avantgardnerio commented on issue #4683: URL: https://github.com/apache/arrow-datafusion/issues/4683#issuecomment-1360414998
Thanks @alamb for writing up a ticket, and thank you @NGA-TRAN for building out an implementation! My personal backstory is that I'm trying to run https://github.com/AgilData/tpcc which defines a number of prepared statements: https://gist.github.com/avantgardnerio/5bdf77297956add71cbf664d3366224b When the JDBC client calls the server, it doesn't use the `prepare ...` syntax - it just passes the query verbatim to `do_action_create_prepared_statement()` which accepts a request: ``` pub struct ActionCreatePreparedStatementRequest { /// The valid SQL string to create a prepared statement for. #[prost(string, tag = "1")] pub query: ::prost::alloc::string::String, } ``` and returns a response: ``` pub struct ActionCreatePreparedStatementResult { /// Opaque handle for the prepared statement on the server. #[prost(bytes = "vec", tag = "1")] pub prepared_statement_handle: ::prost::alloc::vec::Vec<u8>, /// If a result set generating query was provided, dataset_schema contains the /// schema of the dataset as described in Schema.fbs::Schema, it is serialized as an IPC message. #[prost(bytes = "vec", tag = "2")] pub dataset_schema: ::prost::alloc::vec::Vec<u8>, /// If the query provided contained parameters, parameter_schema contains the /// schema of the expected parameters as described in Schema.fbs::Schema, it is serialized as an IPC message. #[prost(bytes = "vec", tag = "3")] pub parameter_schema: ::prost::alloc::vec::Vec<u8>, } ``` This leads me to believe that in order to be compatible with the FlightSQL JDBC client and apps that use it, the type inference has to happen in the DataFusion planner. That being said, implementation might not be too difficult: based upon the statements in the git gist above, types always seem to be: 1. in a values clause and infer-able from what they are being inserted into 2. or in an assignment clause, and infer-able from the LHS I think it would be possible to take: ``` Placeholder { /// The identifier of the parameter (e.g, $1 or $foo) id: String, /// The type the parameter will be filled in with data_type: DataType, }, ``` and make `data_type: Option<DataType>`. This way the placeholders could serve for a 1st pass, then a second inference pass could walk the tree, infer the types and set them if possible or error if it can't figure them out. Just a suggestion - let me know if I can help out more. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
