builtin-blame.c

   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 "quote.h"
  17 #include "xdiff-interface.h"
  18 #include "cache-tree.h"
  19 #include "path-list.h"
  20 #include "mailmap.h"
  21
  22 static char blame_usage[] =
  23 "git-blame [-c] [-b] [-l] [--root] [-t] [-f] [-n] [-s] [-p] [-w] [-L n,m] [-S <revs-file>] [-M] [-C] [-C] [--contents <filename>] [--incremental] [commit] [--] file\n"
  24 "  -c                  Use the same output mode as git-annotate (Default: off)\n"
  25 "  -b                  Show blank SHA-1 for boundary commits (Default: off)\n"
  26 "  -l                  Show long commit SHA1 (Default: off)\n"
  27 "  --root              Do not treat root commits as boundaries (Default: off)\n"
  28 "  -t                  Show raw timestamp (Default: off)\n"
  29 "  -f, --show-name     Show original filename (Default: auto)\n"
  30 "  -n, --show-number   Show original linenumber (Default: off)\n"
  31 "  -s                  Suppress author name and timestamp (Default: off)\n"
  32 "  -p, --porcelain     Show in a format designed for machine consumption\n"
  33 "  -w                  Ignore whitespace differences\n"
  34 "  -L n,m              Process only line range n,m, counting from 1\n"
  35 "  -M, -C              Find line movements within and across files\n"
  36 "  --incremental       Show blame entries as we find them, incrementally\n"
  37 "  --contents file     Use <file>'s contents as the final image\n"
  38 "  -S revs-file        Use revisions from revs-file instead of calling git-rev-list\n";
  39
  40 static int longest_file;
  41 static int longest_author;
  42 static int max_orig_digits;
  43 static int max_digits;
  44 static int max_score_digits;
  45 static int show_root;
  46 static int blank_boundary;
  47 static int incremental;
  48 static int cmd_is_annotate;
  49 static int xdl_opts = XDF_NEED_MINIMAL;
  50 static struct path_list mailmap;
  51
  52 #ifndef DEBUG
  53 #define DEBUG 0
  54 #endif
  55
  56 /* stats */
  57 static int num_read_blob;
  58 static int num_get_patch;
  59 static int num_commits;
  60
  61 #define PICKAXE_BLAME_MOVE              01
  62 #define PICKAXE_BLAME_COPY              02
  63 #define PICKAXE_BLAME_COPY_HARDER       04
  64 #define PICKAXE_BLAME_COPY_HARDEST      010
  65
  66 /*
  67  * blame for a blame_entry with score lower than these thresholds
  68  * is not passed to the parent using move/copy logic.
  69  */
  70 static unsigned blame_move_score;
  71 static unsigned blame_copy_score;
  72 #define BLAME_DEFAULT_MOVE_SCORE        20
  73 #define BLAME_DEFAULT_COPY_SCORE        40
  74
  75 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
  76 #define METAINFO_SHOWN          (1u<<12)
  77 #define MORE_THAN_ONE_PATH      (1u<<13)
  78
  79 /*
  80  * One blob in a commit that is being suspected
  81  */
  82 struct origin {
  83         int refcnt;
  84         struct commit *commit;
  85         mmfile_t file;
  86         unsigned char blob_sha1[20];
  87         char path[FLEX_ARRAY];
  88 };
  89
  90 /*
  91  * Given an origin, prepare mmfile_t structure to be used by the
  92  * diff machinery
  93  */
  94 static char *fill_origin_blob(struct origin *o, mmfile_t *file)
  95 {
  96         if (!o->file.ptr) {
  97                 enum object_type type;
  98                 num_read_blob++;
  99                 file->ptr = read_sha1_file(o->blob_sha1, &type,
 100                                            (unsigned long *)(&(file->size)));
 101                 if (!file->ptr)
 102                         die("Cannot read blob %s for path %s",
 103                             sha1_to_hex(o->blob_sha1),
 104                             o->path);
 105                 o->file = *file;
 106         }
 107         else
 108                 *file = o->file;
 109         return file->ptr;
 110 }
 111
 112 /*
 113  * Origin is refcounted and usually we keep the blob contents to be
 114  * reused.
 115  */
 116 static inline struct origin *origin_incref(struct origin *o)
 117 {
 118         if (o)
 119                 o->refcnt++;
 120         return o;
 121 }
 122
 123 static void origin_decref(struct origin *o)
 124 {
 125         if (o && --o->refcnt <= 0) {
 126                 if (o->file.ptr)
 127                         free(o->file.ptr);
 128                 memset(o, 0, sizeof(*o));
 129                 free(o);
 130         }
 131 }
 132
 133 static void drop_origin_blob(struct origin *o)
 134 {
 135         if (o->file.ptr) {
 136                 free(o->file.ptr);
 137                 o->file.ptr = NULL;
 138         }
 139 }
 140
 141 /*
 142  * Each group of lines is described by a blame_entry; it can be split
 143  * as we pass blame to the parents.  They form a linked list in the
 144  * scoreboard structure, sorted by the target line number.
 145  */
 146 struct blame_entry {
 147         struct blame_entry *prev;
 148         struct blame_entry *next;
 149
 150         /* the first line of this group in the final image;
 151          * internally all line numbers are 0 based.
 152          */
 153         int lno;
 154
 155         /* how many lines this group has */
 156         int num_lines;
 157
 158         /* the commit that introduced this group into the final image */
 159         struct origin *suspect;
 160
 161         /* true if the suspect is truly guilty; false while we have not
 162          * checked if the group came from one of its parents.
 163          */
 164         char guilty;
 165
 166         /* the line number of the first line of this group in the
 167          * suspect's file; internally all line numbers are 0 based.
 168          */
 169         int s_lno;
 170
 171         /* how significant this entry is -- cached to avoid
 172          * scanning the lines over and over.
 173          */
 174         unsigned score;
 175 };
 176
 177 /*
 178  * The current state of the blame assignment.
 179  */
 180 struct scoreboard {
 181         /* the final commit (i.e. where we started digging from) */
 182         struct commit *final;
 183
 184         const char *path;
 185
 186         /*
 187          * The contents in the final image.
 188          * Used by many functions to obtain contents of the nth line,
 189          * indexed with scoreboard.lineno[blame_entry.lno].
 190          */
 191         const char *final_buf;
 192         unsigned long final_buf_size;
 193
 194         /* linked list of blames */
 195         struct blame_entry *ent;
 196
 197         /* look-up a line in the final buffer */
 198         int num_lines;
 199         int *lineno;
 200 };
 201
 202 static inline int same_suspect(struct origin *a, struct origin *b)
 203 {
 204         if (a == b)
 205                 return 1;
 206         if (a->commit != b->commit)
 207                 return 0;
 208         return !strcmp(a->path, b->path);
 209 }
 210
 211 static void sanity_check_refcnt(struct scoreboard *);
 212
 213 /*
 214  * If two blame entries that are next to each other came from
 215  * contiguous lines in the same origin (i.e. <commit, path> pair),
 216  * merge them together.
 217  */
 218 static void coalesce(struct scoreboard *sb)
 219 {
 220         struct blame_entry *ent, *next;
 221
 222         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 223                 if (same_suspect(ent->suspect, next->suspect) &&
 224                     ent->guilty == next->guilty &&
 225                     ent->s_lno + ent->num_lines == next->s_lno) {
 226                         ent->num_lines += next->num_lines;
 227                         ent->next = next->next;
 228                         if (ent->next)
 229                                 ent->next->prev = ent;
 230                         origin_decref(next->suspect);
 231                         free(next);
 232                         ent->score = 0;
 233                         next = ent; /* again */
 234                 }
 235         }
 236
 237         if (DEBUG) /* sanity */
 238                 sanity_check_refcnt(sb);
 239 }
 240
 241 /*
 242  * Given a commit and a path in it, create a new origin structure.
 243  * The callers that add blame to the scoreboard should use
 244  * get_origin() to obtain shared, refcounted copy instead of calling
 245  * this function directly.
 246  */
 247 static struct origin *make_origin(struct commit *commit, const char *path)
 248 {
 249         struct origin *o;
 250         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 251         o->commit = commit;
 252         o->refcnt = 1;
 253         strcpy(o->path, path);
 254         return o;
 255 }
 256
 257 /*
 258  * Locate an existing origin or create a new one.
 259  */
 260 static struct origin *get_origin(struct scoreboard *sb,
 261                                  struct commit *commit,
 262                                  const char *path)
 263 {
 264         struct blame_entry *e;
 265
 266         for (e = sb->ent; e; e = e->next) {
 267                 if (e->suspect->commit == commit &&
 268                     !strcmp(e->suspect->path, path))
 269                         return origin_incref(e->suspect);
 270         }
 271         return make_origin(commit, path);
 272 }
 273
 274 /*
 275  * Fill the blob_sha1 field of an origin if it hasn't, so that later
 276  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 277  * for an origin is also used to pass the blame for the entire file to
 278  * the parent to detect the case where a child's blob is identical to
 279  * that of its parent's.
 280  */
 281 static int fill_blob_sha1(struct origin *origin)
 282 {
 283         unsigned mode;
 284
 285         if (!is_null_sha1(origin->blob_sha1))
 286                 return 0;
 287         if (get_tree_entry(origin->commit->object.sha1,
 288                            origin->path,
 289                            origin->blob_sha1, &mode))
 290                 goto error_out;
 291         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 292                 goto error_out;
 293         return 0;
 294  error_out:
 295         hashclr(origin->blob_sha1);
 296         return -1;
 297 }
 298
 299 /*
 300  * We have an origin -- check if the same path exists in the
 301  * parent and return an origin structure to represent it.
 302  */
 303 static struct origin *find_origin(struct scoreboard *sb,
 304                                   struct commit *parent,
 305                                   struct origin *origin)
 306 {
 307         struct origin *porigin = NULL;
 308         struct diff_options diff_opts;
 309         const char *paths[2];
 310
 311         if (parent->util) {
 312                 /*
 313                  * Each commit object can cache one origin in that
 314                  * commit.  This is a freestanding copy of origin and
 315                  * not refcounted.
 316                  */
 317                 struct origin *cached = parent->util;
 318                 if (!strcmp(cached->path, origin->path)) {
 319                         /*
 320                          * The same path between origin and its parent
 321                          * without renaming -- the most common case.
 322                          */
 323                         porigin = get_origin(sb, parent, cached->path);
 324
 325                         /*
 326                          * If the origin was newly created (i.e. get_origin
 327                          * would call make_origin if none is found in the
 328                          * scoreboard), it does not know the blob_sha1,
 329                          * so copy it.  Otherwise porigin was in the
 330                          * scoreboard and already knows blob_sha1.
 331                          */
 332                         if (porigin->refcnt == 1)
 333                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
 334                         return porigin;
 335                 }
 336                 /* otherwise it was not very useful; free it */
 337                 free(parent->util);
 338                 parent->util = NULL;
 339         }
 340
 341         /* See if the origin->path is different between parent
 342          * and origin first.  Most of the time they are the
 343          * same and diff-tree is fairly efficient about this.
 344          */
 345         diff_setup(&diff_opts);
 346         DIFF_OPT_SET(&diff_opts, RECURSIVE);
 347         diff_opts.detect_rename = 0;
 348         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 349         paths[0] = origin->path;
 350         paths[1] = NULL;
 351
 352         diff_tree_setup_paths(paths, &diff_opts);
 353         if (diff_setup_done(&diff_opts) < 0)
 354                 die("diff-setup");
 355
 356         if (is_null_sha1(origin->commit->object.sha1))
 357                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
 358         else
 359                 diff_tree_sha1(parent->tree->object.sha1,
 360                                origin->commit->tree->object.sha1,
 361                                "", &diff_opts);
 362         diffcore_std(&diff_opts);
 363
 364         /* It is either one entry that says "modified", or "created",
 365          * or nothing.
 366          */
 367         if (!diff_queued_diff.nr) {
 368                 /* The path is the same as parent */
 369                 porigin = get_origin(sb, parent, origin->path);
 370                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
 371         }
 372         else if (diff_queued_diff.nr != 1)
 373                 die("internal error in blame::find_origin");
 374         else {
 375                 struct diff_filepair *p = diff_queued_diff.queue[0];
 376                 switch (p->status) {
 377                 default:
 378                         die("internal error in blame::find_origin (%c)",
 379                             p->status);
 380                 case 'M':
 381                         porigin = get_origin(sb, parent, origin->path);
 382                         hashcpy(porigin->blob_sha1, p->one->sha1);
 383                         break;
 384                 case 'A':
 385                 case 'T':
 386                         /* Did not exist in parent, or type changed */
 387                         break;
 388                 }
 389         }
 390         diff_flush(&diff_opts);
 391         diff_tree_release_paths(&diff_opts);
 392         if (porigin) {
 393                 /*
 394                  * Create a freestanding copy that is not part of
 395                  * the refcounted origin found in the scoreboard, and
 396                  * cache it in the commit.
 397                  */
 398                 struct origin *cached;
 399
 400                 cached = make_origin(porigin->commit, porigin->path);
 401                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
 402                 parent->util = cached;
 403         }
 404         return porigin;
 405 }
 406
 407 /*
 408  * We have an origin -- find the path that corresponds to it in its
 409  * parent and return an origin structure to represent it.
 410  */
 411 static struct origin *find_rename(struct scoreboard *sb,
 412                                   struct commit *parent,
 413                                   struct origin *origin)
 414 {
 415         struct origin *porigin = NULL;
 416         struct diff_options diff_opts;
 417         int i;
 418         const char *paths[2];
 419
 420         diff_setup(&diff_opts);
 421         DIFF_OPT_SET(&diff_opts, RECURSIVE);
 422         diff_opts.detect_rename = DIFF_DETECT_RENAME;
 423         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 424         diff_opts.single_follow = origin->path;
 425         paths[0] = NULL;
 426         diff_tree_setup_paths(paths, &diff_opts);
 427         if (diff_setup_done(&diff_opts) < 0)
 428                 die("diff-setup");
 429
 430         if (is_null_sha1(origin->commit->object.sha1))
 431                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
 432         else
 433                 diff_tree_sha1(parent->tree->object.sha1,
 434                                origin->commit->tree->object.sha1,
 435                                "", &diff_opts);
 436         diffcore_std(&diff_opts);
 437
 438         for (i = 0; i < diff_queued_diff.nr; i++) {
 439                 struct diff_filepair *p = diff_queued_diff.queue[i];
 440                 if ((p->status == 'R' || p->status == 'C') &&
 441                     !strcmp(p->two->path, origin->path)) {
 442                         porigin = get_origin(sb, parent, p->one->path);
 443                         hashcpy(porigin->blob_sha1, p->one->sha1);
 444                         break;
 445                 }
 446         }
 447         diff_flush(&diff_opts);
 448         diff_tree_release_paths(&diff_opts);
 449         return porigin;
 450 }
 451
 452 /*
 453  * Parsing of patch chunks...
 454  */
 455 struct chunk {
 456         /* line number in postimage; up to but not including this
 457          * line is the same as preimage
 458          */
 459         int same;
 460
 461         /* preimage line number after this chunk */
 462         int p_next;
 463
 464         /* postimage line number after this chunk */
 465         int t_next;
 466 };
 467
 468 struct patch {
 469         struct chunk *chunks;
 470         int num;
 471 };
 472
 473 struct blame_diff_state {
 474         struct xdiff_emit_state xm;
 475         struct patch *ret;
 476         unsigned hunk_post_context;
 477         unsigned hunk_in_pre_context : 1;
 478 };
 479
 480 static void process_u_diff(void *state_, char *line, unsigned long len)
 481 {
 482         struct blame_diff_state *state = state_;
 483         struct chunk *chunk;
 484         int off1, off2, len1, len2, num;
 485
 486         num = state->ret->num;
 487         if (len < 4 || line[0] != '@' || line[1] != '@') {
 488                 if (state->hunk_in_pre_context && line[0] == ' ')
 489                         state->ret->chunks[num - 1].same++;
 490                 else {
 491                         state->hunk_in_pre_context = 0;
 492                         if (line[0] == ' ')
 493                                 state->hunk_post_context++;
 494                         else
 495                                 state->hunk_post_context = 0;
 496                 }
 497                 return;
 498         }
 499
 500         if (num && state->hunk_post_context) {
 501                 chunk = &state->ret->chunks[num - 1];
 502                 chunk->p_next -= state->hunk_post_context;
 503                 chunk->t_next -= state->hunk_post_context;
 504         }
 505         state->ret->num = ++num;
 506         state->ret->chunks = xrealloc(state->ret->chunks,
 507                                       sizeof(struct chunk) * num);
 508         chunk = &state->ret->chunks[num - 1];
 509         if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
 510                 state->ret->num--;
 511                 return;
 512         }
 513
 514         /* Line numbers in patch output are one based. */
 515         off1--;
 516         off2--;
 517
 518         chunk->same = len2 ? off2 : (off2 + 1);
 519
 520         chunk->p_next = off1 + (len1 ? len1 : 1);
 521         chunk->t_next = chunk->same + len2;
 522         state->hunk_in_pre_context = 1;
 523         state->hunk_post_context = 0;
 524 }
 525
 526 static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
 527                                     int context)
 528 {
 529         struct blame_diff_state state;
 530         xpparam_t xpp;
 531         xdemitconf_t xecfg;
 532         xdemitcb_t ecb;
 533
 534         xpp.flags = xdl_opts;
 535         memset(&xecfg, 0, sizeof(xecfg));
 536         xecfg.ctxlen = context;
 537         ecb.outf = xdiff_outf;
 538         ecb.priv = &state;
 539         memset(&state, 0, sizeof(state));
 540         state.xm.consume = process_u_diff;
 541         state.ret = xmalloc(sizeof(struct patch));
 542         state.ret->chunks = NULL;
 543         state.ret->num = 0;
 544
 545         xdl_diff(file_p, file_o, &xpp, &xecfg, &ecb);
 546
 547         if (state.ret->num) {
 548                 struct chunk *chunk;
 549                 chunk = &state.ret->chunks[state.ret->num - 1];
 550                 chunk->p_next -= state.hunk_post_context;
 551                 chunk->t_next -= state.hunk_post_context;
 552         }
 553         return state.ret;
 554 }
 555
 556 /*
 557  * Run diff between two origins and grab the patch output, so that
 558  * we can pass blame for lines origin is currently suspected for
 559  * to its parent.
 560  */
 561 static struct patch *get_patch(struct origin *parent, struct origin *origin)
 562 {
 563         mmfile_t file_p, file_o;
 564         struct patch *patch;
 565
 566         fill_origin_blob(parent, &file_p);
 567         fill_origin_blob(origin, &file_o);
 568         if (!file_p.ptr || !file_o.ptr)
 569                 return NULL;
 570         patch = compare_buffer(&file_p, &file_o, 0);
 571         num_get_patch++;
 572         return patch;
 573 }
 574
 575 static void free_patch(struct patch *p)
 576 {
 577         free(p->chunks);
 578         free(p);
 579 }
 580
 581 /*
 582  * Link in a new blame entry to the scoreboard.  Entries that cover the
 583  * same line range have been removed from the scoreboard previously.
 584  */
 585 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 586 {
 587         struct blame_entry *ent, *prev = NULL;
 588
 589         origin_incref(e->suspect);
 590
 591         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 592                 prev = ent;
 593
 594         /* prev, if not NULL, is the last one that is below e */
 595         e->prev = prev;
 596         if (prev) {
 597                 e->next = prev->next;
 598                 prev->next = e;
 599         }
 600         else {
 601                 e->next = sb->ent;
 602                 sb->ent = e;
 603         }
 604         if (e->next)
 605                 e->next->prev = e;
 606 }
 607
 608 /*
 609  * src typically is on-stack; we want to copy the information in it to
 610  * an malloced blame_entry that is already on the linked list of the
 611  * scoreboard.  The origin of dst loses a refcnt while the origin of src
 612  * gains one.
 613  */
 614 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 615 {
 616         struct blame_entry *p, *n;
 617
 618         p = dst->prev;
 619         n = dst->next;
 620         origin_incref(src->suspect);
 621         origin_decref(dst->suspect);
 622         memcpy(dst, src, sizeof(*src));
 623         dst->prev = p;
 624         dst->next = n;
 625         dst->score = 0;
 626 }
 627
 628 static const char *nth_line(struct scoreboard *sb, int lno)
 629 {
 630         return sb->final_buf + sb->lineno[lno];
 631 }
 632
 633 /*
 634  * It is known that lines between tlno to same came from parent, and e
 635  * has an overlap with that range.  it also is known that parent's
 636  * line plno corresponds to e's line tlno.
 637  *
 638  *                <---- e ----->
 639  *                   <------>
 640  *                   <------------>
 641  *             <------------>
 642  *             <------------------>
 643  *
 644  * Split e into potentially three parts; before this chunk, the chunk
 645  * to be blamed for the parent, and after that portion.
 646  */
 647 static void split_overlap(struct blame_entry *split,
 648                           struct blame_entry *e,
 649                           int tlno, int plno, int same,
 650                           struct origin *parent)
 651 {
 652         int chunk_end_lno;
 653         memset(split, 0, sizeof(struct blame_entry [3]));
 654
 655         if (e->s_lno < tlno) {
 656                 /* there is a pre-chunk part not blamed on parent */
 657                 split[0].suspect = origin_incref(e->suspect);
 658                 split[0].lno = e->lno;
 659                 split[0].s_lno = e->s_lno;
 660                 split[0].num_lines = tlno - e->s_lno;
 661                 split[1].lno = e->lno + tlno - e->s_lno;
 662                 split[1].s_lno = plno;
 663         }
 664         else {
 665                 split[1].lno = e->lno;
 666                 split[1].s_lno = plno + (e->s_lno - tlno);
 667         }
 668
 669         if (same < e->s_lno + e->num_lines) {
 670                 /* there is a post-chunk part not blamed on parent */
 671                 split[2].suspect = origin_incref(e->suspect);
 672                 split[2].lno = e->lno + (same - e->s_lno);
 673                 split[2].s_lno = e->s_lno + (same - e->s_lno);
 674                 split[2].num_lines = e->s_lno + e->num_lines - same;
 675                 chunk_end_lno = split[2].lno;
 676         }
 677         else
 678                 chunk_end_lno = e->lno + e->num_lines;
 679         split[1].num_lines = chunk_end_lno - split[1].lno;
 680
 681         /*
 682          * if it turns out there is nothing to blame the parent for,
 683          * forget about the splitting.  !split[1].suspect signals this.
 684          */
 685         if (split[1].num_lines < 1)
 686                 return;
 687         split[1].suspect = origin_incref(parent);
 688 }
 689
 690 /*
 691  * split_overlap() divided an existing blame e into up to three parts
 692  * in split.  Adjust the linked list of blames in the scoreboard to
 693  * reflect the split.
 694  */
 695 static void split_blame(struct scoreboard *sb,
 696                         struct blame_entry *split,
 697                         struct blame_entry *e)
 698 {
 699         struct blame_entry *new_entry;
 700
 701         if (split[0].suspect && split[2].suspect) {
 702                 /* The first part (reuse storage for the existing entry e) */
 703                 dup_entry(e, &split[0]);
 704
 705                 /* The last part -- me */
 706                 new_entry = xmalloc(sizeof(*new_entry));
 707                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 708                 add_blame_entry(sb, new_entry);
 709
 710                 /* ... and the middle part -- parent */
 711                 new_entry = xmalloc(sizeof(*new_entry));
 712                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 713                 add_blame_entry(sb, new_entry);
 714         }
 715         else if (!split[0].suspect && !split[2].suspect)
 716                 /*
 717                  * The parent covers the entire area; reuse storage for
 718                  * e and replace it with the parent.
 719                  */
 720                 dup_entry(e, &split[1]);
 721         else if (split[0].suspect) {
 722                 /* me and then parent */
 723                 dup_entry(e, &split[0]);
 724
 725                 new_entry = xmalloc(sizeof(*new_entry));
 726                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 727                 add_blame_entry(sb, new_entry);
 728         }
 729         else {
 730                 /* parent and then me */
 731                 dup_entry(e, &split[1]);
 732
 733                 new_entry = xmalloc(sizeof(*new_entry));
 734                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 735                 add_blame_entry(sb, new_entry);
 736         }
 737
 738         if (DEBUG) { /* sanity */
 739                 struct blame_entry *ent;
 740                 int lno = sb->ent->lno, corrupt = 0;
 741
 742                 for (ent = sb->ent; ent; ent = ent->next) {
 743                         if (lno != ent->lno)
 744                                 corrupt = 1;
 745                         if (ent->s_lno < 0)
 746                                 corrupt = 1;
 747                         lno += ent->num_lines;
 748                 }
 749                 if (corrupt) {
 750                         lno = sb->ent->lno;
 751                         for (ent = sb->ent; ent; ent = ent->next) {
 752                                 printf("L %8d l %8d n %8d\n",
 753                                        lno, ent->lno, ent->num_lines);
 754                                 lno = ent->lno + ent->num_lines;
 755                         }
 756                         die("oops");
 757                 }
 758         }
 759 }
 760
 761 /*
 762  * After splitting the blame, the origins used by the
 763  * on-stack blame_entry should lose one refcnt each.
 764  */
 765 static void decref_split(struct blame_entry *split)
 766 {
 767         int i;
 768
 769         for (i = 0; i < 3; i++)
 770                 origin_decref(split[i].suspect);
 771 }
 772
 773 /*
 774  * Helper for blame_chunk().  blame_entry e is known to overlap with
 775  * the patch hunk; split it and pass blame to the parent.
 776  */
 777 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 778                           int tlno, int plno, int same,
 779                           struct origin *parent)
 780 {
 781         struct blame_entry split[3];
 782
 783         split_overlap(split, e, tlno, plno, same, parent);
 784         if (split[1].suspect)
 785                 split_blame(sb, split, e);
 786         decref_split(split);
 787 }
 788
 789 /*
 790  * Find the line number of the last line the target is suspected for.
 791  */
 792 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 793 {
 794         struct blame_entry *e;
 795         int last_in_target = -1;
 796
 797         for (e = sb->ent; e; e = e->next) {
 798                 if (e->guilty || !same_suspect(e->suspect, target))
 799                         continue;
 800                 if (last_in_target < e->s_lno + e->num_lines)
 801                         last_in_target = e->s_lno + e->num_lines;
 802         }
 803         return last_in_target;
 804 }
 805
 806 /*
 807  * Process one hunk from the patch between the current suspect for
 808  * blame_entry e and its parent.  Find and split the overlap, and
 809  * pass blame to the overlapping part to the parent.
 810  */
 811 static void blame_chunk(struct scoreboard *sb,
 812                         int tlno, int plno, int same,
 813                         struct origin *target, struct origin *parent)
 814 {
 815         struct blame_entry *e;
 816
 817         for (e = sb->ent; e; e = e->next) {
 818                 if (e->guilty || !same_suspect(e->suspect, target))
 819                         continue;
 820                 if (same <= e->s_lno)
 821                         continue;
 822                 if (tlno < e->s_lno + e->num_lines)
 823                         blame_overlap(sb, e, tlno, plno, same, parent);
 824         }
 825 }
 826
 827 /*
 828  * We are looking at the origin 'target' and aiming to pass blame
 829  * for the lines it is suspected to its parent.  Run diff to find
 830  * which lines came from parent and pass blame for them.
 831  */
 832 static int pass_blame_to_parent(struct scoreboard *sb,
 833                                 struct origin *target,
 834                                 struct origin *parent)
 835 {
 836         int i, last_in_target, plno, tlno;
 837         struct patch *patch;
 838
 839         last_in_target = find_last_in_target(sb, target);
 840         if (last_in_target < 0)
 841                 return 1; /* nothing remains for this target */
 842
 843         patch = get_patch(parent, target);
 844         plno = tlno = 0;
 845         for (i = 0; i < patch->num; i++) {
 846                 struct chunk *chunk = &patch->chunks[i];
 847
 848                 blame_chunk(sb, tlno, plno, chunk->same, target, parent);
 849                 plno = chunk->p_next;
 850                 tlno = chunk->t_next;
 851         }
 852         /* The rest (i.e. anything after tlno) are the same as the parent */
 853         blame_chunk(sb, tlno, plno, last_in_target, target, parent);
 854
 855         free_patch(patch);
 856         return 0;
 857 }
 858
 859 /*
 860  * The lines in blame_entry after splitting blames many times can become
 861  * very small and trivial, and at some point it becomes pointless to
 862  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 863  * ordinary C program, and it is not worth to say it was copied from
 864  * totally unrelated file in the parent.
 865  *
 866  * Compute how trivial the lines in the blame_entry are.
 867  */
 868 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 869 {
 870         unsigned score;
 871         const char *cp, *ep;
 872
 873         if (e->score)
 874                 return e->score;
 875
 876         score = 1;
 877         cp = nth_line(sb, e->lno);
 878         ep = nth_line(sb, e->lno + e->num_lines);
 879         while (cp < ep) {
 880                 unsigned ch = *((unsigned char *)cp);
 881                 if (isalnum(ch))
 882                         score++;
 883                 cp++;
 884         }
 885         e->score = score;
 886         return score;
 887 }
 888
 889 /*
 890  * best_so_far[] and this[] are both a split of an existing blame_entry
 891  * that passes blame to the parent.  Maintain best_so_far the best split
 892  * so far, by comparing this and best_so_far and copying this into
 893  * bst_so_far as needed.
 894  */
 895 static void copy_split_if_better(struct scoreboard *sb,
 896                                  struct blame_entry *best_so_far,
 897                                  struct blame_entry *this)
 898 {
 899         int i;
 900
 901         if (!this[1].suspect)
 902                 return;
 903         if (best_so_far[1].suspect) {
 904                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 905                         return;
 906         }
 907
 908         for (i = 0; i < 3; i++)
 909                 origin_incref(this[i].suspect);
 910         decref_split(best_so_far);
 911         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 912 }
 913
 914 /*
 915  * We are looking at a part of the final image represented by
 916  * ent (tlno and same are offset by ent->s_lno).
 917  * tlno is where we are looking at in the final image.
 918  * up to (but not including) same match preimage.
 919  * plno is where we are looking at in the preimage.
 920  *
 921  * <-------------- final image ---------------------->
 922  *       <------ent------>
 923  *         ^tlno ^same
 924  *    <---------preimage----->
 925  *         ^plno
 926  *
 927  * All line numbers are 0-based.
 928  */
 929 static void handle_split(struct scoreboard *sb,
 930                          struct blame_entry *ent,
 931                          int tlno, int plno, int same,
 932                          struct origin *parent,
 933                          struct blame_entry *split)
 934 {
 935         if (ent->num_lines <= tlno)
 936                 return;
 937         if (tlno < same) {
 938                 struct blame_entry this[3];
 939                 tlno += ent->s_lno;
 940                 same += ent->s_lno;
 941                 split_overlap(this, ent, tlno, plno, same, parent);
 942                 copy_split_if_better(sb, split, this);
 943                 decref_split(this);
 944         }
 945 }
 946
 947 /*
 948  * Find the lines from parent that are the same as ent so that
 949  * we can pass blames to it.  file_p has the blob contents for
 950  * the parent.
 951  */
 952 static void find_copy_in_blob(struct scoreboard *sb,
 953                               struct blame_entry *ent,
 954                               struct origin *parent,
 955                               struct blame_entry *split,
 956                               mmfile_t *file_p)
 957 {
 958         const char *cp;
 959         int cnt;
 960         mmfile_t file_o;
 961         struct patch *patch;
 962         int i, plno, tlno;
 963
 964         /*
 965          * Prepare mmfile that contains only the lines in ent.
 966          */
 967         cp = nth_line(sb, ent->lno);
 968         file_o.ptr = (char*) cp;
 969         cnt = ent->num_lines;
 970
 971         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
 972                 if (*cp++ == '\n')
 973                         cnt--;
 974         }
 975         file_o.size = cp - file_o.ptr;
 976
 977         patch = compare_buffer(file_p, &file_o, 1);
 978
 979         /*
 980          * file_o is a part of final image we are annotating.
 981          * file_p partially may match that image.
 982          */
 983         memset(split, 0, sizeof(struct blame_entry [3]));
 984         plno = tlno = 0;
 985         for (i = 0; i < patch->num; i++) {
 986                 struct chunk *chunk = &patch->chunks[i];
 987
 988                 handle_split(sb, ent, tlno, plno, chunk->same, parent, split);
 989                 plno = chunk->p_next;
 990                 tlno = chunk->t_next;
 991         }
 992         /* remainder, if any, all match the preimage */
 993         handle_split(sb, ent, tlno, plno, ent->num_lines, parent, split);
 994         free_patch(patch);
 995 }
 996
 997 /*
 998  * See if lines currently target is suspected for can be attributed to
 999  * parent.
1000  */
1001 static int find_move_in_parent(struct scoreboard *sb,
1002                                struct origin *target,
1003                                struct origin *parent)
1004 {
1005         int last_in_target, made_progress;
1006         struct blame_entry *e, split[3];
1007         mmfile_t file_p;
1008
1009         last_in_target = find_last_in_target(sb, target);
1010         if (last_in_target < 0)
1011                 return 1; /* nothing remains for this target */
1012
1013         fill_origin_blob(parent, &file_p);
1014         if (!file_p.ptr)
1015                 return 0;
1016
1017         made_progress = 1;
1018         while (made_progress) {
1019                 made_progress = 0;
1020                 for (e = sb->ent; e; e = e->next) {
1021                         if (e->guilty || !same_suspect(e->suspect, target))
1022                                 continue;
1023                         find_copy_in_blob(sb, e, parent, split, &file_p);
1024                         if (split[1].suspect &&
1025                             blame_move_score < ent_score(sb, &split[1])) {
1026                                 split_blame(sb, split, e);
1027                                 made_progress = 1;
1028                         }
1029                         decref_split(split);
1030                 }
1031         }
1032         return 0;
1033 }
1034
1035 struct blame_list {
1036         struct blame_entry *ent;
1037         struct blame_entry split[3];
1038 };
1039
1040 /*
1041  * Count the number of entries the target is suspected for,
1042  * and prepare a list of entry and the best split.
1043  */
1044 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1045                                            struct origin *target,
1046                                            int *num_ents_p)
1047 {
1048         struct blame_entry *e;
1049         int num_ents, i;
1050         struct blame_list *blame_list = NULL;
1051
1052         for (e = sb->ent, num_ents = 0; e; e = e->next)
1053                 if (!e->guilty && same_suspect(e->suspect, target))
1054                         num_ents++;
1055         if (num_ents) {
1056                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1057                 for (e = sb->ent, i = 0; e; e = e->next)
1058                         if (!e->guilty && same_suspect(e->suspect, target))
1059                                 blame_list[i++].ent = e;
1060         }
1061         *num_ents_p = num_ents;
1062         return blame_list;
1063 }
1064
1065 /*
1066  * For lines target is suspected for, see if we can find code movement
1067  * across file boundary from the parent commit.  porigin is the path
1068  * in the parent we already tried.
1069  */
1070 static int find_copy_in_parent(struct scoreboard *sb,
1071                                struct origin *target,
1072                                struct commit *parent,
1073                                struct origin *porigin,
1074                                int opt)
1075 {
1076         struct diff_options diff_opts;
1077         const char *paths[1];
1078         int i, j;
1079         int retval;
1080         struct blame_list *blame_list;
1081         int num_ents;
1082
1083         blame_list = setup_blame_list(sb, target, &num_ents);
1084         if (!blame_list)
1085                 return 1; /* nothing remains for this target */
1086
1087         diff_setup(&diff_opts);
1088         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1089         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1090
1091         paths[0] = NULL;
1092         diff_tree_setup_paths(paths, &diff_opts);
1093         if (diff_setup_done(&diff_opts) < 0)
1094                 die("diff-setup");
1095
1096         /* Try "find copies harder" on new path if requested;
1097          * we do not want to use diffcore_rename() actually to
1098          * match things up; find_copies_harder is set only to
1099          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1100          * and this code needs to be after diff_setup_done(), which
1101          * usually makes find-copies-harder imply copy detection.
1102          */
1103         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1104             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1105                 && (!porigin || strcmp(target->path, porigin->path))))
1106                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1107
1108         if (is_null_sha1(target->commit->object.sha1))
1109                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1110         else
1111                 diff_tree_sha1(parent->tree->object.sha1,
1112                                target->commit->tree->object.sha1,
1113                                "", &diff_opts);
1114
1115         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1116                 diffcore_std(&diff_opts);
1117
1118         retval = 0;
1119         while (1) {
1120                 int made_progress = 0;
1121
1122                 for (i = 0; i < diff_queued_diff.nr; i++) {
1123                         struct diff_filepair *p = diff_queued_diff.queue[i];
1124                         struct origin *norigin;
1125                         mmfile_t file_p;
1126                         struct blame_entry this[3];
1127
1128                         if (!DIFF_FILE_VALID(p->one))
1129                                 continue; /* does not exist in parent */
1130                         if (porigin && !strcmp(p->one->path, porigin->path))
1131                                 /* find_move already dealt with this path */
1132                                 continue;
1133
1134                         norigin = get_origin(sb, parent, p->one->path);
1135                         hashcpy(norigin->blob_sha1, p->one->sha1);
1136                         fill_origin_blob(norigin, &file_p);
1137                         if (!file_p.ptr)
1138                                 continue;
1139
1140                         for (j = 0; j < num_ents; j++) {
1141                                 find_copy_in_blob(sb, blame_list[j].ent,
1142                                                   norigin, this, &file_p);
1143                                 copy_split_if_better(sb, blame_list[j].split,
1144                                                      this);
1145                                 decref_split(this);
1146                         }
1147                         origin_decref(norigin);
1148                 }
1149
1150                 for (j = 0; j < num_ents; j++) {
1151                         struct blame_entry *split = blame_list[j].split;
1152                         if (split[1].suspect &&
1153                             blame_copy_score < ent_score(sb, &split[1])) {
1154                                 split_blame(sb, split, blame_list[j].ent);
1155                                 made_progress = 1;
1156                         }
1157                         decref_split(split);
1158                 }
1159                 free(blame_list);
1160
1161                 if (!made_progress)
1162                         break;
1163                 blame_list = setup_blame_list(sb, target, &num_ents);
1164                 if (!blame_list) {
1165                         retval = 1;
1166                         break;
1167                 }
1168         }
1169         diff_flush(&diff_opts);
1170         diff_tree_release_paths(&diff_opts);
1171         return retval;
1172 }
1173
1174 /*
1175  * The blobs of origin and porigin exactly match, so everything
1176  * origin is suspected for can be blamed on the parent.
1177  */
1178 static void pass_whole_blame(struct scoreboard *sb,
1179                              struct origin *origin, struct origin *porigin)
1180 {
1181         struct blame_entry *e;
1182
1183         if (!porigin->file.ptr && origin->file.ptr) {
1184                 /* Steal its file */
1185                 porigin->file = origin->file;
1186                 origin->file.ptr = NULL;
1187         }
1188         for (e = sb->ent; e; e = e->next) {
1189                 if (!same_suspect(e->suspect, origin))
1190                         continue;
1191                 origin_incref(porigin);
1192                 origin_decref(e->suspect);
1193                 e->suspect = porigin;
1194         }
1195 }
1196
1197 #define MAXPARENT 16
1198
1199 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1200 {
1201         int i, pass;
1202         struct commit *commit = origin->commit;
1203         struct commit_list *parent;
1204         struct origin *parent_origin[MAXPARENT], *porigin;
1205
1206         memset(parent_origin, 0, sizeof(parent_origin));
1207
1208         /* The first pass looks for unrenamed path to optimize for
1209          * common cases, then we look for renames in the second pass.
1210          */
1211         for (pass = 0; pass < 2; pass++) {
1212                 struct origin *(*find)(struct scoreboard *,
1213                                        struct commit *, struct origin *);
1214                 find = pass ? find_rename : find_origin;
1215
1216                 for (i = 0, parent = commit->parents;
1217                      i < MAXPARENT && parent;
1218                      parent = parent->next, i++) {
1219                         struct commit *p = parent->item;
1220                         int j, same;
1221
1222                         if (parent_origin[i])
1223                                 continue;
1224                         if (parse_commit(p))
1225                                 continue;
1226                         porigin = find(sb, p, origin);
1227                         if (!porigin)
1228                                 continue;
1229                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1230                                 pass_whole_blame(sb, origin, porigin);
1231                                 origin_decref(porigin);
1232                                 goto finish;
1233                         }
1234                         for (j = same = 0; j < i; j++)
1235                                 if (parent_origin[j] &&
1236                                     !hashcmp(parent_origin[j]->blob_sha1,
1237                                              porigin->blob_sha1)) {
1238                                         same = 1;
1239                                         break;
1240                                 }
1241                         if (!same)
1242                                 parent_origin[i] = porigin;
1243                         else
1244                                 origin_decref(porigin);
1245                 }
1246         }
1247
1248         num_commits++;
1249         for (i = 0, parent = commit->parents;
1250              i < MAXPARENT && parent;
1251              parent = parent->next, i++) {
1252                 struct origin *porigin = parent_origin[i];
1253                 if (!porigin)
1254                         continue;
1255                 if (pass_blame_to_parent(sb, origin, porigin))
1256                         goto finish;
1257         }
1258
1259         /*
1260          * Optionally find moves in parents' files.
1261          */
1262         if (opt & PICKAXE_BLAME_MOVE)
1263                 for (i = 0, parent = commit->parents;
1264                      i < MAXPARENT && parent;
1265                      parent = parent->next, i++) {
1266                         struct origin *porigin = parent_origin[i];
1267                         if (!porigin)
1268                                 continue;
1269                         if (find_move_in_parent(sb, origin, porigin))
1270                                 goto finish;
1271                 }
1272
1273         /*
1274          * Optionally find copies from parents' files.
1275          */
1276         if (opt & PICKAXE_BLAME_COPY)
1277                 for (i = 0, parent = commit->parents;
1278                      i < MAXPARENT && parent;
1279                      parent = parent->next, i++) {
1280                         struct origin *porigin = parent_origin[i];
1281                         if (find_copy_in_parent(sb, origin, parent->item,
1282                                                 porigin, opt))
1283                                 goto finish;
1284                 }
1285
1286  finish:
1287         for (i = 0; i < MAXPARENT; i++) {
1288                 if (parent_origin[i]) {
1289                         drop_origin_blob(parent_origin[i]);
1290                         origin_decref(parent_origin[i]);
1291                 }
1292         }
1293         drop_origin_blob(origin);
1294 }
1295
1296 /*
1297  * Information on commits, used for output.
1298  */
1299 struct commit_info
1300 {
1301         const char *author;
1302         const char *author_mail;
1303         unsigned long author_time;
1304         const char *author_tz;
1305
1306         /* filled only when asked for details */
1307         const char *committer;
1308         const char *committer_mail;
1309         unsigned long committer_time;
1310         const char *committer_tz;
1311
1312         const char *summary;
1313 };
1314
1315 /*
1316  * Parse author/committer line in the commit object buffer
1317  */
1318 static void get_ac_line(const char *inbuf, const char *what,
1319                         int bufsz, char *person, const char **mail,
1320                         unsigned long *time, const char **tz)
1321 {
1322         int len, tzlen, maillen;
1323         char *tmp, *endp, *timepos;
1324
1325         tmp = strstr(inbuf, what);
1326         if (!tmp)
1327                 goto error_out;
1328         tmp += strlen(what);
1329         endp = strchr(tmp, '\n');
1330         if (!endp)
1331                 len = strlen(tmp);
1332         else
1333                 len = endp - tmp;
1334         if (bufsz <= len) {
1335         error_out:
1336                 /* Ugh */
1337                 *mail = *tz = "(unknown)";
1338                 *time = 0;
1339                 return;
1340         }
1341         memcpy(person, tmp, len);
1342
1343         tmp = person;
1344         tmp += len;
1345         *tmp = 0;
1346         while (*tmp != ' ')
1347                 tmp--;
1348         *tz = tmp+1;
1349         tzlen = (person+len)-(tmp+1);
1350
1351         *tmp = 0;
1352         while (*tmp != ' ')
1353                 tmp--;
1354         *time = strtoul(tmp, NULL, 10);
1355         timepos = tmp;
1356
1357         *tmp = 0;
1358         while (*tmp != ' ')
1359                 tmp--;
1360         *mail = tmp + 1;
1361         *tmp = 0;
1362         maillen = timepos - tmp;
1363
1364         if (!mailmap.nr)
1365                 return;
1366
1367         /*
1368          * mailmap expansion may make the name longer.
1369          * make room by pushing stuff down.
1370          */
1371         tmp = person + bufsz - (tzlen + 1);
1372         memmove(tmp, *tz, tzlen);
1373         tmp[tzlen] = 0;
1374         *tz = tmp;
1375
1376         tmp = tmp - (maillen + 1);
1377         memmove(tmp, *mail, maillen);
1378         tmp[maillen] = 0;
1379         *mail = tmp;
1380
1381         /*
1382          * Now, convert e-mail using mailmap
1383          */
1384         map_email(&mailmap, tmp + 1, person, tmp-person-1);
1385 }
1386
1387 static void get_commit_info(struct commit *commit,
1388                             struct commit_info *ret,
1389                             int detailed)
1390 {
1391         int len;
1392         char *tmp, *endp;
1393         static char author_buf[1024];
1394         static char committer_buf[1024];
1395         static char summary_buf[1024];
1396
1397         /*
1398          * We've operated without save_commit_buffer, so
1399          * we now need to populate them for output.
1400          */
1401         if (!commit->buffer) {
1402                 enum object_type type;
1403                 unsigned long size;
1404                 commit->buffer =
1405                         read_sha1_file(commit->object.sha1, &type, &size);
1406                 if (!commit->buffer)
1407                         die("Cannot read commit %s",
1408                             sha1_to_hex(commit->object.sha1));
1409         }
1410         ret->author = author_buf;
1411         get_ac_line(commit->buffer, "\nauthor ",
1412                     sizeof(author_buf), author_buf, &ret->author_mail,
1413                     &ret->author_time, &ret->author_tz);
1414
1415         if (!detailed)
1416                 return;
1417
1418         ret->committer = committer_buf;
1419         get_ac_line(commit->buffer, "\ncommitter ",
1420                     sizeof(committer_buf), committer_buf, &ret->committer_mail,
1421                     &ret->committer_time, &ret->committer_tz);
1422
1423         ret->summary = summary_buf;
1424         tmp = strstr(commit->buffer, "\n\n");
1425         if (!tmp) {
1426         error_out:
1427                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1428                 return;
1429         }
1430         tmp += 2;
1431         endp = strchr(tmp, '\n');
1432         if (!endp)
1433                 endp = tmp + strlen(tmp);
1434         len = endp - tmp;
1435         if (len >= sizeof(summary_buf) || len == 0)
1436                 goto error_out;
1437         memcpy(summary_buf, tmp, len);
1438         summary_buf[len] = 0;
1439 }
1440
1441 /*
1442  * To allow LF and other nonportable characters in pathnames,
1443  * they are c-style quoted as needed.
1444  */
1445 static void write_filename_info(const char *path)
1446 {
1447         printf("filename ");
1448         write_name_quoted(path, stdout, '\n');
1449 }
1450
1451 /*
1452  * The blame_entry is found to be guilty for the range.  Mark it
1453  * as such, and show it in incremental output.
1454  */
1455 static void found_guilty_entry(struct blame_entry *ent)
1456 {
1457         if (ent->guilty)
1458                 return;
1459         ent->guilty = 1;
1460         if (incremental) {
1461                 struct origin *suspect = ent->suspect;
1462
1463                 printf("%s %d %d %d\n",
1464                        sha1_to_hex(suspect->commit->object.sha1),
1465                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1466                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1467                         struct commit_info ci;
1468                         suspect->commit->object.flags |= METAINFO_SHOWN;
1469                         get_commit_info(suspect->commit, &ci, 1);
1470                         printf("author %s\n", ci.author);
1471                         printf("author-mail %s\n", ci.author_mail);
1472                         printf("author-time %lu\n", ci.author_time);
1473                         printf("author-tz %s\n", ci.author_tz);
1474                         printf("committer %s\n", ci.committer);
1475                         printf("committer-mail %s\n", ci.committer_mail);
1476                         printf("committer-time %lu\n", ci.committer_time);
1477                         printf("committer-tz %s\n", ci.committer_tz);
1478                         printf("summary %s\n", ci.summary);
1479                         if (suspect->commit->object.flags & UNINTERESTING)
1480                                 printf("boundary\n");
1481                 }
1482                 write_filename_info(suspect->path);
1483                 maybe_flush_or_die(stdout, "stdout");
1484         }
1485 }
1486
1487 /*
1488  * The main loop -- while the scoreboard has lines whose true origin
1489  * is still unknown, pick one blame_entry, and allow its current
1490  * suspect to pass blames to its parents.
1491  */
1492 static void assign_blame(struct scoreboard *sb, struct rev_info *revs, int opt)
1493 {
1494         while (1) {
1495                 struct blame_entry *ent;
1496                 struct commit *commit;
1497                 struct origin *suspect = NULL;
1498
1499                 /* find one suspect to break down */
1500                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1501                         if (!ent->guilty)
1502                                 suspect = ent->suspect;
1503                 if (!suspect)
1504                         return; /* all done */
1505
1506                 /*
1507                  * We will use this suspect later in the loop,
1508                  * so hold onto it in the meantime.
1509                  */
1510                 origin_incref(suspect);
1511                 commit = suspect->commit;
1512                 if (!commit->object.parsed)
1513                         parse_commit(commit);
1514                 if (!(commit->object.flags & UNINTERESTING) &&
1515                     !(revs->max_age != -1 && commit->date < revs->max_age))
1516                         pass_blame(sb, suspect, opt);
1517                 else {
1518                         commit->object.flags |= UNINTERESTING;
1519                         if (commit->object.parsed)
1520                                 mark_parents_uninteresting(commit);
1521                 }
1522                 /* treat root commit as boundary */
1523                 if (!commit->parents && !show_root)
1524                         commit->object.flags |= UNINTERESTING;
1525
1526                 /* Take responsibility for the remaining entries */
1527                 for (ent = sb->ent; ent; ent = ent->next)
1528                         if (same_suspect(ent->suspect, suspect))
1529                                 found_guilty_entry(ent);
1530                 origin_decref(suspect);
1531
1532                 if (DEBUG) /* sanity */
1533                         sanity_check_refcnt(sb);
1534         }
1535 }
1536
1537 static const char *format_time(unsigned long time, const char *tz_str,
1538                                int show_raw_time)
1539 {
1540         static char time_buf[128];
1541         time_t t = time;
1542         int minutes, tz;
1543         struct tm *tm;
1544
1545         if (show_raw_time) {
1546                 sprintf(time_buf, "%lu %s", time, tz_str);
1547                 return time_buf;
1548         }
1549
1550         tz = atoi(tz_str);
1551         minutes = tz < 0 ? -tz : tz;
1552         minutes = (minutes / 100)*60 + (minutes % 100);
1553         minutes = tz < 0 ? -minutes : minutes;
1554         t = time + minutes * 60;
1555         tm = gmtime(&t);
1556
1557         strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1558         strcat(time_buf, tz_str);
1559         return time_buf;
1560 }
1561
1562 #define OUTPUT_ANNOTATE_COMPAT  001
1563 #define OUTPUT_LONG_OBJECT_NAME 002
1564 #define OUTPUT_RAW_TIMESTAMP    004
1565 #define OUTPUT_PORCELAIN        010
1566 #define OUTPUT_SHOW_NAME        020
1567 #define OUTPUT_SHOW_NUMBER      040
1568 #define OUTPUT_SHOW_SCORE      0100
1569 #define OUTPUT_NO_AUTHOR       0200
1570
1571 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1572 {
1573         int cnt;
1574         const char *cp;
1575         struct origin *suspect = ent->suspect;
1576         char hex[41];
1577
1578         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1579         printf("%s%c%d %d %d\n",
1580                hex,
1581                ent->guilty ? ' ' : '*', // purely for debugging
1582                ent->s_lno + 1,
1583                ent->lno + 1,
1584                ent->num_lines);
1585         if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1586                 struct commit_info ci;
1587                 suspect->commit->object.flags |= METAINFO_SHOWN;
1588                 get_commit_info(suspect->commit, &ci, 1);
1589                 printf("author %s\n", ci.author);
1590                 printf("author-mail %s\n", ci.author_mail);
1591                 printf("author-time %lu\n", ci.author_time);
1592                 printf("author-tz %s\n", ci.author_tz);
1593                 printf("committer %s\n", ci.committer);
1594                 printf("committer-mail %s\n", ci.committer_mail);
1595                 printf("committer-time %lu\n", ci.committer_time);
1596                 printf("committer-tz %s\n", ci.committer_tz);
1597                 write_filename_info(suspect->path);
1598                 printf("summary %s\n", ci.summary);
1599                 if (suspect->commit->object.flags & UNINTERESTING)
1600                         printf("boundary\n");
1601         }
1602         else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1603                 write_filename_info(suspect->path);
1604
1605         cp = nth_line(sb, ent->lno);
1606         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1607                 char ch;
1608                 if (cnt)
1609                         printf("%s %d %d\n", hex,
1610                                ent->s_lno + 1 + cnt,
1611                                ent->lno + 1 + cnt);
1612                 putchar('\t');
1613                 do {
1614                         ch = *cp++;
1615                         putchar(ch);
1616                 } while (ch != '\n' &&
1617                          cp < sb->final_buf + sb->final_buf_size);
1618         }
1619 }
1620
1621 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1622 {
1623         int cnt;
1624         const char *cp;
1625         struct origin *suspect = ent->suspect;
1626         struct commit_info ci;
1627         char hex[41];
1628         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1629
1630         get_commit_info(suspect->commit, &ci, 1);
1631         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1632
1633         cp = nth_line(sb, ent->lno);
1634         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1635                 char ch;
1636                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1637
1638                 if (suspect->commit->object.flags & UNINTERESTING) {
1639                         if (blank_boundary)
1640                                 memset(hex, ' ', length);
1641                         else if (!cmd_is_annotate) {
1642                                 length--;
1643                                 putchar('^');
1644                         }
1645                 }
1646
1647                 printf("%.*s", length, hex);
1648                 if (opt & OUTPUT_ANNOTATE_COMPAT)
1649                         printf("\t(%10s\t%10s\t%d)", ci.author,
1650                                format_time(ci.author_time, ci.author_tz,
1651                                            show_raw_time),
1652                                ent->lno + 1 + cnt);
1653                 else {
1654                         if (opt & OUTPUT_SHOW_SCORE)
1655                                 printf(" %*d %02d",
1656                                        max_score_digits, ent->score,
1657                                        ent->suspect->refcnt);
1658                         if (opt & OUTPUT_SHOW_NAME)
1659                                 printf(" %-*.*s", longest_file, longest_file,
1660                                        suspect->path);
1661                         if (opt & OUTPUT_SHOW_NUMBER)
1662                                 printf(" %*d", max_orig_digits,
1663                                        ent->s_lno + 1 + cnt);
1664
1665                         if (!(opt & OUTPUT_NO_AUTHOR))
1666                                 printf(" (%-*.*s %10s",
1667                                        longest_author, longest_author,
1668                                        ci.author,
1669                                        format_time(ci.author_time,
1670                                                    ci.author_tz,
1671                                                    show_raw_time));
1672                         printf(" %*d) ",
1673                                max_digits, ent->lno + 1 + cnt);
1674                 }
1675                 do {
1676                         ch = *cp++;
1677                         putchar(ch);
1678                 } while (ch != '\n' &&
1679                          cp < sb->final_buf + sb->final_buf_size);
1680         }
1681 }
1682
1683 static void output(struct scoreboard *sb, int option)
1684 {
1685         struct blame_entry *ent;
1686
1687         if (option & OUTPUT_PORCELAIN) {
1688                 for (ent = sb->ent; ent; ent = ent->next) {
1689                         struct blame_entry *oth;
1690                         struct origin *suspect = ent->suspect;
1691                         struct commit *commit = suspect->commit;
1692                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1693                                 continue;
1694                         for (oth = ent->next; oth; oth = oth->next) {
1695                                 if ((oth->suspect->commit != commit) ||
1696                                     !strcmp(oth->suspect->path, suspect->path))
1697                                         continue;
1698                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1699                                 break;
1700                         }
1701                 }
1702         }
1703
1704         for (ent = sb->ent; ent; ent = ent->next) {
1705                 if (option & OUTPUT_PORCELAIN)
1706                         emit_porcelain(sb, ent);
1707                 else {
1708                         emit_other(sb, ent, option);
1709                 }
1710         }
1711 }
1712
1713 /*
1714  * To allow quick access to the contents of nth line in the
1715  * final image, prepare an index in the scoreboard.
1716  */
1717 static int prepare_lines(struct scoreboard *sb)
1718 {
1719         const char *buf = sb->final_buf;
1720         unsigned long len = sb->final_buf_size;
1721         int num = 0, incomplete = 0, bol = 1;
1722
1723         if (len && buf[len-1] != '\n')
1724                 incomplete++; /* incomplete line at the end */
1725         while (len--) {
1726                 if (bol) {
1727                         sb->lineno = xrealloc(sb->lineno,
1728                                               sizeof(int* ) * (num + 1));
1729                         sb->lineno[num] = buf - sb->final_buf;
1730                         bol = 0;
1731                 }
1732                 if (*buf++ == '\n') {
1733                         num++;
1734                         bol = 1;
1735                 }
1736         }
1737         sb->lineno = xrealloc(sb->lineno,
1738                               sizeof(int* ) * (num + incomplete + 1));
1739         sb->lineno[num + incomplete] = buf - sb->final_buf;
1740         sb->num_lines = num + incomplete;
1741         return sb->num_lines;
1742 }
1743
1744 /*
1745  * Add phony grafts for use with -S; this is primarily to
1746  * support git-cvsserver that wants to give a linear history
1747  * to its clients.
1748  */
1749 static int read_ancestry(const char *graft_file)
1750 {
1751         FILE *fp = fopen(graft_file, "r");
1752         char buf[1024];
1753         if (!fp)
1754                 return -1;
1755         while (fgets(buf, sizeof(buf), fp)) {
1756                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1757                 int len = strlen(buf);
1758                 struct commit_graft *graft = read_graft_line(buf, len);
1759                 if (graft)
1760                         register_commit_graft(graft, 0);
1761         }
1762         fclose(fp);
1763         return 0;
1764 }
1765
1766 /*
1767  * How many columns do we need to show line numbers in decimal?
1768  */
1769 static int lineno_width(int lines)
1770 {
1771         int i, width;
1772
1773         for (width = 1, i = 10; i <= lines + 1; width++)
1774                 i *= 10;
1775         return width;
1776 }
1777
1778 /*
1779  * How many columns do we need to show line numbers, authors,
1780  * and filenames?
1781  */
1782 static void find_alignment(struct scoreboard *sb, int *option)
1783 {
1784         int longest_src_lines = 0;
1785         int longest_dst_lines = 0;
1786         unsigned largest_score = 0;
1787         struct blame_entry *e;
1788
1789         for (e = sb->ent; e; e = e->next) {
1790                 struct origin *suspect = e->suspect;
1791                 struct commit_info ci;
1792                 int num;
1793
1794                 if (strcmp(suspect->path, sb->path))
1795                         *option |= OUTPUT_SHOW_NAME;
1796                 num = strlen(suspect->path);
1797                 if (longest_file < num)
1798                         longest_file = num;
1799                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1800                         suspect->commit->object.flags |= METAINFO_SHOWN;
1801                         get_commit_info(suspect->commit, &ci, 1);
1802                         num = strlen(ci.author);
1803                         if (longest_author < num)
1804                                 longest_author = num;
1805                 }
1806                 num = e->s_lno + e->num_lines;
1807                 if (longest_src_lines < num)
1808                         longest_src_lines = num;
1809                 num = e->lno + e->num_lines;
1810                 if (longest_dst_lines < num)
1811                         longest_dst_lines = num;
1812                 if (largest_score < ent_score(sb, e))
1813                         largest_score = ent_score(sb, e);
1814         }
1815         max_orig_digits = lineno_width(longest_src_lines);
1816         max_digits = lineno_width(longest_dst_lines);
1817         max_score_digits = lineno_width(largest_score);
1818 }
1819
1820 /*
1821  * For debugging -- origin is refcounted, and this asserts that
1822  * we do not underflow.
1823  */
1824 static void sanity_check_refcnt(struct scoreboard *sb)
1825 {
1826         int baa = 0;
1827         struct blame_entry *ent;
1828
1829         for (ent = sb->ent; ent; ent = ent->next) {
1830                 /* Nobody should have zero or negative refcnt */
1831                 if (ent->suspect->refcnt <= 0) {
1832                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1833                                 ent->suspect->path,
1834                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1835                                 ent->suspect->refcnt);
1836                         baa = 1;
1837                 }
1838         }
1839         for (ent = sb->ent; ent; ent = ent->next) {
1840                 /* Mark the ones that haven't been checked */
1841                 if (0 < ent->suspect->refcnt)
1842                         ent->suspect->refcnt = -ent->suspect->refcnt;
1843         }
1844         for (ent = sb->ent; ent; ent = ent->next) {
1845                 /*
1846                  * ... then pick each and see if they have the the
1847                  * correct refcnt.
1848                  */
1849                 int found;
1850                 struct blame_entry *e;
1851                 struct origin *suspect = ent->suspect;
1852
1853                 if (0 < suspect->refcnt)
1854                         continue;
1855                 suspect->refcnt = -suspect->refcnt; /* Unmark */
1856                 for (found = 0, e = sb->ent; e; e = e->next) {
1857                         if (e->suspect != suspect)
1858                                 continue;
1859                         found++;
1860                 }
1861                 if (suspect->refcnt != found) {
1862                         fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1863                                 ent->suspect->path,
1864                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1865                                 ent->suspect->refcnt, found);
1866                         baa = 2;
1867                 }
1868         }
1869         if (baa) {
1870                 int opt = 0160;
1871                 find_alignment(sb, &opt);
1872                 output(sb, opt);
1873                 die("Baa %d!", baa);
1874         }
1875 }
1876
1877 /*
1878  * Used for the command line parsing; check if the path exists
1879  * in the working tree.
1880  */
1881 static int has_path_in_work_tree(const char *path)
1882 {
1883         struct stat st;
1884         return !lstat(path, &st);
1885 }
1886
1887 static unsigned parse_score(const char *arg)
1888 {
1889         char *end;
1890         unsigned long score = strtoul(arg, &end, 10);
1891         if (*end)
1892                 return 0;
1893         return score;
1894 }
1895
1896 static const char *add_prefix(const char *prefix, const char *path)
1897 {
1898         if (!prefix || !prefix[0])
1899                 return path;
1900         return prefix_path(prefix, strlen(prefix), path);
1901 }
1902
1903 /*
1904  * Parsing of (comma separated) one item in the -L option
1905  */
1906 static const char *parse_loc(const char *spec,
1907                              struct scoreboard *sb, long lno,
1908                              long begin, long *ret)
1909 {
1910         char *term;
1911         const char *line;
1912         long num;
1913         int reg_error;
1914         regex_t regexp;
1915         regmatch_t match[1];
1916
1917         /* Allow "-L <something>,+20" to mean starting at <something>
1918          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1919          * <something>.
1920          */
1921         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1922                 num = strtol(spec + 1, &term, 10);
1923                 if (term != spec + 1) {
1924                         if (spec[0] == '-')
1925                                 num = 0 - num;
1926                         if (0 < num)
1927                                 *ret = begin + num - 2;
1928                         else if (!num)
1929                                 *ret = begin;
1930                         else
1931                                 *ret = begin + num;
1932                         return term;
1933                 }
1934                 return spec;
1935         }
1936         num = strtol(spec, &term, 10);
1937         if (term != spec) {
1938                 *ret = num;
1939                 return term;
1940         }
1941         if (spec[0] != '/')
1942                 return spec;
1943
1944         /* it could be a regexp of form /.../ */
1945         for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1946                 if (*term == '\\')
1947                         term++;
1948         }
1949         if (*term != '/')
1950                 return spec;
1951
1952         /* try [spec+1 .. term-1] as regexp */
1953         *term = 0;
1954         begin--; /* input is in human terms */
1955         line = nth_line(sb, begin);
1956
1957         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1958             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1959                 const char *cp = line + match[0].rm_so;
1960                 const char *nline;
1961
1962                 while (begin++ < lno) {
1963                         nline = nth_line(sb, begin);
1964                         if (line <= cp && cp < nline)
1965                                 break;
1966                         line = nline;
1967                 }
1968                 *ret = begin;
1969                 regfree(&regexp);
1970                 *term++ = '/';
1971                 return term;
1972         }
1973         else {
1974                 char errbuf[1024];
1975                 regerror(reg_error, &regexp, errbuf, 1024);
1976                 die("-L parameter '%s': %s", spec + 1, errbuf);
1977         }
1978 }
1979
1980 /*
1981  * Parsing of -L option
1982  */
1983 static void prepare_blame_range(struct scoreboard *sb,
1984                                 const char *bottomtop,
1985                                 long lno,
1986                                 long *bottom, long *top)
1987 {
1988         const char *term;
1989
1990         term = parse_loc(bottomtop, sb, lno, 1, bottom);
1991         if (*term == ',') {
1992                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1993                 if (*term)
1994                         usage(blame_usage);
1995         }
1996         if (*term)
1997                 usage(blame_usage);
1998 }
1999
2000 static int git_blame_config(const char *var, const char *value)
2001 {
2002         if (!strcmp(var, "blame.showroot")) {
2003                 show_root = git_config_bool(var, value);
2004                 return 0;
2005         }
2006         if (!strcmp(var, "blame.blankboundary")) {
2007                 blank_boundary = git_config_bool(var, value);
2008                 return 0;
2009         }
2010         return git_default_config(var, value);
2011 }
2012
2013 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
2014 {
2015         struct commit *commit;
2016         struct origin *origin;
2017         unsigned char head_sha1[20];
2018         struct strbuf buf;
2019         const char *ident;
2020         time_t now;
2021         int size, len;
2022         struct cache_entry *ce;
2023         unsigned mode;
2024
2025         if (get_sha1("HEAD", head_sha1))
2026                 die("No such ref: HEAD");
2027
2028         time(&now);
2029         commit = xcalloc(1, sizeof(*commit));
2030         commit->parents = xcalloc(1, sizeof(*commit->parents));
2031         commit->parents->item = lookup_commit_reference(head_sha1);
2032         commit->object.parsed = 1;
2033         commit->date = now;
2034         commit->object.type = OBJ_COMMIT;
2035
2036         origin = make_origin(commit, path);
2037
2038         strbuf_init(&buf, 0);
2039         if (!contents_from || strcmp("-", contents_from)) {
2040                 struct stat st;
2041                 const char *read_from;
2042                 unsigned long fin_size;
2043
2044                 if (contents_from) {
2045                         if (stat(contents_from, &st) < 0)
2046                                 die("Cannot stat %s", contents_from);
2047                         read_from = contents_from;
2048                 }
2049                 else {
2050                         if (lstat(path, &st) < 0)
2051                                 die("Cannot lstat %s", path);
2052                         read_from = path;
2053                 }
2054                 fin_size = xsize_t(st.st_size);
2055                 mode = canon_mode(st.st_mode);
2056                 switch (st.st_mode & S_IFMT) {
2057                 case S_IFREG:
2058                         if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2059                                 die("cannot open or read %s", read_from);
2060                         break;
2061                 case S_IFLNK:
2062                         if (readlink(read_from, buf.buf, buf.alloc) != fin_size)
2063                                 die("cannot readlink %s", read_from);
2064                         buf.len = fin_size;
2065                         break;
2066                 default:
2067                         die("unsupported file type %s", read_from);
2068                 }
2069         }
2070         else {
2071                 /* Reading from stdin */
2072                 contents_from = "standard input";
2073                 mode = 0;
2074                 if (strbuf_read(&buf, 0, 0) < 0)
2075                         die("read error %s from stdin", strerror(errno));
2076         }
2077         convert_to_git(path, buf.buf, buf.len, &buf);
2078         origin->file.ptr = buf.buf;
2079         origin->file.size = buf.len;
2080         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2081         commit->util = origin;
2082
2083         /*
2084          * Read the current index, replace the path entry with
2085          * origin->blob_sha1 without mucking with its mode or type
2086          * bits; we are not going to write this index out -- we just
2087          * want to run "diff-index --cached".
2088          */
2089         discard_cache();
2090         read_cache();
2091
2092         len = strlen(path);
2093         if (!mode) {
2094                 int pos = cache_name_pos(path, len);
2095                 if (0 <= pos)
2096                         mode = ntohl(active_cache[pos]->ce_mode);
2097                 else
2098                         /* Let's not bother reading from HEAD tree */
2099                         mode = S_IFREG | 0644;
2100         }
2101         size = cache_entry_size(len);
2102         ce = xcalloc(1, size);
2103         hashcpy(ce->sha1, origin->blob_sha1);
2104         memcpy(ce->name, path, len);
2105         ce->ce_flags = create_ce_flags(len, 0);
2106         ce->ce_mode = create_ce_mode(mode);
2107         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2108
2109         /*
2110          * We are not going to write this out, so this does not matter
2111          * right now, but someday we might optimize diff-index --cached
2112          * with cache-tree information.
2113          */
2114         cache_tree_invalidate_path(active_cache_tree, path);
2115
2116         commit->buffer = xmalloc(400);
2117         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2118         snprintf(commit->buffer, 400,
2119                 "tree 0000000000000000000000000000000000000000\n"
2120                 "parent %s\n"
2121                 "author %s\n"
2122                 "committer %s\n\n"
2123                 "Version of %s from %s\n",
2124                 sha1_to_hex(head_sha1),
2125                 ident, ident, path, contents_from ? contents_from : path);
2126         return commit;
2127 }
2128
2129 int cmd_blame(int argc, const char **argv, const char *prefix)
2130 {
2131         struct rev_info revs;
2132         const char *path;
2133         struct scoreboard sb;
2134         struct origin *o;
2135         struct blame_entry *ent;
2136         int i, seen_dashdash, unk, opt;
2137         long bottom, top, lno;
2138         int output_option = 0;
2139         int show_stats = 0;
2140         const char *revs_file = NULL;
2141         const char *final_commit_name = NULL;
2142         enum object_type type;
2143         const char *bottomtop = NULL;
2144         const char *contents_from = NULL;
2145
2146         cmd_is_annotate = !strcmp(argv[0], "annotate");
2147
2148         git_config(git_blame_config);
2149         save_commit_buffer = 0;
2150
2151         opt = 0;
2152         seen_dashdash = 0;
2153         for (unk = i = 1; i < argc; i++) {
2154                 const char *arg = argv[i];
2155                 if (*arg != '-')
2156                         break;
2157                 else if (!strcmp("-b", arg))
2158                         blank_boundary = 1;
2159                 else if (!strcmp("--root", arg))
2160                         show_root = 1;
2161                 else if (!strcmp(arg, "--show-stats"))
2162                         show_stats = 1;
2163                 else if (!strcmp("-c", arg))
2164                         output_option |= OUTPUT_ANNOTATE_COMPAT;
2165                 else if (!strcmp("-t", arg))
2166                         output_option |= OUTPUT_RAW_TIMESTAMP;
2167                 else if (!strcmp("-l", arg))
2168                         output_option |= OUTPUT_LONG_OBJECT_NAME;
2169                 else if (!strcmp("-s", arg))
2170                         output_option |= OUTPUT_NO_AUTHOR;
2171                 else if (!strcmp("-w", arg))
2172                         xdl_opts |= XDF_IGNORE_WHITESPACE;
2173                 else if (!strcmp("-S", arg) && ++i < argc)
2174                         revs_file = argv[i];
2175                 else if (!prefixcmp(arg, "-M")) {
2176                         opt |= PICKAXE_BLAME_MOVE;
2177                         blame_move_score = parse_score(arg+2);
2178                 }
2179                 else if (!prefixcmp(arg, "-C")) {
2180                         /*
2181                          * -C enables copy from removed files;
2182                          * -C -C enables copy from existing files, but only
2183                          *       when blaming a new file;
2184                          * -C -C -C enables copy from existing files for
2185                          *          everybody
2186                          */
2187                         if (opt & PICKAXE_BLAME_COPY_HARDER)
2188                                 opt |= PICKAXE_BLAME_COPY_HARDEST;
2189                         if (opt & PICKAXE_BLAME_COPY)
2190                                 opt |= PICKAXE_BLAME_COPY_HARDER;
2191                         opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2192                         blame_copy_score = parse_score(arg+2);
2193                 }
2194                 else if (!prefixcmp(arg, "-L")) {
2195                         if (!arg[2]) {
2196                                 if (++i >= argc)
2197                                         usage(blame_usage);
2198                                 arg = argv[i];
2199                         }
2200                         else
2201                                 arg += 2;
2202                         if (bottomtop)
2203                                 die("More than one '-L n,m' option given");
2204                         bottomtop = arg;
2205                 }
2206                 else if (!strcmp("--contents", arg)) {
2207                         if (++i >= argc)
2208                                 usage(blame_usage);
2209                         contents_from = argv[i];
2210                 }
2211                 else if (!strcmp("--incremental", arg))
2212                         incremental = 1;
2213                 else if (!strcmp("--score-debug", arg))
2214                         output_option |= OUTPUT_SHOW_SCORE;
2215                 else if (!strcmp("-f", arg) ||
2216                          !strcmp("--show-name", arg))
2217                         output_option |= OUTPUT_SHOW_NAME;
2218                 else if (!strcmp("-n", arg) ||
2219                          !strcmp("--show-number", arg))
2220                         output_option |= OUTPUT_SHOW_NUMBER;
2221                 else if (!strcmp("-p", arg) ||
2222                          !strcmp("--porcelain", arg))
2223                         output_option |= OUTPUT_PORCELAIN;
2224                 else if (!strcmp("--", arg)) {
2225                         seen_dashdash = 1;
2226                         i++;
2227                         break;
2228                 }
2229                 else
2230                         argv[unk++] = arg;
2231         }
2232
2233         if (!blame_move_score)
2234                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2235         if (!blame_copy_score)
2236                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2237
2238         /*
2239          * We have collected options unknown to us in argv[1..unk]
2240          * which are to be passed to revision machinery if we are
2241          * going to do the "bottom" processing.
2242          *
2243          * The remaining are:
2244          *
2245          * (1) if seen_dashdash, its either
2246          *     "-options -- <path>" or
2247          *     "-options -- <path> <rev>".
2248          *     but the latter is allowed only if there is no
2249          *     options that we passed to revision machinery.
2250          *
2251          * (2) otherwise, we may have "--" somewhere later and
2252          *     might be looking at the first one of multiple 'rev'
2253          *     parameters (e.g. " master ^next ^maint -- path").
2254          *     See if there is a dashdash first, and give the
2255          *     arguments before that to revision machinery.
2256          *     After that there must be one 'path'.
2257          *
2258          * (3) otherwise, its one of the three:
2259          *     "-options <path> <rev>"
2260          *     "-options <rev> <path>"
2261          *     "-options <path>"
2262          *     but again the first one is allowed only if
2263          *     there is no options that we passed to revision
2264          *     machinery.
2265          */
2266
2267         if (seen_dashdash) {
2268                 /* (1) */
2269                 if (argc <= i)
2270                         usage(blame_usage);
2271                 path = add_prefix(prefix, argv[i]);
2272                 if (i + 1 == argc - 1) {
2273                         if (unk != 1)
2274                                 usage(blame_usage);
2275                         argv[unk++] = argv[i + 1];
2276                 }
2277                 else if (i + 1 != argc)
2278                         /* garbage at end */
2279                         usage(blame_usage);
2280         }
2281         else {
2282                 int j;
2283                 for (j = i; !seen_dashdash && j < argc; j++)
2284                         if (!strcmp(argv[j], "--"))
2285                                 seen_dashdash = j;
2286                 if (seen_dashdash) {
2287                         /* (2) */
2288                         if (seen_dashdash + 1 != argc - 1)
2289                                 usage(blame_usage);
2290                         path = add_prefix(prefix, argv[seen_dashdash + 1]);
2291                         for (j = i; j < seen_dashdash; j++)
2292                                 argv[unk++] = argv[j];
2293                 }
2294                 else {
2295                         /* (3) */
2296                         if (argc <= i)
2297                                 usage(blame_usage);
2298                         path = add_prefix(prefix, argv[i]);
2299                         if (i + 1 == argc - 1) {
2300                                 final_commit_name = argv[i + 1];
2301
2302                                 /* if (unk == 1) we could be getting
2303                                  * old-style
2304                                  */
2305                                 if (unk == 1 && !has_path_in_work_tree(path)) {
2306                                         path = add_prefix(prefix, argv[i + 1]);
2307                                         final_commit_name = argv[i];
2308                                 }
2309                         }
2310                         else if (i != argc - 1)
2311                                 usage(blame_usage); /* garbage at end */
2312
2313                         setup_work_tree();
2314                         if (!has_path_in_work_tree(path))
2315                                 die("cannot stat path %s: %s",
2316                                     path, strerror(errno));
2317                 }
2318         }
2319
2320         if (final_commit_name)
2321                 argv[unk++] = final_commit_name;
2322
2323         /*
2324          * Now we got rev and path.  We do not want the path pruning
2325          * but we may want "bottom" processing.
2326          */
2327         argv[unk++] = "--"; /* terminate the rev name */
2328         argv[unk] = NULL;
2329
2330         init_revisions(&revs, NULL);
2331         setup_revisions(unk, argv, &revs, NULL);
2332         memset(&sb, 0, sizeof(sb));
2333
2334         /*
2335          * There must be one and only one positive commit in the
2336          * revs->pending array.
2337          */
2338         for (i = 0; i < revs.pending.nr; i++) {
2339                 struct object *obj = revs.pending.objects[i].item;
2340                 if (obj->flags & UNINTERESTING)
2341                         continue;
2342                 while (obj->type == OBJ_TAG)
2343                         obj = deref_tag(obj, NULL, 0);
2344                 if (obj->type != OBJ_COMMIT)
2345                         die("Non commit %s?",
2346                             revs.pending.objects[i].name);
2347                 if (sb.final)
2348                         die("More than one commit to dig from %s and %s?",
2349                             revs.pending.objects[i].name,
2350                             final_commit_name);
2351                 sb.final = (struct commit *) obj;
2352                 final_commit_name = revs.pending.objects[i].name;
2353         }
2354
2355         if (!sb.final) {
2356                 /*
2357                  * "--not A B -- path" without anything positive;
2358                  * do not default to HEAD, but use the working tree
2359                  * or "--contents".
2360                  */
2361                 setup_work_tree();
2362                 sb.final = fake_working_tree_commit(path, contents_from);
2363                 add_pending_object(&revs, &(sb.final->object), ":");
2364         }
2365         else if (contents_from)
2366                 die("Cannot use --contents with final commit object name");
2367
2368         /*
2369          * If we have bottom, this will mark the ancestors of the
2370          * bottom commits we would reach while traversing as
2371          * uninteresting.
2372          */
2373         prepare_revision_walk(&revs);
2374
2375         if (is_null_sha1(sb.final->object.sha1)) {
2376                 char *buf;
2377                 o = sb.final->util;
2378                 buf = xmalloc(o->file.size + 1);
2379                 memcpy(buf, o->file.ptr, o->file.size + 1);
2380                 sb.final_buf = buf;
2381                 sb.final_buf_size = o->file.size;
2382         }
2383         else {
2384                 o = get_origin(&sb, sb.final, path);
2385                 if (fill_blob_sha1(o))
2386                         die("no such path %s in %s", path, final_commit_name);
2387
2388                 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2389                                               &sb.final_buf_size);
2390                 if (!sb.final_buf)
2391                         die("Cannot read blob %s for path %s",
2392                             sha1_to_hex(o->blob_sha1),
2393                             path);
2394         }
2395         num_read_blob++;
2396         lno = prepare_lines(&sb);
2397
2398         bottom = top = 0;
2399         if (bottomtop)
2400                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2401         if (bottom && top && top < bottom) {
2402                 long tmp;
2403                 tmp = top; top = bottom; bottom = tmp;
2404         }
2405         if (bottom < 1)
2406                 bottom = 1;
2407         if (top < 1)
2408                 top = lno;
2409         bottom--;
2410         if (lno < top)
2411                 die("file %s has only %lu lines", path, lno);
2412
2413         ent = xcalloc(1, sizeof(*ent));
2414         ent->lno = bottom;
2415         ent->num_lines = top - bottom;
2416         ent->suspect = o;
2417         ent->s_lno = bottom;
2418
2419         sb.ent = ent;
2420         sb.path = path;
2421
2422         if (revs_file && read_ancestry(revs_file))
2423                 die("reading graft file %s failed: %s",
2424                     revs_file, strerror(errno));
2425
2426         read_mailmap(&mailmap, ".mailmap", NULL);
2427
2428         if (!incremental)
2429                 setup_pager();
2430
2431         assign_blame(&sb, &revs, opt);
2432
2433         if (incremental)
2434                 return 0;
2435
2436         coalesce(&sb);
2437
2438         if (!(output_option & OUTPUT_PORCELAIN))
2439                 find_alignment(&sb, &output_option);
2440
2441         output(&sb, output_option);
2442         free((void *)sb.final_buf);
2443         for (ent = sb.ent; ent; ) {
2444                 struct blame_entry *e = ent->next;
2445                 free(ent);
2446                 ent = e;
2447         }
2448
2449         if (show_stats) {
2450                 printf("num read blob: %d\n", num_read_blob);
2451                 printf("num get patch: %d\n", num_get_patch);
2452                 printf("num commits: %d\n", num_commits);
2453         }
2454         return 0;
2455 }