blame: use a helper function in blame_chunk()
[git/git.git] / blame.c
CommitLineData
072bf432
JS
1#include "cache.h"
2#include "refs.h"
cbd53a21 3#include "object-store.h"
072bf432 4#include "cache-tree.h"
b543bb1c
JS
5#include "mergesort.h"
6#include "diff.h"
7#include "diffcore.h"
09002f1b 8#include "tag.h"
f5dd754c 9#include "blame.h"
14ba97f8 10#include "alloc.h"
4e0df4e6
NTND
11#include "commit-slab.h"
12
13define_commit_slab(blame_suspects, struct blame_origin *);
14static struct blame_suspects blame_suspects;
15
16struct blame_origin *get_blame_suspects(struct commit *commit)
17{
18 struct blame_origin **result;
19
20 result = blame_suspects_peek(&blame_suspects, commit);
21
22 return result ? *result : NULL;
23}
24
25static void set_blame_suspects(struct commit *commit, struct blame_origin *origin)
26{
27 *blame_suspects_at(&blame_suspects, commit) = origin;
28}
f5dd754c
JS
29
30void blame_origin_decref(struct blame_origin *o)
31{
32 if (o && --o->refcnt <= 0) {
33 struct blame_origin *p, *l = NULL;
34 if (o->previous)
35 blame_origin_decref(o->previous);
36 free(o->file.ptr);
37 /* Should be present exactly once in commit chain */
4e0df4e6 38 for (p = get_blame_suspects(o->commit); p; l = p, p = p->next) {
f5dd754c
JS
39 if (p == o) {
40 if (l)
41 l->next = p->next;
42 else
4e0df4e6 43 set_blame_suspects(o->commit, p->next);
f5dd754c
JS
44 free(o);
45 return;
46 }
47 }
48 die("internal error in blame_origin_decref");
49 }
50}
51
52/*
53 * Given a commit and a path in it, create a new origin structure.
54 * The callers that add blame to the scoreboard should use
55 * get_origin() to obtain shared, refcounted copy instead of calling
56 * this function directly.
57 */
072bf432 58static struct blame_origin *make_origin(struct commit *commit, const char *path)
f5dd754c
JS
59{
60 struct blame_origin *o;
61 FLEX_ALLOC_STR(o, path, path);
62 o->commit = commit;
63 o->refcnt = 1;
4e0df4e6
NTND
64 o->next = get_blame_suspects(commit);
65 set_blame_suspects(commit, o);
f5dd754c
JS
66 return o;
67}
68
69/*
70 * Locate an existing origin or create a new one.
71 * This moves the origin to front position in the commit util list.
72 */
09002f1b 73static struct blame_origin *get_origin(struct commit *commit, const char *path)
f5dd754c
JS
74{
75 struct blame_origin *o, *l;
76
4e0df4e6 77 for (o = get_blame_suspects(commit), l = NULL; o; l = o, o = o->next) {
f5dd754c
JS
78 if (!strcmp(o->path, path)) {
79 /* bump to front */
80 if (l) {
81 l->next = o->next;
4e0df4e6
NTND
82 o->next = get_blame_suspects(commit);
83 set_blame_suspects(commit, o);
f5dd754c
JS
84 }
85 return blame_origin_incref(o);
86 }
87 }
88 return make_origin(commit, path);
89}
072bf432
JS
90
91
92
a470beea 93static void verify_working_tree_path(struct repository *r,
ecbbc0a5 94 struct commit *work_tree, const char *path)
072bf432
JS
95{
96 struct commit_list *parents;
97 int pos;
98
99 for (parents = work_tree->parents; parents; parents = parents->next) {
100 const struct object_id *commit_oid = &parents->item->object.oid;
101 struct object_id blob_oid;
102 unsigned mode;
103
916bc35b 104 if (!get_tree_entry(commit_oid, path, &blob_oid, &mode) &&
a470beea 105 oid_object_info(r, &blob_oid, NULL) == OBJ_BLOB)
072bf432
JS
106 return;
107 }
108
a470beea 109 pos = index_name_pos(r->index, path, strlen(path));
072bf432
JS
110 if (pos >= 0)
111 ; /* path is in the index */
a470beea
NTND
112 else if (-1 - pos < r->index->cache_nr &&
113 !strcmp(r->index->cache[-1 - pos]->name, path))
072bf432
JS
114 ; /* path is in the index, unmerged */
115 else
116 die("no such path '%s' in HEAD", path);
117}
118
fb998eae
NTND
119static struct commit_list **append_parent(struct repository *r,
120 struct commit_list **tail,
121 const struct object_id *oid)
072bf432
JS
122{
123 struct commit *parent;
124
fb998eae 125 parent = lookup_commit_reference(r, oid);
072bf432
JS
126 if (!parent)
127 die("no such commit %s", oid_to_hex(oid));
128 return &commit_list_insert(parent, tail)->next;
129}
130
fb998eae
NTND
131static void append_merge_parents(struct repository *r,
132 struct commit_list **tail)
072bf432
JS
133{
134 int merge_head;
135 struct strbuf line = STRBUF_INIT;
136
fb998eae 137 merge_head = open(git_path_merge_head(r), O_RDONLY);
072bf432
JS
138 if (merge_head < 0) {
139 if (errno == ENOENT)
140 return;
102de880 141 die("cannot open '%s' for reading",
fb998eae 142 git_path_merge_head(r));
072bf432
JS
143 }
144
145 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
146 struct object_id oid;
147 if (line.len < GIT_SHA1_HEXSZ || get_oid_hex(line.buf, &oid))
102de880 148 die("unknown line in '%s': %s",
fb998eae
NTND
149 git_path_merge_head(r), line.buf);
150 tail = append_parent(r, tail, &oid);
072bf432
JS
151 }
152 close(merge_head);
153 strbuf_release(&line);
154}
155
156/*
157 * This isn't as simple as passing sb->buf and sb->len, because we
158 * want to transfer ownership of the buffer to the commit (so we
159 * must use detach).
160 */
fb998eae
NTND
161static void set_commit_buffer_from_strbuf(struct repository *r,
162 struct commit *c,
163 struct strbuf *sb)
072bf432
JS
164{
165 size_t len;
166 void *buf = strbuf_detach(sb, &len);
fb998eae 167 set_commit_buffer(r, c, buf, len);
072bf432
JS
168}
169
170/*
171 * Prepare a dummy commit that represents the work tree (or staged) item.
172 * Note that annotating work tree item never works in the reverse.
173 */
a470beea 174static struct commit *fake_working_tree_commit(struct repository *r,
ecbbc0a5 175 struct diff_options *opt,
09002f1b
JS
176 const char *path,
177 const char *contents_from)
072bf432
JS
178{
179 struct commit *commit;
180 struct blame_origin *origin;
181 struct commit_list **parent_tail, *parent;
182 struct object_id head_oid;
183 struct strbuf buf = STRBUF_INIT;
184 const char *ident;
185 time_t now;
a849735b 186 int len;
072bf432
JS
187 struct cache_entry *ce;
188 unsigned mode;
189 struct strbuf msg = STRBUF_INIT;
190
a470beea 191 read_index(r->index);
072bf432 192 time(&now);
fb998eae 193 commit = alloc_commit_node(r);
072bf432
JS
194 commit->object.parsed = 1;
195 commit->date = now;
196 parent_tail = &commit->parents;
197
49e61479 198 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
072bf432
JS
199 die("no such ref: HEAD");
200
fb998eae
NTND
201 parent_tail = append_parent(r, parent_tail, &head_oid);
202 append_merge_parents(r, parent_tail);
a470beea 203 verify_working_tree_path(r, commit, path);
072bf432
JS
204
205 origin = make_origin(commit, path);
206
207 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
208 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
209 for (parent = commit->parents; parent; parent = parent->next)
210 strbuf_addf(&msg, "parent %s\n",
211 oid_to_hex(&parent->item->object.oid));
212 strbuf_addf(&msg,
213 "author %s\n"
214 "committer %s\n\n"
215 "Version of %s from %s\n",
216 ident, ident, path,
217 (!contents_from ? path :
218 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
fb998eae 219 set_commit_buffer_from_strbuf(r, commit, &msg);
072bf432
JS
220
221 if (!contents_from || strcmp("-", contents_from)) {
222 struct stat st;
223 const char *read_from;
224 char *buf_ptr;
225 unsigned long buf_len;
226
227 if (contents_from) {
228 if (stat(contents_from, &st) < 0)
229 die_errno("Cannot stat '%s'", contents_from);
230 read_from = contents_from;
231 }
232 else {
233 if (lstat(path, &st) < 0)
234 die_errno("Cannot lstat '%s'", path);
235 read_from = path;
236 }
237 mode = canon_mode(st.st_mode);
238
239 switch (st.st_mode & S_IFMT) {
240 case S_IFREG:
0d1e0e78 241 if (opt->flags.allow_textconv &&
6afaf807 242 textconv_object(r, read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
072bf432
JS
243 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
244 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
245 die_errno("cannot open or read '%s'", read_from);
246 break;
247 case S_IFLNK:
248 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
249 die_errno("cannot readlink '%s'", read_from);
250 break;
251 default:
252 die("unsupported file type %s", read_from);
253 }
254 }
255 else {
256 /* Reading from stdin */
257 mode = 0;
258 if (strbuf_read(&buf, 0, 0) < 0)
259 die_errno("failed to read from stdin");
260 }
a470beea 261 convert_to_git(r->index, path, buf.buf, buf.len, &buf, 0);
072bf432
JS
262 origin->file.ptr = buf.buf;
263 origin->file.size = buf.len;
829e5c3b 264 pretend_object_file(buf.buf, buf.len, OBJ_BLOB, &origin->blob_oid);
072bf432
JS
265
266 /*
267 * Read the current index, replace the path entry with
268 * origin->blob_sha1 without mucking with its mode or type
269 * bits; we are not going to write this index out -- we just
270 * want to run "diff-index --cached".
271 */
a470beea
NTND
272 discard_index(r->index);
273 read_index(r->index);
072bf432
JS
274
275 len = strlen(path);
276 if (!mode) {
a470beea 277 int pos = index_name_pos(r->index, path, len);
072bf432 278 if (0 <= pos)
a470beea 279 mode = r->index->cache[pos]->ce_mode;
072bf432
JS
280 else
281 /* Let's not bother reading from HEAD tree */
282 mode = S_IFREG | 0644;
283 }
a470beea 284 ce = make_empty_cache_entry(r->index, len);
072bf432
JS
285 oidcpy(&ce->oid, &origin->blob_oid);
286 memcpy(ce->name, path, len);
287 ce->ce_flags = create_ce_flags(0);
288 ce->ce_namelen = len;
289 ce->ce_mode = create_ce_mode(mode);
a470beea 290 add_index_entry(r->index, ce,
ecbbc0a5 291 ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
072bf432 292
a470beea 293 cache_tree_invalidate_path(r->index, path);
072bf432
JS
294
295 return commit;
296}
b543bb1c
JS
297
298
299
300static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
301 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
302{
303 xpparam_t xpp = {0};
304 xdemitconf_t xecfg = {0};
305 xdemitcb_t ecb = {NULL};
306
307 xpp.flags = xdl_opts;
308 xecfg.hunk_func = hunk_func;
309 ecb.priv = cb_data;
310 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
311}
312
313/*
314 * Given an origin, prepare mmfile_t structure to be used by the
315 * diff machinery
316 */
317static void fill_origin_blob(struct diff_options *opt,
318 struct blame_origin *o, mmfile_t *file, int *num_read_blob)
319{
320 if (!o->file.ptr) {
321 enum object_type type;
322 unsigned long file_size;
323
324 (*num_read_blob)++;
0d1e0e78 325 if (opt->flags.allow_textconv &&
6afaf807
NTND
326 textconv_object(opt->repo, o->path, o->mode,
327 &o->blob_oid, 1, &file->ptr, &file_size))
b543bb1c
JS
328 ;
329 else
b4f5aca4 330 file->ptr = read_object_file(&o->blob_oid, &type,
331 &file_size);
b543bb1c
JS
332 file->size = file_size;
333
334 if (!file->ptr)
335 die("Cannot read blob %s for path %s",
336 oid_to_hex(&o->blob_oid),
337 o->path);
338 o->file = *file;
339 }
340 else
341 *file = o->file;
342}
343
344static void drop_origin_blob(struct blame_origin *o)
345{
ce528de0 346 FREE_AND_NULL(o->file.ptr);
b543bb1c
JS
347}
348
349/*
350 * Any merge of blames happens on lists of blames that arrived via
351 * different parents in a single suspect. In this case, we want to
352 * sort according to the suspect line numbers as opposed to the final
353 * image line numbers. The function body is somewhat longish because
354 * it avoids unnecessary writes.
355 */
356
357static struct blame_entry *blame_merge(struct blame_entry *list1,
358 struct blame_entry *list2)
359{
360 struct blame_entry *p1 = list1, *p2 = list2,
361 **tail = &list1;
362
363 if (!p1)
364 return p2;
365 if (!p2)
366 return p1;
367
368 if (p1->s_lno <= p2->s_lno) {
369 do {
370 tail = &p1->next;
371 if ((p1 = *tail) == NULL) {
372 *tail = p2;
373 return list1;
374 }
375 } while (p1->s_lno <= p2->s_lno);
376 }
377 for (;;) {
378 *tail = p2;
379 do {
380 tail = &p2->next;
381 if ((p2 = *tail) == NULL) {
382 *tail = p1;
383 return list1;
384 }
385 } while (p1->s_lno > p2->s_lno);
386 *tail = p1;
387 do {
388 tail = &p1->next;
389 if ((p1 = *tail) == NULL) {
390 *tail = p2;
391 return list1;
392 }
393 } while (p1->s_lno <= p2->s_lno);
394 }
395}
396
397static void *get_next_blame(const void *p)
398{
399 return ((struct blame_entry *)p)->next;
400}
401
402static void set_next_blame(void *p1, void *p2)
403{
404 ((struct blame_entry *)p1)->next = p2;
405}
406
407/*
408 * Final image line numbers are all different, so we don't need a
409 * three-way comparison here.
410 */
411
412static int compare_blame_final(const void *p1, const void *p2)
413{
414 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
415 ? 1 : -1;
416}
417
418static int compare_blame_suspect(const void *p1, const void *p2)
419{
420 const struct blame_entry *s1 = p1, *s2 = p2;
421 /*
422 * to allow for collating suspects, we sort according to the
423 * respective pointer value as the primary sorting criterion.
424 * The actual relation is pretty unimportant as long as it
425 * establishes a total order. Comparing as integers gives us
426 * that.
427 */
428 if (s1->suspect != s2->suspect)
429 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
430 if (s1->s_lno == s2->s_lno)
431 return 0;
432 return s1->s_lno > s2->s_lno ? 1 : -1;
433}
434
435void blame_sort_final(struct blame_scoreboard *sb)
436{
437 sb->ent = llist_mergesort(sb->ent, get_next_blame, set_next_blame,
438 compare_blame_final);
439}
440
09002f1b
JS
441static int compare_commits_by_reverse_commit_date(const void *a,
442 const void *b,
443 void *c)
444{
445 return -compare_commits_by_commit_date(a, b, c);
446}
447
b543bb1c
JS
448/*
449 * For debugging -- origin is refcounted, and this asserts that
450 * we do not underflow.
451 */
452static void sanity_check_refcnt(struct blame_scoreboard *sb)
453{
454 int baa = 0;
455 struct blame_entry *ent;
456
457 for (ent = sb->ent; ent; ent = ent->next) {
458 /* Nobody should have zero or negative refcnt */
459 if (ent->suspect->refcnt <= 0) {
460 fprintf(stderr, "%s in %s has negative refcnt %d\n",
461 ent->suspect->path,
462 oid_to_hex(&ent->suspect->commit->object.oid),
463 ent->suspect->refcnt);
464 baa = 1;
465 }
466 }
467 if (baa)
468 sb->on_sanity_fail(sb, baa);
469}
470
471/*
472 * If two blame entries that are next to each other came from
473 * contiguous lines in the same origin (i.e. <commit, path> pair),
474 * merge them together.
475 */
476void blame_coalesce(struct blame_scoreboard *sb)
477{
478 struct blame_entry *ent, *next;
479
480 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
481 if (ent->suspect == next->suspect &&
482 ent->s_lno + ent->num_lines == next->s_lno) {
483 ent->num_lines += next->num_lines;
484 ent->next = next->next;
485 blame_origin_decref(next->suspect);
486 free(next);
487 ent->score = 0;
488 next = ent; /* again */
489 }
490 }
491
492 if (sb->debug) /* sanity */
493 sanity_check_refcnt(sb);
494}
495
496/*
497 * Merge the given sorted list of blames into a preexisting origin.
498 * If there were no previous blames to that commit, it is entered into
499 * the commit priority queue of the score board.
500 */
501
502static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porigin,
503 struct blame_entry *sorted)
504{
505 if (porigin->suspects)
506 porigin->suspects = blame_merge(porigin->suspects, sorted);
507 else {
508 struct blame_origin *o;
4e0df4e6 509 for (o = get_blame_suspects(porigin->commit); o; o = o->next) {
b543bb1c
JS
510 if (o->suspects) {
511 porigin->suspects = sorted;
512 return;
513 }
514 }
515 porigin->suspects = sorted;
516 prio_queue_put(&sb->commits, porigin->commit);
517 }
518}
519
09002f1b
JS
520/*
521 * Fill the blob_sha1 field of an origin if it hasn't, so that later
522 * call to fill_origin_blob() can use it to locate the data. blob_sha1
523 * for an origin is also used to pass the blame for the entire file to
524 * the parent to detect the case where a child's blob is identical to
525 * that of its parent's.
526 *
527 * This also fills origin->mode for corresponding tree path.
528 */
a470beea 529static int fill_blob_sha1_and_mode(struct repository *r,
ecbbc0a5 530 struct blame_origin *origin)
09002f1b
JS
531{
532 if (!is_null_oid(&origin->blob_oid))
533 return 0;
916bc35b 534 if (get_tree_entry(&origin->commit->object.oid, origin->path, &origin->blob_oid, &origin->mode))
09002f1b 535 goto error_out;
a470beea 536 if (oid_object_info(r, &origin->blob_oid, NULL) != OBJ_BLOB)
09002f1b
JS
537 goto error_out;
538 return 0;
539 error_out:
540 oidclr(&origin->blob_oid);
541 origin->mode = S_IFINVALID;
542 return -1;
543}
544
b543bb1c
JS
545/*
546 * We have an origin -- check if the same path exists in the
547 * parent and return an origin structure to represent it.
548 */
e6757652
NTND
549static struct blame_origin *find_origin(struct repository *r,
550 struct commit *parent,
551 struct blame_origin *origin)
b543bb1c
JS
552{
553 struct blame_origin *porigin;
554 struct diff_options diff_opts;
555 const char *paths[2];
556
557 /* First check any existing origins */
4e0df4e6 558 for (porigin = get_blame_suspects(parent); porigin; porigin = porigin->next)
b543bb1c
JS
559 if (!strcmp(porigin->path, origin->path)) {
560 /*
561 * The same path between origin and its parent
562 * without renaming -- the most common case.
563 */
564 return blame_origin_incref (porigin);
565 }
566
567 /* See if the origin->path is different between parent
568 * and origin first. Most of the time they are the
569 * same and diff-tree is fairly efficient about this.
570 */
e6757652 571 repo_diff_setup(r, &diff_opts);
0d1e0e78 572 diff_opts.flags.recursive = 1;
b543bb1c
JS
573 diff_opts.detect_rename = 0;
574 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
575 paths[0] = origin->path;
576 paths[1] = NULL;
577
578 parse_pathspec(&diff_opts.pathspec,
579 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
580 PATHSPEC_LITERAL_PATH, "", paths);
581 diff_setup_done(&diff_opts);
582
583 if (is_null_oid(&origin->commit->object.oid))
2e27bd77 584 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
b543bb1c 585 else
2e27bd77
DS
586 diff_tree_oid(get_commit_tree_oid(parent),
587 get_commit_tree_oid(origin->commit),
a6f38c10 588 "", &diff_opts);
b543bb1c
JS
589 diffcore_std(&diff_opts);
590
591 if (!diff_queued_diff.nr) {
592 /* The path is the same as parent */
593 porigin = get_origin(parent, origin->path);
594 oidcpy(&porigin->blob_oid, &origin->blob_oid);
595 porigin->mode = origin->mode;
596 } else {
597 /*
598 * Since origin->path is a pathspec, if the parent
599 * commit had it as a directory, we will see a whole
600 * bunch of deletion of files in the directory that we
601 * do not care about.
602 */
603 int i;
604 struct diff_filepair *p = NULL;
605 for (i = 0; i < diff_queued_diff.nr; i++) {
606 const char *name;
607 p = diff_queued_diff.queue[i];
608 name = p->one->path ? p->one->path : p->two->path;
609 if (!strcmp(name, origin->path))
610 break;
611 }
612 if (!p)
613 die("internal error in blame::find_origin");
614 switch (p->status) {
615 default:
616 die("internal error in blame::find_origin (%c)",
617 p->status);
618 case 'M':
619 porigin = get_origin(parent, origin->path);
620 oidcpy(&porigin->blob_oid, &p->one->oid);
621 porigin->mode = p->one->mode;
622 break;
623 case 'A':
624 case 'T':
625 /* Did not exist in parent, or type changed */
626 break;
627 }
628 }
629 diff_flush(&diff_opts);
630 clear_pathspec(&diff_opts.pathspec);
631 return porigin;
632}
633
634/*
635 * We have an origin -- find the path that corresponds to it in its
636 * parent and return an origin structure to represent it.
637 */
e6757652
NTND
638static struct blame_origin *find_rename(struct repository *r,
639 struct commit *parent,
640 struct blame_origin *origin)
b543bb1c
JS
641{
642 struct blame_origin *porigin = NULL;
643 struct diff_options diff_opts;
644 int i;
645
e6757652 646 repo_diff_setup(r, &diff_opts);
0d1e0e78 647 diff_opts.flags.recursive = 1;
b543bb1c
JS
648 diff_opts.detect_rename = DIFF_DETECT_RENAME;
649 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
650 diff_opts.single_follow = origin->path;
651 diff_setup_done(&diff_opts);
652
653 if (is_null_oid(&origin->commit->object.oid))
2e27bd77 654 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
b543bb1c 655 else
2e27bd77
DS
656 diff_tree_oid(get_commit_tree_oid(parent),
657 get_commit_tree_oid(origin->commit),
a6f38c10 658 "", &diff_opts);
b543bb1c
JS
659 diffcore_std(&diff_opts);
660
661 for (i = 0; i < diff_queued_diff.nr; i++) {
662 struct diff_filepair *p = diff_queued_diff.queue[i];
663 if ((p->status == 'R' || p->status == 'C') &&
664 !strcmp(p->two->path, origin->path)) {
665 porigin = get_origin(parent, p->one->path);
666 oidcpy(&porigin->blob_oid, &p->one->oid);
667 porigin->mode = p->one->mode;
668 break;
669 }
670 }
671 diff_flush(&diff_opts);
672 clear_pathspec(&diff_opts.pathspec);
673 return porigin;
674}
675
676/*
677 * Append a new blame entry to a given output queue.
678 */
679static void add_blame_entry(struct blame_entry ***queue,
680 const struct blame_entry *src)
681{
682 struct blame_entry *e = xmalloc(sizeof(*e));
683 memcpy(e, src, sizeof(*e));
684 blame_origin_incref(e->suspect);
685
686 e->next = **queue;
687 **queue = e;
688 *queue = &e->next;
689}
690
691/*
692 * src typically is on-stack; we want to copy the information in it to
693 * a malloced blame_entry that gets added to the given queue. The
694 * origin of dst loses a refcnt.
695 */
696static void dup_entry(struct blame_entry ***queue,
697 struct blame_entry *dst, struct blame_entry *src)
698{
699 blame_origin_incref(src->suspect);
700 blame_origin_decref(dst->suspect);
701 memcpy(dst, src, sizeof(*src));
702 dst->next = **queue;
703 **queue = dst;
704 *queue = &dst->next;
705}
706
707const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
708{
709 return sb->final_buf + sb->lineno[lno];
710}
711
712/*
713 * It is known that lines between tlno to same came from parent, and e
714 * has an overlap with that range. it also is known that parent's
715 * line plno corresponds to e's line tlno.
716 *
717 * <---- e ----->
718 * <------>
719 * <------------>
720 * <------------>
721 * <------------------>
722 *
723 * Split e into potentially three parts; before this chunk, the chunk
724 * to be blamed for the parent, and after that portion.
725 */
726static void split_overlap(struct blame_entry *split,
727 struct blame_entry *e,
728 int tlno, int plno, int same,
729 struct blame_origin *parent)
730{
731 int chunk_end_lno;
732 memset(split, 0, sizeof(struct blame_entry [3]));
733
734 if (e->s_lno < tlno) {
735 /* there is a pre-chunk part not blamed on parent */
736 split[0].suspect = blame_origin_incref(e->suspect);
737 split[0].lno = e->lno;
738 split[0].s_lno = e->s_lno;
739 split[0].num_lines = tlno - e->s_lno;
740 split[1].lno = e->lno + tlno - e->s_lno;
741 split[1].s_lno = plno;
742 }
743 else {
744 split[1].lno = e->lno;
745 split[1].s_lno = plno + (e->s_lno - tlno);
746 }
747
748 if (same < e->s_lno + e->num_lines) {
749 /* there is a post-chunk part not blamed on parent */
750 split[2].suspect = blame_origin_incref(e->suspect);
751 split[2].lno = e->lno + (same - e->s_lno);
752 split[2].s_lno = e->s_lno + (same - e->s_lno);
753 split[2].num_lines = e->s_lno + e->num_lines - same;
754 chunk_end_lno = split[2].lno;
755 }
756 else
757 chunk_end_lno = e->lno + e->num_lines;
758 split[1].num_lines = chunk_end_lno - split[1].lno;
759
760 /*
761 * if it turns out there is nothing to blame the parent for,
762 * forget about the splitting. !split[1].suspect signals this.
763 */
764 if (split[1].num_lines < 1)
765 return;
766 split[1].suspect = blame_origin_incref(parent);
767}
768
769/*
770 * split_overlap() divided an existing blame e into up to three parts
771 * in split. Any assigned blame is moved to queue to
772 * reflect the split.
773 */
774static void split_blame(struct blame_entry ***blamed,
775 struct blame_entry ***unblamed,
776 struct blame_entry *split,
777 struct blame_entry *e)
778{
779 if (split[0].suspect && split[2].suspect) {
780 /* The first part (reuse storage for the existing entry e) */
781 dup_entry(unblamed, e, &split[0]);
782
783 /* The last part -- me */
784 add_blame_entry(unblamed, &split[2]);
785
786 /* ... and the middle part -- parent */
787 add_blame_entry(blamed, &split[1]);
788 }
789 else if (!split[0].suspect && !split[2].suspect)
790 /*
791 * The parent covers the entire area; reuse storage for
792 * e and replace it with the parent.
793 */
794 dup_entry(blamed, e, &split[1]);
795 else if (split[0].suspect) {
796 /* me and then parent */
797 dup_entry(unblamed, e, &split[0]);
798 add_blame_entry(blamed, &split[1]);
799 }
800 else {
801 /* parent and then me */
802 dup_entry(blamed, e, &split[1]);
803 add_blame_entry(unblamed, &split[2]);
804 }
805}
806
807/*
808 * After splitting the blame, the origins used by the
809 * on-stack blame_entry should lose one refcnt each.
810 */
811static void decref_split(struct blame_entry *split)
812{
813 int i;
814
815 for (i = 0; i < 3; i++)
816 blame_origin_decref(split[i].suspect);
817}
818
819/*
820 * reverse_blame reverses the list given in head, appending tail.
821 * That allows us to build lists in reverse order, then reverse them
822 * afterwards. This can be faster than building the list in proper
823 * order right away. The reason is that building in proper order
824 * requires writing a link in the _previous_ element, while building
825 * in reverse order just requires placing the list head into the
826 * _current_ element.
827 */
828
829static struct blame_entry *reverse_blame(struct blame_entry *head,
830 struct blame_entry *tail)
831{
832 while (head) {
833 struct blame_entry *next = head->next;
834 head->next = tail;
835 tail = head;
836 head = next;
837 }
838 return tail;
839}
840
55f808fb
BR
841/*
842 * Splits a blame entry into two entries at 'len' lines. The original 'e'
843 * consists of len lines, i.e. [e->lno, e->lno + len), and the second part,
844 * which is returned, consists of the remainder: [e->lno + len, e->lno +
845 * e->num_lines). The caller needs to sort out the reference counting for the
846 * new entry's suspect.
847 */
848static struct blame_entry *split_blame_at(struct blame_entry *e, int len,
849 struct blame_origin *new_suspect)
850{
851 struct blame_entry *n = xcalloc(1, sizeof(struct blame_entry));
852
853 n->suspect = new_suspect;
854 n->lno = e->lno + len;
855 n->s_lno = e->s_lno + len;
856 n->num_lines = e->num_lines - len;
857 e->num_lines = len;
858 e->score = 0;
859 return n;
860}
861
b543bb1c
JS
862/*
863 * Process one hunk from the patch between the current suspect for
864 * blame_entry e and its parent. This first blames any unfinished
865 * entries before the chunk (which is where target and parent start
866 * differing) on the parent, and then splits blame entries at the
867 * start and at the end of the difference region. Since use of -M and
868 * -C options may lead to overlapping/duplicate source line number
869 * ranges, all we can rely on from sorting/merging is the order of the
870 * first suspect line number.
871 */
872static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
873 int tlno, int offset, int same,
874 struct blame_origin *parent)
875{
876 struct blame_entry *e = **srcq;
877 struct blame_entry *samep = NULL, *diffp = NULL;
878
879 while (e && e->s_lno < tlno) {
880 struct blame_entry *next = e->next;
881 /*
882 * current record starts before differing portion. If
883 * it reaches into it, we need to split it up and
884 * examine the second part separately.
885 */
886 if (e->s_lno + e->num_lines > tlno) {
887 /* Move second half to a new record */
55f808fb
BR
888 struct blame_entry *n;
889
890 n = split_blame_at(e, tlno - e->s_lno, e->suspect);
b543bb1c
JS
891 /* Push new record to diffp */
892 n->next = diffp;
893 diffp = n;
894 } else
895 blame_origin_decref(e->suspect);
896 /* Pass blame for everything before the differing
897 * chunk to the parent */
898 e->suspect = blame_origin_incref(parent);
899 e->s_lno += offset;
900 e->next = samep;
901 samep = e;
902 e = next;
903 }
904 /*
905 * As we don't know how much of a common stretch after this
906 * diff will occur, the currently blamed parts are all that we
907 * can assign to the parent for now.
908 */
909
910 if (samep) {
911 **dstq = reverse_blame(samep, **dstq);
912 *dstq = &samep->next;
913 }
914 /*
915 * Prepend the split off portions: everything after e starts
916 * after the blameable portion.
917 */
918 e = reverse_blame(diffp, e);
919
920 /*
921 * Now retain records on the target while parts are different
922 * from the parent.
923 */
924 samep = NULL;
925 diffp = NULL;
926 while (e && e->s_lno < same) {
927 struct blame_entry *next = e->next;
928
929 /*
930 * If current record extends into sameness, need to split.
931 */
932 if (e->s_lno + e->num_lines > same) {
933 /*
934 * Move second half to a new record to be
935 * processed by later chunks
936 */
55f808fb
BR
937 struct blame_entry *n;
938
939 n = split_blame_at(e, same - e->s_lno,
940 blame_origin_incref(e->suspect));
b543bb1c
JS
941 /* Push new record to samep */
942 n->next = samep;
943 samep = n;
944 }
945 e->next = diffp;
946 diffp = e;
947 e = next;
948 }
949 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
950 /* Move across elements that are in the unblamable portion */
951 if (diffp)
952 *srcq = &diffp->next;
953}
954
955struct blame_chunk_cb_data {
956 struct blame_origin *parent;
957 long offset;
958 struct blame_entry **dstq;
959 struct blame_entry **srcq;
960};
961
962/* diff chunks are from parent to target */
963static int blame_chunk_cb(long start_a, long count_a,
964 long start_b, long count_b, void *data)
965{
966 struct blame_chunk_cb_data *d = data;
967 if (start_a - start_b != d->offset)
968 die("internal error in blame::blame_chunk_cb");
969 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
970 start_b + count_b, d->parent);
971 d->offset = start_a + count_a - (start_b + count_b);
972 return 0;
973}
974
975/*
976 * We are looking at the origin 'target' and aiming to pass blame
977 * for the lines it is suspected to its parent. Run diff to find
978 * which lines came from parent and pass blame for them.
979 */
980static void pass_blame_to_parent(struct blame_scoreboard *sb,
981 struct blame_origin *target,
982 struct blame_origin *parent)
983{
984 mmfile_t file_p, file_o;
985 struct blame_chunk_cb_data d;
986 struct blame_entry *newdest = NULL;
987
988 if (!target->suspects)
989 return; /* nothing remains for this target */
990
991 d.parent = parent;
992 d.offset = 0;
993 d.dstq = &newdest; d.srcq = &target->suspects;
994
995 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
996 fill_origin_blob(&sb->revs->diffopt, target, &file_o, &sb->num_read_blob);
997 sb->num_get_patch++;
998
999 if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
1000 die("unable to generate diff (%s -> %s)",
1001 oid_to_hex(&parent->commit->object.oid),
1002 oid_to_hex(&target->commit->object.oid));
1003 /* The rest are the same as the parent */
1004 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
1005 *d.dstq = NULL;
1006 queue_blames(sb, parent, newdest);
1007
1008 return;
1009}
1010
1011/*
1012 * The lines in blame_entry after splitting blames many times can become
1013 * very small and trivial, and at some point it becomes pointless to
1014 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
1015 * ordinary C program, and it is not worth to say it was copied from
1016 * totally unrelated file in the parent.
1017 *
1018 * Compute how trivial the lines in the blame_entry are.
1019 */
1020unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
1021{
1022 unsigned score;
1023 const char *cp, *ep;
1024
1025 if (e->score)
1026 return e->score;
1027
1028 score = 1;
1029 cp = blame_nth_line(sb, e->lno);
1030 ep = blame_nth_line(sb, e->lno + e->num_lines);
1031 while (cp < ep) {
1032 unsigned ch = *((unsigned char *)cp);
1033 if (isalnum(ch))
1034 score++;
1035 cp++;
1036 }
1037 e->score = score;
1038 return score;
1039}
1040
1041/*
abeacb25
BW
1042 * best_so_far[] and potential[] are both a split of an existing blame_entry
1043 * that passes blame to the parent. Maintain best_so_far the best split so
1044 * far, by comparing potential and best_so_far and copying potential into
b543bb1c
JS
1045 * bst_so_far as needed.
1046 */
1047static void copy_split_if_better(struct blame_scoreboard *sb,
1048 struct blame_entry *best_so_far,
abeacb25 1049 struct blame_entry *potential)
b543bb1c
JS
1050{
1051 int i;
1052
abeacb25 1053 if (!potential[1].suspect)
b543bb1c
JS
1054 return;
1055 if (best_so_far[1].suspect) {
abeacb25
BW
1056 if (blame_entry_score(sb, &potential[1]) <
1057 blame_entry_score(sb, &best_so_far[1]))
b543bb1c
JS
1058 return;
1059 }
1060
1061 for (i = 0; i < 3; i++)
abeacb25 1062 blame_origin_incref(potential[i].suspect);
b543bb1c 1063 decref_split(best_so_far);
abeacb25 1064 memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
b543bb1c
JS
1065}
1066
1067/*
1068 * We are looking at a part of the final image represented by
1069 * ent (tlno and same are offset by ent->s_lno).
1070 * tlno is where we are looking at in the final image.
1071 * up to (but not including) same match preimage.
1072 * plno is where we are looking at in the preimage.
1073 *
1074 * <-------------- final image ---------------------->
1075 * <------ent------>
1076 * ^tlno ^same
1077 * <---------preimage----->
1078 * ^plno
1079 *
1080 * All line numbers are 0-based.
1081 */
1082static void handle_split(struct blame_scoreboard *sb,
1083 struct blame_entry *ent,
1084 int tlno, int plno, int same,
1085 struct blame_origin *parent,
1086 struct blame_entry *split)
1087{
1088 if (ent->num_lines <= tlno)
1089 return;
1090 if (tlno < same) {
abeacb25 1091 struct blame_entry potential[3];
b543bb1c
JS
1092 tlno += ent->s_lno;
1093 same += ent->s_lno;
abeacb25
BW
1094 split_overlap(potential, ent, tlno, plno, same, parent);
1095 copy_split_if_better(sb, split, potential);
1096 decref_split(potential);
b543bb1c
JS
1097 }
1098}
1099
1100struct handle_split_cb_data {
1101 struct blame_scoreboard *sb;
1102 struct blame_entry *ent;
1103 struct blame_origin *parent;
1104 struct blame_entry *split;
1105 long plno;
1106 long tlno;
1107};
1108
1109static int handle_split_cb(long start_a, long count_a,
1110 long start_b, long count_b, void *data)
1111{
1112 struct handle_split_cb_data *d = data;
1113 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1114 d->split);
1115 d->plno = start_a + count_a;
1116 d->tlno = start_b + count_b;
1117 return 0;
1118}
1119
1120/*
1121 * Find the lines from parent that are the same as ent so that
1122 * we can pass blames to it. file_p has the blob contents for
1123 * the parent.
1124 */
1125static void find_copy_in_blob(struct blame_scoreboard *sb,
1126 struct blame_entry *ent,
1127 struct blame_origin *parent,
1128 struct blame_entry *split,
1129 mmfile_t *file_p)
1130{
1131 const char *cp;
1132 mmfile_t file_o;
1133 struct handle_split_cb_data d;
1134
1135 memset(&d, 0, sizeof(d));
1136 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1137 /*
1138 * Prepare mmfile that contains only the lines in ent.
1139 */
1140 cp = blame_nth_line(sb, ent->lno);
1141 file_o.ptr = (char *) cp;
1142 file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
1143
1144 /*
1145 * file_o is a part of final image we are annotating.
1146 * file_p partially may match that image.
1147 */
1148 memset(split, 0, sizeof(struct blame_entry [3]));
1149 if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
1150 die("unable to generate diff (%s)",
1151 oid_to_hex(&parent->commit->object.oid));
1152 /* remainder, if any, all match the preimage */
1153 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1154}
1155
1156/* Move all blame entries from list *source that have a score smaller
1157 * than score_min to the front of list *small.
1158 * Returns a pointer to the link pointing to the old head of the small list.
1159 */
1160
1161static struct blame_entry **filter_small(struct blame_scoreboard *sb,
1162 struct blame_entry **small,
1163 struct blame_entry **source,
1164 unsigned score_min)
1165{
1166 struct blame_entry *p = *source;
1167 struct blame_entry *oldsmall = *small;
1168 while (p) {
1169 if (blame_entry_score(sb, p) <= score_min) {
1170 *small = p;
1171 small = &p->next;
1172 p = *small;
1173 } else {
1174 *source = p;
1175 source = &p->next;
1176 p = *source;
1177 }
1178 }
1179 *small = oldsmall;
1180 *source = NULL;
1181 return small;
1182}
1183
1184/*
1185 * See if lines currently target is suspected for can be attributed to
1186 * parent.
1187 */
1188static void find_move_in_parent(struct blame_scoreboard *sb,
1189 struct blame_entry ***blamed,
1190 struct blame_entry **toosmall,
1191 struct blame_origin *target,
1192 struct blame_origin *parent)
1193{
1194 struct blame_entry *e, split[3];
1195 struct blame_entry *unblamed = target->suspects;
1196 struct blame_entry *leftover = NULL;
1197 mmfile_t file_p;
1198
1199 if (!unblamed)
1200 return; /* nothing remains for this target */
1201
1202 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
1203 if (!file_p.ptr)
1204 return;
1205
1206 /* At each iteration, unblamed has a NULL-terminated list of
1207 * entries that have not yet been tested for blame. leftover
1208 * contains the reversed list of entries that have been tested
1209 * without being assignable to the parent.
1210 */
1211 do {
1212 struct blame_entry **unblamedtail = &unblamed;
1213 struct blame_entry *next;
1214 for (e = unblamed; e; e = next) {
1215 next = e->next;
1216 find_copy_in_blob(sb, e, parent, split, &file_p);
1217 if (split[1].suspect &&
1218 sb->move_score < blame_entry_score(sb, &split[1])) {
1219 split_blame(blamed, &unblamedtail, split, e);
1220 } else {
1221 e->next = leftover;
1222 leftover = e;
1223 }
1224 decref_split(split);
1225 }
1226 *unblamedtail = NULL;
1227 toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
1228 } while (unblamed);
1229 target->suspects = reverse_blame(leftover, NULL);
1230}
1231
1232struct blame_list {
1233 struct blame_entry *ent;
1234 struct blame_entry split[3];
1235};
1236
1237/*
1238 * Count the number of entries the target is suspected for,
1239 * and prepare a list of entry and the best split.
1240 */
1241static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1242 int *num_ents_p)
1243{
1244 struct blame_entry *e;
1245 int num_ents, i;
1246 struct blame_list *blame_list = NULL;
1247
1248 for (e = unblamed, num_ents = 0; e; e = e->next)
1249 num_ents++;
1250 if (num_ents) {
1251 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1252 for (e = unblamed, i = 0; e; e = e->next)
1253 blame_list[i++].ent = e;
1254 }
1255 *num_ents_p = num_ents;
1256 return blame_list;
1257}
1258
1259/*
1260 * For lines target is suspected for, see if we can find code movement
1261 * across file boundary from the parent commit. porigin is the path
1262 * in the parent we already tried.
1263 */
1264static void find_copy_in_parent(struct blame_scoreboard *sb,
1265 struct blame_entry ***blamed,
1266 struct blame_entry **toosmall,
1267 struct blame_origin *target,
1268 struct commit *parent,
1269 struct blame_origin *porigin,
1270 int opt)
1271{
1272 struct diff_options diff_opts;
1273 int i, j;
1274 struct blame_list *blame_list;
1275 int num_ents;
1276 struct blame_entry *unblamed = target->suspects;
1277 struct blame_entry *leftover = NULL;
1278
1279 if (!unblamed)
1280 return; /* nothing remains for this target */
1281
e6757652 1282 repo_diff_setup(sb->repo, &diff_opts);
0d1e0e78 1283 diff_opts.flags.recursive = 1;
b543bb1c
JS
1284 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1285
1286 diff_setup_done(&diff_opts);
1287
1288 /* Try "find copies harder" on new path if requested;
1289 * we do not want to use diffcore_rename() actually to
1290 * match things up; find_copies_harder is set only to
a6f38c10 1291 * force diff_tree_oid() to feed all filepairs to diff_queue,
b543bb1c
JS
1292 * and this code needs to be after diff_setup_done(), which
1293 * usually makes find-copies-harder imply copy detection.
1294 */
1295 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1296 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1297 && (!porigin || strcmp(target->path, porigin->path))))
0d1e0e78 1298 diff_opts.flags.find_copies_harder = 1;
b543bb1c
JS
1299
1300 if (is_null_oid(&target->commit->object.oid))
2e27bd77 1301 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
b543bb1c 1302 else
2e27bd77
DS
1303 diff_tree_oid(get_commit_tree_oid(parent),
1304 get_commit_tree_oid(target->commit),
a6f38c10 1305 "", &diff_opts);
b543bb1c 1306
0d1e0e78 1307 if (!diff_opts.flags.find_copies_harder)
b543bb1c
JS
1308 diffcore_std(&diff_opts);
1309
1310 do {
1311 struct blame_entry **unblamedtail = &unblamed;
1312 blame_list = setup_blame_list(unblamed, &num_ents);
1313
1314 for (i = 0; i < diff_queued_diff.nr; i++) {
1315 struct diff_filepair *p = diff_queued_diff.queue[i];
1316 struct blame_origin *norigin;
1317 mmfile_t file_p;
abeacb25 1318 struct blame_entry potential[3];
b543bb1c
JS
1319
1320 if (!DIFF_FILE_VALID(p->one))
1321 continue; /* does not exist in parent */
1322 if (S_ISGITLINK(p->one->mode))
1323 continue; /* ignore git links */
1324 if (porigin && !strcmp(p->one->path, porigin->path))
1325 /* find_move already dealt with this path */
1326 continue;
1327
1328 norigin = get_origin(parent, p->one->path);
1329 oidcpy(&norigin->blob_oid, &p->one->oid);
1330 norigin->mode = p->one->mode;
1331 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p, &sb->num_read_blob);
1332 if (!file_p.ptr)
1333 continue;
1334
1335 for (j = 0; j < num_ents; j++) {
1336 find_copy_in_blob(sb, blame_list[j].ent,
abeacb25 1337 norigin, potential, &file_p);
b543bb1c 1338 copy_split_if_better(sb, blame_list[j].split,
abeacb25
BW
1339 potential);
1340 decref_split(potential);
b543bb1c
JS
1341 }
1342 blame_origin_decref(norigin);
1343 }
1344
1345 for (j = 0; j < num_ents; j++) {
1346 struct blame_entry *split = blame_list[j].split;
1347 if (split[1].suspect &&
1348 sb->copy_score < blame_entry_score(sb, &split[1])) {
1349 split_blame(blamed, &unblamedtail, split,
1350 blame_list[j].ent);
1351 } else {
1352 blame_list[j].ent->next = leftover;
1353 leftover = blame_list[j].ent;
1354 }
1355 decref_split(split);
1356 }
1357 free(blame_list);
1358 *unblamedtail = NULL;
1359 toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
1360 } while (unblamed);
1361 target->suspects = reverse_blame(leftover, NULL);
1362 diff_flush(&diff_opts);
1363 clear_pathspec(&diff_opts.pathspec);
1364}
1365
1366/*
1367 * The blobs of origin and porigin exactly match, so everything
1368 * origin is suspected for can be blamed on the parent.
1369 */
1370static void pass_whole_blame(struct blame_scoreboard *sb,
1371 struct blame_origin *origin, struct blame_origin *porigin)
1372{
1373 struct blame_entry *e, *suspects;
1374
1375 if (!porigin->file.ptr && origin->file.ptr) {
1376 /* Steal its file */
1377 porigin->file = origin->file;
1378 origin->file.ptr = NULL;
1379 }
1380 suspects = origin->suspects;
1381 origin->suspects = NULL;
1382 for (e = suspects; e; e = e->next) {
1383 blame_origin_incref(porigin);
1384 blame_origin_decref(e->suspect);
1385 e->suspect = porigin;
1386 }
1387 queue_blames(sb, porigin, suspects);
1388}
1389
1390/*
1391 * We pass blame from the current commit to its parents. We keep saying
1392 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1393 * exonerate ourselves.
1394 */
1395static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
1396 int reverse)
1397{
1398 if (!reverse) {
1399 if (revs->first_parent_only &&
1400 commit->parents &&
1401 commit->parents->next) {
1402 free_commit_list(commit->parents->next);
1403 commit->parents->next = NULL;
1404 }
1405 return commit->parents;
1406 }
1407 return lookup_decoration(&revs->children, &commit->object);
1408}
1409
1410static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
1411{
1412 struct commit_list *l = first_scapegoat(revs, commit, reverse);
1413 return commit_list_count(l);
1414}
1415
1416/* Distribute collected unsorted blames to the respected sorted lists
1417 * in the various origins.
1418 */
1419static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
1420{
1421 blamed = llist_mergesort(blamed, get_next_blame, set_next_blame,
1422 compare_blame_suspect);
1423 while (blamed)
1424 {
1425 struct blame_origin *porigin = blamed->suspect;
1426 struct blame_entry *suspects = NULL;
1427 do {
1428 struct blame_entry *next = blamed->next;
1429 blamed->next = suspects;
1430 suspects = blamed;
1431 blamed = next;
1432 } while (blamed && blamed->suspect == porigin);
1433 suspects = reverse_blame(suspects, NULL);
1434 queue_blames(sb, porigin, suspects);
1435 }
1436}
1437
1438#define MAXSG 16
1439
1440static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
1441{
1442 struct rev_info *revs = sb->revs;
1443 int i, pass, num_sg;
1444 struct commit *commit = origin->commit;
1445 struct commit_list *sg;
1446 struct blame_origin *sg_buf[MAXSG];
1447 struct blame_origin *porigin, **sg_origin = sg_buf;
1448 struct blame_entry *toosmall = NULL;
1449 struct blame_entry *blames, **blametail = &blames;
1450
1451 num_sg = num_scapegoats(revs, commit, sb->reverse);
1452 if (!num_sg)
1453 goto finish;
1454 else if (num_sg < ARRAY_SIZE(sg_buf))
1455 memset(sg_buf, 0, sizeof(sg_buf));
1456 else
1457 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1458
1459 /*
1460 * The first pass looks for unrenamed path to optimize for
1461 * common cases, then we look for renames in the second pass.
1462 */
1463 for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
e6757652 1464 struct blame_origin *(*find)(struct repository *, struct commit *, struct blame_origin *);
b543bb1c
JS
1465 find = pass ? find_rename : find_origin;
1466
1467 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1468 i < num_sg && sg;
1469 sg = sg->next, i++) {
1470 struct commit *p = sg->item;
1471 int j, same;
1472
1473 if (sg_origin[i])
1474 continue;
1475 if (parse_commit(p))
1476 continue;
e6757652 1477 porigin = find(sb->repo, p, origin);
b543bb1c
JS
1478 if (!porigin)
1479 continue;
4a7e27e9 1480 if (oideq(&porigin->blob_oid, &origin->blob_oid)) {
b543bb1c
JS
1481 pass_whole_blame(sb, origin, porigin);
1482 blame_origin_decref(porigin);
1483 goto finish;
1484 }
1485 for (j = same = 0; j < i; j++)
1486 if (sg_origin[j] &&
4a7e27e9 1487 oideq(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
b543bb1c
JS
1488 same = 1;
1489 break;
1490 }
1491 if (!same)
1492 sg_origin[i] = porigin;
1493 else
1494 blame_origin_decref(porigin);
1495 }
1496 }
1497
1498 sb->num_commits++;
1499 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1500 i < num_sg && sg;
1501 sg = sg->next, i++) {
1502 struct blame_origin *porigin = sg_origin[i];
1503 if (!porigin)
1504 continue;
1505 if (!origin->previous) {
1506 blame_origin_incref(porigin);
1507 origin->previous = porigin;
1508 }
1509 pass_blame_to_parent(sb, origin, porigin);
1510 if (!origin->suspects)
1511 goto finish;
1512 }
1513
1514 /*
1515 * Optionally find moves in parents' files.
1516 */
1517 if (opt & PICKAXE_BLAME_MOVE) {
1518 filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
1519 if (origin->suspects) {
1520 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1521 i < num_sg && sg;
1522 sg = sg->next, i++) {
1523 struct blame_origin *porigin = sg_origin[i];
1524 if (!porigin)
1525 continue;
1526 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1527 if (!origin->suspects)
1528 break;
1529 }
1530 }
1531 }
1532
1533 /*
1534 * Optionally find copies from parents' files.
1535 */
1536 if (opt & PICKAXE_BLAME_COPY) {
1537 if (sb->copy_score > sb->move_score)
1538 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1539 else if (sb->copy_score < sb->move_score) {
1540 origin->suspects = blame_merge(origin->suspects, toosmall);
1541 toosmall = NULL;
1542 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1543 }
1544 if (!origin->suspects)
1545 goto finish;
1546
1547 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1548 i < num_sg && sg;
1549 sg = sg->next, i++) {
1550 struct blame_origin *porigin = sg_origin[i];
1551 find_copy_in_parent(sb, &blametail, &toosmall,
1552 origin, sg->item, porigin, opt);
1553 if (!origin->suspects)
1554 goto finish;
1555 }
1556 }
1557
1558finish:
1559 *blametail = NULL;
1560 distribute_blame(sb, blames);
1561 /*
1562 * prepend toosmall to origin->suspects
1563 *
1564 * There is no point in sorting: this ends up on a big
1565 * unsorted list in the caller anyway.
1566 */
1567 if (toosmall) {
1568 struct blame_entry **tail = &toosmall;
1569 while (*tail)
1570 tail = &(*tail)->next;
1571 *tail = origin->suspects;
1572 origin->suspects = toosmall;
1573 }
1574 for (i = 0; i < num_sg; i++) {
1575 if (sg_origin[i]) {
1576 drop_origin_blob(sg_origin[i]);
1577 blame_origin_decref(sg_origin[i]);
1578 }
1579 }
1580 drop_origin_blob(origin);
1581 if (sg_buf != sg_origin)
1582 free(sg_origin);
1583}
1584
1585/*
1586 * The main loop -- while we have blobs with lines whose true origin
1587 * is still unknown, pick one blob, and allow its lines to pass blames
1588 * to its parents. */
1589void assign_blame(struct blame_scoreboard *sb, int opt)
1590{
1591 struct rev_info *revs = sb->revs;
1592 struct commit *commit = prio_queue_get(&sb->commits);
1593
1594 while (commit) {
1595 struct blame_entry *ent;
4e0df4e6 1596 struct blame_origin *suspect = get_blame_suspects(commit);
b543bb1c
JS
1597
1598 /* find one suspect to break down */
1599 while (suspect && !suspect->suspects)
1600 suspect = suspect->next;
1601
1602 if (!suspect) {
1603 commit = prio_queue_get(&sb->commits);
1604 continue;
1605 }
1606
1607 assert(commit == suspect->commit);
1608
1609 /*
1610 * We will use this suspect later in the loop,
1611 * so hold onto it in the meantime.
1612 */
1613 blame_origin_incref(suspect);
1614 parse_commit(commit);
1615 if (sb->reverse ||
1616 (!(commit->object.flags & UNINTERESTING) &&
1617 !(revs->max_age != -1 && commit->date < revs->max_age)))
1618 pass_blame(sb, suspect, opt);
1619 else {
1620 commit->object.flags |= UNINTERESTING;
1621 if (commit->object.parsed)
1622 mark_parents_uninteresting(commit);
1623 }
1624 /* treat root commit as boundary */
1625 if (!commit->parents && !sb->show_root)
1626 commit->object.flags |= UNINTERESTING;
1627
1628 /* Take responsibility for the remaining entries */
1629 ent = suspect->suspects;
1630 if (ent) {
1631 suspect->guilty = 1;
1632 for (;;) {
1633 struct blame_entry *next = ent->next;
1634 if (sb->found_guilty_entry)
1635 sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
1636 if (next) {
1637 ent = next;
1638 continue;
1639 }
1640 ent->next = sb->ent;
1641 sb->ent = suspect->suspects;
1642 suspect->suspects = NULL;
1643 break;
1644 }
1645 }
1646 blame_origin_decref(suspect);
1647
1648 if (sb->debug) /* sanity */
1649 sanity_check_refcnt(sb);
1650 }
1651}
09002f1b
JS
1652
1653static const char *get_next_line(const char *start, const char *end)
1654{
1655 const char *nl = memchr(start, '\n', end - start);
1656 return nl ? nl + 1 : end;
1657}
1658
1659/*
1660 * To allow quick access to the contents of nth line in the
1661 * final image, prepare an index in the scoreboard.
1662 */
1663static int prepare_lines(struct blame_scoreboard *sb)
1664{
1665 const char *buf = sb->final_buf;
1666 unsigned long len = sb->final_buf_size;
1667 const char *end = buf + len;
1668 const char *p;
1669 int *lineno;
1670 int num = 0;
1671
1672 for (p = buf; p < end; p = get_next_line(p, end))
1673 num++;
1674
1675 ALLOC_ARRAY(sb->lineno, num + 1);
1676 lineno = sb->lineno;
1677
1678 for (p = buf; p < end; p = get_next_line(p, end))
1679 *lineno++ = p - buf;
1680
1681 *lineno = len;
1682
1683 sb->num_lines = num;
1684 return sb->num_lines;
1685}
1686
1687static struct commit *find_single_final(struct rev_info *revs,
1688 const char **name_p)
1689{
1690 int i;
1691 struct commit *found = NULL;
1692 const char *name = NULL;
1693
1694 for (i = 0; i < revs->pending.nr; i++) {
1695 struct object *obj = revs->pending.objects[i].item;
1696 if (obj->flags & UNINTERESTING)
1697 continue;
fb998eae 1698 obj = deref_tag(revs->repo, obj, NULL, 0);
09002f1b
JS
1699 if (obj->type != OBJ_COMMIT)
1700 die("Non commit %s?", revs->pending.objects[i].name);
1701 if (found)
1702 die("More than one commit to dig from %s and %s?",
1703 revs->pending.objects[i].name, name);
1704 found = (struct commit *)obj;
1705 name = revs->pending.objects[i].name;
1706 }
1707 if (name_p)
9e7d8a9b 1708 *name_p = xstrdup_or_null(name);
09002f1b
JS
1709 return found;
1710}
1711
1712static struct commit *dwim_reverse_initial(struct rev_info *revs,
1713 const char **name_p)
1714{
1715 /*
1716 * DWIM "git blame --reverse ONE -- PATH" as
1717 * "git blame --reverse ONE..HEAD -- PATH" but only do so
1718 * when it makes sense.
1719 */
1720 struct object *obj;
1721 struct commit *head_commit;
583c6a22 1722 struct object_id head_oid;
09002f1b
JS
1723
1724 if (revs->pending.nr != 1)
1725 return NULL;
1726
1727 /* Is that sole rev a committish? */
1728 obj = revs->pending.objects[0].item;
fb998eae 1729 obj = deref_tag(revs->repo, obj, NULL, 0);
09002f1b
JS
1730 if (obj->type != OBJ_COMMIT)
1731 return NULL;
1732
1733 /* Do we have HEAD? */
49e61479 1734 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
09002f1b 1735 return NULL;
fb998eae 1736 head_commit = lookup_commit_reference_gently(revs->repo,
21e1ee8f 1737 &head_oid, 1);
09002f1b
JS
1738 if (!head_commit)
1739 return NULL;
1740
1741 /* Turn "ONE" into "ONE..HEAD" then */
1742 obj->flags |= UNINTERESTING;
1743 add_pending_object(revs, &head_commit->object, "HEAD");
1744
1745 if (name_p)
1746 *name_p = revs->pending.objects[0].name;
1747 return (struct commit *)obj;
1748}
1749
1750static struct commit *find_single_initial(struct rev_info *revs,
1751 const char **name_p)
1752{
1753 int i;
1754 struct commit *found = NULL;
1755 const char *name = NULL;
1756
1757 /*
1758 * There must be one and only one negative commit, and it must be
1759 * the boundary.
1760 */
1761 for (i = 0; i < revs->pending.nr; i++) {
1762 struct object *obj = revs->pending.objects[i].item;
1763 if (!(obj->flags & UNINTERESTING))
1764 continue;
fb998eae 1765 obj = deref_tag(revs->repo, obj, NULL, 0);
09002f1b
JS
1766 if (obj->type != OBJ_COMMIT)
1767 die("Non commit %s?", revs->pending.objects[i].name);
1768 if (found)
1769 die("More than one commit to dig up from, %s and %s?",
1770 revs->pending.objects[i].name, name);
1771 found = (struct commit *) obj;
1772 name = revs->pending.objects[i].name;
1773 }
1774
1775 if (!name)
1776 found = dwim_reverse_initial(revs, &name);
1777 if (!name)
1778 die("No commit to dig up from?");
1779
1780 if (name_p)
9e7d8a9b 1781 *name_p = xstrdup(name);
09002f1b
JS
1782 return found;
1783}
1784
1785void init_scoreboard(struct blame_scoreboard *sb)
1786{
1787 memset(sb, 0, sizeof(struct blame_scoreboard));
1788 sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
1789 sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
1790}
1791
ecbbc0a5
NTND
1792void setup_scoreboard(struct blame_scoreboard *sb,
1793 const char *path,
1794 struct blame_origin **orig)
09002f1b
JS
1795{
1796 const char *final_commit_name = NULL;
1797 struct blame_origin *o;
1798 struct commit *final_commit = NULL;
1799 enum object_type type;
1800
4e0df4e6
NTND
1801 init_blame_suspects(&blame_suspects);
1802
09002f1b
JS
1803 if (sb->reverse && sb->contents_from)
1804 die(_("--contents and --reverse do not blend well."));
1805
ecbbc0a5
NTND
1806 if (!sb->repo)
1807 BUG("repo is NULL");
1808
09002f1b
JS
1809 if (!sb->reverse) {
1810 sb->final = find_single_final(sb->revs, &final_commit_name);
1811 sb->commits.compare = compare_commits_by_commit_date;
1812 } else {
1813 sb->final = find_single_initial(sb->revs, &final_commit_name);
1814 sb->commits.compare = compare_commits_by_reverse_commit_date;
1815 }
1816
1817 if (sb->final && sb->contents_from)
1818 die(_("cannot use --contents with final commit object name"));
1819
1820 if (sb->reverse && sb->revs->first_parent_only)
1821 sb->revs->children.name = NULL;
1822
1823 if (!sb->final) {
1824 /*
1825 * "--not A B -- path" without anything positive;
1826 * do not default to HEAD, but use the working tree
1827 * or "--contents".
1828 */
1829 setup_work_tree();
ecbbc0a5
NTND
1830 sb->final = fake_working_tree_commit(sb->repo,
1831 &sb->revs->diffopt,
09002f1b
JS
1832 path, sb->contents_from);
1833 add_pending_object(sb->revs, &(sb->final->object), ":");
1834 }
1835
1836 if (sb->reverse && sb->revs->first_parent_only) {
1837 final_commit = find_single_final(sb->revs, NULL);
1838 if (!final_commit)
1839 die(_("--reverse and --first-parent together require specified latest commit"));
1840 }
1841
1842 /*
1843 * If we have bottom, this will mark the ancestors of the
1844 * bottom commits we would reach while traversing as
1845 * uninteresting.
1846 */
1847 if (prepare_revision_walk(sb->revs))
1848 die(_("revision walk setup failed"));
1849
1850 if (sb->reverse && sb->revs->first_parent_only) {
1851 struct commit *c = final_commit;
1852
1853 sb->revs->children.name = "children";
1854 while (c->parents &&
9001dc2a 1855 !oideq(&c->object.oid, &sb->final->object.oid)) {
09002f1b
JS
1856 struct commit_list *l = xcalloc(1, sizeof(*l));
1857
1858 l->item = c;
1859 if (add_decoration(&sb->revs->children,
1860 &c->parents->item->object, l))
033abf97 1861 BUG("not unique item in first-parent chain");
09002f1b
JS
1862 c = c->parents->item;
1863 }
1864
9001dc2a 1865 if (!oideq(&c->object.oid, &sb->final->object.oid))
09002f1b
JS
1866 die(_("--reverse --first-parent together require range along first-parent chain"));
1867 }
1868
1869 if (is_null_oid(&sb->final->object.oid)) {
4e0df4e6 1870 o = get_blame_suspects(sb->final);
09002f1b
JS
1871 sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
1872 sb->final_buf_size = o->file.size;
1873 }
1874 else {
1875 o = get_origin(sb->final, path);
ecbbc0a5 1876 if (fill_blob_sha1_and_mode(sb->repo, o))
09002f1b
JS
1877 die(_("no such path %s in %s"), path, final_commit_name);
1878
0d1e0e78 1879 if (sb->revs->diffopt.flags.allow_textconv &&
6afaf807 1880 textconv_object(sb->repo, path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
09002f1b
JS
1881 &sb->final_buf_size))
1882 ;
1883 else
b4f5aca4 1884 sb->final_buf = read_object_file(&o->blob_oid, &type,
1885 &sb->final_buf_size);
09002f1b
JS
1886
1887 if (!sb->final_buf)
1888 die(_("cannot read blob %s for path %s"),
1889 oid_to_hex(&o->blob_oid),
1890 path);
1891 }
1892 sb->num_read_blob++;
1893 prepare_lines(sb);
1894
1895 if (orig)
1896 *orig = o;
9e7d8a9b
SG
1897
1898 free((char *)final_commit_name);
09002f1b 1899}
bd481de7
JS
1900
1901
1902
1903struct blame_entry *blame_entry_prepend(struct blame_entry *head,
1904 long start, long end,
1905 struct blame_origin *o)
1906{
1907 struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
1908 new_head->lno = start;
1909 new_head->num_lines = end - start;
1910 new_head->suspect = o;
1911 new_head->s_lno = start;
1912 new_head->next = head;
1913 blame_origin_incref(o);
1914 return new_head;
1915}