Lwjgl3: `apply` functions added in 3.2.3 break existing Kotlin code

[Hugobros3]

Release 3.2.3 brought a new apply method in StructBuffer and thus clashes with the built-in apply extension function available in Kotlin code. That apply function appears to do the same job but has a different calling convention from the Kotlin one (Kotlin apply takes a lambda with this renamed to the object it's called on, while StructBuffer's takes a SAM ).

The result of this is large portions of Kotlin code that use StructBuffers are now broken (example here) and need tedious manual working around the issue. I have to say it's a very unfortunate oversight to have implemented any functionality that clashes with basic Kotlin stdlib functionality, especially given LWJGL itself uses Kotlin for various things internally.

I'm not sure how to fix this, I admit it's been a while since I worked on my Kotlin game engine project. I'd say roll back the changes, but this already shipped so it's a matter of choosing whose code you'll break now. The middle ground might just be to rename the function ?

[Hugobros3]

(I don't know about tons, it's just mine afaik. But apply is a staple of idiomatic Kt code)

[octylFractal]

I don't think this does the same job -- here apply works on a relative element of the buffer (and increments the position!), whereas Kotlin's apply would be targeting the buffer itself. Your code only worked because the buffer also has methods to target the currently pointed-to element of the buffer, not because Kotlin's apply was extracting the element from the buffer.

I do think that it may be slightly misleading for them to have the same name and similar signatures, but do _very_ different things.

[XenoAmess]

So can it build with the latest version of kotlin?
Or just compile error?

[Hugobros3]

Your question is weirdly phrased. This has nothing to do with changes on the Kotlin side, the latest version of LWJGL just breaks my code in plenty of places.

I do very liberal amounts of stackAlloc(1) to get singleton array of structs to pass into Vulkan in plenty of places where the API expects an array, since LWJGL doesn't offer overloads for the non-buffer variants. Using Kotlin's apply there just makes sense and yields the most readable code, which is why I had settled on that pattern.

[XenoAmess]

Your question is weirdly phrased. This has nothing to do with changes on the Kotlin side, the latest version of LWJGL just breaks my code in plenty of places.

I do very liberal amounts of stackAlloc(1) to get singleton array of structs to pass into Vulkan in plenty of places where the API expects an array, since LWJGL doesn't offer overloads for the non-buffer variants. Using Kotlin's apply there just makes sense and yields the most readable code, which is why I had settled on that pattern.

Oh. I see.
This is why I hate kotlin.
Such clash seems will never happen if using java instead.

[Spasi]

Hey @Hugobros3, thanks for opening this issue. Adding these methods to StructBuffer was meant to make it easier for Java code to consume the Vulkan API, but it indeed had this unfortunate side-effect when working in Kotlin. I would probably take it back (or made different) if this was reported earlier, but I'm afraid it's too late now.

However, I would like to point out that there's a ton of flexibility in Kotlin and there are trivial workarounds that don't sacrifice readability. As an example, I refactored some of the code in the posted screenshot and this is how I would have written it:

val colorAttachmentReference = VkAttachmentReference.callocStack(outputsDeclaration.outputs.size)
colorAttachmentReference.forEachIndexed { index, ref -> ref
    .attachment(index)
    .layout(VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL)
}

val subpassDescription = VkSubpassDescription.callocStack(1)
    .pipelineBindPoint(VK_PIPELINE_BIND_POINT_GRAPHICS)
    .pColorAttachments(colorAttachmentReference)
    .colorAttachmentCount(colorAttachmentReference.capacity())
    .pDepthStencilAttachment(if (!depth.enabled) null else VkAttachmentReference.callocStack()
        .attachment(attachmentDescription.capacity() - 1)
        .layout(
            if (depth.write)
                VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
            else
                VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL
        )
    )

This is shorter and I would argue cleaner than the original code.

[Hugobros3]

Well it was reported a year ago, it looks like the issue slipped under the cracks. It doesn't really matter too much, as you say, it's too late now, and personally I have moved in other directions.