gitweb: use new Git::Repo API, and add optional caching
[git/gitweb-caching.git] / builtin-blame.c
blobe4d12de8a994434058745439a86929c0d556a3ec
1 /*
2 * Pickaxe
4 * Copyright (c) 2006, Junio C Hamano
5 */
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"
23 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
25 static const char *blame_opt_usage[] = {
26 blame_usage,
27 "",
28 "[rev-opts] are documented in git-rev-list(1)",
29 NULL
32 static int longest_file;
33 static int longest_author;
34 static int max_orig_digits;
35 static int max_digits;
36 static int max_score_digits;
37 static int show_root;
38 static int reverse;
39 static int blank_boundary;
40 static int incremental;
41 static int cmd_is_annotate;
42 static int xdl_opts = XDF_NEED_MINIMAL;
43 static struct string_list mailmap;
45 #ifndef DEBUG
46 #define DEBUG 0
47 #endif
49 /* stats */
50 static int num_read_blob;
51 static int num_get_patch;
52 static int num_commits;
54 #define PICKAXE_BLAME_MOVE 01
55 #define PICKAXE_BLAME_COPY 02
56 #define PICKAXE_BLAME_COPY_HARDER 04
57 #define PICKAXE_BLAME_COPY_HARDEST 010
60 * blame for a blame_entry with score lower than these thresholds
61 * is not passed to the parent using move/copy logic.
63 static unsigned blame_move_score;
64 static unsigned blame_copy_score;
65 #define BLAME_DEFAULT_MOVE_SCORE 20
66 #define BLAME_DEFAULT_COPY_SCORE 40
68 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
69 #define METAINFO_SHOWN (1u<<12)
70 #define MORE_THAN_ONE_PATH (1u<<13)
73 * One blob in a commit that is being suspected
75 struct origin {
76 int refcnt;
77 struct commit *commit;
78 mmfile_t file;
79 unsigned char blob_sha1[20];
80 char path[FLEX_ARRAY];
84 * Given an origin, prepare mmfile_t structure to be used by the
85 * diff machinery
87 static void fill_origin_blob(struct origin *o, mmfile_t *file)
89 if (!o->file.ptr) {
90 enum object_type type;
91 num_read_blob++;
92 file->ptr = read_sha1_file(o->blob_sha1, &type,
93 (unsigned long *)(&(file->size)));
94 if (!file->ptr)
95 die("Cannot read blob %s for path %s",
96 sha1_to_hex(o->blob_sha1),
97 o->path);
98 o->file = *file;
100 else
101 *file = o->file;
105 * Origin is refcounted and usually we keep the blob contents to be
106 * reused.
108 static inline struct origin *origin_incref(struct origin *o)
110 if (o)
111 o->refcnt++;
112 return o;
115 static void origin_decref(struct origin *o)
117 if (o && --o->refcnt <= 0) {
118 free(o->file.ptr);
119 free(o);
123 static void drop_origin_blob(struct origin *o)
125 if (o->file.ptr) {
126 free(o->file.ptr);
127 o->file.ptr = NULL;
132 * Each group of lines is described by a blame_entry; it can be split
133 * as we pass blame to the parents. They form a linked list in the
134 * scoreboard structure, sorted by the target line number.
136 struct blame_entry {
137 struct blame_entry *prev;
138 struct blame_entry *next;
140 /* the first line of this group in the final image;
141 * internally all line numbers are 0 based.
143 int lno;
145 /* how many lines this group has */
146 int num_lines;
148 /* the commit that introduced this group into the final image */
149 struct origin *suspect;
151 /* true if the suspect is truly guilty; false while we have not
152 * checked if the group came from one of its parents.
154 char guilty;
156 /* true if the entry has been scanned for copies in the current parent
158 char scanned;
160 /* the line number of the first line of this group in the
161 * suspect's file; internally all line numbers are 0 based.
163 int s_lno;
165 /* how significant this entry is -- cached to avoid
166 * scanning the lines over and over.
168 unsigned score;
172 * The current state of the blame assignment.
174 struct scoreboard {
175 /* the final commit (i.e. where we started digging from) */
176 struct commit *final;
177 struct rev_info *revs;
178 const char *path;
181 * The contents in the final image.
182 * Used by many functions to obtain contents of the nth line,
183 * indexed with scoreboard.lineno[blame_entry.lno].
185 const char *final_buf;
186 unsigned long final_buf_size;
188 /* linked list of blames */
189 struct blame_entry *ent;
191 /* look-up a line in the final buffer */
192 int num_lines;
193 int *lineno;
196 static inline int same_suspect(struct origin *a, struct origin *b)
198 if (a == b)
199 return 1;
200 if (a->commit != b->commit)
201 return 0;
202 return !strcmp(a->path, b->path);
205 static void sanity_check_refcnt(struct scoreboard *);
208 * If two blame entries that are next to each other came from
209 * contiguous lines in the same origin (i.e. <commit, path> pair),
210 * merge them together.
212 static void coalesce(struct scoreboard *sb)
214 struct blame_entry *ent, *next;
216 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
217 if (same_suspect(ent->suspect, next->suspect) &&
218 ent->guilty == next->guilty &&
219 ent->s_lno + ent->num_lines == next->s_lno) {
220 ent->num_lines += next->num_lines;
221 ent->next = next->next;
222 if (ent->next)
223 ent->next->prev = ent;
224 origin_decref(next->suspect);
225 free(next);
226 ent->score = 0;
227 next = ent; /* again */
231 if (DEBUG) /* sanity */
232 sanity_check_refcnt(sb);
236 * Given a commit and a path in it, create a new origin structure.
237 * The callers that add blame to the scoreboard should use
238 * get_origin() to obtain shared, refcounted copy instead of calling
239 * this function directly.
241 static struct origin *make_origin(struct commit *commit, const char *path)
243 struct origin *o;
244 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
245 o->commit = commit;
246 o->refcnt = 1;
247 strcpy(o->path, path);
248 return o;
252 * Locate an existing origin or create a new one.
254 static struct origin *get_origin(struct scoreboard *sb,
255 struct commit *commit,
256 const char *path)
258 struct blame_entry *e;
260 for (e = sb->ent; e; e = e->next) {
261 if (e->suspect->commit == commit &&
262 !strcmp(e->suspect->path, path))
263 return origin_incref(e->suspect);
265 return make_origin(commit, path);
269 * Fill the blob_sha1 field of an origin if it hasn't, so that later
270 * call to fill_origin_blob() can use it to locate the data. blob_sha1
271 * for an origin is also used to pass the blame for the entire file to
272 * the parent to detect the case where a child's blob is identical to
273 * that of its parent's.
275 static int fill_blob_sha1(struct origin *origin)
277 unsigned mode;
279 if (!is_null_sha1(origin->blob_sha1))
280 return 0;
281 if (get_tree_entry(origin->commit->object.sha1,
282 origin->path,
283 origin->blob_sha1, &mode))
284 goto error_out;
285 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
286 goto error_out;
287 return 0;
288 error_out:
289 hashclr(origin->blob_sha1);
290 return -1;
294 * We have an origin -- check if the same path exists in the
295 * parent and return an origin structure to represent it.
297 static struct origin *find_origin(struct scoreboard *sb,
298 struct commit *parent,
299 struct origin *origin)
301 struct origin *porigin = NULL;
302 struct diff_options diff_opts;
303 const char *paths[2];
305 if (parent->util) {
307 * Each commit object can cache one origin in that
308 * commit. This is a freestanding copy of origin and
309 * not refcounted.
311 struct origin *cached = parent->util;
312 if (!strcmp(cached->path, origin->path)) {
314 * The same path between origin and its parent
315 * without renaming -- the most common case.
317 porigin = get_origin(sb, parent, cached->path);
320 * If the origin was newly created (i.e. get_origin
321 * would call make_origin if none is found in the
322 * scoreboard), it does not know the blob_sha1,
323 * so copy it. Otherwise porigin was in the
324 * scoreboard and already knows blob_sha1.
326 if (porigin->refcnt == 1)
327 hashcpy(porigin->blob_sha1, cached->blob_sha1);
328 return porigin;
330 /* otherwise it was not very useful; free it */
331 free(parent->util);
332 parent->util = NULL;
335 /* See if the origin->path is different between parent
336 * and origin first. Most of the time they are the
337 * same and diff-tree is fairly efficient about this.
339 diff_setup(&diff_opts);
340 DIFF_OPT_SET(&diff_opts, RECURSIVE);
341 diff_opts.detect_rename = 0;
342 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
343 paths[0] = origin->path;
344 paths[1] = NULL;
346 diff_tree_setup_paths(paths, &diff_opts);
347 if (diff_setup_done(&diff_opts) < 0)
348 die("diff-setup");
350 if (is_null_sha1(origin->commit->object.sha1))
351 do_diff_cache(parent->tree->object.sha1, &diff_opts);
352 else
353 diff_tree_sha1(parent->tree->object.sha1,
354 origin->commit->tree->object.sha1,
355 "", &diff_opts);
356 diffcore_std(&diff_opts);
358 /* It is either one entry that says "modified", or "created",
359 * or nothing.
361 if (!diff_queued_diff.nr) {
362 /* The path is the same as parent */
363 porigin = get_origin(sb, parent, origin->path);
364 hashcpy(porigin->blob_sha1, origin->blob_sha1);
366 else if (diff_queued_diff.nr != 1)
367 die("internal error in blame::find_origin");
368 else {
369 struct diff_filepair *p = diff_queued_diff.queue[0];
370 switch (p->status) {
371 default:
372 die("internal error in blame::find_origin (%c)",
373 p->status);
374 case 'M':
375 porigin = get_origin(sb, parent, origin->path);
376 hashcpy(porigin->blob_sha1, p->one->sha1);
377 break;
378 case 'A':
379 case 'T':
380 /* Did not exist in parent, or type changed */
381 break;
384 diff_flush(&diff_opts);
385 diff_tree_release_paths(&diff_opts);
386 if (porigin) {
388 * Create a freestanding copy that is not part of
389 * the refcounted origin found in the scoreboard, and
390 * cache it in the commit.
392 struct origin *cached;
394 cached = make_origin(porigin->commit, porigin->path);
395 hashcpy(cached->blob_sha1, porigin->blob_sha1);
396 parent->util = cached;
398 return porigin;
402 * We have an origin -- find the path that corresponds to it in its
403 * parent and return an origin structure to represent it.
405 static struct origin *find_rename(struct scoreboard *sb,
406 struct commit *parent,
407 struct origin *origin)
409 struct origin *porigin = NULL;
410 struct diff_options diff_opts;
411 int i;
412 const char *paths[2];
414 diff_setup(&diff_opts);
415 DIFF_OPT_SET(&diff_opts, RECURSIVE);
416 diff_opts.detect_rename = DIFF_DETECT_RENAME;
417 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
418 diff_opts.single_follow = origin->path;
419 paths[0] = NULL;
420 diff_tree_setup_paths(paths, &diff_opts);
421 if (diff_setup_done(&diff_opts) < 0)
422 die("diff-setup");
424 if (is_null_sha1(origin->commit->object.sha1))
425 do_diff_cache(parent->tree->object.sha1, &diff_opts);
426 else
427 diff_tree_sha1(parent->tree->object.sha1,
428 origin->commit->tree->object.sha1,
429 "", &diff_opts);
430 diffcore_std(&diff_opts);
432 for (i = 0; i < diff_queued_diff.nr; i++) {
433 struct diff_filepair *p = diff_queued_diff.queue[i];
434 if ((p->status == 'R' || p->status == 'C') &&
435 !strcmp(p->two->path, origin->path)) {
436 porigin = get_origin(sb, parent, p->one->path);
437 hashcpy(porigin->blob_sha1, p->one->sha1);
438 break;
441 diff_flush(&diff_opts);
442 diff_tree_release_paths(&diff_opts);
443 return porigin;
447 * Parsing of patch chunks...
449 struct chunk {
450 /* line number in postimage; up to but not including this
451 * line is the same as preimage
453 int same;
455 /* preimage line number after this chunk */
456 int p_next;
458 /* postimage line number after this chunk */
459 int t_next;
462 struct patch {
463 struct chunk *chunks;
464 int num;
467 struct blame_diff_state {
468 struct patch *ret;
469 unsigned hunk_post_context;
470 unsigned hunk_in_pre_context : 1;
473 static void process_u_diff(void *state_, char *line, unsigned long len)
475 struct blame_diff_state *state = state_;
476 struct chunk *chunk;
477 int off1, off2, len1, len2, num;
479 num = state->ret->num;
480 if (len < 4 || line[0] != '@' || line[1] != '@') {
481 if (state->hunk_in_pre_context && line[0] == ' ')
482 state->ret->chunks[num - 1].same++;
483 else {
484 state->hunk_in_pre_context = 0;
485 if (line[0] == ' ')
486 state->hunk_post_context++;
487 else
488 state->hunk_post_context = 0;
490 return;
493 if (num && state->hunk_post_context) {
494 chunk = &state->ret->chunks[num - 1];
495 chunk->p_next -= state->hunk_post_context;
496 chunk->t_next -= state->hunk_post_context;
498 state->ret->num = ++num;
499 state->ret->chunks = xrealloc(state->ret->chunks,
500 sizeof(struct chunk) * num);
501 chunk = &state->ret->chunks[num - 1];
502 if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
503 state->ret->num--;
504 return;
507 /* Line numbers in patch output are one based. */
508 off1--;
509 off2--;
511 chunk->same = len2 ? off2 : (off2 + 1);
513 chunk->p_next = off1 + (len1 ? len1 : 1);
514 chunk->t_next = chunk->same + len2;
515 state->hunk_in_pre_context = 1;
516 state->hunk_post_context = 0;
519 static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
520 int context)
522 struct blame_diff_state state;
523 xpparam_t xpp;
524 xdemitconf_t xecfg;
525 xdemitcb_t ecb;
527 xpp.flags = xdl_opts;
528 memset(&xecfg, 0, sizeof(xecfg));
529 xecfg.ctxlen = context;
530 memset(&state, 0, sizeof(state));
531 state.ret = xmalloc(sizeof(struct patch));
532 state.ret->chunks = NULL;
533 state.ret->num = 0;
535 xdi_diff_outf(file_p, file_o, process_u_diff, &state, &xpp, &xecfg, &ecb);
537 if (state.ret->num) {
538 struct chunk *chunk;
539 chunk = &state.ret->chunks[state.ret->num - 1];
540 chunk->p_next -= state.hunk_post_context;
541 chunk->t_next -= state.hunk_post_context;
543 return state.ret;
547 * Run diff between two origins and grab the patch output, so that
548 * we can pass blame for lines origin is currently suspected for
549 * to its parent.
551 static struct patch *get_patch(struct origin *parent, struct origin *origin)
553 mmfile_t file_p, file_o;
554 struct patch *patch;
556 fill_origin_blob(parent, &file_p);
557 fill_origin_blob(origin, &file_o);
558 if (!file_p.ptr || !file_o.ptr)
559 return NULL;
560 patch = compare_buffer(&file_p, &file_o, 0);
561 num_get_patch++;
562 return patch;
565 static void free_patch(struct patch *p)
567 free(p->chunks);
568 free(p);
572 * Link in a new blame entry to the scoreboard. Entries that cover the
573 * same line range have been removed from the scoreboard previously.
575 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
577 struct blame_entry *ent, *prev = NULL;
579 origin_incref(e->suspect);
581 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
582 prev = ent;
584 /* prev, if not NULL, is the last one that is below e */
585 e->prev = prev;
586 if (prev) {
587 e->next = prev->next;
588 prev->next = e;
590 else {
591 e->next = sb->ent;
592 sb->ent = e;
594 if (e->next)
595 e->next->prev = e;
599 * src typically is on-stack; we want to copy the information in it to
600 * a malloced blame_entry that is already on the linked list of the
601 * scoreboard. The origin of dst loses a refcnt while the origin of src
602 * gains one.
604 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
606 struct blame_entry *p, *n;
608 p = dst->prev;
609 n = dst->next;
610 origin_incref(src->suspect);
611 origin_decref(dst->suspect);
612 memcpy(dst, src, sizeof(*src));
613 dst->prev = p;
614 dst->next = n;
615 dst->score = 0;
618 static const char *nth_line(struct scoreboard *sb, int lno)
620 return sb->final_buf + sb->lineno[lno];
624 * It is known that lines between tlno to same came from parent, and e
625 * has an overlap with that range. it also is known that parent's
626 * line plno corresponds to e's line tlno.
628 * <---- e ----->
629 * <------>
630 * <------------>
631 * <------------>
632 * <------------------>
634 * Split e into potentially three parts; before this chunk, the chunk
635 * to be blamed for the parent, and after that portion.
637 static void split_overlap(struct blame_entry *split,
638 struct blame_entry *e,
639 int tlno, int plno, int same,
640 struct origin *parent)
642 int chunk_end_lno;
643 memset(split, 0, sizeof(struct blame_entry [3]));
645 if (e->s_lno < tlno) {
646 /* there is a pre-chunk part not blamed on parent */
647 split[0].suspect = origin_incref(e->suspect);
648 split[0].lno = e->lno;
649 split[0].s_lno = e->s_lno;
650 split[0].num_lines = tlno - e->s_lno;
651 split[1].lno = e->lno + tlno - e->s_lno;
652 split[1].s_lno = plno;
654 else {
655 split[1].lno = e->lno;
656 split[1].s_lno = plno + (e->s_lno - tlno);
659 if (same < e->s_lno + e->num_lines) {
660 /* there is a post-chunk part not blamed on parent */
661 split[2].suspect = origin_incref(e->suspect);
662 split[2].lno = e->lno + (same - e->s_lno);
663 split[2].s_lno = e->s_lno + (same - e->s_lno);
664 split[2].num_lines = e->s_lno + e->num_lines - same;
665 chunk_end_lno = split[2].lno;
667 else
668 chunk_end_lno = e->lno + e->num_lines;
669 split[1].num_lines = chunk_end_lno - split[1].lno;
672 * if it turns out there is nothing to blame the parent for,
673 * forget about the splitting. !split[1].suspect signals this.
675 if (split[1].num_lines < 1)
676 return;
677 split[1].suspect = origin_incref(parent);
681 * split_overlap() divided an existing blame e into up to three parts
682 * in split. Adjust the linked list of blames in the scoreboard to
683 * reflect the split.
685 static void split_blame(struct scoreboard *sb,
686 struct blame_entry *split,
687 struct blame_entry *e)
689 struct blame_entry *new_entry;
691 if (split[0].suspect && split[2].suspect) {
692 /* The first part (reuse storage for the existing entry e) */
693 dup_entry(e, &split[0]);
695 /* The last part -- me */
696 new_entry = xmalloc(sizeof(*new_entry));
697 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
698 add_blame_entry(sb, new_entry);
700 /* ... and the middle part -- parent */
701 new_entry = xmalloc(sizeof(*new_entry));
702 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
703 add_blame_entry(sb, new_entry);
705 else if (!split[0].suspect && !split[2].suspect)
707 * The parent covers the entire area; reuse storage for
708 * e and replace it with the parent.
710 dup_entry(e, &split[1]);
711 else if (split[0].suspect) {
712 /* me and then parent */
713 dup_entry(e, &split[0]);
715 new_entry = xmalloc(sizeof(*new_entry));
716 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
717 add_blame_entry(sb, new_entry);
719 else {
720 /* parent and then me */
721 dup_entry(e, &split[1]);
723 new_entry = xmalloc(sizeof(*new_entry));
724 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
725 add_blame_entry(sb, new_entry);
728 if (DEBUG) { /* sanity */
729 struct blame_entry *ent;
730 int lno = sb->ent->lno, corrupt = 0;
732 for (ent = sb->ent; ent; ent = ent->next) {
733 if (lno != ent->lno)
734 corrupt = 1;
735 if (ent->s_lno < 0)
736 corrupt = 1;
737 lno += ent->num_lines;
739 if (corrupt) {
740 lno = sb->ent->lno;
741 for (ent = sb->ent; ent; ent = ent->next) {
742 printf("L %8d l %8d n %8d\n",
743 lno, ent->lno, ent->num_lines);
744 lno = ent->lno + ent->num_lines;
746 die("oops");
752 * After splitting the blame, the origins used by the
753 * on-stack blame_entry should lose one refcnt each.
755 static void decref_split(struct blame_entry *split)
757 int i;
759 for (i = 0; i < 3; i++)
760 origin_decref(split[i].suspect);
764 * Helper for blame_chunk(). blame_entry e is known to overlap with
765 * the patch hunk; split it and pass blame to the parent.
767 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
768 int tlno, int plno, int same,
769 struct origin *parent)
771 struct blame_entry split[3];
773 split_overlap(split, e, tlno, plno, same, parent);
774 if (split[1].suspect)
775 split_blame(sb, split, e);
776 decref_split(split);
780 * Find the line number of the last line the target is suspected for.
782 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
784 struct blame_entry *e;
785 int last_in_target = -1;
787 for (e = sb->ent; e; e = e->next) {
788 if (e->guilty || !same_suspect(e->suspect, target))
789 continue;
790 if (last_in_target < e->s_lno + e->num_lines)
791 last_in_target = e->s_lno + e->num_lines;
793 return last_in_target;
797 * Process one hunk from the patch between the current suspect for
798 * blame_entry e and its parent. Find and split the overlap, and
799 * pass blame to the overlapping part to the parent.
801 static void blame_chunk(struct scoreboard *sb,
802 int tlno, int plno, int same,
803 struct origin *target, struct origin *parent)
805 struct blame_entry *e;
807 for (e = sb->ent; e; e = e->next) {
808 if (e->guilty || !same_suspect(e->suspect, target))
809 continue;
810 if (same <= e->s_lno)
811 continue;
812 if (tlno < e->s_lno + e->num_lines)
813 blame_overlap(sb, e, tlno, plno, same, parent);
818 * We are looking at the origin 'target' and aiming to pass blame
819 * for the lines it is suspected to its parent. Run diff to find
820 * which lines came from parent and pass blame for them.
822 static int pass_blame_to_parent(struct scoreboard *sb,
823 struct origin *target,
824 struct origin *parent)
826 int i, last_in_target, plno, tlno;
827 struct patch *patch;
829 last_in_target = find_last_in_target(sb, target);
830 if (last_in_target < 0)
831 return 1; /* nothing remains for this target */
833 patch = get_patch(parent, target);
834 plno = tlno = 0;
835 for (i = 0; i < patch->num; i++) {
836 struct chunk *chunk = &patch->chunks[i];
838 blame_chunk(sb, tlno, plno, chunk->same, target, parent);
839 plno = chunk->p_next;
840 tlno = chunk->t_next;
842 /* The rest (i.e. anything after tlno) are the same as the parent */
843 blame_chunk(sb, tlno, plno, last_in_target, target, parent);
845 free_patch(patch);
846 return 0;
850 * The lines in blame_entry after splitting blames many times can become
851 * very small and trivial, and at some point it becomes pointless to
852 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
853 * ordinary C program, and it is not worth to say it was copied from
854 * totally unrelated file in the parent.
856 * Compute how trivial the lines in the blame_entry are.
858 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
860 unsigned score;
861 const char *cp, *ep;
863 if (e->score)
864 return e->score;
866 score = 1;
867 cp = nth_line(sb, e->lno);
868 ep = nth_line(sb, e->lno + e->num_lines);
869 while (cp < ep) {
870 unsigned ch = *((unsigned char *)cp);
871 if (isalnum(ch))
872 score++;
873 cp++;
875 e->score = score;
876 return score;
880 * best_so_far[] and this[] are both a split of an existing blame_entry
881 * that passes blame to the parent. Maintain best_so_far the best split
882 * so far, by comparing this and best_so_far and copying this into
883 * bst_so_far as needed.
885 static void copy_split_if_better(struct scoreboard *sb,
886 struct blame_entry *best_so_far,
887 struct blame_entry *this)
889 int i;
891 if (!this[1].suspect)
892 return;
893 if (best_so_far[1].suspect) {
894 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
895 return;
898 for (i = 0; i < 3; i++)
899 origin_incref(this[i].suspect);
900 decref_split(best_so_far);
901 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
905 * We are looking at a part of the final image represented by
906 * ent (tlno and same are offset by ent->s_lno).
907 * tlno is where we are looking at in the final image.
908 * up to (but not including) same match preimage.
909 * plno is where we are looking at in the preimage.
911 * <-------------- final image ---------------------->
912 * <------ent------>
913 * ^tlno ^same
914 * <---------preimage----->
915 * ^plno
917 * All line numbers are 0-based.
919 static void handle_split(struct scoreboard *sb,
920 struct blame_entry *ent,
921 int tlno, int plno, int same,
922 struct origin *parent,
923 struct blame_entry *split)
925 if (ent->num_lines <= tlno)
926 return;
927 if (tlno < same) {
928 struct blame_entry this[3];
929 tlno += ent->s_lno;
930 same += ent->s_lno;
931 split_overlap(this, ent, tlno, plno, same, parent);
932 copy_split_if_better(sb, split, this);
933 decref_split(this);
938 * Find the lines from parent that are the same as ent so that
939 * we can pass blames to it. file_p has the blob contents for
940 * the parent.
942 static void find_copy_in_blob(struct scoreboard *sb,
943 struct blame_entry *ent,
944 struct origin *parent,
945 struct blame_entry *split,
946 mmfile_t *file_p)
948 const char *cp;
949 int cnt;
950 mmfile_t file_o;
951 struct patch *patch;
952 int i, plno, tlno;
955 * Prepare mmfile that contains only the lines in ent.
957 cp = nth_line(sb, ent->lno);
958 file_o.ptr = (char*) cp;
959 cnt = ent->num_lines;
961 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
962 if (*cp++ == '\n')
963 cnt--;
965 file_o.size = cp - file_o.ptr;
967 patch = compare_buffer(file_p, &file_o, 1);
970 * file_o is a part of final image we are annotating.
971 * file_p partially may match that image.
973 memset(split, 0, sizeof(struct blame_entry [3]));
974 plno = tlno = 0;
975 for (i = 0; i < patch->num; i++) {
976 struct chunk *chunk = &patch->chunks[i];
978 handle_split(sb, ent, tlno, plno, chunk->same, parent, split);
979 plno = chunk->p_next;
980 tlno = chunk->t_next;
982 /* remainder, if any, all match the preimage */
983 handle_split(sb, ent, tlno, plno, ent->num_lines, parent, split);
984 free_patch(patch);
988 * See if lines currently target is suspected for can be attributed to
989 * parent.
991 static int find_move_in_parent(struct scoreboard *sb,
992 struct origin *target,
993 struct origin *parent)
995 int last_in_target, made_progress;
996 struct blame_entry *e, split[3];
997 mmfile_t file_p;
999 last_in_target = find_last_in_target(sb, target);
1000 if (last_in_target < 0)
1001 return 1; /* nothing remains for this target */
1003 fill_origin_blob(parent, &file_p);
1004 if (!file_p.ptr)
1005 return 0;
1007 made_progress = 1;
1008 while (made_progress) {
1009 made_progress = 0;
1010 for (e = sb->ent; e; e = e->next) {
1011 if (e->guilty || !same_suspect(e->suspect, target) ||
1012 ent_score(sb, e) < blame_move_score)
1013 continue;
1014 find_copy_in_blob(sb, e, parent, split, &file_p);
1015 if (split[1].suspect &&
1016 blame_move_score < ent_score(sb, &split[1])) {
1017 split_blame(sb, split, e);
1018 made_progress = 1;
1020 decref_split(split);
1023 return 0;
1026 struct blame_list {
1027 struct blame_entry *ent;
1028 struct blame_entry split[3];
1032 * Count the number of entries the target is suspected for,
1033 * and prepare a list of entry and the best split.
1035 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1036 struct origin *target,
1037 int min_score,
1038 int *num_ents_p)
1040 struct blame_entry *e;
1041 int num_ents, i;
1042 struct blame_list *blame_list = NULL;
1044 for (e = sb->ent, num_ents = 0; e; e = e->next)
1045 if (!e->scanned && !e->guilty &&
1046 same_suspect(e->suspect, target) &&
1047 min_score < ent_score(sb, e))
1048 num_ents++;
1049 if (num_ents) {
1050 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1051 for (e = sb->ent, i = 0; e; e = e->next)
1052 if (!e->scanned && !e->guilty &&
1053 same_suspect(e->suspect, target) &&
1054 min_score < ent_score(sb, e))
1055 blame_list[i++].ent = e;
1057 *num_ents_p = num_ents;
1058 return blame_list;
1062 * Reset the scanned status on all entries.
1064 static void reset_scanned_flag(struct scoreboard *sb)
1066 struct blame_entry *e;
1067 for (e = sb->ent; e; e = e->next)
1068 e->scanned = 0;
1072 * For lines target is suspected for, see if we can find code movement
1073 * across file boundary from the parent commit. porigin is the path
1074 * in the parent we already tried.
1076 static int find_copy_in_parent(struct scoreboard *sb,
1077 struct origin *target,
1078 struct commit *parent,
1079 struct origin *porigin,
1080 int opt)
1082 struct diff_options diff_opts;
1083 const char *paths[1];
1084 int i, j;
1085 int retval;
1086 struct blame_list *blame_list;
1087 int num_ents;
1089 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1090 if (!blame_list)
1091 return 1; /* nothing remains for this target */
1093 diff_setup(&diff_opts);
1094 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1095 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1097 paths[0] = NULL;
1098 diff_tree_setup_paths(paths, &diff_opts);
1099 if (diff_setup_done(&diff_opts) < 0)
1100 die("diff-setup");
1102 /* Try "find copies harder" on new path if requested;
1103 * we do not want to use diffcore_rename() actually to
1104 * match things up; find_copies_harder is set only to
1105 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1106 * and this code needs to be after diff_setup_done(), which
1107 * usually makes find-copies-harder imply copy detection.
1109 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1110 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1111 && (!porigin || strcmp(target->path, porigin->path))))
1112 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1114 if (is_null_sha1(target->commit->object.sha1))
1115 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1116 else
1117 diff_tree_sha1(parent->tree->object.sha1,
1118 target->commit->tree->object.sha1,
1119 "", &diff_opts);
1121 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1122 diffcore_std(&diff_opts);
1124 retval = 0;
1125 while (1) {
1126 int made_progress = 0;
1128 for (i = 0; i < diff_queued_diff.nr; i++) {
1129 struct diff_filepair *p = diff_queued_diff.queue[i];
1130 struct origin *norigin;
1131 mmfile_t file_p;
1132 struct blame_entry this[3];
1134 if (!DIFF_FILE_VALID(p->one))
1135 continue; /* does not exist in parent */
1136 if (porigin && !strcmp(p->one->path, porigin->path))
1137 /* find_move already dealt with this path */
1138 continue;
1140 norigin = get_origin(sb, parent, p->one->path);
1141 hashcpy(norigin->blob_sha1, p->one->sha1);
1142 fill_origin_blob(norigin, &file_p);
1143 if (!file_p.ptr)
1144 continue;
1146 for (j = 0; j < num_ents; j++) {
1147 find_copy_in_blob(sb, blame_list[j].ent,
1148 norigin, this, &file_p);
1149 copy_split_if_better(sb, blame_list[j].split,
1150 this);
1151 decref_split(this);
1153 origin_decref(norigin);
1156 for (j = 0; j < num_ents; j++) {
1157 struct blame_entry *split = blame_list[j].split;
1158 if (split[1].suspect &&
1159 blame_copy_score < ent_score(sb, &split[1])) {
1160 split_blame(sb, split, blame_list[j].ent);
1161 made_progress = 1;
1163 else
1164 blame_list[j].ent->scanned = 1;
1165 decref_split(split);
1167 free(blame_list);
1169 if (!made_progress)
1170 break;
1171 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1172 if (!blame_list) {
1173 retval = 1;
1174 break;
1177 reset_scanned_flag(sb);
1178 diff_flush(&diff_opts);
1179 diff_tree_release_paths(&diff_opts);
1180 return retval;
1184 * The blobs of origin and porigin exactly match, so everything
1185 * origin is suspected for can be blamed on the parent.
1187 static void pass_whole_blame(struct scoreboard *sb,
1188 struct origin *origin, struct origin *porigin)
1190 struct blame_entry *e;
1192 if (!porigin->file.ptr && origin->file.ptr) {
1193 /* Steal its file */
1194 porigin->file = origin->file;
1195 origin->file.ptr = NULL;
1197 for (e = sb->ent; e; e = e->next) {
1198 if (!same_suspect(e->suspect, origin))
1199 continue;
1200 origin_incref(porigin);
1201 origin_decref(e->suspect);
1202 e->suspect = porigin;
1207 * We pass blame from the current commit to its parents. We keep saying
1208 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1209 * exonerate ourselves.
1211 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1213 if (!reverse)
1214 return commit->parents;
1215 return lookup_decoration(&revs->children, &commit->object);
1218 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1220 int cnt;
1221 struct commit_list *l = first_scapegoat(revs, commit);
1222 for (cnt = 0; l; l = l->next)
1223 cnt++;
1224 return cnt;
1227 #define MAXSG 16
1229 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1231 struct rev_info *revs = sb->revs;
1232 int i, pass, num_sg;
1233 struct commit *commit = origin->commit;
1234 struct commit_list *sg;
1235 struct origin *sg_buf[MAXSG];
1236 struct origin *porigin, **sg_origin = sg_buf;
1238 num_sg = num_scapegoats(revs, commit);
1239 if (!num_sg)
1240 goto finish;
1241 else if (num_sg < ARRAY_SIZE(sg_buf))
1242 memset(sg_buf, 0, sizeof(sg_buf));
1243 else
1244 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1247 * The first pass looks for unrenamed path to optimize for
1248 * common cases, then we look for renames in the second pass.
1250 for (pass = 0; pass < 2; pass++) {
1251 struct origin *(*find)(struct scoreboard *,
1252 struct commit *, struct origin *);
1253 find = pass ? find_rename : find_origin;
1255 for (i = 0, sg = first_scapegoat(revs, commit);
1256 i < num_sg && sg;
1257 sg = sg->next, i++) {
1258 struct commit *p = sg->item;
1259 int j, same;
1261 if (sg_origin[i])
1262 continue;
1263 if (parse_commit(p))
1264 continue;
1265 porigin = find(sb, p, origin);
1266 if (!porigin)
1267 continue;
1268 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1269 pass_whole_blame(sb, origin, porigin);
1270 origin_decref(porigin);
1271 goto finish;
1273 for (j = same = 0; j < i; j++)
1274 if (sg_origin[j] &&
1275 !hashcmp(sg_origin[j]->blob_sha1,
1276 porigin->blob_sha1)) {
1277 same = 1;
1278 break;
1280 if (!same)
1281 sg_origin[i] = porigin;
1282 else
1283 origin_decref(porigin);
1287 num_commits++;
1288 for (i = 0, sg = first_scapegoat(revs, commit);
1289 i < num_sg && sg;
1290 sg = sg->next, i++) {
1291 struct origin *porigin = sg_origin[i];
1292 if (!porigin)
1293 continue;
1294 if (pass_blame_to_parent(sb, origin, porigin))
1295 goto finish;
1299 * Optionally find moves in parents' files.
1301 if (opt & PICKAXE_BLAME_MOVE)
1302 for (i = 0, sg = first_scapegoat(revs, commit);
1303 i < num_sg && sg;
1304 sg = sg->next, i++) {
1305 struct origin *porigin = sg_origin[i];
1306 if (!porigin)
1307 continue;
1308 if (find_move_in_parent(sb, origin, porigin))
1309 goto finish;
1313 * Optionally find copies from parents' files.
1315 if (opt & PICKAXE_BLAME_COPY)
1316 for (i = 0, sg = first_scapegoat(revs, commit);
1317 i < num_sg && sg;
1318 sg = sg->next, i++) {
1319 struct origin *porigin = sg_origin[i];
1320 if (find_copy_in_parent(sb, origin, sg->item,
1321 porigin, opt))
1322 goto finish;
1325 finish:
1326 for (i = 0; i < num_sg; i++) {
1327 if (sg_origin[i]) {
1328 drop_origin_blob(sg_origin[i]);
1329 origin_decref(sg_origin[i]);
1332 drop_origin_blob(origin);
1333 if (sg_buf != sg_origin)
1334 free(sg_origin);
1338 * Information on commits, used for output.
1340 struct commit_info
1342 const char *author;
1343 const char *author_mail;
1344 unsigned long author_time;
1345 const char *author_tz;
1347 /* filled only when asked for details */
1348 const char *committer;
1349 const char *committer_mail;
1350 unsigned long committer_time;
1351 const char *committer_tz;
1353 const char *summary;
1357 * Parse author/committer line in the commit object buffer
1359 static void get_ac_line(const char *inbuf, const char *what,
1360 int bufsz, char *person, const char **mail,
1361 unsigned long *time, const char **tz)
1363 int len, tzlen, maillen;
1364 char *tmp, *endp, *timepos;
1366 tmp = strstr(inbuf, what);
1367 if (!tmp)
1368 goto error_out;
1369 tmp += strlen(what);
1370 endp = strchr(tmp, '\n');
1371 if (!endp)
1372 len = strlen(tmp);
1373 else
1374 len = endp - tmp;
1375 if (bufsz <= len) {
1376 error_out:
1377 /* Ugh */
1378 *mail = *tz = "(unknown)";
1379 *time = 0;
1380 return;
1382 memcpy(person, tmp, len);
1384 tmp = person;
1385 tmp += len;
1386 *tmp = 0;
1387 while (*tmp != ' ')
1388 tmp--;
1389 *tz = tmp+1;
1390 tzlen = (person+len)-(tmp+1);
1392 *tmp = 0;
1393 while (*tmp != ' ')
1394 tmp--;
1395 *time = strtoul(tmp, NULL, 10);
1396 timepos = tmp;
1398 *tmp = 0;
1399 while (*tmp != ' ')
1400 tmp--;
1401 *mail = tmp + 1;
1402 *tmp = 0;
1403 maillen = timepos - tmp;
1405 if (!mailmap.nr)
1406 return;
1409 * mailmap expansion may make the name longer.
1410 * make room by pushing stuff down.
1412 tmp = person + bufsz - (tzlen + 1);
1413 memmove(tmp, *tz, tzlen);
1414 tmp[tzlen] = 0;
1415 *tz = tmp;
1417 tmp = tmp - (maillen + 1);
1418 memmove(tmp, *mail, maillen);
1419 tmp[maillen] = 0;
1420 *mail = tmp;
1423 * Now, convert e-mail using mailmap
1425 map_email(&mailmap, tmp + 1, person, tmp-person-1);
1428 static void get_commit_info(struct commit *commit,
1429 struct commit_info *ret,
1430 int detailed)
1432 int len;
1433 char *tmp, *endp;
1434 static char author_buf[1024];
1435 static char committer_buf[1024];
1436 static char summary_buf[1024];
1439 * We've operated without save_commit_buffer, so
1440 * we now need to populate them for output.
1442 if (!commit->buffer) {
1443 enum object_type type;
1444 unsigned long size;
1445 commit->buffer =
1446 read_sha1_file(commit->object.sha1, &type, &size);
1447 if (!commit->buffer)
1448 die("Cannot read commit %s",
1449 sha1_to_hex(commit->object.sha1));
1451 ret->author = author_buf;
1452 get_ac_line(commit->buffer, "\nauthor ",
1453 sizeof(author_buf), author_buf, &ret->author_mail,
1454 &ret->author_time, &ret->author_tz);
1456 if (!detailed)
1457 return;
1459 ret->committer = committer_buf;
1460 get_ac_line(commit->buffer, "\ncommitter ",
1461 sizeof(committer_buf), committer_buf, &ret->committer_mail,
1462 &ret->committer_time, &ret->committer_tz);
1464 ret->summary = summary_buf;
1465 tmp = strstr(commit->buffer, "\n\n");
1466 if (!tmp) {
1467 error_out:
1468 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1469 return;
1471 tmp += 2;
1472 endp = strchr(tmp, '\n');
1473 if (!endp)
1474 endp = tmp + strlen(tmp);
1475 len = endp - tmp;
1476 if (len >= sizeof(summary_buf) || len == 0)
1477 goto error_out;
1478 memcpy(summary_buf, tmp, len);
1479 summary_buf[len] = 0;
1483 * To allow LF and other nonportable characters in pathnames,
1484 * they are c-style quoted as needed.
1486 static void write_filename_info(const char *path)
1488 printf("filename ");
1489 write_name_quoted(path, stdout, '\n');
1493 * The blame_entry is found to be guilty for the range. Mark it
1494 * as such, and show it in incremental output.
1496 static void found_guilty_entry(struct blame_entry *ent)
1498 if (ent->guilty)
1499 return;
1500 ent->guilty = 1;
1501 if (incremental) {
1502 struct origin *suspect = ent->suspect;
1504 printf("%s %d %d %d\n",
1505 sha1_to_hex(suspect->commit->object.sha1),
1506 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1507 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1508 struct commit_info ci;
1509 suspect->commit->object.flags |= METAINFO_SHOWN;
1510 get_commit_info(suspect->commit, &ci, 1);
1511 printf("author %s\n", ci.author);
1512 printf("author-mail %s\n", ci.author_mail);
1513 printf("author-time %lu\n", ci.author_time);
1514 printf("author-tz %s\n", ci.author_tz);
1515 printf("committer %s\n", ci.committer);
1516 printf("committer-mail %s\n", ci.committer_mail);
1517 printf("committer-time %lu\n", ci.committer_time);
1518 printf("committer-tz %s\n", ci.committer_tz);
1519 printf("summary %s\n", ci.summary);
1520 if (suspect->commit->object.flags & UNINTERESTING)
1521 printf("boundary\n");
1523 write_filename_info(suspect->path);
1524 maybe_flush_or_die(stdout, "stdout");
1529 * The main loop -- while the scoreboard has lines whose true origin
1530 * is still unknown, pick one blame_entry, and allow its current
1531 * suspect to pass blames to its parents.
1533 static void assign_blame(struct scoreboard *sb, int opt)
1535 struct rev_info *revs = sb->revs;
1537 while (1) {
1538 struct blame_entry *ent;
1539 struct commit *commit;
1540 struct origin *suspect = NULL;
1542 /* find one suspect to break down */
1543 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1544 if (!ent->guilty)
1545 suspect = ent->suspect;
1546 if (!suspect)
1547 return; /* all done */
1550 * We will use this suspect later in the loop,
1551 * so hold onto it in the meantime.
1553 origin_incref(suspect);
1554 commit = suspect->commit;
1555 if (!commit->object.parsed)
1556 parse_commit(commit);
1557 if (reverse ||
1558 (!(commit->object.flags & UNINTERESTING) &&
1559 !(revs->max_age != -1 && commit->date < revs->max_age)))
1560 pass_blame(sb, suspect, opt);
1561 else {
1562 commit->object.flags |= UNINTERESTING;
1563 if (commit->object.parsed)
1564 mark_parents_uninteresting(commit);
1566 /* treat root commit as boundary */
1567 if (!commit->parents && !show_root)
1568 commit->object.flags |= UNINTERESTING;
1570 /* Take responsibility for the remaining entries */
1571 for (ent = sb->ent; ent; ent = ent->next)
1572 if (same_suspect(ent->suspect, suspect))
1573 found_guilty_entry(ent);
1574 origin_decref(suspect);
1576 if (DEBUG) /* sanity */
1577 sanity_check_refcnt(sb);
1581 static const char *format_time(unsigned long time, const char *tz_str,
1582 int show_raw_time)
1584 static char time_buf[128];
1585 time_t t = time;
1586 int minutes, tz;
1587 struct tm *tm;
1589 if (show_raw_time) {
1590 sprintf(time_buf, "%lu %s", time, tz_str);
1591 return time_buf;
1594 tz = atoi(tz_str);
1595 minutes = tz < 0 ? -tz : tz;
1596 minutes = (minutes / 100)*60 + (minutes % 100);
1597 minutes = tz < 0 ? -minutes : minutes;
1598 t = time + minutes * 60;
1599 tm = gmtime(&t);
1601 strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1602 strcat(time_buf, tz_str);
1603 return time_buf;
1606 #define OUTPUT_ANNOTATE_COMPAT 001
1607 #define OUTPUT_LONG_OBJECT_NAME 002
1608 #define OUTPUT_RAW_TIMESTAMP 004
1609 #define OUTPUT_PORCELAIN 010
1610 #define OUTPUT_SHOW_NAME 020
1611 #define OUTPUT_SHOW_NUMBER 040
1612 #define OUTPUT_SHOW_SCORE 0100
1613 #define OUTPUT_NO_AUTHOR 0200
1615 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1617 int cnt;
1618 const char *cp;
1619 struct origin *suspect = ent->suspect;
1620 char hex[41];
1622 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1623 printf("%s%c%d %d %d\n",
1624 hex,
1625 ent->guilty ? ' ' : '*', // purely for debugging
1626 ent->s_lno + 1,
1627 ent->lno + 1,
1628 ent->num_lines);
1629 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1630 struct commit_info ci;
1631 suspect->commit->object.flags |= METAINFO_SHOWN;
1632 get_commit_info(suspect->commit, &ci, 1);
1633 printf("author %s\n", ci.author);
1634 printf("author-mail %s\n", ci.author_mail);
1635 printf("author-time %lu\n", ci.author_time);
1636 printf("author-tz %s\n", ci.author_tz);
1637 printf("committer %s\n", ci.committer);
1638 printf("committer-mail %s\n", ci.committer_mail);
1639 printf("committer-time %lu\n", ci.committer_time);
1640 printf("committer-tz %s\n", ci.committer_tz);
1641 write_filename_info(suspect->path);
1642 printf("summary %s\n", ci.summary);
1643 if (suspect->commit->object.flags & UNINTERESTING)
1644 printf("boundary\n");
1646 else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1647 write_filename_info(suspect->path);
1649 cp = nth_line(sb, ent->lno);
1650 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1651 char ch;
1652 if (cnt)
1653 printf("%s %d %d\n", hex,
1654 ent->s_lno + 1 + cnt,
1655 ent->lno + 1 + cnt);
1656 putchar('\t');
1657 do {
1658 ch = *cp++;
1659 putchar(ch);
1660 } while (ch != '\n' &&
1661 cp < sb->final_buf + sb->final_buf_size);
1665 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1667 int cnt;
1668 const char *cp;
1669 struct origin *suspect = ent->suspect;
1670 struct commit_info ci;
1671 char hex[41];
1672 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1674 get_commit_info(suspect->commit, &ci, 1);
1675 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1677 cp = nth_line(sb, ent->lno);
1678 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1679 char ch;
1680 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1682 if (suspect->commit->object.flags & UNINTERESTING) {
1683 if (blank_boundary)
1684 memset(hex, ' ', length);
1685 else if (!cmd_is_annotate) {
1686 length--;
1687 putchar('^');
1691 printf("%.*s", length, hex);
1692 if (opt & OUTPUT_ANNOTATE_COMPAT)
1693 printf("\t(%10s\t%10s\t%d)", ci.author,
1694 format_time(ci.author_time, ci.author_tz,
1695 show_raw_time),
1696 ent->lno + 1 + cnt);
1697 else {
1698 if (opt & OUTPUT_SHOW_SCORE)
1699 printf(" %*d %02d",
1700 max_score_digits, ent->score,
1701 ent->suspect->refcnt);
1702 if (opt & OUTPUT_SHOW_NAME)
1703 printf(" %-*.*s", longest_file, longest_file,
1704 suspect->path);
1705 if (opt & OUTPUT_SHOW_NUMBER)
1706 printf(" %*d", max_orig_digits,
1707 ent->s_lno + 1 + cnt);
1709 if (!(opt & OUTPUT_NO_AUTHOR))
1710 printf(" (%-*.*s %10s",
1711 longest_author, longest_author,
1712 ci.author,
1713 format_time(ci.author_time,
1714 ci.author_tz,
1715 show_raw_time));
1716 printf(" %*d) ",
1717 max_digits, ent->lno + 1 + cnt);
1719 do {
1720 ch = *cp++;
1721 putchar(ch);
1722 } while (ch != '\n' &&
1723 cp < sb->final_buf + sb->final_buf_size);
1727 static void output(struct scoreboard *sb, int option)
1729 struct blame_entry *ent;
1731 if (option & OUTPUT_PORCELAIN) {
1732 for (ent = sb->ent; ent; ent = ent->next) {
1733 struct blame_entry *oth;
1734 struct origin *suspect = ent->suspect;
1735 struct commit *commit = suspect->commit;
1736 if (commit->object.flags & MORE_THAN_ONE_PATH)
1737 continue;
1738 for (oth = ent->next; oth; oth = oth->next) {
1739 if ((oth->suspect->commit != commit) ||
1740 !strcmp(oth->suspect->path, suspect->path))
1741 continue;
1742 commit->object.flags |= MORE_THAN_ONE_PATH;
1743 break;
1748 for (ent = sb->ent; ent; ent = ent->next) {
1749 if (option & OUTPUT_PORCELAIN)
1750 emit_porcelain(sb, ent);
1751 else {
1752 emit_other(sb, ent, option);
1758 * To allow quick access to the contents of nth line in the
1759 * final image, prepare an index in the scoreboard.
1761 static int prepare_lines(struct scoreboard *sb)
1763 const char *buf = sb->final_buf;
1764 unsigned long len = sb->final_buf_size;
1765 int num = 0, incomplete = 0, bol = 1;
1767 if (len && buf[len-1] != '\n')
1768 incomplete++; /* incomplete line at the end */
1769 while (len--) {
1770 if (bol) {
1771 sb->lineno = xrealloc(sb->lineno,
1772 sizeof(int* ) * (num + 1));
1773 sb->lineno[num] = buf - sb->final_buf;
1774 bol = 0;
1776 if (*buf++ == '\n') {
1777 num++;
1778 bol = 1;
1781 sb->lineno = xrealloc(sb->lineno,
1782 sizeof(int* ) * (num + incomplete + 1));
1783 sb->lineno[num + incomplete] = buf - sb->final_buf;
1784 sb->num_lines = num + incomplete;
1785 return sb->num_lines;
1789 * Add phony grafts for use with -S; this is primarily to
1790 * support git-cvsserver that wants to give a linear history
1791 * to its clients.
1793 static int read_ancestry(const char *graft_file)
1795 FILE *fp = fopen(graft_file, "r");
1796 char buf[1024];
1797 if (!fp)
1798 return -1;
1799 while (fgets(buf, sizeof(buf), fp)) {
1800 /* The format is just "Commit Parent1 Parent2 ...\n" */
1801 int len = strlen(buf);
1802 struct commit_graft *graft = read_graft_line(buf, len);
1803 if (graft)
1804 register_commit_graft(graft, 0);
1806 fclose(fp);
1807 return 0;
1811 * How many columns do we need to show line numbers in decimal?
1813 static int lineno_width(int lines)
1815 int i, width;
1817 for (width = 1, i = 10; i <= lines + 1; width++)
1818 i *= 10;
1819 return width;
1823 * How many columns do we need to show line numbers, authors,
1824 * and filenames?
1826 static void find_alignment(struct scoreboard *sb, int *option)
1828 int longest_src_lines = 0;
1829 int longest_dst_lines = 0;
1830 unsigned largest_score = 0;
1831 struct blame_entry *e;
1833 for (e = sb->ent; e; e = e->next) {
1834 struct origin *suspect = e->suspect;
1835 struct commit_info ci;
1836 int num;
1838 if (strcmp(suspect->path, sb->path))
1839 *option |= OUTPUT_SHOW_NAME;
1840 num = strlen(suspect->path);
1841 if (longest_file < num)
1842 longest_file = num;
1843 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1844 suspect->commit->object.flags |= METAINFO_SHOWN;
1845 get_commit_info(suspect->commit, &ci, 1);
1846 num = strlen(ci.author);
1847 if (longest_author < num)
1848 longest_author = num;
1850 num = e->s_lno + e->num_lines;
1851 if (longest_src_lines < num)
1852 longest_src_lines = num;
1853 num = e->lno + e->num_lines;
1854 if (longest_dst_lines < num)
1855 longest_dst_lines = num;
1856 if (largest_score < ent_score(sb, e))
1857 largest_score = ent_score(sb, e);
1859 max_orig_digits = lineno_width(longest_src_lines);
1860 max_digits = lineno_width(longest_dst_lines);
1861 max_score_digits = lineno_width(largest_score);
1865 * For debugging -- origin is refcounted, and this asserts that
1866 * we do not underflow.
1868 static void sanity_check_refcnt(struct scoreboard *sb)
1870 int baa = 0;
1871 struct blame_entry *ent;
1873 for (ent = sb->ent; ent; ent = ent->next) {
1874 /* Nobody should have zero or negative refcnt */
1875 if (ent->suspect->refcnt <= 0) {
1876 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1877 ent->suspect->path,
1878 sha1_to_hex(ent->suspect->commit->object.sha1),
1879 ent->suspect->refcnt);
1880 baa = 1;
1883 for (ent = sb->ent; ent; ent = ent->next) {
1884 /* Mark the ones that haven't been checked */
1885 if (0 < ent->suspect->refcnt)
1886 ent->suspect->refcnt = -ent->suspect->refcnt;
1888 for (ent = sb->ent; ent; ent = ent->next) {
1890 * ... then pick each and see if they have the the
1891 * correct refcnt.
1893 int found;
1894 struct blame_entry *e;
1895 struct origin *suspect = ent->suspect;
1897 if (0 < suspect->refcnt)
1898 continue;
1899 suspect->refcnt = -suspect->refcnt; /* Unmark */
1900 for (found = 0, e = sb->ent; e; e = e->next) {
1901 if (e->suspect != suspect)
1902 continue;
1903 found++;
1905 if (suspect->refcnt != found) {
1906 fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1907 ent->suspect->path,
1908 sha1_to_hex(ent->suspect->commit->object.sha1),
1909 ent->suspect->refcnt, found);
1910 baa = 2;
1913 if (baa) {
1914 int opt = 0160;
1915 find_alignment(sb, &opt);
1916 output(sb, opt);
1917 die("Baa %d!", baa);
1922 * Used for the command line parsing; check if the path exists
1923 * in the working tree.
1925 static int has_string_in_work_tree(const char *path)
1927 struct stat st;
1928 return !lstat(path, &st);
1931 static unsigned parse_score(const char *arg)
1933 char *end;
1934 unsigned long score = strtoul(arg, &end, 10);
1935 if (*end)
1936 return 0;
1937 return score;
1940 static const char *add_prefix(const char *prefix, const char *path)
1942 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1946 * Parsing of (comma separated) one item in the -L option
1948 static const char *parse_loc(const char *spec,
1949 struct scoreboard *sb, long lno,
1950 long begin, long *ret)
1952 char *term;
1953 const char *line;
1954 long num;
1955 int reg_error;
1956 regex_t regexp;
1957 regmatch_t match[1];
1959 /* Allow "-L <something>,+20" to mean starting at <something>
1960 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1961 * <something>.
1963 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1964 num = strtol(spec + 1, &term, 10);
1965 if (term != spec + 1) {
1966 if (spec[0] == '-')
1967 num = 0 - num;
1968 if (0 < num)
1969 *ret = begin + num - 2;
1970 else if (!num)
1971 *ret = begin;
1972 else
1973 *ret = begin + num;
1974 return term;
1976 return spec;
1978 num = strtol(spec, &term, 10);
1979 if (term != spec) {
1980 *ret = num;
1981 return term;
1983 if (spec[0] != '/')
1984 return spec;
1986 /* it could be a regexp of form /.../ */
1987 for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1988 if (*term == '\\')
1989 term++;
1991 if (*term != '/')
1992 return spec;
1994 /* try [spec+1 .. term-1] as regexp */
1995 *term = 0;
1996 begin--; /* input is in human terms */
1997 line = nth_line(sb, begin);
1999 if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
2000 !(reg_error = regexec(&regexp, line, 1, match, 0))) {
2001 const char *cp = line + match[0].rm_so;
2002 const char *nline;
2004 while (begin++ < lno) {
2005 nline = nth_line(sb, begin);
2006 if (line <= cp && cp < nline)
2007 break;
2008 line = nline;
2010 *ret = begin;
2011 regfree(&regexp);
2012 *term++ = '/';
2013 return term;
2015 else {
2016 char errbuf[1024];
2017 regerror(reg_error, &regexp, errbuf, 1024);
2018 die("-L parameter '%s': %s", spec + 1, errbuf);
2023 * Parsing of -L option
2025 static void prepare_blame_range(struct scoreboard *sb,
2026 const char *bottomtop,
2027 long lno,
2028 long *bottom, long *top)
2030 const char *term;
2032 term = parse_loc(bottomtop, sb, lno, 1, bottom);
2033 if (*term == ',') {
2034 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2035 if (*term)
2036 usage(blame_usage);
2038 if (*term)
2039 usage(blame_usage);
2042 static int git_blame_config(const char *var, const char *value, void *cb)
2044 if (!strcmp(var, "blame.showroot")) {
2045 show_root = git_config_bool(var, value);
2046 return 0;
2048 if (!strcmp(var, "blame.blankboundary")) {
2049 blank_boundary = git_config_bool(var, value);
2050 return 0;
2052 return git_default_config(var, value, cb);
2056 * Prepare a dummy commit that represents the work tree (or staged) item.
2057 * Note that annotating work tree item never works in the reverse.
2059 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
2061 struct commit *commit;
2062 struct origin *origin;
2063 unsigned char head_sha1[20];
2064 struct strbuf buf;
2065 const char *ident;
2066 time_t now;
2067 int size, len;
2068 struct cache_entry *ce;
2069 unsigned mode;
2071 if (get_sha1("HEAD", head_sha1))
2072 die("No such ref: HEAD");
2074 time(&now);
2075 commit = xcalloc(1, sizeof(*commit));
2076 commit->parents = xcalloc(1, sizeof(*commit->parents));
2077 commit->parents->item = lookup_commit_reference(head_sha1);
2078 commit->object.parsed = 1;
2079 commit->date = now;
2080 commit->object.type = OBJ_COMMIT;
2082 origin = make_origin(commit, path);
2084 strbuf_init(&buf, 0);
2085 if (!contents_from || strcmp("-", contents_from)) {
2086 struct stat st;
2087 const char *read_from;
2088 unsigned long fin_size;
2090 if (contents_from) {
2091 if (stat(contents_from, &st) < 0)
2092 die("Cannot stat %s", contents_from);
2093 read_from = contents_from;
2095 else {
2096 if (lstat(path, &st) < 0)
2097 die("Cannot lstat %s", path);
2098 read_from = path;
2100 fin_size = xsize_t(st.st_size);
2101 mode = canon_mode(st.st_mode);
2102 switch (st.st_mode & S_IFMT) {
2103 case S_IFREG:
2104 if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2105 die("cannot open or read %s", read_from);
2106 break;
2107 case S_IFLNK:
2108 if (readlink(read_from, buf.buf, buf.alloc) != fin_size)
2109 die("cannot readlink %s", read_from);
2110 buf.len = fin_size;
2111 break;
2112 default:
2113 die("unsupported file type %s", read_from);
2116 else {
2117 /* Reading from stdin */
2118 contents_from = "standard input";
2119 mode = 0;
2120 if (strbuf_read(&buf, 0, 0) < 0)
2121 die("read error %s from stdin", strerror(errno));
2123 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2124 origin->file.ptr = buf.buf;
2125 origin->file.size = buf.len;
2126 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2127 commit->util = origin;
2130 * Read the current index, replace the path entry with
2131 * origin->blob_sha1 without mucking with its mode or type
2132 * bits; we are not going to write this index out -- we just
2133 * want to run "diff-index --cached".
2135 discard_cache();
2136 read_cache();
2138 len = strlen(path);
2139 if (!mode) {
2140 int pos = cache_name_pos(path, len);
2141 if (0 <= pos)
2142 mode = active_cache[pos]->ce_mode;
2143 else
2144 /* Let's not bother reading from HEAD tree */
2145 mode = S_IFREG | 0644;
2147 size = cache_entry_size(len);
2148 ce = xcalloc(1, size);
2149 hashcpy(ce->sha1, origin->blob_sha1);
2150 memcpy(ce->name, path, len);
2151 ce->ce_flags = create_ce_flags(len, 0);
2152 ce->ce_mode = create_ce_mode(mode);
2153 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2156 * We are not going to write this out, so this does not matter
2157 * right now, but someday we might optimize diff-index --cached
2158 * with cache-tree information.
2160 cache_tree_invalidate_path(active_cache_tree, path);
2162 commit->buffer = xmalloc(400);
2163 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2164 snprintf(commit->buffer, 400,
2165 "tree 0000000000000000000000000000000000000000\n"
2166 "parent %s\n"
2167 "author %s\n"
2168 "committer %s\n\n"
2169 "Version of %s from %s\n",
2170 sha1_to_hex(head_sha1),
2171 ident, ident, path, contents_from ? contents_from : path);
2172 return commit;
2175 static const char *prepare_final(struct scoreboard *sb)
2177 int i;
2178 const char *final_commit_name = NULL;
2179 struct rev_info *revs = sb->revs;
2182 * There must be one and only one positive commit in the
2183 * revs->pending array.
2185 for (i = 0; i < revs->pending.nr; i++) {
2186 struct object *obj = revs->pending.objects[i].item;
2187 if (obj->flags & UNINTERESTING)
2188 continue;
2189 while (obj->type == OBJ_TAG)
2190 obj = deref_tag(obj, NULL, 0);
2191 if (obj->type != OBJ_COMMIT)
2192 die("Non commit %s?", revs->pending.objects[i].name);
2193 if (sb->final)
2194 die("More than one commit to dig from %s and %s?",
2195 revs->pending.objects[i].name,
2196 final_commit_name);
2197 sb->final = (struct commit *) obj;
2198 final_commit_name = revs->pending.objects[i].name;
2200 return final_commit_name;
2203 static const char *prepare_initial(struct scoreboard *sb)
2205 int i;
2206 const char *final_commit_name = NULL;
2207 struct rev_info *revs = sb->revs;
2210 * There must be one and only one negative commit, and it must be
2211 * the boundary.
2213 for (i = 0; i < revs->pending.nr; i++) {
2214 struct object *obj = revs->pending.objects[i].item;
2215 if (!(obj->flags & UNINTERESTING))
2216 continue;
2217 while (obj->type == OBJ_TAG)
2218 obj = deref_tag(obj, NULL, 0);
2219 if (obj->type != OBJ_COMMIT)
2220 die("Non commit %s?", revs->pending.objects[i].name);
2221 if (sb->final)
2222 die("More than one commit to dig down to %s and %s?",
2223 revs->pending.objects[i].name,
2224 final_commit_name);
2225 sb->final = (struct commit *) obj;
2226 final_commit_name = revs->pending.objects[i].name;
2228 if (!final_commit_name)
2229 die("No commit to dig down to?");
2230 return final_commit_name;
2233 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2235 int *opt = option->value;
2238 * -C enables copy from removed files;
2239 * -C -C enables copy from existing files, but only
2240 * when blaming a new file;
2241 * -C -C -C enables copy from existing files for
2242 * everybody
2244 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2245 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2246 if (*opt & PICKAXE_BLAME_COPY)
2247 *opt |= PICKAXE_BLAME_COPY_HARDER;
2248 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2250 if (arg)
2251 blame_copy_score = parse_score(arg);
2252 return 0;
2255 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2257 int *opt = option->value;
2259 *opt |= PICKAXE_BLAME_MOVE;
2261 if (arg)
2262 blame_move_score = parse_score(arg);
2263 return 0;
2266 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2268 const char **bottomtop = option->value;
2269 if (!arg)
2270 return -1;
2271 if (*bottomtop)
2272 die("More than one '-L n,m' option given");
2273 *bottomtop = arg;
2274 return 0;
2277 int cmd_blame(int argc, const char **argv, const char *prefix)
2279 struct rev_info revs;
2280 const char *path;
2281 struct scoreboard sb;
2282 struct origin *o;
2283 struct blame_entry *ent;
2284 long dashdash_pos, bottom, top, lno;
2285 const char *final_commit_name = NULL;
2286 enum object_type type;
2288 static const char *bottomtop = NULL;
2289 static int output_option = 0, opt = 0;
2290 static int show_stats = 0;
2291 static const char *revs_file = NULL;
2292 static const char *contents_from = NULL;
2293 static const struct option options[] = {
2294 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2295 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2296 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2297 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2298 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2299 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2300 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2301 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2302 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2303 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2304 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2305 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2306 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2307 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2308 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2309 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2310 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2311 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2312 OPT_END()
2315 struct parse_opt_ctx_t ctx;
2317 cmd_is_annotate = !strcmp(argv[0], "annotate");
2319 git_config(git_blame_config, NULL);
2320 init_revisions(&revs, NULL);
2321 save_commit_buffer = 0;
2322 dashdash_pos = 0;
2324 parse_options_start(&ctx, argc, argv, PARSE_OPT_KEEP_DASHDASH |
2325 PARSE_OPT_KEEP_ARGV0);
2326 for (;;) {
2327 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2328 case PARSE_OPT_HELP:
2329 exit(129);
2330 case PARSE_OPT_DONE:
2331 if (ctx.argv[0])
2332 dashdash_pos = ctx.cpidx;
2333 goto parse_done;
2336 if (!strcmp(ctx.argv[0], "--reverse")) {
2337 ctx.argv[0] = "--children";
2338 reverse = 1;
2340 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2342 parse_done:
2343 argc = parse_options_end(&ctx);
2345 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2346 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2347 PICKAXE_BLAME_COPY_HARDER);
2349 if (!blame_move_score)
2350 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2351 if (!blame_copy_score)
2352 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2355 * We have collected options unknown to us in argv[1..unk]
2356 * which are to be passed to revision machinery if we are
2357 * going to do the "bottom" processing.
2359 * The remaining are:
2361 * (1) if dashdash_pos != 0, its either
2362 * "blame [revisions] -- <path>" or
2363 * "blame -- <path> <rev>"
2365 * (2) otherwise, its one of the two:
2366 * "blame [revisions] <path>"
2367 * "blame <path> <rev>"
2369 * Note that we must strip out <path> from the arguments: we do not
2370 * want the path pruning but we may want "bottom" processing.
2372 if (dashdash_pos) {
2373 switch (argc - dashdash_pos - 1) {
2374 case 2: /* (1b) */
2375 if (argc != 4)
2376 usage_with_options(blame_opt_usage, options);
2377 /* reorder for the new way: <rev> -- <path> */
2378 argv[1] = argv[3];
2379 argv[3] = argv[2];
2380 argv[2] = "--";
2381 /* FALLTHROUGH */
2382 case 1: /* (1a) */
2383 path = add_prefix(prefix, argv[--argc]);
2384 argv[argc] = NULL;
2385 break;
2386 default:
2387 usage_with_options(blame_opt_usage, options);
2389 } else {
2390 if (argc < 2)
2391 usage_with_options(blame_opt_usage, options);
2392 path = add_prefix(prefix, argv[argc - 1]);
2393 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2394 path = add_prefix(prefix, argv[1]);
2395 argv[1] = argv[2];
2397 argv[argc - 1] = "--";
2399 setup_work_tree();
2400 if (!has_string_in_work_tree(path))
2401 die("cannot stat path %s: %s", path, strerror(errno));
2404 setup_revisions(argc, argv, &revs, NULL);
2405 memset(&sb, 0, sizeof(sb));
2407 sb.revs = &revs;
2408 if (!reverse)
2409 final_commit_name = prepare_final(&sb);
2410 else if (contents_from)
2411 die("--contents and --children do not blend well.");
2412 else
2413 final_commit_name = prepare_initial(&sb);
2415 if (!sb.final) {
2417 * "--not A B -- path" without anything positive;
2418 * do not default to HEAD, but use the working tree
2419 * or "--contents".
2421 setup_work_tree();
2422 sb.final = fake_working_tree_commit(path, contents_from);
2423 add_pending_object(&revs, &(sb.final->object), ":");
2425 else if (contents_from)
2426 die("Cannot use --contents with final commit object name");
2429 * If we have bottom, this will mark the ancestors of the
2430 * bottom commits we would reach while traversing as
2431 * uninteresting.
2433 if (prepare_revision_walk(&revs))
2434 die("revision walk setup failed");
2436 if (is_null_sha1(sb.final->object.sha1)) {
2437 char *buf;
2438 o = sb.final->util;
2439 buf = xmalloc(o->file.size + 1);
2440 memcpy(buf, o->file.ptr, o->file.size + 1);
2441 sb.final_buf = buf;
2442 sb.final_buf_size = o->file.size;
2444 else {
2445 o = get_origin(&sb, sb.final, path);
2446 if (fill_blob_sha1(o))
2447 die("no such path %s in %s", path, final_commit_name);
2449 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2450 &sb.final_buf_size);
2451 if (!sb.final_buf)
2452 die("Cannot read blob %s for path %s",
2453 sha1_to_hex(o->blob_sha1),
2454 path);
2456 num_read_blob++;
2457 lno = prepare_lines(&sb);
2459 bottom = top = 0;
2460 if (bottomtop)
2461 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2462 if (bottom && top && top < bottom) {
2463 long tmp;
2464 tmp = top; top = bottom; bottom = tmp;
2466 if (bottom < 1)
2467 bottom = 1;
2468 if (top < 1)
2469 top = lno;
2470 bottom--;
2471 if (lno < top)
2472 die("file %s has only %lu lines", path, lno);
2474 ent = xcalloc(1, sizeof(*ent));
2475 ent->lno = bottom;
2476 ent->num_lines = top - bottom;
2477 ent->suspect = o;
2478 ent->s_lno = bottom;
2480 sb.ent = ent;
2481 sb.path = path;
2483 if (revs_file && read_ancestry(revs_file))
2484 die("reading graft file %s failed: %s",
2485 revs_file, strerror(errno));
2487 read_mailmap(&mailmap, ".mailmap", NULL);
2489 if (!incremental)
2490 setup_pager();
2492 assign_blame(&sb, opt);
2494 if (incremental)
2495 return 0;
2497 coalesce(&sb);
2499 if (!(output_option & OUTPUT_PORCELAIN))
2500 find_alignment(&sb, &output_option);
2502 output(&sb, output_option);
2503 free((void *)sb.final_buf);
2504 for (ent = sb.ent; ent; ) {
2505 struct blame_entry *e = ent->next;
2506 free(ent);
2507 ent = e;
2510 if (show_stats) {
2511 printf("num read blob: %d\n", num_read_blob);
2512 printf("num get patch: %d\n", num_get_patch);
2513 printf("num commits: %d\n", num_commits);
2515 return 0;