double val= je->s.cs->strntod((char *) je->value, je->value_len, &end; You do this even if the value_type is not number. Not very safe. Better to avoid doing

Hi, Rucha. I decided i dislike the overall design of the struct json_schema. Seems better to do it with C++ classes using virtual methods. So instead of structures we'd have classes as json_schema_const json_schema_enum json_scnema_string json_schema_number etc. they all inherited from the class json_schema_item { virtual parse() virtual check() } Parsing the schema we can behave like this: json_parse_schema() checks all the keys on the current level looking for 'const' 'enum' and 'typ[e' keywords only. If it gets 'const' or 'enum' it creates the json_schema_const/enum respectively For the "type" it creates the appropriate json_schema_xxx and callst ->parse() with the same JSON object. So the parse would go over it once again looking for the type-specific keywords. That way we don't need the hash_list as the destructor of the json_schema_xxx class would free it's resources. Minor comments. You compare keys like strncmp((const char*)curr_key, "const", key_len) all over the code. But it is not correct. That way 'cons' or 'con' or 'co' or even 'c' - they all will be considered equal to the 'const'. Just add the 'key_cmp()' function that would compare key lengths then do the strncmp. that here at all. That queriy just crashes the server. Didn't yet found why SET @schema_array= '{ "typ": "arra", "prefixItem": [ { "typ": "objec"}, { "typ": "objec", "propertie": { "number1": {"typ":"numbe"}, "string1": {"typ":"strin"}, "obj1" : {"typ":"objec", "propertie": { "key1" : {"type":"string"} } } }, "required":["obj1"] } ], "items":false }'; SELECT JSON_SCHEMA_VALID(@schema_array, '[{"key1":"value1"}, {"number1":1, "string1":"str1", "obj1":{"key1":"val1"}}]'); Feel free to discuss the design details! Best regards. HF