Linux: pet chat/talk buttons clipped by the window shape at fractional display scaling

[pinoybear]

On KDE Plasma with the display at 115%, the chat button on the pet doesn't show at all, and the talk button is cut off along its top edge. At 100% both show normally, chat on top and talk below it. I got the same result with two different pets (the default and Purl) and at two pet sizes, so it looks like the scaling.

The buttons look like they're drawn but clipped by the X11 window shape from applyLinuxPetWindowShape: the talk button is cut along a straight edge even with the pet in the middle of the screen. At 115% Electron reports scaleFactor=1.15625, so my guess is the shape's pet hitbox and the pet hitbox in the page end up slightly out of line at fractional scale, and the top of the button column falls outside the shape. That would explain chat (the first slot) being completely gone and talk (the second) only partly cut.

From the log at 115% with pet scale 0.5:

linux window shape applied … scale=0.5 hudScale=1.1 scaleFactor=1.15625 shape=[{"x":"114","y":"311","width":"176","height":"185"}, …]

Reproduce

1. KDE Plasma Wayland session (OpenPets on XWayland with the default --ozone-platform=x11), display scale set to 115%.
2. Control Center → Assistant → Chat & Voice, turn on Chat button and Talk button.
3. With no status bubble showing, the chat button isn't visible and the talk button is cut off at the top.
4. Set the display scale to 100% and restart OpenPets: both buttons show.

Environment: CachyOS, kernel 7.2.7-1-cachyos, KDE Plasma 6.7.5 / KWin 6.7.5 (Wayland session), xorg-xwayland 24.1.13, OpenPets 4.0.0 (openpets-bin from the AUR, built from the v4.0.0 .deb), 2880x1920 display. My other laptop at 100% (main f24f934) shows both buttons fine.

Happy to test a fix at 115% if that helps.

[pinoybear]

Correction to my guess above: I measured the actual page layout against the applied window shape, and the shape math is correct, not mismatched.

Converting the pet page's on-screen positions (from the browser, in DIP) to physical pixels using the reported scaleFactor=1.15625 reproduces the exact shape logged (x=112, y=308, width=176, height=185) almost exactly. The chat button's own rectangle also sits entirely inside that hitbox rectangle in DIP coordinates, so by the numbers it should already be covered by the shape.

But it's still visibly clipped on screen. So this isn't the shape being computed wrong from the page — it looks more like a mismatch between the shape setShape() is told to apply and how X11/XWayland actually renders/positions that mask at this fractional scale. I also tested setShape() directly with a minimal non-OpenPets window (no pet code at all) and got the shaped region landing slightly off from where the math predicts, by roughly the same amount regardless of whether the window was focusable — so it doesn't look tied to #228/#234's focusable change either.

Not sure this is a quick fix. Happy to help test anything if it'd be useful.