:exec statements can generate unused return type struct, leading to compile errors

[DavidS-ovm]

Version

1.23.0

What happened?

The attached playground generates code that can't be compiled because of the missing pgtype import. This is because the ResetTaskRunRow struct is not used in ResetTaskRun, because the query is specified as exec.

Either sqlc should not generate the struct it is not using OR complain about the mismatch between the :exec annotation and the fact that the statement returns data. I would prefer the latter as this is now the second time I've ran into the issue of confusion because I changed the SQL of a statement without fixing the annotation, and each time it's been a royal pain to find.

Relevant log output

No response

Database schema

CREATE TABLE IF NOT EXISTS job
(
    task_name text NOT NULL,
    last_run timestamp with time zone DEFAULT now() NOT NULL
);

SQL queries

-- name: ResetTaskRun :exec
UPDATE job
SET last_run = NOW()
FROM (
        SELECT last_run,
            task_name
        FROM job AS o
        WHERE o.task_name = $1 
  		FOR UPDATE
    ) AS old_values
WHERE job.task_name = $1
RETURNING job.last_run AS this_run,
    old_values.last_run AS last_run;

Configuration

{
    "version": "2",
    "sql": [
        {
            "schema": "schema.sql",
            "queries": "query.sql",
            "engine": "postgresql",
            "gen": {
                "go": {
                    "out": "db",
                    "sql_package": "pgx/v5",
                }
            }
        }
    ]
}

Playground URL

https://play.sqlc.dev/p/3968fbdbe70621e70093ca0efca43a1e3bbb4ad5835d60b57d0167f5397bb0d6

What operating system are you using?

No response

What database engines are you using?

PostgreSQL

What type of code are you generating?

Go

[andrewmbenton]

Either sqlc should not generate the struct it is not using OR complain about the mismatch between the :exec annotation and the fact that the statement returns data.

Agreed. We've been planning to make :exec even more permissive than it already is, so I don't think we'd have sqlc error on an :exec annotation with a RETURNING... clause. But I think it would be reasonable to not generate the unused struct and print a warning.

In the short term, you can probably work around this with the :omit_unused_structs config flag.

[DavidS-ovm]

Either sqlc should not generate the struct it is not using OR complain about the mismatch between the :exec annotation and the fact that the statement returns data.

Agreed. We've been planning to make :exec even more permissive than it already is, so I don't think we'd have sqlc error on an :exec annotation with a RETURNING... clause. But I think it would be reasonable to not generate the unused struct and print a warning.

a warning would be just fine. Anything to warn me off that I'm holding it wrong! Maybe a check for sqlc vet that could be enabled?

In the short term, you can probably work around this with the :omit_unused_structs config flag.

I've already got that set, no dice.