Merge branch 'jc/rename'
authorJunio C Hamano <gitster@pobox.com>
Wed, 9 Apr 2008 07:46:40 +0000 (00:46 -0700)
committerJunio C Hamano <gitster@pobox.com>
Wed, 9 Apr 2008 08:09:12 +0000 (01:09 -0700)
* 'jc/rename' (early part):
Optimize rename detection for a huge diff

1  2 
diffcore-rename.c
diff --combined diffcore-rename.c
index 31941bcbbf457fe02097f8c46d25f1f9fc7c4f9f,5974362d3b38a1445f02c35130f52c0661861b5e..1369a5ec45349c1ef04e4df2d7d7d4dd28cd3596
@@@ -112,8 -112,8 +112,8 @@@ static int basename_same(struct diff_fi
  struct diff_score {
        int src; /* index in rename_src */
        int dst; /* index in rename_dst */
-       int score;
-       int name_score;
+       unsigned short score;
+       short name_score;
  };
  
  static int estimate_similarity(struct diff_filespec *src,
@@@ -223,6 -223,12 +223,12 @@@ static int score_compare(const void *a_
  {
        const struct diff_score *a = a_, *b = b_;
  
+       /* sink the unused ones to the bottom */
+       if (a->dst < 0)
+               return (0 <= b->dst);
+       else if (b->dst < 0)
+               return -1;
        if (a->score == b->score)
                return b->name_score - a->name_score;
  
@@@ -387,6 -393,22 +393,22 @@@ static int find_exact_renames(void
        return i;
  }
  
+ #define NUM_CANDIDATE_PER_DST 4
+ static void record_if_better(struct diff_score m[], struct diff_score *o)
+ {
+       int i, worst;
+       /* find the worst one */
+       worst = 0;
+       for (i = 1; i < NUM_CANDIDATE_PER_DST; i++)
+               if (score_compare(&m[i], &m[worst]) > 0)
+                       worst = i;
+       /* is it better than the worst one? */
+       if (score_compare(&m[worst], o) > 0)
+               m[worst] = *o;
+ }
  void diffcore_rename(struct diff_options *options)
  {
        int detect_rename = options->detect_rename;
         */
        if (rename_limit <= 0 || rename_limit > 32767)
                rename_limit = 32767;
 -      if (num_create > rename_limit && num_src > rename_limit)
 -              goto cleanup;
 -      if (num_create * num_src > rename_limit * rename_limit)
 +      if ((num_create > rename_limit && num_src > rename_limit) ||
 +          (num_create * num_src > rename_limit * rename_limit)) {
 +              warning("too many files, skipping inexact rename detection");
                goto cleanup;
 +      }
  
-       mx = xmalloc(sizeof(*mx) * num_create * num_src);
+       mx = xcalloc(num_create * NUM_CANDIDATE_PER_DST, sizeof(*mx));
        for (dst_cnt = i = 0; i < rename_dst_nr; i++) {
-               int base = dst_cnt * num_src;
                struct diff_filespec *two = rename_dst[i].two;
+               struct diff_score *m;
                if (rename_dst[i].pair)
                        continue; /* dealt with exact match already. */
+               m = &mx[dst_cnt * NUM_CANDIDATE_PER_DST];
+               for (j = 0; j < NUM_CANDIDATE_PER_DST; j++)
+                       m[j].dst = -1;
                for (j = 0; j < rename_src_nr; j++) {
                        struct diff_filespec *one = rename_src[j].one;
-                       struct diff_score *m = &mx[base+j];
-                       m->src = j;
-                       m->dst = i;
-                       m->score = estimate_similarity(one, two,
-                                                      minimum_score);
-                       m->name_score = basename_same(one, two);
+                       struct diff_score this_src;
+                       this_src.score = estimate_similarity(one, two,
+                                                            minimum_score);
+                       this_src.name_score = basename_same(one, two);
+                       this_src.dst = i;
+                       this_src.src = j;
+                       record_if_better(m, &this_src);
                        diff_free_filespec_blob(one);
                }
                /* We do not need the text anymore */
                diff_free_filespec_blob(two);
                dst_cnt++;
        }
        /* cost matrix sorted by most to least similar pair */
-       qsort(mx, num_create * num_src, sizeof(*mx), score_compare);
-       for (i = 0; i < num_create * num_src; i++) {
-               struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
-               struct diff_filespec *src;
+       qsort(mx, dst_cnt * NUM_CANDIDATE_PER_DST, sizeof(*mx), score_compare);
+       for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
+               struct diff_rename_dst *dst;
+               if ((mx[i].dst < 0) ||
+                   (mx[i].score < minimum_score))
+                       break; /* there is no more usable pair. */
+               dst = &rename_dst[mx[i].dst];
                if (dst->pair)
                        continue; /* already done, either exact or fuzzy. */
-               if (mx[i].score < minimum_score)
-                       break; /* there is no more usable pair. */
-               src = rename_src[mx[i].src].one;
-               if (src->rename_used)
+               if (rename_src[mx[i].src].one->rename_used)
                        continue;
                record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
                rename_count++;
        }
-       for (i = 0; i < num_create * num_src; i++) {
-               struct diff_rename_dst *dst = &rename_dst[mx[i].dst];
+       for (i = 0; i < dst_cnt * NUM_CANDIDATE_PER_DST; i++) {
+               struct diff_rename_dst *dst;
+               if ((mx[i].dst < 0) ||
+                   (mx[i].score < minimum_score))
+                       break; /* there is no more usable pair. */
+               dst = &rename_dst[mx[i].dst];
                if (dst->pair)
                        continue; /* already done, either exact or fuzzy. */
-               if (mx[i].score < minimum_score)
-                       break; /* there is no more usable pair. */
                record_rename_pair(mx[i].dst, mx[i].src, mx[i].score);
                rename_count++;
        }