Control: block 1121012 by 1121516 On 2025-11-28 09:12, Emilio Pozuelo Monfort wrote:
On 20/11/2025 15:04, Drew Parsons wrote:On 2025-11-20 12:13, Emilio Pozuelo Monfort wrote:Control: tags -1 confirmed On 19/11/2025 16:03, Drew Parsons wrote:I'd like to proceed with ABI transitions of the numerical library stack, scotch to petsc. scotch 7.0 → 7.0c (same ABI but soname renamed by switch to cmake) superlu-dist 8 → 9 (finally fixed) hypre 2.32.0 → 2.33.0 mumps 5.7 → 5.8 petsc 3.22 → 3.24 (with slepc)...Sounds good, please go ahead.Thanks. scotch is uploaded now, I'll work down the list once each reaches the unstable archive.There are a few autopkgtest failures blocking the transition, see: https://tracker.debian.org/pkg/scotch
scotch is acknowledged, Bug#1121516. ppc64el. Upstream is looking into it.
Push come to shove we can skip the failing test if necessary. opm-grid/opm-simulators is being worked on, Bug#1121356
https://tracker.debian.org/pkg/petsc
sundials. This I don't understand. sundials is rebuilt successfully, 7.1.1+dfsg1-10+b1 But the petsc migration test with sundials tests the old version, 7.1.1+dfsg1-10 not 7.1.1+dfsg1-10+b1
So of course it fails. Why isn't it testing sundials 7.1.1+dfsg1-10+b1?feenox. Bug#1121362. The problem is a clash between gmsh and opencascade, Bug#1118271, blocked by freecad failure.
petsc will be freed if feenox is removed from testing and sundials tests using 7.1.1+dfsg1-10+b1
https://tracker.debian.org/pkg/fenics-dolfinx
adios4dolfinx migration tests. Looks like a transient failure not related to the transition, it previously passed. adios4dolfinx test_checkpointing.py appears to be a little flaky, not sure how best to manage it (short of skipping altogether). Raised upstream https://github.com/jorgensd/adios4dolfinx/issues/147
dolfinx-mpc. Similar to the sundials petsc problem. Rebuilt successfully 0.9.3-2+b2, but the dolfinx migration test tests dolfinx-mpc 0.9.3-2+b1 not 0.9.3-2+b2. What's the point of the binNMU if it's not being used for transition migration?
Drew