Make the KA-frame change load against the KiteUtils 0.13 that shipped

[1-Bort-1]

TL;DR

#50 was written before KiteUtils 0.13 was released and calls a two-argument quat2viewer that the release does not define, which stops KiteViewers precompiling at all; the caller now converts a KS attitude itself with fromKS2KA. On top of that, this branch moves to the released dependency stack: KiteUtils 0.13, KiteModels 0.12 (which writes KA orientations to SysState) and Julia 1.12/1.13. CI is green.

The call that is not there

KiteUtils v0.13.0 was released today with quat2viewer(attitude) and no second method — src/transformations.jl:140. A KS orientation is the caller's job to convert, which is what the release notes of OpenSourceAWE/KiteUtils.jl#130 say: quat2viewer(fromKS2KA(q)).

This is not a wrong-looking kite on an old log. update_system is called from the @compile_workload in src/KiteViewers.jl:139, so the method that does not exist is reached while the package is being precompiled, and using KiteViewers fails outright:

ERROR: LoadError: MethodError: no method matching
  quat2viewer(::KiteUtils.FrameQuat{KiteUtils.SysState{11,...}, Float32}, ::KiteUtils.FrameConvention)
Closest candidates are: quat2viewer(::Any)
  @ KiteUtils ~/.julia/packages/KiteUtils/xNJLj/src/transformations.jl:140
Stacktrace:
 [1] update_system(kv::KiteViewers.Viewer3D, state::KiteUtils.SysState{11,...}; scale, kite_scale, frame, wind)
 ...
 @ ~/worktrees/KiteViewers.jl-54/src/KiteViewers.jl:139
Failed to precompile KiteViewers [2e593061-95e7-45e4-95f4-df0491f2e601]

test/test_frames.jl pins what #50 promises and what this restores — that a KS attitude and the KA attitude it converts to are drawn the same way round. It reads back KiteViewers.quat[], the quaternion the kite mesh is actually rotated by, so it fails on the wrong conversion as well as on no conversion, and it does so over five attitudes rather than the demo state's one: the demo attitude, the identity, two with all four components at 0.5, and one more about a single axis. Each is written into state.orient as KA, drawn, then written again as its fromKA2KS image and drawn with frame=KS. Deleting the conversion in src/common.jl turns all five red; restoring it turns all five green. atol=1e-6 rather than the default relative tolerance, which components sitting at zero can never meet.

Moving to KiteModels 0.12 and Julia 1.12/1.13

• Project.toml, examples/Project.toml: KiteModels = "0.12", julia = "1.12, 1.13", Pkg = "1.12, 1.13". KiteUtils 0.13 and the WinchModels and AtmosphericModels releases that accept it require Julia 1.12.
• Manifest-v1.11.toml.default removed; Manifest-v1.12.toml.default updated by a minimal resolve: KiteModels 0.12.0, KiteUtils 0.13.1, WinchModels 0.3.12, AtmosphericModels 0.3.12, KitePodModels 0.4.2 (plus two JLLs from Julia 1.12.7). All registered versions.
• bin/release no longer checks the 1.11 manifest; README and docs say Julia 1.12 or 1.13; CHANGELOG.md has two new Unreleased entries.
• Tests: test_frames.jl and test_parking.jl share one Viewer3D (created only if viewer is not defined yet), because the viewer state is module-global and a second viewer in one session is unsupported (review comment on test/runtests.jl).

Before #50 can merge

This PR targets frame-unification (#50). After it merges there, frame-unification needs main merged in:

• bin/release conflicts: main (Support Julia 1.12 and 1.13, drop Julia 1.11 #61) checks the 1.12 and 1.13 manifests, this branch only 1.12. Take main's version.
• CHANGELOG.md conflicts: both add an Unreleased section; combine them.
• Add a Manifest-v1.13.toml.default resolved for KiteModels 0.12 / KiteUtils 0.13 (main has one for the old stack).

Then KiteViewers can be released as v0.7.0 (breaking: update_system takes frame= instead of ned), after which OpenSourceAWE/KiteModels.jl can drop its [sources] lines that point at this branch.

Found on the way, not fixed here

• The KiteModels-driven examples (steering_4p.jl, depower_simple.jl, joystick.jl, reelout_*.jl, the *_bench_video.jl pair) and test_parking.jl/test_steering.jl all call update_system with the default frame. That is correct now that KiteModels.jl#312 is released (KiteModels 0.12 writes KA), and these examples need KiteModels 0.12 anyway, so there is nothing to write differently.
• .github/workflows/CI.yml carries fail-fast: false. One matrix cell today, so it costs nothing yet.
• No CHANGELOG.md entry for the quat2viewer fix: Take the orientation convention as a keyword, not ned #50's Unreleased entry already describes the behaviour this makes true. The entries this branch adds are for the KiteModels 0.12 and Julia 1.12/1.13 requirements.
• bin/install was rewritten in cleanup: bin/install takes -y, -u and -h and touches nothing outside the repo #55 and moved to Julia 1.12 and 1.13 in Support Julia 1.12 and 1.13, drop Julia 1.11 #61, both merged into main; this branch is stacked on Take the orientation convention as a keyword, not ned #50 and picks them up once frame-unification is updated from main (see below).

Verification

• Reproduced first: Failed to precompile KiteViewers, MethodError: no method matching quat2viewer(::FrameQuat, ::FrameConvention) — full output above
• test/test_frames.jl red before, green after, in a KiteUtils 0.13.0 environment (5 passed; 0 passed, 5 failed with the conversion removed)
• Local full suite on 7113008 with the registered packages (KiteModels 0.12.0, KiteUtils 0.13.1): test_frames 5/5, KiteViewers.jl 3/3; re-run from the tracked manifest with no local overrides (Pkg.resolve() moved nothing; Julia 1.12.7, examples env, exit 0) and as Pkg.test() on Julia 1.13.0 (PASS, exit 0)
• GitHub CI green on 7048901 (first green run of this branch)
• Docs build: not run; no public symbol is added or renamed, and update_system is already on docs/src/reference.md
• Risk: test_parking.jl drives the viewer from a real KPS4 run writing KA, but asserts position and force, not the drawn orientation; that the kite is drawn the right way round is proven only for the five hand-written attitudes in test_frames.jl.

Scope

+58 / −2017 across 12 files against frame-unification; −1984 of that is Manifest-v1.11.toml.default. The code change is src/common.jl (two lines of conversion and the deletion of a commented-out three-line-kite block); test/test_frames.jl is new.

Opened elsewhere

• Pull request: Build SysState with the KiteUtils 0.12 constructor and one-slot winch fields KiteModels.jl#321
• Pull request: Accept KiteUtils 0.13, support Julia 1.12/1.13 only, test every line of src/ aenarete/WinchModels.jl#26

Opened by 1-Bort-1, an AI agent working for @1-Bart-1.
Closes #54 · task KiteViewers.jl-54

[1-Bort-1]

Independent review (advisory)

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

Good

• The two-line fix matches the released KiteUtils 0.13 API — src/common.jl:373-374 now converts with fromKS2KA and calls the single-argument quat2viewer, which is what the precompile workload (src/KiteViewers.jl:139, default frame=KA) hits, so using KiteViewers can precompile again.
• The new test pins the actual rendered value: it reads KiteViewers.quat[], the quaternion the kite mesh is rotated by, not an intermediate, so it goes red on a missing conversion.
• Scope is exactly the stated idea (+17/-9, four files); the only extra is deleting a commented-out three-line-kite block inside the same function, which §6 asks for and the card names.
• examples/Project.toml following the root project to KiteUtils = "0.13" is consistent — leaving it at 0.11.13, 0.12 inside one workspace would be unresolvable either way.
• No CHANGELOG entry is the right call and is argued: #50's Unreleased entry already states the behaviour, and a line saying the previous commit did not precompile is about the branch, not the release.

Not good

• test/runtests.jl:8 — test_frames.jl creates a Viewer3D at top level and test_parking.jl then creates a second one in the same session, which CLAUDE.md calls unsupported — the two share quat, textnode and the rest of the module globals. The assertions survive only because test_frames reads its quaternion before test_parking starts; a later reorder or a second frame test would break silently.
• The test compares the KS path against the KA path; if the KA↔KS transform is an involution (as the card implies) it would also pass with fromKA2KS in src/common.jl:373 — numerically equivalent, but the test does not pin the name.
• frame == KS ? ... : state.orient treats every non-KS member of FrameConvention as KA silently; harmless if the enum has exactly two members, worth a thought if it ever grows.
• One attitude from demo_state is the whole coverage for frame=KS — flagged in the card, and acceptable while nothing else resolves.
• Neither the new test nor the suite can actually run until KiteModels ships its 0.13 support, so this lands unverified by CI; the card is explicit about it and the PR stays draft.

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

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[ufechner7]

@1-Bort-1 I updated the description, you overwrote it. Why?

[1-Bort-1]

Not on purpose: the bot rewrites the description from its own card at the end of every turn, and that card was older than your edit. Restored your description from the edit history (14:43:57) as the card, so the next write keeps it; only Verification gains this turn's runs.

[1-Bort-1]

Local full suite: PASS (1 min, Julia 1.13.0, one cell of the matrix)