FIX Prevent segfaults in CLI runner (pyodide#4836)

After many hours of debugging, I minimized the problem down to this: emscripten-core/emscripten#22052

Thanks to @henryiii for reporting. See thread here for my comments while I was debugging:
scikit-hep/boost-histogram#938

Resolves pyodide#2964.

docs/project/changelog.md

@@ -39,6 +39,10 @@ myst:
   "Emscripten" and not as "emscripten".
   {pr}`4812`
 
+- {{ Fix }} Resolution of JavaScript symbols in dynamic libraries doesn't fail
+  anymore.
+  {pr}`4836`
+
 ### Packages
 
 - New Packages: `pytest-asyncio` {pr}`4819`

packages/cpp-exceptions-test2/meta.yaml

@@ -0,0 +1,11 @@
+package:
+  name: cpp-exceptions-test2
+  version: "1.0"
+  tag:
+    - core
+    - pyodide.test
+source:
+  path: src
+build:
+  cxxflags: -fexceptions
+  ldflags: -fexceptions

packages/cpp-exceptions-test2/src/cpp_exceptions_test2.cpp

@@ -0,0 +1,13 @@
+#include "Python.h"
+#include <stdexcept>
+
+extern "C" __attribute__((visibility("default"))) PyObject*
+PyInit_cpp_exceptions_test2()
+{
+  try {
+    throw std::runtime_error("something bad?");
+  } catch (const std::exception& e) {
+    PyErr_SetString(PyExc_ImportError, "oops");
+  }
+  return nullptr;
+}

packages/cpp-exceptions-test2/src/setup.py

@@ -0,0 +1,16 @@
+from setuptools import Extension, setup
+
+setup(
+    name="cpp-exceptions-test2",
+    version="1.0",
+    ext_modules=[
+        Extension(
+            name="cpp_exceptions_test2",  # as it would be imported
+            # may include packages/namespaces separated by `.`
+            language="c++",
+            sources=[
+                "cpp_exceptions_test2.cpp"
+            ],  # all sources are compiled into a single binary file
+        ),
+    ],
+)

src/core/error_handling.c

@@ -119,13 +119,13 @@ EM_JS(JsVal, restore_stderr, (void), {
 /**
  * Wrap the exception in a JavaScript PythonError object.
  *
- * The return value of this function is always a valid hiwire ID to an error
- * object. It never returns NULL.
+ * The return value of this function is always a JavaScript error object. It
+ * never returns null.
  *
  * We are cautious about leaking the Python stack frame, so we don't increment
  * the reference count on the exception object, we just store a pointer to it.
- * Later we can check if this pointer is equal to sys.last_exc and if so
- * restore the exception (see restore_sys_last_exception).
+ * Later we can check if this pointer is equal to sys.last_exc and if so restore
+ * the exception (see restore_sys_last_exception).
  *
  * WARNING: dereferencing the error pointer stored on the PythonError is a
  * use-after-free bug.

src/js/load-package.ts

@@ -555,6 +555,10 @@ export async function loadPackage(
       Array.from(toLoad.values()).map(({ installPromise }) => installPromise),
     );
 
+    // Warning: this sounds like it might not do anything important, but it
+    // fills in the GOT. There can be segfaults if we leave it out.
+    // See https://github.com/emscripten-core/emscripten/issues/22052
+    // TODO: Fix Emscripten so this isn't needed
     Module.reportUndefinedSymbols();
     if (loadedPackageData.size > 0) {
       const successNames = Array.from(loadedPackageData, (pkg) => pkg.name)

src/templates/python

@@ -142,6 +142,11 @@ async function main() {
             console.error(e);
         }
     }
+    // Warning: this sounds like it might not do anything important, but it
+    // fills in the GOT. There can be segfaults if we leave it out.
+    // See https://github.com/emscripten-core/emscripten/issues/22052
+    // TODO: Fix Emscripten so this isn't needed
+    py._module.reportUndefinedSymbols();
 
     py.runPython(
         `
@@ -176,14 +181,10 @@ async function main() {
     try {
         errcode = py._module._run_main();
     } catch (e) {
-        // If someone called exit, just exit with the right return code.
         if(e.constructor.name === "ExitStatus"){
             process.exit(e.status);
         }
-        // Otherwise if there is some sort of error, include the Python
-        // tracebook in addition to the JavaScript traceback
-        py._module._dump_traceback();
-        throw e;
+        py._api.fatal_error(e);
     }
     if (errcode) {
         process.exit(errcode);

src/tests/test_cmdline_runner.py

@@ -502,3 +502,19 @@ def test_sys_exit(selenium, venv):
     assert result.returncode == 12
     assert result.stdout == ""
     assert result.stderr == ""
+
+
+def test_cpp_exceptions(selenium, venv):
+    result = install_pkg(venv, "cpp-exceptions-test2")
+    print(result.stdout)
+    print(result.stderr)
+    assert result.returncode == 0
+    result = subprocess.run(
+        [venv / "bin/python", "-c", "import cpp_exceptions_test2"],
+        capture_output=True,
+        encoding="utf-8",
+    )
+    print(result.stdout)
+    print(result.stderr)
+    assert result.returncode == 1
+    assert "ImportError: oops" in result.stderr