Figwheel crash calling js\require in complex circumstances

[deg]

I'm seeing what looks like a Figwheel crash:

[exp] undefined is not an object (evaluating 'receipts_native.core.init.call')
* js/figwheel-bridge.js:117:14 in <unknown>
- node_modules/promise/setimmediate/core.js:37:14 in tryCallOne
- node_modules/promise/setimmediate/core.js:123:25 in <unknown>
- ... 10 more stack frames from framework internals

[exp] Unknown named module: 'native-base'
- node_modules/metro-bundler/src/Resolver/polyfills/require.js:96:12 in _require
* js/figwheel-bridge.js:162:26 in require
* js/figwheel-bridge.js:117:14 in <unknown>
- node_modules/promise/setimmediate/core.js:37:14 in tryCallOne
- node_modules/promise/setimmediate/core.js:123:25 in <unknown>
- ... 10 more stack frames from framework internals

This is triggered by trying to js\require a module from code generated by a macro.

I've branched my project to create a small example that demonstrates the issue. See https://github.com/deg/receipts-native/tree/figwheel-require. The key lines are https://github.com/deg/receipts-native/blob/e39a22b39c69d28ae153b84f028dab304e930279/src/receipts_native/macros.cljc#L23-L34

#?(:clj
   (defmacro def-native-components
     "Extract one or more React classes from a JavaScript module. Adapt them as Reagent components and name them.

      [TODO] Weird bug that the js/require here seems to crash figwheel, unless the same module was previously
      require'd from .cljs. If this bug is real, I'll probably have to remove the require from this macro."
     [module-name names]
     (let [module (gensym "module-")]
       `(let [~module (js/require ~module-name)]
          ~@(map (fn [name]
                   `(def ~name (get-class ~module ~(str name))))
                   names)))))

As it stands now, my code is minimal. The one complication is that it requires an Expo React Native setup. I hope this is reduced to something small enough that you can see the problem. If not, I can try to create a test case that does not need React Native (though this is the first context in which I've had to use js\require, so I'll be treading on some unfamiliar ground).

[deg]

Looks like defmacro was a red herring. I don't know much more yet, but wanted to give you an alert that I don't fully understand the problem yet.

It seems to be enough to have the js/require in some unusual position, rather than at top-level. I'm still not certain what I'm seeing.

Does this ring any bells? Is there something magical about how js/require is handled?

[deg]

Well, defmacro was a red herring. I've simplified the code, and can reproduce the crash just wrapping the js/require in a let. No need to define a macro, or any other complexity.

Still in the same branch: https://github.com/deg/receipts-native/tree/figwheel-require, and the relevant code is now at https://github.com/deg/receipts-native/blob/figwheel-require/src/receipts_native/core.cljs#L11-L26:

;;; Restoring this next line removes the crash.
;;; It seems that Figwheel (or something) is unhappy when module is first required in the
;;; expansion of a macro???

;; (defonce native-base (js/require "native-base"))

(defn get-class
       "Extract React class from JavaScript module and adapt as Reagent component."
       [module class-name]
     (r/adapt-react-class (aget module class-name)))

(let [module (js/require "native-base")]
  (def Body      (get-class module "Body"))
  (def Container (get-class module "Container"))
  (def Header    (get-class module "Header"))
  (def Title     (get-class module "Title")))

I've still only reproduced this in the context of Expo and React-Native. I'll try to find a simpler case.