ROOT 6 planning
ROOT 6 planning (22 Oct 2014)
Present: The ROOT Team, Marcin, David L., Peter, Marco, Bill (R)
News:
o bad alloc of cms: pin-pointed the origin. MVAComputer::addProcessor(const VarProcessor* proc) uses the TypeWIthDict interface to clone the object pointed by proc, which could be a daughter class of VarProcessor.
Peter:
o Some issues found in the master of 5 and 6 release cycles: reported in jira
o After some adjustments for the R6 build, reco of raw data is running again. Next week perf numbers will be reported.
Marcin:
o Some Coral classes have dictionaries in Atlas libraries. It is not yet clear how these classes are used.
o Issue with function in namespaces: workaround is there.
o Online test: missing system header. Fixed ~5 days ago: need to make the build available soon.
o Two solutions to make the headers available: make headers available feeding the paths to gInterpreter. If no python available, this has to be done via the ROOT_INCLUDE_PATH. Unfortunately for Atlas Offline, this env variable is 180 KB long. Some solutions are possible on the root or atlas side: it's not a blocker
Marco:
o No blockers at the moment
o Caching of LoKi functors: working on a prototype in the past week. By next week numbers about stripping can be hopefully provided
o Circulated a rootfile: a checksum mismatch. Fixed in the newest version of ROOT but the issue stays for the files produced with the latest newest versions of ROOT.
Bill:
o Invested most of its time removing TType usage. TType caused parsing and this inflated memory. Now also TType creation for Enums and refs is suppressed. Pointers and C style arrays are left: pointers could be easy, C style arrays are not hard but could be ugly - need to discuss with Philippe. TType will then be gone for good.
o std bad alloc: might have a hunch - it could be a wrong number of indirection in the cms patch of TClingCallFunc. This is an important one: it could hide other errors.
o IB: errors which are not reproducible. For example bytecount mismatches
o ~TGenericClassInfo crash: not seen anymore, likely to have been fixed.