6.7 samples - #963

Ran the six samples on this branch to capture assets for the 6.7 post. All six build 0 warnings / 0 errors in Debug and Release, and five behave exactly as described. Two reproducible issues in SystemBackdropShowcase, both of which look like Window.SystemBackdrop behaviour on the Skia Win32 head rather than anything in the sample code.

Environment: Uno.Sdk 6.7.0-dev.154 (the branch global.json), net10.0-desktop, Release, Windows 11 build 26200. Measurements are mean RGB and distinct colour count over a 320x60 region of empty page ground, captured with PrintWindow(PW_RENDERFULLCONTENT).

1. Once the backdrop is set to None, no material renders again

On a fresh window, Mica and Desktop Acrylic both render. Select None, then select either material again, and the window stays opaque white for the rest of its life.

Switching Mica to Desktop Acrylic directly works every time, so the trigger is specifically the trip through SystemBackdrop = null. The sample's subtitle promises "Pick a material and this window changes live", which holds until someone picks None and then stops holding.

2. While a backdrop is active the client area is never cleared, so repaints stack

Resizing the window leaves the previous layout painted underneath, and it compounds: after one resize the text is doubled, after a second resize there are three overlapping copies. The same resize sequence with None selected renders perfectly clean, and the other five samples resize cleanly with no backdrop involved.

It is on screen rather than in the capture. PrintWindow and a direct screen copy show it identically. Selecting None repaints opaquely and clears it instantly.

My resizes came from SetWindowPos. A user drag-resize goes through the same WM_SIZE path, but I have not confirmed that by hand.

One note for anyone else capturing this sample

The material needs a moment to compose after the window appears or after a selection changes. A screenshot taken too early reads as flat #F3F3F3 for every option, which looks exactly like the material failing to render. Allowing time to settle before capturing avoids chasing a bug that is not there. This cost me a while, so it seemed worth writing down.

Happy to share the screenshots and the measurement scripts, or to open these as issues on the main repo if that is a better home for them.

@mtmattei thanks for the measurements. Both are framework bugs, as you suspected, but only one of them is fixed so far.

1. No material after a trip through None. unoplatform/uno#24810 (6.7) / #24806 (6.8) should cover this. Until now, setting a backdrop walked the visual tree, swapped every opaque background for a transparent one, and tried to restore them on null. The None round trip goes through that save/restore bookkeeping. Those PRs remove the walk completely. The island root now switches between three modes (none / transparent / fallback) on each assignment, so None followed by Mica just flips it back to transparent. I haven't run your exact None -> Mica sequence against that build yet, so please re-measure once it ships.

There's a catch that affects this sample: once the walk is gone, Uno stops rewriting app backgrounds, the same as WinUI, so an opaque page hides the material. The page had ApplicationPageBackgroundThemeBrush, so with the fix every option would have looked like flat #F3F3F3. Fixed in 75a1d4e: the page is now Background="Transparent", and the explanation text no longer says Uno makes your backgrounds transparent. This works with the current 6.7.0-dev.154 SDK too.

2. Stacked repaints on resize while a backdrop is active. Not covered by #24810. This is the damage-region / retained-layer blit compositing over its own previous frame once the root is transparent. That's why None hides it: an opaque root covers the stale pixels. It's the Win32 counterpart of unoplatform/uno#24495 (macOS). unoplatform/uno#24546 fixes the shared RetainedLayer blit, but the Win32 OpenGL renderer has its own inline copy that it doesn't touch yet, and master's fix(hosts): clear the frame to the window's background again isn't on 6.7. Your SetWindowPos repro is a nice clean one, so feel free to drop it on #24495 or in a new Win32 issue on the main repo.

The tip about letting the material settle before capturing is a good one. A flat #F3F3F3 screenshot is the most convincing fake bug there is.

🤖 Addressed PR comments up to this point

• @mtmattei: SystemBackdropShowcase None round trip + resize ghosting. Sample page made transparent in 75a1d4e. Bug fixes #1 should be covered by fix(win32): Make SystemBackdrop work and stop rewriting app backgrounds (6.7) uno#24810 / #24806; Add CI #2 is a separate Win32 blit bug (see reply above).