Add the ability to export symbols from executables, not just dylibs, via a new compiler flag: -C export-executable-symbols.


Java and C# can't statically link against C/Rust code. Both require dylib symbols for their common native interop solution. Which is fine if you let their executables call your dylib, but is a problem if you want your Rust executable to load a JVM instance, and let it call back into your executable. You might want to do this to allow you to:

  • Load multiple language runtimes into the same process (Rust + C# + Java + Lua anyone? Only one of them can be the entry executable...)
  • Display user-friendly error messages if language runtimes are missing (maybe even a download link!)
  • #[test] Java/Rust interop via cargo test.

For this last case, I manually export executable symbols via LINK. This is ugly, brittle, and rustc already knows how to do this automatically, across more platforms, and better.

Guide-level explanation could gain:

## export-executable-symbols

This flag causes `rustc` to export symbols from executables, as if they were dynamic libraries.

You might use this to allow the JVM or MSCLR to call back into your executable's
Rust code from Java/C# when embedding their runtimes into your Rust executable.

rustc -C help could gain:

    -C    export-executable-symbols -- export symbols from executables, as if they were dynamic libraries.

My Java interop Quick Start would start recommending a .cargo/config with:

rustflags = ["-C", "export-executable-symbols"]

Reference-level explanation

On a technical level, this just involves preventing an early bailout when calling fn export_symbols on executables with MSVC or GNU linker backends. Other linker backends (EmLinker, WasmLd, PtxLinker) do not have this early bailout in the first place, and remain unaffected.


  • Options bloat
  • The burden of supporting a niche use-case in hideously platform specific code

Rationale and alternatives

This is very simple to implement, leverages existing code to enable it to do exactly what it was meant to do, and has few drawbacks.


  • Unconditionally export symbols from executables instead of introducing a new compiler flag.
  • Introduce a crate-level attribute instead of a compiler flag (#![export_all_symbols]? #![export_symbols]?)
  • Write yet another cargo subcommand to install/remember for interop testing instead of using cargo test.
  • Write interop tests exclusively as integration tests, in an entirely separate crate, that can load the testee as a dylib.
  • Continue abusing LINK, writing a tool to auto-generate .defs via build scripts - possibly by reading metadata from other tools.
  • Use nightly link-args instead of LINK, but still write a .def generator.
  • Remember to always cargo build a dylib copy of a crate manually before cargo test ing, and load that instead. (That would also add a whole second copy of all functions and static vars in the same unit test process!)

Prior art

C and C++ compilers can already do this via __declspec(dllexport) annotations. Most people don't really notice it, for good or for ill.

Unresolved questions

  • Is this a good name for it?
  • Should it be more general and export when limit_rdylib_exports or crate_type == ProcMacro?

Future possibilities

We could introduce a new source annotation, #[export]. For backwards compatibility with current behavior, #[no_mangle] symbols could be exported by default - and possibly disabled with #[export(false)]. This would reduce the need to hide this change to compiler/linker behavior behind a compiler flag or crate annotation.

Maybe other options to control what symbols get exported? Although I'd fear turning rustc into yet another linker script implementation, so maybe not.

My own building atop this in the wider language ecosystem would be for improved Java/Rust interop/testing, with the eventual goal of improved Android API support for Rust. Many APIs are only exposed via Java, and I'd like said APIs to be usable in a safe and sound fashion.