error: fix macOS terminate hang + dump ObjC NSException details

[lugt]

Summary

Two related fixes for the terminate handler on Apple platforms, so uncaught exceptions (especially ObjC NSExceptions crossing the C++ boundary) crash with a non-zero exit and are diagnosable. Currently on macOS they hang forever in an infinite loop.

1. std::terminate() → std::abort() (the real bug)

On macOS the default terminate handler is libobjc's _objc_terminate(). terminate_handler restores it (std::set_terminate(old_terminate_handler)) then calls std::terminate(). But _objc_terminate does:

type = __cxa_current_exception_type();
if (type) __cxa_rethrow();   // rethrow the in-flight exception

The rethrow_exception(e_ptr) + catch(...) dispatch earlier in terminate_handler does not clear the exception's __cxa "uncaught" state, so _objc_terminate sees an in-flight exception, __cxa_rethrows it, the rethrow is again uncaught → std::terminate → _objc_terminate → __cxa_rethrow → ... an infinite std::terminate ↔ __terminate loop instead of aborting.

This affects all uncaught exceptions on macOS (verified with a plain std::runtime_error), not just ObjC. On Linux the default handler aborts directly without inspecting the exception, so it was never visible there.

Fix: std::abort() terminates unconditionally (SIGABRT, exit 134). On other platforms the default handler already aborts, so behaviour is equivalent.

2. Dump ObjC NSException details from catch(...)

An NSException thrown across a C++ boundary does not derive from std::exception and lands in catch(...) with its name/reason/callStack lost.

• install_objc_exception_tracker() installs an objc_setExceptionPreprocessor that records the most recent NSException and its throw-time call stack ([NSThread callStackSymbols]) per thread.
• dump_objc_exception() prints them from the catch(...) arm.
• Recording at throw time (rather than rethrowing inside catch(...)) avoids disturbing the C++ exception runtime — an earlier rethrow-based attempt (objc_exception_rethrow / std::rethrow_exception + @catch) turned out to be the same class of bug as Age of Empires II (and Definitive Edition) is being preferred over openage #1.

The .mm is built only on Apple (enable_language(OBJCXX) guarded by APPLE); other platforms get no-op stubs.

Testing (macOS 26.3, Apple Silicon; Qt 6.11, LLVM clang 22)

scenario	before	after
uncaught std::runtime_error	infinite loop (hang)	FATAL + exception info + stack, exit 134
uncaught ObjC NSException	infinite loop (hang)	FATAL + name/reason/callStack, exit 134

lldb confirmed the loop root cause: old_terminate_handler resolves to libobjc.A.dylib_objc_terminate(), which __cxa_rethrow`s the in-flight exception.

Default-handler baseline (no libopenage linked): an uncaught ObjC NSException aborts with exit 134 via ObjC's own uncaught-exception path — confirming libopenage's terminate_handler is what introduced the hang.

Changes

• libopenage/error/handlers.cpp — std::terminate() → std::abort(); call dump_objc_exception() in catch(...); install/uninstall the tracker alongside the existing SIGSEGV/terminate handlers.
• libopenage/error/objc_exception.h / .mm — install_objc_exception_tracker() / uninstall_objc_exception_tracker() / dump_objc_exception().
• CMakeLists.txt — enable_language(OBJCXX) on Apple.
• libopenage/error/CMakeLists.txt — build objc_exception.mm on Apple.

Relation to PR #1812

Independent — based on master (9a5a7ccb), zero file overlap with #1812. Can be merged separately, in either order.

🤖 Generated with Claude Code