Just to clarify this PR is NOT FINISHED YET. I need more clarification of the scope of this issue.

Update:
I think I know what files to change now.

Hi @alamb and @blaginin ,

This PR is mostly done. There are still some areas (non-trivial test cases migration) where I'm not sure if I should or how should I migrate them to insta. For example:

The assert_expected_plan in roundtrip_logical_plan.rs accepts a hard-coded expected_plan_str, and uses assert_eq! internally. I wanted to replace the assert_eq! with assert_snapshot!, but Rust function does not accept insta inline snapshot syntax like @r, meaning that I can't pass a raw string to assert_expected_plan to perform a inline snapshot assertion.

And there are many more functions like this, I can't simply change them as they accepts dynamically generated plan_str and plan.schema(). Changing them into assert_snapshot! with inline snapshot may change their original behaviour.

Just want to know if I need to migrate those non-trivial test cases as well, thanks.

And there are many more functions like this, I can't simply change them as they accepts dynamically generated plan_str and plan.schema(). Changing them into assert_snapshot! with inline snapshot may change their original behaviour.

Just want to know if I need to migrate those non-trivial test cases as well, thanks.

In general I suggest we get the basic easy to port tests done and then work on the others as a follow on PR

For those changes, I think you would have to restructure the tests

From

        let expected_plan_str = format!(
            "Projection: count(Int64(1)) AS {syntax}\
        \n  Aggregate: groupBy=[[]], aggr=[[count(Int64(1))]]\
        \n    Projection: \
        \n      LeftSemi Join: data.a = data2.a\
        \n        Aggregate: groupBy=[[data.a]], aggr=[[]]\
        \n          TableScan: data projection=[a]\
        \n        TableScan: data2 projection=[a]"
        );

        assert_expected_plan(
            &format!("SELECT {syntax} FROM (SELECT data.a FROM data INTERSECT SELECT data2.a FROM data2);"),
            &expected_plan_str,
            true
        ).await

Perhaps something more like

         assert_snapshot!(
        do_conversion(&format!("SELECT {syntax} FROM (SELECT data.a FROM data INTERSECT SELECT data2.a FROM data2);),
        @r#"
            "Projection: count(Int64(1)) AS {syntax}\
        \n  Aggregate: groupBy=[[]], aggr=[[count(Int64(1))]]\
        \n    Projection: \
        \n      LeftSemi Join: data.a = data2.a\
        \n        Aggregate: groupBy=[[data.a]], aggr=[[]]\
        \n          TableScan: data projection=[a]\
        \n        TableScan: data2 projection=[a]"
        );
            "#
                );
        Ok(())

Where do_conversion returns a string and does what assert-epected_plan does, something like

async fn do_conversion(
    sql: &str,
    assert_schema: bool,
) -> Result<String> {
    let ctx = create_context().await?;
    let df = ctx.sql(sql).await?;
    let plan = df.into_optimized_plan()?;
    let proto = to_substrait_plan(&plan, &ctx.state())?;
    let plan2 = from_substrait_plan(&ctx.state(), &proto).await?;
    let plan2 = ctx.state().optimize(&plan2)?;

    if assert_schema {
        assert_eq!(plan.schema(), plan2.schema());
    }

    Ok(format!("{plan2}");)
}

Closes #15398.

If you plan to do more prs, consider changing it to "related", so the ticket won't get auto closed

In general I suggest we get the basic easy to port tests done and then work on the others as a follow on PR

Got it, thanks @alamb and @blaginin for the help.

I think I've finished migrating all the "easy" portion now, please review this PR. The only thing left are the non-trivial cases in roundtrip_logical_plan.rs.

I will create another PR for that as suggested.