Documentation: add Packfile URIs design doc
[git.git] / combine-diff.c
blob002e0e5438bc27472d032ac562a2f3a29ec20a68
1 #include "cache.h"
2 #include "object-store.h"
3 #include "commit.h"
4 #include "blob.h"
5 #include "diff.h"
6 #include "diffcore.h"
7 #include "quote.h"
8 #include "xdiff-interface.h"
9 #include "xdiff/xmacros.h"
10 #include "log-tree.h"
11 #include "refs.h"
12 #include "userdiff.h"
13 #include "oid-array.h"
14 #include "revision.h"
16 static int compare_paths(const struct combine_diff_path *one,
17 const struct diff_filespec *two)
19 if (!S_ISDIR(one->mode) && !S_ISDIR(two->mode))
20 return strcmp(one->path, two->path);
22 return base_name_compare(one->path, strlen(one->path), one->mode,
23 two->path, strlen(two->path), two->mode);
26 static int filename_changed(char status)
28 return status == 'R' || status == 'C';
31 static struct combine_diff_path *intersect_paths(
32 struct combine_diff_path *curr,
33 int n,
34 int num_parent,
35 int combined_all_paths)
37 struct diff_queue_struct *q = &diff_queued_diff;
38 struct combine_diff_path *p, **tail = &curr;
39 int i, j, cmp;
41 if (!n) {
42 for (i = 0; i < q->nr; i++) {
43 int len;
44 const char *path;
45 if (diff_unmodified_pair(q->queue[i]))
46 continue;
47 path = q->queue[i]->two->path;
48 len = strlen(path);
49 p = xmalloc(combine_diff_path_size(num_parent, len));
50 p->path = (char *) &(p->parent[num_parent]);
51 memcpy(p->path, path, len);
52 p->path[len] = 0;
53 p->next = NULL;
54 memset(p->parent, 0,
55 sizeof(p->parent[0]) * num_parent);
57 oidcpy(&p->oid, &q->queue[i]->two->oid);
58 p->mode = q->queue[i]->two->mode;
59 oidcpy(&p->parent[n].oid, &q->queue[i]->one->oid);
60 p->parent[n].mode = q->queue[i]->one->mode;
61 p->parent[n].status = q->queue[i]->status;
63 if (combined_all_paths &&
64 filename_changed(p->parent[n].status)) {
65 strbuf_init(&p->parent[n].path, 0);
66 strbuf_addstr(&p->parent[n].path,
67 q->queue[i]->one->path);
69 *tail = p;
70 tail = &p->next;
72 return curr;
76 * paths in curr (linked list) and q->queue[] (array) are
77 * both sorted in the tree order.
79 i = 0;
80 while ((p = *tail) != NULL) {
81 cmp = ((i >= q->nr)
82 ? -1 : compare_paths(p, q->queue[i]->two));
84 if (cmp < 0) {
85 /* p->path not in q->queue[]; drop it */
86 *tail = p->next;
87 for (j = 0; j < num_parent; j++)
88 if (combined_all_paths &&
89 filename_changed(p->parent[j].status))
90 strbuf_release(&p->parent[j].path);
91 free(p);
92 continue;
95 if (cmp > 0) {
96 /* q->queue[i] not in p->path; skip it */
97 i++;
98 continue;
101 oidcpy(&p->parent[n].oid, &q->queue[i]->one->oid);
102 p->parent[n].mode = q->queue[i]->one->mode;
103 p->parent[n].status = q->queue[i]->status;
104 if (combined_all_paths &&
105 filename_changed(p->parent[n].status))
106 strbuf_addstr(&p->parent[n].path,
107 q->queue[i]->one->path);
109 tail = &p->next;
110 i++;
112 return curr;
115 /* Lines lost from parent */
116 struct lline {
117 struct lline *next, *prev;
118 int len;
119 unsigned long parent_map;
120 char line[FLEX_ARRAY];
123 /* Lines lost from current parent (before coalescing) */
124 struct plost {
125 struct lline *lost_head, *lost_tail;
126 int len;
129 /* Lines surviving in the merge result */
130 struct sline {
131 /* Accumulated and coalesced lost lines */
132 struct lline *lost;
133 int lenlost;
134 struct plost plost;
135 char *bol;
136 int len;
137 /* bit 0 up to (N-1) are on if the parent has this line (i.e.
138 * we did not change it).
139 * bit N is used for "interesting" lines, including context.
140 * bit (N+1) is used for "do not show deletion before this".
142 unsigned long flag;
143 unsigned long *p_lno;
146 static int match_string_spaces(const char *line1, int len1,
147 const char *line2, int len2,
148 long flags)
150 if (flags & XDF_WHITESPACE_FLAGS) {
151 for (; len1 > 0 && XDL_ISSPACE(line1[len1 - 1]); len1--);
152 for (; len2 > 0 && XDL_ISSPACE(line2[len2 - 1]); len2--);
155 if (!(flags & (XDF_IGNORE_WHITESPACE | XDF_IGNORE_WHITESPACE_CHANGE)))
156 return (len1 == len2 && !memcmp(line1, line2, len1));
158 while (len1 > 0 && len2 > 0) {
159 len1--;
160 len2--;
161 if (XDL_ISSPACE(line1[len1]) || XDL_ISSPACE(line2[len2])) {
162 if ((flags & XDF_IGNORE_WHITESPACE_CHANGE) &&
163 (!XDL_ISSPACE(line1[len1]) || !XDL_ISSPACE(line2[len2])))
164 return 0;
166 for (; len1 > 0 && XDL_ISSPACE(line1[len1]); len1--);
167 for (; len2 > 0 && XDL_ISSPACE(line2[len2]); len2--);
169 if (line1[len1] != line2[len2])
170 return 0;
173 if (flags & XDF_IGNORE_WHITESPACE) {
174 /* Consume remaining spaces */
175 for (; len1 > 0 && XDL_ISSPACE(line1[len1 - 1]); len1--);
176 for (; len2 > 0 && XDL_ISSPACE(line2[len2 - 1]); len2--);
179 /* We matched full line1 and line2 */
180 if (!len1 && !len2)
181 return 1;
183 return 0;
186 enum coalesce_direction { MATCH, BASE, NEW };
188 /* Coalesce new lines into base by finding LCS */
189 static struct lline *coalesce_lines(struct lline *base, int *lenbase,
190 struct lline *newline, int lennew,
191 unsigned long parent, long flags)
193 int **lcs;
194 enum coalesce_direction **directions;
195 struct lline *baseend, *newend = NULL;
196 int i, j, origbaselen = *lenbase;
198 if (newline == NULL)
199 return base;
201 if (base == NULL) {
202 *lenbase = lennew;
203 return newline;
207 * Coalesce new lines into base by finding the LCS
208 * - Create the table to run dynamic programming
209 * - Compute the LCS
210 * - Then reverse read the direction structure:
211 * - If we have MATCH, assign parent to base flag, and consume
212 * both baseend and newend
213 * - Else if we have BASE, consume baseend
214 * - Else if we have NEW, insert newend lline into base and
215 * consume newend
217 lcs = xcalloc(st_add(origbaselen, 1), sizeof(int*));
218 directions = xcalloc(st_add(origbaselen, 1), sizeof(enum coalesce_direction*));
219 for (i = 0; i < origbaselen + 1; i++) {
220 lcs[i] = xcalloc(st_add(lennew, 1), sizeof(int));
221 directions[i] = xcalloc(st_add(lennew, 1), sizeof(enum coalesce_direction));
222 directions[i][0] = BASE;
224 for (j = 1; j < lennew + 1; j++)
225 directions[0][j] = NEW;
227 for (i = 1, baseend = base; i < origbaselen + 1; i++) {
228 for (j = 1, newend = newline; j < lennew + 1; j++) {
229 if (match_string_spaces(baseend->line, baseend->len,
230 newend->line, newend->len, flags)) {
231 lcs[i][j] = lcs[i - 1][j - 1] + 1;
232 directions[i][j] = MATCH;
233 } else if (lcs[i][j - 1] >= lcs[i - 1][j]) {
234 lcs[i][j] = lcs[i][j - 1];
235 directions[i][j] = NEW;
236 } else {
237 lcs[i][j] = lcs[i - 1][j];
238 directions[i][j] = BASE;
240 if (newend->next)
241 newend = newend->next;
243 if (baseend->next)
244 baseend = baseend->next;
247 for (i = 0; i < origbaselen + 1; i++)
248 free(lcs[i]);
249 free(lcs);
251 /* At this point, baseend and newend point to the end of each lists */
252 i--;
253 j--;
254 while (i != 0 || j != 0) {
255 if (directions[i][j] == MATCH) {
256 baseend->parent_map |= 1<<parent;
257 baseend = baseend->prev;
258 newend = newend->prev;
259 i--;
260 j--;
261 } else if (directions[i][j] == NEW) {
262 struct lline *lline;
264 lline = newend;
265 /* Remove lline from new list and update newend */
266 if (lline->prev)
267 lline->prev->next = lline->next;
268 else
269 newline = lline->next;
270 if (lline->next)
271 lline->next->prev = lline->prev;
273 newend = lline->prev;
274 j--;
276 /* Add lline to base list */
277 if (baseend) {
278 lline->next = baseend->next;
279 lline->prev = baseend;
280 if (lline->prev)
281 lline->prev->next = lline;
283 else {
284 lline->next = base;
285 base = lline;
287 (*lenbase)++;
289 if (lline->next)
290 lline->next->prev = lline;
292 } else {
293 baseend = baseend->prev;
294 i--;
298 newend = newline;
299 while (newend) {
300 struct lline *lline = newend;
301 newend = newend->next;
302 free(lline);
305 for (i = 0; i < origbaselen + 1; i++)
306 free(directions[i]);
307 free(directions);
309 return base;
312 static char *grab_blob(struct repository *r,
313 const struct object_id *oid, unsigned int mode,
314 unsigned long *size, struct userdiff_driver *textconv,
315 const char *path)
317 char *blob;
318 enum object_type type;
320 if (S_ISGITLINK(mode)) {
321 struct strbuf buf = STRBUF_INIT;
322 strbuf_addf(&buf, "Subproject commit %s\n", oid_to_hex(oid));
323 *size = buf.len;
324 blob = strbuf_detach(&buf, NULL);
325 } else if (is_null_oid(oid)) {
326 /* deleted blob */
327 *size = 0;
328 return xcalloc(1, 1);
329 } else if (textconv) {
330 struct diff_filespec *df = alloc_filespec(path);
331 fill_filespec(df, oid, 1, mode);
332 *size = fill_textconv(r, textconv, df, &blob);
333 free_filespec(df);
334 } else {
335 blob = read_object_file(oid, &type, size);
336 if (type != OBJ_BLOB)
337 die("object '%s' is not a blob!", oid_to_hex(oid));
339 return blob;
342 static void append_lost(struct sline *sline, int n, const char *line, int len)
344 struct lline *lline;
345 unsigned long this_mask = (1UL<<n);
346 if (line[len-1] == '\n')
347 len--;
349 FLEX_ALLOC_MEM(lline, line, line, len);
350 lline->len = len;
351 lline->next = NULL;
352 lline->prev = sline->plost.lost_tail;
353 if (lline->prev)
354 lline->prev->next = lline;
355 else
356 sline->plost.lost_head = lline;
357 sline->plost.lost_tail = lline;
358 sline->plost.len++;
359 lline->parent_map = this_mask;
362 struct combine_diff_state {
363 unsigned int lno;
364 int ob, on, nb, nn;
365 unsigned long nmask;
366 int num_parent;
367 int n;
368 struct sline *sline;
369 struct sline *lost_bucket;
372 static void consume_hunk(void *state_,
373 long ob, long on,
374 long nb, long nn,
375 const char *funcline, long funclen)
377 struct combine_diff_state *state = state_;
379 state->ob = ob;
380 state->on = on;
381 state->nb = nb;
382 state->nn = nn;
383 state->lno = state->nb;
384 if (state->nn == 0) {
385 /* @@ -X,Y +N,0 @@ removed Y lines
386 * that would have come *after* line N
387 * in the result. Our lost buckets hang
388 * to the line after the removed lines,
390 * Note that this is correct even when N == 0,
391 * in which case the hunk removes the first
392 * line in the file.
394 state->lost_bucket = &state->sline[state->nb];
395 if (!state->nb)
396 state->nb = 1;
397 } else {
398 state->lost_bucket = &state->sline[state->nb-1];
400 if (!state->sline[state->nb-1].p_lno)
401 state->sline[state->nb-1].p_lno =
402 xcalloc(state->num_parent, sizeof(unsigned long));
403 state->sline[state->nb-1].p_lno[state->n] = state->ob;
406 static void consume_line(void *state_, char *line, unsigned long len)
408 struct combine_diff_state *state = state_;
409 if (!state->lost_bucket)
410 return; /* not in any hunk yet */
411 switch (line[0]) {
412 case '-':
413 append_lost(state->lost_bucket, state->n, line+1, len-1);
414 break;
415 case '+':
416 state->sline[state->lno-1].flag |= state->nmask;
417 state->lno++;
418 break;
422 static void combine_diff(struct repository *r,
423 const struct object_id *parent, unsigned int mode,
424 mmfile_t *result_file,
425 struct sline *sline, unsigned int cnt, int n,
426 int num_parent, int result_deleted,
427 struct userdiff_driver *textconv,
428 const char *path, long flags)
430 unsigned int p_lno, lno;
431 unsigned long nmask = (1UL << n);
432 xpparam_t xpp;
433 xdemitconf_t xecfg;
434 mmfile_t parent_file;
435 struct combine_diff_state state;
436 unsigned long sz;
438 if (result_deleted)
439 return; /* result deleted */
441 parent_file.ptr = grab_blob(r, parent, mode, &sz, textconv, path);
442 parent_file.size = sz;
443 memset(&xpp, 0, sizeof(xpp));
444 xpp.flags = flags;
445 memset(&xecfg, 0, sizeof(xecfg));
446 memset(&state, 0, sizeof(state));
447 state.nmask = nmask;
448 state.sline = sline;
449 state.lno = 1;
450 state.num_parent = num_parent;
451 state.n = n;
453 if (xdi_diff_outf(&parent_file, result_file, consume_hunk,
454 consume_line, &state, &xpp, &xecfg))
455 die("unable to generate combined diff for %s",
456 oid_to_hex(parent));
457 free(parent_file.ptr);
459 /* Assign line numbers for this parent.
461 * sline[lno].p_lno[n] records the first line number
462 * (counting from 1) for parent N if the final hunk display
463 * started by showing sline[lno] (possibly showing the lost
464 * lines attached to it first).
466 for (lno = 0, p_lno = 1; lno <= cnt; lno++) {
467 struct lline *ll;
468 sline[lno].p_lno[n] = p_lno;
470 /* Coalesce new lines */
471 if (sline[lno].plost.lost_head) {
472 struct sline *sl = &sline[lno];
473 sl->lost = coalesce_lines(sl->lost, &sl->lenlost,
474 sl->plost.lost_head,
475 sl->plost.len, n, flags);
476 sl->plost.lost_head = sl->plost.lost_tail = NULL;
477 sl->plost.len = 0;
480 /* How many lines would this sline advance the p_lno? */
481 ll = sline[lno].lost;
482 while (ll) {
483 if (ll->parent_map & nmask)
484 p_lno++; /* '-' means parent had it */
485 ll = ll->next;
487 if (lno < cnt && !(sline[lno].flag & nmask))
488 p_lno++; /* no '+' means parent had it */
490 sline[lno].p_lno[n] = p_lno; /* trailer */
493 static unsigned long context = 3;
494 static char combine_marker = '@';
496 static int interesting(struct sline *sline, unsigned long all_mask)
498 /* If some parents lost lines here, or if we have added to
499 * some parent, it is interesting.
501 return ((sline->flag & all_mask) || sline->lost);
504 static unsigned long adjust_hunk_tail(struct sline *sline,
505 unsigned long all_mask,
506 unsigned long hunk_begin,
507 unsigned long i)
509 /* i points at the first uninteresting line. If the last line
510 * of the hunk was interesting only because it has some
511 * deletion, then it is not all that interesting for the
512 * purpose of giving trailing context lines. This is because
513 * we output '-' line and then unmodified sline[i-1] itself in
514 * that case which gives us one extra context line.
516 if ((hunk_begin + 1 <= i) && !(sline[i-1].flag & all_mask))
517 i--;
518 return i;
521 static unsigned long find_next(struct sline *sline,
522 unsigned long mark,
523 unsigned long i,
524 unsigned long cnt,
525 int look_for_uninteresting)
527 /* We have examined up to i-1 and are about to look at i.
528 * Find next interesting or uninteresting line. Here,
529 * "interesting" does not mean interesting(), but marked by
530 * the give_context() function below (i.e. it includes context
531 * lines that are not interesting to interesting() function
532 * that are surrounded by interesting() ones.
534 while (i <= cnt)
535 if (look_for_uninteresting
536 ? !(sline[i].flag & mark)
537 : (sline[i].flag & mark))
538 return i;
539 else
540 i++;
541 return i;
544 static int give_context(struct sline *sline, unsigned long cnt, int num_parent)
546 unsigned long all_mask = (1UL<<num_parent) - 1;
547 unsigned long mark = (1UL<<num_parent);
548 unsigned long no_pre_delete = (2UL<<num_parent);
549 unsigned long i;
551 /* Two groups of interesting lines may have a short gap of
552 * uninteresting lines. Connect such groups to give them a
553 * bit of context.
555 * We first start from what the interesting() function says,
556 * and mark them with "mark", and paint context lines with the
557 * mark. So interesting() would still say false for such context
558 * lines but they are treated as "interesting" in the end.
560 i = find_next(sline, mark, 0, cnt, 0);
561 if (cnt < i)
562 return 0;
564 while (i <= cnt) {
565 unsigned long j = (context < i) ? (i - context) : 0;
566 unsigned long k;
568 /* Paint a few lines before the first interesting line. */
569 while (j < i) {
570 if (!(sline[j].flag & mark))
571 sline[j].flag |= no_pre_delete;
572 sline[j++].flag |= mark;
575 again:
576 /* we know up to i is to be included. where does the
577 * next uninteresting one start?
579 j = find_next(sline, mark, i, cnt, 1);
580 if (cnt < j)
581 break; /* the rest are all interesting */
583 /* lookahead context lines */
584 k = find_next(sline, mark, j, cnt, 0);
585 j = adjust_hunk_tail(sline, all_mask, i, j);
587 if (k < j + context) {
588 /* k is interesting and [j,k) are not, but
589 * paint them interesting because the gap is small.
591 while (j < k)
592 sline[j++].flag |= mark;
593 i = k;
594 goto again;
597 /* j is the first uninteresting line and there is
598 * no overlap beyond it within context lines. Paint
599 * the trailing edge a bit.
601 i = k;
602 k = (j + context < cnt+1) ? j + context : cnt+1;
603 while (j < k)
604 sline[j++].flag |= mark;
606 return 1;
609 static int make_hunks(struct sline *sline, unsigned long cnt,
610 int num_parent, int dense)
612 unsigned long all_mask = (1UL<<num_parent) - 1;
613 unsigned long mark = (1UL<<num_parent);
614 unsigned long i;
615 int has_interesting = 0;
617 for (i = 0; i <= cnt; i++) {
618 if (interesting(&sline[i], all_mask))
619 sline[i].flag |= mark;
620 else
621 sline[i].flag &= ~mark;
623 if (!dense)
624 return give_context(sline, cnt, num_parent);
626 /* Look at each hunk, and if we have changes from only one
627 * parent, or the changes are the same from all but one
628 * parent, mark that uninteresting.
630 i = 0;
631 while (i <= cnt) {
632 unsigned long j, hunk_begin, hunk_end;
633 unsigned long same_diff;
634 while (i <= cnt && !(sline[i].flag & mark))
635 i++;
636 if (cnt < i)
637 break; /* No more interesting hunks */
638 hunk_begin = i;
639 for (j = i + 1; j <= cnt; j++) {
640 if (!(sline[j].flag & mark)) {
641 /* Look beyond the end to see if there
642 * is an interesting line after this
643 * hunk within context span.
645 unsigned long la; /* lookahead */
646 int contin = 0;
647 la = adjust_hunk_tail(sline, all_mask,
648 hunk_begin, j);
649 la = (la + context < cnt + 1) ?
650 (la + context) : cnt + 1;
651 while (la && j <= --la) {
652 if (sline[la].flag & mark) {
653 contin = 1;
654 break;
657 if (!contin)
658 break;
659 j = la;
662 hunk_end = j;
664 /* [i..hunk_end) are interesting. Now is it really
665 * interesting? We check if there are only two versions
666 * and the result matches one of them. That is, we look
667 * at:
668 * (+) line, which records lines added to which parents;
669 * this line appears in the result.
670 * (-) line, which records from what parents the line
671 * was removed; this line does not appear in the result.
672 * then check the set of parents the result has difference
673 * from, from all lines. If there are lines that has
674 * different set of parents that the result has differences
675 * from, that means we have more than two versions.
677 * Even when we have only two versions, if the result does
678 * not match any of the parents, the it should be considered
679 * interesting. In such a case, we would have all '+' line.
680 * After passing the above "two versions" test, that would
681 * appear as "the same set of parents" to be "all parents".
683 same_diff = 0;
684 has_interesting = 0;
685 for (j = i; j < hunk_end && !has_interesting; j++) {
686 unsigned long this_diff = sline[j].flag & all_mask;
687 struct lline *ll = sline[j].lost;
688 if (this_diff) {
689 /* This has some changes. Is it the
690 * same as others?
692 if (!same_diff)
693 same_diff = this_diff;
694 else if (same_diff != this_diff) {
695 has_interesting = 1;
696 break;
699 while (ll && !has_interesting) {
700 /* Lost this line from these parents;
701 * who are they? Are they the same?
703 this_diff = ll->parent_map;
704 if (!same_diff)
705 same_diff = this_diff;
706 else if (same_diff != this_diff) {
707 has_interesting = 1;
709 ll = ll->next;
713 if (!has_interesting && same_diff != all_mask) {
714 /* This hunk is not that interesting after all */
715 for (j = hunk_begin; j < hunk_end; j++)
716 sline[j].flag &= ~mark;
718 i = hunk_end;
721 has_interesting = give_context(sline, cnt, num_parent);
722 return has_interesting;
725 static void show_parent_lno(struct sline *sline, unsigned long l0, unsigned long l1, int n, unsigned long null_context)
727 l0 = sline[l0].p_lno[n];
728 l1 = sline[l1].p_lno[n];
729 printf(" -%lu,%lu", l0, l1-l0-null_context);
732 static int hunk_comment_line(const char *bol)
734 int ch;
736 if (!bol)
737 return 0;
738 ch = *bol & 0xff;
739 return (isalpha(ch) || ch == '_' || ch == '$');
742 static void show_line_to_eol(const char *line, int len, const char *reset)
744 int saw_cr_at_eol = 0;
745 if (len < 0)
746 len = strlen(line);
747 saw_cr_at_eol = (len && line[len-1] == '\r');
749 printf("%.*s%s%s\n", len - saw_cr_at_eol, line,
750 reset,
751 saw_cr_at_eol ? "\r" : "");
754 static void dump_sline(struct sline *sline, const char *line_prefix,
755 unsigned long cnt, int num_parent,
756 int use_color, int result_deleted)
758 unsigned long mark = (1UL<<num_parent);
759 unsigned long no_pre_delete = (2UL<<num_parent);
760 int i;
761 unsigned long lno = 0;
762 const char *c_frag = diff_get_color(use_color, DIFF_FRAGINFO);
763 const char *c_func = diff_get_color(use_color, DIFF_FUNCINFO);
764 const char *c_new = diff_get_color(use_color, DIFF_FILE_NEW);
765 const char *c_old = diff_get_color(use_color, DIFF_FILE_OLD);
766 const char *c_context = diff_get_color(use_color, DIFF_CONTEXT);
767 const char *c_reset = diff_get_color(use_color, DIFF_RESET);
769 if (result_deleted)
770 return; /* result deleted */
772 while (1) {
773 unsigned long hunk_end;
774 unsigned long rlines;
775 const char *hunk_comment = NULL;
776 unsigned long null_context = 0;
778 while (lno <= cnt && !(sline[lno].flag & mark)) {
779 if (hunk_comment_line(sline[lno].bol))
780 hunk_comment = sline[lno].bol;
781 lno++;
783 if (cnt < lno)
784 break;
785 else {
786 for (hunk_end = lno + 1; hunk_end <= cnt; hunk_end++)
787 if (!(sline[hunk_end].flag & mark))
788 break;
790 rlines = hunk_end - lno;
791 if (cnt < hunk_end)
792 rlines--; /* pointing at the last delete hunk */
794 if (!context) {
796 * Even when running with --unified=0, all
797 * lines in the hunk needs to be processed in
798 * the loop below in order to show the
799 * deletion recorded in lost_head. However,
800 * we do not want to show the resulting line
801 * with all blank context markers in such a
802 * case. Compensate.
804 unsigned long j;
805 for (j = lno; j < hunk_end; j++)
806 if (!(sline[j].flag & (mark-1)))
807 null_context++;
808 rlines -= null_context;
811 printf("%s%s", line_prefix, c_frag);
812 for (i = 0; i <= num_parent; i++) putchar(combine_marker);
813 for (i = 0; i < num_parent; i++)
814 show_parent_lno(sline, lno, hunk_end, i, null_context);
815 printf(" +%lu,%lu ", lno+1, rlines);
816 for (i = 0; i <= num_parent; i++) putchar(combine_marker);
818 if (hunk_comment) {
819 int comment_end = 0;
820 for (i = 0; i < 40; i++) {
821 int ch = hunk_comment[i] & 0xff;
822 if (!ch || ch == '\n')
823 break;
824 if (!isspace(ch))
825 comment_end = i;
827 if (comment_end)
828 printf("%s%s %s%s", c_reset,
829 c_context, c_reset,
830 c_func);
831 for (i = 0; i < comment_end; i++)
832 putchar(hunk_comment[i]);
835 printf("%s\n", c_reset);
836 while (lno < hunk_end) {
837 struct lline *ll;
838 int j;
839 unsigned long p_mask;
840 struct sline *sl = &sline[lno++];
841 ll = (sl->flag & no_pre_delete) ? NULL : sl->lost;
842 while (ll) {
843 printf("%s%s", line_prefix, c_old);
844 for (j = 0; j < num_parent; j++) {
845 if (ll->parent_map & (1UL<<j))
846 putchar('-');
847 else
848 putchar(' ');
850 show_line_to_eol(ll->line, -1, c_reset);
851 ll = ll->next;
853 if (cnt < lno)
854 break;
855 p_mask = 1;
856 fputs(line_prefix, stdout);
857 if (!(sl->flag & (mark-1))) {
859 * This sline was here to hang the
860 * lost lines in front of it.
862 if (!context)
863 continue;
864 fputs(c_context, stdout);
866 else
867 fputs(c_new, stdout);
868 for (j = 0; j < num_parent; j++) {
869 if (p_mask & sl->flag)
870 putchar('+');
871 else
872 putchar(' ');
873 p_mask <<= 1;
875 show_line_to_eol(sl->bol, sl->len, c_reset);
880 static void reuse_combine_diff(struct sline *sline, unsigned long cnt,
881 int i, int j)
883 /* We have already examined parent j and we know parent i
884 * and parent j are the same, so reuse the combined result
885 * of parent j for parent i.
887 unsigned long lno, imask, jmask;
888 imask = (1UL<<i);
889 jmask = (1UL<<j);
891 for (lno = 0; lno <= cnt; lno++) {
892 struct lline *ll = sline->lost;
893 sline->p_lno[i] = sline->p_lno[j];
894 while (ll) {
895 if (ll->parent_map & jmask)
896 ll->parent_map |= imask;
897 ll = ll->next;
899 if (sline->flag & jmask)
900 sline->flag |= imask;
901 sline++;
903 /* the overall size of the file (sline[cnt]) */
904 sline->p_lno[i] = sline->p_lno[j];
907 static void dump_quoted_path(const char *head,
908 const char *prefix,
909 const char *path,
910 const char *line_prefix,
911 const char *c_meta, const char *c_reset)
913 static struct strbuf buf = STRBUF_INIT;
915 strbuf_reset(&buf);
916 strbuf_addstr(&buf, line_prefix);
917 strbuf_addstr(&buf, c_meta);
918 strbuf_addstr(&buf, head);
919 quote_two_c_style(&buf, prefix, path, 0);
920 strbuf_addstr(&buf, c_reset);
921 puts(buf.buf);
924 static void show_combined_header(struct combine_diff_path *elem,
925 int num_parent,
926 int dense,
927 struct rev_info *rev,
928 const char *line_prefix,
929 int mode_differs,
930 int show_file_header)
932 struct diff_options *opt = &rev->diffopt;
933 int abbrev = opt->flags.full_index ? the_hash_algo->hexsz : DEFAULT_ABBREV;
934 const char *a_prefix = opt->a_prefix ? opt->a_prefix : "a/";
935 const char *b_prefix = opt->b_prefix ? opt->b_prefix : "b/";
936 const char *c_meta = diff_get_color_opt(opt, DIFF_METAINFO);
937 const char *c_reset = diff_get_color_opt(opt, DIFF_RESET);
938 const char *abb;
939 int added = 0;
940 int deleted = 0;
941 int i;
943 if (rev->loginfo && !rev->no_commit_id)
944 show_log(rev);
946 dump_quoted_path(dense ? "diff --cc " : "diff --combined ",
947 "", elem->path, line_prefix, c_meta, c_reset);
948 printf("%s%sindex ", line_prefix, c_meta);
949 for (i = 0; i < num_parent; i++) {
950 abb = find_unique_abbrev(&elem->parent[i].oid,
951 abbrev);
952 printf("%s%s", i ? "," : "", abb);
954 abb = find_unique_abbrev(&elem->oid, abbrev);
955 printf("..%s%s\n", abb, c_reset);
957 if (mode_differs) {
958 deleted = !elem->mode;
960 /* We say it was added if nobody had it */
961 added = !deleted;
962 for (i = 0; added && i < num_parent; i++)
963 if (elem->parent[i].status !=
964 DIFF_STATUS_ADDED)
965 added = 0;
966 if (added)
967 printf("%s%snew file mode %06o",
968 line_prefix, c_meta, elem->mode);
969 else {
970 if (deleted)
971 printf("%s%sdeleted file ",
972 line_prefix, c_meta);
973 printf("mode ");
974 for (i = 0; i < num_parent; i++) {
975 printf("%s%06o", i ? "," : "",
976 elem->parent[i].mode);
978 if (elem->mode)
979 printf("..%06o", elem->mode);
981 printf("%s\n", c_reset);
984 if (!show_file_header)
985 return;
987 if (rev->combined_all_paths) {
988 for (i = 0; i < num_parent; i++) {
989 char *path = filename_changed(elem->parent[i].status)
990 ? elem->parent[i].path.buf : elem->path;
991 if (elem->parent[i].status == DIFF_STATUS_ADDED)
992 dump_quoted_path("--- ", "", "/dev/null",
993 line_prefix, c_meta, c_reset);
994 else
995 dump_quoted_path("--- ", a_prefix, path,
996 line_prefix, c_meta, c_reset);
998 } else {
999 if (added)
1000 dump_quoted_path("--- ", "", "/dev/null",
1001 line_prefix, c_meta, c_reset);
1002 else
1003 dump_quoted_path("--- ", a_prefix, elem->path,
1004 line_prefix, c_meta, c_reset);
1006 if (deleted)
1007 dump_quoted_path("+++ ", "", "/dev/null",
1008 line_prefix, c_meta, c_reset);
1009 else
1010 dump_quoted_path("+++ ", b_prefix, elem->path,
1011 line_prefix, c_meta, c_reset);
1014 static void show_patch_diff(struct combine_diff_path *elem, int num_parent,
1015 int dense, int working_tree_file,
1016 struct rev_info *rev)
1018 struct diff_options *opt = &rev->diffopt;
1019 unsigned long result_size, cnt, lno;
1020 int result_deleted = 0;
1021 char *result, *cp;
1022 struct sline *sline; /* survived lines */
1023 int mode_differs = 0;
1024 int i, show_hunks;
1025 mmfile_t result_file;
1026 struct userdiff_driver *userdiff;
1027 struct userdiff_driver *textconv = NULL;
1028 int is_binary;
1029 const char *line_prefix = diff_line_prefix(opt);
1031 context = opt->context;
1032 userdiff = userdiff_find_by_path(opt->repo->index, elem->path);
1033 if (!userdiff)
1034 userdiff = userdiff_find_by_name("default");
1035 if (opt->flags.allow_textconv)
1036 textconv = userdiff_get_textconv(opt->repo, userdiff);
1038 /* Read the result of merge first */
1039 if (!working_tree_file)
1040 result = grab_blob(opt->repo, &elem->oid, elem->mode, &result_size,
1041 textconv, elem->path);
1042 else {
1043 /* Used by diff-tree to read from the working tree */
1044 struct stat st;
1045 int fd = -1;
1047 if (lstat(elem->path, &st) < 0)
1048 goto deleted_file;
1050 if (S_ISLNK(st.st_mode)) {
1051 struct strbuf buf = STRBUF_INIT;
1053 if (strbuf_readlink(&buf, elem->path, st.st_size) < 0) {
1054 error_errno("readlink(%s)", elem->path);
1055 return;
1057 result_size = buf.len;
1058 result = strbuf_detach(&buf, NULL);
1059 elem->mode = canon_mode(st.st_mode);
1060 } else if (S_ISDIR(st.st_mode)) {
1061 struct object_id oid;
1062 if (resolve_gitlink_ref(elem->path, "HEAD", &oid) < 0)
1063 result = grab_blob(opt->repo, &elem->oid,
1064 elem->mode, &result_size,
1065 NULL, NULL);
1066 else
1067 result = grab_blob(opt->repo, &oid, elem->mode,
1068 &result_size, NULL, NULL);
1069 } else if (textconv) {
1070 struct diff_filespec *df = alloc_filespec(elem->path);
1071 fill_filespec(df, &null_oid, 0, st.st_mode);
1072 result_size = fill_textconv(opt->repo, textconv, df, &result);
1073 free_filespec(df);
1074 } else if (0 <= (fd = open(elem->path, O_RDONLY))) {
1075 size_t len = xsize_t(st.st_size);
1076 ssize_t done;
1077 int is_file, i;
1079 elem->mode = canon_mode(st.st_mode);
1080 /* if symlinks don't work, assume symlink if all parents
1081 * are symlinks
1083 is_file = has_symlinks;
1084 for (i = 0; !is_file && i < num_parent; i++)
1085 is_file = !S_ISLNK(elem->parent[i].mode);
1086 if (!is_file)
1087 elem->mode = canon_mode(S_IFLNK);
1089 result_size = len;
1090 result = xmallocz(len);
1092 done = read_in_full(fd, result, len);
1093 if (done < 0)
1094 die_errno("read error '%s'", elem->path);
1095 else if (done < len)
1096 die("early EOF '%s'", elem->path);
1098 /* If not a fake symlink, apply filters, e.g. autocrlf */
1099 if (is_file) {
1100 struct strbuf buf = STRBUF_INIT;
1102 if (convert_to_git(rev->diffopt.repo->index,
1103 elem->path, result, len, &buf, global_conv_flags_eol)) {
1104 free(result);
1105 result = strbuf_detach(&buf, &len);
1106 result_size = len;
1110 else {
1111 deleted_file:
1112 result_deleted = 1;
1113 result_size = 0;
1114 elem->mode = 0;
1115 result = xcalloc(1, 1);
1118 if (0 <= fd)
1119 close(fd);
1122 for (i = 0; i < num_parent; i++) {
1123 if (elem->parent[i].mode != elem->mode) {
1124 mode_differs = 1;
1125 break;
1129 if (textconv)
1130 is_binary = 0;
1131 else if (userdiff->binary != -1)
1132 is_binary = userdiff->binary;
1133 else {
1134 is_binary = buffer_is_binary(result, result_size);
1135 for (i = 0; !is_binary && i < num_parent; i++) {
1136 char *buf;
1137 unsigned long size;
1138 buf = grab_blob(opt->repo,
1139 &elem->parent[i].oid,
1140 elem->parent[i].mode,
1141 &size, NULL, NULL);
1142 if (buffer_is_binary(buf, size))
1143 is_binary = 1;
1144 free(buf);
1147 if (is_binary) {
1148 show_combined_header(elem, num_parent, dense, rev,
1149 line_prefix, mode_differs, 0);
1150 printf("Binary files differ\n");
1151 free(result);
1152 return;
1155 for (cnt = 0, cp = result; cp < result + result_size; cp++) {
1156 if (*cp == '\n')
1157 cnt++;
1159 if (result_size && result[result_size-1] != '\n')
1160 cnt++; /* incomplete line */
1162 sline = xcalloc(st_add(cnt, 2), sizeof(*sline));
1163 sline[0].bol = result;
1164 for (lno = 0, cp = result; cp < result + result_size; cp++) {
1165 if (*cp == '\n') {
1166 sline[lno].len = cp - sline[lno].bol;
1167 lno++;
1168 if (lno < cnt)
1169 sline[lno].bol = cp + 1;
1172 if (result_size && result[result_size-1] != '\n')
1173 sline[cnt-1].len = result_size - (sline[cnt-1].bol - result);
1175 result_file.ptr = result;
1176 result_file.size = result_size;
1178 /* Even p_lno[cnt+1] is valid -- that is for the end line number
1179 * for deletion hunk at the end.
1181 sline[0].p_lno = xcalloc(st_mult(st_add(cnt, 2), num_parent), sizeof(unsigned long));
1182 for (lno = 0; lno <= cnt; lno++)
1183 sline[lno+1].p_lno = sline[lno].p_lno + num_parent;
1185 for (i = 0; i < num_parent; i++) {
1186 int j;
1187 for (j = 0; j < i; j++) {
1188 if (oideq(&elem->parent[i].oid,
1189 &elem->parent[j].oid)) {
1190 reuse_combine_diff(sline, cnt, i, j);
1191 break;
1194 if (i <= j)
1195 combine_diff(opt->repo,
1196 &elem->parent[i].oid,
1197 elem->parent[i].mode,
1198 &result_file, sline,
1199 cnt, i, num_parent, result_deleted,
1200 textconv, elem->path, opt->xdl_opts);
1203 show_hunks = make_hunks(sline, cnt, num_parent, dense);
1205 if (show_hunks || mode_differs || working_tree_file) {
1206 show_combined_header(elem, num_parent, dense, rev,
1207 line_prefix, mode_differs, 1);
1208 dump_sline(sline, line_prefix, cnt, num_parent,
1209 opt->use_color, result_deleted);
1211 free(result);
1213 for (lno = 0; lno < cnt; lno++) {
1214 if (sline[lno].lost) {
1215 struct lline *ll = sline[lno].lost;
1216 while (ll) {
1217 struct lline *tmp = ll;
1218 ll = ll->next;
1219 free(tmp);
1223 free(sline[0].p_lno);
1224 free(sline);
1227 static void show_raw_diff(struct combine_diff_path *p, int num_parent, struct rev_info *rev)
1229 struct diff_options *opt = &rev->diffopt;
1230 int line_termination, inter_name_termination, i;
1231 const char *line_prefix = diff_line_prefix(opt);
1233 line_termination = opt->line_termination;
1234 inter_name_termination = '\t';
1235 if (!line_termination)
1236 inter_name_termination = 0;
1238 if (rev->loginfo && !rev->no_commit_id)
1239 show_log(rev);
1242 if (opt->output_format & DIFF_FORMAT_RAW) {
1243 printf("%s", line_prefix);
1245 /* As many colons as there are parents */
1246 for (i = 0; i < num_parent; i++)
1247 putchar(':');
1249 /* Show the modes */
1250 for (i = 0; i < num_parent; i++)
1251 printf("%06o ", p->parent[i].mode);
1252 printf("%06o", p->mode);
1254 /* Show sha1's */
1255 for (i = 0; i < num_parent; i++)
1256 printf(" %s", diff_aligned_abbrev(&p->parent[i].oid,
1257 opt->abbrev));
1258 printf(" %s ", diff_aligned_abbrev(&p->oid, opt->abbrev));
1261 if (opt->output_format & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS)) {
1262 for (i = 0; i < num_parent; i++)
1263 putchar(p->parent[i].status);
1264 putchar(inter_name_termination);
1267 for (i = 0; i < num_parent; i++)
1268 if (rev->combined_all_paths) {
1269 if (filename_changed(p->parent[i].status))
1270 write_name_quoted(p->parent[i].path.buf, stdout,
1271 inter_name_termination);
1272 else
1273 write_name_quoted(p->path, stdout,
1274 inter_name_termination);
1276 write_name_quoted(p->path, stdout, line_termination);
1280 * The result (p->elem) is from the working tree and their
1281 * parents are typically from multiple stages during a merge
1282 * (i.e. diff-files) or the state in HEAD and in the index
1283 * (i.e. diff-index).
1285 void show_combined_diff(struct combine_diff_path *p,
1286 int num_parent,
1287 int dense,
1288 struct rev_info *rev)
1290 struct diff_options *opt = &rev->diffopt;
1292 if (opt->output_format & (DIFF_FORMAT_RAW |
1293 DIFF_FORMAT_NAME |
1294 DIFF_FORMAT_NAME_STATUS))
1295 show_raw_diff(p, num_parent, rev);
1296 else if (opt->output_format & DIFF_FORMAT_PATCH)
1297 show_patch_diff(p, num_parent, dense, 1, rev);
1300 static void free_combined_pair(struct diff_filepair *pair)
1302 free(pair->two);
1303 free(pair);
1307 * A combine_diff_path expresses N parents on the LHS against 1 merge
1308 * result. Synthesize a diff_filepair that has N entries on the "one"
1309 * side and 1 entry on the "two" side.
1311 * In the future, we might want to add more data to combine_diff_path
1312 * so that we can fill fields we are ignoring (most notably, size) here,
1313 * but currently nobody uses it, so this should suffice for now.
1315 static struct diff_filepair *combined_pair(struct combine_diff_path *p,
1316 int num_parent)
1318 int i;
1319 struct diff_filepair *pair;
1320 struct diff_filespec *pool;
1322 pair = xmalloc(sizeof(*pair));
1323 pool = xcalloc(st_add(num_parent, 1), sizeof(struct diff_filespec));
1324 pair->one = pool + 1;
1325 pair->two = pool;
1327 for (i = 0; i < num_parent; i++) {
1328 pair->one[i].path = p->path;
1329 pair->one[i].mode = p->parent[i].mode;
1330 oidcpy(&pair->one[i].oid, &p->parent[i].oid);
1331 pair->one[i].oid_valid = !is_null_oid(&p->parent[i].oid);
1332 pair->one[i].has_more_entries = 1;
1334 pair->one[num_parent - 1].has_more_entries = 0;
1336 pair->two->path = p->path;
1337 pair->two->mode = p->mode;
1338 oidcpy(&pair->two->oid, &p->oid);
1339 pair->two->oid_valid = !is_null_oid(&p->oid);
1340 return pair;
1343 static void handle_combined_callback(struct diff_options *opt,
1344 struct combine_diff_path *paths,
1345 int num_parent,
1346 int num_paths)
1348 struct combine_diff_path *p;
1349 struct diff_queue_struct q;
1350 int i;
1352 q.queue = xcalloc(num_paths, sizeof(struct diff_filepair *));
1353 q.alloc = num_paths;
1354 q.nr = num_paths;
1355 for (i = 0, p = paths; p; p = p->next)
1356 q.queue[i++] = combined_pair(p, num_parent);
1357 opt->format_callback(&q, opt, opt->format_callback_data);
1358 for (i = 0; i < num_paths; i++)
1359 free_combined_pair(q.queue[i]);
1360 free(q.queue);
1363 static const char *path_path(void *obj)
1365 struct combine_diff_path *path = (struct combine_diff_path *)obj;
1367 return path->path;
1371 * Diff stat formats which we always compute solely against the first parent.
1373 #define STAT_FORMAT_MASK (DIFF_FORMAT_NUMSTAT \
1374 | DIFF_FORMAT_SHORTSTAT \
1375 | DIFF_FORMAT_SUMMARY \
1376 | DIFF_FORMAT_DIRSTAT \
1377 | DIFF_FORMAT_DIFFSTAT)
1379 /* find set of paths that every parent touches */
1380 static struct combine_diff_path *find_paths_generic(const struct object_id *oid,
1381 const struct oid_array *parents,
1382 struct diff_options *opt,
1383 int combined_all_paths)
1385 struct combine_diff_path *paths = NULL;
1386 int i, num_parent = parents->nr;
1388 int output_format = opt->output_format;
1389 const char *orderfile = opt->orderfile;
1391 opt->output_format = DIFF_FORMAT_NO_OUTPUT;
1392 /* tell diff_tree to emit paths in sorted (=tree) order */
1393 opt->orderfile = NULL;
1395 /* D(A,P1...Pn) = D(A,P1) ^ ... ^ D(A,Pn) (wrt paths) */
1396 for (i = 0; i < num_parent; i++) {
1398 * show stat against the first parent even when doing
1399 * combined diff.
1401 int stat_opt = output_format & STAT_FORMAT_MASK;
1402 if (i == 0 && stat_opt)
1403 opt->output_format = stat_opt;
1404 else
1405 opt->output_format = DIFF_FORMAT_NO_OUTPUT;
1406 diff_tree_oid(&parents->oid[i], oid, "", opt);
1407 diffcore_std(opt);
1408 paths = intersect_paths(paths, i, num_parent,
1409 combined_all_paths);
1411 /* if showing diff, show it in requested order */
1412 if (opt->output_format != DIFF_FORMAT_NO_OUTPUT &&
1413 orderfile) {
1414 diffcore_order(orderfile);
1417 diff_flush(opt);
1420 opt->output_format = output_format;
1421 opt->orderfile = orderfile;
1422 return paths;
1427 * find set of paths that everybody touches, assuming diff is run without
1428 * rename/copy detection, etc, comparing all trees simultaneously (= faster).
1430 static struct combine_diff_path *find_paths_multitree(
1431 const struct object_id *oid, const struct oid_array *parents,
1432 struct diff_options *opt)
1434 int i, nparent = parents->nr;
1435 const struct object_id **parents_oid;
1436 struct combine_diff_path paths_head;
1437 struct strbuf base;
1439 ALLOC_ARRAY(parents_oid, nparent);
1440 for (i = 0; i < nparent; i++)
1441 parents_oid[i] = &parents->oid[i];
1443 /* fake list head, so worker can assume it is non-NULL */
1444 paths_head.next = NULL;
1446 strbuf_init(&base, PATH_MAX);
1447 diff_tree_paths(&paths_head, oid, parents_oid, nparent, &base, opt);
1449 strbuf_release(&base);
1450 free(parents_oid);
1451 return paths_head.next;
1455 void diff_tree_combined(const struct object_id *oid,
1456 const struct oid_array *parents,
1457 int dense,
1458 struct rev_info *rev)
1460 struct diff_options *opt = &rev->diffopt;
1461 struct diff_options diffopts;
1462 struct combine_diff_path *p, *paths;
1463 int i, num_paths, needsep, show_log_first, num_parent = parents->nr;
1464 int need_generic_pathscan;
1466 /* nothing to do, if no parents */
1467 if (!num_parent)
1468 return;
1470 show_log_first = !!rev->loginfo && !rev->no_commit_id;
1471 needsep = 0;
1472 if (show_log_first) {
1473 show_log(rev);
1475 if (rev->verbose_header && opt->output_format &&
1476 opt->output_format != DIFF_FORMAT_NO_OUTPUT &&
1477 !commit_format_is_empty(rev->commit_format))
1478 printf("%s%c", diff_line_prefix(opt),
1479 opt->line_termination);
1482 diffopts = *opt;
1483 copy_pathspec(&diffopts.pathspec, &opt->pathspec);
1484 diffopts.flags.recursive = 1;
1485 diffopts.flags.allow_external = 0;
1487 /* find set of paths that everybody touches
1489 * NOTE
1491 * Diffcore transformations are bound to diff_filespec and logic
1492 * comparing two entries - i.e. they do not apply directly to combine
1493 * diff.
1495 * If some of such transformations is requested - we launch generic
1496 * path scanning, which works significantly slower compared to
1497 * simultaneous all-trees-in-one-go scan in find_paths_multitree().
1499 * TODO some of the filters could be ported to work on
1500 * combine_diff_paths - i.e. all functionality that skips paths, so in
1501 * theory, we could end up having only multitree path scanning.
1503 * NOTE please keep this semantically in sync with diffcore_std()
1505 need_generic_pathscan = opt->skip_stat_unmatch ||
1506 opt->flags.follow_renames ||
1507 opt->break_opt != -1 ||
1508 opt->detect_rename ||
1509 (opt->pickaxe_opts & DIFF_PICKAXE_KINDS_MASK) ||
1510 opt->filter;
1513 if (need_generic_pathscan) {
1515 * NOTE generic case also handles --stat, as it computes
1516 * diff(sha1,parent_i) for all i to do the job, specifically
1517 * for parent0.
1519 paths = find_paths_generic(oid, parents, &diffopts,
1520 rev->combined_all_paths);
1522 else {
1523 int stat_opt;
1524 paths = find_paths_multitree(oid, parents, &diffopts);
1527 * show stat against the first parent even
1528 * when doing combined diff.
1530 stat_opt = opt->output_format & STAT_FORMAT_MASK;
1531 if (stat_opt) {
1532 diffopts.output_format = stat_opt;
1534 diff_tree_oid(&parents->oid[0], oid, "", &diffopts);
1535 diffcore_std(&diffopts);
1536 if (opt->orderfile)
1537 diffcore_order(opt->orderfile);
1538 diff_flush(&diffopts);
1542 /* find out number of surviving paths */
1543 for (num_paths = 0, p = paths; p; p = p->next)
1544 num_paths++;
1546 /* order paths according to diffcore_order */
1547 if (opt->orderfile && num_paths) {
1548 struct obj_order *o;
1550 ALLOC_ARRAY(o, num_paths);
1551 for (i = 0, p = paths; p; p = p->next, i++)
1552 o[i].obj = p;
1553 order_objects(opt->orderfile, path_path, o, num_paths);
1554 for (i = 0; i < num_paths - 1; i++) {
1555 p = o[i].obj;
1556 p->next = o[i+1].obj;
1559 p = o[num_paths-1].obj;
1560 p->next = NULL;
1561 paths = o[0].obj;
1562 free(o);
1566 if (num_paths) {
1567 if (opt->output_format & (DIFF_FORMAT_RAW |
1568 DIFF_FORMAT_NAME |
1569 DIFF_FORMAT_NAME_STATUS)) {
1570 for (p = paths; p; p = p->next)
1571 show_raw_diff(p, num_parent, rev);
1572 needsep = 1;
1574 else if (opt->output_format & STAT_FORMAT_MASK)
1575 needsep = 1;
1576 else if (opt->output_format & DIFF_FORMAT_CALLBACK)
1577 handle_combined_callback(opt, paths, num_parent, num_paths);
1579 if (opt->output_format & DIFF_FORMAT_PATCH) {
1580 if (needsep)
1581 printf("%s%c", diff_line_prefix(opt),
1582 opt->line_termination);
1583 for (p = paths; p; p = p->next)
1584 show_patch_diff(p, num_parent, dense,
1585 0, rev);
1589 /* Clean things up */
1590 while (paths) {
1591 struct combine_diff_path *tmp = paths;
1592 paths = paths->next;
1593 for (i = 0; i < num_parent; i++)
1594 if (rev->combined_all_paths &&
1595 filename_changed(tmp->parent[i].status))
1596 strbuf_release(&tmp->parent[i].path);
1597 free(tmp);
1600 clear_pathspec(&diffopts.pathspec);
1603 void diff_tree_combined_merge(const struct commit *commit, int dense,
1604 struct rev_info *rev)
1606 struct commit_list *parent = get_saved_parents(rev, commit);
1607 struct oid_array parents = OID_ARRAY_INIT;
1609 while (parent) {
1610 oid_array_append(&parents, &parent->item->object.oid);
1611 parent = parent->next;
1613 diff_tree_combined(&commit->object.oid, &parents, dense, rev);
1614 oid_array_clear(&parents);