[Qgis-developer] qgspostgresprovider and cursors

[Stefanie Tellex]

Hi,

I had a strange bug where the transaction got aborted because of a bad 
SQL where clause, and after it was aborted, future queries were being 
ignored.

The attached patch makes it go away by ensuring that the current 
transaction is closed and all existing cursors are closed before setting 
a new subset query.  It also makes it possible to close a cursor more 
than once without a problem by preventing openCursors from falling below 
zero.

However the code for managing cursors and transactions in 
QgsPostgresProvider was pretty confusing, and I suspect there are 
similar bugs hiding out still.

It's against revision 9563.

Thanks,

Stefanie
-------------- next part --------------
Index: qgis/qgis_unstable/src/providers/postgres/qgspostgresprovider.cpp
===================================================================
--- qgis.orig/qgis_unstable/src/providers/postgres/qgspostgresprovider.cpp	2008-11-06 11:51:42.000000000 -0500
+++ qgis/qgis_unstable/src/providers/postgres/qgspostgresprovider.cpp	2008-11-06 11:51:44.000000000 -0500
@@ -2173,6 +2173,7 @@
 
 void QgsPostgresProvider::setSubsetString( QString theSQL )
 {
+  connectionRO->closeCursor( QString( "qgisf%1" ).arg( providerId ) );
   sqlWhereClause = theSQL;
   // Update datasource uri too
   mUri.setSql( theSQL );
@@ -2619,8 +2620,11 @@
 bool QgsPostgresProvider::Conn::closeCursor( QString cursorName )
 {
   bool res = PQexecNR( QString( "CLOSE %1" ).arg( cursorName ) );
-
-  if ( --openCursors == 0 )
+  --openCursors;
+  if (openCursors < 0) {
+    openCursors = 0;
+  }
+  if (openCursors == 0 )
   {
     QgsDebugMsg( "Commiting read-only transaction" );
     PQexecNR( "COMMIT" );