fix(dc_measurements): retract the QR in-plane rotation limitation

[Minipada]

What this is

#297 asked three questions in order: does a newer ZXing-C++ fix the reported in-plane rotation failure, does it matter in practice, and if so is a bounded rotation-retry worth its per-frame cost.

The answers are no, no, and moot — the reported limitation does not exist. #123 and the camera.md warning it left behind were both measuring the harness, not the library.

What was actually happening

The benchmark rotated the demo texture inside its own 290x365 bounds. qrcode_0001.png's module area is 210 px at (45, 100) — the printed label takes the rest of the height, so the code sits above the centre the rotation turns about. Swing that and a corner of the module area, finder pattern and all, leaves the frame.

Forward-mapping the four module corners through the same getRotationMatrix2D the harness used says exactly when:

angle	module corners outside the frame	decodes
0°	0/4	✅
15°	0/4	✅
30°	1/4	❌
45°	1/4	❌
60°	1/4	❌
75°	1/4	❌
90°	0/4	✅

That is, to the angle, the reported pass/fail pattern. A QR the frame has cut a finder pattern off is unreadable at any angle by any decoder. The causes #297 ruled out — interpolation blur, module size, every Binarizer, tryDownscale — were all ruled out correctly; the one variable nobody varied was the frame.

Confirmed against the production library, not a binding

A standalone C++ program calling ZXing::ReadBarcodes the way camera.cpp does, compiled inside localhost/dc-workspace:latest against the exact libzxing-dev 2.2.1-3 that rosdep resolves on Noble/Jazzy:

clipped (rotated in the texture's own bounds)  0:ok 15:ok 30:MISS 45:MISS 60:MISS 75:MISS 90:ok
fully framed (padded to the diagonal)          0:ok 15:ok 30:ok   45:ok   60:ok   75:ok   90:ok

The PyPI bindings at 2.3.0 and 3.1.1 give identical results in both columns. So there is nothing for a newer release to fix — no vendoring, no rosdep pin — and the rotation-retry mitigation would not have worked anyway: re-rotating an image whose corner was already cropped cannot put the pixels back.

What changed

• dc_measurements/test/test_barcode_rotation.cpp (new) — two tests over the real reader, with no dc_simulation asset dependency; it reconstructs the texture's layout to the pixel with ZXing's own writer, the way test_code_pose already does.
  • DecodesACodeAtEveryInPlaneRotation — fully framed, decodes at every 5° from 0° to 90°. This is the assertion the retracted claim would fail.
  • LosesACodeTheFrameClips — rotated inside the texture's own bounds, asserts decode succeeds iff all four module corners stay inside the frame, so the test pins the cause and not just the symptom.
• doc/src/dc/measurements/camera.md — the admonish block goes from a warning describing a library defect to an info describing the framing requirement that actually matters: keep the whole code plus its quiet zone in the image, and in-plane angle is a non-issue. It also notes rotation_angle is unrelated, since whole quarter turns never crop. The perspective/tilt half of the Move QR/barcode detection to a more reliable C++ library (ZXing-C++) #123 comparison (100% vs. 0% for zbar at 35% skew) was independently measured and stands unchanged.
• No change to camera.cpp — nothing to mitigate, so no new parameter and no per-frame cost.

Verification

• colcon build --packages-up-to dc_measurements in localhost/dc-workspace:latest with this worktree bind-mounted: clean.
• colcon test --packages-select dc_measurements --ctest-args -R "test_barcode_rotation |test_code_pose" → 10 tests, 0 errors, 0 failures, 0 skipped (8 gtest cases across the two binaries plus their 2 ctest entries), against libzxing 2.2.1.
• prek run --all-files --skip build-doc → all 24 hooks pass, including clang-format and the REUSE licence check on the new file.

Closes #297

🤖 Generated with Claude Code

https://claude.ai/code/session_01L6PGKrx3v8rc8E1b6yrNpd

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.20%. Comparing base (f4cad58) to head (f75c6d8).

Additional details and impacted files

@@           Coverage Diff           @@
##            jazzy     #359   +/-   ##
=======================================
  Coverage   67.20%   67.20%           
=======================================
  Files          96       96           
  Lines        5981     5981           
=======================================
  Hits         4019     4019           
  Misses       1962     1962           

Flag	Coverage Δ
cpp-jazzy	67.20% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.