Native code generation

The native compiler selects LLVM’s host target triple and creates a TargetMachine for that target. Its data layout is attached to the generated module before IR construction, so isize and usize use the target pointer width rather than a compiler-host constant.

NativeCompiler::EmitObject runs LLVM’s legacy code-emission pass pipeline and writes a relocatable object. NativeCompiler::LinkExecutable invokes the system C++ linker driver without a shell, which supplies the platform startup objects and standard executable-linking conventions. Unix hosts prefer c++; Windows hosts prefer clang++. BuildExecutable uses a temporary object with the host object suffix and removes it after linking.

Commands:

flecha --emit-object program.fl -o program.o
flecha build program.fl -o program

The current implementation targets the machine running the compiler. Installed distributions find the version-matched runtime archive relative to the compiler executable; FLECHA_RUNTIME_LIBRARY provides a development override. Explicit cross-compilation triples, sysroots, optimization levels, and debug information are future work.

Run the native end-to-end smoke test with:

make native-poc

It compiles examples/llvm_poc.fl, confirms that both the object and executable were produced, executes the result, and requires exit code 42.