git.oblomov.eu Git - git/blob - diffcore-delta.c

   1 #include "cache.h"
   2 #include "diff.h"
   3 #include "diffcore.h"
   4
   5 /*
   6  * Idea here is very simple.
   7  *
   8  * Almost all data we are interested in are text, but sometimes we have
   9  * to deal with binary data.  So we cut them into chunks delimited by
  10  * LF byte, or 64-byte sequence, whichever comes first, and hash them.
  11  *
  12  * For those chunks, if the source buffer has more instances of it
  13  * than the destination buffer, that means the difference are the
  14  * number of bytes not copied from source to destination.  If the
  15  * counts are the same, everything was copied from source to
  16  * destination.  If the destination has more, everything was copied,
  17  * and destination added more.
  18  *
  19  * We are doing an approximation so we do not really have to waste
  20  * memory by actually storing the sequence.  We just hash them into
  21  * somewhere around 2^16 hashbuckets and count the occurrences.
  22  */
  23
  24 /* Wild guess at the initial hash size */
  25 #define INITIAL_HASH_SIZE 9
  26
  27 /* We leave more room in smaller hash but do not let it
  28  * grow to have unused hole too much.
  29  */
  30 #define INITIAL_FREE(sz_log2) ((1<<(sz_log2))*(sz_log2-3)/(sz_log2))
  31
  32 /* A prime rather carefully chosen between 2^16..2^17, so that
  33  * HASHBASE < INITIAL_FREE(17).  We want to keep the maximum hashtable
  34  * size under the current 2<<17 maximum, which can hold this many
  35  * different values before overflowing to hashtable of size 2<<18.
  36  */
  37 #define HASHBASE 107927
  38
  39 struct spanhash {
  40         unsigned int hashval;
  41         unsigned int cnt;
  42 };
  43 struct spanhash_top {
  44         int alloc_log2;
  45         int free;
  46         struct spanhash data[FLEX_ARRAY];
  47 };
  48
  49 static struct spanhash_top *spanhash_rehash(struct spanhash_top *orig)
  50 {
  51         struct spanhash_top *new_spanhash;
  52         int i;
  53         int osz = 1 << orig->alloc_log2;
  54         int sz = osz << 1;
  55
  56         new_spanhash = xmalloc(st_add(sizeof(*orig),
  57                              st_mult(sizeof(struct spanhash), sz)));
  58         new_spanhash->alloc_log2 = orig->alloc_log2 + 1;
  59         new_spanhash->free = INITIAL_FREE(new_spanhash->alloc_log2);
  60         memset(new_spanhash->data, 0, sizeof(struct spanhash) * sz);
  61         for (i = 0; i < osz; i++) {
  62                 struct spanhash *o = &(orig->data[i]);
  63                 int bucket;
  64                 if (!o->cnt)
  65                         continue;
  66                 bucket = o->hashval & (sz - 1);
  67                 while (1) {
  68                         struct spanhash *h = &(new_spanhash->data[bucket++]);
  69                         if (!h->cnt) {
  70                                 h->hashval = o->hashval;
  71                                 h->cnt = o->cnt;
  72                                 new_spanhash->free--;
  73                                 break;
  74                         }
  75                         if (sz <= bucket)
  76                                 bucket = 0;
  77                 }
  78         }
  79         free(orig);
  80         return new_spanhash;
  81 }
  82
  83 static struct spanhash_top *add_spanhash(struct spanhash_top *top,
  84                                          unsigned int hashval, int cnt)
  85 {
  86         int bucket, lim;
  87         struct spanhash *h;
  88
  89         lim = (1 << top->alloc_log2);
  90         bucket = hashval & (lim - 1);
  91         while (1) {
  92                 h = &(top->data[bucket++]);
  93                 if (!h->cnt) {
  94                         h->hashval = hashval;
  95                         h->cnt = cnt;
  96                         top->free--;
  97                         if (top->free < 0)
  98                                 return spanhash_rehash(top);
  99                         return top;
 100                 }
 101                 if (h->hashval == hashval) {
 102                         h->cnt += cnt;
 103                         return top;
 104                 }
 105                 if (lim <= bucket)
 106                         bucket = 0;
 107         }
 108 }
 109
 110 static int spanhash_cmp(const void *a_, const void *b_)
 111 {
 112         const struct spanhash *a = a_;
 113         const struct spanhash *b = b_;
 114
 115         /* A count of zero compares at the end.. */
 116         if (!a->cnt)
 117                 return !b->cnt ? 0 : 1;
 118         if (!b->cnt)
 119                 return -1;
 120         return a->hashval < b->hashval ? -1 :
 121                 a->hashval > b->hashval ? 1 : 0;
 122 }
 123
 124 static struct spanhash_top *hash_chars(struct repository *r,
 125                                        struct diff_filespec *one)
 126 {
 127         int i, n;
 128         unsigned int accum1, accum2, hashval;
 129         struct spanhash_top *hash;
 130         unsigned char *buf = one->data;
 131         unsigned int sz = one->size;
 132         int is_text = !diff_filespec_is_binary(r, one);
 133
 134         i = INITIAL_HASH_SIZE;
 135         hash = xmalloc(st_add(sizeof(*hash),
 136                               st_mult(sizeof(struct spanhash), 1<<i)));
 137         hash->alloc_log2 = i;
 138         hash->free = INITIAL_FREE(i);
 139         memset(hash->data, 0, sizeof(struct spanhash) * (1<<i));
 140
 141         n = 0;
 142         accum1 = accum2 = 0;
 143         while (sz) {
 144                 unsigned int c = *buf++;
 145                 unsigned int old_1 = accum1;
 146                 sz--;
 147
 148                 /* Ignore CR in CRLF sequence if text */
 149                 if (is_text && c == '\r' && sz && *buf == '\n')
 150                         continue;
 151
 152                 accum1 = (accum1 << 7) ^ (accum2 >> 25);
 153                 accum2 = (accum2 << 7) ^ (old_1 >> 25);
 154                 accum1 += c;
 155                 if (++n < 64 && c != '\n')
 156                         continue;
 157                 hashval = (accum1 + accum2 * 0x61) % HASHBASE;
 158                 hash = add_spanhash(hash, hashval, n);
 159                 n = 0;
 160                 accum1 = accum2 = 0;
 161         }
 162         QSORT(hash->data, 1ul << hash->alloc_log2, spanhash_cmp);
 163         return hash;
 164 }
 165
 166 int diffcore_count_changes(struct repository *r,
 167                            struct diff_filespec *src,
 168                            struct diff_filespec *dst,
 169                            void **src_count_p,
 170                            void **dst_count_p,
 171                            unsigned long *src_copied,
 172                            unsigned long *literal_added)
 173 {
 174         struct spanhash *s, *d;
 175         struct spanhash_top *src_count, *dst_count;
 176         unsigned long sc, la;
 177
 178         src_count = dst_count = NULL;
 179         if (src_count_p)
 180                 src_count = *src_count_p;
 181         if (!src_count) {
 182                 src_count = hash_chars(r, src);
 183                 if (src_count_p)
 184                         *src_count_p = src_count;
 185         }
 186         if (dst_count_p)
 187                 dst_count = *dst_count_p;
 188         if (!dst_count) {
 189                 dst_count = hash_chars(r, dst);
 190                 if (dst_count_p)
 191                         *dst_count_p = dst_count;
 192         }
 193         sc = la = 0;
 194
 195         s = src_count->data;
 196         d = dst_count->data;
 197         for (;;) {
 198                 unsigned dst_cnt, src_cnt;
 199                 if (!s->cnt)
 200                         break; /* we checked all in src */
 201                 while (d->cnt) {
 202                         if (d->hashval >= s->hashval)
 203                                 break;
 204                         la += d->cnt;
 205                         d++;
 206                 }
 207                 src_cnt = s->cnt;
 208                 dst_cnt = 0;
 209                 if (d->cnt && d->hashval == s->hashval) {
 210                         dst_cnt = d->cnt;
 211                         d++;
 212                 }
 213                 if (src_cnt < dst_cnt) {
 214                         la += dst_cnt - src_cnt;
 215                         sc += src_cnt;
 216                 }
 217                 else
 218                         sc += dst_cnt;
 219                 s++;
 220         }
 221         while (d->cnt) {
 222                 la += d->cnt;
 223                 d++;
 224         }
 225
 226         if (!src_count_p)
 227                 free(src_count);
 228         if (!dst_count_p)
 229                 free(dst_count);
 230         *src_copied = sc;
 231         *literal_added = la;
 232         return 0;
 233 }