Per-spoke solver for xhat spokes + close solver-options-redesign §4

[DLWoodruff]

Draft. Bundles two related pieces of "solver work" on the solver-options surface.

1. Per-spoke solver for the xhat inner-bound spokes (code)

The outer-bound spokes already accept --<spoke>-solver-name, -solver-options, and -solver-options-file (lagrangian, reduced_costs, subgradient, relaxed_ph, ph_dual — all via Config.add_solver_specs(prefix)), and FWPH has its own --fwph-mip-solver-name / --fwph-qp-solver-name. The xhat inner-bound spokes did not: their *_args helpers never called add_solver_specs, and their factories never called apply_solver_specs.

This adds the full trio to the xhat spokes:

• Register — add_solver_specs("<prefix>") in xhatlooper_args, xhatshuffle_args, xhatspecific_args, xhatxbar_args, xhatlshaped_args.
• Consume — apply_solver_specs("<prefix>", spoke, cfg) in the five matching spoke factories in cfg_vanilla.py.

apply_solver_specs updates iter0/iterk_solver_options in place (never reassigns), so each xhat factory's nested xhat_solver_options reference to iterk stays valid. Because the three flags register independently, "different options but the inherited solver" also works (supply only --<xhat>-solver-options).

So, e.g.:

generic_cylinders ... --solver-name gurobi --xhatshuffle --xhatshuffle-solver-name xpress

now runs the PH hub on gurobi and the xhatshuffle inner-bound spoke on xpress.

Regression tests added to mpisppy/tests/test_solver_options_layers.py (already wired into the coverage harness):

• all five xhat *_args register the solver trio;
• xhatshuffle_spoke routes per-spoke solver name + options (and the nested xhat_solver_options) into the spoke.

Corrects and will close #805 (whose original problem statement overstated the gap — the outer-bound spokes already had these flags).

2. solver_options_redesign.md §4 open questions closed (docs)

The solver-options redesign shipped a while ago (PRs #696–#703). This marks §4's open questions #1–#4 Resolved, matching the shipped implementation (#5 was already marked). No behavior change.

Notes / not done here

• xhatlooper / xhatspecific now register the flags too, but those spokes aren't wired into generic_cylinders (they're used by other drivers) — intentional.
• The doc's §1 "as-is" section and top "implementation in progress" status line are now stale (the redesign fully merged); left as a follow-up rather than rewritten here.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR extends mpi-sppy’s solver-options “per-spoke” configuration surface to the xhat inner-bound spokes (bringing them in line with existing outer-bound spokes), and updates the solver-options redesign design doc to mark several previously-open questions as resolved.

Changes:

• Register per-spoke solver specification flags for xhat spokes by adding add_solver_specs("<xhat>") in the relevant Config *_args helpers.
• Apply per-spoke solver specifications inside the corresponding xhat spoke factories via apply_solver_specs("<xhat>", spoke_dict, cfg).
• Add regression tests verifying (a) xhat args register the solver trio and (b) xhatshuffle_spoke routes per-spoke solver name/options into the spoke options; plus doc updates resolving solver-options redesign §4 questions.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
mpisppy/utils/config.py	Registers per-xhat-spoke solver flags via add_solver_specs() in the xhat *_args helpers.
mpisppy/utils/cfg_vanilla.py	Ensures xhat spoke factories consume per-spoke solver specs by calling apply_solver_specs() before building xhat-specific option sub-dicts.
mpisppy/tests/test_solver_options_layers.py	Adds regression tests for xhat per-spoke solver flag registration and routing (via xhatshuffle_spoke).
doc/designs/solver_options_redesign.md	Marks solver-options redesign §4 questions as resolved and references the shipped implementation behavior.

---

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