pack-objects: rework check_delta_limit usage
[gitweb.git] / builtin-pack-objects.c
index 099dea0e1eea0165f9acd5fb37764249aea033c6..d44b8f4c01ff74d4726276f8ccc81b9db21ccd0c 100644 (file)
@@ -27,7 +27,6 @@ struct object_entry {
                                 * nonzero if already written.
                                 */
        unsigned int depth;     /* delta depth */
-       unsigned int delta_limit;       /* base adjustment for in-pack delta */
        unsigned int hash;      /* name hint hash */
        enum object_type type;
        enum object_type in_pack_type;  /* could be delta */
@@ -233,12 +232,6 @@ static struct revindex_entry * find_packed_object(struct packed_git *p,
        die("internal error: pack revindex corrupt");
 }
 
-static off_t find_packed_object_size(struct packed_git *p, off_t ofs)
-{
-       struct revindex_entry *entry = find_packed_object(p, ofs);
-       return entry[1].offset - ofs;
-}
-
 static const unsigned char *find_packed_object_name(struct packed_git *p,
                                                    off_t ofs)
 {
@@ -321,6 +314,28 @@ static int check_pack_inflate(struct packed_git *p,
                stream.total_in == len) ? 0 : -1;
 }
 
+static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
+                         off_t offset, off_t len, unsigned int nr)
+{
+       const uint32_t *index_crc;
+       uint32_t data_crc = crc32(0, Z_NULL, 0);
+
+       do {
+               unsigned int avail;
+               void *data = use_pack(p, w_curs, offset, &avail);
+               if (avail > len)
+                       avail = len;
+               data_crc = crc32(data_crc, data, avail);
+               offset += avail;
+               len -= avail;
+       } while (len);
+
+       index_crc = p->index_data;
+       index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
+
+       return data_crc != ntohl(*index_crc);
+}
+
 static void copy_pack_data(struct sha1file *f,
                struct packed_git *p,
                struct pack_window **w_curs,
@@ -485,6 +500,7 @@ static unsigned long write_object(struct sha1file *f,
        else {
                struct packed_git *p = entry->in_pack;
                struct pack_window *w_curs = NULL;
+               struct revindex_entry *revidx;
                off_t offset;
 
                if (entry->delta) {
@@ -507,12 +523,17 @@ static unsigned long write_object(struct sha1file *f,
                        hdrlen += 20;
                }
 
-               offset = entry->in_pack_offset + entry->in_pack_header_size;
-               datalen = find_packed_object_size(p, entry->in_pack_offset)
-                               - entry->in_pack_header_size;
-               if (!pack_to_stdout && check_pack_inflate(p, &w_curs,
-                               offset, datalen, entry->size))
-                       die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
+               offset = entry->in_pack_offset;
+               revidx = find_packed_object(p, offset);
+               datalen = revidx[1].offset - offset;
+               if (!pack_to_stdout && p->index_version > 1 &&
+                   check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
+                       die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
+               offset += entry->in_pack_header_size;
+               datalen -= entry->in_pack_header_size;
+               if (!pack_to_stdout && p->index_version == 1 &&
+                   check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
+                       die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
                copy_pack_data(f, p, &w_curs, offset, datalen);
                unuse_pack(&w_curs);
                reused++;
@@ -759,12 +780,19 @@ static unsigned name_hash(const char *name)
 
 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
 {
-       uint32_t idx = nr_objects;
        struct object_entry *entry;
-       struct packed_git *p;
+       struct packed_git *p, *found_pack = NULL;
        off_t found_offset = 0;
-       struct packed_git *found_pack = NULL;
-       int ix, status = 0;
+       int ix;
+
+       ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
+       if (ix >= 0) {
+               if (exclude) {
+                       entry = objects + object_ix[ix] - 1;
+                       entry->preferred_base = 1;
+               }
+               return 0;
+       }
 
        if (!exclude) {
                for (p = packed_git; p; p = p->next) {
@@ -781,43 +809,34 @@ static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclud
                        }
                }
        }
-       if ((entry = locate_object_entry(sha1)) != NULL)
-               goto already_added;
 
-       if (idx >= nr_alloc) {
-               nr_alloc = (idx + 1024) * 3 / 2;
+       if (nr_objects >= nr_alloc) {
+               nr_alloc = (nr_alloc  + 1024) * 3 / 2;
                objects = xrealloc(objects, nr_alloc * sizeof(*entry));
        }
-       entry = objects + idx;
-       nr_objects = idx + 1;
+
+       entry = objects + nr_objects++;
        memset(entry, 0, sizeof(*entry));
        hashcpy(entry->sha1, sha1);
        entry->hash = hash;
+       if (exclude)
+               entry->preferred_base = 1;
+       if (found_pack) {
+               entry->in_pack = found_pack;
+               entry->in_pack_offset = found_offset;
+       }
 
        if (object_ix_hashsz * 3 <= nr_objects * 4)
                rehash_objects();
-       else {
-               ix = locate_object_entry_hash(entry->sha1);
-               if (0 <= ix)
-                       die("internal error in object hashing.");
-               object_ix[-1 - ix] = idx + 1;
-       }
-       status = 1;
+       else
+               object_ix[-1 - ix] = nr_objects;
 
- already_added:
        if (progress_update) {
                fprintf(stderr, "Counting objects...%u\r", nr_objects);
                progress_update = 0;
        }
-       if (exclude)
-               entry->preferred_base = 1;
-       else {
-               if (found_pack) {
-                       entry->in_pack = found_pack;
-                       entry->in_pack_offset = found_offset;
-               }
-       }
-       return status;
+
+       return 1;
 }
 
 struct pbase_tree_cache {
@@ -939,22 +958,21 @@ static void add_pbase_object(struct tree_desc *tree,
                             const char *fullname)
 {
        struct name_entry entry;
+       int cmp;
 
        while (tree_entry(tree,&entry)) {
-               unsigned long size;
-               enum object_type type;
-
-               if (tree_entry_len(entry.path, entry.sha1) != cmplen ||
-                   memcmp(entry.path, name, cmplen) ||
-                   !has_sha1_file(entry.sha1) ||
-                   (type = sha1_object_info(entry.sha1, &size)) < 0)
+               cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
+                     memcmp(name, entry.path, cmplen);
+               if (cmp > 0)
                        continue;
+               if (cmp < 0)
+                       return;
                if (name[cmplen] != '/') {
                        unsigned hash = name_hash(fullname);
                        add_object_entry(entry.sha1, hash, 1);
                        return;
                }
-               if (type == OBJ_TREE) {
+               if (S_ISDIR(entry.mode)) {
                        struct tree_desc sub;
                        struct pbase_tree_cache *tree;
                        const char *down = name+cmplen+1;
@@ -1014,15 +1032,15 @@ static int check_pbase_path(unsigned hash)
 static void add_preferred_base_object(const char *name, unsigned hash)
 {
        struct pbase_tree *it;
-       int cmplen = name_cmp_len(name);
+       int cmplen;
 
-       if (check_pbase_path(hash))
+       if (!num_preferred_base || check_pbase_path(hash))
                return;
 
+       cmplen = name_cmp_len(name);
        for (it = pbase_tree; it; it = it->next) {
                if (cmplen == 0) {
-                       hash = name_hash("");
-                       add_object_entry(it->pcache.sha1, hash, 1);
+                       add_object_entry(it->pcache.sha1, 0, 1);
                }
                else {
                        struct tree_desc tree;
@@ -1153,19 +1171,6 @@ static void check_object(struct object_entry *entry)
                    sha1_to_hex(entry->sha1));
 }
 
-static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
-{
-       struct object_entry *child = me->delta_child;
-       unsigned int m = n;
-       while (child) {
-               unsigned int c = check_delta_limit(child, n + 1);
-               if (m < c)
-                       m = c;
-               child = child->delta_sibling;
-       }
-       return m;
-}
-
 static void get_object_details(void)
 {
        uint32_t i;
@@ -1174,23 +1179,6 @@ static void get_object_details(void)
        prepare_pack_ix();
        for (i = 0, entry = objects; i < nr_objects; i++, entry++)
                check_object(entry);
-
-       if (nr_objects == nr_result) {
-               /*
-                * Depth of objects that depend on the entry -- this
-                * is subtracted from depth-max to break too deep
-                * delta chain because of delta data reusing.
-                * However, we loosen this restriction when we know we
-                * are creating a thin pack -- it will have to be
-                * expanded on the other end anyway, so do not
-                * artificially cut the delta chain and let it go as
-                * deep as it wants.
-                */
-               for (i = 0, entry = objects; i < nr_objects; i++, entry++)
-                       if (!entry->delta && entry->delta_child)
-                               entry->delta_limit =
-                                       check_delta_limit(entry, 1);
-       }
 }
 
 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
@@ -1257,7 +1245,7 @@ static int type_size_sort(const struct object_entry *a, const struct object_entr
                return -1;
        if (a->size > b->size)
                return 1;
-       return a < b ? -1 : (a > b);
+       return a > b ? -1 : (a < b);  /* newest last */
 }
 
 struct unpacked {
@@ -1303,16 +1291,7 @@ static int try_delta(struct unpacked *trg, struct unpacked *src,
            trg_entry->in_pack_type != OBJ_OFS_DELTA)
                return 0;
 
-       /*
-        * If the current object is at pack edge, take the depth the
-        * objects that depend on the current object into account --
-        * otherwise they would become too deep.
-        */
-       if (trg_entry->delta_child) {
-               if (max_depth <= trg_entry->delta_limit)
-                       return 0;
-               max_depth -= trg_entry->delta_limit;
-       }
+       /* Let's not bust the allowed depth. */
        if (src_entry->depth >= max_depth)
                return 0;
 
@@ -1359,9 +1338,17 @@ static int try_delta(struct unpacked *trg, struct unpacked *src,
        return 1;
 }
 
-static void progress_interval(int signum)
+static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
 {
-       progress_update = 1;
+       struct object_entry *child = me->delta_child;
+       unsigned int m = n;
+       while (child) {
+               unsigned int c = check_delta_limit(child, n + 1);
+               if (m < c)
+                       m = c;
+               child = child->delta_sibling;
+       }
+       return m;
 }
 
 static void find_deltas(struct object_entry **list, int window, int depth)
@@ -1370,6 +1357,7 @@ static void find_deltas(struct object_entry **list, int window, int depth)
        unsigned int array_size = window * sizeof(struct unpacked);
        struct unpacked *array;
        unsigned last_percent = 999;
+       int max_depth;
 
        if (!nr_objects)
                return;
@@ -1410,6 +1398,18 @@ static void find_deltas(struct object_entry **list, int window, int depth)
                n->data = NULL;
                n->entry = entry;
 
+               /*
+                * If the current object is at pack edge, take the depth the
+                * objects that depend on the current object into account
+                * otherwise they would become too deep.
+                */
+               max_depth = depth;
+               if (entry->delta_child) {
+                       max_depth -= check_delta_limit(entry, 0);
+                       if (max_depth <= 0)
+                               goto next;
+               }
+
                j = window;
                while (--j > 0) {
                        uint32_t other_idx = idx + j;
@@ -1419,9 +1419,10 @@ static void find_deltas(struct object_entry **list, int window, int depth)
                        m = array + other_idx;
                        if (!m->entry)
                                break;
-                       if (try_delta(n, m, depth) < 0)
+                       if (try_delta(n, m, max_depth) < 0)
                                break;
                }
+
                /* if we made n a delta, and if n is already at max
                 * depth, leaving it in the window is pointless.  we
                 * should evict it first.
@@ -1429,6 +1430,7 @@ static void find_deltas(struct object_entry **list, int window, int depth)
                if (entry->delta && depth <= entry->depth)
                        continue;
 
+               next:
                idx++;
                if (idx >= window)
                        idx = 0;
@@ -1506,6 +1508,11 @@ static int reuse_cached_pack(unsigned char *sha1)
        return 1;
 }
 
+static void progress_interval(int signum)
+{
+       progress_update = 1;
+}
+
 static void setup_progress_signal(void)
 {
        struct sigaction sa;
@@ -1567,9 +1574,7 @@ static void read_object_list_from_stdin(void)
 
 static void show_commit(struct commit *commit)
 {
-       unsigned hash = name_hash("");
-       add_preferred_base_object("", hash);
-       add_object_entry(commit->object.sha1, hash, 0);
+       add_object_entry(commit->object.sha1, 0, 0);
 }
 
 static void show_object(struct object_array_entry *p)