The Promise and Peril of Kotlin Multiplatform

Transcript

1. Justin Brooks The Promise and Peril of Kotlin Multiplatform A New Way to Share Code
2. The Status Quo
3. CPU-Native Code C++ / C / Swift • Debugging is annoying • C interop is sometimes a pain (JNI, Swift/C++ interop is new) • C interop is sometimes not as free as you think (Java Native Interface) • Misses advantages of virtual machines with JIT (like the Android Runtime) • Can bloat binary size unnecessarily • Misses JIT-level optimizations (especially pro fi le guided optimizations)
4. JavaScript • Debugging is painful • Performance characteristics can depend on the host VM • Misses platform-speci fi c optimizations • Interop overhead • “Smart View” problems
5. Xamarin / Mono • Debugging is painful • Build-time complexity incorporating msbuild • Slow, generational stop-the-world GC • Misses advantages of platform-native virtual machines • ART pro fi le-guided optimizations • LLVM JIT negatively impacts startup performance, which is already bad
6. Kotlin Multiplatform
7. Not a library. Not a framework.
8. A Quick Tangent on Compilers
9. int square(int num) { return num * num; } de fi %2 = alloca i32, align 4 store i32 %0, ptr %2, align 4 call void @llvm.dbg.declare(metadata ptr %2, metadata !16), !dbg !17 %3 = load i32, ptr %2, align 4, !dbg !18 %4 = load i32, ptr %2, align 4, !dbg !19 %5 = mul nsw i32 %3, %4, !dbg !20 ret i32 %5, !dbg !21 }
10. int square(int num) { return num * num; } de fi ne dso_local noundef i32 @_Z6squarei(i32 noundef %0) #0 !dbg !10 { %2 = alloca i32, align 4 store i32 %0, ptr %2, align 4 call void @llvm.dbg.declare(metadata ptr %2, metadata !16), !dbg !17 %3 = load i32, ptr %2, align 4, !dbg !18 %4 = load i32, ptr %2, align 4, !dbg !19 %5 = mul nsw i32 %3, %4, !dbg !20 ret i32 %5, !dbg !21 }
11. MODULE_FRAGMENT name:<main> FILE fqName:<root> fi FUN name:square visibility:public modality:FINAL <> (n:kotlin.Int) returnType:kotlin.Int VALUE_PARAMETER name:n index:0 type:kotlin.Int BLOCK_BODY RETURN type=kotlin.Nothing from='public fi CALL 'public fi $this: GET_VAR 'n: kotlin.Int declared in <root>.square' type=kotlin.Int origin=null other: GET_VAR 'n: kotlin.Int declared in <root>.square' type=kotlin.Int origin=null fun square(n: Int): Int { return n * n }
12. MODULE_FRAGMENT name:<main> FILE fqName:<root> fi leName:sources/main.kt FUN name:square visibility:public modality:FINAL <> (n:kotlin.Int) returnType:kotlin.Int VALUE_PARAMETER name:n index:0 type:kotlin.Int BLOCK_BODY RETURN type=kotlin.Nothing from='public fi nal fun square (n: kotlin.Int): kotlin.Int declared in <root>' CALL 'public fi nal fun times (other: kotlin.Int): kotlin.Int [operator] declared in kotlin.Int' type=kotlin.Int $this: GET_VAR 'n: kotlin.Int declared in <root>.square' type=kotlin.Int origin=null other: GET_VAR 'n: kotlin.Int declared in <root>.square' type=kotlin.Int origin=null fun square(n: Int): Int { return n * n }
13. RISC-V ARM X64
14. RISC-V ARM X64
15. Compiler Backends: Stable *
16. KMP is compiler orchestration (a build system plugin and IDE tooling)
17. Kotlin Mutliplatform (KMP)
18. expect & actual Delegating to Platforms
19. > Task :common:compileDebugKotlinAndroid FAILED e: Expected class ‘Platform' has no actual declaration in module <common_debug> for JVM expect class Platform {}
20. • All Kotlin Common • Delegate to platforms with expect & actual • Glue for platform-native libraries • Application • Kotlin all the way down (Compose Multiplatform) • Library • For Kotlin Multiplatform • For platform targets (XCFramework, Android Archive, NPM Package) How What Lots of Options
21. Some Good Third-Party Libraries • cashapp/SQLDelight • square/okio • ktorio/ktor • And a weird one (cashapp/ Redwood) • Kotlin/kotlinx.serialization • Kotlin/kotlinx-datetime • Kotlin/kotlinx-io • Kotlin/kotlinx.collections.immutable Some Good First Party Libraries
22. Peril • N+1 Problem • Kotlin/Native doesn’t support vararg parameters - so Kotlin common is missing String.format. • Building JS-native NPM packages is onerous. • Sometimes you want (or need) API boundaries to be di ff erent between platform targets. Doable but a little awkward. • Pure Swift dependencies aren’t supported for Apple targets.
23. Knapsack (Demo)
24. common/ • Authentication initial password grant & refreshing access token • Network http & json serialization • Caching SQLite & simple key-value storage • Text processing HTML → platform-speci fi c rich text representations
25. Directory Files Lines Blanks Comments Code Complexity Common 32 Kotlin 971 148 8 815 64 knapsack- android 12 Kotlin 784 79 29 676 17 knapsack-iOS 10 Swift 442 76 68 298 10
26. Share when it helps; otherwise, don’t.
28. https://github.com/Faithlife/FaithlifeOAuth-Mobile https://net fl ixtechblog.com/net fl ix-android-and-ios-studio-apps-kotlin-multiplatform-d6d4d8d25d23 https://kotlinlang.org/lp/mobile/case-studies/cash-app https://youtrack.jetbrains.com/issue/KT-47241/Kotlin-Variation-That-Compiles-Down-To-IL https://github.com/jzbrooks/Knapsack