Merge branch 'jk/cat-file-batch-optim'
[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 = 0; /* output cursor */
114
115         qsort(rs->ranges, rs->nr, sizeof(struct range), range_cmp);
116
117         for (i = 0; i < rs->nr; i++) {
118                 if (rs->ranges[i].start == rs->ranges[i].end)
119                         continue;
120                 if (o > 0 && rs->ranges[i].start <= rs->ranges[o-1].end) {
121                         if (rs->ranges[o-1].end < rs->ranges[i].end)
122                                 rs->ranges[o-1].end = rs->ranges[i].end;
123                 } else {
124                         rs->ranges[o].start = rs->ranges[i].start;
125                         rs->ranges[o].end = rs->ranges[i].end;
126                         o++;
127                 }
128         }
129         assert(o <= rs->nr);
130         rs->nr = o;
131
132         range_set_check_invariants(rs);
133 }
134
135 /*
136  * Union of range sets (i.e., sets of line numbers).  Used to merge
137  * them when searches meet at a common ancestor.
138  *
139  * This is also where the ranges are consolidated into canonical form:
140  * overlapping and adjacent ranges are merged, and empty ranges are
141  * removed.
142  */
143 static void range_set_union(struct range_set *out,
144                              struct range_set *a, struct range_set *b)
145 {
146         int i = 0, j = 0, o = 0;
147         struct range *ra = a->ranges;
148         struct range *rb = b->ranges;
149         /* cannot make an alias of out->ranges: it may change during grow */
150
151         assert(out->nr == 0);
152         while (i < a->nr || j < b->nr) {
153                 struct range *new;
154                 if (i < a->nr && j < b->nr) {
155                         if (ra[i].start < rb[j].start)
156                                 new = &ra[i++];
157                         else if (ra[i].start > rb[j].start)
158                                 new = &rb[j++];
159                         else if (ra[i].end < rb[j].end)
160                                 new = &ra[i++];
161                         else
162                                 new = &rb[j++];
163                 } else if (i < a->nr)      /* b exhausted */
164                         new = &ra[i++];
165                 else                       /* a exhausted */
166                         new = &rb[j++];
167                 if (new->start == new->end)
168                         ; /* empty range */
169                 else if (!o || out->ranges[o-1].end < new->start) {
170                         range_set_grow(out, 1);
171                         out->ranges[o].start = new->start;
172                         out->ranges[o].end = new->end;
173                         o++;
174                 } else if (out->ranges[o-1].end < new->end) {
175                         out->ranges[o-1].end = new->end;
176                 }
177         }
178         out->nr = o;
179 }
180
181 /*
182  * Difference of range sets (out = a \ b).  Pass the "interesting"
183  * ranges as 'a' and the target side of the diff as 'b': it removes
184  * the ranges for which the commit is responsible.
185  */
186 static void range_set_difference(struct range_set *out,
187                                   struct range_set *a, struct range_set *b)
188 {
189         int i, j =  0;
190         for (i = 0; i < a->nr; i++) {
191                 long start = a->ranges[i].start;
192                 long end = a->ranges[i].end;
193                 while (start < end) {
194                         while (j < b->nr && start >= b->ranges[j].end)
195                                 /*
196                                  * a:         |-------
197                                  * b: ------|
198                                  */
199                                 j++;
200                         if (j >= b->nr || end < b->ranges[j].start) {
201                                 /*
202                                  * b exhausted, or
203                                  * a:  ----|
204                                  * b:         |----
205                                  */
206                                 range_set_append(out, start, end);
207                                 break;
208                         }
209                         if (start >= b->ranges[j].start) {
210                                 /*
211                                  * a:     |--????
212                                  * b: |------|
213                                  */
214                                 start = b->ranges[j].end;
215                         } else if (end > b->ranges[j].start) {
216                                 /*
217                                  * a: |-----|
218                                  * b:    |--?????
219                                  */
220                                 if (start < b->ranges[j].start)
221                                         range_set_append(out, start, b->ranges[j].start);
222                                 start = b->ranges[j].end;
223                         }
224                 }
225         }
226 }
227
228 static void diff_ranges_init(struct diff_ranges *diff)
229 {
230         range_set_init(&diff->parent, 0);
231         range_set_init(&diff->target, 0);
232 }
233
234 static void diff_ranges_release(struct diff_ranges *diff)
235 {
236         range_set_release(&diff->parent);
237         range_set_release(&diff->target);
238 }
239
240 void line_log_data_init(struct line_log_data *r)
241 {
242         memset(r, 0, sizeof(struct line_log_data));
243         range_set_init(&r->ranges, 0);
244 }
245
246 static void line_log_data_clear(struct line_log_data *r)
247 {
248         range_set_release(&r->ranges);
249         if (r->pair)
250                 diff_free_filepair(r->pair);
251 }
252
253 static void free_line_log_data(struct line_log_data *r)
254 {
255         while (r) {
256                 struct line_log_data *next = r->next;
257                 line_log_data_clear(r);
258                 free(r);
259                 r = next;
260         }
261 }
262
263 static struct line_log_data *
264 search_line_log_data(struct line_log_data *list, const char *path,
265                      struct line_log_data **insertion_point)
266 {
267         struct line_log_data *p = list;
268         if (insertion_point)
269                 *insertion_point = NULL;
270         while (p) {
271                 int cmp = strcmp(p->path, path);
272                 if (!cmp)
273                         return p;
274                 if (insertion_point && cmp < 0)
275                         *insertion_point = p;
276                 p = p->next;
277         }
278         return NULL;
279 }
280
281 /*
282  * Note: takes ownership of 'path', which happens to be what the only
283  * caller needs.
284  */
285 static void line_log_data_insert(struct line_log_data **list,
286                                  char *path,
287                                  long begin, long end)
288 {
289         struct line_log_data *ip;
290         struct line_log_data *p = search_line_log_data(*list, path, &ip);
291
292         if (p) {
293                 range_set_append_unsafe(&p->ranges, begin, end);
294                 sort_and_merge_range_set(&p->ranges);
295                 free(path);
296                 return;
297         }
298
299         p = xcalloc(1, sizeof(struct line_log_data));
300         p->path = path;
301         range_set_append(&p->ranges, begin, end);
302         sort_and_merge_range_set(&p->ranges);
303         if (ip) {
304                 p->next = ip->next;
305                 ip->next = p;
306         } else {
307                 p->next = *list;
308                 *list = p;
309         }
310 }
311
312 struct collect_diff_cbdata {
313         struct diff_ranges *diff;
314 };
315
316 static int collect_diff_cb(long start_a, long count_a,
317                            long start_b, long count_b,
318                            void *data)
319 {
320         struct collect_diff_cbdata *d = data;
321
322         if (count_a >= 0)
323                 range_set_append(&d->diff->parent, start_a, start_a + count_a);
324         if (count_b >= 0)
325                 range_set_append(&d->diff->target, start_b, start_b + count_b);
326
327         return 0;
328 }
329
330 static void collect_diff(mmfile_t *parent, mmfile_t *target, struct diff_ranges *out)
331 {
332         struct collect_diff_cbdata cbdata = {NULL};
333         xpparam_t xpp;
334         xdemitconf_t xecfg;
335         xdemitcb_t ecb;
336
337         memset(&xpp, 0, sizeof(xpp));
338         memset(&xecfg, 0, sizeof(xecfg));
339         xecfg.ctxlen = xecfg.interhunkctxlen = 0;
340
341         cbdata.diff = out;
342         xecfg.hunk_func = collect_diff_cb;
343         memset(&ecb, 0, sizeof(ecb));
344         ecb.priv = &cbdata;
345         xdi_diff(parent, target, &xpp, &xecfg, &ecb);
346 }
347
348 /*
349  * These are handy for debugging.  Removing them with #if 0 silences
350  * the "unused function" warning.
351  */
352 #if 0
353 static void dump_range_set(struct range_set *rs, const char *desc)
354 {
355         int i;
356         printf("range set %s (%d items):\n", desc, rs->nr);
357         for (i = 0; i < rs->nr; i++)
358                 printf("\t[%ld,%ld]\n", rs->ranges[i].start, rs->ranges[i].end);
359 }
360
361 static void dump_line_log_data(struct line_log_data *r)
362 {
363         char buf[4096];
364         while (r) {
365                 snprintf(buf, 4096, "file %s\n", r->path);
366                 dump_range_set(&r->ranges, buf);
367                 r = r->next;
368         }
369 }
370
371 static void dump_diff_ranges(struct diff_ranges *diff, const char *desc)
372 {
373         int i;
374         assert(diff->parent.nr == diff->target.nr);
375         printf("diff ranges %s (%d items):\n", desc, diff->parent.nr);
376         printf("\tparent\ttarget\n");
377         for (i = 0; i < diff->parent.nr; i++) {
378                 printf("\t[%ld,%ld]\t[%ld,%ld]\n",
379                        diff->parent.ranges[i].start,
380                        diff->parent.ranges[i].end,
381                        diff->target.ranges[i].start,
382                        diff->target.ranges[i].end);
383         }
384 }
385 #endif
386
387
388 static int ranges_overlap(struct range *a, struct range *b)
389 {
390         return !(a->end <= b->start || b->end <= a->start);
391 }
392
393 /*
394  * Given a diff and the set of interesting ranges, determine all hunks
395  * of the diff which touch (overlap) at least one of the interesting
396  * ranges in the target.
397  */
398 static void diff_ranges_filter_touched(struct diff_ranges *out,
399                                        struct diff_ranges *diff,
400                                        struct range_set *rs)
401 {
402         int i, j = 0;
403
404         assert(out->target.nr == 0);
405
406         for (i = 0; i < diff->target.nr; i++) {
407                 while (diff->target.ranges[i].start > rs->ranges[j].end) {
408                         j++;
409                         if (j == rs->nr)
410                                 return;
411                 }
412                 if (ranges_overlap(&diff->target.ranges[i], &rs->ranges[j])) {
413                         range_set_append(&out->parent,
414                                          diff->parent.ranges[i].start,
415                                          diff->parent.ranges[i].end);
416                         range_set_append(&out->target,
417                                          diff->target.ranges[i].start,
418                                          diff->target.ranges[i].end);
419                 }
420         }
421 }
422
423 /*
424  * Adjust the line counts in 'rs' to account for the lines
425  * added/removed in the diff.
426  */
427 static void range_set_shift_diff(struct range_set *out,
428                                  struct range_set *rs,
429                                  struct diff_ranges *diff)
430 {
431         int i, j = 0;
432         long offset = 0;
433         struct range *src = rs->ranges;
434         struct range *target = diff->target.ranges;
435         struct range *parent = diff->parent.ranges;
436
437         for (i = 0; i < rs->nr; i++) {
438                 while (j < diff->target.nr && src[i].start >= target[j].start) {
439                         offset += (parent[j].end-parent[j].start)
440                                 - (target[j].end-target[j].start);
441                         j++;
442                 }
443                 range_set_append(out, src[i].start+offset, src[i].end+offset);
444         }
445 }
446
447 /*
448  * Given a diff and the set of interesting ranges, map the ranges
449  * across the diff.  That is: observe that the target commit takes
450  * blame for all the + (target-side) ranges.  So for every pair of
451  * ranges in the diff that was touched, we remove the latter and add
452  * its parent side.
453  */
454 static void range_set_map_across_diff(struct range_set *out,
455                                       struct range_set *rs,
456                                       struct diff_ranges *diff,
457                                       struct diff_ranges **touched_out)
458 {
459         struct diff_ranges *touched = xmalloc(sizeof(*touched));
460         struct range_set tmp1 = RANGE_SET_INIT;
461         struct range_set tmp2 = RANGE_SET_INIT;
462
463         diff_ranges_init(touched);
464         diff_ranges_filter_touched(touched, diff, rs);
465         range_set_difference(&tmp1, rs, &touched->target);
466         range_set_shift_diff(&tmp2, &tmp1, diff);
467         range_set_union(out, &tmp2, &touched->parent);
468         range_set_release(&tmp1);
469         range_set_release(&tmp2);
470
471         *touched_out = touched;
472 }
473
474 static struct commit *check_single_commit(struct rev_info *revs)
475 {
476         struct object *commit = NULL;
477         int found = -1;
478         int i;
479
480         for (i = 0; i < revs->pending.nr; i++) {
481                 struct object *obj = revs->pending.objects[i].item;
482                 if (obj->flags & UNINTERESTING)
483                         continue;
484                 while (obj->type == OBJ_TAG)
485                         obj = deref_tag(obj, NULL, 0);
486                 if (obj->type != OBJ_COMMIT)
487                         die("Non commit %s?", revs->pending.objects[i].name);
488                 if (commit)
489                         die("More than one commit to dig from: %s and %s?",
490                             revs->pending.objects[i].name,
491                             revs->pending.objects[found].name);
492                 commit = obj;
493                 found = i;
494         }
495
496         if (!commit)
497                 die("No commit specified?");
498
499         return (struct commit *) commit;
500 }
501
502 static void fill_blob_sha1(struct commit *commit, struct diff_filespec *spec)
503 {
504         unsigned mode;
505         unsigned char sha1[20];
506
507         if (get_tree_entry(commit->object.sha1, spec->path,
508                            sha1, &mode))
509                 die("There is no path %s in the commit", spec->path);
510         fill_filespec(spec, sha1, 1, mode);
511
512         return;
513 }
514
515 static void fill_line_ends(struct diff_filespec *spec, long *lines,
516                            unsigned long **line_ends)
517 {
518         int num = 0, size = 50;
519         long cur = 0;
520         unsigned long *ends = NULL;
521         char *data = NULL;
522
523         if (diff_populate_filespec(spec, 0))
524                 die("Cannot read blob %s", sha1_to_hex(spec->sha1));
525
526         ends = xmalloc(size * sizeof(*ends));
527         ends[cur++] = 0;
528         data = spec->data;
529         while (num < spec->size) {
530                 if (data[num] == '\n' || num == spec->size - 1) {
531                         ALLOC_GROW(ends, (cur + 1), size);
532                         ends[cur++] = num;
533                 }
534                 num++;
535         }
536
537         /* shrink the array to fit the elements */
538         ends = xrealloc(ends, cur * sizeof(*ends));
539         *lines = cur-1;
540         *line_ends = ends;
541 }
542
543 struct nth_line_cb {
544         struct diff_filespec *spec;
545         long lines;
546         unsigned long *line_ends;
547 };
548
549 static const char *nth_line(void *data, long line)
550 {
551         struct nth_line_cb *d = data;
552         assert(d && line <= d->lines);
553         assert(d->spec && d->spec->data);
554
555         if (line == 0)
556                 return (char *)d->spec->data;
557         else
558                 return (char *)d->spec->data + d->line_ends[line] + 1;
559 }
560
561 static struct line_log_data *
562 parse_lines(struct commit *commit, const char *prefix, struct string_list *args)
563 {
564         long lines = 0;
565         unsigned long *ends = NULL;
566         struct nth_line_cb cb_data;
567         struct string_list_item *item;
568         struct line_log_data *ranges = NULL;
569
570         for_each_string_list_item(item, args) {
571                 const char *name_part, *range_part;
572                 char *full_name;
573                 struct diff_filespec *spec;
574                 long begin = 0, end = 0;
575
576                 name_part = skip_range_arg(item->string);
577                 if (!name_part || *name_part != ':' || !name_part[1])
578                         die("-L argument '%s' not of the form start,end:file",
579                             item->string);
580                 range_part = xstrndup(item->string, name_part - item->string);
581                 name_part++;
582
583                 full_name = prefix_path(prefix, prefix ? strlen(prefix) : 0,
584                                         name_part);
585
586                 spec = alloc_filespec(full_name);
587                 fill_blob_sha1(commit, spec);
588                 fill_line_ends(spec, &lines, &ends);
589                 cb_data.spec = spec;
590                 cb_data.lines = lines;
591                 cb_data.line_ends = ends;
592
593                 if (parse_range_arg(range_part, nth_line, &cb_data,
594                                     lines, &begin, &end,
595                                     full_name))
596                         die("malformed -L argument '%s'", range_part);
597                 if (begin < 1)
598                         begin = 1;
599                 if (end < 1)
600                         end = lines;
601                 begin--;
602                 if (lines < end || lines < begin)
603                         die("file %s has only %ld lines", name_part, lines);
604                 line_log_data_insert(&ranges, full_name, begin, end);
605
606                 free_filespec(spec);
607                 free(ends);
608                 ends = NULL;
609         }
610
611         return ranges;
612 }
613
614 static struct line_log_data *line_log_data_copy_one(struct line_log_data *r)
615 {
616         struct line_log_data *ret = xmalloc(sizeof(*ret));
617
618         assert(r);
619         line_log_data_init(ret);
620         range_set_copy(&ret->ranges, &r->ranges);
621
622         ret->path = xstrdup(r->path);
623
624         return ret;
625 }
626
627 static struct line_log_data *
628 line_log_data_copy(struct line_log_data *r)
629 {
630         struct line_log_data *ret = NULL;
631         struct line_log_data *tmp = NULL, *prev = NULL;
632
633         assert(r);
634         ret = tmp = prev = line_log_data_copy_one(r);
635         r = r->next;
636         while (r) {
637                 tmp = line_log_data_copy_one(r);
638                 prev->next = tmp;
639                 prev = tmp;
640                 r = r->next;
641         }
642
643         return ret;
644 }
645
646 /* merge two range sets across files */
647 static struct line_log_data *line_log_data_merge(struct line_log_data *a,
648                                                  struct line_log_data *b)
649 {
650         struct line_log_data *head = NULL, **pp = &head;
651
652         while (a || b) {
653                 struct line_log_data *src;
654                 struct line_log_data *src2 = NULL;
655                 struct line_log_data *d;
656                 int cmp;
657                 if (!a)
658                         cmp = 1;
659                 else if (!b)
660                         cmp = -1;
661                 else
662                         cmp = strcmp(a->path, b->path);
663                 if (cmp < 0) {
664                         src = a;
665                         a = a->next;
666                 } else if (cmp == 0) {
667                         src = a;
668                         a = a->next;
669                         src2 = b;
670                         b = b->next;
671                 } else {
672                         src = b;
673                         b = b->next;
674                 }
675                 d = xmalloc(sizeof(struct line_log_data));
676                 line_log_data_init(d);
677                 d->path = xstrdup(src->path);
678                 *pp = d;
679                 pp = &d->next;
680                 if (src2)
681                         range_set_union(&d->ranges, &src->ranges, &src2->ranges);
682                 else
683                         range_set_copy(&d->ranges, &src->ranges);
684         }
685
686         return head;
687 }
688
689 static void add_line_range(struct rev_info *revs, struct commit *commit,
690                            struct line_log_data *range)
691 {
692         struct line_log_data *old = NULL;
693         struct line_log_data *new = NULL;
694
695         old = lookup_decoration(&revs->line_log_data, &commit->object);
696         if (old && range) {
697                 new = line_log_data_merge(old, range);
698                 free_line_log_data(old);
699         } else if (range)
700                 new = line_log_data_copy(range);
701
702         if (new)
703                 add_decoration(&revs->line_log_data, &commit->object, new);
704 }
705
706 static void clear_commit_line_range(struct rev_info *revs, struct commit *commit)
707 {
708         struct line_log_data *r;
709         r = lookup_decoration(&revs->line_log_data, &commit->object);
710         if (!r)
711                 return;
712         free_line_log_data(r);
713         add_decoration(&revs->line_log_data, &commit->object, NULL);
714 }
715
716 static struct line_log_data *lookup_line_range(struct rev_info *revs,
717                                                struct commit *commit)
718 {
719         struct line_log_data *ret = NULL;
720         struct line_log_data *d;
721
722         ret = lookup_decoration(&revs->line_log_data, &commit->object);
723
724         for (d = ret; d; d = d->next)
725                 range_set_check_invariants(&d->ranges);
726
727         return ret;
728 }
729
730 void line_log_init(struct rev_info *rev, const char *prefix, struct string_list *args)
731 {
732         struct commit *commit = NULL;
733         struct line_log_data *range;
734
735         commit = check_single_commit(rev);
736         range = parse_lines(commit, prefix, args);
737         add_line_range(rev, commit, range);
738
739         if (!rev->diffopt.detect_rename) {
740                 int i, count = 0;
741                 struct line_log_data *r = range;
742                 const char **paths;
743                 while (r) {
744                         count++;
745                         r = r->next;
746                 }
747                 paths = xmalloc((count+1)*sizeof(char *));
748                 r = range;
749                 for (i = 0; i < count; i++) {
750                         paths[i] = xstrdup(r->path);
751                         r = r->next;
752                 }
753                 paths[count] = NULL;
754                 init_pathspec(&rev->diffopt.pathspec, paths);
755                 free(paths);
756         }
757 }
758
759 static void load_tree_desc(struct tree_desc *desc, void **tree,
760                            const unsigned char *sha1)
761 {
762         unsigned long size;
763         *tree = read_object_with_reference(sha1, tree_type, &size, NULL);
764         if (!*tree)
765                 die("Unable to read tree (%s)", sha1_to_hex(sha1));
766         init_tree_desc(desc, *tree, size);
767 }
768
769 static int count_parents(struct commit *commit)
770 {
771         struct commit_list *parents = commit->parents;
772         int count = 0;
773         while (parents) {
774                 count++;
775                 parents = parents->next;
776         }
777         return count;
778 }
779
780 static void move_diff_queue(struct diff_queue_struct *dst,
781                             struct diff_queue_struct *src)
782 {
783         assert(src != dst);
784         memcpy(dst, src, sizeof(struct diff_queue_struct));
785         DIFF_QUEUE_CLEAR(src);
786 }
787
788 static void filter_diffs_for_paths(struct line_log_data *range, int keep_deletions)
789 {
790         int i;
791         struct diff_queue_struct outq;
792         DIFF_QUEUE_CLEAR(&outq);
793
794         for (i = 0; i < diff_queued_diff.nr; i++) {
795                 struct diff_filepair *p = diff_queued_diff.queue[i];
796                 struct line_log_data *rg = NULL;
797
798                 if (!DIFF_FILE_VALID(p->two)) {
799                         if (keep_deletions)
800                                 diff_q(&outq, p);
801                         else
802                                 diff_free_filepair(p);
803                         continue;
804                 }
805                 for (rg = range; rg; rg = rg->next) {
806                         if (!strcmp(rg->path, p->two->path))
807                                 break;
808                 }
809                 if (rg)
810                         diff_q(&outq, p);
811                 else
812                         diff_free_filepair(p);
813         }
814         free(diff_queued_diff.queue);
815         diff_queued_diff = outq;
816 }
817
818 static inline int diff_might_be_rename(void)
819 {
820         int i;
821         for (i = 0; i < diff_queued_diff.nr; i++)
822                 if (!DIFF_FILE_VALID(diff_queued_diff.queue[i]->one)) {
823                         /* fprintf(stderr, "diff_might_be_rename found creation of: %s\n", */
824                         /*      diff_queued_diff.queue[i]->two->path); */
825                         return 1;
826                 }
827         return 0;
828 }
829
830 static void queue_diffs(struct line_log_data *range,
831                         struct diff_options *opt,
832                         struct diff_queue_struct *queue,
833                         struct commit *commit, struct commit *parent)
834 {
835         void *tree1 = NULL, *tree2 = NULL;
836         struct tree_desc desc1, desc2;
837
838         assert(commit);
839         load_tree_desc(&desc2, &tree2, commit->tree->object.sha1);
840         if (parent)
841                 load_tree_desc(&desc1, &tree1, parent->tree->object.sha1);
842         else
843                 init_tree_desc(&desc1, "", 0);
844
845         DIFF_QUEUE_CLEAR(&diff_queued_diff);
846         diff_tree(&desc1, &desc2, "", opt);
847         if (opt->detect_rename) {
848                 filter_diffs_for_paths(range, 1);
849                 if (diff_might_be_rename())
850                         diffcore_std(opt);
851                 filter_diffs_for_paths(range, 0);
852         }
853         move_diff_queue(queue, &diff_queued_diff);
854
855         if (tree1)
856                 free(tree1);
857         if (tree2)
858                 free(tree2);
859 }
860
861 static char *get_nth_line(long line, unsigned long *ends, void *data)
862 {
863         if (line == 0)
864                 return (char *)data;
865         else
866                 return (char *)data + ends[line] + 1;
867 }
868
869 static void print_line(const char *prefix, char first,
870                        long line, unsigned long *ends, void *data,
871                        const char *color, const char *reset)
872 {
873         char *begin = get_nth_line(line, ends, data);
874         char *end = get_nth_line(line+1, ends, data);
875         int had_nl = 0;
876
877         if (end > begin && end[-1] == '\n') {
878                 end--;
879                 had_nl = 1;
880         }
881
882         fputs(prefix, stdout);
883         fputs(color, stdout);
884         putchar(first);
885         fwrite(begin, 1, end-begin, stdout);
886         fputs(reset, stdout);
887         putchar('\n');
888         if (!had_nl)
889                 fputs("\\ No newline at end of file\n", stdout);
890 }
891
892 static char *output_prefix(struct diff_options *opt)
893 {
894         char *prefix = "";
895
896         if (opt->output_prefix) {
897                 struct strbuf *sb = opt->output_prefix(opt, opt->output_prefix_data);
898                 prefix = sb->buf;
899         }
900
901         return prefix;
902 }
903
904 static void dump_diff_hacky_one(struct rev_info *rev, struct line_log_data *range)
905 {
906         int i, j = 0;
907         long p_lines, t_lines;
908         unsigned long *p_ends = NULL, *t_ends = NULL;
909         struct diff_filepair *pair = range->pair;
910         struct diff_ranges *diff = &range->diff;
911
912         struct diff_options *opt = &rev->diffopt;
913         char *prefix = output_prefix(opt);
914         const char *c_reset = diff_get_color(opt->use_color, DIFF_RESET);
915         const char *c_frag = diff_get_color(opt->use_color, DIFF_FRAGINFO);
916         const char *c_meta = diff_get_color(opt->use_color, DIFF_METAINFO);
917         const char *c_old = diff_get_color(opt->use_color, DIFF_FILE_OLD);
918         const char *c_new = diff_get_color(opt->use_color, DIFF_FILE_NEW);
919         const char *c_plain = diff_get_color(opt->use_color, DIFF_PLAIN);
920
921         if (!pair || !diff)
922                 return;
923
924         if (pair->one->sha1_valid)
925                 fill_line_ends(pair->one, &p_lines, &p_ends);
926         fill_line_ends(pair->two, &t_lines, &t_ends);
927
928         printf("%s%sdiff --git a/%s b/%s%s\n", prefix, c_meta, pair->one->path, pair->two->path, c_reset);
929         printf("%s%s--- %s%s%s\n", prefix, c_meta,
930                pair->one->sha1_valid ? "a/" : "",
931                pair->one->sha1_valid ? pair->one->path : "/dev/null",
932                c_reset);
933         printf("%s%s+++ b/%s%s\n", prefix, c_meta, pair->two->path, c_reset);
934         for (i = 0; i < range->ranges.nr; i++) {
935                 long p_start, p_end;
936                 long t_start = range->ranges.ranges[i].start;
937                 long t_end = range->ranges.ranges[i].end;
938                 long t_cur = t_start;
939                 int j_last;
940
941                 while (j < diff->target.nr && diff->target.ranges[j].end < t_start)
942                         j++;
943                 if (j == diff->target.nr || diff->target.ranges[j].start > t_end)
944                         continue;
945
946                 /* Scan ahead to determine the last diff that falls in this range */
947                 j_last = j;
948                 while (j_last < diff->target.nr && diff->target.ranges[j_last].start < t_end)
949                         j_last++;
950                 if (j_last > j)
951                         j_last--;
952
953                 /*
954                  * Compute parent hunk headers: we know that the diff
955                  * has the correct line numbers (but not all hunks).
956                  * So it suffices to shift the start/end according to
957                  * the line numbers of the first/last hunk(s) that
958                  * fall in this range.
959                  */
960                 if (t_start < diff->target.ranges[j].start)
961                         p_start = diff->parent.ranges[j].start - (diff->target.ranges[j].start-t_start);
962                 else
963                         p_start = diff->parent.ranges[j].start;
964                 if (t_end > diff->target.ranges[j_last].end)
965                         p_end = diff->parent.ranges[j_last].end + (t_end-diff->target.ranges[j_last].end);
966                 else
967                         p_end = diff->parent.ranges[j_last].end;
968
969                 if (!p_start && !p_end) {
970                         p_start = -1;
971                         p_end = -1;
972                 }
973
974                 /* Now output a diff hunk for this range */
975                 printf("%s%s@@ -%ld,%ld +%ld,%ld @@%s\n",
976                        prefix, c_frag,
977                        p_start+1, p_end-p_start, t_start+1, t_end-t_start,
978                        c_reset);
979                 while (j < diff->target.nr && diff->target.ranges[j].start < t_end) {
980                         int k;
981                         for (; t_cur < diff->target.ranges[j].start; t_cur++)
982                                 print_line(prefix, ' ', t_cur, t_ends, pair->two->data,
983                                            c_plain, c_reset);
984                         for (k = diff->parent.ranges[j].start; k < diff->parent.ranges[j].end; k++)
985                                 print_line(prefix, '-', k, p_ends, pair->one->data,
986                                            c_old, c_reset);
987                         for (; t_cur < diff->target.ranges[j].end && t_cur < t_end; t_cur++)
988                                 print_line(prefix, '+', t_cur, t_ends, pair->two->data,
989                                            c_new, c_reset);
990                         j++;
991                 }
992                 for (; t_cur < t_end; t_cur++)
993                         print_line(prefix, ' ', t_cur, t_ends, pair->two->data,
994                                    c_plain, c_reset);
995         }
996
997         free(p_ends);
998         free(t_ends);
999 }
1000
1001 /*
1002  * NEEDSWORK: manually building a diff here is not the Right
1003  * Thing(tm).  log -L should be built into the diff pipeline.
1004  */
1005 static void dump_diff_hacky(struct rev_info *rev, struct line_log_data *range)
1006 {
1007         puts(output_prefix(&rev->diffopt));
1008         while (range) {
1009                 dump_diff_hacky_one(rev, range);
1010                 range = range->next;
1011         }
1012 }
1013
1014 /*
1015  * Unlike most other functions, this destructively operates on
1016  * 'range'.
1017  */
1018 static int process_diff_filepair(struct rev_info *rev,
1019                                  struct diff_filepair *pair,
1020                                  struct line_log_data *range,
1021                                  struct diff_ranges **diff_out)
1022 {
1023         struct line_log_data *rg = range;
1024         struct range_set tmp;
1025         struct diff_ranges diff;
1026         mmfile_t file_parent, file_target;
1027
1028         assert(pair->two->path);
1029         while (rg) {
1030                 assert(rg->path);
1031                 if (!strcmp(rg->path, pair->two->path))
1032                         break;
1033                 rg = rg->next;
1034         }
1035
1036         if (!rg)
1037                 return 0;
1038         if (rg->ranges.nr == 0)
1039                 return 0;
1040
1041         assert(pair->two->sha1_valid);
1042         diff_populate_filespec(pair->two, 0);
1043         file_target.ptr = pair->two->data;
1044         file_target.size = pair->two->size;
1045
1046         if (pair->one->sha1_valid) {
1047                 diff_populate_filespec(pair->one, 0);
1048                 file_parent.ptr = pair->one->data;
1049                 file_parent.size = pair->one->size;
1050         } else {
1051                 file_parent.ptr = "";
1052                 file_parent.size = 0;
1053         }
1054
1055         diff_ranges_init(&diff);
1056         collect_diff(&file_parent, &file_target, &diff);
1057
1058         /* NEEDSWORK should apply some heuristics to prevent mismatches */
1059         free(rg->path);
1060         rg->path = xstrdup(pair->one->path);
1061
1062         range_set_init(&tmp, 0);
1063         range_set_map_across_diff(&tmp, &rg->ranges, &diff, diff_out);
1064         range_set_release(&rg->ranges);
1065         range_set_move(&rg->ranges, &tmp);
1066
1067         diff_ranges_release(&diff);
1068
1069         return ((*diff_out)->parent.nr > 0);
1070 }
1071
1072 static struct diff_filepair *diff_filepair_dup(struct diff_filepair *pair)
1073 {
1074         struct diff_filepair *new = xmalloc(sizeof(struct diff_filepair));
1075         new->one = pair->one;
1076         new->two = pair->two;
1077         new->one->count++;
1078         new->two->count++;
1079         return new;
1080 }
1081
1082 static void free_diffqueues(int n, struct diff_queue_struct *dq)
1083 {
1084         int i, j;
1085         for (i = 0; i < n; i++)
1086                 for (j = 0; j < dq[i].nr; j++)
1087                         diff_free_filepair(dq[i].queue[j]);
1088         free(dq);
1089 }
1090
1091 static int process_all_files(struct line_log_data **range_out,
1092                              struct rev_info *rev,
1093                              struct diff_queue_struct *queue,
1094                              struct line_log_data *range)
1095 {
1096         int i, changed = 0;
1097
1098         *range_out = line_log_data_copy(range);
1099
1100         for (i = 0; i < queue->nr; i++) {
1101                 struct diff_ranges *pairdiff = NULL;
1102                 struct diff_filepair *pair = queue->queue[i];
1103                 if (process_diff_filepair(rev, pair, *range_out, &pairdiff)) {
1104                         /*
1105                          * Store away the diff for later output.  We
1106                          * tuck it in the ranges we got as _input_,
1107                          * since that's the commit that caused the
1108                          * diff.
1109                          *
1110                          * NEEDSWORK not enough when we get around to
1111                          * doing something interesting with merges;
1112                          * currently each invocation on a merge parent
1113                          * trashes the previous one's diff.
1114                          *
1115                          * NEEDSWORK tramples over data structures not owned here
1116                          */
1117                         struct line_log_data *rg = range;
1118                         changed++;
1119                         while (rg && strcmp(rg->path, pair->two->path))
1120                                 rg = rg->next;
1121                         assert(rg);
1122                         rg->pair = diff_filepair_dup(queue->queue[i]);
1123                         memcpy(&rg->diff, pairdiff, sizeof(struct diff_ranges));
1124                 }
1125         }
1126
1127         return changed;
1128 }
1129
1130 int line_log_print(struct rev_info *rev, struct commit *commit)
1131 {
1132         struct line_log_data *range = lookup_line_range(rev, commit);
1133
1134         show_log(rev);
1135         dump_diff_hacky(rev, range);
1136         return 1;
1137 }
1138
1139 static int process_ranges_ordinary_commit(struct rev_info *rev, struct commit *commit,
1140                                           struct line_log_data *range)
1141 {
1142         struct commit *parent = NULL;
1143         struct diff_queue_struct queue;
1144         struct line_log_data *parent_range;
1145         int changed;
1146
1147         if (commit->parents)
1148                 parent = commit->parents->item;
1149
1150         queue_diffs(range, &rev->diffopt, &queue, commit, parent);
1151         changed = process_all_files(&parent_range, rev, &queue, range);
1152         if (parent)
1153                 add_line_range(rev, parent, parent_range);
1154         return changed;
1155 }
1156
1157 static int process_ranges_merge_commit(struct rev_info *rev, struct commit *commit,
1158                                        struct line_log_data *range)
1159 {
1160         struct diff_queue_struct *diffqueues;
1161         struct line_log_data **cand;
1162         struct commit **parents;
1163         struct commit_list *p;
1164         int i;
1165         int nparents = count_parents(commit);
1166
1167         diffqueues = xmalloc(nparents * sizeof(*diffqueues));
1168         cand = xmalloc(nparents * sizeof(*cand));
1169         parents = xmalloc(nparents * sizeof(*parents));
1170
1171         p = commit->parents;
1172         for (i = 0; i < nparents; i++) {
1173                 parents[i] = p->item;
1174                 p = p->next;
1175                 queue_diffs(range, &rev->diffopt, &diffqueues[i], commit, parents[i]);
1176         }
1177
1178         for (i = 0; i < nparents; i++) {
1179                 int changed;
1180                 cand[i] = NULL;
1181                 changed = process_all_files(&cand[i], rev, &diffqueues[i], range);
1182                 if (!changed) {
1183                         /*
1184                          * This parent can take all the blame, so we
1185                          * don't follow any other path in history
1186                          */
1187                         add_line_range(rev, parents[i], cand[i]);
1188                         clear_commit_line_range(rev, commit);
1189                         commit->parents = xmalloc(sizeof(struct commit_list));
1190                         commit->parents->item = parents[i];
1191                         commit->parents->next = NULL;
1192                         free(parents);
1193                         free(cand);
1194                         free_diffqueues(nparents, diffqueues);
1195                         /* NEEDSWORK leaking like a sieve */
1196                         return 0;
1197                 }
1198         }
1199
1200         /*
1201          * No single parent took the blame.  We add the candidates
1202          * from the above loop to the parents.
1203          */
1204         for (i = 0; i < nparents; i++) {
1205                 add_line_range(rev, parents[i], cand[i]);
1206         }
1207
1208         clear_commit_line_range(rev, commit);
1209         free(parents);
1210         free(cand);
1211         free_diffqueues(nparents, diffqueues);
1212         return 1;
1213
1214         /* NEEDSWORK evil merge detection stuff */
1215         /* NEEDSWORK leaking like a sieve */
1216 }
1217
1218 static int process_ranges_arbitrary_commit(struct rev_info *rev, struct commit *commit)
1219 {
1220         struct line_log_data *range = lookup_line_range(rev, commit);
1221         int changed = 0;
1222
1223         if (range) {
1224                 if (!commit->parents || !commit->parents->next)
1225                         changed = process_ranges_ordinary_commit(rev, commit, range);
1226                 else
1227                         changed = process_ranges_merge_commit(rev, commit, range);
1228         }
1229
1230         if (!changed)
1231                 commit->object.flags |= TREESAME;
1232
1233         return changed;
1234 }
1235
1236 static enum rewrite_result line_log_rewrite_one(struct rev_info *rev, struct commit **pp)
1237 {
1238         for (;;) {
1239                 struct commit *p = *pp;
1240                 if (p->parents && p->parents->next)
1241                         return rewrite_one_ok;
1242                 if (p->object.flags & UNINTERESTING)
1243                         return rewrite_one_ok;
1244                 if (!(p->object.flags & TREESAME))
1245                         return rewrite_one_ok;
1246                 if (!p->parents)
1247                         return rewrite_one_noparents;
1248                 *pp = p->parents->item;
1249         }
1250 }
1251
1252 int line_log_filter(struct rev_info *rev)
1253 {
1254         struct commit *commit;
1255         struct commit_list *list = rev->commits;
1256         struct commit_list *out = NULL, **pp = &out;
1257
1258         while (list) {
1259                 struct commit_list *to_free = NULL;
1260                 commit = list->item;
1261                 if (process_ranges_arbitrary_commit(rev, commit)) {
1262                         *pp = list;
1263                         pp = &list->next;
1264                 } else
1265                         to_free = list;
1266                 list = list->next;
1267                 free(to_free);
1268         }
1269         *pp = NULL;
1270
1271         for (list = out; list; list = list->next)
1272                 rewrite_parents(rev, list->item, line_log_rewrite_one);
1273
1274         rev->commits = out;
1275
1276         return 0;
1277 }