How to speed up the task `compileKotlinMetadata` ? I set `org.gradle.parallel=true` in settings and...

# multiplatform

[Stefan Oltmann]

How to speed up the task compileKotlinMetadata ?

I set org.gradle.parallel=true in settings and even call it with --parallel, but this tasks takes 3 minutes for me - even on a beefy M1 Pro (8 performance cores) it tooks as long as on an M1.

So this tasks does not benefit from more cores...

What can I do to cut down compiling time here?

[christophsturm]

you could try running a build scan

[Stefan Oltmann]

So far I run a --profile and this task comes up as the single one that takes so long and does not seem to benefit from parallelisation.

[Stefan Oltmann]

Ok, did that... It has no suggestions for me. 🤷‍♂️

[Stefan Oltmann]

This just takes very long and I don't see how to speed up the task.

[Stefan Oltmann]

[Grégory Lureau]

Not 100% sure but I believe the compilation of 1 module (gradle) is mono-threaded. So if you can cut your shared modules in more modules (without too much dependencies between them) it should be able to use multi-threads.

[Stefan Oltmann]

Ok, sounds reasonable. So in my case multiple XCFrameworks instead of just one big. 🤔

[Grégory Lureau]

You can have multi-modules and get a big XCFramework. For example I use:

val xcf = XCFramework("MyLib")
ios {
    binaries {
        framework("MyLib") {
            isStatic = true
            projectsToExport().forEach { export(it) }
            xcf.add(this)

and

fun projectsToExport() =
    rootProject.allprojects
        .filter {
            it.childProjects.isEmpty() &&
                <your export rules> &&
                it != project
        }

[Grégory Lureau]

I have that configuration in my "top-level module" (mainly used for exporting rules) that depends on many libraries.

[Grégory Lureau]

without export(..) the XCFramework will contain classes transitively used by the "top-level module" but renamed with the module name. Adding export(..) ensure all classes of the package will be provided, and without the name trick.

[Stefan Oltmann]

Thank you 🙂