refactor: Old Planner Never See Again (Part X)

[Xuanwo]

I hereby agree to the terms of the CLA available at: https://databend.rs/dev/policies/cla/

Summary

This PR is trying to remove the ExprParser.

After this PR, our storage will don't depend on sqlparser anymore.

This PR introduces a dangerous function, as I said in comment:

/// unchecked_expressions_analyze will analyzer given expr str into `Expression`.
///
/// `unchecked` means there is no correction checks for expr:
///
/// - We don't if the expr is **safe**.
/// - We don't if the expr is **existed**.
/// - We don't if the expr is **valid**.
///
/// We just convert the given expr into Expression, and hoping everything works.
///
/// This function should only be used by very limited use-cases like fuse engines
/// `cluster_keys` which already checked by our parsers.
///
/// **FBI WARNING: DON'T USE THIS FUNCTION UNLESS YOU KNOW WHAT I MEAN**
///
/// Any usage to this function should be reviewed carefully.
///
/// # Future Plan
///
/// This function should be removed once our new expression framework is ready.
pub fn unchecked_expressions_analyze(expr: &str) -> Result<Vec<Expression>> { ... }

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated
databend	✅ Ready (Inspect)	Visit Preview	Sep 17, 2022 at 1:39PM (UTC)

[Xuanwo]

unchecked_expressions_analyze can't parse complex cluster key, I'm seeking new ways to address this problem.

[Xuanwo]

@mergify update

[mergify]

update

✅ Branch has been successfully updated

[Xuanwo]

This PR is blocked by Planner.

Now, planner is implemented inside query/service/src/sql/planner, and It depends on types like crate::evaluator::Evaluator to parse DataBlock at bind time (aka, ValueSourceV2). Thus, we can't move Planner out to be a standalone crate common-planner.

However, common-storages-fuse needs Planner to parse exprs correctly (and safely).

Ideally, we should not parse DataBlock during bind. Can we move that logic to interpreter-insert?

Cc @leiysky @sundy-li

[sundy-li]

Ideally, we should not parse DataBlock during bind. Can we move that logic to interpreter-insert?

Yes, I think it'll be done in #7613

[Xuanwo]

Yes, I think it'll be done in #7613

Cc @youngsofun for confirmation

[sundy-li]

I'll try to apply a fast hot pr to improve this.

[Xuanwo]

This PR is out-of-date, I will start a new to replace it.