Login

194th ROOT Parallelism, Performance and Programming Model Meeting

→ Europe/Zurich
32/S-C22 (CERN)

32/S-C22

CERN

17
Show room on map
Marta Czurylo (CERN), Vincenzo Eduardo Padulano (CERN)
ROOT PPP Room
Zoom Meeting ID
61666320058
Host
Marta Czurylo
Useful links
Join via phone
Zoom URL
Hide

  • [Stephan Hageboeck] Slides 16-17, could you just read the entire object instead of the data members?

    • Yes, that would work both for TTree and RNTuple. But I'm showing these examples explicitly to show some cases where an analysis works for TTree but doesn't for RNTuple.

  • [Giacomo Parolini] When you say std::array do you also mean C-style arrays?

    • Not considered so far, but it would be hard to enforce Rule 2 in that case...

  • [Giacomo Parolini] Technical challenge to implement RVec<std::vector>?

    • Not really, other than the proliferation of corner cases to support.

  • [Jonas Hahnfeld] Do you want these rules to be exhaustive? RNTuple already supports all the three rules and more, do you want to limit?

    • I like the idea of codifying the RDataFrame behaviour so it is clear, documented and easily explainable whenever a new issue/question arises. And yes, part of this means potentially limiting RNTuple capabilities when processed via RDataFrame

  • [Jonas Hahnfeld] I would propose that we leave everything that is now inconsistent as it is, modulo fixing certain RNTupleDS bugs. We should make sure that TTree-->RNTuple is compatible. Even though the inconsistencies in the combinations of sequence container types are not ideal to look at, I don't think we need to also ensure RNTuple-->TTree compatibility. Thus, I would propose to not codify these rules, and just find the happy path that fixes the FCC schema issue for TTree in a way that the same RDataFrame can also work with an RNTuple data source.

  • [Giacomo Parolini] I agree, and I think we should also document the limitations of the interaction between RDataFrame and the TTree data source.

Conclusion:

  • Do not codify the rules
  • Fix the original FCC schema issue for TTree, ensuring that the same RDataFrame code can work for an RNTuple data source with the same dataset schema.
  • Document limitations of the interaction between the TTree data source and RDataFrame.
There are minutes attached to this event. Show them.
    • 16:00 → 17:00
      Rules for I/O of sequence types in RDataFrame 1h
      Speaker: Dr Vincenzo Eduardo Padulano (CERN)
      PPP_2026_03_12