Move buffer_is_binary() to xdiff-interface.h
[gitweb.git] / builtin-pack-objects.c
index a4370bbee175ecb7ae95d6e8d93dea3c677c1b8c..ccb25f6a9c298bd977c5f55a84dd9a20a1e54e23 100644 (file)
@@ -1,5 +1,6 @@
 #include "builtin.h"
 #include "cache.h"
+#include "attr.h"
 #include "object.h"
 #include "blob.h"
 #include "commit.h"
@@ -35,14 +36,16 @@ struct object_entry {
        struct object_entry *delta_sibling; /* other deltified objects who
                                             * uses the same base as me
                                             */
+       void *delta_data;       /* cached delta (uncompressed) */
        unsigned long delta_size;       /* delta data size (uncompressed) */
        enum object_type type;
        enum object_type in_pack_type;  /* could be delta */
        unsigned char in_pack_header_size;
        unsigned char preferred_base; /* we do not pack this, but is available
-                                      * to be used as the base objectto delta
+                                      * to be used as the base object to delta
                                       * objects against.
                                       */
+       unsigned char no_try_delta;
 };
 
 /*
@@ -74,6 +77,10 @@ static struct progress progress_state;
 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
 static int pack_compression_seen;
 
+static unsigned long delta_cache_size = 0;
+static unsigned long max_delta_cache_size = 0;
+static unsigned long cache_max_small_delta_size = 1000;
+
 /*
  * The object names in objects array are hashed with this hashtable,
  * to help looking up the entry by object name.
@@ -403,24 +410,24 @@ static unsigned long write_object(struct sha1file *f,
                z_stream stream;
                unsigned long maxsize;
                void *out;
-               buf = read_sha1_file(entry->sha1, &type, &size);
-               if (!buf)
-                       die("unable to read %s", sha1_to_hex(entry->sha1));
-               if (size != entry->size)
-                       die("object %s size inconsistency (%lu vs %lu)",
-                           sha1_to_hex(entry->sha1), size, entry->size);
-               if (usable_delta) {
-                       buf = delta_against(buf, size, entry);
+               if (!usable_delta) {
+                       buf = read_sha1_file(entry->sha1, &obj_type, &size);
+                       if (!buf)
+                               die("unable to read %s", sha1_to_hex(entry->sha1));
+               } else if (entry->delta_data) {
                        size = entry->delta_size;
+                       buf = entry->delta_data;
+                       entry->delta_data = NULL;
                        obj_type = (allow_ofs_delta && entry->delta->offset) ?
                                OBJ_OFS_DELTA : OBJ_REF_DELTA;
                } else {
-                       /*
-                        * recover real object type in case
-                        * check_object() wanted to re-use a delta,
-                        * but we couldn't since base was in previous split pack
-                        */
-                       obj_type = type;
+                       buf = read_sha1_file(entry->sha1, &type, &size);
+                       if (!buf)
+                               die("unable to read %s", sha1_to_hex(entry->sha1));
+                       buf = delta_against(buf, size, entry);
+                       size = entry->delta_size;
+                       obj_type = (allow_ofs_delta && entry->delta->offset) ?
+                               OBJ_OFS_DELTA : OBJ_REF_DELTA;
                }
                /* compress the data to store and put compressed length in datalen */
                memset(&stream, 0, sizeof(stream));
@@ -860,6 +867,9 @@ static unsigned name_hash(const char *name)
        unsigned char c;
        unsigned hash = 0;
 
+       if (!name)
+               return 0;
+
        /*
         * This effectively just creates a sortable number from the
         * last sixteen non-whitespace characters. Last characters
@@ -873,13 +883,36 @@ static unsigned name_hash(const char *name)
        return hash;
 }
 
+static void setup_delta_attr_check(struct git_attr_check *check)
+{
+       static struct git_attr *attr_delta;
+
+       if (!attr_delta)
+               attr_delta = git_attr("delta", 5);
+
+       check[0].attr = attr_delta;
+}
+
+static int no_try_delta(const char *path)
+{
+       struct git_attr_check check[1];
+
+       setup_delta_attr_check(check);
+       if (git_checkattr(path, ARRAY_SIZE(check), check))
+               return 0;
+       if (ATTR_FALSE(check->value))
+               return 1;
+       return 0;
+}
+
 static int add_object_entry(const unsigned char *sha1, enum object_type type,
-                           unsigned hash, int exclude)
+                           const char *name, int exclude)
 {
        struct object_entry *entry;
        struct packed_git *p, *found_pack = NULL;
        off_t found_offset = 0;
        int ix;
+       unsigned hash = name_hash(name);
 
        ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
        if (ix >= 0) {
@@ -936,6 +969,9 @@ static int add_object_entry(const unsigned char *sha1, enum object_type type,
        if (progress)
                display_progress(&progress_state, nr_objects);
 
+       if (name && no_try_delta(name))
+               entry->no_try_delta = 1;
+
        return 1;
 }
 
@@ -1068,10 +1104,9 @@ static void add_pbase_object(struct tree_desc *tree,
                if (cmp < 0)
                        return;
                if (name[cmplen] != '/') {
-                       unsigned hash = name_hash(fullname);
                        add_object_entry(entry.sha1,
                                         S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
-                                        hash, 1);
+                                        fullname, 1);
                        return;
                }
                if (S_ISDIR(entry.mode)) {
@@ -1131,10 +1166,11 @@ static int check_pbase_path(unsigned hash)
        return 0;
 }
 
-static void add_preferred_base_object(const char *name, unsigned hash)
+static void add_preferred_base_object(const char *name)
 {
        struct pbase_tree *it;
        int cmplen;
+       unsigned hash = name_hash(name);
 
        if (!num_preferred_base || check_pbase_path(hash))
                return;
@@ -1142,7 +1178,7 @@ static void add_preferred_base_object(const char *name, unsigned hash)
        cmplen = name_cmp_len(name);
        for (it = pbase_tree; it; it = it->next) {
                if (cmplen == 0) {
-                       add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
+                       add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
                }
                else {
                        struct tree_desc tree;
@@ -1354,6 +1390,23 @@ struct unpacked {
        struct delta_index *index;
 };
 
+static int delta_cacheable(struct unpacked *trg, struct unpacked *src,
+                           unsigned long src_size, unsigned long trg_size,
+                           unsigned long delta_size)
+{
+       if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
+               return 0;
+
+       if (delta_size < cache_max_small_delta_size)
+               return 1;
+
+       /* cache delta, if objects are large enough compared to delta size */
+       if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
+               return 1;
+
+       return 0;
+}
+
 /*
  * We search for deltas _backwards_ in a list sorted by type and
  * by size, so that we see progressively smaller and smaller files.
@@ -1423,18 +1476,32 @@ static int try_delta(struct unpacked *trg, struct unpacked *src,
        }
        if (!src->index) {
                src->index = create_delta_index(src->data, src_size);
-               if (!src->index)
-                       die("out of memory");
+               if (!src->index) {
+                       static int warned = 0;
+                       if (!warned++)
+                               warning("suboptimal pack - out of memory");
+                       return 0;
+               }
        }
 
        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
        if (!delta_buf)
                return 0;
 
+       if (trg_entry->delta_data) {
+               delta_cache_size -= trg_entry->delta_size;
+               free(trg_entry->delta_data);
+       }
+       trg_entry->delta_data = 0;
        trg_entry->delta = src_entry;
        trg_entry->delta_size = delta_size;
        trg_entry->depth = src_entry->depth + 1;
-       free(delta_buf);
+
+       if (delta_cacheable(src, trg, src_size, trg_size, delta_size)) {
+               trg_entry->delta_data = xrealloc(delta_buf, delta_size);
+               delta_cache_size += trg_entry->delta_size;
+       } else
+               free(delta_buf);
        return 1;
 }
 
@@ -1484,6 +1551,10 @@ static void find_deltas(struct object_entry **list, int window, int depth)
 
                if (entry->size < 50)
                        continue;
+
+               if (entry->no_try_delta)
+                       continue;
+
                free_delta_index(n->index);
                n->index = NULL;
                free(n->data);
@@ -1576,6 +1647,14 @@ static int git_pack_config(const char *k, const char *v)
                pack_compression_seen = 1;
                return 0;
        }
+       if (!strcmp(k, "pack.deltacachesize")) {
+               max_delta_cache_size = git_config_int(k, v);
+               return 0;
+       }
+       if (!strcmp(k, "pack.deltacachelimit")) {
+               cache_max_small_delta_size = git_config_int(k, v);
+               return 0;
+       }
        return git_default_config(k, v);
 }
 
@@ -1583,7 +1662,6 @@ static void read_object_list_from_stdin(void)
 {
        char line[40 + 1 + PATH_MAX + 2];
        unsigned char sha1[20];
-       unsigned hash;
 
        for (;;) {
                if (!fgets(line, sizeof(line), stdin)) {
@@ -1606,22 +1684,20 @@ static void read_object_list_from_stdin(void)
                if (get_sha1_hex(line, sha1))
                        die("expected sha1, got garbage:\n %s", line);
 
-               hash = name_hash(line+41);
-               add_preferred_base_object(line+41, hash);
-               add_object_entry(sha1, 0, hash, 0);
+               add_preferred_base_object(line+41);
+               add_object_entry(sha1, 0, line+41, 0);
        }
 }
 
 static void show_commit(struct commit *commit)
 {
-       add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
+       add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
 }
 
 static void show_object(struct object_array_entry *p)
 {
-       unsigned hash = name_hash(p->name);
-       add_preferred_base_object(p->name, hash);
-       add_object_entry(p->item->sha1, p->item->type, hash, 0);
+       add_preferred_base_object(p->name);
+       add_object_entry(p->item->sha1, p->item->type, p->name, 0);
 }
 
 static void show_edge(struct commit *commit)