Merge branch 'jc/clean-exclude-doc'
[git] / combine-diff.c
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 "log-tree.h"
9 #include "refs.h"
10 #include "userdiff.h"
11
12 static struct combine_diff_path *intersect_paths(struct combine_diff_path *curr, int n, int num_parent)
13 {
14         struct diff_queue_struct *q = &diff_queued_diff;
15         struct combine_diff_path *p;
16         int i;
17
18         if (!n) {
19                 struct combine_diff_path *list = NULL, **tail = &list;
20                 for (i = 0; i < q->nr; i++) {
21                         int len;
22                         const char *path;
23                         if (diff_unmodified_pair(q->queue[i]))
24                                 continue;
25                         path = q->queue[i]->two->path;
26                         len = strlen(path);
27                         p = xmalloc(combine_diff_path_size(num_parent, len));
28                         p->path = (char *) &(p->parent[num_parent]);
29                         memcpy(p->path, path, len);
30                         p->path[len] = 0;
31                         p->len = len;
32                         p->next = NULL;
33                         memset(p->parent, 0,
34                                sizeof(p->parent[0]) * num_parent);
35
36                         hashcpy(p->sha1, q->queue[i]->two->sha1);
37                         p->mode = q->queue[i]->two->mode;
38                         hashcpy(p->parent[n].sha1, q->queue[i]->one->sha1);
39                         p->parent[n].mode = q->queue[i]->one->mode;
40                         p->parent[n].status = q->queue[i]->status;
41                         *tail = p;
42                         tail = &p->next;
43                 }
44                 return list;
45         }
46
47         for (p = curr; p; p = p->next) {
48                 int found = 0;
49                 if (!p->len)
50                         continue;
51                 for (i = 0; i < q->nr; i++) {
52                         const char *path;
53                         int len;
54
55                         if (diff_unmodified_pair(q->queue[i]))
56                                 continue;
57                         path = q->queue[i]->two->path;
58                         len = strlen(path);
59                         if (len == p->len && !memcmp(path, p->path, len)) {
60                                 found = 1;
61                                 hashcpy(p->parent[n].sha1, q->queue[i]->one->sha1);
62                                 p->parent[n].mode = q->queue[i]->one->mode;
63                                 p->parent[n].status = q->queue[i]->status;
64                                 break;
65                         }
66                 }
67                 if (!found)
68                         p->len = 0;
69         }
70         return curr;
71 }
72
73 /* Lines lost from parent */
74 struct lline {
75         struct lline *next;
76         int len;
77         unsigned long parent_map;
78         char line[FLEX_ARRAY];
79 };
80
81 /* Lines surviving in the merge result */
82 struct sline {
83         struct lline *lost_head, **lost_tail;
84         struct lline *next_lost;
85         char *bol;
86         int len;
87         /* bit 0 up to (N-1) are on if the parent has this line (i.e.
88          * we did not change it).
89          * bit N is used for "interesting" lines, including context.
90          * bit (N+1) is used for "do not show deletion before this".
91          */
92         unsigned long flag;
93         unsigned long *p_lno;
94 };
95
96 static char *grab_blob(const unsigned char *sha1, unsigned int mode,
97                        unsigned long *size, struct userdiff_driver *textconv,
98                        const char *path)
99 {
100         char *blob;
101         enum object_type type;
102
103         if (S_ISGITLINK(mode)) {
104                 blob = xmalloc(100);
105                 *size = snprintf(blob, 100,
106                                  "Subproject commit %s\n", sha1_to_hex(sha1));
107         } else if (is_null_sha1(sha1)) {
108                 /* deleted blob */
109                 *size = 0;
110                 return xcalloc(1, 1);
111         } else if (textconv) {
112                 struct diff_filespec *df = alloc_filespec(path);
113                 fill_filespec(df, sha1, mode);
114                 *size = fill_textconv(textconv, df, &blob);
115                 free_filespec(df);
116         } else {
117                 blob = read_sha1_file(sha1, &type, size);
118                 if (type != OBJ_BLOB)
119                         die("object '%s' is not a blob!", sha1_to_hex(sha1));
120         }
121         return blob;
122 }
123
124 static void append_lost(struct sline *sline, int n, const char *line, int len)
125 {
126         struct lline *lline;
127         unsigned long this_mask = (1UL<<n);
128         if (line[len-1] == '\n')
129                 len--;
130
131         /* Check to see if we can squash things */
132         if (sline->lost_head) {
133                 lline = sline->next_lost;
134                 while (lline) {
135                         if (lline->len == len &&
136                             !memcmp(lline->line, line, len)) {
137                                 lline->parent_map |= this_mask;
138                                 sline->next_lost = lline->next;
139                                 return;
140                         }
141                         lline = lline->next;
142                 }
143         }
144
145         lline = xmalloc(sizeof(*lline) + len + 1);
146         lline->len = len;
147         lline->next = NULL;
148         lline->parent_map = this_mask;
149         memcpy(lline->line, line, len);
150         lline->line[len] = 0;
151         *sline->lost_tail = lline;
152         sline->lost_tail = &lline->next;
153         sline->next_lost = NULL;
154 }
155
156 struct combine_diff_state {
157         unsigned int lno;
158         int ob, on, nb, nn;
159         unsigned long nmask;
160         int num_parent;
161         int n;
162         struct sline *sline;
163         struct sline *lost_bucket;
164 };
165
166 static void consume_line(void *state_, char *line, unsigned long len)
167 {
168         struct combine_diff_state *state = state_;
169         if (5 < len && !memcmp("@@ -", line, 4)) {
170                 if (parse_hunk_header(line, len,
171                                       &state->ob, &state->on,
172                                       &state->nb, &state->nn))
173                         return;
174                 state->lno = state->nb;
175                 if (state->nn == 0) {
176                         /* @@ -X,Y +N,0 @@ removed Y lines
177                          * that would have come *after* line N
178                          * in the result.  Our lost buckets hang
179                          * to the line after the removed lines,
180                          *
181                          * Note that this is correct even when N == 0,
182                          * in which case the hunk removes the first
183                          * line in the file.
184                          */
185                         state->lost_bucket = &state->sline[state->nb];
186                         if (!state->nb)
187                                 state->nb = 1;
188                 } else {
189                         state->lost_bucket = &state->sline[state->nb-1];
190                 }
191                 if (!state->sline[state->nb-1].p_lno)
192                         state->sline[state->nb-1].p_lno =
193                                 xcalloc(state->num_parent,
194                                         sizeof(unsigned long));
195                 state->sline[state->nb-1].p_lno[state->n] = state->ob;
196                 state->lost_bucket->next_lost = state->lost_bucket->lost_head;
197                 return;
198         }
199         if (!state->lost_bucket)
200                 return; /* not in any hunk yet */
201         switch (line[0]) {
202         case '-':
203                 append_lost(state->lost_bucket, state->n, line+1, len-1);
204                 break;
205         case '+':
206                 state->sline[state->lno-1].flag |= state->nmask;
207                 state->lno++;
208                 break;
209         }
210 }
211
212 static void combine_diff(const unsigned char *parent, unsigned int mode,
213                          mmfile_t *result_file,
214                          struct sline *sline, unsigned int cnt, int n,
215                          int num_parent, int result_deleted,
216                          struct userdiff_driver *textconv,
217                          const char *path)
218 {
219         unsigned int p_lno, lno;
220         unsigned long nmask = (1UL << n);
221         xpparam_t xpp;
222         xdemitconf_t xecfg;
223         mmfile_t parent_file;
224         struct combine_diff_state state;
225         unsigned long sz;
226
227         if (result_deleted)
228                 return; /* result deleted */
229
230         parent_file.ptr = grab_blob(parent, mode, &sz, textconv, path);
231         parent_file.size = sz;
232         memset(&xpp, 0, sizeof(xpp));
233         xpp.flags = 0;
234         memset(&xecfg, 0, sizeof(xecfg));
235         memset(&state, 0, sizeof(state));
236         state.nmask = nmask;
237         state.sline = sline;
238         state.lno = 1;
239         state.num_parent = num_parent;
240         state.n = n;
241
242         xdi_diff_outf(&parent_file, result_file, consume_line, &state,
243                       &xpp, &xecfg);
244         free(parent_file.ptr);
245
246         /* Assign line numbers for this parent.
247          *
248          * sline[lno].p_lno[n] records the first line number
249          * (counting from 1) for parent N if the final hunk display
250          * started by showing sline[lno] (possibly showing the lost
251          * lines attached to it first).
252          */
253         for (lno = 0,  p_lno = 1; lno <= cnt; lno++) {
254                 struct lline *ll;
255                 sline[lno].p_lno[n] = p_lno;
256
257                 /* How many lines would this sline advance the p_lno? */
258                 ll = sline[lno].lost_head;
259                 while (ll) {
260                         if (ll->parent_map & nmask)
261                                 p_lno++; /* '-' means parent had it */
262                         ll = ll->next;
263                 }
264                 if (lno < cnt && !(sline[lno].flag & nmask))
265                         p_lno++; /* no '+' means parent had it */
266         }
267         sline[lno].p_lno[n] = p_lno; /* trailer */
268 }
269
270 static unsigned long context = 3;
271 static char combine_marker = '@';
272
273 static int interesting(struct sline *sline, unsigned long all_mask)
274 {
275         /* If some parents lost lines here, or if we have added to
276          * some parent, it is interesting.
277          */
278         return ((sline->flag & all_mask) || sline->lost_head);
279 }
280
281 static unsigned long adjust_hunk_tail(struct sline *sline,
282                                       unsigned long all_mask,
283                                       unsigned long hunk_begin,
284                                       unsigned long i)
285 {
286         /* i points at the first uninteresting line.  If the last line
287          * of the hunk was interesting only because it has some
288          * deletion, then it is not all that interesting for the
289          * purpose of giving trailing context lines.  This is because
290          * we output '-' line and then unmodified sline[i-1] itself in
291          * that case which gives us one extra context line.
292          */
293         if ((hunk_begin + 1 <= i) && !(sline[i-1].flag & all_mask))
294                 i--;
295         return i;
296 }
297
298 static unsigned long find_next(struct sline *sline,
299                                unsigned long mark,
300                                unsigned long i,
301                                unsigned long cnt,
302                                int look_for_uninteresting)
303 {
304         /* We have examined up to i-1 and are about to look at i.
305          * Find next interesting or uninteresting line.  Here,
306          * "interesting" does not mean interesting(), but marked by
307          * the give_context() function below (i.e. it includes context
308          * lines that are not interesting to interesting() function
309          * that are surrounded by interesting() ones.
310          */
311         while (i <= cnt)
312                 if (look_for_uninteresting
313                     ? !(sline[i].flag & mark)
314                     : (sline[i].flag & mark))
315                         return i;
316                 else
317                         i++;
318         return i;
319 }
320
321 static int give_context(struct sline *sline, unsigned long cnt, int num_parent)
322 {
323         unsigned long all_mask = (1UL<<num_parent) - 1;
324         unsigned long mark = (1UL<<num_parent);
325         unsigned long no_pre_delete = (2UL<<num_parent);
326         unsigned long i;
327
328         /* Two groups of interesting lines may have a short gap of
329          * uninteresting lines.  Connect such groups to give them a
330          * bit of context.
331          *
332          * We first start from what the interesting() function says,
333          * and mark them with "mark", and paint context lines with the
334          * mark.  So interesting() would still say false for such context
335          * lines but they are treated as "interesting" in the end.
336          */
337         i = find_next(sline, mark, 0, cnt, 0);
338         if (cnt < i)
339                 return 0;
340
341         while (i <= cnt) {
342                 unsigned long j = (context < i) ? (i - context) : 0;
343                 unsigned long k;
344
345                 /* Paint a few lines before the first interesting line. */
346                 while (j < i)
347                         sline[j++].flag |= mark | no_pre_delete;
348
349         again:
350                 /* we know up to i is to be included.  where does the
351                  * next uninteresting one start?
352                  */
353                 j = find_next(sline, mark, i, cnt, 1);
354                 if (cnt < j)
355                         break; /* the rest are all interesting */
356
357                 /* lookahead context lines */
358                 k = find_next(sline, mark, j, cnt, 0);
359                 j = adjust_hunk_tail(sline, all_mask, i, j);
360
361                 if (k < j + context) {
362                         /* k is interesting and [j,k) are not, but
363                          * paint them interesting because the gap is small.
364                          */
365                         while (j < k)
366                                 sline[j++].flag |= mark;
367                         i = k;
368                         goto again;
369                 }
370
371                 /* j is the first uninteresting line and there is
372                  * no overlap beyond it within context lines.  Paint
373                  * the trailing edge a bit.
374                  */
375                 i = k;
376                 k = (j + context < cnt+1) ? j + context : cnt+1;
377                 while (j < k)
378                         sline[j++].flag |= mark;
379         }
380         return 1;
381 }
382
383 static int make_hunks(struct sline *sline, unsigned long cnt,
384                        int num_parent, int dense)
385 {
386         unsigned long all_mask = (1UL<<num_parent) - 1;
387         unsigned long mark = (1UL<<num_parent);
388         unsigned long i;
389         int has_interesting = 0;
390
391         for (i = 0; i <= cnt; i++) {
392                 if (interesting(&sline[i], all_mask))
393                         sline[i].flag |= mark;
394                 else
395                         sline[i].flag &= ~mark;
396         }
397         if (!dense)
398                 return give_context(sline, cnt, num_parent);
399
400         /* Look at each hunk, and if we have changes from only one
401          * parent, or the changes are the same from all but one
402          * parent, mark that uninteresting.
403          */
404         i = 0;
405         while (i <= cnt) {
406                 unsigned long j, hunk_begin, hunk_end;
407                 unsigned long same_diff;
408                 while (i <= cnt && !(sline[i].flag & mark))
409                         i++;
410                 if (cnt < i)
411                         break; /* No more interesting hunks */
412                 hunk_begin = i;
413                 for (j = i + 1; j <= cnt; j++) {
414                         if (!(sline[j].flag & mark)) {
415                                 /* Look beyond the end to see if there
416                                  * is an interesting line after this
417                                  * hunk within context span.
418                                  */
419                                 unsigned long la; /* lookahead */
420                                 int contin = 0;
421                                 la = adjust_hunk_tail(sline, all_mask,
422                                                      hunk_begin, j);
423                                 la = (la + context < cnt + 1) ?
424                                         (la + context) : cnt + 1;
425                                 while (j <= --la) {
426                                         if (sline[la].flag & mark) {
427                                                 contin = 1;
428                                                 break;
429                                         }
430                                 }
431                                 if (!contin)
432                                         break;
433                                 j = la;
434                         }
435                 }
436                 hunk_end = j;
437
438                 /* [i..hunk_end) are interesting.  Now is it really
439                  * interesting?  We check if there are only two versions
440                  * and the result matches one of them.  That is, we look
441                  * at:
442                  *   (+) line, which records lines added to which parents;
443                  *       this line appears in the result.
444                  *   (-) line, which records from what parents the line
445                  *       was removed; this line does not appear in the result.
446                  * then check the set of parents the result has difference
447                  * from, from all lines.  If there are lines that has
448                  * different set of parents that the result has differences
449                  * from, that means we have more than two versions.
450                  *
451                  * Even when we have only two versions, if the result does
452                  * not match any of the parents, the it should be considered
453                  * interesting.  In such a case, we would have all '+' line.
454                  * After passing the above "two versions" test, that would
455                  * appear as "the same set of parents" to be "all parents".
456                  */
457                 same_diff = 0;
458                 has_interesting = 0;
459                 for (j = i; j < hunk_end && !has_interesting; j++) {
460                         unsigned long this_diff = sline[j].flag & all_mask;
461                         struct lline *ll = sline[j].lost_head;
462                         if (this_diff) {
463                                 /* This has some changes.  Is it the
464                                  * same as others?
465                                  */
466                                 if (!same_diff)
467                                         same_diff = this_diff;
468                                 else if (same_diff != this_diff) {
469                                         has_interesting = 1;
470                                         break;
471                                 }
472                         }
473                         while (ll && !has_interesting) {
474                                 /* Lost this line from these parents;
475                                  * who are they?  Are they the same?
476                                  */
477                                 this_diff = ll->parent_map;
478                                 if (!same_diff)
479                                         same_diff = this_diff;
480                                 else if (same_diff != this_diff) {
481                                         has_interesting = 1;
482                                 }
483                                 ll = ll->next;
484                         }
485                 }
486
487                 if (!has_interesting && same_diff != all_mask) {
488                         /* This hunk is not that interesting after all */
489                         for (j = hunk_begin; j < hunk_end; j++)
490                                 sline[j].flag &= ~mark;
491                 }
492                 i = hunk_end;
493         }
494
495         has_interesting = give_context(sline, cnt, num_parent);
496         return has_interesting;
497 }
498
499 static void show_parent_lno(struct sline *sline, unsigned long l0, unsigned long l1, int n, unsigned long null_context)
500 {
501         l0 = sline[l0].p_lno[n];
502         l1 = sline[l1].p_lno[n];
503         printf(" -%lu,%lu", l0, l1-l0-null_context);
504 }
505
506 static int hunk_comment_line(const char *bol)
507 {
508         int ch;
509
510         if (!bol)
511                 return 0;
512         ch = *bol & 0xff;
513         return (isalpha(ch) || ch == '_' || ch == '$');
514 }
515
516 static void show_line_to_eol(const char *line, int len, const char *reset)
517 {
518         int saw_cr_at_eol = 0;
519         if (len < 0)
520                 len = strlen(line);
521         saw_cr_at_eol = (len && line[len-1] == '\r');
522
523         printf("%.*s%s%s\n", len - saw_cr_at_eol, line,
524                reset,
525                saw_cr_at_eol ? "\r" : "");
526 }
527
528 static void dump_sline(struct sline *sline, unsigned long cnt, int num_parent,
529                        int use_color, int result_deleted)
530 {
531         unsigned long mark = (1UL<<num_parent);
532         unsigned long no_pre_delete = (2UL<<num_parent);
533         int i;
534         unsigned long lno = 0;
535         const char *c_frag = diff_get_color(use_color, DIFF_FRAGINFO);
536         const char *c_func = diff_get_color(use_color, DIFF_FUNCINFO);
537         const char *c_new = diff_get_color(use_color, DIFF_FILE_NEW);
538         const char *c_old = diff_get_color(use_color, DIFF_FILE_OLD);
539         const char *c_plain = diff_get_color(use_color, DIFF_PLAIN);
540         const char *c_reset = diff_get_color(use_color, DIFF_RESET);
541
542         if (result_deleted)
543                 return; /* result deleted */
544
545         while (1) {
546                 unsigned long hunk_end;
547                 unsigned long rlines;
548                 const char *hunk_comment = NULL;
549                 unsigned long null_context = 0;
550
551                 while (lno <= cnt && !(sline[lno].flag & mark)) {
552                         if (hunk_comment_line(sline[lno].bol))
553                                 hunk_comment = sline[lno].bol;
554                         lno++;
555                 }
556                 if (cnt < lno)
557                         break;
558                 else {
559                         for (hunk_end = lno + 1; hunk_end <= cnt; hunk_end++)
560                                 if (!(sline[hunk_end].flag & mark))
561                                         break;
562                 }
563                 rlines = hunk_end - lno;
564                 if (cnt < hunk_end)
565                         rlines--; /* pointing at the last delete hunk */
566
567                 if (!context) {
568                         /*
569                          * Even when running with --unified=0, all
570                          * lines in the hunk needs to be processed in
571                          * the loop below in order to show the
572                          * deletion recorded in lost_head.  However,
573                          * we do not want to show the resulting line
574                          * with all blank context markers in such a
575                          * case.  Compensate.
576                          */
577                         unsigned long j;
578                         for (j = lno; j < hunk_end; j++)
579                                 if (!(sline[j].flag & (mark-1)))
580                                         null_context++;
581                         rlines -= null_context;
582                 }
583
584                 fputs(c_frag, stdout);
585                 for (i = 0; i <= num_parent; i++) putchar(combine_marker);
586                 for (i = 0; i < num_parent; i++)
587                         show_parent_lno(sline, lno, hunk_end, i, null_context);
588                 printf(" +%lu,%lu ", lno+1, rlines);
589                 for (i = 0; i <= num_parent; i++) putchar(combine_marker);
590
591                 if (hunk_comment) {
592                         int comment_end = 0;
593                         for (i = 0; i < 40; i++) {
594                                 int ch = hunk_comment[i] & 0xff;
595                                 if (!ch || ch == '\n')
596                                         break;
597                                 if (!isspace(ch))
598                                     comment_end = i;
599                         }
600                         if (comment_end)
601                                 printf("%s%s %s%s", c_reset,
602                                                     c_plain, c_reset,
603                                                     c_func);
604                         for (i = 0; i < comment_end; i++)
605                                 putchar(hunk_comment[i]);
606                 }
607
608                 printf("%s\n", c_reset);
609                 while (lno < hunk_end) {
610                         struct lline *ll;
611                         int j;
612                         unsigned long p_mask;
613                         struct sline *sl = &sline[lno++];
614                         ll = (sl->flag & no_pre_delete) ? NULL : sl->lost_head;
615                         while (ll) {
616                                 fputs(c_old, stdout);
617                                 for (j = 0; j < num_parent; j++) {
618                                         if (ll->parent_map & (1UL<<j))
619                                                 putchar('-');
620                                         else
621                                                 putchar(' ');
622                                 }
623                                 show_line_to_eol(ll->line, -1, c_reset);
624                                 ll = ll->next;
625                         }
626                         if (cnt < lno)
627                                 break;
628                         p_mask = 1;
629                         if (!(sl->flag & (mark-1))) {
630                                 /*
631                                  * This sline was here to hang the
632                                  * lost lines in front of it.
633                                  */
634                                 if (!context)
635                                         continue;
636                                 fputs(c_plain, stdout);
637                         }
638                         else
639                                 fputs(c_new, stdout);
640                         for (j = 0; j < num_parent; j++) {
641                                 if (p_mask & sl->flag)
642                                         putchar('+');
643                                 else
644                                         putchar(' ');
645                                 p_mask <<= 1;
646                         }
647                         show_line_to_eol(sl->bol, sl->len, c_reset);
648                 }
649         }
650 }
651
652 static void reuse_combine_diff(struct sline *sline, unsigned long cnt,
653                                int i, int j)
654 {
655         /* We have already examined parent j and we know parent i
656          * and parent j are the same, so reuse the combined result
657          * of parent j for parent i.
658          */
659         unsigned long lno, imask, jmask;
660         imask = (1UL<<i);
661         jmask = (1UL<<j);
662
663         for (lno = 0; lno <= cnt; lno++) {
664                 struct lline *ll = sline->lost_head;
665                 sline->p_lno[i] = sline->p_lno[j];
666                 while (ll) {
667                         if (ll->parent_map & jmask)
668                                 ll->parent_map |= imask;
669                         ll = ll->next;
670                 }
671                 if (sline->flag & jmask)
672                         sline->flag |= imask;
673                 sline++;
674         }
675         /* the overall size of the file (sline[cnt]) */
676         sline->p_lno[i] = sline->p_lno[j];
677 }
678
679 static void dump_quoted_path(const char *head,
680                              const char *prefix,
681                              const char *path,
682                              const char *c_meta, const char *c_reset)
683 {
684         static struct strbuf buf = STRBUF_INIT;
685
686         strbuf_reset(&buf);
687         strbuf_addstr(&buf, c_meta);
688         strbuf_addstr(&buf, head);
689         quote_two_c_style(&buf, prefix, path, 0);
690         strbuf_addstr(&buf, c_reset);
691         puts(buf.buf);
692 }
693
694 static void show_combined_header(struct combine_diff_path *elem,
695                                  int num_parent,
696                                  int dense,
697                                  struct rev_info *rev,
698                                  int mode_differs,
699                                  int show_file_header)
700 {
701         struct diff_options *opt = &rev->diffopt;
702         int abbrev = DIFF_OPT_TST(opt, FULL_INDEX) ? 40 : DEFAULT_ABBREV;
703         const char *a_prefix = opt->a_prefix ? opt->a_prefix : "a/";
704         const char *b_prefix = opt->b_prefix ? opt->b_prefix : "b/";
705         const char *c_meta = diff_get_color_opt(opt, DIFF_METAINFO);
706         const char *c_reset = diff_get_color_opt(opt, DIFF_RESET);
707         const char *abb;
708         int added = 0;
709         int deleted = 0;
710         int i;
711
712         if (rev->loginfo && !rev->no_commit_id)
713                 show_log(rev);
714
715         dump_quoted_path(dense ? "diff --cc " : "diff --combined ",
716                          "", elem->path, c_meta, c_reset);
717         printf("%sindex ", c_meta);
718         for (i = 0; i < num_parent; i++) {
719                 abb = find_unique_abbrev(elem->parent[i].sha1,
720                                          abbrev);
721                 printf("%s%s", i ? "," : "", abb);
722         }
723         abb = find_unique_abbrev(elem->sha1, abbrev);
724         printf("..%s%s\n", abb, c_reset);
725
726         if (mode_differs) {
727                 deleted = !elem->mode;
728
729                 /* We say it was added if nobody had it */
730                 added = !deleted;
731                 for (i = 0; added && i < num_parent; i++)
732                         if (elem->parent[i].status !=
733                             DIFF_STATUS_ADDED)
734                                 added = 0;
735                 if (added)
736                         printf("%snew file mode %06o",
737                                c_meta, elem->mode);
738                 else {
739                         if (deleted)
740                                 printf("%sdeleted file ", c_meta);
741                         printf("mode ");
742                         for (i = 0; i < num_parent; i++) {
743                                 printf("%s%06o", i ? "," : "",
744                                        elem->parent[i].mode);
745                         }
746                         if (elem->mode)
747                                 printf("..%06o", elem->mode);
748                 }
749                 printf("%s\n", c_reset);
750         }
751
752         if (!show_file_header)
753                 return;
754
755         if (added)
756                 dump_quoted_path("--- ", "", "/dev/null",
757                                  c_meta, c_reset);
758         else
759                 dump_quoted_path("--- ", a_prefix, elem->path,
760                                  c_meta, c_reset);
761         if (deleted)
762                 dump_quoted_path("+++ ", "", "/dev/null",
763                                  c_meta, c_reset);
764         else
765                 dump_quoted_path("+++ ", b_prefix, elem->path,
766                                  c_meta, c_reset);
767 }
768
769 static void show_patch_diff(struct combine_diff_path *elem, int num_parent,
770                             int dense, int working_tree_file,
771                             struct rev_info *rev)
772 {
773         struct diff_options *opt = &rev->diffopt;
774         unsigned long result_size, cnt, lno;
775         int result_deleted = 0;
776         char *result, *cp;
777         struct sline *sline; /* survived lines */
778         int mode_differs = 0;
779         int i, show_hunks;
780         mmfile_t result_file;
781         struct userdiff_driver *userdiff;
782         struct userdiff_driver *textconv = NULL;
783         int is_binary;
784
785         context = opt->context;
786         userdiff = userdiff_find_by_path(elem->path);
787         if (!userdiff)
788                 userdiff = userdiff_find_by_name("default");
789         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV))
790                 textconv = userdiff_get_textconv(userdiff);
791
792         /* Read the result of merge first */
793         if (!working_tree_file)
794                 result = grab_blob(elem->sha1, elem->mode, &result_size,
795                                    textconv, elem->path);
796         else {
797                 /* Used by diff-tree to read from the working tree */
798                 struct stat st;
799                 int fd = -1;
800
801                 if (lstat(elem->path, &st) < 0)
802                         goto deleted_file;
803
804                 if (S_ISLNK(st.st_mode)) {
805                         struct strbuf buf = STRBUF_INIT;
806
807                         if (strbuf_readlink(&buf, elem->path, st.st_size) < 0) {
808                                 error("readlink(%s): %s", elem->path,
809                                       strerror(errno));
810                                 return;
811                         }
812                         result_size = buf.len;
813                         result = strbuf_detach(&buf, NULL);
814                         elem->mode = canon_mode(st.st_mode);
815                 } else if (S_ISDIR(st.st_mode)) {
816                         unsigned char sha1[20];
817                         if (resolve_gitlink_ref(elem->path, "HEAD", sha1) < 0)
818                                 result = grab_blob(elem->sha1, elem->mode,
819                                                    &result_size, NULL, NULL);
820                         else
821                                 result = grab_blob(sha1, elem->mode,
822                                                    &result_size, NULL, NULL);
823                 } else if (textconv) {
824                         struct diff_filespec *df = alloc_filespec(elem->path);
825                         fill_filespec(df, null_sha1, st.st_mode);
826                         result_size = fill_textconv(textconv, df, &result);
827                         free_filespec(df);
828                 } else if (0 <= (fd = open(elem->path, O_RDONLY))) {
829                         size_t len = xsize_t(st.st_size);
830                         ssize_t done;
831                         int is_file, i;
832
833                         elem->mode = canon_mode(st.st_mode);
834                         /* if symlinks don't work, assume symlink if all parents
835                          * are symlinks
836                          */
837                         is_file = has_symlinks;
838                         for (i = 0; !is_file && i < num_parent; i++)
839                                 is_file = !S_ISLNK(elem->parent[i].mode);
840                         if (!is_file)
841                                 elem->mode = canon_mode(S_IFLNK);
842
843                         result_size = len;
844                         result = xmalloc(len + 1);
845
846                         done = read_in_full(fd, result, len);
847                         if (done < 0)
848                                 die_errno("read error '%s'", elem->path);
849                         else if (done < len)
850                                 die("early EOF '%s'", elem->path);
851
852                         result[len] = 0;
853
854                         /* If not a fake symlink, apply filters, e.g. autocrlf */
855                         if (is_file) {
856                                 struct strbuf buf = STRBUF_INIT;
857
858                                 if (convert_to_git(elem->path, result, len, &buf, safe_crlf)) {
859                                         free(result);
860                                         result = strbuf_detach(&buf, &len);
861                                         result_size = len;
862                                 }
863                         }
864                 }
865                 else {
866                 deleted_file:
867                         result_deleted = 1;
868                         result_size = 0;
869                         elem->mode = 0;
870                         result = xcalloc(1, 1);
871                 }
872
873                 if (0 <= fd)
874                         close(fd);
875         }
876
877         for (i = 0; i < num_parent; i++) {
878                 if (elem->parent[i].mode != elem->mode) {
879                         mode_differs = 1;
880                         break;
881                 }
882         }
883
884         if (textconv)
885                 is_binary = 0;
886         else if (userdiff->binary != -1)
887                 is_binary = userdiff->binary;
888         else {
889                 is_binary = buffer_is_binary(result, result_size);
890                 for (i = 0; !is_binary && i < num_parent; i++) {
891                         char *buf;
892                         unsigned long size;
893                         buf = grab_blob(elem->parent[i].sha1,
894                                         elem->parent[i].mode,
895                                         &size, NULL, NULL);
896                         if (buffer_is_binary(buf, size))
897                                 is_binary = 1;
898                         free(buf);
899                 }
900         }
901         if (is_binary) {
902                 show_combined_header(elem, num_parent, dense, rev,
903                                      mode_differs, 0);
904                 printf("Binary files differ\n");
905                 free(result);
906                 return;
907         }
908
909         for (cnt = 0, cp = result; cp < result + result_size; cp++) {
910                 if (*cp == '\n')
911                         cnt++;
912         }
913         if (result_size && result[result_size-1] != '\n')
914                 cnt++; /* incomplete line */
915
916         sline = xcalloc(cnt+2, sizeof(*sline));
917         sline[0].bol = result;
918         for (lno = 0; lno <= cnt + 1; lno++) {
919                 sline[lno].lost_tail = &sline[lno].lost_head;
920                 sline[lno].flag = 0;
921         }
922         for (lno = 0, cp = result; cp < result + result_size; cp++) {
923                 if (*cp == '\n') {
924                         sline[lno].len = cp - sline[lno].bol;
925                         lno++;
926                         if (lno < cnt)
927                                 sline[lno].bol = cp + 1;
928                 }
929         }
930         if (result_size && result[result_size-1] != '\n')
931                 sline[cnt-1].len = result_size - (sline[cnt-1].bol - result);
932
933         result_file.ptr = result;
934         result_file.size = result_size;
935
936         /* Even p_lno[cnt+1] is valid -- that is for the end line number
937          * for deletion hunk at the end.
938          */
939         sline[0].p_lno = xcalloc((cnt+2) * num_parent, sizeof(unsigned long));
940         for (lno = 0; lno <= cnt; lno++)
941                 sline[lno+1].p_lno = sline[lno].p_lno + num_parent;
942
943         for (i = 0; i < num_parent; i++) {
944                 int j;
945                 for (j = 0; j < i; j++) {
946                         if (!hashcmp(elem->parent[i].sha1,
947                                      elem->parent[j].sha1)) {
948                                 reuse_combine_diff(sline, cnt, i, j);
949                                 break;
950                         }
951                 }
952                 if (i <= j)
953                         combine_diff(elem->parent[i].sha1,
954                                      elem->parent[i].mode,
955                                      &result_file, sline,
956                                      cnt, i, num_parent, result_deleted,
957                                      textconv, elem->path);
958         }
959
960         show_hunks = make_hunks(sline, cnt, num_parent, dense);
961
962         if (show_hunks || mode_differs || working_tree_file) {
963                 show_combined_header(elem, num_parent, dense, rev,
964                                      mode_differs, 1);
965                 dump_sline(sline, cnt, num_parent,
966                            opt->use_color, result_deleted);
967         }
968         free(result);
969
970         for (lno = 0; lno < cnt; lno++) {
971                 if (sline[lno].lost_head) {
972                         struct lline *ll = sline[lno].lost_head;
973                         while (ll) {
974                                 struct lline *tmp = ll;
975                                 ll = ll->next;
976                                 free(tmp);
977                         }
978                 }
979         }
980         free(sline[0].p_lno);
981         free(sline);
982 }
983
984 #define COLONS "::::::::::::::::::::::::::::::::"
985
986 static void show_raw_diff(struct combine_diff_path *p, int num_parent, struct rev_info *rev)
987 {
988         struct diff_options *opt = &rev->diffopt;
989         int i, offset;
990         const char *prefix;
991         int line_termination, inter_name_termination;
992
993         line_termination = opt->line_termination;
994         inter_name_termination = '\t';
995         if (!line_termination)
996                 inter_name_termination = 0;
997
998         if (rev->loginfo && !rev->no_commit_id)
999                 show_log(rev);
1000
1001         if (opt->output_format & DIFF_FORMAT_RAW) {
1002                 offset = strlen(COLONS) - num_parent;
1003                 if (offset < 0)
1004                         offset = 0;
1005                 prefix = COLONS + offset;
1006
1007                 /* Show the modes */
1008                 for (i = 0; i < num_parent; i++) {
1009                         printf("%s%06o", prefix, p->parent[i].mode);
1010                         prefix = " ";
1011                 }
1012                 printf("%s%06o", prefix, p->mode);
1013
1014                 /* Show sha1's */
1015                 for (i = 0; i < num_parent; i++)
1016                         printf(" %s", diff_unique_abbrev(p->parent[i].sha1,
1017                                                          opt->abbrev));
1018                 printf(" %s ", diff_unique_abbrev(p->sha1, opt->abbrev));
1019         }
1020
1021         if (opt->output_format & (DIFF_FORMAT_RAW | DIFF_FORMAT_NAME_STATUS)) {
1022                 for (i = 0; i < num_parent; i++)
1023                         putchar(p->parent[i].status);
1024                 putchar(inter_name_termination);
1025         }
1026
1027         write_name_quoted(p->path, stdout, line_termination);
1028 }
1029
1030 /*
1031  * The result (p->elem) is from the working tree and their
1032  * parents are typically from multiple stages during a merge
1033  * (i.e. diff-files) or the state in HEAD and in the index
1034  * (i.e. diff-index).
1035  */
1036 void show_combined_diff(struct combine_diff_path *p,
1037                        int num_parent,
1038                        int dense,
1039                        struct rev_info *rev)
1040 {
1041         struct diff_options *opt = &rev->diffopt;
1042         if (!p->len)
1043                 return;
1044         if (opt->output_format & (DIFF_FORMAT_RAW |
1045                                   DIFF_FORMAT_NAME |
1046                                   DIFF_FORMAT_NAME_STATUS))
1047                 show_raw_diff(p, num_parent, rev);
1048         else if (opt->output_format & DIFF_FORMAT_PATCH)
1049                 show_patch_diff(p, num_parent, dense, 1, rev);
1050 }
1051
1052 static void free_combined_pair(struct diff_filepair *pair)
1053 {
1054         free(pair->two);
1055         free(pair);
1056 }
1057
1058 /*
1059  * A combine_diff_path expresses N parents on the LHS against 1 merge
1060  * result. Synthesize a diff_filepair that has N entries on the "one"
1061  * side and 1 entry on the "two" side.
1062  *
1063  * In the future, we might want to add more data to combine_diff_path
1064  * so that we can fill fields we are ignoring (most notably, size) here,
1065  * but currently nobody uses it, so this should suffice for now.
1066  */
1067 static struct diff_filepair *combined_pair(struct combine_diff_path *p,
1068                                            int num_parent)
1069 {
1070         int i;
1071         struct diff_filepair *pair;
1072         struct diff_filespec *pool;
1073
1074         pair = xmalloc(sizeof(*pair));
1075         pool = xcalloc(num_parent + 1, sizeof(struct diff_filespec));
1076         pair->one = pool + 1;
1077         pair->two = pool;
1078
1079         for (i = 0; i < num_parent; i++) {
1080                 pair->one[i].path = p->path;
1081                 pair->one[i].mode = p->parent[i].mode;
1082                 hashcpy(pair->one[i].sha1, p->parent[i].sha1);
1083                 pair->one[i].sha1_valid = !is_null_sha1(p->parent[i].sha1);
1084                 pair->one[i].has_more_entries = 1;
1085         }
1086         pair->one[num_parent - 1].has_more_entries = 0;
1087
1088         pair->two->path = p->path;
1089         pair->two->mode = p->mode;
1090         hashcpy(pair->two->sha1, p->sha1);
1091         pair->two->sha1_valid = !is_null_sha1(p->sha1);
1092         return pair;
1093 }
1094
1095 static void handle_combined_callback(struct diff_options *opt,
1096                                      struct combine_diff_path *paths,
1097                                      int num_parent,
1098                                      int num_paths)
1099 {
1100         struct combine_diff_path *p;
1101         struct diff_queue_struct q;
1102         int i;
1103
1104         q.queue = xcalloc(num_paths, sizeof(struct diff_filepair *));
1105         q.alloc = num_paths;
1106         q.nr = num_paths;
1107         for (i = 0, p = paths; p; p = p->next) {
1108                 if (!p->len)
1109                         continue;
1110                 q.queue[i++] = combined_pair(p, num_parent);
1111         }
1112         opt->format_callback(&q, opt, opt->format_callback_data);
1113         for (i = 0; i < num_paths; i++)
1114                 free_combined_pair(q.queue[i]);
1115         free(q.queue);
1116 }
1117
1118 void diff_tree_combined(const unsigned char *sha1,
1119                         const unsigned char parent[][20],
1120                         int num_parent,
1121                         int dense,
1122                         struct rev_info *rev)
1123 {
1124         struct diff_options *opt = &rev->diffopt;
1125         struct diff_options diffopts;
1126         struct combine_diff_path *p, *paths = NULL;
1127         int i, num_paths, needsep, show_log_first;
1128
1129         diffopts = *opt;
1130         diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
1131         DIFF_OPT_SET(&diffopts, RECURSIVE);
1132         DIFF_OPT_CLR(&diffopts, ALLOW_EXTERNAL);
1133
1134         show_log_first = !!rev->loginfo && !rev->no_commit_id;
1135         needsep = 0;
1136         /* find set of paths that everybody touches */
1137         for (i = 0; i < num_parent; i++) {
1138                 /* show stat against the first parent even
1139                  * when doing combined diff.
1140                  */
1141                 int stat_opt = (opt->output_format &
1142                                 (DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIFFSTAT));
1143                 if (i == 0 && stat_opt)
1144                         diffopts.output_format = stat_opt;
1145                 else
1146                         diffopts.output_format = DIFF_FORMAT_NO_OUTPUT;
1147                 diff_tree_sha1(parent[i], sha1, "", &diffopts);
1148                 diffcore_std(&diffopts);
1149                 paths = intersect_paths(paths, i, num_parent);
1150
1151                 if (show_log_first && i == 0) {
1152                         show_log(rev);
1153                         if (rev->verbose_header && opt->output_format)
1154                                 putchar(opt->line_termination);
1155                 }
1156                 diff_flush(&diffopts);
1157         }
1158
1159         /* find out surviving paths */
1160         for (num_paths = 0, p = paths; p; p = p->next) {
1161                 if (p->len)
1162                         num_paths++;
1163         }
1164         if (num_paths) {
1165                 if (opt->output_format & (DIFF_FORMAT_RAW |
1166                                           DIFF_FORMAT_NAME |
1167                                           DIFF_FORMAT_NAME_STATUS)) {
1168                         for (p = paths; p; p = p->next) {
1169                                 if (p->len)
1170                                         show_raw_diff(p, num_parent, rev);
1171                         }
1172                         needsep = 1;
1173                 }
1174                 else if (opt->output_format &
1175                          (DIFF_FORMAT_NUMSTAT|DIFF_FORMAT_DIFFSTAT))
1176                         needsep = 1;
1177                 else if (opt->output_format & DIFF_FORMAT_CALLBACK)
1178                         handle_combined_callback(opt, paths, num_parent, num_paths);
1179
1180                 if (opt->output_format & DIFF_FORMAT_PATCH) {
1181                         if (needsep)
1182                                 putchar(opt->line_termination);
1183                         for (p = paths; p; p = p->next) {
1184                                 if (p->len)
1185                                         show_patch_diff(p, num_parent, dense,
1186                                                         0, rev);
1187                         }
1188                 }
1189         }
1190
1191         /* Clean things up */
1192         while (paths) {
1193                 struct combine_diff_path *tmp = paths;
1194                 paths = paths->next;
1195                 free(tmp);
1196         }
1197 }
1198
1199 void diff_tree_combined_merge(const unsigned char *sha1,
1200                              int dense, struct rev_info *rev)
1201 {
1202         int num_parent;
1203         const unsigned char (*parent)[20];
1204         struct commit *commit = lookup_commit(sha1);
1205         struct commit_list *parents;
1206
1207         /* count parents */
1208         for (parents = commit->parents, num_parent = 0;
1209              parents;
1210              parents = parents->next, num_parent++)
1211                 ; /* nothing */
1212
1213         parent = xmalloc(num_parent * sizeof(*parent));
1214         for (parents = commit->parents, num_parent = 0;
1215              parents;
1216              parents = parents->next, num_parent++)
1217                 hashcpy((unsigned char *)(parent + num_parent),
1218                         parents->item->object.sha1);
1219         diff_tree_combined(sha1, parent, num_parent, dense, rev);
1220 }