Document gitattributes(5)
[gitweb.git] / fast-import.c
index 3856c87c4e03a91ad0cea36e5a9d00676e80833a..cdd629d6bc547609daabab6c1f3b73939fddb7b9 100644 (file)
+/*
+Format of STDIN stream:
+
+  stream ::= cmd*;
+
+  cmd ::= new_blob
+        | new_commit
+        | new_tag
+        | reset_branch
+        | checkpoint
+        ;
+
+  new_blob ::= 'blob' lf
+       mark?
+    file_content;
+  file_content ::= data;
+
+  new_commit ::= 'commit' sp ref_str lf
+    mark?
+    ('author' sp name '<' email '>' when lf)?
+    'committer' sp name '<' email '>' when 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;
+
+  file_change ::= file_clr | file_del | file_obm | file_inm;
+  file_clr ::= 'deleteall' lf;
+  file_del ::= 'D' sp path_str lf;
+  file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
+  file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
+    data;
+
+  new_tag ::= 'tag' sp tag_str lf
+    'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
+       'tagger' sp name '<' email '>' when lf
+    tag_msg;
+  tag_msg ::= data;
+
+  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
+     # the stream parser to reserve space for the gapped values.  An
+        # idnum can be updated in the future to a new object by issuing
+     # a new mark directive with the old idnum.
+        #
+  mark ::= 'mark' sp idnum lf;
+  data ::= (delimited_data | exact_data)
+    lf;
+
+    # note: delim may be any string but must not contain lf.
+    # data_line may contain any data but must not be exactly
+    # delim.
+  delimited_data ::= 'data' sp '<<' delim lf
+    (data_line lf)*
+       delim lf;
+
+     # note: declen indicates the length of binary_data in bytes.
+     # declen does not include the lf preceeding the binary data.
+     #
+  exact_data ::= 'data' sp declen lf
+    binary_data;
+
+     # note: quoted strings are C-style quoting supporting \c for
+     # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
+        # is the signed byte value in octal.  Note that the only
+     # characters which must actually be escaped to protect the
+     # stream formatting is: \, " and LF.  Otherwise these values
+        # are UTF8.
+     #
+  ref_str     ::= ref;
+  sha1exp_str ::= sha1exp;
+  tag_str     ::= tag;
+  path_str    ::= path    | '"' quoted(path)    '"' ;
+  mode        ::= '100644' | '644'
+                | '100755' | '755'
+                | '120000'
+                ;
+
+  declen ::= # unsigned 32 bit value, ascii base10 notation;
+  bigint ::= # unsigned integer value, ascii base10 notation;
+  binary_data ::= # file content, not interpreted;
+
+  when         ::= raw_when | rfc2822_when;
+  raw_when     ::= ts sp tz;
+  rfc2822_when ::= # Valid RFC 2822 date and time;
+
+  sp ::= # ASCII space character;
+  lf ::= # ASCII newline (LF) character;
+
+     # note: a colon (':') must precede the numerical value assigned to
+        # an idnum.  This is to distinguish it from a ref or tag name as
+     # GIT does not permit ':' in ref or tag strings.
+        #
+  idnum   ::= ':' bigint;
+  path    ::= # GIT style file path, e.g. "a/b/c";
+  ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
+  tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
+  sha1exp ::= # Any valid GIT SHA1 expression;
+  hexsha1 ::= # SHA1 in hexadecimal format;
+
+     # note: name and email are UTF8 strings, however name must not
+        # contain '<' or lf and email must not contain any of the
+     # following: '<', '>', lf.
+        #
+  name  ::= # valid GIT author/committer name;
+  email ::= # valid GIT author/committer email;
+  ts    ::= # time since the epoch in seconds, ascii base10 notation;
+  tz    ::= # GIT style timezone;
+*/
+
 #include "builtin.h"
 #include "cache.h"
 #include "object.h"
 #include "blob.h"
+#include "tree.h"
+#include "commit.h"
 #include "delta.h"
 #include "pack.h"
+#include "refs.h"
 #include "csum-file.h"
+#include "strbuf.h"
+#include "quote.h"
+
+#define PACK_ID_BITS 16
+#define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
 
 struct object_entry
 {
        struct object_entry *next;
-       unsigned long offset;
+       uint32_t offset;
+       unsigned type : TYPE_BITS;
+       unsigned pack_id : PACK_ID_BITS;
        unsigned char sha1[20];
 };
 
-struct object_entry_block
+struct object_entry_pool
 {
-       struct object_entry_block *next_block;
+       struct object_entry_pool *next_pool;
        struct object_entry *next_free;
        struct object_entry *end;
-       struct object_entry entries[0];
+       struct object_entry entries[FLEX_ARRAY]; /* more */
+};
+
+struct mark_set
+{
+       union {
+               struct object_entry *marked[1024];
+               struct mark_set *sets[1024];
+       } data;
+       unsigned int shift;
+};
+
+struct last_object
+{
+       void *data;
+       unsigned long len;
+       uint32_t offset;
+       unsigned int depth;
+       unsigned no_free:1;
+};
+
+struct mem_pool
+{
+       struct mem_pool *next_pool;
+       char *next_free;
+       char *end;
+       char space[FLEX_ARRAY]; /* more */
+};
+
+struct atom_str
+{
+       struct atom_str *next_atom;
+       unsigned short str_len;
+       char str_dat[FLEX_ARRAY]; /* more */
+};
+
+struct tree_content;
+struct tree_entry
+{
+       struct tree_content *tree;
+       struct atom_str* name;
+       struct tree_entry_ms
+       {
+               uint16_t mode;
+               unsigned char sha1[20];
+       } versions[2];
+};
+
+struct tree_content
+{
+       unsigned int entry_capacity; /* must match avail_tree_content */
+       unsigned int entry_count;
+       unsigned int delta_depth;
+       struct tree_entry *entries[FLEX_ARRAY]; /* more */
+};
+
+struct avail_tree_content
+{
+       unsigned int entry_capacity; /* must match tree_content */
+       struct avail_tree_content *next_avail;
+};
+
+struct branch
+{
+       struct branch *table_next_branch;
+       struct branch *active_next_branch;
+       const char *name;
+       struct tree_entry branch_tree;
+       uintmax_t last_commit;
+       unsigned active : 1;
+       unsigned pack_id : PACK_ID_BITS;
+       unsigned char sha1[20];
+};
+
+struct tag
+{
+       struct tag *next_tag;
+       const char *name;
+       unsigned int pack_id;
+       unsigned char sha1[20];
+};
+
+struct dbuf
+{
+       void *buffer;
+       size_t capacity;
+};
+
+struct hash_list
+{
+       struct hash_list *next;
+       unsigned char sha1[20];
 };
 
-static int max_depth = 10;
-static unsigned long alloc_count;
+typedef enum {
+       WHENSPEC_RAW = 1,
+       WHENSPEC_RFC2822,
+       WHENSPEC_NOW,
+} whenspec_type;
+
+/* Configured limits on output */
+static unsigned long max_depth = 10;
+static off_t max_packsize = (1LL << 32) - 1;
+static int force_update;
+
+/* Stats and misc. counters */
+static uintmax_t alloc_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 object_count;
-static unsigned long duplicate_count;
-static unsigned long packoff;
-static int packfd;
-static int current_depth;
-static void *lastdat;
-static unsigned long lastdatlen;
-static unsigned char lastsha1[20];
-static unsigned char packsha1[20];
-struct object_entry *object_table[1 << 16];
-struct object_entry_block *blocks;
-
-static void alloc_objects(int cnt)
-{
-       struct object_entry_block *b;
-
-       b = xmalloc(sizeof(struct object_entry_block)
+static unsigned long branch_count;
+static unsigned long branch_load_count;
+static int failure;
+static FILE *pack_edges;
+
+/* Memory pools */
+static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
+static size_t total_allocd;
+static struct mem_pool *mem_pool;
+
+/* Atom management */
+static unsigned int atom_table_sz = 4451;
+static unsigned int atom_cnt;
+static struct atom_str **atom_table;
+
+/* The .pack file being generated */
+static unsigned int pack_id;
+static struct packed_git *pack_data;
+static struct packed_git **all_packs;
+static unsigned long pack_size;
+
+/* Table of objects we've written. */
+static unsigned int object_entry_alloc = 5000;
+static struct object_entry_pool *blocks;
+static struct object_entry *object_table[1 << 16];
+static struct mark_set *marks;
+static const char* mark_file;
+
+/* Our last blob */
+static struct last_object last_blob;
+
+/* Tree management */
+static unsigned int tree_entry_alloc = 1000;
+static void *avail_tree_entry;
+static unsigned int avail_tree_table_sz = 100;
+static struct avail_tree_content **avail_tree_table;
+static struct dbuf old_tree;
+static struct dbuf new_tree;
+
+/* Branch data */
+static unsigned long max_active_branches = 5;
+static unsigned long cur_active_branches;
+static unsigned long branch_table_sz = 1039;
+static struct branch **branch_table;
+static struct branch *active_branches;
+
+/* Tag data */
+static struct tag *first_tag;
+static struct tag *last_tag;
+
+/* Input stream parsing */
+static whenspec_type whenspec = WHENSPEC_RAW;
+static struct strbuf command_buf;
+static uintmax_t next_mark;
+static struct dbuf new_data;
+
+
+static void alloc_objects(unsigned int cnt)
+{
+       struct object_entry_pool *b;
+
+       b = xmalloc(sizeof(struct object_entry_pool)
                + cnt * sizeof(struct object_entry));
-       b->next_block = blocks;
+       b->next_pool = blocks;
        b->next_free = b->entries;
        b->end = b->entries + cnt;
        blocks = b;
        alloc_count += cnt;
 }
 
-static struct object_entrynew_object(unsigned char *sha1)
+static struct object_entry *new_object(unsigned char *sha1)
 {
        struct object_entry *e;
 
        if (blocks->next_free == blocks->end)
-               alloc_objects(1000);
+               alloc_objects(object_entry_alloc);
 
        e = blocks->next_free++;
-       memcpy(e->sha1, sha1, sizeof(e->sha1));
+       hashcpy(e->sha1, sha1);
        return e;
 }
 
-static struct object_entry* insert_object(unsigned char *sha1)
+static struct object_entry *find_object(unsigned char *sha1)
+{
+       unsigned int h = sha1[0] << 8 | sha1[1];
+       struct object_entry *e;
+       for (e = object_table[h]; e; e = e->next)
+               if (!hashcmp(sha1, e->sha1))
+                       return e;
+       return NULL;
+}
+
+static struct object_entry *insert_object(unsigned char *sha1)
 {
        unsigned int h = sha1[0] << 8 | sha1[1];
        struct object_entry *e = object_table[h];
-       struct object_entry *p = 0;
+       struct object_entry *p = NULL;
 
        while (e) {
-               if (!memcmp(sha1, e->sha1, sizeof(e->sha1)))
+               if (!hashcmp(sha1, e->sha1))
                        return e;
                p = e;
                e = e->next;
        }
 
        e = new_object(sha1);
-       e->next = 0;
+       e->next = NULL;
        e->offset = 0;
        if (p)
                p->next = e;
@@ -83,185 +374,346 @@ static struct object_entry* insert_object(unsigned char *sha1)
        return e;
 }
 
-static ssize_t yread(int fd, void *buffer, size_t length)
+static unsigned int hc_str(const char *s, size_t len)
 {
-       ssize_t ret = 0;
-       while (ret < length) {
-               ssize_t size = xread(fd, (char *) buffer + ret, length - ret);
-               if (size < 0) {
-                       return size;
-               }
-               if (size == 0) {
-                       return ret;
+       unsigned int r = 0;
+       while (len-- > 0)
+               r = r * 31 + *s++;
+       return r;
+}
+
+static void *pool_alloc(size_t len)
+{
+       struct mem_pool *p;
+       void *r;
+
+       for (p = mem_pool; p; p = p->next_pool)
+               if ((p->end - p->next_free >= len))
+                       break;
+
+       if (!p) {
+               if (len >= (mem_pool_alloc/2)) {
+                       total_allocd += len;
+                       return xmalloc(len);
                }
-               ret += size;
+               total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
+               p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
+               p->next_pool = mem_pool;
+               p->next_free = p->space;
+               p->end = p->next_free + mem_pool_alloc;
+               mem_pool = p;
+       }
+
+       r = p->next_free;
+       /* round out to a pointer alignment */
+       if (len & (sizeof(void*) - 1))
+               len += sizeof(void*) - (len & (sizeof(void*) - 1));
+       p->next_free += len;
+       return r;
+}
+
+static void *pool_calloc(size_t count, size_t size)
+{
+       size_t len = count * size;
+       void *r = pool_alloc(len);
+       memset(r, 0, len);
+       return r;
+}
+
+static char *pool_strdup(const char *s)
+{
+       char *r = pool_alloc(strlen(s) + 1);
+       strcpy(r, s);
+       return r;
+}
+
+static void size_dbuf(struct dbuf *b, size_t maxlen)
+{
+       if (b->buffer) {
+               if (b->capacity >= maxlen)
+                       return;
+               free(b->buffer);
        }
-       return ret;
+       b->capacity = ((maxlen / 1024) + 1) * 1024;
+       b->buffer = xmalloc(b->capacity);
 }
 
-static ssize_t ywrite(int fd, void *buffer, size_t length)
+static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 {
-       ssize_t ret = 0;
-       while (ret < length) {
-               ssize_t size = xwrite(fd, (char *) buffer + ret, length - ret);
-               if (size < 0) {
-                       return size;
+       struct mark_set *s = marks;
+       while ((idnum >> s->shift) >= 1024) {
+               s = pool_calloc(1, sizeof(struct mark_set));
+               s->shift = marks->shift + 10;
+               s->data.sets[0] = marks;
+               marks = s;
+       }
+       while (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));
+                       s->data.sets[i]->shift = s->shift - 10;
                }
-               if (size == 0) {
-                       return ret;
+               s = s->data.sets[i];
+       }
+       if (!s->data.marked[idnum])
+               marks_set_count++;
+       s->data.marked[idnum] = oe;
+}
+
+static struct object_entry *find_mark(uintmax_t 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) {
+                       uintmax_t i = idnum >> s->shift;
+                       idnum -= i << s->shift;
+                       s = s->data.sets[i];
                }
-               ret += size;
+               if (s)
+                       oe = s->data.marked[idnum];
        }
-       return ret;
+       if (!oe)
+               die("mark :%" PRIuMAX " not declared", orig_idnum);
+       return oe;
 }
 
-static unsigned long encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
+static struct atom_str *to_atom(const char *s, unsigned short len)
 {
-       int n = 1;
-       unsigned char c;
+       unsigned int hc = hc_str(s, len) % atom_table_sz;
+       struct atom_str *c;
 
-       if (type < OBJ_COMMIT || type > OBJ_DELTA)
-               die("bad type %d", type);
+       for (c = atom_table[hc]; c; c = c->next_atom)
+               if (c->str_len == len && !strncmp(s, c->str_dat, len))
+                       return c;
 
-       c = (type << 4) | (size & 15);
-       size >>= 4;
-       while (size) {
-               *hdr++ = c | 0x80;
-               c = size & 0x7f;
-               size >>= 7;
-               n++;
-       }
-       *hdr = c;
-       return n;
+       c = pool_alloc(sizeof(struct atom_str) + len + 1);
+       c->str_len = len;
+       strncpy(c->str_dat, s, len);
+       c->str_dat[len] = 0;
+       c->next_atom = atom_table[hc];
+       atom_table[hc] = c;
+       atom_cnt++;
+       return c;
 }
 
-static void write_blob(void *dat, unsigned long datlen)
+static struct branch *lookup_branch(const char *name)
 {
-       z_stream s;
-       void *out, *delta;
-       unsigned char hdr[64];
-       unsigned long hdrlen, deltalen;
+       unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
+       struct branch *b;
 
-       if (lastdat && current_depth < max_depth) {
-               delta = diff_delta(lastdat, lastdatlen,
-                       dat, datlen,
-                       &deltalen, 0);
-       } else
-               delta = 0;
+       for (b = branch_table[hc]; b; b = b->table_next_branch)
+               if (!strcmp(name, b->name))
+                       return b;
+       return NULL;
+}
 
-       memset(&s, 0, sizeof(s));
-       deflateInit(&s, zlib_compression_level);
+static struct branch *new_branch(const char *name)
+{
+       unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
+       struct branch* b = lookup_branch(name);
 
-       if (delta) {
-               current_depth++;
-               s.next_in = delta;
-               s.avail_in = deltalen;
-               hdrlen = encode_header(OBJ_DELTA, deltalen, hdr);
-               if (ywrite(packfd, hdr, hdrlen) != hdrlen)
-                       die("Can't write object header: %s", strerror(errno));
-               if (ywrite(packfd, lastsha1, sizeof(lastsha1)) != sizeof(lastsha1))
-                       die("Can't write object base: %s", strerror(errno));
-               packoff += hdrlen + sizeof(lastsha1);
+       if (b)
+               die("Invalid attempt to create duplicate branch: %s", name);
+       if (check_ref_format(name))
+               die("Branch name doesn't conform to GIT standards: %s", 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;
+       b->active = 0;
+       b->pack_id = MAX_PACK_ID;
+       branch_table[hc] = b;
+       branch_count++;
+       return b;
+}
+
+static unsigned int hc_entries(unsigned int cnt)
+{
+       cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
+       return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
+}
+
+static struct tree_content *new_tree_content(unsigned int cnt)
+{
+       struct avail_tree_content *f, *l = NULL;
+       struct tree_content *t;
+       unsigned int hc = hc_entries(cnt);
+
+       for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
+               if (f->entry_capacity >= cnt)
+                       break;
+
+       if (f) {
+               if (l)
+                       l->next_avail = f->next_avail;
+               else
+                       avail_tree_table[hc] = f->next_avail;
        } else {
-               current_depth = 0;
-               s.next_in = dat;
-               s.avail_in = datlen;
-               hdrlen = encode_header(OBJ_BLOB, datlen, hdr);
-               if (ywrite(packfd, hdr, hdrlen) != hdrlen)
-                       die("Can't write object header: %s", strerror(errno));
-               packoff += hdrlen;
+               cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
+               f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
+               f->entry_capacity = cnt;
        }
 
-       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);
+       t = (struct tree_content*)f;
+       t->entry_count = 0;
+       t->delta_depth = 0;
+       return t;
+}
 
-       if (ywrite(packfd, out, s.total_out) != s.total_out)
-               die("Failed writing compressed data %s", strerror(errno));
-       packoff += s.total_out;
+static void release_tree_entry(struct tree_entry *e);
+static void release_tree_content(struct tree_content *t)
+{
+       struct avail_tree_content *f = (struct avail_tree_content*)t;
+       unsigned int hc = hc_entries(f->entry_capacity);
+       f->next_avail = avail_tree_table[hc];
+       avail_tree_table[hc] = f;
+}
 
-       free(out);
-       if (delta)
-               free(delta);
+static void release_tree_content_recursive(struct tree_content *t)
+{
+       unsigned int i;
+       for (i = 0; i < t->entry_count; i++)
+               release_tree_entry(t->entries[i]);
+       release_tree_content(t);
+}
+
+static struct tree_content *grow_tree_content(
+       struct tree_content *t,
+       int amt)
+{
+       struct tree_content *r = new_tree_content(t->entry_count + amt);
+       r->entry_count = t->entry_count;
+       r->delta_depth = t->delta_depth;
+       memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
+       release_tree_content(t);
+       return r;
+}
+
+static struct tree_entry *new_tree_entry(void)
+{
+       struct tree_entry *e;
+
+       if (!avail_tree_entry) {
+               unsigned int n = tree_entry_alloc;
+               total_allocd += n * sizeof(struct tree_entry);
+               avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
+               while (n-- > 1) {
+                       *((void**)e) = e + 1;
+                       e++;
+               }
+               *((void**)e) = NULL;
+       }
+
+       e = avail_tree_entry;
+       avail_tree_entry = *((void**)e);
+       return e;
+}
+
+static void release_tree_entry(struct tree_entry *e)
+{
+       if (e->tree)
+               release_tree_content_recursive(e->tree);
+       *((void**)e) = avail_tree_entry;
+       avail_tree_entry = e;
 }
 
-static void init_pack_header()
+static void start_packfile(void)
 {
-       const char* magic = "PACK";
-       unsigned long version = 2;
-       unsigned long zero = 0;
+       static char tmpfile[PATH_MAX];
+       struct packed_git *p;
+       struct pack_header hdr;
+       int pack_fd;
+
+       snprintf(tmpfile, sizeof(tmpfile),
+               "%s/tmp_pack_XXXXXX", get_object_directory());
+       pack_fd = mkstemp(tmpfile);
+       if (pack_fd < 0)
+               die("Can't create %s: %s", tmpfile, strerror(errno));
+       p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
+       strcpy(p->pack_name, tmpfile);
+       p->pack_fd = pack_fd;
+
+       hdr.hdr_signature = htonl(PACK_SIGNATURE);
+       hdr.hdr_version = htonl(2);
+       hdr.hdr_entries = 0;
+       write_or_die(p->pack_fd, &hdr, sizeof(hdr));
 
-       version = htonl(version);
+       pack_data = p;
+       pack_size = sizeof(hdr);
+       object_count = 0;
 
-       if (ywrite(packfd, (char*)magic, 4) != 4)
-               die("Can't write pack magic: %s", strerror(errno));
-       if (ywrite(packfd, &version, 4) != 4)
-               die("Can't write pack version: %s", strerror(errno));
-       if (ywrite(packfd, &zero, 4) != 4)
-               die("Can't write 0 object count: %s", strerror(errno));
-       packoff = 4 * 3;
+       all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
+       all_packs[pack_id] = p;
 }
 
-static void fixup_header_footer()
+static void fixup_header_footer(void)
 {
+       static const int buf_sz = 128 * 1024;
+       int pack_fd = pack_data->pack_fd;
        SHA_CTX c;
-       char hdr[8];
-       unsigned long cnt;
+       struct pack_header hdr;
        char *buf;
-       size_t n;
 
-       if (lseek(packfd, 0, SEEK_SET) != 0)
+       if (lseek(pack_fd, 0, SEEK_SET) != 0)
                die("Failed seeking to start: %s", strerror(errno));
+       if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
+               die("Unable to reread header of %s", pack_data->pack_name);
+       if (lseek(pack_fd, 0, SEEK_SET) != 0)
+               die("Failed seeking to start: %s", strerror(errno));
+       hdr.hdr_entries = htonl(object_count);
+       write_or_die(pack_fd, &hdr, sizeof(hdr));
 
        SHA1_Init(&c);
-       if (yread(packfd, hdr, 8) != 8)
-               die("Failed reading header: %s", strerror(errno));
-       SHA1_Update(&c, hdr, 8);
-
-       cnt = htonl(object_count);
-       SHA1_Update(&c, &cnt, 4);
-       if (ywrite(packfd, &cnt, 4) != 4)
-               die("Failed writing object count: %s", strerror(errno));
+       SHA1_Update(&c, &hdr, sizeof(hdr));
 
-       buf = xmalloc(128 * 1024);
+       buf = xmalloc(buf_sz);
        for (;;) {
-               n = xread(packfd, buf, 128 * 1024);
-               if (n <= 0)
+               size_t n = xread(pack_fd, buf, buf_sz);
+               if (!n)
                        break;
+               if (n < 0)
+                       die("Failed to checksum %s", pack_data->pack_name);
                SHA1_Update(&c, buf, n);
        }
        free(buf);
 
-       SHA1_Final(packsha1, &c);
-       if (ywrite(packfd, packsha1, sizeof(packsha1)) != sizeof(packsha1))
-               die("Failed writing pack checksum: %s", strerror(errno));
+       SHA1_Final(pack_data->sha1, &c);
+       write_or_die(pack_fd, pack_data->sha1, sizeof(pack_data->sha1));
+       close(pack_fd);
 }
 
-static int oecmp (const void *_a, const void *_b)
+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 memcmp(a->sha1, b->sha1, sizeof(a->sha1));
+       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)
+static char *create_index(void)
 {
+       static char tmpfile[PATH_MAX];
+       SHA_CTX ctx;
        struct sha1file *f;
-       struct object_entry **idx, **c, **last;
-       struct object_entry *e;
-       struct object_entry_block *o;
-       unsigned int array[256];
-       int i;
+       struct object_entry **idx, **c, **last, *e;
+       struct object_entry_pool *o;
+       uint32_t array[256];
+       int i, idx_fd;
 
        /* Build the sorted table of object IDs. */
        idx = xmalloc(object_count * sizeof(struct object_entry*));
        c = idx;
-       for (o = blocks; o; o = o->next_block)
-               for (e = o->entries; e != o->next_free; e++)
-                       *c++ = e;
+       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. */
@@ -277,81 +729,1417 @@ static void write_index(const char *idx_name)
                c = next;
        }
 
-       f = sha1create("%s", idx_name);
+       snprintf(tmpfile, sizeof(tmpfile),
+               "%s/tmp_idx_XXXXXX", get_object_directory());
+       idx_fd = mkstemp(tmpfile);
+       if (idx_fd < 0)
+               die("Can't create %s: %s", tmpfile, strerror(errno));
+       f = sha1fd(idx_fd, tmpfile);
        sha1write(f, array, 256 * sizeof(int));
+       SHA1_Init(&ctx);
        for (c = idx; c != last; c++) {
-               unsigned int offset = htonl((*c)->offset);
+               uint32_t offset = htonl((*c)->offset);
                sha1write(f, &offset, 4);
                sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
+               SHA1_Update(&ctx, (*c)->sha1, 20);
        }
-       sha1write(f, packsha1, sizeof(packsha1));
+       sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
        sha1close(f, NULL, 1);
        free(idx);
+       SHA1_Final(pack_data->sha1, &ctx);
+       return tmpfile;
 }
 
-int main(int argc, const char **argv)
+static char *keep_pack(char *curr_index_name)
 {
-       const char *base_name = argv[1];
-       int est_obj_cnt = atoi(argv[2]);
-       char *pack_name;
-       char *idx_name;
+       static char name[PATH_MAX];
+       static const char *keep_msg = "fast-import";
+       int keep_fd;
 
-       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);
+       chmod(pack_data->pack_name, 0444);
+       chmod(curr_index_name, 0444);
 
-       packfd = open(pack_name, O_RDWR|O_CREAT|O_EXCL, 0666);
-       if (packfd < 0)
-               die("Can't create pack file %s: %s", pack_name, strerror(errno));
+       snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
+                get_object_directory(), sha1_to_hex(pack_data->sha1));
+       keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
+       if (keep_fd < 0)
+               die("cannot create keep file");
+       write(keep_fd, keep_msg, strlen(keep_msg));
+       close(keep_fd);
 
-       alloc_objects(est_obj_cnt);
-       init_pack_header();
-       for (;;) {
-               unsigned long datlen;
-               int hdrlen;
-               void *dat;
-               char hdr[128];
-               unsigned char sha1[20];
-               SHA_CTX c;
-               struct object_entry *e;
+       snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
+                get_object_directory(), sha1_to_hex(pack_data->sha1));
+       if (move_temp_to_file(pack_data->pack_name, name))
+               die("cannot store pack file");
 
-               if (yread(0, &datlen, 4) != 4)
+       snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
+                get_object_directory(), sha1_to_hex(pack_data->sha1));
+       if (move_temp_to_file(curr_index_name, name))
+               die("cannot store index file");
+       return name;
+}
 
-                       break;
+static void unkeep_all_packs(void)
+{
+       static char name[PATH_MAX];
+       int k;
 
-               dat = xmalloc(datlen);
-               if (yread(0, dat, datlen) != datlen)
-                       break;
+       for (k = 0; k < pack_id; k++) {
+               struct packed_git *p = all_packs[k];
+               snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
+                        get_object_directory(), sha1_to_hex(p->sha1));
+               unlink(name);
+       }
+}
 
-               hdrlen = sprintf(hdr, "blob %lu", datlen) + 1;
-               SHA1_Init(&c);
-               SHA1_Update(&c, hdr, hdrlen);
-               SHA1_Update(&c, dat, datlen);
-               SHA1_Final(sha1, &c);
-
-               e = insert_object(sha1);
-               if (!e->offset) {
-                       e->offset = packoff;
-                       write_blob(dat, datlen);
-                       object_count++;
-
-                       if (lastdat)
-                               free(lastdat);
-                       lastdat = dat;
-                       lastdatlen = datlen;
-                       memcpy(lastsha1, sha1, sizeof(sha1));
-               } else {
-                       duplicate_count++;
-                       free(dat);
+static void end_packfile(void)
+{
+       struct packed_git *old_p = pack_data, *new_p;
+
+       if (object_count) {
+               char *idx_name;
+               int i;
+               struct branch *b;
+               struct tag *t;
+
+               fixup_header_footer();
+               idx_name = keep_pack(create_index());
+
+               /* 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;
+               all_packs[pack_id] = new_p;
+               install_packed_git(new_p);
+
+               /* Print the boundary */
+               if (pack_edges) {
+                       fprintf(pack_edges, "%s:", new_p->pack_name);
+                       for (i = 0; i < branch_table_sz; i++) {
+                               for (b = branch_table[i]; b; b = b->table_next_branch) {
+                                       if (b->pack_id == pack_id)
+                                               fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
+                               }
+                       }
+                       for (t = first_tag; t; t = t->next_tag) {
+                               if (t->pack_id == pack_id)
+                                       fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
+                       }
+                       fputc('\n', pack_edges);
+                       fflush(pack_edges);
+               }
+
+               pack_id++;
+       }
+       else
+               unlink(old_p->pack_name);
+       free(old_p);
+
+       /* 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 cycle_packfile(void)
+{
+       end_packfile();
+       start_packfile();
+}
+
+static size_t encode_header(
+       enum object_type type,
+       size_t size,
+       unsigned char *hdr)
+{
+       int n = 1;
+       unsigned char c;
+
+       if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
+               die("bad type %d", type);
+
+       c = (type << 4) | (size & 15);
+       size >>= 4;
+       while (size) {
+               *hdr++ = c | 0x80;
+               c = size & 0x7f;
+               size >>= 7;
+               n++;
+       }
+       *hdr = c;
+       return n;
+}
+
+static int store_object(
+       enum object_type type,
+       void *dat,
+       size_t datlen,
+       struct last_object *last,
+       unsigned char *sha1out,
+       uintmax_t mark)
+{
+       void *out, *delta;
+       struct object_entry *e;
+       unsigned char hdr[96];
+       unsigned char sha1[20];
+       unsigned long hdrlen, deltalen;
+       SHA_CTX c;
+       z_stream s;
+
+       hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
+               (unsigned long)datlen) + 1;
+       SHA1_Init(&c);
+       SHA1_Update(&c, hdr, hdrlen);
+       SHA1_Update(&c, dat, datlen);
+       SHA1_Final(sha1, &c);
+       if (sha1out)
+               hashcpy(sha1out, sha1);
+
+       e = insert_object(sha1);
+       if (mark)
+               insert_mark(mark, e);
+       if (e->offset) {
+               duplicate_count_by_type[type]++;
+               return 1;
+       }
+
+       if (last && last->data && last->depth < max_depth) {
+               delta = diff_delta(last->data, last->len,
+                       dat, datlen,
+                       &deltalen, 0);
+               if (delta && deltalen >= datlen) {
+                       free(delta);
+                       delta = NULL;
+               }
+       } else
+               delta = NULL;
+
+       memset(&s, 0, sizeof(s));
+       deflateInit(&s, zlib_compression_level);
+       if (delta) {
+               s.next_in = delta;
+               s.avail_in = deltalen;
+       } else {
+               s.next_in = dat;
+               s.avail_in = datlen;
+       }
+       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);
+
+       /* Determine if we should auto-checkpoint. */
+       if ((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;
+               cycle_packfile();
+
+               /* 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);
                }
        }
-       fixup_header_footer();
-       close(packfd);
-       write_index(idx_name);
 
-       fprintf(stderr, "%lu objects, %lu duplicates, %lu allocated (%lu overflow)\n",
-               object_count, duplicate_count, alloc_count, alloc_count - est_obj_cnt);
+       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_data->pack_fd, hdr, hdrlen);
+               pack_size += hdrlen;
+
+               hdr[pos] = ofs & 127;
+               while (ofs >>= 7)
+                       hdr[--pos] = 128 | (--ofs & 127);
+               write_or_die(pack_data->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_data->pack_fd, hdr, hdrlen);
+               pack_size += hdrlen;
+       }
+
+       write_or_die(pack_data->pack_fd, out, s.total_out);
+       pack_size += s.total_out;
 
+       free(out);
+       free(delta);
+       if (last) {
+               if (!last->no_free)
+                       free(last->data);
+               last->data = dat;
+               last->offset = e->offset;
+               last->len = datlen;
+       }
        return 0;
 }
+
+static void *gfi_unpack_entry(
+       struct object_entry *oe,
+       unsigned long *sizep)
+{
+       enum object_type type;
+       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, uint16_t *modep)
+{
+       unsigned char c;
+       uint16_t mode = 0;
+
+       while ((c = *str++) != ' ') {
+               if (c < '0' || c > '7')
+                       return NULL;
+               mode = (mode << 3) + (c - '0');
+       }
+       *modep = mode;
+       return str;
+}
+
+static void load_tree(struct tree_entry *root)
+{
+       unsigned char* sha1 = root->versions[1].sha1;
+       struct object_entry *myoe;
+       struct tree_content *t;
+       unsigned long size;
+       char *buf;
+       const char *c;
+
+       root->tree = t = new_tree_content(8);
+       if (is_null_sha1(sha1))
+               return;
+
+       myoe = find_object(sha1);
+       if (myoe) {
+               if (myoe->type != OBJ_TREE)
+                       die("Not a tree: %s", sha1_to_hex(sha1));
+               t->delta_depth = 0;
+               buf = gfi_unpack_entry(myoe, &size);
+       } else {
+               enum object_type type;
+               buf = read_sha1_file(sha1, &type, &size);
+               if (!buf || type != OBJ_TREE)
+                       die("Can't load tree %s", sha1_to_hex(sha1));
+       }
+
+       c = buf;
+       while (c != (buf + size)) {
+               struct tree_entry *e = new_tree_entry();
+
+               if (t->entry_count == t->entry_capacity)
+                       root->tree = t = grow_tree_content(t, t->entry_count);
+               t->entries[t->entry_count++] = e;
+
+               e->tree = NULL;
+               c = get_mode(c, &e->versions[1].mode);
+               if (!c)
+                       die("Corrupt mode in %s", sha1_to_hex(sha1));
+               e->versions[0].mode = e->versions[1].mode;
+               e->name = to_atom(c, strlen(c));
+               c += e->name->str_len + 1;
+               hashcpy(e->versions[0].sha1, (unsigned char*)c);
+               hashcpy(e->versions[1].sha1, (unsigned char*)c);
+               c += 20;
+       }
+       free(buf);
+}
+
+static int tecmp0 (const void *_a, const void *_b)
+{
+       struct tree_entry *a = *((struct tree_entry**)_a);
+       struct tree_entry *b = *((struct tree_entry**)_b);
+       return base_name_compare(
+               a->name->str_dat, a->name->str_len, a->versions[0].mode,
+               b->name->str_dat, b->name->str_len, b->versions[0].mode);
+}
+
+static int tecmp1 (const void *_a, const void *_b)
+{
+       struct tree_entry *a = *((struct tree_entry**)_a);
+       struct tree_entry *b = *((struct tree_entry**)_b);
+       return base_name_compare(
+               a->name->str_dat, a->name->str_len, a->versions[1].mode,
+               b->name->str_dat, b->name->str_len, b->versions[1].mode);
+}
+
+static void mktree(struct tree_content *t,
+       int v,
+       unsigned long *szp,
+       struct dbuf *b)
+{
+       size_t maxlen = 0;
+       unsigned int i;
+       char *c;
+
+       if (!v)
+               qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
+       else
+               qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
+
+       for (i = 0; i < t->entry_count; i++) {
+               if (t->entries[i]->versions[v].mode)
+                       maxlen += t->entries[i]->name->str_len + 34;
+       }
+
+       size_dbuf(b, maxlen);
+       c = b->buffer;
+       for (i = 0; i < t->entry_count; i++) {
+               struct tree_entry *e = t->entries[i];
+               if (!e->versions[v].mode)
+                       continue;
+               c += sprintf(c, "%o", (unsigned int)e->versions[v].mode);
+               *c++ = ' ';
+               strcpy(c, e->name->str_dat);
+               c += e->name->str_len + 1;
+               hashcpy((unsigned char*)c, e->versions[v].sha1);
+               c += 20;
+       }
+       *szp = c - (char*)b->buffer;
+}
+
+static void store_tree(struct tree_entry *root)
+{
+       struct tree_content *t = root->tree;
+       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;
+
+       for (i = 0; i < t->entry_count; i++) {
+               if (t->entries[i]->tree)
+                       store_tree(t->entries[i]);
+       }
+
+       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;
+       }
+
+       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;
+       for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
+               struct tree_entry *e = t->entries[i];
+               if (e->versions[1].mode) {
+                       e->versions[0].mode = e->versions[1].mode;
+                       hashcpy(e->versions[0].sha1, e->versions[1].sha1);
+                       t->entries[j++] = e;
+               } else {
+                       release_tree_entry(e);
+                       del++;
+               }
+       }
+       t->entry_count -= del;
+}
+
+static int tree_content_set(
+       struct tree_entry *root,
+       const char *p,
+       const unsigned char *sha1,
+       const uint16_t mode)
+{
+       struct tree_content *t = root->tree;
+       const char *slash1;
+       unsigned int i, n;
+       struct tree_entry *e;
+
+       slash1 = strchr(p, '/');
+       if (slash1)
+               n = slash1 - p;
+       else
+               n = strlen(p);
+
+       for (i = 0; i < t->entry_count; i++) {
+               e = t->entries[i];
+               if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
+                       if (!slash1) {
+                               if (e->versions[1].mode == mode
+                                               && !hashcmp(e->versions[1].sha1, sha1))
+                                       return 0;
+                               e->versions[1].mode = mode;
+                               hashcpy(e->versions[1].sha1, sha1);
+                               if (e->tree) {
+                                       release_tree_content_recursive(e->tree);
+                                       e->tree = NULL;
+                               }
+                               hashclr(root->versions[1].sha1);
+                               return 1;
+                       }
+                       if (!S_ISDIR(e->versions[1].mode)) {
+                               e->tree = new_tree_content(8);
+                               e->versions[1].mode = S_IFDIR;
+                       }
+                       if (!e->tree)
+                               load_tree(e);
+                       if (tree_content_set(e, slash1 + 1, sha1, mode)) {
+                               hashclr(root->versions[1].sha1);
+                               return 1;
+                       }
+                       return 0;
+               }
+       }
+
+       if (t->entry_count == t->entry_capacity)
+               root->tree = t = grow_tree_content(t, t->entry_count);
+       e = new_tree_entry();
+       e->name = to_atom(p, n);
+       e->versions[0].mode = 0;
+       hashclr(e->versions[0].sha1);
+       t->entries[t->entry_count++] = e;
+       if (slash1) {
+               e->tree = new_tree_content(8);
+               e->versions[1].mode = S_IFDIR;
+               tree_content_set(e, slash1 + 1, sha1, mode);
+       } else {
+               e->tree = NULL;
+               e->versions[1].mode = mode;
+               hashcpy(e->versions[1].sha1, sha1);
+       }
+       hashclr(root->versions[1].sha1);
+       return 1;
+}
+
+static int tree_content_remove(struct tree_entry *root, const char *p)
+{
+       struct tree_content *t = root->tree;
+       const char *slash1;
+       unsigned int i, n;
+       struct tree_entry *e;
+
+       slash1 = strchr(p, '/');
+       if (slash1)
+               n = slash1 - p;
+       else
+               n = strlen(p);
+
+       for (i = 0; i < t->entry_count; i++) {
+               e = t->entries[i];
+               if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
+                       if (!slash1 || !S_ISDIR(e->versions[1].mode))
+                               goto del_entry;
+                       if (!e->tree)
+                               load_tree(e);
+                       if (tree_content_remove(e, slash1 + 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;
+               }
+       }
+       return 0;
+
+del_entry:
+       if (e->tree) {
+               release_tree_content_recursive(e->tree);
+               e->tree = NULL;
+       }
+       e->versions[1].mode = 0;
+       hashclr(e->versions[1].sha1);
+       hashclr(root->versions[1].sha1);
+       return 1;
+}
+
+static int update_branch(struct branch *b)
+{
+       static const char *msg = "fast-import";
+       struct ref_lock *lock;
+       unsigned char old_sha1[20];
+
+       if (read_ref(b->name, old_sha1))
+               hashclr(old_sha1);
+       lock = lock_any_ref_for_update(b->name, old_sha1);
+       if (!lock)
+               return error("Unable to lock %s", b->name);
+       if (!force_update && !is_null_sha1(old_sha1)) {
+               struct commit *old_cmit, *new_cmit;
+
+               old_cmit = lookup_commit_reference_gently(old_sha1, 0);
+               new_cmit = lookup_commit_reference_gently(b->sha1, 0);
+               if (!old_cmit || !new_cmit) {
+                       unlock_ref(lock);
+                       return error("Branch %s is missing commits.", b->name);
+               }
+
+               if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
+                       unlock_ref(lock);
+                       warning("Not updating %s"
+                               " (new tip %s does not contain %s)",
+                               b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
+                       return -1;
+               }
+       }
+       if (write_ref_sha1(lock, b->sha1, msg) < 0)
+               return error("Unable to update %s", b->name);
+       return 0;
+}
+
+static void dump_branches(void)
+{
+       unsigned int i;
+       struct branch *b;
+
+       for (i = 0; i < branch_table_sz; i++) {
+               for (b = branch_table[i]; b; b = b->table_next_branch)
+                       failure |= update_branch(b);
+       }
+}
+
+static void dump_tags(void)
+{
+       static const char *msg = "fast-import";
+       struct tag *t;
+       struct ref_lock *lock;
+       char ref_name[PATH_MAX];
+
+       for (t = first_tag; t; t = t->next_tag) {
+               sprintf(ref_name, "tags/%s", t->name);
+               lock = lock_ref_sha1(ref_name, NULL);
+               if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
+                       failure |= error("Unable to update %s", ref_name);
+       }
+}
+
+static void dump_marks_helper(FILE *f,
+       uintmax_t base,
+       struct mark_set *m)
+{
+       uintmax_t k;
+       if (m->shift) {
+               for (k = 0; k < 1024; k++) {
+                       if (m->data.sets[k])
+                               dump_marks_helper(f, (base + k) << m->shift,
+                                       m->data.sets[k]);
+               }
+       } else {
+               for (k = 0; k < 1024; k++) {
+                       if (m->data.marked[k])
+                               fprintf(f, ":%" PRIuMAX " %s\n", base + k,
+                                       sha1_to_hex(m->data.marked[k]->sha1));
+               }
+       }
+}
+
+static void dump_marks(void)
+{
+       static struct lock_file mark_lock;
+       int mark_fd;
+       FILE *f;
+
+       if (!mark_file)
+               return;
+
+       mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
+       if (mark_fd < 0) {
+               failure |= error("Unable to write marks file %s: %s",
+                       mark_file, strerror(errno));
+               return;
+       }
+
+       f = fdopen(mark_fd, "w");
+       if (!f) {
+               rollback_lock_file(&mark_lock);
+               failure |= error("Unable to write marks file %s: %s",
+                       mark_file, strerror(errno));
+               return;
+       }
+
+       dump_marks_helper(f, 0, marks);
+       fclose(f);
+       if (commit_lock_file(&mark_lock))
+               failure |= error("Unable to write marks file %s: %s",
+                       mark_file, strerror(errno));
+}
+
+static void read_next_command(void)
+{
+       read_line(&command_buf, stdin, '\n');
+}
+
+static void cmd_mark(void)
+{
+       if (!prefixcmp(command_buf.buf, "mark :")) {
+               next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
+               read_next_command();
+       }
+       else
+               next_mark = 0;
+}
+
+static void *cmd_data (size_t *size)
+{
+       size_t length;
+       char *buffer;
+
+       if (prefixcmp(command_buf.buf, "data "))
+               die("Expected 'data n' command, found: %s", command_buf.buf);
+
+       if (!prefixcmp(command_buf.buf + 5, "<<")) {
+               char *term = xstrdup(command_buf.buf + 5 + 2);
+               size_t sz = 8192, term_len = command_buf.len - 5 - 2;
+               length = 0;
+               buffer = xmalloc(sz);
+               for (;;) {
+                       read_next_command();
+                       if (command_buf.eof)
+                               die("EOF in data (terminator '%s' not found)", term);
+                       if (term_len == command_buf.len
+                               && !strcmp(term, command_buf.buf))
+                               break;
+                       if (sz < (length + command_buf.len)) {
+                               sz = sz * 3 / 2 + 16;
+                               if (sz < (length + command_buf.len))
+                                       sz = length + command_buf.len;
+                               buffer = xrealloc(buffer, sz);
+                       }
+                       memcpy(buffer + length,
+                               command_buf.buf,
+                               command_buf.len - 1);
+                       length += command_buf.len - 1;
+                       buffer[length++] = '\n';
+               }
+               free(term);
+       }
+       else {
+               size_t n = 0;
+               length = strtoul(command_buf.buf + 5, NULL, 10);
+               buffer = xmalloc(length);
+               while (n < length) {
+                       size_t s = fread(buffer + n, 1, length - n, stdin);
+                       if (!s && feof(stdin))
+                               die("EOF in data (%lu bytes remaining)",
+                                       (unsigned long)(length - n));
+                       n += s;
+               }
+       }
+
+       if (fgetc(stdin) != '\n')
+               die("An lf did not trail the binary data as expected.");
+
+       *size = length;
+       return buffer;
+}
+
+static int validate_raw_date(const char *src, char *result, int maxlen)
+{
+       const char *orig_src = src;
+       char *endp, sign;
+
+       strtoul(src, &endp, 10);
+       if (endp == src || *endp != ' ')
+               return -1;
+
+       src = endp + 1;
+       if (*src != '-' && *src != '+')
+               return -1;
+       sign = *src;
+
+       strtoul(src + 1, &endp, 10);
+       if (endp == src || *endp || (endp - orig_src) >= maxlen)
+               return -1;
+
+       strcpy(result, orig_src);
+       return 0;
+}
+
+static char *parse_ident(const char *buf)
+{
+       const char *gt;
+       size_t name_len;
+       char *ident;
+
+       gt = strrchr(buf, '>');
+       if (!gt)
+               die("Missing > in ident string: %s", buf);
+       gt++;
+       if (*gt != ' ')
+               die("Missing space after > in ident string: %s", buf);
+       gt++;
+       name_len = gt - buf;
+       ident = xmalloc(name_len + 24);
+       strncpy(ident, buf, name_len);
+
+       switch (whenspec) {
+       case WHENSPEC_RAW:
+               if (validate_raw_date(gt, ident + name_len, 24) < 0)
+                       die("Invalid raw date \"%s\" in ident: %s", gt, buf);
+               break;
+       case WHENSPEC_RFC2822:
+               if (parse_date(gt, ident + name_len, 24) < 0)
+                       die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
+               break;
+       case WHENSPEC_NOW:
+               if (strcmp("now", gt))
+                       die("Date in ident must be 'now': %s", buf);
+               datestamp(ident + name_len, 24);
+               break;
+       }
+
+       return ident;
+}
+
+static void cmd_new_blob(void)
+{
+       size_t l;
+       void *d;
+
+       read_next_command();
+       cmd_mark();
+       d = cmd_data(&l);
+
+       if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
+               free(d);
+}
+
+static void unload_one_branch(void)
+{
+       while (cur_active_branches
+               && cur_active_branches >= max_active_branches) {
+               uintmax_t min_commit = ULONG_MAX;
+               struct branch *e, *l = NULL, *p = NULL;
+
+               for (e = active_branches; e; e = e->active_next_branch) {
+                       if (e->last_commit < min_commit) {
+                               p = l;
+                               min_commit = e->last_commit;
+                       }
+                       l = e;
+               }
+
+               if (p) {
+                       e = p->active_next_branch;
+                       p->active_next_branch = e->active_next_branch;
+               } else {
+                       e = active_branches;
+                       active_branches = e->active_next_branch;
+               }
+               e->active = 0;
+               e->active_next_branch = NULL;
+               if (e->branch_tree.tree) {
+                       release_tree_content_recursive(e->branch_tree.tree);
+                       e->branch_tree.tree = NULL;
+               }
+               cur_active_branches--;
+       }
+}
+
+static void load_branch(struct branch *b)
+{
+       load_tree(&b->branch_tree);
+       if (!b->active) {
+               b->active = 1;
+               b->active_next_branch = active_branches;
+               active_branches = b;
+               cur_active_branches++;
+               branch_load_count++;
+       }
+}
+
+static void file_change_m(struct branch *b)
+{
+       const char *p = command_buf.buf + 2;
+       char *p_uq;
+       const char *endp;
+       struct object_entry *oe = oe;
+       unsigned char sha1[20];
+       uint16_t mode, inline_data = 0;
+
+       p = get_mode(p, &mode);
+       if (!p)
+               die("Corrupt mode: %s", command_buf.buf);
+       switch (mode) {
+       case S_IFREG | 0644:
+       case S_IFREG | 0755:
+       case S_IFLNK:
+       case 0644:
+       case 0755:
+               /* ok */
+               break;
+       default:
+               die("Corrupt mode: %s", command_buf.buf);
+       }
+
+       if (*p == ':') {
+               char *x;
+               oe = find_mark(strtoumax(p + 1, &x, 10));
+               hashcpy(sha1, oe->sha1);
+               p = x;
+       } else if (!prefixcmp(p, "inline")) {
+               inline_data = 1;
+               p += 6;
+       } else {
+               if (get_sha1_hex(p, sha1))
+                       die("Invalid SHA1: %s", command_buf.buf);
+               oe = find_object(sha1);
+               p += 40;
+       }
+       if (*p++ != ' ')
+               die("Missing space after SHA1: %s", command_buf.buf);
+
+       p_uq = unquote_c_style(p, &endp);
+       if (p_uq) {
+               if (*endp)
+                       die("Garbage after path in: %s", command_buf.buf);
+               p = p_uq;
+       }
+
+       if (inline_data) {
+               size_t l;
+               void *d;
+               if (!p_uq)
+                       p = p_uq = xstrdup(p);
+               read_next_command();
+               d = cmd_data(&l);
+               if (store_object(OBJ_BLOB, d, l, &last_blob, sha1, 0))
+                       free(d);
+       } else if (oe) {
+               if (oe->type != OBJ_BLOB)
+                       die("Not a blob (actually a %s): %s",
+                               command_buf.buf, typename(oe->type));
+       } else {
+               enum object_type type = sha1_object_info(sha1, NULL);
+               if (type < 0)
+                       die("Blob not found: %s", command_buf.buf);
+               if (type != OBJ_BLOB)
+                       die("Not a blob (actually a %s): %s",
+                           typename(type), command_buf.buf);
+       }
+
+       tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
+       free(p_uq);
+}
+
+static void file_change_d(struct branch *b)
+{
+       const char *p = command_buf.buf + 2;
+       char *p_uq;
+       const char *endp;
+
+       p_uq = unquote_c_style(p, &endp);
+       if (p_uq) {
+               if (*endp)
+                       die("Garbage after path in: %s", command_buf.buf);
+               p = p_uq;
+       }
+       tree_content_remove(&b->branch_tree, p);
+       free(p_uq);
+}
+
+static void file_change_deleteall(struct branch *b)
+{
+       release_tree_content_recursive(b->branch_tree.tree);
+       hashclr(b->branch_tree.versions[0].sha1);
+       hashclr(b->branch_tree.versions[1].sha1);
+       load_tree(&b->branch_tree);
+}
+
+static void cmd_from(struct branch *b)
+{
+       const char *from;
+       struct branch *s;
+
+       if (prefixcmp(command_buf.buf, "from "))
+               return;
+
+       if (b->branch_tree.tree) {
+               release_tree_content_recursive(b->branch_tree.tree);
+               b->branch_tree.tree = NULL;
+       }
+
+       from = strchr(command_buf.buf, ' ') + 1;
+       s = lookup_branch(from);
+       if (b == s)
+               die("Can't create a branch from itself: %s", b->name);
+       else if (s) {
+               unsigned char *t = s->branch_tree.versions[1].sha1;
+               hashcpy(b->sha1, s->sha1);
+               hashcpy(b->branch_tree.versions[0].sha1, t);
+               hashcpy(b->branch_tree.versions[1].sha1, t);
+       } else if (*from == ':') {
+               uintmax_t idnum = strtoumax(from + 1, NULL, 10);
+               struct object_entry *oe = find_mark(idnum);
+               unsigned long size;
+               char *buf;
+               if (oe->type != OBJ_COMMIT)
+                       die("Mark :%" PRIuMAX " not a commit", idnum);
+               hashcpy(b->sha1, oe->sha1);
+               buf = gfi_unpack_entry(oe, &size);
+               if (!buf || size < 46)
+                       die("Not a valid commit: %s", from);
+               if (memcmp("tree ", buf, 5)
+                       || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
+                       die("The commit %s is corrupt", sha1_to_hex(b->sha1));
+               free(buf);
+               hashcpy(b->branch_tree.versions[0].sha1,
+                       b->branch_tree.versions[1].sha1);
+       } else if (!get_sha1(from, b->sha1)) {
+               if (is_null_sha1(b->sha1)) {
+                       hashclr(b->branch_tree.versions[0].sha1);
+                       hashclr(b->branch_tree.versions[1].sha1);
+               } else {
+                       unsigned long size;
+                       char *buf;
+
+                       buf = read_object_with_reference(b->sha1,
+                               commit_type, &size, b->sha1);
+                       if (!buf || size < 46)
+                               die("Not a valid commit: %s", from);
+                       if (memcmp("tree ", buf, 5)
+                               || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
+                               die("The commit %s is corrupt", sha1_to_hex(b->sha1));
+                       free(buf);
+                       hashcpy(b->branch_tree.versions[0].sha1,
+                               b->branch_tree.versions[1].sha1);
+               }
+       } else
+               die("Invalid ref name or SHA1 expression: %s", from);
+
+       read_next_command();
+}
+
+static struct hash_list *cmd_merge(unsigned int *count)
+{
+       struct hash_list *list = NULL, *n, *e = e;
+       const char *from;
+       struct branch *s;
+
+       *count = 0;
+       while (!prefixcmp(command_buf.buf, "merge ")) {
+               from = strchr(command_buf.buf, ' ') + 1;
+               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 :%" PRIuMAX " not a commit", idnum);
+                       hashcpy(n->sha1, oe->sha1);
+               } else if (!get_sha1(from, n->sha1)) {
+                       unsigned long size;
+                       char *buf = read_object_with_reference(n->sha1,
+                               commit_type, &size, n->sha1);
+                       if (!buf || size < 46)
+                               die("Not a valid commit: %s", from);
+                       free(buf);
+               } else
+                       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(void)
+{
+       struct branch *b;
+       void *msg;
+       size_t msglen;
+       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;
+       b = lookup_branch(sp);
+       if (!b)
+               b = new_branch(sp);
+
+       read_next_command();
+       cmd_mark();
+       if (!prefixcmp(command_buf.buf, "author ")) {
+               author = parse_ident(command_buf.buf + 7);
+               read_next_command();
+       }
+       if (!prefixcmp(command_buf.buf, "committer ")) {
+               committer = parse_ident(command_buf.buf + 10);
+               read_next_command();
+       }
+       if (!committer)
+               die("Expected committer but didn't get one");
+       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) {
+               unload_one_branch();
+               load_branch(b);
+       }
+
+       /* file_change* */
+       for (;;) {
+               if (1 == command_buf.len)
+                       break;
+               else if (!prefixcmp(command_buf.buf, "M "))
+                       file_change_m(b);
+               else if (!prefixcmp(command_buf.buf, "D "))
+                       file_change_d(b);
+               else if (!strcmp("deleteall", command_buf.buf))
+                       file_change_deleteall(b);
+               else
+                       die("Unsupported file_change: %s", command_buf.buf);
+               read_next_command();
+       }
+
+       /* 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, 114 + msglen
+               + merge_count * 49
+               + (author
+                       ? strlen(author) + strlen(committer)
+                       : 2 * strlen(committer)));
+       sp = new_data.buffer;
+       sp += sprintf(sp, "tree %s\n",
+               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;
+       }
+       sp += sprintf(sp, "author %s\n", author ? author : committer);
+       sp += sprintf(sp, "committer %s\n", committer);
+       *sp++ = '\n';
+       memcpy(sp, msg, msglen);
+       sp += msglen;
+       free(author);
+       free(committer);
+       free(msg);
+
+       if (!store_object(OBJ_COMMIT,
+               new_data.buffer, sp - (char*)new_data.buffer,
+               NULL, b->sha1, next_mark))
+               b->pack_id = pack_id;
+       b->last_commit = object_count_by_type[OBJ_COMMIT];
+}
+
+static void cmd_new_tag(void)
+{
+       char *sp;
+       const char *from;
+       char *tagger;
+       struct branch *s;
+       void *msg;
+       size_t msglen;
+       struct tag *t;
+       uintmax_t from_mark = 0;
+       unsigned char sha1[20];
+
+       /* Obtain the new tag name from the rest of our command */
+       sp = strchr(command_buf.buf, ' ') + 1;
+       t = pool_alloc(sizeof(struct tag));
+       t->next_tag = NULL;
+       t->name = pool_strdup(sp);
+       if (last_tag)
+               last_tag->next_tag = t;
+       else
+               first_tag = t;
+       last_tag = t;
+       read_next_command();
+
+       /* from ... */
+       if (prefixcmp(command_buf.buf, "from "))
+               die("Expected from command, got %s", command_buf.buf);
+       from = strchr(command_buf.buf, ' ') + 1;
+       s = lookup_branch(from);
+       if (s) {
+               hashcpy(sha1, s->sha1);
+       } else if (*from == ':') {
+               struct object_entry *oe;
+               from_mark = strtoumax(from + 1, NULL, 10);
+               oe = find_mark(from_mark);
+               if (oe->type != OBJ_COMMIT)
+                       die("Mark :%" PRIuMAX " not a commit", from_mark);
+               hashcpy(sha1, oe->sha1);
+       } else if (!get_sha1(from, sha1)) {
+               unsigned long size;
+               char *buf;
+
+               buf = read_object_with_reference(sha1,
+                       commit_type, &size, sha1);
+               if (!buf || size < 46)
+                       die("Not a valid commit: %s", from);
+               free(buf);
+       } else
+               die("Invalid ref name or SHA1 expression: %s", from);
+       read_next_command();
+
+       /* tagger ... */
+       if (prefixcmp(command_buf.buf, "tagger "))
+               die("Expected tagger command, got %s", command_buf.buf);
+       tagger = parse_ident(command_buf.buf + 7);
+
+       /* tag payload/message */
+       read_next_command();
+       msg = cmd_data(&msglen);
+
+       /* build the tag object */
+       size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
+       sp = new_data.buffer;
+       sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
+       sp += sprintf(sp, "type %s\n", commit_type);
+       sp += sprintf(sp, "tag %s\n", t->name);
+       sp += sprintf(sp, "tagger %s\n", tagger);
+       *sp++ = '\n';
+       memcpy(sp, msg, msglen);
+       sp += msglen;
+       free(tagger);
+       free(msg);
+
+       if (store_object(OBJ_TAG, new_data.buffer,
+               sp - (char*)new_data.buffer,
+               NULL, t->sha1, 0))
+               t->pack_id = MAX_PACK_ID;
+       else
+               t->pack_id = pack_id;
+}
+
+static void cmd_reset_branch(void)
+{
+       struct branch *b;
+       char *sp;
+
+       /* Obtain the branch name from the rest of our command */
+       sp = strchr(command_buf.buf, ' ') + 1;
+       b = lookup_branch(sp);
+       if (b) {
+               hashclr(b->sha1);
+               hashclr(b->branch_tree.versions[0].sha1);
+               hashclr(b->branch_tree.versions[1].sha1);
+               if (b->branch_tree.tree) {
+                       release_tree_content_recursive(b->branch_tree.tree);
+                       b->branch_tree.tree = NULL;
+               }
+       }
+       else
+               b = new_branch(sp);
+       read_next_command();
+       cmd_from(b);
+}
+
+static void cmd_checkpoint(void)
+{
+       if (object_count) {
+               cycle_packfile();
+               dump_branches();
+               dump_tags();
+               dump_marks();
+       }
+       read_next_command();
+}
+
+static void import_marks(const char *input_file)
+{
+       char line[512];
+       FILE *f = fopen(input_file, "r");
+       if (!f)
+               die("cannot read %s: %s", input_file, strerror(errno));
+       while (fgets(line, sizeof(line), f)) {
+               uintmax_t mark;
+               char *end;
+               unsigned char sha1[20];
+               struct object_entry *e;
+
+               end = strchr(line, '\n');
+               if (line[0] != ':' || !end)
+                       die("corrupt mark line: %s", line);
+               *end = 0;
+               mark = strtoumax(line + 1, &end, 10);
+               if (!mark || end == line + 1
+                       || *end != ' ' || get_sha1(end + 1, sha1))
+                       die("corrupt mark line: %s", line);
+               e = find_object(sha1);
+               if (!e) {
+                       enum object_type type = sha1_object_info(sha1, NULL);
+                       if (type < 0)
+                               die("object not found: %s", sha1_to_hex(sha1));
+                       e = insert_object(sha1);
+                       e->type = type;
+                       e->pack_id = MAX_PACK_ID;
+               }
+               insert_mark(mark, e);
+       }
+       fclose(f);
+}
+
+static const char fast_import_usage[] =
+"git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
+
+int main(int argc, const char **argv)
+{
+       int i, show_stats = 1;
+
+       git_config(git_default_config);
+       alloc_objects(object_entry_alloc);
+       strbuf_init(&command_buf);
+       atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
+       branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
+       avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
+       marks = pool_calloc(1, sizeof(struct mark_set));
+
+       for (i = 1; i < argc; i++) {
+               const char *a = argv[i];
+
+               if (*a != '-' || !strcmp(a, "--"))
+                       break;
+               else if (!prefixcmp(a, "--date-format=")) {
+                       const char *fmt = a + 14;
+                       if (!strcmp(fmt, "raw"))
+                               whenspec = WHENSPEC_RAW;
+                       else if (!strcmp(fmt, "rfc2822"))
+                               whenspec = WHENSPEC_RFC2822;
+                       else if (!strcmp(fmt, "now"))
+                               whenspec = WHENSPEC_NOW;
+                       else
+                               die("unknown --date-format argument %s", fmt);
+               }
+               else if (!prefixcmp(a, "--max-pack-size="))
+                       max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
+               else if (!prefixcmp(a, "--depth="))
+                       max_depth = strtoul(a + 8, NULL, 0);
+               else if (!prefixcmp(a, "--active-branches="))
+                       max_active_branches = strtoul(a + 18, NULL, 0);
+               else if (!prefixcmp(a, "--import-marks="))
+                       import_marks(a + 15);
+               else if (!prefixcmp(a, "--export-marks="))
+                       mark_file = a + 15;
+               else if (!prefixcmp(a, "--export-pack-edges=")) {
+                       if (pack_edges)
+                               fclose(pack_edges);
+                       pack_edges = fopen(a + 20, "a");
+                       if (!pack_edges)
+                               die("Cannot open %s: %s", a + 20, strerror(errno));
+               } else if (!strcmp(a, "--force"))
+                       force_update = 1;
+               else if (!strcmp(a, "--quiet"))
+                       show_stats = 0;
+               else if (!strcmp(a, "--stats"))
+                       show_stats = 1;
+               else
+                       die("unknown option %s", a);
+       }
+       if (i != argc)
+               usage(fast_import_usage);
+
+       start_packfile();
+       for (;;) {
+               read_next_command();
+               if (command_buf.eof)
+                       break;
+               else if (!strcmp("blob", command_buf.buf))
+                       cmd_new_blob();
+               else if (!prefixcmp(command_buf.buf, "commit "))
+                       cmd_new_commit();
+               else if (!prefixcmp(command_buf.buf, "tag "))
+                       cmd_new_tag();
+               else if (!prefixcmp(command_buf.buf, "reset "))
+                       cmd_reset_branch();
+               else if (!strcmp("checkpoint", command_buf.buf))
+                       cmd_checkpoint();
+               else
+                       die("Unsupported command: %s", command_buf.buf);
+       }
+       end_packfile();
+
+       dump_branches();
+       dump_tags();
+       unkeep_all_packs();
+       dump_marks();
+
+       if (pack_edges)
+               fclose(pack_edges);
+
+       if (show_stats) {
+               uintmax_t total_count = 0, duplicate_count = 0;
+               for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
+                       total_count += object_count_by_type[i];
+               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: %10" PRIuMAX "\n", alloc_count);
+               fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
+               fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
+               fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
+               fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
+               fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " 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:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
+               fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
+               fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
+               fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
+               fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
+               fprintf(stderr, "---------------------------------------------------------------------\n");
+               pack_report();
+               fprintf(stderr, "---------------------------------------------------------------------\n");
+               fprintf(stderr, "\n");
+       }
+
+       return failure ? 1 : 0;
+}