Why compile to TypeScript

Bynk does not have its own runtime. It compiles to TypeScript and runs on the JavaScript platform — specifically targeting Cloudflare Workers. This page explains what that choice buys and what it costs.

What it buys

Reach without a new ecosystem. A language is only as useful as the libraries, tools, and platforms it can use. By emitting TypeScript, Bynk inherits npm, the JavaScript tooling world, and a mature, globally distributed serverless runtime on day one, instead of asking anyone to adopt a bespoke VM.

Readable, debuggable output. The generated TypeScript is meant to be read. Every commons, type, and handler maps to recognisable TypeScript — branded types, discriminated unions, plain functions. When something misbehaves you can open the emitted code and see exactly what runs; there is no opaque bytecode in between. The compiler also emits a source map (<file>.ts.map) alongside each generated .ts, so stack traces and a debugger resolve back to the original .bynk lines; the .bynk source is embedded for local builds and kept out of deployed Workers by default.

A deployment model that matches the language model. Bynk’s units line up with the platform’s units: a context becomes a Worker, and a stateful agent becomes a Durable Object. The architecture you express in the language is the architecture you deploy, with no impedance mismatch to paper over.

A type system in front of a dynamic runtime. JavaScript is permissive; Bynk puts a strict, refinement-aware type system ahead of it, then emits TypeScript whose own types reinforce the guarantees. You get the platform’s flexibility with a much stronger static contract.

What it costs

You inherit JavaScript’s runtime semantics. Numbers are IEEE-754 doubles; Int is number. Bynk’s types prevent many mistakes, but the underlying value domain is still JavaScript’s, and that occasionally shows through.

Boundaries need validation. Data arriving from the network — an HTTP body, a cross-context call on the workers target — is untyped JSON until proven otherwise. Bynk generates validation at those boundaries, which is correct but is work the runtime does that a single-process typed language might avoid.

Two targets to reason about. The same Bynk program emits differently for bundle (direct calls) and workers (calls over Service Bindings). The source is identical, but the operational characteristics differ, and you have to know which you are building. See Target Cloudflare Workers.

The bet

The wager is that for the services Bynk targets, reach and a proven runtime are worth more than control over a bespoke runtime — and that a strong type system plus readable output recovers most of what you might otherwise miss. See Why Bynk exists for the wider motivation.