How to structure different dialects

[max-sixty]

Continuing from #347

What's the best approach here? I'd propose we have a Trait in translator.rs which is implemented by each dialect — any thoughts?

[max-sixty]

I had a quick go at including a struct in the definition of a Query, but got some rust type errors around object safety which I didn't take the time to figure out. Here's where I paused (but is maybe not even helpful to start from) https://github.com/prql/prql/compare/dialect

One naive approach would be to keep the Enum we currently have, and then have a lookup to find the respective struct. Then the appropriate functions in translator.rs could part of a Trait which structs would implement. That seems worse than implementing them on the object which is a field of Query though.

[max-sixty]

This also applies to versions.

Are there approaches that won't lead to nD x nV implementations, where nD is the number of dialects and nV is the number of versions? Are there other libraries that do this well?

The rust language manages editions — I haven't looked into the mechanics, but the basics are here. And sqlparser-rs has a nice simple way of handling dialects

[aljazerzen]

Having dyn Dialect in AST will be quite painful because of serialization and de-serialization.

What about having:

trait DialectHandler {
   ...
}

enum Dialect { ... }

impl DialectEnum {
  fn handler() -> Box<dyn DialectHandler> {
  }
}

so Dialect enum can be in AST, and DialectHandler trait is used for nice polymorphic behavior?

I'm pushing such commit now, feel free to revert it.

[max-sixty]

Closed by #398, thanks to @aljazerzen ; I'll open a new issue for versions