Support KSP in factory generation

[ZacSweers]

This issue is to track KSP support in factory gen, and different from #704

• AnvilMergeAnnotationDetectorCheck.kt Update AnvilMergeAnnotationDetector for ksp #756
• AssistedFactoryGenerator.kt Add KSP support to InjectConstructorFactory, MembersInjectorCodeGen, and assisted injection #795
• AssistedInjectGenerator.kt Add KSP support to InjectConstructorFactory, MembersInjectorCodeGen, and assisted injection #795
• BindsMethodValidator.kt
• ComponentDetectorCheck.kt Update ComponentDetectorCheck for ksp #753
• InjectConstructorFactoryGenerator.kt Add KSP support to InjectConstructorFactory, MembersInjectorCodeGen, and assisted injection #795
• MapKeyCreatorGenerator.kt Support KSP in MapKeyCreatorGenerator #788
• MembersInjectorGenerator.kt Add KSP support to InjectConstructorFactory, MembersInjectorCodeGen, and assisted injection #795
• ProvidesMethodFactoryGenerator.kt Migrate ProvidesMethodFactoryCodeGen to KSP #806
• ContributesMultibinding Implement KSP support for ContributesMultibinding #740
• ContributesTo Implement KSP support for @ContributesTo #742
• AnvilAnnotationDetectorCheck Start KSP dual implementation #722
• ContributesBinding Start KSP dual implementation #722
• BindingModuleGenerator Restructure binding module code gen #877
• ContributesSubcomponentGenerator Support KSP in ContributesSubcomponentGenerator #828

[jhperezd]

👋 I'll be taking ComponentDetectorCheck.kt and AnvilMergeAnnotationDetectorCheck.kt to begin with.

[r0adkll]

👋🏻 I'll be taking MembersInjectorGenerator.kt to start with as well

[jhperezd]

👋 PRs for ComponentDetectorCheck.kt (#753) and AnvilMergeAnnotationDetectorCheck.kt (#756) are up. I'll take InjectConstructorFactoryGenerator.kt next.

[ursusursus]

May I ask?

Is this about not having to use

anvil {
  generateDaggerFactories = true // default is false
}

such that one would then use ksp dagger everywhere instead, therefore avoiding the incremental compiliation issue?

[ZacSweers]

That wouldn't be enough as you still need anvil's hint generation for contributed elements

[IlyaGulya]

@ZacSweers
Hello!
Would you accept help from the community to speed up this issue resolution?

[ZacSweers]

Sure! The last two are gnarly, so it make take a bigger lift. I can write up more about them tomorrow

[ZacSweers]

So the main issue is that both of those generates require doing class scanning, which complicates things in KSP due to it processing nodes that are not in the compilation source set.

This is obviously necessary for module/component merging, but ideally shouldn't be necessary for any of the factory generators.

BindingModuleGenerator

This is the one that requires the most change. Currently Anvil will scan within the package and generate one module for all the contributed bindings. This is a greedy aggregating step, and after some discussion with @vRallev we agreed this should be simplified to just generate 1:1 mappings of a @ContributesBinding class to a generated @ContributesTo-annotated binding module, rather than one aggregated one. This refactoring would affect both the existing Anvil PSI support as well as KSP.

The bulk of the work would be to adopt this new model, then implement dual KSP support afterward.

ContributesSubcomponentHandlerGenerator and ContributesSubcomponentGenerator

This is support for @ContributesSubcomponent, which is a fairly involved case and requires doing class scanning. I think ContributesSubcomponentHandlerGenerator can be saved for the dagger-KSP PR as it deals with merging, and ContributesSubcomponentGenerator is just a simple generator. I've updated the task list to remove ContributesSubcomponentHandlerGenerator now that I've looked at it more.

[ZacSweers]

I'm gonna take ContributesSubcomponentGenerator next since it's trivial

[ZacSweers]

BindsMethodValidator should be a simple one to do. There is also a bug with the current implementation in PSI (#750) that may get fixed by KSP since it has built-in assignability APIs

[IlyaGulya]

BindsMethodValidator should be a simple one to do. There is also a bug with the current implementation in PSI (#750) that may get fixed by KSP since it has built-in assignability APIs

Thanks!
Then, I'll take BindsMethodValidator.

[IlyaGulya]

@ZacSweers Is there anything that should be done? I can take another task if there's one 🙂

[ZacSweers]

The last big one will be BindingModuleGenerator. But in this case, anvil needs to change how this works and no longer merge modules to create one large module. Instead, binding module hints need to be updated to generate 1:1 binding -> binding module to simplify things. If you want to tackle that that would be helpful, but you may need more guidance from @vRallev and @JoelWilcox on how that implementation should look. This change would need to work in Embedded too

[ZacSweers]

@IlyaGulya fyi I've actually started working on BindingModuleGenerator after some discussion with the square folks and Ralf

[gildor]

Does #831 merge means that all points are done now? Maybe there are some other follow-up tasks.
Thanks!

[ZacSweers]

yup!

[ZacSweers]

There's still some work to do around subcomponent contribution, but that's more about the plugin wiring side rather than the impl itself