Adjust for 2.19.x series
authorJunio C Hamano <gitster@pobox.com>
Wed, 31 Oct 2018 04:12:12 +0000 (13:12 +0900)
committerJunio C Hamano <gitster@pobox.com>
Wed, 31 Oct 2018 04:12:12 +0000 (13:12 +0900)
* jk/detect-truncated-zlib-input
cat-file: handle streaming failures consistently
check_stream_sha1(): handle input underflow
t1450: check large blob in trailing-garbage test

1  2 
builtin/cat-file.c
sha1-file.c
t/t1450-fsck.sh
diff --combined builtin/cat-file.c
index 64ec1745ab2c20ef18a9292ef3b3c82efd46de17,f11732863789b5636c6c069bead9d7f7a8a2677c..0520cecc9a1d963841b384de6173d752e6d78fb7
@@@ -4,16 -4,12 +4,16 @@@
   * Copyright (C) Linus Torvalds, 2005
   */
  #include "cache.h"
 +#include "config.h"
  #include "builtin.h"
 +#include "diff.h"
  #include "parse-options.h"
  #include "userdiff.h"
  #include "streaming.h"
  #include "tree-walk.h"
  #include "sha1-array.h"
 +#include "packfile.h"
 +#include "object-store.h"
  
  struct batch_options {
        int enabled;
@@@ -21,7 -17,6 +21,7 @@@
        int print_contents;
        int buffer_output;
        int all_objects;
 +      int unordered;
        int cmdmode; /* may be 'w' or 'c' for --filters or --textconv */
        const char *format;
  };
@@@ -34,13 -29,13 +34,13 @@@ static int filter_object(const char *pa
  {
        enum object_type type;
  
 -      *buf = read_sha1_file(oid->hash, &type, size);
 +      *buf = read_object_file(oid, &type, size);
        if (!*buf)
                return error(_("cannot read object %s '%s'"),
                             oid_to_hex(oid), path);
        if ((type == OBJ_BLOB) && S_ISREG(mode)) {
                struct strbuf strbuf = STRBUF_INIT;
 -              if (convert_to_working_tree(path, *buf, *size, &strbuf)) {
 +              if (convert_to_working_tree(&the_index, path, *buf, *size, &strbuf)) {
                        free(*buf);
                        *size = strbuf.len;
                        *buf = strbuf_detach(&strbuf, NULL);
        return 0;
  }
  
+ static int stream_blob(const struct object_id *oid)
+ {
+       if (stream_blob_to_fd(1, oid, NULL, 0))
+               die("unable to stream %s to stdout", oid_to_hex(oid));
+       return 0;
+ }
  static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
                        int unknown_type)
  {
        struct object_context obj_context;
        struct object_info oi = OBJECT_INFO_INIT;
        struct strbuf sb = STRBUF_INIT;
 -      unsigned flags = LOOKUP_REPLACE_OBJECT;
 +      unsigned flags = OBJECT_INFO_LOOKUP_REPLACE;
        const char *path = force_path;
  
        if (unknown_type)
 -              flags |= LOOKUP_UNKNOWN_OBJECT;
 +              flags |= OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
  
 -      if (get_sha1_with_context(obj_name, 0, oid.hash, &obj_context))
 +      if (get_oid_with_context(obj_name, GET_OID_RECORD_PATH,
 +                               &oid, &obj_context))
                die("Not a valid object name %s", obj_name);
  
        if (!path)
@@@ -78,8 -79,8 +85,8 @@@
        buf = NULL;
        switch (opt) {
        case 't':
 -              oi.typename = &sb;
 -              if (sha1_object_info_extended(oid.hash, &oi, flags) < 0)
 +              oi.type_name = &sb;
 +              if (oid_object_info_extended(the_repository, &oid, &oi, flags) < 0)
                        die("git cat-file: could not get object info");
                if (sb.len) {
                        printf("%s\n", sb.buf);
@@@ -90,7 -91,7 +97,7 @@@
  
        case 's':
                oi.sizep = &size;
 -              if (sha1_object_info_extended(oid.hash, &oi, flags) < 0)
 +              if (oid_object_info_extended(the_repository, &oid, &oi, flags) < 0)
                        die("git cat-file: could not get object info");
                printf("%lu\n", size);
                return 0;
                return !has_object_file(&oid);
  
        case 'w':
 -              if (!path[0])
 +              if (!path)
                        die("git cat-file --filters %s: <object> must be "
                            "<sha1:path>", obj_name);
  
                break;
  
        case 'c':
 -              if (!path[0])
 +              if (!path)
                        die("git cat-file --textconv %s: <object> must be <sha1:path>",
                            obj_name);
  
                if (textconv_object(path, obj_context.mode, &oid, 1, &buf, &size))
                        break;
 +              /* else fallthrough */
  
        case 'p':
 -              type = sha1_object_info(oid.hash, NULL);
 +              type = oid_object_info(the_repository, &oid, NULL);
                if (type < 0)
                        die("Not a valid object name %s", obj_name);
  
                }
  
                if (type == OBJ_BLOB)
-                       return stream_blob_to_fd(1, &oid, NULL, 0);
+                       return stream_blob(&oid);
 -              buf = read_sha1_file(oid.hash, &type, &size);
 +              buf = read_object_file(&oid, &type, &size);
                if (!buf)
                        die("Cannot read object %s", obj_name);
  
        case 0:
                if (type_from_string(exp_type) == OBJ_BLOB) {
                        struct object_id blob_oid;
 -                      if (sha1_object_info(oid.hash, NULL) == OBJ_TAG) {
 -                              char *buffer = read_sha1_file(oid.hash, &type, &size);
 +                      if (oid_object_info(the_repository, &oid, NULL) == OBJ_TAG) {
 +                              char *buffer = read_object_file(&oid, &type,
 +                                                              &size);
                                const char *target;
                                if (!skip_prefix(buffer, "object ", &target) ||
                                    get_oid_hex(target, &blob_oid))
                        } else
                                oidcpy(&blob_oid, &oid);
  
 -                      if (sha1_object_info(blob_oid.hash, NULL) == OBJ_BLOB)
 +                      if (oid_object_info(the_repository, &blob_oid, NULL) == OBJ_BLOB)
-                               return stream_blob_to_fd(1, &blob_oid, NULL, 0);
+                               return stream_blob(&blob_oid);
                        /*
                         * we attempted to dereference a tag to a blob
                         * and failed; there may be new dereference
                         * fall-back to the usual case.
                         */
                }
 -              buf = read_object_with_reference(oid.hash, exp_type, &size, NULL);
 +              buf = read_object_with_reference(&oid, exp_type, &size, NULL);
                break;
  
        default:
                die("git cat-file %s: bad file", obj_name);
  
        write_or_die(1, buf, size);
 +      free(buf);
 +      free(obj_context.path);
        return 0;
  }
  
@@@ -232,7 -229,7 +239,7 @@@ static void expand_atom(struct strbuf *
                if (data->mark_query)
                        data->info.typep = &data->type;
                else
 -                      strbuf_addstr(sb, typename(data->type));
 +                      strbuf_addstr(sb, type_name(data->type));
        } else if (is_atom("objectsize", atom, len)) {
                if (data->mark_query)
                        data->info.sizep = &data->size;
@@@ -307,25 -304,25 +314,26 @@@ static void print_object_or_die(struct 
                                enum object_type type;
                                if (!textconv_object(data->rest, 0100644, oid,
                                                     1, &contents, &size))
 -                                      contents = read_sha1_file(oid->hash, &type,
 -                                                                &size);
 +                                      contents = read_object_file(oid,
 +                                                                  &type,
 +                                                                  &size);
                                if (!contents)
                                        die("could not convert '%s' %s",
                                            oid_to_hex(oid), data->rest);
                        } else
 -                              die("BUG: invalid cmdmode: %c", opt->cmdmode);
 +                              BUG("invalid cmdmode: %c", opt->cmdmode);
                        batch_write(opt, contents, size);
                        free(contents);
-               } else if (stream_blob_to_fd(1, oid, NULL, 0) < 0)
-                       die("unable to stream %s to stdout", oid_to_hex(oid));
+               } else {
+                       stream_blob(oid);
+               }
        }
        else {
                enum object_type type;
                unsigned long size;
                void *contents;
  
 -              contents = read_sha1_file(oid->hash, &type, &size);
 +              contents = read_object_file(oid, &type, &size);
                if (!contents)
                        die("object %s disappeared", oid_to_hex(oid));
                if (type != data->type)
        }
  }
  
 -static void batch_object_write(const char *obj_name, struct batch_options *opt,
 +static void batch_object_write(const char *obj_name,
 +                             struct strbuf *scratch,
 +                             struct batch_options *opt,
                               struct expand_data *data)
  {
 -      struct strbuf buf = STRBUF_INIT;
 -
        if (!data->skip_object_info &&
 -          sha1_object_info_extended(data->oid.hash, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
 +          oid_object_info_extended(the_repository, &data->oid, &data->info,
 +                                   OBJECT_INFO_LOOKUP_REPLACE) < 0) {
                printf("%s missing\n",
                       obj_name ? obj_name : oid_to_hex(&data->oid));
                fflush(stdout);
                return;
        }
  
 -      strbuf_expand(&buf, opt->format, expand_format, data);
 -      strbuf_addch(&buf, '\n');
 -      batch_write(opt, buf.buf, buf.len);
 -      strbuf_release(&buf);
 +      strbuf_reset(scratch);
 +      strbuf_expand(scratch, opt->format, expand_format, data);
 +      strbuf_addch(scratch, '\n');
 +      batch_write(opt, scratch->buf, scratch->len);
  
        if (opt->print_contents) {
                print_object_or_die(opt, data);
        }
  }
  
 -static void batch_one_object(const char *obj_name, struct batch_options *opt,
 +static void batch_one_object(const char *obj_name,
 +                           struct strbuf *scratch,
 +                           struct batch_options *opt,
                             struct expand_data *data)
  {
        struct object_context ctx;
 -      int flags = opt->follow_symlinks ? GET_SHA1_FOLLOW_SYMLINKS : 0;
 +      int flags = opt->follow_symlinks ? GET_OID_FOLLOW_SYMLINKS : 0;
        enum follow_symlinks_result result;
  
 -      result = get_sha1_with_context(obj_name, flags, data->oid.hash, &ctx);
 +      result = get_oid_with_context(obj_name, flags, &data->oid, &ctx);
        if (result != FOUND) {
                switch (result) {
                case MISSING_OBJECT:
                               (uintmax_t)strlen(obj_name), obj_name);
                        break;
                default:
 -                      die("BUG: unknown get_sha1_with_context result %d\n",
 +                      BUG("unknown get_sha1_with_context result %d\n",
                               result);
                        break;
                }
                return;
        }
  
 -      batch_object_write(obj_name, opt, data);
 +      batch_object_write(obj_name, scratch, opt, data);
  }
  
  struct object_cb_data {
        struct batch_options *opt;
        struct expand_data *expand;
 +      struct oidset *seen;
 +      struct strbuf *scratch;
  };
  
 -static int batch_object_cb(const unsigned char sha1[20], void *vdata)
 +static int batch_object_cb(const struct object_id *oid, void *vdata)
  {
        struct object_cb_data *data = vdata;
 -      hashcpy(data->expand->oid.hash, sha1);
 -      batch_object_write(NULL, data->opt, data->expand);
 +      oidcpy(&data->expand->oid, oid);
 +      batch_object_write(NULL, data->scratch, data->opt, data->expand);
        return 0;
  }
  
 -static int batch_loose_object(const unsigned char *sha1,
 -                            const char *path,
 -                            void *data)
 +static int collect_loose_object(const struct object_id *oid,
 +                              const char *path,
 +                              void *data)
  {
 -      sha1_array_append(data, sha1);
 +      oid_array_append(data, oid);
        return 0;
  }
  
 -static int batch_packed_object(const unsigned char *sha1,
 -                             struct packed_git *pack,
 -                             uint32_t pos,
 -                             void *data)
 +static int collect_packed_object(const struct object_id *oid,
 +                               struct packed_git *pack,
 +                               uint32_t pos,
 +                               void *data)
  {
 -      sha1_array_append(data, sha1);
 +      oid_array_append(data, oid);
        return 0;
  }
  
 +static int batch_unordered_object(const struct object_id *oid, void *vdata)
 +{
 +      struct object_cb_data *data = vdata;
 +
 +      if (oidset_insert(data->seen, oid))
 +              return 0;
 +
 +      return batch_object_cb(oid, data);
 +}
 +
 +static int batch_unordered_loose(const struct object_id *oid,
 +                               const char *path,
 +                               void *data)
 +{
 +      return batch_unordered_object(oid, data);
 +}
 +
 +static int batch_unordered_packed(const struct object_id *oid,
 +                                struct packed_git *pack,
 +                                uint32_t pos,
 +                                void *data)
 +{
 +      return batch_unordered_object(oid, data);
 +}
 +
  static int batch_objects(struct batch_options *opt)
  {
 -      struct strbuf buf = STRBUF_INIT;
 +      struct strbuf input = STRBUF_INIT;
 +      struct strbuf output = STRBUF_INIT;
        struct expand_data data;
        int save_warning;
        int retval = 0;
         */
        memset(&data, 0, sizeof(data));
        data.mark_query = 1;
 -      strbuf_expand(&buf, opt->format, expand_format, &data);
 +      strbuf_expand(&output, opt->format, expand_format, &data);
        data.mark_query = 0;
 +      strbuf_release(&output);
        if (opt->cmdmode)
                data.split_on_whitespace = 1;
  
                data.info.typep = &data.type;
  
        if (opt->all_objects) {
 -              struct sha1_array sa = SHA1_ARRAY_INIT;
                struct object_cb_data cb;
  
 -              for_each_loose_object(batch_loose_object, &sa, 0);
 -              for_each_packed_object(batch_packed_object, &sa, 0);
 +              if (repository_format_partial_clone)
 +                      warning("This repository has extensions.partialClone set. Some objects may not be loaded.");
  
                cb.opt = opt;
                cb.expand = &data;
 -              sha1_array_for_each_unique(&sa, batch_object_cb, &cb);
 +              cb.scratch = &output;
 +
 +              if (opt->unordered) {
 +                      struct oidset seen = OIDSET_INIT;
 +
 +                      cb.seen = &seen;
 +
 +                      for_each_loose_object(batch_unordered_loose, &cb, 0);
 +                      for_each_packed_object(batch_unordered_packed, &cb,
 +                                             FOR_EACH_OBJECT_PACK_ORDER);
 +
 +                      oidset_clear(&seen);
 +              } else {
 +                      struct oid_array sa = OID_ARRAY_INIT;
 +
 +                      for_each_loose_object(collect_loose_object, &sa, 0);
 +                      for_each_packed_object(collect_packed_object, &sa, 0);
 +
 +                      oid_array_for_each_unique(&sa, batch_object_cb, &cb);
 +
 +                      oid_array_clear(&sa);
 +              }
  
 -              sha1_array_clear(&sa);
 +              strbuf_release(&output);
                return 0;
        }
  
        save_warning = warn_on_object_refname_ambiguity;
        warn_on_object_refname_ambiguity = 0;
  
 -      while (strbuf_getline(&buf, stdin) != EOF) {
 +      while (strbuf_getline(&input, stdin) != EOF) {
                if (data.split_on_whitespace) {
                        /*
                         * Split at first whitespace, tying off the beginning
                         * of the string and saving the remainder (or NULL) in
                         * data.rest.
                         */
 -                      char *p = strpbrk(buf.buf, " \t");
 +                      char *p = strpbrk(input.buf, " \t");
                        if (p) {
                                while (*p && strchr(" \t", *p))
                                        *p++ = '\0';
                        data.rest = p;
                }
  
 -              batch_one_object(buf.buf, opt, &data);
 +              batch_one_object(input.buf, &output, opt, &data);
        }
  
 -      strbuf_release(&buf);
 +      strbuf_release(&input);
 +      strbuf_release(&output);
        warn_on_object_refname_ambiguity = save_warning;
        return retval;
  }
@@@ -637,8 -581,6 +645,8 @@@ int cmd_cat_file(int argc, const char *
                         N_("follow in-tree symlinks (used with --batch or --batch-check)")),
                OPT_BOOL(0, "batch-all-objects", &batch.all_objects,
                         N_("show all objects with --batch or --batch-check")),
 +              OPT_BOOL(0, "unordered", &batch.unordered,
 +                       N_("do not order --batch-all-objects output")),
                OPT_END()
        };
  
diff --combined sha1-file.c
index 97b74238483e00c3f07bd5ab0879eb84bf5c8dfa,0000000000000000000000000000000000000000..6d6c3af6598c33623d10a930a39afaf9de74db1a
mode 100644,000000..100644
--- /dev/null
@@@ -1,2282 -1,0 +1,2283 @@@
-              (status == Z_OK || status == Z_BUF_ERROR)) {
 +/*
 + * GIT - The information manager from hell
 + *
 + * Copyright (C) Linus Torvalds, 2005
 + *
 + * This handles basic git sha1 object files - packing, unpacking,
 + * creation etc.
 + */
 +#include "cache.h"
 +#include "config.h"
 +#include "string-list.h"
 +#include "lockfile.h"
 +#include "delta.h"
 +#include "pack.h"
 +#include "blob.h"
 +#include "commit.h"
 +#include "run-command.h"
 +#include "tag.h"
 +#include "tree.h"
 +#include "tree-walk.h"
 +#include "refs.h"
 +#include "pack-revindex.h"
 +#include "sha1-lookup.h"
 +#include "bulk-checkin.h"
 +#include "repository.h"
 +#include "replace-object.h"
 +#include "streaming.h"
 +#include "dir.h"
 +#include "list.h"
 +#include "mergesort.h"
 +#include "quote.h"
 +#include "packfile.h"
 +#include "fetch-object.h"
 +#include "object-store.h"
 +
 +/* The maximum size for an object header. */
 +#define MAX_HEADER_LEN 32
 +
 +
 +#define EMPTY_TREE_SHA1_BIN_LITERAL \
 +       "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
 +       "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
 +
 +#define EMPTY_BLOB_SHA1_BIN_LITERAL \
 +      "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
 +      "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
 +
 +const unsigned char null_sha1[GIT_MAX_RAWSZ];
 +const struct object_id null_oid;
 +static const struct object_id empty_tree_oid = {
 +      EMPTY_TREE_SHA1_BIN_LITERAL
 +};
 +static const struct object_id empty_blob_oid = {
 +      EMPTY_BLOB_SHA1_BIN_LITERAL
 +};
 +
 +static void git_hash_sha1_init(git_hash_ctx *ctx)
 +{
 +      git_SHA1_Init(&ctx->sha1);
 +}
 +
 +static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
 +{
 +      git_SHA1_Update(&ctx->sha1, data, len);
 +}
 +
 +static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
 +{
 +      git_SHA1_Final(hash, &ctx->sha1);
 +}
 +
 +static void git_hash_unknown_init(git_hash_ctx *ctx)
 +{
 +      BUG("trying to init unknown hash");
 +}
 +
 +static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
 +{
 +      BUG("trying to update unknown hash");
 +}
 +
 +static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
 +{
 +      BUG("trying to finalize unknown hash");
 +}
 +
 +const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
 +      {
 +              NULL,
 +              0x00000000,
 +              0,
 +              0,
 +              git_hash_unknown_init,
 +              git_hash_unknown_update,
 +              git_hash_unknown_final,
 +              NULL,
 +              NULL,
 +      },
 +      {
 +              "sha-1",
 +              /* "sha1", big-endian */
 +              0x73686131,
 +              GIT_SHA1_RAWSZ,
 +              GIT_SHA1_HEXSZ,
 +              git_hash_sha1_init,
 +              git_hash_sha1_update,
 +              git_hash_sha1_final,
 +              &empty_tree_oid,
 +              &empty_blob_oid,
 +      },
 +};
 +
 +const char *empty_tree_oid_hex(void)
 +{
 +      static char buf[GIT_MAX_HEXSZ + 1];
 +      return oid_to_hex_r(buf, the_hash_algo->empty_tree);
 +}
 +
 +const char *empty_blob_oid_hex(void)
 +{
 +      static char buf[GIT_MAX_HEXSZ + 1];
 +      return oid_to_hex_r(buf, the_hash_algo->empty_blob);
 +}
 +
 +/*
 + * This is meant to hold a *small* number of objects that you would
 + * want read_sha1_file() to be able to return, but yet you do not want
 + * to write them into the object store (e.g. a browse-only
 + * application).
 + */
 +static struct cached_object {
 +      struct object_id oid;
 +      enum object_type type;
 +      void *buf;
 +      unsigned long size;
 +} *cached_objects;
 +static int cached_object_nr, cached_object_alloc;
 +
 +static struct cached_object empty_tree = {
 +      { EMPTY_TREE_SHA1_BIN_LITERAL },
 +      OBJ_TREE,
 +      "",
 +      0
 +};
 +
 +static struct cached_object *find_cached_object(const struct object_id *oid)
 +{
 +      int i;
 +      struct cached_object *co = cached_objects;
 +
 +      for (i = 0; i < cached_object_nr; i++, co++) {
 +              if (!oidcmp(&co->oid, oid))
 +                      return co;
 +      }
 +      if (!oidcmp(oid, the_hash_algo->empty_tree))
 +              return &empty_tree;
 +      return NULL;
 +}
 +
 +
 +static int get_conv_flags(unsigned flags)
 +{
 +      if (flags & HASH_RENORMALIZE)
 +              return CONV_EOL_RENORMALIZE;
 +      else if (flags & HASH_WRITE_OBJECT)
 +              return global_conv_flags_eol | CONV_WRITE_OBJECT;
 +      else
 +              return 0;
 +}
 +
 +
 +int mkdir_in_gitdir(const char *path)
 +{
 +      if (mkdir(path, 0777)) {
 +              int saved_errno = errno;
 +              struct stat st;
 +              struct strbuf sb = STRBUF_INIT;
 +
 +              if (errno != EEXIST)
 +                      return -1;
 +              /*
 +               * Are we looking at a path in a symlinked worktree
 +               * whose original repository does not yet have it?
 +               * e.g. .git/rr-cache pointing at its original
 +               * repository in which the user hasn't performed any
 +               * conflict resolution yet?
 +               */
 +              if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
 +                  strbuf_readlink(&sb, path, st.st_size) ||
 +                  !is_absolute_path(sb.buf) ||
 +                  mkdir(sb.buf, 0777)) {
 +                      strbuf_release(&sb);
 +                      errno = saved_errno;
 +                      return -1;
 +              }
 +              strbuf_release(&sb);
 +      }
 +      return adjust_shared_perm(path);
 +}
 +
 +enum scld_error safe_create_leading_directories(char *path)
 +{
 +      char *next_component = path + offset_1st_component(path);
 +      enum scld_error ret = SCLD_OK;
 +
 +      while (ret == SCLD_OK && next_component) {
 +              struct stat st;
 +              char *slash = next_component, slash_character;
 +
 +              while (*slash && !is_dir_sep(*slash))
 +                      slash++;
 +
 +              if (!*slash)
 +                      break;
 +
 +              next_component = slash + 1;
 +              while (is_dir_sep(*next_component))
 +                      next_component++;
 +              if (!*next_component)
 +                      break;
 +
 +              slash_character = *slash;
 +              *slash = '\0';
 +              if (!stat(path, &st)) {
 +                      /* path exists */
 +                      if (!S_ISDIR(st.st_mode)) {
 +                              errno = ENOTDIR;
 +                              ret = SCLD_EXISTS;
 +                      }
 +              } else if (mkdir(path, 0777)) {
 +                      if (errno == EEXIST &&
 +                          !stat(path, &st) && S_ISDIR(st.st_mode))
 +                              ; /* somebody created it since we checked */
 +                      else if (errno == ENOENT)
 +                              /*
 +                               * Either mkdir() failed because
 +                               * somebody just pruned the containing
 +                               * directory, or stat() failed because
 +                               * the file that was in our way was
 +                               * just removed.  Either way, inform
 +                               * the caller that it might be worth
 +                               * trying again:
 +                               */
 +                              ret = SCLD_VANISHED;
 +                      else
 +                              ret = SCLD_FAILED;
 +              } else if (adjust_shared_perm(path)) {
 +                      ret = SCLD_PERMS;
 +              }
 +              *slash = slash_character;
 +      }
 +      return ret;
 +}
 +
 +enum scld_error safe_create_leading_directories_const(const char *path)
 +{
 +      int save_errno;
 +      /* path points to cache entries, so xstrdup before messing with it */
 +      char *buf = xstrdup(path);
 +      enum scld_error result = safe_create_leading_directories(buf);
 +
 +      save_errno = errno;
 +      free(buf);
 +      errno = save_errno;
 +      return result;
 +}
 +
 +int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
 +{
 +      /*
 +       * The number of times we will try to remove empty directories
 +       * in the way of path. This is only 1 because if another
 +       * process is racily creating directories that conflict with
 +       * us, we don't want to fight against them.
 +       */
 +      int remove_directories_remaining = 1;
 +
 +      /*
 +       * The number of times that we will try to create the
 +       * directories containing path. We are willing to attempt this
 +       * more than once, because another process could be trying to
 +       * clean up empty directories at the same time as we are
 +       * trying to create them.
 +       */
 +      int create_directories_remaining = 3;
 +
 +      /* A scratch copy of path, filled lazily if we need it: */
 +      struct strbuf path_copy = STRBUF_INIT;
 +
 +      int ret, save_errno;
 +
 +      /* Sanity check: */
 +      assert(*path);
 +
 +retry_fn:
 +      ret = fn(path, cb);
 +      save_errno = errno;
 +      if (!ret)
 +              goto out;
 +
 +      if (errno == EISDIR && remove_directories_remaining-- > 0) {
 +              /*
 +               * A directory is in the way. Maybe it is empty; try
 +               * to remove it:
 +               */
 +              if (!path_copy.len)
 +                      strbuf_addstr(&path_copy, path);
 +
 +              if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
 +                      goto retry_fn;
 +      } else if (errno == ENOENT && create_directories_remaining-- > 0) {
 +              /*
 +               * Maybe the containing directory didn't exist, or
 +               * maybe it was just deleted by a process that is
 +               * racing with us to clean up empty directories. Try
 +               * to create it:
 +               */
 +              enum scld_error scld_result;
 +
 +              if (!path_copy.len)
 +                      strbuf_addstr(&path_copy, path);
 +
 +              do {
 +                      scld_result = safe_create_leading_directories(path_copy.buf);
 +                      if (scld_result == SCLD_OK)
 +                              goto retry_fn;
 +              } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
 +      }
 +
 +out:
 +      strbuf_release(&path_copy);
 +      errno = save_errno;
 +      return ret;
 +}
 +
 +static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 +{
 +      int i;
 +      for (i = 0; i < the_hash_algo->rawsz; i++) {
 +              static char hex[] = "0123456789abcdef";
 +              unsigned int val = sha1[i];
 +              strbuf_addch(buf, hex[val >> 4]);
 +              strbuf_addch(buf, hex[val & 0xf]);
 +              if (!i)
 +                      strbuf_addch(buf, '/');
 +      }
 +}
 +
 +void sha1_file_name(struct repository *r, struct strbuf *buf, const unsigned char *sha1)
 +{
 +      strbuf_addstr(buf, r->objects->objectdir);
 +      strbuf_addch(buf, '/');
 +      fill_sha1_path(buf, sha1);
 +}
 +
 +struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
 +{
 +      strbuf_setlen(&alt->scratch, alt->base_len);
 +      return &alt->scratch;
 +}
 +
 +static const char *alt_sha1_path(struct alternate_object_database *alt,
 +                               const unsigned char *sha1)
 +{
 +      struct strbuf *buf = alt_scratch_buf(alt);
 +      fill_sha1_path(buf, sha1);
 +      return buf->buf;
 +}
 +
 +/*
 + * Return non-zero iff the path is usable as an alternate object database.
 + */
 +static int alt_odb_usable(struct raw_object_store *o,
 +                        struct strbuf *path,
 +                        const char *normalized_objdir)
 +{
 +      struct alternate_object_database *alt;
 +
 +      /* Detect cases where alternate disappeared */
 +      if (!is_directory(path->buf)) {
 +              error(_("object directory %s does not exist; "
 +                      "check .git/objects/info/alternates"),
 +                    path->buf);
 +              return 0;
 +      }
 +
 +      /*
 +       * Prevent the common mistake of listing the same
 +       * thing twice, or object directory itself.
 +       */
 +      for (alt = o->alt_odb_list; alt; alt = alt->next) {
 +              if (!fspathcmp(path->buf, alt->path))
 +                      return 0;
 +      }
 +      if (!fspathcmp(path->buf, normalized_objdir))
 +              return 0;
 +
 +      return 1;
 +}
 +
 +/*
 + * Prepare alternate object database registry.
 + *
 + * The variable alt_odb_list points at the list of struct
 + * alternate_object_database.  The elements on this list come from
 + * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
 + * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
 + * whose contents is similar to that environment variable but can be
 + * LF separated.  Its base points at a statically allocated buffer that
 + * contains "/the/directory/corresponding/to/.git/objects/...", while
 + * its name points just after the slash at the end of ".git/objects/"
 + * in the example above, and has enough space to hold 40-byte hex
 + * SHA1, an extra slash for the first level indirection, and the
 + * terminating NUL.
 + */
 +static void read_info_alternates(struct repository *r,
 +                               const char *relative_base,
 +                               int depth);
 +static int link_alt_odb_entry(struct repository *r, const char *entry,
 +      const char *relative_base, int depth, const char *normalized_objdir)
 +{
 +      struct alternate_object_database *ent;
 +      struct strbuf pathbuf = STRBUF_INIT;
 +
 +      if (!is_absolute_path(entry) && relative_base) {
 +              strbuf_realpath(&pathbuf, relative_base, 1);
 +              strbuf_addch(&pathbuf, '/');
 +      }
 +      strbuf_addstr(&pathbuf, entry);
 +
 +      if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
 +              error(_("unable to normalize alternate object path: %s"),
 +                    pathbuf.buf);
 +              strbuf_release(&pathbuf);
 +              return -1;
 +      }
 +
 +      /*
 +       * The trailing slash after the directory name is given by
 +       * this function at the end. Remove duplicates.
 +       */
 +      while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
 +              strbuf_setlen(&pathbuf, pathbuf.len - 1);
 +
 +      if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
 +              strbuf_release(&pathbuf);
 +              return -1;
 +      }
 +
 +      ent = alloc_alt_odb(pathbuf.buf);
 +
 +      /* add the alternate entry */
 +      *r->objects->alt_odb_tail = ent;
 +      r->objects->alt_odb_tail = &(ent->next);
 +      ent->next = NULL;
 +
 +      /* recursively add alternates */
 +      read_info_alternates(r, pathbuf.buf, depth + 1);
 +
 +      strbuf_release(&pathbuf);
 +      return 0;
 +}
 +
 +static const char *parse_alt_odb_entry(const char *string,
 +                                     int sep,
 +                                     struct strbuf *out)
 +{
 +      const char *end;
 +
 +      strbuf_reset(out);
 +
 +      if (*string == '#') {
 +              /* comment; consume up to next separator */
 +              end = strchrnul(string, sep);
 +      } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
 +              /*
 +               * quoted path; unquote_c_style has copied the
 +               * data for us and set "end". Broken quoting (e.g.,
 +               * an entry that doesn't end with a quote) falls
 +               * back to the unquoted case below.
 +               */
 +      } else {
 +              /* normal, unquoted path */
 +              end = strchrnul(string, sep);
 +              strbuf_add(out, string, end - string);
 +      }
 +
 +      if (*end)
 +              end++;
 +      return end;
 +}
 +
 +static void link_alt_odb_entries(struct repository *r, const char *alt,
 +                               int sep, const char *relative_base, int depth)
 +{
 +      struct strbuf objdirbuf = STRBUF_INIT;
 +      struct strbuf entry = STRBUF_INIT;
 +
 +      if (!alt || !*alt)
 +              return;
 +
 +      if (depth > 5) {
 +              error(_("%s: ignoring alternate object stores, nesting too deep"),
 +                              relative_base);
 +              return;
 +      }
 +
 +      strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
 +      if (strbuf_normalize_path(&objdirbuf) < 0)
 +              die(_("unable to normalize object directory: %s"),
 +                  objdirbuf.buf);
 +
 +      while (*alt) {
 +              alt = parse_alt_odb_entry(alt, sep, &entry);
 +              if (!entry.len)
 +                      continue;
 +              link_alt_odb_entry(r, entry.buf,
 +                                 relative_base, depth, objdirbuf.buf);
 +      }
 +      strbuf_release(&entry);
 +      strbuf_release(&objdirbuf);
 +}
 +
 +static void read_info_alternates(struct repository *r,
 +                               const char *relative_base,
 +                               int depth)
 +{
 +      char *path;
 +      struct strbuf buf = STRBUF_INIT;
 +
 +      path = xstrfmt("%s/info/alternates", relative_base);
 +      if (strbuf_read_file(&buf, path, 1024) < 0) {
 +              warn_on_fopen_errors(path);
 +              free(path);
 +              return;
 +      }
 +
 +      link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
 +      strbuf_release(&buf);
 +      free(path);
 +}
 +
 +struct alternate_object_database *alloc_alt_odb(const char *dir)
 +{
 +      struct alternate_object_database *ent;
 +
 +      FLEX_ALLOC_STR(ent, path, dir);
 +      strbuf_init(&ent->scratch, 0);
 +      strbuf_addf(&ent->scratch, "%s/", dir);
 +      ent->base_len = ent->scratch.len;
 +
 +      return ent;
 +}
 +
 +void add_to_alternates_file(const char *reference)
 +{
 +      struct lock_file lock = LOCK_INIT;
 +      char *alts = git_pathdup("objects/info/alternates");
 +      FILE *in, *out;
 +      int found = 0;
 +
 +      hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
 +      out = fdopen_lock_file(&lock, "w");
 +      if (!out)
 +              die_errno(_("unable to fdopen alternates lockfile"));
 +
 +      in = fopen(alts, "r");
 +      if (in) {
 +              struct strbuf line = STRBUF_INIT;
 +
 +              while (strbuf_getline(&line, in) != EOF) {
 +                      if (!strcmp(reference, line.buf)) {
 +                              found = 1;
 +                              break;
 +                      }
 +                      fprintf_or_die(out, "%s\n", line.buf);
 +              }
 +
 +              strbuf_release(&line);
 +              fclose(in);
 +      }
 +      else if (errno != ENOENT)
 +              die_errno(_("unable to read alternates file"));
 +
 +      if (found) {
 +              rollback_lock_file(&lock);
 +      } else {
 +              fprintf_or_die(out, "%s\n", reference);
 +              if (commit_lock_file(&lock))
 +                      die_errno(_("unable to move new alternates file into place"));
 +              if (the_repository->objects->alt_odb_tail)
 +                      link_alt_odb_entries(the_repository, reference,
 +                                           '\n', NULL, 0);
 +      }
 +      free(alts);
 +}
 +
 +void add_to_alternates_memory(const char *reference)
 +{
 +      /*
 +       * Make sure alternates are initialized, or else our entry may be
 +       * overwritten when they are.
 +       */
 +      prepare_alt_odb(the_repository);
 +
 +      link_alt_odb_entries(the_repository, reference,
 +                           '\n', NULL, 0);
 +}
 +
 +/*
 + * Compute the exact path an alternate is at and returns it. In case of
 + * error NULL is returned and the human readable error is added to `err`
 + * `path` may be relative and should point to $GIT_DIR.
 + * `err` must not be null.
 + */
 +char *compute_alternate_path(const char *path, struct strbuf *err)
 +{
 +      char *ref_git = NULL;
 +      const char *repo, *ref_git_s;
 +      int seen_error = 0;
 +
 +      ref_git_s = real_path_if_valid(path);
 +      if (!ref_git_s) {
 +              seen_error = 1;
 +              strbuf_addf(err, _("path '%s' does not exist"), path);
 +              goto out;
 +      } else
 +              /*
 +               * Beware: read_gitfile(), real_path() and mkpath()
 +               * return static buffer
 +               */
 +              ref_git = xstrdup(ref_git_s);
 +
 +      repo = read_gitfile(ref_git);
 +      if (!repo)
 +              repo = read_gitfile(mkpath("%s/.git", ref_git));
 +      if (repo) {
 +              free(ref_git);
 +              ref_git = xstrdup(repo);
 +      }
 +
 +      if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
 +              char *ref_git_git = mkpathdup("%s/.git", ref_git);
 +              free(ref_git);
 +              ref_git = ref_git_git;
 +      } else if (!is_directory(mkpath("%s/objects", ref_git))) {
 +              struct strbuf sb = STRBUF_INIT;
 +              seen_error = 1;
 +              if (get_common_dir(&sb, ref_git)) {
 +                      strbuf_addf(err,
 +                                  _("reference repository '%s' as a linked "
 +                                    "checkout is not supported yet."),
 +                                  path);
 +                      goto out;
 +              }
 +
 +              strbuf_addf(err, _("reference repository '%s' is not a "
 +                                      "local repository."), path);
 +              goto out;
 +      }
 +
 +      if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
 +              strbuf_addf(err, _("reference repository '%s' is shallow"),
 +                          path);
 +              seen_error = 1;
 +              goto out;
 +      }
 +
 +      if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
 +              strbuf_addf(err,
 +                          _("reference repository '%s' is grafted"),
 +                          path);
 +              seen_error = 1;
 +              goto out;
 +      }
 +
 +out:
 +      if (seen_error) {
 +              FREE_AND_NULL(ref_git);
 +      }
 +
 +      return ref_git;
 +}
 +
 +int foreach_alt_odb(alt_odb_fn fn, void *cb)
 +{
 +      struct alternate_object_database *ent;
 +      int r = 0;
 +
 +      prepare_alt_odb(the_repository);
 +      for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
 +              r = fn(ent, cb);
 +              if (r)
 +                      break;
 +      }
 +      return r;
 +}
 +
 +void prepare_alt_odb(struct repository *r)
 +{
 +      if (r->objects->alt_odb_tail)
 +              return;
 +
 +      r->objects->alt_odb_tail = &r->objects->alt_odb_list;
 +      link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
 +
 +      read_info_alternates(r, r->objects->objectdir, 0);
 +}
 +
 +/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
 +static int freshen_file(const char *fn)
 +{
 +      struct utimbuf t;
 +      t.actime = t.modtime = time(NULL);
 +      return !utime(fn, &t);
 +}
 +
 +/*
 + * All of the check_and_freshen functions return 1 if the file exists and was
 + * freshened (if freshening was requested), 0 otherwise. If they return
 + * 0, you should not assume that it is safe to skip a write of the object (it
 + * either does not exist on disk, or has a stale mtime and may be subject to
 + * pruning).
 + */
 +int check_and_freshen_file(const char *fn, int freshen)
 +{
 +      if (access(fn, F_OK))
 +              return 0;
 +      if (freshen && !freshen_file(fn))
 +              return 0;
 +      return 1;
 +}
 +
 +static int check_and_freshen_local(const struct object_id *oid, int freshen)
 +{
 +      static struct strbuf buf = STRBUF_INIT;
 +
 +      strbuf_reset(&buf);
 +      sha1_file_name(the_repository, &buf, oid->hash);
 +
 +      return check_and_freshen_file(buf.buf, freshen);
 +}
 +
 +static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
 +{
 +      struct alternate_object_database *alt;
 +      prepare_alt_odb(the_repository);
 +      for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
 +              const char *path = alt_sha1_path(alt, oid->hash);
 +              if (check_and_freshen_file(path, freshen))
 +                      return 1;
 +      }
 +      return 0;
 +}
 +
 +static int check_and_freshen(const struct object_id *oid, int freshen)
 +{
 +      return check_and_freshen_local(oid, freshen) ||
 +             check_and_freshen_nonlocal(oid, freshen);
 +}
 +
 +int has_loose_object_nonlocal(const struct object_id *oid)
 +{
 +      return check_and_freshen_nonlocal(oid, 0);
 +}
 +
 +static int has_loose_object(const struct object_id *oid)
 +{
 +      return check_and_freshen(oid, 0);
 +}
 +
 +static void mmap_limit_check(size_t length)
 +{
 +      static size_t limit = 0;
 +      if (!limit) {
 +              limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
 +              if (!limit)
 +                      limit = SIZE_MAX;
 +      }
 +      if (length > limit)
 +              die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
 +                  (uintmax_t)length, (uintmax_t)limit);
 +}
 +
 +void *xmmap_gently(void *start, size_t length,
 +                int prot, int flags, int fd, off_t offset)
 +{
 +      void *ret;
 +
 +      mmap_limit_check(length);
 +      ret = mmap(start, length, prot, flags, fd, offset);
 +      if (ret == MAP_FAILED) {
 +              if (!length)
 +                      return NULL;
 +              release_pack_memory(length);
 +              ret = mmap(start, length, prot, flags, fd, offset);
 +      }
 +      return ret;
 +}
 +
 +void *xmmap(void *start, size_t length,
 +      int prot, int flags, int fd, off_t offset)
 +{
 +      void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
 +      if (ret == MAP_FAILED)
 +              die_errno(_("mmap failed"));
 +      return ret;
 +}
 +
 +/*
 + * With an in-core object data in "map", rehash it to make sure the
 + * object name actually matches "sha1" to detect object corruption.
 + * With "map" == NULL, try reading the object named with "sha1" using
 + * the streaming interface and rehash it to do the same.
 + */
 +int check_object_signature(const struct object_id *oid, void *map,
 +                         unsigned long size, const char *type)
 +{
 +      struct object_id real_oid;
 +      enum object_type obj_type;
 +      struct git_istream *st;
 +      git_hash_ctx c;
 +      char hdr[MAX_HEADER_LEN];
 +      int hdrlen;
 +
 +      if (map) {
 +              hash_object_file(map, size, type, &real_oid);
 +              return oidcmp(oid, &real_oid) ? -1 : 0;
 +      }
 +
 +      st = open_istream(oid, &obj_type, &size, NULL);
 +      if (!st)
 +              return -1;
 +
 +      /* Generate the header */
 +      hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
 +
 +      /* Sha1.. */
 +      the_hash_algo->init_fn(&c);
 +      the_hash_algo->update_fn(&c, hdr, hdrlen);
 +      for (;;) {
 +              char buf[1024 * 16];
 +              ssize_t readlen = read_istream(st, buf, sizeof(buf));
 +
 +              if (readlen < 0) {
 +                      close_istream(st);
 +                      return -1;
 +              }
 +              if (!readlen)
 +                      break;
 +              the_hash_algo->update_fn(&c, buf, readlen);
 +      }
 +      the_hash_algo->final_fn(real_oid.hash, &c);
 +      close_istream(st);
 +      return oidcmp(oid, &real_oid) ? -1 : 0;
 +}
 +
 +int git_open_cloexec(const char *name, int flags)
 +{
 +      int fd;
 +      static int o_cloexec = O_CLOEXEC;
 +
 +      fd = open(name, flags | o_cloexec);
 +      if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
 +              /* Try again w/o O_CLOEXEC: the kernel might not support it */
 +              o_cloexec &= ~O_CLOEXEC;
 +              fd = open(name, flags | o_cloexec);
 +      }
 +
 +#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
 +      {
 +              static int fd_cloexec = FD_CLOEXEC;
 +
 +              if (!o_cloexec && 0 <= fd && fd_cloexec) {
 +                      /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
 +                      int flags = fcntl(fd, F_GETFD);
 +                      if (fcntl(fd, F_SETFD, flags | fd_cloexec))
 +                              fd_cloexec = 0;
 +              }
 +      }
 +#endif
 +      return fd;
 +}
 +
 +/*
 + * Find "sha1" as a loose object in the local repository or in an alternate.
 + * Returns 0 on success, negative on failure.
 + *
 + * The "path" out-parameter will give the path of the object we found (if any).
 + * Note that it may point to static storage and is only valid until another
 + * call to sha1_file_name(), etc.
 + */
 +static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
 +                        struct stat *st, const char **path)
 +{
 +      struct alternate_object_database *alt;
 +      static struct strbuf buf = STRBUF_INIT;
 +
 +      strbuf_reset(&buf);
 +      sha1_file_name(r, &buf, sha1);
 +      *path = buf.buf;
 +
 +      if (!lstat(*path, st))
 +              return 0;
 +
 +      prepare_alt_odb(r);
 +      errno = ENOENT;
 +      for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
 +              *path = alt_sha1_path(alt, sha1);
 +              if (!lstat(*path, st))
 +                      return 0;
 +      }
 +
 +      return -1;
 +}
 +
 +/*
 + * Like stat_sha1_file(), but actually open the object and return the
 + * descriptor. See the caveats on the "path" parameter above.
 + */
 +static int open_sha1_file(struct repository *r,
 +                        const unsigned char *sha1, const char **path)
 +{
 +      int fd;
 +      struct alternate_object_database *alt;
 +      int most_interesting_errno;
 +      static struct strbuf buf = STRBUF_INIT;
 +
 +      strbuf_reset(&buf);
 +      sha1_file_name(r, &buf, sha1);
 +      *path = buf.buf;
 +
 +      fd = git_open(*path);
 +      if (fd >= 0)
 +              return fd;
 +      most_interesting_errno = errno;
 +
 +      prepare_alt_odb(r);
 +      for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
 +              *path = alt_sha1_path(alt, sha1);
 +              fd = git_open(*path);
 +              if (fd >= 0)
 +                      return fd;
 +              if (most_interesting_errno == ENOENT)
 +                      most_interesting_errno = errno;
 +      }
 +      errno = most_interesting_errno;
 +      return -1;
 +}
 +
 +/*
 + * Map the loose object at "path" if it is not NULL, or the path found by
 + * searching for a loose object named "sha1".
 + */
 +static void *map_sha1_file_1(struct repository *r, const char *path,
 +                           const unsigned char *sha1, unsigned long *size)
 +{
 +      void *map;
 +      int fd;
 +
 +      if (path)
 +              fd = git_open(path);
 +      else
 +              fd = open_sha1_file(r, sha1, &path);
 +      map = NULL;
 +      if (fd >= 0) {
 +              struct stat st;
 +
 +              if (!fstat(fd, &st)) {
 +                      *size = xsize_t(st.st_size);
 +                      if (!*size) {
 +                              /* mmap() is forbidden on empty files */
 +                              error(_("object file %s is empty"), path);
 +                              return NULL;
 +                      }
 +                      map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
 +              }
 +              close(fd);
 +      }
 +      return map;
 +}
 +
 +void *map_sha1_file(struct repository *r,
 +                  const unsigned char *sha1, unsigned long *size)
 +{
 +      return map_sha1_file_1(r, NULL, sha1, size);
 +}
 +
 +static int unpack_sha1_short_header(git_zstream *stream,
 +                                  unsigned char *map, unsigned long mapsize,
 +                                  void *buffer, unsigned long bufsiz)
 +{
 +      /* Get the data stream */
 +      memset(stream, 0, sizeof(*stream));
 +      stream->next_in = map;
 +      stream->avail_in = mapsize;
 +      stream->next_out = buffer;
 +      stream->avail_out = bufsiz;
 +
 +      git_inflate_init(stream);
 +      return git_inflate(stream, 0);
 +}
 +
 +int unpack_sha1_header(git_zstream *stream,
 +                     unsigned char *map, unsigned long mapsize,
 +                     void *buffer, unsigned long bufsiz)
 +{
 +      int status = unpack_sha1_short_header(stream, map, mapsize,
 +                                            buffer, bufsiz);
 +
 +      if (status < Z_OK)
 +              return status;
 +
 +      /* Make sure we have the terminating NUL */
 +      if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 +              return -1;
 +      return 0;
 +}
 +
 +static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
 +                                      unsigned long mapsize, void *buffer,
 +                                      unsigned long bufsiz, struct strbuf *header)
 +{
 +      int status;
 +
 +      status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
 +      if (status < Z_OK)
 +              return -1;
 +
 +      /*
 +       * Check if entire header is unpacked in the first iteration.
 +       */
 +      if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 +              return 0;
 +
 +      /*
 +       * buffer[0..bufsiz] was not large enough.  Copy the partial
 +       * result out to header, and then append the result of further
 +       * reading the stream.
 +       */
 +      strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
 +      stream->next_out = buffer;
 +      stream->avail_out = bufsiz;
 +
 +      do {
 +              status = git_inflate(stream, 0);
 +              strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
 +              if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
 +                      return 0;
 +              stream->next_out = buffer;
 +              stream->avail_out = bufsiz;
 +      } while (status != Z_STREAM_END);
 +      return -1;
 +}
 +
 +static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
 +{
 +      int bytes = strlen(buffer) + 1;
 +      unsigned char *buf = xmallocz(size);
 +      unsigned long n;
 +      int status = Z_OK;
 +
 +      n = stream->total_out - bytes;
 +      if (n > size)
 +              n = size;
 +      memcpy(buf, (char *) buffer + bytes, n);
 +      bytes = n;
 +      if (bytes <= size) {
 +              /*
 +               * The above condition must be (bytes <= size), not
 +               * (bytes < size).  In other words, even though we
 +               * expect no more output and set avail_out to zero,
 +               * the input zlib stream may have bytes that express
 +               * "this concludes the stream", and we *do* want to
 +               * eat that input.
 +               *
 +               * Otherwise we would not be able to test that we
 +               * consumed all the input to reach the expected size;
 +               * we also want to check that zlib tells us that all
 +               * went well with status == Z_STREAM_END at the end.
 +               */
 +              stream->next_out = buf + bytes;
 +              stream->avail_out = size - bytes;
 +              while (status == Z_OK)
 +                      status = git_inflate(stream, Z_FINISH);
 +      }
 +      if (status == Z_STREAM_END && !stream->avail_in) {
 +              git_inflate_end(stream);
 +              return buf;
 +      }
 +
 +      if (status < 0)
 +              error(_("corrupt loose object '%s'"), sha1_to_hex(sha1));
 +      else if (stream->avail_in)
 +              error(_("garbage at end of loose object '%s'"),
 +                    sha1_to_hex(sha1));
 +      free(buf);
 +      return NULL;
 +}
 +
 +/*
 + * We used to just use "sscanf()", but that's actually way
 + * too permissive for what we want to check. So do an anal
 + * object header parse by hand.
 + */
 +static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
 +                             unsigned int flags)
 +{
 +      const char *type_buf = hdr;
 +      unsigned long size;
 +      int type, type_len = 0;
 +
 +      /*
 +       * The type can be of any size but is followed by
 +       * a space.
 +       */
 +      for (;;) {
 +              char c = *hdr++;
 +              if (!c)
 +                      return -1;
 +              if (c == ' ')
 +                      break;
 +              type_len++;
 +      }
 +
 +      type = type_from_string_gently(type_buf, type_len, 1);
 +      if (oi->type_name)
 +              strbuf_add(oi->type_name, type_buf, type_len);
 +      /*
 +       * Set type to 0 if its an unknown object and
 +       * we're obtaining the type using '--allow-unknown-type'
 +       * option.
 +       */
 +      if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
 +              type = 0;
 +      else if (type < 0)
 +              die(_("invalid object type"));
 +      if (oi->typep)
 +              *oi->typep = type;
 +
 +      /*
 +       * The length must follow immediately, and be in canonical
 +       * decimal format (ie "010" is not valid).
 +       */
 +      size = *hdr++ - '0';
 +      if (size > 9)
 +              return -1;
 +      if (size) {
 +              for (;;) {
 +                      unsigned long c = *hdr - '0';
 +                      if (c > 9)
 +                              break;
 +                      hdr++;
 +                      size = size * 10 + c;
 +              }
 +      }
 +
 +      if (oi->sizep)
 +              *oi->sizep = size;
 +
 +      /*
 +       * The length must be followed by a zero byte
 +       */
 +      return *hdr ? -1 : type;
 +}
 +
 +int parse_sha1_header(const char *hdr, unsigned long *sizep)
 +{
 +      struct object_info oi = OBJECT_INFO_INIT;
 +
 +      oi.sizep = sizep;
 +      return parse_sha1_header_extended(hdr, &oi, 0);
 +}
 +
 +static int sha1_loose_object_info(struct repository *r,
 +                                const unsigned char *sha1,
 +                                struct object_info *oi, int flags)
 +{
 +      int status = 0;
 +      unsigned long mapsize;
 +      void *map;
 +      git_zstream stream;
 +      char hdr[MAX_HEADER_LEN];
 +      struct strbuf hdrbuf = STRBUF_INIT;
 +      unsigned long size_scratch;
 +
 +      if (oi->delta_base_sha1)
 +              hashclr(oi->delta_base_sha1);
 +
 +      /*
 +       * If we don't care about type or size, then we don't
 +       * need to look inside the object at all. Note that we
 +       * do not optimize out the stat call, even if the
 +       * caller doesn't care about the disk-size, since our
 +       * return value implicitly indicates whether the
 +       * object even exists.
 +       */
 +      if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
 +              const char *path;
 +              struct stat st;
 +              if (stat_sha1_file(r, sha1, &st, &path) < 0)
 +                      return -1;
 +              if (oi->disk_sizep)
 +                      *oi->disk_sizep = st.st_size;
 +              return 0;
 +      }
 +
 +      map = map_sha1_file(r, sha1, &mapsize);
 +      if (!map)
 +              return -1;
 +
 +      if (!oi->sizep)
 +              oi->sizep = &size_scratch;
 +
 +      if (oi->disk_sizep)
 +              *oi->disk_sizep = mapsize;
 +      if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
 +              if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
 +                      status = error(_("unable to unpack %s header with --allow-unknown-type"),
 +                                     sha1_to_hex(sha1));
 +      } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
 +              status = error(_("unable to unpack %s header"),
 +                             sha1_to_hex(sha1));
 +      if (status < 0)
 +              ; /* Do nothing */
 +      else if (hdrbuf.len) {
 +              if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
 +                      status = error(_("unable to parse %s header with --allow-unknown-type"),
 +                                     sha1_to_hex(sha1));
 +      } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
 +              status = error(_("unable to parse %s header"), sha1_to_hex(sha1));
 +
 +      if (status >= 0 && oi->contentp) {
 +              *oi->contentp = unpack_sha1_rest(&stream, hdr,
 +                                               *oi->sizep, sha1);
 +              if (!*oi->contentp) {
 +                      git_inflate_end(&stream);
 +                      status = -1;
 +              }
 +      } else
 +              git_inflate_end(&stream);
 +
 +      munmap(map, mapsize);
 +      if (status && oi->typep)
 +              *oi->typep = status;
 +      if (oi->sizep == &size_scratch)
 +              oi->sizep = NULL;
 +      strbuf_release(&hdrbuf);
 +      oi->whence = OI_LOOSE;
 +      return (status < 0) ? status : 0;
 +}
 +
 +int fetch_if_missing = 1;
 +
 +int oid_object_info_extended(struct repository *r, const struct object_id *oid,
 +                           struct object_info *oi, unsigned flags)
 +{
 +      static struct object_info blank_oi = OBJECT_INFO_INIT;
 +      struct pack_entry e;
 +      int rtype;
 +      const struct object_id *real = oid;
 +      int already_retried = 0;
 +
 +      if (flags & OBJECT_INFO_LOOKUP_REPLACE)
 +              real = lookup_replace_object(r, oid);
 +
 +      if (is_null_oid(real))
 +              return -1;
 +
 +      if (!oi)
 +              oi = &blank_oi;
 +
 +      if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
 +              struct cached_object *co = find_cached_object(real);
 +              if (co) {
 +                      if (oi->typep)
 +                              *(oi->typep) = co->type;
 +                      if (oi->sizep)
 +                              *(oi->sizep) = co->size;
 +                      if (oi->disk_sizep)
 +                              *(oi->disk_sizep) = 0;
 +                      if (oi->delta_base_sha1)
 +                              hashclr(oi->delta_base_sha1);
 +                      if (oi->type_name)
 +                              strbuf_addstr(oi->type_name, type_name(co->type));
 +                      if (oi->contentp)
 +                              *oi->contentp = xmemdupz(co->buf, co->size);
 +                      oi->whence = OI_CACHED;
 +                      return 0;
 +              }
 +      }
 +
 +      while (1) {
 +              if (find_pack_entry(r, real, &e))
 +                      break;
 +
 +              if (flags & OBJECT_INFO_IGNORE_LOOSE)
 +                      return -1;
 +
 +              /* Most likely it's a loose object. */
 +              if (!sha1_loose_object_info(r, real->hash, oi, flags))
 +                      return 0;
 +
 +              /* Not a loose object; someone else may have just packed it. */
 +              if (!(flags & OBJECT_INFO_QUICK)) {
 +                      reprepare_packed_git(r);
 +                      if (find_pack_entry(r, real, &e))
 +                              break;
 +              }
 +
 +              /* Check if it is a missing object */
 +              if (fetch_if_missing && repository_format_partial_clone &&
 +                  !already_retried && r == the_repository) {
 +                      /*
 +                       * TODO Investigate having fetch_object() return
 +                       * TODO error/success and stopping the music here.
 +                       * TODO Pass a repository struct through fetch_object,
 +                       * such that arbitrary repositories work.
 +                       */
 +                      fetch_object(repository_format_partial_clone, real->hash);
 +                      already_retried = 1;
 +                      continue;
 +              }
 +
 +              return -1;
 +      }
 +
 +      if (oi == &blank_oi)
 +              /*
 +               * We know that the caller doesn't actually need the
 +               * information below, so return early.
 +               */
 +              return 0;
 +      rtype = packed_object_info(r, e.p, e.offset, oi);
 +      if (rtype < 0) {
 +              mark_bad_packed_object(e.p, real->hash);
 +              return oid_object_info_extended(r, real, oi, 0);
 +      } else if (oi->whence == OI_PACKED) {
 +              oi->u.packed.offset = e.offset;
 +              oi->u.packed.pack = e.p;
 +              oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
 +                                       rtype == OBJ_OFS_DELTA);
 +      }
 +
 +      return 0;
 +}
 +
 +/* returns enum object_type or negative */
 +int oid_object_info(struct repository *r,
 +                  const struct object_id *oid,
 +                  unsigned long *sizep)
 +{
 +      enum object_type type;
 +      struct object_info oi = OBJECT_INFO_INIT;
 +
 +      oi.typep = &type;
 +      oi.sizep = sizep;
 +      if (oid_object_info_extended(r, oid, &oi,
 +                                    OBJECT_INFO_LOOKUP_REPLACE) < 0)
 +              return -1;
 +      return type;
 +}
 +
 +static void *read_object(const unsigned char *sha1, enum object_type *type,
 +                       unsigned long *size)
 +{
 +      struct object_id oid;
 +      struct object_info oi = OBJECT_INFO_INIT;
 +      void *content;
 +      oi.typep = type;
 +      oi.sizep = size;
 +      oi.contentp = &content;
 +
 +      hashcpy(oid.hash, sha1);
 +
 +      if (oid_object_info_extended(the_repository, &oid, &oi, 0) < 0)
 +              return NULL;
 +      return content;
 +}
 +
 +int pretend_object_file(void *buf, unsigned long len, enum object_type type,
 +                      struct object_id *oid)
 +{
 +      struct cached_object *co;
 +
 +      hash_object_file(buf, len, type_name(type), oid);
 +      if (has_sha1_file(oid->hash) || find_cached_object(oid))
 +              return 0;
 +      ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
 +      co = &cached_objects[cached_object_nr++];
 +      co->size = len;
 +      co->type = type;
 +      co->buf = xmalloc(len);
 +      memcpy(co->buf, buf, len);
 +      oidcpy(&co->oid, oid);
 +      return 0;
 +}
 +
 +/*
 + * This function dies on corrupt objects; the callers who want to
 + * deal with them should arrange to call read_object() and give error
 + * messages themselves.
 + */
 +void *read_object_file_extended(const struct object_id *oid,
 +                              enum object_type *type,
 +                              unsigned long *size,
 +                              int lookup_replace)
 +{
 +      void *data;
 +      const struct packed_git *p;
 +      const char *path;
 +      struct stat st;
 +      const struct object_id *repl = lookup_replace ?
 +              lookup_replace_object(the_repository, oid) : oid;
 +
 +      errno = 0;
 +      data = read_object(repl->hash, type, size);
 +      if (data)
 +              return data;
 +
 +      if (errno && errno != ENOENT)
 +              die_errno(_("failed to read object %s"), oid_to_hex(oid));
 +
 +      /* die if we replaced an object with one that does not exist */
 +      if (repl != oid)
 +              die(_("replacement %s not found for %s"),
 +                  oid_to_hex(repl), oid_to_hex(oid));
 +
 +      if (!stat_sha1_file(the_repository, repl->hash, &st, &path))
 +              die(_("loose object %s (stored in %s) is corrupt"),
 +                  oid_to_hex(repl), path);
 +
 +      if ((p = has_packed_and_bad(repl->hash)) != NULL)
 +              die(_("packed object %s (stored in %s) is corrupt"),
 +                  oid_to_hex(repl), p->pack_name);
 +
 +      return NULL;
 +}
 +
 +void *read_object_with_reference(const struct object_id *oid,
 +                               const char *required_type_name,
 +                               unsigned long *size,
 +                               struct object_id *actual_oid_return)
 +{
 +      enum object_type type, required_type;
 +      void *buffer;
 +      unsigned long isize;
 +      struct object_id actual_oid;
 +
 +      required_type = type_from_string(required_type_name);
 +      oidcpy(&actual_oid, oid);
 +      while (1) {
 +              int ref_length = -1;
 +              const char *ref_type = NULL;
 +
 +              buffer = read_object_file(&actual_oid, &type, &isize);
 +              if (!buffer)
 +                      return NULL;
 +              if (type == required_type) {
 +                      *size = isize;
 +                      if (actual_oid_return)
 +                              oidcpy(actual_oid_return, &actual_oid);
 +                      return buffer;
 +              }
 +              /* Handle references */
 +              else if (type == OBJ_COMMIT)
 +                      ref_type = "tree ";
 +              else if (type == OBJ_TAG)
 +                      ref_type = "object ";
 +              else {
 +                      free(buffer);
 +                      return NULL;
 +              }
 +              ref_length = strlen(ref_type);
 +
 +              if (ref_length + the_hash_algo->hexsz > isize ||
 +                  memcmp(buffer, ref_type, ref_length) ||
 +                  get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
 +                      free(buffer);
 +                      return NULL;
 +              }
 +              free(buffer);
 +              /* Now we have the ID of the referred-to object in
 +               * actual_oid.  Check again. */
 +      }
 +}
 +
 +static void write_object_file_prepare(const void *buf, unsigned long len,
 +                                    const char *type, struct object_id *oid,
 +                                    char *hdr, int *hdrlen)
 +{
 +      git_hash_ctx c;
 +
 +      /* Generate the header */
 +      *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
 +
 +      /* Sha1.. */
 +      the_hash_algo->init_fn(&c);
 +      the_hash_algo->update_fn(&c, hdr, *hdrlen);
 +      the_hash_algo->update_fn(&c, buf, len);
 +      the_hash_algo->final_fn(oid->hash, &c);
 +}
 +
 +/*
 + * Move the just written object into its final resting place.
 + */
 +int finalize_object_file(const char *tmpfile, const char *filename)
 +{
 +      int ret = 0;
 +
 +      if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
 +              goto try_rename;
 +      else if (link(tmpfile, filename))
 +              ret = errno;
 +
 +      /*
 +       * Coda hack - coda doesn't like cross-directory links,
 +       * so we fall back to a rename, which will mean that it
 +       * won't be able to check collisions, but that's not a
 +       * big deal.
 +       *
 +       * The same holds for FAT formatted media.
 +       *
 +       * When this succeeds, we just return.  We have nothing
 +       * left to unlink.
 +       */
 +      if (ret && ret != EEXIST) {
 +      try_rename:
 +              if (!rename(tmpfile, filename))
 +                      goto out;
 +              ret = errno;
 +      }
 +      unlink_or_warn(tmpfile);
 +      if (ret) {
 +              if (ret != EEXIST) {
 +                      return error_errno(_("unable to write sha1 filename %s"), filename);
 +              }
 +              /* FIXME!!! Collision check here ? */
 +      }
 +
 +out:
 +      if (adjust_shared_perm(filename))
 +              return error(_("unable to set permission to '%s'"), filename);
 +      return 0;
 +}
 +
 +static int write_buffer(int fd, const void *buf, size_t len)
 +{
 +      if (write_in_full(fd, buf, len) < 0)
 +              return error_errno(_("file write error"));
 +      return 0;
 +}
 +
 +int hash_object_file(const void *buf, unsigned long len, const char *type,
 +                   struct object_id *oid)
 +{
 +      char hdr[MAX_HEADER_LEN];
 +      int hdrlen = sizeof(hdr);
 +      write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
 +      return 0;
 +}
 +
 +/* Finalize a file on disk, and close it. */
 +static void close_sha1_file(int fd)
 +{
 +      if (fsync_object_files)
 +              fsync_or_die(fd, "sha1 file");
 +      if (close(fd) != 0)
 +              die_errno(_("error when closing sha1 file"));
 +}
 +
 +/* Size of directory component, including the ending '/' */
 +static inline int directory_size(const char *filename)
 +{
 +      const char *s = strrchr(filename, '/');
 +      if (!s)
 +              return 0;
 +      return s - filename + 1;
 +}
 +
 +/*
 + * This creates a temporary file in the same directory as the final
 + * 'filename'
 + *
 + * We want to avoid cross-directory filename renames, because those
 + * can have problems on various filesystems (FAT, NFS, Coda).
 + */
 +static int create_tmpfile(struct strbuf *tmp, const char *filename)
 +{
 +      int fd, dirlen = directory_size(filename);
 +
 +      strbuf_reset(tmp);
 +      strbuf_add(tmp, filename, dirlen);
 +      strbuf_addstr(tmp, "tmp_obj_XXXXXX");
 +      fd = git_mkstemp_mode(tmp->buf, 0444);
 +      if (fd < 0 && dirlen && errno == ENOENT) {
 +              /*
 +               * Make sure the directory exists; note that the contents
 +               * of the buffer are undefined after mkstemp returns an
 +               * error, so we have to rewrite the whole buffer from
 +               * scratch.
 +               */
 +              strbuf_reset(tmp);
 +              strbuf_add(tmp, filename, dirlen - 1);
 +              if (mkdir(tmp->buf, 0777) && errno != EEXIST)
 +                      return -1;
 +              if (adjust_shared_perm(tmp->buf))
 +                      return -1;
 +
 +              /* Try again */
 +              strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
 +              fd = git_mkstemp_mode(tmp->buf, 0444);
 +      }
 +      return fd;
 +}
 +
 +static int write_loose_object(const struct object_id *oid, char *hdr,
 +                            int hdrlen, const void *buf, unsigned long len,
 +                            time_t mtime)
 +{
 +      int fd, ret;
 +      unsigned char compressed[4096];
 +      git_zstream stream;
 +      git_hash_ctx c;
 +      struct object_id parano_oid;
 +      static struct strbuf tmp_file = STRBUF_INIT;
 +      static struct strbuf filename = STRBUF_INIT;
 +
 +      strbuf_reset(&filename);
 +      sha1_file_name(the_repository, &filename, oid->hash);
 +
 +      fd = create_tmpfile(&tmp_file, filename.buf);
 +      if (fd < 0) {
 +              if (errno == EACCES)
 +                      return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
 +              else
 +                      return error_errno(_("unable to create temporary file"));
 +      }
 +
 +      /* Set it up */
 +      git_deflate_init(&stream, zlib_compression_level);
 +      stream.next_out = compressed;
 +      stream.avail_out = sizeof(compressed);
 +      the_hash_algo->init_fn(&c);
 +
 +      /* First header.. */
 +      stream.next_in = (unsigned char *)hdr;
 +      stream.avail_in = hdrlen;
 +      while (git_deflate(&stream, 0) == Z_OK)
 +              ; /* nothing */
 +      the_hash_algo->update_fn(&c, hdr, hdrlen);
 +
 +      /* Then the data itself.. */
 +      stream.next_in = (void *)buf;
 +      stream.avail_in = len;
 +      do {
 +              unsigned char *in0 = stream.next_in;
 +              ret = git_deflate(&stream, Z_FINISH);
 +              the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
 +              if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
 +                      die(_("unable to write sha1 file"));
 +              stream.next_out = compressed;
 +              stream.avail_out = sizeof(compressed);
 +      } while (ret == Z_OK);
 +
 +      if (ret != Z_STREAM_END)
 +              die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
 +                  ret);
 +      ret = git_deflate_end_gently(&stream);
 +      if (ret != Z_OK)
 +              die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
 +                  ret);
 +      the_hash_algo->final_fn(parano_oid.hash, &c);
 +      if (oidcmp(oid, &parano_oid) != 0)
 +              die(_("confused by unstable object source data for %s"),
 +                  oid_to_hex(oid));
 +
 +      close_sha1_file(fd);
 +
 +      if (mtime) {
 +              struct utimbuf utb;
 +              utb.actime = mtime;
 +              utb.modtime = mtime;
 +              if (utime(tmp_file.buf, &utb) < 0)
 +                      warning_errno(_("failed utime() on %s"), tmp_file.buf);
 +      }
 +
 +      return finalize_object_file(tmp_file.buf, filename.buf);
 +}
 +
 +static int freshen_loose_object(const struct object_id *oid)
 +{
 +      return check_and_freshen(oid, 1);
 +}
 +
 +static int freshen_packed_object(const struct object_id *oid)
 +{
 +      struct pack_entry e;
 +      if (!find_pack_entry(the_repository, oid, &e))
 +              return 0;
 +      if (e.p->freshened)
 +              return 1;
 +      if (!freshen_file(e.p->pack_name))
 +              return 0;
 +      e.p->freshened = 1;
 +      return 1;
 +}
 +
 +int write_object_file(const void *buf, unsigned long len, const char *type,
 +                    struct object_id *oid)
 +{
 +      char hdr[MAX_HEADER_LEN];
 +      int hdrlen = sizeof(hdr);
 +
 +      /* Normally if we have it in the pack then we do not bother writing
 +       * it out into .git/objects/??/?{38} file.
 +       */
 +      write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
 +      if (freshen_packed_object(oid) || freshen_loose_object(oid))
 +              return 0;
 +      return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
 +}
 +
 +int hash_object_file_literally(const void *buf, unsigned long len,
 +                             const char *type, struct object_id *oid,
 +                             unsigned flags)
 +{
 +      char *header;
 +      int hdrlen, status = 0;
 +
 +      /* type string, SP, %lu of the length plus NUL must fit this */
 +      hdrlen = strlen(type) + MAX_HEADER_LEN;
 +      header = xmalloc(hdrlen);
 +      write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
 +
 +      if (!(flags & HASH_WRITE_OBJECT))
 +              goto cleanup;
 +      if (freshen_packed_object(oid) || freshen_loose_object(oid))
 +              goto cleanup;
 +      status = write_loose_object(oid, header, hdrlen, buf, len, 0);
 +
 +cleanup:
 +      free(header);
 +      return status;
 +}
 +
 +int force_object_loose(const struct object_id *oid, time_t mtime)
 +{
 +      void *buf;
 +      unsigned long len;
 +      enum object_type type;
 +      char hdr[MAX_HEADER_LEN];
 +      int hdrlen;
 +      int ret;
 +
 +      if (has_loose_object(oid))
 +              return 0;
 +      buf = read_object(oid->hash, &type, &len);
 +      if (!buf)
 +              return error(_("cannot read sha1_file for %s"), oid_to_hex(oid));
 +      hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
 +      ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
 +      free(buf);
 +
 +      return ret;
 +}
 +
 +int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
 +{
 +      struct object_id oid;
 +      if (!startup_info->have_repository)
 +              return 0;
 +      hashcpy(oid.hash, sha1);
 +      return oid_object_info_extended(the_repository, &oid, NULL,
 +                                      flags | OBJECT_INFO_SKIP_CACHED) >= 0;
 +}
 +
 +int has_object_file(const struct object_id *oid)
 +{
 +      return has_sha1_file(oid->hash);
 +}
 +
 +int has_object_file_with_flags(const struct object_id *oid, int flags)
 +{
 +      return has_sha1_file_with_flags(oid->hash, flags);
 +}
 +
 +static void check_tree(const void *buf, size_t size)
 +{
 +      struct tree_desc desc;
 +      struct name_entry entry;
 +
 +      init_tree_desc(&desc, buf, size);
 +      while (tree_entry(&desc, &entry))
 +              /* do nothing
 +               * tree_entry() will die() on malformed entries */
 +              ;
 +}
 +
 +static void check_commit(const void *buf, size_t size)
 +{
 +      struct commit c;
 +      memset(&c, 0, sizeof(c));
 +      if (parse_commit_buffer(the_repository, &c, buf, size, 0))
 +              die(_("corrupt commit"));
 +}
 +
 +static void check_tag(const void *buf, size_t size)
 +{
 +      struct tag t;
 +      memset(&t, 0, sizeof(t));
 +      if (parse_tag_buffer(the_repository, &t, buf, size))
 +              die(_("corrupt tag"));
 +}
 +
 +static int index_mem(struct object_id *oid, void *buf, size_t size,
 +                   enum object_type type,
 +                   const char *path, unsigned flags)
 +{
 +      int ret, re_allocated = 0;
 +      int write_object = flags & HASH_WRITE_OBJECT;
 +
 +      if (!type)
 +              type = OBJ_BLOB;
 +
 +      /*
 +       * Convert blobs to git internal format
 +       */
 +      if ((type == OBJ_BLOB) && path) {
 +              struct strbuf nbuf = STRBUF_INIT;
 +              if (convert_to_git(&the_index, path, buf, size, &nbuf,
 +                                 get_conv_flags(flags))) {
 +                      buf = strbuf_detach(&nbuf, &size);
 +                      re_allocated = 1;
 +              }
 +      }
 +      if (flags & HASH_FORMAT_CHECK) {
 +              if (type == OBJ_TREE)
 +                      check_tree(buf, size);
 +              if (type == OBJ_COMMIT)
 +                      check_commit(buf, size);
 +              if (type == OBJ_TAG)
 +                      check_tag(buf, size);
 +      }
 +
 +      if (write_object)
 +              ret = write_object_file(buf, size, type_name(type), oid);
 +      else
 +              ret = hash_object_file(buf, size, type_name(type), oid);
 +      if (re_allocated)
 +              free(buf);
 +      return ret;
 +}
 +
 +static int index_stream_convert_blob(struct object_id *oid, int fd,
 +                                   const char *path, unsigned flags)
 +{
 +      int ret;
 +      const int write_object = flags & HASH_WRITE_OBJECT;
 +      struct strbuf sbuf = STRBUF_INIT;
 +
 +      assert(path);
 +      assert(would_convert_to_git_filter_fd(&the_index, path));
 +
 +      convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
 +                               get_conv_flags(flags));
 +
 +      if (write_object)
 +              ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
 +                                      oid);
 +      else
 +              ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
 +                                     oid);
 +      strbuf_release(&sbuf);
 +      return ret;
 +}
 +
 +static int index_pipe(struct object_id *oid, int fd, enum object_type type,
 +                    const char *path, unsigned flags)
 +{
 +      struct strbuf sbuf = STRBUF_INIT;
 +      int ret;
 +
 +      if (strbuf_read(&sbuf, fd, 4096) >= 0)
 +              ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
 +      else
 +              ret = -1;
 +      strbuf_release(&sbuf);
 +      return ret;
 +}
 +
 +#define SMALL_FILE_SIZE (32*1024)
 +
 +static int index_core(struct object_id *oid, int fd, size_t size,
 +                    enum object_type type, const char *path,
 +                    unsigned flags)
 +{
 +      int ret;
 +
 +      if (!size) {
 +              ret = index_mem(oid, "", size, type, path, flags);
 +      } else if (size <= SMALL_FILE_SIZE) {
 +              char *buf = xmalloc(size);
 +              ssize_t read_result = read_in_full(fd, buf, size);
 +              if (read_result < 0)
 +                      ret = error_errno(_("read error while indexing %s"),
 +                                        path ? path : "<unknown>");
 +              else if (read_result != size)
 +                      ret = error(_("short read while indexing %s"),
 +                                  path ? path : "<unknown>");
 +              else
 +                      ret = index_mem(oid, buf, size, type, path, flags);
 +              free(buf);
 +      } else {
 +              void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
 +              ret = index_mem(oid, buf, size, type, path, flags);
 +              munmap(buf, size);
 +      }
 +      return ret;
 +}
 +
 +/*
 + * This creates one packfile per large blob unless bulk-checkin
 + * machinery is "plugged".
 + *
 + * This also bypasses the usual "convert-to-git" dance, and that is on
 + * purpose. We could write a streaming version of the converting
 + * functions and insert that before feeding the data to fast-import
 + * (or equivalent in-core API described above). However, that is
 + * somewhat complicated, as we do not know the size of the filter
 + * result, which we need to know beforehand when writing a git object.
 + * Since the primary motivation for trying to stream from the working
 + * tree file and to avoid mmaping it in core is to deal with large
 + * binary blobs, they generally do not want to get any conversion, and
 + * callers should avoid this code path when filters are requested.
 + */
 +static int index_stream(struct object_id *oid, int fd, size_t size,
 +                      enum object_type type, const char *path,
 +                      unsigned flags)
 +{
 +      return index_bulk_checkin(oid, fd, size, type, path, flags);
 +}
 +
 +int index_fd(struct object_id *oid, int fd, struct stat *st,
 +           enum object_type type, const char *path, unsigned flags)
 +{
 +      int ret;
 +
 +      /*
 +       * Call xsize_t() only when needed to avoid potentially unnecessary
 +       * die() for large files.
 +       */
 +      if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(&the_index, path))
 +              ret = index_stream_convert_blob(oid, fd, path, flags);
 +      else if (!S_ISREG(st->st_mode))
 +              ret = index_pipe(oid, fd, type, path, flags);
 +      else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
 +               (path && would_convert_to_git(&the_index, path)))
 +              ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
 +                               flags);
 +      else
 +              ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
 +                                 flags);
 +      close(fd);
 +      return ret;
 +}
 +
 +int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
 +{
 +      int fd;
 +      struct strbuf sb = STRBUF_INIT;
 +      int rc = 0;
 +
 +      switch (st->st_mode & S_IFMT) {
 +      case S_IFREG:
 +              fd = open(path, O_RDONLY);
 +              if (fd < 0)
 +                      return error_errno("open(\"%s\")", path);
 +              if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
 +                      return error(_("%s: failed to insert into database"),
 +                                   path);
 +              break;
 +      case S_IFLNK:
 +              if (strbuf_readlink(&sb, path, st->st_size))
 +                      return error_errno("readlink(\"%s\")", path);
 +              if (!(flags & HASH_WRITE_OBJECT))
 +                      hash_object_file(sb.buf, sb.len, blob_type, oid);
 +              else if (write_object_file(sb.buf, sb.len, blob_type, oid))
 +                      rc = error(_("%s: failed to insert into database"), path);
 +              strbuf_release(&sb);
 +              break;
 +      case S_IFDIR:
 +              return resolve_gitlink_ref(path, "HEAD", oid);
 +      default:
 +              return error(_("%s: unsupported file type"), path);
 +      }
 +      return rc;
 +}
 +
 +int read_pack_header(int fd, struct pack_header *header)
 +{
 +      if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
 +              /* "eof before pack header was fully read" */
 +              return PH_ERROR_EOF;
 +
 +      if (header->hdr_signature != htonl(PACK_SIGNATURE))
 +              /* "protocol error (pack signature mismatch detected)" */
 +              return PH_ERROR_PACK_SIGNATURE;
 +      if (!pack_version_ok(header->hdr_version))
 +              /* "protocol error (pack version unsupported)" */
 +              return PH_ERROR_PROTOCOL;
 +      return 0;
 +}
 +
 +void assert_oid_type(const struct object_id *oid, enum object_type expect)
 +{
 +      enum object_type type = oid_object_info(the_repository, oid, NULL);
 +      if (type < 0)
 +              die(_("%s is not a valid object"), oid_to_hex(oid));
 +      if (type != expect)
 +              die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
 +                  type_name(expect));
 +}
 +
 +int for_each_file_in_obj_subdir(unsigned int subdir_nr,
 +                              struct strbuf *path,
 +                              each_loose_object_fn obj_cb,
 +                              each_loose_cruft_fn cruft_cb,
 +                              each_loose_subdir_fn subdir_cb,
 +                              void *data)
 +{
 +      size_t origlen, baselen;
 +      DIR *dir;
 +      struct dirent *de;
 +      int r = 0;
 +      struct object_id oid;
 +
 +      if (subdir_nr > 0xff)
 +              BUG("invalid loose object subdirectory: %x", subdir_nr);
 +
 +      origlen = path->len;
 +      strbuf_complete(path, '/');
 +      strbuf_addf(path, "%02x", subdir_nr);
 +
 +      dir = opendir(path->buf);
 +      if (!dir) {
 +              if (errno != ENOENT)
 +                      r = error_errno(_("unable to open %s"), path->buf);
 +              strbuf_setlen(path, origlen);
 +              return r;
 +      }
 +
 +      oid.hash[0] = subdir_nr;
 +      strbuf_addch(path, '/');
 +      baselen = path->len;
 +
 +      while ((de = readdir(dir))) {
 +              size_t namelen;
 +              if (is_dot_or_dotdot(de->d_name))
 +                      continue;
 +
 +              namelen = strlen(de->d_name);
 +              strbuf_setlen(path, baselen);
 +              strbuf_add(path, de->d_name, namelen);
 +              if (namelen == the_hash_algo->hexsz - 2 &&
 +                  !hex_to_bytes(oid.hash + 1, de->d_name,
 +                                the_hash_algo->rawsz - 1)) {
 +                      if (obj_cb) {
 +                              r = obj_cb(&oid, path->buf, data);
 +                              if (r)
 +                                      break;
 +                      }
 +                      continue;
 +              }
 +
 +              if (cruft_cb) {
 +                      r = cruft_cb(de->d_name, path->buf, data);
 +                      if (r)
 +                              break;
 +              }
 +      }
 +      closedir(dir);
 +
 +      strbuf_setlen(path, baselen - 1);
 +      if (!r && subdir_cb)
 +              r = subdir_cb(subdir_nr, path->buf, data);
 +
 +      strbuf_setlen(path, origlen);
 +
 +      return r;
 +}
 +
 +int for_each_loose_file_in_objdir_buf(struct strbuf *path,
 +                          each_loose_object_fn obj_cb,
 +                          each_loose_cruft_fn cruft_cb,
 +                          each_loose_subdir_fn subdir_cb,
 +                          void *data)
 +{
 +      int r = 0;
 +      int i;
 +
 +      for (i = 0; i < 256; i++) {
 +              r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
 +                                              subdir_cb, data);
 +              if (r)
 +                      break;
 +      }
 +
 +      return r;
 +}
 +
 +int for_each_loose_file_in_objdir(const char *path,
 +                                each_loose_object_fn obj_cb,
 +                                each_loose_cruft_fn cruft_cb,
 +                                each_loose_subdir_fn subdir_cb,
 +                                void *data)
 +{
 +      struct strbuf buf = STRBUF_INIT;
 +      int r;
 +
 +      strbuf_addstr(&buf, path);
 +      r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
 +                                            subdir_cb, data);
 +      strbuf_release(&buf);
 +
 +      return r;
 +}
 +
 +struct loose_alt_odb_data {
 +      each_loose_object_fn *cb;
 +      void *data;
 +};
 +
 +static int loose_from_alt_odb(struct alternate_object_database *alt,
 +                            void *vdata)
 +{
 +      struct loose_alt_odb_data *data = vdata;
 +      struct strbuf buf = STRBUF_INIT;
 +      int r;
 +
 +      strbuf_addstr(&buf, alt->path);
 +      r = for_each_loose_file_in_objdir_buf(&buf,
 +                                            data->cb, NULL, NULL,
 +                                            data->data);
 +      strbuf_release(&buf);
 +      return r;
 +}
 +
 +int for_each_loose_object(each_loose_object_fn cb, void *data,
 +                        enum for_each_object_flags flags)
 +{
 +      struct loose_alt_odb_data alt;
 +      int r;
 +
 +      r = for_each_loose_file_in_objdir(get_object_directory(),
 +                                        cb, NULL, NULL, data);
 +      if (r)
 +              return r;
 +
 +      if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
 +              return 0;
 +
 +      alt.cb = cb;
 +      alt.data = data;
 +      return foreach_alt_odb(loose_from_alt_odb, &alt);
 +}
 +
 +static int check_stream_sha1(git_zstream *stream,
 +                           const char *hdr,
 +                           unsigned long size,
 +                           const char *path,
 +                           const unsigned char *expected_sha1)
 +{
 +      git_hash_ctx c;
 +      unsigned char real_sha1[GIT_MAX_RAWSZ];
 +      unsigned char buf[4096];
 +      unsigned long total_read;
 +      int status = Z_OK;
 +
 +      the_hash_algo->init_fn(&c);
 +      the_hash_algo->update_fn(&c, hdr, stream->total_out);
 +
 +      /*
 +       * We already read some bytes into hdr, but the ones up to the NUL
 +       * do not count against the object's content size.
 +       */
 +      total_read = stream->total_out - strlen(hdr) - 1;
 +
 +      /*
 +       * This size comparison must be "<=" to read the final zlib packets;
 +       * see the comment in unpack_sha1_rest for details.
 +       */
 +      while (total_read <= size &&
++             (status == Z_OK ||
++              (status == Z_BUF_ERROR && !stream->avail_out))) {
 +              stream->next_out = buf;
 +              stream->avail_out = sizeof(buf);
 +              if (size - total_read < stream->avail_out)
 +                      stream->avail_out = size - total_read;
 +              status = git_inflate(stream, Z_FINISH);
 +              the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
 +              total_read += stream->next_out - buf;
 +      }
 +      git_inflate_end(stream);
 +
 +      if (status != Z_STREAM_END) {
 +              error(_("corrupt loose object '%s'"), sha1_to_hex(expected_sha1));
 +              return -1;
 +      }
 +      if (stream->avail_in) {
 +              error(_("garbage at end of loose object '%s'"),
 +                    sha1_to_hex(expected_sha1));
 +              return -1;
 +      }
 +
 +      the_hash_algo->final_fn(real_sha1, &c);
 +      if (hashcmp(expected_sha1, real_sha1)) {
 +              error(_("sha1 mismatch for %s (expected %s)"), path,
 +                    sha1_to_hex(expected_sha1));
 +              return -1;
 +      }
 +
 +      return 0;
 +}
 +
 +int read_loose_object(const char *path,
 +                    const struct object_id *expected_oid,
 +                    enum object_type *type,
 +                    unsigned long *size,
 +                    void **contents)
 +{
 +      int ret = -1;
 +      void *map = NULL;
 +      unsigned long mapsize;
 +      git_zstream stream;
 +      char hdr[MAX_HEADER_LEN];
 +
 +      *contents = NULL;
 +
 +      map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
 +      if (!map) {
 +              error_errno(_("unable to mmap %s"), path);
 +              goto out;
 +      }
 +
 +      if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
 +              error(_("unable to unpack header of %s"), path);
 +              goto out;
 +      }
 +
 +      *type = parse_sha1_header(hdr, size);
 +      if (*type < 0) {
 +              error(_("unable to parse header of %s"), path);
 +              git_inflate_end(&stream);
 +              goto out;
 +      }
 +
 +      if (*type == OBJ_BLOB && *size > big_file_threshold) {
 +              if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
 +                      goto out;
 +      } else {
 +              *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
 +              if (!*contents) {
 +                      error(_("unable to unpack contents of %s"), path);
 +                      git_inflate_end(&stream);
 +                      goto out;
 +              }
 +              if (check_object_signature(expected_oid, *contents,
 +                                       *size, type_name(*type))) {
 +                      error(_("sha1 mismatch for %s (expected %s)"), path,
 +                            oid_to_hex(expected_oid));
 +                      free(*contents);
 +                      goto out;
 +              }
 +      }
 +
 +      ret = 0; /* everything checks out */
 +
 +out:
 +      if (map)
 +              munmap(map, mapsize);
 +      return ret;
 +}
diff --combined t/t1450-fsck.sh
index 0f2dd26f74b4b2c19474ad563e7c34701838cb46,1f245bbc5f6f1b825647b7a8b5198219339f2740..b5677d26a4268505139ec6cb7d65fe35db7debbe
@@@ -16,7 -16,8 +16,7 @@@ test_expect_success setup 
        git checkout HEAD^0 &&
        test_commit B fileB two &&
        git tag -d A B &&
 -      git reflog expire --expire=now --all &&
 -      >empty
 +      git reflog expire --expire=now --all
  '
  
  test_expect_success 'loose objects borrowed from alternate are not missing' '
                test_commit C fileC one &&
                git fsck --no-dangling >../actual 2>&1
        ) &&
 -      test_cmp empty actual
 +      test_must_be_empty actual
  '
  
  test_expect_success 'HEAD is part of refs, valid objects appear valid' '
        git fsck >actual 2>&1 &&
 -      test_cmp empty actual
 +      test_must_be_empty actual
  '
  
  # Corruption tests follow.  Make sure to remove all traces of the
@@@ -221,28 -222,6 +221,28 @@@ test_expect_success 'unparseable tree o
        test_i18ngrep ! "fatal: empty filename in tree entry" out
  '
  
 +hex2oct() {
 +      perl -ne 'printf "\\%03o", hex for /../g'
 +}
 +
 +test_expect_success 'tree entry with type mismatch' '
 +      test_when_finished "remove_object \$blob" &&
 +      test_when_finished "remove_object \$tree" &&
 +      test_when_finished "remove_object \$commit" &&
 +      test_when_finished "git update-ref -d refs/heads/type_mismatch" &&
 +      blob=$(echo blob | git hash-object -w --stdin) &&
 +      blob_bin=$(echo $blob | hex2oct) &&
 +      tree=$(
 +              printf "40000 dir\0${blob_bin}100644 file\0${blob_bin}" |
 +              git hash-object -t tree --stdin -w --literally
 +      ) &&
 +      commit=$(git commit-tree $tree) &&
 +      git update-ref refs/heads/type_mismatch $commit &&
 +      test_must_fail git fsck >out 2>&1 &&
 +      test_i18ngrep "is a blob, not a tree" out &&
 +      test_i18ngrep ! "dangling blob" out
 +'
 +
  test_expect_success 'tag pointing to nonexistent' '
        cat >invalid-tag <<-\EOF &&
        object ffffffffffffffffffffffffffffffffffffffff
@@@ -345,12 -324,12 +345,12 @@@ test_expect_success 'tag with NUL in he
  
  test_expect_success 'cleaned up' '
        git fsck >actual 2>&1 &&
 -      test_cmp empty actual
 +      test_must_be_empty actual
  '
  
  test_expect_success 'rev-list --verify-objects' '
        git rev-list --verify-objects --all >/dev/null 2>out &&
 -      test_cmp empty out
 +      test_must_be_empty out
  '
  
  test_expect_success 'rev-list --verify-objects with bad sha1' '
  
        test_might_fail git rev-list --verify-objects refs/heads/bogus >/dev/null 2>out &&
        cat out &&
 -      grep -q "error: sha1 mismatch 63ffffffffffffffffffffffffffffffffffffff" out
 +      test_i18ngrep -q "error: sha1 mismatch 63ffffffffffffffffffffffffffffffffffffff" out
  '
  
  test_expect_success 'force fsck to ignore double author' '
@@@ -594,7 -573,7 +594,7 @@@ test_expect_success 'fsck --name-object
                remove_object $(git rev-parse julius:caesar.t) &&
                test_must_fail git fsck --name-objects >out &&
                tree=$(git rev-parse --verify julius:) &&
 -              grep "$tree (\(refs/heads/master\|HEAD\)@{[0-9]*}:" out
 +              egrep "$tree \((refs/heads/master|HEAD)@\{[0-9]*\}:" out
        )
  '
  
@@@ -629,22 -608,6 +629,22 @@@ test_expect_success 'fsck errors in pac
        ! grep corrupt out
  '
  
 +test_expect_success 'fsck fails on corrupt packfile' '
 +      hsh=$(git commit-tree -m mycommit HEAD^{tree}) &&
 +      pack=$(echo $hsh | git pack-objects .git/objects/pack/pack) &&
 +
 +      # Corrupt the first byte of the first object. (It contains 3 type bits,
 +      # at least one of which is not zero, so setting the first byte to 0 is
 +      # sufficient.)
 +      chmod a+w .git/objects/pack/pack-$pack.pack &&
 +      printf '\0' | dd of=.git/objects/pack/pack-$pack.pack bs=1 conv=notrunc seek=12 &&
 +
 +      test_when_finished "rm -f .git/objects/pack/pack-$pack.*" &&
 +      remove_object $hsh &&
 +      test_must_fail git fsck 2>out &&
 +      test_i18ngrep "checksum mismatch" out
 +'
 +
  test_expect_success 'fsck finds problems in duplicate loose objects' '
        rm -rf broken-duplicate &&
        git init broken-duplicate &&
@@@ -673,16 -636,35 +673,35 @@@ test_expect_success 'fsck detects trail
        test_i18ngrep "garbage.*$commit" out
  '
  
- test_expect_success 'fsck detects trailing loose garbage (blob)' '
+ test_expect_success 'fsck detects trailing loose garbage (large blob)' '
        blob=$(echo trailing | git hash-object -w --stdin) &&
        file=$(sha1_file $blob) &&
        test_when_finished "remove_object $blob" &&
        chmod +w "$file" &&
        echo garbage >>"$file" &&
-       test_must_fail git fsck 2>out &&
+       test_must_fail git -c core.bigfilethreshold=5 fsck 2>out &&
        test_i18ngrep "garbage.*$blob" out
  '
  
 -      test-genrandom truncate 4096 >file &&
+ test_expect_success 'fsck detects truncated loose object' '
+       # make it big enough that we know we will truncate in the data
+       # portion, not the header
++      test-tool genrandom truncate 4096 >file &&
+       blob=$(git hash-object -w file) &&
+       file=$(sha1_file $blob) &&
+       test_when_finished "remove_object $blob" &&
+       test_copy_bytes 1024 <"$file" >tmp &&
+       rm "$file" &&
+       mv -f tmp "$file" &&
+       # check both regular and streaming code paths
+       test_must_fail git fsck 2>out &&
+       test_i18ngrep corrupt.*$blob out &&
+       test_must_fail git -c core.bigfilethreshold=128 fsck 2>out &&
+       test_i18ngrep corrupt.*$blob out
+ '
  # for each of type, we have one version which is referenced by another object
  # (and so while unreachable, not dangling), and another variant which really is
  # dangling.
@@@ -712,7 -694,7 +731,7 @@@ test_expect_success 'fsck notices dangl
  
  test_expect_success 'fsck $name notices bogus $name' '
        test_must_fail git fsck bogus &&
 -      test_must_fail git fsck $_z40
 +      test_must_fail git fsck $ZERO_OID
  '
  
  test_expect_success 'bogus head does not fallback to all heads' '
        blob=$(git rev-parse :foo) &&
        test_when_finished "git rm --cached foo" &&
        remove_object $blob &&
 -      test_must_fail git fsck $_z40 >out 2>&1 &&
 +      test_must_fail git fsck $ZERO_OID >out 2>&1 &&
        ! grep $blob out
  '
  
 +# Corrupt the checksum on the index.
 +# Add 1 to the last byte in the SHA.
 +corrupt_index_checksum () {
 +    perl -w -e '
 +      use Fcntl ":seek";
 +      open my $fh, "+<", ".git/index" or die "open: $!";
 +      binmode $fh;
 +      seek $fh, -1, SEEK_END or die "seek: $!";
 +      read $fh, my $in_byte, 1 or die "read: $!";
 +
 +      $in_value = unpack("C", $in_byte);
 +      $out_value = ($in_value + 1) & 255;
 +
 +      $out_byte = pack("C", $out_value);
 +
 +      seek $fh, -1, SEEK_END or die "seek: $!";
 +      print $fh $out_byte;
 +      close $fh or die "close: $!";
 +    '
 +}
 +
 +# Corrupt the checksum on the index and then
 +# verify that only fsck notices.
 +test_expect_success 'detect corrupt index file in fsck' '
 +      cp .git/index .git/index.backup &&
 +      test_when_finished "mv .git/index.backup .git/index" &&
 +      corrupt_index_checksum &&
 +      test_must_fail git fsck --cache 2>errors &&
 +      grep "bad index file" errors
 +'
 +
  test_done