[Rivet-svn] r1745 - in trunk: include/Rivet src

[blackhole at projects.hepforge.org]

Author: fsiegert
Date: Wed Aug  5 16:34:46 2009
New Revision: 1745

Log:
When running the rivet executable with multiple analyses I had
reproducably (but not with all analysis combinations?) the effect that
the _info object of one of the analyses had been reset to 0x6666666
right before event analysis started, causing a crash.
The only way I found to fix this is to not use a smart pointer for the
_info object, but a regular pointer. This works right away (with no
other changes).
If somebody has a reason for using a smart pointer here and any idea how
to debug this, let me know and I can re-test with a smart pointer.

Modified:
   trunk/include/Rivet/Analysis.hh
   trunk/src/Analysis.cc

Modified: trunk/include/Rivet/Analysis.hh
==============================================================================
--- trunk/include/Rivet/Analysis.hh	Wed Aug  5 15:02:59 2009	(r1744)
+++ trunk/include/Rivet/Analysis.hh	Wed Aug  5 16:34:46 2009	(r1745)
@@ -59,7 +59,7 @@
     Analysis(const std::string& name);
 
     /// The destructor.
-    virtual ~Analysis() { }
+    virtual ~Analysis();
     //@}
 
   public:
@@ -401,7 +401,7 @@
   protected:
 
     /// Pointer to analysis metadata object
-    shared_ptr<AnalysisInfo> _info;
+    AnalysisInfo * _info;
 
     
   private:

Modified: trunk/src/Analysis.cc
==============================================================================
--- trunk/src/Analysis.cc	Wed Aug  5 15:02:59 2009	(r1744)
+++ trunk/src/Analysis.cc	Wed Aug  5 16:34:46 2009	(r1745)
@@ -17,10 +17,14 @@
       _analysishandler(0),
       _madeHistoDir(false)
   {
-    _info.reset( AnalysisInfo::make(name) );
+    _info = AnalysisInfo::make(name);
     setBeams(ANY, ANY);
   }
-  
+
+  Analysis::~Analysis()
+  {
+    if (_info) delete _info;
+  }
 
   IAnalysisFactory& Analysis::analysisFactory() {
     return handler().analysisFactory();