Use base-R standalone files for functionality copied in Positron?

[lionel-]

It might be helpful if the functionality we copied over to Ark for tool calls in Positron was implemented in base-R only standalone files. This way the functionality is updated and tested in one place, usethis can be used to manage updates (https://usethis.r-lib.org/reference/use_standalone.html), and we wouldn't need to adapt the code to base R when we need to update.

You can see which functions we adapted in posit-dev/ark#868 (files help.R and llm_tools.R).

This is just a suggestion, I'd understand if you think this would be impractical or too constraining on your end.

[gadenbuie]

Thanks @lionel-. From the linked PR here are the key changes that were made:

Specifically, the changes from btw are as follows:

• remove dependency on tidyverse (dplyr, etc.) by rewriting transformations in base R
• return plain-text results rather than throwing errors in most cases
• remove dependency on rmarkdown by bundling the needed pandoc conversion bits

We used dplyr by convenience and I could certainly see us moving dplyr to Suggests. That said, it'd be a nontrivial undertaking and I don't want to hold up releasing btw at this point.

Similarly, in the btw + ellmer context, we prefer throwing errors. And the pandoc conversion bits are equally non-trivial.

I'm going to leave this issue open because I do think it'd be worth the effort in the future to use base R for transformations as well as to revisit the parts that use pandoc, but just not at this time.