L-shaped: fix subproblem indexing when root_scenarios is used

[DLWoodruff]

Fixes #551 (also explains the crash in discussion #542).

The bug

farmer_lshapedhub.py (like any L-shaped run using the root_scenarios option) builds one scenario directly into the root problem. Such scenarios have no eta variable and no Benders subproblem — root.eta is indexed by the non-root scenarios only. But set_ls in mpisppy/utils/lshaped_cuts.py sized global_subproblem_count and the local-to-global subproblem index map over all scenario names.

With a single hub rank the misindexing is coincidentally harmless (the positional keys and the dropped root scenario line up, and the extra array slot stays zero) — which is why the 2-process run in run_all.py never crashed. With two or more hub ranks, a rank owning a scenario that comes after a root scenario writes its cut data to a global slot past the end of all_root_etas, producing the reported

File ".../pyomo/contrib/benders/benders_cuts.py", line 379, in generate_cut
    root_eta = self.all_root_etas[global_subproblem_ndx]
IndexError: list index out of range

Note this is narrower than the diagnosis in the issue thread: mpi-sppy's LShapedCutGeneratorData subclass already bypasses Pyomo's call-everywhere/argmin rank-assignment protocol by imposing mpi-sppy's own scenario-to-rank map — the map was just wrong when root_scenarios was nonempty.

The fix

set_ls now indexes over the non-root scenarios only, matching the eta ordering built by lshaped.py (getattr(ls, "has_root_scens", False) so the cross_scen_spoke.py caller, whose opt object has no such attribute and uses all scenarios, keeps its exact previous behavior).

Tests

• mpisppy/tests/test_lshaped_cuts.py (new, serial, no solver): unit tests of the set_ls bookkeeping, including per-rank views of the failing 3-rank configuration. The root-scenario cases fail on main and pass with the fix. Wired into run_coverage.bash and the CI unit-test job.
• straight_tests.py: new end-to-end run of farmer_lshapedhub.py with a 3-rank hub (one scenario per rank, middle scenario in the root problem) — the exact shape of the L-shaped/benders cuts Index error #551 repro. Reproduces the IndexError on main; with the fix it converges to the farmer optimum.

Verified locally with gurobi_persistent: the original -np 6 ... --xhatlshaped repro from #551, plus -np 4 and -np 2, all converge to −108390 with gap 0; netdes with --cross-scenario-cuts still runs clean; test_maximization.py (5 L-shaped tests) passes.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 76.44%. Comparing base (3749661) to head (32bb9f9).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #817      +/-   ##
==========================================
+ Coverage   76.33%   76.44%   +0.10%     
==========================================
  Files         170      170              
  Lines       22628    22631       +3     
==========================================
+ Hits        17274    17301      +27     
+ Misses       5354     5330      -24     

☔ 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.

[bknueven]

This code obviously isn't used much by us 🤦‍♂️