org.apache.hadoop.hive.ql.exec.DDLTask handles DDL operations in Hive.
Some operations (those exposed as REST API by Templeton) in this class, write a JSON object to stdout (on failure) which includes "error" and "errorCode" fields. "errorCode" is set to values like 404, 500, etc. i.e. Http Status codes. I believe this was done to support Templeton REST API. This is not the right place to introduce Http codes.
DDLTask should use canonical error messages defined in org.apache.hadoop.hive.ql.ErrorMsg and propagate it up to HCatCli which can decide how to encode the error and to which stream. In particular it can create a parseable JSON object which Templeton can then interpret. JSON object can also be generated here for errors from the parser/resolver (e.g. trying to describe a table which doesn't exist).
Then Templeton can map the canonical error code to appropriate Http Code.