Repository navigation
Conversation
The popup movement test was checking the popup’s position right after moving its parent, but Qt doesn’t actually reposition it until the next polish pass. So I changed those checks to wait for that, and left the expected coordinates as they were. There’s also a quirk with Qt’s `minimal` backend: it updates window geometry but never sends a move notification. The test now sends that notification itself and then processes the popup’s pending polish. No production code was touched. To validate, I ran all nine tests on both `offscreen` and `minimal`, and they pass. Formatting is clean too. Linting the full file gives 13 diagnostics, but the upstream baseline has the same 13, so this change doesn’t introduce any new ones.
AI-generated CI validation evidenceFollow-up commit: 8a79e95. The existing Archlinux job now enables Verified results:
Reproduction, with the dependencies listed in the Archlinux workflow installed: cmake -GNinja -B build -DVENDOR_CPPTRACE=ON -DBUILD_TESTING=ON
cmake --build build
QT_QUICK_BACKEND=software QT_QPA_PLATFORM=offscreen ctest --test-dir build -R '^popupwindow$' --no-tests=error --timeout 60 --verbose
QT_QUICK_BACKEND=software QT_QPA_PLATFORM=minimal ctest --test-dir build -R '^popupwindow$' --no-tests=error --timeout 60 --verboseThe anchored filter and This establishes compilation and automated headless regression execution, not graphical desktop/compositor behavior. The earlier local nine-target test results are separate evidence; this CI step executes only popupwindow on two backends. No new local C++ test run is claimed. |
The popup movement test was checking the popup’s position right after moving its parent, but Qt doesn’t actually reposition it until the next polish pass. So I changed those checks to wait for that, and left the expected coordinates as they were.
There’s also a quirk with Qt’s
minimalbackend: it updates window geometry but never sends a move notification. The test now sends that notification itself and then processes the popup’s pending polish. No production code was touched.To validate, I ran all nine tests on both
offscreenandminimal, and they pass. Formatting is clean too. Linting the full file gives 13 diagnostics, but the upstream baseline has the same 13, so this change doesn’t introduce any new ones.