f12b2d45447d42b604fec12169678124a7ce984c
   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/* stats */
  44static int num_read_blob;
  45static int num_get_patch;
  46static int num_commits;
  47
  48#define PICKAXE_BLAME_MOVE              01
  49#define PICKAXE_BLAME_COPY              02
  50#define PICKAXE_BLAME_COPY_HARDER       04
  51
  52/*
  53 * blame for a blame_entry with score lower than these thresholds
  54 * is not passed to the parent using move/copy logic.
  55 */
  56static unsigned blame_move_score;
  57static unsigned blame_copy_score;
  58#define BLAME_DEFAULT_MOVE_SCORE        20
  59#define BLAME_DEFAULT_COPY_SCORE        40
  60
  61/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
  62#define METAINFO_SHOWN          (1u<<12)
  63#define MORE_THAN_ONE_PATH      (1u<<13)
  64
  65/*
  66 * One blob in a commit that is being suspected
  67 */
  68struct origin {
  69        int refcnt;
  70        struct commit *commit;
  71        mmfile_t file;
  72        unsigned char blob_sha1[20];
  73        char path[FLEX_ARRAY];
  74};
  75
  76static char *fill_origin_blob(struct origin *o, mmfile_t *file)
  77{
  78        if (!o->file.ptr) {
  79                char type[10];
  80                num_read_blob++;
  81                file->ptr = read_sha1_file(o->blob_sha1, type,
  82                                           (unsigned long *)(&(file->size)));
  83                o->file = *file;
  84        }
  85        else
  86                *file = o->file;
  87        return file->ptr;
  88}
  89
  90static inline struct origin *origin_incref(struct origin *o)
  91{
  92        if (o)
  93                o->refcnt++;
  94        return o;
  95}
  96
  97static void origin_decref(struct origin *o)
  98{
  99        if (o && --o->refcnt <= 0) {
 100                if (o->file.ptr)
 101                        free(o->file.ptr);
 102                memset(o, 0, sizeof(*o));
 103                free(o);
 104        }
 105}
 106
 107struct blame_entry {
 108        struct blame_entry *prev;
 109        struct blame_entry *next;
 110
 111        /* the first line of this group in the final image;
 112         * internally all line numbers are 0 based.
 113         */
 114        int lno;
 115
 116        /* how many lines this group has */
 117        int num_lines;
 118
 119        /* the commit that introduced this group into the final image */
 120        struct origin *suspect;
 121
 122        /* true if the suspect is truly guilty; false while we have not
 123         * checked if the group came from one of its parents.
 124         */
 125        char guilty;
 126
 127        /* the line number of the first line of this group in the
 128         * suspect's file; internally all line numbers are 0 based.
 129         */
 130        int s_lno;
 131
 132        /* how significant this entry is -- cached to avoid
 133         * scanning the lines over and over
 134         */
 135        unsigned score;
 136};
 137
 138struct scoreboard {
 139        /* the final commit (i.e. where we started digging from) */
 140        struct commit *final;
 141
 142        const char *path;
 143
 144        /* the contents in the final; pointed into by buf pointers of
 145         * blame_entries
 146         */
 147        const char *final_buf;
 148        unsigned long final_buf_size;
 149
 150        /* linked list of blames */
 151        struct blame_entry *ent;
 152
 153        /* look-up a line in the final buffer */
 154        int num_lines;
 155        int *lineno;
 156};
 157
 158static int cmp_suspect(struct origin *a, struct origin *b)
 159{
 160        int cmp = hashcmp(a->commit->object.sha1, b->commit->object.sha1);
 161        if (cmp)
 162                return cmp;
 163        return strcmp(a->path, b->path);
 164}
 165
 166#define cmp_suspect(a, b) ( ((a)==(b)) ? 0 : cmp_suspect(a,b) )
 167
 168static void sanity_check_refcnt(struct scoreboard *);
 169
 170static void coalesce(struct scoreboard *sb)
 171{
 172        struct blame_entry *ent, *next;
 173
 174        for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 175                if (!cmp_suspect(ent->suspect, next->suspect) &&
 176                    ent->guilty == next->guilty &&
 177                    ent->s_lno + ent->num_lines == next->s_lno) {
 178                        ent->num_lines += next->num_lines;
 179                        ent->next = next->next;
 180                        if (ent->next)
 181                                ent->next->prev = ent;
 182                        origin_decref(next->suspect);
 183                        free(next);
 184                        ent->score = 0;
 185                        next = ent; /* again */
 186                }
 187        }
 188
 189        if (DEBUG) /* sanity */
 190                sanity_check_refcnt(sb);
 191}
 192
 193static struct origin *make_origin(struct commit *commit, const char *path)
 194{
 195        struct origin *o;
 196        o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 197        o->commit = commit;
 198        o->refcnt = 1;
 199        strcpy(o->path, path);
 200        return o;
 201}
 202
 203static struct origin *get_origin(struct scoreboard *sb,
 204                                 struct commit *commit,
 205                                 const char *path)
 206{
 207        struct blame_entry *e;
 208
 209        for (e = sb->ent; e; e = e->next) {
 210                if (e->suspect->commit == commit &&
 211                    !strcmp(e->suspect->path, path))
 212                        return origin_incref(e->suspect);
 213        }
 214        return make_origin(commit, path);
 215}
 216
 217static int fill_blob_sha1(struct origin *origin)
 218{
 219        unsigned mode;
 220        char type[10];
 221
 222        if (!is_null_sha1(origin->blob_sha1))
 223                return 0;
 224        if (get_tree_entry(origin->commit->object.sha1,
 225                           origin->path,
 226                           origin->blob_sha1, &mode))
 227                goto error_out;
 228        if (sha1_object_info(origin->blob_sha1, type, NULL) ||
 229            strcmp(type, blob_type))
 230                goto error_out;
 231        return 0;
 232 error_out:
 233        hashclr(origin->blob_sha1);
 234        return -1;
 235}
 236
 237static struct origin *find_origin(struct scoreboard *sb,
 238                                  struct commit *parent,
 239                                  struct origin *origin)
 240{
 241        struct origin *porigin = NULL;
 242        struct diff_options diff_opts;
 243        const char *paths[2];
 244
 245        if (parent->util) {
 246                /* This is a freestanding copy of origin and not
 247                 * refcounted.
 248                 */
 249                struct origin *cached = parent->util;
 250                if (!strcmp(cached->path, origin->path)) {
 251                        porigin = get_origin(sb, parent, cached->path);
 252                        if (porigin->refcnt == 1)
 253                                hashcpy(porigin->blob_sha1, cached->blob_sha1);
 254                        return porigin;
 255                }
 256                /* otherwise it was not very useful; free it */
 257                free(parent->util);
 258                parent->util = NULL;
 259        }
 260
 261        /* See if the origin->path is different between parent
 262         * and origin first.  Most of the time they are the
 263         * same and diff-tree is fairly efficient about this.
 264         */
 265        diff_setup(&diff_opts);
 266        diff_opts.recursive = 1;
 267        diff_opts.detect_rename = 0;
 268        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 269        paths[0] = origin->path;
 270        paths[1] = NULL;
 271
 272        diff_tree_setup_paths(paths, &diff_opts);
 273        if (diff_setup_done(&diff_opts) < 0)
 274                die("diff-setup");
 275        diff_tree_sha1(parent->tree->object.sha1,
 276                       origin->commit->tree->object.sha1,
 277                       "", &diff_opts);
 278        diffcore_std(&diff_opts);
 279
 280        /* It is either one entry that says "modified", or "created",
 281         * or nothing.
 282         */
 283        if (!diff_queued_diff.nr) {
 284                /* The path is the same as parent */
 285                porigin = get_origin(sb, parent, origin->path);
 286                hashcpy(porigin->blob_sha1, origin->blob_sha1);
 287        }
 288        else if (diff_queued_diff.nr != 1)
 289                die("internal error in pickaxe::find_origin");
 290        else {
 291                struct diff_filepair *p = diff_queued_diff.queue[0];
 292                switch (p->status) {
 293                default:
 294                        die("internal error in pickaxe::find_origin (%c)",
 295                            p->status);
 296                case 'M':
 297                        porigin = get_origin(sb, parent, origin->path);
 298                        hashcpy(porigin->blob_sha1, p->one->sha1);
 299                        break;
 300                case 'A':
 301                case 'T':
 302                        /* Did not exist in parent, or type changed */
 303                        break;
 304                }
 305        }
 306        diff_flush(&diff_opts);
 307        if (porigin) {
 308                struct origin *cached;
 309                cached = make_origin(porigin->commit, porigin->path);
 310                hashcpy(cached->blob_sha1, porigin->blob_sha1);
 311                parent->util = cached;
 312        }
 313        return porigin;
 314}
 315
 316static struct origin *find_rename(struct scoreboard *sb,
 317                                  struct commit *parent,
 318                                  struct origin *origin)
 319{
 320        struct origin *porigin = NULL;
 321        struct diff_options diff_opts;
 322        int i;
 323        const char *paths[2];
 324
 325        diff_setup(&diff_opts);
 326        diff_opts.recursive = 1;
 327        diff_opts.detect_rename = DIFF_DETECT_RENAME;
 328        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 329        diff_opts.single_follow = origin->path;
 330        paths[0] = NULL;
 331        diff_tree_setup_paths(paths, &diff_opts);
 332        if (diff_setup_done(&diff_opts) < 0)
 333                die("diff-setup");
 334        diff_tree_sha1(parent->tree->object.sha1,
 335                       origin->commit->tree->object.sha1,
 336                       "", &diff_opts);
 337        diffcore_std(&diff_opts);
 338
 339        for (i = 0; i < diff_queued_diff.nr; i++) {
 340                struct diff_filepair *p = diff_queued_diff.queue[i];
 341                if ((p->status == 'R' || p->status == 'C') &&
 342                    !strcmp(p->two->path, origin->path)) {
 343                        porigin = get_origin(sb, parent, p->one->path);
 344                        hashcpy(porigin->blob_sha1, p->one->sha1);
 345                        break;
 346                }
 347        }
 348        diff_flush(&diff_opts);
 349        return porigin;
 350}
 351
 352struct chunk {
 353        /* line number in postimage; up to but not including this
 354         * line is the same as preimage
 355         */
 356        int same;
 357
 358        /* preimage line number after this chunk */
 359        int p_next;
 360
 361        /* postimage line number after this chunk */
 362        int t_next;
 363};
 364
 365struct patch {
 366        struct chunk *chunks;
 367        int num;
 368};
 369
 370struct blame_diff_state {
 371        struct xdiff_emit_state xm;
 372        struct patch *ret;
 373        unsigned hunk_post_context;
 374        unsigned hunk_in_pre_context : 1;
 375};
 376
 377static void process_u_diff(void *state_, char *line, unsigned long len)
 378{
 379        struct blame_diff_state *state = state_;
 380        struct chunk *chunk;
 381        int off1, off2, len1, len2, num;
 382
 383        num = state->ret->num;
 384        if (len < 4 || line[0] != '@' || line[1] != '@') {
 385                if (state->hunk_in_pre_context && line[0] == ' ')
 386                        state->ret->chunks[num - 1].same++;
 387                else {
 388                        state->hunk_in_pre_context = 0;
 389                        if (line[0] == ' ')
 390                                state->hunk_post_context++;
 391                        else
 392                                state->hunk_post_context = 0;
 393                }
 394                return;
 395        }
 396
 397        if (num && state->hunk_post_context) {
 398                chunk = &state->ret->chunks[num - 1];
 399                chunk->p_next -= state->hunk_post_context;
 400                chunk->t_next -= state->hunk_post_context;
 401        }
 402        state->ret->num = ++num;
 403        state->ret->chunks = xrealloc(state->ret->chunks,
 404                                      sizeof(struct chunk) * num);
 405        chunk = &state->ret->chunks[num - 1];
 406        if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
 407                state->ret->num--;
 408                return;
 409        }
 410
 411        /* Line numbers in patch output are one based. */
 412        off1--;
 413        off2--;
 414
 415        chunk->same = len2 ? off2 : (off2 + 1);
 416
 417        chunk->p_next = off1 + (len1 ? len1 : 1);
 418        chunk->t_next = chunk->same + len2;
 419        state->hunk_in_pre_context = 1;
 420        state->hunk_post_context = 0;
 421}
 422
 423static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
 424                                    int context)
 425{
 426        struct blame_diff_state state;
 427        xpparam_t xpp;
 428        xdemitconf_t xecfg;
 429        xdemitcb_t ecb;
 430
 431        xpp.flags = XDF_NEED_MINIMAL;
 432        xecfg.ctxlen = context;
 433        xecfg.flags = 0;
 434        ecb.outf = xdiff_outf;
 435        ecb.priv = &state;
 436        memset(&state, 0, sizeof(state));
 437        state.xm.consume = process_u_diff;
 438        state.ret = xmalloc(sizeof(struct patch));
 439        state.ret->chunks = NULL;
 440        state.ret->num = 0;
 441
 442        xdl_diff(file_p, file_o, &xpp, &xecfg, &ecb);
 443
 444        if (state.ret->num) {
 445                struct chunk *chunk;
 446                chunk = &state.ret->chunks[state.ret->num - 1];
 447                chunk->p_next -= state.hunk_post_context;
 448                chunk->t_next -= state.hunk_post_context;
 449        }
 450        return state.ret;
 451}
 452
 453static struct patch *get_patch(struct origin *parent, struct origin *origin)
 454{
 455        mmfile_t file_p, file_o;
 456        struct patch *patch;
 457
 458        fill_origin_blob(parent, &file_p);
 459        fill_origin_blob(origin, &file_o);
 460        if (!file_p.ptr || !file_o.ptr)
 461                return NULL;
 462        patch = compare_buffer(&file_p, &file_o, 0);
 463        num_get_patch++;
 464        return patch;
 465}
 466
 467static void free_patch(struct patch *p)
 468{
 469        free(p->chunks);
 470        free(p);
 471}
 472
 473static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 474{
 475        struct blame_entry *ent, *prev = NULL;
 476
 477        origin_incref(e->suspect);
 478
 479        for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 480                prev = ent;
 481
 482        /* prev, if not NULL, is the last one that is below e */
 483        e->prev = prev;
 484        if (prev) {
 485                e->next = prev->next;
 486                prev->next = e;
 487        }
 488        else {
 489                e->next = sb->ent;
 490                sb->ent = e;
 491        }
 492        if (e->next)
 493                e->next->prev = e;
 494}
 495
 496static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 497{
 498        struct blame_entry *p, *n;
 499
 500        p = dst->prev;
 501        n = dst->next;
 502        origin_incref(src->suspect);
 503        origin_decref(dst->suspect);
 504        memcpy(dst, src, sizeof(*src));
 505        dst->prev = p;
 506        dst->next = n;
 507        dst->score = 0;
 508}
 509
 510static const char *nth_line(struct scoreboard *sb, int lno)
 511{
 512        return sb->final_buf + sb->lineno[lno];
 513}
 514
 515static void split_overlap(struct blame_entry *split,
 516                          struct blame_entry *e,
 517                          int tlno, int plno, int same,
 518                          struct origin *parent)
 519{
 520        /* it is known that lines between tlno to same came from
 521         * parent, and e has an overlap with that range.  it also is
 522         * known that parent's line plno corresponds to e's line tlno.
 523         *
 524         *                <---- e ----->
 525         *                   <------>
 526         *                   <------------>
 527         *             <------------>
 528         *             <------------------>
 529         *
 530         * Potentially we need to split e into three parts; before
 531         * this chunk, the chunk to be blamed for parent, and after
 532         * that portion.
 533         */
 534        int chunk_end_lno;
 535        memset(split, 0, sizeof(struct blame_entry [3]));
 536
 537        if (e->s_lno < tlno) {
 538                /* there is a pre-chunk part not blamed on parent */
 539                split[0].suspect = origin_incref(e->suspect);
 540                split[0].lno = e->lno;
 541                split[0].s_lno = e->s_lno;
 542                split[0].num_lines = tlno - e->s_lno;
 543                split[1].lno = e->lno + tlno - e->s_lno;
 544                split[1].s_lno = plno;
 545        }
 546        else {
 547                split[1].lno = e->lno;
 548                split[1].s_lno = plno + (e->s_lno - tlno);
 549        }
 550
 551        if (same < e->s_lno + e->num_lines) {
 552                /* there is a post-chunk part not blamed on parent */
 553                split[2].suspect = origin_incref(e->suspect);
 554                split[2].lno = e->lno + (same - e->s_lno);
 555                split[2].s_lno = e->s_lno + (same - e->s_lno);
 556                split[2].num_lines = e->s_lno + e->num_lines - same;
 557                chunk_end_lno = split[2].lno;
 558        }
 559        else
 560                chunk_end_lno = e->lno + e->num_lines;
 561        split[1].num_lines = chunk_end_lno - split[1].lno;
 562
 563        if (split[1].num_lines < 1)
 564                return;
 565        split[1].suspect = origin_incref(parent);
 566}
 567
 568static void split_blame(struct scoreboard *sb,
 569                        struct blame_entry *split,
 570                        struct blame_entry *e)
 571{
 572        struct blame_entry *new_entry;
 573
 574        if (split[0].suspect && split[2].suspect) {
 575                /* we need to split e into two and add another for parent */
 576                dup_entry(e, &split[0]);
 577
 578                new_entry = xmalloc(sizeof(*new_entry));
 579                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 580                add_blame_entry(sb, new_entry);
 581
 582                new_entry = xmalloc(sizeof(*new_entry));
 583                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 584                add_blame_entry(sb, new_entry);
 585        }
 586        else if (!split[0].suspect && !split[2].suspect)
 587                /* parent covers the entire area */
 588                dup_entry(e, &split[1]);
 589        else if (split[0].suspect) {
 590                dup_entry(e, &split[0]);
 591
 592                new_entry = xmalloc(sizeof(*new_entry));
 593                memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 594                add_blame_entry(sb, new_entry);
 595        }
 596        else {
 597                dup_entry(e, &split[1]);
 598
 599                new_entry = xmalloc(sizeof(*new_entry));
 600                memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 601                add_blame_entry(sb, new_entry);
 602        }
 603
 604        if (DEBUG) { /* sanity */
 605                struct blame_entry *ent;
 606                int lno = sb->ent->lno, corrupt = 0;
 607
 608                for (ent = sb->ent; ent; ent = ent->next) {
 609                        if (lno != ent->lno)
 610                                corrupt = 1;
 611                        if (ent->s_lno < 0)
 612                                corrupt = 1;
 613                        lno += ent->num_lines;
 614                }
 615                if (corrupt) {
 616                        lno = sb->ent->lno;
 617                        for (ent = sb->ent; ent; ent = ent->next) {
 618                                printf("L %8d l %8d n %8d\n",
 619                                       lno, ent->lno, ent->num_lines);
 620                                lno = ent->lno + ent->num_lines;
 621                        }
 622                        die("oops");
 623                }
 624        }
 625}
 626
 627static void decref_split(struct blame_entry *split)
 628{
 629        int i;
 630
 631        for (i = 0; i < 3; i++)
 632                origin_decref(split[i].suspect);
 633}
 634
 635static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 636                          int tlno, int plno, int same,
 637                          struct origin *parent)
 638{
 639        struct blame_entry split[3];
 640
 641        split_overlap(split, e, tlno, plno, same, parent);
 642        if (split[1].suspect)
 643                split_blame(sb, split, e);
 644        decref_split(split);
 645}
 646
 647static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 648{
 649        struct blame_entry *e;
 650        int last_in_target = -1;
 651
 652        for (e = sb->ent; e; e = e->next) {
 653                if (e->guilty || cmp_suspect(e->suspect, target))
 654                        continue;
 655                if (last_in_target < e->s_lno + e->num_lines)
 656                        last_in_target = e->s_lno + e->num_lines;
 657        }
 658        return last_in_target;
 659}
 660
 661static void blame_chunk(struct scoreboard *sb,
 662                        int tlno, int plno, int same,
 663                        struct origin *target, struct origin *parent)
 664{
 665        struct blame_entry *e;
 666
 667        for (e = sb->ent; e; e = e->next) {
 668                if (e->guilty || cmp_suspect(e->suspect, target))
 669                        continue;
 670                if (same <= e->s_lno)
 671                        continue;
 672                if (tlno < e->s_lno + e->num_lines)
 673                        blame_overlap(sb, e, tlno, plno, same, parent);
 674        }
 675}
 676
 677static int pass_blame_to_parent(struct scoreboard *sb,
 678                                struct origin *target,
 679                                struct origin *parent)
 680{
 681        int i, last_in_target, plno, tlno;
 682        struct patch *patch;
 683
 684        last_in_target = find_last_in_target(sb, target);
 685        if (last_in_target < 0)
 686                return 1; /* nothing remains for this target */
 687
 688        patch = get_patch(parent, target);
 689        plno = tlno = 0;
 690        for (i = 0; i < patch->num; i++) {
 691                struct chunk *chunk = &patch->chunks[i];
 692
 693                blame_chunk(sb, tlno, plno, chunk->same, target, parent);
 694                plno = chunk->p_next;
 695                tlno = chunk->t_next;
 696        }
 697        /* rest (i.e. anything above tlno) are the same as parent */
 698        blame_chunk(sb, tlno, plno, last_in_target, target, parent);
 699
 700        free_patch(patch);
 701        return 0;
 702}
 703
 704static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 705{
 706        unsigned score;
 707        const char *cp, *ep;
 708
 709        if (e->score)
 710                return e->score;
 711
 712        score = 1;
 713        cp = nth_line(sb, e->lno);
 714        ep = nth_line(sb, e->lno + e->num_lines);
 715        while (cp < ep) {
 716                unsigned ch = *((unsigned char *)cp);
 717                if (isalnum(ch))
 718                        score++;
 719                cp++;
 720        }
 721        e->score = score;
 722        return score;
 723}
 724
 725static void copy_split_if_better(struct scoreboard *sb,
 726                                 struct blame_entry *best_so_far,
 727                                 struct blame_entry *this)
 728{
 729        int i;
 730
 731        if (!this[1].suspect)
 732                return;
 733        if (best_so_far[1].suspect) {
 734                if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 735                        return;
 736        }
 737
 738        for (i = 0; i < 3; i++)
 739                origin_incref(this[i].suspect);
 740        decref_split(best_so_far);
 741        memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 742}
 743
 744static void find_copy_in_blob(struct scoreboard *sb,
 745                              struct blame_entry *ent,
 746                              struct origin *parent,
 747                              struct blame_entry *split,
 748                              mmfile_t *file_p)
 749{
 750        const char *cp;
 751        int cnt;
 752        mmfile_t file_o;
 753        struct patch *patch;
 754        int i, plno, tlno;
 755
 756        cp = nth_line(sb, ent->lno);
 757        file_o.ptr = (char*) cp;
 758        cnt = ent->num_lines;
 759
 760        while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 761                if (*cp++ == '\n')
 762                        cnt--;
 763        }
 764        file_o.size = cp - file_o.ptr;
 765
 766        patch = compare_buffer(file_p, &file_o, 1);
 767
 768        memset(split, 0, sizeof(struct blame_entry [3]));
 769        plno = tlno = 0;
 770        for (i = 0; i < patch->num; i++) {
 771                struct chunk *chunk = &patch->chunks[i];
 772
 773                /* tlno to chunk->same are the same as ent */
 774                if (ent->num_lines <= tlno)
 775                        break;
 776                if (tlno < chunk->same) {
 777                        struct blame_entry this[3];
 778                        split_overlap(this, ent,
 779                                      tlno + ent->s_lno, plno,
 780                                      chunk->same + ent->s_lno,
 781                                      parent);
 782                        copy_split_if_better(sb, split, this);
 783                        decref_split(this);
 784                }
 785                plno = chunk->p_next;
 786                tlno = chunk->t_next;
 787        }
 788        free_patch(patch);
 789}
 790
 791static int find_move_in_parent(struct scoreboard *sb,
 792                               struct origin *target,
 793                               struct origin *parent)
 794{
 795        int last_in_target, made_progress;
 796        struct blame_entry *e, split[3];
 797        mmfile_t file_p;
 798
 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(parent, &file_p);
 804        if (!file_p.ptr)
 805                return 0;
 806
 807        made_progress = 1;
 808        while (made_progress) {
 809                made_progress = 0;
 810                for (e = sb->ent; e; e = e->next) {
 811                        if (e->guilty || cmp_suspect(e->suspect, target))
 812                                continue;
 813                        find_copy_in_blob(sb, e, parent, split, &file_p);
 814                        if (split[1].suspect &&
 815                            blame_move_score < ent_score(sb, &split[1])) {
 816                                split_blame(sb, split, e);
 817                                made_progress = 1;
 818                        }
 819                        decref_split(split);
 820                }
 821        }
 822        return 0;
 823}
 824
 825
 826struct blame_list {
 827        struct blame_entry *ent;
 828        struct blame_entry split[3];
 829};
 830
 831static struct blame_list *setup_blame_list(struct scoreboard *sb,
 832                                           struct origin *target,
 833                                           int *num_ents_p)
 834{
 835        struct blame_entry *e;
 836        int num_ents, i;
 837        struct blame_list *blame_list = NULL;
 838
 839        /* Count the number of entries the target is suspected for,
 840         * and prepare a list of entry and the best split.
 841         */
 842        for (e = sb->ent, num_ents = 0; e; e = e->next)
 843                if (!e->guilty && !cmp_suspect(e->suspect, target))
 844                        num_ents++;
 845        if (num_ents) {
 846                blame_list = xcalloc(num_ents, sizeof(struct blame_list));
 847                for (e = sb->ent, i = 0; e; e = e->next)
 848                        if (!e->guilty && !cmp_suspect(e->suspect, target))
 849                                blame_list[i++].ent = e;
 850        }
 851        *num_ents_p = num_ents;
 852        return blame_list;
 853}
 854
 855static int find_copy_in_parent(struct scoreboard *sb,
 856                               struct origin *target,
 857                               struct commit *parent,
 858                               struct origin *porigin,
 859                               int opt)
 860{
 861        struct diff_options diff_opts;
 862        const char *paths[1];
 863        int i, j;
 864        int retval;
 865        struct blame_list *blame_list;
 866        int num_ents;
 867
 868        blame_list = setup_blame_list(sb, target, &num_ents);
 869        if (!blame_list)
 870                return 1; /* nothing remains for this target */
 871
 872        diff_setup(&diff_opts);
 873        diff_opts.recursive = 1;
 874        diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 875
 876        paths[0] = NULL;
 877        diff_tree_setup_paths(paths, &diff_opts);
 878        if (diff_setup_done(&diff_opts) < 0)
 879                die("diff-setup");
 880
 881        /* Try "find copies harder" on new path if requested;
 882         * we do not want to use diffcore_rename() actually to
 883         * match things up; find_copies_harder is set only to
 884         * force diff_tree_sha1() to feed all filepairs to diff_queue,
 885         * and this code needs to be after diff_setup_done(), which
 886         * usually makes find-copies-harder imply copy detection.
 887         */
 888        if ((opt & PICKAXE_BLAME_COPY_HARDER) &&
 889            (!porigin || strcmp(target->path, porigin->path)))
 890                diff_opts.find_copies_harder = 1;
 891
 892        diff_tree_sha1(parent->tree->object.sha1,
 893                       target->commit->tree->object.sha1,
 894                       "", &diff_opts);
 895
 896        if (!diff_opts.find_copies_harder)
 897                diffcore_std(&diff_opts);
 898
 899        retval = 0;
 900        while (1) {
 901                int made_progress = 0;
 902
 903                for (i = 0; i < diff_queued_diff.nr; i++) {
 904                        struct diff_filepair *p = diff_queued_diff.queue[i];
 905                        struct origin *norigin;
 906                        mmfile_t file_p;
 907                        struct blame_entry this[3];
 908
 909                        if (!DIFF_FILE_VALID(p->one))
 910                                continue; /* does not exist in parent */
 911                        if (porigin && !strcmp(p->one->path, porigin->path))
 912                                /* find_move already dealt with this path */
 913                                continue;
 914
 915                        norigin = get_origin(sb, parent, p->one->path);
 916                        hashcpy(norigin->blob_sha1, p->one->sha1);
 917                        fill_origin_blob(norigin, &file_p);
 918                        if (!file_p.ptr)
 919                                continue;
 920
 921                        for (j = 0; j < num_ents; j++) {
 922                                find_copy_in_blob(sb, blame_list[j].ent,
 923                                                  norigin, this, &file_p);
 924                                copy_split_if_better(sb, blame_list[j].split,
 925                                                     this);
 926                                decref_split(this);
 927                        }
 928                        origin_decref(norigin);
 929                }
 930
 931                for (j = 0; j < num_ents; j++) {
 932                        struct blame_entry *split = blame_list[j].split;
 933                        if (split[1].suspect &&
 934                            blame_copy_score < ent_score(sb, &split[1])) {
 935                                split_blame(sb, split, blame_list[j].ent);
 936                                made_progress = 1;
 937                        }
 938                        decref_split(split);
 939                }
 940                free(blame_list);
 941
 942                if (!made_progress)
 943                        break;
 944                blame_list = setup_blame_list(sb, target, &num_ents);
 945                if (!blame_list) {
 946                        retval = 1;
 947                        break;
 948                }
 949        }
 950        diff_flush(&diff_opts);
 951
 952        return retval;
 953}
 954
 955/* The blobs of origin and porigin exactly match, so everything
 956 * origin is suspected for can be blamed on the parent.
 957 */
 958static void pass_whole_blame(struct scoreboard *sb,
 959                             struct origin *origin, struct origin *porigin)
 960{
 961        struct blame_entry *e;
 962
 963        if (!porigin->file.ptr && origin->file.ptr) {
 964                /* Steal its file */
 965                porigin->file = origin->file;
 966                origin->file.ptr = NULL;
 967        }
 968        for (e = sb->ent; e; e = e->next) {
 969                if (cmp_suspect(e->suspect, origin))
 970                        continue;
 971                origin_incref(porigin);
 972                origin_decref(e->suspect);
 973                e->suspect = porigin;
 974        }
 975}
 976
 977#define MAXPARENT 16
 978
 979static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
 980{
 981        int i, pass;
 982        struct commit *commit = origin->commit;
 983        struct commit_list *parent;
 984        struct origin *parent_origin[MAXPARENT], *porigin;
 985
 986        memset(parent_origin, 0, sizeof(parent_origin));
 987
 988        /* The first pass looks for unrenamed path to optimize for
 989         * common cases, then we look for renames in the second pass.
 990         */
 991        for (pass = 0; pass < 2; pass++) {
 992                struct origin *(*find)(struct scoreboard *,
 993                                       struct commit *, struct origin *);
 994                find = pass ? find_rename : find_origin;
 995
 996                for (i = 0, parent = commit->parents;
 997                     i < MAXPARENT && parent;
 998                     parent = parent->next, i++) {
 999                        struct commit *p = parent->item;
1000                        int j, same;
1001
1002                        if (parent_origin[i])
1003                                continue;
1004                        if (parse_commit(p))
1005                                continue;
1006                        porigin = find(sb, p, origin);
1007                        if (!porigin)
1008                                continue;
1009                        if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1010                                pass_whole_blame(sb, origin, porigin);
1011                                origin_decref(porigin);
1012                                goto finish;
1013                        }
1014                        for (j = same = 0; j < i; j++)
1015                                if (parent_origin[j] &&
1016                                    !hashcmp(parent_origin[j]->blob_sha1,
1017                                             porigin->blob_sha1)) {
1018                                        same = 1;
1019                                        break;
1020                                }
1021                        if (!same)
1022                                parent_origin[i] = porigin;
1023                        else
1024                                origin_decref(porigin);
1025                }
1026        }
1027
1028        num_commits++;
1029        for (i = 0, parent = commit->parents;
1030             i < MAXPARENT && parent;
1031             parent = parent->next, i++) {
1032                struct origin *porigin = parent_origin[i];
1033                if (!porigin)
1034                        continue;
1035                if (pass_blame_to_parent(sb, origin, porigin))
1036                        goto finish;
1037        }
1038
1039        /*
1040         * Optionally run "miff" to find moves in parents' files here.
1041         */
1042        if (opt & PICKAXE_BLAME_MOVE)
1043                for (i = 0, parent = commit->parents;
1044                     i < MAXPARENT && parent;
1045                     parent = parent->next, i++) {
1046                        struct origin *porigin = parent_origin[i];
1047                        if (!porigin)
1048                                continue;
1049                        if (find_move_in_parent(sb, origin, porigin))
1050                                goto finish;
1051                }
1052
1053        /*
1054         * Optionally run "ciff" to find copies from parents' files here.
1055         */
1056        if (opt & PICKAXE_BLAME_COPY)
1057                for (i = 0, parent = commit->parents;
1058                     i < MAXPARENT && parent;
1059                     parent = parent->next, i++) {
1060                        struct origin *porigin = parent_origin[i];
1061                        if (find_copy_in_parent(sb, origin, parent->item,
1062                                                porigin, opt))
1063                                goto finish;
1064                }
1065
1066 finish:
1067        for (i = 0; i < MAXPARENT; i++)
1068                origin_decref(parent_origin[i]);
1069}
1070
1071static void assign_blame(struct scoreboard *sb, struct rev_info *revs, int opt)
1072{
1073        while (1) {
1074                struct blame_entry *ent;
1075                struct commit *commit;
1076                struct origin *suspect = NULL;
1077
1078                /* find one suspect to break down */
1079                for (ent = sb->ent; !suspect && ent; ent = ent->next)
1080                        if (!ent->guilty)
1081                                suspect = ent->suspect;
1082                if (!suspect)
1083                        return; /* all done */
1084
1085                origin_incref(suspect);
1086                commit = suspect->commit;
1087                if (!commit->object.parsed)
1088                        parse_commit(commit);
1089                if (!(commit->object.flags & UNINTERESTING) &&
1090                    !(revs->max_age != -1 && commit->date  < revs->max_age))
1091                        pass_blame(sb, suspect, opt);
1092
1093                /* Take responsibility for the remaining entries */
1094                for (ent = sb->ent; ent; ent = ent->next)
1095                        if (!cmp_suspect(ent->suspect, suspect))
1096                                ent->guilty = 1;
1097                origin_decref(suspect);
1098
1099                if (DEBUG) /* sanity */
1100                        sanity_check_refcnt(sb);
1101        }
1102}
1103
1104static const char *format_time(unsigned long time, const char *tz_str,
1105                               int show_raw_time)
1106{
1107        static char time_buf[128];
1108        time_t t = time;
1109        int minutes, tz;
1110        struct tm *tm;
1111
1112        if (show_raw_time) {
1113                sprintf(time_buf, "%lu %s", time, tz_str);
1114                return time_buf;
1115        }
1116
1117        tz = atoi(tz_str);
1118        minutes = tz < 0 ? -tz : tz;
1119        minutes = (minutes / 100)*60 + (minutes % 100);
1120        minutes = tz < 0 ? -minutes : minutes;
1121        t = time + minutes * 60;
1122        tm = gmtime(&t);
1123
1124        strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1125        strcat(time_buf, tz_str);
1126        return time_buf;
1127}
1128
1129struct commit_info
1130{
1131        char *author;
1132        char *author_mail;
1133        unsigned long author_time;
1134        char *author_tz;
1135
1136        /* filled only when asked for details */
1137        char *committer;
1138        char *committer_mail;
1139        unsigned long committer_time;
1140        char *committer_tz;
1141
1142        char *summary;
1143};
1144
1145static void get_ac_line(const char *inbuf, const char *what,
1146                        int bufsz, char *person, char **mail,
1147                        unsigned long *time, char **tz)
1148{
1149        int len;
1150        char *tmp, *endp;
1151
1152        tmp = strstr(inbuf, what);
1153        if (!tmp)
1154                goto error_out;
1155        tmp += strlen(what);
1156        endp = strchr(tmp, '\n');
1157        if (!endp)
1158                len = strlen(tmp);
1159        else
1160                len = endp - tmp;
1161        if (bufsz <= len) {
1162        error_out:
1163                /* Ugh */
1164                person = *mail = *tz = "(unknown)";
1165                *time = 0;
1166                return;
1167        }
1168        memcpy(person, tmp, len);
1169
1170        tmp = person;
1171        tmp += len;
1172        *tmp = 0;
1173        while (*tmp != ' ')
1174                tmp--;
1175        *tz = tmp+1;
1176
1177        *tmp = 0;
1178        while (*tmp != ' ')
1179                tmp--;
1180        *time = strtoul(tmp, NULL, 10);
1181
1182        *tmp = 0;
1183        while (*tmp != ' ')
1184                tmp--;
1185        *mail = tmp + 1;
1186        *tmp = 0;
1187}
1188
1189static void get_commit_info(struct commit *commit,
1190                            struct commit_info *ret,
1191                            int detailed)
1192{
1193        int len;
1194        char *tmp, *endp;
1195        static char author_buf[1024];
1196        static char committer_buf[1024];
1197        static char summary_buf[1024];
1198
1199        /* We've operated without save_commit_buffer, so
1200         * we now need to populate them for output.
1201         */
1202        if (!commit->buffer) {
1203                char type[20];
1204                unsigned long size;
1205                commit->buffer =
1206                        read_sha1_file(commit->object.sha1, type, &size);
1207        }
1208        ret->author = author_buf;
1209        get_ac_line(commit->buffer, "\nauthor ",
1210                    sizeof(author_buf), author_buf, &ret->author_mail,
1211                    &ret->author_time, &ret->author_tz);
1212
1213        if (!detailed)
1214                return;
1215
1216        ret->committer = committer_buf;
1217        get_ac_line(commit->buffer, "\ncommitter ",
1218                    sizeof(committer_buf), committer_buf, &ret->committer_mail,
1219                    &ret->committer_time, &ret->committer_tz);
1220
1221        ret->summary = summary_buf;
1222        tmp = strstr(commit->buffer, "\n\n");
1223        if (!tmp) {
1224        error_out:
1225                sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1226                return;
1227        }
1228        tmp += 2;
1229        endp = strchr(tmp, '\n');
1230        if (!endp)
1231                goto error_out;
1232        len = endp - tmp;
1233        if (len >= sizeof(summary_buf))
1234                goto error_out;
1235        memcpy(summary_buf, tmp, len);
1236        summary_buf[len] = 0;
1237}
1238
1239#define OUTPUT_ANNOTATE_COMPAT  001
1240#define OUTPUT_LONG_OBJECT_NAME 002
1241#define OUTPUT_RAW_TIMESTAMP    004
1242#define OUTPUT_PORCELAIN        010
1243#define OUTPUT_SHOW_NAME        020
1244#define OUTPUT_SHOW_NUMBER      040
1245#define OUTPUT_SHOW_SCORE      0100
1246
1247static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1248{
1249        int cnt;
1250        const char *cp;
1251        struct origin *suspect = ent->suspect;
1252        char hex[41];
1253
1254        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1255        printf("%s%c%d %d %d\n",
1256               hex,
1257               ent->guilty ? ' ' : '*', // purely for debugging
1258               ent->s_lno + 1,
1259               ent->lno + 1,
1260               ent->num_lines);
1261        if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1262                struct commit_info ci;
1263                suspect->commit->object.flags |= METAINFO_SHOWN;
1264                get_commit_info(suspect->commit, &ci, 1);
1265                printf("author %s\n", ci.author);
1266                printf("author-mail %s\n", ci.author_mail);
1267                printf("author-time %lu\n", ci.author_time);
1268                printf("author-tz %s\n", ci.author_tz);
1269                printf("committer %s\n", ci.committer);
1270                printf("committer-mail %s\n", ci.committer_mail);
1271                printf("committer-time %lu\n", ci.committer_time);
1272                printf("committer-tz %s\n", ci.committer_tz);
1273                printf("filename %s\n", suspect->path);
1274                printf("summary %s\n", ci.summary);
1275        }
1276        else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1277                printf("filename %s\n", suspect->path);
1278
1279        cp = nth_line(sb, ent->lno);
1280        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1281                char ch;
1282                if (cnt)
1283                        printf("%s %d %d\n", hex,
1284                               ent->s_lno + 1 + cnt,
1285                               ent->lno + 1 + cnt);
1286                putchar('\t');
1287                do {
1288                        ch = *cp++;
1289                        putchar(ch);
1290                } while (ch != '\n' &&
1291                         cp < sb->final_buf + sb->final_buf_size);
1292        }
1293}
1294
1295static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1296{
1297        int cnt;
1298        const char *cp;
1299        struct origin *suspect = ent->suspect;
1300        struct commit_info ci;
1301        char hex[41];
1302        int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1303
1304        get_commit_info(suspect->commit, &ci, 1);
1305        strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1306
1307        cp = nth_line(sb, ent->lno);
1308        for (cnt = 0; cnt < ent->num_lines; cnt++) {
1309                char ch;
1310
1311                printf("%.*s", (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8, hex);
1312                if (opt & OUTPUT_ANNOTATE_COMPAT)
1313                        printf("\t(%10s\t%10s\t%d)", ci.author,
1314                               format_time(ci.author_time, ci.author_tz,
1315                                           show_raw_time),
1316                               ent->lno + 1 + cnt);
1317                else {
1318                        if (opt & OUTPUT_SHOW_SCORE)
1319                                printf(" %*d %02d",
1320                                       max_score_digits, ent->score,
1321                                       ent->suspect->refcnt);
1322                        if (opt & OUTPUT_SHOW_NAME)
1323                                printf(" %-*.*s", longest_file, longest_file,
1324                                       suspect->path);
1325                        if (opt & OUTPUT_SHOW_NUMBER)
1326                                printf(" %*d", max_orig_digits,
1327                                       ent->s_lno + 1 + cnt);
1328                        printf(" (%-*.*s %10s %*d) ",
1329                               longest_author, longest_author, ci.author,
1330                               format_time(ci.author_time, ci.author_tz,
1331                                           show_raw_time),
1332                               max_digits, ent->lno + 1 + cnt);
1333                }
1334                do {
1335                        ch = *cp++;
1336                        putchar(ch);
1337                } while (ch != '\n' &&
1338                         cp < sb->final_buf + sb->final_buf_size);
1339        }
1340}
1341
1342static void output(struct scoreboard *sb, int option)
1343{
1344        struct blame_entry *ent;
1345
1346        if (option & OUTPUT_PORCELAIN) {
1347                for (ent = sb->ent; ent; ent = ent->next) {
1348                        struct blame_entry *oth;
1349                        struct origin *suspect = ent->suspect;
1350                        struct commit *commit = suspect->commit;
1351                        if (commit->object.flags & MORE_THAN_ONE_PATH)
1352                                continue;
1353                        for (oth = ent->next; oth; oth = oth->next) {
1354                                if ((oth->suspect->commit != commit) ||
1355                                    !strcmp(oth->suspect->path, suspect->path))
1356                                        continue;
1357                                commit->object.flags |= MORE_THAN_ONE_PATH;
1358                                break;
1359                        }
1360                }
1361        }
1362
1363        for (ent = sb->ent; ent; ent = ent->next) {
1364                if (option & OUTPUT_PORCELAIN)
1365                        emit_porcelain(sb, ent);
1366                else {
1367                        emit_other(sb, ent, option);
1368                }
1369        }
1370}
1371
1372static int prepare_lines(struct scoreboard *sb)
1373{
1374        const char *buf = sb->final_buf;
1375        unsigned long len = sb->final_buf_size;
1376        int num = 0, incomplete = 0, bol = 1;
1377
1378        if (len && buf[len-1] != '\n')
1379                incomplete++; /* incomplete line at the end */
1380        while (len--) {
1381                if (bol) {
1382                        sb->lineno = xrealloc(sb->lineno,
1383                                              sizeof(int* ) * (num + 1));
1384                        sb->lineno[num] = buf - sb->final_buf;
1385                        bol = 0;
1386                }
1387                if (*buf++ == '\n') {
1388                        num++;
1389                        bol = 1;
1390                }
1391        }
1392        sb->lineno = xrealloc(sb->lineno,
1393                              sizeof(int* ) * (num + incomplete + 1));
1394        sb->lineno[num + incomplete] = buf - sb->final_buf;
1395        sb->num_lines = num + incomplete;
1396        return sb->num_lines;
1397}
1398
1399static int read_ancestry(const char *graft_file)
1400{
1401        FILE *fp = fopen(graft_file, "r");
1402        char buf[1024];
1403        if (!fp)
1404                return -1;
1405        while (fgets(buf, sizeof(buf), fp)) {
1406                /* The format is just "Commit Parent1 Parent2 ...\n" */
1407                int len = strlen(buf);
1408                struct commit_graft *graft = read_graft_line(buf, len);
1409                register_commit_graft(graft, 0);
1410        }
1411        fclose(fp);
1412        return 0;
1413}
1414
1415static int lineno_width(int lines)
1416{
1417        int i, width;
1418
1419        for (width = 1, i = 10; i <= lines + 1; width++)
1420                i *= 10;
1421        return width;
1422}
1423
1424static void find_alignment(struct scoreboard *sb, int *option)
1425{
1426        int longest_src_lines = 0;
1427        int longest_dst_lines = 0;
1428        unsigned largest_score = 0;
1429        struct blame_entry *e;
1430
1431        for (e = sb->ent; e; e = e->next) {
1432                struct origin *suspect = e->suspect;
1433                struct commit_info ci;
1434                int num;
1435
1436                if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1437                        suspect->commit->object.flags |= METAINFO_SHOWN;
1438                        get_commit_info(suspect->commit, &ci, 1);
1439                        if (strcmp(suspect->path, sb->path))
1440                                *option |= OUTPUT_SHOW_NAME;
1441                        num = strlen(suspect->path);
1442                        if (longest_file < num)
1443                                longest_file = num;
1444                        num = strlen(ci.author);
1445                        if (longest_author < num)
1446                                longest_author = num;
1447                }
1448                num = e->s_lno + e->num_lines;
1449                if (longest_src_lines < num)
1450                        longest_src_lines = num;
1451                num = e->lno + e->num_lines;
1452                if (longest_dst_lines < num)
1453                        longest_dst_lines = num;
1454                if (largest_score < ent_score(sb, e))
1455                        largest_score = ent_score(sb, e);
1456        }
1457        max_orig_digits = lineno_width(longest_src_lines);
1458        max_digits = lineno_width(longest_dst_lines);
1459        max_score_digits = lineno_width(largest_score);
1460}
1461
1462static void sanity_check_refcnt(struct scoreboard *sb)
1463{
1464        int baa = 0;
1465        struct blame_entry *ent;
1466
1467        for (ent = sb->ent; ent; ent = ent->next) {
1468                /* Nobody should have zero or negative refcnt */
1469                if (ent->suspect->refcnt <= 0) {
1470                        fprintf(stderr, "%s in %s has negative refcnt %d\n",
1471                                ent->suspect->path,
1472                                sha1_to_hex(ent->suspect->commit->object.sha1),
1473                                ent->suspect->refcnt);
1474                        baa = 1;
1475                }
1476        }
1477        for (ent = sb->ent; ent; ent = ent->next) {
1478                /* Mark the ones that haven't been checked */
1479                if (0 < ent->suspect->refcnt)
1480                        ent->suspect->refcnt = -ent->suspect->refcnt;
1481        }
1482        for (ent = sb->ent; ent; ent = ent->next) {
1483                /* then pick each and see if they have the the correct
1484                 * refcnt.
1485                 */
1486                int found;
1487                struct blame_entry *e;
1488                struct origin *suspect = ent->suspect;
1489
1490                if (0 < suspect->refcnt)
1491                        continue;
1492                suspect->refcnt = -suspect->refcnt; /* Unmark */
1493                for (found = 0, e = sb->ent; e; e = e->next) {
1494                        if (e->suspect != suspect)
1495                                continue;
1496                        found++;
1497                }
1498                if (suspect->refcnt != found) {
1499                        fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1500                                ent->suspect->path,
1501                                sha1_to_hex(ent->suspect->commit->object.sha1),
1502                                ent->suspect->refcnt, found);
1503                        baa = 2;
1504                }
1505        }
1506        if (baa) {
1507                int opt = 0160;
1508                find_alignment(sb, &opt);
1509                output(sb, opt);
1510                die("Baa %d!", baa);
1511        }
1512}
1513
1514static int has_path_in_work_tree(const char *path)
1515{
1516        struct stat st;
1517        return !lstat(path, &st);
1518}
1519
1520static unsigned parse_score(const char *arg)
1521{
1522        char *end;
1523        unsigned long score = strtoul(arg, &end, 10);
1524        if (*end)
1525                return 0;
1526        return score;
1527}
1528
1529static const char *add_prefix(const char *prefix, const char *path)
1530{
1531        if (!prefix || !prefix[0])
1532                return path;
1533        return prefix_path(prefix, strlen(prefix), path);
1534}
1535
1536int cmd_pickaxe(int argc, const char **argv, const char *prefix)
1537{
1538        struct rev_info revs;
1539        const char *path;
1540        struct scoreboard sb;
1541        struct origin *o;
1542        struct blame_entry *ent;
1543        int i, seen_dashdash, unk, opt;
1544        long bottom, top, lno;
1545        int output_option = 0;
1546        const char *revs_file = NULL;
1547        const char *final_commit_name = NULL;
1548        char type[10];
1549
1550        save_commit_buffer = 0;
1551
1552        opt = 0;
1553        bottom = top = 0;
1554        seen_dashdash = 0;
1555        for (unk = i = 1; i < argc; i++) {
1556                const char *arg = argv[i];
1557                if (*arg != '-')
1558                        break;
1559                else if (!strcmp("-c", arg))
1560                        output_option |= OUTPUT_ANNOTATE_COMPAT;
1561                else if (!strcmp("-t", arg))
1562                        output_option |= OUTPUT_RAW_TIMESTAMP;
1563                else if (!strcmp("-l", arg))
1564                        output_option |= OUTPUT_LONG_OBJECT_NAME;
1565                else if (!strcmp("-S", arg) && ++i < argc)
1566                        revs_file = argv[i];
1567                else if (!strncmp("-M", arg, 2)) {
1568                        opt |= PICKAXE_BLAME_MOVE;
1569                        blame_move_score = parse_score(arg+2);
1570                }
1571                else if (!strncmp("-C", arg, 2)) {
1572                        if (opt & PICKAXE_BLAME_COPY)
1573                                opt |= PICKAXE_BLAME_COPY_HARDER;
1574                        opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
1575                        blame_copy_score = parse_score(arg+2);
1576                }
1577                else if (!strncmp("-L", arg, 2)) {
1578                        char *term;
1579                        if (!arg[2]) {
1580                                if (++i >= argc)
1581                                        usage(pickaxe_usage);
1582                                arg = argv[i];
1583                        }
1584                        else
1585                                arg += 2;
1586                        if (bottom || top)
1587                                die("More than one '-L n,m' option given");
1588                        bottom = strtol(arg, &term, 10);
1589                        if (*term == ',') {
1590                                top = strtol(term + 1, &term, 10);
1591                                if (*term)
1592                                        usage(pickaxe_usage);
1593                        }
1594                        if (bottom && top && top < bottom) {
1595                                unsigned long tmp;
1596                                tmp = top; top = bottom; bottom = tmp;
1597                        }
1598                }
1599                else if (!strcmp("--score-debug", arg))
1600                        output_option |= OUTPUT_SHOW_SCORE;
1601                else if (!strcmp("-f", arg) ||
1602                         !strcmp("--show-name", arg))
1603                        output_option |= OUTPUT_SHOW_NAME;
1604                else if (!strcmp("-n", arg) ||
1605                         !strcmp("--show-number", arg))
1606                        output_option |= OUTPUT_SHOW_NUMBER;
1607                else if (!strcmp("-p", arg) ||
1608                         !strcmp("--porcelain", arg))
1609                        output_option |= OUTPUT_PORCELAIN;
1610                else if (!strcmp("--", arg)) {
1611                        seen_dashdash = 1;
1612                        i++;
1613                        break;
1614                }
1615                else
1616                        argv[unk++] = arg;
1617        }
1618
1619        if (!blame_move_score)
1620                blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
1621        if (!blame_copy_score)
1622                blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
1623
1624        /* We have collected options unknown to us in argv[1..unk]
1625         * which are to be passed to revision machinery if we are
1626         * going to do the "bottom" procesing.
1627         *
1628         * The remaining are:
1629         *
1630         * (1) if seen_dashdash, its either
1631         *     "-options -- <path>" or
1632         *     "-options -- <path> <rev>".
1633         *     but the latter is allowed only if there is no
1634         *     options that we passed to revision machinery.
1635         *
1636         * (2) otherwise, we may have "--" somewhere later and
1637         *     might be looking at the first one of multiple 'rev'
1638         *     parameters (e.g. " master ^next ^maint -- path").
1639         *     See if there is a dashdash first, and give the
1640         *     arguments before that to revision machinery.
1641         *     After that there must be one 'path'.
1642         *
1643         * (3) otherwise, its one of the three:
1644         *     "-options <path> <rev>"
1645         *     "-options <rev> <path>"
1646         *     "-options <path>"
1647         *     but again the first one is allowed only if
1648         *     there is no options that we passed to revision
1649         *     machinery.
1650         */
1651
1652        if (seen_dashdash) {
1653                /* (1) */
1654                if (argc <= i)
1655                        usage(pickaxe_usage);
1656                path = add_prefix(prefix, argv[i]);
1657                if (i + 1 == argc - 1) {
1658                        if (unk != 1)
1659                                usage(pickaxe_usage);
1660                        argv[unk++] = argv[i + 1];
1661                }
1662                else if (i + 1 != argc)
1663                        /* garbage at end */
1664                        usage(pickaxe_usage);
1665        }
1666        else {
1667                int j;
1668                for (j = i; !seen_dashdash && j < argc; j++)
1669                        if (!strcmp(argv[j], "--"))
1670                                seen_dashdash = j;
1671                if (seen_dashdash) {
1672                        if (seen_dashdash + 1 != argc - 1)
1673                                usage(pickaxe_usage);
1674                        path = add_prefix(prefix, argv[seen_dashdash + 1]);
1675                        for (j = i; j < seen_dashdash; j++)
1676                                argv[unk++] = argv[j];
1677                }
1678                else {
1679                        /* (3) */
1680                        path = add_prefix(prefix, argv[i]);
1681                        if (i + 1 == argc - 1) {
1682                                final_commit_name = argv[i + 1];
1683
1684                                /* if (unk == 1) we could be getting
1685                                 * old-style
1686                                 */
1687                                if (unk == 1 && !has_path_in_work_tree(path)) {
1688                                        path = add_prefix(prefix, argv[i + 1]);
1689                                        final_commit_name = argv[i];
1690                                }
1691                        }
1692                        else if (i != argc - 1)
1693                                usage(pickaxe_usage); /* garbage at end */
1694
1695                        if (!has_path_in_work_tree(path))
1696                                die("cannot stat path %s: %s",
1697                                    path, strerror(errno));
1698                }
1699        }
1700
1701        if (final_commit_name)
1702                argv[unk++] = final_commit_name;
1703
1704        /* Now we got rev and path.  We do not want the path pruning
1705         * but we may want "bottom" processing.
1706         */
1707        argv[unk] = NULL;
1708
1709        init_revisions(&revs, NULL);
1710        setup_revisions(unk, argv, &revs, "HEAD");
1711        memset(&sb, 0, sizeof(sb));
1712
1713        /* There must be one and only one positive commit in the
1714         * revs->pending array.
1715         */
1716        for (i = 0; i < revs.pending.nr; i++) {
1717                struct object *obj = revs.pending.objects[i].item;
1718                if (obj->flags & UNINTERESTING)
1719                        continue;
1720                while (obj->type == OBJ_TAG)
1721                        obj = deref_tag(obj, NULL, 0);
1722                if (obj->type != OBJ_COMMIT)
1723                        die("Non commit %s?",
1724                            revs.pending.objects[i].name);
1725                if (sb.final)
1726                        die("More than one commit to dig from %s and %s?",
1727                            revs.pending.objects[i].name,
1728                            final_commit_name);
1729                sb.final = (struct commit *) obj;
1730                final_commit_name = revs.pending.objects[i].name;
1731        }
1732
1733        if (!sb.final) {
1734                /* "--not A B -- path" without anything positive */
1735                unsigned char head_sha1[20];
1736
1737                final_commit_name = "HEAD";
1738                if (get_sha1(final_commit_name, head_sha1))
1739                        die("No such ref: HEAD");
1740                sb.final = lookup_commit_reference(head_sha1);
1741                add_pending_object(&revs, &(sb.final->object), "HEAD");
1742        }
1743
1744        /* If we have bottom, this will mark the ancestors of the
1745         * bottom commits we would reach while traversing as
1746         * uninteresting.
1747         */
1748        prepare_revision_walk(&revs);
1749
1750        o = get_origin(&sb, sb.final, path);
1751        if (fill_blob_sha1(o))
1752                die("no such path %s in %s", path, final_commit_name);
1753
1754        sb.final_buf = read_sha1_file(o->blob_sha1, type, &sb.final_buf_size);
1755        num_read_blob++;
1756        lno = prepare_lines(&sb);
1757
1758        if (bottom < 1)
1759                bottom = 1;
1760        if (top < 1)
1761                top = lno;
1762        bottom--;
1763        if (lno < top)
1764                die("file %s has only %lu lines", path, lno);
1765
1766        ent = xcalloc(1, sizeof(*ent));
1767        ent->lno = bottom;
1768        ent->num_lines = top - bottom;
1769        ent->suspect = o;
1770        ent->s_lno = bottom;
1771
1772        sb.ent = ent;
1773        sb.path = path;
1774
1775        if (revs_file && read_ancestry(revs_file))
1776                die("reading graft file %s failed: %s",
1777                    revs_file, strerror(errno));
1778
1779        assign_blame(&sb, &revs, opt);
1780
1781        coalesce(&sb);
1782
1783        if (!(output_option & OUTPUT_PORCELAIN))
1784                find_alignment(&sb, &output_option);
1785
1786        output(&sb, output_option);
1787        free((void *)sb.final_buf);
1788        for (ent = sb.ent; ent; ) {
1789                struct blame_entry *e = ent->next;
1790                free(ent);
1791                ent = e;
1792        }
1793
1794        if (DEBUG) {
1795                printf("num read blob: %d\n", num_read_blob);
1796                printf("num get patch: %d\n", num_get_patch);
1797                printf("num commits: %d\n", num_commits);
1798        }
1799        return 0;
1800}