[1-Bort-1]

TL;DR

Adds test/solver/test_polar_vectors.jl, the first test that checks a polar-data wing's solution against its input polar. Other tests already solve such wings (test_verification.jl, test_moment_units.jl, test_wing_directions.jl), but they compare against reference data or frame conventions, not panel by panel against the polar each panel was given.

The test

A flat 3-panel wing, 1 m × 1 m panels at 5°, built from four sections so refine! keeps them one-to-one. Each panel's polar is the average of its two edge sections (build_interps in src/panel.jl), so the design lets that averaging show.

• Each panel solves on its own polar. The +y sections carry cl = 4π·alpha, cd = 0.1, cm = -0.02, the -y sections 2π·alpha, 0.05, -0.01. Solved cl_dist / alpha_dist must be [4π, 3π, 2π], cd_dist [0.1, 0.075, 0.05], cm_dist [-0.02, -0.015, -0.01], and the stronger +y side must roll the wing positive about x (moment[1] > 0, in its own named testset). Every solved alpha falls between table knots, so the interpolation is exercised, not just the knot values.
• A 2π·alpha polar is INVISCID plus profile drag. With cd = 0.05, cm = 0 on every section, circulation and lift match the INVISCID wing to 1e-12. Each panel's force differs from INVISCID by exactly its profile drag, |Δf_i| = drag_dist_i · width_i, and cd_dist is 0.05 on every panel. The symmetric wing has no rolling moment.

I compared per panel on purpose: the norm of the total force difference misses the sum of the panel drags by 6e-6 relative (20.635145 vs 20.635268), because each panel's drag lies along its own induced flow direction.

@ufechner7's suggestion on the thread (3 panels, outer 2π·alpha, cd = 0.05, cm = 0, one outer twice the other) is in there in spirit. A middle panel with a different model from the outer ones cannot be built, since both edge sections of a panel must share a model.

Also fixed

test/bench.jl's gamma-loop allocation test built its POLAR_VECTORS table from alpha = -20:30, in degrees, where radians are expected. It now uses deg2rad. Allocation counts are unchanged (6 allocs, 208 B for both models).

Verification

• Reproduced first: n/a, new test on existing behaviour. To show it catches a defect, I made build_interps take cl from the first section only. The test failed at sol.cl_dist ≈ [4π, 3π, 2π] .* sol.alpha_dist (middle panel 0.4967 vs 0.3725), and passed again (10/10) once src/panel.jl was restored. I did this on the rebased tree.
• test/solver/test_polar_vectors.jl: 10/10 pass on ccd2cd1 after the review round. test/bench.jl: 21/21 pass. Both run in juliaserver, Julia 1.13.0, test environment.
• Rebased on main @ 1b72901. Uses Solver(n_panels, n_unrefined_sections), since Solver(body_aero) is now deprecated, with no warnings.
• Local full suite (agent ci-local, Julia 1.13.0, one matrix cell) on ccd2cd1: PASS in 10 min, exit 0 · GitHub CI on ccd2cd1: PASS (Julia 1.12 ubuntu/macOS/windows, 1.13 ubuntu, docs, setup test)
• Merges cleanly into current main, so not re-synced
• Docs/changelog: n/a, test-only · REUSE: the repo has no bin/reuse_lint.
• Risk: moment[1] > 0 relies on sections being sorted +y first. That holds now, and the cl/cd/cm assertions would fail too if it changed.

Scope

+62 / -1 across 3 files: the new 60-line test, one line registering it in test/runtests.jl, and the one-line unit fix in test/bench.jl. No src/ changes. Searched test/ for POLAR_VECTORS, POLAR_MATRICES, INVISCID and 2π for an existing polar-vs-polar or polar-vs-inviscid solve check. There was none; test/panel/test_panel.jl checks the averaging at panel level only. test_verification.jl has a polar_vectors(alpha_deg, cl, cd, cm) helper; linear_polar does not reuse it because the two share only the deg2rad, and reusing it would mean moving it into test_data_utils.jl and editing test_verification.jl, outside this diff.

Closes #29 · task VortexStepMethod.jl-29

[1-Bort-1]

Independent review (advisory)

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

Good

• Fills a real gap: I read test_verification, test_moment_units and test_wing_directions, which do solve polar wings but only against CFD, scaling or symmetry. None checks panel cl/cd/cm against the polar it was given.
• The 4-section, 3-panel layout makes the middle panel's averaged polar show up as [4π, 3π, 2π] and [0.1, 0.075, 0.05]. The card's mutation test (cl taken from the first section only) shows the assertion catches a broken build_interps.
• Comparing each panel's force against INVISCID is the right precision. The card gives the numbers showing a total-force comparison would be off by 6e-6.
• The bench.jl degrees-to-radians fix is correct: with alphas in degrees the table ran cl up to 2π·30. The card names it under 'Also fixed' and says allocation counts did not change.
• The helpers have docstrings, the testset names describe the behaviour they protect, the new file is registered in runtests.jl, and I read no line over 92 characters.
• The tolerances are strict (rtol 1e-12 on gamma and lift, 1e-10 solver rtol), so nothing was loosened to make it pass.

Not good

• test/solver/test_polar_vectors.jl:53 — drag_dist and lift_dist are each q·c times cd and cl, so drag/lift ≈ cd/cl holds by construction and adds nothing the cd_dist check does not already prove. Checking sol_polar.cd_dist ≈ fill(0.05, 3) would state the same thing directly, or the line could be dropped.
• The card's TL;DR says this is 'the first test that solves a wing built from polar data'. test_verification.jl:113, test_moment_units.jl:16 and test_wing_directions.jl:17 already do. The accurate claim is 'the first to check the solution against the input polar'.
• test_verification.jl:29 already has a polar_vectors helper that builds POLAR_VECTORS tuples from degrees. The card's search list does not mention it, and linear_polar could have reused it or said why it doesn't.
• moment[1] > 0 depends on the +y-first section order and on the sign convention. The card calls this out as a risk; a short testset name like 'stronger +y side rolls positive' would make the intent clear without the card.
• solve_three_panel_wing(INVISCID, nothing, nothing) passes two arguments that only exist to be ignored. A default polar_neg_y=polar_pos_y would shorten both of the second testset's calls, but this is a minor point.

_{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: drag_dist and lift_dist are each q·c times cd and cl, so drag/lift ≈ cd/cl holds by construction and adds nothing the cd_dist check does not already prove. Checking sol_polar.cd_dist ≈ fill(0.05, 3) would state the same thing directly, or the line could be dropped.

[1-Bort-1]

Agreed, it held by construction. Replaced with sol_polar.cd_dist ≈ fill(0.05, 3) in ccd2cd1.

[1-Bort-1]

Local full suite: PASS (6 min, Julia 1.13.0, one cell of the matrix)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!