duckdb: table functions such as read_parquet, read_csv and read_json are unknown #4659

What do you want to change?

A query that reads a file with a DuckDB table function fails with unknown function.
This is a very common way to use DuckDB.
The function list in internal/engine/duckdb/dialect/functions.jsonl has no entry for these table functions.
Their columns depend on the file, so sqlc cannot know them.

Request, one of these or another idea:

• a way to declare the columns of a table function, in the config or in the schema file, or
• documentation that shows the recommended pattern (for example, a stand-in table with the same columns).

Output:

== FROM read_parquet('items.parquet')
# package repro
query.sql:1:1: unknown function "read_parquet"
sqlc exit code: 1
== FROM read_csv('items.csv')
# package repro
query.sql:1:1: unknown function "read_csv"
sqlc exit code: 1
== FROM read_json('items.json')
# package repro
query.sql:1:1: unknown function "read_json"
sqlc exit code: 1

#!/usr/bin/env bash
rm -rf repro5 && mkdir repro5 && cd repro5
cat > sqlc.yaml <<'EOF'
version: "2"
sql:
  - engine: "duckdb"
    schema: "schema.sql"
    queries: "query.sql"
    gen:
      go:
        package: "repro"
        out: "out"
EOF
cat > schema.sql <<'EOF'
CREATE TABLE items (id INTEGER NOT NULL, name VARCHAR NOT NULL);
EOF
for fn in "read_parquet('items.parquet')" "read_csv('items.csv')" "read_json('items.json')"; do
  printf -- "-- name: Q :many\nSELECT id, name FROM %s;\n" "$fn" > query.sql
  echo "== FROM $fn"
  sqlc generate 2>&1 | head -2; echo "sqlc exit code: ${PIPESTATUS[0]}"
done

No patch: the right design is a decision for the maintainers.

What database engines need to be changed?

(none of the listed options)

What programming language backends need to be changed?

Go