Serving R with Ark

Lionel Henry

Where is the in Positron?

• Posit’s new IDE Positron is multilingual

• Focus broadened to Python 🐍🐍🐍
  Where does this leave us with R?

Positron and Ark

• Positron extensible and interoperable
  via independent backends

• Most of the R support is
  implemented in Ark

30 months of work

• Jonathan McPherson and Kevin Ushey started the project in early 2022

• Davis Vaughan, Romain François, Daniel Falbel, Julia Silge, Jenny Bryan, and I joined in 2023

What’s in the Ark?

1. 🤔 An LSP server
2. 🫨 A DAP server
3. 🤯 A Jupyter kernel

Goal: For R support to be independent of Positron

What’s in the Ark?

• Trend of language servers in last decade

• Interoperability via a standard protocol
  

• Great for diversity!
  Same tooling shared by users of Emacs, nvim, VS Code, …

(1) Language Server Protocol

Language Server Protocol

LSP deals with suggestions, diagnostics, formatting, …

• Knows about your source code and its dependencies
  Gets updates when user types
• Parses code (tree-sitter), indexes it, analyses it
• Provides code assistance

LSP: Completion

LSP: SignatureHelp

LSP: Hover

(2) Debug Adapter Protocol

Debug Adapter Protocol

DAP deals with breakpoints and stepping through code

• Injects browser() calls at break points (TODO)
• Manages source references and location
• Creates virtual documents for external functions
• Inspects local variables while stepping

DAP

• Stopped at line 1359 in a virtual document for the whole dplyr namespace
  We’d like to do better by downloading actual sources to preserve folder structure and comments
• From there can go to the next line
• Inspect currently running functions (call stack) and current state of variables

(3) Jupyter kernel

A missing piece?

• Dev-oriented editors get by with a LSP and DAP

• Data-oriented editors need interactive execution

• Assistance => LSP
  Debugging => DAP
  Execution => ???

Two approaches

• Implement IDE API for the language (e.g. RStudio)
  • Specific to each interpreter
  • Ad hoc communication with frontend
    
    
• Communicate with raw text (e.g. ESS)
  • Hard to match inputs to outputs
  • Makes background interaction difficult

Can we do better?

Jupyter

• Mainly known for notebooks but consoles possible
• Extensible via kernels implementing a protocol
  Inputs and outputs are standard messages
• Impressive compatibility list (even… Fortran!)
  https://github.com/jupyter/jupyter/wiki/Jupyter-kernels

A native kernel for R

• IRkernel is an R package
  Easy to install from CRAN
• Ark implements the IDE API
  Manages R externally
• Native behaviour like in RStudio or the terminal

sys.calls()

Three-in-one

Fully reusable infrastructure for R support

• Assistance => LSP

• Debugging => DAP

• Execution => Jupyter

The whole is greater than the sum of its parts

But we plan standalone releases

Multilingualism in Positron

Demo

Stepping through base R at R and C++ levels

Towards shared R support

• Already usable in all existing Jupyter apps
• Still need to build support in other editors

Towards modern R tooling

Exciting opportunities for future development

• Smarter code analysis and diagnostics
• Annotation of package code with declare()

Conclusion

• Ark aims for R to remain a first-class citizen
  in the evolving IDE landscape
  • More collaboration between IDE maintainers
  • Long-term sustainability of R tooling