Clojurescript lenses

Code Lenses for ClojureScript

Layered on top of clojure-lenses (in clojure-skills), which covers the host-neutral Clojure translations. This skill records only the ClojureScript-specific deltas that the baseline does not address: JavaScript interop boundaries, externs and advanced compilation, macro stage separation, the JS event loop, the absence of refs / agents / STM, and the limits of var reification.

The default code-lenses review set is grug, honest-code, tidy-first, and parse-dont-validate. aposd and legacy-code are opt-in (added with +aposd or +legacy-code, or invoked directly). The sections below apply when the corresponding lens is active; APOSD and Legacy Code translations are present so the lens can use them when explicitly invoked, but they are not part of the default trigger set.

When the code-lenses plugin is active in a ClojureScript project, apply the baseline clojure-lenses translations first; reach for this skill for the JavaScript-specific additions below.

Grug Brain

• Default building blocks remain maps, vectors, and sets with namespaced keys. JavaScript objects belong at the interop boundary only; convert with clj->js and js->clj once and operate on Clojure data inside the namespace.
• Reach for deftype only when implementing a JavaScript protocol or interop that requires a named prototype. Use plain functions and data for everything else.
• The macro / function divide is its own complexity: macros must live in .clj or .cljc files and be required via :require-macros, :refer-macros, or :include-macros. Treat that as a reason to prefer functions; reach for macros only when the macro is doing real syntactic work that a function cannot.
• REPL-driven development on ClojureScript means the browser REPL or Node REPL. Spike in the REPL before introducing abstractions. The cost of starting a CLJS REPL is higher than a JVM REPL, so once a REPL is up, keep using it.
• :advanced compilation is a complexity demon waiting to happen. Enable externs inference (:infer-externs true, *warn-on-infer*) early and add ^js / ^js/Foo.Bar hints incrementally so production builds do not surprise you.
• Reagent / re-frame / Fulcro are powerful, but they are abstraction. A vector of hiccup and an atom is simpler when the problem fits. Reach for a framework when the open polymorphism it provides matches a real need.

A Philosophy of Software Design

• Namespaces are still modules. The clojure-lenses rule applies: small public API, private helpers via defn-. ClojureScript adds the externs-inference angle: a deep module hides interop ugliness so the public API does not leak js/ prefixes or property-renaming hazards.
• The interop boundary is a module boundary. Convert at the edges with js->clj and clj->js; inside the module, operate on Clojure data. Callers should not see goog.object/get or aget in the public API of a namespace whose job is application logic.
• Hide framework-specific lifecycle behind namespace boundaries. Reagent components, re-frame events, and React hooks each have their own conventions; the namespace that owns a feature should expose Clojure data in and Clojure data out, not Reagent atoms or React refs.
• Information hiding includes externs. A namespace that wraps a JS library should hide the externs and ^js hints from callers. Downstream code receives Clojure values typed appropriately.
• cljs.spec.alpha and Malli work at boundaries the same way they do on the JVM. Use them at HTTP, WebSocket, and localStorage boundaries to parse raw input into domain maps.

Tidy First

• Guard clauses still mean flattening nested if / when / let with cond, if-let, when-let, some->, and some->>. Reader conditionals (#?(:cljs ...)) are a tidying tool when the same .cljc file must read on the JVM and on the JS host.
• Explaining variables: extract complex JS-interop expressions into named let bindings. (.-userName user) is opaque under advanced compilation; (let [user-name (.-userName user)] ...) is barely better but makes the externs hazard explicit, and a typed (let [^js/UserRecord user user, user-name (.-userName user)] ...) is honest.
• Extract helper: pull blocks from a let body or threading pipeline into a named defn-. ClojureScript-specific helpers often live near the interop boundary so the calling code stays free of js/ prefixes.
• Reading order: ClojureScript files compile top-down the same as JVM Clojure. Helpers-first ordering is natural. The lazy seq versus eager mapv choice still matters; eager wins when side effects matter, which on ClojureScript means DOM writes, localStorage, and fetch.
• Threading macros (->, ->>, as->) remain the primary flattening tool. some-> and some->> are especially useful around JS values that may be nil (which equals undefined and null).
• REPL-guided refactoring: evaluate forms in the browser or Node REPL. The REPL evaluator runs in the live JS host, so refactorings are visible immediately.

Parse, Don't Validate

• The parse layer at JS boundaries: input from JSON, fetch responses, localStorage, URLSearchParams, and DOM form values. Use js->clj :keywordize-keys true once, then conform with cljs.spec.alpha or Malli into a domain map.
• Smart constructors apply: a parse-user function that takes raw JS data and returns either a conformed map or throws ex-info with :reason data. Downstream functions receive parsed data and do not re-validate.
• Namespaced keys remain the marker for parsed data: :user/email, not :email.
• Closed maps via Malli's :closed option still catch typos. Add closed schemas at the boundary to reject unexpected keys from JS payloads.
• Tagged maps for sum types: {:type :loading}, {:type :loaded :data ...}, {:type :error :message ...}. Dispatch with case, cond, or multimethods.
• The ^js/Foo.Bar hint is the externs-aware parse marker at the interop boundary. Without it, the compiler may rename properties under advanced compilation and silently break parse code that worked in development.
• Do not simulate branded types with deftype wrappers; the baseline rule against simulated branded types still applies, and ClojureScript adds the cost of breaking interop.

Honest Code

• Data is data. Clojure maps, vectors, and sets are honest; js-obj and clj->js results belong at the interop boundary. Keep the inside of a namespace Clojure-shaped.
• Pure functions over immutable data are idiomatic. Reagent components are pure functions of state; re-frame subscriptions are pure transformations of the app DB; React function components in Fulcro / Helix are pure. Hidden state is dishonest in any of those.
• One source of truth: ClojureScript apps often centralize state in a single re-frame app-db or a single defonce atom. That is the honest variant. Distributing state across many top-level def atoms makes reasoning fragile.
• Compose flat, never deep: hiccup is flat composition. Threading macros, comp, transducers, and middleware stacks (Reagent / re-frame interceptors) are flat. Avoid deep React component hierarchies that hide their own state.
• Let It Crash, ClojureScript edition: catch JS-typed exceptions where they have a named type (js/TypeError, js/Error) and :default for the catch-all. For background work (web workers, service workers), surface failure to the main thread and let the supervisor decide. There is no Erlang-style supervisor on the JS host.
• Boring tests: cljs.test with (is (= (f input) expected)) is the boring test. Reach for with-redefs-style stubbing rarely; ClojureScript's var semantics do not match JVM's, and dependency injection through function arguments or protocols is more honest.
• Declare what, not how: spec / Malli schemas, re-frame interceptors, and Reagent reactions are declarations. Prefer them to imperative effect chains.
• Atoms remain honest state. The JS host is single-threaded per realm, so swap! semantics simplify: no race conditions inside a synchronous call. core.async channels are honest when the use case requires backpressure or pipelines, but a js/Promise is simpler for a single asynchronous action.

Legacy Code

• Seams in ClojureScript: vars (rebindable in tests with with-redefs, but vars are not reified at runtime, so the seam is shallower than on the JVM), higher-order functions (pass a function instead of hardcoding a call), protocols at boundaries, and system maps (Component / Integrant / Mount work on ClojureScript as well as on the JVM).
• Extract Interface translates to extracting a protocol or, more often, a function parameter. ClojureScript protocols are useful when the same shape must work on plain JS objects and on Clojure values; otherwise function parameters suffice.
• Parameterize Constructor translates to passing a deps map or system key. Functions that touch JS globals (js/window, js/document, js/fetch) should receive those through arguments so tests can substitute.
• Sprout method/class translates to sprouting a new function or namespace. New behavior goes in a new defn called from the legacy code, not into the legacy defn itself.
• Wrap method/class translates to a wrapper function or namespace. Useful for wrapping a third-party JS library so callers see Clojure data and the externs / ^js hints stay inside the wrapper.
• Scratch refactoring maps to REPL-driven exploration in the browser or Node REPL. The compiler emits source maps, so stack traces from REPL evaluations point back to ClojureScript source.
• Effect sketching: trace which namespaces and vars a change touches, paying special attention to anything that calls into js/ globals or uses ^js. clj-kondo covers .cljs and .cljc files; use it to map the dependency graph before changing tightly coupled namespaces.
• Hard dependencies in ClojureScript include direct interop calls on js/window, js/document, js/localStorage, js/fetch, framework singletons (a global re-frame app-db when a parameterized one would do), and namespace-level defonce state that is reached from many places.
• Characterization tests use cljs.test. For UI behavior, snapshot the rendered hiccup; for re-frame flows, dispatch known events and inspect the resulting app-db. The REPL is the fastest way to discover what an existing namespace returns.

Gotchas

• with-redefs looks like a seam, but vars on ClojureScript are not reified at runtime in the same way; the rebind only affects code that resolves the var dynamically. Code that closes over the original value at definition time will keep using the original. Prefer dependency injection.
• Reagent reactions and re-frame subscriptions look like pure functions but participate in a reactive graph. Refactoring that changes when a subscription evaluates can break rendering even if the final output is unchanged.
• React hooks (used through Helix, UIx, or interop) have ordering requirements. "Extract helper" that pulls a hook call into a conditional branch breaks the hook contract. Mention this whenever a refactor touches a hook-bearing function.
• Source maps cover stack traces but not console.log of compiled code. When debugging interop, log Clojure values before they hit clj->js, not after.