3 #include "object-store.h"
4 #include "cache-tree.h"
15 #include "commit-slab.h"
17 #include "commit-graph.h"
19 define_commit_slab(blame_suspects
, struct blame_origin
*);
20 static struct blame_suspects blame_suspects
;
22 struct blame_origin
*get_blame_suspects(struct commit
*commit
)
24 struct blame_origin
**result
;
26 result
= blame_suspects_peek(&blame_suspects
, commit
);
28 return result
? *result
: NULL
;
31 static void set_blame_suspects(struct commit
*commit
, struct blame_origin
*origin
)
33 *blame_suspects_at(&blame_suspects
, commit
) = origin
;
36 void blame_origin_decref(struct blame_origin
*o
)
38 if (o
&& --o
->refcnt
<= 0) {
39 struct blame_origin
*p
, *l
= NULL
;
41 blame_origin_decref(o
->previous
);
43 /* Should be present exactly once in commit chain */
44 for (p
= get_blame_suspects(o
->commit
); p
; l
= p
, p
= p
->next
) {
49 set_blame_suspects(o
->commit
, p
->next
);
54 die("internal error in blame_origin_decref");
59 * Given a commit and a path in it, create a new origin structure.
60 * The callers that add blame to the scoreboard should use
61 * get_origin() to obtain shared, refcounted copy instead of calling
62 * this function directly.
64 static struct blame_origin
*make_origin(struct commit
*commit
, const char *path
)
66 struct blame_origin
*o
;
67 FLEX_ALLOC_STR(o
, path
, path
);
70 o
->next
= get_blame_suspects(commit
);
71 set_blame_suspects(commit
, o
);
76 * Locate an existing origin or create a new one.
77 * This moves the origin to front position in the commit util list.
79 static struct blame_origin
*get_origin(struct commit
*commit
, const char *path
)
81 struct blame_origin
*o
, *l
;
83 for (o
= get_blame_suspects(commit
), l
= NULL
; o
; l
= o
, o
= o
->next
) {
84 if (!strcmp(o
->path
, path
)) {
88 o
->next
= get_blame_suspects(commit
);
89 set_blame_suspects(commit
, o
);
91 return blame_origin_incref(o
);
94 return make_origin(commit
, path
);
99 static void verify_working_tree_path(struct repository
*r
,
100 struct commit
*work_tree
, const char *path
)
102 struct commit_list
*parents
;
105 for (parents
= work_tree
->parents
; parents
; parents
= parents
->next
) {
106 const struct object_id
*commit_oid
= &parents
->item
->object
.oid
;
107 struct object_id blob_oid
;
110 if (!get_tree_entry(r
, commit_oid
, path
, &blob_oid
, &mode
) &&
111 oid_object_info(r
, &blob_oid
, NULL
) == OBJ_BLOB
)
115 pos
= index_name_pos(r
->index
, path
, strlen(path
));
117 ; /* path is in the index */
118 else if (-1 - pos
< r
->index
->cache_nr
&&
119 !strcmp(r
->index
->cache
[-1 - pos
]->name
, path
))
120 ; /* path is in the index, unmerged */
122 die("no such path '%s' in HEAD", path
);
125 static struct commit_list
**append_parent(struct repository
*r
,
126 struct commit_list
**tail
,
127 const struct object_id
*oid
)
129 struct commit
*parent
;
131 parent
= lookup_commit_reference(r
, oid
);
133 die("no such commit %s", oid_to_hex(oid
));
134 return &commit_list_insert(parent
, tail
)->next
;
137 static void append_merge_parents(struct repository
*r
,
138 struct commit_list
**tail
)
141 struct strbuf line
= STRBUF_INIT
;
143 merge_head
= open(git_path_merge_head(r
), O_RDONLY
);
144 if (merge_head
< 0) {
147 die("cannot open '%s' for reading",
148 git_path_merge_head(r
));
151 while (!strbuf_getwholeline_fd(&line
, merge_head
, '\n')) {
152 struct object_id oid
;
153 if (get_oid_hex(line
.buf
, &oid
))
154 die("unknown line in '%s': %s",
155 git_path_merge_head(r
), line
.buf
);
156 tail
= append_parent(r
, tail
, &oid
);
159 strbuf_release(&line
);
163 * This isn't as simple as passing sb->buf and sb->len, because we
164 * want to transfer ownership of the buffer to the commit (so we
167 static void set_commit_buffer_from_strbuf(struct repository
*r
,
172 void *buf
= strbuf_detach(sb
, &len
);
173 set_commit_buffer(r
, c
, buf
, len
);
177 * Prepare a dummy commit that represents the work tree (or staged) item.
178 * Note that annotating work tree item never works in the reverse.
180 static struct commit
*fake_working_tree_commit(struct repository
*r
,
181 struct diff_options
*opt
,
183 const char *contents_from
)
185 struct commit
*commit
;
186 struct blame_origin
*origin
;
187 struct commit_list
**parent_tail
, *parent
;
188 struct object_id head_oid
;
189 struct strbuf buf
= STRBUF_INIT
;
193 struct cache_entry
*ce
;
195 struct strbuf msg
= STRBUF_INIT
;
199 commit
= alloc_commit_node(r
);
200 commit
->object
.parsed
= 1;
202 parent_tail
= &commit
->parents
;
204 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING
, &head_oid
, NULL
))
205 die("no such ref: HEAD");
207 parent_tail
= append_parent(r
, parent_tail
, &head_oid
);
208 append_merge_parents(r
, parent_tail
);
209 verify_working_tree_path(r
, commit
, path
);
211 origin
= make_origin(commit
, path
);
213 ident
= fmt_ident("Not Committed Yet", "not.committed.yet",
214 WANT_BLANK_IDENT
, NULL
, 0);
215 strbuf_addstr(&msg
, "tree 0000000000000000000000000000000000000000\n");
216 for (parent
= commit
->parents
; parent
; parent
= parent
->next
)
217 strbuf_addf(&msg
, "parent %s\n",
218 oid_to_hex(&parent
->item
->object
.oid
));
222 "Version of %s from %s\n",
224 (!contents_from
? path
:
225 (!strcmp(contents_from
, "-") ? "standard input" : contents_from
)));
226 set_commit_buffer_from_strbuf(r
, commit
, &msg
);
228 if (!contents_from
|| strcmp("-", contents_from
)) {
230 const char *read_from
;
232 unsigned long buf_len
;
235 if (stat(contents_from
, &st
) < 0)
236 die_errno("Cannot stat '%s'", contents_from
);
237 read_from
= contents_from
;
240 if (lstat(path
, &st
) < 0)
241 die_errno("Cannot lstat '%s'", path
);
244 mode
= canon_mode(st
.st_mode
);
246 switch (st
.st_mode
& S_IFMT
) {
248 if (opt
->flags
.allow_textconv
&&
249 textconv_object(r
, read_from
, mode
, null_oid(), 0, &buf_ptr
, &buf_len
))
250 strbuf_attach(&buf
, buf_ptr
, buf_len
, buf_len
+ 1);
251 else if (strbuf_read_file(&buf
, read_from
, st
.st_size
) != st
.st_size
)
252 die_errno("cannot open or read '%s'", read_from
);
255 if (strbuf_readlink(&buf
, read_from
, st
.st_size
) < 0)
256 die_errno("cannot readlink '%s'", read_from
);
259 die("unsupported file type %s", read_from
);
263 /* Reading from stdin */
265 if (strbuf_read(&buf
, 0, 0) < 0)
266 die_errno("failed to read from stdin");
268 convert_to_git(r
->index
, path
, buf
.buf
, buf
.len
, &buf
, 0);
269 origin
->file
.ptr
= buf
.buf
;
270 origin
->file
.size
= buf
.len
;
271 pretend_object_file(buf
.buf
, buf
.len
, OBJ_BLOB
, &origin
->blob_oid
);
274 * Read the current index, replace the path entry with
275 * origin->blob_sha1 without mucking with its mode or type
276 * bits; we are not going to write this index out -- we just
277 * want to run "diff-index --cached".
279 discard_index(r
->index
);
284 int pos
= index_name_pos(r
->index
, path
, len
);
286 mode
= r
->index
->cache
[pos
]->ce_mode
;
288 /* Let's not bother reading from HEAD tree */
289 mode
= S_IFREG
| 0644;
291 ce
= make_empty_cache_entry(r
->index
, len
);
292 oidcpy(&ce
->oid
, &origin
->blob_oid
);
293 memcpy(ce
->name
, path
, len
);
294 ce
->ce_flags
= create_ce_flags(0);
295 ce
->ce_namelen
= len
;
296 ce
->ce_mode
= create_ce_mode(mode
);
297 add_index_entry(r
->index
, ce
,
298 ADD_CACHE_OK_TO_ADD
| ADD_CACHE_OK_TO_REPLACE
);
300 cache_tree_invalidate_path(r
->index
, path
);
307 static int diff_hunks(mmfile_t
*file_a
, mmfile_t
*file_b
,
308 xdl_emit_hunk_consume_func_t hunk_func
, void *cb_data
, int xdl_opts
)
311 xdemitconf_t xecfg
= {0};
312 xdemitcb_t ecb
= {NULL
};
314 xpp
.flags
= xdl_opts
;
315 xecfg
.hunk_func
= hunk_func
;
317 return xdi_diff(file_a
, file_b
, &xpp
, &xecfg
, &ecb
);
320 static const char *get_next_line(const char *start
, const char *end
)
322 const char *nl
= memchr(start
, '\n', end
- start
);
324 return nl
? nl
+ 1 : end
;
327 static int find_line_starts(int **line_starts
, const char *buf
,
330 const char *end
= buf
+ len
;
335 for (p
= buf
; p
< end
; p
= get_next_line(p
, end
))
338 ALLOC_ARRAY(*line_starts
, num
+ 1);
339 lineno
= *line_starts
;
341 for (p
= buf
; p
< end
; p
= get_next_line(p
, end
))
349 struct fingerprint_entry
;
351 /* A fingerprint is intended to loosely represent a string, such that two
352 * fingerprints can be quickly compared to give an indication of the similarity
353 * of the strings that they represent.
355 * A fingerprint is represented as a multiset of the lower-cased byte pairs in
356 * the string that it represents. Whitespace is added at each end of the
357 * string. Whitespace pairs are ignored. Whitespace is converted to '\0'.
358 * For example, the string "Darth Radar" will be converted to the following
360 * {"\0d", "da", "da", "ar", "ar", "rt", "th", "h\0", "\0r", "ra", "ad", "r\0"}
362 * The similarity between two fingerprints is the size of the intersection of
363 * their multisets, including repeated elements. See fingerprint_similarity for
366 * For ease of implementation, the fingerprint is implemented as a map
367 * of byte pairs to the count of that byte pair in the string, instead of
368 * allowing repeated elements in a set.
372 /* As we know the maximum number of entries in advance, it's
373 * convenient to store the entries in a single array instead of having
374 * the hashmap manage the memory.
376 struct fingerprint_entry
*entries
;
379 /* A byte pair in a fingerprint. Stores the number of times the byte pair
380 * occurs in the string that the fingerprint represents.
382 struct fingerprint_entry
{
383 /* The hashmap entry - the hash represents the byte pair in its
384 * entirety so we don't need to store the byte pair separately.
386 struct hashmap_entry entry
;
387 /* The number of times the byte pair occurs in the string that the
388 * fingerprint represents.
393 /* See `struct fingerprint` for an explanation of what a fingerprint is.
394 * \param result the fingerprint of the string is stored here. This must be
395 * freed later using free_fingerprint.
396 * \param line_begin the start of the string
397 * \param line_end the end of the string
399 static void get_fingerprint(struct fingerprint
*result
,
400 const char *line_begin
,
401 const char *line_end
)
403 unsigned int hash
, c0
= 0, c1
;
405 int max_map_entry_count
= 1 + line_end
- line_begin
;
406 struct fingerprint_entry
*entry
= xcalloc(max_map_entry_count
,
407 sizeof(struct fingerprint_entry
));
408 struct fingerprint_entry
*found_entry
;
410 hashmap_init(&result
->map
, NULL
, NULL
, max_map_entry_count
);
411 result
->entries
= entry
;
412 for (p
= line_begin
; p
<= line_end
; ++p
, c0
= c1
) {
413 /* Always terminate the string with whitespace.
414 * Normalise whitespace to 0, and normalise letters to
415 * lower case. This won't work for multibyte characters but at
416 * worst will match some unrelated characters.
418 if ((p
== line_end
) || isspace(*p
))
422 hash
= c0
| (c1
<< 8);
423 /* Ignore whitespace pairs */
426 hashmap_entry_init(&entry
->entry
, hash
);
428 found_entry
= hashmap_get_entry(&result
->map
, entry
,
429 /* member name */ entry
, NULL
);
431 found_entry
->count
+= 1;
434 hashmap_add(&result
->map
, &entry
->entry
);
440 static void free_fingerprint(struct fingerprint
*f
)
442 hashmap_clear(&f
->map
);
446 /* Calculates the similarity between two fingerprints as the size of the
447 * intersection of their multisets, including repeated elements. See
448 * `struct fingerprint` for an explanation of the fingerprint representation.
449 * The similarity between "cat mat" and "father rather" is 2 because "at" is
450 * present twice in both strings while the similarity between "tim" and "mit"
453 static int fingerprint_similarity(struct fingerprint
*a
, struct fingerprint
*b
)
455 int intersection
= 0;
456 struct hashmap_iter iter
;
457 const struct fingerprint_entry
*entry_a
, *entry_b
;
459 hashmap_for_each_entry(&b
->map
, &iter
, entry_b
,
460 entry
/* member name */) {
461 entry_a
= hashmap_get_entry(&a
->map
, entry_b
, entry
, NULL
);
463 intersection
+= entry_a
->count
< entry_b
->count
?
464 entry_a
->count
: entry_b
->count
;
470 /* Subtracts byte-pair elements in B from A, modifying A in place.
472 static void fingerprint_subtract(struct fingerprint
*a
, struct fingerprint
*b
)
474 struct hashmap_iter iter
;
475 struct fingerprint_entry
*entry_a
;
476 const struct fingerprint_entry
*entry_b
;
478 hashmap_iter_init(&b
->map
, &iter
);
480 hashmap_for_each_entry(&b
->map
, &iter
, entry_b
,
481 entry
/* member name */) {
482 entry_a
= hashmap_get_entry(&a
->map
, entry_b
, entry
, NULL
);
484 if (entry_a
->count
<= entry_b
->count
)
485 hashmap_remove(&a
->map
, &entry_b
->entry
, NULL
);
487 entry_a
->count
-= entry_b
->count
;
492 /* Calculate fingerprints for a series of lines.
493 * Puts the fingerprints in the fingerprints array, which must have been
494 * preallocated to allow storing line_count elements.
496 static void get_line_fingerprints(struct fingerprint
*fingerprints
,
497 const char *content
, const int *line_starts
,
498 long first_line
, long line_count
)
501 const char *linestart
, *lineend
;
503 line_starts
+= first_line
;
504 for (i
= 0; i
< line_count
; ++i
) {
505 linestart
= content
+ line_starts
[i
];
506 lineend
= content
+ line_starts
[i
+ 1];
507 get_fingerprint(fingerprints
+ i
, linestart
, lineend
);
511 static void free_line_fingerprints(struct fingerprint
*fingerprints
,
516 for (i
= 0; i
< nr_fingerprints
; i
++)
517 free_fingerprint(&fingerprints
[i
]);
520 /* This contains the data necessary to linearly map a line number in one half
521 * of a diff chunk to the line in the other half of the diff chunk that is
522 * closest in terms of its position as a fraction of the length of the chunk.
524 struct line_number_mapping
{
525 int destination_start
, destination_length
,
526 source_start
, source_length
;
529 /* Given a line number in one range, offset and scale it to map it onto the
531 * Essentially this mapping is a simple linear equation but the calculation is
532 * more complicated to allow performing it with integer operations.
533 * Another complication is that if a line could map onto many lines in the
534 * destination range then we want to choose the line at the center of those
536 * Example: if the chunk is 2 lines long in A and 10 lines long in B then the
537 * first 5 lines in B will map onto the first line in the A chunk, while the
538 * last 5 lines will all map onto the second line in the A chunk.
539 * Example: if the chunk is 10 lines long in A and 2 lines long in B then line
540 * 0 in B will map onto line 2 in A, and line 1 in B will map onto line 7 in A.
542 static int map_line_number(int line_number
,
543 const struct line_number_mapping
*mapping
)
545 return ((line_number
- mapping
->source_start
) * 2 + 1) *
546 mapping
->destination_length
/
547 (mapping
->source_length
* 2) +
548 mapping
->destination_start
;
551 /* Get a pointer to the element storing the similarity between a line in A
554 * The similarities are stored in a 2-dimensional array. Each "row" in the
555 * array contains the similarities for a line in B. The similarities stored in
556 * a row are the similarities between the line in B and the nearby lines in A.
557 * To keep the length of each row the same, it is padded out with values of -1
558 * where the search range extends beyond the lines in A.
559 * For example, if max_search_distance_a is 2 and the two sides of a diff chunk
566 * Then the similarity array will contain:
567 * [-1, -1, am, bm, cm,
568 * -1, an, bn, cn, dn,
569 * ao, bo, co, do, eo,
570 * bp, cp, dp, ep, -1,
571 * cq, dq, eq, -1, -1]
572 * Where similarities are denoted either by -1 for invalid, or the
573 * concatenation of the two lines in the diff being compared.
575 * \param similarities array of similarities between lines in A and B
576 * \param line_a the index of the line in A, in the same frame of reference as
578 * \param local_line_b the index of the line in B, relative to the first line
579 * in B that similarities represents.
580 * \param closest_line_a the index of the line in A that is deemed to be
581 * closest to local_line_b. This must be in the same
582 * frame of reference as line_a. This value defines
583 * where similarities is centered for the line in B.
584 * \param max_search_distance_a maximum distance in lines from the closest line
585 * in A for other lines in A for which
586 * similarities may be calculated.
588 static int *get_similarity(int *similarities
,
589 int line_a
, int local_line_b
,
590 int closest_line_a
, int max_search_distance_a
)
592 assert(abs(line_a
- closest_line_a
) <=
593 max_search_distance_a
);
594 return similarities
+ line_a
- closest_line_a
+
595 max_search_distance_a
+
596 local_line_b
* (max_search_distance_a
* 2 + 1);
599 #define CERTAIN_NOTHING_MATCHES -2
600 #define CERTAINTY_NOT_CALCULATED -1
602 /* Given a line in B, first calculate its similarities with nearby lines in A
603 * if not already calculated, then identify the most similar and second most
604 * similar lines. The "certainty" is calculated based on those two
607 * \param start_a the index of the first line of the chunk in A
608 * \param length_a the length in lines of the chunk in A
609 * \param local_line_b the index of the line in B, relative to the first line
611 * \param fingerprints_a array of fingerprints for the chunk in A
612 * \param fingerprints_b array of fingerprints for the chunk in B
613 * \param similarities 2-dimensional array of similarities between lines in A
614 * and B. See get_similarity() for more details.
615 * \param certainties array of values indicating how strongly a line in B is
616 * matched with some line in A.
617 * \param second_best_result array of absolute indices in A for the second
618 * closest match of a line in B.
619 * \param result array of absolute indices in A for the closest match of a line
621 * \param max_search_distance_a maximum distance in lines from the closest line
622 * in A for other lines in A for which
623 * similarities may be calculated.
624 * \param map_line_number_in_b_to_a parameter to map_line_number().
626 static void find_best_line_matches(
631 struct fingerprint
*fingerprints_a
,
632 struct fingerprint
*fingerprints_b
,
635 int *second_best_result
,
637 const int max_search_distance_a
,
638 const struct line_number_mapping
*map_line_number_in_b_to_a
)
641 int i
, search_start
, search_end
, closest_local_line_a
, *similarity
,
642 best_similarity
= 0, second_best_similarity
= 0,
643 best_similarity_index
= 0, second_best_similarity_index
= 0;
645 /* certainty has already been calculated so no need to redo the work */
646 if (certainties
[local_line_b
] != CERTAINTY_NOT_CALCULATED
)
649 closest_local_line_a
= map_line_number(
650 local_line_b
+ start_b
, map_line_number_in_b_to_a
) - start_a
;
652 search_start
= closest_local_line_a
- max_search_distance_a
;
653 if (search_start
< 0)
656 search_end
= closest_local_line_a
+ max_search_distance_a
+ 1;
657 if (search_end
> length_a
)
658 search_end
= length_a
;
660 for (i
= search_start
; i
< search_end
; ++i
) {
661 similarity
= get_similarity(similarities
,
663 closest_local_line_a
,
664 max_search_distance_a
);
665 if (*similarity
== -1) {
666 /* This value will never exceed 10 but assert just in
669 assert(abs(i
- closest_local_line_a
) < 1000);
670 /* scale the similarity by (1000 - distance from
671 * closest line) to act as a tie break between lines
672 * that otherwise are equally similar.
674 *similarity
= fingerprint_similarity(
675 fingerprints_b
+ local_line_b
,
676 fingerprints_a
+ i
) *
677 (1000 - abs(i
- closest_local_line_a
));
679 if (*similarity
> best_similarity
) {
680 second_best_similarity
= best_similarity
;
681 second_best_similarity_index
= best_similarity_index
;
682 best_similarity
= *similarity
;
683 best_similarity_index
= i
;
684 } else if (*similarity
> second_best_similarity
) {
685 second_best_similarity
= *similarity
;
686 second_best_similarity_index
= i
;
690 if (best_similarity
== 0) {
691 /* this line definitely doesn't match with anything. Mark it
692 * with this special value so it doesn't get invalidated and
693 * won't be recalculated.
695 certainties
[local_line_b
] = CERTAIN_NOTHING_MATCHES
;
696 result
[local_line_b
] = -1;
698 /* Calculate the certainty with which this line matches.
699 * If the line matches well with two lines then that reduces
700 * the certainty. However we still want to prioritise matching
701 * a line that matches very well with two lines over matching a
702 * line that matches poorly with one line, hence doubling
704 * This means that if we have
705 * line X that matches only one line with a score of 3,
706 * line Y that matches two lines equally with a score of 5,
707 * and line Z that matches only one line with a score or 2,
708 * then the lines in order of certainty are X, Y, Z.
710 certainties
[local_line_b
] = best_similarity
* 2 -
711 second_best_similarity
;
713 /* We keep both the best and second best results to allow us to
714 * check at a later stage of the matching process whether the
715 * result needs to be invalidated.
717 result
[local_line_b
] = start_a
+ best_similarity_index
;
718 second_best_result
[local_line_b
] =
719 start_a
+ second_best_similarity_index
;
724 * This finds the line that we can match with the most confidence, and
725 * uses it as a partition. It then calls itself on the lines on either side of
726 * that partition. In this way we avoid lines appearing out of order, and
727 * retain a sensible line ordering.
728 * \param start_a index of the first line in A with which lines in B may be
730 * \param start_b index of the first line in B for which matching should be
732 * \param length_a number of lines in A with which lines in B may be compared.
733 * \param length_b number of lines in B for which matching should be done.
734 * \param fingerprints_a mutable array of fingerprints in A. The first element
735 * corresponds to the line at start_a.
736 * \param fingerprints_b array of fingerprints in B. The first element
737 * corresponds to the line at start_b.
738 * \param similarities 2-dimensional array of similarities between lines in A
739 * and B. See get_similarity() for more details.
740 * \param certainties array of values indicating how strongly a line in B is
741 * matched with some line in A.
742 * \param second_best_result array of absolute indices in A for the second
743 * closest match of a line in B.
744 * \param result array of absolute indices in A for the closest match of a line
746 * \param max_search_distance_a maximum distance in lines from the closest line
747 * in A for other lines in A for which
748 * similarities may be calculated.
749 * \param max_search_distance_b an upper bound on the greatest possible
750 * distance between lines in B such that they will
751 * both be compared with the same line in A
752 * according to max_search_distance_a.
753 * \param map_line_number_in_b_to_a parameter to map_line_number().
755 static void fuzzy_find_matching_lines_recurse(
756 int start_a
, int start_b
,
757 int length_a
, int length_b
,
758 struct fingerprint
*fingerprints_a
,
759 struct fingerprint
*fingerprints_b
,
762 int *second_best_result
,
764 int max_search_distance_a
,
765 int max_search_distance_b
,
766 const struct line_number_mapping
*map_line_number_in_b_to_a
)
768 int i
, invalidate_min
, invalidate_max
, offset_b
,
769 second_half_start_a
, second_half_start_b
,
770 second_half_length_a
, second_half_length_b
,
771 most_certain_line_a
, most_certain_local_line_b
= -1,
772 most_certain_line_certainty
= -1,
773 closest_local_line_a
;
775 for (i
= 0; i
< length_b
; ++i
) {
776 find_best_line_matches(start_a
,
786 max_search_distance_a
,
787 map_line_number_in_b_to_a
);
789 if (certainties
[i
] > most_certain_line_certainty
) {
790 most_certain_line_certainty
= certainties
[i
];
791 most_certain_local_line_b
= i
;
796 if (most_certain_local_line_b
== -1)
799 most_certain_line_a
= result
[most_certain_local_line_b
];
802 * Subtract the most certain line's fingerprint in B from the matched
803 * fingerprint in A. This means that other lines in B can't also match
804 * the same parts of the line in A.
806 fingerprint_subtract(fingerprints_a
+ most_certain_line_a
- start_a
,
807 fingerprints_b
+ most_certain_local_line_b
);
809 /* Invalidate results that may be affected by the choice of most
812 invalidate_min
= most_certain_local_line_b
- max_search_distance_b
;
813 invalidate_max
= most_certain_local_line_b
+ max_search_distance_b
+ 1;
814 if (invalidate_min
< 0)
816 if (invalidate_max
> length_b
)
817 invalidate_max
= length_b
;
819 /* As the fingerprint in A has changed, discard previously calculated
820 * similarity values with that fingerprint.
822 for (i
= invalidate_min
; i
< invalidate_max
; ++i
) {
823 closest_local_line_a
= map_line_number(
824 i
+ start_b
, map_line_number_in_b_to_a
) - start_a
;
826 /* Check that the lines in A and B are close enough that there
827 * is a similarity value for them.
829 if (abs(most_certain_line_a
- start_a
- closest_local_line_a
) >
830 max_search_distance_a
) {
834 *get_similarity(similarities
, most_certain_line_a
- start_a
,
835 i
, closest_local_line_a
,
836 max_search_distance_a
) = -1;
839 /* More invalidating of results that may be affected by the choice of
841 * Discard the matches for lines in B that are currently matched with a
842 * line in A such that their ordering contradicts the ordering imposed
843 * by the choice of most certain line.
845 for (i
= most_certain_local_line_b
- 1; i
>= invalidate_min
; --i
) {
846 /* In this loop we discard results for lines in B that are
847 * before most-certain-line-B but are matched with a line in A
848 * that is after most-certain-line-A.
850 if (certainties
[i
] >= 0 &&
851 (result
[i
] >= most_certain_line_a
||
852 second_best_result
[i
] >= most_certain_line_a
)) {
853 certainties
[i
] = CERTAINTY_NOT_CALCULATED
;
856 for (i
= most_certain_local_line_b
+ 1; i
< invalidate_max
; ++i
) {
857 /* In this loop we discard results for lines in B that are
858 * after most-certain-line-B but are matched with a line in A
859 * that is before most-certain-line-A.
861 if (certainties
[i
] >= 0 &&
862 (result
[i
] <= most_certain_line_a
||
863 second_best_result
[i
] <= most_certain_line_a
)) {
864 certainties
[i
] = CERTAINTY_NOT_CALCULATED
;
868 /* Repeat the matching process for lines before the most certain line.
870 if (most_certain_local_line_b
> 0) {
871 fuzzy_find_matching_lines_recurse(
873 most_certain_line_a
+ 1 - start_a
,
874 most_certain_local_line_b
,
875 fingerprints_a
, fingerprints_b
, similarities
,
876 certainties
, second_best_result
, result
,
877 max_search_distance_a
,
878 max_search_distance_b
,
879 map_line_number_in_b_to_a
);
881 /* Repeat the matching process for lines after the most certain line.
883 if (most_certain_local_line_b
+ 1 < length_b
) {
884 second_half_start_a
= most_certain_line_a
;
885 offset_b
= most_certain_local_line_b
+ 1;
886 second_half_start_b
= start_b
+ offset_b
;
887 second_half_length_a
=
888 length_a
+ start_a
- second_half_start_a
;
889 second_half_length_b
=
890 length_b
+ start_b
- second_half_start_b
;
891 fuzzy_find_matching_lines_recurse(
892 second_half_start_a
, second_half_start_b
,
893 second_half_length_a
, second_half_length_b
,
894 fingerprints_a
+ second_half_start_a
- start_a
,
895 fingerprints_b
+ offset_b
,
897 offset_b
* (max_search_distance_a
* 2 + 1),
898 certainties
+ offset_b
,
899 second_best_result
+ offset_b
, result
+ offset_b
,
900 max_search_distance_a
,
901 max_search_distance_b
,
902 map_line_number_in_b_to_a
);
906 /* Find the lines in the parent line range that most closely match the lines in
907 * the target line range. This is accomplished by matching fingerprints in each
908 * blame_origin, and choosing the best matches that preserve the line ordering.
909 * See struct fingerprint for details of fingerprint matching, and
910 * fuzzy_find_matching_lines_recurse for details of preserving line ordering.
912 * The performance is believed to be O(n log n) in the typical case and O(n^2)
913 * in a pathological case, where n is the number of lines in the target range.
915 static int *fuzzy_find_matching_lines(struct blame_origin
*parent
,
916 struct blame_origin
*target
,
917 int tlno
, int parent_slno
, int same
,
920 /* We use the terminology "A" for the left hand side of the diff AKA
921 * parent, and "B" for the right hand side of the diff AKA target. */
922 int start_a
= parent_slno
;
923 int length_a
= parent_len
;
925 int length_b
= same
- tlno
;
927 struct line_number_mapping map_line_number_in_b_to_a
= {
928 start_a
, length_a
, start_b
, length_b
931 struct fingerprint
*fingerprints_a
= parent
->fingerprints
;
932 struct fingerprint
*fingerprints_b
= target
->fingerprints
;
934 int i
, *result
, *second_best_result
,
935 *certainties
, *similarities
, similarity_count
;
938 * max_search_distance_a means that given a line in B, compare it to
939 * the line in A that is closest to its position, and the lines in A
940 * that are no greater than max_search_distance_a lines away from the
943 * max_search_distance_b is an upper bound on the greatest possible
944 * distance between lines in B such that they will both be compared
945 * with the same line in A according to max_search_distance_a.
947 int max_search_distance_a
= 10, max_search_distance_b
;
952 if (max_search_distance_a
>= length_a
)
953 max_search_distance_a
= length_a
? length_a
- 1 : 0;
955 max_search_distance_b
= ((2 * max_search_distance_a
+ 1) * length_b
958 CALLOC_ARRAY(result
, length_b
);
959 CALLOC_ARRAY(second_best_result
, length_b
);
960 CALLOC_ARRAY(certainties
, length_b
);
962 /* See get_similarity() for details of similarities. */
963 similarity_count
= length_b
* (max_search_distance_a
* 2 + 1);
964 CALLOC_ARRAY(similarities
, similarity_count
);
966 for (i
= 0; i
< length_b
; ++i
) {
968 second_best_result
[i
] = -1;
969 certainties
[i
] = CERTAINTY_NOT_CALCULATED
;
972 for (i
= 0; i
< similarity_count
; ++i
)
973 similarities
[i
] = -1;
975 fuzzy_find_matching_lines_recurse(start_a
, start_b
,
977 fingerprints_a
+ start_a
,
978 fingerprints_b
+ start_b
,
983 max_search_distance_a
,
984 max_search_distance_b
,
985 &map_line_number_in_b_to_a
);
989 free(second_best_result
);
994 static void fill_origin_fingerprints(struct blame_origin
*o
)
1000 o
->num_lines
= find_line_starts(&line_starts
, o
->file
.ptr
,
1002 CALLOC_ARRAY(o
->fingerprints
, o
->num_lines
);
1003 get_line_fingerprints(o
->fingerprints
, o
->file
.ptr
, line_starts
,
1008 static void drop_origin_fingerprints(struct blame_origin
*o
)
1010 if (o
->fingerprints
) {
1011 free_line_fingerprints(o
->fingerprints
, o
->num_lines
);
1013 FREE_AND_NULL(o
->fingerprints
);
1018 * Given an origin, prepare mmfile_t structure to be used by the
1021 static void fill_origin_blob(struct diff_options
*opt
,
1022 struct blame_origin
*o
, mmfile_t
*file
,
1023 int *num_read_blob
, int fill_fingerprints
)
1026 enum object_type type
;
1027 unsigned long file_size
;
1030 if (opt
->flags
.allow_textconv
&&
1031 textconv_object(opt
->repo
, o
->path
, o
->mode
,
1032 &o
->blob_oid
, 1, &file
->ptr
, &file_size
))
1035 file
->ptr
= repo_read_object_file(the_repository
,
1036 &o
->blob_oid
, &type
,
1038 file
->size
= file_size
;
1041 die("Cannot read blob %s for path %s",
1042 oid_to_hex(&o
->blob_oid
),
1048 if (fill_fingerprints
)
1049 fill_origin_fingerprints(o
);
1052 static void drop_origin_blob(struct blame_origin
*o
)
1054 FREE_AND_NULL(o
->file
.ptr
);
1055 drop_origin_fingerprints(o
);
1059 * Any merge of blames happens on lists of blames that arrived via
1060 * different parents in a single suspect. In this case, we want to
1061 * sort according to the suspect line numbers as opposed to the final
1062 * image line numbers. The function body is somewhat longish because
1063 * it avoids unnecessary writes.
1066 static struct blame_entry
*blame_merge(struct blame_entry
*list1
,
1067 struct blame_entry
*list2
)
1069 struct blame_entry
*p1
= list1
, *p2
= list2
,
1077 if (p1
->s_lno
<= p2
->s_lno
) {
1080 if (!(p1
= *tail
)) {
1084 } while (p1
->s_lno
<= p2
->s_lno
);
1090 if (!(p2
= *tail
)) {
1094 } while (p1
->s_lno
> p2
->s_lno
);
1098 if (!(p1
= *tail
)) {
1102 } while (p1
->s_lno
<= p2
->s_lno
);
1106 DEFINE_LIST_SORT(static, sort_blame_entries
, struct blame_entry
, next
);
1109 * Final image line numbers are all different, so we don't need a
1110 * three-way comparison here.
1113 static int compare_blame_final(const struct blame_entry
*e1
,
1114 const struct blame_entry
*e2
)
1116 return e1
->lno
> e2
->lno
? 1 : -1;
1119 static int compare_blame_suspect(const struct blame_entry
*s1
,
1120 const struct blame_entry
*s2
)
1123 * to allow for collating suspects, we sort according to the
1124 * respective pointer value as the primary sorting criterion.
1125 * The actual relation is pretty unimportant as long as it
1126 * establishes a total order. Comparing as integers gives us
1129 if (s1
->suspect
!= s2
->suspect
)
1130 return (intptr_t)s1
->suspect
> (intptr_t)s2
->suspect
? 1 : -1;
1131 if (s1
->s_lno
== s2
->s_lno
)
1133 return s1
->s_lno
> s2
->s_lno
? 1 : -1;
1136 void blame_sort_final(struct blame_scoreboard
*sb
)
1138 sort_blame_entries(&sb
->ent
, compare_blame_final
);
1141 static int compare_commits_by_reverse_commit_date(const void *a
,
1145 return -compare_commits_by_commit_date(a
, b
, c
);
1149 * For debugging -- origin is refcounted, and this asserts that
1150 * we do not underflow.
1152 static void sanity_check_refcnt(struct blame_scoreboard
*sb
)
1155 struct blame_entry
*ent
;
1157 for (ent
= sb
->ent
; ent
; ent
= ent
->next
) {
1158 /* Nobody should have zero or negative refcnt */
1159 if (ent
->suspect
->refcnt
<= 0) {
1160 fprintf(stderr
, "%s in %s has negative refcnt %d\n",
1162 oid_to_hex(&ent
->suspect
->commit
->object
.oid
),
1163 ent
->suspect
->refcnt
);
1168 sb
->on_sanity_fail(sb
, baa
);
1172 * If two blame entries that are next to each other came from
1173 * contiguous lines in the same origin (i.e. <commit, path> pair),
1174 * merge them together.
1176 void blame_coalesce(struct blame_scoreboard
*sb
)
1178 struct blame_entry
*ent
, *next
;
1180 for (ent
= sb
->ent
; ent
&& (next
= ent
->next
); ent
= next
) {
1181 if (ent
->suspect
== next
->suspect
&&
1182 ent
->s_lno
+ ent
->num_lines
== next
->s_lno
&&
1183 ent
->lno
+ ent
->num_lines
== next
->lno
&&
1184 ent
->ignored
== next
->ignored
&&
1185 ent
->unblamable
== next
->unblamable
) {
1186 ent
->num_lines
+= next
->num_lines
;
1187 ent
->next
= next
->next
;
1188 blame_origin_decref(next
->suspect
);
1191 next
= ent
; /* again */
1195 if (sb
->debug
) /* sanity */
1196 sanity_check_refcnt(sb
);
1200 * Merge the given sorted list of blames into a preexisting origin.
1201 * If there were no previous blames to that commit, it is entered into
1202 * the commit priority queue of the score board.
1205 static void queue_blames(struct blame_scoreboard
*sb
, struct blame_origin
*porigin
,
1206 struct blame_entry
*sorted
)
1208 if (porigin
->suspects
)
1209 porigin
->suspects
= blame_merge(porigin
->suspects
, sorted
);
1211 struct blame_origin
*o
;
1212 for (o
= get_blame_suspects(porigin
->commit
); o
; o
= o
->next
) {
1214 porigin
->suspects
= sorted
;
1218 porigin
->suspects
= sorted
;
1219 prio_queue_put(&sb
->commits
, porigin
->commit
);
1224 * Fill the blob_sha1 field of an origin if it hasn't, so that later
1225 * call to fill_origin_blob() can use it to locate the data. blob_sha1
1226 * for an origin is also used to pass the blame for the entire file to
1227 * the parent to detect the case where a child's blob is identical to
1228 * that of its parent's.
1230 * This also fills origin->mode for corresponding tree path.
1232 static int fill_blob_sha1_and_mode(struct repository
*r
,
1233 struct blame_origin
*origin
)
1235 if (!is_null_oid(&origin
->blob_oid
))
1237 if (get_tree_entry(r
, &origin
->commit
->object
.oid
, origin
->path
, &origin
->blob_oid
, &origin
->mode
))
1239 if (oid_object_info(r
, &origin
->blob_oid
, NULL
) != OBJ_BLOB
)
1243 oidclr(&origin
->blob_oid
);
1244 origin
->mode
= S_IFINVALID
;
1248 struct blame_bloom_data
{
1250 * Changed-path Bloom filter keys. These can help prevent
1251 * computing diffs against first parents, but we need to
1252 * expand the list as code is moved or files are renamed.
1254 struct bloom_filter_settings
*settings
;
1255 struct bloom_key
**keys
;
1260 static int bloom_count_queries
= 0;
1261 static int bloom_count_no
= 0;
1262 static int maybe_changed_path(struct repository
*r
,
1263 struct blame_origin
*origin
,
1264 struct blame_bloom_data
*bd
)
1267 struct bloom_filter
*filter
;
1272 if (commit_graph_generation(origin
->commit
) == GENERATION_NUMBER_INFINITY
)
1275 filter
= get_bloom_filter(r
, origin
->commit
);
1280 bloom_count_queries
++;
1281 for (i
= 0; i
< bd
->nr
; i
++) {
1282 if (bloom_filter_contains(filter
,
1292 static void add_bloom_key(struct blame_bloom_data
*bd
,
1298 if (bd
->nr
>= bd
->alloc
) {
1300 REALLOC_ARRAY(bd
->keys
, bd
->alloc
);
1303 bd
->keys
[bd
->nr
] = xmalloc(sizeof(struct bloom_key
));
1304 fill_bloom_key(path
, strlen(path
), bd
->keys
[bd
->nr
], bd
->settings
);
1309 * We have an origin -- check if the same path exists in the
1310 * parent and return an origin structure to represent it.
1312 static struct blame_origin
*find_origin(struct repository
*r
,
1313 struct commit
*parent
,
1314 struct blame_origin
*origin
,
1315 struct blame_bloom_data
*bd
)
1317 struct blame_origin
*porigin
;
1318 struct diff_options diff_opts
;
1319 const char *paths
[2];
1321 /* First check any existing origins */
1322 for (porigin
= get_blame_suspects(parent
); porigin
; porigin
= porigin
->next
)
1323 if (!strcmp(porigin
->path
, origin
->path
)) {
1325 * The same path between origin and its parent
1326 * without renaming -- the most common case.
1328 return blame_origin_incref (porigin
);
1331 /* See if the origin->path is different between parent
1332 * and origin first. Most of the time they are the
1333 * same and diff-tree is fairly efficient about this.
1335 repo_diff_setup(r
, &diff_opts
);
1336 diff_opts
.flags
.recursive
= 1;
1337 diff_opts
.detect_rename
= 0;
1338 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
1339 paths
[0] = origin
->path
;
1342 parse_pathspec(&diff_opts
.pathspec
,
1343 PATHSPEC_ALL_MAGIC
& ~PATHSPEC_LITERAL
,
1344 PATHSPEC_LITERAL_PATH
, "", paths
);
1345 diff_setup_done(&diff_opts
);
1347 if (is_null_oid(&origin
->commit
->object
.oid
))
1348 do_diff_cache(get_commit_tree_oid(parent
), &diff_opts
);
1350 int compute_diff
= 1;
1351 if (origin
->commit
->parents
&&
1352 oideq(&parent
->object
.oid
,
1353 &origin
->commit
->parents
->item
->object
.oid
))
1354 compute_diff
= maybe_changed_path(r
, origin
, bd
);
1357 diff_tree_oid(get_commit_tree_oid(parent
),
1358 get_commit_tree_oid(origin
->commit
),
1361 diffcore_std(&diff_opts
);
1363 if (!diff_queued_diff
.nr
) {
1364 /* The path is the same as parent */
1365 porigin
= get_origin(parent
, origin
->path
);
1366 oidcpy(&porigin
->blob_oid
, &origin
->blob_oid
);
1367 porigin
->mode
= origin
->mode
;
1370 * Since origin->path is a pathspec, if the parent
1371 * commit had it as a directory, we will see a whole
1372 * bunch of deletion of files in the directory that we
1373 * do not care about.
1376 struct diff_filepair
*p
= NULL
;
1377 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
1379 p
= diff_queued_diff
.queue
[i
];
1380 name
= p
->one
->path
? p
->one
->path
: p
->two
->path
;
1381 if (!strcmp(name
, origin
->path
))
1385 die("internal error in blame::find_origin");
1386 switch (p
->status
) {
1388 die("internal error in blame::find_origin (%c)",
1391 porigin
= get_origin(parent
, origin
->path
);
1392 oidcpy(&porigin
->blob_oid
, &p
->one
->oid
);
1393 porigin
->mode
= p
->one
->mode
;
1397 /* Did not exist in parent, or type changed */
1401 diff_flush(&diff_opts
);
1406 * We have an origin -- find the path that corresponds to it in its
1407 * parent and return an origin structure to represent it.
1409 static struct blame_origin
*find_rename(struct repository
*r
,
1410 struct commit
*parent
,
1411 struct blame_origin
*origin
,
1412 struct blame_bloom_data
*bd
)
1414 struct blame_origin
*porigin
= NULL
;
1415 struct diff_options diff_opts
;
1418 repo_diff_setup(r
, &diff_opts
);
1419 diff_opts
.flags
.recursive
= 1;
1420 diff_opts
.detect_rename
= DIFF_DETECT_RENAME
;
1421 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
1422 diff_opts
.single_follow
= origin
->path
;
1423 diff_setup_done(&diff_opts
);
1425 if (is_null_oid(&origin
->commit
->object
.oid
))
1426 do_diff_cache(get_commit_tree_oid(parent
), &diff_opts
);
1428 diff_tree_oid(get_commit_tree_oid(parent
),
1429 get_commit_tree_oid(origin
->commit
),
1431 diffcore_std(&diff_opts
);
1433 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
1434 struct diff_filepair
*p
= diff_queued_diff
.queue
[i
];
1435 if ((p
->status
== 'R' || p
->status
== 'C') &&
1436 !strcmp(p
->two
->path
, origin
->path
)) {
1437 add_bloom_key(bd
, p
->one
->path
);
1438 porigin
= get_origin(parent
, p
->one
->path
);
1439 oidcpy(&porigin
->blob_oid
, &p
->one
->oid
);
1440 porigin
->mode
= p
->one
->mode
;
1444 diff_flush(&diff_opts
);
1449 * Append a new blame entry to a given output queue.
1451 static void add_blame_entry(struct blame_entry
***queue
,
1452 const struct blame_entry
*src
)
1454 struct blame_entry
*e
= xmalloc(sizeof(*e
));
1455 memcpy(e
, src
, sizeof(*e
));
1456 blame_origin_incref(e
->suspect
);
1464 * src typically is on-stack; we want to copy the information in it to
1465 * a malloced blame_entry that gets added to the given queue. The
1466 * origin of dst loses a refcnt.
1468 static void dup_entry(struct blame_entry
***queue
,
1469 struct blame_entry
*dst
, struct blame_entry
*src
)
1471 blame_origin_incref(src
->suspect
);
1472 blame_origin_decref(dst
->suspect
);
1473 memcpy(dst
, src
, sizeof(*src
));
1474 dst
->next
= **queue
;
1476 *queue
= &dst
->next
;
1479 const char *blame_nth_line(struct blame_scoreboard
*sb
, long lno
)
1481 return sb
->final_buf
+ sb
->lineno
[lno
];
1485 * It is known that lines between tlno to same came from parent, and e
1486 * has an overlap with that range. it also is known that parent's
1487 * line plno corresponds to e's line tlno.
1493 * <------------------>
1495 * Split e into potentially three parts; before this chunk, the chunk
1496 * to be blamed for the parent, and after that portion.
1498 static void split_overlap(struct blame_entry
*split
,
1499 struct blame_entry
*e
,
1500 int tlno
, int plno
, int same
,
1501 struct blame_origin
*parent
)
1505 memset(split
, 0, sizeof(struct blame_entry
[3]));
1507 for (i
= 0; i
< 3; i
++) {
1508 split
[i
].ignored
= e
->ignored
;
1509 split
[i
].unblamable
= e
->unblamable
;
1512 if (e
->s_lno
< tlno
) {
1513 /* there is a pre-chunk part not blamed on parent */
1514 split
[0].suspect
= blame_origin_incref(e
->suspect
);
1515 split
[0].lno
= e
->lno
;
1516 split
[0].s_lno
= e
->s_lno
;
1517 split
[0].num_lines
= tlno
- e
->s_lno
;
1518 split
[1].lno
= e
->lno
+ tlno
- e
->s_lno
;
1519 split
[1].s_lno
= plno
;
1522 split
[1].lno
= e
->lno
;
1523 split
[1].s_lno
= plno
+ (e
->s_lno
- tlno
);
1526 if (same
< e
->s_lno
+ e
->num_lines
) {
1527 /* there is a post-chunk part not blamed on parent */
1528 split
[2].suspect
= blame_origin_incref(e
->suspect
);
1529 split
[2].lno
= e
->lno
+ (same
- e
->s_lno
);
1530 split
[2].s_lno
= e
->s_lno
+ (same
- e
->s_lno
);
1531 split
[2].num_lines
= e
->s_lno
+ e
->num_lines
- same
;
1532 chunk_end_lno
= split
[2].lno
;
1535 chunk_end_lno
= e
->lno
+ e
->num_lines
;
1536 split
[1].num_lines
= chunk_end_lno
- split
[1].lno
;
1539 * if it turns out there is nothing to blame the parent for,
1540 * forget about the splitting. !split[1].suspect signals this.
1542 if (split
[1].num_lines
< 1)
1544 split
[1].suspect
= blame_origin_incref(parent
);
1548 * split_overlap() divided an existing blame e into up to three parts
1549 * in split. Any assigned blame is moved to queue to
1550 * reflect the split.
1552 static void split_blame(struct blame_entry
***blamed
,
1553 struct blame_entry
***unblamed
,
1554 struct blame_entry
*split
,
1555 struct blame_entry
*e
)
1557 if (split
[0].suspect
&& split
[2].suspect
) {
1558 /* The first part (reuse storage for the existing entry e) */
1559 dup_entry(unblamed
, e
, &split
[0]);
1561 /* The last part -- me */
1562 add_blame_entry(unblamed
, &split
[2]);
1564 /* ... and the middle part -- parent */
1565 add_blame_entry(blamed
, &split
[1]);
1567 else if (!split
[0].suspect
&& !split
[2].suspect
)
1569 * The parent covers the entire area; reuse storage for
1570 * e and replace it with the parent.
1572 dup_entry(blamed
, e
, &split
[1]);
1573 else if (split
[0].suspect
) {
1574 /* me and then parent */
1575 dup_entry(unblamed
, e
, &split
[0]);
1576 add_blame_entry(blamed
, &split
[1]);
1579 /* parent and then me */
1580 dup_entry(blamed
, e
, &split
[1]);
1581 add_blame_entry(unblamed
, &split
[2]);
1586 * After splitting the blame, the origins used by the
1587 * on-stack blame_entry should lose one refcnt each.
1589 static void decref_split(struct blame_entry
*split
)
1593 for (i
= 0; i
< 3; i
++)
1594 blame_origin_decref(split
[i
].suspect
);
1598 * reverse_blame reverses the list given in head, appending tail.
1599 * That allows us to build lists in reverse order, then reverse them
1600 * afterwards. This can be faster than building the list in proper
1601 * order right away. The reason is that building in proper order
1602 * requires writing a link in the _previous_ element, while building
1603 * in reverse order just requires placing the list head into the
1604 * _current_ element.
1607 static struct blame_entry
*reverse_blame(struct blame_entry
*head
,
1608 struct blame_entry
*tail
)
1611 struct blame_entry
*next
= head
->next
;
1620 * Splits a blame entry into two entries at 'len' lines. The original 'e'
1621 * consists of len lines, i.e. [e->lno, e->lno + len), and the second part,
1622 * which is returned, consists of the remainder: [e->lno + len, e->lno +
1623 * e->num_lines). The caller needs to sort out the reference counting for the
1624 * new entry's suspect.
1626 static struct blame_entry
*split_blame_at(struct blame_entry
*e
, int len
,
1627 struct blame_origin
*new_suspect
)
1629 struct blame_entry
*n
= xcalloc(1, sizeof(struct blame_entry
));
1631 n
->suspect
= new_suspect
;
1632 n
->ignored
= e
->ignored
;
1633 n
->unblamable
= e
->unblamable
;
1634 n
->lno
= e
->lno
+ len
;
1635 n
->s_lno
= e
->s_lno
+ len
;
1636 n
->num_lines
= e
->num_lines
- len
;
1642 struct blame_line_tracker
{
1647 static int are_lines_adjacent(struct blame_line_tracker
*first
,
1648 struct blame_line_tracker
*second
)
1650 return first
->is_parent
== second
->is_parent
&&
1651 first
->s_lno
+ 1 == second
->s_lno
;
1654 static int scan_parent_range(struct fingerprint
*p_fps
,
1655 struct fingerprint
*t_fps
, int t_idx
,
1656 int from
, int nr_lines
)
1659 #define FINGERPRINT_FILE_THRESHOLD 10
1660 int best_sim_val
= FINGERPRINT_FILE_THRESHOLD
;
1661 int best_sim_idx
= -1;
1663 for (p_idx
= from
; p_idx
< from
+ nr_lines
; p_idx
++) {
1664 sim
= fingerprint_similarity(&t_fps
[t_idx
], &p_fps
[p_idx
]);
1665 if (sim
< best_sim_val
)
1667 /* Break ties with the closest-to-target line number */
1668 if (sim
== best_sim_val
&& best_sim_idx
!= -1 &&
1669 abs(best_sim_idx
- t_idx
) < abs(p_idx
- t_idx
))
1672 best_sim_idx
= p_idx
;
1674 return best_sim_idx
;
1678 * The first pass checks the blame entry (from the target) against the parent's
1679 * diff chunk. If that fails for a line, the second pass tries to match that
1680 * line to any part of parent file. That catches cases where a change was
1681 * broken into two chunks by 'context.'
1683 static void guess_line_blames(struct blame_origin
*parent
,
1684 struct blame_origin
*target
,
1685 int tlno
, int offset
, int same
, int parent_len
,
1686 struct blame_line_tracker
*line_blames
)
1688 int i
, best_idx
, target_idx
;
1689 int parent_slno
= tlno
+ offset
;
1692 fuzzy_matches
= fuzzy_find_matching_lines(parent
, target
,
1693 tlno
, parent_slno
, same
,
1695 for (i
= 0; i
< same
- tlno
; i
++) {
1696 target_idx
= tlno
+ i
;
1697 if (fuzzy_matches
&& fuzzy_matches
[i
] >= 0) {
1698 best_idx
= fuzzy_matches
[i
];
1700 best_idx
= scan_parent_range(parent
->fingerprints
,
1701 target
->fingerprints
,
1705 if (best_idx
>= 0) {
1706 line_blames
[i
].is_parent
= 1;
1707 line_blames
[i
].s_lno
= best_idx
;
1709 line_blames
[i
].is_parent
= 0;
1710 line_blames
[i
].s_lno
= target_idx
;
1713 free(fuzzy_matches
);
1717 * This decides which parts of a blame entry go to the parent (added to the
1718 * ignoredp list) and which stay with the target (added to the diffp list). The
1719 * actual decision was made in a separate heuristic function, and those answers
1720 * for the lines in 'e' are in line_blames. This consumes e, essentially
1721 * putting it on a list.
1723 * Note that the blame entries on the ignoredp list are not necessarily sorted
1724 * with respect to the parent's line numbers yet.
1726 static void ignore_blame_entry(struct blame_entry
*e
,
1727 struct blame_origin
*parent
,
1728 struct blame_entry
**diffp
,
1729 struct blame_entry
**ignoredp
,
1730 struct blame_line_tracker
*line_blames
)
1732 int entry_len
, nr_lines
, i
;
1735 * We carve new entries off the front of e. Each entry comes from a
1736 * contiguous chunk of lines: adjacent lines from the same origin
1737 * (either the parent or the target).
1740 nr_lines
= e
->num_lines
; /* e changes in the loop */
1741 for (i
= 0; i
< nr_lines
; i
++) {
1742 struct blame_entry
*next
= NULL
;
1745 * We are often adjacent to the next line - only split the blame
1746 * entry when we have to.
1748 if (i
+ 1 < nr_lines
) {
1749 if (are_lines_adjacent(&line_blames
[i
],
1750 &line_blames
[i
+ 1])) {
1754 next
= split_blame_at(e
, entry_len
,
1755 blame_origin_incref(e
->suspect
));
1757 if (line_blames
[i
].is_parent
) {
1759 blame_origin_decref(e
->suspect
);
1760 e
->suspect
= blame_origin_incref(parent
);
1761 e
->s_lno
= line_blames
[i
- entry_len
+ 1].s_lno
;
1762 e
->next
= *ignoredp
;
1766 /* e->s_lno is already in the target's address space. */
1770 assert(e
->num_lines
== entry_len
);
1778 * Process one hunk from the patch between the current suspect for
1779 * blame_entry e and its parent. This first blames any unfinished
1780 * entries before the chunk (which is where target and parent start
1781 * differing) on the parent, and then splits blame entries at the
1782 * start and at the end of the difference region. Since use of -M and
1783 * -C options may lead to overlapping/duplicate source line number
1784 * ranges, all we can rely on from sorting/merging is the order of the
1785 * first suspect line number.
1787 * tlno: line number in the target where this chunk begins
1788 * same: line number in the target where this chunk ends
1789 * offset: add to tlno to get the chunk starting point in the parent
1790 * parent_len: number of lines in the parent chunk
1792 static void blame_chunk(struct blame_entry
***dstq
, struct blame_entry
***srcq
,
1793 int tlno
, int offset
, int same
, int parent_len
,
1794 struct blame_origin
*parent
,
1795 struct blame_origin
*target
, int ignore_diffs
)
1797 struct blame_entry
*e
= **srcq
;
1798 struct blame_entry
*samep
= NULL
, *diffp
= NULL
, *ignoredp
= NULL
;
1799 struct blame_line_tracker
*line_blames
= NULL
;
1801 while (e
&& e
->s_lno
< tlno
) {
1802 struct blame_entry
*next
= e
->next
;
1804 * current record starts before differing portion. If
1805 * it reaches into it, we need to split it up and
1806 * examine the second part separately.
1808 if (e
->s_lno
+ e
->num_lines
> tlno
) {
1809 /* Move second half to a new record */
1810 struct blame_entry
*n
;
1812 n
= split_blame_at(e
, tlno
- e
->s_lno
, e
->suspect
);
1813 /* Push new record to diffp */
1817 blame_origin_decref(e
->suspect
);
1818 /* Pass blame for everything before the differing
1819 * chunk to the parent */
1820 e
->suspect
= blame_origin_incref(parent
);
1827 * As we don't know how much of a common stretch after this
1828 * diff will occur, the currently blamed parts are all that we
1829 * can assign to the parent for now.
1833 **dstq
= reverse_blame(samep
, **dstq
);
1834 *dstq
= &samep
->next
;
1837 * Prepend the split off portions: everything after e starts
1838 * after the blameable portion.
1840 e
= reverse_blame(diffp
, e
);
1843 * Now retain records on the target while parts are different
1849 if (ignore_diffs
&& same
- tlno
> 0) {
1850 CALLOC_ARRAY(line_blames
, same
- tlno
);
1851 guess_line_blames(parent
, target
, tlno
, offset
, same
,
1852 parent_len
, line_blames
);
1855 while (e
&& e
->s_lno
< same
) {
1856 struct blame_entry
*next
= e
->next
;
1859 * If current record extends into sameness, need to split.
1861 if (e
->s_lno
+ e
->num_lines
> same
) {
1863 * Move second half to a new record to be
1864 * processed by later chunks
1866 struct blame_entry
*n
;
1868 n
= split_blame_at(e
, same
- e
->s_lno
,
1869 blame_origin_incref(e
->suspect
));
1870 /* Push new record to samep */
1875 ignore_blame_entry(e
, parent
, &diffp
, &ignoredp
,
1876 line_blames
+ e
->s_lno
- tlno
);
1886 * Note ignoredp is not sorted yet, and thus neither is dstq.
1887 * That list must be sorted before we queue_blames(). We defer
1888 * sorting until after all diff hunks are processed, so that
1889 * guess_line_blames() can pick *any* line in the parent. The
1890 * slight drawback is that we end up sorting all blame entries
1891 * passed to the parent, including those that are unrelated to
1892 * changes made by the ignored commit.
1894 **dstq
= reverse_blame(ignoredp
, **dstq
);
1895 *dstq
= &ignoredp
->next
;
1897 **srcq
= reverse_blame(diffp
, reverse_blame(samep
, e
));
1898 /* Move across elements that are in the unblamable portion */
1900 *srcq
= &diffp
->next
;
1903 struct blame_chunk_cb_data
{
1904 struct blame_origin
*parent
;
1905 struct blame_origin
*target
;
1908 struct blame_entry
**dstq
;
1909 struct blame_entry
**srcq
;
1912 /* diff chunks are from parent to target */
1913 static int blame_chunk_cb(long start_a
, long count_a
,
1914 long start_b
, long count_b
, void *data
)
1916 struct blame_chunk_cb_data
*d
= data
;
1917 if (start_a
- start_b
!= d
->offset
)
1918 die("internal error in blame::blame_chunk_cb");
1919 blame_chunk(&d
->dstq
, &d
->srcq
, start_b
, start_a
- start_b
,
1920 start_b
+ count_b
, count_a
, d
->parent
, d
->target
,
1922 d
->offset
= start_a
+ count_a
- (start_b
+ count_b
);
1927 * We are looking at the origin 'target' and aiming to pass blame
1928 * for the lines it is suspected to its parent. Run diff to find
1929 * which lines came from parent and pass blame for them.
1931 static void pass_blame_to_parent(struct blame_scoreboard
*sb
,
1932 struct blame_origin
*target
,
1933 struct blame_origin
*parent
, int ignore_diffs
)
1935 mmfile_t file_p
, file_o
;
1936 struct blame_chunk_cb_data d
;
1937 struct blame_entry
*newdest
= NULL
;
1939 if (!target
->suspects
)
1940 return; /* nothing remains for this target */
1945 d
.ignore_diffs
= ignore_diffs
;
1946 d
.dstq
= &newdest
; d
.srcq
= &target
->suspects
;
1948 fill_origin_blob(&sb
->revs
->diffopt
, parent
, &file_p
,
1949 &sb
->num_read_blob
, ignore_diffs
);
1950 fill_origin_blob(&sb
->revs
->diffopt
, target
, &file_o
,
1951 &sb
->num_read_blob
, ignore_diffs
);
1952 sb
->num_get_patch
++;
1954 if (diff_hunks(&file_p
, &file_o
, blame_chunk_cb
, &d
, sb
->xdl_opts
))
1955 die("unable to generate diff (%s -> %s)",
1956 oid_to_hex(&parent
->commit
->object
.oid
),
1957 oid_to_hex(&target
->commit
->object
.oid
));
1958 /* The rest are the same as the parent */
1959 blame_chunk(&d
.dstq
, &d
.srcq
, INT_MAX
, d
.offset
, INT_MAX
, 0,
1963 sort_blame_entries(&newdest
, compare_blame_suspect
);
1964 queue_blames(sb
, parent
, newdest
);
1970 * The lines in blame_entry after splitting blames many times can become
1971 * very small and trivial, and at some point it becomes pointless to
1972 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
1973 * ordinary C program, and it is not worth to say it was copied from
1974 * totally unrelated file in the parent.
1976 * Compute how trivial the lines in the blame_entry are.
1978 unsigned blame_entry_score(struct blame_scoreboard
*sb
, struct blame_entry
*e
)
1981 const char *cp
, *ep
;
1987 cp
= blame_nth_line(sb
, e
->lno
);
1988 ep
= blame_nth_line(sb
, e
->lno
+ e
->num_lines
);
1990 unsigned ch
= *((unsigned char *)cp
);
2000 * best_so_far[] and potential[] are both a split of an existing blame_entry
2001 * that passes blame to the parent. Maintain best_so_far the best split so
2002 * far, by comparing potential and best_so_far and copying potential into
2003 * bst_so_far as needed.
2005 static void copy_split_if_better(struct blame_scoreboard
*sb
,
2006 struct blame_entry
*best_so_far
,
2007 struct blame_entry
*potential
)
2011 if (!potential
[1].suspect
)
2013 if (best_so_far
[1].suspect
) {
2014 if (blame_entry_score(sb
, &potential
[1]) <
2015 blame_entry_score(sb
, &best_so_far
[1]))
2019 for (i
= 0; i
< 3; i
++)
2020 blame_origin_incref(potential
[i
].suspect
);
2021 decref_split(best_so_far
);
2022 memcpy(best_so_far
, potential
, sizeof(struct blame_entry
[3]));
2026 * We are looking at a part of the final image represented by
2027 * ent (tlno and same are offset by ent->s_lno).
2028 * tlno is where we are looking at in the final image.
2029 * up to (but not including) same match preimage.
2030 * plno is where we are looking at in the preimage.
2032 * <-------------- final image ---------------------->
2035 * <---------preimage----->
2038 * All line numbers are 0-based.
2040 static void handle_split(struct blame_scoreboard
*sb
,
2041 struct blame_entry
*ent
,
2042 int tlno
, int plno
, int same
,
2043 struct blame_origin
*parent
,
2044 struct blame_entry
*split
)
2046 if (ent
->num_lines
<= tlno
)
2049 struct blame_entry potential
[3];
2052 split_overlap(potential
, ent
, tlno
, plno
, same
, parent
);
2053 copy_split_if_better(sb
, split
, potential
);
2054 decref_split(potential
);
2058 struct handle_split_cb_data
{
2059 struct blame_scoreboard
*sb
;
2060 struct blame_entry
*ent
;
2061 struct blame_origin
*parent
;
2062 struct blame_entry
*split
;
2067 static int handle_split_cb(long start_a
, long count_a
,
2068 long start_b
, long count_b
, void *data
)
2070 struct handle_split_cb_data
*d
= data
;
2071 handle_split(d
->sb
, d
->ent
, d
->tlno
, d
->plno
, start_b
, d
->parent
,
2073 d
->plno
= start_a
+ count_a
;
2074 d
->tlno
= start_b
+ count_b
;
2079 * Find the lines from parent that are the same as ent so that
2080 * we can pass blames to it. file_p has the blob contents for
2083 static void find_copy_in_blob(struct blame_scoreboard
*sb
,
2084 struct blame_entry
*ent
,
2085 struct blame_origin
*parent
,
2086 struct blame_entry
*split
,
2091 struct handle_split_cb_data d
;
2093 memset(&d
, 0, sizeof(d
));
2094 d
.sb
= sb
; d
.ent
= ent
; d
.parent
= parent
; d
.split
= split
;
2096 * Prepare mmfile that contains only the lines in ent.
2098 cp
= blame_nth_line(sb
, ent
->lno
);
2099 file_o
.ptr
= (char *) cp
;
2100 file_o
.size
= blame_nth_line(sb
, ent
->lno
+ ent
->num_lines
) - cp
;
2103 * file_o is a part of final image we are annotating.
2104 * file_p partially may match that image.
2106 memset(split
, 0, sizeof(struct blame_entry
[3]));
2107 if (diff_hunks(file_p
, &file_o
, handle_split_cb
, &d
, sb
->xdl_opts
))
2108 die("unable to generate diff (%s)",
2109 oid_to_hex(&parent
->commit
->object
.oid
));
2110 /* remainder, if any, all match the preimage */
2111 handle_split(sb
, ent
, d
.tlno
, d
.plno
, ent
->num_lines
, parent
, split
);
2114 /* Move all blame entries from list *source that have a score smaller
2115 * than score_min to the front of list *small.
2116 * Returns a pointer to the link pointing to the old head of the small list.
2119 static struct blame_entry
**filter_small(struct blame_scoreboard
*sb
,
2120 struct blame_entry
**small
,
2121 struct blame_entry
**source
,
2124 struct blame_entry
*p
= *source
;
2125 struct blame_entry
*oldsmall
= *small
;
2127 if (blame_entry_score(sb
, p
) <= score_min
) {
2143 * See if lines currently target is suspected for can be attributed to
2146 static void find_move_in_parent(struct blame_scoreboard
*sb
,
2147 struct blame_entry
***blamed
,
2148 struct blame_entry
**toosmall
,
2149 struct blame_origin
*target
,
2150 struct blame_origin
*parent
)
2152 struct blame_entry
*e
, split
[3];
2153 struct blame_entry
*unblamed
= target
->suspects
;
2154 struct blame_entry
*leftover
= NULL
;
2158 return; /* nothing remains for this target */
2160 fill_origin_blob(&sb
->revs
->diffopt
, parent
, &file_p
,
2161 &sb
->num_read_blob
, 0);
2165 /* At each iteration, unblamed has a NULL-terminated list of
2166 * entries that have not yet been tested for blame. leftover
2167 * contains the reversed list of entries that have been tested
2168 * without being assignable to the parent.
2171 struct blame_entry
**unblamedtail
= &unblamed
;
2172 struct blame_entry
*next
;
2173 for (e
= unblamed
; e
; e
= next
) {
2175 find_copy_in_blob(sb
, e
, parent
, split
, &file_p
);
2176 if (split
[1].suspect
&&
2177 sb
->move_score
< blame_entry_score(sb
, &split
[1])) {
2178 split_blame(blamed
, &unblamedtail
, split
, e
);
2183 decref_split(split
);
2185 *unblamedtail
= NULL
;
2186 toosmall
= filter_small(sb
, toosmall
, &unblamed
, sb
->move_score
);
2188 target
->suspects
= reverse_blame(leftover
, NULL
);
2192 struct blame_entry
*ent
;
2193 struct blame_entry split
[3];
2197 * Count the number of entries the target is suspected for,
2198 * and prepare a list of entry and the best split.
2200 static struct blame_list
*setup_blame_list(struct blame_entry
*unblamed
,
2203 struct blame_entry
*e
;
2205 struct blame_list
*blame_list
= NULL
;
2207 for (e
= unblamed
, num_ents
= 0; e
; e
= e
->next
)
2210 CALLOC_ARRAY(blame_list
, num_ents
);
2211 for (e
= unblamed
, i
= 0; e
; e
= e
->next
)
2212 blame_list
[i
++].ent
= e
;
2214 *num_ents_p
= num_ents
;
2219 * For lines target is suspected for, see if we can find code movement
2220 * across file boundary from the parent commit. porigin is the path
2221 * in the parent we already tried.
2223 static void find_copy_in_parent(struct blame_scoreboard
*sb
,
2224 struct blame_entry
***blamed
,
2225 struct blame_entry
**toosmall
,
2226 struct blame_origin
*target
,
2227 struct commit
*parent
,
2228 struct blame_origin
*porigin
,
2231 struct diff_options diff_opts
;
2233 struct blame_list
*blame_list
;
2235 struct blame_entry
*unblamed
= target
->suspects
;
2236 struct blame_entry
*leftover
= NULL
;
2239 return; /* nothing remains for this target */
2241 repo_diff_setup(sb
->repo
, &diff_opts
);
2242 diff_opts
.flags
.recursive
= 1;
2243 diff_opts
.output_format
= DIFF_FORMAT_NO_OUTPUT
;
2245 diff_setup_done(&diff_opts
);
2247 /* Try "find copies harder" on new path if requested;
2248 * we do not want to use diffcore_rename() actually to
2249 * match things up; find_copies_harder is set only to
2250 * force diff_tree_oid() to feed all filepairs to diff_queue,
2251 * and this code needs to be after diff_setup_done(), which
2252 * usually makes find-copies-harder imply copy detection.
2254 if ((opt
& PICKAXE_BLAME_COPY_HARDEST
)
2255 || ((opt
& PICKAXE_BLAME_COPY_HARDER
)
2256 && (!porigin
|| strcmp(target
->path
, porigin
->path
))))
2257 diff_opts
.flags
.find_copies_harder
= 1;
2259 if (is_null_oid(&target
->commit
->object
.oid
))
2260 do_diff_cache(get_commit_tree_oid(parent
), &diff_opts
);
2262 diff_tree_oid(get_commit_tree_oid(parent
),
2263 get_commit_tree_oid(target
->commit
),
2266 if (!diff_opts
.flags
.find_copies_harder
)
2267 diffcore_std(&diff_opts
);
2270 struct blame_entry
**unblamedtail
= &unblamed
;
2271 blame_list
= setup_blame_list(unblamed
, &num_ents
);
2273 for (i
= 0; i
< diff_queued_diff
.nr
; i
++) {
2274 struct diff_filepair
*p
= diff_queued_diff
.queue
[i
];
2275 struct blame_origin
*norigin
;
2277 struct blame_entry potential
[3];
2279 if (!DIFF_FILE_VALID(p
->one
))
2280 continue; /* does not exist in parent */
2281 if (S_ISGITLINK(p
->one
->mode
))
2282 continue; /* ignore git links */
2283 if (porigin
&& !strcmp(p
->one
->path
, porigin
->path
))
2284 /* find_move already dealt with this path */
2287 norigin
= get_origin(parent
, p
->one
->path
);
2288 oidcpy(&norigin
->blob_oid
, &p
->one
->oid
);
2289 norigin
->mode
= p
->one
->mode
;
2290 fill_origin_blob(&sb
->revs
->diffopt
, norigin
, &file_p
,
2291 &sb
->num_read_blob
, 0);
2295 for (j
= 0; j
< num_ents
; j
++) {
2296 find_copy_in_blob(sb
, blame_list
[j
].ent
,
2297 norigin
, potential
, &file_p
);
2298 copy_split_if_better(sb
, blame_list
[j
].split
,
2300 decref_split(potential
);
2302 blame_origin_decref(norigin
);
2305 for (j
= 0; j
< num_ents
; j
++) {
2306 struct blame_entry
*split
= blame_list
[j
].split
;
2307 if (split
[1].suspect
&&
2308 sb
->copy_score
< blame_entry_score(sb
, &split
[1])) {
2309 split_blame(blamed
, &unblamedtail
, split
,
2312 blame_list
[j
].ent
->next
= leftover
;
2313 leftover
= blame_list
[j
].ent
;
2315 decref_split(split
);
2318 *unblamedtail
= NULL
;
2319 toosmall
= filter_small(sb
, toosmall
, &unblamed
, sb
->copy_score
);
2321 target
->suspects
= reverse_blame(leftover
, NULL
);
2322 diff_flush(&diff_opts
);
2326 * The blobs of origin and porigin exactly match, so everything
2327 * origin is suspected for can be blamed on the parent.
2329 static void pass_whole_blame(struct blame_scoreboard
*sb
,
2330 struct blame_origin
*origin
, struct blame_origin
*porigin
)
2332 struct blame_entry
*e
, *suspects
;
2334 if (!porigin
->file
.ptr
&& origin
->file
.ptr
) {
2335 /* Steal its file */
2336 porigin
->file
= origin
->file
;
2337 origin
->file
.ptr
= NULL
;
2339 suspects
= origin
->suspects
;
2340 origin
->suspects
= NULL
;
2341 for (e
= suspects
; e
; e
= e
->next
) {
2342 blame_origin_incref(porigin
);
2343 blame_origin_decref(e
->suspect
);
2344 e
->suspect
= porigin
;
2346 queue_blames(sb
, porigin
, suspects
);
2350 * We pass blame from the current commit to its parents. We keep saying
2351 * "parent" (and "porigin"), but what we mean is to find scapegoat to
2352 * exonerate ourselves.
2354 static struct commit_list
*first_scapegoat(struct rev_info
*revs
, struct commit
*commit
,
2358 if (revs
->first_parent_only
&&
2360 commit
->parents
->next
) {
2361 free_commit_list(commit
->parents
->next
);
2362 commit
->parents
->next
= NULL
;
2364 return commit
->parents
;
2366 return lookup_decoration(&revs
->children
, &commit
->object
);
2369 static int num_scapegoats(struct rev_info
*revs
, struct commit
*commit
, int reverse
)
2371 struct commit_list
*l
= first_scapegoat(revs
, commit
, reverse
);
2372 return commit_list_count(l
);
2375 /* Distribute collected unsorted blames to the respected sorted lists
2376 * in the various origins.
2378 static void distribute_blame(struct blame_scoreboard
*sb
, struct blame_entry
*blamed
)
2380 sort_blame_entries(&blamed
, compare_blame_suspect
);
2383 struct blame_origin
*porigin
= blamed
->suspect
;
2384 struct blame_entry
*suspects
= NULL
;
2386 struct blame_entry
*next
= blamed
->next
;
2387 blamed
->next
= suspects
;
2390 } while (blamed
&& blamed
->suspect
== porigin
);
2391 suspects
= reverse_blame(suspects
, NULL
);
2392 queue_blames(sb
, porigin
, suspects
);
2398 typedef struct blame_origin
*(*blame_find_alg
)(struct repository
*,
2400 struct blame_origin
*,
2401 struct blame_bloom_data
*);
2403 static void pass_blame(struct blame_scoreboard
*sb
, struct blame_origin
*origin
, int opt
)
2405 struct rev_info
*revs
= sb
->revs
;
2406 int i
, pass
, num_sg
;
2407 struct commit
*commit
= origin
->commit
;
2408 struct commit_list
*sg
;
2409 struct blame_origin
*sg_buf
[MAXSG
];
2410 struct blame_origin
*porigin
, **sg_origin
= sg_buf
;
2411 struct blame_entry
*toosmall
= NULL
;
2412 struct blame_entry
*blames
, **blametail
= &blames
;
2414 num_sg
= num_scapegoats(revs
, commit
, sb
->reverse
);
2417 else if (num_sg
< ARRAY_SIZE(sg_buf
))
2418 memset(sg_buf
, 0, sizeof(sg_buf
));
2420 CALLOC_ARRAY(sg_origin
, num_sg
);
2423 * The first pass looks for unrenamed path to optimize for
2424 * common cases, then we look for renames in the second pass.
2426 for (pass
= 0; pass
< 2 - sb
->no_whole_file_rename
; pass
++) {
2427 blame_find_alg find
= pass
? find_rename
: find_origin
;
2429 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2431 sg
= sg
->next
, i
++) {
2432 struct commit
*p
= sg
->item
;
2437 if (repo_parse_commit(the_repository
, p
))
2439 porigin
= find(sb
->repo
, p
, origin
, sb
->bloom_data
);
2442 if (oideq(&porigin
->blob_oid
, &origin
->blob_oid
)) {
2443 pass_whole_blame(sb
, origin
, porigin
);
2444 blame_origin_decref(porigin
);
2447 for (j
= same
= 0; j
< i
; j
++)
2449 oideq(&sg_origin
[j
]->blob_oid
, &porigin
->blob_oid
)) {
2454 sg_origin
[i
] = porigin
;
2456 blame_origin_decref(porigin
);
2461 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2463 sg
= sg
->next
, i
++) {
2464 struct blame_origin
*porigin
= sg_origin
[i
];
2467 if (!origin
->previous
) {
2468 blame_origin_incref(porigin
);
2469 origin
->previous
= porigin
;
2471 pass_blame_to_parent(sb
, origin
, porigin
, 0);
2472 if (!origin
->suspects
)
2477 * Pass remaining suspects for ignored commits to their parents.
2479 if (oidset_contains(&sb
->ignore_list
, &commit
->object
.oid
)) {
2480 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2482 sg
= sg
->next
, i
++) {
2483 struct blame_origin
*porigin
= sg_origin
[i
];
2487 pass_blame_to_parent(sb
, origin
, porigin
, 1);
2489 * Preemptively drop porigin so we can refresh the
2490 * fingerprints if we use the parent again, which can
2491 * occur if you ignore back-to-back commits.
2493 drop_origin_blob(porigin
);
2494 if (!origin
->suspects
)
2500 * Optionally find moves in parents' files.
2502 if (opt
& PICKAXE_BLAME_MOVE
) {
2503 filter_small(sb
, &toosmall
, &origin
->suspects
, sb
->move_score
);
2504 if (origin
->suspects
) {
2505 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2507 sg
= sg
->next
, i
++) {
2508 struct blame_origin
*porigin
= sg_origin
[i
];
2511 find_move_in_parent(sb
, &blametail
, &toosmall
, origin
, porigin
);
2512 if (!origin
->suspects
)
2519 * Optionally find copies from parents' files.
2521 if (opt
& PICKAXE_BLAME_COPY
) {
2522 if (sb
->copy_score
> sb
->move_score
)
2523 filter_small(sb
, &toosmall
, &origin
->suspects
, sb
->copy_score
);
2524 else if (sb
->copy_score
< sb
->move_score
) {
2525 origin
->suspects
= blame_merge(origin
->suspects
, toosmall
);
2527 filter_small(sb
, &toosmall
, &origin
->suspects
, sb
->copy_score
);
2529 if (!origin
->suspects
)
2532 for (i
= 0, sg
= first_scapegoat(revs
, commit
, sb
->reverse
);
2534 sg
= sg
->next
, i
++) {
2535 struct blame_origin
*porigin
= sg_origin
[i
];
2536 find_copy_in_parent(sb
, &blametail
, &toosmall
,
2537 origin
, sg
->item
, porigin
, opt
);
2538 if (!origin
->suspects
)
2545 distribute_blame(sb
, blames
);
2547 * prepend toosmall to origin->suspects
2549 * There is no point in sorting: this ends up on a big
2550 * unsorted list in the caller anyway.
2553 struct blame_entry
**tail
= &toosmall
;
2555 tail
= &(*tail
)->next
;
2556 *tail
= origin
->suspects
;
2557 origin
->suspects
= toosmall
;
2559 for (i
= 0; i
< num_sg
; i
++) {
2561 if (!sg_origin
[i
]->suspects
)
2562 drop_origin_blob(sg_origin
[i
]);
2563 blame_origin_decref(sg_origin
[i
]);
2566 drop_origin_blob(origin
);
2567 if (sg_buf
!= sg_origin
)
2572 * The main loop -- while we have blobs with lines whose true origin
2573 * is still unknown, pick one blob, and allow its lines to pass blames
2574 * to its parents. */
2575 void assign_blame(struct blame_scoreboard
*sb
, int opt
)
2577 struct rev_info
*revs
= sb
->revs
;
2578 struct commit
*commit
= prio_queue_get(&sb
->commits
);
2581 struct blame_entry
*ent
;
2582 struct blame_origin
*suspect
= get_blame_suspects(commit
);
2584 /* find one suspect to break down */
2585 while (suspect
&& !suspect
->suspects
)
2586 suspect
= suspect
->next
;
2589 commit
= prio_queue_get(&sb
->commits
);
2593 assert(commit
== suspect
->commit
);
2596 * We will use this suspect later in the loop,
2597 * so hold onto it in the meantime.
2599 blame_origin_incref(suspect
);
2600 repo_parse_commit(the_repository
, commit
);
2602 (!(commit
->object
.flags
& UNINTERESTING
) &&
2603 !(revs
->max_age
!= -1 && commit
->date
< revs
->max_age
)))
2604 pass_blame(sb
, suspect
, opt
);
2606 commit
->object
.flags
|= UNINTERESTING
;
2607 if (commit
->object
.parsed
)
2608 mark_parents_uninteresting(sb
->revs
, commit
);
2610 /* treat root commit as boundary */
2611 if (!commit
->parents
&& !sb
->show_root
)
2612 commit
->object
.flags
|= UNINTERESTING
;
2614 /* Take responsibility for the remaining entries */
2615 ent
= suspect
->suspects
;
2617 suspect
->guilty
= 1;
2619 struct blame_entry
*next
= ent
->next
;
2620 if (sb
->found_guilty_entry
)
2621 sb
->found_guilty_entry(ent
, sb
->found_guilty_entry_data
);
2626 ent
->next
= sb
->ent
;
2627 sb
->ent
= suspect
->suspects
;
2628 suspect
->suspects
= NULL
;
2632 blame_origin_decref(suspect
);
2634 if (sb
->debug
) /* sanity */
2635 sanity_check_refcnt(sb
);
2640 * To allow quick access to the contents of nth line in the
2641 * final image, prepare an index in the scoreboard.
2643 static int prepare_lines(struct blame_scoreboard
*sb
)
2645 sb
->num_lines
= find_line_starts(&sb
->lineno
, sb
->final_buf
,
2646 sb
->final_buf_size
);
2647 return sb
->num_lines
;
2650 static struct commit
*find_single_final(struct rev_info
*revs
,
2651 const char **name_p
)
2654 struct commit
*found
= NULL
;
2655 const char *name
= NULL
;
2657 for (i
= 0; i
< revs
->pending
.nr
; i
++) {
2658 struct object
*obj
= revs
->pending
.objects
[i
].item
;
2659 if (obj
->flags
& UNINTERESTING
)
2661 obj
= deref_tag(revs
->repo
, obj
, NULL
, 0);
2662 if (!obj
|| obj
->type
!= OBJ_COMMIT
)
2663 die("Non commit %s?", revs
->pending
.objects
[i
].name
);
2665 die("More than one commit to dig from %s and %s?",
2666 revs
->pending
.objects
[i
].name
, name
);
2667 found
= (struct commit
*)obj
;
2668 name
= revs
->pending
.objects
[i
].name
;
2671 *name_p
= xstrdup_or_null(name
);
2675 static struct commit
*dwim_reverse_initial(struct rev_info
*revs
,
2676 const char **name_p
)
2679 * DWIM "git blame --reverse ONE -- PATH" as
2680 * "git blame --reverse ONE..HEAD -- PATH" but only do so
2681 * when it makes sense.
2684 struct commit
*head_commit
;
2685 struct object_id head_oid
;
2687 if (revs
->pending
.nr
!= 1)
2690 /* Is that sole rev a committish? */
2691 obj
= revs
->pending
.objects
[0].item
;
2692 obj
= deref_tag(revs
->repo
, obj
, NULL
, 0);
2693 if (!obj
|| obj
->type
!= OBJ_COMMIT
)
2696 /* Do we have HEAD? */
2697 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING
, &head_oid
, NULL
))
2699 head_commit
= lookup_commit_reference_gently(revs
->repo
,
2704 /* Turn "ONE" into "ONE..HEAD" then */
2705 obj
->flags
|= UNINTERESTING
;
2706 add_pending_object(revs
, &head_commit
->object
, "HEAD");
2709 *name_p
= revs
->pending
.objects
[0].name
;
2710 return (struct commit
*)obj
;
2713 static struct commit
*find_single_initial(struct rev_info
*revs
,
2714 const char **name_p
)
2717 struct commit
*found
= NULL
;
2718 const char *name
= NULL
;
2721 * There must be one and only one negative commit, and it must be
2724 for (i
= 0; i
< revs
->pending
.nr
; i
++) {
2725 struct object
*obj
= revs
->pending
.objects
[i
].item
;
2726 if (!(obj
->flags
& UNINTERESTING
))
2728 obj
= deref_tag(revs
->repo
, obj
, NULL
, 0);
2729 if (!obj
|| obj
->type
!= OBJ_COMMIT
)
2730 die("Non commit %s?", revs
->pending
.objects
[i
].name
);
2732 die("More than one commit to dig up from, %s and %s?",
2733 revs
->pending
.objects
[i
].name
, name
);
2734 found
= (struct commit
*) obj
;
2735 name
= revs
->pending
.objects
[i
].name
;
2739 found
= dwim_reverse_initial(revs
, &name
);
2741 die("No commit to dig up from?");
2744 *name_p
= xstrdup(name
);
2748 void init_scoreboard(struct blame_scoreboard
*sb
)
2750 memset(sb
, 0, sizeof(struct blame_scoreboard
));
2751 sb
->move_score
= BLAME_DEFAULT_MOVE_SCORE
;
2752 sb
->copy_score
= BLAME_DEFAULT_COPY_SCORE
;
2755 void setup_scoreboard(struct blame_scoreboard
*sb
,
2756 struct blame_origin
**orig
)
2758 const char *final_commit_name
= NULL
;
2759 struct blame_origin
*o
;
2760 struct commit
*final_commit
= NULL
;
2761 enum object_type type
;
2763 init_blame_suspects(&blame_suspects
);
2765 if (sb
->reverse
&& sb
->contents_from
)
2766 die(_("--contents and --reverse do not blend well."));
2769 BUG("repo is NULL");
2772 sb
->final
= find_single_final(sb
->revs
, &final_commit_name
);
2773 sb
->commits
.compare
= compare_commits_by_commit_date
;
2775 sb
->final
= find_single_initial(sb
->revs
, &final_commit_name
);
2776 sb
->commits
.compare
= compare_commits_by_reverse_commit_date
;
2779 if (sb
->final
&& sb
->contents_from
)
2780 die(_("cannot use --contents with final commit object name"));
2782 if (sb
->reverse
&& sb
->revs
->first_parent_only
)
2783 sb
->revs
->children
.name
= NULL
;
2787 * "--not A B -- path" without anything positive;
2788 * do not default to HEAD, but use the working tree
2792 sb
->final
= fake_working_tree_commit(sb
->repo
,
2794 sb
->path
, sb
->contents_from
);
2795 add_pending_object(sb
->revs
, &(sb
->final
->object
), ":");
2798 if (sb
->reverse
&& sb
->revs
->first_parent_only
) {
2799 final_commit
= find_single_final(sb
->revs
, NULL
);
2801 die(_("--reverse and --first-parent together require specified latest commit"));
2805 * If we have bottom, this will mark the ancestors of the
2806 * bottom commits we would reach while traversing as
2809 if (prepare_revision_walk(sb
->revs
))
2810 die(_("revision walk setup failed"));
2812 if (sb
->reverse
&& sb
->revs
->first_parent_only
) {
2813 struct commit
*c
= final_commit
;
2815 sb
->revs
->children
.name
= "children";
2816 while (c
->parents
&&
2817 !oideq(&c
->object
.oid
, &sb
->final
->object
.oid
)) {
2818 struct commit_list
*l
= xcalloc(1, sizeof(*l
));
2821 if (add_decoration(&sb
->revs
->children
,
2822 &c
->parents
->item
->object
, l
))
2823 BUG("not unique item in first-parent chain");
2824 c
= c
->parents
->item
;
2827 if (!oideq(&c
->object
.oid
, &sb
->final
->object
.oid
))
2828 die(_("--reverse --first-parent together require range along first-parent chain"));
2831 if (is_null_oid(&sb
->final
->object
.oid
)) {
2832 o
= get_blame_suspects(sb
->final
);
2833 sb
->final_buf
= xmemdupz(o
->file
.ptr
, o
->file
.size
);
2834 sb
->final_buf_size
= o
->file
.size
;
2837 o
= get_origin(sb
->final
, sb
->path
);
2838 if (fill_blob_sha1_and_mode(sb
->repo
, o
))
2839 die(_("no such path %s in %s"), sb
->path
, final_commit_name
);
2841 if (sb
->revs
->diffopt
.flags
.allow_textconv
&&
2842 textconv_object(sb
->repo
, sb
->path
, o
->mode
, &o
->blob_oid
, 1, (char **) &sb
->final_buf
,
2843 &sb
->final_buf_size
))
2846 sb
->final_buf
= repo_read_object_file(the_repository
,
2849 &sb
->final_buf_size
);
2852 die(_("cannot read blob %s for path %s"),
2853 oid_to_hex(&o
->blob_oid
),
2856 sb
->num_read_blob
++;
2862 free((char *)final_commit_name
);
2867 struct blame_entry
*blame_entry_prepend(struct blame_entry
*head
,
2868 long start
, long end
,
2869 struct blame_origin
*o
)
2871 struct blame_entry
*new_head
= xcalloc(1, sizeof(struct blame_entry
));
2872 new_head
->lno
= start
;
2873 new_head
->num_lines
= end
- start
;
2874 new_head
->suspect
= o
;
2875 new_head
->s_lno
= start
;
2876 new_head
->next
= head
;
2877 blame_origin_incref(o
);
2881 void setup_blame_bloom_data(struct blame_scoreboard
*sb
)
2883 struct blame_bloom_data
*bd
;
2884 struct bloom_filter_settings
*bs
;
2886 if (!sb
->repo
->objects
->commit_graph
)
2889 bs
= get_bloom_filter_settings(sb
->repo
);
2893 bd
= xmalloc(sizeof(struct blame_bloom_data
));
2899 ALLOC_ARRAY(bd
->keys
, bd
->alloc
);
2901 add_bloom_key(bd
, sb
->path
);
2903 sb
->bloom_data
= bd
;
2906 void cleanup_scoreboard(struct blame_scoreboard
*sb
)
2908 if (sb
->bloom_data
) {
2910 for (i
= 0; i
< sb
->bloom_data
->nr
; i
++) {
2911 free(sb
->bloom_data
->keys
[i
]->hashes
);
2912 free(sb
->bloom_data
->keys
[i
]);
2914 free(sb
->bloom_data
->keys
);
2915 FREE_AND_NULL(sb
->bloom_data
);
2917 trace2_data_intmax("blame", sb
->repo
,
2918 "bloom/queries", bloom_count_queries
);
2919 trace2_data_intmax("blame", sb
->repo
,
2920 "bloom/response-no", bloom_count_no
);