Re: [RFC WIP PATCH 1/3] tag: factor out tag reading from write_tag_body()

Lucas Oshiro [off-list ref] writes:

+/* 
+ * Returns the tag body of the given oid or NULL, in case of error. If size is
+ * not NULL it is assigned the body size in bytes (excluding the '\0').
+ */
+static char *get_tag_body(const struct object_id *oid, size_t *size) 
 {
+	unsigned long buf_size;
 	enum object_type type;
+	char *buf, *sp, *tag_body;
+	size_t tag_body_size, signature_offset;
 
+	buf = read_object_file(oid, &type, &buf_size);
 	if (!buf)
+		return NULL;
 	/* skip header */
 	sp = strstr(buf, "\n\n");
 
+	if (!sp || !buf_size || type != OBJ_TAG) {
 		free(buf);
+		return NULL;
 	}

Returning early when !buf_size before even attempting to strstr
would be cleaner to read, i.e.

	buf = read_object_file(...);
	if (!buf || !buf_size) {
		free(buf);
		return NULL;
	}
	body = strstr(buf, "\n\n");

FWIW, the type check that is done after this point could also be a
part of the early return, as there is no point scanning for the end
of object header part if the object is not a tag (e.g. if it were a
blob, there is no "header part" and scanning for a blank line is
meaningless).
	

 	sp += 2; /* skip the 2 LFs */
+	signature_offset = parse_signature(sp, buf + buf_size - sp);
+	sp[signature_offset] = '\0';
 
+	/* detach sp from buf */
+	tag_body_size = strlen(sp) + 1;
+	tag_body = xmalloc(tag_body_size);
+	xsnprintf(tag_body, tag_body_size, "%s", sp);

Isn't this essentially

	tag_body = xstrdup(sp);
        tag_body_size = signature_offset;

(my arith may be off by one or two, but does a separate
tag_body_size need to exist?)

 	free(buf);
+	if (size)
+		*size = tag_body_size - 1; /* exclude '\0' */
+	return tag_body;
+}
+
+static void write_tag_body(int fd, const struct object_id *oid)
+{
+	size_t size;
+	const char *tag_body = get_tag_body(oid, &size);
+
+	if (!tag_body) {
+		warning("failed to get tag body for %s", oid->hash);

I do not think the original gives any such warning.

 - Do we want to be unconditionally noisy this way?
 - Should this be a fatal error?  If not, why?
 - Should the message be translatable?

As an interface, is it sensible to force any and all callers of
get_tag_body() to supply a pointer to &size?  Is the returned value
always a NUL-terminated string?  I suspect that people would find it
a more natural interface if its were like:

	const char *body = get_tag_body(oid);

	if (!body)		
		...;

	if (this caller needs size) {
		size_t body_size = strlen(body);
		... use both body and body_size ...
		write_or_die(fd, body, body_size);
	} else {
		... just use body ...
		printf("%s", body);
	}
	

+		return;
+	}
+	printf("tag_body: <%s>\n", tag_body);
+	write_or_die(fd, tag_body, size);

WTH is this double writing?

 }
 
 static int build_tag_object(struct strbuf *buf, int sign, struct object_id *result)