Kotlin Multiplatform

[StefanOltmann]

It would be great to bring libvips to the Kotlin Multiplatform ecosystem. For the JVM, you could integrate it with vips-ffm, while iOS could leverage a native lib through Kotlin's expect/actual mechanism. The approach for Android is less clear, but for JS and wasmJS targets, wasm-vips could be used similarly via expect/actual.

The implementation wouldn't need to cover everything libvips can do (as vips-ffm allows), but focusing on a robust set of common operations would be a practical start.

Currently, SKIA is the only available option. Knowing that libvips exists - and is likely much faster than most alternatives - it would be fantastic to have (for example) a unified thumbnailing API accessible across all platforms.

If it's done, you could present how you accomplished it at KotlinConf. ;)

[lopcode]

I don't have a huge amount of motivation to do the work for this personally, but open to you / anyone else doing some research / proof of concept. After thinking about it a bit my main questions:

• Is there a reason why it couldn't be comprehensive on each platform, the same as JVM? I think code generation is an important aspect of this library - I wonder if it's possible to generate the KMP "API" as well as the implementations on each platform?
• I wouldn't want dependencies to leak in to the "pure Java" implementation - maybe it would need another module for "JVM for KMP"?
• On Android, would it "just" be the JVM version?
• How does KMP deal with native libraries?

This also leaks in to another question for the Java version that's on my mind - I'm wondering if it's worth building libs to give folks an "easy" option for inclusion. But that's potentially a lot of maintenance overhead, and telling folks to use the package manager for their system is easier for the project.

[StefanOltmann]

I don't have a huge amount of motivation to do the work for this personally

I guess primarily because many aspects of how to achieve it are unclear (yet). If I had a concrete understanding of how everything functions and could be implemented, I would have outlined it here.

but open to you / anyone else doing some research / proof of concept

I may eventually experiment with it, but my current workload includes numerous other features for Ashampoo Photo Organizer. This means it could take me up to two or three years to prioritize this task. That said, it presents a performance improvement I could market once I exhaust the list of major new features. ;)

Is there a reason why it couldn't be comprehensive on each platform, the same as JVM?

That would indeed be fantastic.

I think code generation is an important aspect of this library - I wonder if it's possible to generate the KMP "API" as well as the implementations on each platform?

I’d love to see that happen, but I’m uncertain if it’s feasible, as I’m not deeply familiar with Kotlin at that level. I had hoped you might be. :D

If it is feasible, I’m curious why Skiko appears to have been crafted manually (using JNI) rather than generated.

I wouldn't want dependencies to leak in to the "pure Java" implementation - maybe it would need another module for "JVM for KMP"?

Agreed, I wasn’t suggesting adding Kotlin as a dependency to vips-ffm. My vision is for a new repository, vips-kmp, that leverages vips-ffm for the JVM component.

On Android, would it "just" be the JVM version?

Given that Android is still limited to Java 17 (after years stuck on Java 11), it likely won’t support features introduced in Java 22 for quite some time.

As a result, it might need to rely on the slower JNI instead. If that could be generated and avoid dependencies on JVips (which hasn’t been updated in three years), it would be ideal. However, I don’t currently know how to address that.

How does KMP deal with native libraries?

The relevant information is detailed at https://kotlinlang.org/docs/native-c-interop.html.

This applies to native (= non-JVM!) targets like Windows, Linux, macOS, and iOS. In practice, it’s particularly useful for iOS.

I'm wondering if it's worth building libs to give folks an "easy" option for inclusion.

I believe it is. It has tremendous potential. From all the available options, the way you’re using vips provides the fastest known approach for performing graphic operations.

It would easily become the GOTO solutions and beat SKIA for this tasks.

But that's potentially a lot of maintenance overhead

This may be mitigated through automation. A crucial missing piece for seamless integration is bundling vips-ffm with the native libraries, similar to how Skiko operates. Achieving this would require figuring out how to build the libraries for all platforms using GitHub Actions.

[lopcode]

Thinking out loud for future me:

• Square have KotlinPoet and I bet it could generate KMP expect/actual stuff: https://square.github.io/kotlinpoet/
• Don't know if KMP is "all or nothing" but a good first step might be shipping a KMP "Java 22+ only" library to prove the concept - the API, plus a JVM implementation
• JNI is a tricky one, it's not a good thing to bet on and will likely disappear or become increasingly difficult to use as Java encourages folks to move to FFM (though maybe JNI is more likely to stick around on constrained platforms like Android)
• Wondering if anyone has had any success using libvips in an iOS app, I bet they've tried at least - worth looking