R frontend — PARKED (dynamic-runtime language, not a math overlay)

• Type: feature — frontend request
• Status: backlog — PARKED (2026-07-04, user decision)
• Opened: 2026-07-04 (feature request — considered as a frontend candidate)

Why parked

Initial expectation was that R is a thin mathematical/statistical syntax overlay — cheap to add like a C-family frontend. It is not. R is a full dynamic-runtime language, in the same bucket as JavaScript ([[feature-js-frontend-parked]]), NOT the Zig/Rust "C-compatible syntactic sugar" bucket that maps straight onto the shared static IR.

What makes it a runtime project, not a frontend:

• Dynamically typed — variables hold any type → needs a boxed/tagged value model + garbage collector. Does not map onto the statically-typed IR the way C/Zig/Rust do.
• Vector-first — the atomic unit is a vector (a scalar is a length-1 vector); elementwise ops + recycling rules → needs a vector engine.
• Copy-on-modify — pass-by-value with copy-on-write, refcounted (our ARC is only a partial precedent).
• Lazy args + non-standard evaluation — function args are promises; functions capture unevaluated argument expressions (quote/substitute/ eval). Runtime metaprogramming, and pervasive — the tidyverse (dplyr/ggplot2) is built on it. Cut it and you can't run most real R.
• NA missing values propagate through every type/op.
• The value IS the ecosystem — base R + stats + CRAN, most heavy packages being C/Fortran with R glue. "Support real R" implies that stack — the same JS/Cesium-style wall.

The real fork (shared with JS)

R's question is not "R yes/no" — it is "do we want a dynamic-value runtime (tagged values + GC) at all?" That runtime is the big shared investment for BOTH R and JS; the C-family frontends (Zig/Rust) never need it. So:

• C-sugar frontends (Zig, Rust) — tractable, map onto the IR, near-term.
• Dynamic-runtime languages (JS, R) — gated behind the one runtime decision. Build it once → both reachable; skip it → both parked.

Reopen only with

• A concrete pure-numeric R subset goal (scalar/vector math + functions + control flow, no NSE, no data.frames, no CRAN) with a real target program, OR
• A decision to build the dynamic-value runtime (then sequence R + JS together).

Log

• 2026-07-04 — parked per user ("forget about R; thought it was a math overlay, it's more complex"). Recorded the runtime-vs-overlay reasoning + the shared-with-JS runtime fork so it isn't reargued from scratch.