[compiler][gating] Experimental directive based gating

[mofeiZ]

Adds dynamicGating as an experimental option for testing rollout DX at Meta. If specified, this enables dynamic gating which matches use memo if(...) directives.

Example usage

Input file

// @dynamicGating:{"source":"myModule"}
export function MyComponent() {
  'use memo if(isEnabled)';
   return <div>...</div>;
}

Compiler output

import {isEnabled} from 'myModule';
export const MyComponent = isEnabled()
  ? <optimized version>
  : <original version>;

---

Stack created with Sapling. Best reviewed with ReviewStack.

• -> [compiler][gating] Experimental directive based gating #33149
• [compiler][entrypoint] Fix edgecases for noEmit and opt-outs #33148

[josephsavona]

How would this work for real code? If you use the use memo if(...) form, what would we use as the source of the import?

[mofeiZ]

How would this work for real code? If you use the use memo if(...) form, what would we use as the source of the import?

For a large codebase internally, we added a react-compiler-gating-for-<project>.js file which:

• is unconditionally added to our bundles (imported at the same resource tier as React)
• exports gating functions e.g. function enabledForSubproject() {...}

This codebase currently uses a lightweight wrapper around the gating config, but we can extend the same self-service gating function mechanism to use memo if(...) rollouts.

We briefly discussed inserting more analyzable gating (e.g. is_feature_flag_enabled('feature_flag_name')) calls which can be understood by other babel transforms, but decided that the additional complexity is not worth it for now.

Letting developers specify their own filenames (e.g. use memo if(myModuleName)) would also be a pain, as we would need to have some way of ensuring the specified module is included in relevant bundles.

[mofeiZ]

Dynamic gating should also be an opt-in if compilationMode: "annotation" is enabled.

If "use no memo" is set at module scope, use memo if(...) should have the same semantics as use memo, whatever that is

[mofeiZ]

not sure how we feel about ??= when the right hand side is a potentially expensive computation. I'm happy to make this more explicit if that's preferred (e.g. if (referencedBeforeDeclared == null) ...

[josephsavona]

i think this is fine, ??= is a short-circuiting operator just like ||=

[josephsavona]

nice!

[josephsavona]

i think this is fine, ??= is a short-circuiting operator just like ||=