[AGENT++] AgentX++ suagent commiterror let the RowStatus in detroy(6); the row is block forever!

[Claus Klein]

Hi Frank,

my first interpretation of my test was not right. The handling with  
commit failed works fine!

But still there may be a problem if the AgentXSlave 'deallocate  
request' fails.
Too, It is not very efficiently if we register a row only to destroy  
it immediately?


I changed a little bit at the code and improved the RowStatus destroy  
handling.
I hope it is right, I am not sure! Please check it.

Too I improved my test scripts, I attache my patch and the scripts.

With regards,

Claus

claus-kleins-macbook-pro:AgentPro clausklein$ bzr diff -p1
=== modified file 'agent++/src/mib.cpp'
--- old/agent++/src/mib.cpp	2010-10-31 08:30:19 +0000
+++ new/agent++/src/mib.cpp	2010-10-31 10:28:54 +0000
@@ -572,10 +572,10 @@
  		  return ((l == rowNotInService) ||
  			  (l == rowActive) || (l == rowDestroy));
  		default:
-		  return FALSE;
+		  return (l == rowDestroy); // FIXME: added! We have to allow  
destroy in each state! ck
  		}
  	}
-	else
+	else    // FIXME: what state is this? ck
  		return ((l == rowCreateAndGo) || (l == rowCreateAndWait) ||
  			(l == rowDestroy));
  }
@@ -672,7 +672,7 @@
  	  break;
  	case rowDestroy:
  	  set_value(rs);
-	  delete undo; // No undo of row destroy!
+	  delete undo; // FIXME: why, ck? No undo of row destroy!
  	  undo = 0;
  	  break;
  	default:
@@ -2177,10 +2177,12 @@
  			  ok   = TRUE;
  			  wait = TRUE;
  			  break;
+#if 1   // FIXME: Do we allow destroy of a nonexiting row without  
error? YES! ck
  			case rowDestroy:
  			  ok = TRUE;
  			  ignore = TRUE;
  			  break;
+#endif
  			default:
  			  delete[] fulfilled;
  			  delete[] required;
@@ -2236,6 +2238,7 @@
  			pvbs[col] = req->get_value(i);
  		}
  	}
+
  	// test values through prepare
  	MibTableRow* new_row = new MibTableRow(generator);
  	new_row->set_index(new_index);
@@ -2304,14 +2307,15 @@
  	else
  		voting_result = perform_voting(new_row,
  					       rowEmpty, rowDestroy);
-	delete new_row;
-
  	// Unset the ready flag of the current column.
  	// via this one all the other columns are set by commit_set_request
  	req->unset_ready(ind);
  	delete[] fulfilled;
  	delete[] required;
  	delete[] pvbs;
+	delete new_row; // FIXME: same oder as other cleanup blocks; ck
+        // TODO: use a macro for cleanup or better spart pointers? ck
+
  	return voting_result;
  }


claus-kleins-macbook-pro:AgentPro clausklein$
-------------- next part --------------



On 29.10.2010, at 22:59, Claus Klein wrote:

> Hi Frank,
>
> I have found a third issue about the RowStatus handling at the  
> subagent.
> With this script, you can force this error as many times you want,
> simple choose a new row index.
>
> Kindly regards,
> Claus
>