2 * wiggle - apply rejected patches
4 * Copyright (C) 2005 Neil Brown <neilb@cse.unsw.edu.au>
5 * Copyright (C) 2010-2013 Neil Brown <neilb@suse.de>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program.
22 * Email: <neilb@suse.de>
29 * Second attempt at merging....
31 * We want to create a mergelist which identifies 'orig' and 'after'
32 * sections (from a and c) and conflicts (which are ranges of a,b,c which
34 * It is also helpful to differentiate 'orig' sections that aren't
35 * matched in 'b' with orig sections that are.
36 * To help with highlighting, it will be useful to know where
37 * the conflicts match the csl lists.
39 * This can all be achieved with a list of (a,b,c,c1,c1) 5-tuples.
40 * If two consecutive differ in more than one of a,b,c, it is a
42 * If only 'a' differ, it is un-matched original.
43 * If only 'b' differ, it is matched, unchanged original
44 * If only 'c' differ, it is 1
47 static inline int min(int a, int b)
52 static int check_alreadyapplied(struct file af, struct file cf,
58 for (i = 0; i < m->al; i++) {
59 if (af.list[m->a+i].len != cf.list[m->c+i].len)
61 if (strncmp(af.list[m->a+i].start,
62 cf.list[m->c+i].start,
63 af.list[m->a+i].len) != 0)
67 printf("already applied %d,%d,%d - %d,%d,%d\n",
68 m->a, m->b, m->c, m->al, m->bl, m->cl);
69 printf(" %.10s - %.10s\n", af.list[m->a].start,
72 m->type = AlreadyApplied;
76 /* A 'cut-point' is a location in the merger where it is reasonable
77 * the change the mode of display - between displaying the merger
78 * and displaying the separate streams.
79 * A 'conflict' can only be displayed as separate stream so when
80 * one is found, we need to find a preceding and trailing cut-point
81 * and enlarge the conflict to that range.
82 * A suitable location is one where all three streams are at a line-end.
84 static int is_cutpoint(struct merge m,
85 struct file af, struct file bf, struct file cf)
87 return ((m.a == 0 || ends_line(af.list[m.a-1])) &&
88 (m.b == 0 || ends_line(bf.list[m.b-1])) &&
89 (m.c == 0 || ends_line(cf.list[m.c-1])));
92 int isolate_conflicts(struct file af, struct file bf, struct file cf,
93 struct csl *csl1, struct csl *csl2, int words,
94 struct merge *m, int show_wiggles,
97 /* A Conflict indicates that something is definitely wrong
98 * and so we need to be a bit suspicious of nearby apparent matches.
99 * To display a conflict effectively we expands its effect to
100 * include any Extraneous, Unmatched, Changed or AlreadyApplied text.
101 * Also, unless 'words', we need to include any partial lines
102 * in the Unchanged text that forms the border of a conflict.
104 * A Changed text may also border a conflict, but it can
105 * only border one conflict (where as an Unchanged can border
106 * a preceding and a following conflict).
107 * The 'new' section of a Changed text appears in the
108 * conflict as does any part of the original before
111 * A hunk header (Extraneous) is never considered part of a
112 * conflict. It thereby can serve as a separator between
115 * Extended conflicts are marked by setting ->in_conflict in
116 * the "struct merge". This is '1' for an Unchanged, Changed,
117 * or (Extraneous) hunk header which borders the conflict,
118 * '2' for a merger which is truly in conflict, and '3' for
119 * a merger which is causing a 'wiggle'.
120 * When in_conflict == 1, the 'lo' and 'hi' fields indicate
121 * how much of the 'a' file is included in the conflict, the rest
122 * being part of the clean result.
123 * Elements in af from m->a to m->a+m->lo are in the preceding
124 * conflict, from m->a+m->lo to m->a+m->hi are clean, and
125 * m->a+m->hi to m->a+m->al are in the following conflict.
127 * We need to ensure there is adequate context for the conflict.
128 * So ensure there are at least 3 newlines in Extraneous or
129 * Unchanged on both sides of a Conflict - but don't go so far
130 * as including a hunk header.
131 * If there are 3, and they are all in 'Unchanged' sections, then
132 * that much context is not really needed - reduce it a bit.
134 * If a wiggle is adjacent to a conflict then:
135 * - if show_wiggles is set, we just merge them
136 * - if it is not set, we still want to count the wiggle.
139 int cnt = 0, wiggles = 0;
145 for (i = 0; m[i].type != End; i++)
146 m[i].in_conflict = 0;
148 for (i = 0; m[i].type != End; i++) {
149 /* The '3' here is a count of newlines. Once we find
150 * that many newlines of the particular type, we have escaped.
152 if (m[i].type == Changed)
154 if (m[i].type == Unmatched)
156 if (m[i].type == Extraneous && bf.list[m[i].b].start[0])
157 /* hunk headers don't imply wiggles, other
158 * extraneous text does.
162 if (m[i].type != Unchanged && changed && (unmatched || extraneous)) {
169 if ((m[i].type == Conflict) ||
170 (show_wiggles && in_wiggle)) {
171 /* We have a conflict or wiggle here.
172 * First search backwards for an Unchanged marking
173 * things as in_conflict. Then find the
174 * cut-point in the Unchanged. If there isn't one,
177 * Then search forward doing the same thing.
180 m[i].in_conflict = m[i].type == Conflict ? 2 : 3;
185 if (m[j].type == Extraneous &&
186 bf.list[m[j].b].start[0] == '\0')
187 /* hunk header - not conflict any more */
189 if (m[j].in_conflict > 1)
190 /* Merge the conflicts */
192 if (!m[j].in_conflict) {
193 m[j].in_conflict = 1;
196 /* Following must set m[j].hi, or set
199 if (m[j].type == Extraneous) {
200 for (k = m[j].bl; k > 0; k--)
201 if (ends_line(bf.list[m[j].b+k-1]))
205 if (m[j].type != Unchanged &&
206 m[j].type != Changed) {
207 if (m[j].type == Conflict)
208 m[j].in_conflict = 2;
210 m[j].in_conflict = m[i].in_conflict;
213 /* If we find enough newlines in this section,
214 * then we only really need 1, but would rather
215 * it wasn't the first one. 'firstk' allows us
216 * to track which newline we actually use
223 /* need to find the last line-break, which
224 * might be after the last newline, if there
225 * is one, or might be at the start
227 for (k = m[j].al; k > 0; k--) {
228 if (m[j].a + k >= af.elcnt)
229 /* FIXME impossible!*/
231 if (ends_line(af.list[m[j].a+k-1])) {
232 if (firstk > m[j].al)
245 else if (is_cutpoint(m[j], af,bf,cf))
248 /* no start-of-line found... */
251 (m[j].type == Changed)) {
252 /* this can only work if start is
253 * also a line break */
254 if (is_cutpoint(m[j], af,bf,cf))
261 m[j].in_conflict = m[i].in_conflict;
264 /* now the forward search */
266 for (j = i+1; m[j].type != End; j++) {
267 if (m[j].type == Extraneous) {
268 for (k = 0; k < m[j].bl; k++)
269 if (ends_line(bf.list[m[j].b+k]))
272 if (m[j].type != Unchanged &&
273 m[j].type != Changed) {
274 if (m[j].type == Conflict)
275 m[j].in_conflict = 2;
277 m[j].in_conflict = m[i].in_conflict;
280 m[j].in_conflict = 1;
286 /* need to find a line-break, which might be at
287 * the very beginning, or might be after the
288 * first newline - if there is one
290 if (is_cutpoint(m[j], af,bf,cf))
293 /* If we find enough newlines in this section,
294 * then we only really need 1, but would rather
295 * it wasn't the first one. 'firstk' allows us
296 * to track which newline we actually use
299 for (k = 0 ; k < m[j].al ; k++)
300 if (ends_line(af.list[m[j].a+k])) {
311 /* Hit end of file, pretend we found 3 newlines. */
315 m[j+1].type == Unmatched) {
316 /* If this Unmatched exceeds 3 lines, just stop here */
319 for (p = 0; p < m[j+1].al ; p++)
320 if (ends_line(af.list[m[j+1].a+p])) {
331 /* no start-of-line found */
334 if (m[j].lo <= m[j].al+1 &&
335 (m[j].type == Changed)) {
336 /* this can only work if the end is a line break */
337 if (is_cutpoint(m[j+1], af,bf,cf))
342 if (m[j].lo < m[j].al+1)
344 m[j].in_conflict = m[i].in_conflict;
346 if (m[j-1].in_conflict == 1)
349 /* A hunk header bordered the conflict */
352 /* If any of the merges are Changed or Conflict,
353 * then this really is a Conflict or Wiggle.
354 * If not they are just Unchanged, Unmatched,
355 * Extraneous or AlreadyApplied, and so don't
357 * Note that the first/last merges (in_conflict==1)
358 * can be Changed and so much be check separately.
360 if (m[j].type == Changed)
362 for (j = i-1; j >= 0 && m[j].in_conflict > 1; j--)
363 if (m[j].type == Changed || m[j].type == Conflict)
365 if (j >= 0 && m[j].type == Changed)
367 /* False alarm, no real conflict/wiggle here as
368 * nothing changed. */
371 if (m[j].in_conflict == 1) {
374 m[j].in_conflict = 0;
378 m[j++].in_conflict = 0;
380 if (m[i].type == End)
383 for (k = 1; k < m[i].al; k++) {
384 if (m[i].a + k >= af.elcnt)
385 /* FIXME this should be impossible, but
389 if (words || ends_line(af.list[m[i].a+k])) {
401 /* Now count the conflicts and wiggles */
402 for (i = 0; m[i].type != End; i++) {
403 int true_conflict = 0;
404 if (!m[i].in_conflict)
407 for (j = i; m[j].type != End && m[j].in_conflict; j++) {
408 if (m[j].in_conflict == 2)
411 m[j].in_conflict == 1) {
413 if (!m[j+1].in_conflict)
429 struct ci make_merger(struct file af, struct file bf, struct file cf,
430 struct csl *csl1, struct csl *csl2, int words,
431 int ignore_already, int show_wiggles)
433 /* find the wiggles and conflicts between csl1 and csl2
438 int header_checked = -1;
439 int header_found = 0;
441 rv.conflicts = rv.wiggles = rv.ignored = 0;
443 for (i = 0; csl1[i].len; i++)
446 for (i = 0; csl2[i].len; i++)
449 /* maybe a bit of slack at each end */
452 rv.merger = xmalloc(sizeof(struct merge)*l);
454 a = b = c = c1 = c2 = 0;
458 match1 = (a >= csl1[c1].a && b >= csl1[c1].b); /* c1 doesn't match */
459 match2 = (b >= csl2[c2].a && c >= csl2[c2].b);
461 if (header_checked != c2) {
462 /* Check if there is a hunk header in this range */
465 for (j = b; j < csl2[c2].a + csl2[c2].len; j++)
466 if (bf.list[j].start[0] == '\0') {
475 rv.merger[i].c1 = c1;
476 rv.merger[i].c2 = c2;
477 rv.merger[i].in_conflict = 0;
479 if (!match1 && match2) {
480 /* This is either Unmatched or Extraneous - probably both.
481 * If the match2 has a hunk-header Extraneous, it must
482 * align with an end-of-line in 'a', so adjust endpoint
484 int newa = csl1[c1].a;
485 if (header_found >= 0) {
487 !ends_line(af.list[newa-1]))
490 if (a == newa && b == csl1[c1].b)
493 /* some unmatched text */
494 rv.merger[i].type = Unmatched;
495 rv.merger[i].al = newa - a;
500 assert(b < csl1[c1].b);
501 /* some Extraneous text */
502 /* length is min of unmatched on left
503 * and matched on right.
504 * However a hunk-header must be an
505 * Extraneous section by itself, so if this
506 * start with one, the length is 1, and if
507 * there is one in the middle, only take the
508 * text up to there for now.
510 rv.merger[i].type = Extraneous;
514 csl2[c2].len - (b-csl2[c2].a));
515 if (header_found == b) {
518 } else if (header_found > b && header_found < newb) {
524 rv.merger[i].bl = newb - b;
526 } else if (match1 && !match2) {
528 * if 'c' is currently at a suitable cut-point, then
529 * we can look for a triple-cut-point for start.
530 * Also, if csl2[c2].b isn't in a conflict, and is
531 * a suitable cut-point, then we could make a
532 * triple-cut-point for end of a conflict.
535 rv.merger[i].type = Changed;
536 rv.merger[i].bl = min(csl1[c1].b+csl1[c1].len, csl2[c2].a) - b;
537 rv.merger[i].al = rv.merger[i].bl;
538 rv.merger[i].cl = csl2[c2].b - c;
539 } else if (match1 && match2) {
540 /* Some unchanged text
542 rv.merger[i].type = Unchanged;
544 min(csl1[c1].len - (b-csl1[c1].b),
545 csl2[c2].len - (b-csl2[c2].a));
546 rv.merger[i].al = rv.merger[i].cl =
549 /* must be a conflict.
550 * Move a and c to next match, and b to closest of the two
552 rv.merger[i].type = Conflict;
553 rv.merger[i].al = csl1[c1].a - a;
554 rv.merger[i].cl = csl2[c2].b - c;
555 rv.merger[i].bl = min(csl1[c1].b, csl2[c2].a) - b;
556 if (ignore_already &&
557 check_alreadyapplied(af, cf, &rv.merger[i]))
559 else if (rv.merger[i].bl == 0 &&
561 /* As the 'before' stream is empty, this
562 * could look like Unmatched in the
563 * original, and an insertion in the
564 * diff. Reporting it like that is
565 * probably more useful that as a full
567 * Leave the type for the insertion as
568 * Conflict (not Changed) as there is some
569 * real uncertainty here, but allow the
570 * original to become Unmatched.
574 rv.merger[i].oldtype = rv.merger[i].type;
575 a += rv.merger[i].al;
576 b += rv.merger[i].bl;
577 c += rv.merger[i].cl;
580 while (csl1[c1].a + csl1[c1].len <= a && csl1[c1].len)
582 assert(csl1[c1].b + csl1[c1].len >= b);
583 while (csl2[c2].b + csl2[c2].len <= c && csl2[c2].len)
585 assert(csl2[c2].a + csl2[c2].len >= b);
586 if (csl1[c1].len == 0 && csl2[c2].len == 0 &&
587 a == csl1[c1].a && b == csl1[c1].b &&
588 b == csl2[c2].a && c == csl2[c2].b)
591 rv.merger[i].type = End;
592 rv.merger[i].oldtype = End;
596 rv.merger[i].c1 = c1;
597 rv.merger[i].c2 = c2;
598 rv.merger[i].in_conflict = 0;
601 /* Now revert any AlreadyApplied that aren't bounded by
602 * Unchanged or Changed.
604 for (i = 0; rv.merger[i].type != End; i++) {
605 if (rv.merger[i].type != AlreadyApplied)
607 if (i > 0 && rv.merger[i-1].type != Unchanged &&
608 rv.merger[i-1].type != Changed)
609 rv.merger[i].type = Conflict;
610 if (rv.merger[i+1].type != Unchanged &&
611 rv.merger[i+1].type != Changed &&
612 rv.merger[i+1].type != End)
613 rv.merger[i].type = Conflict;
615 rv.conflicts = isolate_conflicts(af, bf, cf, csl1, csl2, words,
616 rv.merger, show_wiggles, &rv.wiggles);
620 static int printrange(FILE *out, struct file *f, int start, int len,
624 while (len > 0 && start < f->elcnt) {
625 struct elmnt e = f->list[start];
627 if (e.start[e.plen-1] == '\n' &&
637 static const char *conflict_types[] = {
638 "", " border"," conflict"," wiggle" };
640 int print_merge(FILE *out, struct file *a, struct file *b, struct file *c,
641 int words, struct merge *merger,
642 struct merge *mpos, int streampos, int offsetpos)
647 int offset = INT_MAX;
650 for (m = merger; m->type != End ; m++) {
653 printf("[%s: %d-%d,%d-%d,%d-%d%s(%d,%d)]\n",
654 m->type==Unmatched ? "Unmatched" :
655 m->type==Unchanged ? "Unchanged" :
656 m->type==Extraneous ? "Extraneous" :
657 m->type==Changed ? "Changed" :
658 m->type==AlreadyApplied ? "AlreadyApplied" :
659 m->type==Conflict ? "Conflict":"unknown",
663 conflict_types[m->in_conflict],
666 while (m->in_conflict) {
667 /* need to print from 'hi' to 'lo' of next
668 * Unchanged which is < it's hi
670 int found_conflict = 0;
672 if (m->in_conflict == 1)
678 if (m->in_conflict == 1 && m->type == Unchanged)
679 lineno += printrange(out, a, m->a+m->lo, m->hi - m->lo, offset - m->lo);
684 for (cm = m; cm->in_conflict; cm++) {
685 printf("{%s: %d-%d,%d-%d,%d-%d%s(%d,%d)}%s\n",
686 cm->type==Unmatched?"Unmatched":
687 cm->type==Unchanged?"Unchanged":
688 cm->type==Extraneous?"Extraneous":
689 cm->type==Changed?"Changed":
690 cm->type==AlreadyApplied?"AlreadyApplied":
691 cm->type==Conflict?"Conflict":"unknown",
692 cm->a, cm->a+cm->al-1,
693 cm->b, cm->b+cm->bl-1,
694 cm->c, cm->c+cm->cl-1,
695 conflict_types[m->in_conflict],
697 (cm->type == Extraneous &&
698 b->list[cm->b].start[0] == '\0') ?
699 b->list[cm->b].start+1: ""
701 if (cm->in_conflict == 1 && cm != m)
705 if (m->in_conflict == 1 &&
706 m[1].in_conflict == 1) {
707 /* Nothing between two conflicts */
712 fputs(words ? "<<<---" : "<<<<<<< found\n", out);
715 for (cm = m; cm->in_conflict; cm++) {
716 if (cm == mpos && streampos == 0)
718 if (cm->type == Conflict)
720 if (cm->in_conflict == 1 && cm != m) {
721 lineno += printrange(out, a, cm->a, cm->lo, offset);
724 lineno += printrange(out, a, cm->a+st1, cm->al-st1, offset-st1);
726 if (cm == mpos && streampos == 0)
729 if (cm == mpos && streampos == 0)
732 fputs(words ? "|||" : "||||||| expected\n", out);
737 for (cm = m; cm->in_conflict; cm++) {
738 if (cm->type == Extraneous &&
739 b->list[cm->b].start[0] == '\0') {
740 /* This is a hunk header, skip it and possibly
747 if (cm->type != Unchanged && cm->type != Unmatched)
749 if (cm == mpos && streampos == 1)
751 if (cm->in_conflict == 1 && cm != m) {
752 lineno += printrange(out, a, cm->a, cm->lo, offset);
755 lineno += printrange(out, b, cm->b+st1, cm->bl-st1, offset-st1);
757 if (cm == mpos && streampos == 1)
760 if (cm == mpos && streampos == 1)
762 fputs(words ? "===" : "=======\n", out);
767 for (cm = m; cm->in_conflict; cm++) {
768 if (cm->type == Extraneous &&
769 b->list[cm->b].start[0] == '\0') {
770 /* This is a hunk header, skip it and possibly
771 * abort this section and restart.
776 /* If remaining merges are all
777 * Extraneous, Unchanged, or Unmatched,
778 * we don't need them.
780 while (cm->in_conflict > 1 &&
781 (cm->type == Extraneous ||
782 cm->type == Unmatched ||
783 cm->type == Unchanged))
785 if (!cm->in_conflict)
786 /* Nothing more to report */
788 if (cm->in_conflict == 1 &&
789 (cm->type == Extraneous ||
790 cm->type == Unmatched ||
791 cm->type == Unchanged))
792 /* border between conflicts, but
793 * still nothing to report.
796 fputs(words ? ">>>" : ">>>>>>> replacement\n", out);
797 fputs(words ? "<<<" : "<<<<<<< found\n", out);
801 if (cm->type != Unchanged && cm->type != Unmatched)
803 if (cm == mpos && streampos == 2)
805 if (cm->in_conflict == 1 && cm != m) {
806 if (cm->type == Unchanged)
807 lineno += printrange(out, a, cm->a, cm->lo, offset);
809 lineno += printrange(out, c, cm->c, cm->cl, offset);
812 if (cm->type == Changed)
813 st1 = 0; /* All of result of change must be printed */
814 lineno += printrange(out, c, cm->c+st1, cm->cl-st1, offset-st1);
816 if (cm == mpos && streampos == 2)
819 if (cm == mpos && streampos == 2)
821 if (!found_conflict) {
822 /* This section was wiggled in successfully,
823 * but full conflict display was requested.
824 * So now print out the wiggled result as well.
826 fputs(words ? "&&&" : "&&&&&&& resolution\n", out);
830 for (cm = m; cm->in_conflict; cm++) {
832 if (cm->in_conflict == 1 && cm != m)
838 lineno += printrange(out, a, cm->a+st1,
839 last ? cm->lo : cm->al-st1, offset-st1);
844 lineno += printrange(out, c, cm->c,
845 last ? cm->lo : cm->cl, offset);
856 fputs(words ? "--->>>" : ">>>>>>> replacement\n", out);
860 if (m->in_conflict == 1 && m[1].in_conflict == 0) {
861 /* End of a conflict, no conflict follows */
864 if (m->type == Unchanged)
865 lineno += printrange(out, a, m->a+m->lo, m->hi-m->lo, offset-m->lo);
872 /* there is always some non-conflict after a conflict,
873 * unless we hit the end
879 printf("<<%s: %d-%d,%d-%d,%d-%d%s(%d,%d)>>\n",
880 m->type==Unmatched?"Unmatched":
881 m->type==Unchanged?"Unchanged":
882 m->type==Extraneous?"Extraneous":
883 m->type==Changed?"Changed":
884 m->type==AlreadyApplied?"AlreadyApplied":
885 m->type==Conflict?"Conflict":"unknown",
889 conflict_types[m->in_conflict],
899 lineno += printrange(out, a, m->a, m->al, offset);
904 lineno += printrange(out, c, m->c, m->cl, offset);
916 int save_merge(struct file a, struct file b, struct file c,
917 struct merge *merger, char *file, int backup)
919 char *replacename = xmalloc(strlen(file) + 20);
920 char *orignew = xmalloc(strlen(file) + 20);
925 strcpy(replacename, file);
926 strcat(replacename, "XXXXXX");
927 strcpy(orignew, file);
928 strcat(orignew, ".porig");
930 fd = mkstemp(replacename);
935 outfile = fdopen(fd, "w");
936 lineno = print_merge(outfile, &a, &b, &c, 0, merger,
939 if (backup && rename(file, orignew) != 0)
941 else if (rename(replacename, file) != 0)
947 return err < 0 ? err : lineno;
950 int save_tmp_merge(struct file a, struct file b, struct file c,
951 struct merge *merger, char **filep,
952 struct merge *mpos, int streampos, int offsetpos)
961 dir = getenv("TMPDIR");
965 asprintf(&fname, "%s/wiggle-tmp-XXXXXX", dir);
969 base = strrchr(dir, '/');
974 asprintf(&fname, "%.*stmp-XXXXXX-%s", (int)(base-dir), dir, base);
975 suffix = strlen(base)+1;
977 fd = mkstemps(fname, suffix);
984 outfile = fdopen(fd, "w");
985 lineno = print_merge(outfile, &a, &b, &c, 0, merger,
986 mpos, streampos, offsetpos);