32dc3932edfa6c007d75b5369917e1fd744ca317
   1/*
   2 * Pickaxe
   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 "xdiff-interface.h"
  17
  18#include <time.h>
  19#include <sys/time.h>
  20
  21static char pickaxe_usage[] =
  22"git-pickaxe [-c] [-l] [-t] [-f] [-n] [-p] [-L n,m] [-S <revs-file>] [-M] [-C] [-C] [commit] [--] file\n"
  23"  -c, --compatibility Use the same output mode as git-annotate (Default: off)\n"
  24"  -l, --long          Show long commit SHA1 (Default: off)\n"
  25"  -t, --time          Show raw timestamp (Default: off)\n"
  26"  -f, --show-name     Show original filename (Default: auto)\n"
  27"  -n, --show-number   Show original linenumber (Default: off)\n"
  28"  -p, --porcelain     Show in a format designed for machine consumption\n"
  29"  -L n,m              Process only line range n,m, counting from 1\n"
  30"  -M, -C              Find line movements within and across files\n"
  31"  -S revs-file        Use revisions from revs-file instead of calling git-rev-list\n";
  32
  33static int longest_file;
  34static int longest_author;
  35static int max_orig_digits;
  36static int max_digits;
  37static int max_score_digits;
  38
  39#ifndef DEBUG
  40#define DEBUG 0
  41#endif
  42
  43#define PICKAXE_BLAME_MOVE              01
  44#define PICKAXE_BLAME_COPY              02
  45#define PICKAXE_BLAME_COPY_HARDER       04
  46
  47/*
  48 * blame for a blame_entry with score lower than these thresholds
  49 * is not passed to the parent using move/copy logic.
  50 */
  51static unsigned blame_move_score;
  52static unsigned blame_copy_score;
  53#define BLAME_DEFAULT_MOVE_SCORE        20
  54#define BLAME_DEFAULT_COPY_SCORE        40
  55
  56/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
  57#define METAINFO_SHOWN          (1u<<12)
  58#define MORE_THAN_ONE_PATH      (1u<<13)
  59
  60/*
  61 * One blob in a commit that is being suspected
  62 */
  63struct origin {
  64        int refcnt;
  65        struct commit *commit;
  66        unsigned char blob_sha1[20];
  67        char path[FLEX_ARRAY];
  68};
  69
  70static inline struct origin *origin_incref(struct origin *o)
  71{
  72        if (o)
  73                o->refcnt++;
  74        return o;
  75}
  76
  77static void origin_decref(struct origin *o)
  78{
  79        if (o && --o->refcnt <= 0) {
  80                memset(o, 0, sizeof(*o));
  81                free(o);
  82        }
  83}
  84
  85struct blame_entry {
  86        struct blame_entry *prev;
  87        struct blame_entry *next;
  88
  89        /* the first line of this group in the final image;
  90         * internally all line numbers are 0 based.
  91         */
  92        int lno;
  93
  94        /* how many lines this group has */
  95        int num_lines;
  96
  97        /* the commit that introduced this group into the final image */
  98        struct origin *suspect;
  99
 100        /* true if the suspect is truly guilty; false while we have not
 101         * checked if the group came from one of its parents.
 102         */
 103        char guilty;
 104
 105        /* the line number of the first line of this group in the
 106         * suspect's file; internally all line numbers are 0 based.
 107         */
 108        int s_lno;
 109
 110        /* how significant this entry is -- cached to avoid
 111         * scanning the lines over and over
 112         */
 113        unsigned score;
 114};
 115
 116struct scoreboard {
 117        /* the final commit (i.e. where we started digging from) */
 118        struct commit *final;
 119
 120        const char *path;
 121
 122        /* the contents in the final; pointed into by buf pointers of
 123         * blame_entries
 124         */
 125        const char *final_buf;
 126        unsigned long final_buf_size;
 127
 128        /* linked list of blames */
 129        struct blame_entry *ent;
 130
 131        /* look-up a line in the final buffer */
 132        int num_lines;
 133        int *lineno;
 134};
 135
 136static int cmp_suspect(struct origin *a, struct origin *b)
 137{
 138        int cmp = hashcmp(a->commit->object.sha1, b->commit->object.sha1);
 139        if (cmp)
 140                return cmp;
 141        return strcmp(a->path, b->path);
 142}
 143
 144static void sanity_check_refcnt(struct scoreboard *);
 145
 146static void coalesce(struct scoreboard *sb)
 147{
 148        struct blame_entry *ent, *next;
 149
 150        for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 151                if (!cmp_suspect(ent->suspect, next->suspect) &&
 152                    ent->guilty == next->guilty &&
 153                    ent->s_lno + ent->num_lines == next->s_lno) {
 154                        ent->num_lines += next->num_lines;
 155                        ent->next = next->next;
 156                        if (ent->next)
 157                                ent->next->prev = ent;
 158                        origin_decref(next->suspect);
 159                        free(next);
 160                        ent->score = 0;
 161                        next = ent; /* again */
 162                }
 163        }
 164
 165        if (DEBUG) /* sanity */
 166                sanity_check_refcnt(sb);
 167}
 168
 169static struct origin *get_origin(struct scoreboard *sb,
 170                                 struct commit *commit,
 171                                 const char *path)
 172{
 173        struct blame_entry *e;
 174        struct origin *o;
 175
 176        for (e = sb->ent; e; e = e->next) {
 177                if (e->suspect->commit == commit &&
 178                    !strcmp(e->suspect->path, path))
 179                        return origin_incref(e->suspect);
 180        }
 181        o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 182        o->commit = commit;
 183        o->refcnt = 1;
 184        strcpy(o->path, path);
 185        return o;
 186}
 187
 188static int fill_blob_sha1(struct origin *origin)
 189{
 190        unsigned mode;
 191        char type[10];
 192
 193        if (!is_null_sha1(origin->blob_sha1))
 194                return 0;
 195        if (get_tree_entry(origin->commit->object.sha1,
 196                           origin->path,
 197                           origin->blob_sha1, &mode))
 198                goto error_out;
 199        if (sha1_object_info(origin->blob_sha1, type, NULL) ||
 200            strcmp(type, blob_type))
 201                goto error_out;
 202        return 0;
 203 error_out:
 204        hashclr(origin->blob_sha1);
 205        return -1;
 206}
 207
 208static struct origin *find_origin(struct scoreboard *sb,
 209                                  struct commit *parent,
 210                                  struct origin *origin)
 211{
 212        struct origin *porigin = NULL;
 213        struct diff_options diff_opts;
 214        const char *paths[2];
 215
 216        /* See if the origin->path is different between parent
 217         * and origin first.  Most of the time they are the
 218         * same and diff-tree is fairly efficient about this.
 219         */
 220        diff_setup(&diff_opts);
 221        diff_opts.recursive = 1;
 222        diff_opts.detect_rename = 0;
 223        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 224        paths[0] = origin->path;
 225        paths[1] = NULL;
 226
 227        diff_tree_setup_paths(paths, &diff_opts);
 228        if (diff_setup_done(&diff_opts) < 0)
 229                die("diff-setup");
 230        diff_tree_sha1(parent->tree->object.sha1,
 231                       origin->commit->tree->object.sha1,
 232                       "", &diff_opts);
 233        diffcore_std(&diff_opts);
 234
 235        /* It is either one entry that says "modified", or "created",
 236         * or nothing.
 237         */
 238        if (!diff_queued_diff.nr) {
 239                /* The path is the same as parent */
 240                porigin = get_origin(sb, parent, origin->path);
 241                hashcpy(porigin->blob_sha1, origin->blob_sha1);
 242        }
 243        else if (diff_queued_diff.nr != 1)
 244                die("internal error in pickaxe::find_origin");
 245        else {
 246                struct diff_filepair *p = diff_queued_diff.queue[0];
 247                switch (p->status) {
 248                default:
 249                        die("internal error in pickaxe::find_origin (%c)",
 250                            p->status);
 251                case 'M':
 252                        porigin = get_origin(sb, parent, origin->path);
 253                        hashcpy(porigin->blob_sha1, p->one->sha1);
 254                        break;
 255                case 'A':
 256                case 'T':
 257                        /* Did not exist in parent, or type changed */
 258                        break;
 259                }
 260        }
 261        diff_flush(&diff_opts);
 262        return porigin;
 263}
 264
 265static struct origin *find_rename(struct scoreboard *sb,
 266                                  struct commit *parent,
 267                                  struct origin *origin)
 268{
 269        struct origin *porigin = NULL;
 270        struct diff_options diff_opts;
 271        int i;
 272        const char *paths[2];
 273
 274        diff_setup(&diff_opts);
 275        diff_opts.recursive = 1;
 276        diff_opts.detect_rename = DIFF_DETECT_RENAME;
 277        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 278        paths[0] = NULL;
 279        diff_tree_setup_paths(paths, &diff_opts);
 280        if (diff_setup_done(&diff_opts) < 0)
 281                die("diff-setup");
 282        diff_tree_sha1(parent->tree->object.sha1,
 283                       origin->commit->tree->object.sha1,
 284                       "", &diff_opts);
 285        diffcore_std(&diff_opts);
 286
 287        for (i = 0; i < diff_queued_diff.nr; i++) {
 288                struct diff_filepair *p = diff_queued_diff.queue[i];
 289                if ((p->status == 'R' || p->status == 'C') &&
 290                    !strcmp(p->two->path, origin->path)) {
 291                        porigin = get_origin(sb, parent, p->one->path);
 292                        hashcpy(porigin->blob_sha1, p->one->sha1);
 293                        break;
 294                }
 295        }
 296        diff_flush(&diff_opts);
 297        return porigin;
 298}
 299
 300struct chunk {
 301        /* line number in postimage; up to but not including this
 302         * line is the same as preimage
 303         */
 304        int same;
 305
 306        /* preimage line number after this chunk */
 307        int p_next;
 308
 309        /* postimage line number after this chunk */
 310        int t_next;
 311};
 312
 313struct patch {
 314        struct chunk *chunks;
 315        int num;
 316};
 317
 318struct blame_diff_state {
 319        struct xdiff_emit_state xm;
 320        struct patch *ret;
 321        unsigned hunk_post_context;
 322        unsigned hunk_in_pre_context : 1;
 323};
 324
 325static void process_u_diff(void *state_, char *line, unsigned long len)
 326{
 327        struct blame_diff_state *state = state_;
 328        struct chunk *chunk;
 329        int off1, off2, len1, len2, num;
 330
 331        num = state->ret->num;
 332        if (len < 4 || line[0] != '@' || line[1] != '@') {
 333                if (state->hunk_in_pre_context && line[0] == ' ')
 334                        state->ret->chunks[num - 1].same++;
 335                else {
 336                        state->hunk_in_pre_context = 0;
 337                        if (line[0] == ' ')
 338                                state->hunk_post_context++;
 339                        else
 340                                state->hunk_post_context = 0;
 341                }
 342                return;
 343        }
 344
 345        if (num && state->hunk_post_context) {
 346                chunk = &state->ret->chunks[num - 1];
 347                chunk->p_next -= state->hunk_post_context;
 348                chunk->t_next -= state->hunk_post_context;
 349        }
 350        state->ret->num = ++num;
 351        state->ret->chunks = xrealloc(state->ret->chunks,
 352                                      sizeof(struct chunk) * num);
 353        chunk = &state->ret->chunks[num - 1];
 354        if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
 355                state->ret->num--;
 356                return;
 357        }
 358
 359        /* Line numbers in patch output are one based. */
 360        off1--;
 361        off2--;
 362
 363        chunk->same = len2 ? off2 : (off2 + 1);
 364
 365        chunk->p_next = off1 + (len1 ? len1 : 1);
 366        chunk->t_next = chunk->same + len2;
 367        state->hunk_in_pre_context = 1;
 368        state->hunk_post_context = 0;
 369}
 370
 371static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
 372                                    int context)
 373{
 374        struct blame_diff_state state;
 375        xpparam_t xpp;
 376        xdemitconf_t xecfg;
 377        xdemitcb_t ecb;
 378
 379        xpp.flags = XDF_NEED_MINIMAL;
 380        xecfg.ctxlen = context;
 381        xecfg.flags = 0;
 382        ecb.outf = xdiff_outf;
 383        ecb.priv = &state;
 384        memset(&state, 0, sizeof(state));
 385        state.xm.consume = process_u_diff;
 386        state.ret = xmalloc(sizeof(struct patch));
 387        state.ret->chunks = NULL;
 388        state.ret->num = 0;
 389
 390        xdl_diff(file_p, file_o, &xpp, &xecfg, &ecb);
 391
 392        if (state.ret->num) {
 393                struct chunk *chunk;
 394                chunk = &state.ret->chunks[state.ret->num - 1];
 395                chunk->p_next -= state.hunk_post_context;
 396                chunk->t_next -= state.hunk_post_context;
 397        }
 398        return state.ret;
 399}
 400
 401static struct patch *get_patch(struct origin *parent, struct origin *origin)
 402{
 403        mmfile_t file_p, file_o;
 404        char type[10];
 405        char *blob_p, *blob_o;
 406        struct patch *patch;
 407
 408        blob_p = read_sha1_file(parent->blob_sha1, type,
 409                                (unsigned long *) &file_p.size);
 410        blob_o = read_sha1_file(origin->blob_sha1, type,
 411                                (unsigned long *) &file_o.size);
 412        file_p.ptr = blob_p;
 413        file_o.ptr = blob_o;
 414        if (!file_p.ptr || !file_o.ptr) {
 415                free(blob_p);
 416                free(blob_o);
 417                return NULL;
 418        }
 419
 420        patch = compare_buffer(&file_p, &file_o, 0);
 421        free(blob_p);
 422        free(blob_o);
 423        return patch;
 424}
 425
 426static void free_patch(struct patch *p)
 427{
 428        free(p->chunks);
 429        free(p);
 430}
 431
 432static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 433{
 434        struct blame_entry *ent, *prev = NULL;
 435
 436        origin_incref(e->suspect);
 437
 438        for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 439                prev = ent;
 440
 441        /* prev, if not NULL, is the last one that is below e */
 442        e->prev = prev;
 443        if (prev) {
 444                e->next = prev->next;
 445                prev->next = e;
 446        }
 447        else {
 448                e->next = sb->ent;
 449                sb->ent = e;
 450        }
 451        if (e->next)
 452                e->next->prev = e;
 453}
 454
 455static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 456{
 457        struct blame_entry *p, *n;
 458
 459        p = dst->prev;
 460        n = dst->next;
 461        origin_incref(src->suspect);
 462        origin_decref(dst->suspect);
 463        memcpy(dst, src, sizeof(*src));
 464        dst->prev = p;
 465        dst->next = n;
 466        dst->score = 0;
 467}
 468
 469static const char *nth_line(struct scoreboard *sb, int lno)
 470{
 471        return sb->final_buf + sb->lineno[lno];
 472}
 473
 474static void split_overlap(struct blame_entry *split,
 475                          struct blame_entry *e,
 476                          int tlno, int plno, int same,
 477                          struct origin *parent)
 478{
 479        /* it is known that lines between tlno to same came from
 480         * parent, and e has an overlap with that range.  it also is
 481         * known that parent's line plno corresponds to e's line tlno.
 482         *
 483         *                <---- e ----->
 484         *                   <------>
 485         *                   <------------>
 486         *             <------------>
 487         *             <------------------>
 488         *
 489         * Potentially we need to split e into three parts; before
 490         * this chunk, the chunk to be blamed for parent, and after
 491         * that portion.
 492         */
 493        int chunk_end_lno;
 494        memset(split, 0, sizeof(struct blame_entry [3]));
 495
 496        if (e->s_lno < tlno) {
 497                /* there is a pre-chunk part not blamed on parent */
 498                split[0].suspect = origin_incref(e->suspect);
 499                split[0].lno = e->lno;
 500                split[0].s_lno = e->s_lno;
 501                split[0].num_lines = tlno - e->s_lno;
 502                split[1].lno = e->lno + tlno - e->s_lno;
 503                split[1].s_lno = plno;
 504        }
 505        else {
 506                split[1].lno = e->lno;
 507                split[1].s_lno = plno + (e->s_lno - tlno);
 508        }
 509
 510        if (same < e->s_lno + e->num_lines) {
 511                /* there is a post-chunk part not blamed on parent */
 512                split[2].suspect = origin_incref(e->suspect);
 513                split[2].lno = e->lno + (same - e->s_lno);
 514                split[2].s_lno = e->s_lno + (same - e->s_lno);
 515                split[2].num_lines = e->s_lno + e->num_lines - same;
 516                chunk_end_lno = split[2].lno;
 517        }
 518        else
 519                chunk_end_lno = e->lno + e->num_lines;
 520        split[1].num_lines = chunk_end_lno - split[1].lno;
 521
 522        if (split[1].num_lines < 1)
 523                return;
 524        split[1].suspect = origin_incref(parent);
 525}
 526
 527static void split_blame(struct scoreboard *sb,
 528                        struct blame_entry *split,
 529                        struct blame_entry *e)
 530{
 531        struct blame_entry *new_entry;
 532
 533        if (split[0].suspect && split[2].suspect) {
 534                /* we need to split e into two and add another for parent */
 535                dup_entry(e, &split[0]);
 536
 537                new_entry = xmalloc(sizeof(*new_entry));
 538                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 539                add_blame_entry(sb, new_entry);
 540
 541                new_entry = xmalloc(sizeof(*new_entry));
 542                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 543                add_blame_entry(sb, new_entry);
 544        }
 545        else if (!split[0].suspect && !split[2].suspect)
 546                /* parent covers the entire area */
 547                dup_entry(e, &split[1]);
 548        else if (split[0].suspect) {
 549                dup_entry(e, &split[0]);
 550
 551                new_entry = xmalloc(sizeof(*new_entry));
 552                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 553                add_blame_entry(sb, new_entry);
 554        }
 555        else {
 556                dup_entry(e, &split[1]);
 557
 558                new_entry = xmalloc(sizeof(*new_entry));
 559                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 560                add_blame_entry(sb, new_entry);
 561        }
 562
 563        if (DEBUG) { /* sanity */
 564                struct blame_entry *ent;
 565                int lno = sb->ent->lno, corrupt = 0;
 566
 567                for (ent = sb->ent; ent; ent = ent->next) {
 568                        if (lno != ent->lno)
 569                                corrupt = 1;
 570                        if (ent->s_lno < 0)
 571                                corrupt = 1;
 572                        lno += ent->num_lines;
 573                }
 574                if (corrupt) {
 575                        lno = sb->ent->lno;
 576                        for (ent = sb->ent; ent; ent = ent->next) {
 577                                printf("L %8d l %8d n %8d\n",
 578                                       lno, ent->lno, ent->num_lines);
 579                                lno = ent->lno + ent->num_lines;
 580                        }
 581                        die("oops");
 582                }
 583        }
 584}
 585
 586static void decref_split(struct blame_entry *split)
 587{
 588        int i;
 589
 590        for (i = 0; i < 3; i++)
 591                origin_decref(split[i].suspect);
 592}
 593
 594static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 595                          int tlno, int plno, int same,
 596                          struct origin *parent)
 597{
 598        struct blame_entry split[3];
 599
 600        split_overlap(split, e, tlno, plno, same, parent);
 601        if (split[1].suspect)
 602                split_blame(sb, split, e);
 603        decref_split(split);
 604}
 605
 606static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 607{
 608        struct blame_entry *e;
 609        int last_in_target = -1;
 610
 611        for (e = sb->ent; e; e = e->next) {
 612                if (e->guilty || cmp_suspect(e->suspect, target))
 613                        continue;
 614                if (last_in_target < e->s_lno + e->num_lines)
 615                        last_in_target = e->s_lno + e->num_lines;
 616        }
 617        return last_in_target;
 618}
 619
 620static void blame_chunk(struct scoreboard *sb,
 621                        int tlno, int plno, int same,
 622                        struct origin *target, struct origin *parent)
 623{
 624        struct blame_entry *e;
 625
 626        for (e = sb->ent; e; e = e->next) {
 627                if (e->guilty || cmp_suspect(e->suspect, target))
 628                        continue;
 629                if (same <= e->s_lno)
 630                        continue;
 631                if (tlno < e->s_lno + e->num_lines)
 632                        blame_overlap(sb, e, tlno, plno, same, parent);
 633        }
 634}
 635
 636static int pass_blame_to_parent(struct scoreboard *sb,
 637                                struct origin *target,
 638                                struct origin *parent)
 639{
 640        int i, last_in_target, plno, tlno;
 641        struct patch *patch;
 642
 643        last_in_target = find_last_in_target(sb, target);
 644        if (last_in_target < 0)
 645                return 1; /* nothing remains for this target */
 646
 647        patch = get_patch(parent, target);
 648        plno = tlno = 0;
 649        for (i = 0; i < patch->num; i++) {
 650                struct chunk *chunk = &patch->chunks[i];
 651
 652                blame_chunk(sb, tlno, plno, chunk->same, target, parent);
 653                plno = chunk->p_next;
 654                tlno = chunk->t_next;
 655        }
 656        /* rest (i.e. anything above tlno) are the same as parent */
 657        blame_chunk(sb, tlno, plno, last_in_target, target, parent);
 658
 659        free_patch(patch);
 660        return 0;
 661}
 662
 663static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 664{
 665        unsigned score;
 666        const char *cp, *ep;
 667
 668        if (e->score)
 669                return e->score;
 670
 671        score = 1;
 672        cp = nth_line(sb, e->lno);
 673        ep = nth_line(sb, e->lno + e->num_lines);
 674        while (cp < ep) {
 675                unsigned ch = *((unsigned char *)cp);
 676                if (isalnum(ch))
 677                        score++;
 678                cp++;
 679        }
 680        e->score = score;
 681        return score;
 682}
 683
 684static void copy_split_if_better(struct scoreboard *sb,
 685                                 struct blame_entry *best_so_far,
 686                                 struct blame_entry *this)
 687{
 688        int i;
 689
 690        if (!this[1].suspect)
 691                return;
 692        if (best_so_far[1].suspect) {
 693                if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 694                        return;
 695        }
 696
 697        for (i = 0; i < 3; i++)
 698                origin_incref(this[i].suspect);
 699        decref_split(best_so_far);
 700        memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 701}
 702
 703static void find_copy_in_blob(struct scoreboard *sb,
 704                              struct blame_entry *ent,
 705                              struct origin *parent,
 706                              struct blame_entry *split,
 707                              mmfile_t *file_p)
 708{
 709        const char *cp;
 710        int cnt;
 711        mmfile_t file_o;
 712        struct patch *patch;
 713        int i, plno, tlno;
 714
 715        cp = nth_line(sb, ent->lno);
 716        file_o.ptr = (char*) cp;
 717        cnt = ent->num_lines;
 718
 719        while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 720                if (*cp++ == '\n')
 721                        cnt--;
 722        }
 723        file_o.size = cp - file_o.ptr;
 724
 725        patch = compare_buffer(file_p, &file_o, 1);
 726
 727        memset(split, 0, sizeof(struct blame_entry [3]));
 728        plno = tlno = 0;
 729        for (i = 0; i < patch->num; i++) {
 730                struct chunk *chunk = &patch->chunks[i];
 731
 732                /* tlno to chunk->same are the same as ent */
 733                if (ent->num_lines <= tlno)
 734                        break;
 735                if (tlno < chunk->same) {
 736                        struct blame_entry this[3];
 737                        split_overlap(this, ent,
 738                                      tlno + ent->s_lno, plno,
 739                                      chunk->same + ent->s_lno,
 740                                      parent);
 741                        copy_split_if_better(sb, split, this);
 742                        decref_split(this);
 743                }
 744                plno = chunk->p_next;
 745                tlno = chunk->t_next;
 746        }
 747        free_patch(patch);
 748}
 749
 750static int find_move_in_parent(struct scoreboard *sb,
 751                               struct origin *target,
 752                               struct origin *parent)
 753{
 754        int last_in_target;
 755        struct blame_entry *e, split[3];
 756        mmfile_t file_p;
 757        char type[10];
 758        char *blob_p;
 759
 760        last_in_target = find_last_in_target(sb, target);
 761        if (last_in_target < 0)
 762                return 1; /* nothing remains for this target */
 763
 764        blob_p = read_sha1_file(parent->blob_sha1, type,
 765                                (unsigned long *) &file_p.size);
 766        file_p.ptr = blob_p;
 767        if (!file_p.ptr) {
 768                free(blob_p);
 769                return 0;
 770        }
 771
 772        for (e = sb->ent; e; e = e->next) {
 773                if (e->guilty || cmp_suspect(e->suspect, target))
 774                        continue;
 775                find_copy_in_blob(sb, e, parent, split, &file_p);
 776                if (split[1].suspect &&
 777                    blame_move_score < ent_score(sb, &split[1]))
 778                        split_blame(sb, split, e);
 779                decref_split(split);
 780        }
 781        free(blob_p);
 782        return 0;
 783}
 784
 785static int find_copy_in_parent(struct scoreboard *sb,
 786                               struct origin *target,
 787                               struct commit *parent,
 788                               struct origin *porigin,
 789                               int opt)
 790{
 791        struct diff_options diff_opts;
 792        const char *paths[1];
 793        struct blame_entry *e;
 794        int i, j;
 795        struct blame_list {
 796                struct blame_entry *ent;
 797                struct blame_entry split[3];
 798        } *blame_list;
 799        int num_ents;
 800
 801        /* Count the number of entries the target is suspected for,
 802         * and prepare a list of entry and the best split.
 803         */
 804        for (e = sb->ent, num_ents = 0; e; e = e->next)
 805                if (!e->guilty && !cmp_suspect(e->suspect, target))
 806                        num_ents++;
 807        if (!num_ents)
 808                return 1; /* nothing remains for this target */
 809
 810        blame_list = xcalloc(num_ents, sizeof(struct blame_list));
 811        for (e = sb->ent, i = 0; e; e = e->next)
 812                if (!e->guilty && !cmp_suspect(e->suspect, target))
 813                        blame_list[i++].ent = e;
 814
 815        diff_setup(&diff_opts);
 816        diff_opts.recursive = 1;
 817        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 818
 819        /* Try "find copies harder" on new path */
 820        if ((opt & PICKAXE_BLAME_COPY_HARDER) &&
 821            (!porigin || strcmp(target->path, porigin->path))) {
 822                diff_opts.detect_rename = DIFF_DETECT_COPY;
 823                diff_opts.find_copies_harder = 1;
 824        }
 825        paths[0] = NULL;
 826        diff_tree_setup_paths(paths, &diff_opts);
 827        if (diff_setup_done(&diff_opts) < 0)
 828                die("diff-setup");
 829        diff_tree_sha1(parent->tree->object.sha1,
 830                       target->commit->tree->object.sha1,
 831                       "", &diff_opts);
 832        diffcore_std(&diff_opts);
 833
 834        for (i = 0; i < diff_queued_diff.nr; i++) {
 835                struct diff_filepair *p = diff_queued_diff.queue[i];
 836                struct origin *norigin;
 837                mmfile_t file_p;
 838                char type[10];
 839                char *blob;
 840                struct blame_entry this[3];
 841
 842                if (!DIFF_FILE_VALID(p->one))
 843                        continue; /* does not exist in parent */
 844                if (porigin && !strcmp(p->one->path, porigin->path))
 845                        /* find_move already dealt with this path */
 846                        continue;
 847
 848                norigin = get_origin(sb, parent, p->one->path);
 849                hashcpy(norigin->blob_sha1, p->one->sha1);
 850                blob = read_sha1_file(norigin->blob_sha1, type,
 851                                      (unsigned long *) &file_p.size);
 852                file_p.ptr = blob;
 853                if (!file_p.ptr)
 854                        continue;
 855
 856                for (j = 0; j < num_ents; j++) {
 857                        find_copy_in_blob(sb, blame_list[j].ent, norigin,
 858                                          this, &file_p);
 859                        copy_split_if_better(sb, blame_list[j].split,
 860                                             this);
 861                        decref_split(this);
 862                }
 863                free(blob);
 864                origin_decref(norigin);
 865        }
 866        diff_flush(&diff_opts);
 867
 868        for (j = 0; j < num_ents; j++) {
 869                struct blame_entry *split = blame_list[j].split;
 870                if (split[1].suspect &&
 871                    blame_copy_score < ent_score(sb, &split[1]))
 872                        split_blame(sb, split, blame_list[j].ent);
 873                decref_split(split);
 874        }
 875        free(blame_list);
 876
 877        return 0;
 878}
 879
 880#define MAXPARENT 16
 881
 882static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
 883{
 884        int i, pass;
 885        struct commit *commit = origin->commit;
 886        struct commit_list *parent;
 887        struct origin *parent_origin[MAXPARENT], *porigin;
 888
 889        memset(parent_origin, 0, sizeof(parent_origin));
 890
 891        /* The first pass looks for unrenamed path to optimize for
 892         * common cases, then we look for renames in the second pass.
 893         */
 894        for (pass = 0; pass < 2; pass++) {
 895                struct origin *(*find)(struct scoreboard *,
 896                                       struct commit *, struct origin *);
 897                find = pass ? find_rename : find_origin;
 898
 899                for (i = 0, parent = commit->parents;
 900                     i < MAXPARENT && parent;
 901                     parent = parent->next, i++) {
 902                        struct commit *p = parent->item;
 903
 904                        if (parent_origin[i])
 905                                continue;
 906                        if (parse_commit(p))
 907                                continue;
 908                        porigin = find(sb, parent->item, origin);
 909                        if (!porigin)
 910                                continue;
 911                        if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
 912                                struct blame_entry *e;
 913                                for (e = sb->ent; e; e = e->next)
 914                                        if (e->suspect == origin) {
 915                                                origin_incref(porigin);
 916                                                origin_decref(e->suspect);
 917                                                e->suspect = porigin;
 918                                        }
 919                                origin_decref(porigin);
 920                                goto finish;
 921                        }
 922                        parent_origin[i] = porigin;
 923                }
 924        }
 925
 926        for (i = 0, parent = commit->parents;
 927             i < MAXPARENT && parent;
 928             parent = parent->next, i++) {
 929                struct origin *porigin = parent_origin[i];
 930                if (!porigin)
 931                        continue;
 932                if (pass_blame_to_parent(sb, origin, porigin))
 933                        goto finish;
 934        }
 935
 936        /*
 937         * Optionally run "miff" to find moves in parents' files here.
 938         */
 939        if (opt & PICKAXE_BLAME_MOVE)
 940                for (i = 0, parent = commit->parents;
 941                     i < MAXPARENT && parent;
 942                     parent = parent->next, i++) {
 943                        struct origin *porigin = parent_origin[i];
 944                        if (!porigin)
 945                                continue;
 946                        if (find_move_in_parent(sb, origin, porigin))
 947                                goto finish;
 948                }
 949
 950        /*
 951         * Optionally run "ciff" to find copies from parents' files here.
 952         */
 953        if (opt & PICKAXE_BLAME_COPY)
 954                for (i = 0, parent = commit->parents;
 955                     i < MAXPARENT && parent;
 956                     parent = parent->next, i++) {
 957                        struct origin *porigin = parent_origin[i];
 958                        if (find_copy_in_parent(sb, origin, parent->item,
 959                                                porigin, opt))
 960                                goto finish;
 961                }
 962
 963 finish:
 964        for (i = 0; i < MAXPARENT; i++)
 965                origin_decref(parent_origin[i]);
 966}
 967
 968static void assign_blame(struct scoreboard *sb, struct rev_info *revs, int opt)
 969{
 970        while (1) {
 971                struct blame_entry *ent;
 972                struct commit *commit;
 973                struct origin *suspect = NULL;
 974
 975                /* find one suspect to break down */
 976                for (ent = sb->ent; !suspect && ent; ent = ent->next)
 977                        if (!ent->guilty)
 978                                suspect = ent->suspect;
 979                if (!suspect)
 980                        return; /* all done */
 981
 982                origin_incref(suspect);
 983                commit = suspect->commit;
 984                parse_commit(commit);
 985                if (!(commit->object.flags & UNINTERESTING) &&
 986                    !(revs->max_age != -1 && commit->date  < revs->max_age))
 987                        pass_blame(sb, suspect, opt);
 988
 989                /* Take responsibility for the remaining entries */
 990                for (ent = sb->ent; ent; ent = ent->next)
 991                        if (!cmp_suspect(ent->suspect, suspect))
 992                                ent->guilty = 1;
 993                origin_decref(suspect);
 994
 995                if (DEBUG) /* sanity */
 996                        sanity_check_refcnt(sb);
 997        }
 998}
 999
1000static const char *format_time(unsigned long time, const char *tz_str,
1001                               int show_raw_time)
1002{
1003        static char time_buf[128];
1004        time_t t = time;
1005        int minutes, tz;
1006        struct tm *tm;
1007
1008        if (show_raw_time) {
1009                sprintf(time_buf, "%lu %s", time, tz_str);
1010                return time_buf;
1011        }
1012
1013        tz = atoi(tz_str);
1014        minutes = tz < 0 ? -tz : tz;
1015        minutes = (minutes / 100)*60 + (minutes % 100);
1016        minutes = tz < 0 ? -minutes : minutes;
1017        t = time + minutes * 60;
1018        tm = gmtime(&t);
1019
1020        strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1021        strcat(time_buf, tz_str);
1022        return time_buf;
1023}
1024
1025struct commit_info
1026{
1027        char *author;
1028        char *author_mail;
1029        unsigned long author_time;
1030        char *author_tz;
1031
1032        /* filled only when asked for details */
1033        char *committer;
1034        char *committer_mail;
1035        unsigned long committer_time;
1036        char *committer_tz;
1037
1038        char *summary;
1039};
1040
1041static void get_ac_line(const char *inbuf, const char *what,
1042                        int bufsz, char *person, char **mail,
1043                        unsigned long *time, char **tz)
1044{
1045        int len;
1046        char *tmp, *endp;
1047
1048        tmp = strstr(inbuf, what);
1049        if (!tmp)
1050                goto error_out;
1051        tmp += strlen(what);
1052        endp = strchr(tmp, '\n');
1053        if (!endp)
1054                len = strlen(tmp);
1055        else
1056                len = endp - tmp;
1057        if (bufsz <= len) {
1058        error_out:
1059                /* Ugh */
1060                person = *mail = *tz = "(unknown)";
1061                *time = 0;
1062                return;
1063        }
1064        memcpy(person, tmp, len);
1065
1066        tmp = person;
1067        tmp += len;
1068        *tmp = 0;
1069        while (*tmp != ' ')
1070                tmp--;
1071        *tz = tmp+1;
1072
1073        *tmp = 0;
1074        while (*tmp != ' ')
1075                tmp--;
1076        *time = strtoul(tmp, NULL, 10);
1077
1078        *tmp = 0;
1079        while (*tmp != ' ')
1080                tmp--;
1081        *mail = tmp + 1;
1082        *tmp = 0;
1083}
1084
1085static void get_commit_info(struct commit *commit,
1086                            struct commit_info *ret,
1087                            int detailed)
1088{
1089        int len;
1090        char *tmp, *endp;
1091        static char author_buf[1024];
1092        static char committer_buf[1024];
1093        static char summary_buf[1024];
1094
1095        /* We've operated without save_commit_buffer, so
1096         * we now need to populate them for output.
1097         */
1098        if (!commit->buffer) {
1099                char type[20];
1100                unsigned long size;
1101                commit->buffer =
1102                        read_sha1_file(commit->object.sha1, type, &size);
1103        }
1104        ret->author = author_buf;
1105        get_ac_line(commit->buffer, "\nauthor ",
1106                    sizeof(author_buf), author_buf, &ret->author_mail,
1107                    &ret->author_time, &ret->author_tz);
1108
1109        if (!detailed)
1110                return;
1111
1112        ret->committer = committer_buf;
1113        get_ac_line(commit->buffer, "\ncommitter ",
1114                    sizeof(committer_buf), committer_buf, &ret->committer_mail,
1115                    &ret->committer_time, &ret->committer_tz);
1116
1117        ret->summary = summary_buf;
1118        tmp = strstr(commit->buffer, "\n\n");
1119        if (!tmp) {
1120        error_out:
1121                sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1122                return;
1123        }
1124        tmp += 2;
1125        endp = strchr(tmp, '\n');
1126        if (!endp)
1127                goto error_out;
1128        len = endp - tmp;
1129        if (len >= sizeof(summary_buf))
1130                goto error_out;
1131        memcpy(summary_buf, tmp, len);
1132        summary_buf[len] = 0;
1133}
1134
1135#define OUTPUT_ANNOTATE_COMPAT  001
1136#define OUTPUT_LONG_OBJECT_NAME 002
1137#define OUTPUT_RAW_TIMESTAMP    004
1138#define OUTPUT_PORCELAIN        010
1139#define OUTPUT_SHOW_NAME        020
1140#define OUTPUT_SHOW_NUMBER      040
1141#define OUTPUT_SHOW_SCORE      0100
1142
1143static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1144{
1145        int cnt;
1146        const char *cp;
1147        struct origin *suspect = ent->suspect;
1148        char hex[41];
1149
1150        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1151        printf("%s%c%d %d %d\n",
1152               hex,
1153               ent->guilty ? ' ' : '*', // purely for debugging
1154               ent->s_lno + 1,
1155               ent->lno + 1,
1156               ent->num_lines);
1157        if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1158                struct commit_info ci;
1159                suspect->commit->object.flags |= METAINFO_SHOWN;
1160                get_commit_info(suspect->commit, &ci, 1);
1161                printf("author %s\n", ci.author);
1162                printf("author-mail %s\n", ci.author_mail);
1163                printf("author-time %lu\n", ci.author_time);
1164                printf("author-tz %s\n", ci.author_tz);
1165                printf("committer %s\n", ci.committer);
1166                printf("committer-mail %s\n", ci.committer_mail);
1167                printf("committer-time %lu\n", ci.committer_time);
1168                printf("committer-tz %s\n", ci.committer_tz);
1169                printf("filename %s\n", suspect->path);
1170                printf("summary %s\n", ci.summary);
1171        }
1172        else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1173                printf("filename %s\n", suspect->path);
1174
1175        cp = nth_line(sb, ent->lno);
1176        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1177                char ch;
1178                if (cnt)
1179                        printf("%s %d %d\n", hex,
1180                               ent->s_lno + 1 + cnt,
1181                               ent->lno + 1 + cnt);
1182                putchar('\t');
1183                do {
1184                        ch = *cp++;
1185                        putchar(ch);
1186                } while (ch != '\n' &&
1187                         cp < sb->final_buf + sb->final_buf_size);
1188        }
1189}
1190
1191static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1192{
1193        int cnt;
1194        const char *cp;
1195        struct origin *suspect = ent->suspect;
1196        struct commit_info ci;
1197        char hex[41];
1198        int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1199
1200        get_commit_info(suspect->commit, &ci, 1);
1201        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1202
1203        cp = nth_line(sb, ent->lno);
1204        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1205                char ch;
1206
1207                printf("%.*s", (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8, hex);
1208                if (opt & OUTPUT_ANNOTATE_COMPAT)
1209                        printf("\t(%10s\t%10s\t%d)", ci.author,
1210                               format_time(ci.author_time, ci.author_tz,
1211                                           show_raw_time),
1212                               ent->lno + 1 + cnt);
1213                else {
1214                        if (opt & OUTPUT_SHOW_SCORE)
1215                                printf(" %*d %02d",
1216                                       max_score_digits, ent->score,
1217                                       ent->suspect->refcnt);
1218                        if (opt & OUTPUT_SHOW_NAME)
1219                                printf(" %-*.*s", longest_file, longest_file,
1220                                       suspect->path);
1221                        if (opt & OUTPUT_SHOW_NUMBER)
1222                                printf(" %*d", max_orig_digits,
1223                                       ent->s_lno + 1 + cnt);
1224                        printf(" (%-*.*s %10s %*d) ",
1225                               longest_author, longest_author, ci.author,
1226                               format_time(ci.author_time, ci.author_tz,
1227                                           show_raw_time),
1228                               max_digits, ent->lno + 1 + cnt);
1229                }
1230                do {
1231                        ch = *cp++;
1232                        putchar(ch);
1233                } while (ch != '\n' &&
1234                         cp < sb->final_buf + sb->final_buf_size);
1235        }
1236}
1237
1238static void output(struct scoreboard *sb, int option)
1239{
1240        struct blame_entry *ent;
1241
1242        if (option & OUTPUT_PORCELAIN) {
1243                for (ent = sb->ent; ent; ent = ent->next) {
1244                        struct blame_entry *oth;
1245                        struct origin *suspect = ent->suspect;
1246                        struct commit *commit = suspect->commit;
1247                        if (commit->object.flags & MORE_THAN_ONE_PATH)
1248                                continue;
1249                        for (oth = ent->next; oth; oth = oth->next) {
1250                                if ((oth->suspect->commit != commit) ||
1251                                    !strcmp(oth->suspect->path, suspect->path))
1252                                        continue;
1253                                commit->object.flags |= MORE_THAN_ONE_PATH;
1254                                break;
1255                        }
1256                }
1257        }
1258
1259        for (ent = sb->ent; ent; ent = ent->next) {
1260                if (option & OUTPUT_PORCELAIN)
1261                        emit_porcelain(sb, ent);
1262                else {
1263                        emit_other(sb, ent, option);
1264                }
1265        }
1266}
1267
1268static int prepare_lines(struct scoreboard *sb)
1269{
1270        const char *buf = sb->final_buf;
1271        unsigned long len = sb->final_buf_size;
1272        int num = 0, incomplete = 0, bol = 1;
1273
1274        if (len && buf[len-1] != '\n')
1275                incomplete++; /* incomplete line at the end */
1276        while (len--) {
1277                if (bol) {
1278                        sb->lineno = xrealloc(sb->lineno,
1279                                              sizeof(int* ) * (num + 1));
1280                        sb->lineno[num] = buf - sb->final_buf;
1281                        bol = 0;
1282                }
1283                if (*buf++ == '\n') {
1284                        num++;
1285                        bol = 1;
1286                }
1287        }
1288        sb->lineno = xrealloc(sb->lineno,
1289                              sizeof(int* ) * (num + incomplete + 1));
1290        sb->lineno[num + incomplete] = buf - sb->final_buf;
1291        sb->num_lines = num + incomplete;
1292        return sb->num_lines;
1293}
1294
1295static int read_ancestry(const char *graft_file)
1296{
1297        FILE *fp = fopen(graft_file, "r");
1298        char buf[1024];
1299        if (!fp)
1300                return -1;
1301        while (fgets(buf, sizeof(buf), fp)) {
1302                /* The format is just "Commit Parent1 Parent2 ...\n" */
1303                int len = strlen(buf);
1304                struct commit_graft *graft = read_graft_line(buf, len);
1305                register_commit_graft(graft, 0);
1306        }
1307        fclose(fp);
1308        return 0;
1309}
1310
1311static int lineno_width(int lines)
1312{
1313        int i, width;
1314
1315        for (width = 1, i = 10; i <= lines + 1; width++)
1316                i *= 10;
1317        return width;
1318}
1319
1320static void find_alignment(struct scoreboard *sb, int *option)
1321{
1322        int longest_src_lines = 0;
1323        int longest_dst_lines = 0;
1324        unsigned largest_score = 0;
1325        struct blame_entry *e;
1326
1327        for (e = sb->ent; e; e = e->next) {
1328                struct origin *suspect = e->suspect;
1329                struct commit_info ci;
1330                int num;
1331
1332                if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1333                        suspect->commit->object.flags |= METAINFO_SHOWN;
1334                        get_commit_info(suspect->commit, &ci, 1);
1335                        if (strcmp(suspect->path, sb->path))
1336                                *option |= OUTPUT_SHOW_NAME;
1337                        num = strlen(suspect->path);
1338                        if (longest_file < num)
1339                                longest_file = num;
1340                        num = strlen(ci.author);
1341                        if (longest_author < num)
1342                                longest_author = num;
1343                }
1344                num = e->s_lno + e->num_lines;
1345                if (longest_src_lines < num)
1346                        longest_src_lines = num;
1347                num = e->lno + e->num_lines;
1348                if (longest_dst_lines < num)
1349                        longest_dst_lines = num;
1350                if (largest_score < ent_score(sb, e))
1351                        largest_score = ent_score(sb, e);
1352        }
1353        max_orig_digits = lineno_width(longest_src_lines);
1354        max_digits = lineno_width(longest_dst_lines);
1355        max_score_digits = lineno_width(largest_score);
1356}
1357
1358static void sanity_check_refcnt(struct scoreboard *sb)
1359{
1360        int baa = 0;
1361        struct blame_entry *ent;
1362
1363        for (ent = sb->ent; ent; ent = ent->next) {
1364                /* Nobody should have zero or negative refcnt */
1365                if (ent->suspect->refcnt <= 0)
1366                        baa = 1;
1367        }
1368        for (ent = sb->ent; ent; ent = ent->next) {
1369                /* Mark the ones that haven't been checked */
1370                if (0 < ent->suspect->refcnt)
1371                        ent->suspect->refcnt = -ent->suspect->refcnt;
1372        }
1373        for (ent = sb->ent; ent; ent = ent->next) {
1374                /* then pick each and see if they have the the
1375                 * correct refcnt
1376                 */
1377                int found;
1378                struct blame_entry *e;
1379                struct origin *suspect = ent->suspect;
1380
1381                if (0 < suspect->refcnt)
1382                        continue;
1383                suspect->refcnt = -suspect->refcnt; /* Unmark */
1384                for (found = 0, e = sb->ent; e; e = e->next) {
1385                        if (e->suspect != suspect)
1386                                continue;
1387                        found++;
1388                }
1389                if (suspect->refcnt != found)
1390                        baa = 1;
1391        }
1392        if (baa) {
1393                int opt = 0160;
1394                find_alignment(sb, &opt);
1395                output(sb, opt);
1396                die("Baa!");
1397        }
1398}
1399
1400static int has_path_in_work_tree(const char *path)
1401{
1402        struct stat st;
1403        return !lstat(path, &st);
1404}
1405
1406static unsigned parse_score(const char *arg)
1407{
1408        char *end;
1409        unsigned long score = strtoul(arg, &end, 10);
1410        if (*end)
1411                return 0;
1412        return score;
1413}
1414
1415int cmd_pickaxe(int argc, const char **argv, const char *prefix)
1416{
1417        struct rev_info revs;
1418        const char *path;
1419        struct scoreboard sb;
1420        struct origin *o;
1421        struct blame_entry *ent;
1422        int i, seen_dashdash, unk, opt;
1423        long bottom, top, lno;
1424        int output_option = 0;
1425        const char *revs_file = NULL;
1426        const char *final_commit_name = NULL;
1427        char type[10];
1428
1429        save_commit_buffer = 0;
1430
1431        opt = 0;
1432        bottom = top = 0;
1433        seen_dashdash = 0;
1434        for (unk = i = 1; i < argc; i++) {
1435                const char *arg = argv[i];
1436                if (*arg != '-')
1437                        break;
1438                else if (!strcmp("-c", arg))
1439                        output_option |= OUTPUT_ANNOTATE_COMPAT;
1440                else if (!strcmp("-t", arg))
1441                        output_option |= OUTPUT_RAW_TIMESTAMP;
1442                else if (!strcmp("-l", arg))
1443                        output_option |= OUTPUT_LONG_OBJECT_NAME;
1444                else if (!strcmp("-S", arg) && ++i < argc)
1445                        revs_file = argv[i];
1446                else if (!strncmp("-M", arg, 2)) {
1447                        opt |= PICKAXE_BLAME_MOVE;
1448                        blame_move_score = parse_score(arg+2);
1449                }
1450                else if (!strncmp("-C", arg, 2)) {
1451                        if (opt & PICKAXE_BLAME_COPY)
1452                                opt |= PICKAXE_BLAME_COPY_HARDER;
1453                        opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
1454                        blame_copy_score = parse_score(arg+2);
1455                }
1456                else if (!strncmp("-L", arg, 2)) {
1457                        char *term;
1458                        if (!arg[2]) {
1459                                if (++i >= argc)
1460                                        usage(pickaxe_usage);
1461                                arg = argv[i];
1462                        }
1463                        else
1464                                arg += 2;
1465                        if (bottom || top)
1466                                die("More than one '-L n,m' option given");
1467                        bottom = strtol(arg, &term, 10);
1468                        if (*term == ',') {
1469                                top = strtol(term + 1, &term, 10);
1470                                if (*term)
1471                                        usage(pickaxe_usage);
1472                        }
1473                        if (bottom && top && top < bottom) {
1474                                unsigned long tmp;
1475                                tmp = top; top = bottom; bottom = tmp;
1476                        }
1477                }
1478                else if (!strcmp("--score-debug", arg))
1479                        output_option |= OUTPUT_SHOW_SCORE;
1480                else if (!strcmp("-f", arg) ||
1481                         !strcmp("--show-name", arg))
1482                        output_option |= OUTPUT_SHOW_NAME;
1483                else if (!strcmp("-n", arg) ||
1484                         !strcmp("--show-number", arg))
1485                        output_option |= OUTPUT_SHOW_NUMBER;
1486                else if (!strcmp("-p", arg) ||
1487                         !strcmp("--porcelain", arg))
1488                        output_option |= OUTPUT_PORCELAIN;
1489                else if (!strcmp("--", arg)) {
1490                        seen_dashdash = 1;
1491                        i++;
1492                        break;
1493                }
1494                else
1495                        argv[unk++] = arg;
1496        }
1497
1498        if (!blame_move_score)
1499                blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
1500        if (!blame_copy_score)
1501                blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
1502
1503        /* We have collected options unknown to us in argv[1..unk]
1504         * which are to be passed to revision machinery if we are
1505         * going to do the "bottom" procesing.
1506         *
1507         * The remaining are:
1508         *
1509         * (1) if seen_dashdash, its either
1510         *     "-options -- <path>" or
1511         *     "-options -- <path> <rev>".
1512         *     but the latter is allowed only if there is no
1513         *     options that we passed to revision machinery.
1514         *
1515         * (2) otherwise, we may have "--" somewhere later and
1516         *     might be looking at the first one of multiple 'rev'
1517         *     parameters (e.g. " master ^next ^maint -- path").
1518         *     See if there is a dashdash first, and give the
1519         *     arguments before that to revision machinery.
1520         *     After that there must be one 'path'.
1521         *
1522         * (3) otherwise, its one of the three:
1523         *     "-options <path> <rev>"
1524         *     "-options <rev> <path>"
1525         *     "-options <path>"
1526         *     but again the first one is allowed only if
1527         *     there is no options that we passed to revision
1528         *     machinery.
1529         */
1530
1531        if (seen_dashdash) {
1532                /* (1) */
1533                if (argc <= i)
1534                        usage(pickaxe_usage);
1535                path = argv[i];
1536                if (i + 1 == argc - 1) {
1537                        if (unk != 1)
1538                                usage(pickaxe_usage);
1539                        argv[unk++] = argv[i + 1];
1540                }
1541                else if (i + 1 != argc)
1542                        /* garbage at end */
1543                        usage(pickaxe_usage);
1544        }
1545        else {
1546                int j;
1547                for (j = i; !seen_dashdash && j < argc; j++)
1548                        if (!strcmp(argv[j], "--"))
1549                                seen_dashdash = j;
1550                if (seen_dashdash) {
1551                        if (seen_dashdash + 1 != argc - 1)
1552                                usage(pickaxe_usage);
1553                        path = argv[seen_dashdash + 1];
1554                        for (j = i; j < seen_dashdash; j++)
1555                                argv[unk++] = argv[j];
1556                }
1557                else {
1558                        /* (3) */
1559                        path = argv[i];
1560                        if (i + 1 == argc - 1) {
1561                                final_commit_name = argv[i + 1];
1562
1563                                /* if (unk == 1) we could be getting
1564                                 * old-style
1565                                 */
1566                                if (unk == 1 && !has_path_in_work_tree(path)) {
1567                                        path = argv[i + 1];
1568                                        final_commit_name = argv[i];
1569                                }
1570                        }
1571                        else if (i != argc - 1)
1572                                usage(pickaxe_usage); /* garbage at end */
1573
1574                        if (!has_path_in_work_tree(path))
1575                                die("cannot stat path %s: %s",
1576                                    path, strerror(errno));
1577                }
1578        }
1579
1580        if (final_commit_name)
1581                argv[unk++] = final_commit_name;
1582
1583        /* Now we got rev and path.  We do not want the path pruning
1584         * but we may want "bottom" processing.
1585         */
1586        argv[unk] = NULL;
1587
1588        init_revisions(&revs, NULL);
1589        setup_revisions(unk, argv, &revs, "HEAD");
1590        memset(&sb, 0, sizeof(sb));
1591
1592        /* There must be one and only one positive commit in the
1593         * revs->pending array.
1594         */
1595        for (i = 0; i < revs.pending.nr; i++) {
1596                struct object *obj = revs.pending.objects[i].item;
1597                if (obj->flags & UNINTERESTING)
1598                        continue;
1599                while (obj->type == OBJ_TAG)
1600                        obj = deref_tag(obj, NULL, 0);
1601                if (obj->type != OBJ_COMMIT)
1602                        die("Non commit %s?",
1603                            revs.pending.objects[i].name);
1604                if (sb.final)
1605                        die("More than one commit to dig from %s and %s?",
1606                            revs.pending.objects[i].name,
1607                            final_commit_name);
1608                sb.final = (struct commit *) obj;
1609                final_commit_name = revs.pending.objects[i].name;
1610        }
1611
1612        if (!sb.final) {
1613                /* "--not A B -- path" without anything positive */
1614                unsigned char head_sha1[20];
1615
1616                final_commit_name = "HEAD";
1617                if (get_sha1(final_commit_name, head_sha1))
1618                        die("No such ref: HEAD");
1619                sb.final = lookup_commit_reference(head_sha1);
1620                add_pending_object(&revs, &(sb.final->object), "HEAD");
1621        }
1622
1623        /* If we have bottom, this will mark the ancestors of the
1624         * bottom commits we would reach while traversing as
1625         * uninteresting.
1626         */
1627        prepare_revision_walk(&revs);
1628
1629        o = get_origin(&sb, sb.final, path);
1630        if (fill_blob_sha1(o))
1631                die("no such path %s in %s", path, final_commit_name);
1632
1633        sb.final_buf = read_sha1_file(o->blob_sha1, type, &sb.final_buf_size);
1634        lno = prepare_lines(&sb);
1635
1636        if (bottom < 1)
1637                bottom = 1;
1638        if (top < 1)
1639                top = lno;
1640        bottom--;
1641        if (lno < top)
1642                die("file %s has only %lu lines", path, lno);
1643
1644        ent = xcalloc(1, sizeof(*ent));
1645        ent->lno = bottom;
1646        ent->num_lines = top - bottom;
1647        ent->suspect = o;
1648        ent->s_lno = bottom;
1649
1650        sb.ent = ent;
1651        sb.path = path;
1652
1653        if (revs_file && read_ancestry(revs_file))
1654                die("reading graft file %s failed: %s",
1655                    revs_file, strerror(errno));
1656
1657        assign_blame(&sb, &revs, opt);
1658
1659        coalesce(&sb);
1660
1661        if (!(output_option & OUTPUT_PORCELAIN))
1662                find_alignment(&sb, &output_option);
1663
1664        output(&sb, output_option);
1665        free((void *)sb.final_buf);
1666        for (ent = sb.ent; ent; ) {
1667                struct blame_entry *e = ent->next;
1668                free(ent);
1669                ent = e;
1670        }
1671        return 0;
1672}