Open mrbrentkelly opened 2 weeks ago
Thanks for reporting the issue, @mrbrentkelly.
I'll try to work on this asap. 0.76.2 is planned for tomorrow, so it's unlikely that the fix would be out. I'll make my best to have it out for 0.76.3.
Meanwhile, you should be able to unblock yourself by manually removing the lines from Xcode.
Hi @mrbrentkelly, I was thinking about this issue more deeply, and I am not sure whether disabling Codegen if the library is excluded from autolinking is the right thing.
Let me try to explain:
pod <library name>
line in your Podfile, if you need some specific customization (use_modular_header for example)package.json
setting the property codegenConfig.includesGeneratedCode
to true, these are some example libraries doing that. A couple of clarification questions:
If you are not linking the library, why can't you just uninstall the dependency instead?
Uninstalling isn't always an option... Say for example an app wants to use a particular RN library on Android and not iOS (or vice versa)... typically the approach would then be to disable the library for the intended platform in the react-native.config.js
so that it doesn't get auto-linked. In my opinion the codegen should not happen if a library hasn't been linked (to avoid compilation errors).
Another example would be you're using a RN library like Sentry where linking the native side is optional depending on which APIs you want to use.
Or perhaps you're in a monorepo of multiple apps and node_modules contains all dependencies for all apps.
Another example would be an RN library that you link for debug builds but not release builds (perhaps via an environment var in utilised in the react-native.config.js).
If we proceed with the fix you suggested (skipping codegen if the package is excluded from autolinking) we are going to break the use case of manual linking into the app
It sounds like the codegen for iOS should be looking at linked pods (after RN auto-linking), instead of relying on what's contained in node_modules. That would then work for manually linked libs and auto-linked ones...
node_modules aren't the single source of truth for a React Native app, and as I've noted above there's multiple variables to consider. So I think we are going to need better control over when the codegen happens?
So I think we are going to need better control over when the codegen happens?
Yes, I believe that codegen and autolinking should be decoupled.
For completeness' sake, autolinking works by asking for some configuration to get a list of libraries that needs to be linked. For projects with the CLI, thats react-native info
command. Expo provides their own commands with the same format.
Codegen, instead, looks at both libraries in your dependencies
, peerDependencies
, devDependencies
and into the react-native.config.js
file to decide what to generate and what not.
Also, starting from 0.77 we will not generate the xxxCls
anymore, so the problem will be probably fix itself.
Given that autolinking is using that file to decide what to link in the app, we can patch 0.76 to remove the library from the list of libraries if the react-native.config.js
exclude the library from linking.
Description
When a third party react-native dependency exists in an app's package.json but has been disabled from auto-linking in the
react-native.config.js
file, react-native-codegen still generates native Objective-C code for the unlinked dependency, leading to build failures.Steps to reproduce
Create a new RN project and add a third party library that contains native code (e.g.
react-native-screens
)Create a
react-native.config.js
file and disablereact-native-screens
from auto-linking on iOS...Run pod install and try and build/run the iOS app
Build will fail because the generated
RCTThirdPartyFabricComponentsProvider
files reference things that are linked in the project (e.g._RNSScreenCls
).React Native Version
0.76.1
Affected Platforms
Runtime - iOS
Areas
Codegen
Output of
npx react-native info
Reproducer
https://github.com/mrbrentkelly/rn-codegen-linking-bug
Screenshots and Videos
No response