3 #include "object-store.h"
4 #include "cache-tree.h"
11 #include "commit-slab.h"
13 #include "commit-graph.h"
15 define_commit_slab(blame_suspects, struct blame_origin *);
16 static struct blame_suspects blame_suspects;
18 struct blame_origin *get_blame_suspects(struct commit *commit)
20 struct blame_origin **result;
22 result = blame_suspects_peek(&blame_suspects, commit);
24 return result ? *result : NULL;
27 static void set_blame_suspects(struct commit *commit, struct blame_origin *origin)
29 *blame_suspects_at(&blame_suspects, commit) = origin;
32 void blame_origin_decref(struct blame_origin *o)
34 if (o && --o->refcnt <= 0) {
35 struct blame_origin *p, *l = NULL;
37 blame_origin_decref(o->previous);
39 /* Should be present exactly once in commit chain */
40 for (p = get_blame_suspects(o->commit); p; l = p, p = p->next) {
45 set_blame_suspects(o->commit, p->next);
50 die("internal error in blame_origin_decref");
55 * Given a commit and a path in it, create a new origin structure.
56 * The callers that add blame to the scoreboard should use
57 * get_origin() to obtain shared, refcounted copy instead of calling
58 * this function directly.
60 static struct blame_origin *make_origin(struct commit *commit, const char *path)
62 struct blame_origin *o;
63 FLEX_ALLOC_STR(o, path, path);
66 o->next = get_blame_suspects(commit);
67 set_blame_suspects(commit, o);
72 * Locate an existing origin or create a new one.
73 * This moves the origin to front position in the commit util list.
75 static struct blame_origin *get_origin(struct commit *commit, const char *path)
77 struct blame_origin *o, *l;
79 for (o = get_blame_suspects(commit), l = NULL; o; l = o, o = o->next) {
80 if (!strcmp(o->path, path)) {
84 o->next = get_blame_suspects(commit);
85 set_blame_suspects(commit, o);
87 return blame_origin_incref(o);
90 return make_origin(commit, path);
95 static void verify_working_tree_path(struct repository *r,
96 struct commit *work_tree, const char *path)
98 struct commit_list *parents;
101 for (parents = work_tree->parents; parents; parents = parents->next) {
102 const struct object_id *commit_oid = &parents->item->object.oid;
103 struct object_id blob_oid;
106 if (!get_tree_entry(r, commit_oid, path, &blob_oid, &mode) &&
107 oid_object_info(r, &blob_oid, NULL) == OBJ_BLOB)
111 pos = index_name_pos(r->index, path, strlen(path));
113 ; /* path is in the index */
114 else if (-1 - pos < r->index->cache_nr &&
115 !strcmp(r->index->cache[-1 - pos]->name, path))
116 ; /* path is in the index, unmerged */
118 die("no such path '%s' in HEAD", path);
121 static struct commit_list **append_parent(struct repository *r,
122 struct commit_list **tail,
123 const struct object_id *oid)
125 struct commit *parent;
127 parent = lookup_commit_reference(r, oid);
129 die("no such commit %s", oid_to_hex(oid));
130 return &commit_list_insert(parent, tail)->next;
133 static void append_merge_parents(struct repository *r,
134 struct commit_list **tail)
137 struct strbuf line = STRBUF_INIT;
139 merge_head = open(git_path_merge_head(r), O_RDONLY);
140 if (merge_head < 0) {
143 die("cannot open '%s' for reading",
144 git_path_merge_head(r));
147 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
148 struct object_id oid;
149 if (get_oid_hex(line.buf, &oid))
150 die("unknown line in '%s': %s",
151 git_path_merge_head(r), line.buf);
152 tail = append_parent(r, tail, &oid);
155 strbuf_release(&line);
159 * This isn't as simple as passing sb->buf and sb->len, because we
160 * want to transfer ownership of the buffer to the commit (so we
163 static void set_commit_buffer_from_strbuf(struct repository *r,
168 void *buf = strbuf_detach(sb, &len);
169 set_commit_buffer(r, c, buf, len);
173 * Prepare a dummy commit that represents the work tree (or staged) item.
174 * Note that annotating work tree item never works in the reverse.
176 static struct commit *fake_working_tree_commit(struct repository *r,
177 struct diff_options *opt,
179 const char *contents_from)
181 struct commit *commit;
182 struct blame_origin *origin;
183 struct commit_list **parent_tail, *parent;
184 struct object_id head_oid;
185 struct strbuf buf = STRBUF_INIT;
189 struct cache_entry *ce;
191 struct strbuf msg = STRBUF_INIT;
195 commit = alloc_commit_node(r);
196 commit->object.parsed = 1;
198 parent_tail = &commit->parents;
200 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
201 die("no such ref: HEAD");
203 parent_tail = append_parent(r, parent_tail, &head_oid);
204 append_merge_parents(r, parent_tail);
205 verify_working_tree_path(r, commit, path);
207 origin = make_origin(commit, path);
209 ident = fmt_ident("Not Committed Yet", "not.committed.yet",
210 WANT_BLANK_IDENT, NULL, 0);
211 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
212 for (parent = commit->parents; parent; parent = parent->next)
213 strbuf_addf(&msg, "parent %s\n",
214 oid_to_hex(&parent->item->object.oid));
218 "Version of %s from %s\n",
220 (!contents_from ? path :
221 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
222 set_commit_buffer_from_strbuf(r, commit, &msg);
224 if (!contents_from || strcmp("-", contents_from)) {
226 const char *read_from;
228 unsigned long buf_len;
231 if (stat(contents_from, &st) < 0)
232 die_errno("Cannot stat '%s'", contents_from);
233 read_from = contents_from;
236 if (lstat(path, &st) < 0)
237 die_errno("Cannot lstat '%s'", path);
240 mode = canon_mode(st.st_mode);
242 switch (st.st_mode & S_IFMT) {
244 if (opt->flags.allow_textconv &&
245 textconv_object(r, read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
246 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
247 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
248 die_errno("cannot open or read '%s'", read_from);
251 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
252 die_errno("cannot readlink '%s'", read_from);
255 die("unsupported file type %s", read_from);
259 /* Reading from stdin */
261 if (strbuf_read(&buf, 0, 0) < 0)
262 die_errno("failed to read from stdin");
264 convert_to_git(r->index, path, buf.buf, buf.len, &buf, 0);
265 origin->file.ptr = buf.buf;
266 origin->file.size = buf.len;
267 pretend_object_file(buf.buf, buf.len, OBJ_BLOB, &origin->blob_oid);
270 * Read the current index, replace the path entry with
271 * origin->blob_sha1 without mucking with its mode or type
272 * bits; we are not going to write this index out -- we just
273 * want to run "diff-index --cached".
275 discard_index(r->index);
280 int pos = index_name_pos(r->index, path, len);
282 mode = r->index->cache[pos]->ce_mode;
284 /* Let's not bother reading from HEAD tree */
285 mode = S_IFREG | 0644;
287 ce = make_empty_cache_entry(r->index, len);
288 oidcpy(&ce->oid, &origin->blob_oid);
289 memcpy(ce->name, path, len);
290 ce->ce_flags = create_ce_flags(0);
291 ce->ce_namelen = len;
292 ce->ce_mode = create_ce_mode(mode);
293 add_index_entry(r->index, ce,
294 ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
296 cache_tree_invalidate_path(r->index, path);
303 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
304 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
307 xdemitconf_t xecfg = {0};
308 xdemitcb_t ecb = {NULL};
310 xpp.flags = xdl_opts;
311 xecfg.hunk_func = hunk_func;
313 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
316 static const char *get_next_line(const char *start, const char *end)
318 const char *nl = memchr(start, '\n', end - start);
320 return nl ? nl + 1 : end;
323 static int find_line_starts(int **line_starts, const char *buf,
326 const char *end = buf + len;
331 for (p = buf; p < end; p = get_next_line(p, end))
334 ALLOC_ARRAY(*line_starts, num + 1);
335 lineno = *line_starts;
337 for (p = buf; p < end; p = get_next_line(p, end))
345 struct fingerprint_entry;
347 /* A fingerprint is intended to loosely represent a string, such that two
348 * fingerprints can be quickly compared to give an indication of the similarity
349 * of the strings that they represent.
351 * A fingerprint is represented as a multiset of the lower-cased byte pairs in
352 * the string that it represents. Whitespace is added at each end of the
353 * string. Whitespace pairs are ignored. Whitespace is converted to '\0'.
354 * For example, the string "Darth Radar" will be converted to the following
356 * {"\0d", "da", "da", "ar", "ar", "rt", "th", "h\0", "\0r", "ra", "ad", "r\0"}
358 * The similarity between two fingerprints is the size of the intersection of
359 * their multisets, including repeated elements. See fingerprint_similarity for
362 * For ease of implementation, the fingerprint is implemented as a map
363 * of byte pairs to the count of that byte pair in the string, instead of
364 * allowing repeated elements in a set.
368 /* As we know the maximum number of entries in advance, it's
369 * convenient to store the entries in a single array instead of having
370 * the hashmap manage the memory.
372 struct fingerprint_entry *entries;
375 /* A byte pair in a fingerprint. Stores the number of times the byte pair
376 * occurs in the string that the fingerprint represents.
378 struct fingerprint_entry {
379 /* The hashmap entry - the hash represents the byte pair in its
380 * entirety so we don't need to store the byte pair separately.
382 struct hashmap_entry entry;
383 /* The number of times the byte pair occurs in the string that the
384 * fingerprint represents.
389 /* See `struct fingerprint` for an explanation of what a fingerprint is.
390 * \param result the fingerprint of the string is stored here. This must be
391 * freed later using free_fingerprint.
392 * \param line_begin the start of the string
393 * \param line_end the end of the string
395 static void get_fingerprint(struct fingerprint *result,
396 const char *line_begin,
397 const char *line_end)
399 unsigned int hash, c0 = 0, c1;
401 int max_map_entry_count = 1 + line_end - line_begin;
402 struct fingerprint_entry *entry = xcalloc(max_map_entry_count,
403 sizeof(struct fingerprint_entry));
404 struct fingerprint_entry *found_entry;
406 hashmap_init(&result->map, NULL, NULL, max_map_entry_count);
407 result->entries = entry;
408 for (p = line_begin; p <= line_end; ++p, c0 = c1) {
409 /* Always terminate the string with whitespace.
410 * Normalise whitespace to 0, and normalise letters to
411 * lower case. This won't work for multibyte characters but at
412 * worst will match some unrelated characters.
414 if ((p == line_end) || isspace(*p))
418 hash = c0 | (c1 << 8);
419 /* Ignore whitespace pairs */
422 hashmap_entry_init(&entry->entry, hash);
424 found_entry = hashmap_get_entry(&result->map, entry,
425 /* member name */ entry, NULL);
427 found_entry->count += 1;
430 hashmap_add(&result->map, &entry->entry);
436 static void free_fingerprint(struct fingerprint *f)
438 hashmap_clear(&f->map);
442 /* Calculates the similarity between two fingerprints as the size of the
443 * intersection of their multisets, including repeated elements. See
444 * `struct fingerprint` for an explanation of the fingerprint representation.
445 * The similarity between "cat mat" and "father rather" is 2 because "at" is
446 * present twice in both strings while the similarity between "tim" and "mit"
449 static int fingerprint_similarity(struct fingerprint *a, struct fingerprint *b)
451 int intersection = 0;
452 struct hashmap_iter iter;
453 const struct fingerprint_entry *entry_a, *entry_b;
455 hashmap_for_each_entry(&b->map, &iter, entry_b,
456 entry /* member name */) {
457 entry_a = hashmap_get_entry(&a->map, entry_b, entry, NULL);
459 intersection += entry_a->count < entry_b->count ?
460 entry_a->count : entry_b->count;
466 /* Subtracts byte-pair elements in B from A, modifying A in place.
468 static void fingerprint_subtract(struct fingerprint *a, struct fingerprint *b)
470 struct hashmap_iter iter;
471 struct fingerprint_entry *entry_a;
472 const struct fingerprint_entry *entry_b;
474 hashmap_iter_init(&b->map, &iter);
476 hashmap_for_each_entry(&b->map, &iter, entry_b,
477 entry /* member name */) {
478 entry_a = hashmap_get_entry(&a->map, entry_b, entry, NULL);
480 if (entry_a->count <= entry_b->count)
481 hashmap_remove(&a->map, &entry_b->entry, NULL);
483 entry_a->count -= entry_b->count;
488 /* Calculate fingerprints for a series of lines.
489 * Puts the fingerprints in the fingerprints array, which must have been
490 * preallocated to allow storing line_count elements.
492 static void get_line_fingerprints(struct fingerprint *fingerprints,
493 const char *content, const int *line_starts,
494 long first_line, long line_count)
497 const char *linestart, *lineend;
499 line_starts += first_line;
500 for (i = 0; i < line_count; ++i) {
501 linestart = content + line_starts[i];
502 lineend = content + line_starts[i + 1];
503 get_fingerprint(fingerprints + i, linestart, lineend);
507 static void free_line_fingerprints(struct fingerprint *fingerprints,
512 for (i = 0; i < nr_fingerprints; i++)
513 free_fingerprint(&fingerprints[i]);
516 /* This contains the data necessary to linearly map a line number in one half
517 * of a diff chunk to the line in the other half of the diff chunk that is
518 * closest in terms of its position as a fraction of the length of the chunk.
520 struct line_number_mapping {
521 int destination_start, destination_length,
522 source_start, source_length;
525 /* Given a line number in one range, offset and scale it to map it onto the
527 * Essentially this mapping is a simple linear equation but the calculation is
528 * more complicated to allow performing it with integer operations.
529 * Another complication is that if a line could map onto many lines in the
530 * destination range then we want to choose the line at the center of those
532 * Example: if the chunk is 2 lines long in A and 10 lines long in B then the
533 * first 5 lines in B will map onto the first line in the A chunk, while the
534 * last 5 lines will all map onto the second line in the A chunk.
535 * Example: if the chunk is 10 lines long in A and 2 lines long in B then line
536 * 0 in B will map onto line 2 in A, and line 1 in B will map onto line 7 in A.
538 static int map_line_number(int line_number,
539 const struct line_number_mapping *mapping)
541 return ((line_number - mapping->source_start) * 2 + 1) *
542 mapping->destination_length /
543 (mapping->source_length * 2) +
544 mapping->destination_start;
547 /* Get a pointer to the element storing the similarity between a line in A
550 * The similarities are stored in a 2-dimensional array. Each "row" in the
551 * array contains the similarities for a line in B. The similarities stored in
552 * a row are the similarities between the line in B and the nearby lines in A.
553 * To keep the length of each row the same, it is padded out with values of -1
554 * where the search range extends beyond the lines in A.
555 * For example, if max_search_distance_a is 2 and the two sides of a diff chunk
562 * Then the similarity array will contain:
563 * [-1, -1, am, bm, cm,
564 * -1, an, bn, cn, dn,
565 * ao, bo, co, do, eo,
566 * bp, cp, dp, ep, -1,
567 * cq, dq, eq, -1, -1]
568 * Where similarities are denoted either by -1 for invalid, or the
569 * concatenation of the two lines in the diff being compared.
571 * \param similarities array of similarities between lines in A and B
572 * \param line_a the index of the line in A, in the same frame of reference as
574 * \param local_line_b the index of the line in B, relative to the first line
575 * in B that similarities represents.
576 * \param closest_line_a the index of the line in A that is deemed to be
577 * closest to local_line_b. This must be in the same
578 * frame of reference as line_a. This value defines
579 * where similarities is centered for the line in B.
580 * \param max_search_distance_a maximum distance in lines from the closest line
581 * in A for other lines in A for which
582 * similarities may be calculated.
584 static int *get_similarity(int *similarities,
585 int line_a, int local_line_b,
586 int closest_line_a, int max_search_distance_a)
588 assert(abs(line_a - closest_line_a) <=
589 max_search_distance_a);
590 return similarities + line_a - closest_line_a +
591 max_search_distance_a +
592 local_line_b * (max_search_distance_a * 2 + 1);
595 #define CERTAIN_NOTHING_MATCHES -2
596 #define CERTAINTY_NOT_CALCULATED -1
598 /* Given a line in B, first calculate its similarities with nearby lines in A
599 * if not already calculated, then identify the most similar and second most
600 * similar lines. The "certainty" is calculated based on those two
603 * \param start_a the index of the first line of the chunk in A
604 * \param length_a the length in lines of the chunk in A
605 * \param local_line_b the index of the line in B, relative to the first line
607 * \param fingerprints_a array of fingerprints for the chunk in A
608 * \param fingerprints_b array of fingerprints for the chunk in B
609 * \param similarities 2-dimensional array of similarities between lines in A
610 * and B. See get_similarity() for more details.
611 * \param certainties array of values indicating how strongly a line in B is
612 * matched with some line in A.
613 * \param second_best_result array of absolute indices in A for the second
614 * closest match of a line in B.
615 * \param result array of absolute indices in A for the closest match of a line
617 * \param max_search_distance_a maximum distance in lines from the closest line
618 * in A for other lines in A for which
619 * similarities may be calculated.
620 * \param map_line_number_in_b_to_a parameter to map_line_number().
622 static void find_best_line_matches(
627 struct fingerprint *fingerprints_a,
628 struct fingerprint *fingerprints_b,
631 int *second_best_result,
633 const int max_search_distance_a,
634 const struct line_number_mapping *map_line_number_in_b_to_a)
637 int i, search_start, search_end, closest_local_line_a, *similarity,
638 best_similarity = 0, second_best_similarity = 0,
639 best_similarity_index = 0, second_best_similarity_index = 0;
641 /* certainty has already been calculated so no need to redo the work */
642 if (certainties[local_line_b] != CERTAINTY_NOT_CALCULATED)
645 closest_local_line_a = map_line_number(
646 local_line_b + start_b, map_line_number_in_b_to_a) - start_a;
648 search_start = closest_local_line_a - max_search_distance_a;
649 if (search_start < 0)
652 search_end = closest_local_line_a + max_search_distance_a + 1;
653 if (search_end > length_a)
654 search_end = length_a;
656 for (i = search_start; i < search_end; ++i) {
657 similarity = get_similarity(similarities,
659 closest_local_line_a,
660 max_search_distance_a);
661 if (*similarity == -1) {
662 /* This value will never exceed 10 but assert just in
665 assert(abs(i - closest_local_line_a) < 1000);
666 /* scale the similarity by (1000 - distance from
667 * closest line) to act as a tie break between lines
668 * that otherwise are equally similar.
670 *similarity = fingerprint_similarity(
671 fingerprints_b + local_line_b,
672 fingerprints_a + i) *
673 (1000 - abs(i - closest_local_line_a));
675 if (*similarity > best_similarity) {
676 second_best_similarity = best_similarity;
677 second_best_similarity_index = best_similarity_index;
678 best_similarity = *similarity;
679 best_similarity_index = i;
680 } else if (*similarity > second_best_similarity) {
681 second_best_similarity = *similarity;
682 second_best_similarity_index = i;
686 if (best_similarity == 0) {
687 /* this line definitely doesn't match with anything. Mark it
688 * with this special value so it doesn't get invalidated and
689 * won't be recalculated.
691 certainties[local_line_b] = CERTAIN_NOTHING_MATCHES;
692 result[local_line_b] = -1;
694 /* Calculate the certainty with which this line matches.
695 * If the line matches well with two lines then that reduces
696 * the certainty. However we still want to prioritise matching
697 * a line that matches very well with two lines over matching a
698 * line that matches poorly with one line, hence doubling
700 * This means that if we have
701 * line X that matches only one line with a score of 3,
702 * line Y that matches two lines equally with a score of 5,
703 * and line Z that matches only one line with a score or 2,
704 * then the lines in order of certainty are X, Y, Z.
706 certainties[local_line_b] = best_similarity * 2 -
707 second_best_similarity;
709 /* We keep both the best and second best results to allow us to
710 * check at a later stage of the matching process whether the
711 * result needs to be invalidated.
713 result[local_line_b] = start_a + best_similarity_index;
714 second_best_result[local_line_b] =
715 start_a + second_best_similarity_index;
720 * This finds the line that we can match with the most confidence, and
721 * uses it as a partition. It then calls itself on the lines on either side of
722 * that partition. In this way we avoid lines appearing out of order, and
723 * retain a sensible line ordering.
724 * \param start_a index of the first line in A with which lines in B may be
726 * \param start_b index of the first line in B for which matching should be
728 * \param length_a number of lines in A with which lines in B may be compared.
729 * \param length_b number of lines in B for which matching should be done.
730 * \param fingerprints_a mutable array of fingerprints in A. The first element
731 * corresponds to the line at start_a.
732 * \param fingerprints_b array of fingerprints in B. The first element
733 * corresponds to the line at start_b.
734 * \param similarities 2-dimensional array of similarities between lines in A
735 * and B. See get_similarity() for more details.
736 * \param certainties array of values indicating how strongly a line in B is
737 * matched with some line in A.
738 * \param second_best_result array of absolute indices in A for the second
739 * closest match of a line in B.
740 * \param result array of absolute indices in A for the closest match of a line
742 * \param max_search_distance_a maximum distance in lines from the closest line
743 * in A for other lines in A for which
744 * similarities may be calculated.
745 * \param max_search_distance_b an upper bound on the greatest possible
746 * distance between lines in B such that they will
747 * both be compared with the same line in A
748 * according to max_search_distance_a.
749 * \param map_line_number_in_b_to_a parameter to map_line_number().
751 static void fuzzy_find_matching_lines_recurse(
752 int start_a, int start_b,
753 int length_a, int length_b,
754 struct fingerprint *fingerprints_a,
755 struct fingerprint *fingerprints_b,
758 int *second_best_result,
760 int max_search_distance_a,
761 int max_search_distance_b,
762 const struct line_number_mapping *map_line_number_in_b_to_a)
764 int i, invalidate_min, invalidate_max, offset_b,
765 second_half_start_a, second_half_start_b,
766 second_half_length_a, second_half_length_b,
767 most_certain_line_a, most_certain_local_line_b = -1,
768 most_certain_line_certainty = -1,
769 closest_local_line_a;
771 for (i = 0; i < length_b; ++i) {
772 find_best_line_matches(start_a,
782 max_search_distance_a,
783 map_line_number_in_b_to_a);
785 if (certainties[i] > most_certain_line_certainty) {
786 most_certain_line_certainty = certainties[i];
787 most_certain_local_line_b = i;
792 if (most_certain_local_line_b == -1)
795 most_certain_line_a = result[most_certain_local_line_b];
798 * Subtract the most certain line's fingerprint in B from the matched
799 * fingerprint in A. This means that other lines in B can't also match
800 * the same parts of the line in A.
802 fingerprint_subtract(fingerprints_a + most_certain_line_a - start_a,
803 fingerprints_b + most_certain_local_line_b);
805 /* Invalidate results that may be affected by the choice of most
808 invalidate_min = most_certain_local_line_b - max_search_distance_b;
809 invalidate_max = most_certain_local_line_b + max_search_distance_b + 1;
810 if (invalidate_min < 0)
812 if (invalidate_max > length_b)
813 invalidate_max = length_b;
815 /* As the fingerprint in A has changed, discard previously calculated
816 * similarity values with that fingerprint.
818 for (i = invalidate_min; i < invalidate_max; ++i) {
819 closest_local_line_a = map_line_number(
820 i + start_b, map_line_number_in_b_to_a) - start_a;
822 /* Check that the lines in A and B are close enough that there
823 * is a similarity value for them.
825 if (abs(most_certain_line_a - start_a - closest_local_line_a) >
826 max_search_distance_a) {
830 *get_similarity(similarities, most_certain_line_a - start_a,
831 i, closest_local_line_a,
832 max_search_distance_a) = -1;
835 /* More invalidating of results that may be affected by the choice of
837 * Discard the matches for lines in B that are currently matched with a
838 * line in A such that their ordering contradicts the ordering imposed
839 * by the choice of most certain line.
841 for (i = most_certain_local_line_b - 1; i >= invalidate_min; --i) {
842 /* In this loop we discard results for lines in B that are
843 * before most-certain-line-B but are matched with a line in A
844 * that is after most-certain-line-A.
846 if (certainties[i] >= 0 &&
847 (result[i] >= most_certain_line_a ||
848 second_best_result[i] >= most_certain_line_a)) {
849 certainties[i] = CERTAINTY_NOT_CALCULATED;
852 for (i = most_certain_local_line_b + 1; i < invalidate_max; ++i) {
853 /* In this loop we discard results for lines in B that are
854 * after most-certain-line-B but are matched with a line in A
855 * that is before most-certain-line-A.
857 if (certainties[i] >= 0 &&
858 (result[i] <= most_certain_line_a ||
859 second_best_result[i] <= most_certain_line_a)) {
860 certainties[i] = CERTAINTY_NOT_CALCULATED;
864 /* Repeat the matching process for lines before the most certain line.
866 if (most_certain_local_line_b > 0) {
867 fuzzy_find_matching_lines_recurse(
869 most_certain_line_a + 1 - start_a,
870 most_certain_local_line_b,
871 fingerprints_a, fingerprints_b, similarities,
872 certainties, second_best_result, result,
873 max_search_distance_a,
874 max_search_distance_b,
875 map_line_number_in_b_to_a);
877 /* Repeat the matching process for lines after the most certain line.
879 if (most_certain_local_line_b + 1 < length_b) {
880 second_half_start_a = most_certain_line_a;
881 offset_b = most_certain_local_line_b + 1;
882 second_half_start_b = start_b + offset_b;
883 second_half_length_a =
884 length_a + start_a - second_half_start_a;
885 second_half_length_b =
886 length_b + start_b - second_half_start_b;
887 fuzzy_find_matching_lines_recurse(
888 second_half_start_a, second_half_start_b,
889 second_half_length_a, second_half_length_b,
890 fingerprints_a + second_half_start_a - start_a,
891 fingerprints_b + offset_b,
893 offset_b * (max_search_distance_a * 2 + 1),
894 certainties + offset_b,
895 second_best_result + offset_b, result + offset_b,
896 max_search_distance_a,
897 max_search_distance_b,
898 map_line_number_in_b_to_a);
902 /* Find the lines in the parent line range that most closely match the lines in
903 * the target line range. This is accomplished by matching fingerprints in each
904 * blame_origin, and choosing the best matches that preserve the line ordering.
905 * See struct fingerprint for details of fingerprint matching, and
906 * fuzzy_find_matching_lines_recurse for details of preserving line ordering.
908 * The performance is believed to be O(n log n) in the typical case and O(n^2)
909 * in a pathological case, where n is the number of lines in the target range.
911 static int *fuzzy_find_matching_lines(struct blame_origin *parent,
912 struct blame_origin *target,
913 int tlno, int parent_slno, int same,
916 /* We use the terminology "A" for the left hand side of the diff AKA
917 * parent, and "B" for the right hand side of the diff AKA target. */
918 int start_a = parent_slno;
919 int length_a = parent_len;
921 int length_b = same - tlno;
923 struct line_number_mapping map_line_number_in_b_to_a = {
924 start_a, length_a, start_b, length_b
927 struct fingerprint *fingerprints_a = parent->fingerprints;
928 struct fingerprint *fingerprints_b = target->fingerprints;
930 int i, *result, *second_best_result,
931 *certainties, *similarities, similarity_count;
934 * max_search_distance_a means that given a line in B, compare it to
935 * the line in A that is closest to its position, and the lines in A
936 * that are no greater than max_search_distance_a lines away from the
939 * max_search_distance_b is an upper bound on the greatest possible
940 * distance between lines in B such that they will both be compared
941 * with the same line in A according to max_search_distance_a.
943 int max_search_distance_a = 10, max_search_distance_b;
948 if (max_search_distance_a >= length_a)
949 max_search_distance_a = length_a ? length_a - 1 : 0;
951 max_search_distance_b = ((2 * max_search_distance_a + 1) * length_b
954 result = xcalloc(sizeof(int), length_b);
955 second_best_result = xcalloc(sizeof(int), length_b);
956 certainties = xcalloc(sizeof(int), length_b);
958 /* See get_similarity() for details of similarities. */
959 similarity_count = length_b * (max_search_distance_a * 2 + 1);
960 similarities = xcalloc(sizeof(int), similarity_count);
962 for (i = 0; i < length_b; ++i) {
964 second_best_result[i] = -1;
965 certainties[i] = CERTAINTY_NOT_CALCULATED;
968 for (i = 0; i < similarity_count; ++i)
969 similarities[i] = -1;
971 fuzzy_find_matching_lines_recurse(start_a, start_b,
973 fingerprints_a + start_a,
974 fingerprints_b + start_b,
979 max_search_distance_a,
980 max_search_distance_b,
981 &map_line_number_in_b_to_a);
985 free(second_best_result);
990 static void fill_origin_fingerprints(struct blame_origin *o)
996 o->num_lines = find_line_starts(&line_starts, o->file.ptr,
998 o->fingerprints = xcalloc(sizeof(struct fingerprint), o->num_lines);
999 get_line_fingerprints(o->fingerprints, o->file.ptr, line_starts,
1004 static void drop_origin_fingerprints(struct blame_origin *o)
1006 if (o->fingerprints) {
1007 free_line_fingerprints(o->fingerprints, o->num_lines);
1009 FREE_AND_NULL(o->fingerprints);
1014 * Given an origin, prepare mmfile_t structure to be used by the
1017 static void fill_origin_blob(struct diff_options *opt,
1018 struct blame_origin *o, mmfile_t *file,
1019 int *num_read_blob, int fill_fingerprints)
1022 enum object_type type;
1023 unsigned long file_size;
1026 if (opt->flags.allow_textconv &&
1027 textconv_object(opt->repo, o->path, o->mode,
1028 &o->blob_oid, 1, &file->ptr, &file_size))
1031 file->ptr = read_object_file(&o->blob_oid, &type,
1033 file->size = file_size;
1036 die("Cannot read blob %s for path %s",
1037 oid_to_hex(&o->blob_oid),
1043 if (fill_fingerprints)
1044 fill_origin_fingerprints(o);
1047 static void drop_origin_blob(struct blame_origin *o)
1049 FREE_AND_NULL(o->file.ptr);
1050 drop_origin_fingerprints(o);
1054 * Any merge of blames happens on lists of blames that arrived via
1055 * different parents in a single suspect. In this case, we want to
1056 * sort according to the suspect line numbers as opposed to the final
1057 * image line numbers. The function body is somewhat longish because
1058 * it avoids unnecessary writes.
1061 static struct blame_entry *blame_merge(struct blame_entry *list1,
1062 struct blame_entry *list2)
1064 struct blame_entry *p1 = list1, *p2 = list2,
1072 if (p1->s_lno <= p2->s_lno) {
1075 if ((p1 = *tail) == NULL) {
1079 } while (p1->s_lno <= p2->s_lno);
1085 if ((p2 = *tail) == NULL) {
1089 } while (p1->s_lno > p2->s_lno);
1093 if ((p1 = *tail) == NULL) {
1097 } while (p1->s_lno <= p2->s_lno);
1101 static void *get_next_blame(const void *p)
1103 return ((struct blame_entry *)p)->next;
1106 static void set_next_blame(void *p1, void *p2)
1108 ((struct blame_entry *)p1)->next = p2;
1112 * Final image line numbers are all different, so we don't need a
1113 * three-way comparison here.
1116 static int compare_blame_final(const void *p1, const void *p2)
1118 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
1122 static int compare_blame_suspect(const void *p1, const void *p2)
1124 const struct blame_entry *s1 = p1, *s2 = p2;
1126 * to allow for collating suspects, we sort according to the
1127 * respective pointer value as the primary sorting criterion.
1128 * The actual relation is pretty unimportant as long as it
1129 * establishes a total order. Comparing as integers gives us
1132 if (s1->suspect != s2->suspect)
1133 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
1134 if (s1->s_lno == s2->s_lno)
1136 return s1->s_lno > s2->s_lno ? 1 : -1;
1139 void blame_sort_final(struct blame_scoreboard *sb)
1141 sb->ent = llist_mergesort(sb->ent, get_next_blame, set_next_blame,
1142 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);
1406 clear_pathspec(&diff_opts.pathspec);
1411 * We have an origin -- find the path that corresponds to it in its
1412 * parent and return an origin structure to represent it.
1414 static struct blame_origin *find_rename(struct repository *r,
1415 struct commit *parent,
1416 struct blame_origin *origin,
1417 struct blame_bloom_data *bd)
1419 struct blame_origin *porigin = NULL;
1420 struct diff_options diff_opts;
1423 repo_diff_setup(r, &diff_opts);
1424 diff_opts.flags.recursive = 1;
1425 diff_opts.detect_rename = DIFF_DETECT_RENAME;
1426 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1427 diff_opts.single_follow = origin->path;
1428 diff_setup_done(&diff_opts);
1430 if (is_null_oid(&origin->commit->object.oid))
1431 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
1433 diff_tree_oid(get_commit_tree_oid(parent),
1434 get_commit_tree_oid(origin->commit),
1436 diffcore_std(&diff_opts);
1438 for (i = 0; i < diff_queued_diff.nr; i++) {
1439 struct diff_filepair *p = diff_queued_diff.queue[i];
1440 if ((p->status == 'R' || p->status == 'C') &&
1441 !strcmp(p->two->path, origin->path)) {
1442 add_bloom_key(bd, p->one->path);
1443 porigin = get_origin(parent, p->one->path);
1444 oidcpy(&porigin->blob_oid, &p->one->oid);
1445 porigin->mode = p->one->mode;
1449 diff_flush(&diff_opts);
1450 clear_pathspec(&diff_opts.pathspec);
1455 * Append a new blame entry to a given output queue.
1457 static void add_blame_entry(struct blame_entry ***queue,
1458 const struct blame_entry *src)
1460 struct blame_entry *e = xmalloc(sizeof(*e));
1461 memcpy(e, src, sizeof(*e));
1462 blame_origin_incref(e->suspect);
1470 * src typically is on-stack; we want to copy the information in it to
1471 * a malloced blame_entry that gets added to the given queue. The
1472 * origin of dst loses a refcnt.
1474 static void dup_entry(struct blame_entry ***queue,
1475 struct blame_entry *dst, struct blame_entry *src)
1477 blame_origin_incref(src->suspect);
1478 blame_origin_decref(dst->suspect);
1479 memcpy(dst, src, sizeof(*src));
1480 dst->next = **queue;
1482 *queue = &dst->next;
1485 const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
1487 return sb->final_buf + sb->lineno[lno];
1491 * It is known that lines between tlno to same came from parent, and e
1492 * has an overlap with that range. it also is known that parent's
1493 * line plno corresponds to e's line tlno.
1499 * <------------------>
1501 * Split e into potentially three parts; before this chunk, the chunk
1502 * to be blamed for the parent, and after that portion.
1504 static void split_overlap(struct blame_entry *split,
1505 struct blame_entry *e,
1506 int tlno, int plno, int same,
1507 struct blame_origin *parent)
1511 memset(split, 0, sizeof(struct blame_entry [3]));
1513 for (i = 0; i < 3; i++) {
1514 split[i].ignored = e->ignored;
1515 split[i].unblamable = e->unblamable;
1518 if (e->s_lno < tlno) {
1519 /* there is a pre-chunk part not blamed on parent */
1520 split[0].suspect = blame_origin_incref(e->suspect);
1521 split[0].lno = e->lno;
1522 split[0].s_lno = e->s_lno;
1523 split[0].num_lines = tlno - e->s_lno;
1524 split[1].lno = e->lno + tlno - e->s_lno;
1525 split[1].s_lno = plno;
1528 split[1].lno = e->lno;
1529 split[1].s_lno = plno + (e->s_lno - tlno);
1532 if (same < e->s_lno + e->num_lines) {
1533 /* there is a post-chunk part not blamed on parent */
1534 split[2].suspect = blame_origin_incref(e->suspect);
1535 split[2].lno = e->lno + (same - e->s_lno);
1536 split[2].s_lno = e->s_lno + (same - e->s_lno);
1537 split[2].num_lines = e->s_lno + e->num_lines - same;
1538 chunk_end_lno = split[2].lno;
1541 chunk_end_lno = e->lno + e->num_lines;
1542 split[1].num_lines = chunk_end_lno - split[1].lno;
1545 * if it turns out there is nothing to blame the parent for,
1546 * forget about the splitting. !split[1].suspect signals this.
1548 if (split[1].num_lines < 1)
1550 split[1].suspect = blame_origin_incref(parent);
1554 * split_overlap() divided an existing blame e into up to three parts
1555 * in split. Any assigned blame is moved to queue to
1556 * reflect the split.
1558 static void split_blame(struct blame_entry ***blamed,
1559 struct blame_entry ***unblamed,
1560 struct blame_entry *split,
1561 struct blame_entry *e)
1563 if (split[0].suspect && split[2].suspect) {
1564 /* The first part (reuse storage for the existing entry e) */
1565 dup_entry(unblamed, e, &split[0]);
1567 /* The last part -- me */
1568 add_blame_entry(unblamed, &split[2]);
1570 /* ... and the middle part -- parent */
1571 add_blame_entry(blamed, &split[1]);
1573 else if (!split[0].suspect && !split[2].suspect)
1575 * The parent covers the entire area; reuse storage for
1576 * e and replace it with the parent.
1578 dup_entry(blamed, e, &split[1]);
1579 else if (split[0].suspect) {
1580 /* me and then parent */
1581 dup_entry(unblamed, e, &split[0]);
1582 add_blame_entry(blamed, &split[1]);
1585 /* parent and then me */
1586 dup_entry(blamed, e, &split[1]);
1587 add_blame_entry(unblamed, &split[2]);
1592 * After splitting the blame, the origins used by the
1593 * on-stack blame_entry should lose one refcnt each.
1595 static void decref_split(struct blame_entry *split)
1599 for (i = 0; i < 3; i++)
1600 blame_origin_decref(split[i].suspect);
1604 * reverse_blame reverses the list given in head, appending tail.
1605 * That allows us to build lists in reverse order, then reverse them
1606 * afterwards. This can be faster than building the list in proper
1607 * order right away. The reason is that building in proper order
1608 * requires writing a link in the _previous_ element, while building
1609 * in reverse order just requires placing the list head into the
1610 * _current_ element.
1613 static struct blame_entry *reverse_blame(struct blame_entry *head,
1614 struct blame_entry *tail)
1617 struct blame_entry *next = head->next;
1626 * Splits a blame entry into two entries at 'len' lines. The original 'e'
1627 * consists of len lines, i.e. [e->lno, e->lno + len), and the second part,
1628 * which is returned, consists of the remainder: [e->lno + len, e->lno +
1629 * e->num_lines). The caller needs to sort out the reference counting for the
1630 * new entry's suspect.
1632 static struct blame_entry *split_blame_at(struct blame_entry *e, int len,
1633 struct blame_origin *new_suspect)
1635 struct blame_entry *n = xcalloc(1, sizeof(struct blame_entry));
1637 n->suspect = new_suspect;
1638 n->ignored = e->ignored;
1639 n->unblamable = e->unblamable;
1640 n->lno = e->lno + len;
1641 n->s_lno = e->s_lno + len;
1642 n->num_lines = e->num_lines - len;
1648 struct blame_line_tracker {
1653 static int are_lines_adjacent(struct blame_line_tracker *first,
1654 struct blame_line_tracker *second)
1656 return first->is_parent == second->is_parent &&
1657 first->s_lno + 1 == second->s_lno;
1660 static int scan_parent_range(struct fingerprint *p_fps,
1661 struct fingerprint *t_fps, int t_idx,
1662 int from, int nr_lines)
1665 #define FINGERPRINT_FILE_THRESHOLD 10
1666 int best_sim_val = FINGERPRINT_FILE_THRESHOLD;
1667 int best_sim_idx = -1;
1669 for (p_idx = from; p_idx < from + nr_lines; p_idx++) {
1670 sim = fingerprint_similarity(&t_fps[t_idx], &p_fps[p_idx]);
1671 if (sim < best_sim_val)
1673 /* Break ties with the closest-to-target line number */
1674 if (sim == best_sim_val && best_sim_idx != -1 &&
1675 abs(best_sim_idx - t_idx) < abs(p_idx - t_idx))
1678 best_sim_idx = p_idx;
1680 return best_sim_idx;
1684 * The first pass checks the blame entry (from the target) against the parent's
1685 * diff chunk. If that fails for a line, the second pass tries to match that
1686 * line to any part of parent file. That catches cases where a change was
1687 * broken into two chunks by 'context.'
1689 static void guess_line_blames(struct blame_origin *parent,
1690 struct blame_origin *target,
1691 int tlno, int offset, int same, int parent_len,
1692 struct blame_line_tracker *line_blames)
1694 int i, best_idx, target_idx;
1695 int parent_slno = tlno + offset;
1698 fuzzy_matches = fuzzy_find_matching_lines(parent, target,
1699 tlno, parent_slno, same,
1701 for (i = 0; i < same - tlno; i++) {
1702 target_idx = tlno + i;
1703 if (fuzzy_matches && fuzzy_matches[i] >= 0) {
1704 best_idx = fuzzy_matches[i];
1706 best_idx = scan_parent_range(parent->fingerprints,
1707 target->fingerprints,
1711 if (best_idx >= 0) {
1712 line_blames[i].is_parent = 1;
1713 line_blames[i].s_lno = best_idx;
1715 line_blames[i].is_parent = 0;
1716 line_blames[i].s_lno = target_idx;
1719 free(fuzzy_matches);
1723 * This decides which parts of a blame entry go to the parent (added to the
1724 * ignoredp list) and which stay with the target (added to the diffp list). The
1725 * actual decision was made in a separate heuristic function, and those answers
1726 * for the lines in 'e' are in line_blames. This consumes e, essentially
1727 * putting it on a list.
1729 * Note that the blame entries on the ignoredp list are not necessarily sorted
1730 * with respect to the parent's line numbers yet.
1732 static void ignore_blame_entry(struct blame_entry *e,
1733 struct blame_origin *parent,
1734 struct blame_entry **diffp,
1735 struct blame_entry **ignoredp,
1736 struct blame_line_tracker *line_blames)
1738 int entry_len, nr_lines, i;
1741 * We carve new entries off the front of e. Each entry comes from a
1742 * contiguous chunk of lines: adjacent lines from the same origin
1743 * (either the parent or the target).
1746 nr_lines = e->num_lines; /* e changes in the loop */
1747 for (i = 0; i < nr_lines; i++) {
1748 struct blame_entry *next = NULL;
1751 * We are often adjacent to the next line - only split the blame
1752 * entry when we have to.
1754 if (i + 1 < nr_lines) {
1755 if (are_lines_adjacent(&line_blames[i],
1756 &line_blames[i + 1])) {
1760 next = split_blame_at(e, entry_len,
1761 blame_origin_incref(e->suspect));
1763 if (line_blames[i].is_parent) {
1765 blame_origin_decref(e->suspect);
1766 e->suspect = blame_origin_incref(parent);
1767 e->s_lno = line_blames[i - entry_len + 1].s_lno;
1768 e->next = *ignoredp;
1772 /* e->s_lno is already in the target's address space. */
1776 assert(e->num_lines == entry_len);
1784 * Process one hunk from the patch between the current suspect for
1785 * blame_entry e and its parent. This first blames any unfinished
1786 * entries before the chunk (which is where target and parent start
1787 * differing) on the parent, and then splits blame entries at the
1788 * start and at the end of the difference region. Since use of -M and
1789 * -C options may lead to overlapping/duplicate source line number
1790 * ranges, all we can rely on from sorting/merging is the order of the
1791 * first suspect line number.
1793 * tlno: line number in the target where this chunk begins
1794 * same: line number in the target where this chunk ends
1795 * offset: add to tlno to get the chunk starting point in the parent
1796 * parent_len: number of lines in the parent chunk
1798 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
1799 int tlno, int offset, int same, int parent_len,
1800 struct blame_origin *parent,
1801 struct blame_origin *target, int ignore_diffs)
1803 struct blame_entry *e = **srcq;
1804 struct blame_entry *samep = NULL, *diffp = NULL, *ignoredp = NULL;
1805 struct blame_line_tracker *line_blames = NULL;
1807 while (e && e->s_lno < tlno) {
1808 struct blame_entry *next = e->next;
1810 * current record starts before differing portion. If
1811 * it reaches into it, we need to split it up and
1812 * examine the second part separately.
1814 if (e->s_lno + e->num_lines > tlno) {
1815 /* Move second half to a new record */
1816 struct blame_entry *n;
1818 n = split_blame_at(e, tlno - e->s_lno, e->suspect);
1819 /* Push new record to diffp */
1823 blame_origin_decref(e->suspect);
1824 /* Pass blame for everything before the differing
1825 * chunk to the parent */
1826 e->suspect = blame_origin_incref(parent);
1833 * As we don't know how much of a common stretch after this
1834 * diff will occur, the currently blamed parts are all that we
1835 * can assign to the parent for now.
1839 **dstq = reverse_blame(samep, **dstq);
1840 *dstq = &samep->next;
1843 * Prepend the split off portions: everything after e starts
1844 * after the blameable portion.
1846 e = reverse_blame(diffp, e);
1849 * Now retain records on the target while parts are different
1855 if (ignore_diffs && same - tlno > 0) {
1856 line_blames = xcalloc(sizeof(struct blame_line_tracker),
1858 guess_line_blames(parent, target, tlno, offset, same,
1859 parent_len, line_blames);
1862 while (e && e->s_lno < same) {
1863 struct blame_entry *next = e->next;
1866 * If current record extends into sameness, need to split.
1868 if (e->s_lno + e->num_lines > same) {
1870 * Move second half to a new record to be
1871 * processed by later chunks
1873 struct blame_entry *n;
1875 n = split_blame_at(e, same - e->s_lno,
1876 blame_origin_incref(e->suspect));
1877 /* Push new record to samep */
1882 ignore_blame_entry(e, parent, &diffp, &ignoredp,
1883 line_blames + e->s_lno - tlno);
1893 * Note ignoredp is not sorted yet, and thus neither is dstq.
1894 * That list must be sorted before we queue_blames(). We defer
1895 * sorting until after all diff hunks are processed, so that
1896 * guess_line_blames() can pick *any* line in the parent. The
1897 * slight drawback is that we end up sorting all blame entries
1898 * passed to the parent, including those that are unrelated to
1899 * changes made by the ignored commit.
1901 **dstq = reverse_blame(ignoredp, **dstq);
1902 *dstq = &ignoredp->next;
1904 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
1905 /* Move across elements that are in the unblamable portion */
1907 *srcq = &diffp->next;
1910 struct blame_chunk_cb_data {
1911 struct blame_origin *parent;
1912 struct blame_origin *target;
1915 struct blame_entry **dstq;
1916 struct blame_entry **srcq;
1919 /* diff chunks are from parent to target */
1920 static int blame_chunk_cb(long start_a, long count_a,
1921 long start_b, long count_b, void *data)
1923 struct blame_chunk_cb_data *d = data;
1924 if (start_a - start_b != d->offset)
1925 die("internal error in blame::blame_chunk_cb");
1926 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
1927 start_b + count_b, count_a, d->parent, d->target,
1929 d->offset = start_a + count_a - (start_b + count_b);
1934 * We are looking at the origin 'target' and aiming to pass blame
1935 * for the lines it is suspected to its parent. Run diff to find
1936 * which lines came from parent and pass blame for them.
1938 static void pass_blame_to_parent(struct blame_scoreboard *sb,
1939 struct blame_origin *target,
1940 struct blame_origin *parent, int ignore_diffs)
1942 mmfile_t file_p, file_o;
1943 struct blame_chunk_cb_data d;
1944 struct blame_entry *newdest = NULL;
1946 if (!target->suspects)
1947 return; /* nothing remains for this target */
1952 d.ignore_diffs = ignore_diffs;
1953 d.dstq = &newdest; d.srcq = &target->suspects;
1955 fill_origin_blob(&sb->revs->diffopt, parent, &file_p,
1956 &sb->num_read_blob, ignore_diffs);
1957 fill_origin_blob(&sb->revs->diffopt, target, &file_o,
1958 &sb->num_read_blob, ignore_diffs);
1959 sb->num_get_patch++;
1961 if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
1962 die("unable to generate diff (%s -> %s)",
1963 oid_to_hex(&parent->commit->object.oid),
1964 oid_to_hex(&target->commit->object.oid));
1965 /* The rest are the same as the parent */
1966 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, 0,
1970 newdest = llist_mergesort(newdest, get_next_blame,
1972 compare_blame_suspect);
1973 queue_blames(sb, parent, newdest);
1979 * The lines in blame_entry after splitting blames many times can become
1980 * very small and trivial, and at some point it becomes pointless to
1981 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
1982 * ordinary C program, and it is not worth to say it was copied from
1983 * totally unrelated file in the parent.
1985 * Compute how trivial the lines in the blame_entry are.
1987 unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
1990 const char *cp, *ep;
1996 cp = blame_nth_line(sb, e->lno);
1997 ep = blame_nth_line(sb, e->lno + e->num_lines);
1999 unsigned ch = *((unsigned char *)cp);
2009 * best_so_far[] and potential[] are both a split of an existing blame_entry
2010 * that passes blame to the parent. Maintain best_so_far the best split so
2011 * far, by comparing potential and best_so_far and copying potential into
2012 * bst_so_far as needed.
2014 static void copy_split_if_better(struct blame_scoreboard *sb,
2015 struct blame_entry *best_so_far,
2016 struct blame_entry *potential)
2020 if (!potential[1].suspect)
2022 if (best_so_far[1].suspect) {
2023 if (blame_entry_score(sb, &potential[1]) <
2024 blame_entry_score(sb, &best_so_far[1]))
2028 for (i = 0; i < 3; i++)
2029 blame_origin_incref(potential[i].suspect);
2030 decref_split(best_so_far);
2031 memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
2035 * We are looking at a part of the final image represented by
2036 * ent (tlno and same are offset by ent->s_lno).
2037 * tlno is where we are looking at in the final image.
2038 * up to (but not including) same match preimage.
2039 * plno is where we are looking at in the preimage.
2041 * <-------------- final image ---------------------->
2044 * <---------preimage----->
2047 * All line numbers are 0-based.
2049 static void handle_split(struct blame_scoreboard *sb,
2050 struct blame_entry *ent,
2051 int tlno, int plno, int same,
2052 struct blame_origin *parent,
2053 struct blame_entry *split)
2055 if (ent->num_lines <= tlno)
2058 struct blame_entry potential[3];
2061 split_overlap(potential, ent, tlno, plno, same, parent);
2062 copy_split_if_better(sb, split, potential);
2063 decref_split(potential);
2067 struct handle_split_cb_data {
2068 struct blame_scoreboard *sb;
2069 struct blame_entry *ent;
2070 struct blame_origin *parent;
2071 struct blame_entry *split;
2076 static int handle_split_cb(long start_a, long count_a,
2077 long start_b, long count_b, void *data)
2079 struct handle_split_cb_data *d = data;
2080 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
2082 d->plno = start_a + count_a;
2083 d->tlno = start_b + count_b;
2088 * Find the lines from parent that are the same as ent so that
2089 * we can pass blames to it. file_p has the blob contents for
2092 static void find_copy_in_blob(struct blame_scoreboard *sb,
2093 struct blame_entry *ent,
2094 struct blame_origin *parent,
2095 struct blame_entry *split,
2100 struct handle_split_cb_data d;
2102 memset(&d, 0, sizeof(d));
2103 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
2105 * Prepare mmfile that contains only the lines in ent.
2107 cp = blame_nth_line(sb, ent->lno);
2108 file_o.ptr = (char *) cp;
2109 file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
2112 * file_o is a part of final image we are annotating.
2113 * file_p partially may match that image.
2115 memset(split, 0, sizeof(struct blame_entry [3]));
2116 if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
2117 die("unable to generate diff (%s)",
2118 oid_to_hex(&parent->commit->object.oid));
2119 /* remainder, if any, all match the preimage */
2120 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
2123 /* Move all blame entries from list *source that have a score smaller
2124 * than score_min to the front of list *small.
2125 * Returns a pointer to the link pointing to the old head of the small list.
2128 static struct blame_entry **filter_small(struct blame_scoreboard *sb,
2129 struct blame_entry **small,
2130 struct blame_entry **source,
2133 struct blame_entry *p = *source;
2134 struct blame_entry *oldsmall = *small;
2136 if (blame_entry_score(sb, p) <= score_min) {
2152 * See if lines currently target is suspected for can be attributed to
2155 static void find_move_in_parent(struct blame_scoreboard *sb,
2156 struct blame_entry ***blamed,
2157 struct blame_entry **toosmall,
2158 struct blame_origin *target,
2159 struct blame_origin *parent)
2161 struct blame_entry *e, split[3];
2162 struct blame_entry *unblamed = target->suspects;
2163 struct blame_entry *leftover = NULL;
2167 return; /* nothing remains for this target */
2169 fill_origin_blob(&sb->revs->diffopt, parent, &file_p,
2170 &sb->num_read_blob, 0);
2174 /* At each iteration, unblamed has a NULL-terminated list of
2175 * entries that have not yet been tested for blame. leftover
2176 * contains the reversed list of entries that have been tested
2177 * without being assignable to the parent.
2180 struct blame_entry **unblamedtail = &unblamed;
2181 struct blame_entry *next;
2182 for (e = unblamed; e; e = next) {
2184 find_copy_in_blob(sb, e, parent, split, &file_p);
2185 if (split[1].suspect &&
2186 sb->move_score < blame_entry_score(sb, &split[1])) {
2187 split_blame(blamed, &unblamedtail, split, e);
2192 decref_split(split);
2194 *unblamedtail = NULL;
2195 toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
2197 target->suspects = reverse_blame(leftover, NULL);
2201 struct blame_entry *ent;
2202 struct blame_entry split[3];
2206 * Count the number of entries the target is suspected for,
2207 * and prepare a list of entry and the best split.
2209 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
2212 struct blame_entry *e;
2214 struct blame_list *blame_list = NULL;
2216 for (e = unblamed, num_ents = 0; e; e = e->next)
2219 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
2220 for (e = unblamed, i = 0; e; e = e->next)
2221 blame_list[i++].ent = e;
2223 *num_ents_p = num_ents;
2228 * For lines target is suspected for, see if we can find code movement
2229 * across file boundary from the parent commit. porigin is the path
2230 * in the parent we already tried.
2232 static void find_copy_in_parent(struct blame_scoreboard *sb,
2233 struct blame_entry ***blamed,
2234 struct blame_entry **toosmall,
2235 struct blame_origin *target,
2236 struct commit *parent,
2237 struct blame_origin *porigin,
2240 struct diff_options diff_opts;
2242 struct blame_list *blame_list;
2244 struct blame_entry *unblamed = target->suspects;
2245 struct blame_entry *leftover = NULL;
2248 return; /* nothing remains for this target */
2250 repo_diff_setup(sb->repo, &diff_opts);
2251 diff_opts.flags.recursive = 1;
2252 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
2254 diff_setup_done(&diff_opts);
2256 /* Try "find copies harder" on new path if requested;
2257 * we do not want to use diffcore_rename() actually to
2258 * match things up; find_copies_harder is set only to
2259 * force diff_tree_oid() to feed all filepairs to diff_queue,
2260 * and this code needs to be after diff_setup_done(), which
2261 * usually makes find-copies-harder imply copy detection.
2263 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
2264 || ((opt & PICKAXE_BLAME_COPY_HARDER)
2265 && (!porigin || strcmp(target->path, porigin->path))))
2266 diff_opts.flags.find_copies_harder = 1;
2268 if (is_null_oid(&target->commit->object.oid))
2269 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
2271 diff_tree_oid(get_commit_tree_oid(parent),
2272 get_commit_tree_oid(target->commit),
2275 if (!diff_opts.flags.find_copies_harder)
2276 diffcore_std(&diff_opts);
2279 struct blame_entry **unblamedtail = &unblamed;
2280 blame_list = setup_blame_list(unblamed, &num_ents);
2282 for (i = 0; i < diff_queued_diff.nr; i++) {
2283 struct diff_filepair *p = diff_queued_diff.queue[i];
2284 struct blame_origin *norigin;
2286 struct blame_entry potential[3];
2288 if (!DIFF_FILE_VALID(p->one))
2289 continue; /* does not exist in parent */
2290 if (S_ISGITLINK(p->one->mode))
2291 continue; /* ignore git links */
2292 if (porigin && !strcmp(p->one->path, porigin->path))
2293 /* find_move already dealt with this path */
2296 norigin = get_origin(parent, p->one->path);
2297 oidcpy(&norigin->blob_oid, &p->one->oid);
2298 norigin->mode = p->one->mode;
2299 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p,
2300 &sb->num_read_blob, 0);
2304 for (j = 0; j < num_ents; j++) {
2305 find_copy_in_blob(sb, blame_list[j].ent,
2306 norigin, potential, &file_p);
2307 copy_split_if_better(sb, blame_list[j].split,
2309 decref_split(potential);
2311 blame_origin_decref(norigin);
2314 for (j = 0; j < num_ents; j++) {
2315 struct blame_entry *split = blame_list[j].split;
2316 if (split[1].suspect &&
2317 sb->copy_score < blame_entry_score(sb, &split[1])) {
2318 split_blame(blamed, &unblamedtail, split,
2321 blame_list[j].ent->next = leftover;
2322 leftover = blame_list[j].ent;
2324 decref_split(split);
2327 *unblamedtail = NULL;
2328 toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
2330 target->suspects = reverse_blame(leftover, NULL);
2331 diff_flush(&diff_opts);
2332 clear_pathspec(&diff_opts.pathspec);
2336 * The blobs of origin and porigin exactly match, so everything
2337 * origin is suspected for can be blamed on the parent.
2339 static void pass_whole_blame(struct blame_scoreboard *sb,
2340 struct blame_origin *origin, struct blame_origin *porigin)
2342 struct blame_entry *e, *suspects;
2344 if (!porigin->file.ptr && origin->file.ptr) {
2345 /* Steal its file */
2346 porigin->file = origin->file;
2347 origin->file.ptr = NULL;
2349 suspects = origin->suspects;
2350 origin->suspects = NULL;
2351 for (e = suspects; e; e = e->next) {
2352 blame_origin_incref(porigin);
2353 blame_origin_decref(e->suspect);
2354 e->suspect = porigin;
2356 queue_blames(sb, porigin, suspects);
2360 * We pass blame from the current commit to its parents. We keep saying
2361 * "parent" (and "porigin"), but what we mean is to find scapegoat to
2362 * exonerate ourselves.
2364 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
2368 if (revs->first_parent_only &&
2370 commit->parents->next) {
2371 free_commit_list(commit->parents->next);
2372 commit->parents->next = NULL;
2374 return commit->parents;
2376 return lookup_decoration(&revs->children, &commit->object);
2379 static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
2381 struct commit_list *l = first_scapegoat(revs, commit, reverse);
2382 return commit_list_count(l);
2385 /* Distribute collected unsorted blames to the respected sorted lists
2386 * in the various origins.
2388 static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
2390 blamed = llist_mergesort(blamed, get_next_blame, set_next_blame,
2391 compare_blame_suspect);
2394 struct blame_origin *porigin = blamed->suspect;
2395 struct blame_entry *suspects = NULL;
2397 struct blame_entry *next = blamed->next;
2398 blamed->next = suspects;
2401 } while (blamed && blamed->suspect == porigin);
2402 suspects = reverse_blame(suspects, NULL);
2403 queue_blames(sb, porigin, suspects);
2409 typedef struct blame_origin *(*blame_find_alg)(struct repository *,
2411 struct blame_origin *,
2412 struct blame_bloom_data *);
2414 static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
2416 struct rev_info *revs = sb->revs;
2417 int i, pass, num_sg;
2418 struct commit *commit = origin->commit;
2419 struct commit_list *sg;
2420 struct blame_origin *sg_buf[MAXSG];
2421 struct blame_origin *porigin, **sg_origin = sg_buf;
2422 struct blame_entry *toosmall = NULL;
2423 struct blame_entry *blames, **blametail = &blames;
2425 num_sg = num_scapegoats(revs, commit, sb->reverse);
2428 else if (num_sg < ARRAY_SIZE(sg_buf))
2429 memset(sg_buf, 0, sizeof(sg_buf));
2431 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
2434 * The first pass looks for unrenamed path to optimize for
2435 * common cases, then we look for renames in the second pass.
2437 for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
2438 blame_find_alg find = pass ? find_rename : find_origin;
2440 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
2442 sg = sg->next, i++) {
2443 struct commit *p = sg->item;
2448 if (parse_commit(p))
2450 porigin = find(sb->repo, p, origin, sb->bloom_data);
2453 if (oideq(&porigin->blob_oid, &origin->blob_oid)) {
2454 pass_whole_blame(sb, origin, porigin);
2455 blame_origin_decref(porigin);
2458 for (j = same = 0; j < i; j++)
2460 oideq(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
2465 sg_origin[i] = porigin;
2467 blame_origin_decref(porigin);
2472 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
2474 sg = sg->next, i++) {
2475 struct blame_origin *porigin = sg_origin[i];
2478 if (!origin->previous) {
2479 blame_origin_incref(porigin);
2480 origin->previous = porigin;
2482 pass_blame_to_parent(sb, origin, porigin, 0);
2483 if (!origin->suspects)
2488 * Pass remaining suspects for ignored commits to their parents.
2490 if (oidset_contains(&sb->ignore_list, &commit->object.oid)) {
2491 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
2493 sg = sg->next, i++) {
2494 struct blame_origin *porigin = sg_origin[i];
2498 pass_blame_to_parent(sb, origin, porigin, 1);
2500 * Preemptively drop porigin so we can refresh the
2501 * fingerprints if we use the parent again, which can
2502 * occur if you ignore back-to-back commits.
2504 drop_origin_blob(porigin);
2505 if (!origin->suspects)
2511 * Optionally find moves in parents' files.
2513 if (opt & PICKAXE_BLAME_MOVE) {
2514 filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
2515 if (origin->suspects) {
2516 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
2518 sg = sg->next, i++) {
2519 struct blame_origin *porigin = sg_origin[i];
2522 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
2523 if (!origin->suspects)
2530 * Optionally find copies from parents' files.
2532 if (opt & PICKAXE_BLAME_COPY) {
2533 if (sb->copy_score > sb->move_score)
2534 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
2535 else if (sb->copy_score < sb->move_score) {
2536 origin->suspects = blame_merge(origin->suspects, toosmall);
2538 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
2540 if (!origin->suspects)
2543 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
2545 sg = sg->next, i++) {
2546 struct blame_origin *porigin = sg_origin[i];
2547 find_copy_in_parent(sb, &blametail, &toosmall,
2548 origin, sg->item, porigin, opt);
2549 if (!origin->suspects)
2556 distribute_blame(sb, blames);
2558 * prepend toosmall to origin->suspects
2560 * There is no point in sorting: this ends up on a big
2561 * unsorted list in the caller anyway.
2564 struct blame_entry **tail = &toosmall;
2566 tail = &(*tail)->next;
2567 *tail = origin->suspects;
2568 origin->suspects = toosmall;
2570 for (i = 0; i < num_sg; i++) {
2572 if (!sg_origin[i]->suspects)
2573 drop_origin_blob(sg_origin[i]);
2574 blame_origin_decref(sg_origin[i]);
2577 drop_origin_blob(origin);
2578 if (sg_buf != sg_origin)
2583 * The main loop -- while we have blobs with lines whose true origin
2584 * is still unknown, pick one blob, and allow its lines to pass blames
2585 * to its parents. */
2586 void assign_blame(struct blame_scoreboard *sb, int opt)
2588 struct rev_info *revs = sb->revs;
2589 struct commit *commit = prio_queue_get(&sb->commits);
2592 struct blame_entry *ent;
2593 struct blame_origin *suspect = get_blame_suspects(commit);
2595 /* find one suspect to break down */
2596 while (suspect && !suspect->suspects)
2597 suspect = suspect->next;
2600 commit = prio_queue_get(&sb->commits);
2604 assert(commit == suspect->commit);
2607 * We will use this suspect later in the loop,
2608 * so hold onto it in the meantime.
2610 blame_origin_incref(suspect);
2611 parse_commit(commit);
2613 (!(commit->object.flags & UNINTERESTING) &&
2614 !(revs->max_age != -1 && commit->date < revs->max_age)))
2615 pass_blame(sb, suspect, opt);
2617 commit->object.flags |= UNINTERESTING;
2618 if (commit->object.parsed)
2619 mark_parents_uninteresting(commit);
2621 /* treat root commit as boundary */
2622 if (!commit->parents && !sb->show_root)
2623 commit->object.flags |= UNINTERESTING;
2625 /* Take responsibility for the remaining entries */
2626 ent = suspect->suspects;
2628 suspect->guilty = 1;
2630 struct blame_entry *next = ent->next;
2631 if (sb->found_guilty_entry)
2632 sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
2637 ent->next = sb->ent;
2638 sb->ent = suspect->suspects;
2639 suspect->suspects = NULL;
2643 blame_origin_decref(suspect);
2645 if (sb->debug) /* sanity */
2646 sanity_check_refcnt(sb);
2651 * To allow quick access to the contents of nth line in the
2652 * final image, prepare an index in the scoreboard.
2654 static int prepare_lines(struct blame_scoreboard *sb)
2656 sb->num_lines = find_line_starts(&sb->lineno, sb->final_buf,
2657 sb->final_buf_size);
2658 return sb->num_lines;
2661 static struct commit *find_single_final(struct rev_info *revs,
2662 const char **name_p)
2665 struct commit *found = NULL;
2666 const char *name = NULL;
2668 for (i = 0; i < revs->pending.nr; i++) {
2669 struct object *obj = revs->pending.objects[i].item;
2670 if (obj->flags & UNINTERESTING)
2672 obj = deref_tag(revs->repo, obj, NULL, 0);
2673 if (!obj || obj->type != OBJ_COMMIT)
2674 die("Non commit %s?", revs->pending.objects[i].name);
2676 die("More than one commit to dig from %s and %s?",
2677 revs->pending.objects[i].name, name);
2678 found = (struct commit *)obj;
2679 name = revs->pending.objects[i].name;
2682 *name_p = xstrdup_or_null(name);
2686 static struct commit *dwim_reverse_initial(struct rev_info *revs,
2687 const char **name_p)
2690 * DWIM "git blame --reverse ONE -- PATH" as
2691 * "git blame --reverse ONE..HEAD -- PATH" but only do so
2692 * when it makes sense.
2695 struct commit *head_commit;
2696 struct object_id head_oid;
2698 if (revs->pending.nr != 1)
2701 /* Is that sole rev a committish? */
2702 obj = revs->pending.objects[0].item;
2703 obj = deref_tag(revs->repo, obj, NULL, 0);
2704 if (!obj || obj->type != OBJ_COMMIT)
2707 /* Do we have HEAD? */
2708 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
2710 head_commit = lookup_commit_reference_gently(revs->repo,
2715 /* Turn "ONE" into "ONE..HEAD" then */
2716 obj->flags |= UNINTERESTING;
2717 add_pending_object(revs, &head_commit->object, "HEAD");
2720 *name_p = revs->pending.objects[0].name;
2721 return (struct commit *)obj;
2724 static struct commit *find_single_initial(struct rev_info *revs,
2725 const char **name_p)
2728 struct commit *found = NULL;
2729 const char *name = NULL;
2732 * There must be one and only one negative commit, and it must be
2735 for (i = 0; i < revs->pending.nr; i++) {
2736 struct object *obj = revs->pending.objects[i].item;
2737 if (!(obj->flags & UNINTERESTING))
2739 obj = deref_tag(revs->repo, obj, NULL, 0);
2740 if (!obj || obj->type != OBJ_COMMIT)
2741 die("Non commit %s?", revs->pending.objects[i].name);
2743 die("More than one commit to dig up from, %s and %s?",
2744 revs->pending.objects[i].name, name);
2745 found = (struct commit *) obj;
2746 name = revs->pending.objects[i].name;
2750 found = dwim_reverse_initial(revs, &name);
2752 die("No commit to dig up from?");
2755 *name_p = xstrdup(name);
2759 void init_scoreboard(struct blame_scoreboard *sb)
2761 memset(sb, 0, sizeof(struct blame_scoreboard));
2762 sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
2763 sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
2766 void setup_scoreboard(struct blame_scoreboard *sb,
2767 struct blame_origin **orig)
2769 const char *final_commit_name = NULL;
2770 struct blame_origin *o;
2771 struct commit *final_commit = NULL;
2772 enum object_type type;
2774 init_blame_suspects(&blame_suspects);
2776 if (sb->reverse && sb->contents_from)
2777 die(_("--contents and --reverse do not blend well."));
2780 BUG("repo is NULL");
2783 sb->final = find_single_final(sb->revs, &final_commit_name);
2784 sb->commits.compare = compare_commits_by_commit_date;
2786 sb->final = find_single_initial(sb->revs, &final_commit_name);
2787 sb->commits.compare = compare_commits_by_reverse_commit_date;
2790 if (sb->final && sb->contents_from)
2791 die(_("cannot use --contents with final commit object name"));
2793 if (sb->reverse && sb->revs->first_parent_only)
2794 sb->revs->children.name = NULL;
2798 * "--not A B -- path" without anything positive;
2799 * do not default to HEAD, but use the working tree
2803 sb->final = fake_working_tree_commit(sb->repo,
2805 sb->path, sb->contents_from);
2806 add_pending_object(sb->revs, &(sb->final->object), ":");
2809 if (sb->reverse && sb->revs->first_parent_only) {
2810 final_commit = find_single_final(sb->revs, NULL);
2812 die(_("--reverse and --first-parent together require specified latest commit"));
2816 * If we have bottom, this will mark the ancestors of the
2817 * bottom commits we would reach while traversing as
2820 if (prepare_revision_walk(sb->revs))
2821 die(_("revision walk setup failed"));
2823 if (sb->reverse && sb->revs->first_parent_only) {
2824 struct commit *c = final_commit;
2826 sb->revs->children.name = "children";
2827 while (c->parents &&
2828 !oideq(&c->object.oid, &sb->final->object.oid)) {
2829 struct commit_list *l = xcalloc(1, sizeof(*l));
2832 if (add_decoration(&sb->revs->children,
2833 &c->parents->item->object, l))
2834 BUG("not unique item in first-parent chain");
2835 c = c->parents->item;
2838 if (!oideq(&c->object.oid, &sb->final->object.oid))
2839 die(_("--reverse --first-parent together require range along first-parent chain"));
2842 if (is_null_oid(&sb->final->object.oid)) {
2843 o = get_blame_suspects(sb->final);
2844 sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
2845 sb->final_buf_size = o->file.size;
2848 o = get_origin(sb->final, sb->path);
2849 if (fill_blob_sha1_and_mode(sb->repo, o))
2850 die(_("no such path %s in %s"), sb->path, final_commit_name);
2852 if (sb->revs->diffopt.flags.allow_textconv &&
2853 textconv_object(sb->repo, sb->path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
2854 &sb->final_buf_size))
2857 sb->final_buf = read_object_file(&o->blob_oid, &type,
2858 &sb->final_buf_size);
2861 die(_("cannot read blob %s for path %s"),
2862 oid_to_hex(&o->blob_oid),
2865 sb->num_read_blob++;
2871 free((char *)final_commit_name);
2876 struct blame_entry *blame_entry_prepend(struct blame_entry *head,
2877 long start, long end,
2878 struct blame_origin *o)
2880 struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
2881 new_head->lno = start;
2882 new_head->num_lines = end - start;
2883 new_head->suspect = o;
2884 new_head->s_lno = start;
2885 new_head->next = head;
2886 blame_origin_incref(o);
2890 void setup_blame_bloom_data(struct blame_scoreboard *sb)
2892 struct blame_bloom_data *bd;
2893 struct bloom_filter_settings *bs;
2895 if (!sb->repo->objects->commit_graph)
2898 bs = get_bloom_filter_settings(sb->repo);
2902 bd = xmalloc(sizeof(struct blame_bloom_data));
2908 ALLOC_ARRAY(bd->keys, bd->alloc);
2910 add_bloom_key(bd, sb->path);
2912 sb->bloom_data = bd;
2915 void cleanup_scoreboard(struct blame_scoreboard *sb)
2917 if (sb->bloom_data) {
2919 for (i = 0; i < sb->bloom_data->nr; i++) {
2920 free(sb->bloom_data->keys[i]->hashes);
2921 free(sb->bloom_data->keys[i]);
2923 free(sb->bloom_data->keys);
2924 FREE_AND_NULL(sb->bloom_data);
2926 trace2_data_intmax("blame", sb->repo,
2927 "bloom/queries", bloom_count_queries);
2928 trace2_data_intmax("blame", sb->repo,
2929 "bloom/response-no", bloom_count_no);