I’ve thought about adding annotations to (the internal representation of) Calcite types. We never got very far with it. It would be tricky because we canonize types, and so if you have two instances that represent, say, DECIMAL(10, 3) with different annotations, are they the same type?
One approach is to add a new kind of metadata. It can propagate through the RelNode graph in parallel with the type information. (Constraints already do this.) Julian > On Aug 19, 2022, at 7:54 AM, Dmitry Sysolyatin <dm.sysolya...@gmail.com> > wrote: > > Hello! > I am working on emulating a Postgres server and am faced with the following > problem: > > Postgres `pg_catalog` schema has custom types like `name`, those types can > be easily mapped to calcite basic types. The problem is that several types > in PG can be mapped to one calcite type (for example: `name` and `varchar` > in PG are mapped to `varchar` type in calcite) but postgres has a unique id > for every type. > Some PG clients like Power BI require the right id for the column. > Planner.validateAndGetType() returns row data type for the query, but I can > not derive pg type id from this data type, because calcite data type can > correspond to several PG types. > > One solution would be putting some extra information to RelDataType when > the table structure is defined and somehow extracting it when the row type > for query is inferred. > > I will be very grateful for your help
