ROOT Team Meeting

Friday 24 Feb 2006, 11:00 → 12:30 Europe/Zurich

11:00 → 11:20 Report from CHEP

11:20 → 11:40 Tour de table

///////////////////////////////////////////////////// // Minutes of ROOT team meeting: 24 February 2006 // ///////////////////////////////////////////////////// CHEP06 post mortem ------------------ Interesting experience at Kolkata and Mumbai. In the Kolkata workshop at VECC, we met many young people ready to collaborate with us. Problem is the life time in HENP. Many young people attracted by IT in Bangalore. Many thanks to all speakers and papers authors for CHEP. ROOT was extremely well represented and mentionned in most talks. Most people pleased with the SEAL+ROOT merge. Interest for CINT/Reflex. Papers to be sent asap. Some immediate side-effects,eg -More CMS discussions about PROOF and an evaluation setup. -Reactions from IT (Bernd Panzer). Meeting next week. -International Linear Collider: IV concept based on ROOT Release 5.10 ------------ Previously announced date (23 February) impossible. Waiting for fixes by Philippe, Axel and Wim. At the Architects Forum (23 February), we decided for a new date (Thursday 2 March). Please send your Release Notes asap in the usual format. Indico/CDS team meeting web site visible from outside ------------------------------------------------------ From: cstrato <cstrato@aon.at> To: Rene Brun <Rene.Brun@cern.ch> Subject: BOOT Dear Rene While reading the rootforum I found a link which brought me finally to the following link: http://indico.cern.ch/conferenceDisplay.py?confId=820 I could download your interesting thoughts about STL and ideas about BOOT. At the moment I have only the following question: Is it intentional that everybody can download slides from the ROOT weekly meeting or is it a mistake? In the first case I understand that I could in principle discuss these issues in the rootforum, correct? In the second case I will not use this knowledge. Please let me know. Best regards Christian Should we restrict access to this web pages? Answer no. However, we will not answer any questions from curious observers. Mails to Forum/Savannah ----------------------- Please close as many open points as possible in Savannah Visitors -------- -Timur Pocheptsov: 26 February --> 11 March -Valeriy Onuchin: 1st March --> 23 March (still to be confirmed) Tour de table ------------- Rene ==== -Fixed a problem introduced in version 5.08. Bins with 0 entries were not taken into account in Likelihood fits. -Implement possibility to specify the x,y axis titles in the TF1,2,3 constructors. Axel ==== * Replacing all occurrences of Cint data structures by the Reflex version. Ignore extern "C" interface for now, allowing us to focus on the merge with Reflex, instead of on how to hide this merge from the C interface. Replacement is already done for typedefs. * To allow C++ API and legacy C API to co-exist, Cint now lives in namespace "Cint". * Implemented roottest reflex merger test suite; to be uploaded to CVS shortly. * I'm waiting for Stefan to implement name lookup in Reflex. Until then I can only test the functionality, not performance, of the Cint modifications. * Major modification if TRef-autoloading for TRefs created during different processes. Philippe ======== For the next release we hope to fix Nick West's problem and to remove a memory leak of the user's object in TBranchElement. Lately we corrected several issues in TTree::Draw related to the handling of std::pair (in emulated mode the memory layout overlayed both members), improved the lookup of the branch (to avoid having to read the full object), improved backward compatibility with the old streamer handling of STL collection, fixed the lookup of the branch's counter for TTreeFormula in some nested cases, fixed the handling of std::map when the pair contains a pointer. Most of those prevent the full reading of HepMC and CMS data files. SetOwner is now properly implemented in TClonesArray. Markus ====== Finished Reference folloowing mechanism (TRef, Pool-refs and Gaudi) Patch with Philippe. Discussing API. Stefan ======= - Helping CMS to fix a bug. The problem intially was a misleading error message in ROOT TBuffer about non available constructors for base classes while a derived class was used, but no dicitonary was available. After fixing this bug the missing dicitonary in CMS code was discovered. - ATLAS reported and submitted a patch for genreflex.py for the generated code for base class offset calculation (extra function used for ROOT I/O). The problem was the usage of a templated function to generate the typename which results in two different typenames for templated classes for stl containers or other classes with default parameter names. - ATLAS reported another bug about non handling of constructors with 33 arguments - LHCb reported a problem with generating dictionaries for boost::shared_ptr which contains a conversion operator for pointer to member. The stub function code for this converter is not produced correctly by genreflex.py - Setup Reflex for XCode (currently problem with weak symbols) Olivier ======= * Richard and Timur made a lot of improvements/new features both in the viewer and in "GL in Pad". This list is too long to be given here. It will be detailed in the coming release notes. * TLatex. Work started again on that long list on item related to this class: - #hbar has been implemented - New example showing the TLatex mathematical symbols (latex5.C) - Help update (some characters were missing) * Axis painting: New options X+ and Y+ to paint the X axis on top of the plot (default is bottom) and the Y axis on the right of the plot (Default is left). * TEllipse: - Paint() now uses the FillStyle to control whether or not a filled area should be painted. It used wrongly the FillColor before. - PaintEllipse didn't use its input parameters to paint the ellipse but directly the class data members. This prevent to use this method directly. * The statistics parameters names can now be changed thanks to settings in the file $ROOTSYS/etc/system.rootrc. * Miscellaneous: - THStack drawing: color problem fixed - In case of subrange TH3::Project3D was wrong - Option FB and BB were not recognize with option ISO - TH2: Take into account underflows/overflows when re-binning - TGaxis: New methods SetTitleFont and SetTitleColor to be consistent with TAxis. - The width of the frame drawn by TPostScript::DrawFrame was 4 times smaller than the one on the screen or the one in a PDF file. PDF output is right. - THistPainter::PaintErrors: Option E1 painted vertical edge bars for empty bins. We had a discussion about internalisation of the GUI and all error messages. Bertrand ======== request to clean the empty DLLS in the lib directory. Fons to implement it for the release. - User support. - Bug fixes in gdk layer. - Integration of Root canvas in non-Root applications/toolkits like Qt, MFC, FOX, PVSS. - SPI : installation of python win32 extensions in external. Request to support ROOT/PVSS with gcc2.96 !! (to be discussed) Ian === implemented all functions for file uploading to PROOF/xrootd. Ilka ==== * Presentations for CHEP06: - slides of "General Status of ROOT GUI" - presented examples taken from real code - slides of "From Task Analysis to the GUI Application Design" * Bug fixes in: - SavePrimitive methods of TF1 and TMultigraph - fitLinear.C macro - SetConstrained methods of triple sliders * Work in progress of TF1Editor design and prototyping: - required knowledge on TF1 & TFormula classes - has to cover all ways of user defined functions * Work in progress on new Fit Panel design: - General tab design - mostly finished - Minimization tab design - required knowledge on TVirtualFitter and related classes * User support: - examples on signals/slots communication; - the use of modifier keys; - GUI for setting of function parameters. Anna ==== - The fftw interface: adding multidimensional transforms has required new data-handling functions and there are still some problems with this, especially for the real2real transforms of even/odd functions. Hopefully they will be solved next week. 1d seems to work fine after some simple testing, but I want to try it on some more complex tests too. - Virtual cuts: I'm looking into it, but can't really speak of any results yet. Guenter ======= 1. olbd performance analysis on the cluster ------------------------------------------- Problem of "very fat tails" solved: in about 10% of executed queries the last worker answered between 2 and 10 sec. Reason: the logging of xrootd have been done on the shared nfs drive. As soon as I did the logging on the local drive of every node this problem disappeared: Current results (980 queries, no load on the machines, time until the last node responded) Avg: 124 ms, Min: 92 ms, Max: 708 ms in 90% of the cases the answer is below 185 ms Currently investigating performance under: - cpu load - io load - network load Some problems running proof on the cluster at the moment: xrootd reports: "SetUserEnvironment: setresuid/gid require super-user privileges: do nothing" when creating a new session 2. TProof --------- Changes in TProof, TProofServ, and TXProofServ: TProofResourcesStatic was directly created in StartSlaves() Now created with TVirtualProofResources* TProofServ::GetProofResources() This method decides according to a flag (member variable of TProofServ) which class to load - after discussion with Gerri: loading should be done via Plugin Manager. 3. xproofd protocol ------------------- To do the communication between xrootd and the olbd I now implemented own classes (before I used the XrdOdc classes which are part from the "Open file system"). I did this because for the communication we have different requirements in the message flow (e.g. 1 TCP request -> n UDP answers) Additional classes XrdProofdPlb* as part of libProofd.so (?) Gerri ===== Still to fix a few bugs to be fixed before the release. Fix ProofBench Other patches will wait for next release. Discussion about the PROOF Users Guide. Fons ==== Patch from Christian Holm for Debian distribution and Redhat rpms. Discussion about InstallShield for Windows. Making a few classes for UUIDs in a multi-threaded environment with multi-core cpus. Richard ======= Fixed several bugs in GL viewer. Code Cleanup. Many thanks to Richard for the few months spent with us. Richard has made great improvements in the GL viewer, functionality and speed. Best wishes to him for his future career.