]> git.ipfire.org Git - thirdparty/git.git/commitdiff
Merge branch 'jh/fsck-promisors'
authorJunio C Hamano <gitster@pobox.com>
Tue, 13 Feb 2018 21:39:03 +0000 (13:39 -0800)
committerJunio C Hamano <gitster@pobox.com>
Tue, 13 Feb 2018 21:39:03 +0000 (13:39 -0800)
In preparation for implementing narrow/partial clone, the machinery
for checking object connectivity used by gc and fsck has been
taught that a missing object is OK when it is referenced by a
packfile specially marked as coming from trusted repository that
promises to make them available on-demand and lazily.

* jh/fsck-promisors:
  gc: do not repack promisor packfiles
  rev-list: support termination at promisor objects
  sha1_file: support lazily fetching missing objects
  introduce fetch-object: fetch one promisor object
  index-pack: refactor writing of .keep files
  fsck: support promisor objects as CLI argument
  fsck: support referenced promisor objects
  fsck: support refs pointing to promisor objects
  fsck: introduce partialclone extension
  extension.partialclone: introduce partial clone extension

32 files changed:
Documentation/git-pack-objects.txt
Documentation/gitremote-helpers.txt
Documentation/rev-list-options.txt
Documentation/technical/repository-version.txt
Makefile
builtin/cat-file.c
builtin/fetch-pack.c
builtin/fsck.c
builtin/gc.c
builtin/index-pack.c
builtin/pack-objects.c
builtin/prune.c
builtin/repack.c
builtin/rev-list.c
cache.h
environment.c
fetch-object.c [new file with mode: 0644]
fetch-object.h [new file with mode: 0644]
fetch-pack.c
fetch-pack.h
list-objects.c
object.c
packfile.c
packfile.h
remote-curl.c
revision.c
revision.h
setup.c
sha1_file.c
t/t0410-partial-clone.sh [new file with mode: 0755]
transport.c
transport.h

index aa403d02f33699cde77c507c0402d9dfa0b7fb2f..81bc490ac52eb9414015979d8c244ce063c838b5 100644 (file)
@@ -255,6 +255,17 @@ a missing object is encountered.  This is the default action.
 The form '--missing=allow-any' will allow object traversal to continue
 if a missing object is encountered.  Missing objects will silently be
 omitted from the results.
++
+The form '--missing=allow-promisor' is like 'allow-any', but will only
+allow object traversal to continue for EXPECTED promisor missing objects.
+Unexpected missing object will raise an error.
+
+--exclude-promisor-objects::
+       Omit objects that are known to be in the promisor remote.  (This
+       option has the purpose of operating only on locally created objects,
+       so that when we repack, we still maintain a distinction between
+       locally created objects [without .promisor] and objects from the
+       promisor remote [with .promisor].)  This is used with partial clone.
 
 SEE ALSO
 --------
index 4a584f3c5d7e40fc5a8a1e5c68a63e8156bdcd64..4b8c93ec59de3db02b9914aed4955d486be5f875 100644 (file)
@@ -466,6 +466,13 @@ set by Git if the remote helper has the 'option' capability.
        Transmit <string> as a push option. As the push option
        must not contain LF or NUL characters, the string is not encoded.
 
+'option from-promisor' {'true'|'false'}::
+       Indicate that these objects are being fetched from a promisor.
+
+'option no-dependents' {'true'|'false'}::
+       Indicate that only the objects wanted need to be fetched, not
+       their dependents.
+
 SEE ALSO
 --------
 linkgit:git-remote[1]
index 22f5c9b43dd01c02710e44e36ee6a9dd2afc9b60..7b273635de2b5bf3e3ba6ade8bcca3068a216395 100644 (file)
@@ -750,10 +750,21 @@ The form '--missing=allow-any' will allow object traversal to continue
 if a missing object is encountered.  Missing objects will silently be
 omitted from the results.
 +
+The form '--missing=allow-promisor' is like 'allow-any', but will only
+allow object traversal to continue for EXPECTED promisor missing objects.
+Unexpected missing objects will raise an error.
++
 The form '--missing=print' is like 'allow-any', but will also print a
 list of the missing objects.  Object IDs are prefixed with a ``?'' character.
 endif::git-rev-list[]
 
+--exclude-promisor-objects::
+       (For internal use only.)  Prefilter object traversal at
+       promisor boundary.  This is used with partial clone.  This is
+       stronger than `--missing=allow-promisor` because it limits the
+       traversal, rather than just silencing errors about missing
+       objects.
+
 --no-walk[=(sorted|unsorted)]::
        Only show the given commits, but do not traverse their ancestors.
        This has no effect if a range is specified. If the argument
index 00ad37986efdcebb440c95629acd16213028c239..e03eaccebc9c28acf29408060b2f76d7161db0d6 100644 (file)
@@ -86,3 +86,15 @@ for testing format-1 compatibility.
 When the config key `extensions.preciousObjects` is set to `true`,
 objects in the repository MUST NOT be deleted (e.g., by `git-prune` or
 `git repack -d`).
+
+`partialclone`
+~~~~~~~~~~~~~~
+
+When the config key `extensions.partialclone` is set, it indicates
+that the repo was created with a partial clone (or later performed
+a partial fetch) and that the remote may have omitted sending
+certain unwanted objects.  Such a remote is called a "promisor remote"
+and it promises that all such omitted objects can be fetched from it
+in the future.
+
+The value of this key is the name of the promisor remote.
index 37e02cec1b614097b418e73f2d07bf7e830ed3c1..eb74e6cf3c4b25ee639c192cefcddf8f235ffc0d 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -802,6 +802,7 @@ LIB_OBJS += ewah/ewah_bitmap.o
 LIB_OBJS += ewah/ewah_io.o
 LIB_OBJS += ewah/ewah_rlw.o
 LIB_OBJS += exec_cmd.o
+LIB_OBJS += fetch-object.o
 LIB_OBJS += fetch-pack.o
 LIB_OBJS += fsck.o
 LIB_OBJS += fsmonitor.o
index f5fa4fd75af26a66cd1b2c0d493116704bc16dd5..cf9ea5c79682d2e345fd6f5db9c386e98300fccd 100644 (file)
@@ -475,6 +475,8 @@ static int batch_objects(struct batch_options *opt)
 
                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;
index 366b9d13f929b7d299a31d9c53fb643220fe64e3..15eeed7b177ace3baa21076eb0a2d493ef1caa76 100644 (file)
@@ -53,6 +53,8 @@ int cmd_fetch_pack(int argc, const char **argv, const char *prefix)
        struct oid_array shallow = OID_ARRAY_INIT;
        struct string_list deepen_not = STRING_LIST_INIT_DUP;
 
+       fetch_if_missing = 0;
+
        packet_trace_identity("fetch-pack");
 
        memset(&args, 0, sizeof(args));
@@ -143,6 +145,14 @@ int cmd_fetch_pack(int argc, const char **argv, const char *prefix)
                        args.update_shallow = 1;
                        continue;
                }
+               if (!strcmp("--from-promisor", arg)) {
+                       args.from_promisor = 1;
+                       continue;
+               }
+               if (!strcmp("--no-dependents", arg)) {
+                       args.no_dependents = 1;
+                       continue;
+               }
                usage(fetch_pack_usage);
        }
        if (deepen_not.nr)
index 04846d46f91d6cde14d379da8e0e3dd458c08dc5..7a8a679d4f3d0cf6d95991927947cbd00359868c 100644 (file)
@@ -149,6 +149,15 @@ static int mark_object(struct object *obj, int type, void *data, struct fsck_opt
        if (obj->flags & REACHABLE)
                return 0;
        obj->flags |= REACHABLE;
+
+       if (is_promisor_object(&obj->oid))
+               /*
+                * Further recursion does not need to be performed on this
+                * object since it is a promisor object (so it does not need to
+                * be added to "pending").
+                */
+               return 0;
+
        if (!(obj->flags & HAS_OBJ)) {
                if (parent && !has_object_file(&obj->oid)) {
                        printf("broken link from %7s %s\n",
@@ -208,6 +217,8 @@ static void check_reachable_object(struct object *obj)
         * do a full fsck
         */
        if (!(obj->flags & HAS_OBJ)) {
+               if (is_promisor_object(&obj->oid))
+                       return;
                if (has_sha1_pack(obj->oid.hash))
                        return; /* it is in pack - forget about it */
                printf("missing %s %s\n", printable_type(obj),
@@ -398,7 +409,7 @@ static void fsck_handle_reflog_oid(const char *refname, struct object_id *oid,
                                        xstrfmt("%s@{%"PRItime"}", refname, timestamp));
                        obj->flags |= USED;
                        mark_object_reachable(obj);
-               } else {
+               } else if (!is_promisor_object(oid)) {
                        error("%s: invalid reflog entry %s", refname, oid_to_hex(oid));
                        errors_found |= ERROR_REACHABLE;
                }
@@ -434,6 +445,14 @@ static int fsck_handle_ref(const char *refname, const struct object_id *oid,
 
        obj = parse_object(oid);
        if (!obj) {
+               if (is_promisor_object(oid)) {
+                       /*
+                        * Increment default_refs anyway, because this is a
+                        * valid ref.
+                        */
+                        default_refs++;
+                        return 0;
+               }
                error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
                errors_found |= ERROR_REACHABLE;
                /* We'll continue with the rest despite the error.. */
@@ -659,6 +678,9 @@ int cmd_fsck(int argc, const char **argv, const char *prefix)
        int i;
        struct alternate_object_database *alt;
 
+       /* fsck knows how to handle missing promisor objects */
+       fetch_if_missing = 0;
+
        errors_found = 0;
        check_replace_refs = 0;
 
@@ -731,6 +753,8 @@ int cmd_fsck(int argc, const char **argv, const char *prefix)
                        struct object *obj = lookup_object(oid.hash);
 
                        if (!obj || !(obj->flags & HAS_OBJ)) {
+                               if (is_promisor_object(&oid))
+                                       continue;
                                error("%s: object missing", oid_to_hex(&oid));
                                errors_found |= ERROR_OBJECT;
                                continue;
index 3c5eae0edf12e461e84c71c0e2988fcfa58d3d44..77fa720bd0bf374db8a24338bde8686ee0d60377 100644 (file)
@@ -458,6 +458,9 @@ int cmd_gc(int argc, const char **argv, const char *prefix)
                        argv_array_push(&prune, prune_expire);
                        if (quiet)
                                argv_array_push(&prune, "--no-progress");
+                       if (repository_format_partial_clone)
+                               argv_array_push(&prune,
+                                               "--exclude-promisor-objects");
                        if (run_command_v_opt(prune.argv, RUN_GIT_CMD))
                                return error(FAILED_RUN, prune.argv[0]);
                }
index 4c51aec81f374d52e7c6a1afa73ab9dea3cf40b8..5ebd370c56d2611494868ad1cf5957b5e9ea636e 100644 (file)
@@ -1389,15 +1389,60 @@ static void fix_unresolved_deltas(struct sha1file *f)
        free(sorted_by_pos);
 }
 
+static const char *derive_filename(const char *pack_name, const char *suffix,
+                                  struct strbuf *buf)
+{
+       size_t len;
+       if (!strip_suffix(pack_name, ".pack", &len))
+               die(_("packfile name '%s' does not end with '.pack'"),
+                   pack_name);
+       strbuf_add(buf, pack_name, len);
+       strbuf_addch(buf, '.');
+       strbuf_addstr(buf, suffix);
+       return buf->buf;
+}
+
+static void write_special_file(const char *suffix, const char *msg,
+                              const char *pack_name, const unsigned char *sha1,
+                              const char **report)
+{
+       struct strbuf name_buf = STRBUF_INIT;
+       const char *filename;
+       int fd;
+       int msg_len = strlen(msg);
+
+       if (pack_name)
+               filename = derive_filename(pack_name, suffix, &name_buf);
+       else
+               filename = odb_pack_name(&name_buf, sha1, suffix);
+
+       fd = odb_pack_keep(filename);
+       if (fd < 0) {
+               if (errno != EEXIST)
+                       die_errno(_("cannot write %s file '%s'"),
+                                 suffix, filename);
+       } else {
+               if (msg_len > 0) {
+                       write_or_die(fd, msg, msg_len);
+                       write_or_die(fd, "\n", 1);
+               }
+               if (close(fd) != 0)
+                       die_errno(_("cannot close written %s file '%s'"),
+                                 suffix, filename);
+               if (report)
+                       *report = suffix;
+       }
+       strbuf_release(&name_buf);
+}
+
 static void final(const char *final_pack_name, const char *curr_pack_name,
                  const char *final_index_name, const char *curr_index_name,
-                 const char *keep_name, const char *keep_msg,
+                 const char *keep_msg, const char *promisor_msg,
                  unsigned char *sha1)
 {
        const char *report = "pack";
        struct strbuf pack_name = STRBUF_INIT;
        struct strbuf index_name = STRBUF_INIT;
-       struct strbuf keep_name_buf = STRBUF_INIT;
        int err;
 
        if (!from_stdin) {
@@ -1409,28 +1454,12 @@ static void final(const char *final_pack_name, const char *curr_pack_name,
                        die_errno(_("error while closing pack file"));
        }
 
-       if (keep_msg) {
-               int keep_fd, keep_msg_len = strlen(keep_msg);
-
-               if (!keep_name)
-                       keep_name = odb_pack_name(&keep_name_buf, sha1, "keep");
-
-               keep_fd = odb_pack_keep(keep_name);
-               if (keep_fd < 0) {
-                       if (errno != EEXIST)
-                               die_errno(_("cannot write keep file '%s'"),
-                                         keep_name);
-               } else {
-                       if (keep_msg_len > 0) {
-                               write_or_die(keep_fd, keep_msg, keep_msg_len);
-                               write_or_die(keep_fd, "\n", 1);
-                       }
-                       if (close(keep_fd) != 0)
-                               die_errno(_("cannot close written keep file '%s'"),
-                                         keep_name);
-                       report = "keep";
-               }
-       }
+       if (keep_msg)
+               write_special_file("keep", keep_msg, final_pack_name, sha1,
+                                  &report);
+       if (promisor_msg)
+               write_special_file("promisor", promisor_msg, final_pack_name,
+                                  sha1, NULL);
 
        if (final_pack_name != curr_pack_name) {
                if (!final_pack_name)
@@ -1472,7 +1501,6 @@ static void final(const char *final_pack_name, const char *curr_pack_name,
 
        strbuf_release(&index_name);
        strbuf_release(&pack_name);
-       strbuf_release(&keep_name_buf);
 }
 
 static int git_index_pack_config(const char *k, const char *v, void *cb)
@@ -1615,32 +1643,26 @@ static void show_pack_info(int stat_only)
        }
 }
 
-static const char *derive_filename(const char *pack_name, const char *suffix,
-                                  struct strbuf *buf)
-{
-       size_t len;
-       if (!strip_suffix(pack_name, ".pack", &len))
-               die(_("packfile name '%s' does not end with '.pack'"),
-                   pack_name);
-       strbuf_add(buf, pack_name, len);
-       strbuf_addstr(buf, suffix);
-       return buf->buf;
-}
-
 int cmd_index_pack(int argc, const char **argv, const char *prefix)
 {
        int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
        const char *curr_index;
        const char *index_name = NULL, *pack_name = NULL;
-       const char *keep_name = NULL, *keep_msg = NULL;
-       struct strbuf index_name_buf = STRBUF_INIT,
-                     keep_name_buf = STRBUF_INIT;
+       const char *keep_msg = NULL;
+       const char *promisor_msg = NULL;
+       struct strbuf index_name_buf = STRBUF_INIT;
        struct pack_idx_entry **idx_objects;
        struct pack_idx_option opts;
        unsigned char pack_sha1[20];
        unsigned foreign_nr = 1;        /* zero is a "good" value, assume bad */
        int report_end_of_input = 0;
 
+       /*
+        * index-pack never needs to fetch missing objects, since it only
+        * accesses the repo to do hash collision checks
+        */
+       fetch_if_missing = 0;
+
        if (argc == 2 && !strcmp(argv[1], "-h"))
                usage(index_pack_usage);
 
@@ -1678,6 +1700,8 @@ int cmd_index_pack(int argc, const char **argv, const char *prefix)
                                stat_only = 1;
                        } else if (skip_to_optional_arg(arg, "--keep", &keep_msg)) {
                                ; /* nothing to do */
+                       } else if (skip_to_optional_arg(arg, "--promisor", &promisor_msg)) {
+                               ; /* already parsed */
                        } else if (starts_with(arg, "--threads=")) {
                                char *end;
                                nr_threads = strtoul(arg+10, &end, 0);
@@ -1740,9 +1764,7 @@ int cmd_index_pack(int argc, const char **argv, const char *prefix)
        if (from_stdin && !startup_info->have_repository)
                die(_("--stdin requires a git repository"));
        if (!index_name && pack_name)
-               index_name = derive_filename(pack_name, ".idx", &index_name_buf);
-       if (keep_msg && !keep_name && pack_name)
-               keep_name = derive_filename(pack_name, ".keep", &keep_name_buf);
+               index_name = derive_filename(pack_name, "idx", &index_name_buf);
 
        if (verify) {
                if (!index_name)
@@ -1790,13 +1812,12 @@ int cmd_index_pack(int argc, const char **argv, const char *prefix)
        if (!verify)
                final(pack_name, curr_pack,
                      index_name, curr_index,
-                     keep_name, keep_msg,
+                     keep_msg, promisor_msg,
                      pack_sha1);
        else
                close(input_fd);
        free(objects);
        strbuf_release(&index_name_buf);
-       strbuf_release(&keep_name_buf);
        if (pack_name == NULL)
                free((void *) curr_pack);
        if (index_name == NULL)
index 6b9cfc289d87b543b747c4024504703a2e0d6641..6c71552cdf994c4e65cb5495d133494371529f80 100644 (file)
@@ -75,6 +75,8 @@ static int use_bitmap_index = -1;
 static int write_bitmap_index;
 static uint16_t write_bitmap_options;
 
+static int exclude_promisor_objects;
+
 static unsigned long delta_cache_size = 0;
 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
 static unsigned long cache_max_small_delta_size = 1000;
@@ -84,8 +86,9 @@ static unsigned long window_memory_limit = 0;
 static struct list_objects_filter_options filter_options;
 
 enum missing_action {
-       MA_ERROR = 0,    /* fail if any missing objects are encountered */
-       MA_ALLOW_ANY,    /* silently allow ALL missing objects */
+       MA_ERROR = 0,      /* fail if any missing objects are encountered */
+       MA_ALLOW_ANY,      /* silently allow ALL missing objects */
+       MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
 };
 static enum missing_action arg_missing_action;
 static show_object_fn fn_show_object;
@@ -2578,6 +2581,20 @@ static void show_object__ma_allow_any(struct object *obj, const char *name, void
        show_object(obj, name, data);
 }
 
+static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
+{
+       assert(arg_missing_action == MA_ALLOW_PROMISOR);
+
+       /*
+        * Quietly ignore EXPECTED missing objects.  This avoids problems with
+        * staging them now and getting an odd error later.
+        */
+       if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
+               return;
+
+       show_object(obj, name, data);
+}
+
 static int option_parse_missing_action(const struct option *opt,
                                       const char *arg, int unset)
 {
@@ -2592,10 +2609,18 @@ static int option_parse_missing_action(const struct option *opt,
 
        if (!strcmp(arg, "allow-any")) {
                arg_missing_action = MA_ALLOW_ANY;
+               fetch_if_missing = 0;
                fn_show_object = show_object__ma_allow_any;
                return 0;
        }
 
+       if (!strcmp(arg, "allow-promisor")) {
+               arg_missing_action = MA_ALLOW_PROMISOR;
+               fetch_if_missing = 0;
+               fn_show_object = show_object__ma_allow_promisor;
+               return 0;
+       }
+
        die(_("invalid value for --missing"));
        return 0;
 }
@@ -3009,6 +3034,8 @@ int cmd_pack_objects(int argc, const char **argv, const char *prefix)
                { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
                  N_("handling for missing objects"), PARSE_OPT_NONEG,
                  option_parse_missing_action },
+               OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
+                        N_("do not pack objects in promisor packfiles")),
                OPT_END(),
        };
 
@@ -3054,6 +3081,12 @@ int cmd_pack_objects(int argc, const char **argv, const char *prefix)
                argv_array_push(&rp, "--unpacked");
        }
 
+       if (exclude_promisor_objects) {
+               use_internal_rev_list = 1;
+               fetch_if_missing = 0;
+               argv_array_push(&rp, "--exclude-promisor-objects");
+       }
+
        if (!reuse_object)
                reuse_delta = 0;
        if (pack_compression_level == -1)
index d2fdae680a1ebe75d55d4fb66a3bdd0b54829666..4cfec82f4076e357048bcb5816ceeb0f4b020192 100644 (file)
@@ -101,12 +101,15 @@ int cmd_prune(int argc, const char **argv, const char *prefix)
 {
        struct rev_info revs;
        struct progress *progress = NULL;
+       int exclude_promisor_objects = 0;
        const struct option options[] = {
                OPT__DRY_RUN(&show_only, N_("do not remove, show only")),
                OPT__VERBOSE(&verbose, N_("report pruned objects")),
                OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
                OPT_EXPIRY_DATE(0, "expire", &expire,
                                N_("expire objects older than <time>")),
+               OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
+                        N_("limit traversal to objects outside promisor packfiles")),
                OPT_END()
        };
        char *s;
@@ -139,6 +142,10 @@ int cmd_prune(int argc, const char **argv, const char *prefix)
                show_progress = isatty(2);
        if (show_progress)
                progress = start_delayed_progress(_("Checking connectivity"), 0);
+       if (exclude_promisor_objects) {
+               fetch_if_missing = 0;
+               revs.exclude_promisor_objects = 1;
+       }
 
        mark_reachable_objects(&revs, 1, expire, progress);
        stop_progress(&progress);
index f17a68a17da960813e6e925837638a5d9a26d5ee..7bdb40142f9261dac6514d98ee01dc44e595ec71 100644 (file)
@@ -83,7 +83,8 @@ static void remove_pack_on_signal(int signo)
 
 /*
  * Adds all packs hex strings to the fname list, which do not
- * have a corresponding .keep file.
+ * have a corresponding .keep or .promisor file. These packs are not to
+ * be kept if we are going to pack everything into one file.
  */
 static void get_non_kept_pack_filenames(struct string_list *fname_list)
 {
@@ -101,7 +102,8 @@ static void get_non_kept_pack_filenames(struct string_list *fname_list)
 
                fname = xmemdupz(e->d_name, len);
 
-               if (!file_exists(mkpath("%s/%s.keep", packdir, fname)))
+               if (!file_exists(mkpath("%s/%s.keep", packdir, fname)) &&
+                   !file_exists(mkpath("%s/%s.promisor", packdir, fname)))
                        string_list_append_nodup(fname_list, fname);
                else
                        free(fname);
@@ -232,6 +234,8 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        argv_array_push(&cmd.args, "--all");
        argv_array_push(&cmd.args, "--reflog");
        argv_array_push(&cmd.args, "--indexed-objects");
+       if (repository_format_partial_clone)
+               argv_array_push(&cmd.args, "--exclude-promisor-objects");
        if (window)
                argv_array_pushf(&cmd.args, "--window=%s", window);
        if (window_memory)
index d5345b6a2e237b550e5cee69e6d46ff8271ade22..e27aa1fc0745afc18e71809535d20481f8cb3a62 100644 (file)
@@ -15,6 +15,7 @@
 #include "progress.h"
 #include "reflog-walk.h"
 #include "oidset.h"
+#include "packfile.h"
 
 static const char rev_list_usage[] =
 "git rev-list [OPTION] <commit-id>... [ -- paths... ]\n"
@@ -67,6 +68,7 @@ enum missing_action {
        MA_ERROR = 0,    /* fail if any missing objects are encountered */
        MA_ALLOW_ANY,    /* silently allow ALL missing objects */
        MA_PRINT,        /* print ALL missing objects in special section */
+       MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
 };
 static enum missing_action arg_missing_action;
 
@@ -197,6 +199,12 @@ static void finish_commit(struct commit *commit, void *data)
 
 static inline void finish_object__ma(struct object *obj)
 {
+       /*
+        * Whether or not we try to dynamically fetch missing objects
+        * from the server, we currently DO NOT have the object.  We
+        * can either print, allow (ignore), or conditionally allow
+        * (ignore) them.
+        */
        switch (arg_missing_action) {
        case MA_ERROR:
                die("missing blob object '%s'", oid_to_hex(&obj->oid));
@@ -209,25 +217,36 @@ static inline void finish_object__ma(struct object *obj)
                oidset_insert(&missing_objects, &obj->oid);
                return;
 
+       case MA_ALLOW_PROMISOR:
+               if (is_promisor_object(&obj->oid))
+                       return;
+               die("unexpected missing blob object '%s'",
+                   oid_to_hex(&obj->oid));
+               return;
+
        default:
                BUG("unhandled missing_action");
                return;
        }
 }
 
-static void finish_object(struct object *obj, const char *name, void *cb_data)
+static int finish_object(struct object *obj, const char *name, void *cb_data)
 {
        struct rev_list_info *info = cb_data;
-       if (obj->type == OBJ_BLOB && !has_object_file(&obj->oid))
+       if (obj->type == OBJ_BLOB && !has_object_file(&obj->oid)) {
                finish_object__ma(obj);
+               return 1;
+       }
        if (info->revs->verify_objects && !obj->parsed && obj->type != OBJ_COMMIT)
                parse_object(&obj->oid);
+       return 0;
 }
 
 static void show_object(struct object *obj, const char *name, void *cb_data)
 {
        struct rev_list_info *info = cb_data;
-       finish_object(obj, name, cb_data);
+       if (finish_object(obj, name, cb_data))
+               return;
        display_progress(progress, ++progress_counter);
        if (info->flags & REV_LIST_QUIET)
                return;
@@ -315,11 +334,19 @@ static inline int parse_missing_action_value(const char *value)
 
        if (!strcmp(value, "allow-any")) {
                arg_missing_action = MA_ALLOW_ANY;
+               fetch_if_missing = 0;
                return 1;
        }
 
        if (!strcmp(value, "print")) {
                arg_missing_action = MA_PRINT;
+               fetch_if_missing = 0;
+               return 1;
+       }
+
+       if (!strcmp(value, "allow-promisor")) {
+               arg_missing_action = MA_ALLOW_PROMISOR;
+               fetch_if_missing = 0;
                return 1;
        }
 
@@ -344,6 +371,35 @@ int cmd_rev_list(int argc, const char **argv, const char *prefix)
        init_revisions(&revs, prefix);
        revs.abbrev = DEFAULT_ABBREV;
        revs.commit_format = CMIT_FMT_UNSPECIFIED;
+
+       /*
+        * Scan the argument list before invoking setup_revisions(), so that we
+        * know if fetch_if_missing needs to be set to 0.
+        *
+        * "--exclude-promisor-objects" acts as a pre-filter on missing objects
+        * by not crossing the boundary from realized objects to promisor
+        * objects.
+        *
+        * Let "--missing" to conditionally set fetch_if_missing.
+        */
+       for (i = 1; i < argc; i++) {
+               const char *arg = argv[i];
+               if (!strcmp(arg, "--exclude-promisor-objects")) {
+                       fetch_if_missing = 0;
+                       revs.exclude_promisor_objects = 1;
+                       break;
+               }
+       }
+       for (i = 1; i < argc; i++) {
+               const char *arg = argv[i];
+               if (skip_prefix(arg, "--missing=", &arg)) {
+                       if (revs.exclude_promisor_objects)
+                               die(_("cannot combine --exclude-promisor-objects and --missing"));
+                       if (parse_missing_action_value(arg))
+                               break;
+               }
+       }
+
        argc = setup_revisions(argc, argv, &revs, NULL);
 
        memset(&info, 0, sizeof(info));
@@ -412,9 +468,10 @@ int cmd_rev_list(int argc, const char **argv, const char *prefix)
                        continue;
                }
 
-               if (skip_prefix(arg, "--missing=", &arg) &&
-                   parse_missing_action_value(arg))
-                       continue;
+               if (!strcmp(arg, "--exclude-promisor-objects"))
+                       continue; /* already handled above */
+               if (skip_prefix(arg, "--missing=", &arg))
+                       continue; /* already handled above */
 
                usage(rev_list_usage);
 
diff --git a/cache.h b/cache.h
index d8b975a5715e8a5ec7c32e176958725cba61d67a..23413fdcb71f484b6dc59bec2c912d9dd495cf4f 100644 (file)
--- a/cache.h
+++ b/cache.h
@@ -914,10 +914,12 @@ extern int grafts_replace_parents;
 #define GIT_REPO_VERSION 0
 #define GIT_REPO_VERSION_READ 1
 extern int repository_format_precious_objects;
+extern char *repository_format_partial_clone;
 
 struct repository_format {
        int version;
        int precious_objects;
+       char *partial_clone; /* value of extensions.partialclone */
        int is_bare;
        int hash_algo;
        char *work_tree;
@@ -1648,7 +1650,8 @@ extern struct packed_git {
        unsigned pack_local:1,
                 pack_keep:1,
                 freshened:1,
-                do_not_close:1;
+                do_not_close:1,
+                pack_promisor:1;
        unsigned char sha1[20];
        struct revindex_entry *revindex;
        /* something like ".git/objects/pack/xxxxx.pack" */
@@ -1787,6 +1790,14 @@ struct object_info {
 #define OBJECT_INFO_QUICK 8
 extern int sha1_object_info_extended(const unsigned char *, struct object_info *, unsigned flags);
 
+/*
+ * Set this to 0 to prevent sha1_object_info_extended() from fetching missing
+ * blobs. This has a difference only if extensions.partialClone is set.
+ *
+ * Its default value is 1.
+ */
+extern int fetch_if_missing;
+
 /* Dumb servers support */
 extern int update_server_info(int);
 
index 63ac38a46f8f01ee3d09ed8e09bd93acba63b21a..835bb75c4ef2380d7fd1208f3ca042eee60a9ae3 100644 (file)
@@ -27,6 +27,7 @@ int warn_ambiguous_refs = 1;
 int warn_on_object_refname_ambiguity = 1;
 int ref_paranoia = -1;
 int repository_format_precious_objects;
+char *repository_format_partial_clone;
 const char *git_commit_encoding;
 const char *git_log_output_encoding;
 const char *apply_default_whitespace;
diff --git a/fetch-object.c b/fetch-object.c
new file mode 100644 (file)
index 0000000..258fcfa
--- /dev/null
@@ -0,0 +1,27 @@
+#include "cache.h"
+#include "packfile.h"
+#include "pkt-line.h"
+#include "strbuf.h"
+#include "transport.h"
+#include "fetch-object.h"
+
+void fetch_object(const char *remote_name, const unsigned char *sha1)
+{
+       struct remote *remote;
+       struct transport *transport;
+       struct ref *ref;
+       int original_fetch_if_missing = fetch_if_missing;
+
+       fetch_if_missing = 0;
+       remote = remote_get(remote_name);
+       if (!remote->url[0])
+               die(_("Remote with no URL"));
+       transport = transport_get(remote, remote->url[0]);
+
+       ref = alloc_ref(sha1_to_hex(sha1));
+       hashcpy(ref->old_oid.hash, sha1);
+       transport_set_option(transport, TRANS_OPT_FROM_PROMISOR, "1");
+       transport_set_option(transport, TRANS_OPT_NO_DEPENDENTS, "1");
+       transport_fetch_refs(transport, ref);
+       fetch_if_missing = original_fetch_if_missing;
+}
diff --git a/fetch-object.h b/fetch-object.h
new file mode 100644 (file)
index 0000000..f371300
--- /dev/null
@@ -0,0 +1,6 @@
+#ifndef FETCH_OBJECT_H
+#define FETCH_OBJECT_H
+
+extern void fetch_object(const char *remote_name, const unsigned char *sha1);
+
+#endif
index 9f6b07ad91f8c2c17e85004a0e71a8b69752d120..7aa1f58995a2d987db0fc69a35de4af783779185 100644 (file)
@@ -450,6 +450,8 @@ static int find_common(struct fetch_pack_args *args,
 
        flushes = 0;
        retval = -1;
+       if (args->no_dependents)
+               goto done;
        while ((oid = get_rev())) {
                packet_buf_write(&req_buf, "have %s\n", oid_to_hex(oid));
                print_verbose(args, "have %s", oid_to_hex(oid));
@@ -735,29 +737,31 @@ static int everything_local(struct fetch_pack_args *args,
                }
        }
 
-       if (!args->deepen) {
-               for_each_ref(mark_complete_oid, NULL);
-               for_each_cached_alternate(mark_alternate_complete);
-               commit_list_sort_by_date(&complete);
-               if (cutoff)
-                       mark_recent_complete_commits(args, cutoff);
-       }
+       if (!args->no_dependents) {
+               if (!args->deepen) {
+                       for_each_ref(mark_complete_oid, NULL);
+                       for_each_cached_alternate(mark_alternate_complete);
+                       commit_list_sort_by_date(&complete);
+                       if (cutoff)
+                               mark_recent_complete_commits(args, cutoff);
+               }
 
-       /*
-        * Mark all complete remote refs as common refs.
-        * Don't mark them common yet; the server has to be told so first.
-        */
-       for (ref = *refs; ref; ref = ref->next) {
-               struct object *o = deref_tag(lookup_object(ref->old_oid.hash),
-                                            NULL, 0);
+               /*
+                * Mark all complete remote refs as common refs.
+                * Don't mark them common yet; the server has to be told so first.
+                */
+               for (ref = *refs; ref; ref = ref->next) {
+                       struct object *o = deref_tag(lookup_object(ref->old_oid.hash),
+                                                    NULL, 0);
 
-               if (!o || o->type != OBJ_COMMIT || !(o->flags & COMPLETE))
-                       continue;
+                       if (!o || o->type != OBJ_COMMIT || !(o->flags & COMPLETE))
+                               continue;
 
-               if (!(o->flags & SEEN)) {
-                       rev_list_push((struct commit *)o, COMMON_REF | SEEN);
+                       if (!(o->flags & SEEN)) {
+                               rev_list_push((struct commit *)o, COMMON_REF | SEEN);
 
-                       mark_common((struct commit *)o, 1, 1);
+                               mark_common((struct commit *)o, 1, 1);
+                       }
                }
        }
 
@@ -833,7 +837,7 @@ static int get_pack(struct fetch_pack_args *args,
                argv_array_push(&cmd.args, alternate_shallow_file);
        }
 
-       if (do_keep) {
+       if (do_keep || args->from_promisor) {
                if (pack_lockfile)
                        cmd.out = -1;
                cmd_name = "index-pack";
@@ -843,7 +847,7 @@ static int get_pack(struct fetch_pack_args *args,
                        argv_array_push(&cmd.args, "-v");
                if (args->use_thin_pack)
                        argv_array_push(&cmd.args, "--fix-thin");
-               if (args->lock_pack || unpack_limit) {
+               if (do_keep && (args->lock_pack || unpack_limit)) {
                        char hostname[HOST_NAME_MAX + 1];
                        if (xgethostname(hostname, sizeof(hostname)))
                                xsnprintf(hostname, sizeof(hostname), "localhost");
@@ -853,6 +857,8 @@ static int get_pack(struct fetch_pack_args *args,
                }
                if (args->check_self_contained_and_connected)
                        argv_array_push(&cmd.args, "--check-self-contained-and-connected");
+               if (args->from_promisor)
+                       argv_array_push(&cmd.args, "--promisor");
        }
        else {
                cmd_name = "unpack-objects";
index b6aeb43a8e21437581e0e834e77754922f8c6ee9..aeac1526445efdbe0c639056e3a4175ab3a697e9 100644 (file)
@@ -29,6 +29,14 @@ struct fetch_pack_args {
        unsigned cloning:1;
        unsigned update_shallow:1;
        unsigned deepen:1;
+       unsigned from_promisor:1;
+
+       /*
+        * If 1, fetch_pack() will also not modify any object flags.
+        * This allows fetch_pack() to safely be called by any function,
+        * regardless of which object flags it uses (if any).
+        */
+       unsigned no_dependents:1;
 };
 
 /*
index 0966cdc9fa8dd448aa9955a2fb5324ae8ebbf97d..168bef688a89489a9d88d3e1f773483dbc1c8860 100644 (file)
@@ -9,6 +9,7 @@
 #include "list-objects.h"
 #include "list-objects-filter.h"
 #include "list-objects-filter-options.h"
+#include "packfile.h"
 
 static void process_blob(struct rev_info *revs,
                         struct blob *blob,
@@ -30,6 +31,20 @@ static void process_blob(struct rev_info *revs,
        if (obj->flags & (UNINTERESTING | SEEN))
                return;
 
+       /*
+        * Pre-filter known-missing objects when explicitly requested.
+        * Otherwise, a missing object error message may be reported
+        * later (depending on other filtering criteria).
+        *
+        * Note that this "--exclude-promisor-objects" pre-filtering
+        * may cause the actual filter to report an incomplete list
+        * of missing objects.
+        */
+       if (revs->exclude_promisor_objects &&
+           !has_object_file(&obj->oid) &&
+           is_promisor_object(&obj->oid))
+               return;
+
        pathlen = path->len;
        strbuf_addstr(path, name);
        if (filter_fn)
@@ -91,6 +106,8 @@ static void process_tree(struct rev_info *revs,
                all_entries_interesting: entry_not_interesting;
        int baselen = base->len;
        enum list_objects_filter_result r = LOFR_MARK_SEEN | LOFR_DO_SHOW;
+       int gently = revs->ignore_missing_links ||
+                    revs->exclude_promisor_objects;
 
        if (!revs->tree_objects)
                return;
@@ -98,9 +115,19 @@ static void process_tree(struct rev_info *revs,
                die("bad tree object");
        if (obj->flags & (UNINTERESTING | SEEN))
                return;
-       if (parse_tree_gently(tree, revs->ignore_missing_links) < 0) {
+       if (parse_tree_gently(tree, gently) < 0) {
                if (revs->ignore_missing_links)
                        return;
+
+               /*
+                * Pre-filter known-missing tree objects when explicitly
+                * requested.  This may cause the actual filter to report
+                * an incomplete list of missing objects.
+                */
+               if (revs->exclude_promisor_objects &&
+                   is_promisor_object(&obj->oid))
+                       return;
+
                die("bad tree object %s", oid_to_hex(&obj->oid));
        }
 
index 0afdfd19b784a541ad6d7237add2fbb1c9ec91ab..9e6f9ff20b03f800edae8ec7adc26e30313a5616 100644 (file)
--- a/object.c
+++ b/object.c
@@ -252,7 +252,7 @@ struct object *parse_object(const struct object_id *oid)
        if (obj && obj->parsed)
                return obj;
 
-       if ((obj && obj->type == OBJ_BLOB) ||
+       if ((obj && obj->type == OBJ_BLOB && has_object_file(oid)) ||
            (!obj && has_object_file(oid) &&
             sha1_object_info(oid->hash, NULL) == OBJ_BLOB)) {
                if (check_sha1_signature(repl, NULL, 0, NULL) < 0) {
index 4a5fe7ab1883843a389ce74bf1c7bd89890d8e51..234797cde78b0103f327000204c69a5fa99c0be3 100644 (file)
@@ -8,6 +8,11 @@
 #include "list.h"
 #include "streaming.h"
 #include "sha1-lookup.h"
+#include "commit.h"
+#include "object.h"
+#include "tag.h"
+#include "tree-walk.h"
+#include "tree.h"
 
 char *odb_pack_name(struct strbuf *buf,
                    const unsigned char *sha1,
@@ -643,10 +648,10 @@ struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
                return NULL;
 
        /*
-        * ".pack" is long enough to hold any suffix we're adding (and
+        * ".promisor" is long enough to hold any suffix we're adding (and
         * the use xsnprintf double-checks that)
         */
-       alloc = st_add3(path_len, strlen(".pack"), 1);
+       alloc = st_add3(path_len, strlen(".promisor"), 1);
        p = alloc_packed_git(alloc);
        memcpy(p->pack_name, path, path_len);
 
@@ -654,6 +659,10 @@ struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
        if (!access(p->pack_name, F_OK))
                p->pack_keep = 1;
 
+       xsnprintf(p->pack_name + path_len, alloc - path_len, ".promisor");
+       if (!access(p->pack_name, F_OK))
+               p->pack_promisor = 1;
+
        xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
        if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
                free(p);
@@ -781,7 +790,8 @@ static void prepare_packed_git_one(char *objdir, int local)
                if (ends_with(de->d_name, ".idx") ||
                    ends_with(de->d_name, ".pack") ||
                    ends_with(de->d_name, ".bitmap") ||
-                   ends_with(de->d_name, ".keep"))
+                   ends_with(de->d_name, ".keep") ||
+                   ends_with(de->d_name, ".promisor"))
                        string_list_append(&garbage, path.buf);
                else
                        report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
@@ -1889,6 +1899,9 @@ int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
        for (p = packed_git; p; p = p->next) {
                if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
                        continue;
+               if ((flags & FOR_EACH_OBJECT_PROMISOR_ONLY) &&
+                   !p->pack_promisor)
+                       continue;
                if (open_pack_index(p)) {
                        pack_errors = 1;
                        continue;
@@ -1899,3 +1912,61 @@ int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
        }
        return r ? r : pack_errors;
 }
+
+static int add_promisor_object(const struct object_id *oid,
+                              struct packed_git *pack,
+                              uint32_t pos,
+                              void *set_)
+{
+       struct oidset *set = set_;
+       struct object *obj = parse_object(oid);
+       if (!obj)
+               return 1;
+
+       oidset_insert(set, oid);
+
+       /*
+        * If this is a tree, commit, or tag, the objects it refers
+        * to are also promisor objects. (Blobs refer to no objects.)
+        */
+       if (obj->type == OBJ_TREE) {
+               struct tree *tree = (struct tree *)obj;
+               struct tree_desc desc;
+               struct name_entry entry;
+               if (init_tree_desc_gently(&desc, tree->buffer, tree->size))
+                       /*
+                        * Error messages are given when packs are
+                        * verified, so do not print any here.
+                        */
+                       return 0;
+               while (tree_entry_gently(&desc, &entry))
+                       oidset_insert(set, entry.oid);
+       } else if (obj->type == OBJ_COMMIT) {
+               struct commit *commit = (struct commit *) obj;
+               struct commit_list *parents = commit->parents;
+
+               oidset_insert(set, &commit->tree->object.oid);
+               for (; parents; parents = parents->next)
+                       oidset_insert(set, &parents->item->object.oid);
+       } else if (obj->type == OBJ_TAG) {
+               struct tag *tag = (struct tag *) obj;
+               oidset_insert(set, &tag->tagged->oid);
+       }
+       return 0;
+}
+
+int is_promisor_object(const struct object_id *oid)
+{
+       static struct oidset promisor_objects;
+       static int promisor_objects_prepared;
+
+       if (!promisor_objects_prepared) {
+               if (repository_format_partial_clone) {
+                       for_each_packed_object(add_promisor_object,
+                                              &promisor_objects,
+                                              FOR_EACH_OBJECT_PROMISOR_ONLY);
+               }
+               promisor_objects_prepared = 1;
+       }
+       return oidset_contains(&promisor_objects, oid);
+}
index 0cdeb54dcd97a67c38285e8f81412ec71273fd7f..a7fca598d672b73010a5fb99e4507da4634002ff 100644 (file)
@@ -1,6 +1,8 @@
 #ifndef PACKFILE_H
 #define PACKFILE_H
 
+#include "oidset.h"
+
 /*
  * Generate the filename to be used for a pack file with checksum "sha1" and
  * extension "ext". The result is written into the strbuf "buf", overwriting
@@ -124,6 +126,11 @@ extern int has_sha1_pack(const unsigned char *sha1);
 
 extern int has_pack_index(const unsigned char *sha1);
 
+/*
+ * Only iterate over packs obtained from the promisor remote.
+ */
+#define FOR_EACH_OBJECT_PROMISOR_ONLY 2
+
 /*
  * Iterate over packed objects in both the local
  * repository and any alternates repositories (unless the
@@ -135,4 +142,10 @@ typedef int each_packed_object_fn(const struct object_id *oid,
                                  void *data);
 extern int for_each_packed_object(each_packed_object_fn, void *, unsigned flags);
 
+/*
+ * Return 1 if an object in a promisor packfile is or refers to the given
+ * object, 0 otherwise.
+ */
+extern int is_promisor_object(const struct object_id *oid);
+
 #endif
index 0053b09549ab419ab8f2da2c519d689f5e0d83b3..431839111a9594174c6648d883f43eea9de89128 100644 (file)
@@ -33,7 +33,9 @@ struct options {
                thin : 1,
                /* One of the SEND_PACK_PUSH_CERT_* constants. */
                push_cert : 2,
-               deepen_relative : 1;
+               deepen_relative : 1,
+               from_promisor : 1,
+               no_dependents : 1;
 };
 static struct options options;
 static struct string_list cas_options = STRING_LIST_INIT_DUP;
@@ -157,6 +159,12 @@ static int set_option(const char *name, const char *value)
                        return -1;
                return 0;
 #endif /* LIBCURL_VERSION_NUM >= 0x070a08 */
+       } else if (!strcmp(name, "from-promisor")) {
+               options.from_promisor = 1;
+               return 0;
+       } else if (!strcmp(name, "no-dependents")) {
+               options.no_dependents = 1;
+               return 0;
        } else {
                return 1 /* unsupported */;
        }
@@ -822,6 +830,10 @@ static int fetch_git(struct discovery *heads,
                                 options.deepen_not.items[i].string);
        if (options.deepen_relative && options.depth)
                argv_array_push(&args, "--deepen-relative");
+       if (options.from_promisor)
+               argv_array_push(&args, "--from-promisor");
+       if (options.no_dependents)
+               argv_array_push(&args, "--no-dependents");
        argv_array_push(&args, url.buf);
 
        for (i = 0; i < nr_heads; i++) {
index a60628fbff96c413f9df5641d0bece45d89b721f..236d0d579f0b6920127e586fe09391d172445318 100644 (file)
@@ -198,6 +198,8 @@ static struct object *get_reference(struct rev_info *revs, const char *name,
        if (!object) {
                if (revs->ignore_missing)
                        return object;
+               if (revs->exclude_promisor_objects && is_promisor_object(oid))
+                       return NULL;
                die("bad object %s", name);
        }
        object->flags |= flags;
@@ -799,9 +801,17 @@ static int add_parents_to_list(struct rev_info *revs, struct commit *commit,
 
        for (parent = commit->parents; parent; parent = parent->next) {
                struct commit *p = parent->item;
-
-               if (parse_commit_gently(p, revs->ignore_missing_links) < 0)
+               int gently = revs->ignore_missing_links ||
+                            revs->exclude_promisor_objects;
+               if (parse_commit_gently(p, gently) < 0) {
+                       if (revs->exclude_promisor_objects &&
+                           is_promisor_object(&p->object.oid)) {
+                               if (revs->first_parent_only)
+                                       break;
+                               continue;
+                       }
                        return -1;
+               }
                if (revs->show_source && !p->util)
                        p->util = commit->util;
                p->object.flags |= left_flag;
@@ -2100,6 +2110,10 @@ static int handle_revision_opt(struct rev_info *revs, int argc, const char **arg
                revs->limited = 1;
        } else if (!strcmp(arg, "--ignore-missing")) {
                revs->ignore_missing = 1;
+       } else if (!strcmp(arg, "--exclude-promisor-objects")) {
+               if (fetch_if_missing)
+                       die("BUG: exclude_promisor_objects can only be used when fetch_if_missing is 0");
+               revs->exclude_promisor_objects = 1;
        } else {
                int opts = diff_opt_parse(&revs->diffopt, argv, argc, revs->prefix);
                if (!opts)
@@ -2845,6 +2859,16 @@ void reset_revision_walk(void)
        clear_object_flags(SEEN | ADDED | SHOWN);
 }
 
+static int mark_uninteresting(const struct object_id *oid,
+                             struct packed_git *pack,
+                             uint32_t pos,
+                             void *unused)
+{
+       struct object *o = parse_object(oid);
+       o->flags |= UNINTERESTING | SEEN;
+       return 0;
+}
+
 int prepare_revision_walk(struct rev_info *revs)
 {
        int i;
@@ -2872,6 +2896,11 @@ int prepare_revision_walk(struct rev_info *revs)
            (revs->limited && limiting_can_increase_treesame(revs)))
                revs->treesame.name = "treesame";
 
+       if (revs->exclude_promisor_objects) {
+               for_each_packed_object(mark_uninteresting, NULL,
+                                      FOR_EACH_OBJECT_PROMISOR_ONLY);
+       }
+
        if (revs->no_walk != REVISION_WALK_NO_WALK_UNSORTED)
                commit_list_sort_by_date(&revs->commits);
        if (revs->no_walk)
index d7a35c8c9e752cea0ea05ee7e13d696814470a83..3dee97bfb97a4abd636ad7055eb84cf5dea79558 100644 (file)
@@ -122,7 +122,10 @@ struct rev_info {
                        ancestry_path:1,
                        first_parent_only:1,
                        line_level_traverse:1,
-                       tree_blobs_in_commit_order:1;
+                       tree_blobs_in_commit_order:1,
+
+                       /* for internal use only */
+                       exclude_promisor_objects:1;
 
        /* Diff flags */
        unsigned int    diff:1,
diff --git a/setup.c b/setup.c
index 8cc34186ce1f918ce5a9c8fc22ea81b7c645ff17..c5d55dcee45ca811def540ca43c1e1eb57532a81 100644 (file)
--- a/setup.c
+++ b/setup.c
@@ -422,7 +422,11 @@ static int check_repo_format(const char *var, const char *value, void *vdata)
                        ;
                else if (!strcmp(ext, "preciousobjects"))
                        data->precious_objects = git_config_bool(var, value);
-               else
+               else if (!strcmp(ext, "partialclone")) {
+                       if (!value)
+                               return config_error_nonbool(var);
+                       data->partial_clone = xstrdup(value);
+               } else
                        string_list_append(&data->unknown_extensions, ext);
        } else if (strcmp(var, "core.bare") == 0) {
                data->is_bare = git_config_bool(var, value);
@@ -464,6 +468,7 @@ static int check_repository_format_gently(const char *gitdir, struct repository_
        }
 
        repository_format_precious_objects = candidate->precious_objects;
+       repository_format_partial_clone = candidate->partial_clone;
        string_list_clear(&candidate->unknown_extensions, 0);
        if (!has_common) {
                if (candidate->is_bare != -1) {
index 3da70ac650a8cdeca6ef8a6a424a7740d38267d5..2e58f5560a63426ab1b9848434ea0e1ac3ba1fba 100644 (file)
@@ -29,6 +29,7 @@
 #include "mergesort.h"
 #include "quote.h"
 #include "packfile.h"
+#include "fetch-object.h"
 
 const unsigned char null_sha1[GIT_MAX_RAWSZ];
 const struct object_id null_oid;
@@ -1213,6 +1214,8 @@ static int sha1_loose_object_info(const unsigned char *sha1,
        return (status < 0) ? status : 0;
 }
 
+int fetch_if_missing = 1;
+
 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
 {
        static struct object_info blank_oi = OBJECT_INFO_INIT;
@@ -1221,6 +1224,7 @@ int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi,
        const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
                                    lookup_replace_object(sha1) :
                                    sha1;
+       int already_retried = 0;
 
        if (is_null_sha1(real))
                return -1;
@@ -1248,19 +1252,32 @@ int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi,
                }
        }
 
-       if (!find_pack_entry(real, &e)) {
+       while (1) {
+               if (find_pack_entry(real, &e))
+                       break;
+
                /* Most likely it's a loose object. */
                if (!sha1_loose_object_info(real, oi, flags))
                        return 0;
 
                /* Not a loose object; someone else may have just packed it. */
-               if (flags & OBJECT_INFO_QUICK) {
-                       return -1;
-               } else {
-                       reprepare_packed_git();
-                       if (!find_pack_entry(real, &e))
-                               return -1;
+               reprepare_packed_git();
+               if (find_pack_entry(real, &e))
+                       break;
+
+               /* Check if it is a missing object */
+               if (fetch_if_missing && repository_format_partial_clone &&
+                   !already_retried) {
+                       /*
+                        * TODO Investigate haveing fetch_object() return
+                        * TODO error/success and stopping the music here.
+                        */
+                       fetch_object(repository_format_partial_clone, real);
+                       already_retried = 1;
+                       continue;
                }
+
+               return -1;
        }
 
        if (oi == &blank_oi)
@@ -1269,7 +1286,6 @@ int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi,
                 * information below, so return early.
                 */
                return 0;
-
        rtype = packed_object_info(e.p, e.offset, oi);
        if (rtype < 0) {
                mark_bad_packed_object(e.p, real);
diff --git a/t/t0410-partial-clone.sh b/t/t0410-partial-clone.sh
new file mode 100755 (executable)
index 0000000..cc18b75
--- /dev/null
@@ -0,0 +1,343 @@
+#!/bin/sh
+
+test_description='partial clone'
+
+. ./test-lib.sh
+
+delete_object () {
+       rm $1/.git/objects/$(echo $2 | sed -e 's|^..|&/|')
+}
+
+pack_as_from_promisor () {
+       HASH=$(git -C repo pack-objects .git/objects/pack/pack) &&
+       >repo/.git/objects/pack/pack-$HASH.promisor &&
+       echo $HASH
+}
+
+promise_and_delete () {
+       HASH=$(git -C repo rev-parse "$1") &&
+       git -C repo tag -a -m message my_annotated_tag "$HASH" &&
+       git -C repo rev-parse my_annotated_tag | pack_as_from_promisor &&
+       # tag -d prints a message to stdout, so redirect it
+       git -C repo tag -d my_annotated_tag >/dev/null &&
+       delete_object repo "$HASH"
+}
+
+test_expect_success 'missing reflog object, but promised by a commit, passes fsck' '
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       A=$(git -C repo commit-tree -m a HEAD^{tree}) &&
+       C=$(git -C repo commit-tree -m c -p $A HEAD^{tree}) &&
+
+       # Reference $A only from reflog, and delete it
+       git -C repo branch my_branch "$A" &&
+       git -C repo branch -f my_branch my_commit &&
+       delete_object repo "$A" &&
+
+       # State that we got $C, which refers to $A, from promisor
+       printf "$C\n" | pack_as_from_promisor &&
+
+       # Normally, it fails
+       test_must_fail git -C repo fsck &&
+
+       # But with the extension, it succeeds
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo fsck
+'
+
+test_expect_success 'missing reflog object, but promised by a tag, passes fsck' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       A=$(git -C repo commit-tree -m a HEAD^{tree}) &&
+       git -C repo tag -a -m d my_tag_name $A &&
+       T=$(git -C repo rev-parse my_tag_name) &&
+       git -C repo tag -d my_tag_name &&
+
+       # Reference $A only from reflog, and delete it
+       git -C repo branch my_branch "$A" &&
+       git -C repo branch -f my_branch my_commit &&
+       delete_object repo "$A" &&
+
+       # State that we got $T, which refers to $A, from promisor
+       printf "$T\n" | pack_as_from_promisor &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo fsck
+'
+
+test_expect_success 'missing reflog object alone fails fsck, even with extension set' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       A=$(git -C repo commit-tree -m a HEAD^{tree}) &&
+       B=$(git -C repo commit-tree -m b HEAD^{tree}) &&
+
+       # Reference $A only from reflog, and delete it
+       git -C repo branch my_branch "$A" &&
+       git -C repo branch -f my_branch my_commit &&
+       delete_object repo "$A" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       test_must_fail git -C repo fsck
+'
+
+test_expect_success 'missing ref object, but promised, passes fsck' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       A=$(git -C repo commit-tree -m a HEAD^{tree}) &&
+
+       # Reference $A only from ref
+       git -C repo branch my_branch "$A" &&
+       promise_and_delete "$A" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo fsck
+'
+
+test_expect_success 'missing object, but promised, passes fsck' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo 1 &&
+       test_commit -C repo 2 &&
+       test_commit -C repo 3 &&
+       git -C repo tag -a annotated_tag -m "annotated tag" &&
+
+       C=$(git -C repo rev-parse 1) &&
+       T=$(git -C repo rev-parse 2^{tree}) &&
+       B=$(git hash-object repo/3.t) &&
+       AT=$(git -C repo rev-parse annotated_tag) &&
+
+       promise_and_delete "$C" &&
+       promise_and_delete "$T" &&
+       promise_and_delete "$B" &&
+       promise_and_delete "$AT" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo fsck
+'
+
+test_expect_success 'missing CLI object, but promised, passes fsck' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       A=$(git -C repo commit-tree -m a HEAD^{tree}) &&
+       promise_and_delete "$A" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo fsck "$A"
+'
+
+test_expect_success 'fetching of missing objects' '
+       rm -rf repo &&
+       test_create_repo server &&
+       test_commit -C server foo &&
+       git -C server repack -a -d --write-bitmap-index &&
+
+       git clone "file://$(pwd)/server" repo &&
+       HASH=$(git -C repo rev-parse foo) &&
+       rm -rf repo/.git/objects/* &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "origin" &&
+       git -C repo cat-file -p "$HASH" &&
+
+       # Ensure that the .promisor file is written, and check that its
+       # associated packfile contains the object
+       ls repo/.git/objects/pack/pack-*.promisor >promisorlist &&
+       test_line_count = 1 promisorlist &&
+       IDX=$(cat promisorlist | sed "s/promisor$/idx/") &&
+       git verify-pack --verbose "$IDX" | grep "$HASH"
+'
+
+test_expect_success 'rev-list stops traversal at missing and promised commit' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo foo &&
+       test_commit -C repo bar &&
+
+       FOO=$(git -C repo rev-parse foo) &&
+       promise_and_delete "$FOO" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo rev-list --exclude-promisor-objects --objects bar >out &&
+       grep $(git -C repo rev-parse bar) out &&
+       ! grep $FOO out
+'
+
+test_expect_success 'rev-list stops traversal at missing and promised tree' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo foo &&
+       mkdir repo/a_dir &&
+       echo something >repo/a_dir/something &&
+       git -C repo add a_dir/something &&
+       git -C repo commit -m bar &&
+
+       # foo^{tree} (tree referenced from commit)
+       TREE=$(git -C repo rev-parse foo^{tree}) &&
+
+       # a tree referenced by HEAD^{tree} (tree referenced from tree)
+       TREE2=$(git -C repo ls-tree HEAD^{tree} | grep " tree " | head -1 | cut -b13-52) &&
+
+       promise_and_delete "$TREE" &&
+       promise_and_delete "$TREE2" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo rev-list --exclude-promisor-objects --objects HEAD >out &&
+       grep $(git -C repo rev-parse foo) out &&
+       ! grep $TREE out &&
+       grep $(git -C repo rev-parse HEAD) out &&
+       ! grep $TREE2 out
+'
+
+test_expect_success 'rev-list stops traversal at missing and promised blob' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       echo something >repo/something &&
+       git -C repo add something &&
+       git -C repo commit -m foo &&
+
+       BLOB=$(git -C repo hash-object -w something) &&
+       promise_and_delete "$BLOB" &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo rev-list --exclude-promisor-objects --objects HEAD >out &&
+       grep $(git -C repo rev-parse HEAD) out &&
+       ! grep $BLOB out
+'
+
+test_expect_success 'rev-list stops traversal at promisor commit, tree, and blob' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo foo &&
+       test_commit -C repo bar &&
+       test_commit -C repo baz &&
+
+       COMMIT=$(git -C repo rev-parse foo) &&
+       TREE=$(git -C repo rev-parse bar^{tree}) &&
+       BLOB=$(git hash-object repo/baz.t) &&
+       printf "%s\n%s\n%s\n" $COMMIT $TREE $BLOB | pack_as_from_promisor &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo rev-list --exclude-promisor-objects --objects HEAD >out &&
+       ! grep $COMMIT out &&
+       ! grep $TREE out &&
+       ! grep $BLOB out &&
+       grep $(git -C repo rev-parse bar) out  # sanity check that some walking was done
+'
+
+test_expect_success 'rev-list accepts missing and promised objects on command line' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo foo &&
+       test_commit -C repo bar &&
+       test_commit -C repo baz &&
+
+       COMMIT=$(git -C repo rev-parse foo) &&
+       TREE=$(git -C repo rev-parse bar^{tree}) &&
+       BLOB=$(git hash-object repo/baz.t) &&
+
+       promise_and_delete $COMMIT &&
+       promise_and_delete $TREE &&
+       promise_and_delete $BLOB &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo rev-list --exclude-promisor-objects --objects "$COMMIT" "$TREE" "$BLOB"
+'
+
+test_expect_success 'gc does not repack promisor objects' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       TREE_HASH=$(git -C repo rev-parse HEAD^{tree}) &&
+       HASH=$(printf "$TREE_HASH\n" | pack_as_from_promisor) &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo gc &&
+
+       # Ensure that the promisor packfile still exists, and remove it
+       test -e repo/.git/objects/pack/pack-$HASH.pack &&
+       rm repo/.git/objects/pack/pack-$HASH.* &&
+
+       # Ensure that the single other pack contains the commit, but not the tree
+       ls repo/.git/objects/pack/pack-*.pack >packlist &&
+       test_line_count = 1 packlist &&
+       git verify-pack repo/.git/objects/pack/pack-*.pack -v >out &&
+       grep "$(git -C repo rev-parse HEAD)" out &&
+       ! grep "$TREE_HASH" out
+'
+
+test_expect_success 'gc stops traversal when a missing but promised object is reached' '
+       rm -rf repo &&
+       test_create_repo repo &&
+       test_commit -C repo my_commit &&
+
+       TREE_HASH=$(git -C repo rev-parse HEAD^{tree}) &&
+       HASH=$(promise_and_delete $TREE_HASH) &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "arbitrary string" &&
+       git -C repo gc &&
+
+       # Ensure that the promisor packfile still exists, and remove it
+       test -e repo/.git/objects/pack/pack-$HASH.pack &&
+       rm repo/.git/objects/pack/pack-$HASH.* &&
+
+       # Ensure that the single other pack contains the commit, but not the tree
+       ls repo/.git/objects/pack/pack-*.pack >packlist &&
+       test_line_count = 1 packlist &&
+       git verify-pack repo/.git/objects/pack/pack-*.pack -v >out &&
+       grep "$(git -C repo rev-parse HEAD)" out &&
+       ! grep "$TREE_HASH" out
+'
+
+LIB_HTTPD_PORT=12345  # default port, 410, cannot be used as non-root
+. "$TEST_DIRECTORY"/lib-httpd.sh
+start_httpd
+
+test_expect_success 'fetching of missing objects from an HTTP server' '
+       rm -rf repo &&
+       SERVER="$HTTPD_DOCUMENT_ROOT_PATH/server" &&
+       test_create_repo "$SERVER" &&
+       test_commit -C "$SERVER" foo &&
+       git -C "$SERVER" repack -a -d --write-bitmap-index &&
+
+       git clone $HTTPD_URL/smart/server repo &&
+       HASH=$(git -C repo rev-parse foo) &&
+       rm -rf repo/.git/objects/* &&
+
+       git -C repo config core.repositoryformatversion 1 &&
+       git -C repo config extensions.partialclone "origin" &&
+       git -C repo cat-file -p "$HASH" &&
+
+       # Ensure that the .promisor file is written, and check that its
+       # associated packfile contains the object
+       ls repo/.git/objects/pack/pack-*.promisor >promisorlist &&
+       test_line_count = 1 promisorlist &&
+       IDX=$(cat promisorlist | sed "s/promisor$/idx/") &&
+       git verify-pack --verbose "$IDX" | grep "$HASH"
+'
+
+stop_httpd
+
+test_done
index fc802260f61c7496f4250b15ef6f5ab8dd316898..e82db773fd2383a998dff45b2b06eed668b5bebc 100644 (file)
@@ -161,6 +161,12 @@ static int set_git_option(struct git_transport_options *opts,
        } else if (!strcmp(name, TRANS_OPT_DEEPEN_RELATIVE)) {
                opts->deepen_relative = !!value;
                return 0;
+       } else if (!strcmp(name, TRANS_OPT_FROM_PROMISOR)) {
+               opts->from_promisor = !!value;
+               return 0;
+       } else if (!strcmp(name, TRANS_OPT_NO_DEPENDENTS)) {
+               opts->no_dependents = !!value;
+               return 0;
        }
        return 1;
 }
@@ -229,6 +235,8 @@ static int fetch_refs_via_pack(struct transport *transport,
                data->options.check_self_contained_and_connected;
        args.cloning = transport->cloning;
        args.update_shallow = data->options.update_shallow;
+       args.from_promisor = data->options.from_promisor;
+       args.no_dependents = data->options.no_dependents;
 
        if (!data->got_remote_heads) {
                connect_setup(transport, 0);
index 731c78b6795740f8e9bfb80741b0170671b37dd3..8c3430a5b98dae733892d6aeb718944cd5dea070 100644 (file)
@@ -15,6 +15,8 @@ struct git_transport_options {
        unsigned self_contained_and_connected : 1;
        unsigned update_shallow : 1;
        unsigned deepen_relative : 1;
+       unsigned from_promisor : 1;
+       unsigned no_dependents : 1;
        int depth;
        const char *deepen_since;
        const struct string_list *deepen_not;
@@ -159,6 +161,15 @@ void transport_check_allowed(const char *type);
 /* Send push certificates */
 #define TRANS_OPT_PUSH_CERT "pushcert"
 
+/* Indicate that these objects are being fetched by a promisor */
+#define TRANS_OPT_FROM_PROMISOR "from-promisor"
+
+/*
+ * Indicate that only the objects wanted need to be fetched, not their
+ * dependents
+ */
+#define TRANS_OPT_NO_DEPENDENTS "no-dependents"
+
 /**
  * Returns 0 if the option was used, non-zero otherwise. Prints a
  * message to stderr if the option is not used.