Merge branch 'cc/multi-promisor'
authorJunio C Hamano <gitster@pobox.com>
Wed, 18 Sep 2019 18:50:09 +0000 (11:50 -0700)
committerJunio C Hamano <gitster@pobox.com>
Wed, 18 Sep 2019 18:50:09 +0000 (11:50 -0700)
Teach the lazy clone machinery that there can be more than one
promisor remote and consult them in order when downloading missing
objects on demand.

* cc/multi-promisor:
Move core_partial_clone_filter_default to promisor-remote.c
Move repository_format_partial_clone to promisor-remote.c
Remove fetch-object.{c,h} in favor of promisor-remote.{c,h}
remote: add promisor and partial clone config to the doc
partial-clone: add multiple remotes in the doc
t0410: test fetching from many promisor remotes
builtin/fetch: remove unique promisor remote limitation
promisor-remote: parse remote.*.partialclonefilter
Use promisor_remote_get_direct() and has_promisor_remote()
promisor-remote: use repository_format_partial_clone
promisor-remote: add promisor_remote_reinit()
promisor-remote: implement promisor_remote_get_direct()
Add initial support for many promisor remotes
fetch-object: make functions return an error code
t0410: remove pipes after git commands

27 files changed:
Documentation/config/remote.txt
Documentation/technical/partial-clone.txt
Makefile
builtin/cat-file.c
builtin/fetch.c
builtin/gc.c
builtin/index-pack.c
builtin/repack.c
cache-tree.c
cache.h
config.c
connected.c
diff.c
environment.c
fetch-object.c [deleted file]
fetch-object.h [deleted file]
list-objects-filter-options.c
list-objects-filter-options.h
packfile.c
promisor-remote.c [new file with mode: 0644]
promisor-remote.h [new file with mode: 0644]
setup.c
sha1-file.c
t/t0410-partial-clone.sh
t/t5601-clone.sh
t/t5616-partial-clone.sh
unpack-trees.c
index 6c4cad83a2c9f41206e6c1032573d82fa7f3090c..a8e6437a903592934198fb212ede8241d6acebca 100644 (file)
@@ -76,3 +76,11 @@ remote.<name>.pruneTags::
 +
 See also `remote.<name>.prune` and the PRUNING section of
 linkgit:git-fetch[1].
+
+remote.<name>.promisor::
+       When set to true, this remote will be used to fetch promisor
+       objects.
+
+remote.<name>.partialclonefilter::
+       The filter that will be applied when fetching from this
+       promisor remote.
index 896c7b3878869de8d3564b42abcef9d4ea5dedec..210373e258890d7204c6dcf4a1befbd31e21139b 100644 (file)
@@ -30,12 +30,20 @@ advance* during clone and fetch operations and thereby reduce download
 times and disk usage.  Missing objects can later be "demand fetched"
 if/when needed.
 
+A remote that can later provide the missing objects is called a
+promisor remote, as it promises to send the objects when
+requested. Initialy Git supported only one promisor remote, the origin
+remote from which the user cloned and that was configured in the
+"extensions.partialClone" config option. Later support for more than
+one promisor remote has been implemented.
+
 Use of partial clone requires that the user be online and the origin
-remote be available for on-demand fetching of missing objects.  This may
-or may not be problematic for the user.  For example, if the user can
-stay within the pre-selected subset of the source tree, they may not
-encounter any missing objects.  Alternatively, the user could try to
-pre-fetch various objects if they know that they are going offline.
+remote or other promisor remotes be available for on-demand fetching
+of missing objects.  This may or may not be problematic for the user.
+For example, if the user can stay within the pre-selected subset of
+the source tree, they may not encounter any missing objects.
+Alternatively, the user could try to pre-fetch various objects if they
+know that they are going offline.
 
 
 Non-Goals
@@ -100,18 +108,18 @@ or commits that reference missing trees.
 Handling Missing Objects
 ------------------------
 
-- An object may be missing due to a partial clone or fetch, or missing due
-  to repository corruption.  To differentiate these cases, the local
-  repository specially indicates such filtered packfiles obtained from the
-  promisor remote as "promisor packfiles".
+- An object may be missing due to a partial clone or fetch, or missing
+  due to repository corruption.  To differentiate these cases, the
+  local repository specially indicates such filtered packfiles
+  obtained from promisor remotes as "promisor packfiles".
 +
 These promisor packfiles consist of a "<name>.promisor" file with
 arbitrary contents (like the "<name>.keep" files), in addition to
 their "<name>.pack" and "<name>.idx" files.
 
 - The local repository considers a "promisor object" to be an object that
-  it knows (to the best of its ability) that the promisor remote has promised
-  that it has, either because the local repository has that object in one of
+  it knows (to the best of its ability) that promisor remotes have promised
+  that they have, either because the local repository has that object in one of
   its promisor packfiles, or because another promisor object refers to it.
 +
 When Git encounters a missing object, Git can see if it is a promisor object
@@ -123,12 +131,12 @@ expensive-to-modify list of missing objects.[a]
 - Since almost all Git code currently expects any referenced object to be
   present locally and because we do not want to force every command to do
   a dry-run first, a fallback mechanism is added to allow Git to attempt
-  to dynamically fetch missing objects from the promisor remote.
+  to dynamically fetch missing objects from promisor remotes.
 +
 When the normal object lookup fails to find an object, Git invokes
-fetch-object to try to get the object from the server and then retry
-the object lookup.  This allows objects to be "faulted in" without
-complicated prediction algorithms.
+promisor_remote_get_direct() to try to get the object from a promisor
+remote and then retry the object lookup.  This allows objects to be
+"faulted in" without complicated prediction algorithms.
 +
 For efficiency reasons, no check as to whether the missing object is
 actually a promisor object is performed.
@@ -157,8 +165,7 @@ and prefetch those objects in bulk.
 +
 We are not happy with this global variable and would like to remove it,
 but that requires significant refactoring of the object code to pass an
-additional flag.  We hope that concurrent efforts to add an ODB API can
-encompass this.
+additional flag.
 
 
 Fetching Missing Objects
@@ -182,21 +189,63 @@ has been updated to not use any object flags when the corresponding argument
   though they are not necessary.
 
 
+Using many promisor remotes
+---------------------------
+
+Many promisor remotes can be configured and used.
+
+This allows for example a user to have multiple geographically-close
+cache servers for fetching missing blobs while continuing to do
+filtered `git-fetch` commands from the central server.
+
+When fetching objects, promisor remotes are tried one after the other
+until all the objects have been fetched.
+
+Remotes that are considered "promisor" remotes are those specified by
+the following configuration variables:
+
+- `extensions.partialClone = <name>`
+
+- `remote.<name>.promisor = true`
+
+- `remote.<name>.partialCloneFilter = ...`
+
+Only one promisor remote can be configured using the
+`extensions.partialClone` config variable. This promisor remote will
+be the last one tried when fetching objects.
+
+We decided to make it the last one we try, because it is likely that
+someone using many promisor remotes is doing so because the other
+promisor remotes are better for some reason (maybe they are closer or
+faster for some kind of objects) than the origin, and the origin is
+likely to be the remote specified by extensions.partialClone.
+
+This justification is not very strong, but one choice had to be made,
+and anyway the long term plan should be to make the order somehow
+fully configurable.
+
+For now though the other promisor remotes will be tried in the order
+they appear in the config file.
+
 Current Limitations
 -------------------
 
-- The remote used for a partial clone (or the first partial fetch
-  following a regular clone) is marked as the "promisor remote".
+- It is not possible to specify the order in which the promisor
+  remotes are tried in other ways than the order in which they appear
+  in the config file.
 +
-We are currently limited to a single promisor remote and only that
-remote may be used for subsequent partial fetches.
+It is also not possible to specify an order to be used when fetching
+from one remote and a different order when fetching from another
+remote.
+
+- It is not possible to push only specific objects to a promisor
+  remote.
 +
-We accept this limitation because we believe initial users of this
-feature will be using it on repositories with a strong single central
-server.
+It is not possible to push at the same time to multiple promisor
+remote in a specific order.
 
-- Dynamic object fetching will only ask the promisor remote for missing
-  objects.  We assume that the promisor remote has a complete view of the
+- Dynamic object fetching will only ask promisor remotes for missing
+  objects.  We assume that promisor remotes have a complete view of the
   repository and can satisfy all such requests.
 
 - Repack essentially treats promisor and non-promisor packfiles as 2
@@ -218,15 +267,17 @@ server.
 Future Work
 -----------
 
-- Allow more than one promisor remote and define a strategy for fetching
-  missing objects from specific promisor remotes or of iterating over the
-  set of promisor remotes until a missing object is found.
+- Improve the way to specify the order in which promisor remotes are
+  tried.
 +
-A user might want to have multiple geographically-close cache servers
-for fetching missing blobs while continuing to do filtered `git-fetch`
-commands from the central server, for example.
+For example this could allow to specify explicitly something like:
+"When fetching from this remote, I want to use these promisor remotes
+in this order, though, when pushing or fetching to that remote, I want
+to use those promisor remotes in that order."
+
+- Allow pushing to promisor remotes.
 +
-Or the user might want to work in a triangular work flow with multiple
+The user might want to work in a triangular work flow with multiple
 promisor remotes that each have an incomplete view of the repository.
 
 - Allow repack to work on promisor packfiles (while keeping them distinct
index ad71ae12194e1d27f2adcd4014d0048ee4a5ebcd..f879697ea3b23f6ad6308afb33ae0f08ec262591 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -884,7 +884,6 @@ LIB_OBJS += ewah/ewah_io.o
 LIB_OBJS += ewah/ewah_rlw.o
 LIB_OBJS += exec-cmd.o
 LIB_OBJS += fetch-negotiator.o
-LIB_OBJS += fetch-object.o
 LIB_OBJS += fetch-pack.o
 LIB_OBJS += fsck.o
 LIB_OBJS += fsmonitor.o
@@ -948,6 +947,7 @@ LIB_OBJS += preload-index.o
 LIB_OBJS += pretty.o
 LIB_OBJS += prio-queue.o
 LIB_OBJS += progress.o
+LIB_OBJS += promisor-remote.o
 LIB_OBJS += prompt.o
 LIB_OBJS += protocol.o
 LIB_OBJS += quote.o
index 995d47c85aad24a645786ed6480bd659c755997e..d6a1aa74cd41fe17fdb9d92f6b55fb7d67b31726 100644 (file)
@@ -15,6 +15,7 @@
 #include "sha1-array.h"
 #include "packfile.h"
 #include "object-store.h"
+#include "promisor-remote.h"
 
 struct batch_options {
        int enabled;
@@ -524,8 +525,8 @@ static int batch_objects(struct batch_options *opt)
        if (opt->all_objects) {
                struct object_cb_data cb;
 
-               if (repository_format_partial_clone)
-                       warning("This repository has extensions.partialClone set. Some objects may not be loaded.");
+               if (has_promisor_remote())
+                       warning("This repository uses promisor remotes. Some objects may not be loaded.");
 
                cb.opt = opt;
                cb.expand = &data;
index 54d6b018929159f7eb1649dbaf33eb1b7f251140..538f0e72073fdc92af1b0d79604856cf80a045f6 100644 (file)
@@ -24,6 +24,7 @@
 #include "list-objects-filter-options.h"
 #include "commit-reach.h"
 #include "branch.h"
+#include "promisor-remote.h"
 
 #define FORCED_UPDATES_DELAY_WARNING_IN_MS (10 * 1000)
 
@@ -1559,37 +1560,27 @@ static inline void fetch_one_setup_partial(struct remote *remote)
         * If no prior partial clone/fetch and the current fetch DID NOT
         * request a partial-fetch, do a normal fetch.
         */
-       if (!repository_format_partial_clone && !filter_options.choice)
+       if (!has_promisor_remote() && !filter_options.choice)
                return;
 
        /*
-        * If this is the FIRST partial-fetch request, we enable partial
-        * on this repo and remember the given filter-spec as the default
-        * for subsequent fetches to this remote.
+        * If this is a partial-fetch request, we enable partial on
+        * this repo if not already enabled and remember the given
+        * filter-spec as the default for subsequent fetches to this
+        * remote.
         */
-       if (!repository_format_partial_clone && filter_options.choice) {
+       if (filter_options.choice) {
                partial_clone_register(remote->name, &filter_options);
                return;
        }
 
-       /*
-        * We are currently limited to only ONE promisor remote and only
-        * allow partial-fetches from the promisor remote.
-        */
-       if (strcmp(remote->name, repository_format_partial_clone)) {
-               if (filter_options.choice)
-                       die(_("--filter can only be used with the remote "
-                             "configured in extensions.partialClone"));
-               return;
-       }
-
        /*
         * Do a partial-fetch from the promisor remote using either the
         * explicitly given filter-spec or inherit the filter-spec from
         * the config.
         */
        if (!filter_options.choice)
-               partial_clone_get_default_filter_spec(&filter_options);
+               partial_clone_get_default_filter_spec(&filter_options, remote->name);
        return;
 }
 
@@ -1710,7 +1701,7 @@ int cmd_fetch(int argc, const char **argv, const char *prefix)
        if (depth || deepen_since || deepen_not.nr)
                deepen = 1;
 
-       if (filter_options.choice && !repository_format_partial_clone)
+       if (filter_options.choice && !has_promisor_remote())
                die("--filter can only be used when extensions.partialClone is set");
 
        if (all) {
@@ -1744,7 +1735,7 @@ int cmd_fetch(int argc, const char **argv, const char *prefix)
        }
 
        if (remote) {
-               if (filter_options.choice || repository_format_partial_clone)
+               if (filter_options.choice || has_promisor_remote())
                        fetch_one_setup_partial(remote);
                result = fetch_one(remote, argc, argv, prune_tags_ok);
        } else {
index a22b6ff683465a3addeb059c89775e080197f155..fadb45489f34a760f4c8f6c96c3cc5a5c5c115bd 100644 (file)
@@ -27,6 +27,7 @@
 #include "pack-objects.h"
 #include "blob.h"
 #include "tree.h"
+#include "promisor-remote.h"
 
 #define FAILED_RUN "failed to run %s"
 
@@ -659,7 +660,7 @@ 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)
+                       if (has_promisor_remote())
                                argv_array_push(&prune,
                                                "--exclude-promisor-objects");
                        if (run_command_v_opt(prune.argv, RUN_GIT_CMD))
index 0d55f73b0b443b60dccc31096675d40691b3eb56..a23454da6ef9b3a219a48971ef9045772d328155 100644 (file)
@@ -14,7 +14,7 @@
 #include "thread-utils.h"
 #include "packfile.h"
 #include "object-store.h"
-#include "fetch-object.h"
+#include "promisor-remote.h"
 
 static const char index_pack_usage[] =
 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
@@ -1352,7 +1352,7 @@ static void fix_unresolved_deltas(struct hashfile *f)
                sorted_by_pos[i] = &ref_deltas[i];
        QSORT(sorted_by_pos, nr_ref_deltas, delta_pos_compare);
 
-       if (repository_format_partial_clone) {
+       if (has_promisor_remote()) {
                /*
                 * Prefetch the delta bases.
                 */
@@ -1366,8 +1366,8 @@ static void fix_unresolved_deltas(struct hashfile *f)
                        oid_array_append(&to_fetch, &d->oid);
                }
                if (to_fetch.nr)
-                       fetch_objects(repository_format_partial_clone,
-                                     to_fetch.oid, to_fetch.nr);
+                       promisor_remote_get_direct(the_repository,
+                                                  to_fetch.oid, to_fetch.nr);
                oid_array_clear(&to_fetch);
        }
 
index 632c0c0a79422a229d52c83665331501e8c54e29..3b3dd1437299a53a9031572820c12f3269922695 100644 (file)
@@ -11,6 +11,7 @@
 #include "midx.h"
 #include "packfile.h"
 #include "object-store.h"
+#include "promisor-remote.h"
 
 static int delta_base_offset = 1;
 static int pack_kept_objects = -1;
@@ -361,7 +362,7 @@ 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)
+       if (has_promisor_remote())
                argv_array_push(&cmd.args, "--exclude-promisor-objects");
        if (write_bitmaps > 0)
                argv_array_push(&cmd.args, "--write-bitmap-index");
index c22161f987152ea18715b00815e987847786c5ff..0e5724fad752e9157dbcf1caba0438b42386ee3d 100644 (file)
@@ -5,6 +5,7 @@
 #include "cache-tree.h"
 #include "object-store.h"
 #include "replace-object.h"
+#include "promisor-remote.h"
 
 #ifndef DEBUG_CACHE_TREE
 #define DEBUG_CACHE_TREE 0
@@ -357,7 +358,7 @@ static int update_one(struct cache_tree *it,
                }
 
                ce_missing_ok = mode == S_IFGITLINK || missing_ok ||
-                       (repository_format_partial_clone &&
+                       (has_promisor_remote() &&
                         ce_skip_worktree(ce));
                if (is_null_oid(oid) ||
                    (!ce_missing_ok && !has_object_file(oid))) {
diff --git a/cache.h b/cache.h
index b1da1ab08faad3da19657a9a5dcf5f2592c2127c..3cbad5b603ec03ce9e4ec26be84de653e049c980 100644 (file)
--- a/cache.h
+++ b/cache.h
@@ -937,8 +937,6 @@ 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;
-extern const char *core_partial_clone_filter_default;
 extern int repository_format_worktree_config;
 
 /*
index b61c258d6da5266a66caa55ef821de0f8e0d75aa..743e4570ee38cb6fffe7be7ed06ecdb590481016 100644 (file)
--- a/config.c
+++ b/config.c
@@ -1379,11 +1379,6 @@ static int git_default_core_config(const char *var, const char *value, void *cb)
                return 0;
        }
 
-       if (!strcmp(var, "core.partialclonefilter")) {
-               return git_config_string(&core_partial_clone_filter_default,
-                                        var, value);
-       }
-
        if (!strcmp(var, "core.usereplacerefs")) {
                read_replace_refs = git_config_bool(var, value);
                return 0;
index cd9b324afa5a33be7eced6a420061905d52c211f..971db009b327e63cf71c82b496d70604cecf6415 100644 (file)
@@ -5,6 +5,7 @@
 #include "connected.h"
 #include "transport.h"
 #include "packfile.h"
+#include "promisor-remote.h"
 
 /*
  * If we feed all the commits we want to verify to this command
@@ -73,7 +74,7 @@ int check_connected(oid_iterate_fn fn, void *cb_data,
        argv_array_push(&rev_list.args,"rev-list");
        argv_array_push(&rev_list.args, "--objects");
        argv_array_push(&rev_list.args, "--stdin");
-       if (repository_format_partial_clone)
+       if (has_promisor_remote())
                argv_array_push(&rev_list.args, "--exclude-promisor-objects");
        if (!opt->is_deepening_fetch) {
                argv_array_push(&rev_list.args, "--not");
diff --git a/diff.c b/diff.c
index e28b463f5757e7dd56752702db13d8195a3d39bc..6db6927369e5c18ae3cb7ea27a0ebce37fd2b357 100644 (file)
--- a/diff.c
+++ b/diff.c
@@ -25,7 +25,7 @@
 #include "packfile.h"
 #include "parse-options.h"
 #include "help.h"
-#include "fetch-object.h"
+#include "promisor-remote.h"
 
 #ifdef NO_FAST_WORKING_DIRECTORY
 #define FAST_WORKING_DIRECTORY 0
@@ -6520,8 +6520,7 @@ static void add_if_missing(struct repository *r,
 
 void diffcore_std(struct diff_options *options)
 {
-       if (options->repo == the_repository &&
-           repository_format_partial_clone) {
+       if (options->repo == the_repository && has_promisor_remote()) {
                /*
                 * Prefetch the diff pairs that are about to be flushed.
                 */
@@ -6538,8 +6537,8 @@ void diffcore_std(struct diff_options *options)
                        /*
                         * NEEDSWORK: Consider deduplicating the OIDs sent.
                         */
-                       fetch_objects(repository_format_partial_clone,
-                                     to_fetch.oid, to_fetch.nr);
+                       promisor_remote_get_direct(options->repo,
+                                                  to_fetch.oid, to_fetch.nr);
                oid_array_clear(&to_fetch);
        }
 
index 89af47cb8504903b90460d5faa91be9f61bd2fc3..efa072680a2bca0d317b66554a325b73c227c866 100644 (file)
@@ -31,8 +31,6 @@ 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 *core_partial_clone_filter_default;
 int repository_format_worktree_config;
 const char *git_commit_encoding;
 const char *git_log_output_encoding;
diff --git a/fetch-object.c b/fetch-object.c
deleted file mode 100644 (file)
index 4266548..0000000
+++ /dev/null
@@ -1,40 +0,0 @@
-#include "cache.h"
-#include "packfile.h"
-#include "pkt-line.h"
-#include "strbuf.h"
-#include "transport.h"
-#include "fetch-object.h"
-
-static void fetch_refs(const char *remote_name, struct ref *ref)
-{
-       struct remote *remote;
-       struct transport *transport;
-       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]);
-
-       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;
-}
-
-void fetch_objects(const char *remote_name, const struct object_id *oids,
-                  int oid_nr)
-{
-       struct ref *ref = NULL;
-       int i;
-
-       for (i = 0; i < oid_nr; i++) {
-               struct ref *new_ref = alloc_ref(oid_to_hex(&oids[i]));
-               oidcpy(&new_ref->old_oid, &oids[i]);
-               new_ref->exact_oid = 1;
-               new_ref->next = ref;
-               ref = new_ref;
-       }
-       fetch_refs(remote_name, ref);
-}
diff --git a/fetch-object.h b/fetch-object.h
deleted file mode 100644 (file)
index d6444ca..0000000
+++ /dev/null
@@ -1,9 +0,0 @@
-#ifndef FETCH_OBJECT_H
-#define FETCH_OBJECT_H
-
-struct object_id;
-
-void fetch_objects(const char *remote_name, const struct object_id *oids,
-                  int oid_nr);
-
-#endif
index 1cb20c659c82b151a652da0528d0673ac629cc6c..28c571f922e46cd6ff7aff659677d9003cf3c948 100644 (file)
@@ -6,6 +6,7 @@
 #include "list-objects.h"
 #include "list-objects-filter.h"
 #include "list-objects-filter-options.h"
+#include "promisor-remote.h"
 
 /*
  * Parse value of the argument to the "filter" keyword.
@@ -29,6 +30,9 @@ static int gently_parse_list_objects_filter(
 {
        const char *v0;
 
+       if (!arg)
+               return 0;
+
        if (filter_options->choice) {
                if (errbuf) {
                        strbuf_addstr(
@@ -146,41 +150,42 @@ void partial_clone_register(
        const char *remote,
        const struct list_objects_filter_options *filter_options)
 {
-       /*
-        * Record the name of the partial clone remote in the
-        * config and in the global variable -- the latter is
-        * used throughout to indicate that partial clone is
-        * enabled and to expect missing objects.
-        */
-       if (repository_format_partial_clone &&
-           *repository_format_partial_clone &&
-           strcmp(remote, repository_format_partial_clone))
-               die(_("cannot change partial clone promisor remote"));
+       char *cfg_name;
+       char *filter_name;
 
-       git_config_set("core.repositoryformatversion", "1");
-       git_config_set("extensions.partialclone", remote);
+       /* Check if it is already registered */
+       if (!promisor_remote_find(remote)) {
+               git_config_set("core.repositoryformatversion", "1");
 
-       repository_format_partial_clone = xstrdup(remote);
+               /* Add promisor config for the remote */
+               cfg_name = xstrfmt("remote.%s.promisor", remote);
+               git_config_set(cfg_name, "true");
+               free(cfg_name);
+       }
 
        /*
         * Record the initial filter-spec in the config as
         * the default for subsequent fetches from this remote.
         */
-       core_partial_clone_filter_default =
-               xstrdup(filter_options->filter_spec);
-       git_config_set("core.partialclonefilter",
-                      core_partial_clone_filter_default);
+       filter_name = xstrfmt("remote.%s.partialclonefilter", remote);
+       git_config_set(filter_name, filter_options->filter_spec);
+       free(filter_name);
+
+       /* Make sure the config info are reset */
+       promisor_remote_reinit();
 }
 
 void partial_clone_get_default_filter_spec(
-       struct list_objects_filter_options *filter_options)
+       struct list_objects_filter_options *filter_options,
+       const char *remote)
 {
+       struct promisor_remote *promisor = promisor_remote_find(remote);
+
        /*
         * Parse default value, but silently ignore it if it is invalid.
         */
-       if (!core_partial_clone_filter_default)
-               return;
-       gently_parse_list_objects_filter(filter_options,
-                                        core_partial_clone_filter_default,
-                                        NULL);
+       if (promisor)
+               gently_parse_list_objects_filter(filter_options,
+                                                promisor->partial_clone_filter,
+                                                NULL);
 }
index c54f0000fbade5608e81345a9c5dbd54137d352d..8deaa287b57cbbfd290ce4aa040883b8a5ebcd42 100644 (file)
@@ -87,6 +87,7 @@ void partial_clone_register(
        const char *remote,
        const struct list_objects_filter_options *filter_options);
 void partial_clone_get_default_filter_spec(
-       struct list_objects_filter_options *filter_options);
+       struct list_objects_filter_options *filter_options,
+       const char *remote);
 
 #endif /* LIST_OBJECTS_FILTER_OPTIONS_H */
index d98ac2287683f971a448b931262ef3d459e145d4..1a7d69fe32a8808dd34063c48daf61e32886b47a 100644 (file)
@@ -17,6 +17,7 @@
 #include "object-store.h"
 #include "midx.h"
 #include "commit-graph.h"
+#include "promisor-remote.h"
 
 char *odb_pack_name(struct strbuf *buf,
                    const unsigned char *sha1,
@@ -2132,7 +2133,7 @@ int is_promisor_object(const struct object_id *oid)
        static int promisor_objects_prepared;
 
        if (!promisor_objects_prepared) {
-               if (repository_format_partial_clone) {
+               if (has_promisor_remote()) {
                        for_each_packed_object(add_promisor_object,
                                               &promisor_objects,
                                               FOR_EACH_OBJECT_PROMISOR_ONLY);
diff --git a/promisor-remote.c b/promisor-remote.c
new file mode 100644 (file)
index 0000000..9bc296c
--- /dev/null
@@ -0,0 +1,265 @@
+#include "cache.h"
+#include "object-store.h"
+#include "promisor-remote.h"
+#include "config.h"
+#include "transport.h"
+
+static char *repository_format_partial_clone;
+static const char *core_partial_clone_filter_default;
+
+void set_repository_format_partial_clone(char *partial_clone)
+{
+       repository_format_partial_clone = xstrdup_or_null(partial_clone);
+}
+
+static int fetch_refs(const char *remote_name, struct ref *ref)
+{
+       struct remote *remote;
+       struct transport *transport;
+       int original_fetch_if_missing = fetch_if_missing;
+       int res;
+
+       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]);
+
+       transport_set_option(transport, TRANS_OPT_FROM_PROMISOR, "1");
+       transport_set_option(transport, TRANS_OPT_NO_DEPENDENTS, "1");
+       res = transport_fetch_refs(transport, ref);
+       fetch_if_missing = original_fetch_if_missing;
+
+       return res;
+}
+
+static int fetch_objects(const char *remote_name,
+                        const struct object_id *oids,
+                        int oid_nr)
+{
+       struct ref *ref = NULL;
+       int i;
+
+       for (i = 0; i < oid_nr; i++) {
+               struct ref *new_ref = alloc_ref(oid_to_hex(&oids[i]));
+               oidcpy(&new_ref->old_oid, &oids[i]);
+               new_ref->exact_oid = 1;
+               new_ref->next = ref;
+               ref = new_ref;
+       }
+       return fetch_refs(remote_name, ref);
+}
+
+static struct promisor_remote *promisors;
+static struct promisor_remote **promisors_tail = &promisors;
+
+static struct promisor_remote *promisor_remote_new(const char *remote_name)
+{
+       struct promisor_remote *r;
+
+       if (*remote_name == '/') {
+               warning(_("promisor remote name cannot begin with '/': %s"),
+                       remote_name);
+               return NULL;
+       }
+
+       FLEX_ALLOC_STR(r, name, remote_name);
+
+       *promisors_tail = r;
+       promisors_tail = &r->next;
+
+       return r;
+}
+
+static struct promisor_remote *promisor_remote_lookup(const char *remote_name,
+                                                     struct promisor_remote **previous)
+{
+       struct promisor_remote *r, *p;
+
+       for (p = NULL, r = promisors; r; p = r, r = r->next)
+               if (!strcmp(r->name, remote_name)) {
+                       if (previous)
+                               *previous = p;
+                       return r;
+               }
+
+       return NULL;
+}
+
+static void promisor_remote_move_to_tail(struct promisor_remote *r,
+                                        struct promisor_remote *previous)
+{
+       if (previous)
+               previous->next = r->next;
+       else
+               promisors = r->next ? r->next : r;
+       r->next = NULL;
+       *promisors_tail = r;
+       promisors_tail = &r->next;
+}
+
+static int promisor_remote_config(const char *var, const char *value, void *data)
+{
+       const char *name;
+       int namelen;
+       const char *subkey;
+
+       if (!strcmp(var, "core.partialclonefilter"))
+               return git_config_string(&core_partial_clone_filter_default,
+                                        var, value);
+
+       if (parse_config_key(var, "remote", &name, &namelen, &subkey) < 0)
+               return 0;
+
+       if (!strcmp(subkey, "promisor")) {
+               char *remote_name;
+
+               if (!git_config_bool(var, value))
+                       return 0;
+
+               remote_name = xmemdupz(name, namelen);
+
+               if (!promisor_remote_lookup(remote_name, NULL))
+                       promisor_remote_new(remote_name);
+
+               free(remote_name);
+               return 0;
+       }
+       if (!strcmp(subkey, "partialclonefilter")) {
+               struct promisor_remote *r;
+               char *remote_name = xmemdupz(name, namelen);
+
+               r = promisor_remote_lookup(remote_name, NULL);
+               if (!r)
+                       r = promisor_remote_new(remote_name);
+
+               free(remote_name);
+
+               if (!r)
+                       return 0;
+
+               return git_config_string(&r->partial_clone_filter, var, value);
+       }
+
+       return 0;
+}
+
+static int initialized;
+
+static void promisor_remote_init(void)
+{
+       if (initialized)
+               return;
+       initialized = 1;
+
+       git_config(promisor_remote_config, NULL);
+
+       if (repository_format_partial_clone) {
+               struct promisor_remote *o, *previous;
+
+               o = promisor_remote_lookup(repository_format_partial_clone,
+                                          &previous);
+               if (o)
+                       promisor_remote_move_to_tail(o, previous);
+               else
+                       promisor_remote_new(repository_format_partial_clone);
+       }
+}
+
+static void promisor_remote_clear(void)
+{
+       while (promisors) {
+               struct promisor_remote *r = promisors;
+               promisors = promisors->next;
+               free(r);
+       }
+
+       promisors_tail = &promisors;
+}
+
+void promisor_remote_reinit(void)
+{
+       initialized = 0;
+       promisor_remote_clear();
+       promisor_remote_init();
+}
+
+struct promisor_remote *promisor_remote_find(const char *remote_name)
+{
+       promisor_remote_init();
+
+       if (!remote_name)
+               return promisors;
+
+       return promisor_remote_lookup(remote_name, NULL);
+}
+
+int has_promisor_remote(void)
+{
+       return !!promisor_remote_find(NULL);
+}
+
+static int remove_fetched_oids(struct repository *repo,
+                              struct object_id **oids,
+                              int oid_nr, int to_free)
+{
+       int i, remaining_nr = 0;
+       int *remaining = xcalloc(oid_nr, sizeof(*remaining));
+       struct object_id *old_oids = *oids;
+       struct object_id *new_oids;
+
+       for (i = 0; i < oid_nr; i++)
+               if (oid_object_info_extended(repo, &old_oids[i], NULL,
+                                            OBJECT_INFO_SKIP_FETCH_OBJECT)) {
+                       remaining[i] = 1;
+                       remaining_nr++;
+               }
+
+       if (remaining_nr) {
+               int j = 0;
+               new_oids = xcalloc(remaining_nr, sizeof(*new_oids));
+               for (i = 0; i < oid_nr; i++)
+                       if (remaining[i])
+                               oidcpy(&new_oids[j++], &old_oids[i]);
+               *oids = new_oids;
+               if (to_free)
+                       free(old_oids);
+       }
+
+       free(remaining);
+
+       return remaining_nr;
+}
+
+int promisor_remote_get_direct(struct repository *repo,
+                              const struct object_id *oids,
+                              int oid_nr)
+{
+       struct promisor_remote *r;
+       struct object_id *remaining_oids = (struct object_id *)oids;
+       int remaining_nr = oid_nr;
+       int to_free = 0;
+       int res = -1;
+
+       promisor_remote_init();
+
+       for (r = promisors; r; r = r->next) {
+               if (fetch_objects(r->name, remaining_oids, remaining_nr) < 0) {
+                       if (remaining_nr == 1)
+                               continue;
+                       remaining_nr = remove_fetched_oids(repo, &remaining_oids,
+                                                        remaining_nr, to_free);
+                       if (remaining_nr) {
+                               to_free = 1;
+                               continue;
+                       }
+               }
+               res = 0;
+               break;
+       }
+
+       if (to_free)
+               free(remaining_oids);
+
+       return res;
+}
diff --git a/promisor-remote.h b/promisor-remote.h
new file mode 100644 (file)
index 0000000..8200dfc
--- /dev/null
@@ -0,0 +1,31 @@
+#ifndef PROMISOR_REMOTE_H
+#define PROMISOR_REMOTE_H
+
+struct object_id;
+
+/*
+ * Promisor remote linked list
+ *
+ * Information in its fields come from remote.XXX config entries or
+ * from extensions.partialclone or core.partialclonefilter.
+ */
+struct promisor_remote {
+       struct promisor_remote *next;
+       const char *partial_clone_filter;
+       const char name[FLEX_ARRAY];
+};
+
+extern void promisor_remote_reinit(void);
+extern struct promisor_remote *promisor_remote_find(const char *remote_name);
+extern int has_promisor_remote(void);
+extern int promisor_remote_get_direct(struct repository *repo,
+                                     const struct object_id *oids,
+                                     int oid_nr);
+
+/*
+ * This should be used only once from setup.c to set the value we got
+ * from the extensions.partialclone config option.
+ */
+extern void set_repository_format_partial_clone(char *partial_clone);
+
+#endif /* PROMISOR_REMOTE_H */
diff --git a/setup.c b/setup.c
index 8dcb4631f7d330a290ab0bb6810f24999f65a1b7..25a3038277cdaaca736a82191dd6b7bb82cd9ae2 100644 (file)
--- a/setup.c
+++ b/setup.c
@@ -4,6 +4,7 @@
 #include "dir.h"
 #include "string-list.h"
 #include "chdir-notify.h"
+#include "promisor-remote.h"
 
 static int inside_git_dir = -1;
 static int inside_work_tree = -1;
@@ -478,7 +479,7 @@ static int check_repository_format_gently(const char *gitdir, struct repository_
        }
 
        repository_format_precious_objects = candidate->precious_objects;
-       repository_format_partial_clone = xstrdup_or_null(candidate->partial_clone);
+       set_repository_format_partial_clone(candidate->partial_clone);
        repository_format_worktree_config = candidate->worktree_config;
        string_list_clear(&candidate->unknown_extensions, 0);
 
index 4895408e1ed309241a044c019e11f3addeb722a6..e85f249a5db027594d25f9882d85d1253ddbf36f 100644 (file)
@@ -30,8 +30,8 @@
 #include "mergesort.h"
 #include "quote.h"
 #include "packfile.h"
-#include "fetch-object.h"
 #include "object-store.h"
+#include "promisor-remote.h"
 
 /* The maximum size for an object header. */
 #define MAX_HEADER_LEN 32
@@ -1471,16 +1471,17 @@ int oid_object_info_extended(struct repository *r, const struct object_id *oid,
                }
 
                /* Check if it is a missing object */
-               if (fetch_if_missing && repository_format_partial_clone &&
+               if (fetch_if_missing && has_promisor_remote() &&
                    !already_retried && r == the_repository &&
                    !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
                        /*
-                        * 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.
+                        * TODO Investigate checking promisor_remote_get_direct()
+                        * TODO return value and stopping on error here.
+                        * TODO Pass a repository struct through
+                        * promisor_remote_get_direct(), such that arbitrary
+                        * repositories work.
                         */
-                       fetch_objects(repository_format_partial_clone, real, 1);
+                       promisor_remote_get_direct(r, real, 1);
                        already_retried = 1;
                        continue;
                }
index 33e0aa4a896575f14442e5b1a9029640d4d7ac50..d4b7e535ea134569ed8bf5643edf8d435a22bc1a 100755 (executable)
@@ -26,7 +26,7 @@ promise_and_delete () {
 test_expect_success 'extensions.partialclone without filter' '
        test_create_repo server &&
        git clone --filter="blob:none" "file://$(pwd)/server" client &&
-       git -C client config --unset core.partialclonefilter &&
+       git -C client config --unset remote.origin.partialclonefilter &&
        git -C client fetch origin
 '
 
@@ -166,8 +166,9 @@ test_expect_success 'fetching of missing objects' '
        # 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"
+       IDX=$(sed "s/promisor$/idx/" promisorlist) &&
+       git verify-pack --verbose "$IDX" >out &&
+       grep "$HASH" out
 '
 
 test_expect_success 'fetching of missing objects works with ref-in-want enabled' '
@@ -182,8 +183,55 @@ test_expect_success 'fetching of missing objects works with ref-in-want enabled'
        grep "git< fetch=.*ref-in-want" trace
 '
 
+test_expect_success 'fetching of missing objects from another promisor remote' '
+       git clone "file://$(pwd)/server" server2 &&
+       test_commit -C server2 bar &&
+       git -C server2 repack -a -d --write-bitmap-index &&
+       HASH2=$(git -C server2 rev-parse bar) &&
+
+       git -C repo remote add server2 "file://$(pwd)/server2" &&
+       git -C repo config remote.server2.promisor true &&
+       git -C repo cat-file -p "$HASH2" &&
+
+       git -C repo fetch server2 &&
+       rm -rf repo/.git/objects/* &&
+       git -C repo cat-file -p "$HASH2" &&
+
+       # Ensure that the .promisor file is written, and check that its
+       # associated packfile contains the object
+       ls repo/.git/objects/pack/pack-*.promisor >promisorlist &&
+       test_line_count = 1 promisorlist &&
+       IDX=$(sed "s/promisor$/idx/" promisorlist) &&
+       git verify-pack --verbose "$IDX" >out &&
+       grep "$HASH2" out
+'
+
+test_expect_success 'fetching of missing objects configures a promisor remote' '
+       git clone "file://$(pwd)/server" server3 &&
+       test_commit -C server3 baz &&
+       git -C server3 repack -a -d --write-bitmap-index &&
+       HASH3=$(git -C server3 rev-parse baz) &&
+       git -C server3 config uploadpack.allowfilter 1 &&
+
+       rm repo/.git/objects/pack/pack-*.promisor &&
+
+       git -C repo remote add server3 "file://$(pwd)/server3" &&
+       git -C repo fetch --filter="blob:none" server3 $HASH3 &&
+
+       test_cmp_config -C repo true remote.server3.promisor &&
+
+       # Ensure that the .promisor file is written, and check that its
+       # associated packfile contains the object
+       ls repo/.git/objects/pack/pack-*.promisor >promisorlist &&
+       test_line_count = 1 promisorlist &&
+       IDX=$(sed "s/promisor$/idx/" promisorlist) &&
+       git verify-pack --verbose "$IDX" >out &&
+       grep "$HASH3" out
+'
+
 test_expect_success 'fetching of missing blobs works' '
-       rm -rf server repo &&
+       rm -rf server server2 repo &&
+       rm -rf server server3 repo &&
        test_create_repo server &&
        test_commit -C server foo &&
        git -C server repack -a -d --write-bitmap-index &&
@@ -514,8 +562,9 @@ test_expect_success 'fetching of missing objects from an HTTP server' '
        # associated packfile contains the object
        ls repo/.git/objects/pack/pack-*.promisor >promisorlist &&
        test_line_count = 1 promisorlist &&
-       IDX=$(cat promisorlist | sed "s/promisor$/idx/") &&
-       git verify-pack --verbose "$IDX" | grep "$HASH"
+       IDX=$(sed "s/promisor$/idx/" promisorlist) &&
+       git verify-pack --verbose "$IDX" >out &&
+       grep "$HASH" out
 '
 
 # DO NOT add non-httpd-specific tests here, because the last part of this
index 4a3b901f067c2cb024247f2dc9299a6215939019..2e94354e5b11b3dfcf5104e4911cb37003a1c03e 100755 (executable)
@@ -654,7 +654,8 @@ partial_clone () {
        git -C client fsck &&
 
        # Ensure that unneeded blobs are not inadvertently fetched.
-       test_config -C client extensions.partialclone "not a remote" &&
+       test_config -C client remote.origin.promisor "false" &&
+       git -C client config --unset remote.origin.partialclonefilter &&
        test_must_fail git -C client cat-file -e "$HASH1" &&
 
        # But this blob was fetched, because clone performs an initial checkout
index 565254558f362397270bcf13bd3566993e1b0f50..73cd95812f143bbea40df9242eed64d78e0bf53f 100755 (executable)
@@ -42,8 +42,8 @@ test_expect_success 'do partial clone 1' '
 
        test_cmp expect_1.oids observed.oids &&
        test "$(git -C pc1 config --local core.repositoryformatversion)" = "1" &&
-       test "$(git -C pc1 config --local extensions.partialclone)" = "origin" &&
-       test "$(git -C pc1 config --local core.partialclonefilter)" = "blob:none"
+       test "$(git -C pc1 config --local remote.origin.promisor)" = "true" &&
+       test "$(git -C pc1 config --local remote.origin.partialclonefilter)" = "blob:none"
 '
 
 # checkout master to force dynamic object fetch of blobs at HEAD.
index 50f257bd5c768ae87f8e526c587fa11afad98ad6..9c25126aecaabc8f0e3d0841ee1aa959acdf8dbd 100644 (file)
@@ -16,7 +16,7 @@
 #include "submodule-config.h"
 #include "fsmonitor.h"
 #include "object-store.h"
-#include "fetch-object.h"
+#include "promisor-remote.h"
 
 /*
  * Error messages expected by scripts out of plumbing commands such as
@@ -400,7 +400,7 @@ static int check_updates(struct unpack_trees_options *o)
                load_gitmodules_file(index, &state);
 
        enable_delayed_checkout(&state);
-       if (repository_format_partial_clone && o->update && !o->dry_run) {
+       if (has_promisor_remote() && o->update && !o->dry_run) {
                /*
                 * Prefetch the objects that are to be checked out in the loop
                 * below.
@@ -419,8 +419,8 @@ static int check_updates(struct unpack_trees_options *o)
                        oid_array_append(&to_fetch, &ce->oid);
                }
                if (to_fetch.nr)
-                       fetch_objects(repository_format_partial_clone,
-                                     to_fetch.oid, to_fetch.nr);
+                       promisor_remote_get_direct(the_repository,
+                                                  to_fetch.oid, to_fetch.nr);
                oid_array_clear(&to_fetch);
        }
        for (i = 0; i < index->cache_nr; i++) {