static const& variable within function leads to exception on exit

[semushin-s]

Version of emscripten/emsdk:

emcc (Emscripten gcc/clang-like replacement + linker emulating GNU ld) 4.0.10 (b7dc6e5747465580df5984e723b9d1f10d8e804b)
clang version 21.0.0git (https:/github.com/llvm/llvm-project 8f7e57485ee73205e108d74abb5565d5c63beaca)
Target: wasm32-unknown-emscripten
Thread model: posix
InstalledDir: C:\devel\emsdk\upstream\bin

Compile the code below as follows:
em++ main.cpp -sEXIT_RUNTIME=1 -o main.html

#include <iostream>
#include <string>

const std::string& f()
{
  static const std::string& x = {};
  return x;
}

int main()
{
    std::cout << f() << '\n';
}

Running in browser results in the following error in console:

Uncaught RuntimeError: null function or function signature mismatch
    at main.wasm:0xf65
    at exitRuntime (main.js:1:3158)
    at exitJS (main.js:1:57013)
    at callMain (main.js:1:58576)
    at doRun (main.js:1:58955)
    at main.js:1:59092

Seems like for other targets this code compiles and runs fine, so it shouldn't actually be UB

[kripken]

Removing the & from

  static const std::string& x = {};

fixes things.

I'm not enough of an expert on C++ UB, but yeah, that might not be UB (using lifetime extension). UBSan finds nothing.

Might be worth trying to reduce this to a testcase not using std::string and iostream, as those make the wasm quite large and hard to debug.

[semushin-s]

Could be simplified to the following:

struct C {
    ~C(){ t = 1; }
    int t;
};

void f()
{
  static const C& x = {};
}

int main()
{
    f();
}

[sbc100]

Even simpler:

struct C {
  ~C(){ t = 1; }
  int t;
};

const C& x = {};

int main() {
}

[sbc100]

Interesting. I looks like ___cxa_atexit (and the whole atexit mechanism in system/lib/libc/musl/src/exit/atexit.c assumes that the callback functions take a void* and return void.

However, the C++ destructor in this case (C::~C()) is returning a void*. I guess normally its not the actual destructor itself that is registered by rather __cxx_global_array_dtor .. which the compiler generates and then calls C::~C().