split-index.con commit branch: add a --copy (-c) option to go with --move (-m) (52d59cc)
   1#include "cache.h"
   2#include "split-index.h"
   3#include "ewah/ewok.h"
   4
   5struct split_index *init_split_index(struct index_state *istate)
   6{
   7        if (!istate->split_index) {
   8                istate->split_index = xcalloc(1, sizeof(*istate->split_index));
   9                istate->split_index->refcount = 1;
  10        }
  11        return istate->split_index;
  12}
  13
  14int read_link_extension(struct index_state *istate,
  15                         const void *data_, unsigned long sz)
  16{
  17        const unsigned char *data = data_;
  18        struct split_index *si;
  19        int ret;
  20
  21        if (sz < 20)
  22                return error("corrupt link extension (too short)");
  23        si = init_split_index(istate);
  24        hashcpy(si->base_sha1, data);
  25        data += 20;
  26        sz -= 20;
  27        if (!sz)
  28                return 0;
  29        si->delete_bitmap = ewah_new();
  30        ret = ewah_read_mmap(si->delete_bitmap, data, sz);
  31        if (ret < 0)
  32                return error("corrupt delete bitmap in link extension");
  33        data += ret;
  34        sz -= ret;
  35        si->replace_bitmap = ewah_new();
  36        ret = ewah_read_mmap(si->replace_bitmap, data, sz);
  37        if (ret < 0)
  38                return error("corrupt replace bitmap in link extension");
  39        if (ret != sz)
  40                return error("garbage at the end of link extension");
  41        return 0;
  42}
  43
  44int write_link_extension(struct strbuf *sb,
  45                         struct index_state *istate)
  46{
  47        struct split_index *si = istate->split_index;
  48        strbuf_add(sb, si->base_sha1, 20);
  49        if (!si->delete_bitmap && !si->replace_bitmap)
  50                return 0;
  51        ewah_serialize_strbuf(si->delete_bitmap, sb);
  52        ewah_serialize_strbuf(si->replace_bitmap, sb);
  53        return 0;
  54}
  55
  56static void mark_base_index_entries(struct index_state *base)
  57{
  58        int i;
  59        /*
  60         * To keep track of the shared entries between
  61         * istate->base->cache[] and istate->cache[], base entry
  62         * position is stored in each base entry. All positions start
  63         * from 1 instead of 0, which is reserved to say "this is a new
  64         * entry".
  65         */
  66        for (i = 0; i < base->cache_nr; i++)
  67                base->cache[i]->index = i + 1;
  68}
  69
  70void move_cache_to_base_index(struct index_state *istate)
  71{
  72        struct split_index *si = istate->split_index;
  73        int i;
  74
  75        /*
  76         * If "si" is shared with another index_state (e.g. by
  77         * unpack-trees code), we will need to duplicate split_index
  78         * struct. It's not happening now though, luckily.
  79         */
  80        assert(si->refcount <= 1);
  81
  82        unshare_split_index(istate, 0);
  83        if (si->base) {
  84                discard_index(si->base);
  85                free(si->base);
  86        }
  87        si->base = xcalloc(1, sizeof(*si->base));
  88        si->base->version = istate->version;
  89        /* zero timestamp disables racy test in ce_write_index() */
  90        si->base->timestamp = istate->timestamp;
  91        ALLOC_GROW(si->base->cache, istate->cache_nr, si->base->cache_alloc);
  92        si->base->cache_nr = istate->cache_nr;
  93        COPY_ARRAY(si->base->cache, istate->cache, istate->cache_nr);
  94        mark_base_index_entries(si->base);
  95        for (i = 0; i < si->base->cache_nr; i++)
  96                si->base->cache[i]->ce_flags &= ~CE_UPDATE_IN_BASE;
  97}
  98
  99static void mark_entry_for_delete(size_t pos, void *data)
 100{
 101        struct index_state *istate = data;
 102        if (pos >= istate->cache_nr)
 103                die("position for delete %d exceeds base index size %d",
 104                    (int)pos, istate->cache_nr);
 105        istate->cache[pos]->ce_flags |= CE_REMOVE;
 106        istate->split_index->nr_deletions = 1;
 107}
 108
 109static void replace_entry(size_t pos, void *data)
 110{
 111        struct index_state *istate = data;
 112        struct split_index *si = istate->split_index;
 113        struct cache_entry *dst, *src;
 114
 115        if (pos >= istate->cache_nr)
 116                die("position for replacement %d exceeds base index size %d",
 117                    (int)pos, istate->cache_nr);
 118        if (si->nr_replacements >= si->saved_cache_nr)
 119                die("too many replacements (%d vs %d)",
 120                    si->nr_replacements, si->saved_cache_nr);
 121        dst = istate->cache[pos];
 122        if (dst->ce_flags & CE_REMOVE)
 123                die("entry %d is marked as both replaced and deleted",
 124                    (int)pos);
 125        src = si->saved_cache[si->nr_replacements];
 126        if (ce_namelen(src))
 127                die("corrupt link extension, entry %d should have "
 128                    "zero length name", (int)pos);
 129        src->index = pos + 1;
 130        src->ce_flags |= CE_UPDATE_IN_BASE;
 131        src->ce_namelen = dst->ce_namelen;
 132        copy_cache_entry(dst, src);
 133        free(src);
 134        si->nr_replacements++;
 135}
 136
 137void merge_base_index(struct index_state *istate)
 138{
 139        struct split_index *si = istate->split_index;
 140        unsigned int i;
 141
 142        mark_base_index_entries(si->base);
 143
 144        si->saved_cache     = istate->cache;
 145        si->saved_cache_nr  = istate->cache_nr;
 146        istate->cache_nr    = si->base->cache_nr;
 147        istate->cache       = NULL;
 148        istate->cache_alloc = 0;
 149        ALLOC_GROW(istate->cache, istate->cache_nr, istate->cache_alloc);
 150        COPY_ARRAY(istate->cache, si->base->cache, istate->cache_nr);
 151
 152        si->nr_deletions = 0;
 153        si->nr_replacements = 0;
 154        ewah_each_bit(si->replace_bitmap, replace_entry, istate);
 155        ewah_each_bit(si->delete_bitmap, mark_entry_for_delete, istate);
 156        if (si->nr_deletions)
 157                remove_marked_cache_entries(istate);
 158
 159        for (i = si->nr_replacements; i < si->saved_cache_nr; i++) {
 160                if (!ce_namelen(si->saved_cache[i]))
 161                        die("corrupt link extension, entry %d should "
 162                            "have non-zero length name", i);
 163                add_index_entry(istate, si->saved_cache[i],
 164                                ADD_CACHE_OK_TO_ADD |
 165                                ADD_CACHE_KEEP_CACHE_TREE |
 166                                /*
 167                                 * we may have to replay what
 168                                 * merge-recursive.c:update_stages()
 169                                 * does, which has this flag on
 170                                 */
 171                                ADD_CACHE_SKIP_DFCHECK);
 172                si->saved_cache[i] = NULL;
 173        }
 174
 175        ewah_free(si->delete_bitmap);
 176        ewah_free(si->replace_bitmap);
 177        free(si->saved_cache);
 178        si->delete_bitmap  = NULL;
 179        si->replace_bitmap = NULL;
 180        si->saved_cache    = NULL;
 181        si->saved_cache_nr = 0;
 182}
 183
 184void prepare_to_write_split_index(struct index_state *istate)
 185{
 186        struct split_index *si = init_split_index(istate);
 187        struct cache_entry **entries = NULL, *ce;
 188        int i, nr_entries = 0, nr_alloc = 0;
 189
 190        si->delete_bitmap = ewah_new();
 191        si->replace_bitmap = ewah_new();
 192
 193        if (si->base) {
 194                /* Go through istate->cache[] and mark CE_MATCHED to
 195                 * entry with positive index. We'll go through
 196                 * base->cache[] later to delete all entries in base
 197                 * that are not marked with either CE_MATCHED or
 198                 * CE_UPDATE_IN_BASE. If istate->cache[i] is a
 199                 * duplicate, deduplicate it.
 200                 */
 201                for (i = 0; i < istate->cache_nr; i++) {
 202                        struct cache_entry *base;
 203                        /* namelen is checked separately */
 204                        const unsigned int ondisk_flags =
 205                                CE_STAGEMASK | CE_VALID | CE_EXTENDED_FLAGS;
 206                        unsigned int ce_flags, base_flags, ret;
 207                        ce = istate->cache[i];
 208                        if (!ce->index)
 209                                continue;
 210                        if (ce->index > si->base->cache_nr) {
 211                                ce->index = 0;
 212                                continue;
 213                        }
 214                        ce->ce_flags |= CE_MATCHED; /* or "shared" */
 215                        base = si->base->cache[ce->index - 1];
 216                        if (ce == base)
 217                                continue;
 218                        if (ce->ce_namelen != base->ce_namelen ||
 219                            strcmp(ce->name, base->name)) {
 220                                ce->index = 0;
 221                                continue;
 222                        }
 223                        ce_flags = ce->ce_flags;
 224                        base_flags = base->ce_flags;
 225                        /* only on-disk flags matter */
 226                        ce->ce_flags   &= ondisk_flags;
 227                        base->ce_flags &= ondisk_flags;
 228                        ret = memcmp(&ce->ce_stat_data, &base->ce_stat_data,
 229                                     offsetof(struct cache_entry, name) -
 230                                     offsetof(struct cache_entry, ce_stat_data));
 231                        ce->ce_flags = ce_flags;
 232                        base->ce_flags = base_flags;
 233                        if (ret)
 234                                ce->ce_flags |= CE_UPDATE_IN_BASE;
 235                        free(base);
 236                        si->base->cache[ce->index - 1] = ce;
 237                }
 238                for (i = 0; i < si->base->cache_nr; i++) {
 239                        ce = si->base->cache[i];
 240                        if ((ce->ce_flags & CE_REMOVE) ||
 241                            !(ce->ce_flags & CE_MATCHED))
 242                                ewah_set(si->delete_bitmap, i);
 243                        else if (ce->ce_flags & CE_UPDATE_IN_BASE) {
 244                                ewah_set(si->replace_bitmap, i);
 245                                ce->ce_flags |= CE_STRIP_NAME;
 246                                ALLOC_GROW(entries, nr_entries+1, nr_alloc);
 247                                entries[nr_entries++] = ce;
 248                        }
 249                }
 250        }
 251
 252        for (i = 0; i < istate->cache_nr; i++) {
 253                ce = istate->cache[i];
 254                if ((!si->base || !ce->index) && !(ce->ce_flags & CE_REMOVE)) {
 255                        assert(!(ce->ce_flags & CE_STRIP_NAME));
 256                        ALLOC_GROW(entries, nr_entries+1, nr_alloc);
 257                        entries[nr_entries++] = ce;
 258                }
 259                ce->ce_flags &= ~CE_MATCHED;
 260        }
 261
 262        /*
 263         * take cache[] out temporarily, put entries[] in its place
 264         * for writing
 265         */
 266        si->saved_cache = istate->cache;
 267        si->saved_cache_nr = istate->cache_nr;
 268        istate->cache = entries;
 269        istate->cache_nr = nr_entries;
 270}
 271
 272void finish_writing_split_index(struct index_state *istate)
 273{
 274        struct split_index *si = init_split_index(istate);
 275
 276        ewah_free(si->delete_bitmap);
 277        ewah_free(si->replace_bitmap);
 278        si->delete_bitmap = NULL;
 279        si->replace_bitmap = NULL;
 280        free(istate->cache);
 281        istate->cache = si->saved_cache;
 282        istate->cache_nr = si->saved_cache_nr;
 283}
 284
 285void unshare_split_index(struct index_state *istate, int discard)
 286{
 287        struct split_index *si = istate->split_index;
 288        int i;
 289
 290        if (!si || !si->base)
 291                return;
 292
 293        for (i = 0; i < istate->cache_nr; i++) {
 294                struct cache_entry *ce = istate->cache[i];
 295                struct cache_entry *new = NULL;
 296
 297                if (!ce->index ||
 298                    ce->index > si->base->cache_nr ||
 299                    ce != si->base->cache[ce->index - 1])
 300                        continue;
 301
 302                if (!discard) {
 303                        int len = ce_namelen(ce);
 304                        new = xcalloc(1, cache_entry_size(len));
 305                        copy_cache_entry(new, ce);
 306                        memcpy(new->name, ce->name, len);
 307                        new->index = 0;
 308                }
 309                istate->cache[i] = new;
 310        }
 311}
 312
 313
 314void discard_split_index(struct index_state *istate)
 315{
 316        struct split_index *si = istate->split_index;
 317        if (!si)
 318                return;
 319        unshare_split_index(istate, 0);
 320        istate->split_index = NULL;
 321        si->refcount--;
 322        if (si->refcount)
 323                return;
 324        if (si->base) {
 325                discard_index(si->base);
 326                free(si->base);
 327        }
 328        free(si);
 329}
 330
 331void save_or_free_index_entry(struct index_state *istate, struct cache_entry *ce)
 332{
 333        if (ce->index &&
 334            istate->split_index &&
 335            istate->split_index->base &&
 336            ce->index <= istate->split_index->base->cache_nr &&
 337            ce == istate->split_index->base->cache[ce->index - 1])
 338                ce->ce_flags |= CE_REMOVE;
 339        else
 340                free(ce);
 341}
 342
 343void replace_index_entry_in_base(struct index_state *istate,
 344                                 struct cache_entry *old,
 345                                 struct cache_entry *new)
 346{
 347        if (old->index &&
 348            istate->split_index &&
 349            istate->split_index->base &&
 350            old->index <= istate->split_index->base->cache_nr) {
 351                new->index = old->index;
 352                if (old != istate->split_index->base->cache[new->index - 1])
 353                        free(istate->split_index->base->cache[new->index - 1]);
 354                istate->split_index->base->cache[new->index - 1] = new;
 355        }
 356}
 357
 358void add_split_index(struct index_state *istate)
 359{
 360        if (!istate->split_index) {
 361                init_split_index(istate);
 362                istate->cache_changed |= SPLIT_INDEX_ORDERED;
 363        }
 364}
 365
 366void remove_split_index(struct index_state *istate)
 367{
 368        if (!istate->split_index)
 369                return;
 370        discard_split_index(istate);
 371        istate->cache_changed |= SOMETHING_CHANGED;
 372}