Merge branch 'bw/submodule-helper-usage-fix'
[git/git.git] / tree-diff.c
1 /*
2 * Helper functions for tree diff generation
3 */
4 #include "cache.h"
5 #include "diff.h"
6 #include "diffcore.h"
7 #include "tree.h"
8
9 /*
10 * internal mode marker, saying a tree entry != entry of tp[imin]
11 * (see ll_diff_tree_paths for what it means there)
12 *
13 * we will update/use/emit entry for diff only with it unset.
14 */
15 #define S_IFXMIN_NEQ S_DIFFTREE_IFXMIN_NEQ
16
17 #define FAST_ARRAY_ALLOC(x, nr) do { \
18 if ((nr) <= 2) \
19 (x) = xalloca((nr) * sizeof(*(x))); \
20 else \
21 ALLOC_ARRAY((x), nr); \
22 } while(0)
23 #define FAST_ARRAY_FREE(x, nr) do { \
24 if ((nr) > 2) \
25 free((x)); \
26 } while(0)
27
28 static struct combine_diff_path *ll_diff_tree_paths(
29 struct combine_diff_path *p, const struct object_id *oid,
30 const struct object_id **parents_oid, int nparent,
31 struct strbuf *base, struct diff_options *opt);
32 static int ll_diff_tree_oid(const struct object_id *old_oid,
33 const struct object_id *new_oid,
34 struct strbuf *base, struct diff_options *opt);
35
36 /*
37 * Compare two tree entries, taking into account only path/S_ISDIR(mode),
38 * but not their sha1's.
39 *
40 * NOTE files and directories *always* compare differently, even when having
41 * the same name - thanks to base_name_compare().
42 *
43 * NOTE empty (=invalid) descriptor(s) take part in comparison as +infty,
44 * so that they sort *after* valid tree entries.
45 *
46 * Due to this convention, if trees are scanned in sorted order, all
47 * non-empty descriptors will be processed first.
48 */
49 static int tree_entry_pathcmp(struct tree_desc *t1, struct tree_desc *t2)
50 {
51 struct name_entry *e1, *e2;
52 int cmp;
53
54 /* empty descriptors sort after valid tree entries */
55 if (!t1->size)
56 return t2->size ? 1 : 0;
57 else if (!t2->size)
58 return -1;
59
60 e1 = &t1->entry;
61 e2 = &t2->entry;
62 cmp = base_name_compare(e1->path, tree_entry_len(e1), e1->mode,
63 e2->path, tree_entry_len(e2), e2->mode);
64 return cmp;
65 }
66
67
68 /*
69 * convert path -> opt->diff_*() callbacks
70 *
71 * emits diff to first parent only, and tells diff tree-walker that we are done
72 * with p and it can be freed.
73 */
74 static int emit_diff_first_parent_only(struct diff_options *opt, struct combine_diff_path *p)
75 {
76 struct combine_diff_parent *p0 = &p->parent[0];
77 if (p->mode && p0->mode) {
78 opt->change(opt, p0->mode, p->mode, &p0->oid, &p->oid,
79 1, 1, p->path, 0, 0);
80 }
81 else {
82 const struct object_id *oid;
83 unsigned int mode;
84 int addremove;
85
86 if (p->mode) {
87 addremove = '+';
88 oid = &p->oid;
89 mode = p->mode;
90 } else {
91 addremove = '-';
92 oid = &p0->oid;
93 mode = p0->mode;
94 }
95
96 opt->add_remove(opt, addremove, mode, oid, 1, p->path, 0);
97 }
98
99 return 0; /* we are done with p */
100 }
101
102
103 /*
104 * Make a new combine_diff_path from path/mode/sha1
105 * and append it to paths list tail.
106 *
107 * Memory for created elements could be reused:
108 *
109 * - if last->next == NULL, the memory is allocated;
110 *
111 * - if last->next != NULL, it is assumed that p=last->next was returned
112 * earlier by this function, and p->next was *not* modified.
113 * The memory is then reused from p.
114 *
115 * so for clients,
116 *
117 * - if you do need to keep the element
118 *
119 * p = path_appendnew(p, ...);
120 * process(p);
121 * p->next = NULL;
122 *
123 * - if you don't need to keep the element after processing
124 *
125 * pprev = p;
126 * p = path_appendnew(p, ...);
127 * process(p);
128 * p = pprev;
129 * ; don't forget to free tail->next in the end
130 *
131 * p->parent[] remains uninitialized.
132 */
133 static struct combine_diff_path *path_appendnew(struct combine_diff_path *last,
134 int nparent, const struct strbuf *base, const char *path, int pathlen,
135 unsigned mode, const struct object_id *oid)
136 {
137 struct combine_diff_path *p;
138 size_t len = st_add(base->len, pathlen);
139 size_t alloclen = combine_diff_path_size(nparent, len);
140
141 /* if last->next is !NULL - it is a pre-allocated memory, we can reuse */
142 p = last->next;
143 if (p && (alloclen > (intptr_t)p->next)) {
144 FREE_AND_NULL(p);
145 }
146
147 if (!p) {
148 p = xmalloc(alloclen);
149
150 /*
151 * until we go to it next round, .next holds how many bytes we
152 * allocated (for faster realloc - we don't need copying old data).
153 */
154 p->next = (struct combine_diff_path *)(intptr_t)alloclen;
155 }
156
157 last->next = p;
158
159 p->path = (char *)&(p->parent[nparent]);
160 memcpy(p->path, base->buf, base->len);
161 memcpy(p->path + base->len, path, pathlen);
162 p->path[len] = 0;
163 p->mode = mode;
164 oidcpy(&p->oid, oid ? oid : &null_oid);
165
166 return p;
167 }
168
169 /*
170 * new path should be added to combine diff
171 *
172 * 3 cases on how/when it should be called and behaves:
173 *
174 * t, !tp -> path added, all parents lack it
175 * !t, tp -> path removed from all parents
176 * t, tp -> path modified/added
177 * (M for tp[i]=tp[imin], A otherwise)
178 */
179 static struct combine_diff_path *emit_path(struct combine_diff_path *p,
180 struct strbuf *base, struct diff_options *opt, int nparent,
181 struct tree_desc *t, struct tree_desc *tp,
182 int imin)
183 {
184 unsigned short mode;
185 const char *path;
186 const struct object_id *oid;
187 int pathlen;
188 int old_baselen = base->len;
189 int i, isdir, recurse = 0, emitthis = 1;
190
191 /* at least something has to be valid */
192 assert(t || tp);
193
194 if (t) {
195 /* path present in resulting tree */
196 oid = tree_entry_extract(t, &path, &mode);
197 pathlen = tree_entry_len(&t->entry);
198 isdir = S_ISDIR(mode);
199 } else {
200 /*
201 * a path was removed - take path from imin parent. Also take
202 * mode from that parent, to decide on recursion(1).
203 *
204 * 1) all modes for tp[i]=tp[imin] should be the same wrt
205 * S_ISDIR, thanks to base_name_compare().
206 */
207 tree_entry_extract(&tp[imin], &path, &mode);
208 pathlen = tree_entry_len(&tp[imin].entry);
209
210 isdir = S_ISDIR(mode);
211 oid = NULL;
212 mode = 0;
213 }
214
215 if (opt->flags.recursive && isdir) {
216 recurse = 1;
217 emitthis = opt->flags.tree_in_recursive;
218 }
219
220 if (emitthis) {
221 int keep;
222 struct combine_diff_path *pprev = p;
223 p = path_appendnew(p, nparent, base, path, pathlen, mode, oid);
224
225 for (i = 0; i < nparent; ++i) {
226 /*
227 * tp[i] is valid, if present and if tp[i]==tp[imin] -
228 * otherwise, we should ignore it.
229 */
230 int tpi_valid = tp && !(tp[i].entry.mode & S_IFXMIN_NEQ);
231
232 const struct object_id *oid_i;
233 unsigned mode_i;
234
235 p->parent[i].status =
236 !t ? DIFF_STATUS_DELETED :
237 tpi_valid ?
238 DIFF_STATUS_MODIFIED :
239 DIFF_STATUS_ADDED;
240
241 if (tpi_valid) {
242 oid_i = &tp[i].entry.oid;
243 mode_i = tp[i].entry.mode;
244 }
245 else {
246 oid_i = &null_oid;
247 mode_i = 0;
248 }
249
250 p->parent[i].mode = mode_i;
251 oidcpy(&p->parent[i].oid, oid_i);
252 }
253
254 keep = 1;
255 if (opt->pathchange)
256 keep = opt->pathchange(opt, p);
257
258 /*
259 * If a path was filtered or consumed - we don't need to add it
260 * to the list and can reuse its memory, leaving it as
261 * pre-allocated element on the tail.
262 *
263 * On the other hand, if path needs to be kept, we need to
264 * correct its .next to NULL, as it was pre-initialized to how
265 * much memory was allocated.
266 *
267 * see path_appendnew() for details.
268 */
269 if (!keep)
270 p = pprev;
271 else
272 p->next = NULL;
273 }
274
275 if (recurse) {
276 const struct object_id **parents_oid;
277
278 FAST_ARRAY_ALLOC(parents_oid, nparent);
279 for (i = 0; i < nparent; ++i) {
280 /* same rule as in emitthis */
281 int tpi_valid = tp && !(tp[i].entry.mode & S_IFXMIN_NEQ);
282
283 parents_oid[i] = tpi_valid ? &tp[i].entry.oid : NULL;
284 }
285
286 strbuf_add(base, path, pathlen);
287 strbuf_addch(base, '/');
288 p = ll_diff_tree_paths(p, oid, parents_oid, nparent, base, opt);
289 FAST_ARRAY_FREE(parents_oid, nparent);
290 }
291
292 strbuf_setlen(base, old_baselen);
293 return p;
294 }
295
296 static void skip_uninteresting(struct tree_desc *t, struct strbuf *base,
297 struct diff_options *opt)
298 {
299 enum interesting match;
300
301 while (t->size) {
302 match = tree_entry_interesting(opt->repo->index, &t->entry,
303 base, 0, &opt->pathspec);
304 if (match) {
305 if (match == all_entries_not_interesting)
306 t->size = 0;
307 break;
308 }
309 update_tree_entry(t);
310 }
311 }
312
313
314 /*
315 * generate paths for combined diff D(sha1,parents_oid[])
316 *
317 * Resulting paths are appended to combine_diff_path linked list, and also, are
318 * emitted on the go via opt->pathchange() callback, so it is possible to
319 * process the result as batch or incrementally.
320 *
321 * The paths are generated scanning new tree and all parents trees
322 * simultaneously, similarly to what diff_tree() was doing for 2 trees.
323 * The theory behind such scan is as follows:
324 *
325 *
326 * D(T,P1...Pn) calculation scheme
327 * -------------------------------
328 *
329 * D(T,P1...Pn) = D(T,P1) ^ ... ^ D(T,Pn) (regarding resulting paths set)
330 *
331 * D(T,Pj) - diff between T..Pj
332 * D(T,P1...Pn) - combined diff from T to parents P1,...,Pn
333 *
334 *
335 * We start from all trees, which are sorted, and compare their entries in
336 * lock-step:
337 *
338 * T P1 Pn
339 * - - -
340 * |t| |p1| |pn|
341 * |-| |--| ... |--| imin = argmin(p1...pn)
342 * | | | | | |
343 * |-| |--| |--|
344 * |.| |. | |. |
345 * . . .
346 * . . .
347 *
348 * at any time there could be 3 cases:
349 *
350 * 1) t < p[imin];
351 * 2) t > p[imin];
352 * 3) t = p[imin].
353 *
354 * Schematic deduction of what every case means, and what to do, follows:
355 *
356 * 1) t < p[imin] -> ∀j t ∉ Pj -> "+t" ∈ D(T,Pj) -> D += "+t"; t↓
357 *
358 * 2) t > p[imin]
359 *
360 * 2.1) ∃j: pj > p[imin] -> "-p[imin]" ∉ D(T,Pj) -> D += ø; ∀ pi=p[imin] pi↓
361 * 2.2) ∀i pi = p[imin] -> pi ∉ T -> "-pi" ∈ D(T,Pi) -> D += "-p[imin]"; ∀i pi↓
362 *
363 * 3) t = p[imin]
364 *
365 * 3.1) ∃j: pj > p[imin] -> "+t" ∈ D(T,Pj) -> only pi=p[imin] remains to investigate
366 * 3.2) pi = p[imin] -> investigate δ(t,pi)
367 * |
368 * |
369 * v
370 *
371 * 3.1+3.2) looking at δ(t,pi) ∀i: pi=p[imin] - if all != ø ->
372 *
373 * ⎧δ(t,pi) - if pi=p[imin]
374 * -> D += ⎨
375 * ⎩"+t" - if pi>p[imin]
376 *
377 *
378 * in any case t↓ ∀ pi=p[imin] pi↓
379 *
380 *
381 * ~~~~~~~~
382 *
383 * NOTE
384 *
385 * Usual diff D(A,B) is by definition the same as combined diff D(A,[B]),
386 * so this diff paths generator can, and is used, for plain diffs
387 * generation too.
388 *
389 * Please keep attention to the common D(A,[B]) case when working on the
390 * code, in order not to slow it down.
391 *
392 * NOTE
393 * nparent must be > 0.
394 */
395
396
397 /* ∀ pi=p[imin] pi↓ */
398 static inline void update_tp_entries(struct tree_desc *tp, int nparent)
399 {
400 int i;
401 for (i = 0; i < nparent; ++i)
402 if (!(tp[i].entry.mode & S_IFXMIN_NEQ))
403 update_tree_entry(&tp[i]);
404 }
405
406 static struct combine_diff_path *ll_diff_tree_paths(
407 struct combine_diff_path *p, const struct object_id *oid,
408 const struct object_id **parents_oid, int nparent,
409 struct strbuf *base, struct diff_options *opt)
410 {
411 struct tree_desc t, *tp;
412 void *ttree, **tptree;
413 int i;
414
415 FAST_ARRAY_ALLOC(tp, nparent);
416 FAST_ARRAY_ALLOC(tptree, nparent);
417
418 /*
419 * load parents first, as they are probably already cached.
420 *
421 * ( log_tree_diff() parses commit->parent before calling here via
422 * diff_tree_oid(parent, commit) )
423 */
424 for (i = 0; i < nparent; ++i)
425 tptree[i] = fill_tree_descriptor(opt->repo, &tp[i], parents_oid[i]);
426 ttree = fill_tree_descriptor(opt->repo, &t, oid);
427
428 /* Enable recursion indefinitely */
429 opt->pathspec.recursive = opt->flags.recursive;
430
431 for (;;) {
432 int imin, cmp;
433
434 if (diff_can_quit_early(opt))
435 break;
436
437 if (opt->pathspec.nr) {
438 skip_uninteresting(&t, base, opt);
439 for (i = 0; i < nparent; i++)
440 skip_uninteresting(&tp[i], base, opt);
441 }
442
443 /* comparing is finished when all trees are done */
444 if (!t.size) {
445 int done = 1;
446 for (i = 0; i < nparent; ++i)
447 if (tp[i].size) {
448 done = 0;
449 break;
450 }
451 if (done)
452 break;
453 }
454
455 /*
456 * lookup imin = argmin(p1...pn),
457 * mark entries whether they =p[imin] along the way
458 */
459 imin = 0;
460 tp[0].entry.mode &= ~S_IFXMIN_NEQ;
461
462 for (i = 1; i < nparent; ++i) {
463 cmp = tree_entry_pathcmp(&tp[i], &tp[imin]);
464 if (cmp < 0) {
465 imin = i;
466 tp[i].entry.mode &= ~S_IFXMIN_NEQ;
467 }
468 else if (cmp == 0) {
469 tp[i].entry.mode &= ~S_IFXMIN_NEQ;
470 }
471 else {
472 tp[i].entry.mode |= S_IFXMIN_NEQ;
473 }
474 }
475
476 /* fixup markings for entries before imin */
477 for (i = 0; i < imin; ++i)
478 tp[i].entry.mode |= S_IFXMIN_NEQ; /* pi > p[imin] */
479
480
481
482 /* compare t vs p[imin] */
483 cmp = tree_entry_pathcmp(&t, &tp[imin]);
484
485 /* t = p[imin] */
486 if (cmp == 0) {
487 /* are either pi > p[imin] or diff(t,pi) != ø ? */
488 if (!opt->flags.find_copies_harder) {
489 for (i = 0; i < nparent; ++i) {
490 /* p[i] > p[imin] */
491 if (tp[i].entry.mode & S_IFXMIN_NEQ)
492 continue;
493
494 /* diff(t,pi) != ø */
495 if (!oideq(&t.entry.oid, &tp[i].entry.oid) ||
496 (t.entry.mode != tp[i].entry.mode))
497 continue;
498
499 goto skip_emit_t_tp;
500 }
501 }
502
503 /* D += {δ(t,pi) if pi=p[imin]; "+a" if pi > p[imin]} */
504 p = emit_path(p, base, opt, nparent,
505 &t, tp, imin);
506
507 skip_emit_t_tp:
508 /* t↓, ∀ pi=p[imin] pi↓ */
509 update_tree_entry(&t);
510 update_tp_entries(tp, nparent);
511 }
512
513 /* t < p[imin] */
514 else if (cmp < 0) {
515 /* D += "+t" */
516 p = emit_path(p, base, opt, nparent,
517 &t, /*tp=*/NULL, -1);
518
519 /* t↓ */
520 update_tree_entry(&t);
521 }
522
523 /* t > p[imin] */
524 else {
525 /* ∀i pi=p[imin] -> D += "-p[imin]" */
526 if (!opt->flags.find_copies_harder) {
527 for (i = 0; i < nparent; ++i)
528 if (tp[i].entry.mode & S_IFXMIN_NEQ)
529 goto skip_emit_tp;
530 }
531
532 p = emit_path(p, base, opt, nparent,
533 /*t=*/NULL, tp, imin);
534
535 skip_emit_tp:
536 /* ∀ pi=p[imin] pi↓ */
537 update_tp_entries(tp, nparent);
538 }
539 }
540
541 free(ttree);
542 for (i = nparent-1; i >= 0; i--)
543 free(tptree[i]);
544 FAST_ARRAY_FREE(tptree, nparent);
545 FAST_ARRAY_FREE(tp, nparent);
546
547 return p;
548 }
549
550 struct combine_diff_path *diff_tree_paths(
551 struct combine_diff_path *p, const struct object_id *oid,
552 const struct object_id **parents_oid, int nparent,
553 struct strbuf *base, struct diff_options *opt)
554 {
555 p = ll_diff_tree_paths(p, oid, parents_oid, nparent, base, opt);
556
557 /*
558 * free pre-allocated last element, if any
559 * (see path_appendnew() for details about why)
560 */
561 FREE_AND_NULL(p->next);
562
563 return p;
564 }
565
566 /*
567 * Does it look like the resulting diff might be due to a rename?
568 * - single entry
569 * - not a valid previous file
570 */
571 static inline int diff_might_be_rename(void)
572 {
573 return diff_queued_diff.nr == 1 &&
574 !DIFF_FILE_VALID(diff_queued_diff.queue[0]->one);
575 }
576
577 static void try_to_follow_renames(const struct object_id *old_oid,
578 const struct object_id *new_oid,
579 struct strbuf *base, struct diff_options *opt)
580 {
581 struct diff_options diff_opts;
582 struct diff_queue_struct *q = &diff_queued_diff;
583 struct diff_filepair *choice;
584 int i;
585
586 /*
587 * follow-rename code is very specific, we need exactly one
588 * path. Magic that matches more than one path is not
589 * supported.
590 */
591 GUARD_PATHSPEC(&opt->pathspec, PATHSPEC_FROMTOP | PATHSPEC_LITERAL);
592 #if 0
593 /*
594 * We should reject wildcards as well. Unfortunately we
595 * haven't got a reliable way to detect that 'foo\*bar' in
596 * fact has no wildcards. nowildcard_len is merely a hint for
597 * optimization. Let it slip for now until wildmatch is taught
598 * about dry-run mode and returns wildcard info.
599 */
600 if (opt->pathspec.has_wildcard)
601 die("BUG:%s:%d: wildcards are not supported",
602 __FILE__, __LINE__);
603 #endif
604
605 /* Remove the file creation entry from the diff queue, and remember it */
606 choice = q->queue[0];
607 q->nr = 0;
608
609 repo_diff_setup(opt->repo, &diff_opts);
610 diff_opts.flags.recursive = 1;
611 diff_opts.flags.find_copies_harder = 1;
612 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
613 diff_opts.single_follow = opt->pathspec.items[0].match;
614 diff_opts.break_opt = opt->break_opt;
615 diff_opts.rename_score = opt->rename_score;
616 diff_setup_done(&diff_opts);
617 ll_diff_tree_oid(old_oid, new_oid, base, &diff_opts);
618 diffcore_std(&diff_opts);
619 clear_pathspec(&diff_opts.pathspec);
620
621 /* Go through the new set of filepairing, and see if we find a more interesting one */
622 opt->found_follow = 0;
623 for (i = 0; i < q->nr; i++) {
624 struct diff_filepair *p = q->queue[i];
625
626 /*
627 * Found a source? Not only do we use that for the new
628 * diff_queued_diff, we will also use that as the path in
629 * the future!
630 */
631 if ((p->status == 'R' || p->status == 'C') &&
632 !strcmp(p->two->path, opt->pathspec.items[0].match)) {
633 const char *path[2];
634
635 /* Switch the file-pairs around */
636 q->queue[i] = choice;
637 choice = p;
638
639 /* Update the path we use from now on.. */
640 path[0] = p->one->path;
641 path[1] = NULL;
642 clear_pathspec(&opt->pathspec);
643 parse_pathspec(&opt->pathspec,
644 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
645 PATHSPEC_LITERAL_PATH, "", path);
646
647 /*
648 * The caller expects us to return a set of vanilla
649 * filepairs to let a later call to diffcore_std()
650 * it makes to sort the renames out (among other
651 * things), but we already have found renames
652 * ourselves; signal diffcore_std() not to muck with
653 * rename information.
654 */
655 opt->found_follow = 1;
656 break;
657 }
658 }
659
660 /*
661 * Then, discard all the non-relevant file pairs...
662 */
663 for (i = 0; i < q->nr; i++) {
664 struct diff_filepair *p = q->queue[i];
665 diff_free_filepair(p);
666 }
667
668 /*
669 * .. and re-instate the one we want (which might be either the
670 * original one, or the rename/copy we found)
671 */
672 q->queue[0] = choice;
673 q->nr = 1;
674 }
675
676 static int ll_diff_tree_oid(const struct object_id *old_oid,
677 const struct object_id *new_oid,
678 struct strbuf *base, struct diff_options *opt)
679 {
680 struct combine_diff_path phead, *p;
681 pathchange_fn_t pathchange_old = opt->pathchange;
682
683 phead.next = NULL;
684 opt->pathchange = emit_diff_first_parent_only;
685 diff_tree_paths(&phead, new_oid, &old_oid, 1, base, opt);
686
687 for (p = phead.next; p;) {
688 struct combine_diff_path *pprev = p;
689 p = p->next;
690 free(pprev);
691 }
692
693 opt->pathchange = pathchange_old;
694 return 0;
695 }
696
697 int diff_tree_oid(const struct object_id *old_oid,
698 const struct object_id *new_oid,
699 const char *base_str, struct diff_options *opt)
700 {
701 struct strbuf base;
702 int retval;
703
704 strbuf_init(&base, PATH_MAX);
705 strbuf_addstr(&base, base_str);
706
707 retval = ll_diff_tree_oid(old_oid, new_oid, &base, opt);
708 if (!*base_str && opt->flags.follow_renames && diff_might_be_rename())
709 try_to_follow_renames(old_oid, new_oid, &base, opt);
710
711 strbuf_release(&base);
712
713 return retval;
714 }
715
716 int diff_root_tree_oid(const struct object_id *new_oid, const char *base, struct diff_options *opt)
717 {
718 return diff_tree_oid(NULL, new_oid, base, opt);
719 }