Hey folks! I'm working on a KMP project that utilises KSP for code generation. As usual, I'm using a custom annotation to tag code within the common target module and KSP processor generates new code from that. Project is targeting many platforms (Android, iOS, Web and Desktop). While the project compiles successfully, AS/IntelliJ IDEA isn't able to resolve references to the generated code, impacting the development experience. Additionally, if I change config to only targeting one platform, symbols are resolved successfully. Is there any work around for it and make dev-experience more pleasant?

Encountered the same thing, fwiw. Didn't find a solution

tl;dr: this will be supported by KSP 2 next year. For now, only symbols that are generated for the target source sets are known by the IDE details: KSP adds generated sources to the default source set of each target compilation, but not common / metadata compilations. IDE knows those source sets via gradle tooling api and IDE plugins. Because of how source sets in KMP are compiled, if KSP adds them to intermediate source sets, they'll be observable in downstream compilations and processors. They would not only be redundant but also break processors' assumptions badly. In KSP 2, this will be changed. Processors will only process the default source set but not their dependencies (which are still resolvable). Generated sources can then be added to the corresponding source sets and IDE will be aware of everything generated. This will be a breaking change. Detailed design and documentation will be published for comments and discussions later.

Thanks @Ting-Yuan Huang for your response and the details. I'm eager to use KSP 2 to address and resolve our issue.

This is great news, Break all the things please 🙏🏾, the current situation is far less than ideal.

Does KSP 2 change the way it does incremental processing? it feels like it's running on almost every change, and I am absolutely sure I have incremental processing setup correctly

@Fudge Can you give some examples that could be improved?

Are there gonna be big changes to our KSP processors when we migrate to KSP 2?

@Ting-Yuan Huang I'll maybe try to create some examples later. But right now KSP is running for me for every small change, no matter where it is. Which is insane considering I even only use

aggregating = false

.

@Fudge Thanks for pointing that out. Yeah, the limitation is still there. I hope that might change with K2. We'll do some experiments to see if a better trade-off is possible.

Awesome