t4201: make use of abbreviation in the test more robust
[gitweb.git] / builtin / cat-file.c
index 54db1184a003897294b6aaa46bc6bf29faa919bc..30383e9eb4befb31e0d165d5863615484bcc9da3 100644 (file)
@@ -17,32 +17,63 @@ struct batch_options {
        int print_contents;
        int buffer_output;
        int all_objects;
+       int cmdmode; /* may be 'w' or 'c' for --filters or --textconv */
        const char *format;
 };
 
+static const char *force_path;
+
+static int filter_object(const char *path, unsigned mode,
+                        const struct object_id *oid,
+                        char **buf, unsigned long *size)
+{
+       enum object_type type;
+
+       *buf = read_sha1_file(oid->hash, &type, size);
+       if (!*buf)
+               return error(_("cannot read object %s '%s'"),
+                            oid_to_hex(oid), path);
+       if ((type == OBJ_BLOB) && S_ISREG(mode)) {
+               struct strbuf strbuf = STRBUF_INIT;
+               if (convert_to_working_tree(path, *buf, *size, &strbuf)) {
+                       free(*buf);
+                       *size = strbuf.len;
+                       *buf = strbuf_detach(&strbuf, NULL);
+               }
+       }
+
+       return 0;
+}
+
 static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
                        int unknown_type)
 {
-       unsigned char sha1[20];
+       struct object_id oid;
        enum object_type type;
        char *buf;
        unsigned long size;
        struct object_context obj_context;
-       struct object_info oi = {NULL};
+       struct object_info oi = OBJECT_INFO_INIT;
        struct strbuf sb = STRBUF_INIT;
        unsigned flags = LOOKUP_REPLACE_OBJECT;
+       const char *path = force_path;
 
        if (unknown_type)
                flags |= LOOKUP_UNKNOWN_OBJECT;
 
-       if (get_sha1_with_context(obj_name, 0, sha1, &obj_context))
+       if (get_sha1_with_context(obj_name, 0, oid.hash, &obj_context))
                die("Not a valid object name %s", obj_name);
 
+       if (!path)
+               path = obj_context.path;
+       if (obj_context.mode == S_IFINVALID)
+               obj_context.mode = 0100644;
+
        buf = NULL;
        switch (opt) {
        case 't':
                oi.typename = &sb;
-               if (sha1_object_info_extended(sha1, &oi, flags) < 0)
+               if (sha1_object_info_extended(oid.hash, &oi, flags) < 0)
                        die("git cat-file: could not get object info");
                if (sb.len) {
                        printf("%s\n", sb.buf);
@@ -53,24 +84,34 @@ static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
 
        case 's':
                oi.sizep = &size;
-               if (sha1_object_info_extended(sha1, &oi, flags) < 0)
+               if (sha1_object_info_extended(oid.hash, &oi, flags) < 0)
                        die("git cat-file: could not get object info");
                printf("%lu\n", size);
                return 0;
 
        case 'e':
-               return !has_sha1_file(sha1);
+               return !has_object_file(&oid);
+
+       case 'w':
+               if (!path[0])
+                       die("git cat-file --filters %s: <object> must be "
+                           "<sha1:path>", obj_name);
+
+               if (filter_object(path, obj_context.mode,
+                                 &oid, &buf, &size))
+                       return -1;
+               break;
 
        case 'c':
-               if (!obj_context.path[0])
+               if (!path[0])
                        die("git cat-file --textconv %s: <object> must be <sha1:path>",
                            obj_name);
 
-               if (textconv_object(obj_context.path, obj_context.mode, sha1, 1, &buf, &size))
+               if (textconv_object(path, obj_context.mode, &oid, 1, &buf, &size))
                        break;
 
        case 'p':
-               type = sha1_object_info(sha1, NULL);
+               type = sha1_object_info(oid.hash, NULL);
                if (type < 0)
                        die("Not a valid object name %s", obj_name);
 
@@ -83,8 +124,8 @@ static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
                }
 
                if (type == OBJ_BLOB)
-                       return stream_blob_to_fd(1, sha1, NULL, 0);
-               buf = read_sha1_file(sha1, &type, &size);
+                       return stream_blob_to_fd(1, &oid, NULL, 0);
+               buf = read_sha1_file(oid.hash, &type, &size);
                if (!buf)
                        die("Cannot read object %s", obj_name);
 
@@ -93,19 +134,19 @@ static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
 
        case 0:
                if (type_from_string(exp_type) == OBJ_BLOB) {
-                       unsigned char blob_sha1[20];
-                       if (sha1_object_info(sha1, NULL) == OBJ_TAG) {
-                               char *buffer = read_sha1_file(sha1, &type, &size);
+                       struct object_id blob_oid;
+                       if (sha1_object_info(oid.hash, NULL) == OBJ_TAG) {
+                               char *buffer = read_sha1_file(oid.hash, &type, &size);
                                const char *target;
                                if (!skip_prefix(buffer, "object ", &target) ||
-                                   get_sha1_hex(target, blob_sha1))
-                                       die("%s not a valid tag", sha1_to_hex(sha1));
+                                   get_oid_hex(target, &blob_oid))
+                                       die("%s not a valid tag", oid_to_hex(&oid));
                                free(buffer);
                        } else
-                               hashcpy(blob_sha1, sha1);
+                               oidcpy(&blob_oid, &oid);
 
-                       if (sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
-                               return stream_blob_to_fd(1, blob_sha1, NULL, 0);
+                       if (sha1_object_info(blob_oid.hash, NULL) == OBJ_BLOB)
+                               return stream_blob_to_fd(1, &blob_oid, NULL, 0);
                        /*
                         * we attempted to dereference a tag to a blob
                         * and failed; there may be new dereference
@@ -113,7 +154,7 @@ static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
                         * fall-back to the usual case.
                         */
                }
-               buf = read_object_with_reference(sha1, exp_type, &size, NULL);
+               buf = read_object_with_reference(oid.hash, exp_type, &size, NULL);
                break;
 
        default:
@@ -128,12 +169,12 @@ static int cat_one_file(int opt, const char *exp_type, const char *obj_name,
 }
 
 struct expand_data {
-       unsigned char sha1[20];
+       struct object_id oid;
        enum object_type type;
        unsigned long size;
-       unsigned long disk_size;
+       off_t disk_size;
        const char *rest;
-       unsigned char delta_base_sha1[20];
+       struct object_id delta_base_oid;
 
        /*
         * If mark_query is true, we do not expand anything, but rather
@@ -154,6 +195,13 @@ struct expand_data {
         * elements above, so you can retrieve the response from there.
         */
        struct object_info info;
+
+       /*
+        * This flag will be true if the requested batch format and options
+        * don't require us to call sha1_object_info, which can then be
+        * optimized out.
+        */
+       unsigned skip_object_info : 1;
 };
 
 static int is_atom(const char *atom, const char *s, int slen)
@@ -169,7 +217,7 @@ static void expand_atom(struct strbuf *sb, const char *atom, int len,
 
        if (is_atom("objectname", atom, len)) {
                if (!data->mark_query)
-                       strbuf_addstr(sb, sha1_to_hex(data->sha1));
+                       strbuf_addstr(sb, oid_to_hex(&data->oid));
        } else if (is_atom("objecttype", atom, len)) {
                if (data->mark_query)
                        data->info.typep = &data->type;
@@ -184,7 +232,7 @@ static void expand_atom(struct strbuf *sb, const char *atom, int len,
                if (data->mark_query)
                        data->info.disk_sizep = &data->disk_size;
                else
-                       strbuf_addf(sb, "%lu", data->disk_size);
+                       strbuf_addf(sb, "%"PRIuMAX, (uintmax_t)data->disk_size);
        } else if (is_atom("rest", atom, len)) {
                if (data->mark_query)
                        data->split_on_whitespace = 1;
@@ -192,9 +240,10 @@ static void expand_atom(struct strbuf *sb, const char *atom, int len,
                        strbuf_addstr(sb, data->rest);
        } else if (is_atom("deltabase", atom, len)) {
                if (data->mark_query)
-                       data->info.delta_base_sha1 = data->delta_base_sha1;
+                       data->info.delta_base_sha1 = data->delta_base_oid.hash;
                else
-                       strbuf_addstr(sb, sha1_to_hex(data->delta_base_sha1));
+                       strbuf_addstr(sb,
+                                     oid_to_hex(&data->delta_base_oid));
        } else
                die("unknown format element: %.*s", len, atom);
 }
@@ -225,28 +274,53 @@ static void batch_write(struct batch_options *opt, const void *data, int len)
 
 static void print_object_or_die(struct batch_options *opt, struct expand_data *data)
 {
-       const unsigned char *sha1 = data->sha1;
+       const struct object_id *oid = &data->oid;
 
        assert(data->info.typep);
 
        if (data->type == OBJ_BLOB) {
                if (opt->buffer_output)
                        fflush(stdout);
-               if (stream_blob_to_fd(1, sha1, NULL, 0) < 0)
-                       die("unable to stream %s to stdout", sha1_to_hex(sha1));
+               if (opt->cmdmode) {
+                       char *contents;
+                       unsigned long size;
+
+                       if (!data->rest)
+                               die("missing path for '%s'", oid_to_hex(oid));
+
+                       if (opt->cmdmode == 'w') {
+                               if (filter_object(data->rest, 0100644, oid,
+                                                 &contents, &size))
+                                       die("could not convert '%s' %s",
+                                           oid_to_hex(oid), data->rest);
+                       } else if (opt->cmdmode == 'c') {
+                               enum object_type type;
+                               if (!textconv_object(data->rest, 0100644, oid,
+                                                    1, &contents, &size))
+                                       contents = read_sha1_file(oid->hash, &type,
+                                                                 &size);
+                               if (!contents)
+                                       die("could not convert '%s' %s",
+                                           oid_to_hex(oid), data->rest);
+                       } else
+                               die("BUG: invalid cmdmode: %c", opt->cmdmode);
+                       batch_write(opt, contents, size);
+                       free(contents);
+               } else if (stream_blob_to_fd(1, oid, NULL, 0) < 0)
+                       die("unable to stream %s to stdout", oid_to_hex(oid));
        }
        else {
                enum object_type type;
                unsigned long size;
                void *contents;
 
-               contents = read_sha1_file(sha1, &type, &size);
+               contents = read_sha1_file(oid->hash, &type, &size);
                if (!contents)
-                       die("object %s disappeared", sha1_to_hex(sha1));
+                       die("object %s disappeared", oid_to_hex(oid));
                if (type != data->type)
-                       die("object %s changed type!?", sha1_to_hex(sha1));
+                       die("object %s changed type!?", oid_to_hex(oid));
                if (data->info.sizep && size != data->size)
-                       die("object %s changed size!?", sha1_to_hex(sha1));
+                       die("object %s changed size!?", oid_to_hex(oid));
 
                batch_write(opt, contents, size);
                free(contents);
@@ -258,8 +332,10 @@ static void batch_object_write(const char *obj_name, struct batch_options *opt,
 {
        struct strbuf buf = STRBUF_INIT;
 
-       if (sha1_object_info_extended(data->sha1, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
-               printf("%s missing\n", obj_name ? obj_name : sha1_to_hex(data->sha1));
+       if (!data->skip_object_info &&
+           sha1_object_info_extended(data->oid.hash, &data->info, LOOKUP_REPLACE_OBJECT) < 0) {
+               printf("%s missing\n",
+                      obj_name ? obj_name : oid_to_hex(&data->oid));
                fflush(stdout);
                return;
        }
@@ -282,7 +358,7 @@ static void batch_one_object(const char *obj_name, struct batch_options *opt,
        int flags = opt->follow_symlinks ? GET_SHA1_FOLLOW_SYMLINKS : 0;
        enum follow_symlinks_result result;
 
-       result = get_sha1_with_context(obj_name, flags, data->sha1, &ctx);
+       result = get_sha1_with_context(obj_name, flags, data->oid.hash, &ctx);
        if (result != FOUND) {
                switch (result) {
                case MISSING_OBJECT:
@@ -325,11 +401,12 @@ struct object_cb_data {
        struct expand_data *expand;
 };
 
-static void batch_object_cb(const unsigned char sha1[20], void *vdata)
+static int batch_object_cb(const unsigned char sha1[20], void *vdata)
 {
        struct object_cb_data *data = vdata;
-       hashcpy(data->expand->sha1, sha1);
+       hashcpy(data->expand->oid.hash, sha1);
        batch_object_write(NULL, data->opt, data->expand);
+       return 0;
 }
 
 static int batch_loose_object(const unsigned char *sha1,
@@ -368,6 +445,14 @@ static int batch_objects(struct batch_options *opt)
        data.mark_query = 1;
        strbuf_expand(&buf, opt->format, expand_format, &data);
        data.mark_query = 0;
+       if (opt->cmdmode)
+               data.split_on_whitespace = 1;
+
+       if (opt->all_objects) {
+               struct object_info empty = OBJECT_INFO_INIT;
+               if (!memcmp(&data.info, &empty, sizeof(empty)))
+                       data.skip_object_info = 1;
+       }
 
        /*
         * If we are printing out the object, then always fill in the type,
@@ -425,8 +510,8 @@ static int batch_objects(struct batch_options *opt)
 }
 
 static const char * const cat_file_usage[] = {
-       N_("git cat-file (-t [--allow-unknown-type]|-s [--allow-unknown-type]|-e|-p|<type>|--textconv) <object>"),
-       N_("git cat-file (--batch | --batch-check) [--follow-symlinks]"),
+       N_("git cat-file (-t [--allow-unknown-type] | -s [--allow-unknown-type] | -e | -p | <type> | --textconv | --filters) [--path=<path>] <object>"),
+       N_("git cat-file (--batch | --batch-check) [--follow-symlinks] [--textconv | --filters]"),
        NULL
 };
 
@@ -471,6 +556,10 @@ int cmd_cat_file(int argc, const char **argv, const char *prefix)
                OPT_CMDMODE('p', NULL, &opt, N_("pretty-print object's content"), 'p'),
                OPT_CMDMODE(0, "textconv", &opt,
                            N_("for blob objects, run textconv on object's content"), 'c'),
+               OPT_CMDMODE(0, "filters", &opt,
+                           N_("for blob objects, run filters on object's content"), 'w'),
+               OPT_STRING(0, "path", &force_path, N_("blob"),
+                          N_("use a specific path for --textconv/--filters")),
                OPT_BOOL(0, "allow-unknown-type", &unknown_type,
                          N_("allow -s and -t to work with broken/corrupt objects")),
                OPT_BOOL(0, "buffer", &batch.buffer_output, N_("buffer --batch output")),
@@ -489,10 +578,13 @@ int cmd_cat_file(int argc, const char **argv, const char *prefix)
 
        git_config(git_cat_file_config, NULL);
 
+       batch.buffer_output = -1;
        argc = parse_options(argc, argv, prefix, options, cat_file_usage, 0);
 
        if (opt) {
-               if (argc == 1)
+               if (batch.enabled && (opt == 'c' || opt == 'w'))
+                       batch.cmdmode = opt;
+               else if (argc == 1)
                        obj_name = argv[0];
                else
                        usage_with_options(cat_file_usage, options);
@@ -504,14 +596,31 @@ int cmd_cat_file(int argc, const char **argv, const char *prefix)
                } else
                        usage_with_options(cat_file_usage, options);
        }
-       if (batch.enabled && (opt || argc)) {
-               usage_with_options(cat_file_usage, options);
+       if (batch.enabled) {
+               if (batch.cmdmode != opt || argc)
+                       usage_with_options(cat_file_usage, options);
+               if (batch.cmdmode && batch.all_objects)
+                       die("--batch-all-objects cannot be combined with "
+                           "--textconv nor with --filters");
        }
 
        if ((batch.follow_symlinks || batch.all_objects) && !batch.enabled) {
                usage_with_options(cat_file_usage, options);
        }
 
+       if (force_path && opt != 'c' && opt != 'w') {
+               error("--path=<path> needs --textconv or --filters");
+               usage_with_options(cat_file_usage, options);
+       }
+
+       if (force_path && batch.enabled) {
+               error("--path=<path> incompatible with --batch");
+               usage_with_options(cat_file_usage, options);
+       }
+
+       if (batch.buffer_output < 0)
+               batch.buffer_output = batch.all_objects;
+
        if (batch.enabled)
                return batch_objects(&batch);