Compiler

Skotch compiles Kotlin source files through a shared front-end and then dispatches to target-specific backends. The entire pipeline runs in a single Rust binary with no external tool dependencies (except clang for native linking).

Pipeline overview

.kt source
  → Lexer        (hand-rolled, ~250 lines)
  → Parser       (recursive descent, ~1200 lines)
  → Resolver     (name resolution, forward references)
  → Type Checker (two-pass: signatures first, then bodies)
  → MIR Lowering (AST → three-address-code IR)
  → Backend      (target-specific code generation)

All targets share the first five stages. The MIR (Mid-level Intermediate Representation) serves as the “waist” of the compiler — backends only need to consume MIR, not the AST.

Backends

JVM

Produces Java 17 class files (major version 61). One wrapper class per source file (e.g., hello.kt → HelloKt.class). Top-level functions become static methods. The constant pool, bytecode, and class structure are written directly using byteorder — no dependency on javac or ASM.

skotch emit --target jvm hello.kt -o HelloKt.class
java -cp . HelloKt

DEX

Produces Dalvik Executable format (v035). Uses the same MIR input as the JVM backend but emits DEX bytecode with its index-heavy structure (string pools, type IDs, method IDs). Written from scratch — no dependency on d8 or dx.

skotch emit --target dex hello.kt -o classes.dex

klib

Produces a .klib ZIP archive containing serialized MIR as JSON, a manifest, and source copies. This is the intermediate format used by the LLVM IR and native pipelines.

Note

Skotch klibs use a JSON-based IR format and are not interchangeable with klibs produced by kotlinc-native.

LLVM IR

Produces textual LLVM IR (version 19+). Skotch writes LLVM IR as plain formatted strings — there is no inkwell or llvm-sys dependency, which avoids the libLLVM system requirement.

The runtime is libc only: println(String) maps to puts, println(Int) maps to printf("%d\n").

skotch emit --target llvm hello.kt -o hello.ll
cat hello.ll

Native

Chains the LLVM IR backend with a clang link step to produce a host executable. This is the only place Skotch invokes an external tool.

skotch emit --target native hello.kt -o hello
./hello

Multi-file compilation

When using skotch build, the driver compiles each .kt file to a separate MIR module, then merges them (remapping string IDs to avoid collisions) into a single combined module before handing it to the backend. This allows functions in one file to call functions in another.

Validation

Every supported test fixture is compiled by Skotch and by the corresponding reference tool (kotlinc, d8, kotlinc-native). Both outputs are committed to git so CI can verify agreement without needing the JDK or Android SDK installed. Normalized text forms strip cosmetic differences (constant pool ordering, debug attributes, target triples) to avoid false positives when diffing.