RacketCon 2026 lands in its sixteenth iteration with a program that reads like a thesis statement for language-oriented programming. The keynote comes from Pat Hanrahan — Turing Award recipient, RenderMan architect, Tableau co-founder — speaking on notation itself. That's a deliberate framing choice: before the talks about type systems and effect handlers, the conference asks you to think about representation as a design problem. The Saturday lineup hits hard on Racket's central tension: how do you get the expressive power of macros and DSLs without sacrificing the tooling, types, and performance that production use demands? Fred Fu's talk on type inference for untyped languages tackles the pain point where Typed Racket chokes on macros imported from untyped modules — a real usability cliff that pushes users back to manual rewrites. His approach combines occurrence typing with algebraic subtyping, which is a genuine theoretical move, not just engineering polish. Lucas Myers' Pille project may be the most ambitious entry. It grafts Rhombus's enforestation process onto an LLVM-backed compiler, effectively asking: what if language-oriented programming worked on microcontrollers? This is the kind of research that either opens a new corridor or dies on the vine — the gap between Racket's runtime and bare-metal targets is enormous, and bridging it without losing the macro system is a real engineering challenge. Mike Delmonaco's Treason addresses a problem every IDE developer knows: macros break tooling. His approach — spec-driven subexpression expansion that recovers from errors mid-expansion — is the kind of infrastructure work that makes an ecosystem livable rather than just intellectually interesting. Ryan Culpepper's BrandX similarly targets ergonomic gaps, rethinking OOP in Racket with proper contract support for interfaces and generics. Pavel Panchekha's Herbie talk offers a ten-year retrospective on a floating-point accuracy compiler written in Racket — the kind of longevity case study that conferences rarely get. Matthew Flatt's ffi2 talk traces a line from Eli Barzilay's 2004 FFI through Andy Keep's Chez Scheme ftypes to a new unified layer, which is essentially the story of Racket's runtime evolution in miniature. The supporting program — JJ's effect handler library cio, Sam Phillips' immutable dataframe library Uke, and the Sunday town hall — fills out a picture of a community that's small but technically dense. Early registration runs until September 13 with a discount, the conference will be livestreamed, and recordings go to YouTube afterward. The evening social is at 2325 Broadway. What makes this program distinctive is not any single talk but the coherence of the agenda: every presentation orbits the question of whether extensible languages can move from research artifact to practical tool across domains from web to embedded to scientific computing. Sixteen years in, the Racket community is still making that argument with new evidence rather than retreating to nostalgia.
