[ggorman]

Under DEVITO_MPI=overlap, overlap2 and full, an equation over a SubDomainSet skipped points near the box edges: the halo scheme only honored SubDimension thicknesses, so the box thickness was applied relative to each CORE/OWNED region instead of being intersected with it. basic and diag2 were fine.

• HaloScheme now honors MultiSubDimension thicknesses too, and records them as runtime thicknesses (they are set inside the Operator, one subdomain at a time). This alone fixes overlap.
• overlap2/full fill the OWNED region struct at argument time, when those values don't exist yet, so the remainder applies them relative to the region bounds instead.

Found through DevitoPRO's ABox (devitocodespro/devitopro#941), which gave wrong FWI gradients under full. New test: test_subdomainset_stencil_mpi fails under overlap/overlap2/full on main and passes for all five modes here. Generated code for regular SubDomains is unchanged.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 31.57895% with 39 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.06%. Comparing base (8bbe0d3) to head (0a1835a).

Files with missing lines	Patch %	Lines
tests/test_subdomains.py	8.33%	22 Missing ⚠️
devito/mpi/routines.py	5.55%	17 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3075      +/-   ##
==========================================
- Coverage   84.23%   84.06%   -0.17%     
==========================================
  Files         258      258              
  Lines       57074    57126      +52     
  Branches     4875     4882       +7     
==========================================
- Hits        48076    48024      -52     
- Misses       8170     8271     +101     
- Partials      828      831       +3     

Flag	Coverage Δ
pytest-gpu-aomp-amdgpuX	68.47% <42.42%> (-0.04%)	⬇️
pytest-gpu-gcc-	78.96% <31.57%> (-0.05%)	⬇️
pytest-gpu-icx-	78.86% <31.57%> (-0.07%)	⬇️
pytest-gpu-nvc-nvidiaX	69.04% <42.42%> (-0.05%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mloubout]

This doesn't look right, having to track MultiSubs seaprately looks overly complicated, and having a haloscheme having separate arguments and runtime_tkns is not right.

[ggorman]

The osx failure is the Apple-silicon detection flake, not this change: the job header reads platform: cpu64 on the arm64 runner, so the cpu64-arm skip does not apply and test_nested_nthreads fails compiling x86 intrinsic headers (https://github.com/devitocodes/devito/actions/runs/39303223750/job/118233531439). #3048 fixes the detection. The other 41 jobs pass.