ROOT Team Meeting
ACTIONS
Lorenzo: provide a statement about some workflows parallelisable with ROOTMPI in the area of math/stats/ML
Lorenzo: We need to clean up the automake remnants of the math packages
Lorenzo: Port tests of GenVector to CMake
Guilhermne: Ask witek a debug session to install gsl-dev on macitois17 to make tests succeed
Danilo: fix for i686 assertROOT7027 which is failing
Martin: Eliminate all old icc builds (icc version < 17)
Kim: make PR705 being accepted by Lorenzo
Vassil: point to the presentation for Intel where the tool is used on the open PR
DECISIONS
(not really a decision...) There is consensus about the need of a fixed date for discontinuing the classic build.
Shift
o Olivier is next. Friday he will not be there: we'll cover this day.
Root MPI
o Who is interested in ROOTMPI? ET+GA
- Apparently we do not have any usecase
- Omar suggests fitting and machine learning tasks
o What are the alternatives: ROOT per se or kept as an external project? BB
o Perhaps we shall avoid the dump of 10k lines with an author which leaves. At least this must be split as many PRs GA+VV
- Downside: requires work by the author DP
o Perfect candidate to start the modularisation GA
- Can we make it less intrusive? E.g. at the level of plugin manager? Can we avoid to extend something? VV
o Perhaps we can start as an external package BB
o We shall understand what is the minimal patch in order to compile ROOTMPI against ROOT VV
Failing tests
o GA worked on the failing tests and warnings, during last week: quite some progress made
o GA+VV We need to have a timeline to discontinue the classic
- BB Does not work on Windows
- GA We need to test the classic in the PR since it breaks too often
- VV Some packages even have automake to build (See actions)
- Makefile can be used for small satellite packages distribnuted with ROOT
- BB We need to review the missing features like alien...
Google benchmark
o DP We need to have a machine dedicated to us, physical only to do performance
o VV We need to decide if we want to track performance regressions, and Vassil thinks this is the case
Round Table
Vassil:
- Fix the case where mathvectypes falls back to scalar types if VC is not available: Double_v becamse Double_ty
- Fix nasty ABI issue (incompatible between Clang and Gcc) when passing a tuple as function argument. Pushing it to LLVM
- Fix regression due to our patches. Next LLVM we can use the LLVM proper. We "just" build clang on top.
Olivier:
- Bug in TGraph2D painter fixed
- Include the list of libs relative to a class: thanks to VV we have a way to retrieve the info. Doxygen though cannot offer the poissibility to add custom sessions but we can postprocess the HTML replacing the collaboration diagram
David:
- Measure the phases of ROOT IO on experiments jobs. For example unzip, deserialisation and reads.
- Tool used instrumenting ROOT
- VV Perhaps this can be used upstream to measure the performance of IO in a more streamlied way?
- DS This is at the moment a bit ad hoc but why not
Harry:
- NTR
Guilherme:
- Committed to have CDash completely green
- About to fix some warnings for v6.10
Kim:
- TMVA xvalidation: structural changes needed
- One pending PR 705 touching TH1 (see actions)
Enrico:
- Following items for 6.12 TDF sprint. Most interesting: nice and transparent way to fill histos with a good range and binning.
- Opened ticket about rms changing when zooming, permanently
Rafael:
- Adding the reflection information pcms
Martin:
- Back from vacation, already got issues ironed out in the infrastructure
- Refactoring the stresshistogram: replacing eg the raw pointers and stack allocation with heap allocation
Enric:
- Teaching effort: the ROOT tutorial
- Still thinks we need to re-organise so that they are ready to run the exercises
- Teaching effort 2: lectures for the CSC
- Mediating between Beams and Librarians to include NXCals software in LCG89
- 6.12 sprint: evaluating the usage of TMemFile to implement cache of TDF
Bertrand:
- The JS gui is working. Using openUI, not very easy to usbut nice
- Reducing failing tests cling on Win: 3 different ways of running tests and 3 different numbers of failures. Kind of fragile and takes time.
- Compile ROOT on windows: very difficult
Danilo:
- Fixed with Enrico loose ends of the TDF
- Released 6.10/02 and debug jenikns
- Fix issue with terdown of ROOT when using imt on mac
- Heavy load on the forum
- Teaching, ROOT course and CSC slides preparation