Roundtable
Alex
Studying on differential forms.
Doga
Dataset. Too many secondary vertices in the light jets. Visible in b-jets, where we only expect 2 vertices. Continue investigating.
Carlo
Seeding and filtering. Looking at fake seeds and started a NN. CPU performance improved but memory consumption went up.
Bump to 43 on Friday created a small issue.
Paul: 43 had some breaking changes also with cmake, so going back and forth is a bit cumbersome.
Andreas Stefl: Maybe bisect only in ACTS standalone. Not guaranteed thought that the slowdown is also visible there.
Noemi
Andreas Salzburger
(Big beautiful) Python bindings are progressing well. cmake now generate a python plugin named file and can already fail there if something is missing.
Benjamin
Progress on GSF side: integrated a multistepper, so also sympy can be used. Curious how it improves performance in Athena.
Merge pipeline seems slow at the moment, what is the issue?
Paul: When everyone presses "update" it slows down. Also CI-failurate is non-zero (maybe 5% failure rate?). This adds up. Somehow also the final step does not work with the webhook. Also our microsoft allocation is shared between all our repositories. Also sonarcloud got a bit slower because it depends on 2 jobs.
Maybe get rid of the gitlab connection and put it only on github and hope no-one exploits it.
In github it is decided if docker or host. gitlab is always docker. That's why github always starts a new virtual machine to get rid of the problem.
Module map: we would need to get rid of every phi slice. Discussion with Alexis on how to do it. Alexis is unsure, if it would work, because phi-modules are not extremely stable.
Attila: Who is responsible? -> There is an ATLAS group in Toulouse. They are invested in the module map.
Their code repo is probably not integrateable into ACTS right now. A bit hard to interact with for outsiders.
Paul
Worked on detray conversion and understood the material map a bit better.
got the surface array svg writing a bit better.
apogee CI was broken for a bit but works now. 3 unittests now time out. (regional data prep).
A bit dangerous, so maybe we need to revert and cherry-pick to a newer version?
Attila
Data model on all levels. Trying to look at RD datamodel. Introduced a new class structure to see how far it can be pushed. Planned discussions with PF.
release 21 is still used for generating events, so got back to use gcc6
PF
Layer0-b-tagging for NGT. started to revamp stuff christoff sauer was doing.
Plan to work on checking the changes how to treat the strip hits.
For the next report, DL is in October 10th:
Paul: Maybe freeze the workflow at a fixed date and don't change until the report is done.
PF: reasonable DL for that desired.
Andreas Salzburger: It doesn't make sense to squeeze 5% out if they are untested.
PF: If we are close to a threshold, we can push a bit more
Paul: we are already 20% inside the threshold.
Joana
Andreas Stefl
