fix: set Type field on INSERT parameter columns for correct schema resolution

[rayakame]

I'm writing an external sqlc plugin (sqlc-gen-better-python) and ran into a bug where :copyfrom queries targeting tables in custom schemas send incorrect type information to plugins.

When using a schema like this:

CREATE SCHEMA IF NOT EXISTS custom;
CREATE TYPE custom.mood AS ENUM ('sad', 'ok', 'happy');
CREATE TABLE custom.test_enum (
    id int PRIMARY KEY NOT NULL,
    m  custom.mood NOT NULL
);

And a query like:

-- name: InsertStuff :copyfrom
INSERT INTO custom.test_enum (id, m) VALUES ($1, $2);

The plugin receives the parameter type as name: "custom.mood" instead of schema: "custom", name: "mood". This meant my plugin couldn't resolve the enum type to its definition in the catalog, since the schema was baked into the name string rather than being properly separated.

I noticed that :many SELECT queries on the exact same table returned correct type info, so the bug had to be specific to how INSERT parameter columns are built.

Cause

The *ast.ResTarget case in resolveCatalogRefs (internal/compiler/resolve.go) builds Column structs with DataType (a string like "custom.mood") but doesn't set the structured Type field (*ast.TypeName with separate Schema and Name). The shim in internal/cmd/shim.go checks Type first and falls back to putting DataType into Name with an empty schema when Type is nil.

Previous fix / possible unnecessary code

In /internal/codegen/golang/postgresql_type.go there is the function func parseIdentifierString(name string) (*plugin.Identifier, error) which previously parsed the string generated by sdk.DataType into a new *plugin.Identifier. As far as I've tested this, this function can just be removed now and the column.Type can be used directly.

I could of course implement the same workaround in my own plugin, but I think fixing this at the source is the right approach.

[rayakame]

As far as I can see the failing test is only something that went wrong with docker