Dune Developer Meeting 2023 in Dresden

Date

The developer meeting is held on September 19th and 20th, 2023 at the Technische Universität Dresden.

The meeting is co-located with the Dune User Meeting 2023.

Participants

Simon Praetorius
Oliver Sander
Robert
Carsten
Christian Engwer
Santiago Ospina
Christoph Grüninger (only Wednesday, only remote)
Andreas Dedner (remote)
Markus Blatt
Antonella Rirtorto
Timo Koch

Video conference

The developer meeting will be streamed and allows online participation via BigBlueButton.

Start BBB videoconference

Proposed topics

Please use the edit function to add your topics or write an email to the organizers.

General

CGü: Propose appointing Santiago Ospina De Los Rios ( Heidelberg University, IWR) as a Dune Core Developer
RK: Developers (old and new) and future management structure!
RK: Stability of the gitlab infrastructure. How can we help to improve it?
CGü: I’d like to see an hour of cleaning up. Together we scan (one or two of)
- merge requests
- bugs
- the remaining open FlySpray tasks
- Git branches and decide which can be closed / deleted right away and which to keep.
CGrü: Release 2.9.1 is blocked by two topics, I did not receive too much feedback or support.
AD: 2.9.2 release and new release paper

dune-common

OS: Simon and Santiago have invested a lot of time thinking about improvements to the build system. Can we decide on a way forward? Can some of the work be merged?
SO: MPIHelper dune-common:1274
- RK: What about MPI_Comm_dup while we are at it?
SO: Add a fixed-size multi index: Dune::TypeTree::HybridTreePath -> Dune::HybridMultiIndex.
SO: What is a Matrix? dune-common:#253
CGü: Who is interested in SIMD (Vc-devel)? It is currently broken in master and releases/2.9, blocking the 2.9.1 release and std::simd is making further progress.
CGr: Get rid of greedy generic implementations (operator<< and FieldTraits).

dune-geometry

AM: GeometryType::none default to simplex behaviour. Is this on purpose? dune-grid!32
AM: Higher order derivative information and transformation

dune-grid

VirtualGrid (aka PolymorphicGrid) (roadmap?) dune-grid:661
OS: Can PolymorphicGrid simplify the Python bindings?
OS: Can we get dune-gmsh into dune-grid?
OS: Can we get dune-vtk into dune-grid?

dune-localfunctions

RK: Shape function evaluations should be thread safe. dune-localfunctions:22
SO: RT shape functions only for tets dune-localfunctions:20
OS: Support for matrix-valued FE like the Arnold-Winther element

dune-istl

Discretization modules (FEM, PDELab, …)

Python support

OS: What is the current state of the Python bindings?
OS: Can we get some/better documentation about how to write bindings for you own code? (For example, dune-functions bases?)
OS: Are Julia bindings conceivable?`

etc

RK: To twist or not to twist? I would like to hear opinions dune-alugrid:89
AD: Perhaps we should consider if we have ideas for GSoC this time.

Protocol

Vote on Santiago Ospina De Los Rios ( Heidelberg University, IWR) becoming a Dune Core Developer
9 yes, 0 no, 0 abstentions
Infrastructure
- Jö is still maintaining the mailinglist server. Can we transfer this? -> Markus offers to take over maintainance
- gitlab is now fully running in Dresden. Ansgar is part-time taking care.
- some runners are broken, see Issue ???
Cleanup MPIHelper
- dune-common:1274
- Should we use MPI_Comm_dup? -> transition code that can be enabled/disabled
- MPI_Comm_dup should be fixed along the semantic interposition problem (different instances of the object in different ocmpilation units).
SIMD interface to Vc
- latest Vc versions don’t work with Dune anymore (see core/dune-common#350).
- first we should check which Vc versions work with the old interface
- second we should add interfaces to the new versions
- potentially bump the requirements or even suppport both versions
- Christian will take care of this.
Santiago will prepare a proposal for a HybridMultiIndex.
Cleanup CMake
- no body is using ${ProjectName}-config.cmake.in directly, so Santiago and Simon are free to change that part.
VirtualGrid … Roadmap?
- Naming? yes: PolymorphicGrid
- Communication doesn’t yet work fully. -> to be checked
- A couple of optimizations are proposed. -> postpone
- Should this be merged into dune-grid (in namespace Dune::experimental), in particular as Samuel left? -> not yet, leave the MR open
Can we merge dune-gmsh4 and dune-vtk into dune-grid?
- we start with a small change in dune-grid to suggest dune-gmsh4 and potentially forward when reading gmsh4 files.
- there is common interest to merge gmsh4 on the longer run and we are willing to invest work.
- go in multiple steps to (1) merge a “minimal” Gmsh4 reader with out higher order support (2) check where the grid factory interface needs to be extended (3) merge/update the remaining higher order stuff…
See issue core/dune-localfunctions#22
- we must properly document the guarantees (add to the main dune-localfunctions documentation) (-> Santiago & Carsten)
- shape function evaluations should be thread safe if the local functions objects are thread local
- this implies that shared state between different instances is at least dangerous and must be handled with great care.
- potential raise conditions due to shared state (e.g. std::shared_ptr in the monimial class of the generic finite elements) must be fixed. (-> Robert)
Future development process / project management processes
1. Santiago: The entrance level is quite high and the mailing list is not really used anymore. How can we lower the entrance level?
  - can we make gitlab more open? -> service desk? -> enable github,google,etc. login -> ?
  - can we offe some other communication platform? -> we will try a matrix channel (via the official matrix server).
  - we want to improve the “Getting involved” information.
2. should we formalize certain processes, e.g.
  - how do I become a (core) developer?
  - which responsibilities come with core dev status?
  - when/how do people become alumni?
  - how are interface changes decided?
  - how do we decide on new feature?
  - -> a lot of discussion not written down.
How is the general opinion on Julia bindings? WIAS is considering building upon DUNE for their Julia PDE solvers -> we welcome contributions, at least initially as a dune-julia module/repository.
Can we decide a way forward for drop or simplify pip & venv magic during configure (dune-common:330)?
During CMake python packages are installed implicitly from the web, this is dangerous and inconvenient (e.g. DUNE does not build without internet access (dune-common:329)](https://gitlab.dune-project.org/core/dune-common/-/issues/329)).
- get rid of automatic pip install; unclear what to do without it
- Request to document the different ways to setup Python bindings / venv. With other words: How should one use dunecontrol to build core modules without breaking separated builds, if the automatic download of PIP packages is disabled.
Improve CMake, we have two goals:
1. Have working find_package(dune-grid) for Dune modules.
2. Export pkg-config files for projects outside Dune.
3. Dune superbuild support Steps:
- Invers the way the config.h files are generated: do the generation in the module not downstream
- This may have some implications and needs to be investigated in detail. Requires localization of #defines and find_package calls. Decisions:
- Move detection of SuiteSparse and according config.h variable definition remain dune-common
- Sandiago and Simon are preparing three pull requests and will notify the community when they are considered to be ready for merging
General
- MB: allow branch opm in core modules?
  - No objections
  - use rebase instead of merge to stay up-to-date with branch branched from
  - make opm branch protected in gitlab
- AD: 2.9.2 release and new release paper
  - Timo and Simon discuss a release paper
  - maybe in the same journal as deal.II
Dune-common:
- CGr: Get rid of greedy generic implementations (operator<< and FieldTraits).
  1. operator<<
    - Use basic_ostream as first argument
    - Maybe remove/deprecate this operator<< completely. Carsten prepares a MR.
  2. FieldTraits
    - Not-specialized FieldTraits just forward the arguments.
    - Consensus that Field[Vector|Matrix] should not be parametrized with something not related to a “field”, but the precise definition of “field” is still open.
    - Try to add a deprecation warning for the default/fallback implementation of FieldTraits and inspect the outcode, maybe fix the cases where FieldVector is used in a strange way.
Dune-geometry:
- AM: GeometryType::none default to simplex behaviour. Is this on purpose?
  - using GeometryType::none in referenceElement() is UB
  - it is not a bug
- AM: Higher order derivative information and transformation
  - what would be the return type of a second-derivative?
  - Alternative: method to apply the second derivative to a vector or matrix
  - we need a concrete proposal, e.g. a MR!
Dune-grid:
- OS: HierarchicalGrid vs Grid in python bindings
  - proposal: add method grid() in the gridView object in addition to the hierarchicalGrid() method, for consistency
  - This addition should not be a problem. MR or feature-request required.
Dune-localfunctions:
- MP: matrix-valued FE?
  - LocalBasisTraits? dim-range and range-type, why are there both?
  - Range-type does not need to be a FieldVector
  - dim-range could be the dimension of e.g. a FieldMatrix
  - Traits interface is actually not precisely defined and thus allows to use something else
  - Maybe update the tests
- MP: Hermite-type / transformed FE. Where to put these?
  - Does not really fit into dune-localfunctions
  - Leave the implementation currently in dune-functions
etc:
- RK: twist-free grids
  - currently some issues with hex grids
  - there is a branch in alugrid but not yet merged
- SP: periodic refinement in ALUGrid, Status? Roadmap?
  - RK: will try to look into it.
- AD: Perhaps we should consider if we have ideas for GSoC this time.
  - Maybe helpful in juliy bindings development?
  - SO: use grid concepts to more easily implement a new grid wrapper (e.g. gmsh, petsc grids)
  - CG: would do the management again
  - about 2 month before the application deadline, we start collecting the project ideas.

| Legal Statements / Impressum | Hosted by TU Dresden | generated with Hugo v0.111.3 (Nov 20, 23:25, 2024)