]> git.ipfire.org Git - thirdparty/git.git/blobdiff - fast-import.c
Use uintmax_t for marks in fast-import.
[thirdparty/git.git] / fast-import.c
index 3d991020057c549d7cfbdcb917853ced07c44582..ebffa7c904bc892dc086aa8a53257253764d2453 100644 (file)
@@ -20,6 +20,7 @@ Format of STDIN stream:
     'committer' sp name '<' email '>' ts tz lf
     commit_msg
     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
+    ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
     file_change*
     lf;
   commit_msg ::= data;
@@ -35,7 +36,12 @@ Format of STDIN stream:
     tag_msg;
   tag_msg ::= data;
 
-  reset_branch ::= 'reset' sp ref_str lf;
+  reset_branch ::= 'reset' sp ref_str lf
+    ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
+    lf;
+
+  checkpoint ::= 'checkpoint' lf
+    lf;
 
      # note: the first idnum in a stream should be 1 and subsequent
      # idnums should not have gaps between values as this will cause
@@ -107,8 +113,9 @@ Format of STDIN stream:
 struct object_entry
 {
        struct object_entry *next;
-       enum object_type type;
        unsigned long offset;
+       unsigned type : TYPE_BITS;
+       unsigned pack_id : 16;
        unsigned char sha1[20];
 };
 
@@ -133,9 +140,9 @@ struct last_object
 {
        void *data;
        unsigned long len;
+       unsigned long offset;
        unsigned int depth;
-       int no_free;
-       unsigned char sha1[20];
+       unsigned no_free:1;
 };
 
 struct mem_pool
@@ -202,19 +209,26 @@ struct dbuf
        size_t capacity;
 };
 
+struct hash_list
+{
+       struct hash_list *next;
+       unsigned char sha1[20];
+};
+
+/* Configured limits on output */
+static unsigned long max_depth = 10;
+static unsigned long max_packsize = -1;
+static uintmax_t max_objects = -1;
 
 /* Stats and misc. counters */
-static unsigned long max_depth = 10;
-static unsigned long alloc_count;
+static uintmax_t alloc_count;
+static uintmax_t object_count;
+static uintmax_t marks_set_count;
+static uintmax_t object_count_by_type[1 << TYPE_BITS];
+static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
+static uintmax_t delta_count_by_type[1 << TYPE_BITS];
 static unsigned long branch_count;
 static unsigned long branch_load_count;
-static unsigned long remap_count;
-static unsigned long object_count;
-static unsigned long duplicate_count;
-static unsigned long marks_set_count;
-static unsigned long object_count_by_type[9];
-static unsigned long duplicate_count_by_type[9];
-static unsigned long delta_count_by_type[9];
 
 /* Memory pools */
 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
@@ -227,13 +241,14 @@ static unsigned int atom_cnt;
 static struct atom_str **atom_table;
 
 /* The .pack file being generated */
+static const char *base_name;
+static unsigned int pack_id;
+static char *idx_name;
+static struct packed_git *pack_data;
+static struct packed_git **all_packs;
 static int pack_fd;
 static unsigned long pack_size;
 static unsigned char pack_sha1[20];
-static unsigned char* pack_base;
-static unsigned long pack_moff;
-static unsigned long pack_mlen = 128*1024*1024;
-static unsigned long page_size;
 
 /* Table of objects we've written. */
 static unsigned int object_entry_alloc = 5000;
@@ -266,12 +281,12 @@ static struct tag *last_tag;
 
 /* Input stream parsing */
 static struct strbuf command_buf;
-static unsigned long next_mark;
+static uintmax_t next_mark;
 static struct dbuf new_data;
 static FILE* branch_log;
 
 
-static void alloc_objects(int cnt)
+static void alloc_objects(unsigned int cnt)
 {
        struct object_entry_pool *b;
 
@@ -393,7 +408,7 @@ static void size_dbuf(struct dbuf *b, size_t maxlen)
        b->buffer = xmalloc(b->capacity);
 }
 
-static void insert_mark(unsigned long idnum, struct object_entry *oe)
+static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 {
        struct mark_set *s = marks;
        while ((idnum >> s->shift) >= 1024) {
@@ -403,7 +418,7 @@ static void insert_mark(unsigned long idnum, struct object_entry *oe)
                marks = s;
        }
        while (s->shift) {
-               unsigned long i = idnum >> s->shift;
+               uintmax_t i = idnum >> s->shift;
                idnum -= i << s->shift;
                if (!s->data.sets[i]) {
                        s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
@@ -416,14 +431,14 @@ static void insert_mark(unsigned long idnum, struct object_entry *oe)
        s->data.marked[idnum] = oe;
 }
 
-static struct object_entry* find_mark(unsigned long idnum)
+static struct object_entry* find_mark(uintmax_t idnum)
 {
-       unsigned long orig_idnum = idnum;
+       uintmax_t orig_idnum = idnum;
        struct mark_set *s = marks;
        struct object_entry *oe = NULL;
        if ((idnum >> s->shift) < 1024) {
                while (s && s->shift) {
-                       unsigned long i = idnum >> s->shift;
+                       uintmax_t i = idnum >> s->shift;
                        idnum -= i << s->shift;
                        s = s->data.sets[i];
                }
@@ -431,7 +446,7 @@ static struct object_entry* find_mark(unsigned long idnum)
                        oe = s->data.marked[idnum];
        }
        if (!oe)
-               die("mark :%lu not declared", orig_idnum);
+               die("mark :%ju not declared", orig_idnum);
        return oe;
 }
 
@@ -478,6 +493,8 @@ static struct branch* new_branch(const char *name)
        b = pool_calloc(1, sizeof(struct branch));
        b->name = pool_strdup(name);
        b->table_next_branch = branch_table[hc];
+       b->branch_tree.versions[0].mode = S_IFDIR;
+       b->branch_tree.versions[1].mode = S_IFDIR;
        branch_table[hc] = b;
        branch_count++;
        return b;
@@ -586,17 +603,155 @@ static void yread(int fd, void *buffer, size_t length)
        }
 }
 
-static void ywrite(int fd, void *buffer, size_t length)
+static void start_packfile()
 {
-       ssize_t ret = 0;
-       while (ret < length) {
-               ssize_t size = xwrite(fd, (char *) buffer + ret, length - ret);
-               if (!size)
-                       die("Write to descriptor %i: end of file", fd);
-               if (size < 0)
-                       die("Write to descriptor %i: %s", fd, strerror(errno));
-               ret += size;
+       struct packed_git *p;
+       struct pack_header hdr;
+
+       idx_name = xmalloc(strlen(base_name) + 11);
+       p = xcalloc(1, sizeof(*p) + strlen(base_name) + 13);
+       sprintf(p->pack_name, "%s%5.5i.pack", base_name, pack_id + 1);
+       sprintf(idx_name, "%s%5.5i.idx", base_name, pack_id + 1);
+
+       pack_fd = open(p->pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
+       if (pack_fd < 0)
+               die("Can't create %s: %s", p->pack_name, strerror(errno));
+       p->pack_fd = pack_fd;
+
+       hdr.hdr_signature = htonl(PACK_SIGNATURE);
+       hdr.hdr_version = htonl(2);
+       hdr.hdr_entries = 0;
+       write_or_die(pack_fd, &hdr, sizeof(hdr));
+
+       pack_data = p;
+       pack_size = sizeof(hdr);
+       object_count = 0;
+
+       all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
+       all_packs[pack_id] = p;
+}
+
+static void fixup_header_footer()
+{
+       SHA_CTX c;
+       char hdr[8];
+       unsigned long cnt;
+       char *buf;
+
+       if (lseek(pack_fd, 0, SEEK_SET) != 0)
+               die("Failed seeking to start: %s", strerror(errno));
+
+       SHA1_Init(&c);
+       yread(pack_fd, hdr, 8);
+       SHA1_Update(&c, hdr, 8);
+
+       cnt = htonl(object_count);
+       SHA1_Update(&c, &cnt, 4);
+       write_or_die(pack_fd, &cnt, 4);
+
+       buf = xmalloc(128 * 1024);
+       for (;;) {
+               size_t n = xread(pack_fd, buf, 128 * 1024);
+               if (n <= 0)
+                       break;
+               SHA1_Update(&c, buf, n);
+       }
+       free(buf);
+
+       SHA1_Final(pack_sha1, &c);
+       write_or_die(pack_fd, pack_sha1, sizeof(pack_sha1));
+}
+
+static int oecmp (const void *a_, const void *b_)
+{
+       struct object_entry *a = *((struct object_entry**)a_);
+       struct object_entry *b = *((struct object_entry**)b_);
+       return hashcmp(a->sha1, b->sha1);
+}
+
+static void write_index(const char *idx_name)
+{
+       struct sha1file *f;
+       struct object_entry **idx, **c, **last, *e;
+       struct object_entry_pool *o;
+       unsigned int array[256];
+       int i;
+
+       /* Build the sorted table of object IDs. */
+       idx = xmalloc(object_count * sizeof(struct object_entry*));
+       c = idx;
+       for (o = blocks; o; o = o->next_pool)
+               for (e = o->next_free; e-- != o->entries;)
+                       if (pack_id == e->pack_id)
+                               *c++ = e;
+       last = idx + object_count;
+       if (c != last)
+               die("internal consistency error creating the index");
+       qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
+
+       /* Generate the fan-out array. */
+       c = idx;
+       for (i = 0; i < 256; i++) {
+               struct object_entry **next = c;;
+               while (next < last) {
+                       if ((*next)->sha1[0] != i)
+                               break;
+                       next++;
+               }
+               array[i] = htonl(next - idx);
+               c = next;
+       }
+
+       f = sha1create("%s", idx_name);
+       sha1write(f, array, 256 * sizeof(int));
+       for (c = idx; c != last; c++) {
+               unsigned int offset = htonl((*c)->offset);
+               sha1write(f, &offset, 4);
+               sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
+       }
+       sha1write(f, pack_sha1, sizeof(pack_sha1));
+       sha1close(f, NULL, 1);
+       free(idx);
+}
+
+static void end_packfile()
+{
+       struct packed_git *old_p = pack_data, *new_p;
+
+       if (object_count) {
+               fixup_header_footer();
+               write_index(idx_name);
+               fprintf(stdout, "%s\n", old_p->pack_name);
+               fflush(stdout);
+
+               /* Register the packfile with core git's machinary. */
+               new_p = add_packed_git(idx_name, strlen(idx_name), 1);
+               if (!new_p)
+                       die("core git rejected index %s", idx_name);
+               new_p->windows = old_p->windows;
+               new_p->pack_fd = old_p->pack_fd;
+               all_packs[pack_id++] = new_p;
+               install_packed_git(new_p);
        }
+       else {
+               close(pack_fd);
+               unlink(old_p->pack_name);
+       }
+       free(old_p);
+       free(idx_name);
+
+       /* We can't carry a delta across packfiles. */
+       free(last_blob.data);
+       last_blob.data = NULL;
+       last_blob.len = 0;
+       last_blob.offset = 0;
+       last_blob.depth = 0;
+}
+
+static void checkpoint()
+{
+       end_packfile();
+       start_packfile();
 }
 
 static size_t encode_header(
@@ -607,7 +762,7 @@ static size_t encode_header(
        int n = 1;
        unsigned char c;
 
-       if (type < OBJ_COMMIT || type > OBJ_DELTA)
+       if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
                die("bad type %d", type);
 
        c = (type << 4) | (size & 15);
@@ -628,7 +783,7 @@ static int store_object(
        size_t datlen,
        struct last_object *last,
        unsigned char *sha1out,
-       unsigned long mark)
+       uintmax_t mark)
 {
        void *out, *delta;
        struct object_entry *e;
@@ -650,51 +805,94 @@ static int store_object(
        if (mark)
                insert_mark(mark, e);
        if (e->offset) {
-               duplicate_count++;
                duplicate_count_by_type[type]++;
                return 1;
        }
-       e->type = type;
-       e->offset = pack_size;
-       object_count++;
-       object_count_by_type[type]++;
 
-       if (last && last->data && last->depth < max_depth)
+       if (last && last->data && last->depth < max_depth) {
                delta = diff_delta(last->data, last->len,
                        dat, datlen,
                        &deltalen, 0);
-       else
-               delta = 0;
+               if (delta && deltalen >= datlen) {
+                       free(delta);
+                       delta = NULL;
+               }
+       } else
+               delta = NULL;
 
        memset(&s, 0, sizeof(s));
        deflateInit(&s, zlib_compression_level);
-
        if (delta) {
-               delta_count_by_type[type]++;
-               last->depth++;
                s.next_in = delta;
                s.avail_in = deltalen;
-               hdrlen = encode_header(OBJ_DELTA, deltalen, hdr);
-               ywrite(pack_fd, hdr, hdrlen);
-               ywrite(pack_fd, last->sha1, sizeof(sha1));
-               pack_size += hdrlen + sizeof(sha1);
        } else {
-               if (last)
-                       last->depth = 0;
                s.next_in = dat;
                s.avail_in = datlen;
-               hdrlen = encode_header(type, datlen, hdr);
-               ywrite(pack_fd, hdr, hdrlen);
-               pack_size += hdrlen;
        }
-
        s.avail_out = deflateBound(&s, s.avail_in);
        s.next_out = out = xmalloc(s.avail_out);
        while (deflate(&s, Z_FINISH) == Z_OK)
                /* nothing */;
        deflateEnd(&s);
 
-       ywrite(pack_fd, out, s.total_out);
+       /* Determine if we should auto-checkpoint. */
+       if ((object_count + 1) > max_objects
+               || (object_count + 1) < object_count
+               || (pack_size + 60 + s.total_out) > max_packsize
+               || (pack_size + 60 + s.total_out) < pack_size) {
+
+               /* This new object needs to *not* have the current pack_id. */
+               e->pack_id = pack_id + 1;
+               checkpoint();
+
+               /* We cannot carry a delta into the new pack. */
+               if (delta) {
+                       free(delta);
+                       delta = NULL;
+
+                       memset(&s, 0, sizeof(s));
+                       deflateInit(&s, zlib_compression_level);
+                       s.next_in = dat;
+                       s.avail_in = datlen;
+                       s.avail_out = deflateBound(&s, s.avail_in);
+                       s.next_out = out = xrealloc(out, s.avail_out);
+                       while (deflate(&s, Z_FINISH) == Z_OK)
+                               /* nothing */;
+                       deflateEnd(&s);
+               }
+       }
+
+       e->type = type;
+       e->pack_id = pack_id;
+       e->offset = pack_size;
+       object_count++;
+       object_count_by_type[type]++;
+
+       if (delta) {
+               unsigned long ofs = e->offset - last->offset;
+               unsigned pos = sizeof(hdr) - 1;
+
+               delta_count_by_type[type]++;
+               last->depth++;
+
+               hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
+               write_or_die(pack_fd, hdr, hdrlen);
+               pack_size += hdrlen;
+
+               hdr[pos] = ofs & 127;
+               while (ofs >>= 7)
+                       hdr[--pos] = 128 | (--ofs & 127);
+               write_or_die(pack_fd, hdr + pos, sizeof(hdr) - pos);
+               pack_size += sizeof(hdr) - pos;
+       } else {
+               if (last)
+                       last->depth = 0;
+               hdrlen = encode_header(type, datlen, hdr);
+               write_or_die(pack_fd, hdr, hdrlen);
+               pack_size += hdrlen;
+       }
+
+       write_or_die(pack_fd, out, s.total_out);
        pack_size += s.total_out;
 
        free(out);
@@ -704,139 +902,21 @@ static int store_object(
                if (last->data && !last->no_free)
                        free(last->data);
                last->data = dat;
+               last->offset = e->offset;
                last->len = datlen;
-               hashcpy(last->sha1, sha1);
        }
        return 0;
 }
 
-static unsigned char* map_pack(unsigned long offset, unsigned int *left)
-{
-       if (offset >= pack_size)
-               die("object offset outside of pack file");
-       if (!pack_base
-                       || offset < pack_moff
-                       || (offset + 20) >= (pack_moff + pack_mlen)) {
-               if (pack_base)
-                       munmap(pack_base, pack_mlen);
-               pack_moff = (offset / page_size) * page_size;
-               pack_base = mmap(NULL,pack_mlen,PROT_READ,MAP_SHARED,
-                       pack_fd,pack_moff);
-               if (pack_base == MAP_FAILED)
-                       die("Failed to map generated pack: %s", strerror(errno));
-               remap_count++;
-       }
-       offset -= pack_moff;
-       if (left)
-               *left = pack_mlen - offset;
-       return pack_base + offset;
-}
-
-static unsigned long unpack_object_header(unsigned long offset,
-       enum object_type *type,
+static void *gfi_unpack_entry(
+       struct object_entry *oe,
        unsigned long *sizep)
 {
-       unsigned shift;
-       unsigned char c;
-       unsigned long size;
-
-       c = *map_pack(offset++, NULL);
-       *type = (c >> 4) & 7;
-       size = c & 15;
-       shift = 4;
-       while (c & 0x80) {
-               c = *map_pack(offset++, NULL);
-               size += (c & 0x7f) << shift;
-               shift += 7;
-       }
-       *sizep = size;
-       return offset;
-}
-
-static void *unpack_non_delta_entry(unsigned long o, unsigned long sz)
-{
-       z_stream stream;
-       unsigned char *result;
-
-       result = xmalloc(sz + 1);
-       result[sz] = 0;
-
-       memset(&stream, 0, sizeof(stream));
-       stream.next_in = map_pack(o, &stream.avail_in);
-       stream.next_out = result;
-       stream.avail_out = sz;
-
-       inflateInit(&stream);
-       for (;;) {
-               int st = inflate(&stream, Z_FINISH);
-               if (st == Z_STREAM_END)
-                       break;
-               if (st == Z_OK || st == Z_BUF_ERROR) {
-                       o = stream.next_in - pack_base + pack_moff;
-                       stream.next_in = map_pack(o, &stream.avail_in);
-                       continue;
-               }
-               die("Error %i from zlib during inflate.", st);
-       }
-       inflateEnd(&stream);
-       if (stream.total_out != sz)
-               die("Error after inflate: sizes mismatch");
-       return result;
-}
-
-static void *unpack_entry(unsigned long offset,
-       unsigned long *sizep,
-       unsigned int *delta_depth);
-
-static void *unpack_delta_entry(unsigned long offset,
-       unsigned long delta_size,
-       unsigned long *sizep,
-       unsigned int *delta_depth)
-{
-       struct object_entry *base_oe;
-       unsigned char *base_sha1;
-       void *delta_data, *base, *result;
-       unsigned long base_size, result_size;
-
-       base_sha1 = map_pack(offset, NULL);
-       base_oe = find_object(base_sha1);
-       if (!base_oe)
-               die("I'm broken; I can't find a base I know must be here.");
-       base = unpack_entry(base_oe->offset, &base_size, delta_depth);
-       delta_data = unpack_non_delta_entry(offset + 20, delta_size);
-       result = patch_delta(base, base_size,
-                            delta_data, delta_size,
-                            &result_size);
-       if (!result)
-               die("failed to apply delta");
-       free(delta_data);
-       free(base);
-       *sizep = result_size;
-       (*delta_depth)++;
-       return result;
-}
-
-static void *unpack_entry(unsigned long offset,
-       unsigned long *sizep,
-       unsigned int *delta_depth)
-{
-       unsigned long size;
-       enum object_type kind;
-
-       offset = unpack_object_header(offset, &kind, &size);
-       switch (kind) {
-       case OBJ_DELTA:
-               return unpack_delta_entry(offset, size, sizep, delta_depth);
-       case OBJ_COMMIT:
-       case OBJ_TREE:
-       case OBJ_BLOB:
-       case OBJ_TAG:
-               *sizep = size;
-               *delta_depth = 0;
-               return unpack_non_delta_entry(offset, size);
-       default:
-               die("I created an object I can't read!");
-       }
+       static char type[20];
+       struct packed_git *p = all_packs[oe->pack_id];
+       if (p == pack_data)
+               p->pack_size = pack_size + 20;
+       return unpack_entry(p, oe->offset, type, sizep);
 }
 
 static const char *get_mode(const char *str, unsigned int *modep)
@@ -870,7 +950,8 @@ static void load_tree(struct tree_entry *root)
        if (myoe) {
                if (myoe->type != OBJ_TREE)
                        die("Not a tree: %s", sha1_to_hex(sha1));
-               buf = unpack_entry(myoe->offset, &size, &t->delta_depth);
+               t->delta_depth = 0;
+               buf = gfi_unpack_entry(myoe, &size);
        } else {
                char type[20];
                buf = read_sha1_file(sha1, type, &size);
@@ -959,6 +1040,7 @@ static void store_tree(struct tree_entry *root)
        unsigned int i, j, del;
        unsigned long new_len;
        struct last_object lo;
+       struct object_entry *le;
 
        if (!is_null_sha1(root->versions[1].sha1))
                return;
@@ -968,24 +1050,25 @@ static void store_tree(struct tree_entry *root)
                        store_tree(t->entries[i]);
        }
 
-       if (is_null_sha1(root->versions[0].sha1)
-                       || !find_object(root->versions[0].sha1)) {
+       le = find_object(root->versions[0].sha1);
+       if (!S_ISDIR(root->versions[0].mode)
+               || !le
+               || le->pack_id != pack_id) {
                lo.data = NULL;
                lo.depth = 0;
        } else {
                mktree(t, 0, &lo.len, &old_tree);
                lo.data = old_tree.buffer;
+               lo.offset = le->offset;
                lo.depth = t->delta_depth;
                lo.no_free = 1;
-               hashcpy(lo.sha1, root->versions[0].sha1);
        }
-       mktree(t, 1, &new_len, &new_tree);
 
+       mktree(t, 1, &new_len, &new_tree);
        store_object(OBJ_TREE, new_tree.buffer, new_len,
                &lo, root->versions[1].sha1, 0);
 
        t->delta_depth = lo.depth;
-       hashcpy(root->versions[0].sha1, root->versions[1].sha1);
        for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
                struct tree_entry *e = t->entries[i];
                if (e->versions[1].mode) {
@@ -1088,10 +1171,13 @@ static int tree_content_remove(struct tree_entry *root, const char *p)
                        if (!e->tree)
                                load_tree(e);
                        if (tree_content_remove(e, slash1 + 1)) {
-                               if (!e->tree->entry_count)
-                                       goto del_entry;
-                               hashclr(root->versions[1].sha1);
-                               return 1;
+                               for (n = 0; n < e->tree->entry_count; n++) {
+                                       if (e->tree->entries[n]->versions[1].mode) {
+                                               hashclr(root->versions[1].sha1);
+                                               return 1;
+                                       }
+                               }
+                               goto del_entry;
                        }
                        return 0;
                }
@@ -1109,100 +1195,6 @@ del_entry:
        return 1;
 }
 
-static void init_pack_header()
-{
-       struct pack_header hdr;
-
-       hdr.hdr_signature = htonl(PACK_SIGNATURE);
-       hdr.hdr_version = htonl(2);
-       hdr.hdr_entries = 0;
-
-       ywrite(pack_fd, &hdr, sizeof(hdr));
-       pack_size = sizeof(hdr);
-}
-
-static void fixup_header_footer()
-{
-       SHA_CTX c;
-       char hdr[8];
-       unsigned long cnt;
-       char *buf;
-       size_t n;
-
-       if (lseek(pack_fd, 0, SEEK_SET) != 0)
-               die("Failed seeking to start: %s", strerror(errno));
-
-       SHA1_Init(&c);
-       yread(pack_fd, hdr, 8);
-       SHA1_Update(&c, hdr, 8);
-
-       cnt = htonl(object_count);
-       SHA1_Update(&c, &cnt, 4);
-       ywrite(pack_fd, &cnt, 4);
-
-       buf = xmalloc(128 * 1024);
-       for (;;) {
-               n = xread(pack_fd, buf, 128 * 1024);
-               if (n <= 0)
-                       break;
-               SHA1_Update(&c, buf, n);
-       }
-       free(buf);
-
-       SHA1_Final(pack_sha1, &c);
-       ywrite(pack_fd, pack_sha1, sizeof(pack_sha1));
-}
-
-static int oecmp (const void *_a, const void *_b)
-{
-       struct object_entry *a = *((struct object_entry**)_a);
-       struct object_entry *b = *((struct object_entry**)_b);
-       return hashcmp(a->sha1, b->sha1);
-}
-
-static void write_index(const char *idx_name)
-{
-       struct sha1file *f;
-       struct object_entry **idx, **c, **last;
-       struct object_entry *e;
-       struct object_entry_pool *o;
-       unsigned int array[256];
-       int i;
-
-       /* Build the sorted table of object IDs. */
-       idx = xmalloc(object_count * sizeof(struct object_entry*));
-       c = idx;
-       for (o = blocks; o; o = o->next_pool)
-               for (e = o->entries; e != o->next_free; e++)
-                       *c++ = e;
-       last = idx + object_count;
-       qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
-
-       /* Generate the fan-out array. */
-       c = idx;
-       for (i = 0; i < 256; i++) {
-               struct object_entry **next = c;;
-               while (next < last) {
-                       if ((*next)->sha1[0] != i)
-                               break;
-                       next++;
-               }
-               array[i] = htonl(next - idx);
-               c = next;
-       }
-
-       f = sha1create("%s", idx_name);
-       sha1write(f, array, 256 * sizeof(int));
-       for (c = idx; c != last; c++) {
-               unsigned int offset = htonl((*c)->offset);
-               sha1write(f, &offset, 4);
-               sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
-       }
-       sha1write(f, pack_sha1, sizeof(pack_sha1));
-       sha1close(f, NULL, 1);
-       free(idx);
-}
-
 static void dump_branches()
 {
        static const char *msg = "fast-import";
@@ -1212,7 +1204,7 @@ static void dump_branches()
 
        for (i = 0; i < branch_table_sz; i++) {
                for (b = branch_table[i]; b; b = b->table_next_branch) {
-                       lock = lock_any_ref_for_update(b->name, NULL, 0);
+                       lock = lock_any_ref_for_update(b->name, NULL);
                        if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
                                die("Can't write %s", b->name);
                }
@@ -1228,17 +1220,17 @@ static void dump_tags()
 
        for (t = first_tag; t; t = t->next_tag) {
                sprintf(path, "refs/tags/%s", t->name);
-               lock = lock_any_ref_for_update(path, NULL, 0);
+               lock = lock_any_ref_for_update(path, NULL);
                if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
                        die("Can't write %s", path);
        }
 }
 
 static void dump_marks_helper(FILE *f,
-       unsigned long base,
+       uintmax_t base,
        struct mark_set *m)
 {
-       int k;
+       uintmax_t k;
        if (m->shift) {
                for (k = 0; k < 1024; k++) {
                        if (m->data.sets[k])
@@ -1248,7 +1240,7 @@ static void dump_marks_helper(FILE *f,
        } else {
                for (k = 0; k < 1024; k++) {
                        if (m->data.marked[k])
-                               fprintf(f, ":%lu %s\n", base + k,
+                               fprintf(f, ":%ju %s\n", base + k,
                                        sha1_to_hex(m->data.marked[k]->sha1));
                }
        }
@@ -1272,7 +1264,7 @@ static void read_next_command()
 static void cmd_mark()
 {
        if (!strncmp("mark :", command_buf.buf, 6)) {
-               next_mark = strtoul(command_buf.buf + 6, NULL, 10);
+               next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
                read_next_command();
        }
        else
@@ -1385,7 +1377,8 @@ static void file_change_m(struct branch *b)
 
        if (*p == ':') {
                char *x;
-               oe = find_mark(strtoul(p + 1, &x, 10));
+               oe = find_mark(strtoumax(p + 1, &x, 10));
+               hashcpy(sha1, oe->sha1);
                p = x;
        } else {
                if (get_sha1_hex(p, sha1))
@@ -1467,15 +1460,14 @@ static void cmd_from(struct branch *b)
                hashcpy(b->branch_tree.versions[0].sha1, t);
                hashcpy(b->branch_tree.versions[1].sha1, t);
        } else if (*from == ':') {
-               unsigned long idnum = strtoul(from + 1, NULL, 10);
+               uintmax_t idnum = strtoumax(from + 1, NULL, 10);
                struct object_entry *oe = find_mark(idnum);
                unsigned long size;
-               unsigned int depth;
                char *buf;
                if (oe->type != OBJ_COMMIT)
-                       die("Mark :%lu not a commit", idnum);
+                       die("Mark :%ju not a commit", idnum);
                hashcpy(b->sha1, oe->sha1);
-               buf = unpack_entry(oe->offset, &size, &depth);
+               buf = gfi_unpack_entry(oe, &size);
                if (!buf || size < 46)
                        die("Not a valid commit: %s", from);
                if (memcmp("tree ", buf, 5)
@@ -1509,6 +1501,48 @@ static void cmd_from(struct branch *b)
        read_next_command();
 }
 
+static struct hash_list* cmd_merge(unsigned int *count)
+{
+       struct hash_list *list = NULL, *n, *e;
+       const char *from, *endp;
+       char *str_uq;
+       struct branch *s;
+
+       *count = 0;
+       while (!strncmp("merge ", command_buf.buf, 6)) {
+               from = strchr(command_buf.buf, ' ') + 1;
+               str_uq = unquote_c_style(from, &endp);
+               if (str_uq) {
+                       if (*endp)
+                               die("Garbage after string in: %s", command_buf.buf);
+                       from = str_uq;
+               }
+
+               n = xmalloc(sizeof(*n));
+               s = lookup_branch(from);
+               if (s)
+                       hashcpy(n->sha1, s->sha1);
+               else if (*from == ':') {
+                       uintmax_t idnum = strtoumax(from + 1, NULL, 10);
+                       struct object_entry *oe = find_mark(idnum);
+                       if (oe->type != OBJ_COMMIT)
+                               die("Mark :%ju not a commit", idnum);
+                       hashcpy(n->sha1, oe->sha1);
+               } else if (get_sha1(from, n->sha1))
+                       die("Invalid ref name or SHA1 expression: %s", from);
+
+               n->next = NULL;
+               if (list)
+                       e->next = n;
+               else
+                       list = n;
+               e = n;
+               *count++;
+               read_next_command();
+       }
+       return list;
+}
+
 static void cmd_new_commit()
 {
        struct branch *b;
@@ -1519,6 +1553,8 @@ static void cmd_new_commit()
        char *sp;
        char *author = NULL;
        char *committer = NULL;
+       struct hash_list *merge_list = NULL;
+       unsigned int merge_count;
 
        /* Obtain the branch name from the rest of our command */
        sp = strchr(command_buf.buf, ' ') + 1;
@@ -1549,6 +1585,7 @@ static void cmd_new_commit()
        msg = cmd_data(&msglen);
        read_next_command();
        cmd_from(b);
+       merge_list = cmd_merge(&merge_count);
 
        /* ensure the branch is active/loaded */
        if (!b->branch_tree.tree || !max_active_branches) {
@@ -1571,7 +1608,10 @@ static void cmd_new_commit()
 
        /* build the tree and the commit */
        store_tree(&b->branch_tree);
+       hashcpy(b->branch_tree.versions[0].sha1,
+               b->branch_tree.versions[1].sha1);
        size_dbuf(&new_data, 97 + msglen
+               + merge_count * 49
                + (author
                        ? strlen(author) + strlen(committer)
                        : 2 * strlen(committer)));
@@ -1580,6 +1620,12 @@ static void cmd_new_commit()
                sha1_to_hex(b->branch_tree.versions[1].sha1));
        if (!is_null_sha1(b->sha1))
                sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
+       while (merge_list) {
+               struct hash_list *next = merge_list->next;
+               sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
+               free(merge_list);
+               merge_list = next;
+       }
        if (author)
                sp += sprintf(sp, "%s\n", author);
        else
@@ -1606,7 +1652,7 @@ static void cmd_new_commit()
                        fputc('"', branch_log);
                } else
                        fprintf(branch_log, "%s", b->name);
-               fprintf(branch_log," :%lu %s\n",next_mark,sha1_to_hex(b->sha1));
+               fprintf(branch_log," :%ju %s\n",next_mark,sha1_to_hex(b->sha1));
        }
 }
 
@@ -1621,7 +1667,7 @@ static void cmd_new_tag()
        void *msg;
        size_t msglen;
        struct tag *t;
-       unsigned long from_mark = 0;
+       uintmax_t from_mark = 0;
        unsigned char sha1[20];
 
        /* Obtain the new tag name from the rest of our command */
@@ -1660,10 +1706,10 @@ static void cmd_new_tag()
        if (s) {
                hashcpy(sha1, s->sha1);
        } else if (*from == ':') {
-               from_mark = strtoul(from + 1, NULL, 10);
+               from_mark = strtoumax(from + 1, NULL, 10);
                struct object_entry *oe = find_mark(from_mark);
                if (oe->type != OBJ_COMMIT)
-                       die("Mark :%lu not a commit", from_mark);
+                       die("Mark :%ju not a commit", from_mark);
                hashcpy(sha1, oe->sha1);
        } else if (!get_sha1(from, sha1)) {
                unsigned long size;
@@ -1714,7 +1760,7 @@ static void cmd_new_tag()
                        fputc('"', branch_log);
                } else
                        fprintf(branch_log, "%s", t->name);
-               fprintf(branch_log," :%lu %s\n",from_mark,sha1_to_hex(t->sha1));
+               fprintf(branch_log," :%ju %s\n",from_mark,sha1_to_hex(t->sha1));
        }
 }
 
@@ -1741,8 +1787,19 @@ static void cmd_reset_branch()
                        b->branch_tree.tree = NULL;
                }
        }
+       else
+               b = new_branch(sp);
        if (str_uq)
                free(str_uq);
+       read_next_command();
+       cmd_from(b);
+}
+
+static void cmd_checkpoint()
+{
+       if (object_count)
+               checkpoint();
+       read_next_command();
 }
 
 static const char fast_import_usage[] =
@@ -1750,16 +1807,12 @@ static const char fast_import_usage[] =
 
 int main(int argc, const char **argv)
 {
-       const char *base_name;
        int i;
-       unsigned long est_obj_cnt = object_entry_alloc;
-       char *pack_name;
-       char *idx_name;
-       struct stat sb;
+       uintmax_t est_obj_cnt = object_entry_alloc;
+       uintmax_t duplicate_count;
 
        setup_ident();
        git_config(git_default_config);
-       page_size = getpagesize();
 
        for (i = 1; i < argc; i++) {
                const char *a = argv[i];
@@ -1767,7 +1820,11 @@ int main(int argc, const char **argv)
                if (*a != '-' || !strcmp(a, "--"))
                        break;
                else if (!strncmp(a, "--objects=", 10))
-                       est_obj_cnt = strtoul(a + 10, NULL, 0);
+                       est_obj_cnt = strtoumax(a + 10, NULL, 0);
+               else if (!strncmp(a, "--max-objects-per-pack=", 23))
+                       max_objects = strtoumax(a + 23, NULL, 0);
+               else if (!strncmp(a, "--max-pack-size=", 16))
+                       max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
                else if (!strncmp(a, "--depth=", 8))
                        max_depth = strtoul(a + 8, NULL, 0);
                else if (!strncmp(a, "--active-branches=", 18))
@@ -1786,16 +1843,6 @@ int main(int argc, const char **argv)
                usage(fast_import_usage);
        base_name = argv[i];
 
-       pack_name = xmalloc(strlen(base_name) + 6);
-       sprintf(pack_name, "%s.pack", base_name);
-       idx_name = xmalloc(strlen(base_name) + 5);
-       sprintf(idx_name, "%s.idx", base_name);
-
-       pack_fd = open(pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
-       if (pack_fd < 0)
-               die("Can't create %s: %s", pack_name, strerror(errno));
-
-       init_pack_header();
        alloc_objects(est_obj_cnt);
        strbuf_init(&command_buf);
 
@@ -1804,6 +1851,7 @@ int main(int argc, const char **argv)
        avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
        marks = pool_calloc(1, sizeof(struct mark_set));
 
+       start_packfile();
        for (;;) {
                read_next_command();
                if (command_buf.eof)
@@ -1816,41 +1864,40 @@ int main(int argc, const char **argv)
                        cmd_new_tag();
                else if (!strncmp("reset ", command_buf.buf, 6))
                        cmd_reset_branch();
+               else if (!strcmp("checkpoint", command_buf.buf))
+                       cmd_checkpoint();
                else
                        die("Unsupported command: %s", command_buf.buf);
        }
+       end_packfile();
 
-       fixup_header_footer();
-       close(pack_fd);
-       write_index(idx_name);
        dump_branches();
        dump_tags();
        dump_marks();
        if (branch_log)
                fclose(branch_log);
 
+       duplicate_count = 0;
+       for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
+               duplicate_count += duplicate_count_by_type[i];
+
        fprintf(stderr, "%s statistics:\n", argv[0]);
-       fprintf(stderr, "---------------------------------------------------\n");
-       fprintf(stderr, "Alloc'd objects: %10lu (%10lu overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
-       fprintf(stderr, "Total objects:   %10lu (%10lu duplicates)\n", object_count, duplicate_count);
-       fprintf(stderr, "      blobs  :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
-       fprintf(stderr, "      trees  :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
-       fprintf(stderr, "      commits:   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
-       fprintf(stderr, "      tags   :   %10lu (%10lu duplicates %10lu deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
+       fprintf(stderr, "---------------------------------------------------------------------\n");
+       fprintf(stderr, "Alloc'd objects: %10ju (%10ju overflow  )\n", alloc_count, alloc_count - est_obj_cnt);
+       fprintf(stderr, "Total objects:   %10ju (%10ju duplicates                  )\n", object_count, duplicate_count);
+       fprintf(stderr, "      blobs  :   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
+       fprintf(stderr, "      trees  :   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
+       fprintf(stderr, "      commits:   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
+       fprintf(stderr, "      tags   :   %10ju (%10ju duplicates %10ju deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
        fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
-       fprintf(stderr, "      marks:     %10u (%10lu unique    )\n", (1 << marks->shift) * 1024, marks_set_count);
+       fprintf(stderr, "      marks:     %10ju (%10ju unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
        fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
-       fprintf(stderr, "Memory total:    %10lu KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
+       fprintf(stderr, "Memory total:    %10ju KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
        fprintf(stderr, "       pools:    %10lu KiB\n", total_allocd/1024);
-       fprintf(stderr, "     objects:    %10lu KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
-       fprintf(stderr, "Pack remaps:     %10lu\n", remap_count);
-       fprintf(stderr, "---------------------------------------------------\n");
-
-       stat(pack_name, &sb);
-       fprintf(stderr, "Pack size:       %10lu KiB\n", (unsigned long)(sb.st_size/1024));
-       stat(idx_name, &sb);
-       fprintf(stderr, "Index size:      %10lu KiB\n", (unsigned long)(sb.st_size/1024));
-
+       fprintf(stderr, "     objects:    %10ju KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
+       fprintf(stderr, "---------------------------------------------------------------------\n");
+       pack_report();
+       fprintf(stderr, "---------------------------------------------------------------------\n");
        fprintf(stderr, "\n");
 
        return 0;