builtin/blame.c

   1 /*
   2  * Blame
   3  *
   4  * Copyright (c) 2006, Junio C Hamano
   5  */
   6
   7 #include "cache.h"
   8 #include "builtin.h"
   9 #include "blob.h"
  10 #include "commit.h"
  11 #include "tag.h"
  12 #include "tree-walk.h"
  13 #include "diff.h"
  14 #include "diffcore.h"
  15 #include "revision.h"
  16 #include "quote.h"
  17 #include "xdiff-interface.h"
  18 #include "cache-tree.h"
  19 #include "string-list.h"
  20 #include "mailmap.h"
  21 #include "parse-options.h"
  22 #include "utf8.h"
  23 #include "userdiff.h"
  24 #include "line-range.h"
  25 #include "line-log.h"
  26
  27 static char blame_usage[] = N_("git blame [options] [rev-opts] [rev] [--] file");
  28
  29 static const char *blame_opt_usage[] = {
  30         blame_usage,
  31         "",
  32         N_("[rev-opts] are documented in git-rev-list(1)"),
  33         NULL
  34 };
  35
  36 static int longest_file;
  37 static int longest_author;
  38 static int max_orig_digits;
  39 static int max_digits;
  40 static int max_score_digits;
  41 static int show_root;
  42 static int reverse;
  43 static int blank_boundary;
  44 static int incremental;
  45 static int xdl_opts;
  46 static int abbrev = -1;
  47 static int no_whole_file_rename;
  48
  49 static enum date_mode blame_date_mode = DATE_ISO8601;
  50 static size_t blame_date_width;
  51
  52 static struct string_list mailmap;
  53
  54 #ifndef DEBUG
  55 #define DEBUG 0
  56 #endif
  57
  58 /* stats */
  59 static int num_read_blob;
  60 static int num_get_patch;
  61 static int num_commits;
  62
  63 #define PICKAXE_BLAME_MOVE              01
  64 #define PICKAXE_BLAME_COPY              02
  65 #define PICKAXE_BLAME_COPY_HARDER       04
  66 #define PICKAXE_BLAME_COPY_HARDEST      010
  67
  68 /*
  69  * blame for a blame_entry with score lower than these thresholds
  70  * is not passed to the parent using move/copy logic.
  71  */
  72 static unsigned blame_move_score;
  73 static unsigned blame_copy_score;
  74 #define BLAME_DEFAULT_MOVE_SCORE        20
  75 #define BLAME_DEFAULT_COPY_SCORE        40
  76
  77 /* Remember to update object flag allocation in object.h */
  78 #define METAINFO_SHOWN          (1u<<12)
  79 #define MORE_THAN_ONE_PATH      (1u<<13)
  80
  81 /*
  82  * One blob in a commit that is being suspected
  83  */
  84 struct origin {
  85         int refcnt;
  86         struct origin *previous;
  87         struct commit *commit;
  88         mmfile_t file;
  89         unsigned char blob_sha1[20];
  90         unsigned mode;
  91         char path[FLEX_ARRAY];
  92 };
  93
  94 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
  95                       xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
  96 {
  97         xpparam_t xpp = {0};
  98         xdemitconf_t xecfg = {0};
  99         xdemitcb_t ecb = {NULL};
 100
 101         xpp.flags = xdl_opts;
 102         xecfg.ctxlen = ctxlen;
 103         xecfg.hunk_func = hunk_func;
 104         ecb.priv = cb_data;
 105         return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
 106 }
 107
 108 /*
 109  * Prepare diff_filespec and convert it using diff textconv API
 110  * if the textconv driver exists.
 111  * Return 1 if the conversion succeeds, 0 otherwise.
 112  */
 113 int textconv_object(const char *path,
 114                     unsigned mode,
 115                     const unsigned char *sha1,
 116                     int sha1_valid,
 117                     char **buf,
 118                     unsigned long *buf_size)
 119 {
 120         struct diff_filespec *df;
 121         struct userdiff_driver *textconv;
 122
 123         df = alloc_filespec(path);
 124         fill_filespec(df, sha1, sha1_valid, mode);
 125         textconv = get_textconv(df);
 126         if (!textconv) {
 127                 free_filespec(df);
 128                 return 0;
 129         }
 130
 131         *buf_size = fill_textconv(textconv, df, buf);
 132         free_filespec(df);
 133         return 1;
 134 }
 135
 136 /*
 137  * Given an origin, prepare mmfile_t structure to be used by the
 138  * diff machinery
 139  */
 140 static void fill_origin_blob(struct diff_options *opt,
 141                              struct origin *o, mmfile_t *file)
 142 {
 143         if (!o->file.ptr) {
 144                 enum object_type type;
 145                 unsigned long file_size;
 146
 147                 num_read_blob++;
 148                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
 149                     textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
 150                         ;
 151                 else
 152                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
 153                 file->size = file_size;
 154
 155                 if (!file->ptr)
 156                         die("Cannot read blob %s for path %s",
 157                             sha1_to_hex(o->blob_sha1),
 158                             o->path);
 159                 o->file = *file;
 160         }
 161         else
 162                 *file = o->file;
 163 }
 164
 165 /*
 166  * Origin is refcounted and usually we keep the blob contents to be
 167  * reused.
 168  */
 169 static inline struct origin *origin_incref(struct origin *o)
 170 {
 171         if (o)
 172                 o->refcnt++;
 173         return o;
 174 }
 175
 176 static void origin_decref(struct origin *o)
 177 {
 178         if (o && --o->refcnt <= 0) {
 179                 if (o->previous)
 180                         origin_decref(o->previous);
 181                 free(o->file.ptr);
 182                 free(o);
 183         }
 184 }
 185
 186 static void drop_origin_blob(struct origin *o)
 187 {
 188         if (o->file.ptr) {
 189                 free(o->file.ptr);
 190                 o->file.ptr = NULL;
 191         }
 192 }
 193
 194 /*
 195  * Each group of lines is described by a blame_entry; it can be split
 196  * as we pass blame to the parents.  They form a linked list in the
 197  * scoreboard structure, sorted by the target line number.
 198  */
 199 struct blame_entry {
 200         struct blame_entry *next;
 201
 202         /* the first line of this group in the final image;
 203          * internally all line numbers are 0 based.
 204          */
 205         int lno;
 206
 207         /* how many lines this group has */
 208         int num_lines;
 209
 210         /* the commit that introduced this group into the final image */
 211         struct origin *suspect;
 212
 213         /* true if the suspect is truly guilty; false while we have not
 214          * checked if the group came from one of its parents.
 215          */
 216         char guilty;
 217
 218         /* true if the entry has been scanned for copies in the current parent
 219          */
 220         char scanned;
 221
 222         /* the line number of the first line of this group in the
 223          * suspect's file; internally all line numbers are 0 based.
 224          */
 225         int s_lno;
 226
 227         /* how significant this entry is -- cached to avoid
 228          * scanning the lines over and over.
 229          */
 230         unsigned score;
 231 };
 232
 233 /*
 234  * The current state of the blame assignment.
 235  */
 236 struct scoreboard {
 237         /* the final commit (i.e. where we started digging from) */
 238         struct commit *final;
 239         struct rev_info *revs;
 240         const char *path;
 241
 242         /*
 243          * The contents in the final image.
 244          * Used by many functions to obtain contents of the nth line,
 245          * indexed with scoreboard.lineno[blame_entry.lno].
 246          */
 247         const char *final_buf;
 248         unsigned long final_buf_size;
 249
 250         /* linked list of blames */
 251         struct blame_entry *ent;
 252
 253         /* look-up a line in the final buffer */
 254         int num_lines;
 255         int *lineno;
 256 };
 257
 258 static void sanity_check_refcnt(struct scoreboard *);
 259
 260 /*
 261  * If two blame entries that are next to each other came from
 262  * contiguous lines in the same origin (i.e. <commit, path> pair),
 263  * merge them together.
 264  */
 265 static void coalesce(struct scoreboard *sb)
 266 {
 267         struct blame_entry *ent, *next;
 268
 269         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
 270                 if (ent->suspect == next->suspect &&
 271                     ent->guilty == next->guilty &&
 272                     ent->s_lno + ent->num_lines == next->s_lno) {
 273                         ent->num_lines += next->num_lines;
 274                         ent->next = next->next;
 275                         origin_decref(next->suspect);
 276                         free(next);
 277                         ent->score = 0;
 278                         next = ent; /* again */
 279                 }
 280         }
 281
 282         if (DEBUG) /* sanity */
 283                 sanity_check_refcnt(sb);
 284 }
 285
 286 /*
 287  * Given a commit and a path in it, create a new origin structure.
 288  * The callers that add blame to the scoreboard should use
 289  * get_origin() to obtain shared, refcounted copy instead of calling
 290  * this function directly.
 291  */
 292 static struct origin *make_origin(struct commit *commit, const char *path)
 293 {
 294         struct origin *o;
 295         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
 296         o->commit = commit;
 297         o->refcnt = 1;
 298         strcpy(o->path, path);
 299         return o;
 300 }
 301
 302 /*
 303  * Locate an existing origin or create a new one.
 304  */
 305 static struct origin *get_origin(struct scoreboard *sb,
 306                                  struct commit *commit,
 307                                  const char *path)
 308 {
 309         struct blame_entry *e;
 310
 311         for (e = sb->ent; e; e = e->next) {
 312                 if (e->suspect->commit == commit &&
 313                     !strcmp(e->suspect->path, path))
 314                         return origin_incref(e->suspect);
 315         }
 316         return make_origin(commit, path);
 317 }
 318
 319 /*
 320  * Fill the blob_sha1 field of an origin if it hasn't, so that later
 321  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
 322  * for an origin is also used to pass the blame for the entire file to
 323  * the parent to detect the case where a child's blob is identical to
 324  * that of its parent's.
 325  *
 326  * This also fills origin->mode for corresponding tree path.
 327  */
 328 static int fill_blob_sha1_and_mode(struct origin *origin)
 329 {
 330         if (!is_null_sha1(origin->blob_sha1))
 331                 return 0;
 332         if (get_tree_entry(origin->commit->object.sha1,
 333                            origin->path,
 334                            origin->blob_sha1, &origin->mode))
 335                 goto error_out;
 336         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
 337                 goto error_out;
 338         return 0;
 339  error_out:
 340         hashclr(origin->blob_sha1);
 341         origin->mode = S_IFINVALID;
 342         return -1;
 343 }
 344
 345 /*
 346  * We have an origin -- check if the same path exists in the
 347  * parent and return an origin structure to represent it.
 348  */
 349 static struct origin *find_origin(struct scoreboard *sb,
 350                                   struct commit *parent,
 351                                   struct origin *origin)
 352 {
 353         struct origin *porigin = NULL;
 354         struct diff_options diff_opts;
 355         const char *paths[2];
 356
 357         if (parent->util) {
 358                 /*
 359                  * Each commit object can cache one origin in that
 360                  * commit.  This is a freestanding copy of origin and
 361                  * not refcounted.
 362                  */
 363                 struct origin *cached = parent->util;
 364                 if (!strcmp(cached->path, origin->path)) {
 365                         /*
 366                          * The same path between origin and its parent
 367                          * without renaming -- the most common case.
 368                          */
 369                         porigin = get_origin(sb, parent, cached->path);
 370
 371                         /*
 372                          * If the origin was newly created (i.e. get_origin
 373                          * would call make_origin if none is found in the
 374                          * scoreboard), it does not know the blob_sha1/mode,
 375                          * so copy it.  Otherwise porigin was in the
 376                          * scoreboard and already knows blob_sha1/mode.
 377                          */
 378                         if (porigin->refcnt == 1) {
 379                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
 380                                 porigin->mode = cached->mode;
 381                         }
 382                         return porigin;
 383                 }
 384                 /* otherwise it was not very useful; free it */
 385                 free(parent->util);
 386                 parent->util = NULL;
 387         }
 388
 389         /* See if the origin->path is different between parent
 390          * and origin first.  Most of the time they are the
 391          * same and diff-tree is fairly efficient about this.
 392          */
 393         diff_setup(&diff_opts);
 394         DIFF_OPT_SET(&diff_opts, RECURSIVE);
 395         diff_opts.detect_rename = 0;
 396         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 397         paths[0] = origin->path;
 398         paths[1] = NULL;
 399
 400         parse_pathspec(&diff_opts.pathspec,
 401                        PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
 402                        PATHSPEC_LITERAL_PATH, "", paths);
 403         diff_setup_done(&diff_opts);
 404
 405         if (is_null_sha1(origin->commit->object.sha1))
 406                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
 407         else
 408                 diff_tree_sha1(parent->tree->object.sha1,
 409                                origin->commit->tree->object.sha1,
 410                                "", &diff_opts);
 411         diffcore_std(&diff_opts);
 412
 413         if (!diff_queued_diff.nr) {
 414                 /* The path is the same as parent */
 415                 porigin = get_origin(sb, parent, origin->path);
 416                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
 417                 porigin->mode = origin->mode;
 418         } else {
 419                 /*
 420                  * Since origin->path is a pathspec, if the parent
 421                  * commit had it as a directory, we will see a whole
 422                  * bunch of deletion of files in the directory that we
 423                  * do not care about.
 424                  */
 425                 int i;
 426                 struct diff_filepair *p = NULL;
 427                 for (i = 0; i < diff_queued_diff.nr; i++) {
 428                         const char *name;
 429                         p = diff_queued_diff.queue[i];
 430                         name = p->one->path ? p->one->path : p->two->path;
 431                         if (!strcmp(name, origin->path))
 432                                 break;
 433                 }
 434                 if (!p)
 435                         die("internal error in blame::find_origin");
 436                 switch (p->status) {
 437                 default:
 438                         die("internal error in blame::find_origin (%c)",
 439                             p->status);
 440                 case 'M':
 441                         porigin = get_origin(sb, parent, origin->path);
 442                         hashcpy(porigin->blob_sha1, p->one->sha1);
 443                         porigin->mode = p->one->mode;
 444                         break;
 445                 case 'A':
 446                 case 'T':
 447                         /* Did not exist in parent, or type changed */
 448                         break;
 449                 }
 450         }
 451         diff_flush(&diff_opts);
 452         free_pathspec(&diff_opts.pathspec);
 453         if (porigin) {
 454                 /*
 455                  * Create a freestanding copy that is not part of
 456                  * the refcounted origin found in the scoreboard, and
 457                  * cache it in the commit.
 458                  */
 459                 struct origin *cached;
 460
 461                 cached = make_origin(porigin->commit, porigin->path);
 462                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
 463                 cached->mode = porigin->mode;
 464                 parent->util = cached;
 465         }
 466         return porigin;
 467 }
 468
 469 /*
 470  * We have an origin -- find the path that corresponds to it in its
 471  * parent and return an origin structure to represent it.
 472  */
 473 static struct origin *find_rename(struct scoreboard *sb,
 474                                   struct commit *parent,
 475                                   struct origin *origin)
 476 {
 477         struct origin *porigin = NULL;
 478         struct diff_options diff_opts;
 479         int i;
 480
 481         diff_setup(&diff_opts);
 482         DIFF_OPT_SET(&diff_opts, RECURSIVE);
 483         diff_opts.detect_rename = DIFF_DETECT_RENAME;
 484         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
 485         diff_opts.single_follow = origin->path;
 486         diff_setup_done(&diff_opts);
 487
 488         if (is_null_sha1(origin->commit->object.sha1))
 489                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
 490         else
 491                 diff_tree_sha1(parent->tree->object.sha1,
 492                                origin->commit->tree->object.sha1,
 493                                "", &diff_opts);
 494         diffcore_std(&diff_opts);
 495
 496         for (i = 0; i < diff_queued_diff.nr; i++) {
 497                 struct diff_filepair *p = diff_queued_diff.queue[i];
 498                 if ((p->status == 'R' || p->status == 'C') &&
 499                     !strcmp(p->two->path, origin->path)) {
 500                         porigin = get_origin(sb, parent, p->one->path);
 501                         hashcpy(porigin->blob_sha1, p->one->sha1);
 502                         porigin->mode = p->one->mode;
 503                         break;
 504                 }
 505         }
 506         diff_flush(&diff_opts);
 507         free_pathspec(&diff_opts.pathspec);
 508         return porigin;
 509 }
 510
 511 /*
 512  * Link in a new blame entry to the scoreboard.  Entries that cover the
 513  * same line range have been removed from the scoreboard previously.
 514  */
 515 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
 516 {
 517         struct blame_entry *ent, *prev = NULL;
 518
 519         origin_incref(e->suspect);
 520
 521         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
 522                 prev = ent;
 523
 524         /* prev, if not NULL, is the last one that is below e */
 525
 526         if (prev) {
 527                 e->next = prev->next;
 528                 prev->next = e;
 529         }
 530         else {
 531                 e->next = sb->ent;
 532                 sb->ent = e;
 533         }
 534 }
 535
 536 /*
 537  * src typically is on-stack; we want to copy the information in it to
 538  * a malloced blame_entry that is already on the linked list of the
 539  * scoreboard.  The origin of dst loses a refcnt while the origin of src
 540  * gains one.
 541  */
 542 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
 543 {
 544         struct blame_entry *n;
 545
 546         n = dst->next;
 547         origin_incref(src->suspect);
 548         origin_decref(dst->suspect);
 549         memcpy(dst, src, sizeof(*src));
 550         dst->next = n;
 551         dst->score = 0;
 552 }
 553
 554 static const char *nth_line(struct scoreboard *sb, long lno)
 555 {
 556         return sb->final_buf + sb->lineno[lno];
 557 }
 558
 559 static const char *nth_line_cb(void *data, long lno)
 560 {
 561         return nth_line((struct scoreboard *)data, lno);
 562 }
 563
 564 /*
 565  * It is known that lines between tlno to same came from parent, and e
 566  * has an overlap with that range.  it also is known that parent's
 567  * line plno corresponds to e's line tlno.
 568  *
 569  *                <---- e ----->
 570  *                   <------>
 571  *                   <------------>
 572  *             <------------>
 573  *             <------------------>
 574  *
 575  * Split e into potentially three parts; before this chunk, the chunk
 576  * to be blamed for the parent, and after that portion.
 577  */
 578 static void split_overlap(struct blame_entry *split,
 579                           struct blame_entry *e,
 580                           int tlno, int plno, int same,
 581                           struct origin *parent)
 582 {
 583         int chunk_end_lno;
 584         memset(split, 0, sizeof(struct blame_entry [3]));
 585
 586         if (e->s_lno < tlno) {
 587                 /* there is a pre-chunk part not blamed on parent */
 588                 split[0].suspect = origin_incref(e->suspect);
 589                 split[0].lno = e->lno;
 590                 split[0].s_lno = e->s_lno;
 591                 split[0].num_lines = tlno - e->s_lno;
 592                 split[1].lno = e->lno + tlno - e->s_lno;
 593                 split[1].s_lno = plno;
 594         }
 595         else {
 596                 split[1].lno = e->lno;
 597                 split[1].s_lno = plno + (e->s_lno - tlno);
 598         }
 599
 600         if (same < e->s_lno + e->num_lines) {
 601                 /* there is a post-chunk part not blamed on parent */
 602                 split[2].suspect = origin_incref(e->suspect);
 603                 split[2].lno = e->lno + (same - e->s_lno);
 604                 split[2].s_lno = e->s_lno + (same - e->s_lno);
 605                 split[2].num_lines = e->s_lno + e->num_lines - same;
 606                 chunk_end_lno = split[2].lno;
 607         }
 608         else
 609                 chunk_end_lno = e->lno + e->num_lines;
 610         split[1].num_lines = chunk_end_lno - split[1].lno;
 611
 612         /*
 613          * if it turns out there is nothing to blame the parent for,
 614          * forget about the splitting.  !split[1].suspect signals this.
 615          */
 616         if (split[1].num_lines < 1)
 617                 return;
 618         split[1].suspect = origin_incref(parent);
 619 }
 620
 621 /*
 622  * split_overlap() divided an existing blame e into up to three parts
 623  * in split.  Adjust the linked list of blames in the scoreboard to
 624  * reflect the split.
 625  */
 626 static void split_blame(struct scoreboard *sb,
 627                         struct blame_entry *split,
 628                         struct blame_entry *e)
 629 {
 630         struct blame_entry *new_entry;
 631
 632         if (split[0].suspect && split[2].suspect) {
 633                 /* The first part (reuse storage for the existing entry e) */
 634                 dup_entry(e, &split[0]);
 635
 636                 /* The last part -- me */
 637                 new_entry = xmalloc(sizeof(*new_entry));
 638                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 639                 add_blame_entry(sb, new_entry);
 640
 641                 /* ... and the middle part -- parent */
 642                 new_entry = xmalloc(sizeof(*new_entry));
 643                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 644                 add_blame_entry(sb, new_entry);
 645         }
 646         else if (!split[0].suspect && !split[2].suspect)
 647                 /*
 648                  * The parent covers the entire area; reuse storage for
 649                  * e and replace it with the parent.
 650                  */
 651                 dup_entry(e, &split[1]);
 652         else if (split[0].suspect) {
 653                 /* me and then parent */
 654                 dup_entry(e, &split[0]);
 655
 656                 new_entry = xmalloc(sizeof(*new_entry));
 657                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
 658                 add_blame_entry(sb, new_entry);
 659         }
 660         else {
 661                 /* parent and then me */
 662                 dup_entry(e, &split[1]);
 663
 664                 new_entry = xmalloc(sizeof(*new_entry));
 665                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
 666                 add_blame_entry(sb, new_entry);
 667         }
 668
 669         if (DEBUG) { /* sanity */
 670                 struct blame_entry *ent;
 671                 int lno = sb->ent->lno, corrupt = 0;
 672
 673                 for (ent = sb->ent; ent; ent = ent->next) {
 674                         if (lno != ent->lno)
 675                                 corrupt = 1;
 676                         if (ent->s_lno < 0)
 677                                 corrupt = 1;
 678                         lno += ent->num_lines;
 679                 }
 680                 if (corrupt) {
 681                         lno = sb->ent->lno;
 682                         for (ent = sb->ent; ent; ent = ent->next) {
 683                                 printf("L %8d l %8d n %8d\n",
 684                                        lno, ent->lno, ent->num_lines);
 685                                 lno = ent->lno + ent->num_lines;
 686                         }
 687                         die("oops");
 688                 }
 689         }
 690 }
 691
 692 /*
 693  * After splitting the blame, the origins used by the
 694  * on-stack blame_entry should lose one refcnt each.
 695  */
 696 static void decref_split(struct blame_entry *split)
 697 {
 698         int i;
 699
 700         for (i = 0; i < 3; i++)
 701                 origin_decref(split[i].suspect);
 702 }
 703
 704 /*
 705  * Helper for blame_chunk().  blame_entry e is known to overlap with
 706  * the patch hunk; split it and pass blame to the parent.
 707  */
 708 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
 709                           int tlno, int plno, int same,
 710                           struct origin *parent)
 711 {
 712         struct blame_entry split[3];
 713
 714         split_overlap(split, e, tlno, plno, same, parent);
 715         if (split[1].suspect)
 716                 split_blame(sb, split, e);
 717         decref_split(split);
 718 }
 719
 720 /*
 721  * Find the line number of the last line the target is suspected for.
 722  */
 723 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
 724 {
 725         struct blame_entry *e;
 726         int last_in_target = -1;
 727
 728         for (e = sb->ent; e; e = e->next) {
 729                 if (e->guilty || e->suspect != target)
 730                         continue;
 731                 if (last_in_target < e->s_lno + e->num_lines)
 732                         last_in_target = e->s_lno + e->num_lines;
 733         }
 734         return last_in_target;
 735 }
 736
 737 /*
 738  * Process one hunk from the patch between the current suspect for
 739  * blame_entry e and its parent.  Find and split the overlap, and
 740  * pass blame to the overlapping part to the parent.
 741  */
 742 static void blame_chunk(struct scoreboard *sb,
 743                         int tlno, int plno, int same,
 744                         struct origin *target, struct origin *parent)
 745 {
 746         struct blame_entry *e;
 747
 748         for (e = sb->ent; e; e = e->next) {
 749                 if (e->guilty || e->suspect != target)
 750                         continue;
 751                 if (same <= e->s_lno)
 752                         continue;
 753                 if (tlno < e->s_lno + e->num_lines)
 754                         blame_overlap(sb, e, tlno, plno, same, parent);
 755         }
 756 }
 757
 758 struct blame_chunk_cb_data {
 759         struct scoreboard *sb;
 760         struct origin *target;
 761         struct origin *parent;
 762         long plno;
 763         long tlno;
 764 };
 765
 766 static int blame_chunk_cb(long start_a, long count_a,
 767                           long start_b, long count_b, void *data)
 768 {
 769         struct blame_chunk_cb_data *d = data;
 770         blame_chunk(d->sb, d->tlno, d->plno, start_b, d->target, d->parent);
 771         d->plno = start_a + count_a;
 772         d->tlno = start_b + count_b;
 773         return 0;
 774 }
 775
 776 /*
 777  * We are looking at the origin 'target' and aiming to pass blame
 778  * for the lines it is suspected to its parent.  Run diff to find
 779  * which lines came from parent and pass blame for them.
 780  */
 781 static int pass_blame_to_parent(struct scoreboard *sb,
 782                                 struct origin *target,
 783                                 struct origin *parent)
 784 {
 785         int last_in_target;
 786         mmfile_t file_p, file_o;
 787         struct blame_chunk_cb_data d;
 788
 789         memset(&d, 0, sizeof(d));
 790         d.sb = sb; d.target = target; d.parent = parent;
 791         last_in_target = find_last_in_target(sb, target);
 792         if (last_in_target < 0)
 793                 return 1; /* nothing remains for this target */
 794
 795         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 796         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
 797         num_get_patch++;
 798
 799         diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
 800         /* The rest (i.e. anything after tlno) are the same as the parent */
 801         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
 802
 803         return 0;
 804 }
 805
 806 /*
 807  * The lines in blame_entry after splitting blames many times can become
 808  * very small and trivial, and at some point it becomes pointless to
 809  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
 810  * ordinary C program, and it is not worth to say it was copied from
 811  * totally unrelated file in the parent.
 812  *
 813  * Compute how trivial the lines in the blame_entry are.
 814  */
 815 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
 816 {
 817         unsigned score;
 818         const char *cp, *ep;
 819
 820         if (e->score)
 821                 return e->score;
 822
 823         score = 1;
 824         cp = nth_line(sb, e->lno);
 825         ep = nth_line(sb, e->lno + e->num_lines);
 826         while (cp < ep) {
 827                 unsigned ch = *((unsigned char *)cp);
 828                 if (isalnum(ch))
 829                         score++;
 830                 cp++;
 831         }
 832         e->score = score;
 833         return score;
 834 }
 835
 836 /*
 837  * best_so_far[] and this[] are both a split of an existing blame_entry
 838  * that passes blame to the parent.  Maintain best_so_far the best split
 839  * so far, by comparing this and best_so_far and copying this into
 840  * bst_so_far as needed.
 841  */
 842 static void copy_split_if_better(struct scoreboard *sb,
 843                                  struct blame_entry *best_so_far,
 844                                  struct blame_entry *this)
 845 {
 846         int i;
 847
 848         if (!this[1].suspect)
 849                 return;
 850         if (best_so_far[1].suspect) {
 851                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
 852                         return;
 853         }
 854
 855         for (i = 0; i < 3; i++)
 856                 origin_incref(this[i].suspect);
 857         decref_split(best_so_far);
 858         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
 859 }
 860
 861 /*
 862  * We are looking at a part of the final image represented by
 863  * ent (tlno and same are offset by ent->s_lno).
 864  * tlno is where we are looking at in the final image.
 865  * up to (but not including) same match preimage.
 866  * plno is where we are looking at in the preimage.
 867  *
 868  * <-------------- final image ---------------------->
 869  *       <------ent------>
 870  *         ^tlno ^same
 871  *    <---------preimage----->
 872  *         ^plno
 873  *
 874  * All line numbers are 0-based.
 875  */
 876 static void handle_split(struct scoreboard *sb,
 877                          struct blame_entry *ent,
 878                          int tlno, int plno, int same,
 879                          struct origin *parent,
 880                          struct blame_entry *split)
 881 {
 882         if (ent->num_lines <= tlno)
 883                 return;
 884         if (tlno < same) {
 885                 struct blame_entry this[3];
 886                 tlno += ent->s_lno;
 887                 same += ent->s_lno;
 888                 split_overlap(this, ent, tlno, plno, same, parent);
 889                 copy_split_if_better(sb, split, this);
 890                 decref_split(this);
 891         }
 892 }
 893
 894 struct handle_split_cb_data {
 895         struct scoreboard *sb;
 896         struct blame_entry *ent;
 897         struct origin *parent;
 898         struct blame_entry *split;
 899         long plno;
 900         long tlno;
 901 };
 902
 903 static int handle_split_cb(long start_a, long count_a,
 904                            long start_b, long count_b, void *data)
 905 {
 906         struct handle_split_cb_data *d = data;
 907         handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
 908                      d->split);
 909         d->plno = start_a + count_a;
 910         d->tlno = start_b + count_b;
 911         return 0;
 912 }
 913
 914 /*
 915  * Find the lines from parent that are the same as ent so that
 916  * we can pass blames to it.  file_p has the blob contents for
 917  * the parent.
 918  */
 919 static void find_copy_in_blob(struct scoreboard *sb,
 920                               struct blame_entry *ent,
 921                               struct origin *parent,
 922                               struct blame_entry *split,
 923                               mmfile_t *file_p)
 924 {
 925         const char *cp;
 926         mmfile_t file_o;
 927         struct handle_split_cb_data d;
 928
 929         memset(&d, 0, sizeof(d));
 930         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
 931         /*
 932          * Prepare mmfile that contains only the lines in ent.
 933          */
 934         cp = nth_line(sb, ent->lno);
 935         file_o.ptr = (char *) cp;
 936         file_o.size = nth_line(sb, ent->lno + ent->num_lines) - cp;
 937
 938         /*
 939          * file_o is a part of final image we are annotating.
 940          * file_p partially may match that image.
 941          */
 942         memset(split, 0, sizeof(struct blame_entry [3]));
 943         diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
 944         /* remainder, if any, all match the preimage */
 945         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
 946 }
 947
 948 /*
 949  * See if lines currently target is suspected for can be attributed to
 950  * parent.
 951  */
 952 static int find_move_in_parent(struct scoreboard *sb,
 953                                struct origin *target,
 954                                struct origin *parent)
 955 {
 956         int last_in_target, made_progress;
 957         struct blame_entry *e, split[3];
 958         mmfile_t file_p;
 959
 960         last_in_target = find_last_in_target(sb, target);
 961         if (last_in_target < 0)
 962                 return 1; /* nothing remains for this target */
 963
 964         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
 965         if (!file_p.ptr)
 966                 return 0;
 967
 968         made_progress = 1;
 969         while (made_progress) {
 970                 made_progress = 0;
 971                 for (e = sb->ent; e; e = e->next) {
 972                         if (e->guilty || e->suspect != target ||
 973                             ent_score(sb, e) < blame_move_score)
 974                                 continue;
 975                         find_copy_in_blob(sb, e, parent, split, &file_p);
 976                         if (split[1].suspect &&
 977                             blame_move_score < ent_score(sb, &split[1])) {
 978                                 split_blame(sb, split, e);
 979                                 made_progress = 1;
 980                         }
 981                         decref_split(split);
 982                 }
 983         }
 984         return 0;
 985 }
 986
 987 struct blame_list {
 988         struct blame_entry *ent;
 989         struct blame_entry split[3];
 990 };
 991
 992 /*
 993  * Count the number of entries the target is suspected for,
 994  * and prepare a list of entry and the best split.
 995  */
 996 static struct blame_list *setup_blame_list(struct scoreboard *sb,
 997                                            struct origin *target,
 998                                            int min_score,
 999                                            int *num_ents_p)
1000 {
1001         struct blame_entry *e;
1002         int num_ents, i;
1003         struct blame_list *blame_list = NULL;
1004
1005         for (e = sb->ent, num_ents = 0; e; e = e->next)
1006                 if (!e->scanned && !e->guilty &&
1007                     e->suspect == target &&
1008                     min_score < ent_score(sb, e))
1009                         num_ents++;
1010         if (num_ents) {
1011                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1012                 for (e = sb->ent, i = 0; e; e = e->next)
1013                         if (!e->scanned && !e->guilty &&
1014                             e->suspect == target &&
1015                             min_score < ent_score(sb, e))
1016                                 blame_list[i++].ent = e;
1017         }
1018         *num_ents_p = num_ents;
1019         return blame_list;
1020 }
1021
1022 /*
1023  * Reset the scanned status on all entries.
1024  */
1025 static void reset_scanned_flag(struct scoreboard *sb)
1026 {
1027         struct blame_entry *e;
1028         for (e = sb->ent; e; e = e->next)
1029                 e->scanned = 0;
1030 }
1031
1032 /*
1033  * For lines target is suspected for, see if we can find code movement
1034  * across file boundary from the parent commit.  porigin is the path
1035  * in the parent we already tried.
1036  */
1037 static int find_copy_in_parent(struct scoreboard *sb,
1038                                struct origin *target,
1039                                struct commit *parent,
1040                                struct origin *porigin,
1041                                int opt)
1042 {
1043         struct diff_options diff_opts;
1044         int i, j;
1045         int retval;
1046         struct blame_list *blame_list;
1047         int num_ents;
1048
1049         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1050         if (!blame_list)
1051                 return 1; /* nothing remains for this target */
1052
1053         diff_setup(&diff_opts);
1054         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1055         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1056
1057         diff_setup_done(&diff_opts);
1058
1059         /* Try "find copies harder" on new path if requested;
1060          * we do not want to use diffcore_rename() actually to
1061          * match things up; find_copies_harder is set only to
1062          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1063          * and this code needs to be after diff_setup_done(), which
1064          * usually makes find-copies-harder imply copy detection.
1065          */
1066         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1067             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1068                 && (!porigin || strcmp(target->path, porigin->path))))
1069                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1070
1071         if (is_null_sha1(target->commit->object.sha1))
1072                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1073         else
1074                 diff_tree_sha1(parent->tree->object.sha1,
1075                                target->commit->tree->object.sha1,
1076                                "", &diff_opts);
1077
1078         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1079                 diffcore_std(&diff_opts);
1080
1081         retval = 0;
1082         while (1) {
1083                 int made_progress = 0;
1084
1085                 for (i = 0; i < diff_queued_diff.nr; i++) {
1086                         struct diff_filepair *p = diff_queued_diff.queue[i];
1087                         struct origin *norigin;
1088                         mmfile_t file_p;
1089                         struct blame_entry this[3];
1090
1091                         if (!DIFF_FILE_VALID(p->one))
1092                                 continue; /* does not exist in parent */
1093                         if (S_ISGITLINK(p->one->mode))
1094                                 continue; /* ignore git links */
1095                         if (porigin && !strcmp(p->one->path, porigin->path))
1096                                 /* find_move already dealt with this path */
1097                                 continue;
1098
1099                         norigin = get_origin(sb, parent, p->one->path);
1100                         hashcpy(norigin->blob_sha1, p->one->sha1);
1101                         norigin->mode = p->one->mode;
1102                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1103                         if (!file_p.ptr)
1104                                 continue;
1105
1106                         for (j = 0; j < num_ents; j++) {
1107                                 find_copy_in_blob(sb, blame_list[j].ent,
1108                                                   norigin, this, &file_p);
1109                                 copy_split_if_better(sb, blame_list[j].split,
1110                                                      this);
1111                                 decref_split(this);
1112                         }
1113                         origin_decref(norigin);
1114                 }
1115
1116                 for (j = 0; j < num_ents; j++) {
1117                         struct blame_entry *split = blame_list[j].split;
1118                         if (split[1].suspect &&
1119                             blame_copy_score < ent_score(sb, &split[1])) {
1120                                 split_blame(sb, split, blame_list[j].ent);
1121                                 made_progress = 1;
1122                         }
1123                         else
1124                                 blame_list[j].ent->scanned = 1;
1125                         decref_split(split);
1126                 }
1127                 free(blame_list);
1128
1129                 if (!made_progress)
1130                         break;
1131                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1132                 if (!blame_list) {
1133                         retval = 1;
1134                         break;
1135                 }
1136         }
1137         reset_scanned_flag(sb);
1138         diff_flush(&diff_opts);
1139         free_pathspec(&diff_opts.pathspec);
1140         return retval;
1141 }
1142
1143 /*
1144  * The blobs of origin and porigin exactly match, so everything
1145  * origin is suspected for can be blamed on the parent.
1146  */
1147 static void pass_whole_blame(struct scoreboard *sb,
1148                              struct origin *origin, struct origin *porigin)
1149 {
1150         struct blame_entry *e;
1151
1152         if (!porigin->file.ptr && origin->file.ptr) {
1153                 /* Steal its file */
1154                 porigin->file = origin->file;
1155                 origin->file.ptr = NULL;
1156         }
1157         for (e = sb->ent; e; e = e->next) {
1158                 if (e->suspect != origin)
1159                         continue;
1160                 origin_incref(porigin);
1161                 origin_decref(e->suspect);
1162                 e->suspect = porigin;
1163         }
1164 }
1165
1166 /*
1167  * We pass blame from the current commit to its parents.  We keep saying
1168  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1169  * exonerate ourselves.
1170  */
1171 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1172 {
1173         if (!reverse)
1174                 return commit->parents;
1175         return lookup_decoration(&revs->children, &commit->object);
1176 }
1177
1178 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1179 {
1180         int cnt;
1181         struct commit_list *l = first_scapegoat(revs, commit);
1182         for (cnt = 0; l; l = l->next)
1183                 cnt++;
1184         return cnt;
1185 }
1186
1187 #define MAXSG 16
1188
1189 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1190 {
1191         struct rev_info *revs = sb->revs;
1192         int i, pass, num_sg;
1193         struct commit *commit = origin->commit;
1194         struct commit_list *sg;
1195         struct origin *sg_buf[MAXSG];
1196         struct origin *porigin, **sg_origin = sg_buf;
1197
1198         num_sg = num_scapegoats(revs, commit);
1199         if (!num_sg)
1200                 goto finish;
1201         else if (num_sg < ARRAY_SIZE(sg_buf))
1202                 memset(sg_buf, 0, sizeof(sg_buf));
1203         else
1204                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1205
1206         /*
1207          * The first pass looks for unrenamed path to optimize for
1208          * common cases, then we look for renames in the second pass.
1209          */
1210         for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1211                 struct origin *(*find)(struct scoreboard *,
1212                                        struct commit *, struct origin *);
1213                 find = pass ? find_rename : find_origin;
1214
1215                 for (i = 0, sg = first_scapegoat(revs, commit);
1216                      i < num_sg && sg;
1217                      sg = sg->next, i++) {
1218                         struct commit *p = sg->item;
1219                         int j, same;
1220
1221                         if (sg_origin[i])
1222                                 continue;
1223                         if (parse_commit(p))
1224                                 continue;
1225                         porigin = find(sb, p, origin);
1226                         if (!porigin)
1227                                 continue;
1228                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1229                                 pass_whole_blame(sb, origin, porigin);
1230                                 origin_decref(porigin);
1231                                 goto finish;
1232                         }
1233                         for (j = same = 0; j < i; j++)
1234                                 if (sg_origin[j] &&
1235                                     !hashcmp(sg_origin[j]->blob_sha1,
1236                                              porigin->blob_sha1)) {
1237                                         same = 1;
1238                                         break;
1239                                 }
1240                         if (!same)
1241                                 sg_origin[i] = porigin;
1242                         else
1243                                 origin_decref(porigin);
1244                 }
1245         }
1246
1247         num_commits++;
1248         for (i = 0, sg = first_scapegoat(revs, commit);
1249              i < num_sg && sg;
1250              sg = sg->next, i++) {
1251                 struct origin *porigin = sg_origin[i];
1252                 if (!porigin)
1253                         continue;
1254                 if (!origin->previous) {
1255                         origin_incref(porigin);
1256                         origin->previous = porigin;
1257                 }
1258                 if (pass_blame_to_parent(sb, origin, porigin))
1259                         goto finish;
1260         }
1261
1262         /*
1263          * Optionally find moves in parents' files.
1264          */
1265         if (opt & PICKAXE_BLAME_MOVE)
1266                 for (i = 0, sg = first_scapegoat(revs, commit);
1267                      i < num_sg && sg;
1268                      sg = sg->next, i++) {
1269                         struct origin *porigin = sg_origin[i];
1270                         if (!porigin)
1271                                 continue;
1272                         if (find_move_in_parent(sb, origin, porigin))
1273                                 goto finish;
1274                 }
1275
1276         /*
1277          * Optionally find copies from parents' files.
1278          */
1279         if (opt & PICKAXE_BLAME_COPY)
1280                 for (i = 0, sg = first_scapegoat(revs, commit);
1281                      i < num_sg && sg;
1282                      sg = sg->next, i++) {
1283                         struct origin *porigin = sg_origin[i];
1284                         if (find_copy_in_parent(sb, origin, sg->item,
1285                                                 porigin, opt))
1286                                 goto finish;
1287                 }
1288
1289  finish:
1290         for (i = 0; i < num_sg; i++) {
1291                 if (sg_origin[i]) {
1292                         drop_origin_blob(sg_origin[i]);
1293                         origin_decref(sg_origin[i]);
1294                 }
1295         }
1296         drop_origin_blob(origin);
1297         if (sg_buf != sg_origin)
1298                 free(sg_origin);
1299 }
1300
1301 /*
1302  * Information on commits, used for output.
1303  */
1304 struct commit_info {
1305         struct strbuf author;
1306         struct strbuf author_mail;
1307         unsigned long author_time;
1308         struct strbuf author_tz;
1309
1310         /* filled only when asked for details */
1311         struct strbuf committer;
1312         struct strbuf committer_mail;
1313         unsigned long committer_time;
1314         struct strbuf committer_tz;
1315
1316         struct strbuf summary;
1317 };
1318
1319 /*
1320  * Parse author/committer line in the commit object buffer
1321  */
1322 static void get_ac_line(const char *inbuf, const char *what,
1323         struct strbuf *name, struct strbuf *mail,
1324         unsigned long *time, struct strbuf *tz)
1325 {
1326         struct ident_split ident;
1327         size_t len, maillen, namelen;
1328         char *tmp, *endp;
1329         const char *namebuf, *mailbuf;
1330
1331         tmp = strstr(inbuf, what);
1332         if (!tmp)
1333                 goto error_out;
1334         tmp += strlen(what);
1335         endp = strchr(tmp, '\n');
1336         if (!endp)
1337                 len = strlen(tmp);
1338         else
1339                 len = endp - tmp;
1340
1341         if (split_ident_line(&ident, tmp, len)) {
1342         error_out:
1343                 /* Ugh */
1344                 tmp = "(unknown)";
1345                 strbuf_addstr(name, tmp);
1346                 strbuf_addstr(mail, tmp);
1347                 strbuf_addstr(tz, tmp);
1348                 *time = 0;
1349                 return;
1350         }
1351
1352         namelen = ident.name_end - ident.name_begin;
1353         namebuf = ident.name_begin;
1354
1355         maillen = ident.mail_end - ident.mail_begin;
1356         mailbuf = ident.mail_begin;
1357
1358         if (ident.date_begin && ident.date_end)
1359                 *time = strtoul(ident.date_begin, NULL, 10);
1360         else
1361                 *time = 0;
1362
1363         if (ident.tz_begin && ident.tz_end)
1364                 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1365         else
1366                 strbuf_addstr(tz, "(unknown)");
1367
1368         /*
1369          * Now, convert both name and e-mail using mailmap
1370          */
1371         map_user(&mailmap, &mailbuf, &maillen,
1372                  &namebuf, &namelen);
1373
1374         strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1375         strbuf_add(name, namebuf, namelen);
1376 }
1377
1378 static void commit_info_init(struct commit_info *ci)
1379 {
1380
1381         strbuf_init(&ci->author, 0);
1382         strbuf_init(&ci->author_mail, 0);
1383         strbuf_init(&ci->author_tz, 0);
1384         strbuf_init(&ci->committer, 0);
1385         strbuf_init(&ci->committer_mail, 0);
1386         strbuf_init(&ci->committer_tz, 0);
1387         strbuf_init(&ci->summary, 0);
1388 }
1389
1390 static void commit_info_destroy(struct commit_info *ci)
1391 {
1392
1393         strbuf_release(&ci->author);
1394         strbuf_release(&ci->author_mail);
1395         strbuf_release(&ci->author_tz);
1396         strbuf_release(&ci->committer);
1397         strbuf_release(&ci->committer_mail);
1398         strbuf_release(&ci->committer_tz);
1399         strbuf_release(&ci->summary);
1400 }
1401
1402 static void get_commit_info(struct commit *commit,
1403                             struct commit_info *ret,
1404                             int detailed)
1405 {
1406         int len;
1407         const char *subject, *encoding;
1408         const char *message;
1409
1410         commit_info_init(ret);
1411
1412         encoding = get_log_output_encoding();
1413         message = logmsg_reencode(commit, NULL, encoding);
1414         get_ac_line(message, "\nauthor ",
1415                     &ret->author, &ret->author_mail,
1416                     &ret->author_time, &ret->author_tz);
1417
1418         if (!detailed) {
1419                 unuse_commit_buffer(commit, message);
1420                 return;
1421         }
1422
1423         get_ac_line(message, "\ncommitter ",
1424                     &ret->committer, &ret->committer_mail,
1425                     &ret->committer_time, &ret->committer_tz);
1426
1427         len = find_commit_subject(message, &subject);
1428         if (len)
1429                 strbuf_add(&ret->summary, subject, len);
1430         else
1431                 strbuf_addf(&ret->summary, "(%s)", sha1_to_hex(commit->object.sha1));
1432
1433         unuse_commit_buffer(commit, message);
1434 }
1435
1436 /*
1437  * To allow LF and other nonportable characters in pathnames,
1438  * they are c-style quoted as needed.
1439  */
1440 static void write_filename_info(const char *path)
1441 {
1442         printf("filename ");
1443         write_name_quoted(path, stdout, '\n');
1444 }
1445
1446 /*
1447  * Porcelain/Incremental format wants to show a lot of details per
1448  * commit.  Instead of repeating this every line, emit it only once,
1449  * the first time each commit appears in the output (unless the
1450  * user has specifically asked for us to repeat).
1451  */
1452 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1453 {
1454         struct commit_info ci;
1455
1456         if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1457                 return 0;
1458
1459         suspect->commit->object.flags |= METAINFO_SHOWN;
1460         get_commit_info(suspect->commit, &ci, 1);
1461         printf("author %s\n", ci.author.buf);
1462         printf("author-mail %s\n", ci.author_mail.buf);
1463         printf("author-time %lu\n", ci.author_time);
1464         printf("author-tz %s\n", ci.author_tz.buf);
1465         printf("committer %s\n", ci.committer.buf);
1466         printf("committer-mail %s\n", ci.committer_mail.buf);
1467         printf("committer-time %lu\n", ci.committer_time);
1468         printf("committer-tz %s\n", ci.committer_tz.buf);
1469         printf("summary %s\n", ci.summary.buf);
1470         if (suspect->commit->object.flags & UNINTERESTING)
1471                 printf("boundary\n");
1472         if (suspect->previous) {
1473                 struct origin *prev = suspect->previous;
1474                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1475                 write_name_quoted(prev->path, stdout, '\n');
1476         }
1477
1478         commit_info_destroy(&ci);
1479
1480         return 1;
1481 }
1482
1483 /*
1484  * The blame_entry is found to be guilty for the range.  Mark it
1485  * as such, and show it in incremental output.
1486  */
1487 static void found_guilty_entry(struct blame_entry *ent)
1488 {
1489         if (ent->guilty)
1490                 return;
1491         ent->guilty = 1;
1492         if (incremental) {
1493                 struct origin *suspect = ent->suspect;
1494
1495                 printf("%s %d %d %d\n",
1496                        sha1_to_hex(suspect->commit->object.sha1),
1497                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1498                 emit_one_suspect_detail(suspect, 0);
1499                 write_filename_info(suspect->path);
1500                 maybe_flush_or_die(stdout, "stdout");
1501         }
1502 }
1503
1504 /*
1505  * The main loop -- while the scoreboard has lines whose true origin
1506  * is still unknown, pick one blame_entry, and allow its current
1507  * suspect to pass blames to its parents.
1508  */
1509 static void assign_blame(struct scoreboard *sb, int opt)
1510 {
1511         struct rev_info *revs = sb->revs;
1512
1513         while (1) {
1514                 struct blame_entry *ent;
1515                 struct commit *commit;
1516                 struct origin *suspect = NULL;
1517
1518                 /* find one suspect to break down */
1519                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1520                         if (!ent->guilty)
1521                                 suspect = ent->suspect;
1522                 if (!suspect)
1523                         return; /* all done */
1524
1525                 /*
1526                  * We will use this suspect later in the loop,
1527                  * so hold onto it in the meantime.
1528                  */
1529                 origin_incref(suspect);
1530                 commit = suspect->commit;
1531                 parse_commit(commit);
1532                 if (reverse ||
1533                     (!(commit->object.flags & UNINTERESTING) &&
1534                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1535                         pass_blame(sb, suspect, opt);
1536                 else {
1537                         commit->object.flags |= UNINTERESTING;
1538                         if (commit->object.parsed)
1539                                 mark_parents_uninteresting(commit);
1540                 }
1541                 /* treat root commit as boundary */
1542                 if (!commit->parents && !show_root)
1543                         commit->object.flags |= UNINTERESTING;
1544
1545                 /* Take responsibility for the remaining entries */
1546                 for (ent = sb->ent; ent; ent = ent->next)
1547                         if (ent->suspect == suspect)
1548                                 found_guilty_entry(ent);
1549                 origin_decref(suspect);
1550
1551                 if (DEBUG) /* sanity */
1552                         sanity_check_refcnt(sb);
1553         }
1554 }
1555
1556 static const char *format_time(unsigned long time, const char *tz_str,
1557                                int show_raw_time)
1558 {
1559         static struct strbuf time_buf = STRBUF_INIT;
1560
1561         strbuf_reset(&time_buf);
1562         if (show_raw_time) {
1563                 strbuf_addf(&time_buf, "%lu %s", time, tz_str);
1564         }
1565         else {
1566                 const char *time_str;
1567                 size_t time_width;
1568                 int tz;
1569                 tz = atoi(tz_str);
1570                 time_str = show_date(time, tz, blame_date_mode);
1571                 strbuf_addstr(&time_buf, time_str);
1572                 /*
1573                  * Add space paddings to time_buf to display a fixed width
1574                  * string, and use time_width for display width calibration.
1575                  */
1576                 for (time_width = utf8_strwidth(time_str);
1577                      time_width < blame_date_width;
1578                      time_width++)
1579                         strbuf_addch(&time_buf, ' ');
1580         }
1581         return time_buf.buf;
1582 }
1583
1584 #define OUTPUT_ANNOTATE_COMPAT  001
1585 #define OUTPUT_LONG_OBJECT_NAME 002
1586 #define OUTPUT_RAW_TIMESTAMP    004
1587 #define OUTPUT_PORCELAIN        010
1588 #define OUTPUT_SHOW_NAME        020
1589 #define OUTPUT_SHOW_NUMBER      040
1590 #define OUTPUT_SHOW_SCORE      0100
1591 #define OUTPUT_NO_AUTHOR       0200
1592 #define OUTPUT_SHOW_EMAIL       0400
1593 #define OUTPUT_LINE_PORCELAIN 01000
1594
1595 static void emit_porcelain_details(struct origin *suspect, int repeat)
1596 {
1597         if (emit_one_suspect_detail(suspect, repeat) ||
1598             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1599                 write_filename_info(suspect->path);
1600 }
1601
1602 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1603                            int opt)
1604 {
1605         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1606         int cnt;
1607         const char *cp;
1608         struct origin *suspect = ent->suspect;
1609         char hex[41];
1610
1611         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1612         printf("%s%c%d %d %d\n",
1613                hex,
1614                ent->guilty ? ' ' : '*', /* purely for debugging */
1615                ent->s_lno + 1,
1616                ent->lno + 1,
1617                ent->num_lines);
1618         emit_porcelain_details(suspect, repeat);
1619
1620         cp = nth_line(sb, ent->lno);
1621         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1622                 char ch;
1623                 if (cnt) {
1624                         printf("%s %d %d\n", hex,
1625                                ent->s_lno + 1 + cnt,
1626                                ent->lno + 1 + cnt);
1627                         if (repeat)
1628                                 emit_porcelain_details(suspect, 1);
1629                 }
1630                 putchar('\t');
1631                 do {
1632                         ch = *cp++;
1633                         putchar(ch);
1634                 } while (ch != '\n' &&
1635                          cp < sb->final_buf + sb->final_buf_size);
1636         }
1637
1638         if (sb->final_buf_size && cp[-1] != '\n')
1639                 putchar('\n');
1640 }
1641
1642 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1643 {
1644         int cnt;
1645         const char *cp;
1646         struct origin *suspect = ent->suspect;
1647         struct commit_info ci;
1648         char hex[41];
1649         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1650
1651         get_commit_info(suspect->commit, &ci, 1);
1652         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1653
1654         cp = nth_line(sb, ent->lno);
1655         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1656                 char ch;
1657                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1658
1659                 if (suspect->commit->object.flags & UNINTERESTING) {
1660                         if (blank_boundary)
1661                                 memset(hex, ' ', length);
1662                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1663                                 length--;
1664                                 putchar('^');
1665                         }
1666                 }
1667
1668                 printf("%.*s", length, hex);
1669                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1670                         const char *name;
1671                         if (opt & OUTPUT_SHOW_EMAIL)
1672                                 name = ci.author_mail.buf;
1673                         else
1674                                 name = ci.author.buf;
1675                         printf("\t(%10s\t%10s\t%d)", name,
1676                                format_time(ci.author_time, ci.author_tz.buf,
1677                                            show_raw_time),
1678                                ent->lno + 1 + cnt);
1679                 } else {
1680                         if (opt & OUTPUT_SHOW_SCORE)
1681                                 printf(" %*d %02d",
1682                                        max_score_digits, ent->score,
1683                                        ent->suspect->refcnt);
1684                         if (opt & OUTPUT_SHOW_NAME)
1685                                 printf(" %-*.*s", longest_file, longest_file,
1686                                        suspect->path);
1687                         if (opt & OUTPUT_SHOW_NUMBER)
1688                                 printf(" %*d", max_orig_digits,
1689                                        ent->s_lno + 1 + cnt);
1690
1691                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1692                                 const char *name;
1693                                 int pad;
1694                                 if (opt & OUTPUT_SHOW_EMAIL)
1695                                         name = ci.author_mail.buf;
1696                                 else
1697                                         name = ci.author.buf;
1698                                 pad = longest_author - utf8_strwidth(name);
1699                                 printf(" (%s%*s %10s",
1700                                        name, pad, "",
1701                                        format_time(ci.author_time,
1702                                                    ci.author_tz.buf,
1703                                                    show_raw_time));
1704                         }
1705                         printf(" %*d) ",
1706                                max_digits, ent->lno + 1 + cnt);
1707                 }
1708                 do {
1709                         ch = *cp++;
1710                         putchar(ch);
1711                 } while (ch != '\n' &&
1712                          cp < sb->final_buf + sb->final_buf_size);
1713         }
1714
1715         if (sb->final_buf_size && cp[-1] != '\n')
1716                 putchar('\n');
1717
1718         commit_info_destroy(&ci);
1719 }
1720
1721 static void output(struct scoreboard *sb, int option)
1722 {
1723         struct blame_entry *ent;
1724
1725         if (option & OUTPUT_PORCELAIN) {
1726                 for (ent = sb->ent; ent; ent = ent->next) {
1727                         struct blame_entry *oth;
1728                         struct origin *suspect = ent->suspect;
1729                         struct commit *commit = suspect->commit;
1730                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1731                                 continue;
1732                         for (oth = ent->next; oth; oth = oth->next) {
1733                                 if ((oth->suspect->commit != commit) ||
1734                                     !strcmp(oth->suspect->path, suspect->path))
1735                                         continue;
1736                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1737                                 break;
1738                         }
1739                 }
1740         }
1741
1742         for (ent = sb->ent; ent; ent = ent->next) {
1743                 if (option & OUTPUT_PORCELAIN)
1744                         emit_porcelain(sb, ent, option);
1745                 else {
1746                         emit_other(sb, ent, option);
1747                 }
1748         }
1749 }
1750
1751 /*
1752  * To allow quick access to the contents of nth line in the
1753  * final image, prepare an index in the scoreboard.
1754  */
1755 static int prepare_lines(struct scoreboard *sb)
1756 {
1757         const char *buf = sb->final_buf;
1758         unsigned long len = sb->final_buf_size;
1759         const char *end = buf + len;
1760         const char *p;
1761         int *lineno;
1762         int num = 0, incomplete = 0;
1763
1764         for (p = buf;;) {
1765                 p = memchr(p, '\n', end - p);
1766                 if (p) {
1767                         p++;
1768                         num++;
1769                         continue;
1770                 }
1771                 break;
1772         }
1773
1774         if (len && end[-1] != '\n')
1775                 incomplete++; /* incomplete line at the end */
1776
1777         sb->lineno = xmalloc(sizeof(*sb->lineno) * (num + incomplete + 1));
1778         lineno = sb->lineno;
1779
1780         *lineno++ = 0;
1781         for (p = buf;;) {
1782                 p = memchr(p, '\n', end - p);
1783                 if (p) {
1784                         p++;
1785                         *lineno++ = p - buf;
1786                         continue;
1787                 }
1788                 break;
1789         }
1790
1791         if (incomplete)
1792                 *lineno++ = len;
1793
1794         sb->num_lines = num + incomplete;
1795         return sb->num_lines;
1796 }
1797
1798 /*
1799  * Add phony grafts for use with -S; this is primarily to
1800  * support git's cvsserver that wants to give a linear history
1801  * to its clients.
1802  */
1803 static int read_ancestry(const char *graft_file)
1804 {
1805         FILE *fp = fopen(graft_file, "r");
1806         struct strbuf buf = STRBUF_INIT;
1807         if (!fp)
1808                 return -1;
1809         while (!strbuf_getwholeline(&buf, fp, '\n')) {
1810                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1811                 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
1812                 if (graft)
1813                         register_commit_graft(graft, 0);
1814         }
1815         fclose(fp);
1816         strbuf_release(&buf);
1817         return 0;
1818 }
1819
1820 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1821 {
1822         const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1823                                               auto_abbrev);
1824         int len = strlen(uniq);
1825         if (auto_abbrev < len)
1826                 return len;
1827         return auto_abbrev;
1828 }
1829
1830 /*
1831  * How many columns do we need to show line numbers, authors,
1832  * and filenames?
1833  */
1834 static void find_alignment(struct scoreboard *sb, int *option)
1835 {
1836         int longest_src_lines = 0;
1837         int longest_dst_lines = 0;
1838         unsigned largest_score = 0;
1839         struct blame_entry *e;
1840         int compute_auto_abbrev = (abbrev < 0);
1841         int auto_abbrev = default_abbrev;
1842
1843         for (e = sb->ent; e; e = e->next) {
1844                 struct origin *suspect = e->suspect;
1845                 struct commit_info ci;
1846                 int num;
1847
1848                 if (compute_auto_abbrev)
1849                         auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1850                 if (strcmp(suspect->path, sb->path))
1851                         *option |= OUTPUT_SHOW_NAME;
1852                 num = strlen(suspect->path);
1853                 if (longest_file < num)
1854                         longest_file = num;
1855                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1856                         suspect->commit->object.flags |= METAINFO_SHOWN;
1857                         get_commit_info(suspect->commit, &ci, 1);
1858                         if (*option & OUTPUT_SHOW_EMAIL)
1859                                 num = utf8_strwidth(ci.author_mail.buf);
1860                         else
1861                                 num = utf8_strwidth(ci.author.buf);
1862                         if (longest_author < num)
1863                                 longest_author = num;
1864                 }
1865                 num = e->s_lno + e->num_lines;
1866                 if (longest_src_lines < num)
1867                         longest_src_lines = num;
1868                 num = e->lno + e->num_lines;
1869                 if (longest_dst_lines < num)
1870                         longest_dst_lines = num;
1871                 if (largest_score < ent_score(sb, e))
1872                         largest_score = ent_score(sb, e);
1873
1874                 commit_info_destroy(&ci);
1875         }
1876         max_orig_digits = decimal_width(longest_src_lines);
1877         max_digits = decimal_width(longest_dst_lines);
1878         max_score_digits = decimal_width(largest_score);
1879
1880         if (compute_auto_abbrev)
1881                 /* one more abbrev length is needed for the boundary commit */
1882                 abbrev = auto_abbrev + 1;
1883 }
1884
1885 /*
1886  * For debugging -- origin is refcounted, and this asserts that
1887  * we do not underflow.
1888  */
1889 static void sanity_check_refcnt(struct scoreboard *sb)
1890 {
1891         int baa = 0;
1892         struct blame_entry *ent;
1893
1894         for (ent = sb->ent; ent; ent = ent->next) {
1895                 /* Nobody should have zero or negative refcnt */
1896                 if (ent->suspect->refcnt <= 0) {
1897                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1898                                 ent->suspect->path,
1899                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1900                                 ent->suspect->refcnt);
1901                         baa = 1;
1902                 }
1903         }
1904         if (baa) {
1905                 int opt = 0160;
1906                 find_alignment(sb, &opt);
1907                 output(sb, opt);
1908                 die("Baa %d!", baa);
1909         }
1910 }
1911
1912 /*
1913  * Used for the command line parsing; check if the path exists
1914  * in the working tree.
1915  */
1916 static int has_string_in_work_tree(const char *path)
1917 {
1918         struct stat st;
1919         return !lstat(path, &st);
1920 }
1921
1922 static unsigned parse_score(const char *arg)
1923 {
1924         char *end;
1925         unsigned long score = strtoul(arg, &end, 10);
1926         if (*end)
1927                 return 0;
1928         return score;
1929 }
1930
1931 static const char *add_prefix(const char *prefix, const char *path)
1932 {
1933         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1934 }
1935
1936 static int git_blame_config(const char *var, const char *value, void *cb)
1937 {
1938         if (!strcmp(var, "blame.showroot")) {
1939                 show_root = git_config_bool(var, value);
1940                 return 0;
1941         }
1942         if (!strcmp(var, "blame.blankboundary")) {
1943                 blank_boundary = git_config_bool(var, value);
1944                 return 0;
1945         }
1946         if (!strcmp(var, "blame.date")) {
1947                 if (!value)
1948                         return config_error_nonbool(var);
1949                 blame_date_mode = parse_date_format(value);
1950                 return 0;
1951         }
1952
1953         if (userdiff_config(var, value) < 0)
1954                 return -1;
1955
1956         return git_default_config(var, value, cb);
1957 }
1958
1959 static void verify_working_tree_path(struct commit *work_tree, const char *path)
1960 {
1961         struct commit_list *parents;
1962
1963         for (parents = work_tree->parents; parents; parents = parents->next) {
1964                 const unsigned char *commit_sha1 = parents->item->object.sha1;
1965                 unsigned char blob_sha1[20];
1966                 unsigned mode;
1967
1968                 if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
1969                     sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
1970                         return;
1971         }
1972         die("no such path '%s' in HEAD", path);
1973 }
1974
1975 static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
1976 {
1977         struct commit *parent;
1978
1979         parent = lookup_commit_reference(sha1);
1980         if (!parent)
1981                 die("no such commit %s", sha1_to_hex(sha1));
1982         return &commit_list_insert(parent, tail)->next;
1983 }
1984
1985 static void append_merge_parents(struct commit_list **tail)
1986 {
1987         int merge_head;
1988         const char *merge_head_file = git_path("MERGE_HEAD");
1989         struct strbuf line = STRBUF_INIT;
1990
1991         merge_head = open(merge_head_file, O_RDONLY);
1992         if (merge_head < 0) {
1993                 if (errno == ENOENT)
1994                         return;
1995                 die("cannot open '%s' for reading", merge_head_file);
1996         }
1997
1998         while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
1999                 unsigned char sha1[20];
2000                 if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2001                         die("unknown line in '%s': %s", merge_head_file, line.buf);
2002                 tail = append_parent(tail, sha1);
2003         }
2004         close(merge_head);
2005         strbuf_release(&line);
2006 }
2007
2008 /*
2009  * This isn't as simple as passing sb->buf and sb->len, because we
2010  * want to transfer ownership of the buffer to the commit (so we
2011  * must use detach).
2012  */
2013 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
2014 {
2015         size_t len;
2016         void *buf = strbuf_detach(sb, &len);
2017         set_commit_buffer(c, buf, len);
2018 }
2019
2020 /*
2021  * Prepare a dummy commit that represents the work tree (or staged) item.
2022  * Note that annotating work tree item never works in the reverse.
2023  */
2024 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2025                                                const char *path,
2026                                                const char *contents_from)
2027 {
2028         struct commit *commit;
2029         struct origin *origin;
2030         struct commit_list **parent_tail, *parent;
2031         unsigned char head_sha1[20];
2032         struct strbuf buf = STRBUF_INIT;
2033         const char *ident;
2034         time_t now;
2035         int size, len;
2036         struct cache_entry *ce;
2037         unsigned mode;
2038         struct strbuf msg = STRBUF_INIT;
2039
2040         time(&now);
2041         commit = alloc_commit_node();
2042         commit->object.parsed = 1;
2043         commit->date = now;
2044         commit->object.type = OBJ_COMMIT;
2045         parent_tail = &commit->parents;
2046
2047         if (!resolve_ref_unsafe("HEAD", head_sha1, 1, NULL))
2048                 die("no such ref: HEAD");
2049
2050         parent_tail = append_parent(parent_tail, head_sha1);
2051         append_merge_parents(parent_tail);
2052         verify_working_tree_path(commit, path);
2053
2054         origin = make_origin(commit, path);
2055
2056         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2057         strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2058         for (parent = commit->parents; parent; parent = parent->next)
2059                 strbuf_addf(&msg, "parent %s\n",
2060                             sha1_to_hex(parent->item->object.sha1));
2061         strbuf_addf(&msg,
2062                     "author %s\n"
2063                     "committer %s\n\n"
2064                     "Version of %s from %s\n",
2065                     ident, ident, path,
2066                     (!contents_from ? path :
2067                      (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2068         set_commit_buffer_from_strbuf(commit, &msg);
2069
2070         if (!contents_from || strcmp("-", contents_from)) {
2071                 struct stat st;
2072                 const char *read_from;
2073                 char *buf_ptr;
2074                 unsigned long buf_len;
2075
2076                 if (contents_from) {
2077                         if (stat(contents_from, &st) < 0)
2078                                 die_errno("Cannot stat '%s'", contents_from);
2079                         read_from = contents_from;
2080                 }
2081                 else {
2082                         if (lstat(path, &st) < 0)
2083                                 die_errno("Cannot lstat '%s'", path);
2084                         read_from = path;
2085                 }
2086                 mode = canon_mode(st.st_mode);
2087
2088                 switch (st.st_mode & S_IFMT) {
2089                 case S_IFREG:
2090                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2091                             textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2092                                 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2093                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2094                                 die_errno("cannot open or read '%s'", read_from);
2095                         break;
2096                 case S_IFLNK:
2097                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2098                                 die_errno("cannot readlink '%s'", read_from);
2099                         break;
2100                 default:
2101                         die("unsupported file type %s", read_from);
2102                 }
2103         }
2104         else {
2105                 /* Reading from stdin */
2106                 mode = 0;
2107                 if (strbuf_read(&buf, 0, 0) < 0)
2108                         die_errno("failed to read from stdin");
2109         }
2110         origin->file.ptr = buf.buf;
2111         origin->file.size = buf.len;
2112         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2113         commit->util = origin;
2114
2115         /*
2116          * Read the current index, replace the path entry with
2117          * origin->blob_sha1 without mucking with its mode or type
2118          * bits; we are not going to write this index out -- we just
2119          * want to run "diff-index --cached".
2120          */
2121         discard_cache();
2122         read_cache();
2123
2124         len = strlen(path);
2125         if (!mode) {
2126                 int pos = cache_name_pos(path, len);
2127                 if (0 <= pos)
2128                         mode = active_cache[pos]->ce_mode;
2129                 else
2130                         /* Let's not bother reading from HEAD tree */
2131                         mode = S_IFREG | 0644;
2132         }
2133         size = cache_entry_size(len);
2134         ce = xcalloc(1, size);
2135         hashcpy(ce->sha1, origin->blob_sha1);
2136         memcpy(ce->name, path, len);
2137         ce->ce_flags = create_ce_flags(0);
2138         ce->ce_namelen = len;
2139         ce->ce_mode = create_ce_mode(mode);
2140         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2141
2142         /*
2143          * We are not going to write this out, so this does not matter
2144          * right now, but someday we might optimize diff-index --cached
2145          * with cache-tree information.
2146          */
2147         cache_tree_invalidate_path(active_cache_tree, path);
2148
2149         return commit;
2150 }
2151
2152 static const char *prepare_final(struct scoreboard *sb)
2153 {
2154         int i;
2155         const char *final_commit_name = NULL;
2156         struct rev_info *revs = sb->revs;
2157
2158         /*
2159          * There must be one and only one positive commit in the
2160          * revs->pending array.
2161          */
2162         for (i = 0; i < revs->pending.nr; i++) {
2163                 struct object *obj = revs->pending.objects[i].item;
2164                 if (obj->flags & UNINTERESTING)
2165                         continue;
2166                 while (obj->type == OBJ_TAG)
2167                         obj = deref_tag(obj, NULL, 0);
2168                 if (obj->type != OBJ_COMMIT)
2169                         die("Non commit %s?", revs->pending.objects[i].name);
2170                 if (sb->final)
2171                         die("More than one commit to dig from %s and %s?",
2172                             revs->pending.objects[i].name,
2173                             final_commit_name);
2174                 sb->final = (struct commit *) obj;
2175                 final_commit_name = revs->pending.objects[i].name;
2176         }
2177         return final_commit_name;
2178 }
2179
2180 static const char *prepare_initial(struct scoreboard *sb)
2181 {
2182         int i;
2183         const char *final_commit_name = NULL;
2184         struct rev_info *revs = sb->revs;
2185
2186         /*
2187          * There must be one and only one negative commit, and it must be
2188          * the boundary.
2189          */
2190         for (i = 0; i < revs->pending.nr; i++) {
2191                 struct object *obj = revs->pending.objects[i].item;
2192                 if (!(obj->flags & UNINTERESTING))
2193                         continue;
2194                 while (obj->type == OBJ_TAG)
2195                         obj = deref_tag(obj, NULL, 0);
2196                 if (obj->type != OBJ_COMMIT)
2197                         die("Non commit %s?", revs->pending.objects[i].name);
2198                 if (sb->final)
2199                         die("More than one commit to dig down to %s and %s?",
2200                             revs->pending.objects[i].name,
2201                             final_commit_name);
2202                 sb->final = (struct commit *) obj;
2203                 final_commit_name = revs->pending.objects[i].name;
2204         }
2205         if (!final_commit_name)
2206                 die("No commit to dig down to?");
2207         return final_commit_name;
2208 }
2209
2210 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2211 {
2212         int *opt = option->value;
2213
2214         /*
2215          * -C enables copy from removed files;
2216          * -C -C enables copy from existing files, but only
2217          *       when blaming a new file;
2218          * -C -C -C enables copy from existing files for
2219          *          everybody
2220          */
2221         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2222                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2223         if (*opt & PICKAXE_BLAME_COPY)
2224                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2225         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2226
2227         if (arg)
2228                 blame_copy_score = parse_score(arg);
2229         return 0;
2230 }
2231
2232 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2233 {
2234         int *opt = option->value;
2235
2236         *opt |= PICKAXE_BLAME_MOVE;
2237
2238         if (arg)
2239                 blame_move_score = parse_score(arg);
2240         return 0;
2241 }
2242
2243 int cmd_blame(int argc, const char **argv, const char *prefix)
2244 {
2245         struct rev_info revs;
2246         const char *path;
2247         struct scoreboard sb;
2248         struct origin *o;
2249         struct blame_entry *ent = NULL;
2250         long dashdash_pos, lno;
2251         const char *final_commit_name = NULL;
2252         enum object_type type;
2253
2254         static struct string_list range_list;
2255         static int output_option = 0, opt = 0;
2256         static int show_stats = 0;
2257         static const char *revs_file = NULL;
2258         static const char *contents_from = NULL;
2259         static const struct option options[] = {
2260                 OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2261                 OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2262                 OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2263                 OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2264                 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2265                 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2266                 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2267                 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2268                 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2269                 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2270                 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2271                 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2272                 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2273                 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2274                 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2275                 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2276                 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2277                 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2278                 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2279                 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2280                 OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2281                 OPT__ABBREV(&abbrev),
2282                 OPT_END()
2283         };
2284
2285         struct parse_opt_ctx_t ctx;
2286         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2287         struct range_set ranges;
2288         unsigned int range_i;
2289         long anchor;
2290
2291         git_config(git_blame_config, NULL);
2292         init_revisions(&revs, NULL);
2293         revs.date_mode = blame_date_mode;
2294         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2295         DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2296
2297         save_commit_buffer = 0;
2298         dashdash_pos = 0;
2299
2300         parse_options_start(&ctx, argc, argv, prefix, options,
2301                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2302         for (;;) {
2303                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2304                 case PARSE_OPT_HELP:
2305                         exit(129);
2306                 case PARSE_OPT_DONE:
2307                         if (ctx.argv[0])
2308                                 dashdash_pos = ctx.cpidx;
2309                         goto parse_done;
2310                 }
2311
2312                 if (!strcmp(ctx.argv[0], "--reverse")) {
2313                         ctx.argv[0] = "--children";
2314                         reverse = 1;
2315                 }
2316                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2317         }
2318 parse_done:
2319         no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2320         DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2321         argc = parse_options_end(&ctx);
2322
2323         if (0 < abbrev)
2324                 /* one more abbrev length is needed for the boundary commit */
2325                 abbrev++;
2326
2327         if (revs_file && read_ancestry(revs_file))
2328                 die_errno("reading graft file '%s' failed", revs_file);
2329
2330         if (cmd_is_annotate) {
2331                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2332                 blame_date_mode = DATE_ISO8601;
2333         } else {
2334                 blame_date_mode = revs.date_mode;
2335         }
2336
2337         /* The maximum width used to show the dates */
2338         switch (blame_date_mode) {
2339         case DATE_RFC2822:
2340                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2341                 break;
2342         case DATE_ISO8601:
2343                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2344                 break;
2345         case DATE_RAW:
2346                 blame_date_width = sizeof("1161298804 -0700");
2347                 break;
2348         case DATE_SHORT:
2349                 blame_date_width = sizeof("2006-10-19");
2350                 break;
2351         case DATE_RELATIVE:
2352                 /* TRANSLATORS: This string is used to tell us the maximum
2353                    display width for a relative timestamp in "git blame"
2354                    output.  For C locale, "4 years, 11 months ago", which
2355                    takes 22 places, is the longest among various forms of
2356                    relative timestamps, but your language may need more or
2357                    fewer display columns. */
2358                 blame_date_width = utf8_strwidth(_("4 years, 11 months ago")) + 1; /* add the null */
2359                 break;
2360         case DATE_LOCAL:
2361         case DATE_NORMAL:
2362                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2363                 break;
2364         }
2365         blame_date_width -= 1; /* strip the null */
2366
2367         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2368                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2369                         PICKAXE_BLAME_COPY_HARDER);
2370
2371         if (!blame_move_score)
2372                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2373         if (!blame_copy_score)
2374                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2375
2376         /*
2377          * We have collected options unknown to us in argv[1..unk]
2378          * which are to be passed to revision machinery if we are
2379          * going to do the "bottom" processing.
2380          *
2381          * The remaining are:
2382          *
2383          * (1) if dashdash_pos != 0, it is either
2384          *     "blame [revisions] -- <path>" or
2385          *     "blame -- <path> <rev>"
2386          *
2387          * (2) otherwise, it is one of the two:
2388          *     "blame [revisions] <path>"
2389          *     "blame <path> <rev>"
2390          *
2391          * Note that we must strip out <path> from the arguments: we do not
2392          * want the path pruning but we may want "bottom" processing.
2393          */
2394         if (dashdash_pos) {
2395                 switch (argc - dashdash_pos - 1) {
2396                 case 2: /* (1b) */
2397                         if (argc != 4)
2398                                 usage_with_options(blame_opt_usage, options);
2399                         /* reorder for the new way: <rev> -- <path> */
2400                         argv[1] = argv[3];
2401                         argv[3] = argv[2];
2402                         argv[2] = "--";
2403                         /* FALLTHROUGH */
2404                 case 1: /* (1a) */
2405                         path = add_prefix(prefix, argv[--argc]);
2406                         argv[argc] = NULL;
2407                         break;
2408                 default:
2409                         usage_with_options(blame_opt_usage, options);
2410                 }
2411         } else {
2412                 if (argc < 2)
2413                         usage_with_options(blame_opt_usage, options);
2414                 path = add_prefix(prefix, argv[argc - 1]);
2415                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2416                         path = add_prefix(prefix, argv[1]);
2417                         argv[1] = argv[2];
2418                 }
2419                 argv[argc - 1] = "--";
2420
2421                 setup_work_tree();
2422                 if (!has_string_in_work_tree(path))
2423                         die_errno("cannot stat path '%s'", path);
2424         }
2425
2426         revs.disable_stdin = 1;
2427         setup_revisions(argc, argv, &revs, NULL);
2428         memset(&sb, 0, sizeof(sb));
2429
2430         sb.revs = &revs;
2431         if (!reverse)
2432                 final_commit_name = prepare_final(&sb);
2433         else if (contents_from)
2434                 die("--contents and --children do not blend well.");
2435         else
2436                 final_commit_name = prepare_initial(&sb);
2437
2438         if (!sb.final) {
2439                 /*
2440                  * "--not A B -- path" without anything positive;
2441                  * do not default to HEAD, but use the working tree
2442                  * or "--contents".
2443                  */
2444                 setup_work_tree();
2445                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2446                                                     path, contents_from);
2447                 add_pending_object(&revs, &(sb.final->object), ":");
2448         }
2449         else if (contents_from)
2450                 die("Cannot use --contents with final commit object name");
2451
2452         /*
2453          * If we have bottom, this will mark the ancestors of the
2454          * bottom commits we would reach while traversing as
2455          * uninteresting.
2456          */
2457         if (prepare_revision_walk(&revs))
2458                 die("revision walk setup failed");
2459
2460         if (is_null_sha1(sb.final->object.sha1)) {
2461                 char *buf;
2462                 o = sb.final->util;
2463                 buf = xmalloc(o->file.size + 1);
2464                 memcpy(buf, o->file.ptr, o->file.size + 1);
2465                 sb.final_buf = buf;
2466                 sb.final_buf_size = o->file.size;
2467         }
2468         else {
2469                 o = get_origin(&sb, sb.final, path);
2470                 if (fill_blob_sha1_and_mode(o))
2471                         die("no such path %s in %s", path, final_commit_name);
2472
2473                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2474                     textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2475                                     &sb.final_buf_size))
2476                         ;
2477                 else
2478                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2479                                                       &sb.final_buf_size);
2480
2481                 if (!sb.final_buf)
2482                         die("Cannot read blob %s for path %s",
2483                             sha1_to_hex(o->blob_sha1),
2484                             path);
2485         }
2486         num_read_blob++;
2487         lno = prepare_lines(&sb);
2488
2489         if (lno && !range_list.nr)
2490                 string_list_append(&range_list, xstrdup("1"));
2491
2492         anchor = 1;
2493         range_set_init(&ranges, range_list.nr);
2494         for (range_i = 0; range_i < range_list.nr; ++range_i) {
2495                 long bottom, top;
2496                 if (parse_range_arg(range_list.items[range_i].string,
2497                                     nth_line_cb, &sb, lno, anchor,
2498                                     &bottom, &top, sb.path))
2499                         usage(blame_usage);
2500                 if (lno < top || ((lno || bottom) && lno < bottom))
2501                         die("file %s has only %lu lines", path, lno);
2502                 if (bottom < 1)
2503                         bottom = 1;
2504                 if (top < 1)
2505                         top = lno;
2506                 bottom--;
2507                 range_set_append_unsafe(&ranges, bottom, top);
2508                 anchor = top + 1;
2509         }
2510         sort_and_merge_range_set(&ranges);
2511
2512         for (range_i = ranges.nr; range_i > 0; --range_i) {
2513                 const struct range *r = &ranges.ranges[range_i - 1];
2514                 long bottom = r->start;
2515                 long top = r->end;
2516                 struct blame_entry *next = ent;
2517                 ent = xcalloc(1, sizeof(*ent));
2518                 ent->lno = bottom;
2519                 ent->num_lines = top - bottom;
2520                 ent->suspect = o;
2521                 ent->s_lno = bottom;
2522                 ent->next = next;
2523                 origin_incref(o);
2524         }
2525         origin_decref(o);
2526
2527         range_set_release(&ranges);
2528         string_list_clear(&range_list, 0);
2529
2530         sb.ent = ent;
2531         sb.path = path;
2532
2533         read_mailmap(&mailmap, NULL);
2534
2535         if (!incremental)
2536                 setup_pager();
2537
2538         assign_blame(&sb, opt);
2539
2540         if (incremental)
2541                 return 0;
2542
2543         coalesce(&sb);
2544
2545         if (!(output_option & OUTPUT_PORCELAIN))
2546                 find_alignment(&sb, &output_option);
2547
2548         output(&sb, output_option);
2549         free((void *)sb.final_buf);
2550         for (ent = sb.ent; ent; ) {
2551                 struct blame_entry *e = ent->next;
2552                 free(ent);
2553                 ent = e;
2554         }
2555
2556         if (show_stats) {
2557                 printf("num read blob: %d\n", num_read_blob);
2558                 printf("num get patch: %d\n", num_get_patch);
2559                 printf("num commits: %d\n", num_commits);
2560         }
2561         return 0;
2562 }