range_set: fix coalescing bug when range is a subset of another
[git] / line-log.c
1 #include "git-compat-util.h"
2 #include "line-range.h"
3 #include "cache.h"
4 #include "tag.h"
5 #include "blob.h"
6 #include "tree.h"
7 #include "diff.h"
8 #include "commit.h"
9 #include "decorate.h"
10 #include "revision.h"
11 #include "xdiff-interface.h"
12 #include "strbuf.h"
13 #include "log-tree.h"
14 #include "graph.h"
15 #include "userdiff.h"
16 #include "line-log.h"
17
18 static void range_set_grow(struct range_set *rs, size_t extra)
19 {
20         ALLOC_GROW(rs->ranges, rs->nr + extra, rs->alloc);
21 }
22
23 /* Either initialization would be fine */
24 #define RANGE_SET_INIT {0}
25
26 static void range_set_init(struct range_set *rs, size_t prealloc)
27 {
28         rs->alloc = rs->nr = 0;
29         rs->ranges = NULL;
30         if (prealloc)
31                 range_set_grow(rs, prealloc);
32 }
33
34 static void range_set_release(struct range_set *rs)
35 {
36         free(rs->ranges);
37         rs->alloc = rs->nr = 0;
38         rs->ranges = NULL;
39 }
40
41 /* dst must be uninitialized! */
42 static void range_set_copy(struct range_set *dst, struct range_set *src)
43 {
44         range_set_init(dst, src->nr);
45         memcpy(dst->ranges, src->ranges, src->nr*sizeof(struct range_set));
46         dst->nr = src->nr;
47 }
48 static void range_set_move(struct range_set *dst, struct range_set *src)
49 {
50         range_set_release(dst);
51         dst->ranges = src->ranges;
52         dst->nr = src->nr;
53         dst->alloc = src->alloc;
54         src->ranges = NULL;
55         src->alloc = src->nr = 0;
56 }
57
58 /* tack on a _new_ range _at the end_ */
59 static void range_set_append_unsafe(struct range_set *rs, long a, long b)
60 {
61         assert(a <= b);
62         range_set_grow(rs, 1);
63         rs->ranges[rs->nr].start = a;
64         rs->ranges[rs->nr].end = b;
65         rs->nr++;
66 }
67
68 static void range_set_append(struct range_set *rs, long a, long b)
69 {
70         assert(rs->nr == 0 || rs->ranges[rs->nr-1].end <= a);
71         range_set_append_unsafe(rs, a, b);
72 }
73
74 static int range_cmp(const void *_r, const void *_s)
75 {
76         const struct range *r = _r;
77         const struct range *s = _s;
78
79         /* this could be simply 'return r.start-s.start', but for the types */
80         if (r->start == s->start)
81                 return 0;
82         if (r->start < s->start)
83                 return -1;
84         return 1;
85 }
86
87 /*
88  * Check that the ranges are non-empty, sorted and non-overlapping
89  */
90 static void range_set_check_invariants(struct range_set *rs)
91 {
92         int i;
93
94         if (!rs)
95                 return;
96
97         if (rs->nr)
98                 assert(rs->ranges[0].start < rs->ranges[0].end);
99
100         for (i = 1; i < rs->nr; i++) {
101                 assert(rs->ranges[i-1].end < rs->ranges[i].start);
102                 assert(rs->ranges[i].start < rs->ranges[i].end);
103         }
104 }
105
106 /*
107  * In-place pass of sorting and merging the ranges in the range set,
108  * to establish the invariants when we get the ranges from the user
109  */
110 static void sort_and_merge_range_set(struct range_set *rs)
111 {
112         int i;
113         int o = 1; /* output cursor */
114
115         qsort(rs->ranges, rs->nr, sizeof(struct range), range_cmp);
116
117         for (i = 1; i < rs->nr; i++) {
118                 if (rs->ranges[i].start <= rs->ranges[o-1].end) {
119                         if (rs->ranges[o-1].end < rs->ranges[i].end)
120                                 rs->ranges[o-1].end = rs->ranges[i].end;
121                 } else {
122                         rs->ranges[o].start = rs->ranges[i].start;
123                         rs->ranges[o].end = rs->ranges[i].end;
124                         o++;
125                 }
126         }
127         assert(o <= rs->nr);
128         rs->nr = o;
129
130         range_set_check_invariants(rs);
131 }
132
133 /*
134  * Union of range sets (i.e., sets of line numbers).  Used to merge
135  * them when searches meet at a common ancestor.
136  *
137  * This is also where the ranges are consolidated into canonical form:
138  * overlapping and adjacent ranges are merged, and empty ranges are
139  * removed.
140  */
141 static void range_set_union(struct range_set *out,
142                              struct range_set *a, struct range_set *b)
143 {
144         int i = 0, j = 0, o = 0;
145         struct range *ra = a->ranges;
146         struct range *rb = b->ranges;
147         /* cannot make an alias of out->ranges: it may change during grow */
148
149         assert(out->nr == 0);
150         while (i < a->nr || j < b->nr) {
151                 struct range *new;
152                 if (i < a->nr && j < b->nr) {
153                         if (ra[i].start < rb[j].start)
154                                 new = &ra[i++];
155                         else if (ra[i].start > rb[j].start)
156                                 new = &rb[j++];
157                         else if (ra[i].end < rb[j].end)
158                                 new = &ra[i++];
159                         else
160                                 new = &rb[j++];
161                 } else if (i < a->nr)      /* b exhausted */
162                         new = &ra[i++];
163                 else                       /* a exhausted */
164                         new = &rb[j++];
165                 if (new->start == new->end)
166                         ; /* empty range */
167                 else if (!o || out->ranges[o-1].end < new->start) {
168                         range_set_grow(out, 1);
169                         out->ranges[o].start = new->start;
170                         out->ranges[o].end = new->end;
171                         o++;
172                 } else if (out->ranges[o-1].end < new->end) {
173                         out->ranges[o-1].end = new->end;
174                 }
175         }
176         out->nr = o;
177 }
178
179 /*
180  * Difference of range sets (out = a \ b).  Pass the "interesting"
181  * ranges as 'a' and the target side of the diff as 'b': it removes
182  * the ranges for which the commit is responsible.
183  */
184 static void range_set_difference(struct range_set *out,
185                                   struct range_set *a, struct range_set *b)
186 {
187         int i, j =  0;
188         for (i = 0; i < a->nr; i++) {
189                 long start = a->ranges[i].start;
190                 long end = a->ranges[i].end;
191                 while (start < end) {
192                         while (j < b->nr && start >= b->ranges[j].end)
193                                 /*
194                                  * a:         |-------
195                                  * b: ------|
196                                  */
197                                 j++;
198                         if (j >= b->nr || end < b->ranges[j].start) {
199                                 /*
200                                  * b exhausted, or
201                                  * a:  ----|
202                                  * b:         |----
203                                  */
204                                 range_set_append(out, start, end);
205                                 break;
206                         }
207                         if (start >= b->ranges[j].start) {
208                                 /*
209                                  * a:     |--????
210                                  * b: |------|
211                                  */
212                                 start = b->ranges[j].end;
213                         } else if (end > b->ranges[j].start) {
214                                 /*
215                                  * a: |-----|
216                                  * b:    |--?????
217                                  */
218                                 if (start < b->ranges[j].start)
219                                         range_set_append(out, start, b->ranges[j].start);
220                                 start = b->ranges[j].end;
221                         }
222                 }
223         }
224 }
225
226 static void diff_ranges_init(struct diff_ranges *diff)
227 {
228         range_set_init(&diff->parent, 0);
229         range_set_init(&diff->target, 0);
230 }
231
232 static void diff_ranges_release(struct diff_ranges *diff)
233 {
234         range_set_release(&diff->parent);
235         range_set_release(&diff->target);
236 }
237
238 void line_log_data_init(struct line_log_data *r)
239 {
240         memset(r, 0, sizeof(struct line_log_data));
241         range_set_init(&r->ranges, 0);
242 }
243
244 static void line_log_data_clear(struct line_log_data *r)
245 {
246         range_set_release(&r->ranges);
247         if (r->pair)
248                 diff_free_filepair(r->pair);
249 }
250
251 static void free_line_log_data(struct line_log_data *r)
252 {
253         while (r) {
254                 struct line_log_data *next = r->next;
255                 line_log_data_clear(r);
256                 free(r);
257                 r = next;
258         }
259 }
260
261 static struct line_log_data *
262 search_line_log_data(struct line_log_data *list, const char *path,
263                      struct line_log_data **insertion_point)
264 {
265         struct line_log_data *p = list;
266         if (insertion_point)
267                 *insertion_point = NULL;
268         while (p) {
269                 int cmp = strcmp(p->path, path);
270                 if (!cmp)
271                         return p;
272                 if (insertion_point && cmp < 0)
273                         *insertion_point = p;
274                 p = p->next;
275         }
276         return NULL;
277 }
278
279 /*
280  * Note: takes ownership of 'path', which happens to be what the only
281  * caller needs.
282  */
283 static void line_log_data_insert(struct line_log_data **list,
284                                  char *path,
285                                  long begin, long end)
286 {
287         struct line_log_data *ip;
288         struct line_log_data *p = search_line_log_data(*list, path, &ip);
289
290         if (p) {
291                 range_set_append_unsafe(&p->ranges, begin, end);
292                 sort_and_merge_range_set(&p->ranges);
293                 free(path);
294                 return;
295         }
296
297         p = xcalloc(1, sizeof(struct line_log_data));
298         p->path = path;
299         range_set_append(&p->ranges, begin, end);
300         if (ip) {
301                 p->next = ip->next;
302                 ip->next = p;
303         } else {
304                 p->next = *list;
305                 *list = p;
306         }
307 }
308
309 struct collect_diff_cbdata {
310         struct diff_ranges *diff;
311 };
312
313 static int collect_diff_cb(long start_a, long count_a,
314                            long start_b, long count_b,
315                            void *data)
316 {
317         struct collect_diff_cbdata *d = data;
318
319         if (count_a >= 0)
320                 range_set_append(&d->diff->parent, start_a, start_a + count_a);
321         if (count_b >= 0)
322                 range_set_append(&d->diff->target, start_b, start_b + count_b);
323
324         return 0;
325 }
326
327 static void collect_diff(mmfile_t *parent, mmfile_t *target, struct diff_ranges *out)
328 {
329         struct collect_diff_cbdata cbdata = {NULL};
330         xpparam_t xpp;
331         xdemitconf_t xecfg;
332         xdemitcb_t ecb;
333
334         memset(&xpp, 0, sizeof(xpp));
335         memset(&xecfg, 0, sizeof(xecfg));
336         xecfg.ctxlen = xecfg.interhunkctxlen = 0;
337
338         cbdata.diff = out;
339         xecfg.hunk_func = collect_diff_cb;
340         memset(&ecb, 0, sizeof(ecb));
341         ecb.priv = &cbdata;
342         xdi_diff(parent, target, &xpp, &xecfg, &ecb);
343 }
344
345 /*
346  * These are handy for debugging.  Removing them with #if 0 silences
347  * the "unused function" warning.
348  */
349 #if 0
350 static void dump_range_set(struct range_set *rs, const char *desc)
351 {
352         int i;
353         printf("range set %s (%d items):\n", desc, rs->nr);
354         for (i = 0; i < rs->nr; i++)
355                 printf("\t[%ld,%ld]\n", rs->ranges[i].start, rs->ranges[i].end);
356 }
357
358 static void dump_line_log_data(struct line_log_data *r)
359 {
360         char buf[4096];
361         while (r) {
362                 snprintf(buf, 4096, "file %s\n", r->path);
363                 dump_range_set(&r->ranges, buf);
364                 r = r->next;
365         }
366 }
367
368 static void dump_diff_ranges(struct diff_ranges *diff, const char *desc)
369 {
370         int i;
371         assert(diff->parent.nr == diff->target.nr);
372         printf("diff ranges %s (%d items):\n", desc, diff->parent.nr);
373         printf("\tparent\ttarget\n");
374         for (i = 0; i < diff->parent.nr; i++) {
375                 printf("\t[%ld,%ld]\t[%ld,%ld]\n",
376                        diff->parent.ranges[i].start,
377                        diff->parent.ranges[i].end,
378                        diff->target.ranges[i].start,
379                        diff->target.ranges[i].end);
380         }
381 }
382 #endif
383
384
385 static int ranges_overlap(struct range *a, struct range *b)
386 {
387         return !(a->end <= b->start || b->end <= a->start);
388 }
389
390 /*
391  * Given a diff and the set of interesting ranges, determine all hunks
392  * of the diff which touch (overlap) at least one of the interesting
393  * ranges in the target.
394  */
395 static void diff_ranges_filter_touched(struct diff_ranges *out,
396                                        struct diff_ranges *diff,
397                                        struct range_set *rs)
398 {
399         int i, j = 0;
400
401         assert(out->target.nr == 0);
402
403         for (i = 0; i < diff->target.nr; i++) {
404                 while (diff->target.ranges[i].start > rs->ranges[j].end) {
405                         j++;
406                         if (j == rs->nr)
407                                 return;
408                 }
409                 if (ranges_overlap(&diff->target.ranges[i], &rs->ranges[j])) {
410                         range_set_append(&out->parent,
411                                          diff->parent.ranges[i].start,
412                                          diff->parent.ranges[i].end);
413                         range_set_append(&out->target,
414                                          diff->target.ranges[i].start,
415                                          diff->target.ranges[i].end);
416                 }
417         }
418 }
419
420 /*
421  * Adjust the line counts in 'rs' to account for the lines
422  * added/removed in the diff.
423  */
424 static void range_set_shift_diff(struct range_set *out,
425                                  struct range_set *rs,
426                                  struct diff_ranges *diff)
427 {
428         int i, j = 0;
429         long offset = 0;
430         struct range *src = rs->ranges;
431         struct range *target = diff->target.ranges;
432         struct range *parent = diff->parent.ranges;
433
434         for (i = 0; i < rs->nr; i++) {
435                 while (j < diff->target.nr && src[i].start >= target[j].start) {
436                         offset += (parent[j].end-parent[j].start)
437                                 - (target[j].end-target[j].start);
438                         j++;
439                 }
440                 range_set_append(out, src[i].start+offset, src[i].end+offset);
441         }
442 }
443
444 /*
445  * Given a diff and the set of interesting ranges, map the ranges
446  * across the diff.  That is: observe that the target commit takes
447  * blame for all the + (target-side) ranges.  So for every pair of
448  * ranges in the diff that was touched, we remove the latter and add
449  * its parent side.
450  */
451 static void range_set_map_across_diff(struct range_set *out,
452                                       struct range_set *rs,
453                                       struct diff_ranges *diff,
454                                       struct diff_ranges **touched_out)
455 {
456         struct diff_ranges *touched = xmalloc(sizeof(*touched));
457         struct range_set tmp1 = RANGE_SET_INIT;
458         struct range_set tmp2 = RANGE_SET_INIT;
459
460         diff_ranges_init(touched);
461         diff_ranges_filter_touched(touched, diff, rs);
462         range_set_difference(&tmp1, rs, &touched->target);
463         range_set_shift_diff(&tmp2, &tmp1, diff);
464         range_set_union(out, &tmp2, &touched->parent);
465         range_set_release(&tmp1);
466         range_set_release(&tmp2);
467
468         *touched_out = touched;
469 }
470
471 static struct commit *check_single_commit(struct rev_info *revs)
472 {
473         struct object *commit = NULL;
474         int found = -1;
475         int i;
476
477         for (i = 0; i < revs->pending.nr; i++) {
478                 struct object *obj = revs->pending.objects[i].item;
479                 if (obj->flags & UNINTERESTING)
480                         continue;
481                 while (obj->type == OBJ_TAG)
482                         obj = deref_tag(obj, NULL, 0);
483                 if (obj->type != OBJ_COMMIT)
484                         die("Non commit %s?", revs->pending.objects[i].name);
485                 if (commit)
486                         die("More than one commit to dig from: %s and %s?",
487                             revs->pending.objects[i].name,
488                             revs->pending.objects[found].name);
489                 commit = obj;
490                 found = i;
491         }
492
493         if (!commit)
494                 die("No commit specified?");
495
496         return (struct commit *) commit;
497 }
498
499 static void fill_blob_sha1(struct commit *commit, struct diff_filespec *spec)
500 {
501         unsigned mode;
502         unsigned char sha1[20];
503
504         if (get_tree_entry(commit->object.sha1, spec->path,
505                            sha1, &mode))
506                 die("There is no path %s in the commit", spec->path);
507         fill_filespec(spec, sha1, 1, mode);
508
509         return;
510 }
511
512 static void fill_line_ends(struct diff_filespec *spec, long *lines,
513                            unsigned long **line_ends)
514 {
515         int num = 0, size = 50;
516         long cur = 0;
517         unsigned long *ends = NULL;
518         char *data = NULL;
519
520         if (diff_populate_filespec(spec, 0))
521                 die("Cannot read blob %s", sha1_to_hex(spec->sha1));
522
523         ends = xmalloc(size * sizeof(*ends));
524         ends[cur++] = 0;
525         data = spec->data;
526         while (num < spec->size) {
527                 if (data[num] == '\n' || num == spec->size - 1) {
528                         ALLOC_GROW(ends, (cur + 1), size);
529                         ends[cur++] = num;
530                 }
531                 num++;
532         }
533
534         /* shrink the array to fit the elements */
535         ends = xrealloc(ends, cur * sizeof(*ends));
536         *lines = cur-1;
537         *line_ends = ends;
538 }
539
540 struct nth_line_cb {
541         struct diff_filespec *spec;
542         long lines;
543         unsigned long *line_ends;
544 };
545
546 static const char *nth_line(void *data, long line)
547 {
548         struct nth_line_cb *d = data;
549         assert(d && line <= d->lines);
550         assert(d->spec && d->spec->data);
551
552         if (line == 0)
553                 return (char *)d->spec->data;
554         else
555                 return (char *)d->spec->data + d->line_ends[line] + 1;
556 }
557
558 static struct line_log_data *
559 parse_lines(struct commit *commit, const char *prefix, struct string_list *args)
560 {
561         long lines = 0;
562         unsigned long *ends = NULL;
563         struct nth_line_cb cb_data;
564         struct string_list_item *item;
565         struct line_log_data *ranges = NULL;
566
567         for_each_string_list_item(item, args) {
568                 const char *name_part, *range_part;
569                 char *full_name;
570                 struct diff_filespec *spec;
571                 long begin = 0, end = 0;
572
573                 name_part = skip_range_arg(item->string);
574                 if (!name_part || *name_part != ':' || !name_part[1])
575                         die("-L argument '%s' not of the form start,end:file",
576                             item->string);
577                 range_part = xstrndup(item->string, name_part - item->string);
578                 name_part++;
579
580                 full_name = prefix_path(prefix, prefix ? strlen(prefix) : 0,
581                                         name_part);
582
583                 spec = alloc_filespec(full_name);
584                 fill_blob_sha1(commit, spec);
585                 fill_line_ends(spec, &lines, &ends);
586                 cb_data.spec = spec;
587                 cb_data.lines = lines;
588                 cb_data.line_ends = ends;
589
590                 if (parse_range_arg(range_part, nth_line, &cb_data,
591                                     lines, &begin, &end,
592                                     full_name))
593                         die("malformed -L argument '%s'", range_part);
594                 if (begin < 1)
595                         begin = 1;
596                 if (end < 1)
597                         end = lines;
598                 begin--;
599                 if (lines < end || lines < begin)
600                         die("file %s has only %ld lines", name_part, lines);
601                 line_log_data_insert(&ranges, full_name, begin, end);
602
603                 free_filespec(spec);
604                 free(ends);
605                 ends = NULL;
606         }
607
608         return ranges;
609 }
610
611 static struct line_log_data *line_log_data_copy_one(struct line_log_data *r)
612 {
613         struct line_log_data *ret = xmalloc(sizeof(*ret));
614
615         assert(r);
616         line_log_data_init(ret);
617         range_set_copy(&ret->ranges, &r->ranges);
618
619         ret->path = xstrdup(r->path);
620
621         return ret;
622 }
623
624 static struct line_log_data *
625 line_log_data_copy(struct line_log_data *r)
626 {
627         struct line_log_data *ret = NULL;
628         struct line_log_data *tmp = NULL, *prev = NULL;
629
630         assert(r);
631         ret = tmp = prev = line_log_data_copy_one(r);
632         r = r->next;
633         while (r) {
634                 tmp = line_log_data_copy_one(r);
635                 prev->next = tmp;
636                 prev = tmp;
637                 r = r->next;
638         }
639
640         return ret;
641 }
642
643 /* merge two range sets across files */
644 static struct line_log_data *line_log_data_merge(struct line_log_data *a,
645                                                  struct line_log_data *b)
646 {
647         struct line_log_data *head = NULL, **pp = &head;
648
649         while (a || b) {
650                 struct line_log_data *src;
651                 struct line_log_data *src2 = NULL;
652                 struct line_log_data *d;
653                 int cmp;
654                 if (!a)
655                         cmp = 1;
656                 else if (!b)
657                         cmp = -1;
658                 else
659                         cmp = strcmp(a->path, b->path);
660                 if (cmp < 0) {
661                         src = a;
662                         a = a->next;
663                 } else if (cmp == 0) {
664                         src = a;
665                         a = a->next;
666                         src2 = b;
667                         b = b->next;
668                 } else {
669                         src = b;
670                         b = b->next;
671                 }
672                 d = xmalloc(sizeof(struct line_log_data));
673                 line_log_data_init(d);
674                 d->path = xstrdup(src->path);
675                 *pp = d;
676                 pp = &d->next;
677                 if (src2)
678                         range_set_union(&d->ranges, &src->ranges, &src2->ranges);
679                 else
680                         range_set_copy(&d->ranges, &src->ranges);
681         }
682
683         return head;
684 }
685
686 static void add_line_range(struct rev_info *revs, struct commit *commit,
687                            struct line_log_data *range)
688 {
689         struct line_log_data *old = NULL;
690         struct line_log_data *new = NULL;
691
692         old = lookup_decoration(&revs->line_log_data, &commit->object);
693         if (old && range) {
694                 new = line_log_data_merge(old, range);
695                 free_line_log_data(old);
696         } else if (range)
697                 new = line_log_data_copy(range);
698
699         if (new)
700                 add_decoration(&revs->line_log_data, &commit->object, new);
701 }
702
703 static void clear_commit_line_range(struct rev_info *revs, struct commit *commit)
704 {
705         struct line_log_data *r;
706         r = lookup_decoration(&revs->line_log_data, &commit->object);
707         if (!r)
708                 return;
709         free_line_log_data(r);
710         add_decoration(&revs->line_log_data, &commit->object, NULL);
711 }
712
713 static struct line_log_data *lookup_line_range(struct rev_info *revs,
714                                                struct commit *commit)
715 {
716         struct line_log_data *ret = NULL;
717         struct line_log_data *d;
718
719         ret = lookup_decoration(&revs->line_log_data, &commit->object);
720
721         for (d = ret; d; d = d->next)
722                 range_set_check_invariants(&d->ranges);
723
724         return ret;
725 }
726
727 void line_log_init(struct rev_info *rev, const char *prefix, struct string_list *args)
728 {
729         struct commit *commit = NULL;
730         struct line_log_data *range;
731
732         commit = check_single_commit(rev);
733         range = parse_lines(commit, prefix, args);
734         add_line_range(rev, commit, range);
735
736         if (!rev->diffopt.detect_rename) {
737                 int i, count = 0;
738                 struct line_log_data *r = range;
739                 const char **paths;
740                 while (r) {
741                         count++;
742                         r = r->next;
743                 }
744                 paths = xmalloc((count+1)*sizeof(char *));
745                 r = range;
746                 for (i = 0; i < count; i++) {
747                         paths[i] = xstrdup(r->path);
748                         r = r->next;
749                 }
750                 paths[count] = NULL;
751                 init_pathspec(&rev->diffopt.pathspec, paths);
752                 free(paths);
753         }
754 }
755
756 static void load_tree_desc(struct tree_desc *desc, void **tree,
757                            const unsigned char *sha1)
758 {
759         unsigned long size;
760         *tree = read_object_with_reference(sha1, tree_type, &size, NULL);
761         if (!*tree)
762                 die("Unable to read tree (%s)", sha1_to_hex(sha1));
763         init_tree_desc(desc, *tree, size);
764 }
765
766 static int count_parents(struct commit *commit)
767 {
768         struct commit_list *parents = commit->parents;
769         int count = 0;
770         while (parents) {
771                 count++;
772                 parents = parents->next;
773         }
774         return count;
775 }
776
777 static void move_diff_queue(struct diff_queue_struct *dst,
778                             struct diff_queue_struct *src)
779 {
780         assert(src != dst);
781         memcpy(dst, src, sizeof(struct diff_queue_struct));
782         DIFF_QUEUE_CLEAR(src);
783 }
784
785 static void filter_diffs_for_paths(struct line_log_data *range, int keep_deletions)
786 {
787         int i;
788         struct diff_queue_struct outq;
789         DIFF_QUEUE_CLEAR(&outq);
790
791         for (i = 0; i < diff_queued_diff.nr; i++) {
792                 struct diff_filepair *p = diff_queued_diff.queue[i];
793                 struct line_log_data *rg = NULL;
794
795                 if (!DIFF_FILE_VALID(p->two)) {
796                         if (keep_deletions)
797                                 diff_q(&outq, p);
798                         else
799                                 diff_free_filepair(p);
800                         continue;
801                 }
802                 for (rg = range; rg; rg = rg->next) {
803                         if (!strcmp(rg->path, p->two->path))
804                                 break;
805                 }
806                 if (rg)
807                         diff_q(&outq, p);
808                 else
809                         diff_free_filepair(p);
810         }
811         free(diff_queued_diff.queue);
812         diff_queued_diff = outq;
813 }
814
815 static inline int diff_might_be_rename(void)
816 {
817         int i;
818         for (i = 0; i < diff_queued_diff.nr; i++)
819                 if (!DIFF_FILE_VALID(diff_queued_diff.queue[i]->one)) {
820                         /* fprintf(stderr, "diff_might_be_rename found creation of: %s\n", */
821                         /*      diff_queued_diff.queue[i]->two->path); */
822                         return 1;
823                 }
824         return 0;
825 }
826
827 static void queue_diffs(struct line_log_data *range,
828                         struct diff_options *opt,
829                         struct diff_queue_struct *queue,
830                         struct commit *commit, struct commit *parent)
831 {
832         void *tree1 = NULL, *tree2 = NULL;
833         struct tree_desc desc1, desc2;
834
835         assert(commit);
836         load_tree_desc(&desc2, &tree2, commit->tree->object.sha1);
837         if (parent)
838                 load_tree_desc(&desc1, &tree1, parent->tree->object.sha1);
839         else
840                 init_tree_desc(&desc1, "", 0);
841
842         DIFF_QUEUE_CLEAR(&diff_queued_diff);
843         diff_tree(&desc1, &desc2, "", opt);
844         if (opt->detect_rename) {
845                 filter_diffs_for_paths(range, 1);
846                 if (diff_might_be_rename())
847                         diffcore_std(opt);
848                 filter_diffs_for_paths(range, 0);
849         }
850         move_diff_queue(queue, &diff_queued_diff);
851
852         if (tree1)
853                 free(tree1);
854         if (tree2)
855                 free(tree2);
856 }
857
858 static char *get_nth_line(long line, unsigned long *ends, void *data)
859 {
860         if (line == 0)
861                 return (char *)data;
862         else
863                 return (char *)data + ends[line] + 1;
864 }
865
866 static void print_line(const char *prefix, char first,
867                        long line, unsigned long *ends, void *data,
868                        const char *color, const char *reset)
869 {
870         char *begin = get_nth_line(line, ends, data);
871         char *end = get_nth_line(line+1, ends, data);
872         int had_nl = 0;
873
874         if (end > begin && end[-1] == '\n') {
875                 end--;
876                 had_nl = 1;
877         }
878
879         fputs(prefix, stdout);
880         fputs(color, stdout);
881         putchar(first);
882         fwrite(begin, 1, end-begin, stdout);
883         fputs(reset, stdout);
884         putchar('\n');
885         if (!had_nl)
886                 fputs("\\ No newline at end of file\n", stdout);
887 }
888
889 static char *output_prefix(struct diff_options *opt)
890 {
891         char *prefix = "";
892
893         if (opt->output_prefix) {
894                 struct strbuf *sb = opt->output_prefix(opt, opt->output_prefix_data);
895                 prefix = sb->buf;
896         }
897
898         return prefix;
899 }
900
901 static void dump_diff_hacky_one(struct rev_info *rev, struct line_log_data *range)
902 {
903         int i, j = 0;
904         long p_lines, t_lines;
905         unsigned long *p_ends = NULL, *t_ends = NULL;
906         struct diff_filepair *pair = range->pair;
907         struct diff_ranges *diff = &range->diff;
908
909         struct diff_options *opt = &rev->diffopt;
910         char *prefix = output_prefix(opt);
911         const char *c_reset = diff_get_color(opt->use_color, DIFF_RESET);
912         const char *c_frag = diff_get_color(opt->use_color, DIFF_FRAGINFO);
913         const char *c_meta = diff_get_color(opt->use_color, DIFF_METAINFO);
914         const char *c_old = diff_get_color(opt->use_color, DIFF_FILE_OLD);
915         const char *c_new = diff_get_color(opt->use_color, DIFF_FILE_NEW);
916         const char *c_plain = diff_get_color(opt->use_color, DIFF_PLAIN);
917
918         if (!pair || !diff)
919                 return;
920
921         if (pair->one->sha1_valid)
922                 fill_line_ends(pair->one, &p_lines, &p_ends);
923         fill_line_ends(pair->two, &t_lines, &t_ends);
924
925         printf("%s%sdiff --git a/%s b/%s%s\n", prefix, c_meta, pair->one->path, pair->two->path, c_reset);
926         printf("%s%s--- %s%s%s\n", prefix, c_meta,
927                pair->one->sha1_valid ? "a/" : "",
928                pair->one->sha1_valid ? pair->one->path : "/dev/null",
929                c_reset);
930         printf("%s%s+++ b/%s%s\n", prefix, c_meta, pair->two->path, c_reset);
931         for (i = 0; i < range->ranges.nr; i++) {
932                 long p_start, p_end;
933                 long t_start = range->ranges.ranges[i].start;
934                 long t_end = range->ranges.ranges[i].end;
935                 long t_cur = t_start;
936                 int j_last;
937
938                 while (j < diff->target.nr && diff->target.ranges[j].end < t_start)
939                         j++;
940                 if (j == diff->target.nr || diff->target.ranges[j].start > t_end)
941                         continue;
942
943                 /* Scan ahead to determine the last diff that falls in this range */
944                 j_last = j;
945                 while (j_last < diff->target.nr && diff->target.ranges[j_last].start < t_end)
946                         j_last++;
947                 if (j_last > j)
948                         j_last--;
949
950                 /*
951                  * Compute parent hunk headers: we know that the diff
952                  * has the correct line numbers (but not all hunks).
953                  * So it suffices to shift the start/end according to
954                  * the line numbers of the first/last hunk(s) that
955                  * fall in this range.
956                  */
957                 if (t_start < diff->target.ranges[j].start)
958                         p_start = diff->parent.ranges[j].start - (diff->target.ranges[j].start-t_start);
959                 else
960                         p_start = diff->parent.ranges[j].start;
961                 if (t_end > diff->target.ranges[j_last].end)
962                         p_end = diff->parent.ranges[j_last].end + (t_end-diff->target.ranges[j_last].end);
963                 else
964                         p_end = diff->parent.ranges[j_last].end;
965
966                 if (!p_start && !p_end) {
967                         p_start = -1;
968                         p_end = -1;
969                 }
970
971                 /* Now output a diff hunk for this range */
972                 printf("%s%s@@ -%ld,%ld +%ld,%ld @@%s\n",
973                        prefix, c_frag,
974                        p_start+1, p_end-p_start, t_start+1, t_end-t_start,
975                        c_reset);
976                 while (j < diff->target.nr && diff->target.ranges[j].start < t_end) {
977                         int k;
978                         for (; t_cur < diff->target.ranges[j].start; t_cur++)
979                                 print_line(prefix, ' ', t_cur, t_ends, pair->two->data,
980                                            c_plain, c_reset);
981                         for (k = diff->parent.ranges[j].start; k < diff->parent.ranges[j].end; k++)
982                                 print_line(prefix, '-', k, p_ends, pair->one->data,
983                                            c_old, c_reset);
984                         for (; t_cur < diff->target.ranges[j].end && t_cur < t_end; t_cur++)
985                                 print_line(prefix, '+', t_cur, t_ends, pair->two->data,
986                                            c_new, c_reset);
987                         j++;
988                 }
989                 for (; t_cur < t_end; t_cur++)
990                         print_line(prefix, ' ', t_cur, t_ends, pair->two->data,
991                                    c_plain, c_reset);
992         }
993
994         free(p_ends);
995         free(t_ends);
996 }
997
998 /*
999  * NEEDSWORK: manually building a diff here is not the Right
1000  * Thing(tm).  log -L should be built into the diff pipeline.
1001  */
1002 static void dump_diff_hacky(struct rev_info *rev, struct line_log_data *range)
1003 {
1004         puts(output_prefix(&rev->diffopt));
1005         while (range) {
1006                 dump_diff_hacky_one(rev, range);
1007                 range = range->next;
1008         }
1009 }
1010
1011 /*
1012  * Unlike most other functions, this destructively operates on
1013  * 'range'.
1014  */
1015 static int process_diff_filepair(struct rev_info *rev,
1016                                  struct diff_filepair *pair,
1017                                  struct line_log_data *range,
1018                                  struct diff_ranges **diff_out)
1019 {
1020         struct line_log_data *rg = range;
1021         struct range_set tmp;
1022         struct diff_ranges diff;
1023         mmfile_t file_parent, file_target;
1024
1025         assert(pair->two->path);
1026         while (rg) {
1027                 assert(rg->path);
1028                 if (!strcmp(rg->path, pair->two->path))
1029                         break;
1030                 rg = rg->next;
1031         }
1032
1033         if (!rg)
1034                 return 0;
1035         if (rg->ranges.nr == 0)
1036                 return 0;
1037
1038         assert(pair->two->sha1_valid);
1039         diff_populate_filespec(pair->two, 0);
1040         file_target.ptr = pair->two->data;
1041         file_target.size = pair->two->size;
1042
1043         if (pair->one->sha1_valid) {
1044                 diff_populate_filespec(pair->one, 0);
1045                 file_parent.ptr = pair->one->data;
1046                 file_parent.size = pair->one->size;
1047         } else {
1048                 file_parent.ptr = "";
1049                 file_parent.size = 0;
1050         }
1051
1052         diff_ranges_init(&diff);
1053         collect_diff(&file_parent, &file_target, &diff);
1054
1055         /* NEEDSWORK should apply some heuristics to prevent mismatches */
1056         free(rg->path);
1057         rg->path = xstrdup(pair->one->path);
1058
1059         range_set_init(&tmp, 0);
1060         range_set_map_across_diff(&tmp, &rg->ranges, &diff, diff_out);
1061         range_set_release(&rg->ranges);
1062         range_set_move(&rg->ranges, &tmp);
1063
1064         diff_ranges_release(&diff);
1065
1066         return ((*diff_out)->parent.nr > 0);
1067 }
1068
1069 static struct diff_filepair *diff_filepair_dup(struct diff_filepair *pair)
1070 {
1071         struct diff_filepair *new = xmalloc(sizeof(struct diff_filepair));
1072         new->one = pair->one;
1073         new->two = pair->two;
1074         new->one->count++;
1075         new->two->count++;
1076         return new;
1077 }
1078
1079 static void free_diffqueues(int n, struct diff_queue_struct *dq)
1080 {
1081         int i, j;
1082         for (i = 0; i < n; i++)
1083                 for (j = 0; j < dq[i].nr; j++)
1084                         diff_free_filepair(dq[i].queue[j]);
1085         free(dq);
1086 }
1087
1088 static int process_all_files(struct line_log_data **range_out,
1089                              struct rev_info *rev,
1090                              struct diff_queue_struct *queue,
1091                              struct line_log_data *range)
1092 {
1093         int i, changed = 0;
1094
1095         *range_out = line_log_data_copy(range);
1096
1097         for (i = 0; i < queue->nr; i++) {
1098                 struct diff_ranges *pairdiff = NULL;
1099                 struct diff_filepair *pair = queue->queue[i];
1100                 if (process_diff_filepair(rev, pair, *range_out, &pairdiff)) {
1101                         /*
1102                          * Store away the diff for later output.  We
1103                          * tuck it in the ranges we got as _input_,
1104                          * since that's the commit that caused the
1105                          * diff.
1106                          *
1107                          * NEEDSWORK not enough when we get around to
1108                          * doing something interesting with merges;
1109                          * currently each invocation on a merge parent
1110                          * trashes the previous one's diff.
1111                          *
1112                          * NEEDSWORK tramples over data structures not owned here
1113                          */
1114                         struct line_log_data *rg = range;
1115                         changed++;
1116                         while (rg && strcmp(rg->path, pair->two->path))
1117                                 rg = rg->next;
1118                         assert(rg);
1119                         rg->pair = diff_filepair_dup(queue->queue[i]);
1120                         memcpy(&rg->diff, pairdiff, sizeof(struct diff_ranges));
1121                 }
1122         }
1123
1124         return changed;
1125 }
1126
1127 int line_log_print(struct rev_info *rev, struct commit *commit)
1128 {
1129         struct line_log_data *range = lookup_line_range(rev, commit);
1130
1131         show_log(rev);
1132         dump_diff_hacky(rev, range);
1133         return 1;
1134 }
1135
1136 static int process_ranges_ordinary_commit(struct rev_info *rev, struct commit *commit,
1137                                           struct line_log_data *range)
1138 {
1139         struct commit *parent = NULL;
1140         struct diff_queue_struct queue;
1141         struct line_log_data *parent_range;
1142         int changed;
1143
1144         if (commit->parents)
1145                 parent = commit->parents->item;
1146
1147         queue_diffs(range, &rev->diffopt, &queue, commit, parent);
1148         changed = process_all_files(&parent_range, rev, &queue, range);
1149         if (parent)
1150                 add_line_range(rev, parent, parent_range);
1151         return changed;
1152 }
1153
1154 static int process_ranges_merge_commit(struct rev_info *rev, struct commit *commit,
1155                                        struct line_log_data *range)
1156 {
1157         struct diff_queue_struct *diffqueues;
1158         struct line_log_data **cand;
1159         struct commit **parents;
1160         struct commit_list *p;
1161         int i;
1162         int nparents = count_parents(commit);
1163
1164         diffqueues = xmalloc(nparents * sizeof(*diffqueues));
1165         cand = xmalloc(nparents * sizeof(*cand));
1166         parents = xmalloc(nparents * sizeof(*parents));
1167
1168         p = commit->parents;
1169         for (i = 0; i < nparents; i++) {
1170                 parents[i] = p->item;
1171                 p = p->next;
1172                 queue_diffs(range, &rev->diffopt, &diffqueues[i], commit, parents[i]);
1173         }
1174
1175         for (i = 0; i < nparents; i++) {
1176                 int changed;
1177                 cand[i] = NULL;
1178                 changed = process_all_files(&cand[i], rev, &diffqueues[i], range);
1179                 if (!changed) {
1180                         /*
1181                          * This parent can take all the blame, so we
1182                          * don't follow any other path in history
1183                          */
1184                         add_line_range(rev, parents[i], cand[i]);
1185                         clear_commit_line_range(rev, commit);
1186                         commit->parents = xmalloc(sizeof(struct commit_list));
1187                         commit->parents->item = parents[i];
1188                         commit->parents->next = NULL;
1189                         free(parents);
1190                         free(cand);
1191                         free_diffqueues(nparents, diffqueues);
1192                         /* NEEDSWORK leaking like a sieve */
1193                         return 0;
1194                 }
1195         }
1196
1197         /*
1198          * No single parent took the blame.  We add the candidates
1199          * from the above loop to the parents.
1200          */
1201         for (i = 0; i < nparents; i++) {
1202                 add_line_range(rev, parents[i], cand[i]);
1203         }
1204
1205         clear_commit_line_range(rev, commit);
1206         free(parents);
1207         free(cand);
1208         free_diffqueues(nparents, diffqueues);
1209         return 1;
1210
1211         /* NEEDSWORK evil merge detection stuff */
1212         /* NEEDSWORK leaking like a sieve */
1213 }
1214
1215 static int process_ranges_arbitrary_commit(struct rev_info *rev, struct commit *commit)
1216 {
1217         struct line_log_data *range = lookup_line_range(rev, commit);
1218         int changed = 0;
1219
1220         if (range) {
1221                 if (!commit->parents || !commit->parents->next)
1222                         changed = process_ranges_ordinary_commit(rev, commit, range);
1223                 else
1224                         changed = process_ranges_merge_commit(rev, commit, range);
1225         }
1226
1227         if (!changed)
1228                 commit->object.flags |= TREESAME;
1229
1230         return changed;
1231 }
1232
1233 static enum rewrite_result line_log_rewrite_one(struct rev_info *rev, struct commit **pp)
1234 {
1235         for (;;) {
1236                 struct commit *p = *pp;
1237                 if (p->parents && p->parents->next)
1238                         return rewrite_one_ok;
1239                 if (p->object.flags & UNINTERESTING)
1240                         return rewrite_one_ok;
1241                 if (!(p->object.flags & TREESAME))
1242                         return rewrite_one_ok;
1243                 if (!p->parents)
1244                         return rewrite_one_noparents;
1245                 *pp = p->parents->item;
1246         }
1247 }
1248
1249 int line_log_filter(struct rev_info *rev)
1250 {
1251         struct commit *commit;
1252         struct commit_list *list = rev->commits;
1253         struct commit_list *out = NULL, **pp = &out;
1254
1255         while (list) {
1256                 struct commit_list *to_free = NULL;
1257                 commit = list->item;
1258                 if (process_ranges_arbitrary_commit(rev, commit)) {
1259                         *pp = list;
1260                         pp = &list->next;
1261                 } else
1262                         to_free = list;
1263                 list = list->next;
1264                 free(to_free);
1265         }
1266         *pp = NULL;
1267
1268         for (list = out; list; list = list->next)
1269                 rewrite_parents(rev, list->item, line_log_rewrite_one);
1270
1271         rev->commits = out;
1272
1273         return 0;
1274 }