]> git.ipfire.org Git - thirdparty/git.git/commitdiff
builtin/repack.c: support writing a MIDX while repacking
authorTaylor Blau <me@ttaylorr.com>
Wed, 29 Sep 2021 01:55:18 +0000 (21:55 -0400)
committerJunio C Hamano <gitster@pobox.com>
Wed, 29 Sep 2021 04:20:56 +0000 (21:20 -0700)
Teach `git repack` a new `--write-midx` option for callers that wish to
persist a multi-pack index in their repository while repacking.

There are two existing alternatives to this new flag, but they don't
cover our particular use-case. These alternatives are:

  - Call 'git multi-pack-index write' after running 'git repack', or

  - Set 'GIT_TEST_MULTI_PACK_INDEX=1' in your environment when running
    'git repack'.

The former works, but introduces a gap in bitmap coverage between
repacking and writing a new MIDX (since the repack may have deleted a
pack included in the existing MIDX, invalidating it altogether).

Setting the 'GIT_TEST_' environment variable is obviously unsupported.
In fact, even if it were supported officially, it still wouldn't work,
because it generates the MIDX *after* redundant packs have been dropped,
leading to the same issue as above.

Introduce a new option which eliminates this race by teaching `git
repack` to generate the MIDX at the critical point: after the new packs
have been written and moved into place, but before the redundant packs
have been removed.

This option is compatible with `git repack`'s '--bitmap' option (it
changes the interpretation to be: "write a bitmap corresponding to the
MIDX after one has been generated").

There is a little bit of additional noise in the patch below to avoid
repeating ourselves when selecting which packs to delete. Instead of a
single loop as before (where we iterate over 'existing_packs', decide if
a pack is worth deleting, and if so, delete it), we have two loops (the
first where we decide which ones are worth deleting, and the second
where we actually do the deleting). This makes it so we have a single
check we can make consistently when (1) telling the MIDX which packs we
want to exclude, and (2) actually unlinking the redundant packs.

There is also a tiny change to short-circuit the body of
write_midx_included_packs() when no packs remain in the case of an empty
repository. The MIDX code does not handle this, so avoid trying to
generate a MIDX covering zero packs in the first place.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Documentation/git-repack.txt
builtin/repack.c
t/lib-midx.sh [new file with mode: 0644]
t/t7700-repack.sh
t/t7703-repack-geometric.sh

index 24c00c9384f40b3d6ff487d21d74d8fc19292bee..0f2d235ca559929f092eb3969d5a48cfcb13630d 100644 (file)
@@ -9,7 +9,7 @@ git-repack - Pack unpacked objects in a repository
 SYNOPSIS
 --------
 [verse]
-'git repack' [-a] [-A] [-d] [-f] [-F] [-l] [-n] [-q] [-b] [--window=<n>] [--depth=<n>] [--threads=<n>] [--keep-pack=<pack-name>]
+'git repack' [-a] [-A] [-d] [-f] [-F] [-l] [-n] [-q] [-b] [-m] [--window=<n>] [--depth=<n>] [--threads=<n>] [--keep-pack=<pack-name>] [--write-midx]
 
 DESCRIPTION
 -----------
@@ -128,10 +128,11 @@ depth is 4095.
 -b::
 --write-bitmap-index::
        Write a reachability bitmap index as part of the repack. This
-       only makes sense when used with `-a` or `-A`, as the bitmaps
+       only makes sense when used with `-a`, `-A` or `-m`, as the bitmaps
        must be able to refer to all reachable objects. This option
-       overrides the setting of `repack.writeBitmaps`.  This option
-       has no effect if multiple packfiles are created.
+       overrides the setting of `repack.writeBitmaps`. This option
+       has no effect if multiple packfiles are created, unless writing a
+       MIDX (in which case a multi-pack bitmap is created).
 
 --pack-kept-objects::
        Include objects in `.keep` files when repacking.  Note that we
@@ -190,6 +191,11 @@ to change in the future. This option (implying a drastically different
 repack mode) is not guaranteed to work with all other combinations of
 option to `git repack`.
 
+-m::
+--write-midx::
+       Write a multi-pack index (see linkgit:git-multi-pack-index[1])
+       containing the non-redundant packs.
+
 CONFIGURATION
 -------------
 
index ead4f48fdaac16e08231f30befa45c5e097ed210..dbbb14b3b4b1ca30fb0c53ce666ea4aa728830ac 100644 (file)
@@ -434,6 +434,76 @@ static void clear_pack_geometry(struct pack_geometry *geometry)
        geometry->split = 0;
 }
 
+static void midx_included_packs(struct string_list *include,
+                               struct string_list *existing_nonkept_packs,
+                               struct string_list *existing_kept_packs,
+                               struct string_list *names,
+                               struct pack_geometry *geometry)
+{
+       struct string_list_item *item;
+
+       for_each_string_list_item(item, existing_kept_packs)
+               string_list_insert(include, xstrfmt("%s.idx", item->string));
+       for_each_string_list_item(item, names)
+               string_list_insert(include, xstrfmt("pack-%s.idx", item->string));
+       if (geometry) {
+               struct strbuf buf = STRBUF_INIT;
+               uint32_t i;
+               for (i = geometry->split; i < geometry->pack_nr; i++) {
+                       struct packed_git *p = geometry->pack[i];
+
+                       strbuf_addstr(&buf, pack_basename(p));
+                       strbuf_strip_suffix(&buf, ".pack");
+                       strbuf_addstr(&buf, ".idx");
+
+                       string_list_insert(include, strbuf_detach(&buf, NULL));
+               }
+       } else {
+               for_each_string_list_item(item, existing_nonkept_packs) {
+                       if (item->util)
+                               continue;
+                       string_list_insert(include, xstrfmt("%s.idx", item->string));
+               }
+       }
+}
+
+static int write_midx_included_packs(struct string_list *include,
+                                    int show_progress, int write_bitmaps)
+{
+       struct child_process cmd = CHILD_PROCESS_INIT;
+       struct string_list_item *item;
+       FILE *in;
+       int ret;
+
+       if (!include->nr)
+               return 0;
+
+       cmd.in = -1;
+       cmd.git_cmd = 1;
+
+       strvec_push(&cmd.args, "multi-pack-index");
+       strvec_pushl(&cmd.args, "write", "--stdin-packs", NULL);
+
+       if (show_progress)
+               strvec_push(&cmd.args, "--progress");
+       else
+               strvec_push(&cmd.args, "--no-progress");
+
+       if (write_bitmaps)
+               strvec_push(&cmd.args, "--bitmap");
+
+       ret = start_command(&cmd);
+       if (ret)
+               return ret;
+
+       in = xfdopen(cmd.in, "w");
+       for_each_string_list_item(item, include)
+               fprintf(in, "%s\n", item->string);
+       fclose(in);
+
+       return finish_command(&cmd);
+}
+
 int cmd_repack(int argc, const char **argv, const char *prefix)
 {
        struct child_process cmd = CHILD_PROCESS_INIT;
@@ -457,6 +527,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        int no_update_server_info = 0;
        struct pack_objects_args po_args = {NULL};
        int geometric_factor = 0;
+       int write_midx = 0;
 
        struct option builtin_repack_options[] = {
                OPT_BIT('a', NULL, &pack_everything,
@@ -499,6 +570,8 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
                                N_("do not repack this pack")),
                OPT_INTEGER('g', "geometric", &geometric_factor,
                            N_("find a geometric progression with factor <N>")),
+               OPT_BOOL('m', "write-midx", &write_midx,
+                          N_("write a multi-pack index of the resulting packs")),
                OPT_END()
        };
 
@@ -515,8 +588,8 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
                die(_("--keep-unreachable and -A are incompatible"));
 
        if (write_bitmaps < 0) {
-               if (!(pack_everything & ALL_INTO_ONE) ||
-                   !is_bare_repository())
+               if (!write_midx &&
+                   (!(pack_everything & ALL_INTO_ONE) || !is_bare_repository()))
                        write_bitmaps = 0;
        } else if (write_bitmaps &&
                   git_env_bool(GIT_TEST_MULTI_PACK_INDEX, 0) &&
@@ -526,7 +599,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        if (pack_kept_objects < 0)
                pack_kept_objects = write_bitmaps > 0;
 
-       if (write_bitmaps && !(pack_everything & ALL_INTO_ONE))
+       if (write_bitmaps && !(pack_everything & ALL_INTO_ONE) && !write_midx)
                die(_(incremental_bitmap_conflict_error));
 
        if (geometric_factor) {
@@ -568,10 +641,12 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        }
        if (has_promisor_remote())
                strvec_push(&cmd.args, "--exclude-promisor-objects");
-       if (write_bitmaps > 0)
-               strvec_push(&cmd.args, "--write-bitmap-index");
-       else if (write_bitmaps < 0)
-               strvec_push(&cmd.args, "--write-bitmap-index-quiet");
+       if (!write_midx) {
+               if (write_bitmaps > 0)
+                       strvec_push(&cmd.args, "--write-bitmap-index");
+               else if (write_bitmaps < 0)
+                       strvec_push(&cmd.args, "--write-bitmap-index-quiet");
+       }
        if (use_delta_islands)
                strvec_push(&cmd.args, "--delta-islands");
 
@@ -684,22 +759,47 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        }
        /* End of pack replacement. */
 
+       if (delete_redundant && pack_everything & ALL_INTO_ONE) {
+               const int hexsz = the_hash_algo->hexsz;
+               string_list_sort(&names);
+               for_each_string_list_item(item, &existing_nonkept_packs) {
+                       char *sha1;
+                       size_t len = strlen(item->string);
+                       if (len < hexsz)
+                               continue;
+                       sha1 = item->string + len - hexsz;
+                       /*
+                        * Mark this pack for deletion, which ensures that this
+                        * pack won't be included in a MIDX (if `--write-midx`
+                        * was given) and that we will actually delete this pack
+                        * (if `-d` was given).
+                        */
+                       item->util = (void*)(intptr_t)!string_list_has_string(&names, sha1);
+               }
+       }
+
+       if (write_midx) {
+               struct string_list include = STRING_LIST_INIT_NODUP;
+               midx_included_packs(&include, &existing_nonkept_packs,
+                                   &existing_kept_packs, &names, geometry);
+
+               ret = write_midx_included_packs(&include,
+                                               show_progress, write_bitmaps > 0);
+
+               string_list_clear(&include, 0);
+
+               if (ret)
+                       return ret;
+       }
+
        reprepare_packed_git(the_repository);
 
        if (delete_redundant) {
                int opts = 0;
-               if (pack_everything & ALL_INTO_ONE) {
-                       const int hexsz = the_hash_algo->hexsz;
-                       string_list_sort(&names);
-                       for_each_string_list_item(item, &existing_nonkept_packs) {
-                               char *sha1;
-                               size_t len = strlen(item->string);
-                               if (len < hexsz)
-                                       continue;
-                               sha1 = item->string + len - hexsz;
-                               if (!string_list_has_string(&names, sha1))
-                                       remove_redundant_pack(packdir, item->string);
-                       }
+               for_each_string_list_item(item, &existing_nonkept_packs) {
+                       if (!item->util)
+                               continue;
+                       remove_redundant_pack(packdir, item->string);
                }
 
                if (geometry) {
diff --git a/t/lib-midx.sh b/t/lib-midx.sh
new file mode 100644 (file)
index 0000000..1261994
--- /dev/null
@@ -0,0 +1,8 @@
+# test_midx_consistent <objdir>
+test_midx_consistent () {
+       ls $1/pack/pack-*.idx | xargs -n 1 basename | sort >expect &&
+       test-tool read-midx $1 | grep ^pack-.*\.idx$ | sort >actual &&
+
+       test_cmp expect actual &&
+       git multi-pack-index --object-dir=$1 verify
+}
index 98eda3bfeb56c4cd685ed352ee0fe2a885061c31..6792531dfd85786753db854fa7e1d8a74747e22d 100755 (executable)
@@ -3,6 +3,8 @@
 test_description='git repack works correctly'
 
 . ./test-lib.sh
+. "${TEST_DIRECTORY}/lib-bitmap.sh"
+. "${TEST_DIRECTORY}/lib-midx.sh"
 
 commit_and_pack () {
        test_commit "$@" 1>&2 &&
@@ -234,4 +236,98 @@ test_expect_success 'auto-bitmaps do not complain if unavailable' '
        test_must_be_empty actual
 '
 
+objdir=.git/objects
+midx=$objdir/pack/multi-pack-index
+
+test_expect_success 'setup for --write-midx tests' '
+       git init midx &&
+       (
+               cd midx &&
+               git config core.multiPackIndex true &&
+
+               test_commit base
+       )
+'
+
+test_expect_success '--write-midx unchanged' '
+       (
+               cd midx &&
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack &&
+               test_path_is_missing $midx &&
+               test_path_is_missing $midx-*.bitmap &&
+
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack --write-midx &&
+
+               test_path_is_file $midx &&
+               test_path_is_missing $midx-*.bitmap &&
+               test_midx_consistent $objdir
+       )
+'
+
+test_expect_success '--write-midx with a new pack' '
+       (
+               cd midx &&
+               test_commit loose &&
+
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack --write-midx &&
+
+               test_path_is_file $midx &&
+               test_path_is_missing $midx-*.bitmap &&
+               test_midx_consistent $objdir
+       )
+'
+
+test_expect_success '--write-midx with -b' '
+       (
+               cd midx &&
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack -mb &&
+
+               test_path_is_file $midx &&
+               test_path_is_file $midx-*.bitmap &&
+               test_midx_consistent $objdir
+       )
+'
+
+test_expect_success '--write-midx with -d' '
+       (
+               cd midx &&
+               test_commit repack &&
+
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack -Ad --write-midx &&
+
+               test_path_is_file $midx &&
+               test_path_is_missing $midx-*.bitmap &&
+               test_midx_consistent $objdir
+       )
+'
+
+test_expect_success 'cleans up MIDX when appropriate' '
+       (
+               cd midx &&
+
+               test_commit repack-2 &&
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack -Adb --write-midx &&
+
+               checksum=$(midx_checksum $objdir) &&
+               test_path_is_file $midx &&
+               test_path_is_file $midx-$checksum.bitmap &&
+               test_path_is_file $midx-$checksum.rev &&
+
+               test_commit repack-3 &&
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack -Adb --write-midx &&
+
+               test_path_is_file $midx &&
+               test_path_is_missing $midx-$checksum.bitmap &&
+               test_path_is_missing $midx-$checksum.rev &&
+               test_path_is_file $midx-$(midx_checksum $objdir).bitmap &&
+               test_path_is_file $midx-$(midx_checksum $objdir).rev &&
+
+               test_commit repack-4 &&
+               GIT_TEST_MULTI_PACK_INDEX=0 git repack -Adb &&
+
+               find $objdir/pack -type f -name "multi-pack-index*" >files &&
+               test_must_be_empty files
+       )
+'
+
 test_done
index 5ccaa440e0c140e8ea34112c49c62248539d9678..67049f763762ee70945b81b6cfcdb923372ebb6a 100755 (executable)
@@ -15,7 +15,7 @@ test_expect_success '--geometric with no packs' '
        (
                cd geometric &&
 
-               git repack --geometric 2 >out &&
+               git repack --write-midx --geometric 2 >out &&
                test_i18ngrep "Nothing new to pack" out
        )
 '