App::Shutdown clears m_systems before detaching RED4ext's own hooks

[JustinMDotNet]

Summary

App::Shutdown clears m_systems while RED4ext's own game-side detours are still attached. The detours are not removed until App::Destruct, which is a separate later call. Any in-flight hook callback that runs in that window calls App::Get()->GetXxxSystem(), which evaluates m_systems.at(...) on an empty vector and throws an uncaught std::out_of_range straight into the game.

Where

• src/dll/App.cpp:166-180 — Shutdown walks systems in reverse, then m_systems.clear();.
• src/dll/App.cpp:116-144 — Destruct runs a separate DetourTransaction to remove the hooks. It only runs from the normal exit path; QuickExit does not call it.
• src/dll/App.cpp:182-210 — every GetXxxSystem() does m_systems.at(static_cast<size_t>(...)), which throws on an empty vector.
• src/dll/Hooks/QuickExit.cpp:15-36 — _QuickExit_fnc calls app->Shutdown() and then chains the original quick_exit. Hooks remain installed for the rest of process teardown.

Hooks that demonstrably reach back into App::GetXxxSystem():

• src/dll/Hooks/InitScripts.cpp:27 — App::Get()->GetScriptCompilationSystem()
• src/dll/Hooks/LoadScripts.cpp:20 — same
• src/dll/Hooks/CGameApplication.cpp:19-77 — States::*::Attach calls (the state hooks indirectly need the systems)
• Anything plugin-installed via HookingSystem that calls back into App::Get() from the detour body

Why it matters

Scenario:

1. quick_exit fires on the main thread.
2. _QuickExit_fnc calls App::Get()->Shutdown(). m_systems is now empty.
3. Before the chained quick_exit actually returns, another thread (or a queued continuation on the same thread) hits one of the still-installed detours, e.g. _CBaseEngine_LoadScripts.
4. The detour calls App::Get()->GetScriptCompilationSystem() → m_systems.at(static_cast<size_t>(ESystemType::Script)) → std::out_of_range.
5. No try/catch on the game side. std::terminate, ugly exit dialog, "RED4ext crashed the game on quit" report from the user.

Suggested fix

Two reasonable options, both fine:

1. Detach hooks first. Move the DetourTransaction + Hooks::*::Detach() block from the top of App::Destruct to the top of App::Shutdown. Then the hooks are gone before any system is shut down.
2. Make accessors safe. Change every GetXxxSystem() to return nullptr when m_systems is empty (or when m_systems.size() <= index), and audit every detour body to handle a null system. This is a bigger change but also defends against any future bug in this area.

Option 1 is simpler and fixes the actual ordering bug. Option 2 is defense in depth.

Severity

High. Reachable from a normal quick_exit path. Bug surfaces as "the game crashes on quit," which is exactly the kind of thing users blame the loader for whether or not it is RED4ext's fault.