How to create a target that isn't a SQL dialect?

[vanillajonathan]

It is quite obvious how to create a new target that is a SQL dialect, just edit prql-compiler/src/sql/dialect.rs and create a struct that implements the DialectHandler trait then provide some functions that override functions provided by the trait.

But how would one go about creating a target that isn't a SQL dialect?

[max-sixty]

We'd have to re-architect that part of the code to do that — probably we'd extract the translator code into a Trait, which would implement the translation from RQ to the target.

It's very possible, but it would be a sizable refactor...

[snth]

If you just care about the output and data, then I guess no and it's probably sufficient for my purposes. However for winning over existing Pandas users and more data scientists and ML engineers, I think demonstrably generating Pandas code would be more useful. I wouldn't make it a priority because of that but it's something I've considered a fair amount.

[max-sixty]

Yes, I def think it's possible. It would require someone who wanted to champion this — including refactoring the existing translator code to Traits.

I also don't think it's high-priority for the project, but it would still be positive.

If anyone wants to do this, let's write a proposal for the refactor...

[eitsupi]

Perhaps we can start by implementing the conversion from the PL (json) output by the compiler, right?

[max-sixty]

Yes, from the RQ! (the PL is the first output, then RQ is second). This is the translator phase.

It's a good point — it wouldn't actually require a refactor, someone could implement a translator from RQ -> Pandas in lots of ways; including in any language.

(TBC, I think this offers almost no practical benefit over using DuckDB to target pandas, and I would vote against supporting it in the project, but as ever, open-minded :) )

[vanillajonathan]

More interestingly than Pandas, it opens up possibilities for other query languages,