notes.con commit Teach notes code to properly preserve non-notes in the notes tree (851c2b3)
   1#include "cache.h"
   2#include "notes.h"
   3#include "blob.h"
   4#include "tree.h"
   5#include "utf8.h"
   6#include "strbuf.h"
   7#include "tree-walk.h"
   8
   9/*
  10 * Use a non-balancing simple 16-tree structure with struct int_node as
  11 * internal nodes, and struct leaf_node as leaf nodes. Each int_node has a
  12 * 16-array of pointers to its children.
  13 * The bottom 2 bits of each pointer is used to identify the pointer type
  14 * - ptr & 3 == 0 - NULL pointer, assert(ptr == NULL)
  15 * - ptr & 3 == 1 - pointer to next internal node - cast to struct int_node *
  16 * - ptr & 3 == 2 - pointer to note entry - cast to struct leaf_node *
  17 * - ptr & 3 == 3 - pointer to subtree entry - cast to struct leaf_node *
  18 *
  19 * The root node is a statically allocated struct int_node.
  20 */
  21struct int_node {
  22        void *a[16];
  23};
  24
  25/*
  26 * Leaf nodes come in two variants, note entries and subtree entries,
  27 * distinguished by the LSb of the leaf node pointer (see above).
  28 * As a note entry, the key is the SHA1 of the referenced object, and the
  29 * value is the SHA1 of the note object.
  30 * As a subtree entry, the key is the prefix SHA1 (w/trailing NULs) of the
  31 * referenced object, using the last byte of the key to store the length of
  32 * the prefix. The value is the SHA1 of the tree object containing the notes
  33 * subtree.
  34 */
  35struct leaf_node {
  36        unsigned char key_sha1[20];
  37        unsigned char val_sha1[20];
  38};
  39
  40/*
  41 * A notes tree may contain entries that are not notes, and that do not follow
  42 * the naming conventions of notes. There are typically none/few of these, but
  43 * we still need to keep track of them. Keep a simple linked list sorted alpha-
  44 * betically on the non-note path. The list is populated when parsing tree
  45 * objects in load_subtree(), and the non-notes are correctly written back into
  46 * the tree objects produced by write_notes_tree().
  47 */
  48struct non_note {
  49        struct non_note *next; /* grounded (last->next == NULL) */
  50        char *path;
  51        unsigned int mode;
  52        unsigned char sha1[20];
  53};
  54
  55#define PTR_TYPE_NULL     0
  56#define PTR_TYPE_INTERNAL 1
  57#define PTR_TYPE_NOTE     2
  58#define PTR_TYPE_SUBTREE  3
  59
  60#define GET_PTR_TYPE(ptr)       ((uintptr_t) (ptr) & 3)
  61#define CLR_PTR_TYPE(ptr)       ((void *) ((uintptr_t) (ptr) & ~3))
  62#define SET_PTR_TYPE(ptr, type) ((void *) ((uintptr_t) (ptr) | (type)))
  63
  64#define GET_NIBBLE(n, sha1) (((sha1[(n) >> 1]) >> ((~(n) & 0x01) << 2)) & 0x0f)
  65
  66#define SUBTREE_SHA1_PREFIXCMP(key_sha1, subtree_sha1) \
  67        (memcmp(key_sha1, subtree_sha1, subtree_sha1[19]))
  68
  69struct notes_tree default_notes_tree;
  70
  71static void load_subtree(struct notes_tree *t, struct leaf_node *subtree,
  72                struct int_node *node, unsigned int n);
  73
  74/*
  75 * Search the tree until the appropriate location for the given key is found:
  76 * 1. Start at the root node, with n = 0
  77 * 2. If a[0] at the current level is a matching subtree entry, unpack that
  78 *    subtree entry and remove it; restart search at the current level.
  79 * 3. Use the nth nibble of the key as an index into a:
  80 *    - If a[n] is an int_node, recurse from #2 into that node and increment n
  81 *    - If a matching subtree entry, unpack that subtree entry (and remove it);
  82 *      restart search at the current level.
  83 *    - Otherwise, we have found one of the following:
  84 *      - a subtree entry which does not match the key
  85 *      - a note entry which may or may not match the key
  86 *      - an unused leaf node (NULL)
  87 *      In any case, set *tree and *n, and return pointer to the tree location.
  88 */
  89static void **note_tree_search(struct notes_tree *t, struct int_node **tree,
  90                unsigned char *n, const unsigned char *key_sha1)
  91{
  92        struct leaf_node *l;
  93        unsigned char i;
  94        void *p = (*tree)->a[0];
  95
  96        if (GET_PTR_TYPE(p) == PTR_TYPE_SUBTREE) {
  97                l = (struct leaf_node *) CLR_PTR_TYPE(p);
  98                if (!SUBTREE_SHA1_PREFIXCMP(key_sha1, l->key_sha1)) {
  99                        /* unpack tree and resume search */
 100                        (*tree)->a[0] = NULL;
 101                        load_subtree(t, l, *tree, *n);
 102                        free(l);
 103                        return note_tree_search(t, tree, n, key_sha1);
 104                }
 105        }
 106
 107        i = GET_NIBBLE(*n, key_sha1);
 108        p = (*tree)->a[i];
 109        switch (GET_PTR_TYPE(p)) {
 110        case PTR_TYPE_INTERNAL:
 111                *tree = CLR_PTR_TYPE(p);
 112                (*n)++;
 113                return note_tree_search(t, tree, n, key_sha1);
 114        case PTR_TYPE_SUBTREE:
 115                l = (struct leaf_node *) CLR_PTR_TYPE(p);
 116                if (!SUBTREE_SHA1_PREFIXCMP(key_sha1, l->key_sha1)) {
 117                        /* unpack tree and resume search */
 118                        (*tree)->a[i] = NULL;
 119                        load_subtree(t, l, *tree, *n);
 120                        free(l);
 121                        return note_tree_search(t, tree, n, key_sha1);
 122                }
 123                /* fall through */
 124        default:
 125                return &((*tree)->a[i]);
 126        }
 127}
 128
 129/*
 130 * To find a leaf_node:
 131 * Search to the tree location appropriate for the given key:
 132 * If a note entry with matching key, return the note entry, else return NULL.
 133 */
 134static struct leaf_node *note_tree_find(struct notes_tree *t,
 135                struct int_node *tree, unsigned char n,
 136                const unsigned char *key_sha1)
 137{
 138        void **p = note_tree_search(t, &tree, &n, key_sha1);
 139        if (GET_PTR_TYPE(*p) == PTR_TYPE_NOTE) {
 140                struct leaf_node *l = (struct leaf_node *) CLR_PTR_TYPE(*p);
 141                if (!hashcmp(key_sha1, l->key_sha1))
 142                        return l;
 143        }
 144        return NULL;
 145}
 146
 147/*
 148 * To insert a leaf_node:
 149 * Search to the tree location appropriate for the given leaf_node's key:
 150 * - If location is unused (NULL), store the tweaked pointer directly there
 151 * - If location holds a note entry that matches the note-to-be-inserted, then
 152 *   combine the two notes (by calling the given combine_notes function).
 153 * - If location holds a note entry that matches the subtree-to-be-inserted,
 154 *   then unpack the subtree-to-be-inserted into the location.
 155 * - If location holds a matching subtree entry, unpack the subtree at that
 156 *   location, and restart the insert operation from that level.
 157 * - Else, create a new int_node, holding both the node-at-location and the
 158 *   node-to-be-inserted, and store the new int_node into the location.
 159 */
 160static void note_tree_insert(struct notes_tree *t, struct int_node *tree,
 161                unsigned char n, struct leaf_node *entry, unsigned char type,
 162                combine_notes_fn combine_notes)
 163{
 164        struct int_node *new_node;
 165        struct leaf_node *l;
 166        void **p = note_tree_search(t, &tree, &n, entry->key_sha1);
 167
 168        assert(GET_PTR_TYPE(entry) == 0); /* no type bits set */
 169        l = (struct leaf_node *) CLR_PTR_TYPE(*p);
 170        switch (GET_PTR_TYPE(*p)) {
 171        case PTR_TYPE_NULL:
 172                assert(!*p);
 173                *p = SET_PTR_TYPE(entry, type);
 174                return;
 175        case PTR_TYPE_NOTE:
 176                switch (type) {
 177                case PTR_TYPE_NOTE:
 178                        if (!hashcmp(l->key_sha1, entry->key_sha1)) {
 179                                /* skip concatenation if l == entry */
 180                                if (!hashcmp(l->val_sha1, entry->val_sha1))
 181                                        return;
 182
 183                                if (combine_notes(l->val_sha1, entry->val_sha1))
 184                                        die("failed to combine notes %s and %s"
 185                                            " for object %s",
 186                                            sha1_to_hex(l->val_sha1),
 187                                            sha1_to_hex(entry->val_sha1),
 188                                            sha1_to_hex(l->key_sha1));
 189                                free(entry);
 190                                return;
 191                        }
 192                        break;
 193                case PTR_TYPE_SUBTREE:
 194                        if (!SUBTREE_SHA1_PREFIXCMP(l->key_sha1,
 195                                                    entry->key_sha1)) {
 196                                /* unpack 'entry' */
 197                                load_subtree(t, entry, tree, n);
 198                                free(entry);
 199                                return;
 200                        }
 201                        break;
 202                }
 203                break;
 204        case PTR_TYPE_SUBTREE:
 205                if (!SUBTREE_SHA1_PREFIXCMP(entry->key_sha1, l->key_sha1)) {
 206                        /* unpack 'l' and restart insert */
 207                        *p = NULL;
 208                        load_subtree(t, l, tree, n);
 209                        free(l);
 210                        note_tree_insert(t, tree, n, entry, type,
 211                                         combine_notes);
 212                        return;
 213                }
 214                break;
 215        }
 216
 217        /* non-matching leaf_node */
 218        assert(GET_PTR_TYPE(*p) == PTR_TYPE_NOTE ||
 219               GET_PTR_TYPE(*p) == PTR_TYPE_SUBTREE);
 220        new_node = (struct int_node *) xcalloc(sizeof(struct int_node), 1);
 221        note_tree_insert(t, new_node, n + 1, l, GET_PTR_TYPE(*p),
 222                         combine_notes);
 223        *p = SET_PTR_TYPE(new_node, PTR_TYPE_INTERNAL);
 224        note_tree_insert(t, new_node, n + 1, entry, type, combine_notes);
 225}
 226
 227/*
 228 * How to consolidate an int_node:
 229 * If there are > 1 non-NULL entries, give up and return non-zero.
 230 * Otherwise replace the int_node at the given index in the given parent node
 231 * with the only entry (or a NULL entry if no entries) from the given tree,
 232 * and return 0.
 233 */
 234static int note_tree_consolidate(struct int_node *tree,
 235        struct int_node *parent, unsigned char index)
 236{
 237        unsigned int i;
 238        void *p = NULL;
 239
 240        assert(tree && parent);
 241        assert(CLR_PTR_TYPE(parent->a[index]) == tree);
 242
 243        for (i = 0; i < 16; i++) {
 244                if (GET_PTR_TYPE(tree->a[i]) != PTR_TYPE_NULL) {
 245                        if (p) /* more than one entry */
 246                                return -2;
 247                        p = tree->a[i];
 248                }
 249        }
 250
 251        /* replace tree with p in parent[index] */
 252        parent->a[index] = p;
 253        free(tree);
 254        return 0;
 255}
 256
 257/*
 258 * To remove a leaf_node:
 259 * Search to the tree location appropriate for the given leaf_node's key:
 260 * - If location does not hold a matching entry, abort and do nothing.
 261 * - Replace the matching leaf_node with a NULL entry (and free the leaf_node).
 262 * - Consolidate int_nodes repeatedly, while walking up the tree towards root.
 263 */
 264static void note_tree_remove(struct notes_tree *t, struct int_node *tree,
 265                unsigned char n, struct leaf_node *entry)
 266{
 267        struct leaf_node *l;
 268        struct int_node *parent_stack[20];
 269        unsigned char i, j;
 270        void **p = note_tree_search(t, &tree, &n, entry->key_sha1);
 271
 272        assert(GET_PTR_TYPE(entry) == 0); /* no type bits set */
 273        if (GET_PTR_TYPE(*p) != PTR_TYPE_NOTE)
 274                return; /* type mismatch, nothing to remove */
 275        l = (struct leaf_node *) CLR_PTR_TYPE(*p);
 276        if (hashcmp(l->key_sha1, entry->key_sha1))
 277                return; /* key mismatch, nothing to remove */
 278
 279        /* we have found a matching entry */
 280        free(l);
 281        *p = SET_PTR_TYPE(NULL, PTR_TYPE_NULL);
 282
 283        /* consolidate this tree level, and parent levels, if possible */
 284        if (!n)
 285                return; /* cannot consolidate top level */
 286        /* first, build stack of ancestors between root and current node */
 287        parent_stack[0] = t->root;
 288        for (i = 0; i < n; i++) {
 289                j = GET_NIBBLE(i, entry->key_sha1);
 290                parent_stack[i + 1] = CLR_PTR_TYPE(parent_stack[i]->a[j]);
 291        }
 292        assert(i == n && parent_stack[i] == tree);
 293        /* next, unwind stack until note_tree_consolidate() is done */
 294        while (i > 0 &&
 295               !note_tree_consolidate(parent_stack[i], parent_stack[i - 1],
 296                                      GET_NIBBLE(i - 1, entry->key_sha1)))
 297                i--;
 298}
 299
 300/* Free the entire notes data contained in the given tree */
 301static void note_tree_free(struct int_node *tree)
 302{
 303        unsigned int i;
 304        for (i = 0; i < 16; i++) {
 305                void *p = tree->a[i];
 306                switch (GET_PTR_TYPE(p)) {
 307                case PTR_TYPE_INTERNAL:
 308                        note_tree_free(CLR_PTR_TYPE(p));
 309                        /* fall through */
 310                case PTR_TYPE_NOTE:
 311                case PTR_TYPE_SUBTREE:
 312                        free(CLR_PTR_TYPE(p));
 313                }
 314        }
 315}
 316
 317/*
 318 * Convert a partial SHA1 hex string to the corresponding partial SHA1 value.
 319 * - hex      - Partial SHA1 segment in ASCII hex format
 320 * - hex_len  - Length of above segment. Must be multiple of 2 between 0 and 40
 321 * - sha1     - Partial SHA1 value is written here
 322 * - sha1_len - Max #bytes to store in sha1, Must be >= hex_len / 2, and < 20
 323 * Returns -1 on error (invalid arguments or invalid SHA1 (not in hex format)).
 324 * Otherwise, returns number of bytes written to sha1 (i.e. hex_len / 2).
 325 * Pads sha1 with NULs up to sha1_len (not included in returned length).
 326 */
 327static int get_sha1_hex_segment(const char *hex, unsigned int hex_len,
 328                unsigned char *sha1, unsigned int sha1_len)
 329{
 330        unsigned int i, len = hex_len >> 1;
 331        if (hex_len % 2 != 0 || len > sha1_len)
 332                return -1;
 333        for (i = 0; i < len; i++) {
 334                unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
 335                if (val & ~0xff)
 336                        return -1;
 337                *sha1++ = val;
 338                hex += 2;
 339        }
 340        for (; i < sha1_len; i++)
 341                *sha1++ = 0;
 342        return len;
 343}
 344
 345static int non_note_cmp(const struct non_note *a, const struct non_note *b)
 346{
 347        return strcmp(a->path, b->path);
 348}
 349
 350static void add_non_note(struct notes_tree *t, const char *path,
 351                unsigned int mode, const unsigned char *sha1)
 352{
 353        struct non_note *p = t->prev_non_note, *n;
 354        n = (struct non_note *) xmalloc(sizeof(struct non_note));
 355        n->next = NULL;
 356        n->path = xstrdup(path);
 357        n->mode = mode;
 358        hashcpy(n->sha1, sha1);
 359        t->prev_non_note = n;
 360
 361        if (!t->first_non_note) {
 362                t->first_non_note = n;
 363                return;
 364        }
 365
 366        if (non_note_cmp(p, n) < 0)
 367                ; /* do nothing  */
 368        else if (non_note_cmp(t->first_non_note, n) <= 0)
 369                p = t->first_non_note;
 370        else {
 371                /* n sorts before t->first_non_note */
 372                n->next = t->first_non_note;
 373                t->first_non_note = n;
 374                return;
 375        }
 376
 377        /* n sorts equal or after p */
 378        while (p->next && non_note_cmp(p->next, n) <= 0)
 379                p = p->next;
 380
 381        if (non_note_cmp(p, n) == 0) { /* n ~= p; overwrite p with n */
 382                assert(strcmp(p->path, n->path) == 0);
 383                p->mode = n->mode;
 384                hashcpy(p->sha1, n->sha1);
 385                free(n);
 386                t->prev_non_note = p;
 387                return;
 388        }
 389
 390        /* n sorts between p and p->next */
 391        n->next = p->next;
 392        p->next = n;
 393}
 394
 395static void load_subtree(struct notes_tree *t, struct leaf_node *subtree,
 396                struct int_node *node, unsigned int n)
 397{
 398        unsigned char object_sha1[20];
 399        unsigned int prefix_len;
 400        void *buf;
 401        struct tree_desc desc;
 402        struct name_entry entry;
 403        int len, path_len;
 404        unsigned char type;
 405        struct leaf_node *l;
 406
 407        buf = fill_tree_descriptor(&desc, subtree->val_sha1);
 408        if (!buf)
 409                die("Could not read %s for notes-index",
 410                     sha1_to_hex(subtree->val_sha1));
 411
 412        prefix_len = subtree->key_sha1[19];
 413        assert(prefix_len * 2 >= n);
 414        memcpy(object_sha1, subtree->key_sha1, prefix_len);
 415        while (tree_entry(&desc, &entry)) {
 416                path_len = strlen(entry.path);
 417                len = get_sha1_hex_segment(entry.path, path_len,
 418                                object_sha1 + prefix_len, 20 - prefix_len);
 419                if (len < 0)
 420                        goto handle_non_note; /* entry.path is not a SHA1 */
 421                len += prefix_len;
 422
 423                /*
 424                 * If object SHA1 is complete (len == 20), assume note object
 425                 * If object SHA1 is incomplete (len < 20), and current
 426                 * component consists of 2 hex chars, assume note subtree
 427                 */
 428                if (len <= 20) {
 429                        type = PTR_TYPE_NOTE;
 430                        l = (struct leaf_node *)
 431                                xcalloc(sizeof(struct leaf_node), 1);
 432                        hashcpy(l->key_sha1, object_sha1);
 433                        hashcpy(l->val_sha1, entry.sha1);
 434                        if (len < 20) {
 435                                if (!S_ISDIR(entry.mode) || path_len != 2)
 436                                        goto handle_non_note; /* not subtree */
 437                                l->key_sha1[19] = (unsigned char) len;
 438                                type = PTR_TYPE_SUBTREE;
 439                        }
 440                        note_tree_insert(t, node, n, l, type,
 441                                         combine_notes_concatenate);
 442                }
 443                continue;
 444
 445handle_non_note:
 446                /*
 447                 * Determine full path for this non-note entry:
 448                 * The filename is already found in entry.path, but the
 449                 * directory part of the path must be deduced from the subtree
 450                 * containing this entry. We assume here that the overall notes
 451                 * tree follows a strict byte-based progressive fanout
 452                 * structure (i.e. using 2/38, 2/2/36, etc. fanouts, and not
 453                 * e.g. 4/36 fanout). This means that if a non-note is found at
 454                 * path "dead/beef", the following code will register it as
 455                 * being found on "de/ad/beef".
 456                 * On the other hand, if you use such non-obvious non-note
 457                 * paths in the middle of a notes tree, you deserve what's
 458                 * coming to you ;). Note that for non-notes that are not
 459                 * SHA1-like at the top level, there will be no problems.
 460                 *
 461                 * To conclude, it is strongly advised to make sure non-notes
 462                 * have at least one non-hex character in the top-level path
 463                 * component.
 464                 */
 465                {
 466                        char non_note_path[PATH_MAX];
 467                        char *p = non_note_path;
 468                        const char *q = sha1_to_hex(subtree->key_sha1);
 469                        int i;
 470                        for (i = 0; i < prefix_len; i++) {
 471                                *p++ = *q++;
 472                                *p++ = *q++;
 473                                *p++ = '/';
 474                        }
 475                        strcpy(p, entry.path);
 476                        add_non_note(t, non_note_path, entry.mode, entry.sha1);
 477                }
 478        }
 479        free(buf);
 480}
 481
 482/*
 483 * Determine optimal on-disk fanout for this part of the notes tree
 484 *
 485 * Given a (sub)tree and the level in the internal tree structure, determine
 486 * whether or not the given existing fanout should be expanded for this
 487 * (sub)tree.
 488 *
 489 * Values of the 'fanout' variable:
 490 * - 0: No fanout (all notes are stored directly in the root notes tree)
 491 * - 1: 2/38 fanout
 492 * - 2: 2/2/36 fanout
 493 * - 3: 2/2/2/34 fanout
 494 * etc.
 495 */
 496static unsigned char determine_fanout(struct int_node *tree, unsigned char n,
 497                unsigned char fanout)
 498{
 499        /*
 500         * The following is a simple heuristic that works well in practice:
 501         * For each even-numbered 16-tree level (remember that each on-disk
 502         * fanout level corresponds to _two_ 16-tree levels), peek at all 16
 503         * entries at that tree level. If all of them are either int_nodes or
 504         * subtree entries, then there are likely plenty of notes below this
 505         * level, so we return an incremented fanout.
 506         */
 507        unsigned int i;
 508        if ((n % 2) || (n > 2 * fanout))
 509                return fanout;
 510        for (i = 0; i < 16; i++) {
 511                switch (GET_PTR_TYPE(tree->a[i])) {
 512                case PTR_TYPE_SUBTREE:
 513                case PTR_TYPE_INTERNAL:
 514                        continue;
 515                default:
 516                        return fanout;
 517                }
 518        }
 519        return fanout + 1;
 520}
 521
 522static void construct_path_with_fanout(const unsigned char *sha1,
 523                unsigned char fanout, char *path)
 524{
 525        unsigned int i = 0, j = 0;
 526        const char *hex_sha1 = sha1_to_hex(sha1);
 527        assert(fanout < 20);
 528        while (fanout) {
 529                path[i++] = hex_sha1[j++];
 530                path[i++] = hex_sha1[j++];
 531                path[i++] = '/';
 532                fanout--;
 533        }
 534        strcpy(path + i, hex_sha1 + j);
 535}
 536
 537static int for_each_note_helper(struct notes_tree *t, struct int_node *tree,
 538                unsigned char n, unsigned char fanout, int flags,
 539                each_note_fn fn, void *cb_data)
 540{
 541        unsigned int i;
 542        void *p;
 543        int ret = 0;
 544        struct leaf_node *l;
 545        static char path[40 + 19 + 1];  /* hex SHA1 + 19 * '/' + NUL */
 546
 547        fanout = determine_fanout(tree, n, fanout);
 548        for (i = 0; i < 16; i++) {
 549redo:
 550                p = tree->a[i];
 551                switch (GET_PTR_TYPE(p)) {
 552                case PTR_TYPE_INTERNAL:
 553                        /* recurse into int_node */
 554                        ret = for_each_note_helper(t, CLR_PTR_TYPE(p), n + 1,
 555                                fanout, flags, fn, cb_data);
 556                        break;
 557                case PTR_TYPE_SUBTREE:
 558                        l = (struct leaf_node *) CLR_PTR_TYPE(p);
 559                        /*
 560                         * Subtree entries in the note tree represent parts of
 561                         * the note tree that have not yet been explored. There
 562                         * is a direct relationship between subtree entries at
 563                         * level 'n' in the tree, and the 'fanout' variable:
 564                         * Subtree entries at level 'n <= 2 * fanout' should be
 565                         * preserved, since they correspond exactly to a fanout
 566                         * directory in the on-disk structure. However, subtree
 567                         * entries at level 'n > 2 * fanout' should NOT be
 568                         * preserved, but rather consolidated into the above
 569                         * notes tree level. We achieve this by unconditionally
 570                         * unpacking subtree entries that exist below the
 571                         * threshold level at 'n = 2 * fanout'.
 572                         */
 573                        if (n <= 2 * fanout &&
 574                            flags & FOR_EACH_NOTE_YIELD_SUBTREES) {
 575                                /* invoke callback with subtree */
 576                                unsigned int path_len =
 577                                        l->key_sha1[19] * 2 + fanout;
 578                                assert(path_len < 40 + 19);
 579                                construct_path_with_fanout(l->key_sha1, fanout,
 580                                                           path);
 581                                /* Create trailing slash, if needed */
 582                                if (path[path_len - 1] != '/')
 583                                        path[path_len++] = '/';
 584                                path[path_len] = '\0';
 585                                ret = fn(l->key_sha1, l->val_sha1, path,
 586                                         cb_data);
 587                        }
 588                        if (n > fanout * 2 ||
 589                            !(flags & FOR_EACH_NOTE_DONT_UNPACK_SUBTREES)) {
 590                                /* unpack subtree and resume traversal */
 591                                tree->a[i] = NULL;
 592                                load_subtree(t, l, tree, n);
 593                                free(l);
 594                                goto redo;
 595                        }
 596                        break;
 597                case PTR_TYPE_NOTE:
 598                        l = (struct leaf_node *) CLR_PTR_TYPE(p);
 599                        construct_path_with_fanout(l->key_sha1, fanout, path);
 600                        ret = fn(l->key_sha1, l->val_sha1, path, cb_data);
 601                        break;
 602                }
 603                if (ret)
 604                        return ret;
 605        }
 606        return 0;
 607}
 608
 609struct tree_write_stack {
 610        struct tree_write_stack *next;
 611        struct strbuf buf;
 612        char path[2]; /* path to subtree in next, if any */
 613};
 614
 615static inline int matches_tree_write_stack(struct tree_write_stack *tws,
 616                const char *full_path)
 617{
 618        return  full_path[0] == tws->path[0] &&
 619                full_path[1] == tws->path[1] &&
 620                full_path[2] == '/';
 621}
 622
 623static void write_tree_entry(struct strbuf *buf, unsigned int mode,
 624                const char *path, unsigned int path_len, const
 625                unsigned char *sha1)
 626{
 627                strbuf_addf(buf, "%06o %.*s%c", mode, path_len, path, '\0');
 628                strbuf_add(buf, sha1, 20);
 629}
 630
 631static void tree_write_stack_init_subtree(struct tree_write_stack *tws,
 632                const char *path)
 633{
 634        struct tree_write_stack *n;
 635        assert(!tws->next);
 636        assert(tws->path[0] == '\0' && tws->path[1] == '\0');
 637        n = (struct tree_write_stack *)
 638                xmalloc(sizeof(struct tree_write_stack));
 639        n->next = NULL;
 640        strbuf_init(&n->buf, 256 * (32 + 40)); /* assume 256 entries per tree */
 641        n->path[0] = n->path[1] = '\0';
 642        tws->next = n;
 643        tws->path[0] = path[0];
 644        tws->path[1] = path[1];
 645}
 646
 647static int tree_write_stack_finish_subtree(struct tree_write_stack *tws)
 648{
 649        int ret;
 650        struct tree_write_stack *n = tws->next;
 651        unsigned char s[20];
 652        if (n) {
 653                ret = tree_write_stack_finish_subtree(n);
 654                if (ret)
 655                        return ret;
 656                ret = write_sha1_file(n->buf.buf, n->buf.len, tree_type, s);
 657                if (ret)
 658                        return ret;
 659                strbuf_release(&n->buf);
 660                free(n);
 661                tws->next = NULL;
 662                write_tree_entry(&tws->buf, 040000, tws->path, 2, s);
 663                tws->path[0] = tws->path[1] = '\0';
 664        }
 665        return 0;
 666}
 667
 668static int write_each_note_helper(struct tree_write_stack *tws,
 669                const char *path, unsigned int mode,
 670                const unsigned char *sha1)
 671{
 672        size_t path_len = strlen(path);
 673        unsigned int n = 0;
 674        int ret;
 675
 676        /* Determine common part of tree write stack */
 677        while (tws && 3 * n < path_len &&
 678               matches_tree_write_stack(tws, path + 3 * n)) {
 679                n++;
 680                tws = tws->next;
 681        }
 682
 683        /* tws point to last matching tree_write_stack entry */
 684        ret = tree_write_stack_finish_subtree(tws);
 685        if (ret)
 686                return ret;
 687
 688        /* Start subtrees needed to satisfy path */
 689        while (3 * n + 2 < path_len && path[3 * n + 2] == '/') {
 690                tree_write_stack_init_subtree(tws, path + 3 * n);
 691                n++;
 692                tws = tws->next;
 693        }
 694
 695        /* There should be no more directory components in the given path */
 696        assert(memchr(path + 3 * n, '/', path_len - (3 * n)) == NULL);
 697
 698        /* Finally add given entry to the current tree object */
 699        write_tree_entry(&tws->buf, mode, path + 3 * n, path_len - (3 * n),
 700                         sha1);
 701
 702        return 0;
 703}
 704
 705struct write_each_note_data {
 706        struct tree_write_stack *root;
 707        struct non_note *next_non_note;
 708};
 709
 710static int write_each_non_note_until(const char *note_path,
 711                struct write_each_note_data *d)
 712{
 713        struct non_note *n = d->next_non_note;
 714        int cmp, ret;
 715        while (n && (!note_path || (cmp = strcmp(n->path, note_path)) <= 0)) {
 716                if (note_path && cmp == 0)
 717                        ; /* do nothing, prefer note to non-note */
 718                else {
 719                        ret = write_each_note_helper(d->root, n->path, n->mode,
 720                                                     n->sha1);
 721                        if (ret)
 722                                return ret;
 723                }
 724                n = n->next;
 725        }
 726        d->next_non_note = n;
 727        return 0;
 728}
 729
 730static int write_each_note(const unsigned char *object_sha1,
 731                const unsigned char *note_sha1, char *note_path,
 732                void *cb_data)
 733{
 734        struct write_each_note_data *d =
 735                (struct write_each_note_data *) cb_data;
 736        size_t note_path_len = strlen(note_path);
 737        unsigned int mode = 0100644;
 738
 739        if (note_path[note_path_len - 1] == '/') {
 740                /* subtree entry */
 741                note_path_len--;
 742                note_path[note_path_len] = '\0';
 743                mode = 040000;
 744        }
 745        assert(note_path_len <= 40 + 19);
 746
 747        /* Weave non-note entries into note entries */
 748        return  write_each_non_note_until(note_path, d) ||
 749                write_each_note_helper(d->root, note_path, mode, note_sha1);
 750}
 751
 752int combine_notes_concatenate(unsigned char *cur_sha1,
 753                const unsigned char *new_sha1)
 754{
 755        char *cur_msg = NULL, *new_msg = NULL, *buf;
 756        unsigned long cur_len, new_len, buf_len;
 757        enum object_type cur_type, new_type;
 758        int ret;
 759
 760        /* read in both note blob objects */
 761        if (!is_null_sha1(new_sha1))
 762                new_msg = read_sha1_file(new_sha1, &new_type, &new_len);
 763        if (!new_msg || !new_len || new_type != OBJ_BLOB) {
 764                free(new_msg);
 765                return 0;
 766        }
 767        if (!is_null_sha1(cur_sha1))
 768                cur_msg = read_sha1_file(cur_sha1, &cur_type, &cur_len);
 769        if (!cur_msg || !cur_len || cur_type != OBJ_BLOB) {
 770                free(cur_msg);
 771                free(new_msg);
 772                hashcpy(cur_sha1, new_sha1);
 773                return 0;
 774        }
 775
 776        /* we will separate the notes by a newline anyway */
 777        if (cur_msg[cur_len - 1] == '\n')
 778                cur_len--;
 779
 780        /* concatenate cur_msg and new_msg into buf */
 781        buf_len = cur_len + 1 + new_len;
 782        buf = (char *) xmalloc(buf_len);
 783        memcpy(buf, cur_msg, cur_len);
 784        buf[cur_len] = '\n';
 785        memcpy(buf + cur_len + 1, new_msg, new_len);
 786        free(cur_msg);
 787        free(new_msg);
 788
 789        /* create a new blob object from buf */
 790        ret = write_sha1_file(buf, buf_len, blob_type, cur_sha1);
 791        free(buf);
 792        return ret;
 793}
 794
 795int combine_notes_overwrite(unsigned char *cur_sha1,
 796                const unsigned char *new_sha1)
 797{
 798        hashcpy(cur_sha1, new_sha1);
 799        return 0;
 800}
 801
 802int combine_notes_ignore(unsigned char *cur_sha1,
 803                const unsigned char *new_sha1)
 804{
 805        return 0;
 806}
 807
 808void init_notes(struct notes_tree *t, const char *notes_ref,
 809                combine_notes_fn combine_notes, int flags)
 810{
 811        unsigned char sha1[20], object_sha1[20];
 812        unsigned mode;
 813        struct leaf_node root_tree;
 814
 815        if (!t)
 816                t = &default_notes_tree;
 817        assert(!t->initialized);
 818
 819        if (!notes_ref)
 820                notes_ref = getenv(GIT_NOTES_REF_ENVIRONMENT);
 821        if (!notes_ref)
 822                notes_ref = notes_ref_name; /* value of core.notesRef config */
 823        if (!notes_ref)
 824                notes_ref = GIT_NOTES_DEFAULT_REF;
 825
 826        if (!combine_notes)
 827                combine_notes = combine_notes_concatenate;
 828
 829        t->root = (struct int_node *) xcalloc(sizeof(struct int_node), 1);
 830        t->first_non_note = NULL;
 831        t->prev_non_note = NULL;
 832        t->ref = notes_ref ? xstrdup(notes_ref) : NULL;
 833        t->combine_notes = combine_notes;
 834        t->initialized = 1;
 835
 836        if (flags & NOTES_INIT_EMPTY || !notes_ref ||
 837            read_ref(notes_ref, object_sha1))
 838                return;
 839        if (get_tree_entry(object_sha1, "", sha1, &mode))
 840                die("Failed to read notes tree referenced by %s (%s)",
 841                    notes_ref, object_sha1);
 842
 843        hashclr(root_tree.key_sha1);
 844        hashcpy(root_tree.val_sha1, sha1);
 845        load_subtree(t, &root_tree, t->root, 0);
 846}
 847
 848void add_note(struct notes_tree *t, const unsigned char *object_sha1,
 849                const unsigned char *note_sha1, combine_notes_fn combine_notes)
 850{
 851        struct leaf_node *l;
 852
 853        if (!t)
 854                t = &default_notes_tree;
 855        assert(t->initialized);
 856        if (!combine_notes)
 857                combine_notes = t->combine_notes;
 858        l = (struct leaf_node *) xmalloc(sizeof(struct leaf_node));
 859        hashcpy(l->key_sha1, object_sha1);
 860        hashcpy(l->val_sha1, note_sha1);
 861        note_tree_insert(t, t->root, 0, l, PTR_TYPE_NOTE, combine_notes);
 862}
 863
 864void remove_note(struct notes_tree *t, const unsigned char *object_sha1)
 865{
 866        struct leaf_node l;
 867
 868        if (!t)
 869                t = &default_notes_tree;
 870        assert(t->initialized);
 871        hashcpy(l.key_sha1, object_sha1);
 872        hashclr(l.val_sha1);
 873        return note_tree_remove(t, t->root, 0, &l);
 874}
 875
 876const unsigned char *get_note(struct notes_tree *t,
 877                const unsigned char *object_sha1)
 878{
 879        struct leaf_node *found;
 880
 881        if (!t)
 882                t = &default_notes_tree;
 883        assert(t->initialized);
 884        found = note_tree_find(t, t->root, 0, object_sha1);
 885        return found ? found->val_sha1 : NULL;
 886}
 887
 888int for_each_note(struct notes_tree *t, int flags, each_note_fn fn,
 889                void *cb_data)
 890{
 891        if (!t)
 892                t = &default_notes_tree;
 893        assert(t->initialized);
 894        return for_each_note_helper(t, t->root, 0, 0, flags, fn, cb_data);
 895}
 896
 897int write_notes_tree(struct notes_tree *t, unsigned char *result)
 898{
 899        struct tree_write_stack root;
 900        struct write_each_note_data cb_data;
 901        int ret;
 902
 903        if (!t)
 904                t = &default_notes_tree;
 905        assert(t->initialized);
 906
 907        /* Prepare for traversal of current notes tree */
 908        root.next = NULL; /* last forward entry in list is grounded */
 909        strbuf_init(&root.buf, 256 * (32 + 40)); /* assume 256 entries */
 910        root.path[0] = root.path[1] = '\0';
 911        cb_data.root = &root;
 912        cb_data.next_non_note = t->first_non_note;
 913
 914        /* Write tree objects representing current notes tree */
 915        ret = for_each_note(t, FOR_EACH_NOTE_DONT_UNPACK_SUBTREES |
 916                                FOR_EACH_NOTE_YIELD_SUBTREES,
 917                        write_each_note, &cb_data) ||
 918                write_each_non_note_until(NULL, &cb_data) ||
 919                tree_write_stack_finish_subtree(&root) ||
 920                write_sha1_file(root.buf.buf, root.buf.len, tree_type, result);
 921        strbuf_release(&root.buf);
 922        return ret;
 923}
 924
 925void free_notes(struct notes_tree *t)
 926{
 927        if (!t)
 928                t = &default_notes_tree;
 929        if (t->root)
 930                note_tree_free(t->root);
 931        free(t->root);
 932        while (t->first_non_note) {
 933                t->prev_non_note = t->first_non_note->next;
 934                free(t->first_non_note->path);
 935                free(t->first_non_note);
 936                t->first_non_note = t->prev_non_note;
 937        }
 938        free(t->ref);
 939        memset(t, 0, sizeof(struct notes_tree));
 940}
 941
 942void format_note(struct notes_tree *t, const unsigned char *object_sha1,
 943                struct strbuf *sb, const char *output_encoding, int flags)
 944{
 945        static const char utf8[] = "utf-8";
 946        const unsigned char *sha1;
 947        char *msg, *msg_p;
 948        unsigned long linelen, msglen;
 949        enum object_type type;
 950
 951        if (!t)
 952                t = &default_notes_tree;
 953        if (!t->initialized)
 954                init_notes(t, NULL, NULL, 0);
 955
 956        sha1 = get_note(t, object_sha1);
 957        if (!sha1)
 958                return;
 959
 960        if (!(msg = read_sha1_file(sha1, &type, &msglen)) || !msglen ||
 961                        type != OBJ_BLOB) {
 962                free(msg);
 963                return;
 964        }
 965
 966        if (output_encoding && *output_encoding &&
 967                        strcmp(utf8, output_encoding)) {
 968                char *reencoded = reencode_string(msg, output_encoding, utf8);
 969                if (reencoded) {
 970                        free(msg);
 971                        msg = reencoded;
 972                        msglen = strlen(msg);
 973                }
 974        }
 975
 976        /* we will end the annotation by a newline anyway */
 977        if (msglen && msg[msglen - 1] == '\n')
 978                msglen--;
 979
 980        if (flags & NOTES_SHOW_HEADER)
 981                strbuf_addstr(sb, "\nNotes:\n");
 982
 983        for (msg_p = msg; msg_p < msg + msglen; msg_p += linelen + 1) {
 984                linelen = strchrnul(msg_p, '\n') - msg_p;
 985
 986                if (flags & NOTES_INDENT)
 987                        strbuf_addstr(sb, "    ");
 988                strbuf_add(sb, msg_p, linelen);
 989                strbuf_addch(sb, '\n');
 990        }
 991
 992        free(msg);
 993}