CalculiX stateless target: refuse iorien / beta, guard the extern "C" boundary #162

Context

CalculiXExternalTarget (#127) emits a stateless plugin for CalculiX's native call_external_umat_user hook. #160 covers the stateful path (STATEV offset, TIME rebasing, building on numsim-materials' umat/calculix_interface.h). It deliberately keeps the stateless plugin free of numsim-materials.

This issue covers what remains for the stateless path. Three arguments of the native hook change the right answer even without state, but the emitted entry comments them out. A fourth gap is an unguarded exception boundary.

Gaps (compared with numsim-materials umat/calculix_interface.h, same ABI)

1. iorien is ignored. On the native hook, umat_main.f does not rotate into the material frame (umat_user.f:86-104): that is the material's job. An anisotropic elastic recipe used with *ORIENTATION is evaluated in the global frame, and nothing reports it. Isotropic recipes are not affected.
   Suggested: refuse it (zero the outputs, report, terminate) as calculix_interface.h does, or rotate when the recipe is anisotropic.
2. beta (initial stress, *INITIAL CONDITIONS, TYPE=STRESS) is ignored. A preloaded model is then wrong from the first increment.
   Suggested: refuse a nonzero beta.
3. ielas is ignored. This is harmless for a purely elastic recipe. The check becomes necessary once a non-elastic recipe can reach this target, so add it with the scope check.
4. No exception boundary. The entry point is extern "C" and called from C/Fortran. If anything in <Model>_compute or the linear-algebra emitter can throw, the throw terminates ccx.
   Suggested: wrap the call in try { ... } catch (...), zero the outputs, and set PNEWDT (e.g. 0.25) for a cutback. Exceptions stay out of Fortran this way, as umat_dispatch does.

Note for #160 (STATEV)

The header comment in calculix_external.h ("STATEV0/1 = history old/new", "per-integration-point history flowing through STATEV each call") reads as if ccx passed a per-point slice. It passes the full xstate(nstate_, mi(1), *) array, so the callee must index at nstatv*((iint-1) + mi1*(iel-1)). #160 item 1 describes this; the comment should be corrected so nobody extends the stateless entry by hand from it.

On the numsim-materials side, the host-owned path the stateful mode will use is being consolidated into one material-point core (numsim-materials #184, phase 2: load_history / store_history / history_width). calculix_interface.h keeps its interface.

Validation

#128 (golden-file harness): add a case with *ORIENTATION on an anisotropic elastic recipe, and a case with an initial stress. Each should fail loudly, not give a wrong answer.

Related

• CalculiX target: dlopen'd .so material (no ccx recompile per material) #127 (stateless target), CalculiX stateful path: emit against numsim-materials' calculix_interface.h instead of hand-rolling STATEV/TIME/orientation #160 (stateful path), CalculiX material test harness: golden-file suite over tests/calculix/<material>/ #128 (harness)
• numsim-materials include/numsim-materials/umat/calculix_interface.h