[bronze-ff]

Add a Quit row to the Insert menu's Gameplay tab. The first activation shows Confirm. A second activation closes the game through the normal window-close path. B or Esc cancels the confirmation. The same action is available with Enter and mouse clicks.

This addresses the in-menu exit requested in #1203 and tracked in #1007.

Validation

• Built from source and ran the game on CachyOS with Hyprland/Wayland, SDL3, and an RTX 5060.
• Used a virtual USB Xbox Series controller through Linux uinput. View + Menu opened the overlay, RB selected Gameplay, and the D-pad selected Quit. Holding A left the game running with Confirm displayed. B cancelled it. Two separate A presses then closed the game with exit code 0 through the normal window-close handler.
• Passed the three focused source regressions for menu hit testing, the rumble menu, and background input on Linux. Steam Game Mode and a physical controller still need a hardware check.

Assisted by OpenAI Codex.

[hckrman101]

Ah I was also going to add this. Can we also add a 'Restart' option?

[hckrman101]

Nice, this is a clean reuse of the existing close path. It posts WM_CLOSE to the window, which goes through rx_close_asked() and hos_window_closed_exit() just like the X button (and like Restart now, minus the relaunch). That also works on Windows with no extra change, since display.c already handles WM_CLOSE. The new Row line fits main's Row layout, the confirm state is cleared on B/Esc, on moving off the row, switching tab and closing the menu, and nothing touches the ini or presets. CI is green and it merges cleanly.

Two small nits, nothing blocking:

• Restart now calls save() and pipeline_cache_save(true) before it posts its message. Quit goes straight to WM_CLOSE. If the normal window close already covers that, fine, but it is worth a quick look.
• kQuitRow = kCheatsRow - 1 relies on Quit staying right before Cheats in the table. A comment or a static_assert on the row name would keep a later reorder from silently moving it.

[hckrman101]

Restart now does save() + pipeline_cache_save(true) before posting its message. Does the plain WM_CLOSE path already cover that, or should Quit do the same here?

[bronze-ff]

Good catch, it doesn't. The plain close only runs rx_close_asked(), which stops a shader preparation still under way, and then hos_window_closed_exit() exits. Nothing forces the pipeline cache, so it is only written by the every-300-presents save in vk_replay.cpp. Settings were already safe, since set_row saves each change.

a7b27de makes Quit call save() and pipeline_cache_save(true) before WM_CLOSE, the same as Restart now. On CachyOS, mkw_vk_pipelines.bin was rewritten in the same second as the Quit exit.

[bronze-ff]

Thanks for the review! Both points are in a7b27de:

• Saving first: Quit now does save() and pipeline_cache_save(true) before WM_CLOSE, the same as Restart now. Details are in the inline thread.
• Row order: a static_assert on the name won't compile here, because rows[] isn't constexpr (its rows point at opts). A comment by the table now says Quit and Cheats stay its last two rows, since kQuitRow and kCheatsRow count back from its end.

Re-tested on CachyOS (RTX 5060, Hyprland) with a virtual Xbox Series controller, on a build of main a1c4bcd plus this branch. The menu tests pass (test_menu_hit, test_menu_restart, test_rumble_menu, test_input_background). In the game, View + Menu, RB and D-pad up reach Quit. Holding A shows Confirm without quitting, B cancels, and A twice exits with code 0.

About the Restart option you asked for: I'll send it as a separate PR, since CONTRIBUTING asks for one topic per pull request. It will reuse the same confirm step and the restart question's Restart now path (WM_APP + 9).

[bronze-ff]

The Restart option is now #1266. It is built on this branch, and only its last commit is new.

[hckrman101]

Thanks, both covered: Quit now saves the settings and forces the pipeline cache before closing, like Restart now, and the comment by the table pins the row order. LGTM.