fix(constrained): guard empty stratum IS in interior-multiplier reduction (#291)

[lmoresi]

Closes #291.

Summary

`Stokes_Constrained` segfaulted at np>1 in
`_constrain_interior_multipliers_in_section`
(`src/underworld3/cython/petsc_generic_snes_solvers.pyx:7001`).

The root cause was the boundary-IS guard. On a rank that owns zero points
with a given boundary label value,
`dm.getLabel("UW_Boundaries").getStratumIS(bvalue)` returns a valid PETSc
IS with `getSize() == 0` — not None, non-zero handle, and
`bool(IS) == True`. So the existing guard `if bd_is is not None:` passed
through, and the subsequent `bd_is.getIndices()` crashed with SIGSEGV.

Change

`petsc_generic_snes_solvers.pyx:7001` — replace

```python
if bd_is is not None:
```

with

```python
if bd_is and bd_is.getSize() > 0:
```

`bool(bd_is)` short-circuits the null-handle case (the same idiom
`utilities/boundary_flux.py:70` uses); `.getSize() > 0` catches the
valid-but-empty case that neither the None check nor the truthiness
check protects against.

Verification

Lightweight unit-viscosity repro (16×16 quad, sinusoidal body force,
free-slip on all four walls; `_reduce_interior_multiplier = True` default):

| np | `|u|_inf` | wall | pre-fix |
|----|-------------|-------|---------------|
| 1 | 2.533e-02 | 6.6 s | (unchanged) |
| 2 | 2.533e-02 | 8.5 s | SIGSEGV |
| 4 | 2.533e-02 | 5.0 s | SIGSEGV |

Bit-identical parallel-vs-serial in this regime.

`tests/test_1062_constrained_solcx.py` (canonical SolCx `eta_B=1e6` test):
serial passes (198 s); the np=2 with-mpi rerun exceeded the 300 s timeout
under this session's compute (`Stokes_Constrained` is 34× slower than
Nitsche per #244), so a longer-timeout parallel variant of this test
(probably `tests/parallel/test_1062_constrained_solcx_parallel.py`) is
warranted as follow-up. Not blocking — the guard is a strict guard-widening
that cannot produce wrong physics; it only stops a previously-crashing rank
from calling `.getIndices()` on an empty IS.

Related

• Issue Stokes_Constrained segfaults at np>1 in the interior-multiplier section reduction #291 — this fix.
• Stokes_Constrained is slow in parallel despite low KSP iterations #244 — the `Stokes_Constrained` parallel-slow issue that stops
  `test_1062` np=2 fitting a 300 s timeout on this machine.

Underworld development team with AI support from Claude Code

[Copilot]

Pull request overview

Fixes a parallel crash in Stokes_Constrained interior-multiplier reduction by correctly guarding against valid-but-empty PETSc stratum IS objects (a case that occurs when a rank owns zero points for a given boundary value). Also adds explicit rank-0 warnings when the rotated free-slip helper KSPs diverge, preventing silent “last-iterate” returns.

Changes:

• Widened the getStratumIS guard in _constrain_interior_multipliers_in_section to skip empty strata and avoid getIndices() segfaults in MPI.
• Added _warn_if_ksp_diverged() and invoked it in both direct-LU and fieldsplit-Schur rotated free-slip solve paths.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
src/underworld3/cython/petsc_generic_snes_solvers.pyx	Prevents MPI segfault by guarding empty boundary stratum IS before calling getIndices().
src/underworld3/utilities/rotated_bc.py	Adds rank-0 warnings on KSP divergence to avoid silently returning a partial rotated free-slip solution.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[lmoresi]

Thanks @copilot-pull-request-reviewer — added in 958f6f4:

1. New lightweight parallel regression at tests/parallel/test_1062_constrained_stratum_guard_parallel.py — unit-viscosity 16×16 quad box, free-slip on all four walls at np≥2. On a 2-way axis-aligned partition this puts one full boundary on each rank and zero on the other, triggering the empty-stratum path from the guard. Runs in ~10 s at np=2 and asserts |u|_inf matches the serial GOLDEN to rtol=1e-3. Pre-fix this test SIGSEGVs; post-fix it passes.
2. Un-skipped test_parallel_custom_fmg_stokes_constrained (tests/parallel/test_1017_custom_mg_parallel_mpi.py) — its skip reason explicitly said "auto-enables once Stokes_Constrained segfaults at np>1 in the interior-multiplier section reduction #291 is fixed." Now marked slow (the SolCx eta_B=1e6 benchmark it uses is slow, per Stokes_Constrained is slow in parallel despite low KSP iterations #244) so CI can select via -m "not slow" if needed.