Detailed Summary of Testing of Root 3.00.06 with various Glast programs

March 7 - 

• TBIOROOT v1r1 executes without problem [linux, Windows]
• DbCnv has a compile error [linux]
• TBEvent - the automatic rootcint has a warning which stops the build, but doesn't seem to be a serious problem [linux]
• ROOTWriter v2r5p1 executes, but generates a lot of warnings [linux]

March 8 - 

• tb_recon muon_test executes without problems [linux]

March 20  

• the testing of the root macros has uncovered some troubles [linux].  When I try and load the muon root file produced by ROOTWriter, with the RootTreeAnalysis macro, then I get

   WARNING, class:Event StreamerInfo read from file:/nfs/slac/g/glast/users/glground/tlindner/packages2/ROOTWriter_ROOt3/ROOTWriter/v2r5p1/src/test/tb_muons_5_gev_normal.root
  has the same version:1 than the active class
  but a different checksum.
  You should update the version to ClassDef(Event,2).
  Do not try to write objects with the current class definition,
  the files will not be readable.

March 21

• DbCnv has been fixed so that RootDb will compile.  First regenerated the TBlobDict files using Root 3.00.06 and rootcint.  Then modified RootSelect.cpp as follows:      
  • #include TStreamerInfo.h
  • renamed kLong to TStreamerInfo::kLong
  • renamed kDouble to TStreamerInfo::kDouble
  • renamed EReturnType to TMethodCall::EReturnType

   

• RootHistCnv generation of Root ntuples has been checked on Windows.
  • No modifications required on Windows.
  • GuiApp runs and creates a readable aoRecon.root file that contains data.  Here is an eps file of Veto_DOCA generated with GuiApp, RootHistCnv, and Root 3.00.06.  Here is an eps file of Veto_DOCA generated with GuiApp, RootHistCnv, and Root 2.25.03.

   

• ROOTWriter's muon_test generates the following warning messages:
  • Warning in <ACDTile::Compare>: this method must be overridden!
  • Warning in <CalHit::Compare>: this method must be overridden!
  • Warning in <TkrLayer::Compare>: this method must be overridden!

  The problem is actually in TBEvent, due to changes in Root 3.00 to become more const correct, see this RootTalk post.  So the Compare routines had to be modified in ACDTile, CalHit and TkrLayer.

  Running ROOTWriter muon_test produces a Root file without producing any warning messages (on both Windows and Linux).

March 22

• To produce extensive Root debugging information, set the Root.Debug flag > 0 in the system.rootrc file.
• Fixed ACDTile source file - the streamer was missing a return statement.
• Sent message to RootTalk concerning the problems reading the TBEvent root file.  The data appears ok - despite the warnings concerning the StreamerInfo checksum values.

March 23

• Correspondence with Rene Brun reveals that the problem stems from the MyEvent, MyAnalysis classes and not from the TBEvent root file itself.  Here is a test to see how the checksum is set for the Event class, and in the StreamerInfo stored in the root file itself:

Load the DLL and see the value of checksum for the Event class:

root [1]gSystem->Load("TBEventshr.dll")
(int)0
root [2] Event::Class()->GetCheckSum()
(const unsigned int)929237328

Open the file (without loading the DLL) and then check the StreamerInfo data for the Event class:

root [0] TFile f("I:/tb_muons_5_gev_normal.root")
root [1] gROOT->GetListOfStreamerInfo()->FindObject("Event")->Dump()
fCheckSum 929237328 checksum of original class

If one runs a similar test within the MyEvent constructor - we find the following:

UInt_t eventCheck = Event::Class()->GetCheckSum();
printf("Event checksum = %u\n", eventCheck);
outputs:
Event checksum = 3089147232

gROOT->GetListOfStreamerInfo()->FindObject("Event")->Dump();
outputs:
fCheckSum 929237328 checksum of original class

So somehow, MyEvent sees a different version of the Event class than the version loaded via the DLL....another message was sent to Rene, and we await his reply.

March 27

• New message from Rene, reveals that we should be providing forward declarations of user defined classes in RootTreeAnalysis (MyAnalysis).  So forward declarations of the Event and Recon classes were added.  This did not completely solve the problem however...ultimately we needed to rename the source file from *.c to *.cxx.  Then the warning messages about the checksum disappeared.