Segfaults in periodic boundary processing

[bmunguia]

Describe the bug
I frequently get a segfault when running a simple problem with periodic boundaries (inviscid vortex on square mesh). It only happens with some meshes (e.g. the attached TriAdapt1.su2, but not TriInit.su2). CPhysicalGeometry::MatchPeriodic() seems to give the correct number of matches, but then the segfault occurs in CGeometry::PreprocessPeriodicComms():

==99060== Invalid write of size 8
==99060== at 0xD26DC8: CGeometry::PreprocessPeriodicComms(CGeometry*, CConfig*) (CGeometry.cpp:973)
...

(and in 3 other locations). It only occurs when running in parallel.

I've attached the full valgrind outputs (serial and parallel), as well as the meshes and config I used.

Bug report checklist

• Consulting the build instructions (https://su2code.github.io/docs_v7/Build-SU2-Linux-MacOS/).
• Looking for similar problems on GitHub or CFD-Online (https://www.cfd-online.com/Forums/su2/).
• Updating to the newest version of SU2 (either develop, master, or the pre-built executables https://su2code.github.io/download.html).

Desktop (please complete the following information):

• OS: Linux (Fedora 42 and Ubuntu 22.04)
• C++ compiler and version: GCC 15.1.1 and 11.4.0
• MPI implementation and version: MPICH 4.2.2 and 4.0
• SU2 Version: 8.2.0 (develop)

config.zip
meshes.zip
valgrind_out.zip

[bmunguia]

Mach contours after 15 time steps

Serial

Parallel

[bigfooted]

For the parallel run, can you visualize the mpi ranks (paraview output RANK)? I guess the mpi interface is at the location of the discrepancy?

[bmunguia]

Yeah the point on the left boundary is also on one of the MPI interfaces.

[bmunguia]

The screen output of CPhysicalGeometry::MatchPeriodic() is

Matching the periodic boundary points for marker pair 1.
 Matched 6 points with a max distance of: 0.
Matching the periodic boundary points for marker pair 2.
 Matched 12 points with a max distance of: 0.

I added a print statement before the loop where the segfault occurs, printing the number of periodic send/recv points for relevant ranks

rank: 4
  nPeriodicSend: 1
  nPeriodicRecv: 1
  nPoint_PeriodicSend[nPeriodicSend]:  2
  nPoint_PeriodicRecv[nPeriodicRecv]:  2
rank: 6
  nPeriodicSend: 1
  nPeriodicRecv: 1
  nPoint_PeriodicSend[nPeriodicSend]:  4
  nPoint_PeriodicRecv[nPeriodicRecv]:  5
rank: 7
  nPeriodicSend: 3
  nPeriodicRecv: 3
  nPoint_PeriodicSend[nPeriodicSend]: 11
  nPoint_PeriodicRecv[nPeriodicRecv]: 10

(should nPoint_PeriodicSend == nPoint_PeriodicRecv?)
And from another print statement I added after the loop, checking for ranks where the index exceeds the allocated array sizes

ii exceeded nPoint_PeriodicSend!!!
  rank: 6
  ii: 5
  nPoint_PeriodicSend[nPeriodicSend]: 4
  nPoint_PeriodicRecv[nPeriodicRecv]: 5