Merge branch 'master' into sp/gfi
authorShawn O. Pearce <spearce@spearce.org>
Tue, 30 Jan 2007 16:07:24 +0000 (11:07 -0500)
committerShawn O. Pearce <spearce@spearce.org>
Tue, 30 Jan 2007 16:07:24 +0000 (11:07 -0500)
git-fast-import requires use of inttypes.h, but the master branch has
added it to git-compat-util differently than git-fast-import originally
had used it.  This merge back of master to the fast-import topic is to
get (and use) inttypes.h the way master is using it.

This is a partially evil merge to remove the call to setup_ident(),
as the master branch now contains a change which makes this implicit
and therefore removed the function declaration. (commit 01754769).

Conflicts:

git-compat-util.h

1  2 
.gitignore
Makefile
fast-import.c

diff --cc .gitignore
Simple merge
diff --cc Makefile
+++ b/Makefile
@@@ -192,9 -192,8 +192,9 @@@ SCRIPTS = $(patsubst %.sh,%,$(SCRIPT_SH
  
  # ... and all the rest that could be moved out of bindir to gitexecdir
  PROGRAMS = \
-       git-convert-objects$X git-fetch-pack$X git-fsck-objects$X \
+       git-convert-objects$X git-fetch-pack$X git-fsck$X \
        git-hash-object$X git-index-pack$X git-local-fetch$X \
 +      git-fast-import$X \
        git-merge-base$X \
        git-daemon$X \
        git-merge-index$X git-mktag$X git-mktree$X git-patch-id$X \
diff --cc fast-import.c
index 487a91a,0000000..f62a5a9
mode 100644,000000..100644
--- /dev/null
@@@ -1,2019 -1,0 +1,2018 @@@
-       setup_ident();
 +/*
 +Format of STDIN stream:
 +
 +  stream ::= cmd*;
 +
 +  cmd ::= new_blob
 +        | new_commit
 +        | new_tag
 +        | reset_branch
 +        ;
 +
 +  new_blob ::= 'blob' lf
 +      mark?
 +    file_content;
 +  file_content ::= data;
 +
 +  new_commit ::= 'commit' sp ref_str lf
 +    mark?
 +    ('author' sp name '<' email '>' ts tz lf)?
 +    '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;
 +
 +  file_change ::= file_del | file_obm | file_inm;
 +  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 '>' ts tz 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     | '"' quoted(ref)     '"' ;
 +  sha1exp_str ::= sha1exp | '"' quoted(sha1exp) '"' ;
 +  tag_str     ::= tag     | '"' quoted(tag)     '"' ;
 +  path_str    ::= path    | '"' quoted(path)    '"' ;
 +  mode        ::= '100644' | '644'
 +                | '100755' | '755'
 +                | '140000'
 +                ;
 +
 +  declen ::= # unsigned 32 bit value, ascii base10 notation;
 +  bigint ::= # unsigned integer value, ascii base10 notation;
 +  binary_data ::= # file content, not interpreted;
 +
 +  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 "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;
 +      unsigned type : TYPE_BITS;
 +      unsigned pack_id : PACK_ID_BITS;
 +      unsigned char sha1[20];
 +};
 +
 +struct object_entry_pool
 +{
 +      struct object_entry_pool *next_pool;
 +      struct object_entry *next_free;
 +      struct object_entry *end;
 +      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;
 +      unsigned long 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 int 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
 +      {
 +              unsigned int 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 int pack_id;
 +      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];
 +};
 +
 +/* Configured limits on output */
 +static unsigned long max_depth = 10;
 +static unsigned long max_packsize = (1LL << 32) - 1;
 +
 +/* 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 branch_count;
 +static unsigned long branch_load_count;
 +
 +/* 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 struct strbuf command_buf;
 +static uintmax_t next_mark;
 +static struct dbuf new_data;
 +static FILE* branch_log;
 +
 +
 +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_pool = blocks;
 +      b->next_free = b->entries;
 +      b->end = b->entries + cnt;
 +      blocks = b;
 +      alloc_count += cnt;
 +}
 +
 +static struct object_entry* new_object(unsigned char *sha1)
 +{
 +      struct object_entry *e;
 +
 +      if (blocks->next_free == blocks->end)
 +              alloc_objects(object_entry_alloc);
 +
 +      e = blocks->next_free++;
 +      hashcpy(e->sha1, sha1);
 +      return e;
 +}
 +
 +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 = NULL;
 +
 +      while (e) {
 +              if (!hashcmp(sha1, e->sha1))
 +                      return e;
 +              p = e;
 +              e = e->next;
 +      }
 +
 +      e = new_object(sha1);
 +      e->next = NULL;
 +      e->offset = 0;
 +      if (p)
 +              p->next = e;
 +      else
 +              object_table[h] = e;
 +      return e;
 +}
 +
 +static unsigned int hc_str(const char *s, size_t len)
 +{
 +      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);
 +              }
 +              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);
 +      }
 +      b->capacity = ((maxlen / 1024) + 1) * 1024;
 +      b->buffer = xmalloc(b->capacity);
 +}
 +
 +static void insert_mark(uintmax_t idnum, struct object_entry *oe)
 +{
 +      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;
 +              }
 +              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];
 +              }
 +              if (s)
 +                      oe = s->data.marked[idnum];
 +      }
 +      if (!oe)
 +              die("mark :%ju not declared", orig_idnum);
 +      return oe;
 +}
 +
 +static struct atom_str* to_atom(const char *s, size_t len)
 +{
 +      unsigned int hc = hc_str(s, len) % atom_table_sz;
 +      struct atom_str *c;
 +
 +      for (c = atom_table[hc]; c; c = c->next_atom)
 +              if (c->str_len == len && !strncmp(s, c->str_dat, len))
 +                      return c;
 +
 +      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 struct branch* lookup_branch(const char *name)
 +{
 +      unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
 +      struct branch *b;
 +
 +      for (b = branch_table[hc]; b; b = b->table_next_branch)
 +              if (!strcmp(name, b->name))
 +                      return b;
 +      return NULL;
 +}
 +
 +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 (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->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 {
 +              cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
 +              f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
 +              f->entry_capacity = cnt;
 +      }
 +
 +      t = (struct tree_content*)f;
 +      t->entry_count = 0;
 +      t->delta_depth = 0;
 +      return t;
 +}
 +
 +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;
 +}
 +
 +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 start_packfile(void)
 +{
 +      static char tmpfile[PATH_MAX];
 +      struct packed_git *p;
 +      struct pack_header hdr;
 +      int pack_fd;
 +
 +      snprintf(tmpfile, sizeof(tmpfile),
 +              "%s/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));
 +
 +      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(void)
 +{
 +      static const int buf_sz = 128 * 1024;
 +      int pack_fd = pack_data->pack_fd;
 +      SHA_CTX c;
 +      struct pack_header hdr;
 +      char *buf;
 +
 +      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);
 +      SHA1_Update(&c, &hdr, sizeof(hdr));
 +
 +      buf = xmalloc(buf_sz);
 +      for (;;) {
 +              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(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_)
 +{
 +      struct object_entry *a = *((struct object_entry**)a_);
 +      struct object_entry *b = *((struct object_entry**)b_);
 +      return hashcmp(a->sha1, b->sha1);
 +}
 +
 +static char* create_index(void)
 +{
 +      static char tmpfile[PATH_MAX];
 +      SHA_CTX ctx;
 +      struct sha1file *f;
 +      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_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;
 +      }
 +
 +      snprintf(tmpfile, sizeof(tmpfile),
 +              "%s/index_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++) {
 +              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, pack_data->sha1, sizeof(pack_data->sha1));
 +      sha1close(f, NULL, 1);
 +      free(idx);
 +      SHA1_Final(pack_data->sha1, &ctx);
 +      return tmpfile;
 +}
 +
 +static char* keep_pack(char *curr_index_name)
 +{
 +      static char name[PATH_MAX];
 +      static char *keep_msg = "fast-import";
 +      int keep_fd;
 +
 +      chmod(pack_data->pack_name, 0444);
 +      chmod(curr_index_name, 0444);
 +
 +      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);
 +
 +      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");
 +
 +      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;
 +}
 +
 +static void unkeep_all_packs(void)
 +{
 +      static char name[PATH_MAX];
 +      int k;
 +
 +      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);
 +      }
 +}
 +
 +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 */
 +              fprintf(stdout, "%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(stdout, " %s", sha1_to_hex(b->sha1));
 +                      }
 +              }
 +              for (t = first_tag; t; t = t->next_tag) {
 +                      if (t->pack_id == pack_id)
 +                              fprintf(stdout, " %s", sha1_to_hex(t->sha1));
 +              }
 +              fputc('\n', stdout);
 +
 +              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 checkpoint(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",type_names[type],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;
 +              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_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);
 +      if (delta)
 +              free(delta);
 +      if (last) {
 +              if (last->data && !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)
 +{
 +      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)
 +{
 +      unsigned char c;
 +      unsigned int 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 {
 +              char type[20];
 +              buf = read_sha1_file(sha1, type, &size);
 +              if (!buf || strcmp(type, tree_type))
 +                      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, 8);
 +              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", 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 unsigned int 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, 8);
 +      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 void dump_branches(void)
 +{
 +      static const char *msg = "fast-import";
 +      unsigned int i;
 +      struct branch *b;
 +      struct ref_lock *lock;
 +
 +      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);
 +                      if (!lock || write_ref_sha1(lock, b->sha1, msg) < 0)
 +                              die("Can't write %s", b->name);
 +              }
 +      }
 +}
 +
 +static void dump_tags(void)
 +{
 +      static const char *msg = "fast-import";
 +      struct tag *t;
 +      struct ref_lock *lock;
 +      char path[PATH_MAX];
 +
 +      for (t = first_tag; t; t = t->next_tag) {
 +              sprintf(path, "refs/tags/%s", t->name);
 +              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,
 +      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, ":%ju %s\n", base + k,
 +                                      sha1_to_hex(m->data.marked[k]->sha1));
 +              }
 +      }
 +}
 +
 +static void dump_marks(void)
 +{
 +      if (mark_file)
 +      {
 +              FILE *f = fopen(mark_file, "w");
 +              dump_marks_helper(f, 0, marks);
 +              fclose(f);
 +      }
 +}
 +
 +static void read_next_command(void)
 +{
 +      read_line(&command_buf, stdin, '\n');
 +}
 +
 +static void cmd_mark(void)
 +{
 +      if (!strncmp("mark :", command_buf.buf, 6)) {
 +              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 (strncmp("data ", command_buf.buf, 5))
 +              die("Expected 'data n' command, found: %s", command_buf.buf);
 +
 +      if (!strncmp("<<", command_buf.buf + 5, 2)) {
 +              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)", length - n);
 +                      n += s;
 +              }
 +      }
 +
 +      if (fgetc(stdin) != '\n')
 +              die("An lf did not trail the binary data as expected.");
 +
 +      *size = length;
 +      return buffer;
 +}
 +
 +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) {
 +              unsigned long 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_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);
 +      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;
 +      unsigned char sha1[20];
 +      unsigned int mode, inline_data = 0;
 +      char type[20];
 +
 +      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 (!strncmp("inline", p, 6)) {
 +              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, type_names[oe->type]);
 +      } else {
 +              if (sha1_object_info(sha1, type, NULL))
 +                      die("Blob not found: %s", command_buf.buf);
 +              if (strcmp(blob_type, type))
 +                      die("Not a blob (actually a %s): %s",
 +                              command_buf.buf, type);
 +      }
 +
 +      tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
 +
 +      if (p_uq)
 +              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);
 +      if (p_uq)
 +              free(p_uq);
 +}
 +
 +static void cmd_from(struct branch *b)
 +{
 +      const char *from, *endp;
 +      char *str_uq;
 +      struct branch *s;
 +
 +      if (strncmp("from ", command_buf.buf, 5))
 +              return;
 +
 +      if (b->last_commit)
 +              die("Can't reinitailize branch %s", b->name);
 +
 +      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;
 +      }
 +
 +      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 :%ju 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,
 +                              type_names[OBJ_COMMIT], &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;
 +      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(void)
 +{
 +      struct branch *b;
 +      void *msg;
 +      size_t msglen;
 +      char *str_uq;
 +      const char *endp;
 +      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;
 +      str_uq = unquote_c_style(sp, &endp);
 +      if (str_uq) {
 +              if (*endp)
 +                      die("Garbage after ref in: %s", command_buf.buf);
 +              sp = str_uq;
 +      }
 +      b = lookup_branch(sp);
 +      if (!b)
 +              b = new_branch(sp);
 +      if (str_uq)
 +              free(str_uq);
 +
 +      read_next_command();
 +      cmd_mark();
 +      if (!strncmp("author ", command_buf.buf, 7)) {
 +              author = strdup(command_buf.buf);
 +              read_next_command();
 +      }
 +      if (!strncmp("committer ", command_buf.buf, 10)) {
 +              committer = strdup(command_buf.buf);
 +              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 (!strncmp("M ", command_buf.buf, 2))
 +                      file_change_m(b);
 +              else if (!strncmp("D ", command_buf.buf, 2))
 +                      file_change_d(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, 97 + 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;
 +      }
 +      if (author)
 +              sp += sprintf(sp, "%s\n", author);
 +      else
 +              sp += sprintf(sp, "author %s\n", committer + 10);
 +      sp += sprintf(sp, "%s\n\n", committer);
 +      memcpy(sp, msg, msglen);
 +      sp += msglen;
 +      if (author)
 +              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];
 +
 +      if (branch_log) {
 +              int need_dq = quote_c_style(b->name, NULL, NULL, 0);
 +              fprintf(branch_log, "commit ");
 +              if (need_dq) {
 +                      fputc('"', branch_log);
 +                      quote_c_style(b->name, NULL, branch_log, 0);
 +                      fputc('"', branch_log);
 +              } else
 +                      fprintf(branch_log, "%s", b->name);
 +              fprintf(branch_log," :%ju %s\n",next_mark,sha1_to_hex(b->sha1));
 +      }
 +}
 +
 +static void cmd_new_tag(void)
 +{
 +      char *str_uq;
 +      const char *endp;
 +      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;
 +      str_uq = unquote_c_style(sp, &endp);
 +      if (str_uq) {
 +              if (*endp)
 +                      die("Garbage after tag name in: %s", command_buf.buf);
 +              sp = str_uq;
 +      }
 +      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;
 +      if (str_uq)
 +              free(str_uq);
 +      read_next_command();
 +
 +      /* from ... */
 +      if (strncmp("from ", command_buf.buf, 5))
 +              die("Expected from command, got %s", command_buf.buf);
 +
 +      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;
 +      }
 +
 +      s = lookup_branch(from);
 +      if (s) {
 +              hashcpy(sha1, s->sha1);
 +      } else if (*from == ':') {
 +              from_mark = strtoumax(from + 1, NULL, 10);
 +              struct object_entry *oe = find_mark(from_mark);
 +              if (oe->type != OBJ_COMMIT)
 +                      die("Mark :%ju 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,
 +                      type_names[OBJ_COMMIT], &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);
 +
 +      if (str_uq)
 +              free(str_uq);
 +      read_next_command();
 +
 +      /* tagger ... */
 +      if (strncmp("tagger ", command_buf.buf, 7))
 +              die("Expected tagger command, got %s", command_buf.buf);
 +      tagger = strdup(command_buf.buf);
 +
 +      /* 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", type_names[OBJ_COMMIT]);
 +      sp += sprintf(sp, "tag %s\n", t->name);
 +      sp += sprintf(sp, "%s\n\n", tagger);
 +      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;
 +
 +      if (branch_log) {
 +              int need_dq = quote_c_style(t->name, NULL, NULL, 0);
 +              fprintf(branch_log, "tag ");
 +              if (need_dq) {
 +                      fputc('"', branch_log);
 +                      quote_c_style(t->name, NULL, branch_log, 0);
 +                      fputc('"', branch_log);
 +              } else
 +                      fprintf(branch_log, "%s", t->name);
 +              fprintf(branch_log," :%ju %s\n",from_mark,sha1_to_hex(t->sha1));
 +      }
 +}
 +
 +static void cmd_reset_branch(void)
 +{
 +      struct branch *b;
 +      char *str_uq;
 +      const char *endp;
 +      char *sp;
 +
 +      /* Obtain the branch name from the rest of our command */
 +      sp = strchr(command_buf.buf, ' ') + 1;
 +      str_uq = unquote_c_style(sp, &endp);
 +      if (str_uq) {
 +              if (*endp)
 +                      die("Garbage after ref in: %s", command_buf.buf);
 +              sp = str_uq;
 +      }
 +      b = lookup_branch(sp);
 +      if (b) {
 +              b->last_commit = 0;
 +              if (b->branch_tree.tree) {
 +                      release_tree_content_recursive(b->branch_tree.tree);
 +                      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(void)
 +{
 +      if (object_count)
 +              checkpoint();
 +      read_next_command();
 +}
 +
 +static const char fast_import_usage[] =
 +"git-fast-import [--depth=n] [--active-branches=n] [--export-marks=marks.file] [--branch-log=log]";
 +
 +int main(int argc, const char **argv)
 +{
 +      int i;
 +      uintmax_t total_count, duplicate_count;
 +
 +      git_config(git_default_config);
 +
 +      for (i = 1; i < argc; i++) {
 +              const char *a = argv[i];
 +
 +              if (*a != '-' || !strcmp(a, "--"))
 +                      break;
 +              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))
 +                      max_active_branches = strtoul(a + 18, NULL, 0);
 +              else if (!strncmp(a, "--export-marks=", 15))
 +                      mark_file = a + 15;
 +              else if (!strncmp(a, "--branch-log=", 13)) {
 +                      branch_log = fopen(a + 13, "w");
 +                      if (!branch_log)
 +                              die("Can't create %s: %s", a + 13, strerror(errno));
 +              }
 +              else
 +                      die("unknown option %s", a);
 +      }
 +      if (i != argc)
 +              usage(fast_import_usage);
 +
 +      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));
 +
 +      start_packfile();
 +      for (;;) {
 +              read_next_command();
 +              if (command_buf.eof)
 +                      break;
 +              else if (!strcmp("blob", command_buf.buf))
 +                      cmd_new_blob();
 +              else if (!strncmp("commit ", command_buf.buf, 7))
 +                      cmd_new_commit();
 +              else if (!strncmp("tag ", command_buf.buf, 4))
 +                      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();
 +
 +      dump_branches();
 +      dump_tags();
 +      unkeep_all_packs();
 +      dump_marks();
 +      if (branch_log)
 +              fclose(branch_log);
 +
 +      total_count = 0;
 +      for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
 +              total_count += object_count_by_type[i];
 +      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: %10ju\n", alloc_count);
 +      fprintf(stderr, "Total objects:   %10ju (%10ju duplicates                  )\n", total_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:     %10ju (%10ju unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
 +      fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
 +      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:    %10ju KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
 +      fprintf(stderr, "---------------------------------------------------------------------\n");
 +      pack_report();
 +      fprintf(stderr, "---------------------------------------------------------------------\n");
 +      fprintf(stderr, "\n");
 +
 +      return 0;
 +}