Merge branch 'sg/split-index-racefix'
[gitweb.git] / builtin / repack.c
index 7bdb40142f9261dac6514d98ee01dc44e595ec71..c6a7943d5cb108dbccdfd502d799ab71a9e7e146 100644 (file)
@@ -8,10 +8,14 @@
 #include "strbuf.h"
 #include "string-list.h"
 #include "argv-array.h"
+#include "midx.h"
+#include "packfile.h"
+#include "object-store.h"
 
 static int delta_base_offset = 1;
 static int pack_kept_objects = -1;
 static int write_bitmaps;
+static int use_delta_islands;
 static char *packdir, *packtmp;
 
 static const char *const git_repack_usage[] = {
@@ -40,6 +44,10 @@ static int repack_config(const char *var, const char *value, void *cb)
                write_bitmaps = git_config_bool(var, value);
                return 0;
        }
+       if (!strcmp(var, "repack.usedeltaislands")) {
+               use_delta_islands = git_config_bool(var, value);
+               return 0;
+       }
        return git_default_config(var, value, cb);
 }
 
@@ -83,10 +91,11 @@ static void remove_pack_on_signal(int signo)
 
 /*
  * Adds all packs hex strings to the fname list, which do not
- * have a corresponding .keep or .promisor file. These packs are not to
+ * have a corresponding .keep 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)
+static void get_non_kept_pack_filenames(struct string_list *fname_list,
+                                       const struct string_list *extra_keep)
 {
        DIR *dir;
        struct dirent *e;
@@ -97,13 +106,20 @@ static void get_non_kept_pack_filenames(struct string_list *fname_list)
 
        while ((e = readdir(dir)) != NULL) {
                size_t len;
+               int i;
+
+               for (i = 0; i < extra_keep->nr; i++)
+                       if (!fspathcmp(e->d_name, extra_keep->items[i].string))
+                               break;
+               if (extra_keep->nr > 0 && i < extra_keep->nr)
+                       continue;
+
                if (!strip_suffix(e->d_name, ".pack", &len))
                        continue;
 
                fname = xmemdupz(e->d_name, len);
 
-               if (!file_exists(mkpath("%s/%s.keep", packdir, fname)) &&
-                   !file_exists(mkpath("%s/%s.promisor", packdir, fname)))
+               if (!file_exists(mkpath("%s/%s.keep", packdir, fname)))
                        string_list_append_nodup(fname_list, fname);
                else
                        free(fname);
@@ -113,7 +129,7 @@ static void get_non_kept_pack_filenames(struct string_list *fname_list)
 
 static void remove_redundant_pack(const char *dir_name, const char *base_name)
 {
-       const char *exts[] = {".pack", ".idx", ".keep", ".bitmap"};
+       const char *exts[] = {".pack", ".idx", ".keep", ".bitmap", ".promisor"};
        int i;
        struct strbuf buf = STRBUF_INIT;
        size_t plen;
@@ -129,6 +145,117 @@ static void remove_redundant_pack(const char *dir_name, const char *base_name)
        strbuf_release(&buf);
 }
 
+struct pack_objects_args {
+       const char *window;
+       const char *window_memory;
+       const char *depth;
+       const char *threads;
+       const char *max_pack_size;
+       int no_reuse_delta;
+       int no_reuse_object;
+       int quiet;
+       int local;
+};
+
+static void prepare_pack_objects(struct child_process *cmd,
+                                const struct pack_objects_args *args)
+{
+       argv_array_push(&cmd->args, "pack-objects");
+       if (args->window)
+               argv_array_pushf(&cmd->args, "--window=%s", args->window);
+       if (args->window_memory)
+               argv_array_pushf(&cmd->args, "--window-memory=%s", args->window_memory);
+       if (args->depth)
+               argv_array_pushf(&cmd->args, "--depth=%s", args->depth);
+       if (args->threads)
+               argv_array_pushf(&cmd->args, "--threads=%s", args->threads);
+       if (args->max_pack_size)
+               argv_array_pushf(&cmd->args, "--max-pack-size=%s", args->max_pack_size);
+       if (args->no_reuse_delta)
+               argv_array_pushf(&cmd->args, "--no-reuse-delta");
+       if (args->no_reuse_object)
+               argv_array_pushf(&cmd->args, "--no-reuse-object");
+       if (args->local)
+               argv_array_push(&cmd->args,  "--local");
+       if (args->quiet)
+               argv_array_push(&cmd->args,  "--quiet");
+       if (delta_base_offset)
+               argv_array_push(&cmd->args,  "--delta-base-offset");
+       argv_array_push(&cmd->args, packtmp);
+       cmd->git_cmd = 1;
+       cmd->out = -1;
+}
+
+/*
+ * Write oid to the given struct child_process's stdin, starting it first if
+ * necessary.
+ */
+static int write_oid(const struct object_id *oid, struct packed_git *pack,
+                    uint32_t pos, void *data)
+{
+       struct child_process *cmd = data;
+
+       if (cmd->in == -1) {
+               if (start_command(cmd))
+                       die("Could not start pack-objects to repack promisor objects");
+       }
+
+       xwrite(cmd->in, oid_to_hex(oid), GIT_SHA1_HEXSZ);
+       xwrite(cmd->in, "\n", 1);
+       return 0;
+}
+
+static void repack_promisor_objects(const struct pack_objects_args *args,
+                                   struct string_list *names)
+{
+       struct child_process cmd = CHILD_PROCESS_INIT;
+       FILE *out;
+       struct strbuf line = STRBUF_INIT;
+
+       prepare_pack_objects(&cmd, args);
+       cmd.in = -1;
+
+       /*
+        * NEEDSWORK: Giving pack-objects only the OIDs without any ordering
+        * hints may result in suboptimal deltas in the resulting pack. See if
+        * the OIDs can be sent with fake paths such that pack-objects can use a
+        * {type -> existing pack order} ordering when computing deltas instead
+        * of a {type -> size} ordering, which may produce better deltas.
+        */
+       for_each_packed_object(write_oid, &cmd,
+                              FOR_EACH_OBJECT_PROMISOR_ONLY);
+
+       if (cmd.in == -1)
+               /* No packed objects; cmd was never started */
+               return;
+
+       close(cmd.in);
+
+       out = xfdopen(cmd.out, "r");
+       while (strbuf_getline_lf(&line, out) != EOF) {
+               char *promisor_name;
+               int fd;
+               if (line.len != 40)
+                       die("repack: Expecting 40 character sha1 lines only from pack-objects.");
+               string_list_append(names, line.buf);
+
+               /*
+                * pack-objects creates the .pack and .idx files, but not the
+                * .promisor file. Create the .promisor file, which is empty.
+                */
+               promisor_name = mkpathdup("%s-%s.promisor", packtmp,
+                                         line.buf);
+               fd = open(promisor_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
+               if (fd < 0)
+                       die_errno("unable to create '%s'", promisor_name);
+               close(fd);
+               free(promisor_name);
+       }
+       fclose(out);
+       if (finish_command(&cmd))
+               die("Could not finish pack-objects to repack promisor objects");
+}
+
 #define ALL_INTO_ONE 1
 #define LOOSEN_UNREACHABLE 2
 
@@ -141,6 +268,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
                {".pack"},
                {".idx"},
                {".bitmap", 1},
+               {".promisor", 1},
        };
        struct child_process cmd = CHILD_PROCESS_INIT;
        struct string_list_item *item;
@@ -148,7 +276,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        struct string_list rollback = STRING_LIST_INIT_NODUP;
        struct string_list existing_packs = STRING_LIST_INIT_DUP;
        struct strbuf line = STRBUF_INIT;
-       int ext, ret, failed;
+       int i, ext, ret, failed;
        FILE *out;
 
        /* variables to be filled by option parsing */
@@ -156,14 +284,10 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        int delete_redundant = 0;
        const char *unpack_unreachable = NULL;
        int keep_unreachable = 0;
-       const char *window = NULL, *window_memory = NULL;
-       const char *depth = NULL;
-       const char *threads = NULL;
-       const char *max_pack_size = NULL;
-       int no_reuse_delta = 0, no_reuse_object = 0;
+       struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
        int no_update_server_info = 0;
-       int quiet = 0;
-       int local = 0;
+       int midx_cleared = 0;
+       struct pack_objects_args po_args = {NULL};
 
        struct option builtin_repack_options[] = {
                OPT_BIT('a', NULL, &pack_everything,
@@ -173,33 +297,37 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
                                   LOOSEN_UNREACHABLE | ALL_INTO_ONE),
                OPT_BOOL('d', NULL, &delete_redundant,
                                N_("remove redundant packs, and run git-prune-packed")),
-               OPT_BOOL('f', NULL, &no_reuse_delta,
+               OPT_BOOL('f', NULL, &po_args.no_reuse_delta,
                                N_("pass --no-reuse-delta to git-pack-objects")),
-               OPT_BOOL('F', NULL, &no_reuse_object,
+               OPT_BOOL('F', NULL, &po_args.no_reuse_object,
                                N_("pass --no-reuse-object to git-pack-objects")),
                OPT_BOOL('n', NULL, &no_update_server_info,
                                N_("do not run git-update-server-info")),
-               OPT__QUIET(&quiet, N_("be quiet")),
-               OPT_BOOL('l', "local", &local,
+               OPT__QUIET(&po_args.quiet, N_("be quiet")),
+               OPT_BOOL('l', "local", &po_args.local,
                                N_("pass --local to git-pack-objects")),
                OPT_BOOL('b', "write-bitmap-index", &write_bitmaps,
                                N_("write bitmap index")),
+               OPT_BOOL('i', "delta-islands", &use_delta_islands,
+                               N_("pass --delta-islands to git-pack-objects")),
                OPT_STRING(0, "unpack-unreachable", &unpack_unreachable, N_("approxidate"),
                                N_("with -A, do not loosen objects older than this")),
                OPT_BOOL('k', "keep-unreachable", &keep_unreachable,
                                N_("with -a, repack unreachable objects")),
-               OPT_STRING(0, "window", &window, N_("n"),
+               OPT_STRING(0, "window", &po_args.window, N_("n"),
                                N_("size of the window used for delta compression")),
-               OPT_STRING(0, "window-memory", &window_memory, N_("bytes"),
+               OPT_STRING(0, "window-memory", &po_args.window_memory, N_("bytes"),
                                N_("same as the above, but limit memory size instead of entries count")),
-               OPT_STRING(0, "depth", &depth, N_("n"),
+               OPT_STRING(0, "depth", &po_args.depth, N_("n"),
                                N_("limits the maximum delta depth")),
-               OPT_STRING(0, "threads", &threads, N_("n"),
+               OPT_STRING(0, "threads", &po_args.threads, N_("n"),
                                N_("limits the maximum number of threads")),
-               OPT_STRING(0, "max-pack-size", &max_pack_size, N_("bytes"),
+               OPT_STRING(0, "max-pack-size", &po_args.max_pack_size, N_("bytes"),
                                N_("maximum size of each packfile")),
                OPT_BOOL(0, "pack-kept-objects", &pack_kept_objects,
                                N_("repack objects in packs marked with .keep")),
+               OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
+                               N_("do not repack this pack")),
                OPT_END()
        };
 
@@ -226,35 +354,29 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
 
        sigchain_push_common(remove_pack_on_signal);
 
-       argv_array_push(&cmd.args, "pack-objects");
+       prepare_pack_objects(&cmd, &po_args);
+
        argv_array_push(&cmd.args, "--keep-true-parents");
        if (!pack_kept_objects)
                argv_array_push(&cmd.args, "--honor-pack-keep");
+       for (i = 0; i < keep_pack_list.nr; i++)
+               argv_array_pushf(&cmd.args, "--keep-pack=%s",
+                                keep_pack_list.items[i].string);
        argv_array_push(&cmd.args, "--non-empty");
        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)
-               argv_array_pushf(&cmd.args, "--window-memory=%s", window_memory);
-       if (depth)
-               argv_array_pushf(&cmd.args, "--depth=%s", depth);
-       if (threads)
-               argv_array_pushf(&cmd.args, "--threads=%s", threads);
-       if (max_pack_size)
-               argv_array_pushf(&cmd.args, "--max-pack-size=%s", max_pack_size);
-       if (no_reuse_delta)
-               argv_array_pushf(&cmd.args, "--no-reuse-delta");
-       if (no_reuse_object)
-               argv_array_pushf(&cmd.args, "--no-reuse-object");
        if (write_bitmaps)
                argv_array_push(&cmd.args, "--write-bitmap-index");
+       if (use_delta_islands)
+               argv_array_push(&cmd.args, "--delta-islands");
 
        if (pack_everything & ALL_INTO_ONE) {
-               get_non_kept_pack_filenames(&existing_packs);
+               get_non_kept_pack_filenames(&existing_packs, &keep_pack_list);
+
+               repack_promisor_objects(&po_args, &names);
 
                if (existing_packs.nr && delete_redundant) {
                        if (unpack_unreachable) {
@@ -277,17 +399,6 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
                argv_array_push(&cmd.args, "--incremental");
        }
 
-       if (local)
-               argv_array_push(&cmd.args,  "--local");
-       if (quiet)
-               argv_array_push(&cmd.args,  "--quiet");
-       if (delta_base_offset)
-               argv_array_push(&cmd.args,  "--delta-base-offset");
-
-       argv_array_push(&cmd.args, packtmp);
-
-       cmd.git_cmd = 1;
-       cmd.out = -1;
        cmd.no_stdin = 1;
 
        ret = start_command(&cmd);
@@ -305,7 +416,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        if (ret)
                return ret;
 
-       if (!names.nr && !quiet)
+       if (!names.nr && !po_args.quiet)
                printf("Nothing new to pack.\n");
 
        /*
@@ -318,6 +429,13 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
        for_each_string_list_item(item, &names) {
                for (ext = 0; ext < ARRAY_SIZE(exts); ext++) {
                        char *fname, *fname_old;
+
+                       if (!midx_cleared) {
+                               /* if we move a packfile, it will invalidated the midx */
+                               clear_midx_file(get_object_directory());
+                               midx_cleared = 1;
+                       }
+
                        fname = mkpathdup("%s/pack-%s%s", packdir,
                                                item->string, exts[ext].name);
                        if (!file_exists(fname)) {
@@ -414,6 +532,8 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
 
        /* End of pack replacement. */
 
+       reprepare_packed_git(the_repository);
+
        if (delete_redundant) {
                int opts = 0;
                string_list_sort(&names);
@@ -426,7 +546,7 @@ int cmd_repack(int argc, const char **argv, const char *prefix)
                        if (!string_list_has_string(&names, sha1))
                                remove_redundant_pack(packdir, item->string);
                }
-               if (!quiet && isatty(2))
+               if (!po_args.quiet && isatty(2))
                        opts |= PRUNE_PACKED_VERBOSE;
                prune_packed_objects(opts);
        }