lno - split[1].lno;
 625
 626        /*
 627         * if it turns out there is nothing to blame the parent for,
 628         * forget about the splitting.  !split[1].suspect signals this.
 629         */
 630        if (split[1].num_lines < 1)
 631                return;
 632        split[1].suspect = origin_incref(parent);
 633}
 634
 635/*
 636 * split_overlap() divided an existing blame e into up to three parts
 637 * in split.  Adjust the linked list of blames in the scoreboard to
 638 * reflect the split.
 639 */
 640static void split_blame(struct scoreboard *sb,
 641                        struct blame_entry *split,
 642                        struct blame_entry *e)
 643{
 644        struct blame_entry *new_entry;
 645
 646        if (split[0].suspect && split[2].suspect) {
 647                /* The first part (reuse storage for the existing entry e) */
 648                dup_entry(e, &split[0]);
 649
 650                /* The last part -- me */
 651                new_entry = xmalloc(sizeof(*new_entry));
 652                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 653                add_blame_entry(sb, new_entry);
 654
 655                /* ... and the middle part -- parent */
 656                new_entry = xmalloc(sizeof(*new_entry));
 657                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 658                add_blame_entry(sb, new_entry);
 659        }
 660        else if (!split[0].suspect && !split[2].suspect)
 661                /*
 662                 * The parent covers the entire area; reuse storage for
 663                 * e and replace it with the parent.
 664                 */
 665                dup_entry(e, &split[1]);
 666        else if (split[0].suspect) {
 667                /* me and then parent */
 668                dup_entry(e, &split[0]);
 669
 670                new_entry = xmalloc(sizeof(*new_entry));
 671                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 672                add_blame_entry(sb, new_entry);
 673        }
 674        else {
 675                /* parent and then me */
 676                dup_entry(e, &split[1]);
 677
 678                new_entry = xmalloc(sizeof(*new_entry));
 679                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 680                add_blame_entry(sb, new_entry);
 681        }
 682
 683        if (DEBUG) { /* sanity */
 684                struct blame_entry *ent;
 685                int lno = sb->ent->lno, corrupt = 0;
 686
 687                for (ent = sb->ent; ent; ent = ent->next) {
 688                        if (lno != ent->lno)
 689                                corrupt = 1;
 690                        if (ent->s_lno < 0)
 691                                corrupt = 1;
 692                        lno += ent->num_lines;
 693                }
 694                if (corrupt) {
 695                        lno = sb->ent->lno;
 696                        for (ent = sb->ent; ent; ent = ent->next) {
 697                                printf("L %8d l %8d n %8d\n",
 698                                       lno, ent->lno, ent->num_lines);
 699                                lno = ent->lno + ent->num_lines;
 700                        }
 701                        die("oops");
 702                }
 703        }
 704}
 705
 706/*
 707 * After splitting the blame, the origins used by the
 708 * on-stack blame_entry should lose one refcnt each.
 709 */
 710static void decref_split(struct blame_entry *split)
 711{
 712        int i;
 713
 714        for (i = 0; i < 3; i++)
 715                origin_decref(split[i].suspect);
 716}
 717
 718/*
 719 * Helper for blame_chunk().  blame_entry e is known to overlap with
 720 * the patch hunk; split it and pass blame to the parent.
 721 */
 722static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 723                          int tlno, int plno, int same,
 724                          struct origin *parent)
 725{
 726        struct blame_entry split[3];
 727
 728        split_overlap(split, e, tlno, plno, same, parent);
 729        if (split[1].suspect)
 730                split_blame(sb, split, e);
 731        decref_split(split);
 732}
 733
 734/*
 735 * Find the line number of the last line the target is suspected for.
 736 */
 737static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 738{
 739        struct blame_entry *e;
 740        int last_in_target = -1;
 741
 742        for (e = sb->ent; e; e = e->next) {
 743                if (e->guilty || !same_suspect(e->suspect, target))
 744                        continue;
 745                if (last_in_target < e->s_lno + e->num_lines)
 746                        last_in_target = e->s_lno + e->num_lines;
 747        }
 748        return last_in_target;
 749}
 750
 751/*
 752 * Process one hunk from the patch between the current suspect for
 753 * blame_entry e and its parent.  Find and split the overlap, and
 754 * pass blame to the overlapping part to the parent.
 755 */
 756static void blame_chunk(struct scoreboard *sb,
 757                        int tlno, int plno, int same,
 758                        struct origin *target, struct origin *parent)
 759{
 760        struct blame_entry *e;
 761
 762        for (e = sb->ent; e; e = e->next) {
 763                if (e->guilty || !same_suspect(e->suspect, target))
 764                        continue;
 765                if (same <= e->s_lno)
 766                        continue;
 767                if (tlno < e->s_lno + e->num_lines)
 768                        blame_overlap(sb, e, tlno, plno, same, parent);
 769        }
 770}
 771
 772struct blame_chunk_cb_data {
 773        struct scoreboard *sb;
 774        struct origin *target;
 775        struct origin *parent;
 776        long plno;
 777        long tlno;
 778};
 779
 780static int blame_chunk_cb(long start_a, long count_a,
 781                          long start_b, long count_b, void *data)
 782{
 783        struct blame_chunk_cb_data *d = data;
 784        blame_chunk(d->sb, d->tlno, d->plno, start_b, d->target, d->parent);
 785        d->plno = start_a + count_a;
 786        d->tlno = start_b + count_b;
 787        return 0;
 788}
 789
 790/*
 791 * We are looking at the origin 'target' and aiming to pass blame
 792 * for the lines it is suspected to its parent.  Run diff to find
 793 * which lines came from parent and pass blame for them.
 794 */
 795static int pass_blame_to_parent(struct scoreboard *sb,
 796                                struct origin *target,
 797                                struct origin *parent)
 798{
 799        int last_in_target;
 800        mmfile_t file_p, file_o;
 801        struct blame_chunk_cb_data d;
 802
 803        memset(&d, 0, sizeof(d));
 804        d.sb = sb; d.target = target; d.parent = parent;
 805        last_in_target = find_last_in_target(sb, target);
 806        if (last_in_target < 0)
 807                return 1; /* nothing remains for this target */
 808
 809        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 810        fill_origin_blob(&sb->revs->diffopt, target, &file_o);
 811        num_get_patch++;
 812
 813        diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
 814        /* The rest (i.e. anything after tlno) are the same as the parent */
 815        blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
 816
 817        return 0;
 818}
 819
 820/*
 821 * The lines in blame_entry after splitting blames many times can become
 822 * very small and trivial, and at some point it becomes pointless to
 823 * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 824 * ordinary C program, and it is not worth to say it was copied from
 825 * totally unrelated file in the parent.
 826 *
 827 * Compute how trivial the lines in the blame_entry are.
 828 */
 829static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 830{
 831        unsigned score;
 832        const char *cp, *ep;
 833
 834        if (e->score)
 835                return e->score;
 836
 837        score = 1;
 838        cp = nth_line(sb, e->lno);
 839        ep = nth_line(sb, e->lno + e->num_lines);
 840        while (cp < ep) {
 841                unsigned ch = *((unsigned char *)cp);
 842                if (isalnum(ch))
 843                        score++;
 844                cp++;
 845        }
 846        e->score = score;
 847        return score;
 848}
 849
 850/*
 851 * best_so_far[] and this[] are both a split of an existing blame_entry
 852 * that passes blame to the parent.  Maintain best_so_far the best split
 853 * so far, by comparing this and best_so_far and copying this into
 854 * bst_so_far as needed.
 855 */
 856static void copy_split_if_better(struct scoreboard *sb,
 857                                 struct blame_entry *best_so_far,
 858                                 struct blame_entry *this)
 859{
 860        int i;
 861
 862        if (!this[1].suspect)
 863                return;
 864        if (best_so_far[1].suspect) {
 865                if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 866                        return;
 867        }
 868
 869        for (i = 0; i < 3; i++)
 870                origin_incref(this[i].suspect);
 871        decref_split(best_so_far);
 872        memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 873}
 874
 875/*
 876 * We are looking at a part of the final image represented by
 877 * ent (tlno and same are offset by ent->s_lno).
 878 * tlno is where we are looking at in the final image.
 879 * up to (but not including) same match preimage.
 880 * plno is where we are looking at in the preimage.
 881 *
 882 * <-------------- final image ---------------------->
 883 *       <------ent------>
 884 *         ^tlno ^same
 885 *    <---------preimage----->
 886 *         ^plno
 887 *
 888 * All line numbers are 0-based.
 889 */
 890static void handle_split(struct scoreboard *sb,
 891                         struct blame_entry *ent,
 892                         int tlno, int plno, int same,
 893                         struct origin *parent,
 894                         struct blame_entry *split)
 895{
 896        if (ent->num_lines <= tlno)
 897                return;
 898        if (tlno < same) {
 899                struct blame_entry this[3];
 900                tlno += ent->s_lno;
 901                same += ent->s_lno;
 902                split_overlap(this, ent, tlno, plno, same, parent);
 903                copy_split_if_better(sb, split, this);
 904                decref_split(this);
 905        }
 906}
 907
 908struct handle_split_cb_data {
 909        struct scoreboard *sb;
 910        struct blame_entry *ent;
 911        struct origin *parent;
 912        struct blame_entry *split;
 913        long plno;
 914        long tlno;
 915};
 916
 917static int handle_split_cb(long start_a, long count_a,
 918                           long start_b, long count_b, void *data)
 919{
 920        struct handle_split_cb_data *d = data;
 921        handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
 922                     d->split);
 923        d->plno = start_a + count_a;
 924        d->tlno = start_b + count_b;
 925        return 0;
 926}
 927
 928/*
 929 * Find the lines from parent that are the same as ent so that
 930 * we can pass blames to it.  file_p has the blob contents for
 931 * the parent.
 932 */
 933static void find_copy_in_blob(struct scoreboard *sb,
 934                              struct blame_entry *ent,
 935                              struct origin *parent,
 936                              struct blame_entry *split,
 937                              mmfile_t *file_p)
 938{
 939        const char *cp;
 940        int cnt;
 941        mmfile_t file_o;
 942        struct handle_split_cb_data d;
 943
 944        memset(&d, 0, sizeof(d));
 945        d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
 946        /*
 947         * Prepare mmfile that contains only the lines in ent.
 948         */
 949        cp = nth_line(sb, ent->lno);
 950        file_o.ptr = (char *) cp;
 951        cnt = ent->num_lines;
 952
 953        while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 954                if (*cp++ == '\n')
 955                        cnt--;
 956        }
 957        file_o.size = cp - file_o.ptr;
 958
 959        /*
 960         * file_o is a part of final image we are annotating.
 961         * file_p partially may match that image.
 962         */
 963        memset(split, 0, sizeof(struct blame_entry [3]));
 964        diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
 965        /* remainder, if any, all match the preimage */
 966        handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
 967}
 968
 969/*
 970 * See if lines currently target is suspected for can be attributed to
 971 * parent.
 972 */
 973static int find_move_in_parent(struct scoreboard *sb,
 974                               struct origin *target,
 975                               struct origin *parent)
 976{
 977        int last_in_target, made_progress;
 978        struct blame_entry *e, split[3];
 979        mmfile_t file_p;
 980
 981        last_in_target = find_last_in_target(sb, target);
 982        if (last_in_target < 0)
 983                return 1; /* nothing remains for this target */
 984
 985        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 986        if (!file_p.ptr)
 987                return 0;
 988
 989        made_progress = 1;
 990        while (made_progress) {
 991                made_progress = 0;
 992                for (e = sb->ent; e; e = e->next) {
 993                        if (e->guilty || !same_suspect(e->suspect, target) ||
 994                            ent_score(sb, e) < blame_move_score)
 995                                continue;
 996                        find_copy_in_blob(sb, e, parent, split, &file_p);
 997                        if (split[1].suspect &&
 998                            blame_move_score < ent_score(sb, &split[1])) {
 999                                split_blame(sb, split, e);
1000                                made_progress = 1;
1001                        }
1002                        decref_split(split);
1003                }
1004        }
1005        return 0;
1006}
1007
1008struct blame_list {
1009        struct blame_entry *ent;
1010        struct blame_entry split[3];
1011};
1012
1013/*
1014 * Count the number of entries the target is suspected for,
1015 * and prepare a list of entry and the best split.
1016 */
1017static struct blame_list *setup_blame_list(struct scoreboard *sb,
1018                                           struct origin *target,
1019                                           int min_score,
1020                                           int *num_ents_p)
1021{
1022        struct blame_entry *e;
1023        int num_ents, i;
1024        struct blame_list *blame_list = NULL;
1025
1026        for (e = sb->ent, num_ents = 0; e; e = e->next)
1027                if (!e->scanned && !e->guilty &&
1028                    same_suspect(e->suspect, target) &&
1029                    min_score < ent_score(sb, e))
1030                        num_ents++;
1031        if (num_ents) {
1032                blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1033                for (e = sb->ent, i = 0; e; e = e->next)
1034                        if (!e->scanned && !e->guilty &&
1035                            same_suspect(e->suspect, target) &&
1036                            min_score < ent_score(sb, e))
1037                                blame_list[i++].ent = e;
1038        }
1039        *num_ents_p = num_ents;
1040        return blame_list;
1041}
1042
1043/*
1044 * Reset the scanned status on all entries.
1045 */
1046static void reset_scanned_flag(struct scoreboard *sb)
1047{
1048        struct blame_entry *e;
1049        for (e = sb->ent; e; e = e->next)
1050                e->scanned = 0;
1051}
1052
1053/*
1054 * For lines target is suspected for, see if we can find code movement
1055 * across file boundary from the parent commit.  porigin is the path
1056 * in the parent we already tried.
1057 */
1058static int find_copy_in_parent(struct scoreboard *sb,
1059                               struct origin *target,
1060                               struct commit *parent,
1061                               struct origin *porigin,
1062                               int opt)
1063{
1064        struct diff_options diff_opts;
1065        int i, j;
1066        int retval;
1067        struct blame_list *blame_list;
1068        int num_ents;
1069
1070        blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1071        if (!blame_list)
1072                return 1; /* nothing remains for this target */
1073
1074        diff_setup(&diff_opts);
1075        DIFF_OPT_SET(&diff_opts, RECURSIVE);
1076        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1077
1078        diff_setup_done(&diff_opts);
1079
1080        /* Try "find copies harder" on new path if requested;
1081         * we do not want to use diffcore_rename() actually to
1082         * match things up; find_copies_harder is set only to
1083         * force diff_tree_sha1() to feed all filepairs to diff_queue,
1084         * and this code needs to be after diff_setup_done(), which
1085         * usually makes find-copies-harder imply copy detection.
1086         */
1087        if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1088            || ((opt & PICKAXE_BLAME_COPY_HARDER)
1089                && (!porigin || strcmp(target->path, porigin->path))))
1090                DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1091
1092        if (is_null_sha1(target->commit->object.sha1))
1093                do_diff_cache(parent->tree->object.sha1, &diff_opts);
1094        else
1095                diff_tree_sha1(parent->tree->object.sha1,
1096                               target->commit->tree->object.sha1,
1097                               "", &diff_opts);
1098
1099        if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1100                diffcore_std(&diff_opts);
1101
1102        retval = 0;
1103        while (1) {
1104                int made_progress = 0;
1105
1106                for (i = 0; i < diff_queued_diff.nr; i++) {
1107                        struct diff_filepair *p = diff_queued_diff.queue[i];
1108                        struct origin *norigin;
1109                        mmfile_t file_p;
1110                        struct blame_entry this[3];
1111
1112                        if (!DIFF_FILE_VALID(p->one))
1113                                continue; /* does not exist in parent */
1114                        if (S_ISGITLINK(p->one->mode))
1115                                continue; /* ignore git links */
1116                        if (porigin && !strcmp(p->one->path, porigin->path))
1117                                /* find_move already dealt with this path */
1118                                continue;
1119
1120                        norigin = get_origin(sb, parent, p->one->path);
1121                        hashcpy(norigin->blob_sha1, p->one->sha1);
1122                        norigin->mode = p->one->mode;
1123                        fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1124                        if (!file_p.ptr)
1125                                continue;
1126
1127                        for (j = 0; j < num_ents; j++) {
1128                                find_copy_in_blob(sb, blame_list[j].ent,
1129                                                  norigin, this, &file_p);
1130                                copy_split_if_better(sb, blame_list[j].split,
1131                                                     this);
1132                                decref_split(this);
1133                        }
1134                        origin_decref(norigin);
1135                }
1136
1137                for (j = 0; j < num_ents; j++) {
1138                        struct blame_entry *split = blame_list[j].split;
1139                        if (split[1].suspect &&
1140                            blame_copy_score < ent_score(sb, &split[1])) {
1141                                split_blame(sb, split, blame_list[j].ent);
1142                                made_progress = 1;
1143                        }
1144                        else
1145                                blame_list[j].ent->scanned = 1;
1146                        decref_split(split);
1147                }
1148                free(blame_list);
1149
1150                if (!made_progress)
1151                        break;
1152                blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1153                if (!blame_list) {
1154                        retval = 1;
1155                        break;
1156                }
1157        }
1158        reset_scanned_flag(sb);
1159        diff_flush(&diff_opts);
1160        free_pathspec(&diff_opts.pathspec);
1161        return retval;
1162}
1163
1164/*
1165 * The blobs of origin and porigin exactly match, so everything
1166 * origin is suspected for can be blamed on the parent.
1167 */
1168static void pass_whole_blame(struct scoreboard *sb,
1169                             struct origin *origin, struct origin *porigin)
1170{
1171        struct blame_entry *e;
1172
1173        if (!porigin->file.ptr && origin->file.ptr) {
1174                /* Steal its file */
1175                porigin->file = origin->file;
1176                origin->file.ptr = NULL;
1177        }
1178        for (e = sb->ent; e; e = e->next) {
1179                if (!same_suspect(e->suspect, origin))
1180                        continue;
1181                origin_incref(porigin);
1182                origin_decref(e->suspect);
1183                e->suspect = porigin;
1184        }
1185}
1186
1187/*
1188 * We pass blame from the current commit to its parents.  We keep saying
1189 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1190 * exonerate ourselves.
1191 */
1192static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1193{
1194        if (!reverse)
1195                return commit->parents;
1196        return lookup_decoration(&revs->children, &commit->object);
1197}
1198
1199static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1200{
1201        int cnt;
1202        struct commit_list *l = first_scapegoat(revs, commit);
1203        for (cnt = 0; l; l = l->next)
1204                cnt++;
1205        return cnt;
1206}
1207
1208#define MAXSG 16
1209
1210static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1211{
1212        struct rev_info *revs = sb->revs;
1213        int i, pass, num_sg;
1214        struct commit *commit = origin->commit;
1215        struct commit_list *sg;
1216        struct origin *sg_buf[MAXSG];
1217        struct origin *porigin, **sg_origin = sg_buf;
1218
1219        num_sg = num_scapegoats(revs, commit);
1220        if (!num_sg)
1221                goto finish;
1222        else if (num_sg < ARRAY_SIZE(sg_buf))
1223                memset(sg_buf, 0, sizeof(sg_buf));
1224        else
1225                sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1226
1227        /*
1228         * The first pass looks for unrenamed path to optimize for
1229         * common cases, then we look for renames in the second pass.
1230         */
1231        for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1232                struct origin *(*find)(struct scoreboard *,
1233                                       struct commit *, struct origin *);
1234                find = pass ? find_rename : find_origin;
1235
1236                for (i = 0, sg = first_scapegoat(revs, commit);
1237                     i < num_sg && sg;
1238                     sg = sg->next, i++) {
1239                        struct commit *p = sg->item;
1240                        int j, same;
1241
1242                        if (sg_origin[i])
1243                                continue;
1244                        if (parse_commit(p))
1245                                continue;
1246                        porigin = find(sb, p, origin);
1247                        if (!porigin)
1248                                continue;
1249                        if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1250                                pass_whole_blame(sb, origin, porigin);
1251                                origin_decref(porigin);
1252                                goto finish;
1253                        }
1254                        for (j = same = 0; j < i; j++)
1255                                if (sg_origin[j] &&
1256                                    !hashcmp(sg_origin[j]->blob_sha1,
1257                                             porigin->blob_sha1)) {
1258                                        same = 1;
1259                                        break;
1260                                }
1261                        if (!same)
1262                                sg_origin[i] = porigin;
1263                        else
1264                                origin_decref(porigin);
1265                }
1266        }
1267
1268        num_commits++;
1269        for (i = 0, sg = first_scapegoat(revs, commit);
1270             i < num_sg && sg;
1271             sg = sg->next, i++) {
1272                struct origin *porigin = sg_origin[i];
1273                if (!porigin)
1274                        continue;
1275                if (!origin->previous) {
1276                        origin_incref(porigin);
1277                        origin->previous = porigin;
1278                }
1279                if (pass_blame_to_parent(sb, origin, porigin))
1280                        goto finish;
1281        }
1282
1283        /*
1284         * Optionally find moves in parents' files.
1285         */
1286        if (opt & PICKAXE_BLAME_MOVE)
1287                for (i = 0, sg = first_scapegoat(revs, commit);
1288                     i < num_sg && sg;
1289                     sg = sg->next, i++) {
1290                        struct origin *porigin = sg_origin[i];
1291                        if (!porigin)
1292                                continue;
1293                        if (find_move_in_parent(sb, origin, porigin))
1294                                goto finish;
1295                }
1296
1297        /*
1298         * Optionally find copies from parents' files.
1299         */
1300        if (opt & PICKAXE_BLAME_COPY)
1301                for (i = 0, sg = first_scapegoat(revs, commit);
1302                     i < num_sg && sg;
1303                     sg = sg->next, i++) {
1304                        struct origin *porigin = sg_origin[i];
1305                        if (find_copy_in_parent(sb, origin, sg->item,
1306                                                porigin, opt))
1307                                goto finish;
1308                }
1309
1310 finish:
1311        for (i = 0; i < num_sg; i++) {
1312                if (sg_origin[i]) {
1313                        drop_origin_blob(sg_origin[i]);
1314                        origin_decref(sg_origin[i]);
1315                }
1316        }
1317        drop_origin_blob(origin);
1318        if (sg_buf != sg_origin)
1319                free(sg_origin);
1320}
1321
1322/*
1323 * Information on commits, used for output.
1324 */
1325struct commit_info {
1326        struct strbuf author;
1327        struct strbuf author_mail;
1328        unsigned long author_time;
1329        struct strbuf author_tz;
1330
1331        /* filled only when asked for details */
1332        struct strbuf committer;
1333        struct strbuf committer_mail;
1334        unsigned long committer_time;
1335        struct strbuf committer_tz;
1336
1337        struct strbuf summary;
1338};
1339
1340/*
1341 * Parse author/committer line in the commit object buffer
1342 */
1343static void get_ac_line(const char *inbuf, const char *what,
1344        struct strbuf *name, struct strbuf *mail,
1345        unsigned long *time, struct strbuf *tz)
1346{
1347        struct ident_split ident;
1348        size_t len, maillen, namelen;
1349        char *tmp, *endp;
1350        const char *namebuf, *mailbuf;
1351
1352        tmp = strstr(inbuf, what);
1353        if (!tmp)
1354                goto error_out;
1355        tmp += strlen(what);
1356        endp = strchr(tmp, '\n');
1357        if (!endp)
1358                len = strlen(tmp);
1359        else
1360                len = endp - tmp;
1361
1362        if (split_ident_line(&ident, tmp, len)) {
1363        error_out:
1364                /* Ugh */
1365                tmp = "(unknown)";
1366                strbuf_addstr(name, tmp);
1367                strbuf_addstr(mail, tmp);
1368                strbuf_addstr(tz, tmp);
1369                *time = 0;
1370                return;
1371        }
1372
1373        namelen = ident.name_end - ident.name_begin;
1374        namebuf = ident.name_begin;
1375
1376        maillen = ident.mail_end - ident.mail_begin;
1377        mailbuf = ident.mail_begin;
1378
1379        if (ident.date_begin && ident.date_end)
1380                *time = strtoul(ident.date_begin, NULL, 10);
1381        else
1382                *time = 0;
1383
1384        if (ident.tz_begin && ident.tz_end)
1385                strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1386        else
1387                strbuf_addstr(tz, "(unknown)");
1388
1389        /*
1390         * Now, convert both name and e-mail using mailmap
1391         */
1392        map_user(&mailmap, &mailbuf, &maillen,
1393                 &namebuf, &namelen);
1394
1395        strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1396        strbuf_add(name, namebuf, namelen);
1397}
1398
1399static void commit_info_init(struct commit_info *ci)
1400{
1401
1402        strbuf_init(&ci->author, 0);
1403        strbuf_init(&ci->author_mail, 0);
1404        strbuf_init(&ci->author_tz, 0);
1405        strbuf_init(&ci->committer, 0);
1406        strbuf_init(&ci->committer_mail, 0);
1407        strbuf_init(&ci->committer_tz, 0);
1408        strbuf_init(&ci->summary, 0);
1409}
1410
1411static void commit_info_destroy(struct commit_info *ci)
1412{
1413
1414        strbuf_release(&ci->author);
1415        strbuf_release(&ci->author_mail);
1416        strbuf_release(&ci->author_tz);
1417        strbuf_release(&ci->committer);
1418        strbuf_release(&ci->committer_mail);
1419        strbuf_release(&ci->committer_tz);
1420        strbuf_release(&ci->summary);
1421}
1422
1423static void get_commit_info(struct commit *commit,
1424                            struct commit_info *ret,
1425                            int detailed)
1426{
1427        int len;
1428        const char *subject, *encoding;
1429        char *message;
1430
1431        commit_info_init(ret);
1432
1433        encoding = get_log_output_encoding();
1434        message = logmsg_reencode(commit, NULL, encoding);
1435        get_ac_line(message, "\nauthor ",
1436                    &ret->author, &ret->author_mail,
1437                    &ret->author_time, &ret->author_tz);
1438
1439        if (!detailed) {
1440                logmsg_free(message, commit);
1441                return;
1442        }
1443
1444        get_ac_line(message, "\ncommitter ",
1445                    &ret->committer, &ret->committer_mail,
1446                    &ret->committer_time, &ret->committer_tz);
1447
1448        len = find_commit_subject(message, &subject);
1449        if (len)
1450                strbuf_add(&ret->summary, subject, len);
1451        else
1452                strbuf_addf(&ret->summary, "(%s)", sha1_to_hex(commit->object.sha1));
1453
1454        logmsg_free(message, commit);
1455}
1456
1457/*
1458 * To allow LF and other nonportable characters in pathnames,
1459 * they are c-style quoted as needed.
1460 */
1461static void write_filename_info(const char *path)
1462{
1463        printf("filename ");
1464        write_name_quoted(path, stdout, '\n');
1465}
1466
1467/*
1468 * Porcelain/Incremental format wants to show a lot of details per
1469 * commit.  Instead of repeating this every line, emit it only once,
1470 * the first time each commit appears in the output (unless the
1471 * user has specifically asked for us to repeat).
1472 */
1473static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1474{
1475        struct commit_info ci;
1476
1477        if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1478                return 0;
1479
1480        suspect->commit->object.flags |= METAINFO_SHOWN;
1481        get_commit_info(suspect->commit, &ci, 1);
1482        printf("author %s\n", ci.author.buf);
1483        printf("author-mail %s\n", ci.author_mail.buf);
1484        printf("author-time %lu\n", ci.author_time);
1485        printf("author-tz %s\n", ci.author_tz.buf);
1486        printf("committer %s\n", ci.committer.buf);
1487        printf("committer-mail %s\n", ci.committer_mail.buf);
1488        printf("committer-time %lu\n", ci.committer_time);
1489        printf("committer-tz %s\n", ci.committer_tz.buf);
1490        printf("summary %s\n", ci.summary.buf);
1491        if (suspect->commit->object.flags & UNINTERESTING)
1492                printf("boundary\n");
1493        if (suspect->previous) {
1494                struct origin *prev = suspect->previous;
1495                printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1496                write_name_quoted(prev->path, stdout, '\n');
1497        }
1498
1499        commit_info_destroy(&ci);
1500
1501        return 1;
1502}
1503
1504/*
1505 * The blame_entry is found to be guilty for the range.  Mark it
1506 * as such, and show it in incremental output.
1507 */
1508static void found_guilty_entry(struct blame_entry *ent)
1509{
1510        if (ent->guilty)
1511                return;
1512        ent->guilty = 1;
1513        if (incremental) {
1514                struct origin *suspect = ent->suspect;
1515
1516                printf("%s %d %d %d\n",
1517                       sha1_to_hex(suspect->commit->object.sha1),
1518                       ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1519                emit_one_suspect_detail(suspect, 0);
1520                write_filename_info(suspect->path);
1521                maybe_flush_or_die(stdout, "stdout");
1522        }
1523}
1524
1525/*
1526 * The main loop -- while the scoreboard has lines whose true origin
1527 * is still unknown, pick one blame_entry, and allow its current
1528 * suspect to pass blames to its parents.
1529 */
1530static void assign_blame(struct scoreboard *sb, int opt)
1531{
1532        struct rev_info *revs = sb->revs;
1533
1534        while (1) {
1535                struct blame_entry *ent;
1536                struct commit *commit;
1537                struct origin *suspect = NULL;
1538
1539                /* find one suspect to break down */
1540                for (ent = sb->ent; !suspect && ent; ent = ent->next)
1541                        if (!ent->guilty)
1542                                suspect = ent->suspect;
1543                if (!suspect)
1544                        return; /* all done */
1545
1546                /*
1547                 * We will use this suspect later in the loop,
1548                 * so hold onto it in the meantime.
1549                 */
1550                origin_incref(suspect);
1551                commit = suspect->commit;
1552                if (!commit->object.parsed)
1553                        parse_commit(commit);
1554                if (reverse ||
1555                    (!(commit->object.flags & UNINTERESTING) &&
1556                     !(revs->max_age != -1 && commit->date < revs->max_age)))
1557                        pass_blame(sb, suspect, opt);
1558                else {
1559                        commit->object.flags |= UNINTERESTING;
1560                        if (commit->object.parsed)
1561                                mark_parents_uninteresting(commit);
1562                }
1563                /* treat root commit as boundary */
1564                if (!commit->parents && !show_root)
1565                        commit->object.flags |= UNINTERESTING;
1566
1567                /* Take responsibility for the remaining entries */
1568                for (ent = sb->ent; ent; ent = ent->next)
1569                        if (same_suspect(ent->suspect, suspect))
1570                                found_guilty_entry(ent);
1571                origin_decref(suspect);
1572
1573                if (DEBUG) /* sanity */
1574                        sanity_check_refcnt(sb);
1575        }
1576}
1577
1578static const char *format_time(unsigned long time, const char *tz_str,
1579                               int show_raw_time)
1580{
1581        static char time_buf[128];
1582        const char *time_str;
1583        int time_len;
1584        int tz;
1585
1586        if (show_raw_time) {
1587                snprintf(time_buf, sizeof(time_buf), "%lu %s", time, tz_str);
1588        }
1589        else {
1590                tz = atoi(tz_str);
1591                time_str = show_date(time, tz, blame_date_mode);
1592                time_len = strlen(time_str);
1593                memcpy(time_buf, time_str, time_len);
1594                memset(time_buf + time_len, ' ', blame_date_width - time_len);
1595        }
1596        return time_buf;
1597}
1598
1599#define OUTPUT_ANNOTATE_COMPAT  001
1600#define OUTPUT_LONG_OBJECT_NAME 002
1601#define OUTPUT_RAW_TIMESTAMP    004
1602#define OUTPUT_PORCELAIN        010
1603#define OUTPUT_SHOW_NAME        020
1604#define OUTPUT_SHOW_NUMBER      040
1605#define OUTPUT_SHOW_SCORE      0100
1606#define OUTPUT_NO_AUTHOR       0200
1607#define OUTPUT_SHOW_EMAIL       0400
1608#define OUTPUT_LINE_PORCELAIN 01000
1609
1610static void emit_porcelain_details(struct origin *suspect, int repeat)
1611{
1612        if (emit_one_suspect_detail(suspect, repeat) ||
1613            (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1614                write_filename_info(suspect->path);
1615}
1616
1617static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1618                           int opt)
1619{
1620        int repeat = opt & OUTPUT_LINE_PORCELAIN;
1621        int cnt;
1622        const char *cp;
1623        struct origin *suspect = ent->suspect;
1624        char hex[41];
1625
1626        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1627        printf("%s%c%d %d %d\n",
1628               hex,
1629               ent->guilty ? ' ' : '*', /* purely for debugging */
1630               ent->s_lno + 1,
1631               ent->lno + 1,
1632               ent->num_lines);
1633        emit_porcelain_details(suspect, repeat);
1634
1635        cp = nth_line(sb, ent->lno);
1636        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1637                char ch;
1638                if (cnt) {
1639                        printf("%s %d %d\n", hex,
1640                               ent->s_lno + 1 + cnt,
1641                               ent->lno + 1 + cnt);
1642                        if (repeat)
1643                                emit_porcelain_details(suspect, 1);
1644                }
1645                putchar('\t');
1646                do {
1647                        ch = *cp++;
1648                        putchar(ch);
1649                } while (ch != '\n' &&
1650                         cp < sb->final_buf + sb->final_buf_size);
1651        }
1652
1653        if (sb->final_buf_size && cp[-1] != '\n')
1654                putchar('\n');
1655}
1656
1657static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1658{
1659        int cnt;
1660        const char *cp;
1661        struct origin *suspect = ent->suspect;
1662        struct commit_info ci;
1663        char hex[41];
1664        int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1665
1666        get_commit_info(suspect->commit, &ci, 1);
1667        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1668
1669        cp = nth_line(sb, ent->lno);
1670        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1671                char ch;
1672                int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1673
1674                if (suspect->commit->object.flags & UNINTERESTING) {
1675                        if (blank_boundary)
1676                                memset(hex, ' ', length);
1677                        else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1678                                length--;
1679                                putchar('^');
1680                        }
1681                }
1682
1683                printf("%.*s", length, hex);
1684                if (opt & OUTPUT_ANNOTATE_COMPAT) {
1685                        const char *name;
1686                        if (opt & OUTPUT_SHOW_EMAIL)
1687                                name = ci.author_mail.buf;
1688                        else
1689                                name = ci.author.buf;
1690                        printf("\t(%10s\t%10s\t%d)", name,
1691                               format_time(ci.author_time, ci.author_tz.buf,
1692                                           show_raw_time),
1693                               ent->lno + 1 + cnt);
1694                } else {
1695                        if (opt & OUTPUT_SHOW_SCORE)
1696                                printf(" %*d %02d",
1697                                       max_score_digits, ent->score,
1698                                       ent->suspect->refcnt);
1699                        if (opt & OUTPUT_SHOW_NAME)
1700                                printf(" %-*.*s", longest_file, longest_file,
1701                                       suspect->path);
1702                        if (opt & OUTPUT_SHOW_NUMBER)
1703                                printf(" %*d", max_orig_digits,
1704                                       ent->s_lno + 1 + cnt);
1705
1706                        if (!(opt & OUTPUT_NO_AUTHOR)) {
1707                                const char *name;
1708                                int pad;
1709                                if (opt & OUTPUT_SHOW_EMAIL)
1710                                        name = ci.author_mail.buf;
1711                                else
1712                                        name = ci.author.buf;
1713                                pad = longest_author - utf8_strwidth(name);
1714                                printf(" (%s%*s %10s",
1715                                       name, pad, "",
1716                                       format_time(ci.author_time,
1717                                                   ci.author_tz.buf,
1718                                                   show_raw_time));
1719                        }
1720                        printf(" %*d) ",
1721                               max_digits, ent->lno + 1 + cnt);
1722                }
1723                do {
1724                        ch = *cp++;
1725                        putchar(ch);
1726                } while (ch != '\n' &&
1727                         cp < sb->final_buf + sb->final_buf_size);
1728        }
1729
1730        if (sb->final_buf_size && cp[-1] != '\n')
1731                putchar('\n');
1732
1733        commit_info_destroy(&ci);
1734}
1735
1736static void output(struct scoreboard *sb, int option)
1737{
1738        struct blame_entry *ent;
1739
1740        if (option & OUTPUT_PORCELAIN) {
1741                for (ent = sb->ent; ent; ent = ent->next) {
1742                        struct blame_entry *oth;
1743                        struct origin *suspect = ent->suspect;
1744                        struct commit *commit = suspect->commit;
1745                        if (commit->object.flags & MORE_THAN_ONE_PATH)
1746                                continue;
1747                        for (oth = ent->next; oth; oth = oth->next) {
1748                                if ((oth->suspect->commit != commit) ||
1749                                    !strcmp(oth->suspect->path, suspect->path))
1750                                        continue;
1751                                commit->object.flags |= MORE_THAN_ONE_PATH;
1752                                break;
1753                        }
1754                }
1755        }
1756
1757        for (ent = sb->ent; ent; ent = ent->next) {
1758                if (option & OUTPUT_PORCELAIN)
1759                        emit_porcelain(sb, ent, option);
1760                else {
1761                        emit_other(sb, ent, option);
1762                }
1763        }
1764}
1765
1766/*
1767 * To allow quick access to the contents of nth line in the
1768 * final image, prepare an index in the scoreboard.
1769 */
1770static int prepare_lines(struct scoreboard *sb)
1771{
1772        const char *buf = sb->final_buf;
1773        unsigned long len = sb->final_buf_size;
1774        int num = 0, incomplete = 0, bol = 1;
1775
1776        if (len && buf[len-1] != '\n')
1777                incomplete++; /* incomplete line at the end */
1778        while (len--) {
1779                if (bol) {
1780                        sb->lineno = xrealloc(sb->lineno,
1781                                              sizeof(int *) * (num + 1));
1782                        sb->lineno[num] = buf - sb->final_buf;
1783                        bol = 0;
1784                }
1785                if (*buf++ == '\n') {
1786                        num++;
1787                        bol = 1;
1788                }
1789        }
1790        sb->lineno = xrealloc(sb->lineno,
1791                              sizeof(int *) * (num + incomplete + 1));
1792        sb->lineno[num + incomplete] = buf - sb->final_buf;
1793        sb->num_lines = num + incomplete;
1794        return sb->num_lines;
1795}
1796
1797/*
1798 * Add phony grafts for use with -S; this is primarily to
1799 * support git's cvsserver that wants to give a linear history
1800 * to its clients.
1801 */
1802static int read_ancestry(const char *graft_file)
1803{
1804        FILE *fp = fopen(graft_file, "r");
1805        char buf[1024];
1806        if (!fp)
1807                return -1;
1808        while (fgets(buf, sizeof(buf), fp)) {
1809                /* The format is just "Commit Parent1 Parent2 ...\n" */
1810                int len = strlen(buf);
1811                struct commit_graft *graft = read_graft_line(buf, len);
1812                if (graft)
1813                        register_commit_graft(graft, 0);
1814        }
1815        fclose(fp);
1816        return 0;
1817}
1818
1819static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1820{
1821        const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1822                                              auto_abbrev);
1823        int len = strlen(uniq);
1824        if (auto_abbrev < len)
1825                return len;
1826        return auto_abbrev;
1827}
1828
1829/*
1830 * How many columns do we need to show line numbers, authors,
1831 * and filenames?
1832 */
1833static void find_alignment(struct scoreboard *sb, int *option)
1834{
1835        int longest_src_lines = 0;
1836        int longest_dst_lines = 0;
1837        unsigned largest_score = 0;
1838        struct blame_entry *e;
1839        int compute_auto_abbrev = (abbrev < 0);
1840        int auto_abbrev = default_abbrev;
1841
1842        for (e = sb->ent; e; e = e->next) {
1843                struct origin *suspect = e->suspect;
1844                struct commit_info ci;
1845                int num;
1846
1847                if (compute_auto_abbrev)
1848                        auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1849                if (strcmp(suspect->path, sb->path))
1850                        *option |= OUTPUT_SHOW_NAME;
1851                num = strlen(suspect->path);
1852                if (longest_file < num)
1853                        longest_file = num;
1854                if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1855                        suspect->commit->object.flags |= METAINFO_SHOWN;
1856                        get_commit_info(suspect->commit, &ci, 1);
1857                        if (*option & OUTPUT_SHOW_EMAIL)
1858                                num = utf8_strwidth(ci.author_mail.buf);
1859                        else
1860                                num = utf8_strwidth(ci.author.buf);
1861                        if (longest_author < num)
1862                                longest_author = num;
1863                }
1864                num = e->s_lno + e->num_lines;
1865                if (longest_src_lines < num)
1866                        longest_src_lines = num;
1867                num = e->lno + e->num_lines;
1868                if (longest_dst_lines < num)
1869                        longest_dst_lines = num;
1870                if (largest_score < ent_score(sb, e))
1871                        largest_score = ent_score(sb, e);
1872
1873                commit_info_destroy(&ci);
1874        }
1875        max_orig_digits = decimal_width(longest_src_lines);
1876        max_digits = decimal_width(longest_dst_lines);
1877        max_score_digits = decimal_width(largest_score);
1878
1879        if (compute_auto_abbrev)
1880                /* one more abbrev length is needed for the boundary commit */
1881                abbrev = auto_abbrev + 1;
1882}
1883
1884/*
1885 * For debugging -- origin is refcounted, and this asserts that
1886 * we do not underflow.
1887 */
1888static void sanity_check_refcnt(struct scoreboard *sb)
1889{
1890        int baa = 0;
1891        struct blame_entry *ent;
1892
1893        for (ent = sb->ent; ent; ent = ent->next) {
1894                /* Nobody should have zero or negative refcnt */
1895                if (ent->suspect->refcnt <= 0) {
1896                        fprintf(stderr, "%s in %s has negative refcnt %d\n",
1897                                ent->suspect->path,
1898                                sha1_to_hex(ent->suspect->commit->object.sha1),
1899                                ent->suspect->refcnt);
1900                        baa = 1;
1901                }
1902        }
1903        if (baa) {
1904                int opt = 0160;
1905                find_alignment(sb, &opt);
1906                output(sb, opt);
1907                die("Baa %d!", baa);
1908        }
1909}
1910
1911/*
1912 * Used for the command line parsing; check if the path exists
1913 * in the working tree.
1914 */
1915static int has_string_in_work_tree(const char *path)
1916{
1917        struct stat st;
1918        return !lstat(path, &st);
1919}
1920
1921static unsigned parse_score(const char *arg)
1922{
1923        char *end;
1924        unsigned long score = strtoul(arg, &end, 10);
1925        if (*end)
1926                return 0;
1927        return score;
1928}
1929
1930static const char *add_prefix(const char *prefix, const char *path)
1931{
1932        return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1933}
1934
1935static int git_blame_config(const char *var, const char *value, void *cb)
1936{
1937        if (!strcmp(var, "blame.showroot")) {
1938                show_root = git_config_bool(var, value);
1939                return 0;
1940        }
1941        if (!strcmp(var, "blame.blankboundary")) {
1942                blank_boundary = git_config_bool(var, value);
1943                return 0;
1944        }
1945        if (!strcmp(var, "blame.date")) {
1946                if (!value)
1947                        return config_error_nonbool(var);
1948                blame_date_mode = parse_date_format(value);
1949                return 0;
1950        }
1951
1952        if (userdiff_config(var, value) < 0)
1953                return -1;
1954
1955        return git_default_config(var, value, cb);
1956}
1957
1958static void verify_working_tree_path(struct commit *work_tree, const char *path)
1959{
1960        struct commit_list *parents;
1961
1962        for (parents = work_tree->parents; parents; parents = parents->next) {
1963                const unsigned char *commit_sha1 = parents->item->object.sha1;
1964                unsigned char blob_sha1[20];
1965                unsigned mode;
1966
1967                if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
1968                    sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
1969                        return;
1970        }
1971        die("no such path '%s' in HEAD", path);
1972}
1973
1974static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
1975{
1976        struct commit *parent;
1977
1978        parent = lookup_commit_reference(sha1);
1979        if (!parent)
1980                die("no such commit %s", sha1_to_hex(sha1));
1981        return &commit_list_insert(parent, tail)->next;
1982}
1983
1984static void append_merge_parents(struct commit_list **tail)
1985{
1986        int merge_head;
1987        const char *merge_head_file = git_path("MERGE_HEAD");
1988        struct strbuf line = STRBUF_INIT;
1989
1990        merge_head = open(merge_head_file, O_RDONLY);
1991        if (merge_head < 0) {
1992                if (errno == ENOENT)
1993                        return;
1994                die("cannot open '%s' for reading", merge_head_file);
1995        }
1996
1997        while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
1998                unsigned char sha1[20];
1999                if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2000                        die("unknown line in '%s': %s", merge_head_file, line.buf);
2001                tail = append_parent(tail, sha1);
2002        }
2003        close(merge_head);
2004        strbuf_release(&line);
2005}
2006
2007/*
2008 * Prepare a dummy commit that represents the work tree (or staged) item.
2009 * Note that annotating work tree item never works in the reverse.
2010 */
2011static struct commit *fake_working_tree_commit(struct diff_options *opt,
2012                                               const char *path,
2013                                               const char *contents_from)
2014{
2015        struct commit *commit;
2016        struct origin *origin;
2017        struct commit_list **parent_tail, *parent;
2018        unsigned char head_sha1[20];
2019        struct strbuf buf = STRBUF_INIT;
2020        const char *ident;
2021        time_t now;
2022        int size, len;
2023        struct cache_entry *ce;
2024        unsigned mode;
2025        struct strbuf msg = STRBUF_INIT;
2026
2027        time(&now);
2028        commit = xcalloc(1, sizeof(*commit));
2029        commit->object.parsed = 1;
2030        commit->date = now;
2031        commit->object.type = OBJ_COMMIT;
2032        parent_tail = &commit->parents;
2033
2034        if (!resolve_ref_unsafe("HEAD", head_sha1, 1, NULL))
2035                die("no such ref: HEAD");
2036
2037        parent_tail = append_parent(parent_tail, head_sha1);
2038        append_merge_parents(parent_tail);
2039        verify_working_tree_path(commit, path);
2040
2041        origin = make_origin(commit, path);
2042
2043        ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2044        strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2045        for (parent = commit->parents; parent; parent = parent->next)
2046                strbuf_addf(&msg, "parent %s\n",
2047                            sha1_to_hex(parent->item->object.sha1));
2048        strbuf_addf(&msg,
2049                    "author %s\n"
2050                    "committer %s\n\n"
2051                    "Version of %s from %s\n",
2052                    ident, ident, path,
2053                    (!contents_from ? path :
2054                     (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2055        commit->buffer = strbuf_detach(&msg, NULL);
2056
2057        if (!contents_from || strcmp("-", contents_from)) {
2058                struct stat st;
2059                const char *read_from;
2060                char *buf_ptr;
2061                unsigned long buf_len;
2062
2063                if (contents_from) {
2064                        if (stat(contents_from, &st) < 0)
2065                                die_errno("Cannot stat '%s'", contents_from);
2066                        read_from = contents_from;
2067                }
2068                else {
2069                        if (lstat(path, &st) < 0)
2070                                die_errno("Cannot lstat '%s'", path);
2071                        read_from = path;
2072                }
2073                mode = canon_mode(st.st_mode);
2074
2075                switch (st.st_mode & S_IFMT) {
2076                case S_IFREG:
2077                        if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2078                            textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2079                                strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2080                        else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2081                                die_errno("cannot open or read '%s'", read_from);
2082                        break;
2083                case S_IFLNK:
2084                        if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2085                                die_errno("cannot readlink '%s'", read_from);
2086                        break;
2087                default:
2088                        die("unsupported file type %s", read_from);
2089                }
2090        }
2091        else {
2092                /* Reading from stdin */
2093                mode = 0;
2094                if (strbuf_read(&buf, 0, 0) < 0)
2095                        die_errno("failed to read from stdin");
2096        }
2097        convert_to_git(path, buf.buf, buf.len, &buf, 0);
2098        origin->file.ptr = buf.buf;
2099        origin->file.size = buf.len;
2100        pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2101        commit->util = origin;
2102
2103        /*
2104         * Read the current index, replace the path entry with
2105         * origin->blob_sha1 without mucking with its mode or type
2106         * bits; we are not going to write this index out -- we just
2107         * want to run "diff-index --cached".
2108         */
2109        discard_cache();
2110        read_cache();
2111
2112        len = strlen(path);
2113        if (!mode) {
2114                int pos = cache_name_pos(path, len);
2115                if (0 <= pos)
2116                        mode = active_cache[pos]->ce_mode;
2117                else
2118                        /* Let's not bother reading from HEAD tree */
2119                        mode = S_IFREG | 0644;
2120        }
2121        size = cache_entry_size(len);
2122        ce = xcalloc(1, size);
2123        hashcpy(ce->sha1, origin->blob_sha1);
2124        memcpy(ce->name, path, len);
2125        ce->ce_flags = create_ce_flags(0);
2126        ce->ce_namelen = len;
2127        ce->ce_mode = create_ce_mode(mode);
2128        add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2129
2130        /*
2131         * We are not going to write this out, so this does not matter
2132         * right now, but someday we might optimize diff-index --cached
2133         * with cache-tree information.
2134         */
2135        cache_tree_invalidate_path(active_cache_tree, path);
2136
2137        return commit;
2138}
2139
2140static const char *prepare_final(struct scoreboard *sb)
2141{
2142        int i;
2143        const char *final_commit_name = NULL;
2144        struct rev_info *revs = sb->revs;
2145
2146        /*
2147         * There must be one and only one positive commit in the
2148         * revs->pending array.
2149         */
2150        for (i = 0; i < revs->pending.nr; i++) {
2151                struct object *obj = revs->pending.objects[i].item;
2152                if (obj->flags & UNINTERESTING)
2153                        continue;
2154                while (obj->type == OBJ_TAG)
2155                        obj = deref_tag(obj, NULL, 0);
2156                if (obj->type != OBJ_COMMIT)
2157                        die("Non commit %s?", revs->pending.objects[i].name);
2158                if (sb->final)
2159                        die("More than one commit to dig from %s and %s?",
2160                            revs->pending.objects[i].name,
2161                            final_commit_name);
2162                sb->final = (struct commit *) obj;
2163                final_commit_name = revs->pending.objects[i].name;
2164        }
2165        return final_commit_name;
2166}
2167
2168static const char *prepare_initial(struct scoreboard *sb)
2169{
2170        int i;
2171        const char *final_commit_name = NULL;
2172        struct rev_info *revs = sb->revs;
2173
2174        /*
2175         * There must be one and only one negative commit, and it must be
2176         * the boundary.
2177         */
2178        for (i = 0; i < revs->pending.nr; i++) {
2179                struct object *obj = revs->pending.objects[i].item;
2180                if (!(obj->flags & UNINTERESTING))
2181                        continue;
2182                while (obj->type == OBJ_TAG)
2183                        obj = deref_tag(obj, NULL, 0);
2184                if (obj->type != OBJ_COMMIT)
2185                        die("Non commit %s?", revs->pending.objects[i].name);
2186                if (sb->final)
2187                        die("More than one commit to dig down to %s and %s?",
2188                            revs->pending.objects[i].name,
2189                            final_commit_name);
2190                sb->final = (struct commit *) obj;
2191                final_commit_name = revs->pending.objects[i].name;
2192        }
2193        if (!final_commit_name)
2194                die("No commit to dig down to?");
2195        return final_commit_name;
2196}
2197
2198static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2199{
2200        int *opt = option->value;
2201
2202        /*
2203         * -C enables copy from removed files;
2204         * -C -C enables copy from existing files, but only
2205         *       when blaming a new file;
2206         * -C -C -C enables copy from existing files for
2207         *          everybody
2208         */
2209        if (*opt & PICKAXE_BLAME_COPY_HARDER)
2210                *opt |= PICKAXE_BLAME_COPY_HARDEST;
2211        if (*opt & PICKAXE_BLAME_COPY)
2212                *opt |= PICKAXE_BLAME_COPY_HARDER;
2213        *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2214
2215        if (arg)
2216                blame_copy_score = parse_score(arg);
2217        return 0;
2218}
2219
2220static int blame_move_callback(const struct option *option, const char *arg, int unset)
2221{
2222        int *opt = option->value;
2223
2224        *opt |= PICKAXE_BLAME_MOVE;
2225
2226        if (arg)
2227                blame_move_score = parse_score(arg);
2228        return 0;
2229}
2230
2231int cmd_blame(int argc, const char **argv, const char *prefix)
2232{
2233        struct rev_info revs;
2234        const char *path;
2235        struct scoreboard sb;
2236        struct origin *o;
2237        struct blame_entry *ent = NULL;
2238        long dashdash_pos, lno;
2239        const char *final_commit_name = NULL;
2240        enum object_type type;
2241
2242        static struct string_list range_list;
2243        static int output_option = 0, opt = 0;
2244        static int show_stats = 0;
2245        static const char *revs_file = NULL;
2246        static const char *contents_from = NULL;
2247        static const struct option options[] = {
2248                OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2249                OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2250                OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2251                OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2252                OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2253                OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2254                OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2255                OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2256                OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2257                OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2258                OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2259                OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2260                OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2261                OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2262                OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2263                OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2264                OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2265                OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2266                { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2267                { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2268                OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2269                OPT__ABBREV(&abbrev),
2270                OPT_END()
2271        };
2272
2273        struct parse_opt_ctx_t ctx;
2274        int cmd_is_annotate = !strcmp(argv[0], "annotate");
2275        struct range_set ranges;
2276        unsigned int range_i;
2277        long anchor;
2278
2279        git_config(git_blame_config, NULL);
2280        init_revisions(&revs, NULL);
2281        revs.date_mode = blame_date_mode;
2282        DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2283        DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2284
2285        save_commit_buffer = 0;
2286        dashdash_pos = 0;
2287
2288        parse_options_start(&ctx, argc, argv, prefix, options,
2289                            PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2290        for (;;) {
2291                switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2292                case PARSE_OPT_HELP:
2293                        exit(129);
2294                case PARSE_OPT_DONE:
2295                        if (ctx.argv[0])
2296                                dashdash_pos = ctx.cpidx;
2297                        goto parse_done;
2298                }
2299
2300                if (!strcmp(ctx.argv[0], "--reverse")) {
2301                        ctx.argv[0] = "--children";
2302                        reverse = 1;
2303                }
2304                parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2305        }
2306parse_done:
2307        no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2308        DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2309        argc = parse_options_end(&ctx);
2310
2311        if (0 < abbrev)
2312                /* one more abbrev length is needed for the boundary commit */
2313                abbrev++;
2314
2315        if (revs_file && read_ancestry(revs_file))
2316                die_errno("reading graft file '%s' failed", revs_file);
2317
2318        if (cmd_is_annotate) {
2319                output_option |= OUTPUT_ANNOTATE_COMPAT;
2320                blame_date_mode = DATE_ISO8601;
2321        } else {
2322                blame_date_mode = revs.date_mode;
2323        }
2324
2325        /* The maximum width used to show the dates */
2326        switch (blame_date_mode) {
2327        case DATE_RFC2822:
2328                blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2329                break;
2330        case DATE_ISO8601:
2331                blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2332                break;
2333        case DATE_RAW:
2334                blame_date_width = sizeof("1161298804 -0700");
2335                break;
2336        case DATE_SHORT:
2337                blame_date_width = sizeof("2006-10-19");
2338                break;
2339        case DATE_RELATIVE:
2340                /* "normal" is used as the fallback for "relative" */
2341        case DATE_LOCAL:
2342        case DATE_NORMAL:
2343                blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2344                break;
2345        }
2346        blame_date_width -= 1; /* strip the null */
2347
2348        if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2349                opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2350                        PICKAXE_BLAME_COPY_HARDER);
2351
2352        if (!blame_move_score)
2353                blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2354        if (!blame_copy_score)
2355                blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2356
2357        /*
2358         * We have collected options unknown to us in argv[1..unk]
2359         * which are to be passed to revision machinery if we are
2360         * going to do the "bottom" processing.
2361         *
2362         * The remaining are:
2363         *
2364         * (1) if dashdash_pos != 0, it is either
2365         *     "blame [revisions] -- <path>" or
2366         *     "blame -- <path> <rev>"
2367         *
2368         * (2) otherwise, it is one of the two:
2369         *     "blame [revisions] <path>"
2370         *     "blame <path> <rev>"
2371         *
2372         * Note that we must strip out <path> from the arguments: we do not
2373         * want the path pruning but we may want "bottom" processing.
2374         */
2375        if (dashdash_pos) {
2376                switch (argc - dashdash_pos - 1) {
2377                case 2: /* (1b) */
2378                        if (argc != 4)
2379                                usage_with_options(blame_opt_usage, options);
2380                        /* reorder for the new way: <rev> -- <path> */
2381                        argv[1] = argv[3];
2382                        argv[3] = argv[2];
2383                        argv[2] = "--";
2384                        /* FALLTHROUGH */
2385                case 1: /* (1a) */
2386                        path = add_prefix(prefix, argv[--argc]);
2387                        argv[argc] = NULL;
2388                        break;
2389                default:
2390                        usage_with_options(blame_opt_usage, options);
2391                }
2392        } else {
2393                if (argc < 2)
2394                        usage_with_options(blame_opt_usage, options);
2395                path = add_prefix(prefix, argv[argc - 1]);
2396                if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2397                        path = add_prefix(prefix, argv[1]);
2398                        argv[1] = argv[2];
2399                }
2400                argv[argc - 1] = "--";
2401
2402                setup_work_tree();
2403                if (!has_string_in_work_tree(path))
2404                        die_errno("cannot stat path '%s'", path);
2405        }
2406
2407        revs.disable_stdin = 1;
2408        setup_revisions(argc, argv, &revs, NULL);
2409        memset(&sb, 0, sizeof(sb));
2410
2411        sb.revs = &revs;
2412        if (!reverse)
2413                final_commit_name = prepare_final(&sb);
2414        else if (contents_from)
2415                die("--contents and --children do not blend well.");
2416        else
2417                final_commit_name = prepare_initial(&sb);
2418
2419        if (!sb.final) {
2420                /*
2421                 * "--not A B -- path" without anything positive;
2422                 * do not default to HEAD, but use the working tree
2423                 * or "--contents".
2424                 */
2425                setup_work_tree();
2426                sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2427                                                    path, contents_from);
2428                add_pending_object(&revs, &(sb.final->object), ":");
2429        }
2430        else if (contents_from)
2431                die("Cannot use --contents with final commit object name");
2432
2433        /*
2434         * If we have bottom, this will mark the ancestors of the
2435         * bottom commits we would reach while traversing as
2436         * uninteresting.
2437         */
2438        if (prepare_revision_walk(&revs))
2439                die("revision walk setup failed");
2440
2441        if (is_null_sha1(sb.final->object.sha1)) {
2442                char *buf;
2443                o = sb.final->util;
2444                buf = xmalloc(o->file.size + 1);
2445                memcpy(buf, o->file.ptr, o->file.size + 1);
2446                sb.final_buf = buf;
2447                sb.final_buf_size = o->file.size;
2448        }
2449        else {
2450                o = get_origin(&sb, sb.final, path);
2451                if (fill_blob_sha1_and_mode(o))
2452                        die("no such path %s in %s", path, final_commit_name);
2453
2454                if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2455                    textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2456                                    &sb.final_buf_size))
2457                        ;
2458                else
2459                        sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2460                                                      &sb.final_buf_size);
2461
2462                if (!sb.final_buf)
2463                        die("Cannot read blob %s for path %s",
2464                            sha1_to_hex(o->blob_sha1),
2465                            path);
2466        }
2467        num_read_blob++;
2468        lno = prepare_lines(&sb);
2469
2470        if (lno && !range_list.nr)
2471                string_list_append(&range_list, xstrdup("1"));
2472
2473        anchor = 1;
2474        range_set_init(&ranges, range_list.nr);
2475        for (range_i = 0; range_i < range_list.nr; ++range_i) {
2476                long bottom, top;
2477                if (parse_range_arg(range_list.items[range_i].string,
2478                                    nth_line_cb, &sb, lno, anchor,
2479                                    &bottom, &top, sb.path))
2480                        usage(blame_usage);
2481                if (lno < top || ((lno || bottom) && lno < bottom))
2482                        die("file %s has only %lu lines", path, lno);
2483                if (bottom < 1)
2484                        bottom = 1;
2485                if (top < 1)
2486                        top = lno;
2487                bottom--;
2488                range_set_append_unsafe(&ranges, bottom, top);
2489                anchor = top + 1;
2490        }
2491        sort_and_merge_range_set(&ranges);
2492
2493        for (range_i = ranges.nr; range_i > 0; --range_i) {
2494                const struct range *r = &ranges.ranges[range_i - 1];
2495                long bottom = r->start;
2496                long top = r->end;
2497                struct blame_entry *next = ent;
2498                ent = xcalloc(1, sizeof(*ent));
2499                ent->lno = bottom;
2500                ent->num_lines = top - bottom;
2501                ent->suspect = o;
2502                ent->s_lno = bottom;
2503                ent->next = next;
2504                if (next)
2505                        next->prev = ent;
2506                origin_incref(o);
2507        }
2508        origin_decref(o);
2509
2510        range_set_release(&ranges);
2511        string_list_clear(&range_list, 0);
2512
2513        sb.ent = ent;
2514        sb.path = path;
2515
2516        read_mailmap(&mailmap, NULL);
2517
2518        if (!incremental)
2519                setup_pager();
2520
2521        assign_blame(&sb, opt);
2522
2523        if (incremental)
2524                return 0;
2525
2526        coalesce(&sb);
2527
2528        if (!(output_option & OUTPUT_PORCELAIN))
2529                find_alignment(&sb, &output_option);
2530
2531        output(&sb, output_option);
2532        free((void *)sb.final_buf);
2533        for (ent = sb.ent; ent; ) {
2534                struct blame_entry *e = ent->next;
2535                free(ent);
2536                ent = e;
2537        }
2538
2539        if (show_stats) {
2540                printf("num read blob: %d\n", num_read_blob);
2541                printf("num get patch: %d\n", num_get_patch);
2542                printf("num commits: %d\n", num_commits);
2543        }
2544        return 0;
2545}