Fixed Issue 118: EXCEPTION_ACCESS_VIOLATION on show of 2nd canvas (ATI-specific)

The root cause of this bug was not the new DummyGL code, but severe
bugs in ATI's OpenGL drivers where it appears that if an OpenGL
context is ever made current on more than one thread during the
lifetime of an application, problems begin to occur such as the
SetPixelFormat call failing on the just the next newly-created HDC, or
all subsequent SetPixelFormat calls failing on all
subsequently-created HDCs. This was occurring because the
single-threaded ATI workaround's automatic detection mechanism was not
being enabled until the first time a context was made current, but by
then it was typically too late; the context was made current on the
end user's thread during e.g. Frame.show(), and if it was ever made
current on another thread (like the AWT event queue thread, which is
where all OpenGL operations are performed when the single-threaded
workaround is enabled) then the problems would begin.

The failure has only been seen on Windows so far; ATI's drivers on X11
seem to be better behaved. The workaround is to check for the presence
of ATI's drivers very early during JOGL's initialization, by looking
in the system directory for atioglxx.dll, and enabling the
single-threaded workaround if it was found. This workaround causes the
attached test case to work properly.

Also added some more debugging code to make diagnosing problems like
these easier in the future.


git-svn-id: file:///usr/local/projects/SUN/JOGL/git-svn/svn-server-sync/jogl/trunk@196 232f8b59-042b-4e1e-8c03-345bb8c30851

1 files changed, 51 insertions, 0 deletions

diff --git a/src/net/java/games/jogl/impl/windows/WindowsGLContextFactory.java b/src/net/java/games/jogl/impl/windows/WindowsGLContextFactory.java
index 726091bf1..1c976c355 100644
--- a/src/net/java/games/jogl/impl/windows/WindowsGLContextFactory.java
+++ b/src/net/java/games/jogl/impl/windows/WindowsGLContextFactory.java
@@ -43,6 +43,7 @@ import java.awt.Component;
 import java.awt.GraphicsConfiguration;
 import java.awt.GraphicsDevice;
 import java.awt.Rectangle;
+import java.io.File;
 import java.security.AccessController;
 import java.security.PrivilegedAction;
 import java.util.HashMap;
@@ -74,6 +75,56 @@ public class WindowsGLContextFactory extends GLContextFactory {
     AccessController.doPrivileged( new PrivilegedAction() {
       public Object run() {
         Runtime.getRuntime().addShutdownHook( new ShutdownHook() );
+        
+        // Test for whether we should enable the single-threaded
+        // workaround for ATI cards. It appears that if we make any
+        // OpenGL context current on more than one thread on ATI cards
+        // on Windows then we see random failures like the inability
+        // to create more OpenGL contexts, or having just the next
+        // OpenGL SetPixelFormat operation fail with a GetNextError()
+        // code of 0 (but subsequent ones on subsequently-created
+        // windows succeed). These kinds of failures are obviously due
+        // to bugs in ATI's OpenGL drivers. Through trial and error it
+        // was found that specifying
+        // -DJOGL_SINGLE_THREADED_WORKAROUND=true on the command line
+        // caused these problems to completely disappear. Therefore at
+        // least on Windows we try to enable the single-threaded
+        // workaround before creating any OpenGL contexts. In the
+        // future, if problems are encountered on other platforms and
+        // -DJOGL_SINGLE_THREADED_WORKAROUND=true works around them,
+        // we may want to implement a workaround like this on other
+        // platforms.
+        
+        // The algorithm here is to try to find the system directory
+        // (assuming it is on the same drive as TMPDIR, exposed
+        // through the system property java.io.tmpdir) and see whether
+        // a known file in the ATI drivers is present; if it is, we
+        // enable the single-threaded workaround.
+
+        // If any path down this code fails, we simply bail out -- we
+        // don't go to great lengths to figure out if the ATI drivers
+        // are present. We could add more checks here in the future if
+        // these appear to be insufficient.
+
+        String tmpDirProp = System.getProperty("java.io.tmpdir");
+        if (tmpDirProp != null) {
+          File file = new File(tmpDirProp);
+          if (file.isAbsolute()) {
+            File parent = null;
+            do {
+              parent = file.getParentFile();
+              if (parent != null) {
+                file = parent;
+              }
+            } while (parent != null);
+            // Now the file contains just the drive letter
+            file = new File(new File(new File(file, "windows"), "system32"), "atioglxx.dll");
+            if (file.exists()) {
+              SingleThreadedWorkaround.shouldDoWorkaround();
+            }
+          }
+        }
+
         return( null );
       }
     });