1 #include "git-compat-util.h"
3 #include "object-store-ll.h"
4 #include "cache-tree.h"
12 #include "read-cache.h"
18 #include "commit-slab.h"
20 #include "commit-graph.h"
22 define_commit_slab(blame_suspects
, struct blame_origin
*);
23 static struct blame_suspects blame_suspects
;
25 struct blame_origin
*get_blame_suspects(struct commit
*commit
)
27 struct blame_origin
**result
;
29 result
= blame_suspects_peek(&blame_suspects
, commit
);
31 return result
? *result
: NULL
;
34 static void set_blame_suspects(struct commit
*commit
, struct blame_origin
*origin
)
36 *blame_suspects_at(&blame_suspects
, commit
) = origin
;
39 void blame_origin_decref(struct blame_origin
*o
)
41 if (o
&& --o
->refcnt
<= 0) {
42 struct blame_origin
*p
, *l
= NULL
;
44 blame_origin_decref(o
->previous
);
46 /* Should be present exactly once in commit chain */
47 for (p
= get_blame_suspects(o
->commit
); p
; l
= p
, p
= p
->next
) {
52 set_blame_suspects(o
->commit
, p
->next
);
57 die("internal error in blame_origin_decref");
62 * Given a commit and a path in it, create a new origin structure.
63 * The callers that add blame to the scoreboard should use
64 * get_origin() to obtain shared, refcounted copy instead of calling
65 * this function directly.
67 static struct blame_origin
*make_origin(struct commit
*commit
, const char *path
)
69 struct blame_origin
*o
;
70 FLEX_ALLOC_STR(o
, path
, path
);
73 o
->next
= get_blame_suspects(commit
);
74 set_blame_suspects(commit
, o
);
79 * Locate an existing origin or create a new one.
80 * This moves the origin to front position in the commit util list.
82 static struct blame_origin
*get_origin(struct commit
*commit
, const char *path
)
84 struct blame_origin
*o
, *l
;
86 for (o
= get_blame_suspects(commit
), l
= NULL
; o
; l
= o
, o
= o
->next
) {
87 if (!strcmp(o
->path
, path
)) {
91 o
->next
= get_blame_suspects(commit
);
92 set_blame_suspects(commit
, o
);
94 return blame_origin_incref(o
);
97 return make_origin(commit
, path
);
102 static void verify_working_tree_path(struct repository
*r
,
103 struct commit
*work_tree
, const char *path
)
105 struct commit_list
*parents
;
108 for (parents
= work_tree
->parents
; parents
; parents
= parents
->next
) {
109 const struct object_id
*commit_oid
= &parents
->item
->object
.oid
;
110 struct object_id blob_oid
;
113 if (!get_tree_entry(r
, commit_oid
, path
, &blob_oid
, &mode
) &&
114 oid_object_info(r
, &blob_oid
, NULL
) == OBJ_BLOB
)
118 pos
= index_name_pos(r
->index
, path
, strlen(path
));
120 ; /* path is in the index */
121 else if (-1 - pos
< r
->index
->cache_nr
&&
122 !strcmp(r
->index
->cache
[-1 - pos
]->name
, path
))
123 ; /* path is in the index, unmerged */
125 die("no such path '%s' in HEAD", path
);
128 static struct commit_list
**append_parent(struct repository
*r
,
129 struct commit_list
**tail
,
130 const struct object_id
*oid
)
132 struct commit
*parent
;
134 parent
= lookup_commit_reference(r
, oid
);
136 die("no such commit %s", oid_to_hex(oid
));
137 return &commit_list_insert(parent
, tail
)->next
;
140 static void append_merge_parents(struct repository
*r
,
141 struct commit_list
**tail
)
144 struct strbuf line
= STRBUF_INIT
;
146 merge_head
= open(git_path_merge_head(r
), O_RDONLY
);
147 if (merge_head
< 0) {
150 die("cannot open '%s' for reading",
151 git_path_merge_head(r
));
154 while (!strbuf_getwholeline_fd(&line
, merge_head
, '\n')) {
155 struct object_id oid
;
156 if (get_oid_hex(line
.buf
, &oid
))
157 die("unknown line in '%s': %s",
158 git_path_merge_head(r
), line
.buf
);
159 tail
= append_parent(r
, tail
, &oid
);
162 strbuf_release(&line
);
166 * This isn't as simple as passing sb->buf and sb->len, because we
167 * want to transfer ownership of the buffer to the commit (so we
170 static void set_commit_buffer_from_strbuf(struct repository
*r
,
175 void *buf
= strbuf_detach(sb
, &len
);
176 set_commit_buffer(r
, c
, buf
, len
);
180 * Prepare a dummy commit that represents the work tree (or staged) item.
181 * Note that annotating work tree item never works in the reverse.
183 static struct commit
*fake_working_tree_commit(struct repository
*r
,
184 struct diff_options
*opt
,
186 const char *contents_from
,
187 struct object_id
*oid
)
189 struct commit
*commit
;
190 struct blame_origin
*origin
;
191 struct commit_list
**parent_tail
, *parent
;
192 struct strbuf buf
= STRBUF_INIT
;
196 struct cache_entry
*ce
;
198 struct strbuf msg
= STRBUF_INIT
;
202 commit
= alloc_commit_node(r
);
203 commit
->object
.parsed
= 1;
205 parent_tail
= &commit
->parents
;
207 parent_tail
= append_parent(r
, parent_tail
, oid
);
208 append_merge_parents(r
, parent_tail
);
209 verify_working_tree_path(r
, commit
, path
);
211 origin
= make_origin(commit
, path
);
214 ident
= fmt_ident("External file (--contents)", "external.file",
215 WANT_BLANK_IDENT
, NULL
, 0);
217 ident
= fmt_ident("Not Committed Yet", "not.committed.yet",
218 WANT_BLANK_IDENT
, NULL
, 0);
219 strbuf_addstr(&msg
, "tree 0000000000000000000000000000000000000000\n");
220 for (parent
= commit
->parents
; parent
; parent
= parent
->next
)
221 strbuf_addf(&msg
, "parent %s\n",
222 oid_to_hex(&parent
->item
->object
.oid
));
226 "Version of %s from %s\n",
228 (!contents_from
? path
:
229 (!strcmp(contents_from
, "-") ? "standard input" : contents_from
)));
230 set_commit_buffer_from_strbuf(r
, commit
, &msg
);
232 if (!contents_from
|| strcmp("-", contents_from
)) {
234 const char *read_from
;
236 unsigned long buf_len
;
239 if (stat(contents_from
, &st
) < 0)
240 die_errno("Cannot stat '%s'", contents_from
);
241 read_from
= contents_from
;
244 if (lstat(path
, &st
) < 0)
245 die_errno("Cannot lstat '%s'", path
);
248 mode
= canon_mode(st
.st_mode
);
250 switch (st
.st_mode
& S_IFMT
) {
252 if (opt
->flags
.allow_textconv
&&
253 textconv_object(r
, read_from
, mode
, null_oid(), 0, &buf_ptr
, &buf_len
))
254 strbuf_attach(&buf
, buf_ptr
, buf_len
, buf_len
+ 1);
255 else if (strbuf_read_file(&buf
, read_from
, st
.st_size
) != st
.st_size
)
256 die_errno("cannot open or read '%s'", read_from
);
259 if (strbuf_readlink(&buf
, read_from
, st
.st_size
) < 0)
260 die_errno("cannot readlink '%s'", read_from
);
263 die("unsupported file type %s", read_from
);
267 /* Reading from stdin */
269 if (strbuf_read(&buf
, 0, 0) < 0)
270 die_errno("failed to read from stdin");
272 convert_to_git(r
->index
, path
, buf
.buf
, buf
.len
, &buf
, 0);
273 origin
->file
.ptr
= buf
.buf
;
274 origin
->file
.size
= buf
.len
;
275 pretend_object_file(buf
.buf
, buf
.len
, OBJ_BLOB
, &origin
->blob_oid
);
278 * Read the current index, replace the path entry with
279 * origin->blob_sha1 without mucking with its mode or type
280 * bits; we are not going to write this index out -- we just
281 * want to run "diff-index --cached".
283 discard_index(r
->index
);
288 int pos
= index_name_pos(r
->index
, path
, len
);
290 mode
= r
->index
->cache
[pos
]->ce_mode
;
292 /* Let's not bother reading from HEAD tree */
293 mode
= S_IFREG
| 0644;
295 ce
= make_empty_cache_entry(r
->index
, len
);
296 oidcpy(&ce
->oid
, &origin
->blob_oid
);
297 memcpy(ce
->name
, path
, len
);
298 ce
->ce_flags
= create_ce_flags(0);
299 ce
->ce_namelen
= len
;
300 ce
->ce_mode
= create_ce_mode(mode
);
301 add_index_entry(r
->index
, ce
,
302 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
304 cache_tree_invalidate_path(r
->index
, path
);
311 static int diff_hunks(mmfile_t
*file_a
, mmfile_t
*file_b
,
312 xdl_emit_hunk_consume_func_t hunk_func
, void *cb_data
, int xdl_opts
)
315 xdemitconf_t xecfg
= {0};
316 xdemitcb_t ecb
= {NULL
};
318 xpp
.flags
= xdl_opts
;
319 xecfg
.hunk_func
= hunk_func
;
321 return xdi_diff(file_a
, file_b
, &xpp
, &xecfg
, &ecb
);
324 static const char *get_next_line(const char *start
, const char *end
)
326 const char *nl
= memchr(start
, '\n', end
- start
);
328 return nl
? nl
+ 1 : end
;
331 static int find_line_starts(int **line_starts
, const char *buf
,
334 const char *end
= buf
+ len
;
339 for (p
= buf
; p
< end
; p
= get_next_line(p
, end
))
342 ALLOC_ARRAY(*line_starts
, num
+ 1);
343 lineno
= *line_starts
;
345 for (p
= buf
; p
< end
; p
= get_next_line(p
, end
))
353 struct fingerprint_entry
;
355 /* A fingerprint is intended to loosely represent a string, such that two
356 * fingerprints can be quickly compared to give an indication of the similarity
357 * of the strings that they represent.
359 * A fingerprint is represented as a multiset of the lower-cased byte pairs in
360 * the string that it represents. Whitespace is added at each end of the
361 * string. Whitespace pairs are ignored. Whitespace is converted to '\0'.
362 * For example, the string "Darth Radar" will be converted to the following
364 * {"\0d", "da", "da", "ar", "ar", "rt", "th", "h\0", "\0r", "ra", "ad", "r\0"}
366 * The similarity between two fingerprints is the size of the intersection of
367 * their multisets, including repeated elements. See fingerprint_similarity for
370 * For ease of implementation, the fingerprint is implemented as a map
371 * of byte pairs to the count of that byte pair in the string, instead of
372 * allowing repeated elements in a set.
376 /* As we know the maximum number of entries in advance, it's
377 * convenient to store the entries in a single array instead of having
378 * the hashmap manage the memory.
380 struct fingerprint_entry
*entries
;
383 /* A byte pair in a fingerprint. Stores the number of times the byte pair
384 * occurs in the string that the fingerprint represents.
386 struct fingerprint_entry
{
387 /* The hashmap entry - the hash represents the byte pair in its
388 * entirety so we don't need to store the byte pair separately.
390 struct hashmap_entry entry
;
391 /* The number of times the byte pair occurs in the string that the
392 * fingerprint represents.
397 /* See `struct fingerprint` for an explanation of what a fingerprint is.
398 * \param result the fingerprint of the string is stored here. This must be
399 * freed later using free_fingerprint.
400 * \param line_begin the start of the string
401 * \param line_end the end of the string
403 static void get_fingerprint(struct fingerprint
*result
,
404 const char *line_begin
,
405 const char *line_end
)
407 unsigned int hash
, c0
= 0, c1
;
409 int max_map_entry_count
= 1 + line_end
- line_begin
;
410 struct fingerprint_entry
*entry
= xcalloc(max_map_entry_count
,
411 sizeof(struct fingerprint_entry
));
412 struct fingerprint_entry
*found_entry
;
414 hashmap_init(&result
->map
, NULL
, NULL
, max_map_entry_count
);
415 result
->entries
= entry
;
416 for (p
= line_begin
; p
<= line_end
; ++p
, c0
= c1
) {
417 /* Always terminate the string with whitespace.
418 * Normalise whitespace to 0, and normalise letters to
419 * lower case. This won't work for multibyte characters but at
420 * worst will match some unrelated characters.
422 if ((p
== line_end
) || isspace(*p
))
426 hash
= c0
| (c1
<< 8);
427 /* Ignore whitespace pairs */
430 hashmap_entry_init(&entry
->entry
, hash
);
432 found_entry
= hashmap_get_entry(&result
->map
, entry
,
433 /* member name */ entry
, NULL
);
435 found_entry
->count
+= 1;
438 hashmap_add(&result
->map
, &entry
->entry
);
444 static void free_fingerprint(struct fingerprint
*f
)
446 hashmap_clear(&f
->map
);
450 /* Calculates the similarity between two fingerprints as the size of the
451 * intersection of their multisets, including repeated elements. See
452 * `struct fingerprint` for an explanation of the fingerprint representation.
453 * The similarity between "cat mat" and "father rather" is 2 because "at" is
454 * present twice in both strings while the similarity between "tim" and "mit"
457 static int fingerprint_similarity(struct fingerprint
*a
, struct fingerprint
*b
)
459 int intersection
= 0;
460 struct hashmap_iter iter
;
461 const struct fingerprint_entry
*entry_a
, *entry_b
;
463 hashmap_for_each_entry(&b
->map
, &iter
, entry_b
,
464 entry
/* member name */) {
465 entry_a
= hashmap_get_entry(&a
->map
, entry_b
, entry
, NULL
);
467 intersection
+= entry_a
->count
< entry_b
->count
?
468 entry_a
->count
: entry_b
->count
;
474 /* Subtracts byte-pair elements in B from A, modifying A in place.
476 static void fingerprint_subtract(struct fingerprint
*a
, struct fingerprint
*b
)
478 struct hashmap_iter iter
;
479 struct fingerprint_entry
*entry_a
;
480 const struct fingerprint_entry
*entry_b
;
482 hashmap_iter_init(&b
->map
, &iter
);
484 hashmap_for_each_entry(&b
->map
, &iter
, entry_b
,
485 entry
/* member name */) {
486 entry_a
= hashmap_get_entry(&a
->map
, entry_b
, entry
, NULL
);
488 if (entry_a
->count
<= entry_b
->count
)
489 hashmap_remove(&a
->map
, &entry_b
->entry
, NULL
);
491 entry_a
->count
-= entry_b
->count
;
496 /* Calculate fingerprints for a series of lines.
497 * Puts the fingerprints in the fingerprints array, which must have been
498 * preallocated to allow storing line_count elements.
500 static void get_line_fingerprints(struct fingerprint
*fingerprints
,
501 const char *content
, const int *line_starts
,
502 long first_line
, long line_count
)
505 const char *linestart
, *lineend
;
507 line_starts
+= first_line
;
508 for (i
= 0; i
< line_count
; ++i
) {
509 linestart
= content
+ line_starts
[i
];
510 lineend
= content
+ line_starts
[i
+ 1];
511 get_fingerprint(fingerprints
+ i
, linestart
, lineend
);
515 static void free_line_fingerprints(struct fingerprint
*fingerprints
,
520 for (i
= 0; i
< nr_fingerprints
; i
++)
521 free_fingerprint(&fingerprints
[i
]);
524 /* This contains the data necessary to linearly map a line number in one half
525 * of a diff chunk to the line in the other half of the diff chunk that is
526 * closest in terms of its position as a fraction of the length of the chunk.
528 struct line_number_mapping
{
529 int destination_start
, destination_length
,
530 source_start
, source_length
;
533 /* Given a line number in one range, offset and scale it to map it onto the
535 * Essentially this mapping is a simple linear equation but the calculation is
536 * more complicated to allow performing it with integer operations.
537 * Another complication is that if a line could map onto many lines in the
538 * destination range then we want to choose the line at the center of those
540 * Example: if the chunk is 2 lines long in A and 10 lines long in B then the
541 * first 5 lines in B will map onto the first line in the A chunk, while the
542 * last 5 lines will all map onto the second line in the A chunk.
543 * Example: if the chunk is 10 lines long in A and 2 lines long in B then line
544 * 0 in B will map onto line 2 in A, and line 1 in B will map onto line 7 in A.
546 static int map_line_number(int line_number
,
547 const struct line_number_mapping
*mapping
)
549 return ((line_number
- mapping
->source_start
) * 2 + 1) *
550 mapping
->destination_length
/
551 (mapping
->source_length
* 2) +
552 mapping
->destination_start
;
555 /* Get a pointer to the element storing the similarity between a line in A
558 * The similarities are stored in a 2-dimensional array. Each "row" in the
559 * array contains the similarities for a line in B. The similarities stored in
560 * a row are the similarities between the line in B and the nearby lines in A.
561 * To keep the length of each row the same, it is padded out with values of -1
562 * where the search range extends beyond the lines in A.
563 * For example, if max_search_distance_a is 2 and the two sides of a diff chunk
570 * Then the similarity array will contain:
571 * [-1, -1, am, bm, cm,
572 * -1, an, bn, cn, dn,
573 * ao, bo, co, do, eo,
574 * bp, cp, dp, ep, -1,
575 * cq, dq, eq, -1, -1]
576 * Where similarities are denoted either by -1 for invalid, or the
577 * concatenation of the two lines in the diff being compared.
579 * \param similarities array of similarities between lines in A and B
580 * \param line_a the index of the line in A, in the same frame of reference as
582 * \param local_line_b the index of the line in B, relative to the first line
583 * in B that similarities represents.
584 * \param closest_line_a the index of the line in A that is deemed to be
585 * closest to local_line_b. This must be in the same
586 * frame of reference as line_a. This value defines
587 * where similarities is centered for the line in B.
588 * \param max_search_distance_a maximum distance in lines from the closest line
589 * in A for other lines in A for which
590 * similarities may be calculated.
592 static int *get_similarity(int *similarities
,
593 int line_a
, int local_line_b
,
594 int closest_line_a
, int max_search_distance_a
)
596 assert(abs(line_a
- closest_line_a
) <=
597 max_search_distance_a
);
598 return similarities
+ line_a
- closest_line_a
+
599 max_search_distance_a
+
600 local_line_b
* (max_search_distance_a
* 2 + 1);
603 #define CERTAIN_NOTHING_MATCHES -2
604 #define CERTAINTY_NOT_CALCULATED -1
606 /* Given a line in B, first calculate its similarities with nearby lines in A
607 * if not already calculated, then identify the most similar and second most
608 * similar lines. The "certainty" is calculated based on those two
611 * \param start_a the index of the first line of the chunk in A
612 * \param length_a the length in lines of the chunk in A
613 * \param local_line_b the index of the line in B, relative to the first line
615 * \param fingerprints_a array of fingerprints for the chunk in A
616 * \param fingerprints_b array of fingerprints for the chunk in B
617 * \param similarities 2-dimensional array of similarities between lines in A
618 * and B. See get_similarity() for more details.
619 * \param certainties array of values indicating how strongly a line in B is
620 * matched with some line in A.
621 * \param second_best_result array of absolute indices in A for the second
622 * closest match of a line in B.
623 * \param result array of absolute indices in A for the closest match of a line
625 * \param max_search_distance_a maximum distance in lines from the closest line
626 * in A for other lines in A for which
627 * similarities may be calculated.
628 * \param map_line_number_in_b_to_a parameter to map_line_number().
630 static void find_best_line_matches(
635 struct fingerprint
*fingerprints_a
,
636 struct fingerprint
*fingerprints_b
,
639 int *second_best_result
,
641 const int max_search_distance_a
,
642 const struct line_number_mapping
*map_line_number_in_b_to_a
)
645 int i
, search_start
, search_end
, closest_local_line_a
, *similarity
,
646 best_similarity
= 0, second_best_similarity
= 0,
647 best_similarity_index
= 0, second_best_similarity_index
= 0;
649 /* certainty has already been calculated so no need to redo the work */
650 if (certainties
[local_line_b
] != CERTAINTY_NOT_CALCULATED
)
653 closest_local_line_a
= map_line_number(
654 local_line_b
+ start_b
, map_line_number_in_b_to_a
) - start_a
;
656 search_start
= closest_local_line_a
- max_search_distance_a
;
657 if (search_start
< 0)
660 search_end
= closest_local_line_a
+ max_search_distance_a
+ 1;
661 if (search_end
> length_a
)
662 search_end
= length_a
;
664 for (i
= search_start
; i
< search_end
; ++i
) {
665 similarity
= get_similarity(similarities
,
667 closest_local_line_a
,
668 max_search_distance_a
);
669 if (*similarity
== -1) {
670 /* This value will never exceed 10 but assert just in
673 assert(abs(i
- closest_local_line_a
) < 1000);
674 /* scale the similarity by (1000 - distance from
675 * closest line) to act as a tie break between lines
676 * that otherwise are equally similar.
678 *similarity
= fingerprint_similarity(
679 fingerprints_b
+ local_line_b
,
680 fingerprints_a
+ i
) *
681 (1000 - abs(i
- closest_local_line_a
));
683 if (*similarity
> best_similarity
) {
684 second_best_similarity
= best_similarity
;
685 second_best_similarity_index
= best_similarity_index
;
686 best_similarity
= *similarity
;
687 best_similarity_index
= i
;
688 } else if (*similarity
> second_best_similarity
) {
689 second_best_similarity
= *similarity
;
690 second_best_similarity_index
= i
;
694 if (best_similarity
== 0) {
695 /* this line definitely doesn't match with anything. Mark it
696 * with this special value so it doesn't get invalidated and
697 * won't be recalculated.
699 certainties
[local_line_b
] = CERTAIN_NOTHING_MATCHES
;
700 result
[local_line_b
] = -1;
702 /* Calculate the certainty with which this line matches.
703 * If the line matches well with two lines then that reduces
704 * the certainty. However we still want to prioritise matching
705 * a line that matches very well with two lines over matching a
706 * line that matches poorly with one line, hence doubling
708 * This means that if we have
709 * line X that matches only one line with a score of 3,
710 * line Y that matches two lines equally with a score of 5,
711 * and line Z that matches only one line with a score or 2,
712 * then the lines in order of certainty are X, Y, Z.
714 certainties
[local_line_b
] = best_similarity
* 2 -
715 second_best_similarity
;
717 /* We keep both the best and second best results to allow us to
718 * check at a later stage of the matching process whether the
719 * result needs to be invalidated.
721 result
[local_line_b
] = start_a
+ best_similarity_index
;
722 second_best_result
[local_line_b
] =
723 start_a
+ second_best_similarity_index
;
728 * This finds the line that we can match with the most confidence, and
729 * uses it as a partition. It then calls itself on the lines on either side of
730 * that partition. In this way we avoid lines appearing out of order, and
731 * retain a sensible line ordering.
732 * \param start_a index of the first line in A with which lines in B may be
734 * \param start_b index of the first line in B for which matching should be
736 * \param length_a number of lines in A with which lines in B may be compared.
737 * \param length_b number of lines in B for which matching should be done.
738 * \param fingerprints_a mutable array of fingerprints in A. The first element
739 * corresponds to the line at start_a.
740 * \param fingerprints_b array of fingerprints in B. The first element
741 * corresponds to the line at start_b.
742 * \param similarities 2-dimensional array of similarities between lines in A
743 * and B. See get_similarity() for more details.
744 * \param certainties array of values indicating how strongly a line in B is
745 * matched with some line in A.
746 * \param second_best_result array of absolute indices in A for the second
747 * closest match of a line in B.
748 * \param result array of absolute indices in A for the closest match of a line
750 * \param max_search_distance_a maximum distance in lines from the closest line
751 * in A for other lines in A for which
752 * similarities may be calculated.
753 * \param max_search_distance_b an upper bound on the greatest possible
754 * distance between lines in B such that they will
755 * both be compared with the same line in A
756 * according to max_search_distance_a.
757 * \param map_line_number_in_b_to_a parameter to map_line_number().
759 static void fuzzy_find_matching_lines_recurse(
760 int start_a
, int start_b
,
761 int length_a
, int length_b
,
762 struct fingerprint
*fingerprints_a
,
763 struct fingerprint
*fingerprints_b
,
766 int *second_best_result
,
768 int max_search_distance_a
,
769 int max_search_distance_b
,
770 const struct line_number_mapping
*map_line_number_in_b_to_a
)
772 int i
, invalidate_min
, invalidate_max
, offset_b
,
773 second_half_start_a
, second_half_start_b
,
774 second_half_length_a
, second_half_length_b
,
775 most_certain_line_a
, most_certain_local_line_b
= -1,
776 most_certain_line_certainty
= -1,
777 closest_local_line_a
;
779 for (i
= 0; i
< length_b
; ++i
) {
780 find_best_line_matches(start_a
,
790 max_search_distance_a
,
791 map_line_number_in_b_to_a
);
793 if (certainties
[i
] > most_certain_line_certainty
) {
794 most_certain_line_certainty
= certainties
[i
];
795 most_certain_local_line_b
= i
;
800 if (most_certain_local_line_b
== -1)
803 most_certain_line_a
= result
[most_certain_local_line_b
];
806 * Subtract the most certain line's fingerprint in B from the matched
807 * fingerprint in A. This means that other lines in B can't also match
808 * the same parts of the line in A.
810 fingerprint_subtract(fingerprints_a
+ most_certain_line_a
- start_a
,
811 fingerprints_b
+ most_certain_local_line_b
);
813 /* Invalidate results that may be affected by the choice of most
816 invalidate_min
= most_certain_local_line_b
- max_search_distance_b
;
817 invalidate_max
= most_certain_local_line_b
+ max_search_distance_b
+ 1;
818 if (invalidate_min
< 0)
820 if (invalidate_max
> length_b
)
821 invalidate_max
= length_b
;
823 /* As the fingerprint in A has changed, discard previously calculated
824 * similarity values with that fingerprint.
826 for (i
= invalidate_min
; i
< invalidate_max
; ++i
) {
827 closest_local_line_a
= map_line_number(
828 i
+ start_b
, map_line_number_in_b_to_a
) - start_a
;
830 /* Check that the lines in A and B are close enough that there
831 * is a similarity value for them.
833 if (abs(most_certain_line_a
- start_a
- closest_local_line_a
) >
834 max_search_distance_a
) {
838 *get_similarity(similarities
, most_certain_line_a
- start_a
,
839 i
, closest_local_line_a
,
840 max_search_distance_a
) = -1;
843 /* More invalidating of results that may be affected by the choice of
845 * Discard the matches for lines in B that are currently matched with a
846 * line in A such that their ordering contradicts the ordering imposed
847 * by the choice of most certain line.
849 for (i
= most_certain_local_line_b
- 1; i
>= invalidate_min
; --i
) {
850 /* In this loop we discard results for lines in B that are
851 * before most-certain-line-B but are matched with a line in A
852 * that is after most-certain-line-A.
854 if (certainties
[i
] >= 0 &&
855 (result
[i
] >= most_certain_line_a
||
856 second_best_result
[i
] >= most_certain_line_a
)) {
857 certainties
[i
] = CERTAINTY_NOT_CALCULATED
;
860 for (i
= most_certain_local_line_b
+ 1; i
< invalidate_max
; ++i
) {
861 /* In this loop we discard results for lines in B that are
862 * after most-certain-line-B but are matched with a line in A
863 * that is before most-certain-line-A.
865 if (certainties
[i
] >= 0 &&
866 (result
[i
] <= most_certain_line_a
||
867 second_best_result
[i
] <= most_certain_line_a
)) {
868 certainties
[i
] = CERTAINTY_NOT_CALCULATED
;
872 /* Repeat the matching process for lines before the most certain line.
874 if (most_certain_local_line_b
> 0) {
875 fuzzy_find_matching_lines_recurse(
877 most_certain_line_a
+ 1 - start_a
,
878 most_certain_local_line_b
,
879 fingerprints_a
, fingerprints_b
, similarities
,
880 certainties
, second_best_result
, result
,
881 max_search_distance_a
,
882 max_search_distance_b
,
883 map_line_number_in_b_to_a
);
885 /* Repeat the matching process for lines after the most certain line.
887 if (most_certain_local_line_b
+ 1 < length_b
) {
888 second_half_start_a
= most_certain_line_a
;
889 offset_b
= most_certain_local_line_b
+ 1;
890 second_half_start_b
= start_b
+ offset_b
;
891 second_half_length_a
=
892 length_a
+ start_a
- second_half_start_a
;
893 second_half_length_b
=
894 length_b
+ start_b
- second_half_start_b
;
895 fuzzy_find_matching_lines_recurse(
896 second_half_start_a
, second_half_start_b
,
897 second_half_length_a
, second_half_length_b
,
898 fingerprints_a
+ second_half_start_a
- start_a
,
899 fingerprints_b
+ offset_b
,
901 offset_b
* (max_search_distance_a
* 2 + 1),
902 certainties
+ offset_b
,
903 second_best_result
+ offset_b
, result
+ offset_b
,
904 max_search_distance_a
,
905 max_search_distance_b
,
906 map_line_number_in_b_to_a
);
910 /* Find the lines in the parent line range that most closely match the lines in
911 * the target line range. This is accomplished by matching fingerprints in each
912 * blame_origin, and choosing the best matches that preserve the line ordering.
913 * See struct fingerprint for details of fingerprint matching, and
914 * fuzzy_find_matching_lines_recurse for details of preserving line ordering.
916 * The performance is believed to be O(n log n) in the typical case and O(n^2)
917 * in a pathological case, where n is the number of lines in the target range.
919 static int *fuzzy_find_matching_lines(struct blame_origin
*parent
,
920 struct blame_origin
*target
,
921 int tlno
, int parent_slno
, int same
,
924 /* We use the terminology "A" for the left hand side of the diff AKA
925 * parent, and "B" for the right hand side of the diff AKA target. */
926 int start_a
= parent_slno
;
927 int length_a
= parent_len
;
929 int length_b
= same
- tlno
;
931 struct line_number_mapping map_line_number_in_b_to_a
= {
932 start_a
, length_a
, start_b
, length_b
935 struct fingerprint
*fingerprints_a
= parent
->fingerprints
;
936 struct fingerprint
*fingerprints_b
= target
->fingerprints
;
938 int i
, *result
, *second_best_result
,
939 *certainties
, *similarities
, similarity_count
;
942 * max_search_distance_a means that given a line in B, compare it to
943 * the line in A that is closest to its position, and the lines in A
944 * that are no greater than max_search_distance_a lines away from the
947 * max_search_distance_b is an upper bound on the greatest possible
948 * distance between lines in B such that they will both be compared
949 * with the same line in A according to max_search_distance_a.
951 int max_search_distance_a
= 10, max_search_distance_b
;
956 if (max_search_distance_a
>= length_a
)
957 max_search_distance_a
= length_a
? length_a
- 1 : 0;
959 max_search_distance_b
= ((2 * max_search_distance_a
+ 1) * length_b
962 CALLOC_ARRAY(result
, length_b
);
963 CALLOC_ARRAY(second_best_result
, length_b
);
964 CALLOC_ARRAY(certainties
, length_b
);
966 /* See get_similarity() for details of similarities. */
967 similarity_count
= length_b
* (max_search_distance_a
* 2 + 1);
968 CALLOC_ARRAY(similarities
, similarity_count
);
970 for (i
= 0; i
< length_b
; ++i
) {
972 second_best_result
[i
] = -1;
973 certainties
[i
] = CERTAINTY_NOT_CALCULATED
;
976 for (i
= 0; i
< similarity_count
; ++i
)
977 similarities
[i
] = -1;
979 fuzzy_find_matching_lines_recurse(start_a
, start_b
,
981 fingerprints_a
+ start_a
,
982 fingerprints_b
+ start_b
,
987 max_search_distance_a
,
988 max_search_distance_b
,
989 &map_line_number_in_b_to_a
);
993 free(second_best_result
);
998 static void fill_origin_fingerprints(struct blame_origin
*o
)
1002 if (o
->fingerprints
)
1004 o
->num_lines
= find_line_starts(&line_starts
, o
->file
.ptr
,
1006 CALLOC_ARRAY(o
->fingerprints
, o
->num_lines
);
1007 get_line_fingerprints(o
->fingerprints
, o
->file
.ptr
, line_starts
,
1012 static void drop_origin_fingerprints(struct blame_origin
*o
)
1014 if (o
->fingerprints
) {
1015 free_line_fingerprints(o
->fingerprints
, o
->num_lines
);
1017 FREE_AND_NULL(o
->fingerprints
);
1022 * Given an origin, prepare mmfile_t structure to be used by the
1025 static void fill_origin_blob(struct diff_options
*opt
,
1026 struct blame_origin
*o
, mmfile_t
*file
,
1027 int *num_read_blob
, int fill_fingerprints
)
1030 enum object_type type
;
1031 unsigned long file_size
;
1034 if (opt
->flags
.allow_textconv
&&
1035 textconv_object(opt
->repo
, o
->path
, o
->mode
,
1036 &o
->blob_oid
, 1, &file
->ptr
, &file_size
))
1039 file
->ptr
= repo_read_object_file(the_repository
,
1040 &o
->blob_oid
, &type
,
1042 file
->size
= file_size
;
1045 die("Cannot read blob %s for path %s",
1046 oid_to_hex(&o
->blob_oid
),
1052 if (fill_fingerprints
)
1053 fill_origin_fingerprints(o
);
1056 static void drop_origin_blob(struct blame_origin
*o
)
1058 FREE_AND_NULL(o
->file
.ptr
);
1059 drop_origin_fingerprints(o
);
1063 * Any merge of blames happens on lists of blames that arrived via
1064 * different parents in a single suspect. In this case, we want to
1065 * sort according to the suspect line numbers as opposed to the final
1066 * image line numbers. The function body is somewhat longish because
1067 * it avoids unnecessary writes.
1070 static struct blame_entry
*blame_merge(struct blame_entry
*list1
,
1071 struct blame_entry
*list2
)
1073 struct blame_entry
*p1
= list1
, *p2
= list2
,
1081 if (p1
->s_lno
<= p2
->s_lno
) {
1084 if (!(p1
= *tail
)) {
1088 } while (p1
->s_lno
<= p2
->s_lno
);
1094 if (!(p2
= *tail
)) {
1098 } while (p1
->s_lno
> p2
->s_lno
);
1102 if (!(p1
= *tail
)) {
1106 } while (p1
->s_lno
<= p2
->s_lno
);
1110 DEFINE_LIST_SORT(static, sort_blame_entries
, struct blame_entry
, next
);
1113 * Final image line numbers are all different, so we don't need a
1114 * three-way comparison here.
1117 static int compare_blame_final(const struct blame_entry
*e1
,
1118 const struct blame_entry
*e2
)
1120 return e1
->lno
> e2
->lno
? 1 : -1;
1123 static int compare_blame_suspect(const struct blame_entry
*s1
,
1124 const struct blame_entry
*s2
)
1127 * to allow for collating suspects, we sort according to the
1128 * respective pointer value as the primary sorting criterion.
1129 * The actual relation is pretty unimportant as long as it
1130 * establishes a total order. Comparing as integers gives us
1133 if (s1
->suspect
!= s2
->suspect
)
1134 return (intptr_t)s1
->suspect
> (intptr_t)s2
->suspect
? 1 : -1;
1135 if (s1
->s_lno
== s2
->s_lno
)
1137 return s1
->s_lno
> s2
->s_lno
? 1 : -1;
1140 void blame_sort_final(struct blame_scoreboard
*sb
)
1142 sort_blame_entries(&sb
->ent
, compare_blame_final
);
1145 static int compare_commits_by_reverse_commit_date(const void *a
,
1149 return -compare_commits_by_commit_date(a
, b
, c
);
1153 * For debugging -- origin is refcounted, and this asserts that
1154 * we do not underflow.
1156 static void sanity_check_refcnt(struct blame_scoreboard
*sb
)
1159 struct blame_entry
*ent
;
1161 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
1162 /* Nobody should have zero or negative refcnt */
1163 if (ent
->suspect
->refcnt
<= 0) {
1164 fprintf(stderr
, "%s in %s has negative refcnt %d\n",
1166 oid_to_hex(&ent
->suspect
->commit
->object
.oid
),
1167 ent
->suspect
->refcnt
);
1172 sb
->on_sanity_fail(sb
, baa
);
1176 * If two blame entries that are next to each other came from
1177 * contiguous lines in the same origin (i.e. <commit, path> pair),
1178 * merge them together.
1180 void blame_coalesce(struct blame_scoreboard
*sb
)
1182 struct blame_entry
*ent
, *next
;
1184 for (ent
= sb
->ent
; ent
&& (next
= ent
->next
); ent
= next
) {
1185 if (ent
->suspect
== next
->suspect
&&
1186 ent
->s_lno
+ ent
->num_lines
== next
->s_lno
&&
1187 ent
->lno
+ ent
->num_lines
== next
->lno
&&
1188 ent
->ignored
== next
->ignored
&&
1189 ent
->unblamable
== next
->unblamable
) {
1190 ent
->num_lines
+= next
->num_lines
;
1191 ent
->next
= next
->next
;
1192 blame_origin_decref(next
->suspect
);
1195 next
= ent
; /* again */
1199 if (sb
->debug
) /* sanity */
1200 sanity_check_refcnt(sb
);
1204 * Merge the given sorted list of blames into a preexisting origin.
1205 * If there were no previous blames to that commit, it is entered into
1206 * the commit priority queue of the score board.
1209 static void queue_blames(struct blame_scoreboard
*sb
, struct blame_origin
*porigin
,
1210 struct blame_entry
*sorted
)
1212 if (porigin
->suspects
)
1213 porigin
->suspects
= blame_merge(porigin
->suspects
, sorted
);
1215 struct blame_origin
*o
;
1216 for (o
= get_blame_suspects(porigin
->commit
); o
; o
= o
->next
) {
1218 porigin
->suspects
= sorted
;
1222 porigin
->suspects
= sorted
;
1223 prio_queue_put(&sb
->commits
, porigin
->commit
);
1228 * Fill the blob_sha1 field of an origin if it hasn't, so that later
1229 * call to fill_origin_blob() can use it to locate the data. blob_sha1
1230 * for an origin is also used to pass the blame for the entire file to
1231 * the parent to detect the case where a child's blob is identical to
1232 * that of its parent's.
1234 * This also fills origin->mode for corresponding tree path.
1236 static int fill_blob_sha1_and_mode(struct repository
*r
,
1237 struct blame_origin
*origin
)
1239 if (!is_null_oid(&origin
->blob_oid
))
1241 if (get_tree_entry(r
, &origin
->commit
->object
.oid
, origin
->path
, &origin
->blob_oid
, &origin
->mode
))
1243 if (oid_object_info(r
, &origin
->blob_oid
, NULL
) != OBJ_BLOB
)
1247 oidclr(&origin
->blob_oid
);
1248 origin
->mode
= S_IFINVALID
;
1252 struct blame_bloom_data
{
1254 * Changed-path Bloom filter keys. These can help prevent
1255 * computing diffs against first parents, but we need to
1256 * expand the list as code is moved or files are renamed.
1258 struct bloom_filter_settings
*settings
;
1259 struct bloom_key
**keys
;
1264 static int bloom_count_queries
= 0;
1265 static int bloom_count_no
= 0;
1266 static int maybe_changed_path(struct repository
*r
,
1267 struct blame_origin
*origin
,
1268 struct blame_bloom_data
*bd
)
1271 struct bloom_filter
*filter
;
1276 if (commit_graph_generation(origin
->commit
) == GENERATION_NUMBER_INFINITY
)
1279 filter
= get_bloom_filter(r
, origin
->commit
);
1284 bloom_count_queries
++;
1285 for (i
= 0; i
< bd
->nr
; i
++) {
1286 if (bloom_filter_contains(filter
,
1296 static void add_bloom_key(struct blame_bloom_data
*bd
,
1302 if (bd
->nr
>= bd
->alloc
) {
1304 REALLOC_ARRAY(bd
->keys
, bd
->alloc
);
1307 bd
->keys
[bd
->nr
] = xmalloc(sizeof(struct bloom_key
));
1308 fill_bloom_key(path
, strlen(path
), bd
->keys
[bd
->nr
], bd
->settings
);
1313 * We have an origin -- check if the same path exists in the
1314 * parent and return an origin structure to represent it.
1316 static struct blame_origin
*find_origin(struct repository
*r
,
1317 struct commit
*parent
,
1318 struct blame_origin
*origin
,
1319 struct blame_bloom_data
*bd
)
1321 struct blame_origin
*porigin
;
1322 struct diff_options diff_opts
;
1323 const char *paths
[2];
1325 /* First check any existing origins */
1326 for (porigin
= get_blame_suspects(parent
); porigin
; porigin
= porigin
->next
)
1327 if (!strcmp(porigin
->path
, origin
->path
)) {
1329 * The same path between origin and its parent
1330 * without renaming -- the most common case.
1332 return blame_origin_incref (porigin
);
1335 /* See if the origin->path is different between parent
1336 * and origin first. Most of the time they are the
1337 * same and diff-tree is fairly efficient about this.
1339 repo_diff_setup(r
, &diff_opts
);
1340 diff_opts
.flags
.recursive
= 1;
1341 diff_opts
.detect_rename
= 0;
1342 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
1343 paths
[0] = origin
->path
;
1346 parse_pathspec(&diff_opts
.pathspec
,
1347 PATHSPEC_ALL_MAGIC
& ~PATHSPEC_LITERAL
,
1348 PATHSPEC_LITERAL_PATH
, "", paths
);
1349 diff_setup_done(&diff_opts
);
1351 if (is_null_oid(&origin
->commit
->object
.oid
))
1352 do_diff_cache(get_commit_tree_oid(parent
), &diff_opts
);
1354 int compute_diff
= 1;
1355 if (origin
->commit
->parents
&&
1356 oideq(&parent
->object
.oid
,
1357 &origin
->commit
->parents
->item
->object
.oid
))
1358 compute_diff
= maybe_changed_path(r
, origin
, bd
);
1361 diff_tree_oid(get_commit_tree_oid(parent
),
1362 get_commit_tree_oid(origin
->commit
),
1365 diffcore_std(&diff_opts
);
1367 if (!diff_queued_diff
.nr
) {
1368 /* The path is the same as parent */
1369 porigin
= get_origin(parent
, origin
->path
);
1370 oidcpy(&porigin
->blob_oid
, &origin
->blob_oid
);
1371 porigin
->mode
= origin
->mode
;
1374 * Since origin->path is a pathspec, if the parent
1375 * commit had it as a directory, we will see a whole
1376 * bunch of deletion of files in the directory that we
1377 * do not care about.
1380 struct diff_filepair
*p
= NULL
;
1381 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
1383 p
= diff_queued_diff
.queue
[i
];
1384 name
= p
->one
->path
? p
->one
->path
: p
->two
->path
;
1385 if (!strcmp(name
, origin
->path
))
1389 die("internal error in blame::find_origin");
1390 switch (p
->status
) {
1392 die("internal error in blame::find_origin (%c)",
1395 porigin
= get_origin(parent
, origin
->path
);
1396 oidcpy(&porigin
->blob_oid
, &p
->one
->oid
);
1397 porigin
->mode
= p
->one
->mode
;
1401 /* Did not exist in parent, or type changed */
1405 diff_flush(&diff_opts
);
1410 * We have an origin -- find the path that corresponds to it in its
1411 * parent and return an origin structure to represent it.
1413 static struct blame_origin
*find_rename(struct repository
*r
,
1414 struct commit
*parent
,
1415 struct blame_origin
*origin
,
1416 struct blame_bloom_data
*bd
)
1418 struct blame_origin
*porigin
= NULL
;
1419 struct diff_options diff_opts
;
1422 repo_diff_setup(r
, &diff_opts
);
1423 diff_opts
.flags
.recursive
= 1;
1424 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
1425 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
1426 diff_opts
.single_follow
= origin
->path
;
1427 diff_setup_done(&diff_opts
);
1429 if (is_null_oid(&origin
->commit
->object
.oid
))
1430 do_diff_cache(get_commit_tree_oid(parent
), &diff_opts
);
1432 diff_tree_oid(get_commit_tree_oid(parent
),
1433 get_commit_tree_oid(origin
->commit
),
1435 diffcore_std(&diff_opts
);
1437 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
1438 struct diff_filepair
*p
= diff_queued_diff
.queue
[i
];
1439 if ((p
->status
== 'R' || p
->status
== 'C') &&
1440 !strcmp(p
->two
->path
, origin
->path
)) {
1441 add_bloom_key(bd
, p
->one
->path
);
1442 porigin
= get_origin(parent
, p
->one
->path
);
1443 oidcpy(&porigin
->blob_oid
, &p
->one
->oid
);
1444 porigin
->mode
= p
->one
->mode
;
1448 diff_flush(&diff_opts
);
1453 * Append a new blame entry to a given output queue.
1455 static void add_blame_entry(struct blame_entry
***queue
,
1456 const struct blame_entry
*src
)
1458 struct blame_entry
*e
= xmalloc(sizeof(*e
));
1459 memcpy(e
, src
, sizeof(*e
));
1460 blame_origin_incref(e
->suspect
);
1468 * src typically is on-stack; we want to copy the information in it to
1469 * a malloced blame_entry that gets added to the given queue. The
1470 * origin of dst loses a refcnt.
1472 static void dup_entry(struct blame_entry
***queue
,
1473 struct blame_entry
*dst
, struct blame_entry
*src
)
1475 blame_origin_incref(src
->suspect
);
1476 blame_origin_decref(dst
->suspect
);
1477 memcpy(dst
, src
, sizeof(*src
));
1478 dst
->next
= **queue
;
1480 *queue
= &dst
->next
;
1483 const char *blame_nth_line(struct blame_scoreboard
*sb
, long lno
)
1485 return sb
->final_buf
+ sb
->lineno
[lno
];
1489 * It is known that lines between tlno to same came from parent, and e
1490 * has an overlap with that range. it also is known that parent's
1491 * line plno corresponds to e's line tlno.
1497 * <------------------>
1499 * Split e into potentially three parts; before this chunk, the chunk
1500 * to be blamed for the parent, and after that portion.
1502 static void split_overlap(struct blame_entry
*split
,
1503 struct blame_entry
*e
,
1504 int tlno
, int plno
, int same
,
1505 struct blame_origin
*parent
)
1509 memset(split
, 0, sizeof(struct blame_entry
[3]));
1511 for (i
= 0; i
< 3; i
++) {
1512 split
[i
].ignored
= e
->ignored
;
1513 split
[i
].unblamable
= e
->unblamable
;
1516 if (e
->s_lno
< tlno
) {
1517 /* there is a pre-chunk part not blamed on parent */
1518 split
[0].suspect
= blame_origin_incref(e
->suspect
);
1519 split
[0].lno
= e
->lno
;
1520 split
[0].s_lno
= e
->s_lno
;
1521 split
[0].num_lines
= tlno
- e
->s_lno
;
1522 split
[1].lno
= e
->lno
+ tlno
- e
->s_lno
;
1523 split
[1].s_lno
= plno
;
1526 split
[1].lno
= e
->lno
;
1527 split
[1].s_lno
= plno
+ (e
->s_lno
- tlno
);
1530 if (same
< e
->s_lno
+ e
->num_lines
) {
1531 /* there is a post-chunk part not blamed on parent */
1532 split
[2].suspect
= blame_origin_incref(e
->suspect
);
1533 split
[2].lno
= e
->lno
+ (same
- e
->s_lno
);
1534 split
[2].s_lno
= e
->s_lno
+ (same
- e
->s_lno
);
1535 split
[2].num_lines
= e
->s_lno
+ e
->num_lines
- same
;
1536 chunk_end_lno
= split
[2].lno
;
1539 chunk_end_lno
= e
->lno
+ e
->num_lines
;
1540 split
[1].num_lines
= chunk_end_lno
- split
[1].lno
;
1543 * if it turns out there is nothing to blame the parent for,
1544 * forget about the splitting. !split[1].suspect signals this.
1546 if (split
[1].num_lines
< 1)
1548 split
[1].suspect
= blame_origin_incref(parent
);
1552 * split_overlap() divided an existing blame e into up to three parts
1553 * in split. Any assigned blame is moved to queue to
1554 * reflect the split.
1556 static void split_blame(struct blame_entry
***blamed
,
1557 struct blame_entry
***unblamed
,
1558 struct blame_entry
*split
,
1559 struct blame_entry
*e
)
1561 if (split
[0].suspect
&& split
[2].suspect
) {
1562 /* The first part (reuse storage for the existing entry e) */
1563 dup_entry(unblamed
, e
, &split
[0]);
1565 /* The last part -- me */
1566 add_blame_entry(unblamed
, &split
[2]);
1568 /* ... and the middle part -- parent */
1569 add_blame_entry(blamed
, &split
[1]);
1571 else if (!split
[0].suspect
&& !split
[2].suspect
)
1573 * The parent covers the entire area; reuse storage for
1574 * e and replace it with the parent.
1576 dup_entry(blamed
, e
, &split
[1]);
1577 else if (split
[0].suspect
) {
1578 /* me and then parent */
1579 dup_entry(unblamed
, e
, &split
[0]);
1580 add_blame_entry(blamed
, &split
[1]);
1583 /* parent and then me */
1584 dup_entry(blamed
, e
, &split
[1]);
1585 add_blame_entry(unblamed
, &split
[2]);
1590 * After splitting the blame, the origins used by the
1591 * on-stack blame_entry should lose one refcnt each.
1593 static void decref_split(struct blame_entry
*split
)
1597 for (i
= 0; i
< 3; i
++)
1598 blame_origin_decref(split
[i
].suspect
);
1602 * reverse_blame reverses the list given in head, appending tail.
1603 * That allows us to build lists in reverse order, then reverse them
1604 * afterwards. This can be faster than building the list in proper
1605 * order right away. The reason is that building in proper order
1606 * requires writing a link in the _previous_ element, while building
1607 * in reverse order just requires placing the list head into the
1608 * _current_ element.
1611 static struct blame_entry
*reverse_blame(struct blame_entry
*head
,
1612 struct blame_entry
*tail
)
1615 struct blame_entry
*next
= head
->next
;
1624 * Splits a blame entry into two entries at 'len' lines. The original 'e'
1625 * consists of len lines, i.e. [e->lno, e->lno + len), and the second part,
1626 * which is returned, consists of the remainder: [e->lno + len, e->lno +
1627 * e->num_lines). The caller needs to sort out the reference counting for the
1628 * new entry's suspect.
1630 static struct blame_entry
*split_blame_at(struct blame_entry
*e
, int len
,
1631 struct blame_origin
*new_suspect
)
1633 struct blame_entry
*n
= xcalloc(1, sizeof(struct blame_entry
));
1635 n
->suspect
= new_suspect
;
1636 n
->ignored
= e
->ignored
;
1637 n
->unblamable
= e
->unblamable
;
1638 n
->lno
= e
->lno
+ len
;
1639 n
->s_lno
= e
->s_lno
+ len
;
1640 n
->num_lines
= e
->num_lines
- len
;
1646 struct blame_line_tracker
{
1651 static int are_lines_adjacent(struct blame_line_tracker
*first
,
1652 struct blame_line_tracker
*second
)
1654 return first
->is_parent
== second
->is_parent
&&
1655 first
->s_lno
+ 1 == second
->s_lno
;
1658 static int scan_parent_range(struct fingerprint
*p_fps
,
1659 struct fingerprint
*t_fps
, int t_idx
,
1660 int from
, int nr_lines
)
1663 #define FINGERPRINT_FILE_THRESHOLD 10
1664 int best_sim_val
= FINGERPRINT_FILE_THRESHOLD
;
1665 int best_sim_idx
= -1;
1667 for (p_idx
= from
; p_idx
< from
+ nr_lines
; p_idx
++) {
1668 sim
= fingerprint_similarity(&t_fps
[t_idx
], &p_fps
[p_idx
]);
1669 if (sim
< best_sim_val
)
1671 /* Break ties with the closest-to-target line number */
1672 if (sim
== best_sim_val
&& best_sim_idx
!= -1 &&
1673 abs(best_sim_idx
- t_idx
) < abs(p_idx
- t_idx
))
1676 best_sim_idx
= p_idx
;
1678 return best_sim_idx
;
1682 * The first pass checks the blame entry (from the target) against the parent's
1683 * diff chunk. If that fails for a line, the second pass tries to match that
1684 * line to any part of parent file. That catches cases where a change was
1685 * broken into two chunks by 'context.'
1687 static void guess_line_blames(struct blame_origin
*parent
,
1688 struct blame_origin
*target
,
1689 int tlno
, int offset
, int same
, int parent_len
,
1690 struct blame_line_tracker
*line_blames
)
1692 int i
, best_idx
, target_idx
;
1693 int parent_slno
= tlno
+ offset
;
1696 fuzzy_matches
= fuzzy_find_matching_lines(parent
, target
,
1697 tlno
, parent_slno
, same
,
1699 for (i
= 0; i
< same
- tlno
; i
++) {
1700 target_idx
= tlno
+ i
;
1701 if (fuzzy_matches
&& fuzzy_matches
[i
] >= 0) {
1702 best_idx
= fuzzy_matches
[i
];
1704 best_idx
= scan_parent_range(parent
->fingerprints
,
1705 target
->fingerprints
,
1709 if (best_idx
>= 0) {
1710 line_blames
[i
].is_parent
= 1;
1711 line_blames
[i
].s_lno
= best_idx
;
1713 line_blames
[i
].is_parent
= 0;
1714 line_blames
[i
].s_lno
= target_idx
;
1717 free(fuzzy_matches
);
1721 * This decides which parts of a blame entry go to the parent (added to the
1722 * ignoredp list) and which stay with the target (added to the diffp list). The
1723 * actual decision was made in a separate heuristic function, and those answers
1724 * for the lines in 'e' are in line_blames. This consumes e, essentially
1725 * putting it on a list.
1727 * Note that the blame entries on the ignoredp list are not necessarily sorted
1728 * with respect to the parent's line numbers yet.
1730 static void ignore_blame_entry(struct blame_entry
*e
,
1731 struct blame_origin
*parent
,
1732 struct blame_entry
**diffp
,
1733 struct blame_entry
**ignoredp
,
1734 struct blame_line_tracker
*line_blames
)
1736 int entry_len
, nr_lines
, i
;
1739 * We carve new entries off the front of e. Each entry comes from a
1740 * contiguous chunk of lines: adjacent lines from the same origin
1741 * (either the parent or the target).
1744 nr_lines
= e
->num_lines
; /* e changes in the loop */
1745 for (i
= 0; i
< nr_lines
; i
++) {
1746 struct blame_entry
*next
= NULL
;
1749 * We are often adjacent to the next line - only split the blame
1750 * entry when we have to.
1752 if (i
+ 1 < nr_lines
) {
1753 if (are_lines_adjacent(&line_blames
[i
],
1754 &line_blames
[i
+ 1])) {
1758 next
= split_blame_at(e
, entry_len
,
1759 blame_origin_incref(e
->suspect
));
1761 if (line_blames
[i
].is_parent
) {
1763 blame_origin_decref(e
->suspect
);
1764 e
->suspect
= blame_origin_incref(parent
);
1765 e
->s_lno
= line_blames
[i
- entry_len
+ 1].s_lno
;
1766 e
->next
= *ignoredp
;
1770 /* e->s_lno is already in the target's address space. */
1774 assert(e
->num_lines
== entry_len
);
1782 * Process one hunk from the patch between the current suspect for
1783 * blame_entry e and its parent. This first blames any unfinished
1784 * entries before the chunk (which is where target and parent start
1785 * differing) on the parent, and then splits blame entries at the
1786 * start and at the end of the difference region. Since use of -M and
1787 * -C options may lead to overlapping/duplicate source line number
1788 * ranges, all we can rely on from sorting/merging is the order of the
1789 * first suspect line number.
1791 * tlno: line number in the target where this chunk begins
1792 * same: line number in the target where this chunk ends
1793 * offset: add to tlno to get the chunk starting point in the parent
1794 * parent_len: number of lines in the parent chunk
1796 static void blame_chunk(struct blame_entry
***dstq
, struct blame_entry
***srcq
,
1797 int tlno
, int offset
, int same
, int parent_len
,
1798 struct blame_origin
*parent
,
1799 struct blame_origin
*target
, int ignore_diffs
)
1801 struct blame_entry
*e
= **srcq
;
1802 struct blame_entry
*samep
= NULL
, *diffp
= NULL
, *ignoredp
= NULL
;
1803 struct blame_line_tracker
*line_blames
= NULL
;
1805 while (e
&& e
->s_lno
< tlno
) {
1806 struct blame_entry
*next
= e
->next
;
1808 * current record starts before differing portion. If
1809 * it reaches into it, we need to split it up and
1810 * examine the second part separately.
1812 if (e
->s_lno
+ e
->num_lines
> tlno
) {
1813 /* Move second half to a new record */
1814 struct blame_entry
*n
;
1816 n
= split_blame_at(e
, tlno
- e
->s_lno
, e
->suspect
);
1817 /* Push new record to diffp */
1821 blame_origin_decref(e
->suspect
);
1822 /* Pass blame for everything before the differing
1823 * chunk to the parent */
1824 e
->suspect
= blame_origin_incref(parent
);
1831 * As we don't know how much of a common stretch after this
1832 * diff will occur, the currently blamed parts are all that we
1833 * can assign to the parent for now.
1837 **dstq
= reverse_blame(samep
, **dstq
);
1838 *dstq
= &samep
->next
;
1841 * Prepend the split off portions: everything after e starts
1842 * after the blameable portion.
1844 e
= reverse_blame(diffp
, e
);
1847 * Now retain records on the target while parts are different
1853 if (ignore_diffs
&& same
- tlno
> 0) {
1854 CALLOC_ARRAY(line_blames
, same
- tlno
);
1855 guess_line_blames(parent
, target
, tlno
, offset
, same
,
1856 parent_len
, line_blames
);
1859 while (e
&& e
->s_lno
< same
) {
1860 struct blame_entry
*next
= e
->next
;
1863 * If current record extends into sameness, need to split.
1865 if (e
->s_lno
+ e
->num_lines
> same
) {
1867 * Move second half to a new record to be
1868 * processed by later chunks
1870 struct blame_entry
*n
;
1872 n
= split_blame_at(e
, same
- e
->s_lno
,
1873 blame_origin_incref(e
->suspect
));
1874 /* Push new record to samep */
1879 ignore_blame_entry(e
, parent
, &diffp
, &ignoredp
,
1880 line_blames
+ e
->s_lno
- tlno
);
1890 * Note ignoredp is not sorted yet, and thus neither is dstq.
1891 * That list must be sorted before we queue_blames(). We defer
1892 * sorting until after all diff hunks are processed, so that
1893 * guess_line_blames() can pick *any* line in the parent. The
1894 * slight drawback is that we end up sorting all blame entries
1895 * passed to the parent, including those that are unrelated to
1896 * changes made by the ignored commit.
1898 **dstq
= reverse_blame(ignoredp
, **dstq
);
1899 *dstq
= &ignoredp
->next
;
1901 **srcq
= reverse_blame(diffp
, reverse_blame(samep
, e
));
1902 /* Move across elements that are in the unblamable portion */
1904 *srcq
= &diffp
->next
;
1907 struct blame_chunk_cb_data
{
1908 struct blame_origin
*parent
;
1909 struct blame_origin
*target
;
1912 struct blame_entry
**dstq
;
1913 struct blame_entry
**srcq
;
1916 /* diff chunks are from parent to target */
1917 static int blame_chunk_cb(long start_a
, long count_a
,
1918 long start_b
, long count_b
, void *data
)
1920 struct blame_chunk_cb_data
*d
= data
;
1921 if (start_a
- start_b
!= d
->offset
)
1922 die("internal error in blame::blame_chunk_cb");
1923 blame_chunk(&d
->dstq
, &d
->srcq
, start_b
, start_a
- start_b
,
1924 start_b
+ count_b
, count_a
, d
->parent
, d
->target
,
1926 d
->offset
= start_a
+ count_a
- (start_b
+ count_b
);
1931 * We are looking at the origin 'target' and aiming to pass blame
1932 * for the lines it is suspected to its parent. Run diff to find
1933 * which lines came from parent and pass blame for them.
1935 static void pass_blame_to_parent(struct blame_scoreboard
*sb
,
1936 struct blame_origin
*target
,
1937 struct blame_origin
*parent
, int ignore_diffs
)
1939 mmfile_t file_p
, file_o
;
1940 struct blame_chunk_cb_data d
;
1941 struct blame_entry
*newdest
= NULL
;
1943 if (!target
->suspects
)
1944 return; /* nothing remains for this target */
1949 d
.ignore_diffs
= ignore_diffs
;
1950 d
.dstq
= &newdest
; d
.srcq
= &target
->suspects
;
1952 fill_origin_blob(&sb
->revs
->diffopt
, parent
, &file_p
,
1953 &sb
->num_read_blob
, ignore_diffs
);
1954 fill_origin_blob(&sb
->revs
->diffopt
, target
, &file_o
,
1955 &sb
->num_read_blob
, ignore_diffs
);
1956 sb
->num_get_patch
++;
1958 if (diff_hunks(&file_p
, &file_o
, blame_chunk_cb
, &d
, sb
->xdl_opts
))
1959 die("unable to generate diff (%s -> %s)",
1960 oid_to_hex(&parent
->commit
->object
.oid
),
1961 oid_to_hex(&target
->commit
->object
.oid
));
1962 /* The rest are the same as the parent */
1963 blame_chunk(&d
.dstq
, &d
.srcq
, INT_MAX
, d
.offset
, INT_MAX
, 0,
1967 sort_blame_entries(&newdest
, compare_blame_suspect
);
1968 queue_blames(sb
, parent
, newdest
);
1974 * The lines in blame_entry after splitting blames many times can become
1975 * very small and trivial, and at some point it becomes pointless to
1976 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
1977 * ordinary C program, and it is not worth to say it was copied from
1978 * totally unrelated file in the parent.
1980 * Compute how trivial the lines in the blame_entry are.
1982 unsigned blame_entry_score(struct blame_scoreboard
*sb
, struct blame_entry
*e
)
1985 const char *cp
, *ep
;
1991 cp
= blame_nth_line(sb
, e
->lno
);
1992 ep
= blame_nth_line(sb
, e
->lno
+ e
->num_lines
);
1994 unsigned ch
= *((unsigned char *)cp
);
2004 * best_so_far[] and potential[] are both a split of an existing blame_entry
2005 * that passes blame to the parent. Maintain best_so_far the best split so
2006 * far, by comparing potential and best_so_far and copying potential into
2007 * bst_so_far as needed.
2009 static void copy_split_if_better(struct blame_scoreboard
*sb
,
2010 struct blame_entry
*best_so_far
,
2011 struct blame_entry
*potential
)
2015 if (!potential
[1].suspect
)
2017 if (best_so_far
[1].suspect
) {
2018 if (blame_entry_score(sb
, &potential
[1]) <
2019 blame_entry_score(sb
, &best_so_far
[1]))
2023 for (i
= 0; i
< 3; i
++)
2024 blame_origin_incref(potential
[i
].suspect
);
2025 decref_split(best_so_far
);
2026 memcpy(best_so_far
, potential
, sizeof(struct blame_entry
[3]));
2030 * We are looking at a part of the final image represented by
2031 * ent (tlno and same are offset by ent->s_lno).
2032 * tlno is where we are looking at in the final image.
2033 * up to (but not including) same match preimage.
2034 * plno is where we are looking at in the preimage.
2036 * <-------------- final image ---------------------->
2039 * <---------preimage----->
2042 * All line numbers are 0-based.
2044 static void handle_split(struct blame_scoreboard
*sb
,
2045 struct blame_entry
*ent
,
2046 int tlno
, int plno
, int same
,
2047 struct blame_origin
*parent
,
2048 struct blame_entry
*split
)
2050 if (ent
->num_lines
<= tlno
)
2053 struct blame_entry potential
[3];
2056 split_overlap(potential
, ent
, tlno
, plno
, same
, parent
);
2057 copy_split_if_better(sb
, split
, potential
);
2058 decref_split(potential
);
2062 struct handle_split_cb_data
{
2063 struct blame_scoreboard
*sb
;
2064 struct blame_entry
*ent
;
2065 struct blame_origin
*parent
;
2066 struct blame_entry
*split
;
2071 static int handle_split_cb(long start_a
, long count_a
,
2072 long start_b
, long count_b
, void *data
)
2074 struct handle_split_cb_data
*d
= data
;
2075 handle_split(d
->sb
, d
->ent
, d
->tlno
, d
->plno
, start_b
, d
->parent
,
2077 d
->plno
= start_a
+ count_a
;
2078 d
->tlno
= start_b
+ count_b
;
2083 * Find the lines from parent that are the same as ent so that
2084 * we can pass blames to it. file_p has the blob contents for
2087 static void find_copy_in_blob(struct blame_scoreboard
*sb
,
2088 struct blame_entry
*ent
,
2089 struct blame_origin
*parent
,
2090 struct blame_entry
*split
,
2095 struct handle_split_cb_data d
;
2097 memset(&d
, 0, sizeof(d
));
2098 d
.sb
= sb
; d
.ent
= ent
; d
.parent
= parent
; d
.split
= split
;
2100 * Prepare mmfile that contains only the lines in ent.
2102 cp
= blame_nth_line(sb
, ent
->lno
);
2103 file_o
.ptr
= (char *) cp
;
2104 file_o
.size
= blame_nth_line(sb
, ent
->lno
+ ent
->num_lines
) - cp
;
2107 * file_o is a part of final image we are annotating.
2108 * file_p partially may match that image.
2110 memset(split
, 0, sizeof(struct blame_entry
[3]));
2111 if (diff_hunks(file_p
, &file_o
, handle_split_cb
, &d
, sb
->xdl_opts
))
2112 die("unable to generate diff (%s)",
2113 oid_to_hex(&parent
->commit
->object
.oid
));
2114 /* remainder, if any, all match the preimage */
2115 handle_split(sb
, ent
, d
.tlno
, d
.plno
, ent
->num_lines
, parent
, split
);
2118 /* Move all blame entries from list *source that have a score smaller
2119 * than score_min to the front of list *small.
2120 * Returns a pointer to the link pointing to the old head of the small list.
2123 static struct blame_entry
**filter_small(struct blame_scoreboard
*sb
,
2124 struct blame_entry
**small
,
2125 struct blame_entry
**source
,
2128 struct blame_entry
*p
= *source
;
2129 struct blame_entry
*oldsmall
= *small
;
2131 if (blame_entry_score(sb
, p
) <= score_min
) {
2147 * See if lines currently target is suspected for can be attributed to
2150 static void find_move_in_parent(struct blame_scoreboard
*sb
,
2151 struct blame_entry
***blamed
,
2152 struct blame_entry
**toosmall
,
2153 struct blame_origin
*target
,
2154 struct blame_origin
*parent
)
2156 struct blame_entry
*e
, split
[3];
2157 struct blame_entry
*unblamed
= target
->suspects
;
2158 struct blame_entry
*leftover
= NULL
;
2162 return; /* nothing remains for this target */
2164 fill_origin_blob(&sb
->revs
->diffopt
, parent
, &file_p
,
2165 &sb
->num_read_blob
, 0);
2169 /* At each iteration, unblamed has a NULL-terminated list of
2170 * entries that have not yet been tested for blame. leftover
2171 * contains the reversed list of entries that have been tested
2172 * without being assignable to the parent.
2175 struct blame_entry
**unblamedtail
= &unblamed
;
2176 struct blame_entry
*next
;
2177 for (e
= unblamed
; e
; e
= next
) {
2179 find_copy_in_blob(sb
, e
, parent
, split
, &file_p
);
2180 if (split
[1].suspect
&&
2181 sb
->move_score
< blame_entry_score(sb
, &split
[1])) {
2182 split_blame(blamed
, &unblamedtail
, split
, e
);
2187 decref_split(split
);
2189 *unblamedtail
= NULL
;
2190 toosmall
= filter_small(sb
, toosmall
, &unblamed
, sb
->move_score
);
2192 target
->suspects
= reverse_blame(leftover
, NULL
);
2196 struct blame_entry
*ent
;
2197 struct blame_entry split
[3];
2201 * Count the number of entries the target is suspected for,
2202 * and prepare a list of entry and the best split.
2204 static struct blame_list
*setup_blame_list(struct blame_entry
*unblamed
,
2207 struct blame_entry
*e
;
2209 struct blame_list
*blame_list
= NULL
;
2211 for (e
= unblamed
, num_ents
= 0; e
; e
= e
->next
)
2214 CALLOC_ARRAY(blame_list
, num_ents
);
2215 for (e
= unblamed
, i
= 0; e
; e
= e
->next
)
2216 blame_list
[i
++].ent
= e
;
2218 *num_ents_p
= num_ents
;
2223 * For lines target is suspected for, see if we can find code movement
2224 * across file boundary from the parent commit. porigin is the path
2225 * in the parent we already tried.
2227 static void find_copy_in_parent(struct blame_scoreboard
*sb
,
2228 struct blame_entry
***blamed
,
2229 struct blame_entry
**toosmall
,
2230 struct blame_origin
*target
,
2231 struct commit
*parent
,
2232 struct blame_origin
*porigin
,
2235 struct diff_options diff_opts
;
2237 struct blame_list
*blame_list
;
2239 struct blame_entry
*unblamed
= target
->suspects
;
2240 struct blame_entry
*leftover
= NULL
;
2243 return; /* nothing remains for this target */
2245 repo_diff_setup(sb
->repo
, &diff_opts
);
2246 diff_opts
.flags
.recursive
= 1;
2247 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
2249 diff_setup_done(&diff_opts
);
2251 /* Try "find copies harder" on new path if requested;
2252 * we do not want to use diffcore_rename() actually to
2253 * match things up; find_copies_harder is set only to
2254 * force diff_tree_oid() to feed all filepairs to diff_queue,
2255 * and this code needs to be after diff_setup_done(), which
2256 * usually makes find-copies-harder imply copy detection.
2258 if ((opt
& PICKAXE_BLAME_COPY_HARDEST
)
2259 || ((opt
& PICKAXE_BLAME_COPY_HARDER
)
2260 && (!porigin
|| strcmp(target
->path
, porigin
->path
))))
2261 diff_opts
.flags
.find_copies_harder
= 1;
2263 if (is_null_oid(&target
->commit
->object
.oid
))
2264 do_diff_cache(get_commit_tree_oid(parent
), &diff_opts
);
2266 diff_tree_oid(get_commit_tree_oid(parent
),
2267 get_commit_tree_oid(target
->commit
),
2270 if (!diff_opts
.flags
.find_copies_harder
)
2271 diffcore_std(&diff_opts
);
2274 struct blame_entry
**unblamedtail
= &unblamed
;
2275 blame_list
= setup_blame_list(unblamed
, &num_ents
);
2277 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
2278 struct diff_filepair
*p
= diff_queued_diff
.queue
[i
];
2279 struct blame_origin
*norigin
;
2281 struct blame_entry potential
[3];
2283 if (!DIFF_FILE_VALID(p
->one
))
2284 continue; /* does not exist in parent */
2285 if (S_ISGITLINK(p
->one
->mode
))
2286 continue; /* ignore git links */
2287 if (porigin
&& !strcmp(p
->one
->path
, porigin
->path
))
2288 /* find_move already dealt with this path */
2291 norigin
= get_origin(parent
, p
->one
->path
);
2292 oidcpy(&norigin
->blob_oid
, &p
->one
->oid
);
2293 norigin
->mode
= p
->one
->mode
;
2294 fill_origin_blob(&sb
->revs
->diffopt
, norigin
, &file_p
,
2295 &sb
->num_read_blob
, 0);
2299 for (j
= 0; j
< num_ents
; j
++) {
2300 find_copy_in_blob(sb
, blame_list
[j
].ent
,
2301 norigin
, potential
, &file_p
);
2302 copy_split_if_better(sb
, blame_list
[j
].split
,
2304 decref_split(potential
);
2306 blame_origin_decref(norigin
);
2309 for (j
= 0; j
< num_ents
; j
++) {
2310 struct blame_entry
*split
= blame_list
[j
].split
;
2311 if (split
[1].suspect
&&
2312 sb
->copy_score
< blame_entry_score(sb
, &split
[1])) {
2313 split_blame(blamed
, &unblamedtail
, split
,
2316 blame_list
[j
].ent
->next
= leftover
;
2317 leftover
= blame_list
[j
].ent
;
2319 decref_split(split
);
2322 *unblamedtail
= NULL
;
2323 toosmall
= filter_small(sb
, toosmall
, &unblamed
, sb
->copy_score
);
2325 target
->suspects
= reverse_blame(leftover
, NULL
);
2326 diff_flush(&diff_opts
);
2330 * The blobs of origin and porigin exactly match, so everything
2331 * origin is suspected for can be blamed on the parent.
2333 static void pass_whole_blame(struct blame_scoreboard
*sb
,
2334 struct blame_origin
*origin
, struct blame_origin
*porigin
)
2336 struct blame_entry
*e
, *suspects
;
2338 if (!porigin
->file
.ptr
&& origin
->file
.ptr
) {
2339 /* Steal its file */
2340 porigin
->file
= origin
->file
;
2341 origin
->file
.ptr
= NULL
;
2343 suspects
= origin
->suspects
;
2344 origin
->suspects
= NULL
;
2345 for (e
= suspects
; e
; e
= e
->next
) {
2346 blame_origin_incref(porigin
);
2347 blame_origin_decref(e
->suspect
);
2348 e
->suspect
= porigin
;
2350 queue_blames(sb
, porigin
, suspects
);
2354 * We pass blame from the current commit to its parents. We keep saying
2355 * "parent" (and "porigin"), but what we mean is to find scapegoat to
2356 * exonerate ourselves.
2358 static struct commit_list
*first_scapegoat(struct rev_info
*revs
, struct commit
*commit
,
2362 if (revs
->first_parent_only
&&
2364 commit
->parents
->next
) {
2365 free_commit_list(commit
->parents
->next
);
2366 commit
->parents
->next
= NULL
;
2368 return commit
->parents
;
2370 return lookup_decoration(&revs
->children
, &commit
->object
);
2373 static int num_scapegoats(struct rev_info
*revs
, struct commit
*commit
, int reverse
)
2375 struct commit_list
*l
= first_scapegoat(revs
, commit
, reverse
);
2376 return commit_list_count(l
);
2379 /* Distribute collected unsorted blames to the respected sorted lists
2380 * in the various origins.
2382 static void distribute_blame(struct blame_scoreboard
*sb
, struct blame_entry
*blamed
)
2384 sort_blame_entries(&blamed
, compare_blame_suspect
);
2387 struct blame_origin
*porigin
= blamed
->suspect
;
2388 struct blame_entry
*suspects
= NULL
;
2390 struct blame_entry
*next
= blamed
->next
;
2391 blamed
->next
= suspects
;
2394 } while (blamed
&& blamed
->suspect
== porigin
);
2395 suspects
= reverse_blame(suspects
, NULL
);
2396 queue_blames(sb
, porigin
, suspects
);
2402 typedef struct blame_origin
*(*blame_find_alg
)(struct repository
*,
2404 struct blame_origin
*,
2405 struct blame_bloom_data
*);
2407 static void pass_blame(struct blame_scoreboard
*sb
, struct blame_origin
*origin
, int opt
)
2409 struct rev_info
*revs
= sb
->revs
;
2410 int i
, pass
, num_sg
;
2411 struct commit
*commit
= origin
->commit
;
2412 struct commit_list
*sg
;
2413 struct blame_origin
*sg_buf
[MAXSG
];
2414 struct blame_origin
*porigin
, **sg_origin
= sg_buf
;
2415 struct blame_entry
*toosmall
= NULL
;
2416 struct blame_entry
*blames
, **blametail
= &blames
;
2418 num_sg
= num_scapegoats(revs
, commit
, sb
->reverse
);
2421 else if (num_sg
< ARRAY_SIZE(sg_buf
))
2422 memset(sg_buf
, 0, sizeof(sg_buf
));
2424 CALLOC_ARRAY(sg_origin
, num_sg
);
2427 * The first pass looks for unrenamed path to optimize for
2428 * common cases, then we look for renames in the second pass.
2430 for (pass
= 0; pass
< 2 - sb
->no_whole_file_rename
; pass
++) {
2431 blame_find_alg find
= pass
? find_rename
: find_origin
;
2433 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2435 sg
= sg
->next
, i
++) {
2436 struct commit
*p
= sg
->item
;
2441 if (repo_parse_commit(the_repository
, p
))
2443 porigin
= find(sb
->repo
, p
, origin
, sb
->bloom_data
);
2446 if (oideq(&porigin
->blob_oid
, &origin
->blob_oid
)) {
2447 pass_whole_blame(sb
, origin
, porigin
);
2448 blame_origin_decref(porigin
);
2451 for (j
= same
= 0; j
< i
; j
++)
2453 oideq(&sg_origin
[j
]->blob_oid
, &porigin
->blob_oid
)) {
2458 sg_origin
[i
] = porigin
;
2460 blame_origin_decref(porigin
);
2465 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2467 sg
= sg
->next
, i
++) {
2468 struct blame_origin
*porigin
= sg_origin
[i
];
2471 if (!origin
->previous
) {
2472 blame_origin_incref(porigin
);
2473 origin
->previous
= porigin
;
2475 pass_blame_to_parent(sb
, origin
, porigin
, 0);
2476 if (!origin
->suspects
)
2481 * Pass remaining suspects for ignored commits to their parents.
2483 if (oidset_contains(&sb
->ignore_list
, &commit
->object
.oid
)) {
2484 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2486 sg
= sg
->next
, i
++) {
2487 struct blame_origin
*porigin
= sg_origin
[i
];
2491 pass_blame_to_parent(sb
, origin
, porigin
, 1);
2493 * Preemptively drop porigin so we can refresh the
2494 * fingerprints if we use the parent again, which can
2495 * occur if you ignore back-to-back commits.
2497 drop_origin_blob(porigin
);
2498 if (!origin
->suspects
)
2504 * Optionally find moves in parents' files.
2506 if (opt
& PICKAXE_BLAME_MOVE
) {
2507 filter_small(sb
, &toosmall
, &origin
->suspects
, sb
->move_score
);
2508 if (origin
->suspects
) {
2509 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2511 sg
= sg
->next
, i
++) {
2512 struct blame_origin
*porigin
= sg_origin
[i
];
2515 find_move_in_parent(sb
, &blametail
, &toosmall
, origin
, porigin
);
2516 if (!origin
->suspects
)
2523 * Optionally find copies from parents' files.
2525 if (opt
& PICKAXE_BLAME_COPY
) {
2526 if (sb
->copy_score
> sb
->move_score
)
2527 filter_small(sb
, &toosmall
, &origin
->suspects
, sb
->copy_score
);
2528 else if (sb
->copy_score
< sb
->move_score
) {
2529 origin
->suspects
= blame_merge(origin
->suspects
, toosmall
);
2531 filter_small(sb
, &toosmall
, &origin
->suspects
, sb
->copy_score
);
2533 if (!origin
->suspects
)
2536 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2538 sg
= sg
->next
, i
++) {
2539 struct blame_origin
*porigin
= sg_origin
[i
];
2540 find_copy_in_parent(sb
, &blametail
, &toosmall
,
2541 origin
, sg
->item
, porigin
, opt
);
2542 if (!origin
->suspects
)
2549 distribute_blame(sb
, blames
);
2551 * prepend toosmall to origin->suspects
2553 * There is no point in sorting: this ends up on a big
2554 * unsorted list in the caller anyway.
2557 struct blame_entry
**tail
= &toosmall
;
2559 tail
= &(*tail
)->next
;
2560 *tail
= origin
->suspects
;
2561 origin
->suspects
= toosmall
;
2563 for (i
= 0; i
< num_sg
; i
++) {
2565 if (!sg_origin
[i
]->suspects
)
2566 drop_origin_blob(sg_origin
[i
]);
2567 blame_origin_decref(sg_origin
[i
]);
2570 drop_origin_blob(origin
);
2571 if (sg_buf
!= sg_origin
)
2576 * The main loop -- while we have blobs with lines whose true origin
2577 * is still unknown, pick one blob, and allow its lines to pass blames
2578 * to its parents. */
2579 void assign_blame(struct blame_scoreboard
*sb
, int opt
)
2581 struct rev_info
*revs
= sb
->revs
;
2582 struct commit
*commit
= prio_queue_get(&sb
->commits
);
2585 struct blame_entry
*ent
;
2586 struct blame_origin
*suspect
= get_blame_suspects(commit
);
2588 /* find one suspect to break down */
2589 while (suspect
&& !suspect
->suspects
)
2590 suspect
= suspect
->next
;
2593 commit
= prio_queue_get(&sb
->commits
);
2597 assert(commit
== suspect
->commit
);
2600 * We will use this suspect later in the loop,
2601 * so hold onto it in the meantime.
2603 blame_origin_incref(suspect
);
2604 repo_parse_commit(the_repository
, commit
);
2606 (!(commit
->object
.flags
& UNINTERESTING
) &&
2607 !(revs
->max_age
!= -1 && commit
->date
< revs
->max_age
)))
2608 pass_blame(sb
, suspect
, opt
);
2610 commit
->object
.flags
|= UNINTERESTING
;
2611 if (commit
->object
.parsed
)
2612 mark_parents_uninteresting(sb
->revs
, commit
);
2614 /* treat root commit as boundary */
2615 if (!commit
->parents
&& !sb
->show_root
)
2616 commit
->object
.flags
|= UNINTERESTING
;
2618 /* Take responsibility for the remaining entries */
2619 ent
= suspect
->suspects
;
2621 suspect
->guilty
= 1;
2623 struct blame_entry
*next
= ent
->next
;
2624 if (sb
->found_guilty_entry
)
2625 sb
->found_guilty_entry(ent
, sb
->found_guilty_entry_data
);
2630 ent
->next
= sb
->ent
;
2631 sb
->ent
= suspect
->suspects
;
2632 suspect
->suspects
= NULL
;
2636 blame_origin_decref(suspect
);
2638 if (sb
->debug
) /* sanity */
2639 sanity_check_refcnt(sb
);
2644 * To allow quick access to the contents of nth line in the
2645 * final image, prepare an index in the scoreboard.
2647 static int prepare_lines(struct blame_scoreboard
*sb
)
2649 sb
->num_lines
= find_line_starts(&sb
->lineno
, sb
->final_buf
,
2650 sb
->final_buf_size
);
2651 return sb
->num_lines
;
2654 static struct commit
*find_single_final(struct rev_info
*revs
,
2655 const char **name_p
)
2658 struct commit
*found
= NULL
;
2659 const char *name
= NULL
;
2661 for (i
= 0; i
< revs
->pending
.nr
; i
++) {
2662 struct object
*obj
= revs
->pending
.objects
[i
].item
;
2663 if (obj
->flags
& UNINTERESTING
)
2665 obj
= deref_tag(revs
->repo
, obj
, NULL
, 0);
2666 if (!obj
|| obj
->type
!= OBJ_COMMIT
)
2667 die("Non commit %s?", revs
->pending
.objects
[i
].name
);
2669 die("More than one commit to dig from %s and %s?",
2670 revs
->pending
.objects
[i
].name
, name
);
2671 found
= (struct commit
*)obj
;
2672 name
= revs
->pending
.objects
[i
].name
;
2675 *name_p
= xstrdup_or_null(name
);
2679 static struct commit
*dwim_reverse_initial(struct rev_info
*revs
,
2680 const char **name_p
)
2683 * DWIM "git blame --reverse ONE -- PATH" as
2684 * "git blame --reverse ONE..HEAD -- PATH" but only do so
2685 * when it makes sense.
2688 struct commit
*head_commit
;
2689 struct object_id head_oid
;
2691 if (revs
->pending
.nr
!= 1)
2694 /* Is that sole rev a committish? */
2695 obj
= revs
->pending
.objects
[0].item
;
2696 obj
= deref_tag(revs
->repo
, obj
, NULL
, 0);
2697 if (!obj
|| obj
->type
!= OBJ_COMMIT
)
2700 /* Do we have HEAD? */
2701 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING
, &head_oid
, NULL
))
2703 head_commit
= lookup_commit_reference_gently(revs
->repo
,
2708 /* Turn "ONE" into "ONE..HEAD" then */
2709 obj
->flags
|= UNINTERESTING
;
2710 add_pending_object(revs
, &head_commit
->object
, "HEAD");
2713 *name_p
= revs
->pending
.objects
[0].name
;
2714 return (struct commit
*)obj
;
2717 static struct commit
*find_single_initial(struct rev_info
*revs
,
2718 const char **name_p
)
2721 struct commit
*found
= NULL
;
2722 const char *name
= NULL
;
2725 * There must be one and only one negative commit, and it must be
2728 for (i
= 0; i
< revs
->pending
.nr
; i
++) {
2729 struct object
*obj
= revs
->pending
.objects
[i
].item
;
2730 if (!(obj
->flags
& UNINTERESTING
))
2732 obj
= deref_tag(revs
->repo
, obj
, NULL
, 0);
2733 if (!obj
|| obj
->type
!= OBJ_COMMIT
)
2734 die("Non commit %s?", revs
->pending
.objects
[i
].name
);
2736 die("More than one commit to dig up from, %s and %s?",
2737 revs
->pending
.objects
[i
].name
, name
);
2738 found
= (struct commit
*) obj
;
2739 name
= revs
->pending
.objects
[i
].name
;
2743 found
= dwim_reverse_initial(revs
, &name
);
2745 die("No commit to dig up from?");
2748 *name_p
= xstrdup(name
);
2752 void init_scoreboard(struct blame_scoreboard
*sb
)
2754 memset(sb
, 0, sizeof(struct blame_scoreboard
));
2755 sb
->move_score
= BLAME_DEFAULT_MOVE_SCORE
;
2756 sb
->copy_score
= BLAME_DEFAULT_COPY_SCORE
;
2759 void setup_scoreboard(struct blame_scoreboard
*sb
,
2760 struct blame_origin
**orig
)
2762 const char *final_commit_name
= NULL
;
2763 struct blame_origin
*o
;
2764 struct commit
*final_commit
= NULL
;
2765 enum object_type type
;
2767 init_blame_suspects(&blame_suspects
);
2769 if (sb
->reverse
&& sb
->contents_from
)
2770 die(_("--contents and --reverse do not blend well."));
2773 BUG("repo is NULL");
2776 sb
->final
= find_single_final(sb
->revs
, &final_commit_name
);
2777 sb
->commits
.compare
= compare_commits_by_commit_date
;
2779 sb
->final
= find_single_initial(sb
->revs
, &final_commit_name
);
2780 sb
->commits
.compare
= compare_commits_by_reverse_commit_date
;
2783 if (sb
->reverse
&& sb
->revs
->first_parent_only
)
2784 sb
->revs
->children
.name
= NULL
;
2786 if (sb
->contents_from
|| !sb
->final
) {
2787 struct object_id head_oid
, *parent_oid
;
2790 * Build a fake commit at the top of the history, when
2791 * (1) "git blame [^A] --path", i.e. with no positive end
2792 * of the history range, in which case we build such
2793 * a fake commit on top of the HEAD to blame in-tree
2795 * (2) "git blame --contents=file [A] -- path", with or
2796 * without positive end of the history range but with
2797 * --contents, in which case we pretend that there is
2798 * a fake commit on top of the positive end (defaulting to
2799 * HEAD) that has the given contents in the path.
2802 parent_oid
= &sb
->final
->object
.oid
;
2804 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING
, &head_oid
, NULL
))
2805 die("no such ref: HEAD");
2806 parent_oid
= &head_oid
;
2810 sb
->final
= fake_working_tree_commit(sb
->repo
,
2812 sb
->path
, sb
->contents_from
,
2814 add_pending_object(sb
->revs
, &(sb
->final
->object
), ":");
2817 if (sb
->reverse
&& sb
->revs
->first_parent_only
) {
2818 final_commit
= find_single_final(sb
->revs
, NULL
);
2820 die(_("--reverse and --first-parent together require specified latest commit"));
2824 * If we have bottom, this will mark the ancestors of the
2825 * bottom commits we would reach while traversing as
2828 if (prepare_revision_walk(sb
->revs
))
2829 die(_("revision walk setup failed"));
2831 if (sb
->reverse
&& sb
->revs
->first_parent_only
) {
2832 struct commit
*c
= final_commit
;
2834 sb
->revs
->children
.name
= "children";
2835 while (c
->parents
&&
2836 !oideq(&c
->object
.oid
, &sb
->final
->object
.oid
)) {
2837 struct commit_list
*l
= xcalloc(1, sizeof(*l
));
2840 if (add_decoration(&sb
->revs
->children
,
2841 &c
->parents
->item
->object
, l
))
2842 BUG("not unique item in first-parent chain");
2843 c
= c
->parents
->item
;
2846 if (!oideq(&c
->object
.oid
, &sb
->final
->object
.oid
))
2847 die(_("--reverse --first-parent together require range along first-parent chain"));
2850 if (is_null_oid(&sb
->final
->object
.oid
)) {
2851 o
= get_blame_suspects(sb
->final
);
2852 sb
->final_buf
= xmemdupz(o
->file
.ptr
, o
->file
.size
);
2853 sb
->final_buf_size
= o
->file
.size
;
2856 o
= get_origin(sb
->final
, sb
->path
);
2857 if (fill_blob_sha1_and_mode(sb
->repo
, o
))
2858 die(_("no such path %s in %s"), sb
->path
, final_commit_name
);
2860 if (sb
->revs
->diffopt
.flags
.allow_textconv
&&
2861 textconv_object(sb
->repo
, sb
->path
, o
->mode
, &o
->blob_oid
, 1, (char **) &sb
->final_buf
,
2862 &sb
->final_buf_size
))
2865 sb
->final_buf
= repo_read_object_file(the_repository
,
2868 &sb
->final_buf_size
);
2871 die(_("cannot read blob %s for path %s"),
2872 oid_to_hex(&o
->blob_oid
),
2875 sb
->num_read_blob
++;
2881 free((char *)final_commit_name
);
2886 struct blame_entry
*blame_entry_prepend(struct blame_entry
*head
,
2887 long start
, long end
,
2888 struct blame_origin
*o
)
2890 struct blame_entry
*new_head
= xcalloc(1, sizeof(struct blame_entry
));
2891 new_head
->lno
= start
;
2892 new_head
->num_lines
= end
- start
;
2893 new_head
->suspect
= o
;
2894 new_head
->s_lno
= start
;
2895 new_head
->next
= head
;
2896 blame_origin_incref(o
);
2900 void setup_blame_bloom_data(struct blame_scoreboard
*sb
)
2902 struct blame_bloom_data
*bd
;
2903 struct bloom_filter_settings
*bs
;
2905 if (!sb
->repo
->objects
->commit_graph
)
2908 bs
= get_bloom_filter_settings(sb
->repo
);
2912 bd
= xmalloc(sizeof(struct blame_bloom_data
));
2918 ALLOC_ARRAY(bd
->keys
, bd
->alloc
);
2920 add_bloom_key(bd
, sb
->path
);
2922 sb
->bloom_data
= bd
;
2925 void cleanup_scoreboard(struct blame_scoreboard
*sb
)
2927 if (sb
->bloom_data
) {
2929 for (i
= 0; i
< sb
->bloom_data
->nr
; i
++) {
2930 free(sb
->bloom_data
->keys
[i
]->hashes
);
2931 free(sb
->bloom_data
->keys
[i
]);
2933 free(sb
->bloom_data
->keys
);
2934 FREE_AND_NULL(sb
->bloom_data
);
2936 trace2_data_intmax("blame", sb
->repo
,
2937 "bloom/queries", bloom_count_queries
);
2938 trace2_data_intmax("blame", sb
->repo
,
2939 "bloom/response-no", bloom_count_no
);