Fix PETSc error in mesh.adapt() with extra MeshVariables

[lmoresi]

Summary

• Stale field_ids on old variable lvecs caused createSubDM to fail during variable reinitialization after mesh adaptation
• Fix: destroy all old lvecs upfront before reinitializing any variable, so _setup_ds()'s internal backup/restore loop never encounters stale vectors
• The metric field is now preserved through adaptation (it's user-created and may have external references), rather than being silently deleted

Test plan

• Exact reproducer from issue AMR error after Stokes solve. #48 runs without PETSc errors
• All variables (including metric) survive adaptation with correct field_ids
• Verbose output confirms all variables reinitialize cleanly

Fixes #48

Underworld development team with AI support from Claude Code

[Copilot]

Pull request overview

Fixes a PETSc “Invalid field number …” failure during Mesh.adapt() when additional MeshVariables exist (e.g., created after a solve), by ensuring stale vectors/field IDs cannot be used during variable reinitialization on the adapted DM. Also keeps the user-supplied metric variable registered on the mesh across adaptation.

Changes:

• Preserve all mesh variables (including the metric field) for reinitialization after adaptation, instead of excluding/deleting the metric variable.
• Destroy all existing variable PETSc vectors up front before any _setup_ds() calls, preventing _setup_ds()’s internal backup/restore from encountering stale field IDs (fix for #48).

Comments suppressed due to low confidence (1)

src/underworld3/discretisation/discretisation_mesh.py:3060

• Consider adding a regression test that exercises mesh.adapt() when extra MeshVariables exist (e.g., after a Stokes solve) and verifies that adaptation completes without PETSc field-id errors and that all variables (including the metric field) remain registered with updated field_id values. This change fixes a previously reported failure mode (#48), so a targeted test would help prevent future DM/field ordering regressions (similar to existing DM rebuild regression tests).

        # Destroy ALL old vectors upfront before reinitializing any variable.
        # This is critical because _setup_ds() iterates mesh._vars to backup/restore
        # data — if some variables still hold lvecs with stale field_ids from the
        # pre-adaptation DM, createSubDM will fail on the new DM.  (Fixes #48)
        for old_var in old_vars_data.values():

---

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