cleanup: apply_heading and the two frames the placement fix stopped us

[1-Bort-1]

TL;DR

510a4232 replaced tether-frame placement with min_rotation and made apply_heading! read the body's own orientation, and neither change took its scaffolding out. This drops the now-dead apply_heading helper and the two rotation matrices apply_azimuth_elevation! still built, returned and handed to a callee that never looked at them.

What was left behind

apply_heading(vec, R_t_to_w, curr_R_t_to_w, heading) had no callers. grep -rn apply_heading over the repo found exactly three hits: the definition, the file's header list, and the @docs entry on docs/src/private_functions.md.

apply_azimuth_elevation! computed curr_R_t_to_w = calc_R_t_to_w(rel_pos) and R_t_to_w = calc_R_t_to_w(transform_pos) and returned the pair; reinit! and reposition! each destructured it and threaded both into apply_heading!. The body of apply_heading! reads neither — since 510a4232 it takes its orientation from heading_reference_body and rotates about normalize(reference_body.pos_w - base_pos). So the two calc_R_t_to_w calls existed only to fill a return value nobody consumed, and the docstring line "Returns (curr_R_t_to_w, R_t_to_w) for use in that step" described a contract that had stopped being one.

What is gone

The helper and its docs entry; the two calc_R_t_to_w calls; the return tuple and that docstring line; the two parameters on apply_heading! and the destructuring at both call sites. Nothing else changes: calc_R_t_to_w stays live through calc_heading (and test_heading_calculation.jl exercises it directly), transform_pos is still what min_rotation rotates onto, and rotate_around_z is still used a few lines below. The file header's function list now names apply_heading!, the one that exists.

No CHANGELOG.md entry: nothing observable changes. apply_heading is private and had no callers, and the deleted return values had no readers.

The red job is not this diff, and I did not touch it

Julia 1.12 - ubuntu-latest - x64 - monolith errored in test/test_twist_alignment.jl:123 with AssertionError: VSM solve failed (non-converged or non-finite) on wing 1 — 3143 passed, 0 failed, 1 errored. The other five jobs of that run, Julia 1.11 - ubuntu, Julia 1.12 - macOS, the kernel-backend cell on the very same ubuntu 1.12 image, Documentation and reuse-lint, are green, and so was the full suite on the box.

The deletion is inert on that test, measured rather than argued: I ran test/test_twist_alignment.jl here, then put the parent commit's src/system_structure/transforms.jl back with git checkout c3b2b562^ -- … and ran it again in the same session. Static twists [-0.1336630477312238, -0.0008835202645692225, -0.1319843929300836] and dynamic [0.9221199567195998, -0.0033309221603231233, 0.7965616094657846], identical to the last bit on both. It has to be: the two deleted calc_R_t_to_w calls are pure (smooth_normalize and a cross product, src/generate_system/helpers.jl:210) and nothing read their results.

What the test does is ramp steering for 60 steps and assert on the state it lands in, which its own comment says is next to a divergence: "Holding this steering runs the station twist away until the VSM solve diverges, so a settled operating point does not exist for this 2-plate config." Same code, four platforms, end states from 0.22 to 1.05 and once past the assertion in vsm_aero_coeffs. It last went red the same way on feat/point-flap-delta (#288, 93ac34cd, ubuntu 1.12 monolith and no other cell), whose own later head and merge into main both passed. I opened #316 with the six-run table and what I would do about it. It is a behaviour test with a real fix behind it, so it does not ride inside a deletion PR, and weakening it to get this one green is the thing I must not do.

Where I would push back

The other obvious follow-up — a test pinning what apply_azimuth_elevation! and apply_heading! do near the vertical, where the old frame-based placement was undefined — is #312. Also a behaviour test, also not this diff.

Verification

• Reproduced first: n/a — dead-code removal. The claim it rests on is grep -rn apply_heading, which found the definition, the header line and the docs entry and no call.
• test/test_transform.jl: Transform Tests | 325 325 — all pass, twice (4m31s cold, 17.5s on the warm cache after the last edit). juliaserver, examples env.
• test/test_heading_calculation.jl: Heading - Circular Path | 72 72, Heading - Horizontal Circle | 48 48, Heading - Special Cases | 22 22. This is the file that keeps calc_R_t_to_w honest now that only calc_heading reaches it.
• test/test_twist_alignment.jl (the file CI errored in): green here, and bit-identical with the parent commit's transforms.jl swapped back in — numbers above.
• Local CI mirror (agent ci-local): PASS (70 min, Julia 1.12.7, one cell of the matrix) — weak evidence next to GitHub CI, but it is the cell that went red there.
• GitHub CI: six of the seven jobs pass — Documentation, reuse-lint, Julia 1.11 - ubuntu, Julia 1.12 - windows, Julia 1.12 - macOS and the kernel-backend cell on the same ubuntu 1.12 image. Julia 1.12 - ubuntu monolith is the one red, on test_twist_alignment.jl → test_twist_alignment asserts on a state reached by driving the VSM to the edge of convergence #316, not this diff. I re-ran that cell alone and its six siblings went green a second time; the cell itself was still running when I wrote this, 1h28m into a 180-minute budget, so this PR's own check badge is the live answer and I am not transcribing a guess for it. Either verdict leaves the reading unchanged — green makes it the flake test_twist_alignment asserts on a state reached by driving the VSM to the edge of convergence #316 describes, red makes it the second sighting on this commit, and neither can come from a diff that is bit-identical on this test.
• Docs build: not run here — docs/ carries no manifest and building it would resolve one, which AGENTS.md §2 keeps out of my hands. The @docs entry and the docstring it pointed at go in the same commit, so there is no orphan either way; the Documentation job is green.
• REUSE lint: not run here — reuse is not installed on this box. Neither touched file changes a licence header, and the reuse-lint job is green.
• Up to date with main: the branch was at origin/main when pushed, no rebase needed.
• Benchmark: n/a — two calc_R_t_to_w calls per transform at init only, not on any stepped path.
• Risk: none I can see; the removed values had no readers. If something outside this repo calls SymbolicAWEModels.apply_heading through the private API it breaks — it is not exported and not in the public docs.

Scope

+8 / -30 across 2 files: src/system_structure/transforms.jl and one line off docs/src/private_functions.md. No stack — no open PR touches transforms.jl. Found while answering #254; filed #316 on the way.

Closes #313 · task SymbolicAWEModels.jl-313

[1-Bort-1]

CI: Julia 1.12 - ubuntu-latest - x64 - monolith - pull_request failed on c3b2b562e.

https://github.com/OpenSourceAWE/SymbolicAWEModels.jl/actions/runs/34405402473/job/102647056275

[1-Bort-1]

Local full suite: PASS (29 min, Julia 1.12.7, one cell of the matrix)

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!