mattjala · Oct 2, 2026
@@ -45,7 +45,7 @@ jobs:
   build-and-test:
     name: Build and Test (Linux)
     runs-on: ubuntu-latest
-    timeout-minutes: 15
+    timeout-minutes: 130
 
     strategy:
       fail-fast: false
@@ -252,7 +252,9 @@ jobs:
         hdf.lib.dir=${{ env.HDF4LIB_PATH }}/lib
 
         # Platform Configuration
-        platform.hdf.lib=${{ env.HDF5LIB_PATH }}/lib
+        # Both library directories. This value becomes -Djava.library.path for the test
+        # forks, which replaces the JVM's default.
+        platform.hdf.lib=${{ env.HDF5LIB_PATH }}/lib:${{ env.HDF4LIB_PATH }}/lib
 
         # CI-specific settings
         ci.build=true
@@ -282,10 +284,14 @@ jobs:
 
     - name: Set up Xvfb for Headless GUI Testing
       run: |
-        echo "Installing Xvfb and GTK dependencies for headless SWT testing..."
+        echo "Installing Xvfb, GTK and a window manager for headless SWT testing..."
         sudo apt-get update
-        sudo apt-get install -y xvfb libgtk-3-0 libgtk-3-dev x11-xserver-utils
-        echo "Xvfb and GTK installed successfully"
+        # metacity: SWTBot resolves widgets against Display.getActiveShell(), and on a bare X
+        # server with no window manager nothing ever becomes active, so every lookup that waits
+        # on shell activation would time out. x11-utils supplies xdpyinfo and xprop for the
+        # readiness checks below.
+        sudo apt-get install -y xvfb libgtk-3-0 libgtk-3-dev x11-xserver-utils x11-utils metacity
+        echo "Xvfb, GTK and metacity installed successfully"
 
     - name: Run Tests
       run: |
@@ -300,9 +306,8 @@ jobs:
         # Skip repository install (tries to install files that don't exist in CI)
         mvn install -B -pl object,hdfview -DskipTests -Ddependency-check.skip=true
 
-        # Run object tests only (all 149 tests passing)
-        # HDFView UI tests disabled in CI - require real display, not Xvfb
-        # UI tests should be run locally before merging
+        # Object module tests. These are blocking - they have passed on all three
+        # OSes for a long time and a regression here is a real failure.
         mvn test -B -pl object -Dmaven.test.failure.ignore=false
         echo "::endgroup::"
 
@@ -323,13 +328,95 @@ jobs:
         echo "Target artifacts:"
         find target -name "*.jar" -type f 2>/dev/null | head -5 || echo "No target JARs found"
 
+    - name: Run UI Tests
+      # TBD: This step will not be able to fail the job until
+      # the underlying bugs that CI is detecting are fixed.
+      # Once that is done, remove continue-on-error.
+      continue-on-error: true
+      # Must stay above the surefire fork timeout and below the job timeout.
+      timeout-minutes: 90
+      run: |
+        echo "::group::Run UI Tests (SWTBot under Xvfb)"
+        # Started by hand rather than through xvfb-run because the window manager has to join
+        # the same display, which xvfb-run's single-command wrapper does not allow.
+        Xvfb :99 -screen 0 1280x1024x24 -nolisten tcp &
+        export DISPLAY=:99
+
+        for i in $(seq 1 30); do
+          xdpyinfo -display :99 >/dev/null 2>&1 && break
+          sleep 1
+        done
+        xdpyinfo -display :99 >/dev/null 2>&1 || { echo "Xvfb never accepted connections"; exit 1; }
+        echo "Xvfb is up on :99"
+
+        # Without a window manager no shell ever becomes active, and the waits on shell
+        # activation in closeFile() and openDataObject() time out and fail the test.
+        # Start metacity and wait for it to own the display before running the tests.
+        metacity --sm-disable --display=:99 &
+        for i in $(seq 1 20); do
+          xprop -display :99 -root _NET_SUPPORTING_WM_CHECK 2>/dev/null | grep -q "window id" && break
+          sleep 1
+        done
+        if xprop -display :99 -root _NET_SUPPORTING_WM_CHECK 2>/dev/null | grep -q "window id"; then
+          echo "metacity is managing :99"
+        else
+          echo "WARNING: no window manager on :99 - shell activation waits will time out"
+        fi
+
+        mvn test -B -pl hdfview
+        echo "::endgroup::"
+
+    - name: UI Test Summary
+      if: always()
+      continue-on-error: true
+      run: |
+        # Surface the counts in the job log even when the step above was cut short, so a
+        # run is never just a red X with no numbers attached.
+        python3 - <<'EOF' >> $GITHUB_STEP_SUMMARY
+        import glob, os, xml.etree.ElementTree as ET
+        files = glob.glob("hdfview/target/surefire-reports/*.xml")
+        print("## UI tests\n")
+        if not files:
+            print("No surefire XML was written - the suite did not get far enough to report.")
+        else:
+            tot = fail = err = skip = 0
+            rows = []
+            bad = []
+            for f in sorted(files):
+                # A fork killed by a timeout can leave a half-written XML behind. Report that
+                # file as unreadable rather than letting ParseError abort the whole summary
+                # halfway through the table, which looks like the suite simply stopped.
+                try:
+                    r = ET.parse(f).getroot()
+                except Exception as exc:
+                    bad.append((os.path.basename(f), type(exc).__name__))
+                    continue
+                t = int(r.get("tests", 0)); fl = int(r.get("failures", 0))
+                e = int(r.get("errors", 0)); s = int(r.get("skipped", 0))
+                tot += t; fail += fl; err += e; skip += s
+                rows.append((r.get("name", f), t, fl, e, s))
+            print(f"**{tot} tests, {fail} failures, {err} errors, {skip} skipped** "
+                  f"across {len(rows)} classes\n")
+            print("| Class | Tests | Failures | Errors | Skipped |")
+            print("| --- | --- | --- | --- | --- |")
+            for n, t, fl, e, s in rows:
+                print(f"| {n} | {t} | {fl} | {e} | {s} |")
+            for n, why in bad:
+                print(f"| {n} | unreadable ({why}) | | | |")
+        EOF
+        cat $GITHUB_STEP_SUMMARY
+
+    # Scoped to the object module on purpose. This action publishes a check run, and a check
+    # is not covered by the step's continue-on-error - so globbing '**' would put the
+    # UI failures into a failing check on every PR. The UI tier reports through the
+    # summary step above and the uploaded surefire artifacts below.
     - name: Test Report
       uses: dorny/test-reporter@v1
       if: success() || failure()
       continue-on-error: true
       with:
         name: Maven Test Results (Linux)
-        path: '**/target/surefire-reports/*.xml'
+        path: 'object/target/surefire-reports/*.xml'
         reporter: java-junit
         fail-on-error: false
 

@@ -888,6 +888,18 @@
                     <systemPropertyVariables>
                         <hdfview.workdir>${project.build.testOutputDirectory}/uitest</hdfview.workdir>
                     </systemPropertyVariables>
+                    <!--
+                        One JVM per test class. Each UI test class starts its own HDFView window,
+                        so a class that hangs or crashes the VM would otherwise take every remaining
+                        class down with as well and leave no surefire XML behind.
+                    -->
+                    <reuseForks>false</reuseForks>
+                    <!--
+                        Per-class cap for hangs the JUnit timeouts can't interrupt. Must exceed the
+                        sum of a class's JUnit timeouts (worst case TestHDFViewIntConversions, 56m),
+                        or a killed fork loses that class's results.
+                    -->
+                    <forkedProcessTimeoutInSeconds>3600</forkedProcessTimeoutInSeconds>
                 </configuration>
             </plugin>
 

@@ -29,6 +29,8 @@
 import java.util.Vector;
 import java.util.concurrent.BrokenBarrierException;
 import java.util.concurrent.CyclicBarrier;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.TimeoutException;
 
 import hdf.HDFVersions;
 import hdf.view.HDFView;
@@ -89,6 +91,15 @@ public abstract class AbstractWindowTest {
 
     private static final CyclicBarrier swtBarrier = new CyclicBarrier(2);
 
+    /*
+     * How long a test waits for the main window to come up. Shorter than the 2 minute JUnit
+     * default so a failed launch reports its own message rather than a generic timeout.
+     */
+    private static final int APP_STARTUP_TIMEOUT_SECONDS = 60;
+
+    /* Why the UI thread died, if it did, so that waiters can report the real cause. */
+    private static volatile Throwable appStartupFailure = null;
+
     private static int TEST_DELAY = 10;
 
     private static int open_files = 0;
@@ -102,11 +113,11 @@ protected static enum FILE_MODE { READ_ONLY, READ_WRITE, MULTI_READ_ONLY }
     private static final String objectShellTitleRegex = ".*at.*\\[.*in.*\\]";
 
     @BeforeEach
-    public final void setupSWTBot(TestInfo testInfo) throws InterruptedException, BrokenBarrierException
+    public final void setupSWTBot(TestInfo testInfo) throws InterruptedException
     {
         this.testInfo = testInfo;
         // synchronize with the thread opening the shell
-        swtBarrier.await();
+        awaitAppWindow();
         bot = new SWTBot();
 
         SWTBotPreferences.PLAYBACK_DELAY = TEST_DELAY;
@@ -119,9 +130,33 @@ public void run()
         });
     }
 
+    /**
+     * Waits for the UI thread to open this test's HDFView window. Fails the test if the
+     * window doesn't open within APP_STARTUP_TIMEOUT_SECONDS, or if the UI thread dies first.
+     */
+    private static void awaitAppWindow() throws InterruptedException
+    {
+        try {
+            swtBarrier.await(APP_STARTUP_TIMEOUT_SECONDS, TimeUnit.SECONDS);
+        }
+        catch (TimeoutException e) {
+            fail("HDFView main window did not open within " + APP_STARTUP_TIMEOUT_SECONDS + " seconds",
+                 appStartupFailure);
+        }
+        catch (BrokenBarrierException e) {
+            fail("HDFView UI thread died before the main window opened", appStartupFailure);
+        }
+    }
+
     @AfterEach
     public void closeShell() throws InterruptedException
     {
+        // Nothing to tear down if the window never opened. Bail out before touching Display,
+        // since Display.getDefault() creates a display when none exists, and on a machine with no
+        // usable X server that call will block forever inside the native gdk_threads_enter().
+        if (shell == null || shell.isDisposed())
+            return;
+
         // close the shell
         Display.getDefault().syncExec(new Runnable() {
             @Override
@@ -199,17 +234,26 @@ public void run()
                             window.runMainWindow();
                         }
                     }
-                    catch (Exception e) {
-                        e.printStackTrace();
+                    catch (Throwable t) {
+                        // Record the real cause and break the barrier so that any test parked
+                        // in awaitAppWindow() fails immediately with this stack trace, rather
+                        // than waiting out the startup timeout for a window that will never
+                        // arrive. Throwable, not Exception: a linkage or native-library error
+                        // during SWT startup is exactly the case that used to hang CI.
+                        appStartupFailure = t;
+                        t.printStackTrace();
+                        swtBarrier.reset();
                     }
 
-                    Display.getDefault().syncExec(new Runnable() {
-                        @Override
-                        public void run()
-                        {
-                            shell.getDisplay().dispose();
-                        }
-                    });
+                    if (shell != null) {
+                        Display.getDefault().syncExec(new Runnable() {
+                            @Override
+                            public void run()
+                            {
+                                shell.getDisplay().dispose();
+                            }
+                        });
+                    }
                 }
             });
             uiThread.setDaemon(true);

@@ -3,9 +3,11 @@
 import static org.junit.jupiter.api.Assertions.fail;
 
 import java.io.File;
+import java.util.concurrent.TimeUnit;
 
 import org.junit.jupiter.api.Tag;
 import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.Timeout;
 
 import org.eclipse.swtbot.nebula.nattable.finder.widgets.SWTBotNatTable;
 import org.eclipse.swtbot.swt.finder.widgets.SWTBotShell;
@@ -466,7 +468,9 @@ public void checkHDF5GroupDU16()
         }
     }
 
+    // Walks every cell of a large dataset: ~186s locally, slower on CI runners.
     @Test
+    @Timeout(value = 12, unit = TimeUnit.MINUTES)
     public void checkHDF5GroupDS32()
     {
         String[][] expectedData = {
@@ -772,7 +776,9 @@ public void checkHDF5GroupDS32()
         }
     }
 
+    // Walks every cell of a large dataset: ~186s locally, slower on CI runners.
     @Test
+    @Timeout(value = 12, unit = TimeUnit.MINUTES)
     public void checkHDF5GroupDU32()
     {
         String[][] expectedData = {
@@ -1075,7 +1081,9 @@ public void checkHDF5GroupDU32()
         }
     }
 
+    // Walks every cell of a large dataset: ~372s locally, slower on CI runners.
     @Test
+    @Timeout(value = 12, unit = TimeUnit.MINUTES)
     public void checkHDF5GroupDS64()
     {
         String[][] expectedData    = {{"-1",
@@ -2386,7 +2394,9 @@ public void checkHDF5GroupDS64()
         }
     }
 
+    // Walks every cell of a large dataset: ~372s locally, slower on CI runners.
     @Test
+    @Timeout(value = 12, unit = TimeUnit.MINUTES)
     public void checkHDF5GroupDU64()
     {
         String[][] expectedData = {

@@ -4,9 +4,11 @@
 import static org.junit.jupiter.api.Assertions.fail;
 
 import java.io.File;
+import java.util.concurrent.TimeUnit;
 
 import org.junit.jupiter.api.Tag;
 import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.Timeout;
 
 import org.eclipse.swtbot.nebula.nattable.finder.widgets.SWTBotNatTable;
 import org.eclipse.swtbot.swt.finder.waits.Conditions;
@@ -39,7 +41,9 @@ public class TestTreeViewFilters extends AbstractWindowTest {
                                                 {"170", "171", "172", "173", "174", "175", "176", "177", "178", "179"},
                                                 {"180", "181", "182", "183", "184", "185", "186", "187", "188", "189"},
                                                 {"190", "191", "192", "193", "194", "195", "196", "197", "198", "199"}};
+    // Walks every cell of a large dataset: ~318s locally, slower on CI runners.
     @Test
+    @Timeout(value = 12, unit = TimeUnit.MINUTES)
     public void openHDF5Filters()
     {
         SWTBotShell tableShell = null;

@@ -0,0 +1,4 @@
+# Default timeout for every UI test and lifecycle method, so that a hung test merely fails
+# without blocking. Tests that legitimately run longer can override it with @Timeout. This cannot
+# interrupt a blocked native call.
+junit.jupiter.execution.timeout.default = 2 m