OpenRCT2: the blue screenshots and the console defaults (2026-10-27)

Report. In the data-pilot run (docs/TESTS.md, 10-27 17:43) OpenRCT2 0.5.6+ds-1+ps3.889 with roller-coaster-tycoon2-data 89 (game-data-packager’s import of the user’s CD) showed the title, the windows and Electric Fields in blue and purple in every xwd screenshot, while the 10-19 smoke test (the old rct2-data, build ps3.574) had looked right. It also started in a 1280x720 window with the changelog open.

The colours: the game draws right, the screenshot is wrong

Nothing was wrong with the game, the data or the renderer. Same build, same data, same config (drawing_engine = SOFTWARE_HWD, the default a fresh config gets), started windowed:

• the webcam photo of the TV shows the right colours (turquoise water, green buttons, red-and-yellow helter-skelters), while the xwd taken at the same moment is blue and purple;
• fullscreen (fullscreen_mode = 2, scale 2) the xwd is right too. The 10-19 smoke test ran fullscreen and was judged on the webcam, so it never saw the problem. That’s why it “looked right”.

The cause is the window’s X visual. SDL’s GL window gets visual 0x8d, a DirectColor GLX visual whose masks are byte-swapped: red 0xff00, green 0xff0000, blue 0xff000000. The root’s TrueColor visual has 0xff0000/0xff00/0xff. Twelve DirectColor and eighteen TrueColor visuals the X server lists (xdpyinfo) carry the swapped masks: these are the GLX visuals made from Mesa’s configs on this big-endian host. The pixels in the window are in the root’s layout anyway (X server and scanout copy them unchanged between same-depth drawables, so the TV is right). Only a client that reads the window and decodes it through the visual gets it wrong. xwd does exactly that, -root included (it reads each visual’s region through its own masks and colormap):

• an xwd -id <window> dump decoded with the visual’s masks: pixel ff63bbbb becomes (0xbb, 0x63, 0xff), purple; with the root’s masks it is (0x63, 0xbb, 0xbb), the water’s turquoise. The whole dump decoded with the root’s masks is the right picture;
• the earlier gdp-step1 note “windowed games’ xwd captures came out blue (R/B), fullscreen ones right” is the same thing, for every windowed GL game.

So the data (g1.dat matches gdp’s checksum; it draws right) and the build are fine. No game patch. Consequences:

• screenshots of windowed GL games on the PS3 lie. Judge them on the webcam or capture fullscreen. The harness’s shots (xwd -root, bench/harness/client.sh) are right for our fullscreen recipes;
• the masks are a big-endian bug of the GLX visuals (Mesa’s configs as the X server turns them into visuals: a candidate for the driver track, not fixed here).

The console defaults: patch 0007

On a console the game is started from the menu on a TV. Upstream’s defaults made a fresh configuration start in a 1280x720 window with the changelog over the title screen (last_run_version is empty on a fresh config, which upstream treats as an upgrade). 0007 changes only what a configuration without these settings gets:

• fullscreen_mode defaults to 2, “Fullscreen (desktop)”: SDL’s fullscreen at the desktop’s resolution, read at run time (720p TV 720p, 1080p TV 1080p);
• the changelog opens only after an upgrade (last_run_version set and older), not on the first start;
• the first start’s display check (InferDisplayDPI, once, then saved) also picks the window scale for a TV: desktop height / 540 on a desktop of 1080 lines or more (2 at 1080p: a 960x540 canvas that SDL scales up), else 1. The software renderer paints the whole canvas on the PPE each frame; scale 2 is the setting the benchmark recipe measured (21-24 fps in Electric Fields) and a quarter of 1080p’s pixels.

A configuration that already has the settings keeps them. The player’s Options choices are never overridden. Controllers are not part of this patch: OpenRCT2 0.5.6 has its own gamepad support (scrolling with a stick), not checked yet.

Verified on the PS3

openrct2 0.5.6+ds-1+ps3.940~dev (0001-0007) with roller-coaster-tycoon2-data 89, a fresh user directory (--user-data-path, the object and scenario indexes copied in, no config.ini), started from the desktop’s slice: the window came up 1920x1080 fullscreen, the config it wrote says fullscreen_mode = 2, window_scale = 2.000000, last_run_version = "0.5.6", and no changelog opened. xwd and webcam agree on the colours: the title (turquoise water, green buttons), RCT2’s red scenario list, Electric Fields with its objective window, the Financial Summary and Research & Development windows. The player’s ~/.config/OpenRCT2/config.ini (written by the data-pilot run’s windowed start) was set aside as config.ini.data-pilot-20261027, so the next start from the menu gets the new defaults.