The implementation of FULLTEXT INDEX in the InnoDB storage engine is problematic in many ways. It uses hard-coded partitioning of the inverted index, and fulltext indexes use internal InnoDB tables that are in the same namespace with user tables. The FULLTEXT INDEX code in InnoDB makes use of the InnoDB-internal SQL parser and depends on the InnoDB internal data dictionary tables, both of which we would like to remove.
We currently do not support FULLTEXT INDEX for MyRocks (ENGINE=RocksDB) either.
Because some InnoDB users do need fulltext search functionality, we should develop a replacement before we can remove the old code. Perhaps a replacement could be implemented in the style of a ‘virtual’ or ‘wrapper’ storage engine, similar to ha_partition? We might store the inverted indexes for the fulltext search separately from the main storage engine, and use distributed transactions between the two.