builtin / blame.con commit Merge branch 'js/no-curl-easy-strerror-on-old-curl' (c1310be)
   1/*
   2 * Blame
   3 *
   4 * Copyright (c) 2006, Junio C Hamano
   5 */
   6
   7#include "cache.h"
   8#include "builtin.h"
   9#include "blob.h"
  10#include "commit.h"
  11#include "tag.h"
  12#include "tree-walk.h"
  13#include "diff.h"
  14#include "diffcore.h"
  15#include "revision.h"
  16#include "quote.h"
  17#include "xdiff-interface.h"
  18#include "cache-tree.h"
  19#include "string-list.h"
  20#include "mailmap.h"
  21#include "parse-options.h"
  22#include "utf8.h"
  23#include "userdiff.h"
  24
  25static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
  26
  27static const char *blame_opt_usage[] = {
  28        blame_usage,
  29        "",
  30        "[rev-opts] are documented in git-rev-list(1)",
  31        NULL
  32};
  33
  34static int longest_file;
  35static int longest_author;
  36static int max_orig_digits;
  37static int max_digits;
  38static int max_score_digits;
  39static int show_root;
  40static int reverse;
  41static int blank_boundary;
  42static int incremental;
  43static int xdl_opts;
  44static int abbrev = -1;
  45
  46static enum date_mode blame_date_mode = DATE_ISO8601;
  47static size_t blame_date_width;
  48
  49static struct string_list mailmap;
  50
  51#ifndef DEBUG
  52#define DEBUG 0
  53#endif
  54
  55/* stats */
  56static int num_read_blob;
  57static int num_get_patch;
  58static int num_commits;
  59
  60#define PICKAXE_BLAME_MOVE              01
  61#define PICKAXE_BLAME_COPY              02
  62#define PICKAXE_BLAME_COPY_HARDER       04
  63#define PICKAXE_BLAME_COPY_HARDEST      010
  64
  65/*
  66 * blame for a blame_entry with score lower than these thresholds
  67 * is not passed to the parent using move/copy logic.
  68 */
  69static unsigned blame_move_score;
  70static unsigned blame_copy_score;
  71#define BLAME_DEFAULT_MOVE_SCORE        20
  72#define BLAME_DEFAULT_COPY_SCORE        40
  73
  74/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
  75#define METAINFO_SHOWN          (1u<<12)
  76#define MORE_THAN_ONE_PATH      (1u<<13)
  77
  78/*
  79 * One blob in a commit that is being suspected
  80 */
  81struct origin {
  82        int refcnt;
  83        struct origin *previous;
  84        struct commit *commit;
  85        mmfile_t file;
  86        unsigned char blob_sha1[20];
  87        unsigned mode;
  88        char path[FLEX_ARRAY];
  89};
  90
  91static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
  92                      xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
  93{
  94        xpparam_t xpp = {0};
  95        xdemitconf_t xecfg = {0};
  96        xdemitcb_t ecb = {NULL};
  97
  98        xpp.flags = xdl_opts;
  99        xecfg.ctxlen = ctxlen;
 100        xecfg.hunk_func = hunk_func;
 101        ecb.priv = cb_data;
 102        return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
 103}
 104
 105/*
 106 * Prepare diff_filespec and convert it using diff textconv API
 107 * if the textconv driver exists.
 108 * Return 1 if the conversion succeeds, 0 otherwise.
 109 */
 110int textconv_object(const char *path,
 111                    unsigned mode,
 112                    const unsigned char *sha1,
 113                    int sha1_valid,
 114                    char **buf,
 115                    unsigned long *buf_size)
 116{
 117        struct diff_filespec *df;
 118        struct userdiff_driver *textconv;
 119
 120        df = alloc_filespec(path);
 121        fill_filespec(df, sha1, sha1_valid, mode);
 122        textconv = get_textconv(df);
 123        if (!textconv) {
 124                free_filespec(df);
 125                return 0;
 126        }
 127
 128        *buf_size = fill_textconv(textconv, df, buf);
 129        free_filespec(df);
 130        return 1;
 131}
 132
 133/*
 134 * Given an origin, prepare mmfile_t structure to be used by the
 135 * diff machinery
 136 */
 137static void fill_origin_blob(struct diff_options *opt,
 138                             struct origin *o, mmfile_t *file)
 139{
 140        if (!o->file.ptr) {
 141                enum object_type type;
 142                unsigned long file_size;
 143
 144                num_read_blob++;
 145                if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
 146                    textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
 147                        ;
 148                else
 149                        file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
 150                file->size = file_size;
 151
 152                if (!file->ptr)
 153                        die("Cannot read blob %s for path %s",
 154                            sha1_to_hex(o->blob_sha1),
 155                            o->path);
 156                o->file = *file;
 157        }
 158        else
 159                *file = o->file;
 160}
 161
 162/*
 163 * Origin is refcounted and usually we keep the blob contents to be
 164 * reused.
 165 */
 166static inline struct origin *origin_incref(struct origin *o)
 167{
 168        if (o)
 169                o->refcnt++;
 170        return o;
 171}
 172
 173static void origin_decref(struct origin *o)
 174{
 175        if (o && --o->refcnt <= 0) {
 176                if (o->previous)
 177                        origin_decref(o->previous);
 178                free(o->file.ptr);
 179                free(o);
 180        }
 181}
 182
 183static void drop_origin_blob(struct origin *o)
 184{
 185        if (o->file.ptr) {
 186                free(o->file.ptr);
 187                o->file.ptr = NULL;
 188        }
 189}
 190
 191/*
 192 * Each group of lines is described by a blame_entry; it can be split
 193 * as we pass blame to the parents.  They form a linked list in the
 194 * scoreboard structure, sorted by the target line number.
 195 */
 196struct blame_entry {
 197        struct blame_entry *prev;
 198        struct blame_entry *next;
 199
 200        /* the first line of this group in the final image;
 201         * internally all line numbers are 0 based.
 202         */
 203        int lno;
 204
 205        /* how many lines this group has */
 206        int num_lines;
 207
 208        /* the commit that introduced this group into the final image */
 209        struct origin *suspect;
 210
 211        /* true if the suspect is truly guilty; false while we have not
 212         * checked if the group came from one of its parents.
 213         */
 214        char guilty;
 215
 216        /* true if the entry has been scanned for copies in the current parent
 217         */
 218        char scanned;
 219
 220        /* the line number of the first line of this group in the
 221         * suspect's file; internally all line numbers are 0 based.
 222         */
 223        int s_lno;
 224
 225        /* how significant this entry is -- cached to avoid
 226         * scanning the lines over and over.
 227         */
 228        unsigned score;
 229};
 230
 231/*
 232 * The current state of the blame assignment.
 233 */
 234struct scoreboard {
 235        /* the final commit (i.e. where we started digging from) */
 236        struct commit *final;
 237        struct rev_info *revs;
 238        const char *path;
 239
 240        /*
 241         * The contents in the final image.
 242         * Used by many functions to obtain contents of the nth line,
 243         * indexed with scoreboard.lineno[blame_entry.lno].
 244         */
 245        const char *final_buf;
 246        unsigned long final_buf_size;
 247
 248        /* linked list of blames */
 249        struct blame_entry *ent;
 250
 251        /* look-up a line in the final buffer */
 252        int num_lines;
 253        int *lineno;
 254};
 255
 256static inline int same_suspect(struct origin *a, struct origin *b)
 257{
 258        if (a == b)
 259                return 1;
 260        if (a->commit != b->commit)
 261                return 0;
 262        return !strcmp(a->path, b->path);
 263}
 264
 265static void sanity_check_refcnt(struct scoreboard *);
 266
 267/*
 268 * If two blame entries that are next to each other came from
 269 * contiguous lines in the same origin (i.e. <commit, path> pair),
 270 * merge them together.
 271 */
 272static void coalesce(struct scoreboard *sb)
 273{
 274        struct blame_entry *ent, *next;
 275
 276        for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 277                if (same_suspect(ent->suspect, next->suspect) &&
 278                    ent->guilty == next->guilty &&
 279                    ent->s_lno + ent->num_lines == next->s_lno) {
 280                        ent->num_lines += next->num_lines;
 281                        ent->next = next->next;
 282                        if (ent->next)
 283                                ent->next->prev = ent;
 284                        origin_decref(next->suspect);
 285                        free(next);
 286                        ent->score = 0;
 287                        next = ent; /* again */
 288                }
 289        }
 290
 291        if (DEBUG) /* sanity */
 292                sanity_check_refcnt(sb);
 293}
 294
 295/*
 296 * Given a commit and a path in it, create a new origin structure.
 297 * The callers that add blame to the scoreboard should use
 298 * get_origin() to obtain shared, refcounted copy instead of calling
 299 * this function directly.
 300 */
 301static struct origin *make_origin(struct commit *commit, const char *path)
 302{
 303        struct origin *o;
 304        o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 305        o->commit = commit;
 306        o->refcnt = 1;
 307        strcpy(o->path, path);
 308        return o;
 309}
 310
 311/*
 312 * Locate an existing origin or create a new one.
 313 */
 314static struct origin *get_origin(struct scoreboard *sb,
 315                                 struct commit *commit,
 316                                 const char *path)
 317{
 318        struct blame_entry *e;
 319
 320        for (e = sb->ent; e; e = e->next) {
 321                if (e->suspect->commit == commit &&
 322                    !strcmp(e->suspect->path, path))
 323                        return origin_incref(e->suspect);
 324        }
 325        return make_origin(commit, path);
 326}
 327
 328/*
 329 * Fill the blob_sha1 field of an origin if it hasn't, so that later
 330 * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 331 * for an origin is also used to pass the blame for the entire file to
 332 * the parent to detect the case where a child's blob is identical to
 333 * that of its parent's.
 334 *
 335 * This also fills origin->mode for corresponding tree path.
 336 */
 337static int fill_blob_sha1_and_mode(struct origin *origin)
 338{
 339        if (!is_null_sha1(origin->blob_sha1))
 340                return 0;
 341        if (get_tree_entry(origin->commit->object.sha1,
 342                           origin->path,
 343                           origin->blob_sha1, &origin->mode))
 344                goto error_out;
 345        if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 346                goto error_out;
 347        return 0;
 348 error_out:
 349        hashclr(origin->blob_sha1);
 350        origin->mode = S_IFINVALID;
 351        return -1;
 352}
 353
 354/*
 355 * We have an origin -- check if the same path exists in the
 356 * parent and return an origin structure to represent it.
 357 */
 358static struct origin *find_origin(struct scoreboard *sb,
 359                                  struct commit *parent,
 360                                  struct origin *origin)
 361{
 362        struct origin *porigin = NULL;
 363        struct diff_options diff_opts;
 364        const char *paths[2];
 365
 366        if (parent->util) {
 367                /*
 368                 * Each commit object can cache one origin in that
 369                 * commit.  This is a freestanding copy of origin and
 370                 * not refcounted.
 371                 */
 372                struct origin *cached = parent->util;
 373                if (!strcmp(cached->path, origin->path)) {
 374                        /*
 375                         * The same path between origin and its parent
 376                         * without renaming -- the most common case.
 377                         */
 378                        porigin = get_origin(sb, parent, cached->path);
 379
 380                        /*
 381                         * If the origin was newly created (i.e. get_origin
 382                         * would call make_origin if none is found in the
 383                         * scoreboard), it does not know the blob_sha1/mode,
 384                         * so copy it.  Otherwise porigin was in the
 385                         * scoreboard and already knows blob_sha1/mode.
 386                         */
 387                        if (porigin->refcnt == 1) {
 388                                hashcpy(porigin->blob_sha1, cached->blob_sha1);
 389                                porigin->mode = cached->mode;
 390                        }
 391                        return porigin;
 392                }
 393                /* otherwise it was not very useful; free it */
 394                free(parent->util);
 395                parent->util = NULL;
 396        }
 397
 398        /* See if the origin->path is different between parent
 399         * and origin first.  Most of the time they are the
 400         * same and diff-tree is fairly efficient about this.
 401         */
 402        diff_setup(&diff_opts);
 403        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 404        diff_opts.detect_rename = 0;
 405        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 406        paths[0] = origin->path;
 407        paths[1] = NULL;
 408
 409        diff_tree_setup_paths(paths, &diff_opts);
 410        diff_setup_done(&diff_opts);
 411
 412        if (is_null_sha1(origin->commit->object.sha1))
 413                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 414        else
 415                diff_tree_sha1(parent->tree->object.sha1,
 416                               origin->commit->tree->object.sha1,
 417                               "", &diff_opts);
 418        diffcore_std(&diff_opts);
 419
 420        if (!diff_queued_diff.nr) {
 421                /* The path is the same as parent */
 422                porigin = get_origin(sb, parent, origin->path);
 423                hashcpy(porigin->blob_sha1, origin->blob_sha1);
 424                porigin->mode = origin->mode;
 425        } else {
 426                /*
 427                 * Since origin->path is a pathspec, if the parent
 428                 * commit had it as a directory, we will see a whole
 429                 * bunch of deletion of files in the directory that we
 430                 * do not care about.
 431                 */
 432                int i;
 433                struct diff_filepair *p = NULL;
 434                for (i = 0; i < diff_queued_diff.nr; i++) {
 435                        const char *name;
 436                        p = diff_queued_diff.queue[i];
 437                        name = p->one->path ? p->one->path : p->two->path;
 438                        if (!strcmp(name, origin->path))
 439                                break;
 440                }
 441                if (!p)
 442                        die("internal error in blame::find_origin");
 443                switch (p->status) {
 444                default:
 445                        die("internal error in blame::find_origin (%c)",
 446                            p->status);
 447                case 'M':
 448                        porigin = get_origin(sb, parent, origin->path);
 449                        hashcpy(porigin->blob_sha1, p->one->sha1);
 450                        porigin->mode = p->one->mode;
 451                        break;
 452                case 'A':
 453                case 'T':
 454                        /* Did not exist in parent, or type changed */
 455                        break;
 456                }
 457        }
 458        diff_flush(&diff_opts);
 459        diff_tree_release_paths(&diff_opts);
 460        if (porigin) {
 461                /*
 462                 * Create a freestanding copy that is not part of
 463                 * the refcounted origin found in the scoreboard, and
 464                 * cache it in the commit.
 465                 */
 466                struct origin *cached;
 467
 468                cached = make_origin(porigin->commit, porigin->path);
 469                hashcpy(cached->blob_sha1, porigin->blob_sha1);
 470                cached->mode = porigin->mode;
 471                parent->util = cached;
 472        }
 473        return porigin;
 474}
 475
 476/*
 477 * We have an origin -- find the path that corresponds to it in its
 478 * parent and return an origin structure to represent it.
 479 */
 480static struct origin *find_rename(struct scoreboard *sb,
 481                                  struct commit *parent,
 482                                  struct origin *origin)
 483{
 484        struct origin *porigin = NULL;
 485        struct diff_options diff_opts;
 486        int i;
 487        const char *paths[2];
 488
 489        diff_setup(&diff_opts);
 490        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 491        diff_opts.detect_rename = DIFF_DETECT_RENAME;
 492        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 493        diff_opts.single_follow = origin->path;
 494        paths[0] = NULL;
 495        diff_tree_setup_paths(paths, &diff_opts);
 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        diff_tree_release_paths(&diff_opts);
 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, int lno)
 569{
 570        return sb->final_buf + sb->lineno[lno];
 571}
 572
 573/*
 574 * It is known that lines between tlno to same came from parent, and e
 575 * has an overlap with that range.  it also is known that parent's
 576 * line plno corresponds to e's line tlno.
 577 *
 578 *                <---- e ----->
 579 *                   <------>
 580 *                   <------------>
 581 *             <------------>
 582 *             <------------------>
 583 *
 584 * Split e into potentially three parts; before this chunk, the chunk
 585 * to be blamed for the parent, and after that portion.
 586 */
 587static void split_overlap(struct blame_entry *split,
 588                          struct blame_entry *e,
 589                          int tlno, int plno, int same,
 590                          struct origin *parent)
 591{
 592        int chunk_end_lno;
 593        memset(split, 0, sizeof(struct blame_entry [3]));
 594
 595        if (e->s_lno < tlno) {
 596                /* there is a pre-chunk part not blamed on parent */
 597                split[0].suspect = origin_incref(e->suspect);
 598                split[0].lno = e->lno;
 599                split[0].s_lno = e->s_lno;
 600                split[0].num_lines = tlno - e->s_lno;
 601                split[1].lno = e->lno + tlno - e->s_lno;
 602                split[1].s_lno = plno;
 603        }
 604        else {
 605                split[1].lno = e->lno;
 606                split[1].s_lno = plno + (e->s_lno - tlno);
 607        }
 608
 609        if (same < e->s_lno + e->num_lines) {
 610                /* there is a post-chunk part not blamed on parent */
 611                split[2].suspect = origin_incref(e->suspect);
 612                split[2].lno = e->lno + (same - e->s_lno);
 613                split[2].s_lno = e->s_lno + (same - e->s_lno);
 614                split[2].num_lines = e->s_lno + e->num_lines - same;
 615                chunk_end_lno = split[2].lno;
 616        }
 617        else
 618                chunk_end_lno = e->lno + e->num_lines;
 619        split[1].num_lines = chunk_end_lno - split[1].lno;
 620
 621        /*
 622         * if it turns out there is nothing to blame the parent for,
 623         * forget about the splitting.  !split[1].suspect signals this.
 624         */
 625        if (split[1].num_lines < 1)
 626                return;
 627        split[1].suspect = origin_incref(parent);
 628}
 629
 630/*
 631 * split_overlap() divided an existing blame e into up to three parts
 632 * in split.  Adjust the linked list of blames in the scoreboard to
 633 * reflect the split.
 634 */
 635static void split_blame(struct scoreboard *sb,
 636                        struct blame_entry *split,
 637                        struct blame_entry *e)
 638{
 639        struct blame_entry *new_entry;
 640
 641        if (split[0].suspect && split[2].suspect) {
 642                /* The first part (reuse storage for the existing entry e) */
 643                dup_entry(e, &split[0]);
 644
 645                /* The last part -- me */
 646                new_entry = xmalloc(sizeof(*new_entry));
 647                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 648                add_blame_entry(sb, new_entry);
 649
 650                /* ... and the middle part -- parent */
 651                new_entry = xmalloc(sizeof(*new_entry));
 652                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 653                add_blame_entry(sb, new_entry);
 654        }
 655        else if (!split[0].suspect && !split[2].suspect)
 656                /*
 657                 * The parent covers the entire area; reuse storage for
 658                 * e and replace it with the parent.
 659                 */
 660                dup_entry(e, &split[1]);
 661        else if (split[0].suspect) {
 662                /* me and then parent */
 663                dup_entry(e, &split[0]);
 664
 665                new_entry = xmalloc(sizeof(*new_entry));
 666                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 667                add_blame_entry(sb, new_entry);
 668        }
 669        else {
 670                /* parent and then me */
 671                dup_entry(e, &split[1]);
 672
 673                new_entry = xmalloc(sizeof(*new_entry));
 674                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 675                add_blame_entry(sb, new_entry);
 676        }
 677
 678        if (DEBUG) { /* sanity */
 679                struct blame_entry *ent;
 680                int lno = sb->ent->lno, corrupt = 0;
 681
 682                for (ent = sb->ent; ent; ent = ent->next) {
 683                        if (lno != ent->lno)
 684                                corrupt = 1;
 685                        if (ent->s_lno < 0)
 686                                corrupt = 1;
 687                        lno += ent->num_lines;
 688                }
 689                if (corrupt) {
 690                        lno = sb->ent->lno;
 691                        for (ent = sb->ent; ent; ent = ent->next) {
 692                                printf("L %8d l %8d n %8d\n",
 693                                       lno, ent->lno, ent->num_lines);
 694                                lno = ent->lno + ent->num_lines;
 695                        }
 696                        die("oops");
 697                }
 698        }
 699}
 700
 701/*
 702 * After splitting the blame, the origins used by the
 703 * on-stack blame_entry should lose one refcnt each.
 704 */
 705static void decref_split(struct blame_entry *split)
 706{
 707        int i;
 708
 709        for (i = 0; i < 3; i++)
 710                origin_decref(split[i].suspect);
 711}
 712
 713/*
 714 * Helper for blame_chunk().  blame_entry e is known to overlap with
 715 * the patch hunk; split it and pass blame to the parent.
 716 */
 717static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 718                          int tlno, int plno, int same,
 719                          struct origin *parent)
 720{
 721        struct blame_entry split[3];
 722
 723        split_overlap(split, e, tlno, plno, same, parent);
 724        if (split[1].suspect)
 725                split_blame(sb, split, e);
 726        decref_split(split);
 727}
 728
 729/*
 730 * Find the line number of the last line the target is suspected for.
 731 */
 732static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 733{
 734        struct blame_entry *e;
 735        int last_in_target = -1;
 736
 737        for (e = sb->ent; e; e = e->next) {
 738                if (e->guilty || !same_suspect(e->suspect, target))
 739                        continue;
 740                if (last_in_target < e->s_lno + e->num_lines)
 741                        last_in_target = e->s_lno + e->num_lines;
 742        }
 743        return last_in_target;
 744}
 745
 746/*
 747 * Process one hunk from the patch between the current suspect for
 748 * blame_entry e and its parent.  Find and split the overlap, and
 749 * pass blame to the overlapping part to the parent.
 750 */
 751static void blame_chunk(struct scoreboard *sb,
 752                        int tlno, int plno, int same,
 753                        struct origin *target, struct origin *parent)
 754{
 755        struct blame_entry *e;
 756
 757        for (e = sb->ent; e; e = e->next) {
 758                if (e->guilty || !same_suspect(e->suspect, target))
 759                        continue;
 760                if (same <= e->s_lno)
 761                        continue;
 762                if (tlno < e->s_lno + e->num_lines)
 763                        blame_overlap(sb, e, tlno, plno, same, parent);
 764        }
 765}
 766
 767struct blame_chunk_cb_data {
 768        struct scoreboard *sb;
 769        struct origin *target;
 770        struct origin *parent;
 771        long plno;
 772        long tlno;
 773};
 774
 775static int blame_chunk_cb(long start_a, long count_a,
 776                          long start_b, long count_b, void *data)
 777{
 778        struct blame_chunk_cb_data *d = data;
 779        blame_chunk(d->sb, d->tlno, d->plno, start_b, d->target, d->parent);
 780        d->plno = start_a + count_a;
 781        d->tlno = start_b + count_b;
 782        return 0;
 783}
 784
 785/*
 786 * We are looking at the origin 'target' and aiming to pass blame
 787 * for the lines it is suspected to its parent.  Run diff to find
 788 * which lines came from parent and pass blame for them.
 789 */
 790static int pass_blame_to_parent(struct scoreboard *sb,
 791                                struct origin *target,
 792                                struct origin *parent)
 793{
 794        int last_in_target;
 795        mmfile_t file_p, file_o;
 796        struct blame_chunk_cb_data d;
 797
 798        memset(&d, 0, sizeof(d));
 799        d.sb = sb; d.target = target; d.parent = parent;
 800        last_in_target = find_last_in_target(sb, target);
 801        if (last_in_target < 0)
 802                return 1; /* nothing remains for this target */
 803
 804        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 805        fill_origin_blob(&sb->revs->diffopt, target, &file_o);
 806        num_get_patch++;
 807
 808        diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
 809        /* The rest (i.e. anything after tlno) are the same as the parent */
 810        blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
 811
 812        return 0;
 813}
 814
 815/*
 816 * The lines in blame_entry after splitting blames many times can become
 817 * very small and trivial, and at some point it becomes pointless to
 818 * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 819 * ordinary C program, and it is not worth to say it was copied from
 820 * totally unrelated file in the parent.
 821 *
 822 * Compute how trivial the lines in the blame_entry are.
 823 */
 824static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 825{
 826        unsigned score;
 827        const char *cp, *ep;
 828
 829        if (e->score)
 830                return e->score;
 831
 832        score = 1;
 833        cp = nth_line(sb, e->lno);
 834        ep = nth_line(sb, e->lno + e->num_lines);
 835        while (cp < ep) {
 836                unsigned ch = *((unsigned char *)cp);
 837                if (isalnum(ch))
 838                        score++;
 839                cp++;
 840        }
 841        e->score = score;
 842        return score;
 843}
 844
 845/*
 846 * best_so_far[] and this[] are both a split of an existing blame_entry
 847 * that passes blame to the parent.  Maintain best_so_far the best split
 848 * so far, by comparing this and best_so_far and copying this into
 849 * bst_so_far as needed.
 850 */
 851static void copy_split_if_better(struct scoreboard *sb,
 852                                 struct blame_entry *best_so_far,
 853                                 struct blame_entry *this)
 854{
 855        int i;
 856
 857        if (!this[1].suspect)
 858                return;
 859        if (best_so_far[1].suspect) {
 860                if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 861                        return;
 862        }
 863
 864        for (i = 0; i < 3; i++)
 865                origin_incref(this[i].suspect);
 866        decref_split(best_so_far);
 867        memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 868}
 869
 870/*
 871 * We are looking at a part of the final image represented by
 872 * ent (tlno and same are offset by ent->s_lno).
 873 * tlno is where we are looking at in the final image.
 874 * up to (but not including) same match preimage.
 875 * plno is where we are looking at in the preimage.
 876 *
 877 * <-------------- final image ---------------------->
 878 *       <------ent------>
 879 *         ^tlno ^same
 880 *    <---------preimage----->
 881 *         ^plno
 882 *
 883 * All line numbers are 0-based.
 884 */
 885static void handle_split(struct scoreboard *sb,
 886                         struct blame_entry *ent,
 887                         int tlno, int plno, int same,
 888                         struct origin *parent,
 889                         struct blame_entry *split)
 890{
 891        if (ent->num_lines <= tlno)
 892                return;
 893        if (tlno < same) {
 894                struct blame_entry this[3];
 895                tlno += ent->s_lno;
 896                same += ent->s_lno;
 897                split_overlap(this, ent, tlno, plno, same, parent);
 898                copy_split_if_better(sb, split, this);
 899                decref_split(this);
 900        }
 901}
 902
 903struct handle_split_cb_data {
 904        struct scoreboard *sb;
 905        struct blame_entry *ent;
 906        struct origin *parent;
 907        struct blame_entry *split;
 908        long plno;
 909        long tlno;
 910};
 911
 912static int handle_split_cb(long start_a, long count_a,
 913                           long start_b, long count_b, void *data)
 914{
 915        struct handle_split_cb_data *d = data;
 916        handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
 917                     d->split);
 918        d->plno = start_a + count_a;
 919        d->tlno = start_b + count_b;
 920        return 0;
 921}
 922
 923/*
 924 * Find the lines from parent that are the same as ent so that
 925 * we can pass blames to it.  file_p has the blob contents for
 926 * the parent.
 927 */
 928static void find_copy_in_blob(struct scoreboard *sb,
 929                              struct blame_entry *ent,
 930                              struct origin *parent,
 931                              struct blame_entry *split,
 932                              mmfile_t *file_p)
 933{
 934        const char *cp;
 935        int cnt;
 936        mmfile_t file_o;
 937        struct handle_split_cb_data d;
 938
 939        memset(&d, 0, sizeof(d));
 940        d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
 941        /*
 942         * Prepare mmfile that contains only the lines in ent.
 943         */
 944        cp = nth_line(sb, ent->lno);
 945        file_o.ptr = (char *) cp;
 946        cnt = ent->num_lines;
 947
 948        while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 949                if (*cp++ == '\n')
 950                        cnt--;
 951        }
 952        file_o.size = cp - file_o.ptr;
 953
 954        /*
 955         * file_o is a part of final image we are annotating.
 956         * file_p partially may match that image.
 957         */
 958        memset(split, 0, sizeof(struct blame_entry [3]));
 959        diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
 960        /* remainder, if any, all match the preimage */
 961        handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
 962}
 963
 964/*
 965 * See if lines currently target is suspected for can be attributed to
 966 * parent.
 967 */
 968static int find_move_in_parent(struct scoreboard *sb,
 969                               struct origin *target,
 970                               struct origin *parent)
 971{
 972        int last_in_target, made_progress;
 973        struct blame_entry *e, split[3];
 974        mmfile_t file_p;
 975
 976        last_in_target = find_last_in_target(sb, target);
 977        if (last_in_target < 0)
 978                return 1; /* nothing remains for this target */
 979
 980        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 981        if (!file_p.ptr)
 982                return 0;
 983
 984        made_progress = 1;
 985        while (made_progress) {
 986                made_progress = 0;
 987                for (e = sb->ent; e; e = e->next) {
 988                        if (e->guilty || !same_suspect(e->suspect, target) ||
 989                            ent_score(sb, e) < blame_move_score)
 990                                continue;
 991                        find_copy_in_blob(sb, e, parent, split, &file_p);
 992                        if (split[1].suspect &&
 993                            blame_move_score < ent_score(sb, &split[1])) {
 994                                split_blame(sb, split, e);
 995                                made_progress = 1;
 996                        }
 997                        decref_split(split);
 998                }
 999        }
1000        return 0;
1001}
1002
1003struct blame_list {
1004        struct blame_entry *ent;
1005        struct blame_entry split[3];
1006};
1007
1008/*
1009 * Count the number of entries the target is suspected for,
1010 * and prepare a list of entry and the best split.
1011 */
1012static struct blame_list *setup_blame_list(struct scoreboard *sb,
1013                                           struct origin *target,
1014                                           int min_score,
1015                                           int *num_ents_p)
1016{
1017        struct blame_entry *e;
1018        int num_ents, i;
1019        struct blame_list *blame_list = NULL;
1020
1021        for (e = sb->ent, num_ents = 0; e; e = e->next)
1022                if (!e->scanned && !e->guilty &&
1023                    same_suspect(e->suspect, target) &&
1024                    min_score < ent_score(sb, e))
1025                        num_ents++;
1026        if (num_ents) {
1027                blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1028                for (e = sb->ent, i = 0; e; e = e->next)
1029                        if (!e->scanned && !e->guilty &&
1030                            same_suspect(e->suspect, target) &&
1031                            min_score < ent_score(sb, e))
1032                                blame_list[i++].ent = e;
1033        }
1034        *num_ents_p = num_ents;
1035        return blame_list;
1036}
1037
1038/*
1039 * Reset the scanned status on all entries.
1040 */
1041static void reset_scanned_flag(struct scoreboard *sb)
1042{
1043        struct blame_entry *e;
1044        for (e = sb->ent; e; e = e->next)
1045                e->scanned = 0;
1046}
1047
1048/*
1049 * For lines target is suspected for, see if we can find code movement
1050 * across file boundary from the parent commit.  porigin is the path
1051 * in the parent we already tried.
1052 */
1053static int find_copy_in_parent(struct scoreboard *sb,
1054                               struct origin *target,
1055                               struct commit *parent,
1056                               struct origin *porigin,
1057                               int opt)
1058{
1059        struct diff_options diff_opts;
1060        const char *paths[1];
1061        int i, j;
1062        int retval;
1063        struct blame_list *blame_list;
1064        int num_ents;
1065
1066        blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1067        if (!blame_list)
1068                return 1; /* nothing remains for this target */
1069
1070        diff_setup(&diff_opts);
1071        DIFF_OPT_SET(&diff_opts, RECURSIVE);
1072        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1073
1074        paths[0] = NULL;
1075        diff_tree_setup_paths(paths, &diff_opts);
1076        diff_setup_done(&diff_opts);
1077
1078        /* Try "find copies harder" on new path if requested;
1079         * we do not want to use diffcore_rename() actually to
1080         * match things up; find_copies_harder is set only to
1081         * force diff_tree_sha1() to feed all filepairs to diff_queue,
1082         * and this code needs to be after diff_setup_done(), which
1083         * usually makes find-copies-harder imply copy detection.
1084         */
1085        if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1086            || ((opt & PICKAXE_BLAME_COPY_HARDER)
1087                && (!porigin || strcmp(target->path, porigin->path))))
1088                DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1089
1090        if (is_null_sha1(target->commit->object.sha1))
1091                do_diff_cache(parent->tree->object.sha1, &diff_opts);
1092        else
1093                diff_tree_sha1(parent->tree->object.sha1,
1094                               target->commit->tree->object.sha1,
1095                               "", &diff_opts);
1096
1097        if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1098                diffcore_std(&diff_opts);
1099
1100        retval = 0;
1101        while (1) {
1102                int made_progress = 0;
1103
1104                for (i = 0; i < diff_queued_diff.nr; i++) {
1105                        struct diff_filepair *p = diff_queued_diff.queue[i];
1106                        struct origin *norigin;
1107                        mmfile_t file_p;
1108                        struct blame_entry this[3];
1109
1110                        if (!DIFF_FILE_VALID(p->one))
1111                                continue; /* does not exist in parent */
1112                        if (S_ISGITLINK(p->one->mode))
1113                                continue; /* ignore git links */
1114                        if (porigin && !strcmp(p->one->path, porigin->path))
1115                                /* find_move already dealt with this path */
1116                                continue;
1117
1118                        norigin = get_origin(sb, parent, p->one->path);
1119                        hashcpy(norigin->blob_sha1, p->one->sha1);
1120                        norigin->mode = p->one->mode;
1121                        fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1122                        if (!file_p.ptr)
1123                                continue;
1124
1125                        for (j = 0; j < num_ents; j++) {
1126                                find_copy_in_blob(sb, blame_list[j].ent,
1127                                                  norigin, this, &file_p);
1128                                copy_split_if_better(sb, blame_list[j].split,
1129                                                     this);
1130                                decref_split(this);
1131                        }
1132                        origin_decref(norigin);
1133                }
1134
1135                for (j = 0; j < num_ents; j++) {
1136                        struct blame_entry *split = blame_list[j].split;
1137                        if (split[1].suspect &&
1138                            blame_copy_score < ent_score(sb, &split[1])) {
1139                                split_blame(sb, split, blame_list[j].ent);
1140                                made_progress = 1;
1141                        }
1142                        else
1143                                blame_list[j].ent->scanned = 1;
1144                        decref_split(split);
1145                }
1146                free(blame_list);
1147
1148                if (!made_progress)
1149                        break;
1150                blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1151                if (!blame_list) {
1152                        retval = 1;
1153                        break;
1154                }
1155        }
1156        reset_scanned_flag(sb);
1157        diff_flush(&diff_opts);
1158        diff_tree_release_paths(&diff_opts);
1159        return retval;
1160}
1161
1162/*
1163 * The blobs of origin and porigin exactly match, so everything
1164 * origin is suspected for can be blamed on the parent.
1165 */
1166static void pass_whole_blame(struct scoreboard *sb,
1167                             struct origin *origin, struct origin *porigin)
1168{
1169        struct blame_entry *e;
1170
1171        if (!porigin->file.ptr && origin->file.ptr) {
1172                /* Steal its file */
1173                porigin->file = origin->file;
1174                origin->file.ptr = NULL;
1175        }
1176        for (e = sb->ent; e; e = e->next) {
1177                if (!same_suspect(e->suspect, origin))
1178                        continue;
1179                origin_incref(porigin);
1180                origin_decref(e->suspect);
1181                e->suspect = porigin;
1182        }
1183}
1184
1185/*
1186 * We pass blame from the current commit to its parents.  We keep saying
1187 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1188 * exonerate ourselves.
1189 */
1190static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1191{
1192        if (!reverse)
1193                return commit->parents;
1194        return lookup_decoration(&revs->children, &commit->object);
1195}
1196
1197static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1198{
1199        int cnt;
1200        struct commit_list *l = first_scapegoat(revs, commit);
1201        for (cnt = 0; l; l = l->next)
1202                cnt++;
1203        return cnt;
1204}
1205
1206#define MAXSG 16
1207
1208static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1209{
1210        struct rev_info *revs = sb->revs;
1211        int i, pass, num_sg;
1212        struct commit *commit = origin->commit;
1213        struct commit_list *sg;
1214        struct origin *sg_buf[MAXSG];
1215        struct origin *porigin, **sg_origin = sg_buf;
1216
1217        num_sg = num_scapegoats(revs, commit);
1218        if (!num_sg)
1219                goto finish;
1220        else if (num_sg < ARRAY_SIZE(sg_buf))
1221                memset(sg_buf, 0, sizeof(sg_buf));
1222        else
1223                sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1224
1225        /*
1226         * The first pass looks for unrenamed path to optimize for
1227         * common cases, then we look for renames in the second pass.
1228         */
1229        for (pass = 0; pass < 2; pass++) {
1230                struct origin *(*find)(struct scoreboard *,
1231                                       struct commit *, struct origin *);
1232                find = pass ? find_rename : find_origin;
1233
1234                for (i = 0, sg = first_scapegoat(revs, commit);
1235                     i < num_sg && sg;
1236                     sg = sg->next, i++) {
1237                        struct commit *p = sg->item;
1238                        int j, same;
1239
1240                        if (sg_origin[i])
1241                                continue;
1242                        if (parse_commit(p))
1243                                continue;
1244                        porigin = find(sb, p, origin);
1245                        if (!porigin)
1246                                continue;
1247                        if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1248                                pass_whole_blame(sb, origin, porigin);
1249                                origin_decref(porigin);
1250                                goto finish;
1251                        }
1252                        for (j = same = 0; j < i; j++)
1253                                if (sg_origin[j] &&
1254                                    !hashcmp(sg_origin[j]->blob_sha1,
1255                                             porigin->blob_sha1)) {
1256                                        same = 1;
1257                                        break;
1258                                }
1259                        if (!same)
1260                                sg_origin[i] = porigin;
1261                        else
1262                                origin_decref(porigin);
1263                }
1264        }
1265
1266        num_commits++;
1267        for (i = 0, sg = first_scapegoat(revs, commit);
1268             i < num_sg && sg;
1269             sg = sg->next, i++) {
1270                struct origin *porigin = sg_origin[i];
1271                if (!porigin)
1272                        continue;
1273                if (!origin->previous) {
1274                        origin_incref(porigin);
1275                        origin->previous = porigin;
1276                }
1277                if (pass_blame_to_parent(sb, origin, porigin))
1278                        goto finish;
1279        }
1280
1281        /*
1282         * Optionally find moves in parents' files.
1283         */
1284        if (opt & PICKAXE_BLAME_MOVE)
1285                for (i = 0, sg = first_scapegoat(revs, commit);
1286                     i < num_sg && sg;
1287                     sg = sg->next, i++) {
1288                        struct origin *porigin = sg_origin[i];
1289                        if (!porigin)
1290                                continue;
1291                        if (find_move_in_parent(sb, origin, porigin))
1292                                goto finish;
1293                }
1294
1295        /*
1296         * Optionally find copies from parents' files.
1297         */
1298        if (opt & PICKAXE_BLAME_COPY)
1299                for (i = 0, sg = first_scapegoat(revs, commit);
1300                     i < num_sg && sg;
1301                     sg = sg->next, i++) {
1302                        struct origin *porigin = sg_origin[i];
1303                        if (find_copy_in_parent(sb, origin, sg->item,
1304                                                porigin, opt))
1305                                goto finish;
1306                }
1307
1308 finish:
1309        for (i = 0; i < num_sg; i++) {
1310                if (sg_origin[i]) {
1311                        drop_origin_blob(sg_origin[i]);
1312                        origin_decref(sg_origin[i]);
1313                }
1314        }
1315        drop_origin_blob(origin);
1316        if (sg_buf != sg_origin)
1317                free(sg_origin);
1318}
1319
1320/*
1321 * Information on commits, used for output.
1322 */
1323struct commit_info {
1324        const char *author;
1325        const char *author_mail;
1326        unsigned long author_time;
1327        const char *author_tz;
1328
1329        /* filled only when asked for details */
1330        const char *committer;
1331        const char *committer_mail;
1332        unsigned long committer_time;
1333        const char *committer_tz;
1334
1335        const char *summary;
1336};
1337
1338/*
1339 * Parse author/committer line in the commit object buffer
1340 */
1341static void get_ac_line(const char *inbuf, const char *what,
1342                        int person_len, char *person,
1343                        int mail_len, char *mail,
1344                        unsigned long *time, const char **tz)
1345{
1346        int len, tzlen, maillen;
1347        char *tmp, *endp, *timepos, *mailpos;
1348
1349        tmp = strstr(inbuf, what);
1350        if (!tmp)
1351                goto error_out;
1352        tmp += strlen(what);
1353        endp = strchr(tmp, '\n');
1354        if (!endp)
1355                len = strlen(tmp);
1356        else
1357                len = endp - tmp;
1358        if (person_len <= len) {
1359        error_out:
1360                /* Ugh */
1361                *tz = "(unknown)";
1362                strcpy(person, *tz);
1363                strcpy(mail, *tz);
1364                *time = 0;
1365                return;
1366        }
1367        memcpy(person, tmp, len);
1368
1369        tmp = person;
1370        tmp += len;
1371        *tmp = 0;
1372        while (person < tmp && *tmp != ' ')
1373                tmp--;
1374        if (tmp <= person)
1375                goto error_out;
1376        *tz = tmp+1;
1377        tzlen = (person+len)-(tmp+1);
1378
1379        *tmp = 0;
1380        while (person < tmp && *tmp != ' ')
1381                tmp--;
1382        if (tmp <= person)
1383                goto error_out;
1384        *time = strtoul(tmp, NULL, 10);
1385        timepos = tmp;
1386
1387        *tmp = 0;
1388        while (person < tmp && !(*tmp == ' ' && tmp[1] == '<'))
1389                tmp--;
1390        if (tmp <= person)
1391                return;
1392        mailpos = tmp + 1;
1393        *tmp = 0;
1394        maillen = timepos - tmp;
1395        memcpy(mail, mailpos, maillen);
1396
1397        if (!mailmap.nr)
1398                return;
1399
1400        /*
1401         * mailmap expansion may make the name longer.
1402         * make room by pushing stuff down.
1403         */
1404        tmp = person + person_len - (tzlen + 1);
1405        memmove(tmp, *tz, tzlen);
1406        tmp[tzlen] = 0;
1407        *tz = tmp;
1408
1409        /*
1410         * Now, convert both name and e-mail using mailmap
1411         */
1412        if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1413                /* Add a trailing '>' to email, since map_user returns plain emails
1414                   Note: It already has '<', since we replace from mail+1 */
1415                mailpos = memchr(mail, '\0', mail_len);
1416                if (mailpos && mailpos-mail < mail_len - 1) {
1417                        *mailpos = '>';
1418                        *(mailpos+1) = '\0';
1419                }
1420        }
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;
1429        char *reencoded, *message;
1430        static char author_name[1024];
1431        static char author_mail[1024];
1432        static char committer_name[1024];
1433        static char committer_mail[1024];
1434        static char summary_buf[1024];
1435
1436        /*
1437         * We've operated without save_commit_buffer, so
1438         * we now need to populate them for output.
1439         */
1440        if (!commit->buffer) {
1441                enum object_type type;
1442                unsigned long size;
1443                commit->buffer =
1444                        read_sha1_file(commit->object.sha1, &type, &size);
1445                if (!commit->buffer)
1446                        die("Cannot read commit %s",
1447                            sha1_to_hex(commit->object.sha1));
1448        }
1449        reencoded = reencode_commit_message(commit, NULL);
1450        message   = reencoded ? reencoded : commit->buffer;
1451        ret->author = author_name;
1452        ret->author_mail = author_mail;
1453        get_ac_line(message, "\nauthor ",
1454                    sizeof(author_name), author_name,
1455                    sizeof(author_mail), author_mail,
1456                    &ret->author_time, &ret->author_tz);
1457
1458        if (!detailed) {
1459                free(reencoded);
1460                return;
1461        }
1462
1463        ret->committer = committer_name;
1464        ret->committer_mail = committer_mail;
1465        get_ac_line(message, "\ncommitter ",
1466                    sizeof(committer_name), committer_name,
1467                    sizeof(committer_mail), committer_mail,
1468                    &ret->committer_time, &ret->committer_tz);
1469
1470        ret->summary = summary_buf;
1471        len = find_commit_subject(message, &subject);
1472        if (len && len < sizeof(summary_buf)) {
1473                memcpy(summary_buf, subject, len);
1474                summary_buf[len] = 0;
1475        } else {
1476                sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1477        }
1478        free(reencoded);
1479}
1480
1481/*
1482 * To allow LF and other nonportable characters in pathnames,
1483 * they are c-style quoted as needed.
1484 */
1485static void write_filename_info(const char *path)
1486{
1487        printf("filename ");
1488        write_name_quoted(path, stdout, '\n');
1489}
1490
1491/*
1492 * Porcelain/Incremental format wants to show a lot of details per
1493 * commit.  Instead of repeating this every line, emit it only once,
1494 * the first time each commit appears in the output (unless the
1495 * user has specifically asked for us to repeat).
1496 */
1497static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1498{
1499        struct commit_info ci;
1500
1501        if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1502                return 0;
1503
1504        suspect->commit->object.flags |= METAINFO_SHOWN;
1505        get_commit_info(suspect->commit, &ci, 1);
1506        printf("author %s\n", ci.author);
1507        printf("author-mail %s\n", ci.author_mail);
1508        printf("author-time %lu\n", ci.author_time);
1509        printf("author-tz %s\n", ci.author_tz);
1510        printf("committer %s\n", ci.committer);
1511        printf("committer-mail %s\n", ci.committer_mail);
1512        printf("committer-time %lu\n", ci.committer_time);
1513        printf("committer-tz %s\n", ci.committer_tz);
1514        printf("summary %s\n", ci.summary);
1515        if (suspect->commit->object.flags & UNINTERESTING)
1516                printf("boundary\n");
1517        if (suspect->previous) {
1518                struct origin *prev = suspect->previous;
1519                printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1520                write_name_quoted(prev->path, stdout, '\n');
1521        }
1522        return 1;
1523}
1524
1525/*
1526 * The blame_entry is found to be guilty for the range.  Mark it
1527 * as such, and show it in incremental output.
1528 */
1529static void found_guilty_entry(struct blame_entry *ent)
1530{
1531        if (ent->guilty)
1532                return;
1533        ent->guilty = 1;
1534        if (incremental) {
1535                struct origin *suspect = ent->suspect;
1536
1537                printf("%s %d %d %d\n",
1538                       sha1_to_hex(suspect->commit->object.sha1),
1539                       ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1540                emit_one_suspect_detail(suspect, 0);
1541                write_filename_info(suspect->path);
1542                maybe_flush_or_die(stdout, "stdout");
1543        }
1544}
1545
1546/*
1547 * The main loop -- while the scoreboard has lines whose true origin
1548 * is still unknown, pick one blame_entry, and allow its current
1549 * suspect to pass blames to its parents.
1550 */
1551static void assign_blame(struct scoreboard *sb, int opt)
1552{
1553        struct rev_info *revs = sb->revs;
1554
1555        while (1) {
1556                struct blame_entry *ent;
1557                struct commit *commit;
1558                struct origin *suspect = NULL;
1559
1560                /* find one suspect to break down */
1561                for (ent = sb->ent; !suspect && ent; ent = ent->next)
1562                        if (!ent->guilty)
1563                                suspect = ent->suspect;
1564                if (!suspect)
1565                        return; /* all done */
1566
1567                /*
1568                 * We will use this suspect later in the loop,
1569                 * so hold onto it in the meantime.
1570                 */
1571                origin_incref(suspect);
1572                commit = suspect->commit;
1573                if (!commit->object.parsed)
1574                        parse_commit(commit);
1575                if (reverse ||
1576                    (!(commit->object.flags & UNINTERESTING) &&
1577                     !(revs->max_age != -1 && commit->date < revs->max_age)))
1578                        pass_blame(sb, suspect, opt);
1579                else {
1580                        commit->object.flags |= UNINTERESTING;
1581                        if (commit->object.parsed)
1582                                mark_parents_uninteresting(commit);
1583                }
1584                /* treat root commit as boundary */
1585                if (!commit->parents && !show_root)
1586                        commit->object.flags |= UNINTERESTING;
1587
1588                /* Take responsibility for the remaining entries */
1589                for (ent = sb->ent; ent; ent = ent->next)
1590                        if (same_suspect(ent->suspect, suspect))
1591                                found_guilty_entry(ent);
1592                origin_decref(suspect);
1593
1594                if (DEBUG) /* sanity */
1595                        sanity_check_refcnt(sb);
1596        }
1597}
1598
1599static const char *format_time(unsigned long time, const char *tz_str,
1600                               int show_raw_time)
1601{
1602        static char time_buf[128];
1603        const char *time_str;
1604        int time_len;
1605        int tz;
1606
1607        if (show_raw_time) {
1608                snprintf(time_buf, sizeof(time_buf), "%lu %s", time, tz_str);
1609        }
1610        else {
1611                tz = atoi(tz_str);
1612                time_str = show_date(time, tz, blame_date_mode);
1613                time_len = strlen(time_str);
1614                memcpy(time_buf, time_str, time_len);
1615                memset(time_buf + time_len, ' ', blame_date_width - time_len);
1616        }
1617        return time_buf;
1618}
1619
1620#define OUTPUT_ANNOTATE_COMPAT  001
1621#define OUTPUT_LONG_OBJECT_NAME 002
1622#define OUTPUT_RAW_TIMESTAMP    004
1623#define OUTPUT_PORCELAIN        010
1624#define OUTPUT_SHOW_NAME        020
1625#define OUTPUT_SHOW_NUMBER      040
1626#define OUTPUT_SHOW_SCORE      0100
1627#define OUTPUT_NO_AUTHOR       0200
1628#define OUTPUT_SHOW_EMAIL       0400
1629#define OUTPUT_LINE_PORCELAIN 01000
1630
1631static void emit_porcelain_details(struct origin *suspect, int repeat)
1632{
1633        if (emit_one_suspect_detail(suspect, repeat) ||
1634            (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1635                write_filename_info(suspect->path);
1636}
1637
1638static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1639                           int opt)
1640{
1641        int repeat = opt & OUTPUT_LINE_PORCELAIN;
1642        int cnt;
1643        const char *cp;
1644        struct origin *suspect = ent->suspect;
1645        char hex[41];
1646
1647        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1648        printf("%s%c%d %d %d\n",
1649               hex,
1650               ent->guilty ? ' ' : '*', /* purely for debugging */
1651               ent->s_lno + 1,
1652               ent->lno + 1,
1653               ent->num_lines);
1654        emit_porcelain_details(suspect, repeat);
1655
1656        cp = nth_line(sb, ent->lno);
1657        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1658                char ch;
1659                if (cnt) {
1660                        printf("%s %d %d\n", hex,
1661                               ent->s_lno + 1 + cnt,
1662                               ent->lno + 1 + cnt);
1663                        if (repeat)
1664                                emit_porcelain_details(suspect, 1);
1665                }
1666                putchar('\t');
1667                do {
1668                        ch = *cp++;
1669                        putchar(ch);
1670                } while (ch != '\n' &&
1671                         cp < sb->final_buf + sb->final_buf_size);
1672        }
1673
1674        if (sb->final_buf_size && cp[-1] != '\n')
1675                putchar('\n');
1676}
1677
1678static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1679{
1680        int cnt;
1681        const char *cp;
1682        struct origin *suspect = ent->suspect;
1683        struct commit_info ci;
1684        char hex[41];
1685        int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1686
1687        get_commit_info(suspect->commit, &ci, 1);
1688        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1689
1690        cp = nth_line(sb, ent->lno);
1691        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1692                char ch;
1693                int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1694
1695                if (suspect->commit->object.flags & UNINTERESTING) {
1696                        if (blank_boundary)
1697                                memset(hex, ' ', length);
1698                        else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1699                                length--;
1700                                putchar('^');
1701                        }
1702                }
1703
1704                printf("%.*s", length, hex);
1705                if (opt & OUTPUT_ANNOTATE_COMPAT) {
1706                        const char *name;
1707                        if (opt & OUTPUT_SHOW_EMAIL)
1708                                name = ci.author_mail;
1709                        else
1710                                name = ci.author;
1711                        printf("\t(%10s\t%10s\t%d)", name,
1712                               format_time(ci.author_time, ci.author_tz,
1713                                           show_raw_time),
1714                               ent->lno + 1 + cnt);
1715                } else {
1716                        if (opt & OUTPUT_SHOW_SCORE)
1717                                printf(" %*d %02d",
1718                                       max_score_digits, ent->score,
1719                                       ent->suspect->refcnt);
1720                        if (opt & OUTPUT_SHOW_NAME)
1721                                printf(" %-*.*s", longest_file, longest_file,
1722                                       suspect->path);
1723                        if (opt & OUTPUT_SHOW_NUMBER)
1724                                printf(" %*d", max_orig_digits,
1725                                       ent->s_lno + 1 + cnt);
1726
1727                        if (!(opt & OUTPUT_NO_AUTHOR)) {
1728                                const char *name;
1729                                int pad;
1730                                if (opt & OUTPUT_SHOW_EMAIL)
1731                                        name = ci.author_mail;
1732                                else
1733                                        name = ci.author;
1734                                pad = longest_author - utf8_strwidth(name);
1735                                printf(" (%s%*s %10s",
1736                                       name, pad, "",
1737                                       format_time(ci.author_time,
1738                                                   ci.author_tz,
1739                                                   show_raw_time));
1740                        }
1741                        printf(" %*d) ",
1742                               max_digits, ent->lno + 1 + cnt);
1743                }
1744                do {
1745                        ch = *cp++;
1746                        putchar(ch);
1747                } while (ch != '\n' &&
1748                         cp < sb->final_buf + sb->final_buf_size);
1749        }
1750
1751        if (sb->final_buf_size && cp[-1] != '\n')
1752                putchar('\n');
1753}
1754
1755static void output(struct scoreboard *sb, int option)
1756{
1757        struct blame_entry *ent;
1758
1759        if (option & OUTPUT_PORCELAIN) {
1760                for (ent = sb->ent; ent; ent = ent->next) {
1761                        struct blame_entry *oth;
1762                        struct origin *suspect = ent->suspect;
1763                        struct commit *commit = suspect->commit;
1764                        if (commit->object.flags & MORE_THAN_ONE_PATH)
1765                                continue;
1766                        for (oth = ent->next; oth; oth = oth->next) {
1767                                if ((oth->suspect->commit != commit) ||
1768                                    !strcmp(oth->suspect->path, suspect->path))
1769                                        continue;
1770                                commit->object.flags |= MORE_THAN_ONE_PATH;
1771                                break;
1772                        }
1773                }
1774        }
1775
1776        for (ent = sb->ent; ent; ent = ent->next) {
1777                if (option & OUTPUT_PORCELAIN)
1778                        emit_porcelain(sb, ent, option);
1779                else {
1780                        emit_other(sb, ent, option);
1781                }
1782        }
1783}
1784
1785/*
1786 * To allow quick access to the contents of nth line in the
1787 * final image, prepare an index in the scoreboard.
1788 */
1789static int prepare_lines(struct scoreboard *sb)
1790{
1791        const char *buf = sb->final_buf;
1792        unsigned long len = sb->final_buf_size;
1793        int num = 0, incomplete = 0, bol = 1;
1794
1795        if (len && buf[len-1] != '\n')
1796                incomplete++; /* incomplete line at the end */
1797        while (len--) {
1798                if (bol) {
1799                        sb->lineno = xrealloc(sb->lineno,
1800                                              sizeof(int *) * (num + 1));
1801                        sb->lineno[num] = buf - sb->final_buf;
1802                        bol = 0;
1803                }
1804                if (*buf++ == '\n') {
1805                        num++;
1806                        bol = 1;
1807                }
1808        }
1809        sb->lineno = xrealloc(sb->lineno,
1810                              sizeof(int *) * (num + incomplete + 1));
1811        sb->lineno[num + incomplete] = buf - sb->final_buf;
1812        sb->num_lines = num + incomplete;
1813        return sb->num_lines;
1814}
1815
1816/*
1817 * Add phony grafts for use with -S; this is primarily to
1818 * support git's cvsserver that wants to give a linear history
1819 * to its clients.
1820 */
1821static int read_ancestry(const char *graft_file)
1822{
1823        FILE *fp = fopen(graft_file, "r");
1824        char buf[1024];
1825        if (!fp)
1826                return -1;
1827        while (fgets(buf, sizeof(buf), fp)) {
1828                /* The format is just "Commit Parent1 Parent2 ...\n" */
1829                int len = strlen(buf);
1830                struct commit_graft *graft = read_graft_line(buf, len);
1831                if (graft)
1832                        register_commit_graft(graft, 0);
1833        }
1834        fclose(fp);
1835        return 0;
1836}
1837
1838static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1839{
1840        const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1841                                              auto_abbrev);
1842        int len = strlen(uniq);
1843        if (auto_abbrev < len)
1844                return len;
1845        return auto_abbrev;
1846}
1847
1848/*
1849 * How many columns do we need to show line numbers, authors,
1850 * and filenames?
1851 */
1852static void find_alignment(struct scoreboard *sb, int *option)
1853{
1854        int longest_src_lines = 0;
1855        int longest_dst_lines = 0;
1856        unsigned largest_score = 0;
1857        struct blame_entry *e;
1858        int compute_auto_abbrev = (abbrev < 0);
1859        int auto_abbrev = default_abbrev;
1860
1861        for (e = sb->ent; e; e = e->next) {
1862                struct origin *suspect = e->suspect;
1863                struct commit_info ci;
1864                int num;
1865
1866                if (compute_auto_abbrev)
1867                        auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1868                if (strcmp(suspect->path, sb->path))
1869                        *option |= OUTPUT_SHOW_NAME;
1870                num = strlen(suspect->path);
1871                if (longest_file < num)
1872                        longest_file = num;
1873                if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1874                        suspect->commit->object.flags |= METAINFO_SHOWN;
1875                        get_commit_info(suspect->commit, &ci, 1);
1876                        if (*option & OUTPUT_SHOW_EMAIL)
1877                                num = utf8_strwidth(ci.author_mail);
1878                        else
1879                                num = utf8_strwidth(ci.author);
1880                        if (longest_author < num)
1881                                longest_author = num;
1882                }
1883                num = e->s_lno + e->num_lines;
1884                if (longest_src_lines < num)
1885                        longest_src_lines = num;
1886                num = e->lno + e->num_lines;
1887                if (longest_dst_lines < num)
1888                        longest_dst_lines = num;
1889                if (largest_score < ent_score(sb, e))
1890                        largest_score = ent_score(sb, e);
1891        }
1892        max_orig_digits = decimal_width(longest_src_lines);
1893        max_digits = decimal_width(longest_dst_lines);
1894        max_score_digits = decimal_width(largest_score);
1895
1896        if (compute_auto_abbrev)
1897                /* one more abbrev length is needed for the boundary commit */
1898                abbrev = auto_abbrev + 1;
1899}
1900
1901/*
1902 * For debugging -- origin is refcounted, and this asserts that
1903 * we do not underflow.
1904 */
1905static void sanity_check_refcnt(struct scoreboard *sb)
1906{
1907        int baa = 0;
1908        struct blame_entry *ent;
1909
1910        for (ent = sb->ent; ent; ent = ent->next) {
1911                /* Nobody should have zero or negative refcnt */
1912                if (ent->suspect->refcnt <= 0) {
1913                        fprintf(stderr, "%s in %s has negative refcnt %d\n",
1914                                ent->suspect->path,
1915                                sha1_to_hex(ent->suspect->commit->object.sha1),
1916                                ent->suspect->refcnt);
1917                        baa = 1;
1918                }
1919        }
1920        if (baa) {
1921                int opt = 0160;
1922                find_alignment(sb, &opt);
1923                output(sb, opt);
1924                die("Baa %d!", baa);
1925        }
1926}
1927
1928/*
1929 * Used for the command line parsing; check if the path exists
1930 * in the working tree.
1931 */
1932static int has_string_in_work_tree(const char *path)
1933{
1934        struct stat st;
1935        return !lstat(path, &st);
1936}
1937
1938static unsigned parse_score(const char *arg)
1939{
1940        char *end;
1941        unsigned long score = strtoul(arg, &end, 10);
1942        if (*end)
1943                return 0;
1944        return score;
1945}
1946
1947static const char *add_prefix(const char *prefix, const char *path)
1948{
1949        return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1950}
1951
1952/*
1953 * Parsing of (comma separated) one item in the -L option
1954 */
1955static const char *parse_loc(const char *spec,
1956                             struct scoreboard *sb, long lno,
1957                             long begin, long *ret)
1958{
1959        char *term;
1960        const char *line;
1961        long num;
1962        int reg_error;
1963        regex_t regexp;
1964        regmatch_t match[1];
1965
1966        /* Allow "-L <something>,+20" to mean starting at <something>
1967         * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1968         * <something>.
1969         */
1970        if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1971                num = strtol(spec + 1, &term, 10);
1972                if (term != spec + 1) {
1973                        if (spec[0] == '-')
1974                                num = 0 - num;
1975                        if (0 < num)
1976                                *ret = begin + num - 2;
1977                        else if (!num)
1978                                *ret = begin;
1979                        else
1980                                *ret = begin + num;
1981                        return term;
1982                }
1983                return spec;
1984        }
1985        num = strtol(spec, &term, 10);
1986        if (term != spec) {
1987                *ret = num;
1988                return term;
1989        }
1990        if (spec[0] != '/')
1991                return spec;
1992
1993        /* it could be a regexp of form /.../ */
1994        for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1995                if (*term == '\\')
1996                        term++;
1997        }
1998        if (*term != '/')
1999                return spec;
2000
2001        /* try [spec+1 .. term-1] as regexp */
2002        *term = 0;
2003        begin--; /* input is in human terms */
2004        line = nth_line(sb, begin);
2005
2006        if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
2007            !(reg_error = regexec(&regexp, line, 1, match, 0))) {
2008                const char *cp = line + match[0].rm_so;
2009                const char *nline;
2010
2011                while (begin++ < lno) {
2012                        nline = nth_line(sb, begin);
2013                        if (line <= cp && cp < nline)
2014                                break;
2015                        line = nline;
2016                }
2017                *ret = begin;
2018                regfree(&regexp);
2019                *term++ = '/';
2020                return term;
2021        }
2022        else {
2023                char errbuf[1024];
2024                regerror(reg_error, &regexp, errbuf, 1024);
2025                die("-L parameter '%s': %s", spec + 1, errbuf);
2026        }
2027}
2028
2029/*
2030 * Parsing of -L option
2031 */
2032static void prepare_blame_range(struct scoreboard *sb,
2033                                const char *bottomtop,
2034                                long lno,
2035                                long *bottom, long *top)
2036{
2037        const char *term;
2038
2039        term = parse_loc(bottomtop, sb, lno, 1, bottom);
2040        if (*term == ',') {
2041                term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2042                if (*term)
2043                        usage(blame_usage);
2044        }
2045        if (*term)
2046                usage(blame_usage);
2047}
2048
2049static int git_blame_config(const char *var, const char *value, void *cb)
2050{
2051        if (!strcmp(var, "blame.showroot")) {
2052                show_root = git_config_bool(var, value);
2053                return 0;
2054        }
2055        if (!strcmp(var, "blame.blankboundary")) {
2056                blank_boundary = git_config_bool(var, value);
2057                return 0;
2058        }
2059        if (!strcmp(var, "blame.date")) {
2060                if (!value)
2061                        return config_error_nonbool(var);
2062                blame_date_mode = parse_date_format(value);
2063                return 0;
2064        }
2065
2066        if (userdiff_config(var, value) < 0)
2067                return -1;
2068
2069        return git_default_config(var, value, cb);
2070}
2071
2072/*
2073 * Prepare a dummy commit that represents the work tree (or staged) item.
2074 * Note that annotating work tree item never works in the reverse.
2075 */
2076static struct commit *fake_working_tree_commit(struct diff_options *opt,
2077                                               const char *path,
2078                                               const char *contents_from)
2079{
2080        struct commit *commit;
2081        struct origin *origin;
2082        unsigned char head_sha1[20];
2083        struct strbuf buf = STRBUF_INIT;
2084        const char *ident;
2085        time_t now;
2086        int size, len;
2087        struct cache_entry *ce;
2088        unsigned mode;
2089
2090        if (get_sha1("HEAD", head_sha1))
2091                die("No such ref: HEAD");
2092
2093        time(&now);
2094        commit = xcalloc(1, sizeof(*commit));
2095        commit->parents = xcalloc(1, sizeof(*commit->parents));
2096        commit->parents->item = lookup_commit_reference(head_sha1);
2097        commit->object.parsed = 1;
2098        commit->date = now;
2099        commit->object.type = OBJ_COMMIT;
2100
2101        origin = make_origin(commit, path);
2102
2103        if (!contents_from || strcmp("-", contents_from)) {
2104                struct stat st;
2105                const char *read_from;
2106                char *buf_ptr;
2107                unsigned long buf_len;
2108
2109                if (contents_from) {
2110                        if (stat(contents_from, &st) < 0)
2111                                die_errno("Cannot stat '%s'", contents_from);
2112                        read_from = contents_from;
2113                }
2114                else {
2115                        if (lstat(path, &st) < 0)
2116                                die_errno("Cannot lstat '%s'", path);
2117                        read_from = path;
2118                }
2119                mode = canon_mode(st.st_mode);
2120
2121                switch (st.st_mode & S_IFMT) {
2122                case S_IFREG:
2123                        if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2124                            textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2125                                strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2126                        else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2127                                die_errno("cannot open or read '%s'", read_from);
2128                        break;
2129                case S_IFLNK:
2130                        if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2131                                die_errno("cannot readlink '%s'", read_from);
2132                        break;
2133                default:
2134                        die("unsupported file type %s", read_from);
2135                }
2136        }
2137        else {
2138                /* Reading from stdin */
2139                contents_from = "standard input";
2140                mode = 0;
2141                if (strbuf_read(&buf, 0, 0) < 0)
2142                        die_errno("failed to read from stdin");
2143        }
2144        convert_to_git(path, buf.buf, buf.len, &buf, 0);
2145        origin->file.ptr = buf.buf;
2146        origin->file.size = buf.len;
2147        pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2148        commit->util = origin;
2149
2150        /*
2151         * Read the current index, replace the path entry with
2152         * origin->blob_sha1 without mucking with its mode or type
2153         * bits; we are not going to write this index out -- we just
2154         * want to run "diff-index --cached".
2155         */
2156        discard_cache();
2157        read_cache();
2158
2159        len = strlen(path);
2160        if (!mode) {
2161                int pos = cache_name_pos(path, len);
2162                if (0 <= pos)
2163                        mode = active_cache[pos]->ce_mode;
2164                else
2165                        /* Let's not bother reading from HEAD tree */
2166                        mode = S_IFREG | 0644;
2167        }
2168        size = cache_entry_size(len);
2169        ce = xcalloc(1, size);
2170        hashcpy(ce->sha1, origin->blob_sha1);
2171        memcpy(ce->name, path, len);
2172        ce->ce_flags = create_ce_flags(0);
2173        ce->ce_namelen = len;
2174        ce->ce_mode = create_ce_mode(mode);
2175        add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2176
2177        /*
2178         * We are not going to write this out, so this does not matter
2179         * right now, but someday we might optimize diff-index --cached
2180         * with cache-tree information.
2181         */
2182        cache_tree_invalidate_path(active_cache_tree, path);
2183
2184        commit->buffer = xmalloc(400);
2185        ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2186        snprintf(commit->buffer, 400,
2187                "tree 0000000000000000000000000000000000000000\n"
2188                "parent %s\n"
2189                "author %s\n"
2190                "committer %s\n\n"
2191                "Version of %s from %s\n",
2192                sha1_to_hex(head_sha1),
2193                ident, ident, path, contents_from ? contents_from : path);
2194        return commit;
2195}
2196
2197static const char *prepare_final(struct scoreboard *sb)
2198{
2199        int i;
2200        const char *final_commit_name = NULL;
2201        struct rev_info *revs = sb->revs;
2202
2203        /*
2204         * There must be one and only one positive commit in the
2205         * revs->pending array.
2206         */
2207        for (i = 0; i < revs->pending.nr; i++) {
2208                struct object *obj = revs->pending.objects[i].item;
2209                if (obj->flags & UNINTERESTING)
2210                        continue;
2211                while (obj->type == OBJ_TAG)
2212                        obj = deref_tag(obj, NULL, 0);
2213                if (obj->type != OBJ_COMMIT)
2214                        die("Non commit %s?", revs->pending.objects[i].name);
2215                if (sb->final)
2216                        die("More than one commit to dig from %s and %s?",
2217                            revs->pending.objects[i].name,
2218                            final_commit_name);
2219                sb->final = (struct commit *) obj;
2220                final_commit_name = revs->pending.objects[i].name;
2221        }
2222        return final_commit_name;
2223}
2224
2225static const char *prepare_initial(struct scoreboard *sb)
2226{
2227        int i;
2228        const char *final_commit_name = NULL;
2229        struct rev_info *revs = sb->revs;
2230
2231        /*
2232         * There must be one and only one negative commit, and it must be
2233         * the boundary.
2234         */
2235        for (i = 0; i < revs->pending.nr; i++) {
2236                struct object *obj = revs->pending.objects[i].item;
2237                if (!(obj->flags & UNINTERESTING))
2238                        continue;
2239                while (obj->type == OBJ_TAG)
2240                        obj = deref_tag(obj, NULL, 0);
2241                if (obj->type != OBJ_COMMIT)
2242                        die("Non commit %s?", revs->pending.objects[i].name);
2243                if (sb->final)
2244                        die("More than one commit to dig down to %s and %s?",
2245                            revs->pending.objects[i].name,
2246                            final_commit_name);
2247                sb->final = (struct commit *) obj;
2248                final_commit_name = revs->pending.objects[i].name;
2249        }
2250        if (!final_commit_name)
2251                die("No commit to dig down to?");
2252        return final_commit_name;
2253}
2254
2255static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2256{
2257        int *opt = option->value;
2258
2259        /*
2260         * -C enables copy from removed files;
2261         * -C -C enables copy from existing files, but only
2262         *       when blaming a new file;
2263         * -C -C -C enables copy from existing files for
2264         *          everybody
2265         */
2266        if (*opt & PICKAXE_BLAME_COPY_HARDER)
2267                *opt |= PICKAXE_BLAME_COPY_HARDEST;
2268        if (*opt & PICKAXE_BLAME_COPY)
2269                *opt |= PICKAXE_BLAME_COPY_HARDER;
2270        *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2271
2272        if (arg)
2273                blame_copy_score = parse_score(arg);
2274        return 0;
2275}
2276
2277static int blame_move_callback(const struct option *option, const char *arg, int unset)
2278{
2279        int *opt = option->value;
2280
2281        *opt |= PICKAXE_BLAME_MOVE;
2282
2283        if (arg)
2284                blame_move_score = parse_score(arg);
2285        return 0;
2286}
2287
2288static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2289{
2290        const char **bottomtop = option->value;
2291        if (!arg)
2292                return -1;
2293        if (*bottomtop)
2294                die("More than one '-L n,m' option given");
2295        *bottomtop = arg;
2296        return 0;
2297}
2298
2299int cmd_blame(int argc, const char **argv, const char *prefix)
2300{
2301        struct rev_info revs;
2302        const char *path;
2303        struct scoreboard sb;
2304        struct origin *o;
2305        struct blame_entry *ent;
2306        long dashdash_pos, bottom, top, lno;
2307        const char *final_commit_name = NULL;
2308        enum object_type type;
2309
2310        static const char *bottomtop = NULL;
2311        static int output_option = 0, opt = 0;
2312        static int show_stats = 0;
2313        static const char *revs_file = NULL;
2314        static const char *contents_from = NULL;
2315        static const struct option options[] = {
2316                OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2317                OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2318                OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2319                OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2320                OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2321                OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2322                OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2323                OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2324                OPT_BIT(0, "line-porcelain", &output_option, "Show porcelain format with per-line commit information", OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2325                OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2326                OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2327                OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2328                OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2329                OPT_BIT('e', "show-email", &output_option, "Show author email instead of name (Default: off)", OUTPUT_SHOW_EMAIL),
2330                OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2331                OPT_BIT(0, "minimal", &xdl_opts, "Spend extra cycles to find better match", XDF_NEED_MINIMAL),
2332                OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2333                OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2334                { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2335                { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2336                OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2337                OPT__ABBREV(&abbrev),
2338                OPT_END()
2339        };
2340
2341        struct parse_opt_ctx_t ctx;
2342        int cmd_is_annotate = !strcmp(argv[0], "annotate");
2343
2344        git_config(git_blame_config, NULL);
2345        init_revisions(&revs, NULL);
2346        revs.date_mode = blame_date_mode;
2347        DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2348
2349        save_commit_buffer = 0;
2350        dashdash_pos = 0;
2351
2352        parse_options_start(&ctx, argc, argv, prefix, options,
2353                            PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2354        for (;;) {
2355                switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2356                case PARSE_OPT_HELP:
2357                        exit(129);
2358                case PARSE_OPT_DONE:
2359                        if (ctx.argv[0])
2360                                dashdash_pos = ctx.cpidx;
2361                        goto parse_done;
2362                }
2363
2364                if (!strcmp(ctx.argv[0], "--reverse")) {
2365                        ctx.argv[0] = "--children";
2366                        reverse = 1;
2367                }
2368                parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2369        }
2370parse_done:
2371        argc = parse_options_end(&ctx);
2372
2373        if (0 < abbrev)
2374                /* one more abbrev length is needed for the boundary commit */
2375                abbrev++;
2376
2377        if (revs_file && read_ancestry(revs_file))
2378                die_errno("reading graft file '%s' failed", revs_file);
2379
2380        if (cmd_is_annotate) {
2381                output_option |= OUTPUT_ANNOTATE_COMPAT;
2382                blame_date_mode = DATE_ISO8601;
2383        } else {
2384                blame_date_mode = revs.date_mode;
2385        }
2386
2387        /* The maximum width used to show the dates */
2388        switch (blame_date_mode) {
2389        case DATE_RFC2822:
2390                blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2391                break;
2392        case DATE_ISO8601:
2393                blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2394                break;
2395        case DATE_RAW:
2396                blame_date_width = sizeof("1161298804 -0700");
2397                break;
2398        case DATE_SHORT:
2399                blame_date_width = sizeof("2006-10-19");
2400                break;
2401        case DATE_RELATIVE:
2402                /* "normal" is used as the fallback for "relative" */
2403        case DATE_LOCAL:
2404        case DATE_NORMAL:
2405                blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2406                break;
2407        }
2408        blame_date_width -= 1; /* strip the null */
2409
2410        if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2411                opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2412                        PICKAXE_BLAME_COPY_HARDER);
2413
2414        if (!blame_move_score)
2415                blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2416        if (!blame_copy_score)
2417                blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2418
2419        /*
2420         * We have collected options unknown to us in argv[1..unk]
2421         * which are to be passed to revision machinery if we are
2422         * going to do the "bottom" processing.
2423         *
2424         * The remaining are:
2425         *
2426         * (1) if dashdash_pos != 0, it is either
2427         *     "blame [revisions] -- <path>" or
2428         *     "blame -- <path> <rev>"
2429         *
2430         * (2) otherwise, it is one of the two:
2431         *     "blame [revisions] <path>"
2432         *     "blame <path> <rev>"
2433         *
2434         * Note that we must strip out <path> from the arguments: we do not
2435         * want the path pruning but we may want "bottom" processing.
2436         */
2437        if (dashdash_pos) {
2438                switch (argc - dashdash_pos - 1) {
2439                case 2: /* (1b) */
2440                        if (argc != 4)
2441                                usage_with_options(blame_opt_usage, options);
2442                        /* reorder for the new way: <rev> -- <path> */
2443                        argv[1] = argv[3];
2444                        argv[3] = argv[2];
2445                        argv[2] = "--";
2446                        /* FALLTHROUGH */
2447                case 1: /* (1a) */
2448                        path = add_prefix(prefix, argv[--argc]);
2449                        argv[argc] = NULL;
2450                        break;
2451                default:
2452                        usage_with_options(blame_opt_usage, options);
2453                }
2454        } else {
2455                if (argc < 2)
2456                        usage_with_options(blame_opt_usage, options);
2457                path = add_prefix(prefix, argv[argc - 1]);
2458                if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2459                        path = add_prefix(prefix, argv[1]);
2460                        argv[1] = argv[2];
2461                }
2462                argv[argc - 1] = "--";
2463
2464                setup_work_tree();
2465                if (!has_string_in_work_tree(path))
2466                        die_errno("cannot stat path '%s'", path);
2467        }
2468
2469        revs.disable_stdin = 1;
2470        setup_revisions(argc, argv, &revs, NULL);
2471        memset(&sb, 0, sizeof(sb));
2472
2473        sb.revs = &revs;
2474        if (!reverse)
2475                final_commit_name = prepare_final(&sb);
2476        else if (contents_from)
2477                die("--contents and --children do not blend well.");
2478        else
2479                final_commit_name = prepare_initial(&sb);
2480
2481        if (!sb.final) {
2482                /*
2483                 * "--not A B -- path" without anything positive;
2484                 * do not default to HEAD, but use the working tree
2485                 * or "--contents".
2486                 */
2487                setup_work_tree();
2488                sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2489                                                    path, contents_from);
2490                add_pending_object(&revs, &(sb.final->object), ":");
2491        }
2492        else if (contents_from)
2493                die("Cannot use --contents with final commit object name");
2494
2495        /*
2496         * If we have bottom, this will mark the ancestors of the
2497         * bottom commits we would reach while traversing as
2498         * uninteresting.
2499         */
2500        if (prepare_revision_walk(&revs))
2501                die("revision walk setup failed");
2502
2503        if (is_null_sha1(sb.final->object.sha1)) {
2504                char *buf;
2505                o = sb.final->util;
2506                buf = xmalloc(o->file.size + 1);
2507                memcpy(buf, o->file.ptr, o->file.size + 1);
2508                sb.final_buf = buf;
2509                sb.final_buf_size = o->file.size;
2510        }
2511        else {
2512                o = get_origin(&sb, sb.final, path);
2513                if (fill_blob_sha1_and_mode(o))
2514                        die("no such path %s in %s", path, final_commit_name);
2515
2516                if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2517                    textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2518                                    &sb.final_buf_size))
2519                        ;
2520                else
2521                        sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2522                                                      &sb.final_buf_size);
2523
2524                if (!sb.final_buf)
2525                        die("Cannot read blob %s for path %s",
2526                            sha1_to_hex(o->blob_sha1),
2527                            path);
2528        }
2529        num_read_blob++;
2530        lno = prepare_lines(&sb);
2531
2532        bottom = top = 0;
2533        if (bottomtop)
2534                prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2535        if (bottom && top && top < bottom) {
2536                long tmp;
2537                tmp = top; top = bottom; bottom = tmp;
2538        }
2539        if (bottom < 1)
2540                bottom = 1;
2541        if (top < 1)
2542                top = lno;
2543        bottom--;
2544        if (lno < top || lno < bottom)
2545                die("file %s has only %lu lines", path, lno);
2546
2547        ent = xcalloc(1, sizeof(*ent));
2548        ent->lno = bottom;
2549        ent->num_lines = top - bottom;
2550        ent->suspect = o;
2551        ent->s_lno = bottom;
2552
2553        sb.ent = ent;
2554        sb.path = path;
2555
2556        read_mailmap(&mailmap, NULL);
2557
2558        if (!incremental)
2559                setup_pager();
2560
2561        assign_blame(&sb, opt);
2562
2563        if (incremental)
2564                return 0;
2565
2566        coalesce(&sb);
2567
2568        if (!(output_option & OUTPUT_PORCELAIN))
2569                find_alignment(&sb, &output_option);
2570
2571        output(&sb, output_option);
2572        free((void *)sb.final_buf);
2573        for (ent = sb.ent; ent; ) {
2574                struct blame_entry *e = ent->next;
2575                free(ent);
2576                ent = e;
2577        }
2578
2579        if (show_stats) {
2580                printf("num read blob: %d\n", num_read_blob);
2581                printf("num get patch: %d\n", num_get_patch);
2582                printf("num commits: %d\n", num_commits);
2583        }
2584        return 0;
2585}