On Mon, 16 Jun 2025 16:24:40 GMT, Kevin Rushforth <k...@openjdk.org> wrote:
>> Originally this issue was supposed to resolve problems with some system >> tests (`MenuDoubleShortcutTest`, `TextAreaBehaviorTest` and others) failing >> on my Windows machine. In the process of figuring this out I found out the >> problem is Windows `::SetForegroundWindow()` API refusing to give focus to >> JFX Stage upon calling `Stage.show()`. >> >> The problem happened only when running system tests via Gradle, and with >> more investigation it turned out the culprit is actually running tests via a >> Gradle Daemon, which is the default behavior. According to >> [SetForegroundWindow API >> remarks](https://learn.microsoft.com/en-us/windows/win32/api/winuser/nf-winuser-setforegroundwindow) >> there is a list of conditions a process must meet to be granted a privilege >> of receiving focus, which is supposed to prevent focus stealing. While we do >> meet the required conditions, we don't meet "one of" additional conditions >> listed in the reference: >> - Gradle daemon is a background process, so tests started by it do not meet >> `The calling process is the foreground process.` and `The calling process >> was started by the foreground process.` conditions >> - We most probably run the tests from the terminal, so `There is currently >> no foreground window, and thus no foreground process.` condition fails - the >> foreground window is the Terminal itself. >> - Each test has fresh-started JFX stage so `The calling process received the >> last input event.` condition cannot be met and would require either Robot >> workarounds or manual interaction before each test case. >> - There is no debugger involved in the process (at least most of the time) >> so `Either the foreground process or the calling process is being debugged.` >> is also not met. >> >> As such, Windows refuses to grant JFX Stage focus, which fails some system >> tests relying on it. >> >> While we cannot remedy these conditions in-code (outside of hacky solutions >> I found with `AttachThreadInput` API which I am not a fan of) the only >> solution seems to be running the tests on Windows via either `gradle >> --no-daemon` or by setting `org.gradle.daemon=false` property somewhere in >> `gradle.properties`. >> >> In the process of debugging this problem I wrote a canary test to detect >> whether a Stage receives focus right after calling `show()`. I ran this test >> on all (accessible to me) platforms (Windows, Linux, macOS) - on both Linux >> and macOS the test passes regardless of whether the Gradle deamon is used or >> not. On my Windows machine (Win 11 24H2) it fails when testing... > > tests/system/src/test/java/test/robot/javafx/stage/StageFocusTest.java line > 56: > >> 54: static final double STAGE_Y = 100; >> 55: >> 56: static boolean receivedEvent = false; > > Should this be volatile? It is set on one thread and read from another; > unless you know that this is synchronously set by the robot key press call, I > don't think you can be sure that the write `happens-before` the read. or better an `AtomicBoolean` > tests/system/src/test/java/test/robot/javafx/stage/StageFocusTest.java line > 113: > >> 111: robot.keyPress(KeyCode.A); >> 112: }); >> 113: assertTrue(receivedEvent, "Expected key press has NOT been >> received! Stage did not have focus after showing. Some tests might fail >> because of this." + > > Do you need to sleep before the assertion to ensure that the event has been > delivered? Perhaps we ought to have a utility (latch?) to ensure the sequence? ------------- PR Review Comment: https://git.openjdk.org/jfx/pull/1804#discussion_r2150522663 PR Review Comment: https://git.openjdk.org/jfx/pull/1804#discussion_r2150522303
