merge-recursive: give notice when submodule commit gets fast-forwarded
[gitweb.git] / merge-recursive.c
index ed529f2ceb6fc1064f18f218962232215e0c02af..a9aecccb8c3d59fda25eeae2596ee6b816223bde 100644 (file)
 #include "merge-recursive.h"
 #include "dir.h"
 #include "submodule.h"
+#include "revision.h"
+
+struct path_hashmap_entry {
+       struct hashmap_entry e;
+       char path[FLEX_ARRAY];
+};
+
+static int path_hashmap_cmp(const void *cmp_data,
+                           const void *entry,
+                           const void *entry_or_key,
+                           const void *keydata)
+{
+       const struct path_hashmap_entry *a = entry;
+       const struct path_hashmap_entry *b = entry_or_key;
+       const char *key = keydata;
+
+       if (ignore_case)
+               return strcasecmp(a->path, key ? key : b->path);
+       else
+               return strcmp(a->path, key ? key : b->path);
+}
+
+static unsigned int path_hash(const char *path)
+{
+       return ignore_case ? strihash(path) : strhash(path);
+}
 
 static void flush_output(struct merge_options *o)
 {
@@ -203,7 +229,7 @@ static void output_commit_title(struct merge_options *o, struct commit *commit)
                strbuf_addf(&o->obuf, "virtual %s\n",
                        merge_remote_util(commit)->name);
        else {
-               strbuf_add_unique_abbrev(&o->obuf, commit->object.oid.hash,
+               strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid,
                                         DEFAULT_ABBREV);
                strbuf_addch(&o->obuf, ' ');
                if (parse_commit(commit) != 0)
@@ -310,33 +336,29 @@ struct tree *write_tree_from_memory(struct merge_options *o)
        return result;
 }
 
-static int save_files_dirs(const unsigned char *sha1,
+static int save_files_dirs(const struct object_id *oid,
                struct strbuf *base, const char *path,
                unsigned int mode, int stage, void *context)
 {
+       struct path_hashmap_entry *entry;
        int baselen = base->len;
        struct merge_options *o = context;
 
        strbuf_addstr(base, path);
 
-       if (S_ISDIR(mode))
-               string_list_insert(&o->current_directory_set, base->buf);
-       else
-               string_list_insert(&o->current_file_set, base->buf);
+       FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
+       hashmap_entry_init(entry, path_hash(entry->path));
+       hashmap_add(&o->current_file_dir_set, entry);
 
        strbuf_setlen(base, baselen);
        return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
 }
 
-static int get_files_dirs(struct merge_options *o, struct tree *tree)
+static void get_files_dirs(struct merge_options *o, struct tree *tree)
 {
-       int n;
        struct pathspec match_all;
        memset(&match_all, 0, sizeof(match_all));
-       if (read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o))
-               return 0;
-       n = o->current_file_set.nr + o->current_directory_set.nr;
-       return n;
+       read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
 }
 
 /*
@@ -349,12 +371,12 @@ static struct stage_data *insert_stage_data(const char *path,
 {
        struct string_list_item *item;
        struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
-       get_tree_entry(o->object.oid.hash, path,
-                       e->stages[1].oid.hash, &e->stages[1].mode);
-       get_tree_entry(a->object.oid.hash, path,
-                       e->stages[2].oid.hash, &e->stages[2].mode);
-       get_tree_entry(b->object.oid.hash, path,
-                       e->stages[3].oid.hash, &e->stages[3].mode);
+       get_tree_entry(&o->object.oid, path,
+                       &e->stages[1].oid, &e->stages[1].mode);
+       get_tree_entry(&a->object.oid, path,
+                       &e->stages[2].oid, &e->stages[2].mode);
+       get_tree_entry(&b->object.oid, path,
+                       &e->stages[3].oid, &e->stages[3].mode);
        item = string_list_insert(entries, path);
        item->util = e;
        return e;
@@ -492,6 +514,25 @@ static void record_df_conflict_files(struct merge_options *o,
 
 struct rename {
        struct diff_filepair *pair;
+       /*
+        * Purpose of src_entry and dst_entry:
+        *
+        * If 'before' is renamed to 'after' then src_entry will contain
+        * the versions of 'before' from the merge_base, HEAD, and MERGE in
+        * stages 1, 2, and 3; dst_entry will contain the respective
+        * versions of 'after' in corresponding locations.  Thus, we have a
+        * total of six modes and oids, though some will be null.  (Stage 0
+        * is ignored; we're interested in handling conflicts.)
+        *
+        * Since we don't turn on break-rewrites by default, neither
+        * src_entry nor dst_entry can have all three of their stages have
+        * non-null oids, meaning at most four of the six will be non-null.
+        * Also, since this is a rename, both src_entry and dst_entry will
+        * have at least one non-null oid, meaning at least two will be
+        * non-null.  Of the six oids, a typical rename will have three be
+        * non-null.  Only two implies a rename/delete, and four implies a
+        * rename/add.
+        */
        struct stage_data *src_entry;
        struct stage_data *dst_entry;
        unsigned processed:1;
@@ -519,8 +560,8 @@ static struct string_list *get_renames(struct merge_options *o,
                return renames;
 
        diff_setup(&opts);
-       DIFF_OPT_SET(&opts, RECURSIVE);
-       DIFF_OPT_CLR(&opts, RENAME_EMPTY);
+       opts.flags.recursive = 1;
+       opts.flags.rename_empty = 0;
        opts.detect_rename = DIFF_DETECT_RENAME;
        opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
                            o->diff_rename_limit >= 0 ? o->diff_rename_limit :
@@ -625,7 +666,7 @@ static int remove_file(struct merge_options *o, int clean,
                if (ignore_case) {
                        struct cache_entry *ce;
                        ce = cache_file_exists(path, strlen(path), ignore_case);
-                       if (ce && ce_stage(ce) == 0)
+                       if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
                                return 0;
                }
                if (remove_path(path))
@@ -646,6 +687,7 @@ static void add_flattened_path(struct strbuf *out, const char *s)
 
 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
 {
+       struct path_hashmap_entry *entry;
        struct strbuf newpath = STRBUF_INIT;
        int suffix = 0;
        size_t base_len;
@@ -654,14 +696,16 @@ static char *unique_path(struct merge_options *o, const char *path, const char *
        add_flattened_path(&newpath, branch);
 
        base_len = newpath.len;
-       while (string_list_has_string(&o->current_file_set, newpath.buf) ||
-              string_list_has_string(&o->current_directory_set, newpath.buf) ||
+       while (hashmap_get_from_hash(&o->current_file_dir_set,
+                                    path_hash(newpath.buf), newpath.buf) ||
               (!o->call_depth && file_exists(newpath.buf))) {
                strbuf_setlen(&newpath, base_len);
                strbuf_addf(&newpath, "_%d", suffix++);
        }
 
-       string_list_insert(&o->current_file_set, newpath.buf);
+       FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
+       hashmap_entry_init(entry, path_hash(entry->path));
+       hashmap_add(&o->current_file_dir_set, entry);
        return strbuf_detach(&newpath, NULL);
 }
 
@@ -799,7 +843,7 @@ static int update_file_flags(struct merge_options *o,
                        goto update_index;
                }
 
-               buf = read_sha1_file(oid->hash, &type, &size);
+               buf = read_object_file(oid, &type, &size);
                if (!buf)
                        return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
                if (type != OBJ_BLOB) {
@@ -934,6 +978,185 @@ static int merge_3way(struct merge_options *o,
        return merge_status;
 }
 
+static int find_first_merges(struct object_array *result, const char *path,
+               struct commit *a, struct commit *b)
+{
+       int i, j;
+       struct object_array merges = OBJECT_ARRAY_INIT;
+       struct commit *commit;
+       int contains_another;
+
+       char merged_revision[42];
+       const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
+                                  "--all", merged_revision, NULL };
+       struct rev_info revs;
+       struct setup_revision_opt rev_opts;
+
+       memset(result, 0, sizeof(struct object_array));
+       memset(&rev_opts, 0, sizeof(rev_opts));
+
+       /* get all revisions that merge commit a */
+       xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
+                       oid_to_hex(&a->object.oid));
+       init_revisions(&revs, NULL);
+       rev_opts.submodule = path;
+       /* FIXME: can't handle linked worktrees in submodules yet */
+       revs.single_worktree = path != NULL;
+       setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
+
+       /* save all revisions from the above list that contain b */
+       if (prepare_revision_walk(&revs))
+               die("revision walk setup failed");
+       while ((commit = get_revision(&revs)) != NULL) {
+               struct object *o = &(commit->object);
+               if (in_merge_bases(b, commit))
+                       add_object_array(o, NULL, &merges);
+       }
+       reset_revision_walk();
+
+       /* Now we've got all merges that contain a and b. Prune all
+        * merges that contain another found merge and save them in
+        * result.
+        */
+       for (i = 0; i < merges.nr; i++) {
+               struct commit *m1 = (struct commit *) merges.objects[i].item;
+
+               contains_another = 0;
+               for (j = 0; j < merges.nr; j++) {
+                       struct commit *m2 = (struct commit *) merges.objects[j].item;
+                       if (i != j && in_merge_bases(m2, m1)) {
+                               contains_another = 1;
+                               break;
+                       }
+               }
+
+               if (!contains_another)
+                       add_object_array(merges.objects[i].item, NULL, result);
+       }
+
+       object_array_clear(&merges);
+       return result->nr;
+}
+
+static void print_commit(struct commit *commit)
+{
+       struct strbuf sb = STRBUF_INIT;
+       struct pretty_print_context ctx = {0};
+       ctx.date_mode.type = DATE_NORMAL;
+       format_commit_message(commit, " %h: %m %s", &sb, &ctx);
+       fprintf(stderr, "%s\n", sb.buf);
+       strbuf_release(&sb);
+}
+
+static int merge_submodule(struct merge_options *o,
+                          struct object_id *result, const char *path,
+                          const struct object_id *base, const struct object_id *a,
+                          const struct object_id *b)
+{
+       struct commit *commit_base, *commit_a, *commit_b;
+       int parent_count;
+       struct object_array merges;
+
+       int i;
+       int search = !o->call_depth;
+
+       /* store a in result in case we fail */
+       oidcpy(result, a);
+
+       /* we can not handle deletion conflicts */
+       if (is_null_oid(base))
+               return 0;
+       if (is_null_oid(a))
+               return 0;
+       if (is_null_oid(b))
+               return 0;
+
+       if (add_submodule_odb(path)) {
+               output(o, 1, _("Failed to merge submodule %s (not checked out)"), path);
+               return 0;
+       }
+
+       if (!(commit_base = lookup_commit_reference(base)) ||
+           !(commit_a = lookup_commit_reference(a)) ||
+           !(commit_b = lookup_commit_reference(b))) {
+               output(o, 1, _("Failed to merge submodule %s (commits not present)"), path);
+               return 0;
+       }
+
+       /* check whether both changes are forward */
+       if (!in_merge_bases(commit_base, commit_a) ||
+           !in_merge_bases(commit_base, commit_b)) {
+               output(o, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
+               return 0;
+       }
+
+       /* Case #1: a is contained in b or vice versa */
+       if (in_merge_bases(commit_a, commit_b)) {
+               oidcpy(result, b);
+               if (show(o, 3)) {
+                       output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
+                       output_commit_title(o, commit_b);
+               } else if (show(o, 2))
+                       output(o, 2, _("Fast-forwarding submodule %s to %s"), path, oid_to_hex(b));
+               else
+                       ; /* no output */
+
+               return 1;
+       }
+       if (in_merge_bases(commit_b, commit_a)) {
+               oidcpy(result, a);
+               if (show(o, 3)) {
+                       output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
+                       output_commit_title(o, commit_a);
+               } else if (show(o, 2))
+                       output(o, 2, _("Fast-forwarding submodule %s to %s"), path, oid_to_hex(a));
+               else
+                       ; /* no output */
+
+               return 1;
+       }
+
+       /*
+        * Case #2: There are one or more merges that contain a and b in
+        * the submodule. If there is only one, then present it as a
+        * suggestion to the user, but leave it marked unmerged so the
+        * user needs to confirm the resolution.
+        */
+
+       /* Skip the search if makes no sense to the calling context.  */
+       if (!search)
+               return 0;
+
+       /* find commit which merges them */
+       parent_count = find_first_merges(&merges, path, commit_a, commit_b);
+       switch (parent_count) {
+       case 0:
+               output(o, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
+               break;
+
+       case 1:
+               output(o, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
+               output(o, 2, _("Found a possible merge resolution for the submodule:\n"));
+               print_commit((struct commit *) merges.objects[0].item);
+               output(o, 2, _(
+                       "If this is correct simply add it to the index "
+                       "for example\n"
+                       "by using:\n\n"
+                       "  git update-index --cacheinfo 160000 %s \"%s\"\n\n"
+                       "which will accept this suggestion.\n"),
+                       oid_to_hex(&merges.objects[0].item->oid), path);
+               break;
+
+       default:
+               output(o, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
+               for (i = 0; i < merges.nr; i++)
+                       print_commit((struct commit *) merges.objects[i].item);
+       }
+
+       object_array_clear(&merges);
+       return 0;
+}
+
 static int merge_file_1(struct merge_options *o,
                                           const struct diff_filespec *one,
                                           const struct diff_filespec *a,
@@ -985,8 +1208,9 @@ static int merge_file_1(struct merge_options *o,
                        if ((merge_status < 0) || !result_buf.ptr)
                                ret = err(o, _("Failed to execute internal merge"));
 
-                       if (!ret && write_sha1_file(result_buf.ptr, result_buf.size,
-                                                   blob_type, result->oid.hash))
+                       if (!ret &&
+                           write_object_file(result_buf.ptr, result_buf.size,
+                                             blob_type, &result->oid))
                                ret = err(o, _("Unable to add %s to database"),
                                          a->path);
 
@@ -995,12 +1219,11 @@ static int merge_file_1(struct merge_options *o,
                                return ret;
                        result->clean = (merge_status == 0);
                } else if (S_ISGITLINK(a->mode)) {
-                       result->clean = merge_submodule(&result->oid,
+                       result->clean = merge_submodule(o, &result->oid,
                                                       one->path,
                                                       &one->oid,
                                                       &a->oid,
-                                                      &b->oid,
-                                                      !o->call_depth);
+                                                      &b->oid);
                } else if (S_ISLNK(a->mode)) {
                        switch (o->recursive_variant) {
                        case MERGE_RECURSIVE_NORMAL:
@@ -1612,7 +1835,7 @@ static int read_oid_strbuf(struct merge_options *o,
        void *buf;
        enum object_type type;
        unsigned long size;
-       buf = read_sha1_file(oid->hash, &type, &size);
+       buf = read_object_file(oid, &type, &size);
        if (!buf)
                return err(o, _("cannot read object %s"), oid_to_hex(oid));
        if (type != OBJ_BLOB) {
@@ -1886,8 +2109,9 @@ static int process_entry(struct merge_options *o,
                        oid = b_oid;
                        conf = _("directory/file");
                }
-               if (dir_in_way(path, !o->call_depth,
-                              S_ISGITLINK(a_mode))) {
+               if (dir_in_way(path,
+                              !o->call_depth && !S_ISGITLINK(a_mode),
+                              0)) {
                        char *new_path = unique_path(o, path, add_branch);
                        clean_merge = 0;
                        output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
@@ -1936,7 +2160,14 @@ int merge_trees(struct merge_options *o,
        }
 
        if (oid_eq(&common->object.oid, &merge->object.oid)) {
-               output(o, 0, _("Already up-to-date!"));
+               struct strbuf sb = STRBUF_INIT;
+
+               if (!o->call_depth && index_has_changes(&sb)) {
+                       err(o, _("Dirty index: cannot merge (dirty: %s)"),
+                           sb.buf);
+                       return 0;
+               }
+               output(o, 0, _("Already up to date!"));
                *result = head;
                return 1;
        }
@@ -1954,18 +2185,24 @@ int merge_trees(struct merge_options *o,
        if (unmerged_cache()) {
                struct string_list *entries, *re_head, *re_merge;
                int i;
-               string_list_clear(&o->current_file_set, 1);
-               string_list_clear(&o->current_directory_set, 1);
+               /*
+                * Only need the hashmap while processing entries, so
+                * initialize it here and free it when we are done running
+                * through the entries. Keeping it in the merge_options as
+                * opposed to decaring a local hashmap is for convenience
+                * so that we don't have to pass it to around.
+                */
+               hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
                get_files_dirs(o, head);
                get_files_dirs(o, merge);
 
                entries = get_unmerged();
-               record_df_conflict_files(o, entries);
                re_head  = get_renames(o, head, common, head, merge, entries);
                re_merge = get_renames(o, merge, common, head, merge, entries);
                clean = process_renames(o, re_head, re_merge);
+               record_df_conflict_files(o, entries);
                if (clean < 0)
-                       return clean;
+                       goto cleanup;
                for (i = entries->nr-1; 0 <= i; i--) {
                        const char *path = entries->items[i].string;
                        struct stage_data *e = entries->items[i].util;
@@ -1973,8 +2210,10 @@ int merge_trees(struct merge_options *o,
                                int ret = process_entry(o, path, e);
                                if (!ret)
                                        clean = 0;
-                               else if (ret < 0)
-                                       return ret;
+                               else if (ret < 0) {
+                                       clean = ret;
+                                       goto cleanup;
+                               }
                        }
                }
                for (i = 0; i < entries->nr; i++) {
@@ -1984,13 +2223,19 @@ int merge_trees(struct merge_options *o,
                                    entries->items[i].string);
                }
 
+cleanup:
                string_list_clear(re_merge, 0);
                string_list_clear(re_head, 0);
                string_list_clear(entries, 1);
 
+               hashmap_free(&o->current_file_dir_set, 1);
+
                free(re_merge);
                free(re_head);
                free(entries);
+
+               if (clean < 0)
+                       return clean;
        }
        else
                clean = 1;
@@ -2024,7 +2269,7 @@ int merge_recursive(struct merge_options *o,
 {
        struct commit_list *iter;
        struct commit *merged_common_ancestors;
-       struct tree *mrtree = mrtree;
+       struct tree *mrtree;
        int clean;
 
        if (show(o, 4)) {
@@ -2052,7 +2297,7 @@ int merge_recursive(struct merge_options *o,
                /* if there is no common ancestor, use an empty tree */
                struct tree *tree;
 
-               tree = lookup_tree(&empty_tree_oid);
+               tree = lookup_tree(the_hash_algo->empty_tree);
                merged_common_ancestors = make_virtual_commit(tree, "ancestor");
        }
 
@@ -2133,7 +2378,7 @@ int merge_recursive_generic(struct merge_options *o,
                            struct commit **result)
 {
        int clean;
-       struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
+       struct lock_file lock = LOCK_INIT;
        struct commit *head_commit = get_ref(head, o->branch1);
        struct commit *next_commit = get_ref(merge, o->branch2);
        struct commit_list *ca = NULL;
@@ -2149,14 +2394,16 @@ int merge_recursive_generic(struct merge_options *o,
                }
        }
 
-       hold_locked_index(lock, LOCK_DIE_ON_ERROR);
+       hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
        clean = merge_recursive(o, head_commit, next_commit, ca,
                        result);
-       if (clean < 0)
+       if (clean < 0) {
+               rollback_lock_file(&lock);
                return clean;
+       }
 
-       if (active_cache_changed &&
-           write_locked_index(&the_index, lock, COMMIT_LOCK))
+       if (write_locked_index(&the_index, &lock,
+                              COMMIT_LOCK | SKIP_IF_UNCHANGED))
                return err(o, _("Unable to write index."));
 
        return clean ? 0 : 1;
@@ -2172,6 +2419,7 @@ static void merge_recursive_config(struct merge_options *o)
 
 void init_merge_options(struct merge_options *o)
 {
+       const char *merge_verbosity;
        memset(o, 0, sizeof(struct merge_options));
        o->verbosity = 2;
        o->buffer_output = 1;
@@ -2180,14 +2428,12 @@ void init_merge_options(struct merge_options *o)
        o->renormalize = 0;
        o->detect_rename = 1;
        merge_recursive_config(o);
-       if (getenv("GIT_MERGE_VERBOSITY"))
-               o->verbosity =
-                       strtol(getenv("GIT_MERGE_VERBOSITY"), NULL, 10);
+       merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
+       if (merge_verbosity)
+               o->verbosity = strtol(merge_verbosity, NULL, 10);
        if (o->verbosity >= 5)
                o->buffer_output = 0;
        strbuf_init(&o->obuf, 0);
-       string_list_init(&o->current_file_set, 1);
-       string_list_init(&o->current_directory_set, 1);
        string_list_init(&o->df_conflict_file_set, 1);
 }
 
@@ -2224,6 +2470,8 @@ int parse_merge_opt(struct merge_options *o, const char *s)
                DIFF_XDL_SET(o, IGNORE_WHITESPACE);
        else if (!strcmp(s, "ignore-space-at-eol"))
                DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
+       else if (!strcmp(s, "ignore-cr-at-eol"))
+               DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
        else if (!strcmp(s, "renormalize"))
                o->renormalize = 1;
        else if (!strcmp(s, "no-renormalize"))