branches/zip: row_build_index_entry(): Fix errors introduced in r870.

Do not augment clustered index records.  When computing the prefix,
use dfield (where the prefix was copied), not dfield2 (the record in
the clustered index).

row/row0row.c

@@ -87,9 +87,16 @@ row_build_index_entry(
 
 	if (index->type & DICT_UNIVERSAL) {
 		dtuple_set_n_fields_cmp(entry, entry_len);
+		ut_a(!ext);
 	} else {
 		dtuple_set_n_fields_cmp(
 			entry, dict_index_get_n_unique_in_tree(index));
+		if (dict_index_is_clust(index)) {
+			/* Do not fetch externally stored columns to
+			the clustered index.  Such columns are handled
+			at a higher level. */
+			ext = NULL;
+		}
 	}
 
 	for (i = 0; i < entry_len; i++) {
@@ -119,14 +126,13 @@ row_build_index_entry(
 
 		/* If a column prefix index, take only the prefix */
 		if (ind_field->prefix_len) {
-			if (dfield_get_len(dfield2) != UNIV_SQL_NULL) {
+			if (dfield_get_len(dfield) != UNIV_SQL_NULL) {
 
 				storage_len = dtype_get_at_most_n_mbchars(
 					col->prtype,
 					col->mbminlen, col->mbmaxlen,
 					ind_field->prefix_len,
-					dfield_get_len(dfield2),
-					dfield2->data);
+					dfield->len, dfield->data);
 
 				dfield_set_len(dfield, storage_len);
 			}