linearize and make_dual_shadow take a BodyAerodynamics with several wings

[1-Bort-1]

TL;DR

linearize and make_dual_shadow no longer throw on a body with more than one wing: the dual shadow copies every wing, and theta_idxs/delta_idxs run over the unrefined sections of all wings in order. #330's stability derivatives need this to cover multi-wing bodies.

What changed

Only the one-wing guards stood in the way. solve! already handled several wings, and calc_forces! already filled moment_unrefined_dist wing by wing, offset by the section counts of the wings before. The inputs now use that same order, from the same place.

• make_dual_shadow builds the shadow body from _wing_with_eltype of every wing, not just wings[1]. It also passes on Add the Gaunaa et al. 2024 spanwise-flow viscous drag correction as an opt-in solver setting #352's is_with_viscous_drag_correction, which came in with the merge from main.
• A private unrefined_section_range(body_aero, wing_idx) gives the indices of one wing's unrefined sections. calc_forces! now uses it instead of its own running offset, so the outputs and the inputs share one definition of the order.
• A private method unrefined_deform!(body_aero, theta_angles, delta_angles) hands each wing a view of its own range of angles. linearize calls it, then reinit! as before.
• linearize checks theta_idxs/delta_idxs against U, the solver's total number of unrefined sections, which is also the length of moment_unrefined_dist. The result length is 6 + U.
• Cleanup nearby, no behaviour change:
  • om is renamed omega and be is renamed ad_backend.
  • s is renamed section in _wing_with_eltype.
  • The unused smooth=false argument is no longer passed.
  • A wing_pair test helper replaces four copies of the same two-wing setup.

Tests

Three testsets under solve! on a two-wing body in test/body_aerodynamics/test_body_aerodynamics.jl.

• Each wing gets its own angles. unrefined_deform! on the body with a different twist and deflection for every section must give each wing the same theta_dist and delta_dist as deforming that wing on its own with its range of angles. The expected ranges are written out by hand, not taken from the helper. The INVISCID Jacobians below cannot see deflection, so this is the only check on the deflection split.
• Wings 1e4 m apart. Each wing's twist moves only its own sections' moments. Twisting wing 2 moves wing 2's rows the same way twisting wing 1 moves wing 1's: measured relative difference 1.3e-12, test bound 1e-9. The cross blocks are 5.7e-9 of the diagonal blocks, test bound 1e-7. If the split into wings were wrong, these would be off by O(1).
• Wings one chord apart, so they induce on each other. The ForwardDiff Jacobian matches AutoFiniteDiff (steps 1e-6) to a relative difference of 2.7e-7. The test allows 1e-4.

Notes for the reviewer

• Error message: a body with no unrefined sections and some theta_idxs now gets the length-mismatch ArgumentError instead of "Cannot use theta_idxs…". It is the same exception type.
• Still one wing: solve! and linearize on a multi-wing body throw with type_initial_gamma_distribution=ELLIPTIC, from the guard in calculate_circulation_distribution_elliptical_wing. The default (ZEROS) works. Follow-up: ELLIPTIC initial circulation for a body with more than one wing #357.
• Now reachable through linearize: calc_forces! and calculate_results use wings[1].spanwise_direction for the force directions of every panel, so wings with different span directions (a vertical fin, say) get wrong forces from wing 2 onward. The fix changes results, so it is not made here. Follow-up: calc_forces! and calculate_results use the first wing's spanwise_direction for the panels of every wing #358.
• A wing with no unrefined sections cannot reach unrefined_deform!(body_aero, ...): BodyAerodynamics refuses a wing that has not been refined ("Wing 2 has not been refined"), and refining needs sections.
• CI setting: .github/workflows/CI.yml still sets fail-fast: false. That is a separate cleanup: PR, not done here.

Verification

• Reproduced first: both linearize testsets errored with ArgumentError: Linearization only works for a body_aero with one wing (12 passed, 2 errored), on the first version of those tests.
• Deflection split, red then green: with the body method handing every wing delta_angles[1:3], test_body_aerodynamics.jl gives 4884 passed, 1 failed (the new testset, delta_dist of wing 2). All other tests, the Jacobian tests included, still passed. Restored: 4885/4885.
• At 574b15b (juliaserver): test_body_aerodynamics.jl 4885, test/solver/test_unrefined_dist.jl 40, test/solver/test_forwarddiff.jl 7, test/solver/test_viscous_drag_correction.jl 90 (includes calc_forces! zero-alloc), test/body_aerodynamics/test_results.jl 30, test/solver/test_solver.jl 34. No failures.
• Merged main twice. The first merge brought in 1c493cf (Add the Gaunaa et al. 2024 spanwise-flow viscous drag correction as an opt-in solver setting #352), where only CHANGELOG.md conflicted and both entries are kept. The second brought in 83867a3 (Name the apparent wind va, va_vec and va_dist everywhere except the public API #349's va rename, Zero the damping correction when smooth_circulation! skips smoothing, and cover it and lei_poly_coeffs with unit tests #328), where one set of lines in linearize conflicted: main renamed va to va_vec and this branch renamed om to omega, and both renames are kept.
• At merge 0b28330, on a relaunched session: test_body_aerodynamics.jl 4885, test/solver/test_forwarddiff.jl 7, test/body_aerodynamics/test_results.jl 30, test/solver/test_unrefined_dist.jl 40, test/solver/test_viscous_drag_correction.jl 90, test/solver/test_solver.jl 41. No failures.
• Local CI mirror on 0b28330: PASS (9 min, Julia 1.12.7) · GitHub CI on 0b28330: PASS, all 7 checks. On the earlier head 574b15b, Julia 1.12 - ubuntu-latest failed only test_forwarddiff.jl's POLAR_MATRICES check flakes at ~4%, and the oper #287's POLAR_MATRICES check (0.0460232000169766, table A), and this PR's tests passed in that job.
• Earlier Windows flake, not this change: the first Windows 1.12 run on 1d76212 failed test_forwarddiff.jl's POLAR_MATRICES check flakes at ~4%, and the oper #287's POLAR_MATRICES check at its known value and, in the same process, this PR's FD comparison. A re-run on the same commit passed both. The 16 Jacobian entries in that log are bit-identical to a local run, where the relative error is 2.7e-7. The numbers are on test_forwarddiff.jl's POLAR_MATRICES check flakes at ~4%, and the oper #287.
• Docs build: not run locally, because the docs environment has no manifest here; the CI Documentation job passes on 0b28330. unrefined_section_range is listed in docs/src/private_functions.md. No REUSE in this repo.
• Benchmark: n/a
• Risk: a body whose wings have different panel counts gives the shadow a non-concrete Vector{Wing{P,TD} where P}, as the Float64 body already has. It works but dispatches dynamically, and no test covers mixed panel counts.

Scope

+135 / −47 across 5 files. src/solver.jl shrinks (−13 net): three per-wing checks became one loop, and calc_forces! lost its running offset. src/body_aerodynamics.jl +31 is the range helper and the body method. The test file's +67 net is the three testsets and the wing_pair/linearize_body helpers.

Closes #344 · task VortexStepMethod.jl-344

[1-Bort-1]

Independent review (advisory)

Verdict: APPROVE WITH COMMENTS · 1 inline, 0 off the diff

Good

• Matches the card: the one-wing guards in make_dual_shadow and linearize are removed, and the only new source is unrefined_deform!(body_aero, ...); git diff 246237a --stat shows +103/−41 across the 4 files the card names
• U really is the total number of unrefined sections across wings (src/solver.jl:194 sums n_unrefined_sections), so n_results = 6 + U gives the same size the old length(moment_unrefined_dist) did
• Angles are split per wing in the same order calc_forces! fills moment_unrefined_dist (the unrefined_idx += wing.n_unrefined_sections loop at src/solver.jl:474-520), so inputs and outputs line up
• Three validation branches became one loop with no change in behaviour except the error message the card mentions; no test checks the old 'Cannot use theta_idxs' text (grep finds none)
• Dropping smooth=false changes nothing: it is already the default in unrefined_deform!(wing, ...; smooth=false) at src/wing_geometry.jl:428
• The far-apart testset would catch a wrong offset: if wing 2 were handed wing 1's slice, the second diagonal block would be zero or off by O(1) against own_first
• The renames (om→omega, be→ad_backend, s→section) and the wing_pair helper are named in the card as cleanup and are plain Julia locals and test code, not API
• The new docstring is reachable: the bare unrefined_deform! entry in docs/src/private_functions.md:84 picks up every method's docstring
• Every added line fits in 92 columns; the long lines in src/body_aerodynamics.jl were already there

Not good

• test/body_aerodynamics/test_body_aerodynamics.jl:536 — delta_idxs is split per wing, but the INVISCID wings ignore deflection, so every delta column is zero and no assertion reads them. A swapped or shifted delta offset in unrefined_deform!(body_aero, ...) would pass both testsets.
• The far-apart test uses rtol=1e-4 and a 1e-4 cross-block bound, but the card measured 1.3e-12 and 5.7e-9, so the test allows about 8 orders of magnitude more than it needs to pin the split
• unrefined_deform!(body_aero, ...) is a third place that works out per-wing unrefined offsets (with calc_forces! and calculate_results); a shared per-wing range helper would give one source for that order
• If one wing has no non_deformed_sections, the per-wing call now throws 'Call refine!' even for an empty view, where the old code gave a clear idxs error; this is an edge case and nothing tests it
• The shadow body built from wings with different panel counts dispatches dynamically, and no test covers it; the card flags this and it is not blocking
• The card lists the ELLIPTIC guard and wings[1].spanwise_direction in calc_forces! as known limits but links no follow-up issue; now that linearize reaches the second one, it should have an issue

_{claude, rubric CLEAN_CODE.md. A different lab from the implementer
on purpose: a reviewer sharing its blind spots would not flag its mistakes.}

[1-Bort-1]

MINOR: delta_idxs is split per wing, but the INVISCID wings ignore deflection, so every delta column is zero and no assertion reads them. A swapped or shifted delta offset in unrefined_deform!(body_aero, ...) would pass both testsets.

[1-Bort-1]

Fixed in 574b15b: the new testset "unrefined_deform! hands each wing its own run of angles" fails when wing 2 is handed wing 1's deflections (4884 passed, 1 failed).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[1-Bort-1]

Local full suite: PASS (8 min, Julia 1.12.7, one cell of the matrix)

[1-Bart-1]

Address the comments, fix the conflicts, and open a new issue to allow multi-wing to work with elliptic distribution.

[1-Bort-1]

Conflicts: merged main in 4ccdea6 and again in 0b28330. In the second merge, linearize keeps main's va_vec and this branch's omega. Review comments: 574b15b (deflection-split test, far-apart bounds tightened to 1e-9/1e-7, one unrefined_section_range for calc_forces! and unrefined_deform!). ELLIPTIC on several wings: #357. wings[1].spanwise_direction: #358.