throw no more handles if a timer could not be created

[jonahgraham]

At the moment the code silently fails on calling timerExec if there are no handles left or it fails for other reasons.

Relates to #2806 where we can see that we are running out of handles, but the timerExec does not report a failure.

[jonahgraham]

This is a draft for now because it needs consideration for other platforms and whether after 25 years there is a reason that this case was allowed to fail silently on Windows.

On macOS there is a pointless null check in the same place because an NPE would have already been raised. On GTK I need to investigate more because the functions are not documented as being able to fail.

macOS:

eclipse.platform.swt/bundles/org.eclipse.swt/Eclipse SWT/cocoa/org/eclipse/swt/widgets/Display.java

Lines 5381 to 5390 in bb954ac

	NSTimer timer = NSTimer.scheduledTimerWithTimeInterval(milliseconds / 1000.0, timerDelegate, OS.sel_timerProc_, userInfo, false);
	NSRunLoop runLoop = NSRunLoop.currentRunLoop();
	runLoop.addTimer(timer, OS.NSModalPanelRunLoopMode);
	runLoop.addTimer(timer, OS.NSEventTrackingRunLoopMode);
	timer.retain();
	if (timer != null) {
	nsTimers [index] = timer;
	timerList [index] = runnable;
	}
	}

GTK:

eclipse.platform.swt/bundles/org.eclipse.swt/Eclipse SWT/gtk/org/eclipse/swt/widgets/Display.java

Lines 5750 to 5759 in 286e614

	if (GTK.GTK4) {
	timerId = OS.g_timeout_add (milliseconds, timerProc, index);
	} else {
	timerId = GDK.gdk_threads_add_timeout (milliseconds, timerProc, index);
	}
	if (timerId != 0) {
	timerIds [index] = timerId;
	timerList [index] = runnable;
	}
	}

[github-actions]

Test Results

  172 files  + 3    172 suites  +3   23m 56s ⏱️ - 3m 42s
4 679 tests +56  4 656 ✅ +73  23 💤 +4  0 ❌  - 6 
  466 runs  +38    461 ✅ +36   5 💤 +2  0 ❌ ±0 

Results for commit 74c1657. ± Comparison against base commit 35ce202.

This pull request removes 1 and adds 57 tests. Note that renamed tests count towards both.

org.eclipse.swt.tests.gtk.snippets.Bug336238_ShellSetBoundFailTest ‑ testSetBounds

AllWin32Tests ImageWin32Tests ‑ testDisposeDrawnImageBeforeRequestingTargetForOtherZoom
AllWin32Tests ImageWin32Tests ‑ testDrawImageAtDifferentZooms(boolean)[1] true
AllWin32Tests ImageWin32Tests ‑ testDrawImageAtDifferentZooms(boolean)[2] false
AllWin32Tests ImageWin32Tests ‑ testImageDataForDifferentFractionalZoomsShouldBeDifferent
AllWin32Tests ImageWin32Tests ‑ testImageShouldHaveDimesionAsPerZoomLevel
AllWin32Tests ImageWin32Tests ‑ testRetrieveImageDataAtDifferentZooms(boolean)[1] true
AllWin32Tests ImageWin32Tests ‑ testRetrieveImageDataAtDifferentZooms(boolean)[2] false
AllWin32Tests ImageWin32Tests ‑ test_getImageData_fromCopiedImage
AllWin32Tests ImageWin32Tests ‑ test_getImageData_fromImageForImageDataFromImage
AllWin32Tests TestTreeColumn ‑ test_ColumnOrder
…

This pull request skips 1 test.

org.eclipse.swt.tests.junit.Test_org_eclipse_swt_widgets_Shell ‑ test_activateEventSend

♻️ This comment has been updated with latest results.

[jonahgraham]

This is a draft for now because it needs consideration for other platforms and whether after 25 years there is a reason that this case was allowed to fail silently on Windows.

On macOS there is a pointless null check in the same place because an NPE would have already been raised. On GTK I need to investigate more because the functions are not documented as being able to fail.

My latest commit adds throwing no more handles on macos and gtk if the timer creation routines return 0/null so they also don't silently fail anymore.

I plan to submit this soon if there are no objections.