User Defined Functions with R

[ran-codes]

duckdb/duckdb#12438 (comment)

suggested to ask this in this repo.

Our team uses the tidyverse API and some basics statistics function in R for much of our data work. We have shifted to Duckdb - particualrly the duckdb-dbt workflow - and would be a gamechanger if we could do some of our R work in duckdb via UDF.

I have read duckdb/duckdb#471 but am wonder if there is plans for support for other language UDFs. I think it would be a game changing feature.

[paleolimbot]

A while back I was involved in implementing Arrow UDFs and the challenge was basically that R code must execute on the same thread, so there is a slightly awkward dance where whenever Arrow does something that might execute some R code (like a user-defined function), we have to launch a thread to do "do arrow stuff" and run an event loop on the R thread that waits for R tasks that need executing on the other thread.

Some (not so) light reading that might be relevant (I am not sure how similar DuckDB UDFs are to Arrow UDFs):

https://github.com/apache/arrow/blob/fd11b7a042f0a020a0cde34e043ad2e77752bca6/r/src/safe-call-into-r.h

https://github.com/apache/arrow/blob/fd11b7a042f0a020a0cde34e043ad2e77752bca6/r/src/compute.cpp#L633-L788

[alejandrohagan]

I'm just here to lend my support.

I'm building shiny apps with duckdb back ends and it would be so fantastic if we could get R UDFs.

[ran-codes]

coming back to this. Yes I think happy to contribute as well.

@paleolimbot do you have any PR for other duckdb features similiar to this that potentially itnerested contributors can get a glimpse at just to try to understand at a high level the work that would have to happen for this?

[krlmlr]

I don't think anything like what Dewey described in #181 (comment) currently exists in the duckdb R package. This would be a new development, exciting for sure, but also somewhat difficult.