Choose timezone
Your profile timezone:
ROOT Team Meeting
Description
Zoom link in announcement email; please contact rootdev@cern.ch if you did not receive it!
# ROOT team meeting 2023-07-17
## Find notetaker
Jonas Rembser
## News
No news.
## Shift handover
Vincenzo was on shift, nightly summary reported. It's not too bad. We'll follow up on the Vc warnings.
Pretty busy week on the forum.
The new CI is still red quite often because jobs fail frequently.
Jonas H: there is a way to rerun jobs from the GitHub interface.
Next shifter is Axel who is still away for 2 more days, but he will pick up the shift on Wednesday.
## Meeting Summaries and Plans
### IO meeting
Discussion on inconsistency between https and xrootd file globbing (GitHub issue #13258).
### TMVA
Presentation by GSoC student on what still needs to be implemented for Graph neural networks.
### RooFit
No meeting.
### Planning and godparents
No meeting.
### LIM
Meeting this afternoon, but nothing relevant to report.
### PPP
Dante presenting the RBatchGenerator.
Creating batches of RTensors from TTrees via RDataFrame.
Good discussions, it was concluded that the C++ implementation details should be hidden from the user.
Next monday also presentation by Dante on his work in the SFT group meeting.
## Issues
All issues of last week have been assigned.
## Topics
No topics.
## Round table
* Bertrand: working on making the debug build work on Windows. Not easy because
all the dependencies need to be build in release mode. Already discovered
several bugs with this and pinged the relevant people to fix it. Also working
on making the Windows build nodes less accessible after getting a ping from
IT.
* Philippe: working on the GitHub issues and
* Jonas R.: On vacation last week, this week giving ROOT lecture at the HASCO
school in Goettingen, catching up with forum and students.
* Jonas H.: Working on the LLVM update for LLVM 16. Solving some issues there
with Clad and the atexit transactions. PR later this week.
* Victoria: finished framework for targeted bit flip tests, which will be used
to produce the results for the summer student poster session on the 25th.
* Florine: working on the `std::set` support in RNTuple and on collection
iterators. Trying to figure out the remaining problems with Javier.
* TMVA summer student 1: Working on parsing a Keras model to RModel so that it
can be passed with Sofie. Next will be the same with xgboost.
* TMVA summer student 2: Exploring options to get GNN support into Sofie.
* Silia: Working on live visualization feature for Dask workers. First
prototype working now.
* DistRDF summer student: Working on distributed benchmarking with RDF. Setting
up the software stack, which is not trivial especially on the HPC cluster in
Juelich. Spent some time to understand the existing benchmarks.
* Jolly: benchmarking her GPU decompression and helping people to set up
opensycl. Away for 2 weeks now.
* Martha: finalized progress bar and looked into expanding tutorials. Software
carpentries with Vincenzo last week. Many people were there (about 50).
* Daniel Werner: working on comparing pyhf and roofit, managed to set up the
same statistical workspace in the JSON format in pyhf and RooFit. Next is to
understand why the limit results are different.
* Dante: working on the batch processing PR which takes some time. Maybe
developing new features in parallel.
* Vincenzo: Last week: preparation for the software carpentries and working
with Dante. Was on shift, noticed that people often ask how to fill a TTree
branch from jagged python collections. How to do this with raw TTree, for
example: create a std::vector in PyROOT, register it as a column and fill
event by event, clearing the vectory after each event. There is also another
way with raw arrays and an additional branch for the size.
Jakob: is there no way to do this with RDF? Yes, with the awkward to RDF
conversion, but this is not what users want. They want to add the jagged
branch to the TTree in a file, not only in the RDF view.
Gave a presentation in the LHCb meeting remotely. About 60 people connected
and several questions and comments. For example, again a request for group-by
feature. Also about the export of numpy arrays from `bool` branches. Right
now they convert to numpy arrays of type `object`, which is not ideal. The
proper fix is maybe included in cppyy upstream.
Started the effort of migrating to C++17 as the minimum standard version.
There is still one blocker that is discussed with the relevant people.
This week: PPP meeting at 16:30 with Silia presenting her prototype.
* Jakob: working on internal API improvements of RNtuple.
* ???: Something with CI CD.
## A.O.B.
Vincenzo: Is it intentional that Jenkins CI doesn't run the tutorials on all platforms?
Jonas H: Yes it was, to optimize CI running time. But the new CI will always
run all the tests. And you can also ask Jenkins to run all tests on all
platforms if you want, with the right "@phsft-bot build ..." command.
There are minutes attached to this event.
Show them.