Window manager for properly refreshing and resizing multiple layers of windows

[Qrox]

Summary

SUMMARY: Infrastructure "Window manager for properly refreshing and resizing multiple layers of windows"

Purpose of change

Currently many game menus can be called from multiple places, such as the keybindings menu which can be opened from nearly every game menu. However, when closing some of these menus, keybindings menu included, only the previous menu is redrawn, leaving black space on the screen. Additionally, when resizing the game window, most of the menus aren't correctly resized and re-positioned.

Describe the solution

A new window manager that stores ui_adaptor objects, which UI code can use to encapsulate their initialization and redrawing code as callbacks. When closing a menu or resizing the game window, UIs are invalidated accordingly and redrawn, so all layers of windows are correctly displayed.

This PR only contains the code for the ui manager. Menus need to be migrated individually to be properly refreshed and resized. For example, #38235 migrates the main menu and several related menus using ui_manager.

Describe alternatives you've considered

If the UI code was to be designed from scratch I would use an object-oriented event-based system, but since most of the game's UI code was written to be sequentially executed, this is a solution I found least intrusive.

Testing

Compiled with #38235, triggered a popup in the main game UI, and opened its keybindings menu. Closed the keybindings menu and observed that there was not black space left behind. Resized the game window and observed that the popup was correctly placed in the center of the screen.

[Night-Pryanik]

This should close #8514 once and for all?

[Qrox]

It could, but the related menus should be first migrated to use this system. I'll migrate the menus mentioned there if it turns out to be doable in this PR.

[8street]

Please test it in fullscreen dx9 mode. I remember in fullscreen mode when the game window was maximized after alt-tabbing, an event appeared - SDL_WINDOWEVENT_FOCUS_GAINED but not SDL_WINDOWEVENT_EXPOSED.

[Qrox]

I've tested in fullscreen with direct3d and it was refreshing properly. I didn't see a dx9 mode though, so I'm not sure if that's what you mean.

[8street]

Ok.

[Qrox]

@Night-Pryanik the menus mentioned in that issue are more complex than I expected so I'll leave them to a later PR.

[Qrox]

Also, can someone test if this is working on ncurses?

[anothersimulacrum]

@Qrox Do you need anything specific for testing on curses? Just open the game play around with menus, make a new character, edit settings, make a new world, etc.?

[Qrox]

@anothersimulacrum Thanks! There are two things I would like to test on curses:

1. Refreshing when 3 or more windows are open and the last one is closed. For example, if I press ESC in the main menu, and open the keybindings menu in the exit confirmation popup, when closing the keybindings menu I expect the main menu to be refreshed instead of partly erased.
2. Resizing when 2 or more windows are open and the terminal size is changed. I expect every open windows to be resized according to the new terminal size.

On tiles I also expect the window to be refreshed when the window content is invalidated by system, such as when recovering from a sleep, but I assume on curses this is done by the terminal software and requires no action from the game.

[anothersimulacrum]

Refreshing for that first case does not seem to work

I also tested with xterm, got the same results.
Resizing with no windows open - it's a bit odd

(Recording isn't the highest quality, sorry)
Resizing with windows open does not work (safe mode manager, and then editing the proximity in it are the windows)

Though the first one does work on tiles either

Nor does the second one, it just goes black.

(resizing without having windows opens works)

[Qrox]

Have you applied the changes in #38235? That's where the actual migration takes place.

[anothersimulacrum]

No, that would probably help.

[anothersimulacrum]

Okay, with the changes to make it work applied, it works correctly.

[ZhilkinSerg]

Apparently sometimes g could be empty when there are errors (see #39258), so we need to check for non-emptiness prior to comparing win with members of g, e.g.:

    window_dimensions dim;
    dim.scaled_font_size.x = fontwidth;
    dim.scaled_font_size.y = fontheight;
    if( g ) {
        if( use_tiles && win == g->w_terrain ) {
            // tiles might have different dimensions than standard font
            dim.scaled_font_size.x = tilecontext->get_tile_width();
            dim.scaled_font_size.y = tilecontext->get_tile_height();
        } else if( map_font && win == g->w_terrain ) {
            // map font (if any) might differ from standard font
            dim.scaled_font_size.x = map_font->fontwidth;
            dim.scaled_font_size.y = map_font->fontheight;
        } else if( overmap_font && win == g->w_overmap ) {
            dim.scaled_font_size.x = overmap_font->fontwidth;
            dim.scaled_font_size.y = overmap_font->fontheight;
        }
    }