builtin / blame.con commit Merge branch 'js/gitweb-path-info-unquote' (92cdcc5)
   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                    char **buf,
 114                    unsigned long *buf_size)
 115{
 116        struct diff_filespec *df;
 117        struct userdiff_driver *textconv;
 118
 119        df = alloc_filespec(path);
 120        fill_filespec(df, sha1, mode);
 121        textconv = get_textconv(df);
 122        if (!textconv) {
 123                free_filespec(df);
 124                return 0;
 125        }
 126
 127        *buf_size = fill_textconv(textconv, df, buf);
 128        free_filespec(df);
 129        return 1;
 130}
 131
 132/*
 133 * Given an origin, prepare mmfile_t structure to be used by the
 134 * diff machinery
 135 */
 136static void fill_origin_blob(struct diff_options *opt,
 137                             struct origin *o, mmfile_t *file)
 138{
 139        if (!o->file.ptr) {
 140                enum object_type type;
 141                unsigned long file_size;
 142
 143                num_read_blob++;
 144                if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
 145                    textconv_object(o->path, o->mode, o->blob_sha1, &file->ptr, &file_size))
 146                        ;
 147                else
 148                        file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
 149                file->size = file_size;
 150
 151                if (!file->ptr)
 152                        die("Cannot read blob %s for path %s",
 153                            sha1_to_hex(o->blob_sha1),
 154                            o->path);
 155                o->file = *file;
 156        }
 157        else
 158                *file = o->file;
 159}
 160
 161/*
 162 * Origin is refcounted and usually we keep the blob contents to be
 163 * reused.
 164 */
 165static inline struct origin *origin_incref(struct origin *o)
 166{
 167        if (o)
 168                o->refcnt++;
 169        return o;
 170}
 171
 172static void origin_decref(struct origin *o)
 173{
 174        if (o && --o->refcnt <= 0) {
 175                if (o->previous)
 176                        origin_decref(o->previous);
 177                free(o->file.ptr);
 178                free(o);
 179        }
 180}
 181
 182static void drop_origin_blob(struct origin *o)
 183{
 184        if (o->file.ptr) {
 185                free(o->file.ptr);
 186                o->file.ptr = NULL;
 187        }
 188}
 189
 190/*
 191 * Each group of lines is described by a blame_entry; it can be split
 192 * as we pass blame to the parents.  They form a linked list in the
 193 * scoreboard structure, sorted by the target line number.
 194 */
 195struct blame_entry {
 196        struct blame_entry *prev;
 197        struct blame_entry *next;
 198
 199        /* the first line of this group in the final image;
 200         * internally all line numbers are 0 based.
 201         */
 202        int lno;
 203
 204        /* how many lines this group has */
 205        int num_lines;
 206
 207        /* the commit that introduced this group into the final image */
 208        struct origin *suspect;
 209
 210        /* true if the suspect is truly guilty; false while we have not
 211         * checked if the group came from one of its parents.
 212         */
 213        char guilty;
 214
 215        /* true if the entry has been scanned for copies in the current parent
 216         */
 217        char scanned;
 218
 219        /* the line number of the first line of this group in the
 220         * suspect's file; internally all line numbers are 0 based.
 221         */
 222        int s_lno;
 223
 224        /* how significant this entry is -- cached to avoid
 225         * scanning the lines over and over.
 226         */
 227        unsigned score;
 228};
 229
 230/*
 231 * The current state of the blame assignment.
 232 */
 233struct scoreboard {
 234        /* the final commit (i.e. where we started digging from) */
 235        struct commit *final;
 236        struct rev_info *revs;
 237        const char *path;
 238
 239        /*
 240         * The contents in the final image.
 241         * Used by many functions to obtain contents of the nth line,
 242         * indexed with scoreboard.lineno[blame_entry.lno].
 243         */
 244        const char *final_buf;
 245        unsigned long final_buf_size;
 246
 247        /* linked list of blames */
 248        struct blame_entry *ent;
 249
 250        /* look-up a line in the final buffer */
 251        int num_lines;
 252        int *lineno;
 253};
 254
 255static inline int same_suspect(struct origin *a, struct origin *b)
 256{
 257        if (a == b)
 258                return 1;
 259        if (a->commit != b->commit)
 260                return 0;
 261        return !strcmp(a->path, b->path);
 262}
 263
 264static void sanity_check_refcnt(struct scoreboard *);
 265
 266/*
 267 * If two blame entries that are next to each other came from
 268 * contiguous lines in the same origin (i.e. <commit, path> pair),
 269 * merge them together.
 270 */
 271static void coalesce(struct scoreboard *sb)
 272{
 273        struct blame_entry *ent, *next;
 274
 275        for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 276                if (same_suspect(ent->suspect, next->suspect) &&
 277                    ent->guilty == next->guilty &&
 278                    ent->s_lno + ent->num_lines == next->s_lno) {
 279                        ent->num_lines += next->num_lines;
 280                        ent->next = next->next;
 281                        if (ent->next)
 282                                ent->next->prev = ent;
 283                        origin_decref(next->suspect);
 284                        free(next);
 285                        ent->score = 0;
 286                        next = ent; /* again */
 287                }
 288        }
 289
 290        if (DEBUG) /* sanity */
 291                sanity_check_refcnt(sb);
 292}
 293
 294/*
 295 * Given a commit and a path in it, create a new origin structure.
 296 * The callers that add blame to the scoreboard should use
 297 * get_origin() to obtain shared, refcounted copy instead of calling
 298 * this function directly.
 299 */
 300static struct origin *make_origin(struct commit *commit, const char *path)
 301{
 302        struct origin *o;
 303        o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 304        o->commit = commit;
 305        o->refcnt = 1;
 306        strcpy(o->path, path);
 307        return o;
 308}
 309
 310/*
 311 * Locate an existing origin or create a new one.
 312 */
 313static struct origin *get_origin(struct scoreboard *sb,
 314                                 struct commit *commit,
 315                                 const char *path)
 316{
 317        struct blame_entry *e;
 318
 319        for (e = sb->ent; e; e = e->next) {
 320                if (e->suspect->commit == commit &&
 321                    !strcmp(e->suspect->path, path))
 322                        return origin_incref(e->suspect);
 323        }
 324        return make_origin(commit, path);
 325}
 326
 327/*
 328 * Fill the blob_sha1 field of an origin if it hasn't, so that later
 329 * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 330 * for an origin is also used to pass the blame for the entire file to
 331 * the parent to detect the case where a child's blob is identical to
 332 * that of its parent's.
 333 *
 334 * This also fills origin->mode for corresponding tree path.
 335 */
 336static int fill_blob_sha1_and_mode(struct origin *origin)
 337{
 338        if (!is_null_sha1(origin->blob_sha1))
 339                return 0;
 340        if (get_tree_entry(origin->commit->object.sha1,
 341                           origin->path,
 342                           origin->blob_sha1, &origin->mode))
 343                goto error_out;
 344        if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 345                goto error_out;
 346        return 0;
 347 error_out:
 348        hashclr(origin->blob_sha1);
 349        origin->mode = S_IFINVALID;
 350        return -1;
 351}
 352
 353/*
 354 * We have an origin -- check if the same path exists in the
 355 * parent and return an origin structure to represent it.
 356 */
 357static struct origin *find_origin(struct scoreboard *sb,
 358                                  struct commit *parent,
 359                                  struct origin *origin)
 360{
 361        struct origin *porigin = NULL;
 362        struct diff_options diff_opts;
 363        const char *paths[2];
 364
 365        if (parent->util) {
 366                /*
 367                 * Each commit object can cache one origin in that
 368                 * commit.  This is a freestanding copy of origin and
 369                 * not refcounted.
 370                 */
 371                struct origin *cached = parent->util;
 372                if (!strcmp(cached->path, origin->path)) {
 373                        /*
 374                         * The same path between origin and its parent
 375                         * without renaming -- the most common case.
 376                         */
 377                        porigin = get_origin(sb, parent, cached->path);
 378
 379                        /*
 380                         * If the origin was newly created (i.e. get_origin
 381                         * would call make_origin if none is found in the
 382                         * scoreboard), it does not know the blob_sha1/mode,
 383                         * so copy it.  Otherwise porigin was in the
 384                         * scoreboard and already knows blob_sha1/mode.
 385                         */
 386                        if (porigin->refcnt == 1) {
 387                                hashcpy(porigin->blob_sha1, cached->blob_sha1);
 388                                porigin->mode = cached->mode;
 389                        }
 390                        return porigin;
 391                }
 392                /* otherwise it was not very useful; free it */
 393                free(parent->util);
 394                parent->util = NULL;
 395        }
 396
 397        /* See if the origin->path is different between parent
 398         * and origin first.  Most of the time they are the
 399         * same and diff-tree is fairly efficient about this.
 400         */
 401        diff_setup(&diff_opts);
 402        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 403        diff_opts.detect_rename = 0;
 404        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 405        paths[0] = origin->path;
 406        paths[1] = NULL;
 407
 408        diff_tree_setup_paths(paths, &diff_opts);
 409        diff_setup_done(&diff_opts);
 410
 411        if (is_null_sha1(origin->commit->object.sha1))
 412                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 413        else
 414                diff_tree_sha1(parent->tree->object.sha1,
 415                               origin->commit->tree->object.sha1,
 416                               "", &diff_opts);
 417        diffcore_std(&diff_opts);
 418
 419        if (!diff_queued_diff.nr) {
 420                /* The path is the same as parent */
 421                porigin = get_origin(sb, parent, origin->path);
 422                hashcpy(porigin->blob_sha1, origin->blob_sha1);
 423                porigin->mode = origin->mode;
 424        } else {
 425                /*
 426                 * Since origin->path is a pathspec, if the parent
 427                 * commit had it as a directory, we will see a whole
 428                 * bunch of deletion of files in the directory that we
 429                 * do not care about.
 430                 */
 431                int i;
 432                struct diff_filepair *p = NULL;
 433                for (i = 0; i < diff_queued_diff.nr; i++) {
 434                        const char *name;
 435                        p = diff_queued_diff.queue[i];
 436                        name = p->one->path ? p->one->path : p->two->path;
 437                        if (!strcmp(name, origin->path))
 438                                break;
 439                }
 440                if (!p)
 441                        die("internal error in blame::find_origin");
 442                switch (p->status) {
 443                default:
 444                        die("internal error in blame::find_origin (%c)",
 445                            p->status);
 446                case 'M':
 447                        porigin = get_origin(sb, parent, origin->path);
 448                        hashcpy(porigin->blob_sha1, p->one->sha1);
 449                        porigin->mode = p->one->mode;
 450                        break;
 451                case 'A':
 452                case 'T':
 453                        /* Did not exist in parent, or type changed */
 454                        break;
 455                }
 456        }
 457        diff_flush(&diff_opts);
 458        diff_tree_release_paths(&diff_opts);
 459        if (porigin) {
 460                /*
 461                 * Create a freestanding copy that is not part of
 462                 * the refcounted origin found in the scoreboard, and
 463                 * cache it in the commit.
 464                 */
 465                struct origin *cached;
 466
 467                cached = make_origin(porigin->commit, porigin->path);
 468                hashcpy(cached->blob_sha1, porigin->blob_sha1);
 469                cached->mode = porigin->mode;
 470                parent->util = cached;
 471        }
 472        return porigin;
 473}
 474
 475/*
 476 * We have an origin -- find the path that corresponds to it in its
 477 * parent and return an origin structure to represent it.
 478 */
 479static struct origin *find_rename(struct scoreboard *sb,
 480                                  struct commit *parent,
 481                                  struct origin *origin)
 482{
 483        struct origin *porigin = NULL;
 484        struct diff_options diff_opts;
 485        int i;
 486        const char *paths[2];
 487
 488        diff_setup(&diff_opts);
 489        DIFF_OPT_SET(&diff_opts, RECURSIVE);
 490        diff_opts.detect_rename = DIFF_DETECT_RENAME;
 491        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 492        diff_opts.single_follow = origin->path;
 493        paths[0] = NULL;
 494        diff_tree_setup_paths(paths, &diff_opts);
 495        diff_setup_done(&diff_opts);
 496
 497        if (is_null_sha1(origin->commit->object.sha1))
 498                do_diff_cache(parent->tree->object.sha1, &diff_opts);
 499        else
 500                diff_tree_sha1(parent->tree->object.sha1,
 501                               origin->commit->tree->object.sha1,
 502                               "", &diff_opts);
 503        diffcore_std(&diff_opts);
 504
 505        for (i = 0; i < diff_queued_diff.nr; i++) {
 506                struct diff_filepair *p = diff_queued_diff.queue[i];
 507                if ((p->status == 'R' || p->status == 'C') &&
 508                    !strcmp(p->two->path, origin->path)) {
 509                        porigin = get_origin(sb, parent, p->one->path);
 510                        hashcpy(porigin->blob_sha1, p->one->sha1);
 511                        porigin->mode = p->one->mode;
 512                        break;
 513                }
 514        }
 515        diff_flush(&diff_opts);
 516        diff_tree_release_paths(&diff_opts);
 517        return porigin;
 518}
 519
 520/*
 521 * Link in a new blame entry to the scoreboard.  Entries that cover the
 522 * same line range have been removed from the scoreboard previously.
 523 */
 524static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 525{
 526        struct blame_entry *ent, *prev = NULL;
 527
 528        origin_incref(e->suspect);
 529
 530        for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 531                prev = ent;
 532
 533        /* prev, if not NULL, is the last one that is below e */
 534        e->prev = prev;
 535        if (prev) {
 536                e->next = prev->next;
 537                prev->next = e;
 538        }
 539        else {
 540                e->next = sb->ent;
 541                sb->ent = e;
 542        }
 543        if (e->next)
 544                e->next->prev = e;
 545}
 546
 547/*
 548 * src typically is on-stack; we want to copy the information in it to
 549 * a malloced blame_entry that is already on the linked list of the
 550 * scoreboard.  The origin of dst loses a refcnt while the origin of src
 551 * gains one.
 552 */
 553static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 554{
 555        struct blame_entry *p, *n;
 556
 557        p = dst->prev;
 558        n = dst->next;
 559        origin_incref(src->suspect);
 560        origin_decref(dst->suspect);
 561        memcpy(dst, src, sizeof(*src));
 562        dst->prev = p;
 563        dst->next = n;
 564        dst->score = 0;
 565}
 566
 567static const char *nth_line(struct scoreboard *sb, int lno)
 568{
 569        return sb->final_buf + sb->lineno[lno];
 570}
 571
 572/*
 573 * It is known that lines between tlno to same came from parent, and e
 574 * has an overlap with that range.  it also is known that parent's
 575 * line plno corresponds to e's line tlno.
 576 *
 577 *                <---- e ----->
 578 *                   <------>
 579 *                   <------------>
 580 *             <------------>
 581 *             <------------------>
 582 *
 583 * Split e into potentially three parts; before this chunk, the chunk
 584 * to be blamed for the parent, and after that portion.
 585 */
 586static void split_overlap(struct blame_entry *split,
 587                          struct blame_entry *e,
 588                          int tlno, int plno, int same,
 589                          struct origin *parent)
 590{
 591        int chunk_end_lno;
 592        memset(split, 0, sizeof(struct blame_entry [3]));
 593
 594        if (e->s_lno < tlno) {
 595                /* there is a pre-chunk part not blamed on parent */
 596                split[0].suspect = origin_incref(e->suspect);
 597                split[0].lno = e->lno;
 598                split[0].s_lno = e->s_lno;
 599                split[0].num_lines = tlno - e->s_lno;
 600                split[1].lno = e->lno + tlno - e->s_lno;
 601                split[1].s_lno = plno;
 602        }
 603        else {
 604                split[1].lno = e->lno;
 605                split[1].s_lno = plno + (e->s_lno - tlno);
 606        }
 607
 608        if (same < e->s_lno + e->num_lines) {
 609                /* there is a post-chunk part not blamed on parent */
 610                split[2].suspect = origin_incref(e->suspect);
 611                split[2].lno = e->lno + (same - e->s_lno);
 612                split[2].s_lno = e->s_lno + (same - e->s_lno);
 613                split[2].num_lines = e->s_lno + e->num_lines - same;
 614                chunk_end_lno = split[2].lno;
 615        }
 616        else
 617                chunk_end_lno = e->lno + e->num_lines;
 618        split[1].num_lines = chunk_end_lno - split[1].lno;
 619
 620        /*
 621         * if it turns out there is nothing to blame the parent for,
 622         * forget about the splitting.  !split[1].suspect signals this.
 623         */
 624        if (split[1].num_lines < 1)
 625                return;
 626        split[1].suspect = origin_incref(parent);
 627}
 628
 629/*
 630 * split_overlap() divided an existing blame e into up to three parts
 631 * in split.  Adjust the linked list of blames in the scoreboard to
 632 * reflect the split.
 633 */
 634static void split_blame(struct scoreboard *sb,
 635                        struct blame_entry *split,
 636                        struct blame_entry *e)
 637{
 638        struct blame_entry *new_entry;
 639
 640        if (split[0].suspect && split[2].suspect) {
 641                /* The first part (reuse storage for the existing entry e) */
 642                dup_entry(e, &split[0]);
 643
 644                /* The last part -- me */
 645                new_entry = xmalloc(sizeof(*new_entry));
 646                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 647                add_blame_entry(sb, new_entry);
 648
 649                /* ... and the middle part -- parent */
 650                new_entry = xmalloc(sizeof(*new_entry));
 651                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 652                add_blame_entry(sb, new_entry);
 653        }
 654        else if (!split[0].suspect && !split[2].suspect)
 655                /*
 656                 * The parent covers the entire area; reuse storage for
 657                 * e and replace it with the parent.
 658                 */
 659                dup_entry(e, &split[1]);
 660        else if (split[0].suspect) {
 661                /* me and then parent */
 662                dup_entry(e, &split[0]);
 663
 664                new_entry = xmalloc(sizeof(*new_entry));
 665                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 666                add_blame_entry(sb, new_entry);
 667        }
 668        else {
 669                /* parent and then me */
 670                dup_entry(e, &split[1]);
 671
 672                new_entry = xmalloc(sizeof(*new_entry));
 673                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 674                add_blame_entry(sb, new_entry);
 675        }
 676
 677        if (DEBUG) { /* sanity */
 678                struct blame_entry *ent;
 679                int lno = sb->ent->lno, corrupt = 0;
 680
 681                for (ent = sb->ent; ent; ent = ent->next) {
 682                        if (lno != ent->lno)
 683                                corrupt = 1;
 684                        if (ent->s_lno < 0)
 685                                corrupt = 1;
 686                        lno += ent->num_lines;
 687                }
 688                if (corrupt) {
 689                        lno = sb->ent->lno;
 690                        for (ent = sb->ent; ent; ent = ent->next) {
 691                                printf("L %8d l %8d n %8d\n",
 692                                       lno, ent->lno, ent->num_lines);
 693                                lno = ent->lno + ent->num_lines;
 694                        }
 695                        die("oops");
 696                }
 697        }
 698}
 699
 700/*
 701 * After splitting the blame, the origins used by the
 702 * on-stack blame_entry should lose one refcnt each.
 703 */
 704static void decref_split(struct blame_entry *split)
 705{
 706        int i;
 707
 708        for (i = 0; i < 3; i++)
 709                origin_decref(split[i].suspect);
 710}
 711
 712/*
 713 * Helper for blame_chunk().  blame_entry e is known to overlap with
 714 * the patch hunk; split it and pass blame to the parent.
 715 */
 716static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 717                          int tlno, int plno, int same,
 718                          struct origin *parent)
 719{
 720        struct blame_entry split[3];
 721
 722        split_overlap(split, e, tlno, plno, same, parent);
 723        if (split[1].suspect)
 724                split_blame(sb, split, e);
 725        decref_split(split);
 726}
 727
 728/*
 729 * Find the line number of the last line the target is suspected for.
 730 */
 731static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 732{
 733        struct blame_entry *e;
 734        int last_in_target = -1;
 735
 736        for (e = sb->ent; e; e = e->next) {
 737                if (e->guilty || !same_suspect(e->suspect, target))
 738                        continue;
 739                if (last_in_target < e->s_lno + e->num_lines)
 740                        last_in_target = e->s_lno + e->num_lines;
 741        }
 742        return last_in_target;
 743}
 744
 745/*
 746 * Process one hunk from the patch between the current suspect for
 747 * blame_entry e and its parent.  Find and split the overlap, and
 748 * pass blame to the overlapping part to the parent.
 749 */
 750static void blame_chunk(struct scoreboard *sb,
 751                        int tlno, int plno, int same,
 752                        struct origin *target, struct origin *parent)
 753{
 754        struct blame_entry *e;
 755
 756        for (e = sb->ent; e; e = e->next) {
 757                if (e->guilty || !same_suspect(e->suspect, target))
 758                        continue;
 759                if (same <= e->s_lno)
 760                        continue;
 761                if (tlno < e->s_lno + e->num_lines)
 762                        blame_overlap(sb, e, tlno, plno, same, parent);
 763        }
 764}
 765
 766struct blame_chunk_cb_data {
 767        struct scoreboard *sb;
 768        struct origin *target;
 769        struct origin *parent;
 770        long plno;
 771        long tlno;
 772};
 773
 774static int blame_chunk_cb(long start_a, long count_a,
 775                          long start_b, long count_b, void *data)
 776{
 777        struct blame_chunk_cb_data *d = data;
 778        blame_chunk(d->sb, d->tlno, d->plno, start_b, d->target, d->parent);
 779        d->plno = start_a + count_a;
 780        d->tlno = start_b + count_b;
 781        return 0;
 782}
 783
 784/*
 785 * We are looking at the origin 'target' and aiming to pass blame
 786 * for the lines it is suspected to its parent.  Run diff to find
 787 * which lines came from parent and pass blame for them.
 788 */
 789static int pass_blame_to_parent(struct scoreboard *sb,
 790                                struct origin *target,
 791                                struct origin *parent)
 792{
 793        int last_in_target;
 794        mmfile_t file_p, file_o;
 795        struct blame_chunk_cb_data d;
 796
 797        memset(&d, 0, sizeof(d));
 798        d.sb = sb; d.target = target; d.parent = parent;
 799        last_in_target = find_last_in_target(sb, target);
 800        if (last_in_target < 0)
 801                return 1; /* nothing remains for this target */
 802
 803        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 804        fill_origin_blob(&sb->revs->diffopt, target, &file_o);
 805        num_get_patch++;
 806
 807        diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
 808        /* The rest (i.e. anything after tlno) are the same as the parent */
 809        blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
 810
 811        return 0;
 812}
 813
 814/*
 815 * The lines in blame_entry after splitting blames many times can become
 816 * very small and trivial, and at some point it becomes pointless to
 817 * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 818 * ordinary C program, and it is not worth to say it was copied from
 819 * totally unrelated file in the parent.
 820 *
 821 * Compute how trivial the lines in the blame_entry are.
 822 */
 823static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 824{
 825        unsigned score;
 826        const char *cp, *ep;
 827
 828        if (e->score)
 829                return e->score;
 830
 831        score = 1;
 832        cp = nth_line(sb, e->lno);
 833        ep = nth_line(sb, e->lno + e->num_lines);
 834        while (cp < ep) {
 835                unsigned ch = *((unsigned char *)cp);
 836                if (isalnum(ch))
 837                        score++;
 838                cp++;
 839        }
 840        e->score = score;
 841        return score;
 842}
 843
 844/*
 845 * best_so_far[] and this[] are both a split of an existing blame_entry
 846 * that passes blame to the parent.  Maintain best_so_far the best split
 847 * so far, by comparing this and best_so_far and copying this into
 848 * bst_so_far as needed.
 849 */
 850static void copy_split_if_better(struct scoreboard *sb,
 851                                 struct blame_entry *best_so_far,
 852                                 struct blame_entry *this)
 853{
 854        int i;
 855
 856        if (!this[1].suspect)
 857                return;
 858        if (best_so_far[1].suspect) {
 859                if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 860                        return;
 861        }
 862
 863        for (i = 0; i < 3; i++)
 864                origin_incref(this[i].suspect);
 865        decref_split(best_so_far);
 866        memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 867}
 868
 869/*
 870 * We are looking at a part of the final image represented by
 871 * ent (tlno and same are offset by ent->s_lno).
 872 * tlno is where we are looking at in the final image.
 873 * up to (but not including) same match preimage.
 874 * plno is where we are looking at in the preimage.
 875 *
 876 * <-------------- final image ---------------------->
 877 *       <------ent------>
 878 *         ^tlno ^same
 879 *    <---------preimage----->
 880 *         ^plno
 881 *
 882 * All line numbers are 0-based.
 883 */
 884static void handle_split(struct scoreboard *sb,
 885                         struct blame_entry *ent,
 886                         int tlno, int plno, int same,
 887                         struct origin *parent,
 888                         struct blame_entry *split)
 889{
 890        if (ent->num_lines <= tlno)
 891                return;
 892        if (tlno < same) {
 893                struct blame_entry this[3];
 894                tlno += ent->s_lno;
 895                same += ent->s_lno;
 896                split_overlap(this, ent, tlno, plno, same, parent);
 897                copy_split_if_better(sb, split, this);
 898                decref_split(this);
 899        }
 900}
 901
 902struct handle_split_cb_data {
 903        struct scoreboard *sb;
 904        struct blame_entry *ent;
 905        struct origin *parent;
 906        struct blame_entry *split;
 907        long plno;
 908        long tlno;
 909};
 910
 911static int handle_split_cb(long start_a, long count_a,
 912                           long start_b, long count_b, void *data)
 913{
 914        struct handle_split_cb_data *d = data;
 915        handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
 916                     d->split);
 917        d->plno = start_a + count_a;
 918        d->tlno = start_b + count_b;
 919        return 0;
 920}
 921
 922/*
 923 * Find the lines from parent that are the same as ent so that
 924 * we can pass blames to it.  file_p has the blob contents for
 925 * the parent.
 926 */
 927static void find_copy_in_blob(struct scoreboard *sb,
 928                              struct blame_entry *ent,
 929                              struct origin *parent,
 930                              struct blame_entry *split,
 931                              mmfile_t *file_p)
 932{
 933        const char *cp;
 934        int cnt;
 935        mmfile_t file_o;
 936        struct handle_split_cb_data d;
 937
 938        memset(&d, 0, sizeof(d));
 939        d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
 940        /*
 941         * Prepare mmfile that contains only the lines in ent.
 942         */
 943        cp = nth_line(sb, ent->lno);
 944        file_o.ptr = (char *) cp;
 945        cnt = ent->num_lines;
 946
 947        while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 948                if (*cp++ == '\n')
 949                        cnt--;
 950        }
 951        file_o.size = cp - file_o.ptr;
 952
 953        /*
 954         * file_o is a part of final image we are annotating.
 955         * file_p partially may match that image.
 956         */
 957        memset(split, 0, sizeof(struct blame_entry [3]));
 958        diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
 959        /* remainder, if any, all match the preimage */
 960        handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
 961}
 962
 963/*
 964 * See if lines currently target is suspected for can be attributed to
 965 * parent.
 966 */
 967static int find_move_in_parent(struct scoreboard *sb,
 968                               struct origin *target,
 969                               struct origin *parent)
 970{
 971        int last_in_target, made_progress;
 972        struct blame_entry *e, split[3];
 973        mmfile_t file_p;
 974
 975        last_in_target = find_last_in_target(sb, target);
 976        if (last_in_target < 0)
 977                return 1; /* nothing remains for this target */
 978
 979        fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 980        if (!file_p.ptr)
 981                return 0;
 982
 983        made_progress = 1;
 984        while (made_progress) {
 985                made_progress = 0;
 986                for (e = sb->ent; e; e = e->next) {
 987                        if (e->guilty || !same_suspect(e->suspect, target) ||
 988                            ent_score(sb, e) < blame_move_score)
 989                                continue;
 990                        find_copy_in_blob(sb, e, parent, split, &file_p);
 991                        if (split[1].suspect &&
 992                            blame_move_score < ent_score(sb, &split[1])) {
 993                                split_blame(sb, split, e);
 994                                made_progress = 1;
 995                        }
 996                        decref_split(split);
 997                }
 998        }
 999        return 0;
1000}
1001
1002struct blame_list {
1003        struct blame_entry *ent;
1004        struct blame_entry split[3];
1005};
1006
1007/*
1008 * Count the number of entries the target is suspected for,
1009 * and prepare a list of entry and the best split.
1010 */
1011static struct blame_list *setup_blame_list(struct scoreboard *sb,
1012                                           struct origin *target,
1013                                           int min_score,
1014                                           int *num_ents_p)
1015{
1016        struct blame_entry *e;
1017        int num_ents, i;
1018        struct blame_list *blame_list = NULL;
1019
1020        for (e = sb->ent, num_ents = 0; e; e = e->next)
1021                if (!e->scanned && !e->guilty &&
1022                    same_suspect(e->suspect, target) &&
1023                    min_score < ent_score(sb, e))
1024                        num_ents++;
1025        if (num_ents) {
1026                blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1027                for (e = sb->ent, i = 0; e; e = e->next)
1028                        if (!e->scanned && !e->guilty &&
1029                            same_suspect(e->suspect, target) &&
1030                            min_score < ent_score(sb, e))
1031                                blame_list[i++].ent = e;
1032        }
1033        *num_ents_p = num_ents;
1034        return blame_list;
1035}
1036
1037/*
1038 * Reset the scanned status on all entries.
1039 */
1040static void reset_scanned_flag(struct scoreboard *sb)
1041{
1042        struct blame_entry *e;
1043        for (e = sb->ent; e; e = e->next)
1044                e->scanned = 0;
1045}
1046
1047/*
1048 * For lines target is suspected for, see if we can find code movement
1049 * across file boundary from the parent commit.  porigin is the path
1050 * in the parent we already tried.
1051 */
1052static int find_copy_in_parent(struct scoreboard *sb,
1053                               struct origin *target,
1054                               struct commit *parent,
1055                               struct origin *porigin,
1056                               int opt)
1057{
1058        struct diff_options diff_opts;
1059        const char *paths[1];
1060        int i, j;
1061        int retval;
1062        struct blame_list *blame_list;
1063        int num_ents;
1064
1065        blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1066        if (!blame_list)
1067                return 1; /* nothing remains for this target */
1068
1069        diff_setup(&diff_opts);
1070        DIFF_OPT_SET(&diff_opts, RECURSIVE);
1071        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1072
1073        paths[0] = NULL;
1074        diff_tree_setup_paths(paths, &diff_opts);
1075        diff_setup_done(&diff_opts);
1076
1077        /* Try "find copies harder" on new path if requested;
1078         * we do not want to use diffcore_rename() actually to
1079         * match things up; find_copies_harder is set only to
1080         * force diff_tree_sha1() to feed all filepairs to diff_queue,
1081         * and this code needs to be after diff_setup_done(), which
1082         * usually makes find-copies-harder imply copy detection.
1083         */
1084        if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1085            || ((opt & PICKAXE_BLAME_COPY_HARDER)
1086                && (!porigin || strcmp(target->path, porigin->path))))
1087                DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1088
1089        if (is_null_sha1(target->commit->object.sha1))
1090                do_diff_cache(parent->tree->object.sha1, &diff_opts);
1091        else
1092                diff_tree_sha1(parent->tree->object.sha1,
1093                               target->commit->tree->object.sha1,
1094                               "", &diff_opts);
1095
1096        if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1097                diffcore_std(&diff_opts);
1098
1099        retval = 0;
1100        while (1) {
1101                int made_progress = 0;
1102
1103                for (i = 0; i < diff_queued_diff.nr; i++) {
1104                        struct diff_filepair *p = diff_queued_diff.queue[i];
1105                        struct origin *norigin;
1106                        mmfile_t file_p;
1107                        struct blame_entry this[3];
1108
1109                        if (!DIFF_FILE_VALID(p->one))
1110                                continue; /* does not exist in parent */
1111                        if (S_ISGITLINK(p->one->mode))
1112                                continue; /* ignore git links */
1113                        if (porigin && !strcmp(p->one->path, porigin->path))
1114                                /* find_move already dealt with this path */
1115                                continue;
1116
1117                        norigin = get_origin(sb, parent, p->one->path);
1118                        hashcpy(norigin->blob_sha1, p->one->sha1);
1119                        norigin->mode = p->one->mode;
1120                        fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1121                        if (!file_p.ptr)
1122                                continue;
1123
1124                        for (j = 0; j < num_ents; j++) {
1125                                find_copy_in_blob(sb, blame_list[j].ent,
1126                                                  norigin, this, &file_p);
1127                                copy_split_if_better(sb, blame_list[j].split,
1128                                                     this);
1129                                decref_split(this);
1130                        }
1131                        origin_decref(norigin);
1132                }
1133
1134                for (j = 0; j < num_ents; j++) {
1135                        struct blame_entry *split = blame_list[j].split;
1136                        if (split[1].suspect &&
1137                            blame_copy_score < ent_score(sb, &split[1])) {
1138                                split_blame(sb, split, blame_list[j].ent);
1139                                made_progress = 1;
1140                        }
1141                        else
1142                                blame_list[j].ent->scanned = 1;
1143                        decref_split(split);
1144                }
1145                free(blame_list);
1146
1147                if (!made_progress)
1148                        break;
1149                blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1150                if (!blame_list) {
1151                        retval = 1;
1152                        break;
1153                }
1154        }
1155        reset_scanned_flag(sb);
1156        diff_flush(&diff_opts);
1157        diff_tree_release_paths(&diff_opts);
1158        return retval;
1159}
1160
1161/*
1162 * The blobs of origin and porigin exactly match, so everything
1163 * origin is suspected for can be blamed on the parent.
1164 */
1165static void pass_whole_blame(struct scoreboard *sb,
1166                             struct origin *origin, struct origin *porigin)
1167{
1168        struct blame_entry *e;
1169
1170        if (!porigin->file.ptr && origin->file.ptr) {
1171                /* Steal its file */
1172                porigin->file = origin->file;
1173                origin->file.ptr = NULL;
1174        }
1175        for (e = sb->ent; e; e = e->next) {
1176                if (!same_suspect(e->suspect, origin))
1177                        continue;
1178                origin_incref(porigin);
1179                origin_decref(e->suspect);
1180                e->suspect = porigin;
1181        }
1182}
1183
1184/*
1185 * We pass blame from the current commit to its parents.  We keep saying
1186 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1187 * exonerate ourselves.
1188 */
1189static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1190{
1191        if (!reverse)
1192                return commit->parents;
1193        return lookup_decoration(&revs->children, &commit->object);
1194}
1195
1196static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1197{
1198        int cnt;
1199        struct commit_list *l = first_scapegoat(revs, commit);
1200        for (cnt = 0; l; l = l->next)
1201                cnt++;
1202        return cnt;
1203}
1204
1205#define MAXSG 16
1206
1207static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1208{
1209        struct rev_info *revs = sb->revs;
1210        int i, pass, num_sg;
1211        struct commit *commit = origin->commit;
1212        struct commit_list *sg;
1213        struct origin *sg_buf[MAXSG];
1214        struct origin *porigin, **sg_origin = sg_buf;
1215
1216        num_sg = num_scapegoats(revs, commit);
1217        if (!num_sg)
1218                goto finish;
1219        else if (num_sg < ARRAY_SIZE(sg_buf))
1220                memset(sg_buf, 0, sizeof(sg_buf));
1221        else
1222                sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1223
1224        /*
1225         * The first pass looks for unrenamed path to optimize for
1226         * common cases, then we look for renames in the second pass.
1227         */
1228        for (pass = 0; pass < 2; pass++) {
1229                struct origin *(*find)(struct scoreboard *,
1230                                       struct commit *, struct origin *);
1231                find = pass ? find_rename : find_origin;
1232
1233                for (i = 0, sg = first_scapegoat(revs, commit);
1234                     i < num_sg && sg;
1235                     sg = sg->next, i++) {
1236                        struct commit *p = sg->item;
1237                        int j, same;
1238
1239                        if (sg_origin[i])
1240                                continue;
1241                        if (parse_commit(p))
1242                                continue;
1243                        porigin = find(sb, p, origin);
1244                        if (!porigin)
1245                                continue;
1246                        if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1247                                pass_whole_blame(sb, origin, porigin);
1248                                origin_decref(porigin);
1249                                goto finish;
1250                        }
1251                        for (j = same = 0; j < i; j++)
1252                                if (sg_origin[j] &&
1253                                    !hashcmp(sg_origin[j]->blob_sha1,
1254                                             porigin->blob_sha1)) {
1255                                        same = 1;
1256                                        break;
1257                                }
1258                        if (!same)
1259                                sg_origin[i] = porigin;
1260                        else
1261                                origin_decref(porigin);
1262                }
1263        }
1264
1265        num_commits++;
1266        for (i = 0, sg = first_scapegoat(revs, commit);
1267             i < num_sg && sg;
1268             sg = sg->next, i++) {
1269                struct origin *porigin = sg_origin[i];
1270                if (!porigin)
1271                        continue;
1272                if (!origin->previous) {
1273                        origin_incref(porigin);
1274                        origin->previous = porigin;
1275                }
1276                if (pass_blame_to_parent(sb, origin, porigin))
1277                        goto finish;
1278        }
1279
1280        /*
1281         * Optionally find moves in parents' files.
1282         */
1283        if (opt & PICKAXE_BLAME_MOVE)
1284                for (i = 0, sg = first_scapegoat(revs, commit);
1285                     i < num_sg && sg;
1286                     sg = sg->next, i++) {
1287                        struct origin *porigin = sg_origin[i];
1288                        if (!porigin)
1289                                continue;
1290                        if (find_move_in_parent(sb, origin, porigin))
1291                                goto finish;
1292                }
1293
1294        /*
1295         * Optionally find copies from parents' files.
1296         */
1297        if (opt & PICKAXE_BLAME_COPY)
1298                for (i = 0, sg = first_scapegoat(revs, commit);
1299                     i < num_sg && sg;
1300                     sg = sg->next, i++) {
1301                        struct origin *porigin = sg_origin[i];
1302                        if (find_copy_in_parent(sb, origin, sg->item,
1303                                                porigin, opt))
1304                                goto finish;
1305                }
1306
1307 finish:
1308        for (i = 0; i < num_sg; i++) {
1309                if (sg_origin[i]) {
1310                        drop_origin_blob(sg_origin[i]);
1311                        origin_decref(sg_origin[i]);
1312                }
1313        }
1314        drop_origin_blob(origin);
1315        if (sg_buf != sg_origin)
1316                free(sg_origin);
1317}
1318
1319/*
1320 * Information on commits, used for output.
1321 */
1322struct commit_info {
1323        const char *author;
1324        const char *author_mail;
1325        unsigned long author_time;
1326        const char *author_tz;
1327
1328        /* filled only when asked for details */
1329        const char *committer;
1330        const char *committer_mail;
1331        unsigned long committer_time;
1332        const char *committer_tz;
1333
1334        const char *summary;
1335};
1336
1337/*
1338 * Parse author/committer line in the commit object buffer
1339 */
1340static void get_ac_line(const char *inbuf, const char *what,
1341                        int person_len, char *person,
1342                        int mail_len, char *mail,
1343                        unsigned long *time, const char **tz)
1344{
1345        int len, tzlen, maillen;
1346        char *tmp, *endp, *timepos, *mailpos;
1347
1348        tmp = strstr(inbuf, what);
1349        if (!tmp)
1350                goto error_out;
1351        tmp += strlen(what);
1352        endp = strchr(tmp, '\n');
1353        if (!endp)
1354                len = strlen(tmp);
1355        else
1356                len = endp - tmp;
1357        if (person_len <= len) {
1358        error_out:
1359                /* Ugh */
1360                *tz = "(unknown)";
1361                strcpy(person, *tz);
1362                strcpy(mail, *tz);
1363                *time = 0;
1364                return;
1365        }
1366        memcpy(person, tmp, len);
1367
1368        tmp = person;
1369        tmp += len;
1370        *tmp = 0;
1371        while (person < tmp && *tmp != ' ')
1372                tmp--;
1373        if (tmp <= person)
1374                goto error_out;
1375        *tz = tmp+1;
1376        tzlen = (person+len)-(tmp+1);
1377
1378        *tmp = 0;
1379        while (person < tmp && *tmp != ' ')
1380                tmp--;
1381        if (tmp <= person)
1382                goto error_out;
1383        *time = strtoul(tmp, NULL, 10);
1384        timepos = tmp;
1385
1386        *tmp = 0;
1387        while (person < tmp && !(*tmp == ' ' && tmp[1] == '<'))
1388                tmp--;
1389        if (tmp <= person)
1390                return;
1391        mailpos = tmp + 1;
1392        *tmp = 0;
1393        maillen = timepos - tmp;
1394        memcpy(mail, mailpos, maillen);
1395
1396        if (!mailmap.nr)
1397                return;
1398
1399        /*
1400         * mailmap expansion may make the name longer.
1401         * make room by pushing stuff down.
1402         */
1403        tmp = person + person_len - (tzlen + 1);
1404        memmove(tmp, *tz, tzlen);
1405        tmp[tzlen] = 0;
1406        *tz = tmp;
1407
1408        /*
1409         * Now, convert both name and e-mail using mailmap
1410         */
1411        if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1412                /* Add a trailing '>' to email, since map_user returns plain emails
1413                   Note: It already has '<', since we replace from mail+1 */
1414                mailpos = memchr(mail, '\0', mail_len);
1415                if (mailpos && mailpos-mail < mail_len - 1) {
1416                        *mailpos = '>';
1417                        *(mailpos+1) = '\0';
1418                }
1419        }
1420}
1421
1422static void get_commit_info(struct commit *commit,
1423                            struct commit_info *ret,
1424                            int detailed)
1425{
1426        int len;
1427        const char *subject;
1428        char *reencoded, *message;
1429        static char author_name[1024];
1430        static char author_mail[1024];
1431        static char committer_name[1024];
1432        static char committer_mail[1024];
1433        static char summary_buf[1024];
1434
1435        /*
1436         * We've operated without save_commit_buffer, so
1437         * we now need to populate them for output.
1438         */
1439        if (!commit->buffer) {
1440                enum object_type type;
1441                unsigned long size;
1442                commit->buffer =
1443                        read_sha1_file(commit->object.sha1, &type, &size);
1444                if (!commit->buffer)
1445                        die("Cannot read commit %s",
1446                            sha1_to_hex(commit->object.sha1));
1447        }
1448        reencoded = reencode_commit_message(commit, NULL);
1449        message   = reencoded ? reencoded : commit->buffer;
1450        ret->author = author_name;
1451        ret->author_mail = author_mail;
1452        get_ac_line(message, "\nauthor ",
1453                    sizeof(author_name), author_name,
1454                    sizeof(author_mail), author_mail,
1455                    &ret->author_time, &ret->author_tz);
1456
1457        if (!detailed) {
1458                free(reencoded);
1459                return;
1460        }
1461
1462        ret->committer = committer_name;
1463        ret->committer_mail = committer_mail;
1464        get_ac_line(message, "\ncommitter ",
1465                    sizeof(committer_name), committer_name,
1466                    sizeof(committer_mail), committer_mail,
1467                    &ret->committer_time, &ret->committer_tz);
1468
1469        ret->summary = summary_buf;
1470        len = find_commit_subject(message, &subject);
1471        if (len && len < sizeof(summary_buf)) {
1472                memcpy(summary_buf, subject, len);
1473                summary_buf[len] = 0;
1474        } else {
1475                sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1476        }
1477        free(reencoded);
1478}
1479
1480/*
1481 * To allow LF and other nonportable characters in pathnames,
1482 * they are c-style quoted as needed.
1483 */
1484static void write_filename_info(const char *path)
1485{
1486        printf("filename ");
1487        write_name_quoted(path, stdout, '\n');
1488}
1489
1490/*
1491 * Porcelain/Incremental format wants to show a lot of details per
1492 * commit.  Instead of repeating this every line, emit it only once,
1493 * the first time each commit appears in the output (unless the
1494 * user has specifically asked for us to repeat).
1495 */
1496static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1497{
1498        struct commit_info ci;
1499
1500        if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1501                return 0;
1502
1503        suspect->commit->object.flags |= METAINFO_SHOWN;
1504        get_commit_info(suspect->commit, &ci, 1);
1505        printf("author %s\n", ci.author);
1506        printf("author-mail %s\n", ci.author_mail);
1507        printf("author-time %lu\n", ci.author_time);
1508        printf("author-tz %s\n", ci.author_tz);
1509        printf("committer %s\n", ci.committer);
1510        printf("committer-mail %s\n", ci.committer_mail);
1511        printf("committer-time %lu\n", ci.committer_time);
1512        printf("committer-tz %s\n", ci.committer_tz);
1513        printf("summary %s\n", ci.summary);
1514        if (suspect->commit->object.flags & UNINTERESTING)
1515                printf("boundary\n");
1516        if (suspect->previous) {
1517                struct origin *prev = suspect->previous;
1518                printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1519                write_name_quoted(prev->path, stdout, '\n');
1520        }
1521        return 1;
1522}
1523
1524/*
1525 * The blame_entry is found to be guilty for the range.  Mark it
1526 * as such, and show it in incremental output.
1527 */
1528static void found_guilty_entry(struct blame_entry *ent)
1529{
1530        if (ent->guilty)
1531                return;
1532        ent->guilty = 1;
1533        if (incremental) {
1534                struct origin *suspect = ent->suspect;
1535
1536                printf("%s %d %d %d\n",
1537                       sha1_to_hex(suspect->commit->object.sha1),
1538                       ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1539                emit_one_suspect_detail(suspect, 0);
1540                write_filename_info(suspect->path);
1541                maybe_flush_or_die(stdout, "stdout");
1542        }
1543}
1544
1545/*
1546 * The main loop -- while the scoreboard has lines whose true origin
1547 * is still unknown, pick one blame_entry, and allow its current
1548 * suspect to pass blames to its parents.
1549 */
1550static void assign_blame(struct scoreboard *sb, int opt)
1551{
1552        struct rev_info *revs = sb->revs;
1553
1554        while (1) {
1555                struct blame_entry *ent;
1556                struct commit *commit;
1557                struct origin *suspect = NULL;
1558
1559                /* find one suspect to break down */
1560                for (ent = sb->ent; !suspect && ent; ent = ent->next)
1561                        if (!ent->guilty)
1562                                suspect = ent->suspect;
1563                if (!suspect)
1564                        return; /* all done */
1565
1566                /*
1567                 * We will use this suspect later in the loop,
1568                 * so hold onto it in the meantime.
1569                 */
1570                origin_incref(suspect);
1571                commit = suspect->commit;
1572                if (!commit->object.parsed)
1573                        parse_commit(commit);
1574                if (reverse ||
1575                    (!(commit->object.flags & UNINTERESTING) &&
1576                     !(revs->max_age != -1 && commit->date < revs->max_age)))
1577                        pass_blame(sb, suspect, opt);
1578                else {
1579                        commit->object.flags |= UNINTERESTING;
1580                        if (commit->object.parsed)
1581                                mark_parents_uninteresting(commit);
1582                }
1583                /* treat root commit as boundary */
1584                if (!commit->parents && !show_root)
1585                        commit->object.flags |= UNINTERESTING;
1586
1587                /* Take responsibility for the remaining entries */
1588                for (ent = sb->ent; ent; ent = ent->next)
1589                        if (same_suspect(ent->suspect, suspect))
1590                                found_guilty_entry(ent);
1591                origin_decref(suspect);
1592
1593                if (DEBUG) /* sanity */
1594                        sanity_check_refcnt(sb);
1595        }
1596}
1597
1598static const char *format_time(unsigned long time, const char *tz_str,
1599                               int show_raw_time)
1600{
1601        static char time_buf[128];
1602        const char *time_str;
1603        int time_len;
1604        int tz;
1605
1606        if (show_raw_time) {
1607                snprintf(time_buf, sizeof(time_buf), "%lu %s", time, tz_str);
1608        }
1609        else {
1610                tz = atoi(tz_str);
1611                time_str = show_date(time, tz, blame_date_mode);
1612                time_len = strlen(time_str);
1613                memcpy(time_buf, time_str, time_len);
1614                memset(time_buf + time_len, ' ', blame_date_width - time_len);
1615        }
1616        return time_buf;
1617}
1618
1619#define OUTPUT_ANNOTATE_COMPAT  001
1620#define OUTPUT_LONG_OBJECT_NAME 002
1621#define OUTPUT_RAW_TIMESTAMP    004
1622#define OUTPUT_PORCELAIN        010
1623#define OUTPUT_SHOW_NAME        020
1624#define OUTPUT_SHOW_NUMBER      040
1625#define OUTPUT_SHOW_SCORE      0100
1626#define OUTPUT_NO_AUTHOR       0200
1627#define OUTPUT_SHOW_EMAIL       0400
1628#define OUTPUT_LINE_PORCELAIN 01000
1629
1630static void emit_porcelain_details(struct origin *suspect, int repeat)
1631{
1632        if (emit_one_suspect_detail(suspect, repeat) ||
1633            (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1634                write_filename_info(suspect->path);
1635}
1636
1637static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1638                           int opt)
1639{
1640        int repeat = opt & OUTPUT_LINE_PORCELAIN;
1641        int cnt;
1642        const char *cp;
1643        struct origin *suspect = ent->suspect;
1644        char hex[41];
1645
1646        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1647        printf("%s%c%d %d %d\n",
1648               hex,
1649               ent->guilty ? ' ' : '*', /* purely for debugging */
1650               ent->s_lno + 1,
1651               ent->lno + 1,
1652               ent->num_lines);
1653        emit_porcelain_details(suspect, repeat);
1654
1655        cp = nth_line(sb, ent->lno);
1656        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1657                char ch;
1658                if (cnt) {
1659                        printf("%s %d %d\n", hex,
1660                               ent->s_lno + 1 + cnt,
1661                               ent->lno + 1 + cnt);
1662                        if (repeat)
1663                                emit_porcelain_details(suspect, 1);
1664                }
1665                putchar('\t');
1666                do {
1667                        ch = *cp++;
1668                        putchar(ch);
1669                } while (ch != '\n' &&
1670                         cp < sb->final_buf + sb->final_buf_size);
1671        }
1672
1673        if (sb->final_buf_size && cp[-1] != '\n')
1674                putchar('\n');
1675}
1676
1677static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1678{
1679        int cnt;
1680        const char *cp;
1681        struct origin *suspect = ent->suspect;
1682        struct commit_info ci;
1683        char hex[41];
1684        int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1685
1686        get_commit_info(suspect->commit, &ci, 1);
1687        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1688
1689        cp = nth_line(sb, ent->lno);
1690        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1691                char ch;
1692                int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1693
1694                if (suspect->commit->object.flags & UNINTERESTING) {
1695                        if (blank_boundary)
1696                                memset(hex, ' ', length);
1697                        else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1698                                length--;
1699                                putchar('^');
1700                        }
1701                }
1702
1703                printf("%.*s", length, hex);
1704                if (opt & OUTPUT_ANNOTATE_COMPAT) {
1705                        const char *name;
1706                        if (opt & OUTPUT_SHOW_EMAIL)
1707                                name = ci.author_mail;
1708                        else
1709                                name = ci.author;
1710                        printf("\t(%10s\t%10s\t%d)", name,
1711                               format_time(ci.author_time, ci.author_tz,
1712                                           show_raw_time),
1713                               ent->lno + 1 + cnt);
1714                } else {
1715                        if (opt & OUTPUT_SHOW_SCORE)
1716                                printf(" %*d %02d",
1717                                       max_score_digits, ent->score,
1718                                       ent->suspect->refcnt);
1719                        if (opt & OUTPUT_SHOW_NAME)
1720                                printf(" %-*.*s", longest_file, longest_file,
1721                                       suspect->path);
1722                        if (opt & OUTPUT_SHOW_NUMBER)
1723                                printf(" %*d", max_orig_digits,
1724                                       ent->s_lno + 1 + cnt);
1725
1726                        if (!(opt & OUTPUT_NO_AUTHOR)) {
1727                                const char *name;
1728                                int pad;
1729                                if (opt & OUTPUT_SHOW_EMAIL)
1730                                        name = ci.author_mail;
1731                                else
1732                                        name = ci.author;
1733                                pad = longest_author - utf8_strwidth(name);
1734                                printf(" (%s%*s %10s",
1735                                       name, pad, "",
1736                                       format_time(ci.author_time,
1737                                                   ci.author_tz,
1738                                                   show_raw_time));
1739                        }
1740                        printf(" %*d) ",
1741                               max_digits, ent->lno + 1 + cnt);
1742                }
1743                do {
1744                        ch = *cp++;
1745                        putchar(ch);
1746                } while (ch != '\n' &&
1747                         cp < sb->final_buf + sb->final_buf_size);
1748        }
1749
1750        if (sb->final_buf_size && cp[-1] != '\n')
1751                putchar('\n');
1752}
1753
1754static void output(struct scoreboard *sb, int option)
1755{
1756        struct blame_entry *ent;
1757
1758        if (option & OUTPUT_PORCELAIN) {
1759                for (ent = sb->ent; ent; ent = ent->next) {
1760                        struct blame_entry *oth;
1761                        struct origin *suspect = ent->suspect;
1762                        struct commit *commit = suspect->commit;
1763                        if (commit->object.flags & MORE_THAN_ONE_PATH)
1764                                continue;
1765                        for (oth = ent->next; oth; oth = oth->next) {
1766                                if ((oth->suspect->commit != commit) ||
1767                                    !strcmp(oth->suspect->path, suspect->path))
1768                                        continue;
1769                                commit->object.flags |= MORE_THAN_ONE_PATH;
1770                                break;
1771                        }
1772                }
1773        }
1774
1775        for (ent = sb->ent; ent; ent = ent->next) {
1776                if (option & OUTPUT_PORCELAIN)
1777                        emit_porcelain(sb, ent, option);
1778                else {
1779                        emit_other(sb, ent, option);
1780                }
1781        }
1782}
1783
1784/*
1785 * To allow quick access to the contents of nth line in the
1786 * final image, prepare an index in the scoreboard.
1787 */
1788static int prepare_lines(struct scoreboard *sb)
1789{
1790        const char *buf = sb->final_buf;
1791        unsigned long len = sb->final_buf_size;
1792        int num = 0, incomplete = 0, bol = 1;
1793
1794        if (len && buf[len-1] != '\n')
1795                incomplete++; /* incomplete line at the end */
1796        while (len--) {
1797                if (bol) {
1798                        sb->lineno = xrealloc(sb->lineno,
1799                                              sizeof(int *) * (num + 1));
1800                        sb->lineno[num] = buf - sb->final_buf;
1801                        bol = 0;
1802                }
1803                if (*buf++ == '\n') {
1804                        num++;
1805                        bol = 1;
1806                }
1807        }
1808        sb->lineno = xrealloc(sb->lineno,
1809                              sizeof(int *) * (num + incomplete + 1));
1810        sb->lineno[num + incomplete] = buf - sb->final_buf;
1811        sb->num_lines = num + incomplete;
1812        return sb->num_lines;
1813}
1814
1815/*
1816 * Add phony grafts for use with -S; this is primarily to
1817 * support git's cvsserver that wants to give a linear history
1818 * to its clients.
1819 */
1820static int read_ancestry(const char *graft_file)
1821{
1822        FILE *fp = fopen(graft_file, "r");
1823        char buf[1024];
1824        if (!fp)
1825                return -1;
1826        while (fgets(buf, sizeof(buf), fp)) {
1827                /* The format is just "Commit Parent1 Parent2 ...\n" */
1828                int len = strlen(buf);
1829                struct commit_graft *graft = read_graft_line(buf, len);
1830                if (graft)
1831                        register_commit_graft(graft, 0);
1832        }
1833        fclose(fp);
1834        return 0;
1835}
1836
1837static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1838{
1839        const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1840                                              auto_abbrev);
1841        int len = strlen(uniq);
1842        if (auto_abbrev < len)
1843                return len;
1844        return auto_abbrev;
1845}
1846
1847/*
1848 * How many columns do we need to show line numbers, authors,
1849 * and filenames?
1850 */
1851static void find_alignment(struct scoreboard *sb, int *option)
1852{
1853        int longest_src_lines = 0;
1854        int longest_dst_lines = 0;
1855        unsigned largest_score = 0;
1856        struct blame_entry *e;
1857        int compute_auto_abbrev = (abbrev < 0);
1858        int auto_abbrev = default_abbrev;
1859
1860        for (e = sb->ent; e; e = e->next) {
1861                struct origin *suspect = e->suspect;
1862                struct commit_info ci;
1863                int num;
1864
1865                if (compute_auto_abbrev)
1866                        auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1867                if (strcmp(suspect->path, sb->path))
1868                        *option |= OUTPUT_SHOW_NAME;
1869                num = strlen(suspect->path);
1870                if (longest_file < num)
1871                        longest_file = num;
1872                if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1873                        suspect->commit->object.flags |= METAINFO_SHOWN;
1874                        get_commit_info(suspect->commit, &ci, 1);
1875                        if (*option & OUTPUT_SHOW_EMAIL)
1876                                num = utf8_strwidth(ci.author_mail);
1877                        else
1878                                num = utf8_strwidth(ci.author);
1879                        if (longest_author < num)
1880                                longest_author = num;
1881                }
1882                num = e->s_lno + e->num_lines;
1883                if (longest_src_lines < num)
1884                        longest_src_lines = num;
1885                num = e->lno + e->num_lines;
1886                if (longest_dst_lines < num)
1887                        longest_dst_lines = num;
1888                if (largest_score < ent_score(sb, e))
1889                        largest_score = ent_score(sb, e);
1890        }
1891        max_orig_digits = decimal_width(longest_src_lines);
1892        max_digits = decimal_width(longest_dst_lines);
1893        max_score_digits = decimal_width(largest_score);
1894
1895        if (compute_auto_abbrev)
1896                /* one more abbrev length is needed for the boundary commit */
1897                abbrev = auto_abbrev + 1;
1898}
1899
1900/*
1901 * For debugging -- origin is refcounted, and this asserts that
1902 * we do not underflow.
1903 */
1904static void sanity_check_refcnt(struct scoreboard *sb)
1905{
1906        int baa = 0;
1907        struct blame_entry *ent;
1908
1909        for (ent = sb->ent; ent; ent = ent->next) {
1910                /* Nobody should have zero or negative refcnt */
1911                if (ent->suspect->refcnt <= 0) {
1912                        fprintf(stderr, "%s in %s has negative refcnt %d\n",
1913                                ent->suspect->path,
1914                                sha1_to_hex(ent->suspect->commit->object.sha1),
1915                                ent->suspect->refcnt);
1916                        baa = 1;
1917                }
1918        }
1919        if (baa) {
1920                int opt = 0160;
1921                find_alignment(sb, &opt);
1922                output(sb, opt);
1923                die("Baa %d!", baa);
1924        }
1925}
1926
1927/*
1928 * Used for the command line parsing; check if the path exists
1929 * in the working tree.
1930 */
1931static int has_string_in_work_tree(const char *path)
1932{
1933        struct stat st;
1934        return !lstat(path, &st);
1935}
1936
1937static unsigned parse_score(const char *arg)
1938{
1939        char *end;
1940        unsigned long score = strtoul(arg, &end, 10);
1941        if (*end)
1942                return 0;
1943        return score;
1944}
1945
1946static const char *add_prefix(const char *prefix, const char *path)
1947{
1948        return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1949}
1950
1951/*
1952 * Parsing of (comma separated) one item in the -L option
1953 */
1954static const char *parse_loc(const char *spec,
1955                             struct scoreboard *sb, long lno,
1956                             long begin, long *ret)
1957{
1958        char *term;
1959        const char *line;
1960        long num;
1961        int reg_error;
1962        regex_t regexp;
1963        regmatch_t match[1];
1964
1965        /* Allow "-L <something>,+20" to mean starting at <something>
1966         * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1967         * <something>.
1968         */
1969        if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1970                num = strtol(spec + 1, &term, 10);
1971                if (term != spec + 1) {
1972                        if (spec[0] == '-')
1973                                num = 0 - num;
1974                        if (0 < num)
1975                                *ret = begin + num - 2;
1976                        else if (!num)
1977                                *ret = begin;
1978                        else
1979                                *ret = begin + num;
1980                        return term;
1981                }
1982                return spec;
1983        }
1984        num = strtol(spec, &term, 10);
1985        if (term != spec) {
1986                *ret = num;
1987                return term;
1988        }
1989        if (spec[0] != '/')
1990                return spec;
1991
1992        /* it could be a regexp of form /.../ */
1993        for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1994                if (*term == '\\')
1995                        term++;
1996        }
1997        if (*term != '/')
1998                return spec;
1999
2000        /* try [spec+1 .. term-1] as regexp */
2001        *term = 0;
2002        begin--; /* input is in human terms */
2003        line = nth_line(sb, begin);
2004
2005        if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
2006            !(reg_error = regexec(&regexp, line, 1, match, 0))) {
2007                const char *cp = line + match[0].rm_so;
2008                const char *nline;
2009
2010                while (begin++ < lno) {
2011                        nline = nth_line(sb, begin);
2012                        if (line <= cp && cp < nline)
2013                                break;
2014                        line = nline;
2015                }
2016                *ret = begin;
2017                regfree(&regexp);
2018                *term++ = '/';
2019                return term;
2020        }
2021        else {
2022                char errbuf[1024];
2023                regerror(reg_error, &regexp, errbuf, 1024);
2024                die("-L parameter '%s': %s", spec + 1, errbuf);
2025        }
2026}
2027
2028/*
2029 * Parsing of -L option
2030 */
2031static void prepare_blame_range(struct scoreboard *sb,
2032                                const char *bottomtop,
2033                                long lno,
2034                                long *bottom, long *top)
2035{
2036        const char *term;
2037
2038        term = parse_loc(bottomtop, sb, lno, 1, bottom);
2039        if (*term == ',') {
2040                term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2041                if (*term)
2042                        usage(blame_usage);
2043        }
2044        if (*term)
2045                usage(blame_usage);
2046}
2047
2048static int git_blame_config(const char *var, const char *value, void *cb)
2049{
2050        if (!strcmp(var, "blame.showroot")) {
2051                show_root = git_config_bool(var, value);
2052                return 0;
2053        }
2054        if (!strcmp(var, "blame.blankboundary")) {
2055                blank_boundary = git_config_bool(var, value);
2056                return 0;
2057        }
2058        if (!strcmp(var, "blame.date")) {
2059                if (!value)
2060                        return config_error_nonbool(var);
2061                blame_date_mode = parse_date_format(value);
2062                return 0;
2063        }
2064
2065        if (userdiff_config(var, value) < 0)
2066                return -1;
2067
2068        return git_default_config(var, value, cb);
2069}
2070
2071/*
2072 * Prepare a dummy commit that represents the work tree (or staged) item.
2073 * Note that annotating work tree item never works in the reverse.
2074 */
2075static struct commit *fake_working_tree_commit(struct diff_options *opt,
2076                                               const char *path,
2077                                               const char *contents_from)
2078{
2079        struct commit *commit;
2080        struct origin *origin;
2081        unsigned char head_sha1[20];
2082        struct strbuf buf = STRBUF_INIT;
2083        const char *ident;
2084        time_t now;
2085        int size, len;
2086        struct cache_entry *ce;
2087        unsigned mode;
2088
2089        if (get_sha1("HEAD", head_sha1))
2090                die("No such ref: HEAD");
2091
2092        time(&now);
2093        commit = xcalloc(1, sizeof(*commit));
2094        commit->parents = xcalloc(1, sizeof(*commit->parents));
2095        commit->parents->item = lookup_commit_reference(head_sha1);
2096        commit->object.parsed = 1;
2097        commit->date = now;
2098        commit->object.type = OBJ_COMMIT;
2099
2100        origin = make_origin(commit, path);
2101
2102        if (!contents_from || strcmp("-", contents_from)) {
2103                struct stat st;
2104                const char *read_from;
2105                char *buf_ptr;
2106                unsigned long buf_len;
2107
2108                if (contents_from) {
2109                        if (stat(contents_from, &st) < 0)
2110                                die_errno("Cannot stat '%s'", contents_from);
2111                        read_from = contents_from;
2112                }
2113                else {
2114                        if (lstat(path, &st) < 0)
2115                                die_errno("Cannot lstat '%s'", path);
2116                        read_from = path;
2117                }
2118                mode = canon_mode(st.st_mode);
2119
2120                switch (st.st_mode & S_IFMT) {
2121                case S_IFREG:
2122                        if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2123                            textconv_object(read_from, mode, null_sha1, &buf_ptr, &buf_len))
2124                                strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2125                        else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2126                                die_errno("cannot open or read '%s'", read_from);
2127                        break;
2128                case S_IFLNK:
2129                        if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2130                                die_errno("cannot readlink '%s'", read_from);
2131                        break;
2132                default:
2133                        die("unsupported file type %s", read_from);
2134                }
2135        }
2136        else {
2137                /* Reading from stdin */
2138                contents_from = "standard input";
2139                mode = 0;
2140                if (strbuf_read(&buf, 0, 0) < 0)
2141                        die_errno("failed to read from stdin");
2142        }
2143        convert_to_git(path, buf.buf, buf.len, &buf, 0);
2144        origin->file.ptr = buf.buf;
2145        origin->file.size = buf.len;
2146        pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2147        commit->util = origin;
2148
2149        /*
2150         * Read the current index, replace the path entry with
2151         * origin->blob_sha1 without mucking with its mode or type
2152         * bits; we are not going to write this index out -- we just
2153         * want to run "diff-index --cached".
2154         */
2155        discard_cache();
2156        read_cache();
2157
2158        len = strlen(path);
2159        if (!mode) {
2160                int pos = cache_name_pos(path, len);
2161                if (0 <= pos)
2162                        mode = active_cache[pos]->ce_mode;
2163                else
2164                        /* Let's not bother reading from HEAD tree */
2165                        mode = S_IFREG | 0644;
2166        }
2167        size = cache_entry_size(len);
2168        ce = xcalloc(1, size);
2169        hashcpy(ce->sha1, origin->blob_sha1);
2170        memcpy(ce->name, path, len);
2171        ce->ce_flags = create_ce_flags(0);
2172        ce->ce_namelen = len;
2173        ce->ce_mode = create_ce_mode(mode);
2174        add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2175
2176        /*
2177         * We are not going to write this out, so this does not matter
2178         * right now, but someday we might optimize diff-index --cached
2179         * with cache-tree information.
2180         */
2181        cache_tree_invalidate_path(active_cache_tree, path);
2182
2183        commit->buffer = xmalloc(400);
2184        ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2185        snprintf(commit->buffer, 400,
2186                "tree 0000000000000000000000000000000000000000\n"
2187                "parent %s\n"
2188                "author %s\n"
2189                "committer %s\n\n"
2190                "Version of %s from %s\n",
2191                sha1_to_hex(head_sha1),
2192                ident, ident, path, contents_from ? contents_from : path);
2193        return commit;
2194}
2195
2196static const char *prepare_final(struct scoreboard *sb)
2197{
2198        int i;
2199        const char *final_commit_name = NULL;
2200        struct rev_info *revs = sb->revs;
2201
2202        /*
2203         * There must be one and only one positive commit in the
2204         * revs->pending array.
2205         */
2206        for (i = 0; i < revs->pending.nr; i++) {
2207                struct object *obj = revs->pending.objects[i].item;
2208                if (obj->flags & UNINTERESTING)
2209                        continue;
2210                while (obj->type == OBJ_TAG)
2211                        obj = deref_tag(obj, NULL, 0);
2212                if (obj->type != OBJ_COMMIT)
2213                        die("Non commit %s?", revs->pending.objects[i].name);
2214                if (sb->final)
2215                        die("More than one commit to dig from %s and %s?",
2216                            revs->pending.objects[i].name,
2217                            final_commit_name);
2218                sb->final = (struct commit *) obj;
2219                final_commit_name = revs->pending.objects[i].name;
2220        }
2221        return final_commit_name;
2222}
2223
2224static const char *prepare_initial(struct scoreboard *sb)
2225{
2226        int i;
2227        const char *final_commit_name = NULL;
2228        struct rev_info *revs = sb->revs;
2229
2230        /*
2231         * There must be one and only one negative commit, and it must be
2232         * the boundary.
2233         */
2234        for (i = 0; i < revs->pending.nr; i++) {
2235                struct object *obj = revs->pending.objects[i].item;
2236                if (!(obj->flags & UNINTERESTING))
2237                        continue;
2238                while (obj->type == OBJ_TAG)
2239                        obj = deref_tag(obj, NULL, 0);
2240                if (obj->type != OBJ_COMMIT)
2241                        die("Non commit %s?", revs->pending.objects[i].name);
2242                if (sb->final)
2243                        die("More than one commit to dig down to %s and %s?",
2244                            revs->pending.objects[i].name,
2245                            final_commit_name);
2246                sb->final = (struct commit *) obj;
2247                final_commit_name = revs->pending.objects[i].name;
2248        }
2249        if (!final_commit_name)
2250                die("No commit to dig down to?");
2251        return final_commit_name;
2252}
2253
2254static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2255{
2256        int *opt = option->value;
2257
2258        /*
2259         * -C enables copy from removed files;
2260         * -C -C enables copy from existing files, but only
2261         *       when blaming a new file;
2262         * -C -C -C enables copy from existing files for
2263         *          everybody
2264         */
2265        if (*opt & PICKAXE_BLAME_COPY_HARDER)
2266                *opt |= PICKAXE_BLAME_COPY_HARDEST;
2267        if (*opt & PICKAXE_BLAME_COPY)
2268                *opt |= PICKAXE_BLAME_COPY_HARDER;
2269        *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2270
2271        if (arg)
2272                blame_copy_score = parse_score(arg);
2273        return 0;
2274}
2275
2276static int blame_move_callback(const struct option *option, const char *arg, int unset)
2277{
2278        int *opt = option->value;
2279
2280        *opt |= PICKAXE_BLAME_MOVE;
2281
2282        if (arg)
2283                blame_move_score = parse_score(arg);
2284        return 0;
2285}
2286
2287static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2288{
2289        const char **bottomtop = option->value;
2290        if (!arg)
2291                return -1;
2292        if (*bottomtop)
2293                die("More than one '-L n,m' option given");
2294        *bottomtop = arg;
2295        return 0;
2296}
2297
2298int cmd_blame(int argc, const char **argv, const char *prefix)
2299{
2300        struct rev_info revs;
2301        const char *path;
2302        struct scoreboard sb;
2303        struct origin *o;
2304        struct blame_entry *ent;
2305        long dashdash_pos, bottom, top, lno;
2306        const char *final_commit_name = NULL;
2307        enum object_type type;
2308
2309        static const char *bottomtop = NULL;
2310        static int output_option = 0, opt = 0;
2311        static int show_stats = 0;
2312        static const char *revs_file = NULL;
2313        static const char *contents_from = NULL;
2314        static const struct option options[] = {
2315                OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2316                OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2317                OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2318                OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2319                OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2320                OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2321                OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2322                OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2323                OPT_BIT(0, "line-porcelain", &output_option, "Show porcelain format with per-line commit information", OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2324                OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2325                OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2326                OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2327                OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2328                OPT_BIT('e', "show-email", &output_option, "Show author email instead of name (Default: off)", OUTPUT_SHOW_EMAIL),
2329                OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2330                OPT_BIT(0, "minimal", &xdl_opts, "Spend extra cycles to find better match", XDF_NEED_MINIMAL),
2331                OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2332                OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2333                { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2334                { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2335                OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2336                OPT__ABBREV(&abbrev),
2337                OPT_END()
2338        };
2339
2340        struct parse_opt_ctx_t ctx;
2341        int cmd_is_annotate = !strcmp(argv[0], "annotate");
2342
2343        git_config(git_blame_config, NULL);
2344        init_revisions(&revs, NULL);
2345        revs.date_mode = blame_date_mode;
2346        DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2347
2348        save_commit_buffer = 0;
2349        dashdash_pos = 0;
2350
2351        parse_options_start(&ctx, argc, argv, prefix, options,
2352                            PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2353        for (;;) {
2354                switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2355                case PARSE_OPT_HELP:
2356                        exit(129);
2357                case PARSE_OPT_DONE:
2358                        if (ctx.argv[0])
2359                                dashdash_pos = ctx.cpidx;
2360                        goto parse_done;
2361                }
2362
2363                if (!strcmp(ctx.argv[0], "--reverse")) {
2364                        ctx.argv[0] = "--children";
2365                        reverse = 1;
2366                }
2367                parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2368        }
2369parse_done:
2370        argc = parse_options_end(&ctx);
2371
2372        if (0 < abbrev)
2373                /* one more abbrev length is needed for the boundary commit */
2374                abbrev++;
2375
2376        if (revs_file && read_ancestry(revs_file))
2377                die_errno("reading graft file '%s' failed", revs_file);
2378
2379        if (cmd_is_annotate) {
2380                output_option |= OUTPUT_ANNOTATE_COMPAT;
2381                blame_date_mode = DATE_ISO8601;
2382        } else {
2383                blame_date_mode = revs.date_mode;
2384        }
2385
2386        /* The maximum width used to show the dates */
2387        switch (blame_date_mode) {
2388        case DATE_RFC2822:
2389                blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2390                break;
2391        case DATE_ISO8601:
2392                blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2393                break;
2394        case DATE_RAW:
2395                blame_date_width = sizeof("1161298804 -0700");
2396                break;
2397        case DATE_SHORT:
2398                blame_date_width = sizeof("2006-10-19");
2399                break;
2400        case DATE_RELATIVE:
2401                /* "normal" is used as the fallback for "relative" */
2402        case DATE_LOCAL:
2403        case DATE_NORMAL:
2404                blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2405                break;
2406        }
2407        blame_date_width -= 1; /* strip the null */
2408
2409        if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2410                opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2411                        PICKAXE_BLAME_COPY_HARDER);
2412
2413        if (!blame_move_score)
2414                blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2415        if (!blame_copy_score)
2416                blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2417
2418        /*
2419         * We have collected options unknown to us in argv[1..unk]
2420         * which are to be passed to revision machinery if we are
2421         * going to do the "bottom" processing.
2422         *
2423         * The remaining are:
2424         *
2425         * (1) if dashdash_pos != 0, it is either
2426         *     "blame [revisions] -- <path>" or
2427         *     "blame -- <path> <rev>"
2428         *
2429         * (2) otherwise, it is one of the two:
2430         *     "blame [revisions] <path>"
2431         *     "blame <path> <rev>"
2432         *
2433         * Note that we must strip out <path> from the arguments: we do not
2434         * want the path pruning but we may want "bottom" processing.
2435         */
2436        if (dashdash_pos) {
2437                switch (argc - dashdash_pos - 1) {
2438                case 2: /* (1b) */
2439                        if (argc != 4)
2440                                usage_with_options(blame_opt_usage, options);
2441                        /* reorder for the new way: <rev> -- <path> */
2442                        argv[1] = argv[3];
2443                        argv[3] = argv[2];
2444                        argv[2] = "--";
2445                        /* FALLTHROUGH */
2446                case 1: /* (1a) */
2447                        path = add_prefix(prefix, argv[--argc]);
2448                        argv[argc] = NULL;
2449                        break;
2450                default:
2451                        usage_with_options(blame_opt_usage, options);
2452                }
2453        } else {
2454                if (argc < 2)
2455                        usage_with_options(blame_opt_usage, options);
2456                path = add_prefix(prefix, argv[argc - 1]);
2457                if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2458                        path = add_prefix(prefix, argv[1]);
2459                        argv[1] = argv[2];
2460                }
2461                argv[argc - 1] = "--";
2462
2463                setup_work_tree();
2464                if (!has_string_in_work_tree(path))
2465                        die_errno("cannot stat path '%s'", path);
2466        }
2467
2468        revs.disable_stdin = 1;
2469        setup_revisions(argc, argv, &revs, NULL);
2470        memset(&sb, 0, sizeof(sb));
2471
2472        sb.revs = &revs;
2473        if (!reverse)
2474                final_commit_name = prepare_final(&sb);
2475        else if (contents_from)
2476                die("--contents and --children do not blend well.");
2477        else
2478                final_commit_name = prepare_initial(&sb);
2479
2480        if (!sb.final) {
2481                /*
2482                 * "--not A B -- path" without anything positive;
2483                 * do not default to HEAD, but use the working tree
2484                 * or "--contents".
2485                 */
2486                setup_work_tree();
2487                sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2488                                                    path, contents_from);
2489                add_pending_object(&revs, &(sb.final->object), ":");
2490        }
2491        else if (contents_from)
2492                die("Cannot use --contents with final commit object name");
2493
2494        /*
2495         * If we have bottom, this will mark the ancestors of the
2496         * bottom commits we would reach while traversing as
2497         * uninteresting.
2498         */
2499        if (prepare_revision_walk(&revs))
2500                die("revision walk setup failed");
2501
2502        if (is_null_sha1(sb.final->object.sha1)) {
2503                char *buf;
2504                o = sb.final->util;
2505                buf = xmalloc(o->file.size + 1);
2506                memcpy(buf, o->file.ptr, o->file.size + 1);
2507                sb.final_buf = buf;
2508                sb.final_buf_size = o->file.size;
2509        }
2510        else {
2511                o = get_origin(&sb, sb.final, path);
2512                if (fill_blob_sha1_and_mode(o))
2513                        die("no such path %s in %s", path, final_commit_name);
2514
2515                if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2516                    textconv_object(path, o->mode, o->blob_sha1, (char **) &sb.final_buf,
2517                                    &sb.final_buf_size))
2518                        ;
2519                else
2520                        sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2521                                                      &sb.final_buf_size);
2522
2523                if (!sb.final_buf)
2524                        die("Cannot read blob %s for path %s",
2525                            sha1_to_hex(o->blob_sha1),
2526                            path);
2527        }
2528        num_read_blob++;
2529        lno = prepare_lines(&sb);
2530
2531        bottom = top = 0;
2532        if (bottomtop)
2533                prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2534        if (bottom && top && top < bottom) {
2535                long tmp;
2536                tmp = top; top = bottom; bottom = tmp;
2537        }
2538        if (bottom < 1)
2539                bottom = 1;
2540        if (top < 1)
2541                top = lno;
2542        bottom--;
2543        if (lno < top || lno < bottom)
2544                die("file %s has only %lu lines", path, lno);
2545
2546        ent = xcalloc(1, sizeof(*ent));
2547        ent->lno = bottom;
2548        ent->num_lines = top - bottom;
2549        ent->suspect = o;
2550        ent->s_lno = bottom;
2551
2552        sb.ent = ent;
2553        sb.path = path;
2554
2555        read_mailmap(&mailmap, NULL);
2556
2557        if (!incremental)
2558                setup_pager();
2559
2560        assign_blame(&sb, opt);
2561
2562        if (incremental)
2563                return 0;
2564
2565        coalesce(&sb);
2566
2567        if (!(output_option & OUTPUT_PORCELAIN))
2568                find_alignment(&sb, &output_option);
2569
2570        output(&sb, output_option);
2571        free((void *)sb.final_buf);
2572        for (ent = sb.ent; ent; ) {
2573                struct blame_entry *e = ent->next;
2574                free(ent);
2575                ent = e;
2576        }
2577
2578        if (show_stats) {
2579                printf("num read blob: %d\n", num_read_blob);
2580                printf("num get patch: %d\n", num_get_patch);
2581                printf("num commits: %d\n", num_commits);
2582        }
2583        return 0;
2584}