commit: reject invalid UTF-8 codepoints
[git/mingw/j6t.git] / combine-diff.c
blob6dc06093d3c1eb6f3ab7a9cc79f0f192e16dc617
1 #include "cache.h"
2 #include "commit.h"
3 #include "blob.h"
4 #include "diff.h"
5 #include "diffcore.h"
6 #include "quote.h"
7 #include "xdiff-interface.h"
8 #include "xdiff/xmacros.h"
9 #include "log-tree.h"
10 #include "refs.h"
11 #include "userdiff.h"
12 #include "sha1-array.h"
14 static struct combine_diff_path *intersect_paths(struct combine_diff_path *curr, int n, int num_parent)
16 struct diff_queue_struct *q = &diff_queued_diff;
17 struct combine_diff_path *p;
18 int i;
20 if (!n) {
21 struct combine_diff_path *list = NULL, **tail = &list;
22 for (i = 0; i < q->nr; i++) {
23 int len;
24 const char *path;
25 if (diff_unmodified_pair(q->queue[i]))
26 continue;
27 path = q->queue[i]->two->path;
28 len = strlen(path);
29 p = xmalloc(combine_diff_path_size(num_parent, len));
30 p->path = (char *) &(p->parent[num_parent]);
31 memcpy(p->path, path, len);
32 p->path[len] = 0;
33 p->len = len;
34 p->next = NULL;
35 memset(p->parent, 0,
36 sizeof(p->parent[0]) * num_parent);
38 hashcpy(p->sha1, q->queue[i]->two->sha1);
39 p->mode = q->queue[i]->two->mode;
40 hashcpy(p->parent[n].sha1, q->queue[i]->one->sha1);
41 p->parent[n].mode = q->queue[i]->one->mode;
42 p->parent[n].status = q->queue[i]->status;
43 *tail = p;
44 tail = &p->next;
46 return list;
49 for (p = curr; p; p = p->next) {
50 int found = 0;
51 if (!p->len)
52 continue;
53 for (i = 0; i < q->nr; i++) {
54 const char *path;
55 int len;
57 if (diff_unmodified_pair(q->queue[i]))
58 continue;
59 path = q->queue[i]->two->path;
60 len = strlen(path);
61 if (len == p->len && !memcmp(path, p->path, len)) {
62 found = 1;
63 hashcpy(p->parent[n].sha1, q->queue[i]->one->sha1);
64 p->parent[n].mode = q->queue[i]->one->mode;
65 p->parent[n].status = q->queue[i]->status;
66 break;
69 if (!found)
70 p->len = 0;
72 return curr;
75 /* Lines lost from parent */
76 struct lline {
77 struct lline *next, *prev;
78 int len;
79 unsigned long parent_map;
80 char line[FLEX_ARRAY];
83 /* Lines lost from current parent (before coalescing) */
84 struct plost {
85 struct lline *lost_head, *lost_tail;
86 int len;
89 /* Lines surviving in the merge result */
90 struct sline {
91 /* Accumulated and coalesced lost lines */
92 struct lline *lost;
93 int lenlost;
94 struct plost plost;
95 char *bol;
96 int len;
97 /* bit 0 up to (N-1) are on if the parent has this line (i.e.
98 * we did not change it).
99 * bit N is used for "interesting" lines, including context.
100 * bit (N+1) is used for "do not show deletion before this".
102 unsigned long flag;
103 unsigned long *p_lno;
106 static int match_string_spaces(const char *line1, int len1,
107 const char *line2, int len2,
108 long flags)
110 if (flags & XDF_WHITESPACE_FLAGS) {
111 for (; len1 > 0 && XDL_ISSPACE(line1[len1 - 1]); len1--);
112 for (; len2 > 0 && XDL_ISSPACE(line2[len2 - 1]); len2--);
115 if (!(flags & (XDF_IGNORE_WHITESPACE | XDF_IGNORE_WHITESPACE_CHANGE)))
116 return (len1 == len2 && !memcmp(line1, line2, len1));
118 while (len1 > 0 && len2 > 0) {
119 len1--;
120 len2--;
121 if (XDL_ISSPACE(line1[len1]) || XDL_ISSPACE(line2[len2])) {
122 if ((flags & XDF_IGNORE_WHITESPACE_CHANGE) &&
123 (!XDL_ISSPACE(line1[len1]) || !XDL_ISSPACE(line2[len2])))
124 return 0;
126 for (; len1 > 0 && XDL_ISSPACE(line1[len1]); len1--);
127 for (; len2 > 0 && XDL_ISSPACE(line2[len2]); len2--);
129 if (line1[len1] != line2[len2])
130 return 0;
133 if (flags & XDF_IGNORE_WHITESPACE) {
134 /* Consume remaining spaces */
135 for (; len1 > 0 && XDL_ISSPACE(line1[len1 - 1]); len1--);
136 for (; len2 > 0 && XDL_ISSPACE(line2[len2 - 1]); len2--);
139 /* We matched full line1 and line2 */
140 if (!len1 && !len2)
141 return 1;
143 return 0;
146 enum coalesce_direction { MATCH, BASE, NEW };
148 /* Coalesce new lines into base by finding LCS */
149 static struct lline *coalesce_lines(struct lline *base, int *lenbase,
150 struct lline *new, int lennew,
151 unsigned long parent, long flags)
153 int **lcs;
154 enum coalesce_direction **directions;
155 struct lline *baseend, *newend = NULL;
156 int i, j, origbaselen = *lenbase;
158 if (new == NULL)
159 return base;
161 if (base == NULL) {
162 *lenbase = lennew;
163 return new;
167 * Coalesce new lines into base by finding the LCS
168 * - Create the table to run dynamic programing
169 * - Compute the LCS
170 * - Then reverse read the direction structure:
171 * - If we have MATCH, assign parent to base flag, and consume
172 * both baseend and newend
173 * - Else if we have BASE, consume baseend
174 * - Else if we have NEW, insert newend lline into base and
175 * consume newend
177 lcs = xcalloc(origbaselen + 1, sizeof(int*));
178 directions = xcalloc(origbaselen + 1, sizeof(enum coalesce_direction*));
179 for (i = 0; i < origbaselen + 1; i++) {
180 lcs[i] = xcalloc(lennew + 1, sizeof(int));
181 directions[i] = xcalloc(lennew + 1, sizeof(enum coalesce_direction));
182 directions[i][0] = BASE;
184 for (j = 1; j < lennew + 1; j++)
185 directions[0][j] = NEW;
187 for (i = 1, baseend = base; i < origbaselen + 1; i++) {
188 for (j = 1, newend = new; j < lennew + 1; j++) {
189 if (match_string_spaces(baseend->line, baseend->len,
190 newend->line, newend->len, flags)) {
191 lcs[i][j] = lcs[i - 1][j - 1] + 1;
192 directions[i][j] = MATCH;
193 } else if (lcs[i][j - 1] >= lcs[i - 1][j]) {
194 lcs[i][j] = lcs[i][j - 1];
195 directions[i][j] = NEW;
196 } else {
197 lcs[i][j] = lcs[i - 1][j];
198 directions[i][j] = BASE;
200 if (newend->next)
201 newend = newend->next;
203 if (baseend->next)
204 baseend = baseend->next;
207 for (i = 0; i < origbaselen + 1; i++)
208 free(lcs[i]);
209 free(lcs);
211 /* At this point, baseend and newend point to the end of each lists */
212 i--;
213 j--;
214 while (i != 0 || j != 0) {
215 if (directions[i][j] == MATCH) {
216 baseend->parent_map |= 1<<parent;
217 baseend = baseend->prev;
218 newend = newend->prev;
219 i--;
220 j--;
221 } else if (directions[i][j] == NEW) {
222 struct lline *lline;
224 lline = newend;
225 /* Remove lline from new list and update newend */
226 if (lline->prev)
227 lline->prev->next = lline->next;
228 else
229 new = lline->next;
230 if (lline->next)
231 lline->next->prev = lline->prev;
233 newend = lline->prev;
234 j--;
236 /* Add lline to base list */
237 if (baseend) {
238 lline->next = baseend->next;
239 lline->prev = baseend;
240 if (lline->prev)
241 lline->prev->next = lline;
243 else {
244 lline->next = base;
245 base = lline;
247 (*lenbase)++;
249 if (lline->next)
250 lline->next->prev = lline;
252 } else {
253 baseend = baseend->prev;
254 i--;
258 newend = new;
259 while (newend) {
260 struct lline *lline = newend;
261 newend = newend->next;
262 free(lline);
265 for (i = 0; i < origbaselen + 1; i++)
266 free(directions[i]);
267 free(directions);
269 return base;
272 static char *grab_blob(const unsigned char *sha1, unsigned int mode,
273 unsigned long *size, struct userdiff_driver *textconv,
274 const char *path)
276 char *blob;
277 enum object_type type;
279 if (S_ISGITLINK(mode)) {
280 blob = xmalloc(100);
281 *size = snprintf(blob, 100,
282 "Subproject commit %s\n", sha1_to_hex(sha1));
283 } else if (is_null_sha1(sha1)) {
284 /* deleted blob */
285 *size = 0;
286 return xcalloc(1, 1);
287 } else if (textconv) {
288 struct diff_filespec *df = alloc_filespec(path);
289 fill_filespec(df, sha1, 1, mode);
290 *size = fill_textconv(textconv, df, &blob);
291 free_filespec(df);
292 } else {
293 blob = read_sha1_file(sha1, &type, size);
294 if (type != OBJ_BLOB)
295 die("object '%s' is not a blob!", sha1_to_hex(sha1));
297 return blob;
300 static void append_lost(struct sline *sline, int n, const char *line, int len)
302 struct lline *lline;
303 unsigned long this_mask = (1UL<<n);
304 if (line[len-1] == '\n')
305 len--;
307 lline = xmalloc(sizeof(*lline) + len + 1);
308 lline->len = len;
309 lline->next = NULL;
310 lline->prev = sline->plost.lost_tail;
311 if (lline->prev)
312 lline->prev->next = lline;
313 else
314 sline->plost.lost_head = lline;
315 sline->plost.lost_tail = lline;
316 sline->plost.len++;
317 lline->parent_map = this_mask;
318 memcpy(lline->line, line, len);
319 lline->line[len] = 0;
322 struct combine_diff_state {
323 unsigned int lno;
324 int ob, on, nb, nn;
325 unsigned long nmask;
326 int num_parent;
327 int n;
328 struct sline *sline;
329 struct sline *lost_bucket;
332 static void consume_line(void *state_, char *line, unsigned long len)
334 struct combine_diff_state *state = state_;
335 if (5 < len && !memcmp("@@ -", line, 4)) {
336 if (parse_hunk_header(line, len,
337 &state->ob, &state->on,
338 &state->nb, &state->nn))
339 return;
340 state->lno = state->nb;
341 if (state->nn == 0) {
342 /* @@ -X,Y +N,0 @@ removed Y lines
343 * that would have come *after* line N
344 * in the result. Our lost buckets hang
345 * to the line after the removed lines,
347 * Note that this is correct even when N == 0,
348 * in which case the hunk removes the first
349 * line in the file.
351 state->lost_bucket = &state->sline[state->nb];
352 if (!state->nb)
353 state->nb = 1;
354 } else {
355 state->lost_bucket = &state->sline[state->nb-1];
357 if (!state->sline[state->nb-1].p_lno)
358 state->sline[state->nb-1].p_lno =
359 xcalloc(state->num_parent,
360 sizeof(unsigned long));
361 state->sline[state->nb-1].p_lno[state->n] = state->ob;
362 return;
364 if (!state->lost_bucket)
365 return; /* not in any hunk yet */
366 switch (line[0]) {
367 case '-':
368 append_lost(state->lost_bucket, state->n, line+1, len-1);
369 break;
370 case '+':
371 state->sline[state->lno-1].flag |= state->nmask;
372 state->lno++;
373 break;
377 static void combine_diff(const unsigned char *parent, unsigned int mode,
378 mmfile_t *result_file,
379 struct sline *sline, unsigned int cnt, int n,
380 int num_parent, int result_deleted,
381 struct userdiff_driver *textconv,
382 const char *path, long flags)
384 unsigned int p_lno, lno;
385 unsigned long nmask = (1UL << n);
386 xpparam_t xpp;
387 xdemitconf_t xecfg;
388 mmfile_t parent_file;
389 struct combine_diff_state state;
390 unsigned long sz;
392 if (result_deleted)
393 return; /* result deleted */
395 parent_file.ptr = grab_blob(parent, mode, &sz, textconv, path);
396 parent_file.size = sz;
397 memset(&xpp, 0, sizeof(xpp));
398 xpp.flags = flags;
399 memset(&xecfg, 0, sizeof(xecfg));
400 memset(&state, 0, sizeof(state));
401 state.nmask = nmask;
402 state.sline = sline;
403 state.lno = 1;
404 state.num_parent = num_parent;
405 state.n = n;
407 xdi_diff_outf(&parent_file, result_file, consume_line, &state,
408 &xpp, &xecfg);
409 free(parent_file.ptr);
411 /* Assign line numbers for this parent.
413 * sline[lno].p_lno[n] records the first line number
414 * (counting from 1) for parent N if the final hunk display
415 * started by showing sline[lno] (possibly showing the lost
416 * lines attached to it first).
418 for (lno = 0, p_lno = 1; lno <= cnt; lno++) {
419 struct lline *ll;
420 sline[lno].p_lno[n] = p_lno;
422 /* Coalesce new lines */
423 if (sline[lno].plost.lost_head) {
424 struct sline *sl = &sline[lno];
425 sl->lost = coalesce_lines(sl->lost, &sl->lenlost,
426 sl->plost.lost_head,
427 sl->plost.len, n, flags);
428 sl->plost.lost_head = sl->plost.lost_tail = NULL;
429 sl->plost.len = 0;
432 /* How many lines would this sline advance the p_lno? */
433 ll = sline[lno].lost;
434 while (ll) {
435 if (ll->parent_map & nmask)
436 p_lno++; /* '-' means parent had it */
437 ll = ll->next;
439 if (lno < cnt && !(sline[lno].flag & nmask))
440 p_lno++; /* no '+' means parent had it */
442 sline[lno].p_lno[n] = p_lno; /* trailer */
445 static unsigned long context = 3;
446 static char combine_marker = '@';
448 static int interesting(struct sline *sline, unsigned long all_mask)
450 /* If some parents lost lines here, or if we have added to
451 * some parent, it is interesting.
453 return ((sline->flag & all_mask) || sline->lost);
456 static unsigned long adjust_hunk_tail(struct sline *sline,
457 unsigned long all_mask,
458 unsigned long hunk_begin,
459 unsigned long i)
461 /* i points at the first uninteresting line. If the last line
462 * of the hunk was interesting only because it has some
463 * deletion, then it is not all that interesting for the
464 * purpose of giving trailing context lines. This is because
465 * we output '-' line and then unmodified sline[i-1] itself in
466 * that case which gives us one extra context line.
468 if ((hunk_begin + 1 <= i) && !(sline[i-1].flag & all_mask))
469 i--;
470 return i;
473 static unsigned long find_next(struct sline *sline,
474 unsigned long mark,
475 unsigned long i,
476 unsigned long cnt,
477 int look_for_uninteresting)
479 /* We have examined up to i-1 and are about to look at i.
480 * Find next interesting or uninteresting line. Here,
481 * "interesting" does not mean interesting(), but marked by
482 * the give_context() function below (i.e. it includes context
483 * lines that are not interesting to interesting() function
484 * that are surrounded by interesting() ones.
486 while (i <= cnt)
487 if (look_for_uninteresting
488 ? !(sline[i].flag & mark)
489 : (sline[i].flag & mark))
490 return i;
491 else
492 i++;
493 return i;
496 static int give_context(struct sline *sline, unsigned long cnt, int num_parent)
498 unsigned long all_mask = (1UL<<num_parent) - 1;
499 unsigned long mark = (1UL<<num_parent);
500 unsigned long no_pre_delete = (2UL<<num_parent);
501 unsigned long i;
503 /* Two groups of interesting lines may have a short gap of
504 * uninteresting lines. Connect such groups to give them a
505 * bit of context.
507 * We first start from what the interesting() function says,
508 * and mark them with "mark", and paint context lines with the
509 * mark. So interesting() would still say false for such context
510 * lines but they are treated as "interesting" in the end.
512 i = find_next(sline, mark, 0, cnt, 0);
513 if (cnt < i)
514 return 0;
516 while (i <= cnt) {
517 unsigned long j = (context < i) ? (i - context) : 0;
518 unsigned long k;
520 /* Paint a few lines before the first interesting line. */
521 while (j < i) {
522 if (!(sline[j].flag & mark))
523 sline[j].flag |= no_pre_delete;
524 sline[j++].flag |= mark;
527 again:
528 /* we know up to i is to be included. where does the
529 * next uninteresting one start?
531 j = find_next(sline, mark, i, cnt, 1);
532 if (cnt < j)
533 break; /* the rest are all interesting */
535 /* lookahead context lines */
536 k = find_next(sline, mark, j, cnt, 0);
537 j = adjust_hunk_tail(sline, all_mask, i, j);
539 if (k < j + context) {
540 /* k is interesting and [j,k) are not, but
541 * paint them interesting because the gap is small.
543 while (j < k)
544 sline[j++].flag |= mark;
545 i = k;
546 goto again;
549 /* j is the first uninteresting line and there is
550 * no overlap beyond it within context lines. Paint
551 * the trailing edge a bit.
553 i = k;
554 k = (j + context < cnt+1) ? j + context : cnt+1;
555 while (j < k)
556 sline[j++].flag |= mark;
558 return 1;
561 static int make_hunks(struct sline *sline, unsigned long cnt,
562 int num_parent, int dense)
564 unsigned long all_mask = (1UL<<num_parent) - 1;
565 unsigned long mark = (1UL<<num_parent);
566 unsigned long i;
567 int has_interesting = 0;
569 for (i = 0; i <= cnt; i++) {
570 if (interesting(&sline[i], all_mask))
571 sline[i].flag |= mark;
572 else
573 sline[i].flag &= ~mark;
575 if (!dense)
576 return give_context(sline, cnt, num_parent);
578 /* Look at each hunk, and if we have changes from only one
579 * parent, or the changes are the same from all but one
580 * parent, mark that uninteresting.
582 i = 0;
583 while (i <= cnt) {
584 unsigned long j, hunk_begin, hunk_end;
585 unsigned long same_diff;
586 while (i <= cnt && !(sline[i].flag & mark))
587 i++;
588 if (cnt < i)
589 break; /* No more interesting hunks */
590 hunk_begin = i;
591 for (j = i + 1; j <= cnt; j++) {
592 if (!(sline[j].flag & mark)) {
593 /* Look beyond the end to see if there
594 * is an interesting line after this
595 * hunk within context span.
597 unsigned long la; /* lookahead */
598 int contin = 0;
599 la = adjust_hunk_tail(sline, all_mask,
600 hunk_begin, j);
601 la = (la + context < cnt + 1) ?
602 (la + context) : cnt + 1;
603 while (la && j <= --la) {
604 if (sline[la].flag & mark) {
605 contin = 1;
606 break;
609 if (!contin)
610 break;
611 j = la;
614 hunk_end = j;
616 /* [i..hunk_end) are interesting. Now is it really
617 * interesting? We check if there are only two versions
618 * and the result matches one of them. That is, we look
619 * at:
620 * (+) line, which records lines added to which parents;
621 * this line appears in the result.
622 * (-) line, which records from what parents the line
623 * was removed; this line does not appear in the result.
624 * then check the set of parents the result has difference
625 * from, from all lines. If there are lines that has
626 * different set of parents that the result has differences
627 * from, that means we have more than two versions.
629 * Even when we have only two versions, if the result does
630 * not match any of the parents, the it should be considered
631 * interesting. In such a case, we would have all '+' line.
632 * After passing the above "two versions" test, that would
633 * appear as "the same set of parents" to be "all parents".
635 same_diff = 0;
636 has_interesting = 0;
637 for (j = i; j < hunk_end && !has_interesting; j++) {
638 unsigned long this_diff = sline[j].flag & all_mask;
639 struct lline *ll = sline[j].lost;
640 if (this_diff) {
641 /* This has some changes. Is it the
642 * same as others?
644 if (!same_diff)
645 same_diff = this_diff;
646 else if (same_diff != this_diff) {
647 has_interesting = 1;
648 break;
651 while (ll && !has_interesting) {
652 /* Lost this line from these parents;
653 * who are they? Are they the same?
655 this_diff = ll->parent_map;
656 if (!same_diff)
657 same_diff = this_diff;
658 else if (same_diff != this_diff) {
659 has_interesting = 1;
661 ll = ll->next;
665 if (!has_interesting && same_diff != all_mask) {
666 /* This hunk is not that interesting after all */
667 for (j = hunk_begin; j < hunk_end; j++)
668 sline[j].flag &= ~mark;
670 i = hunk_end;
673 has_interesting = give_context(sline, cnt, num_parent);
674 return has_interesting;
677 static void show_parent_lno(struct sline *sline, unsigned long l0, unsigned long l1, int n, unsigned long null_context)
679 l0 = sline[l0].p_lno[n];
680 l1 = sline[l1].p_lno[n];
681 printf(" -%lu,%lu", l0, l1-l0-null_context);
684 static int hunk_comment_line(const char *bol)
686 int ch;
688 if (!bol)
689 return 0;
690 ch = *bol & 0xff;
691 return (isalpha(ch) || ch == '_' || ch == '$');
694 static void show_line_to_eol(const char *line, int len, const char *reset)
696 int saw_cr_at_eol = 0;
697 if (len < 0)
698 len = strlen(line);
699 saw_cr_at_eol = (len && line[len-1] == '\r');
701 printf("%.*s%s%s\n", len - saw_cr_at_eol, line,
702 reset,
703 saw_cr_at_eol ? "\r" : "");
706 static void dump_sline(struct sline *sline, const char *line_prefix,
707 unsigned long cnt, int num_parent,
708 int use_color, int result_deleted)
710 unsigned long mark = (1UL<<num_parent);
711 unsigned long no_pre_delete = (2UL<<num_parent);
712 int i;
713 unsigned long lno = 0;
714 const char *c_frag = diff_get_color(use_color, DIFF_FRAGINFO);
715 const char *c_func = diff_get_color(use_color, DIFF_FUNCINFO);
716 const char *c_new = diff_get_color(use_color, DIFF_FILE_NEW);
717 const char *c_old = diff_get_color(use_color, DIFF_FILE_OLD);
718 const char *c_plain = diff_get_color(use_color, DIFF_PLAIN);
719 const char *c_reset = diff_get_color(use_color, DIFF_RESET);
721 if (result_deleted)
722 return; /* result deleted */
724 while (1) {
725 unsigned long hunk_end;
726 unsigned long rlines;
727 const char *hunk_comment = NULL;
728 unsigned long null_context = 0;
730 while (lno <= cnt && !(sline[lno].flag & mark)) {
731 if (hunk_comment_line(sline[lno].bol))
732 hunk_comment = sline[lno].bol;
733 lno++;
735 if (cnt < lno)
736 break;
737 else {
738 for (hunk_end = lno + 1; hunk_end <= cnt; hunk_end++)
739 if (!(sline[hunk_end].flag & mark))
740 break;
742 rlines = hunk_end - lno;
743 if (cnt < hunk_end)
744 rlines--; /* pointing at the last delete hunk */
746 if (!context) {
748 * Even when running with --unified=0, all
749 * lines in the hunk needs to be processed in
750 * the loop below in order to show the
751 * deletion recorded in lost_head. However,
752 * we do not want to show the resulting line
753 * with all blank context markers in such a
754 * case. Compensate.
756 unsigned long j;
757 for (j = lno; j < hunk_end; j++)
758 if (!(sline[j].flag & (mark-1)))
759 null_context++;
760 rlines -= null_context;
763 printf("%s%s", line_prefix, c_frag);
764 for (i = 0; i <= num_parent; i++) putchar(combine_marker);
765 for (i = 0; i < num_parent; i++)
766 show_parent_lno(sline, lno, hunk_end, i, null_context);
767 printf(" +%lu,%lu ", lno+1, rlines);
768 for (i = 0; i <= num_parent; i++) putchar(combine_marker);
770 if (hunk_comment) {
771 int comment_end = 0;
772 for (i = 0; i < 40; i++) {
773 int ch = hunk_comment[i] & 0xff;
774 if (!ch || ch == '\n')
775 break;
776 if (!isspace(ch))
777 comment_end = i;
779 if (comment_end)
780 printf("%s%s %s%s", c_reset,
781 c_plain, c_reset,
782 c_func);
783 for (i = 0; i < comment_end; i++)
784 putchar(hunk_comment[i]);
787 printf("%s\n", c_reset);
788 while (lno < hunk_end) {
789 struct lline *ll;
790 int j;
791 unsigned long p_mask;
792 struct sline *sl = &sline[lno++];
793 ll = (sl->flag & no_pre_delete) ? NULL : sl->lost;
794 while (ll) {
795 printf("%s%s", line_prefix, c_old);
796 for (j = 0; j < num_parent; j++) {
797 if (ll->parent_map & (1UL<<j))
798 putchar('-');
799 else
800 putchar(' ');
802 show_line_to_eol(ll->line, -1, c_reset);
803 ll = ll->next;
805 if (cnt < lno)
806 break;
807 p_mask = 1;
808 fputs(line_prefix, stdout);
809 if (!(sl->flag & (mark-1))) {
811 * This sline was here to hang the
812 * lost lines in front of it.
814 if (!context)
815 continue;
816 fputs(c_plain, stdout);
818 else
819 fputs(c_new, stdout);
820 for (j = 0; j < num_parent; j++) {
821 if (p_mask & sl->flag)
822 putchar('+');
823 else
824 putchar(' ');
825 p_mask <<= 1;
827 show_line_to_eol(sl->bol, sl->len, c_reset);
832 static void reuse_combine_diff(struct sline *sline, unsigned long cnt,
833 int i, int j)
835 /* We have already examined parent j and we know parent i
836 * and parent j are the same, so reuse the combined result
837 * of parent j for parent i.
839 unsigned long lno, imask, jmask;
840 imask = (1UL<<i);
841 jmask = (1UL<<j);
843 for (lno = 0; lno <= cnt; lno++) {
844 struct lline *ll = sline->lost;
845 sline->p_lno[i] = sline->p_lno[j];
846 while (ll) {
847 if (ll->parent_map & jmask)
848 ll->parent_map |= imask;
849 ll = ll->next;
851 if (sline->flag & jmask)
852 sline->flag |= imask;
853 sline++;
855 /* the overall size of the file (sline[cnt]) */
856 sline->p_lno[i] = sline->p_lno[j];
859 static void dump_quoted_path(const char *head,
860 const char *prefix,
861 const char *path,
862 const char *line_prefix,
863 const char *c_meta, const char *c_reset)
865 static struct strbuf buf = STRBUF_INIT;
867 strbuf_reset(&buf);
868 strbuf_addstr(&buf, line_prefix);
869 strbuf_addstr(&buf, c_meta);
870 strbuf_addstr(&buf, head);
871 quote_two_c_style(&buf, prefix, path, 0);
872 strbuf_addstr(&buf, c_reset);
873 puts(buf.buf);
876 static void show_combined_header(struct combine_diff_path *elem,
877 int num_parent,
878 int dense,
879 struct rev_info *rev,
880 const char *line_prefix,
881 int mode_differs,
882 int show_file_header)
884 struct diff_options *opt = &rev->diffopt;
885 int abbrev = DIFF_OPT_TST(opt, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
886 const char *a_prefix = opt->a_prefix ? opt->a_prefix : "a/";
887 const char *b_prefix = opt->b_prefix ? opt->b_prefix : "b/";
888 const char *c_meta = diff_get_color_opt(opt, DIFF_METAINFO);
889 const char *c_reset = diff_get_color_opt(opt, DIFF_RESET);
890 const char *abb;
891 int added = 0;
892 int deleted = 0;
893 int i;
895 if (rev->loginfo && !rev->no_commit_id)
896 show_log(rev);
898 dump_quoted_path(dense ? "diff --cc " : "diff --combined ",
899 "", elem->path, line_prefix, c_meta, c_reset);
900 printf("%s%sindex ", line_prefix, c_meta);
901 for (i = 0; i < num_parent; i++) {
902 abb = find_unique_abbrev(elem->parent[i].sha1,
903 abbrev);
904 printf("%s%s", i ? "," : "", abb);
906 abb = find_unique_abbrev(elem->sha1, abbrev);
907 printf("..%s%s\n", abb, c_reset);
909 if (mode_differs) {
910 deleted = !elem->mode;
912 /* We say it was added if nobody had it */
913 added = !deleted;
914 for (i = 0; added && i < num_parent; i++)
915 if (elem->parent[i].status !=
916 DIFF_STATUS_ADDED)
917 added = 0;
918 if (added)
919 printf("%s%snew file mode %06o",
920 line_prefix, c_meta, elem->mode);
921 else {
922 if (deleted)
923 printf("%s%sdeleted file ",
924 line_prefix, c_meta);
925 printf("mode ");
926 for (i = 0; i < num_parent; i++) {
927 printf("%s%06o", i ? "," : "",
928 elem->parent[i].mode);
930 if (elem->mode)
931 printf("..%06o", elem->mode);
933 printf("%s\n", c_reset);
936 if (!show_file_header)
937 return;
939 if (added)
940 dump_quoted_path("--- ", "", "/dev/null",
941 line_prefix, c_meta, c_reset);
942 else
943 dump_quoted_path("--- ", a_prefix, elem->path,
944 line_prefix, c_meta, c_reset);
945 if (deleted)
946 dump_quoted_path("+++ ", "", "/dev/null",
947 line_prefix, c_meta, c_reset);
948 else
949 dump_quoted_path("+++ ", b_prefix, elem->path,
950 line_prefix, c_meta, c_reset);
953 static void show_patch_diff(struct combine_diff_path *elem, int num_parent,
954 int dense, int working_tree_file,
955 struct rev_info *rev)
957 struct diff_options *opt = &rev->diffopt;
958 unsigned long result_size, cnt, lno;
959 int result_deleted = 0;
960 char *result, *cp;
961 struct sline *sline; /* survived lines */
962 int mode_differs = 0;
963 int i, show_hunks;
964 mmfile_t result_file;
965 struct userdiff_driver *userdiff;
966 struct userdiff_driver *textconv = NULL;
967 int is_binary;
968 const char *line_prefix = diff_line_prefix(opt);
970 context = opt->context;
971 userdiff = userdiff_find_by_path(elem->path);
972 if (!userdiff)
973 userdiff = userdiff_find_by_name("default");
974 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV))
975 textconv = userdiff_get_textconv(userdiff);
977 /* Read the result of merge first */
978 if (!working_tree_file)
979 result = grab_blob(elem->sha1, elem->mode, &result_size,
980 textconv, elem->path);
981 else {
982 /* Used by diff-tree to read from the working tree */
983 struct stat st;
984 int fd = -1;
986 if (lstat(elem->path, &st) < 0)
987 goto deleted_file;
989 if (S_ISLNK(st.st_mode)) {
990 struct strbuf buf = STRBUF_INIT;
992 if (strbuf_readlink(&buf, elem->path, st.st_size) < 0) {
993 error("readlink(%s): %s", elem->path,
994 strerror(errno));
995 return;
997 result_size = buf.len;
998 result = strbuf_detach(&buf, NULL);
999 elem->mode = canon_mode(st.st_mode);
1000 } else if (S_ISDIR(st.st_mode)) {
1001 unsigned char sha1[20];
1002 if (resolve_gitlink_ref(elem->path, "HEAD", sha1) < 0)
1003 result = grab_blob(elem->sha1, elem->mode,
1004 &result_size, NULL, NULL);
1005 else
1006 result = grab_blob(sha1, elem->mode,
1007 &result_size, NULL, NULL);
1008 } else if (textconv) {
1009 struct diff_filespec *df = alloc_filespec(elem->path);
1010 fill_filespec(df, null_sha1, 0, st.st_mode);
1011 result_size = fill_textconv(textconv, df, &result);
1012 free_filespec(df);
1013 } else if (0 <= (fd = open(elem->path, O_RDONLY))) {
1014 size_t len = xsize_t(st.st_size);
1015 ssize_t done;
1016 int is_file, i;
1018 elem->mode = canon_mode(st.st_mode);
1019 /* if symlinks don't work, assume symlink if all parents
1020 * are symlinks
1022 is_file = has_symlinks;
1023 for (i = 0; !is_file && i < num_parent; i++)
1024 is_file = !S_ISLNK(elem->parent[i].mode);
1025 if (!is_file)
1026 elem->mode = canon_mode(S_IFLNK);
1028 result_size = len;
1029 result = xmalloc(len + 1);
1031 done = read_in_full(fd, result, len);
1032 if (done < 0)
1033 die_errno("read error '%s'", elem->path);
1034 else if (done < len)
1035 die("early EOF '%s'", elem->path);
1037 result[len] = 0;
1039 /* If not a fake symlink, apply filters, e.g. autocrlf */
1040 if (is_file) {
1041 struct strbuf buf = STRBUF_INIT;
1043 if (convert_to_git(elem->path, result, len, &buf, safe_crlf)) {
1044 free(result);
1045 result = strbuf_detach(&buf, &len);
1046 result_size = len;
1050 else {
1051 deleted_file:
1052 result_deleted = 1;
1053 result_size = 0;
1054 elem->mode = 0;
1055 result = xcalloc(1, 1);
1058 if (0 <= fd)
1059 close(fd);
1062 for (i = 0; i < num_parent; i++) {
1063 if (elem->parent[i].mode != elem->mode) {
1064 mode_differs = 1;
1065 break;
1069 if (textconv)
1070 is_binary = 0;
1071 else if (userdiff->binary != -1)
1072 is_binary = userdiff->binary;
1073 else {
1074 is_binary = buffer_is_binary(result, result_size);
1075 for (i = 0; !is_binary && i < num_parent; i++) {
1076 char *buf;
1077 unsigned long size;
1078 buf = grab_blob(elem->parent[i].sha1,
1079 elem->parent[i].mode,
1080 &size, NULL, NULL);
1081 if (buffer_is_binary(buf, size))
1082 is_binary = 1;
1083 free(buf);
1086 if (is_binary) {
1087 show_combined_header(elem, num_parent, dense, rev,
1088 line_prefix, mode_differs, 0);
1089 printf("Binary files differ\n");
1090 free(result);
1091 return;
1094 for (cnt = 0, cp = result; cp < result + result_size; cp++) {
1095 if (*cp == '\n')
1096 cnt++;
1098 if (result_size && result[result_size-1] != '\n')
1099 cnt++; /* incomplete line */
1101 sline = xcalloc(cnt+2, sizeof(*sline));
1102 sline[0].bol = result;
1103 for (lno = 0, cp = result; cp < result + result_size; cp++) {
1104 if (*cp == '\n') {
1105 sline[lno].len = cp - sline[lno].bol;
1106 lno++;
1107 if (lno < cnt)
1108 sline[lno].bol = cp + 1;
1111 if (result_size && result[result_size-1] != '\n')
1112 sline[cnt-1].len = result_size - (sline[cnt-1].bol - result);
1114 result_file.ptr = result;
1115 result_file.size = result_size;
1117 /* Even p_lno[cnt+1] is valid -- that is for the end line number
1118 * for deletion hunk at the end.
1120 sline[0].p_lno = xcalloc((cnt+2) * num_parent, sizeof(unsigned long));
1121 for (lno = 0; lno <= cnt; lno++)
1122 sline[lno+1].p_lno = sline[lno].p_lno + num_parent;
1124 for (i = 0; i < num_parent; i++) {
1125 int j;
1126 for (j = 0; j < i; j++) {
1127 if (!hashcmp(elem->parent[i].sha1,
1128 elem->parent[j].sha1)) {
1129 reuse_combine_diff(sline, cnt, i, j);
1130 break;
1133 if (i <= j)
1134 combine_diff(elem->parent[i].sha1,
1135 elem->parent[i].mode,
1136 &result_file, sline,
1137 cnt, i, num_parent, result_deleted,
1138 textconv, elem->path, opt->xdl_opts);
1141 show_hunks = make_hunks(sline, cnt, num_parent, dense);
1143 if (show_hunks || mode_differs || working_tree_file) {
1144 show_combined_header(elem, num_parent, dense, rev,
1145 line_prefix, mode_differs, 1);
1146 dump_sline(sline, line_prefix, cnt, num_parent,
1147 opt->use_color, result_deleted);
1149 free(result);
1151 for (lno = 0; lno < cnt; lno++) {
1152 if (sline[lno].lost) {
1153 struct lline *ll = sline[lno].lost;
1154 while (ll) {
1155 struct lline *tmp = ll;
1156 ll = ll->next;
1157 free(tmp);
1161 free(sline[0].p_lno);
1162 free(sline);
1165 static void show_raw_diff(struct combine_diff_path *p, int num_parent, struct rev_info *rev)
1167 struct diff_options *opt = &rev->diffopt;
1168 int line_termination, inter_name_termination, i;
1169 const char *line_prefix = diff_line_prefix(opt);
1171 line_termination = opt->line_termination;
1172 inter_name_termination = '\t';
1173 if (!line_termination)
1174 inter_name_termination = 0;
1176 if (rev->loginfo && !rev->no_commit_id)
1177 show_log(rev);
1180 if (opt->output_format & DIFF_FORMAT_RAW) {
1181 printf("%s", line_prefix);
1183 /* As many colons as there are parents */
1184 for (i = 0; i < num_parent; i++)
1185 putchar(':');
1187 /* Show the modes */
1188 for (i = 0; i < num_parent; i++)
1189 printf("%06o ", p->parent[i].mode);
1190 printf("%06o", p->mode);
1192 /* Show sha1's */
1193 for (i = 0; i < num_parent; i++)
1194 printf(" %s", diff_unique_abbrev(p->parent[i].sha1,
1195 opt->abbrev));
1196 printf(" %s ", diff_unique_abbrev(p->sha1, opt->abbrev));
1199 if (opt->output_format & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS)) {
1200 for (i = 0; i < num_parent; i++)
1201 putchar(p->parent[i].status);
1202 putchar(inter_name_termination);
1205 write_name_quoted(p->path, stdout, line_termination);
1209 * The result (p->elem) is from the working tree and their
1210 * parents are typically from multiple stages during a merge
1211 * (i.e. diff-files) or the state in HEAD and in the index
1212 * (i.e. diff-index).
1214 void show_combined_diff(struct combine_diff_path *p,
1215 int num_parent,
1216 int dense,
1217 struct rev_info *rev)
1219 struct diff_options *opt = &rev->diffopt;
1221 if (!p->len)
1222 return;
1223 if (opt->output_format & (DIFF_FORMAT_RAW |
1224 DIFF_FORMAT_NAME |
1225 DIFF_FORMAT_NAME_STATUS))
1226 show_raw_diff(p, num_parent, rev);
1227 else if (opt->output_format & DIFF_FORMAT_PATCH)
1228 show_patch_diff(p, num_parent, dense, 1, rev);
1231 static void free_combined_pair(struct diff_filepair *pair)
1233 free(pair->two);
1234 free(pair);
1238 * A combine_diff_path expresses N parents on the LHS against 1 merge
1239 * result. Synthesize a diff_filepair that has N entries on the "one"
1240 * side and 1 entry on the "two" side.
1242 * In the future, we might want to add more data to combine_diff_path
1243 * so that we can fill fields we are ignoring (most notably, size) here,
1244 * but currently nobody uses it, so this should suffice for now.
1246 static struct diff_filepair *combined_pair(struct combine_diff_path *p,
1247 int num_parent)
1249 int i;
1250 struct diff_filepair *pair;
1251 struct diff_filespec *pool;
1253 pair = xmalloc(sizeof(*pair));
1254 pool = xcalloc(num_parent + 1, sizeof(struct diff_filespec));
1255 pair->one = pool + 1;
1256 pair->two = pool;
1258 for (i = 0; i < num_parent; i++) {
1259 pair->one[i].path = p->path;
1260 pair->one[i].mode = p->parent[i].mode;
1261 hashcpy(pair->one[i].sha1, p->parent[i].sha1);
1262 pair->one[i].sha1_valid = !is_null_sha1(p->parent[i].sha1);
1263 pair->one[i].has_more_entries = 1;
1265 pair->one[num_parent - 1].has_more_entries = 0;
1267 pair->two->path = p->path;
1268 pair->two->mode = p->mode;
1269 hashcpy(pair->two->sha1, p->sha1);
1270 pair->two->sha1_valid = !is_null_sha1(p->sha1);
1271 return pair;
1274 static void handle_combined_callback(struct diff_options *opt,
1275 struct combine_diff_path *paths,
1276 int num_parent,
1277 int num_paths)
1279 struct combine_diff_path *p;
1280 struct diff_queue_struct q;
1281 int i;
1283 q.queue = xcalloc(num_paths, sizeof(struct diff_filepair *));
1284 q.alloc = num_paths;
1285 q.nr = num_paths;
1286 for (i = 0, p = paths; p; p = p->next) {
1287 if (!p->len)
1288 continue;
1289 q.queue[i++] = combined_pair(p, num_parent);
1291 opt->format_callback(&q, opt, opt->format_callback_data);
1292 for (i = 0; i < num_paths; i++)
1293 free_combined_pair(q.queue[i]);
1294 free(q.queue);
1297 void diff_tree_combined(const unsigned char *sha1,
1298 const struct sha1_array *parents,
1299 int dense,
1300 struct rev_info *rev)
1302 struct diff_options *opt = &rev->diffopt;
1303 struct diff_options diffopts;
1304 struct combine_diff_path *p, *paths = NULL;
1305 int i, num_paths, needsep, show_log_first, num_parent = parents->nr;
1307 diffopts = *opt;
1308 diff_tree_setup_paths(diffopts.pathspec.raw, &diffopts);
1309 diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
1310 DIFF_OPT_SET(&diffopts, RECURSIVE);
1311 DIFF_OPT_CLR(&diffopts, ALLOW_EXTERNAL);
1313 show_log_first = !!rev->loginfo && !rev->no_commit_id;
1314 needsep = 0;
1315 /* find set of paths that everybody touches */
1316 for (i = 0; i < num_parent; i++) {
1317 /* show stat against the first parent even
1318 * when doing combined diff.
1320 int stat_opt = (opt->output_format &
1321 (DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIFFSTAT));
1322 if (i == 0 && stat_opt)
1323 diffopts.output_format = stat_opt;
1324 else
1325 diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
1326 diff_tree_sha1(parents->sha1[i], sha1, "", &diffopts);
1327 diffcore_std(&diffopts);
1328 paths = intersect_paths(paths, i, num_parent);
1330 if (show_log_first && i == 0) {
1331 show_log(rev);
1333 if (rev->verbose_header && opt->output_format)
1334 printf("%s%c", diff_line_prefix(opt),
1335 opt->line_termination);
1337 diff_flush(&diffopts);
1340 /* find out surviving paths */
1341 for (num_paths = 0, p = paths; p; p = p->next) {
1342 if (p->len)
1343 num_paths++;
1345 if (num_paths) {
1346 if (opt->output_format & (DIFF_FORMAT_RAW |
1347 DIFF_FORMAT_NAME |
1348 DIFF_FORMAT_NAME_STATUS)) {
1349 for (p = paths; p; p = p->next) {
1350 if (p->len)
1351 show_raw_diff(p, num_parent, rev);
1353 needsep = 1;
1355 else if (opt->output_format &
1356 (DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIFFSTAT))
1357 needsep = 1;
1358 else if (opt->output_format & DIFF_FORMAT_CALLBACK)
1359 handle_combined_callback(opt, paths, num_parent, num_paths);
1361 if (opt->output_format & DIFF_FORMAT_PATCH) {
1362 if (needsep)
1363 printf("%s%c", diff_line_prefix(opt),
1364 opt->line_termination);
1365 for (p = paths; p; p = p->next) {
1366 if (p->len)
1367 show_patch_diff(p, num_parent, dense,
1368 0, rev);
1373 /* Clean things up */
1374 while (paths) {
1375 struct combine_diff_path *tmp = paths;
1376 paths = paths->next;
1377 free(tmp);
1380 diff_tree_release_paths(&diffopts);
1383 void diff_tree_combined_merge(const struct commit *commit, int dense,
1384 struct rev_info *rev)
1386 struct commit_list *parent = commit->parents;
1387 struct sha1_array parents = SHA1_ARRAY_INIT;
1389 while (parent) {
1390 sha1_array_append(&parents, parent->item->object.sha1);
1391 parent = parent->next;
1393 diff_tree_combined(commit->object.sha1, &parents, dense, rev);
1394 sha1_array_clear(&parents);