Add a keyboard mapping tab to the options screen

[tomconder]

What

Splits the options screen into two tabs, Display and Keyboard, one layer each, with a shared tab strip at the top. Switch tabs with Q/E, the gamepad shoulder buttons, or by clicking a tab.

[Q]   • Display     Keyboard   [E]

The active tab takes the red fill the menu rows use for a selected row ({0.84, 0.04, 0.04, 0.14}) and a small white dot before its label. No border, no rounded corners, so the strip stays lighter than a menu row. The dot is the shared Quad with a half-width corner radius, so it costs no new primitive.

The Keyboard tab lists the rebindable actions with their current key, plus Reset to Defaults and Return.

How rebinding works

The layer asks for a rebind by setting an atomic flag. The key capture, the binding write and the settings write all run on the main thread inside InputManager::update(), because Settings is not thread safe and menu layers run on the render thread. The capture frame resolves no actions, so the captured key cannot also fire the action it is being bound to.

Overrides persist as input.<action> and are re-applied at the end of buildDefaultBindings(). 0 means "no override", so Reset to Defaults writes zeros and the defaults survive.

Deliberate limits, marked in the source:

• Menu navigation is not rebindable. Losing it strands the player on the screen.
• Escape cancels a capture, so nothing can be bound back to Escape by hand. Reset to Defaults restores the ones that had it.
• No conflict check: two actions may share a key.
• Keyboard only. Gamepad and axis bindings are untouched.

Fix that came with the split

ExitLayer, IntroLayer and MazeLayer each asked whether the options screen was open by testing OptionLayer::isActive(). With the screen split across two layers that test read as "closed" on the Keyboard tab, so those layers woke up behind it: ExitLayer swallowed the first Escape, and from the intro menu, Enter-to-rebind would have started a new game. All five call sites now go through Maze::isOptionsOpen().

Gamepad mode no longer bounces back

Using the gamepad in a menu switched the prompts straight back to keyboard glyphs. InputManager::updateActiveDevice() claimed the device for KeyboardMouse on held key state. The Xbox 360 driver aliases the D-pad to arrow keys, so the ghost key was still down on the frame after the gamepad button released, and the device flipped on every press.

Keyboard and mouse now claim the device on a fresh press edge or real cursor motion. A key or mouse button still takes the device from the gamepad the moment you press it. The special cases for Escape and the arrow keys only ordered themselves against the gamepad check, which the generic loop already covers, so they are gone. 19 lines out, 6 in.

Assets

keyboard_q, keyboard_e, xbox_lb and xbox_rb from Kenney Input Prompts (CC0), taken from the same Default folders as the sprites already in the tree. The prompt sprite cache moved out of keyhints.cpp into game::ui::promptSprite() so the tab strip and the hint bar share one cache.

Testing

cmake --build out/build/ci-windows-release --target game and pre-commit run --all-files are both clean.

I drove the app end to end with synthesised key events, then checked the rendered frame and the observable state: the settings file, the window rect, and whether the process exits.

• Tab strip renders with the active tab filled and dotted, the inactive one dimmed, and the prompts swapping with the device. The strip clears the first row at 1920x1080.
• Tab switching. E reaches the Keyboard tab, Q returns to Display.
• Rebinding. Enter on Move Forward shows "Press a key"; K then writes input.moveForward = 75. Escape cancels and restores the previous key.
• Reset to Defaults. Clears every override back to 0.
• Overrides load on restart. Bound Full Screen to K, quit, relaunched: in game, K toggled the window 1936x1119 → 1920x1080 → back, so the stored override was re-applied over the default F.
• The Escape fix. One press from the Keyboard tab returns to the intro menu, confirmed by the next Down+Enter selecting Quit and the process exiting, which happens only when the intro menu has control.

Gamepad stickiness was measured against a physical Xbox 360 pad, since no virtual-pad driver is installed. A watcher sampled the window every 0.4 s, logging the XInputGetState button word, GetCursorPos deltas and GetAsyncKeyState, and classified the hint bar by template diff against a known keyboard frame and a known gamepad frame:

• D-pad press gave gamepad glyphs, and the mode held for 29 consecutive samples (~11.6 s) with no keyboard, mouse or pad input. Before the fix it reverted within a frame.
• The same log shows the root cause: one sample carries btn=0x0001 (D-pad Up) and GetAsyncKeyState(VK_UP) true at the same instant.
• One run did revert at once, but the cursor was moving throughout it (mouseMoved= 7, 18, 20, 296…), which is the intended switch.

I reset the bindings to defaults afterwards, so no test state is left behind.

The tab strip starts at the first row's left edge, so it lines up with the menu block rather than centring on the window. Callers pass that edge in from their own Yoga layout, which keeps the strip and the rows together on resize.

Checked at 1920x1080 windowed. Other resolutions are untested. The gamepad prompts do render, confirmed in the stickiness run above.