4 * Copyright (c) 2006, Junio C Hamano
12 #include "tree-walk.h"
17 #include "xdiff-interface.h"
18 #include "cache-tree.h"
19 #include "string-list.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
[] = {
28 "[rev-opts] are documented in git-rev-list(1)",
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
;
39 static int blank_boundary
;
40 static int incremental
;
41 static int xdl_opts
= XDF_NEED_MINIMAL
;
42 static struct string_list mailmap
;
49 static int num_read_blob
;
50 static int num_get_patch
;
51 static int num_commits
;
53 #define PICKAXE_BLAME_MOVE 01
54 #define PICKAXE_BLAME_COPY 02
55 #define PICKAXE_BLAME_COPY_HARDER 04
56 #define PICKAXE_BLAME_COPY_HARDEST 010
59 * blame for a blame_entry with score lower than these thresholds
60 * is not passed to the parent using move/copy logic.
62 static unsigned blame_move_score
;
63 static unsigned blame_copy_score
;
64 #define BLAME_DEFAULT_MOVE_SCORE 20
65 #define BLAME_DEFAULT_COPY_SCORE 40
67 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
68 #define METAINFO_SHOWN (1u<<12)
69 #define MORE_THAN_ONE_PATH (1u<<13)
72 * One blob in a commit that is being suspected
76 struct origin
*previous
;
77 struct commit
*commit
;
79 unsigned char blob_sha1
[20];
80 char path
[FLEX_ARRAY
];
84 * Given an origin, prepare mmfile_t structure to be used by the
87 static void fill_origin_blob(struct origin
*o
, mmfile_t
*file
)
90 enum object_type type
;
92 file
->ptr
= read_sha1_file(o
->blob_sha1
, &type
,
93 (unsigned long *)(&(file
->size
)));
95 die("Cannot read blob %s for path %s",
96 sha1_to_hex(o
->blob_sha1
),
105 * Origin is refcounted and usually we keep the blob contents to be
108 static inline struct origin
*origin_incref(struct origin
*o
)
115 static void origin_decref(struct origin
*o
)
117 if (o
&& --o
->refcnt
<= 0) {
119 origin_decref(o
->previous
);
125 static void drop_origin_blob(struct origin
*o
)
134 * Each group of lines is described by a blame_entry; it can be split
135 * as we pass blame to the parents. They form a linked list in the
136 * scoreboard structure, sorted by the target line number.
139 struct blame_entry
*prev
;
140 struct blame_entry
*next
;
142 /* the first line of this group in the final image;
143 * internally all line numbers are 0 based.
147 /* how many lines this group has */
150 /* the commit that introduced this group into the final image */
151 struct origin
*suspect
;
153 /* true if the suspect is truly guilty; false while we have not
154 * checked if the group came from one of its parents.
158 /* true if the entry has been scanned for copies in the current parent
162 /* the line number of the first line of this group in the
163 * suspect's file; internally all line numbers are 0 based.
167 /* how significant this entry is -- cached to avoid
168 * scanning the lines over and over.
174 * The current state of the blame assignment.
177 /* the final commit (i.e. where we started digging from) */
178 struct commit
*final
;
179 struct rev_info
*revs
;
183 * The contents in the final image.
184 * Used by many functions to obtain contents of the nth line,
185 * indexed with scoreboard.lineno[blame_entry.lno].
187 const char *final_buf
;
188 unsigned long final_buf_size
;
190 /* linked list of blames */
191 struct blame_entry
*ent
;
193 /* look-up a line in the final buffer */
198 static inline int same_suspect(struct origin
*a
, struct origin
*b
)
202 if (a
->commit
!= b
->commit
)
204 return !strcmp(a
->path
, b
->path
);
207 static void sanity_check_refcnt(struct scoreboard
*);
210 * If two blame entries that are next to each other came from
211 * contiguous lines in the same origin (i.e. <commit, path> pair),
212 * merge them together.
214 static void coalesce(struct scoreboard
*sb
)
216 struct blame_entry
*ent
, *next
;
218 for (ent
= sb
->ent
; ent
&& (next
= ent
->next
); ent
= next
) {
219 if (same_suspect(ent
->suspect
, next
->suspect
) &&
220 ent
->guilty
== next
->guilty
&&
221 ent
->s_lno
+ ent
->num_lines
== next
->s_lno
) {
222 ent
->num_lines
+= next
->num_lines
;
223 ent
->next
= next
->next
;
225 ent
->next
->prev
= ent
;
226 origin_decref(next
->suspect
);
229 next
= ent
; /* again */
233 if (DEBUG
) /* sanity */
234 sanity_check_refcnt(sb
);
238 * Given a commit and a path in it, create a new origin structure.
239 * The callers that add blame to the scoreboard should use
240 * get_origin() to obtain shared, refcounted copy instead of calling
241 * this function directly.
243 static struct origin
*make_origin(struct commit
*commit
, const char *path
)
246 o
= xcalloc(1, sizeof(*o
) + strlen(path
) + 1);
249 strcpy(o
->path
, path
);
254 * Locate an existing origin or create a new one.
256 static struct origin
*get_origin(struct scoreboard
*sb
,
257 struct commit
*commit
,
260 struct blame_entry
*e
;
262 for (e
= sb
->ent
; e
; e
= e
->next
) {
263 if (e
->suspect
->commit
== commit
&&
264 !strcmp(e
->suspect
->path
, path
))
265 return origin_incref(e
->suspect
);
267 return make_origin(commit
, path
);
271 * Fill the blob_sha1 field of an origin if it hasn't, so that later
272 * call to fill_origin_blob() can use it to locate the data. blob_sha1
273 * for an origin is also used to pass the blame for the entire file to
274 * the parent to detect the case where a child's blob is identical to
275 * that of its parent's.
277 static int fill_blob_sha1(struct origin
*origin
)
281 if (!is_null_sha1(origin
->blob_sha1
))
283 if (get_tree_entry(origin
->commit
->object
.sha1
,
285 origin
->blob_sha1
, &mode
))
287 if (sha1_object_info(origin
->blob_sha1
, NULL
) != OBJ_BLOB
)
291 hashclr(origin
->blob_sha1
);
296 * We have an origin -- check if the same path exists in the
297 * parent and return an origin structure to represent it.
299 static struct origin
*find_origin(struct scoreboard
*sb
,
300 struct commit
*parent
,
301 struct origin
*origin
)
303 struct origin
*porigin
= NULL
;
304 struct diff_options diff_opts
;
305 const char *paths
[2];
309 * Each commit object can cache one origin in that
310 * commit. This is a freestanding copy of origin and
313 struct origin
*cached
= parent
->util
;
314 if (!strcmp(cached
->path
, origin
->path
)) {
316 * The same path between origin and its parent
317 * without renaming -- the most common case.
319 porigin
= get_origin(sb
, parent
, cached
->path
);
322 * If the origin was newly created (i.e. get_origin
323 * would call make_origin if none is found in the
324 * scoreboard), it does not know the blob_sha1,
325 * so copy it. Otherwise porigin was in the
326 * scoreboard and already knows blob_sha1.
328 if (porigin
->refcnt
== 1)
329 hashcpy(porigin
->blob_sha1
, cached
->blob_sha1
);
332 /* otherwise it was not very useful; free it */
337 /* See if the origin->path is different between parent
338 * and origin first. Most of the time they are the
339 * same and diff-tree is fairly efficient about this.
341 diff_setup(&diff_opts
);
342 DIFF_OPT_SET(&diff_opts
, RECURSIVE
);
343 diff_opts
.detect_rename
= 0;
344 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
345 paths
[0] = origin
->path
;
348 diff_tree_setup_paths(paths
, &diff_opts
);
349 if (diff_setup_done(&diff_opts
) < 0)
352 if (is_null_sha1(origin
->commit
->object
.sha1
))
353 do_diff_cache(parent
->tree
->object
.sha1
, &diff_opts
);
355 diff_tree_sha1(parent
->tree
->object
.sha1
,
356 origin
->commit
->tree
->object
.sha1
,
358 diffcore_std(&diff_opts
);
360 /* It is either one entry that says "modified", or "created",
363 if (!diff_queued_diff
.nr
) {
364 /* The path is the same as parent */
365 porigin
= get_origin(sb
, parent
, origin
->path
);
366 hashcpy(porigin
->blob_sha1
, origin
->blob_sha1
);
368 else if (diff_queued_diff
.nr
!= 1)
369 die("internal error in blame::find_origin");
371 struct diff_filepair
*p
= diff_queued_diff
.queue
[0];
374 die("internal error in blame::find_origin (%c)",
377 porigin
= get_origin(sb
, parent
, origin
->path
);
378 hashcpy(porigin
->blob_sha1
, p
->one
->sha1
);
382 /* Did not exist in parent, or type changed */
386 diff_flush(&diff_opts
);
387 diff_tree_release_paths(&diff_opts
);
390 * Create a freestanding copy that is not part of
391 * the refcounted origin found in the scoreboard, and
392 * cache it in the commit.
394 struct origin
*cached
;
396 cached
= make_origin(porigin
->commit
, porigin
->path
);
397 hashcpy(cached
->blob_sha1
, porigin
->blob_sha1
);
398 parent
->util
= cached
;
404 * We have an origin -- find the path that corresponds to it in its
405 * parent and return an origin structure to represent it.
407 static struct origin
*find_rename(struct scoreboard
*sb
,
408 struct commit
*parent
,
409 struct origin
*origin
)
411 struct origin
*porigin
= NULL
;
412 struct diff_options diff_opts
;
414 const char *paths
[2];
416 diff_setup(&diff_opts
);
417 DIFF_OPT_SET(&diff_opts
, RECURSIVE
);
418 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
419 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
420 diff_opts
.single_follow
= origin
->path
;
422 diff_tree_setup_paths(paths
, &diff_opts
);
423 if (diff_setup_done(&diff_opts
) < 0)
426 if (is_null_sha1(origin
->commit
->object
.sha1
))
427 do_diff_cache(parent
->tree
->object
.sha1
, &diff_opts
);
429 diff_tree_sha1(parent
->tree
->object
.sha1
,
430 origin
->commit
->tree
->object
.sha1
,
432 diffcore_std(&diff_opts
);
434 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
435 struct diff_filepair
*p
= diff_queued_diff
.queue
[i
];
436 if ((p
->status
== 'R' || p
->status
== 'C') &&
437 !strcmp(p
->two
->path
, origin
->path
)) {
438 porigin
= get_origin(sb
, parent
, p
->one
->path
);
439 hashcpy(porigin
->blob_sha1
, p
->one
->sha1
);
443 diff_flush(&diff_opts
);
444 diff_tree_release_paths(&diff_opts
);
449 * Parsing of patch chunks...
452 /* line number in postimage; up to but not including this
453 * line is the same as preimage
457 /* preimage line number after this chunk */
460 /* postimage line number after this chunk */
465 struct chunk
*chunks
;
469 struct blame_diff_state
{
471 unsigned hunk_post_context
;
472 unsigned hunk_in_pre_context
: 1;
475 static void process_u_diff(void *state_
, char *line
, unsigned long len
)
477 struct blame_diff_state
*state
= state_
;
479 int off1
, off2
, len1
, len2
, num
;
481 num
= state
->ret
->num
;
482 if (len
< 4 || line
[0] != '@' || line
[1] != '@') {
483 if (state
->hunk_in_pre_context
&& line
[0] == ' ')
484 state
->ret
->chunks
[num
- 1].same
++;
486 state
->hunk_in_pre_context
= 0;
488 state
->hunk_post_context
++;
490 state
->hunk_post_context
= 0;
495 if (num
&& state
->hunk_post_context
) {
496 chunk
= &state
->ret
->chunks
[num
- 1];
497 chunk
->p_next
-= state
->hunk_post_context
;
498 chunk
->t_next
-= state
->hunk_post_context
;
500 state
->ret
->num
= ++num
;
501 state
->ret
->chunks
= xrealloc(state
->ret
->chunks
,
502 sizeof(struct chunk
) * num
);
503 chunk
= &state
->ret
->chunks
[num
- 1];
504 if (parse_hunk_header(line
, len
, &off1
, &len1
, &off2
, &len2
)) {
509 /* Line numbers in patch output are one based. */
513 chunk
->same
= len2
? off2
: (off2
+ 1);
515 chunk
->p_next
= off1
+ (len1
? len1
: 1);
516 chunk
->t_next
= chunk
->same
+ len2
;
517 state
->hunk_in_pre_context
= 1;
518 state
->hunk_post_context
= 0;
521 static struct patch
*compare_buffer(mmfile_t
*file_p
, mmfile_t
*file_o
,
524 struct blame_diff_state state
;
529 xpp
.flags
= xdl_opts
;
530 memset(&xecfg
, 0, sizeof(xecfg
));
531 xecfg
.ctxlen
= context
;
532 memset(&state
, 0, sizeof(state
));
533 state
.ret
= xmalloc(sizeof(struct patch
));
534 state
.ret
->chunks
= NULL
;
537 xdi_diff_outf(file_p
, file_o
, process_u_diff
, &state
, &xpp
, &xecfg
, &ecb
);
539 if (state
.ret
->num
) {
541 chunk
= &state
.ret
->chunks
[state
.ret
->num
- 1];
542 chunk
->p_next
-= state
.hunk_post_context
;
543 chunk
->t_next
-= state
.hunk_post_context
;
549 * Run diff between two origins and grab the patch output, so that
550 * we can pass blame for lines origin is currently suspected for
553 static struct patch
*get_patch(struct origin
*parent
, struct origin
*origin
)
555 mmfile_t file_p
, file_o
;
558 fill_origin_blob(parent
, &file_p
);
559 fill_origin_blob(origin
, &file_o
);
560 if (!file_p
.ptr
|| !file_o
.ptr
)
562 patch
= compare_buffer(&file_p
, &file_o
, 0);
567 static void free_patch(struct patch
*p
)
574 * Link in a new blame entry to the scoreboard. Entries that cover the
575 * same line range have been removed from the scoreboard previously.
577 static void add_blame_entry(struct scoreboard
*sb
, struct blame_entry
*e
)
579 struct blame_entry
*ent
, *prev
= NULL
;
581 origin_incref(e
->suspect
);
583 for (ent
= sb
->ent
; ent
&& ent
->lno
< e
->lno
; ent
= ent
->next
)
586 /* prev, if not NULL, is the last one that is below e */
589 e
->next
= prev
->next
;
601 * src typically is on-stack; we want to copy the information in it to
602 * a malloced blame_entry that is already on the linked list of the
603 * scoreboard. The origin of dst loses a refcnt while the origin of src
606 static void dup_entry(struct blame_entry
*dst
, struct blame_entry
*src
)
608 struct blame_entry
*p
, *n
;
612 origin_incref(src
->suspect
);
613 origin_decref(dst
->suspect
);
614 memcpy(dst
, src
, sizeof(*src
));
620 static const char *nth_line(struct scoreboard
*sb
, int lno
)
622 return sb
->final_buf
+ sb
->lineno
[lno
];
626 * It is known that lines between tlno to same came from parent, and e
627 * has an overlap with that range. it also is known that parent's
628 * line plno corresponds to e's line tlno.
634 * <------------------>
636 * Split e into potentially three parts; before this chunk, the chunk
637 * to be blamed for the parent, and after that portion.
639 static void split_overlap(struct blame_entry
*split
,
640 struct blame_entry
*e
,
641 int tlno
, int plno
, int same
,
642 struct origin
*parent
)
645 memset(split
, 0, sizeof(struct blame_entry
[3]));
647 if (e
->s_lno
< tlno
) {
648 /* there is a pre-chunk part not blamed on parent */
649 split
[0].suspect
= origin_incref(e
->suspect
);
650 split
[0].lno
= e
->lno
;
651 split
[0].s_lno
= e
->s_lno
;
652 split
[0].num_lines
= tlno
- e
->s_lno
;
653 split
[1].lno
= e
->lno
+ tlno
- e
->s_lno
;
654 split
[1].s_lno
= plno
;
657 split
[1].lno
= e
->lno
;
658 split
[1].s_lno
= plno
+ (e
->s_lno
- tlno
);
661 if (same
< e
->s_lno
+ e
->num_lines
) {
662 /* there is a post-chunk part not blamed on parent */
663 split
[2].suspect
= origin_incref(e
->suspect
);
664 split
[2].lno
= e
->lno
+ (same
- e
->s_lno
);
665 split
[2].s_lno
= e
->s_lno
+ (same
- e
->s_lno
);
666 split
[2].num_lines
= e
->s_lno
+ e
->num_lines
- same
;
667 chunk_end_lno
= split
[2].lno
;
670 chunk_end_lno
= e
->lno
+ e
->num_lines
;
671 split
[1].num_lines
= chunk_end_lno
- split
[1].lno
;
674 * if it turns out there is nothing to blame the parent for,
675 * forget about the splitting. !split[1].suspect signals this.
677 if (split
[1].num_lines
< 1)
679 split
[1].suspect
= origin_incref(parent
);
683 * split_overlap() divided an existing blame e into up to three parts
684 * in split. Adjust the linked list of blames in the scoreboard to
687 static void split_blame(struct scoreboard
*sb
,
688 struct blame_entry
*split
,
689 struct blame_entry
*e
)
691 struct blame_entry
*new_entry
;
693 if (split
[0].suspect
&& split
[2].suspect
) {
694 /* The first part (reuse storage for the existing entry e) */
695 dup_entry(e
, &split
[0]);
697 /* The last part -- me */
698 new_entry
= xmalloc(sizeof(*new_entry
));
699 memcpy(new_entry
, &(split
[2]), sizeof(struct blame_entry
));
700 add_blame_entry(sb
, new_entry
);
702 /* ... and the middle part -- parent */
703 new_entry
= xmalloc(sizeof(*new_entry
));
704 memcpy(new_entry
, &(split
[1]), sizeof(struct blame_entry
));
705 add_blame_entry(sb
, new_entry
);
707 else if (!split
[0].suspect
&& !split
[2].suspect
)
709 * The parent covers the entire area; reuse storage for
710 * e and replace it with the parent.
712 dup_entry(e
, &split
[1]);
713 else if (split
[0].suspect
) {
714 /* me and then parent */
715 dup_entry(e
, &split
[0]);
717 new_entry
= xmalloc(sizeof(*new_entry
));
718 memcpy(new_entry
, &(split
[1]), sizeof(struct blame_entry
));
719 add_blame_entry(sb
, new_entry
);
722 /* parent and then me */
723 dup_entry(e
, &split
[1]);
725 new_entry
= xmalloc(sizeof(*new_entry
));
726 memcpy(new_entry
, &(split
[2]), sizeof(struct blame_entry
));
727 add_blame_entry(sb
, new_entry
);
730 if (DEBUG
) { /* sanity */
731 struct blame_entry
*ent
;
732 int lno
= sb
->ent
->lno
, corrupt
= 0;
734 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
739 lno
+= ent
->num_lines
;
743 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
744 printf("L %8d l %8d n %8d\n",
745 lno
, ent
->lno
, ent
->num_lines
);
746 lno
= ent
->lno
+ ent
->num_lines
;
754 * After splitting the blame, the origins used by the
755 * on-stack blame_entry should lose one refcnt each.
757 static void decref_split(struct blame_entry
*split
)
761 for (i
= 0; i
< 3; i
++)
762 origin_decref(split
[i
].suspect
);
766 * Helper for blame_chunk(). blame_entry e is known to overlap with
767 * the patch hunk; split it and pass blame to the parent.
769 static void blame_overlap(struct scoreboard
*sb
, struct blame_entry
*e
,
770 int tlno
, int plno
, int same
,
771 struct origin
*parent
)
773 struct blame_entry split
[3];
775 split_overlap(split
, e
, tlno
, plno
, same
, parent
);
776 if (split
[1].suspect
)
777 split_blame(sb
, split
, e
);
782 * Find the line number of the last line the target is suspected for.
784 static int find_last_in_target(struct scoreboard
*sb
, struct origin
*target
)
786 struct blame_entry
*e
;
787 int last_in_target
= -1;
789 for (e
= sb
->ent
; e
; e
= e
->next
) {
790 if (e
->guilty
|| !same_suspect(e
->suspect
, target
))
792 if (last_in_target
< e
->s_lno
+ e
->num_lines
)
793 last_in_target
= e
->s_lno
+ e
->num_lines
;
795 return last_in_target
;
799 * Process one hunk from the patch between the current suspect for
800 * blame_entry e and its parent. Find and split the overlap, and
801 * pass blame to the overlapping part to the parent.
803 static void blame_chunk(struct scoreboard
*sb
,
804 int tlno
, int plno
, int same
,
805 struct origin
*target
, struct origin
*parent
)
807 struct blame_entry
*e
;
809 for (e
= sb
->ent
; e
; e
= e
->next
) {
810 if (e
->guilty
|| !same_suspect(e
->suspect
, target
))
812 if (same
<= e
->s_lno
)
814 if (tlno
< e
->s_lno
+ e
->num_lines
)
815 blame_overlap(sb
, e
, tlno
, plno
, same
, parent
);
820 * We are looking at the origin 'target' and aiming to pass blame
821 * for the lines it is suspected to its parent. Run diff to find
822 * which lines came from parent and pass blame for them.
824 static int pass_blame_to_parent(struct scoreboard
*sb
,
825 struct origin
*target
,
826 struct origin
*parent
)
828 int i
, last_in_target
, plno
, tlno
;
831 last_in_target
= find_last_in_target(sb
, target
);
832 if (last_in_target
< 0)
833 return 1; /* nothing remains for this target */
835 patch
= get_patch(parent
, target
);
837 for (i
= 0; i
< patch
->num
; i
++) {
838 struct chunk
*chunk
= &patch
->chunks
[i
];
840 blame_chunk(sb
, tlno
, plno
, chunk
->same
, target
, parent
);
841 plno
= chunk
->p_next
;
842 tlno
= chunk
->t_next
;
844 /* The rest (i.e. anything after tlno) are the same as the parent */
845 blame_chunk(sb
, tlno
, plno
, last_in_target
, target
, parent
);
852 * The lines in blame_entry after splitting blames many times can become
853 * very small and trivial, and at some point it becomes pointless to
854 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
855 * ordinary C program, and it is not worth to say it was copied from
856 * totally unrelated file in the parent.
858 * Compute how trivial the lines in the blame_entry are.
860 static unsigned ent_score(struct scoreboard
*sb
, struct blame_entry
*e
)
869 cp
= nth_line(sb
, e
->lno
);
870 ep
= nth_line(sb
, e
->lno
+ e
->num_lines
);
872 unsigned ch
= *((unsigned char *)cp
);
882 * best_so_far[] and this[] are both a split of an existing blame_entry
883 * that passes blame to the parent. Maintain best_so_far the best split
884 * so far, by comparing this and best_so_far and copying this into
885 * bst_so_far as needed.
887 static void copy_split_if_better(struct scoreboard
*sb
,
888 struct blame_entry
*best_so_far
,
889 struct blame_entry
*this)
893 if (!this[1].suspect
)
895 if (best_so_far
[1].suspect
) {
896 if (ent_score(sb
, &this[1]) < ent_score(sb
, &best_so_far
[1]))
900 for (i
= 0; i
< 3; i
++)
901 origin_incref(this[i
].suspect
);
902 decref_split(best_so_far
);
903 memcpy(best_so_far
, this, sizeof(struct blame_entry
[3]));
907 * We are looking at a part of the final image represented by
908 * ent (tlno and same are offset by ent->s_lno).
909 * tlno is where we are looking at in the final image.
910 * up to (but not including) same match preimage.
911 * plno is where we are looking at in the preimage.
913 * <-------------- final image ---------------------->
916 * <---------preimage----->
919 * All line numbers are 0-based.
921 static void handle_split(struct scoreboard
*sb
,
922 struct blame_entry
*ent
,
923 int tlno
, int plno
, int same
,
924 struct origin
*parent
,
925 struct blame_entry
*split
)
927 if (ent
->num_lines
<= tlno
)
930 struct blame_entry
this[3];
933 split_overlap(this, ent
, tlno
, plno
, same
, parent
);
934 copy_split_if_better(sb
, split
, this);
940 * Find the lines from parent that are the same as ent so that
941 * we can pass blames to it. file_p has the blob contents for
944 static void find_copy_in_blob(struct scoreboard
*sb
,
945 struct blame_entry
*ent
,
946 struct origin
*parent
,
947 struct blame_entry
*split
,
957 * Prepare mmfile that contains only the lines in ent.
959 cp
= nth_line(sb
, ent
->lno
);
960 file_o
.ptr
= (char*) cp
;
961 cnt
= ent
->num_lines
;
963 while (cnt
&& cp
< sb
->final_buf
+ sb
->final_buf_size
) {
967 file_o
.size
= cp
- file_o
.ptr
;
969 patch
= compare_buffer(file_p
, &file_o
, 1);
972 * file_o is a part of final image we are annotating.
973 * file_p partially may match that image.
975 memset(split
, 0, sizeof(struct blame_entry
[3]));
977 for (i
= 0; i
< patch
->num
; i
++) {
978 struct chunk
*chunk
= &patch
->chunks
[i
];
980 handle_split(sb
, ent
, tlno
, plno
, chunk
->same
, parent
, split
);
981 plno
= chunk
->p_next
;
982 tlno
= chunk
->t_next
;
984 /* remainder, if any, all match the preimage */
985 handle_split(sb
, ent
, tlno
, plno
, ent
->num_lines
, parent
, split
);
990 * See if lines currently target is suspected for can be attributed to
993 static int find_move_in_parent(struct scoreboard
*sb
,
994 struct origin
*target
,
995 struct origin
*parent
)
997 int last_in_target
, made_progress
;
998 struct blame_entry
*e
, split
[3];
1001 last_in_target
= find_last_in_target(sb
, target
);
1002 if (last_in_target
< 0)
1003 return 1; /* nothing remains for this target */
1005 fill_origin_blob(parent
, &file_p
);
1010 while (made_progress
) {
1012 for (e
= sb
->ent
; e
; e
= e
->next
) {
1013 if (e
->guilty
|| !same_suspect(e
->suspect
, target
) ||
1014 ent_score(sb
, e
) < blame_move_score
)
1016 find_copy_in_blob(sb
, e
, parent
, split
, &file_p
);
1017 if (split
[1].suspect
&&
1018 blame_move_score
< ent_score(sb
, &split
[1])) {
1019 split_blame(sb
, split
, e
);
1022 decref_split(split
);
1029 struct blame_entry
*ent
;
1030 struct blame_entry split
[3];
1034 * Count the number of entries the target is suspected for,
1035 * and prepare a list of entry and the best split.
1037 static struct blame_list
*setup_blame_list(struct scoreboard
*sb
,
1038 struct origin
*target
,
1042 struct blame_entry
*e
;
1044 struct blame_list
*blame_list
= NULL
;
1046 for (e
= sb
->ent
, num_ents
= 0; e
; e
= e
->next
)
1047 if (!e
->scanned
&& !e
->guilty
&&
1048 same_suspect(e
->suspect
, target
) &&
1049 min_score
< ent_score(sb
, e
))
1052 blame_list
= xcalloc(num_ents
, sizeof(struct blame_list
));
1053 for (e
= sb
->ent
, i
= 0; e
; e
= e
->next
)
1054 if (!e
->scanned
&& !e
->guilty
&&
1055 same_suspect(e
->suspect
, target
) &&
1056 min_score
< ent_score(sb
, e
))
1057 blame_list
[i
++].ent
= e
;
1059 *num_ents_p
= num_ents
;
1064 * Reset the scanned status on all entries.
1066 static void reset_scanned_flag(struct scoreboard
*sb
)
1068 struct blame_entry
*e
;
1069 for (e
= sb
->ent
; e
; e
= e
->next
)
1074 * For lines target is suspected for, see if we can find code movement
1075 * across file boundary from the parent commit. porigin is the path
1076 * in the parent we already tried.
1078 static int find_copy_in_parent(struct scoreboard
*sb
,
1079 struct origin
*target
,
1080 struct commit
*parent
,
1081 struct origin
*porigin
,
1084 struct diff_options diff_opts
;
1085 const char *paths
[1];
1088 struct blame_list
*blame_list
;
1091 blame_list
= setup_blame_list(sb
, target
, blame_copy_score
, &num_ents
);
1093 return 1; /* nothing remains for this target */
1095 diff_setup(&diff_opts
);
1096 DIFF_OPT_SET(&diff_opts
, RECURSIVE
);
1097 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
1100 diff_tree_setup_paths(paths
, &diff_opts
);
1101 if (diff_setup_done(&diff_opts
) < 0)
1104 /* Try "find copies harder" on new path if requested;
1105 * we do not want to use diffcore_rename() actually to
1106 * match things up; find_copies_harder is set only to
1107 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1108 * and this code needs to be after diff_setup_done(), which
1109 * usually makes find-copies-harder imply copy detection.
1111 if ((opt
& PICKAXE_BLAME_COPY_HARDEST
)
1112 || ((opt
& PICKAXE_BLAME_COPY_HARDER
)
1113 && (!porigin
|| strcmp(target
->path
, porigin
->path
))))
1114 DIFF_OPT_SET(&diff_opts
, FIND_COPIES_HARDER
);
1116 if (is_null_sha1(target
->commit
->object
.sha1
))
1117 do_diff_cache(parent
->tree
->object
.sha1
, &diff_opts
);
1119 diff_tree_sha1(parent
->tree
->object
.sha1
,
1120 target
->commit
->tree
->object
.sha1
,
1123 if (!DIFF_OPT_TST(&diff_opts
, FIND_COPIES_HARDER
))
1124 diffcore_std(&diff_opts
);
1128 int made_progress
= 0;
1130 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
1131 struct diff_filepair
*p
= diff_queued_diff
.queue
[i
];
1132 struct origin
*norigin
;
1134 struct blame_entry
this[3];
1136 if (!DIFF_FILE_VALID(p
->one
))
1137 continue; /* does not exist in parent */
1138 if (S_ISGITLINK(p
->one
->mode
))
1139 continue; /* ignore git links */
1140 if (porigin
&& !strcmp(p
->one
->path
, porigin
->path
))
1141 /* find_move already dealt with this path */
1144 norigin
= get_origin(sb
, parent
, p
->one
->path
);
1145 hashcpy(norigin
->blob_sha1
, p
->one
->sha1
);
1146 fill_origin_blob(norigin
, &file_p
);
1150 for (j
= 0; j
< num_ents
; j
++) {
1151 find_copy_in_blob(sb
, blame_list
[j
].ent
,
1152 norigin
, this, &file_p
);
1153 copy_split_if_better(sb
, blame_list
[j
].split
,
1157 origin_decref(norigin
);
1160 for (j
= 0; j
< num_ents
; j
++) {
1161 struct blame_entry
*split
= blame_list
[j
].split
;
1162 if (split
[1].suspect
&&
1163 blame_copy_score
< ent_score(sb
, &split
[1])) {
1164 split_blame(sb
, split
, blame_list
[j
].ent
);
1168 blame_list
[j
].ent
->scanned
= 1;
1169 decref_split(split
);
1175 blame_list
= setup_blame_list(sb
, target
, blame_copy_score
, &num_ents
);
1181 reset_scanned_flag(sb
);
1182 diff_flush(&diff_opts
);
1183 diff_tree_release_paths(&diff_opts
);
1188 * The blobs of origin and porigin exactly match, so everything
1189 * origin is suspected for can be blamed on the parent.
1191 static void pass_whole_blame(struct scoreboard
*sb
,
1192 struct origin
*origin
, struct origin
*porigin
)
1194 struct blame_entry
*e
;
1196 if (!porigin
->file
.ptr
&& origin
->file
.ptr
) {
1197 /* Steal its file */
1198 porigin
->file
= origin
->file
;
1199 origin
->file
.ptr
= NULL
;
1201 for (e
= sb
->ent
; e
; e
= e
->next
) {
1202 if (!same_suspect(e
->suspect
, origin
))
1204 origin_incref(porigin
);
1205 origin_decref(e
->suspect
);
1206 e
->suspect
= porigin
;
1211 * We pass blame from the current commit to its parents. We keep saying
1212 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1213 * exonerate ourselves.
1215 static struct commit_list
*first_scapegoat(struct rev_info
*revs
, struct commit
*commit
)
1218 return commit
->parents
;
1219 return lookup_decoration(&revs
->children
, &commit
->object
);
1222 static int num_scapegoats(struct rev_info
*revs
, struct commit
*commit
)
1225 struct commit_list
*l
= first_scapegoat(revs
, commit
);
1226 for (cnt
= 0; l
; l
= l
->next
)
1233 static void pass_blame(struct scoreboard
*sb
, struct origin
*origin
, int opt
)
1235 struct rev_info
*revs
= sb
->revs
;
1236 int i
, pass
, num_sg
;
1237 struct commit
*commit
= origin
->commit
;
1238 struct commit_list
*sg
;
1239 struct origin
*sg_buf
[MAXSG
];
1240 struct origin
*porigin
, **sg_origin
= sg_buf
;
1242 num_sg
= num_scapegoats(revs
, commit
);
1245 else if (num_sg
< ARRAY_SIZE(sg_buf
))
1246 memset(sg_buf
, 0, sizeof(sg_buf
));
1248 sg_origin
= xcalloc(num_sg
, sizeof(*sg_origin
));
1251 * The first pass looks for unrenamed path to optimize for
1252 * common cases, then we look for renames in the second pass.
1254 for (pass
= 0; pass
< 2; pass
++) {
1255 struct origin
*(*find
)(struct scoreboard
*,
1256 struct commit
*, struct origin
*);
1257 find
= pass
? find_rename
: find_origin
;
1259 for (i
= 0, sg
= first_scapegoat(revs
, commit
);
1261 sg
= sg
->next
, i
++) {
1262 struct commit
*p
= sg
->item
;
1267 if (parse_commit(p
))
1269 porigin
= find(sb
, p
, origin
);
1272 if (!hashcmp(porigin
->blob_sha1
, origin
->blob_sha1
)) {
1273 pass_whole_blame(sb
, origin
, porigin
);
1274 origin_decref(porigin
);
1277 for (j
= same
= 0; j
< i
; j
++)
1279 !hashcmp(sg_origin
[j
]->blob_sha1
,
1280 porigin
->blob_sha1
)) {
1285 sg_origin
[i
] = porigin
;
1287 origin_decref(porigin
);
1292 for (i
= 0, sg
= first_scapegoat(revs
, commit
);
1294 sg
= sg
->next
, i
++) {
1295 struct origin
*porigin
= sg_origin
[i
];
1298 if (!origin
->previous
) {
1299 origin_incref(porigin
);
1300 origin
->previous
= porigin
;
1302 if (pass_blame_to_parent(sb
, origin
, porigin
))
1307 * Optionally find moves in parents' files.
1309 if (opt
& PICKAXE_BLAME_MOVE
)
1310 for (i
= 0, sg
= first_scapegoat(revs
, commit
);
1312 sg
= sg
->next
, i
++) {
1313 struct origin
*porigin
= sg_origin
[i
];
1316 if (find_move_in_parent(sb
, origin
, porigin
))
1321 * Optionally find copies from parents' files.
1323 if (opt
& PICKAXE_BLAME_COPY
)
1324 for (i
= 0, sg
= first_scapegoat(revs
, commit
);
1326 sg
= sg
->next
, i
++) {
1327 struct origin
*porigin
= sg_origin
[i
];
1328 if (find_copy_in_parent(sb
, origin
, sg
->item
,
1334 for (i
= 0; i
< num_sg
; i
++) {
1336 drop_origin_blob(sg_origin
[i
]);
1337 origin_decref(sg_origin
[i
]);
1340 drop_origin_blob(origin
);
1341 if (sg_buf
!= sg_origin
)
1346 * Information on commits, used for output.
1351 const char *author_mail
;
1352 unsigned long author_time
;
1353 const char *author_tz
;
1355 /* filled only when asked for details */
1356 const char *committer
;
1357 const char *committer_mail
;
1358 unsigned long committer_time
;
1359 const char *committer_tz
;
1361 const char *summary
;
1365 * Parse author/committer line in the commit object buffer
1367 static void get_ac_line(const char *inbuf
, const char *what
,
1368 int bufsz
, char *person
, const char **mail
,
1369 unsigned long *time
, const char **tz
)
1371 int len
, tzlen
, maillen
;
1372 char *tmp
, *endp
, *timepos
;
1374 tmp
= strstr(inbuf
, what
);
1377 tmp
+= strlen(what
);
1378 endp
= strchr(tmp
, '\n');
1386 *mail
= *tz
= "(unknown)";
1390 memcpy(person
, tmp
, len
);
1398 tzlen
= (person
+len
)-(tmp
+1);
1403 *time
= strtoul(tmp
, NULL
, 10);
1411 maillen
= timepos
- tmp
;
1417 * mailmap expansion may make the name longer.
1418 * make room by pushing stuff down.
1420 tmp
= person
+ bufsz
- (tzlen
+ 1);
1421 memmove(tmp
, *tz
, tzlen
);
1425 tmp
= tmp
- (maillen
+ 1);
1426 memmove(tmp
, *mail
, maillen
);
1431 * Now, convert e-mail using mailmap
1433 map_email(&mailmap
, tmp
+ 1, person
, tmp
-person
-1);
1436 static void get_commit_info(struct commit
*commit
,
1437 struct commit_info
*ret
,
1442 static char author_buf
[1024];
1443 static char committer_buf
[1024];
1444 static char summary_buf
[1024];
1447 * We've operated without save_commit_buffer, so
1448 * we now need to populate them for output.
1450 if (!commit
->buffer
) {
1451 enum object_type type
;
1454 read_sha1_file(commit
->object
.sha1
, &type
, &size
);
1455 if (!commit
->buffer
)
1456 die("Cannot read commit %s",
1457 sha1_to_hex(commit
->object
.sha1
));
1459 ret
->author
= author_buf
;
1460 get_ac_line(commit
->buffer
, "\nauthor ",
1461 sizeof(author_buf
), author_buf
, &ret
->author_mail
,
1462 &ret
->author_time
, &ret
->author_tz
);
1467 ret
->committer
= committer_buf
;
1468 get_ac_line(commit
->buffer
, "\ncommitter ",
1469 sizeof(committer_buf
), committer_buf
, &ret
->committer_mail
,
1470 &ret
->committer_time
, &ret
->committer_tz
);
1472 ret
->summary
= summary_buf
;
1473 tmp
= strstr(commit
->buffer
, "\n\n");
1476 sprintf(summary_buf
, "(%s)", sha1_to_hex(commit
->object
.sha1
));
1480 endp
= strchr(tmp
, '\n');
1482 endp
= tmp
+ strlen(tmp
);
1484 if (len
>= sizeof(summary_buf
) || len
== 0)
1486 memcpy(summary_buf
, tmp
, len
);
1487 summary_buf
[len
] = 0;
1491 * To allow LF and other nonportable characters in pathnames,
1492 * they are c-style quoted as needed.
1494 static void write_filename_info(const char *path
)
1496 printf("filename ");
1497 write_name_quoted(path
, stdout
, '\n');
1501 * Porcelain/Incremental format wants to show a lot of details per
1502 * commit. Instead of repeating this every line, emit it only once,
1503 * the first time each commit appears in the output.
1505 static int emit_one_suspect_detail(struct origin
*suspect
)
1507 struct commit_info ci
;
1509 if (suspect
->commit
->object
.flags
& METAINFO_SHOWN
)
1512 suspect
->commit
->object
.flags
|= METAINFO_SHOWN
;
1513 get_commit_info(suspect
->commit
, &ci
, 1);
1514 printf("author %s\n", ci
.author
);
1515 printf("author-mail %s\n", ci
.author_mail
);
1516 printf("author-time %lu\n", ci
.author_time
);
1517 printf("author-tz %s\n", ci
.author_tz
);
1518 printf("committer %s\n", ci
.committer
);
1519 printf("committer-mail %s\n", ci
.committer_mail
);
1520 printf("committer-time %lu\n", ci
.committer_time
);
1521 printf("committer-tz %s\n", ci
.committer_tz
);
1522 printf("summary %s\n", ci
.summary
);
1523 if (suspect
->commit
->object
.flags
& UNINTERESTING
)
1524 printf("boundary\n");
1525 if (suspect
->previous
) {
1526 struct origin
*prev
= suspect
->previous
;
1527 printf("previous %s ", sha1_to_hex(prev
->commit
->object
.sha1
));
1528 write_name_quoted(prev
->path
, stdout
, '\n');
1534 * The blame_entry is found to be guilty for the range. Mark it
1535 * as such, and show it in incremental output.
1537 static void found_guilty_entry(struct blame_entry
*ent
)
1543 struct origin
*suspect
= ent
->suspect
;
1545 printf("%s %d %d %d\n",
1546 sha1_to_hex(suspect
->commit
->object
.sha1
),
1547 ent
->s_lno
+ 1, ent
->lno
+ 1, ent
->num_lines
);
1548 emit_one_suspect_detail(suspect
);
1549 write_filename_info(suspect
->path
);
1550 maybe_flush_or_die(stdout
, "stdout");
1555 * The main loop -- while the scoreboard has lines whose true origin
1556 * is still unknown, pick one blame_entry, and allow its current
1557 * suspect to pass blames to its parents.
1559 static void assign_blame(struct scoreboard
*sb
, int opt
)
1561 struct rev_info
*revs
= sb
->revs
;
1564 struct blame_entry
*ent
;
1565 struct commit
*commit
;
1566 struct origin
*suspect
= NULL
;
1568 /* find one suspect to break down */
1569 for (ent
= sb
->ent
; !suspect
&& ent
; ent
= ent
->next
)
1571 suspect
= ent
->suspect
;
1573 return; /* all done */
1576 * We will use this suspect later in the loop,
1577 * so hold onto it in the meantime.
1579 origin_incref(suspect
);
1580 commit
= suspect
->commit
;
1581 if (!commit
->object
.parsed
)
1582 parse_commit(commit
);
1584 (!(commit
->object
.flags
& UNINTERESTING
) &&
1585 !(revs
->max_age
!= -1 && commit
->date
< revs
->max_age
)))
1586 pass_blame(sb
, suspect
, opt
);
1588 commit
->object
.flags
|= UNINTERESTING
;
1589 if (commit
->object
.parsed
)
1590 mark_parents_uninteresting(commit
);
1592 /* treat root commit as boundary */
1593 if (!commit
->parents
&& !show_root
)
1594 commit
->object
.flags
|= UNINTERESTING
;
1596 /* Take responsibility for the remaining entries */
1597 for (ent
= sb
->ent
; ent
; ent
= ent
->next
)
1598 if (same_suspect(ent
->suspect
, suspect
))
1599 found_guilty_entry(ent
);
1600 origin_decref(suspect
);
1602 if (DEBUG
) /* sanity */
1603 sanity_check_refcnt(sb
);
1607 static const char *format_time(unsigned long time
, const char *tz_str
,
1610 static char time_buf
[128];
1615 if (show_raw_time
) {
1616 sprintf(time_buf
, "%lu %s", time
, tz_str
);
1621 minutes
= tz
< 0 ? -tz
: tz
;
1622 minutes
= (minutes
/ 100)*60 + (minutes
% 100);
1623 minutes
= tz
< 0 ? -minutes
: minutes
;
1624 t
= time
+ minutes
* 60;
1627 strftime(time_buf
, sizeof(time_buf
), "%Y-%m-%d %H:%M:%S ", tm
);
1628 strcat(time_buf
, tz_str
);
1632 #define OUTPUT_ANNOTATE_COMPAT 001
1633 #define OUTPUT_LONG_OBJECT_NAME 002
1634 #define OUTPUT_RAW_TIMESTAMP 004
1635 #define OUTPUT_PORCELAIN 010
1636 #define OUTPUT_SHOW_NAME 020
1637 #define OUTPUT_SHOW_NUMBER 040
1638 #define OUTPUT_SHOW_SCORE 0100
1639 #define OUTPUT_NO_AUTHOR 0200
1641 static void emit_porcelain(struct scoreboard
*sb
, struct blame_entry
*ent
)
1645 struct origin
*suspect
= ent
->suspect
;
1648 strcpy(hex
, sha1_to_hex(suspect
->commit
->object
.sha1
));
1649 printf("%s%c%d %d %d\n",
1651 ent
->guilty
? ' ' : '*', // purely for debugging
1655 if (emit_one_suspect_detail(suspect
) ||
1656 (suspect
->commit
->object
.flags
& MORE_THAN_ONE_PATH
))
1657 write_filename_info(suspect
->path
);
1659 cp
= nth_line(sb
, ent
->lno
);
1660 for (cnt
= 0; cnt
< ent
->num_lines
; cnt
++) {
1663 printf("%s %d %d\n", hex
,
1664 ent
->s_lno
+ 1 + cnt
,
1665 ent
->lno
+ 1 + cnt
);
1670 } while (ch
!= '\n' &&
1671 cp
< sb
->final_buf
+ sb
->final_buf_size
);
1675 static void emit_other(struct scoreboard
*sb
, struct blame_entry
*ent
, int opt
)
1679 struct origin
*suspect
= ent
->suspect
;
1680 struct commit_info ci
;
1682 int show_raw_time
= !!(opt
& OUTPUT_RAW_TIMESTAMP
);
1684 get_commit_info(suspect
->commit
, &ci
, 1);
1685 strcpy(hex
, sha1_to_hex(suspect
->commit
->object
.sha1
));
1687 cp
= nth_line(sb
, ent
->lno
);
1688 for (cnt
= 0; cnt
< ent
->num_lines
; cnt
++) {
1690 int length
= (opt
& OUTPUT_LONG_OBJECT_NAME
) ? 40 : 8;
1692 if (suspect
->commit
->object
.flags
& UNINTERESTING
) {
1694 memset(hex
, ' ', length
);
1695 else if (!(opt
& OUTPUT_ANNOTATE_COMPAT
)) {
1701 printf("%.*s", length
, hex
);
1702 if (opt
& OUTPUT_ANNOTATE_COMPAT
)
1703 printf("\t(%10s\t%10s\t%d)", ci
.author
,
1704 format_time(ci
.author_time
, ci
.author_tz
,
1706 ent
->lno
+ 1 + cnt
);
1708 if (opt
& OUTPUT_SHOW_SCORE
)
1710 max_score_digits
, ent
->score
,
1711 ent
->suspect
->refcnt
);
1712 if (opt
& OUTPUT_SHOW_NAME
)
1713 printf(" %-*.*s", longest_file
, longest_file
,
1715 if (opt
& OUTPUT_SHOW_NUMBER
)
1716 printf(" %*d", max_orig_digits
,
1717 ent
->s_lno
+ 1 + cnt
);
1719 if (!(opt
& OUTPUT_NO_AUTHOR
))
1720 printf(" (%-*.*s %10s",
1721 longest_author
, longest_author
,
1723 format_time(ci
.author_time
,
1727 max_digits
, ent
->lno
+ 1 + cnt
);
1732 } while (ch
!= '\n' &&
1733 cp
< sb
->final_buf
+ sb
->final_buf_size
);
1737 static void output(struct scoreboard
*sb
, int option
)
1739 struct blame_entry
*ent
;
1741 if (option
& OUTPUT_PORCELAIN
) {
1742 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
1743 struct blame_entry
*oth
;
1744 struct origin
*suspect
= ent
->suspect
;
1745 struct commit
*commit
= suspect
->commit
;
1746 if (commit
->object
.flags
& MORE_THAN_ONE_PATH
)
1748 for (oth
= ent
->next
; oth
; oth
= oth
->next
) {
1749 if ((oth
->suspect
->commit
!= commit
) ||
1750 !strcmp(oth
->suspect
->path
, suspect
->path
))
1752 commit
->object
.flags
|= MORE_THAN_ONE_PATH
;
1758 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
1759 if (option
& OUTPUT_PORCELAIN
)
1760 emit_porcelain(sb
, ent
);
1762 emit_other(sb
, ent
, option
);
1768 * To allow quick access to the contents of nth line in the
1769 * final image, prepare an index in the scoreboard.
1771 static int prepare_lines(struct scoreboard
*sb
)
1773 const char *buf
= sb
->final_buf
;
1774 unsigned long len
= sb
->final_buf_size
;
1775 int num
= 0, incomplete
= 0, bol
= 1;
1777 if (len
&& buf
[len
-1] != '\n')
1778 incomplete
++; /* incomplete line at the end */
1781 sb
->lineno
= xrealloc(sb
->lineno
,
1782 sizeof(int* ) * (num
+ 1));
1783 sb
->lineno
[num
] = buf
- sb
->final_buf
;
1786 if (*buf
++ == '\n') {
1791 sb
->lineno
= xrealloc(sb
->lineno
,
1792 sizeof(int* ) * (num
+ incomplete
+ 1));
1793 sb
->lineno
[num
+ incomplete
] = buf
- sb
->final_buf
;
1794 sb
->num_lines
= num
+ incomplete
;
1795 return sb
->num_lines
;
1799 * Add phony grafts for use with -S; this is primarily to
1800 * support git's cvsserver that wants to give a linear history
1803 static int read_ancestry(const char *graft_file
)
1805 FILE *fp
= fopen(graft_file
, "r");
1809 while (fgets(buf
, sizeof(buf
), fp
)) {
1810 /* The format is just "Commit Parent1 Parent2 ...\n" */
1811 int len
= strlen(buf
);
1812 struct commit_graft
*graft
= read_graft_line(buf
, len
);
1814 register_commit_graft(graft
, 0);
1821 * How many columns do we need to show line numbers in decimal?
1823 static int lineno_width(int lines
)
1827 for (width
= 1, i
= 10; i
<= lines
+ 1; width
++)
1833 * How many columns do we need to show line numbers, authors,
1836 static void find_alignment(struct scoreboard
*sb
, int *option
)
1838 int longest_src_lines
= 0;
1839 int longest_dst_lines
= 0;
1840 unsigned largest_score
= 0;
1841 struct blame_entry
*e
;
1843 for (e
= sb
->ent
; e
; e
= e
->next
) {
1844 struct origin
*suspect
= e
->suspect
;
1845 struct commit_info ci
;
1848 if (strcmp(suspect
->path
, sb
->path
))
1849 *option
|= OUTPUT_SHOW_NAME
;
1850 num
= strlen(suspect
->path
);
1851 if (longest_file
< num
)
1853 if (!(suspect
->commit
->object
.flags
& METAINFO_SHOWN
)) {
1854 suspect
->commit
->object
.flags
|= METAINFO_SHOWN
;
1855 get_commit_info(suspect
->commit
, &ci
, 1);
1856 num
= strlen(ci
.author
);
1857 if (longest_author
< num
)
1858 longest_author
= num
;
1860 num
= e
->s_lno
+ e
->num_lines
;
1861 if (longest_src_lines
< num
)
1862 longest_src_lines
= num
;
1863 num
= e
->lno
+ e
->num_lines
;
1864 if (longest_dst_lines
< num
)
1865 longest_dst_lines
= num
;
1866 if (largest_score
< ent_score(sb
, e
))
1867 largest_score
= ent_score(sb
, e
);
1869 max_orig_digits
= lineno_width(longest_src_lines
);
1870 max_digits
= lineno_width(longest_dst_lines
);
1871 max_score_digits
= lineno_width(largest_score
);
1875 * For debugging -- origin is refcounted, and this asserts that
1876 * we do not underflow.
1878 static void sanity_check_refcnt(struct scoreboard
*sb
)
1881 struct blame_entry
*ent
;
1883 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
1884 /* Nobody should have zero or negative refcnt */
1885 if (ent
->suspect
->refcnt
<= 0) {
1886 fprintf(stderr
, "%s in %s has negative refcnt %d\n",
1888 sha1_to_hex(ent
->suspect
->commit
->object
.sha1
),
1889 ent
->suspect
->refcnt
);
1895 find_alignment(sb
, &opt
);
1897 die("Baa %d!", baa
);
1902 * Used for the command line parsing; check if the path exists
1903 * in the working tree.
1905 static int has_string_in_work_tree(const char *path
)
1908 return !lstat(path
, &st
);
1911 static unsigned parse_score(const char *arg
)
1914 unsigned long score
= strtoul(arg
, &end
, 10);
1920 static const char *add_prefix(const char *prefix
, const char *path
)
1922 return prefix_path(prefix
, prefix
? strlen(prefix
) : 0, path
);
1926 * Parsing of (comma separated) one item in the -L option
1928 static const char *parse_loc(const char *spec
,
1929 struct scoreboard
*sb
, long lno
,
1930 long begin
, long *ret
)
1937 regmatch_t match
[1];
1939 /* Allow "-L <something>,+20" to mean starting at <something>
1940 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1943 if (1 < begin
&& (spec
[0] == '+' || spec
[0] == '-')) {
1944 num
= strtol(spec
+ 1, &term
, 10);
1945 if (term
!= spec
+ 1) {
1949 *ret
= begin
+ num
- 2;
1958 num
= strtol(spec
, &term
, 10);
1966 /* it could be a regexp of form /.../ */
1967 for (term
= (char*) spec
+ 1; *term
&& *term
!= '/'; term
++) {
1974 /* try [spec+1 .. term-1] as regexp */
1976 begin
--; /* input is in human terms */
1977 line
= nth_line(sb
, begin
);
1979 if (!(reg_error
= regcomp(®exp
, spec
+ 1, REG_NEWLINE
)) &&
1980 !(reg_error
= regexec(®exp
, line
, 1, match
, 0))) {
1981 const char *cp
= line
+ match
[0].rm_so
;
1984 while (begin
++ < lno
) {
1985 nline
= nth_line(sb
, begin
);
1986 if (line
<= cp
&& cp
< nline
)
1997 regerror(reg_error
, ®exp
, errbuf
, 1024);
1998 die("-L parameter '%s': %s", spec
+ 1, errbuf
);
2003 * Parsing of -L option
2005 static void prepare_blame_range(struct scoreboard
*sb
,
2006 const char *bottomtop
,
2008 long *bottom
, long *top
)
2012 term
= parse_loc(bottomtop
, sb
, lno
, 1, bottom
);
2014 term
= parse_loc(term
+ 1, sb
, lno
, *bottom
+ 1, top
);
2022 static int git_blame_config(const char *var
, const char *value
, void *cb
)
2024 if (!strcmp(var
, "blame.showroot")) {
2025 show_root
= git_config_bool(var
, value
);
2028 if (!strcmp(var
, "blame.blankboundary")) {
2029 blank_boundary
= git_config_bool(var
, value
);
2032 return git_default_config(var
, value
, cb
);
2036 * Prepare a dummy commit that represents the work tree (or staged) item.
2037 * Note that annotating work tree item never works in the reverse.
2039 static struct commit
*fake_working_tree_commit(const char *path
, const char *contents_from
)
2041 struct commit
*commit
;
2042 struct origin
*origin
;
2043 unsigned char head_sha1
[20];
2044 struct strbuf buf
= STRBUF_INIT
;
2048 struct cache_entry
*ce
;
2051 if (get_sha1("HEAD", head_sha1
))
2052 die("No such ref: HEAD");
2055 commit
= xcalloc(1, sizeof(*commit
));
2056 commit
->parents
= xcalloc(1, sizeof(*commit
->parents
));
2057 commit
->parents
->item
= lookup_commit_reference(head_sha1
);
2058 commit
->object
.parsed
= 1;
2060 commit
->object
.type
= OBJ_COMMIT
;
2062 origin
= make_origin(commit
, path
);
2064 if (!contents_from
|| strcmp("-", contents_from
)) {
2066 const char *read_from
;
2067 unsigned long fin_size
;
2069 if (contents_from
) {
2070 if (stat(contents_from
, &st
) < 0)
2071 die("Cannot stat %s", contents_from
);
2072 read_from
= contents_from
;
2075 if (lstat(path
, &st
) < 0)
2076 die("Cannot lstat %s", path
);
2079 fin_size
= xsize_t(st
.st_size
);
2080 mode
= canon_mode(st
.st_mode
);
2081 switch (st
.st_mode
& S_IFMT
) {
2083 if (strbuf_read_file(&buf
, read_from
, st
.st_size
) != st
.st_size
)
2084 die("cannot open or read %s", read_from
);
2087 if (readlink(read_from
, buf
.buf
, buf
.alloc
) != fin_size
)
2088 die("cannot readlink %s", read_from
);
2092 die("unsupported file type %s", read_from
);
2096 /* Reading from stdin */
2097 contents_from
= "standard input";
2099 if (strbuf_read(&buf
, 0, 0) < 0)
2100 die("read error %s from stdin", strerror(errno
));
2102 convert_to_git(path
, buf
.buf
, buf
.len
, &buf
, 0);
2103 origin
->file
.ptr
= buf
.buf
;
2104 origin
->file
.size
= buf
.len
;
2105 pretend_sha1_file(buf
.buf
, buf
.len
, OBJ_BLOB
, origin
->blob_sha1
);
2106 commit
->util
= origin
;
2109 * Read the current index, replace the path entry with
2110 * origin->blob_sha1 without mucking with its mode or type
2111 * bits; we are not going to write this index out -- we just
2112 * want to run "diff-index --cached".
2119 int pos
= cache_name_pos(path
, len
);
2121 mode
= active_cache
[pos
]->ce_mode
;
2123 /* Let's not bother reading from HEAD tree */
2124 mode
= S_IFREG
| 0644;
2126 size
= cache_entry_size(len
);
2127 ce
= xcalloc(1, size
);
2128 hashcpy(ce
->sha1
, origin
->blob_sha1
);
2129 memcpy(ce
->name
, path
, len
);
2130 ce
->ce_flags
= create_ce_flags(len
, 0);
2131 ce
->ce_mode
= create_ce_mode(mode
);
2132 add_cache_entry(ce
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
2135 * We are not going to write this out, so this does not matter
2136 * right now, but someday we might optimize diff-index --cached
2137 * with cache-tree information.
2139 cache_tree_invalidate_path(active_cache_tree
, path
);
2141 commit
->buffer
= xmalloc(400);
2142 ident
= fmt_ident("Not Committed Yet", "not.committed.yet", NULL
, 0);
2143 snprintf(commit
->buffer
, 400,
2144 "tree 0000000000000000000000000000000000000000\n"
2148 "Version of %s from %s\n",
2149 sha1_to_hex(head_sha1
),
2150 ident
, ident
, path
, contents_from
? contents_from
: path
);
2154 static const char *prepare_final(struct scoreboard
*sb
)
2157 const char *final_commit_name
= NULL
;
2158 struct rev_info
*revs
= sb
->revs
;
2161 * There must be one and only one positive commit in the
2162 * revs->pending array.
2164 for (i
= 0; i
< revs
->pending
.nr
; i
++) {
2165 struct object
*obj
= revs
->pending
.objects
[i
].item
;
2166 if (obj
->flags
& UNINTERESTING
)
2168 while (obj
->type
== OBJ_TAG
)
2169 obj
= deref_tag(obj
, NULL
, 0);
2170 if (obj
->type
!= OBJ_COMMIT
)
2171 die("Non commit %s?", revs
->pending
.objects
[i
].name
);
2173 die("More than one commit to dig from %s and %s?",
2174 revs
->pending
.objects
[i
].name
,
2176 sb
->final
= (struct commit
*) obj
;
2177 final_commit_name
= revs
->pending
.objects
[i
].name
;
2179 return final_commit_name
;
2182 static const char *prepare_initial(struct scoreboard
*sb
)
2185 const char *final_commit_name
= NULL
;
2186 struct rev_info
*revs
= sb
->revs
;
2189 * There must be one and only one negative commit, and it must be
2192 for (i
= 0; i
< revs
->pending
.nr
; i
++) {
2193 struct object
*obj
= revs
->pending
.objects
[i
].item
;
2194 if (!(obj
->flags
& UNINTERESTING
))
2196 while (obj
->type
== OBJ_TAG
)
2197 obj
= deref_tag(obj
, NULL
, 0);
2198 if (obj
->type
!= OBJ_COMMIT
)
2199 die("Non commit %s?", revs
->pending
.objects
[i
].name
);
2201 die("More than one commit to dig down to %s and %s?",
2202 revs
->pending
.objects
[i
].name
,
2204 sb
->final
= (struct commit
*) obj
;
2205 final_commit_name
= revs
->pending
.objects
[i
].name
;
2207 if (!final_commit_name
)
2208 die("No commit to dig down to?");
2209 return final_commit_name
;
2212 static int blame_copy_callback(const struct option
*option
, const char *arg
, int unset
)
2214 int *opt
= option
->value
;
2217 * -C enables copy from removed files;
2218 * -C -C enables copy from existing files, but only
2219 * when blaming a new file;
2220 * -C -C -C enables copy from existing files for
2223 if (*opt
& PICKAXE_BLAME_COPY_HARDER
)
2224 *opt
|= PICKAXE_BLAME_COPY_HARDEST
;
2225 if (*opt
& PICKAXE_BLAME_COPY
)
2226 *opt
|= PICKAXE_BLAME_COPY_HARDER
;
2227 *opt
|= PICKAXE_BLAME_COPY
| PICKAXE_BLAME_MOVE
;
2230 blame_copy_score
= parse_score(arg
);
2234 static int blame_move_callback(const struct option
*option
, const char *arg
, int unset
)
2236 int *opt
= option
->value
;
2238 *opt
|= PICKAXE_BLAME_MOVE
;
2241 blame_move_score
= parse_score(arg
);
2245 static int blame_bottomtop_callback(const struct option
*option
, const char *arg
, int unset
)
2247 const char **bottomtop
= option
->value
;
2251 die("More than one '-L n,m' option given");
2256 int cmd_blame(int argc
, const char **argv
, const char *prefix
)
2258 struct rev_info revs
;
2260 struct scoreboard sb
;
2262 struct blame_entry
*ent
;
2263 long dashdash_pos
, bottom
, top
, lno
;
2264 const char *final_commit_name
= NULL
;
2265 enum object_type type
;
2267 static const char *bottomtop
= NULL
;
2268 static int output_option
= 0, opt
= 0;
2269 static int show_stats
= 0;
2270 static const char *revs_file
= NULL
;
2271 static const char *contents_from
= NULL
;
2272 static const struct option options
[] = {
2273 OPT_BOOLEAN(0, "incremental", &incremental
, "Show blame entries as we find them, incrementally"),
2274 OPT_BOOLEAN('b', NULL
, &blank_boundary
, "Show blank SHA-1 for boundary commits (Default: off)"),
2275 OPT_BOOLEAN(0, "root", &show_root
, "Do not treat root commits as boundaries (Default: off)"),
2276 OPT_BOOLEAN(0, "show-stats", &show_stats
, "Show work cost statistics"),
2277 OPT_BIT(0, "score-debug", &output_option
, "Show output score for blame entries", OUTPUT_SHOW_SCORE
),
2278 OPT_BIT('f', "show-name", &output_option
, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME
),
2279 OPT_BIT('n', "show-number", &output_option
, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER
),
2280 OPT_BIT('p', "porcelain", &output_option
, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN
),
2281 OPT_BIT('c', NULL
, &output_option
, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT
),
2282 OPT_BIT('t', NULL
, &output_option
, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP
),
2283 OPT_BIT('l', NULL
, &output_option
, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME
),
2284 OPT_BIT('s', NULL
, &output_option
, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR
),
2285 OPT_BIT('w', NULL
, &xdl_opts
, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE
),
2286 OPT_STRING('S', NULL
, &revs_file
, "file", "Use revisions from <file> instead of calling git-rev-list"),
2287 OPT_STRING(0, "contents", &contents_from
, "file", "Use <file>'s contents as the final image"),
2288 { OPTION_CALLBACK
, 'C', NULL
, &opt
, "score", "Find line copies within and across files", PARSE_OPT_OPTARG
, blame_copy_callback
},
2289 { OPTION_CALLBACK
, 'M', NULL
, &opt
, "score", "Find line movements within and across files", PARSE_OPT_OPTARG
, blame_move_callback
},
2290 OPT_CALLBACK('L', NULL
, &bottomtop
, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback
),
2294 struct parse_opt_ctx_t ctx
;
2295 int cmd_is_annotate
= !strcmp(argv
[0], "annotate");
2297 git_config(git_blame_config
, NULL
);
2298 init_revisions(&revs
, NULL
);
2299 save_commit_buffer
= 0;
2302 parse_options_start(&ctx
, argc
, argv
, PARSE_OPT_KEEP_DASHDASH
|
2303 PARSE_OPT_KEEP_ARGV0
);
2305 switch (parse_options_step(&ctx
, options
, blame_opt_usage
)) {
2306 case PARSE_OPT_HELP
:
2308 case PARSE_OPT_DONE
:
2310 dashdash_pos
= ctx
.cpidx
;
2314 if (!strcmp(ctx
.argv
[0], "--reverse")) {
2315 ctx
.argv
[0] = "--children";
2318 parse_revision_opt(&revs
, &ctx
, options
, blame_opt_usage
);
2321 argc
= parse_options_end(&ctx
);
2323 if (cmd_is_annotate
)
2324 output_option
|= OUTPUT_ANNOTATE_COMPAT
;
2326 if (DIFF_OPT_TST(&revs
.diffopt
, FIND_COPIES_HARDER
))
2327 opt
|= (PICKAXE_BLAME_COPY
| PICKAXE_BLAME_MOVE
|
2328 PICKAXE_BLAME_COPY_HARDER
);
2330 if (!blame_move_score
)
2331 blame_move_score
= BLAME_DEFAULT_MOVE_SCORE
;
2332 if (!blame_copy_score
)
2333 blame_copy_score
= BLAME_DEFAULT_COPY_SCORE
;
2336 * We have collected options unknown to us in argv[1..unk]
2337 * which are to be passed to revision machinery if we are
2338 * going to do the "bottom" processing.
2340 * The remaining are:
2342 * (1) if dashdash_pos != 0, its either
2343 * "blame [revisions] -- <path>" or
2344 * "blame -- <path> <rev>"
2346 * (2) otherwise, its one of the two:
2347 * "blame [revisions] <path>"
2348 * "blame <path> <rev>"
2350 * Note that we must strip out <path> from the arguments: we do not
2351 * want the path pruning but we may want "bottom" processing.
2354 switch (argc
- dashdash_pos
- 1) {
2357 usage_with_options(blame_opt_usage
, options
);
2358 /* reorder for the new way: <rev> -- <path> */
2364 path
= add_prefix(prefix
, argv
[--argc
]);
2368 usage_with_options(blame_opt_usage
, options
);
2372 usage_with_options(blame_opt_usage
, options
);
2373 path
= add_prefix(prefix
, argv
[argc
- 1]);
2374 if (argc
== 3 && !has_string_in_work_tree(path
)) { /* (2b) */
2375 path
= add_prefix(prefix
, argv
[1]);
2378 argv
[argc
- 1] = "--";
2381 if (!has_string_in_work_tree(path
))
2382 die("cannot stat path %s: %s", path
, strerror(errno
));
2385 setup_revisions(argc
, argv
, &revs
, NULL
);
2386 memset(&sb
, 0, sizeof(sb
));
2390 final_commit_name
= prepare_final(&sb
);
2391 else if (contents_from
)
2392 die("--contents and --children do not blend well.");
2394 final_commit_name
= prepare_initial(&sb
);
2398 * "--not A B -- path" without anything positive;
2399 * do not default to HEAD, but use the working tree
2403 sb
.final
= fake_working_tree_commit(path
, contents_from
);
2404 add_pending_object(&revs
, &(sb
.final
->object
), ":");
2406 else if (contents_from
)
2407 die("Cannot use --contents with final commit object name");
2410 * If we have bottom, this will mark the ancestors of the
2411 * bottom commits we would reach while traversing as
2414 if (prepare_revision_walk(&revs
))
2415 die("revision walk setup failed");
2417 if (is_null_sha1(sb
.final
->object
.sha1
)) {
2420 buf
= xmalloc(o
->file
.size
+ 1);
2421 memcpy(buf
, o
->file
.ptr
, o
->file
.size
+ 1);
2423 sb
.final_buf_size
= o
->file
.size
;
2426 o
= get_origin(&sb
, sb
.final
, path
);
2427 if (fill_blob_sha1(o
))
2428 die("no such path %s in %s", path
, final_commit_name
);
2430 sb
.final_buf
= read_sha1_file(o
->blob_sha1
, &type
,
2431 &sb
.final_buf_size
);
2433 die("Cannot read blob %s for path %s",
2434 sha1_to_hex(o
->blob_sha1
),
2438 lno
= prepare_lines(&sb
);
2442 prepare_blame_range(&sb
, bottomtop
, lno
, &bottom
, &top
);
2443 if (bottom
&& top
&& top
< bottom
) {
2445 tmp
= top
; top
= bottom
; bottom
= tmp
;
2453 die("file %s has only %lu lines", path
, lno
);
2455 ent
= xcalloc(1, sizeof(*ent
));
2457 ent
->num_lines
= top
- bottom
;
2459 ent
->s_lno
= bottom
;
2464 if (revs_file
&& read_ancestry(revs_file
))
2465 die("reading graft file %s failed: %s",
2466 revs_file
, strerror(errno
));
2468 read_mailmap(&mailmap
, ".mailmap", NULL
);
2473 assign_blame(&sb
, opt
);
2480 if (!(output_option
& OUTPUT_PORCELAIN
))
2481 find_alignment(&sb
, &output_option
);
2483 output(&sb
, output_option
);
2484 free((void *)sb
.final_buf
);
2485 for (ent
= sb
.ent
; ent
; ) {
2486 struct blame_entry
*e
= ent
->next
;
2492 printf("num read blob: %d\n", num_read_blob
);
2493 printf("num get patch: %d\n", num_get_patch
);
2494 printf("num commits: %d\n", num_commits
);