svn commit: r690140 - /myfaces/orchestra/trunk/core/src/main/java/org/apache/myfaces/orchestra/conversation/jsf/AccessScopePhaseListener.java

[sk...@apache.org]

Author: skitching
Date: Fri Aug 29 00:57:01 2008
New Revision: 690140

URL: http://svn.apache.org/viewvc?rev=690140&view=rev
Log:
Fix for ORCHESTRA-29 - access-scoped conversations always get discarded during ajax requests from the NetAdvantage ajax library.

Modified:
    myfaces/orchestra/trunk/core/src/main/java/org/apache/myfaces/orchestra/conversation/jsf/AccessScopePhaseListener.java

Modified: myfaces/orchestra/trunk/core/src/main/java/org/apache/myfaces/orchestra/conversation/jsf/AccessScopePhaseListener.java
URL: http://svn.apache.org/viewvc/myfaces/orchestra/trunk/core/src/main/java/org/apache/myfaces/orchestra/conversation/jsf/AccessScopePhaseListener.java?rev=690140&r1=690139&r2=690140&view=diff
==============================================================================
--- myfaces/orchestra/trunk/core/src/main/java/org/apache/myfaces/orchestra/conversation/jsf/AccessScopePhaseListener.java (original)
+++ myfaces/orchestra/trunk/core/src/main/java/org/apache/myfaces/orchestra/conversation/jsf/AccessScopePhaseListener.java Fri Aug 29 00:57:01 2008
@@ -89,7 +89,34 @@
     public void beforePhase(PhaseEvent event)
     {
         AccessScopeManager accessManager = AccessScopeManager.getInstance();
-        accessManager.beginRecording();
+        
+        if (event.getFacesContext().getResponseComplete())
+        {
+            // Unusual case. If something set responseComplete before the rendering phase started, then
+            // the rendering phase would simply be skipped and we would not be here. So presumably some
+            // PhaseListener that ran in this phase but before this PhaseListener has decided to implement
+            // its own form of "rendering" itself and then set responseComplete to bypass the normal
+            // rendering phase.
+            //
+            // When this occurs, any accesses to conversation-scoped beans will already have happened.
+            // If we were to call "beginRecording" here as normal then then at afterPhase all conversations
+            // would appear to be unaccessed which is not true - they may have been accessed, but before
+            // this method ran. So instead, just disable checking for this particular request. This
+            // solution isn't 100% correct, but should be close enough, and no other alternative is
+            // simple. The correct solution would possibly be to use a custom Lifecycle class to do the
+            // access-checking rather than a PhaseListener; that would then avoid any dependency on the
+            // order in which PhaseListener classes are registered. Note, however, that some ajax
+            // libraries use a custom Lifecycle class themselves, so again there are potential
+            // registration ordering issues there.
+            //
+            // This situation happens with the NetAdvantage ajax library at least, and perhaps others.
+            accessManager.setIgnoreRequest();
+        }
+        else
+        {
+            // normal case
+            accessManager.beginRecording();
+        }
     }
 
     public void afterPhase(PhaseEvent event)