[lKolabrodl]

Closes #267.

Summary

Add table-driven rejection coverage for every validation guard in the EgoSuite projection geometry:

• zero and non-finite quaternion norms
• zero image width and height boundaries
• zero horizontal and vertical focal lengths
• non-finite focal lengths and principal points
• an empty joint sequence
• non-finite x, y, and z joint coordinates

This is test-only and does not change production behavior.

Why

The existing tests covered valid projections but did not prove that malformed camera or joint data is rejected. A future regression could therefore remove any of the guards without failing the suite.

Verification

• uv run --project examples/egosuite_evaluation ruff check examples/egosuite_evaluation
• uv run --project examples/egosuite_evaluation ruff format --check examples/egosuite_evaluation
• uv run --project examples/egosuite_evaluation ty check examples/egosuite_evaluation
• uv run --project examples/egosuite_evaluation pytest -q examples/egosuite_evaluation/tests (21 passed)
• Mutation-checked all six guards independently: disabling each guard made its corresponding new cases fail while the other tests stayed green.

The native Windows test run used a local no-op fcntl import shim solely for test collection, consistent with the project's documented Windows limitation. These geometry tests do not exercise file locking or storage.

AI assistance

AI assistance was used to implement and validate this change. The final diff and test results were reviewed before submission.

[github-actions]

👋 Hi @lKolabrodl — thank you so much for your first contribution to HFlow!

A maintainer will review your pull request as soon as possible. In the meantime:

• ⭐ Star the repo: https://github.com/Hebbian-Robotics/hflow
• 💬 Join our Discord: https://discord.gg/vacepQvjmg
• 📖 Check CONTRIBUTING.md for the validation commands CI will run
• 📣 Share the project with your network

💡 Tip: one open pull request per contributor at a time. Issues with an assignee are taken; everything else is fair game.

We are excited to have you here and appreciate your help making the project better! 🙌

[kstonekuan]

LGTM, merging. Welcome.

Reran the mutation experiment independently. Every guard has exactly the coverage it should, and no test fires for a guard it does not own:

guard removed             tests that went red
quaternion norm           zero-quaternion-norm, nonfinite-quaternion-norm
width/height positive     zero-width, zero-height
focal lengths positive    zero-horizontal-focal-length, zero-vertical-focal-length
calibration finite        all four nonfinite-focal / nonfinite-principal-point cases
nonempty joints           empty-joint-sequence
joint coords finite        nonfinite-joint-x, nonfinite-joint-y, nonfinite-joint-z

On DoD 4, the boundary answer is stronger than the PR body claims. Zero is the only value that pins these three guards. Weakening them instead of deleting them:

width/height:     <= 0   becomes < 0     -> zero-width, zero-height fail
focal lengths:    <= 0.0 becomes < 0.0   -> both zero-focal cases fail
quaternion norm:  <= 0.0 becomes < 0.0   -> zero-quaternion-norm fails

A width=-1 fixture would have caught none of those. An off-by-one in the comparison is the realistic regression here, so zero is the correct side, not just the convenient one.

Two more choices that are easy to get wrong. math.nan for the finiteness cases rather than -math.inf: the positivity check runs first and -inf <= 0.0 is True, so an infinite focal length lands on "must be positive" and never reaches the guard you meant to cover. And matching the message rather than the type. With the quaternion guard removed:

Quaternion(inf, 0, 0, 1)  ->  ValueError: hand joint coordinates must be finite
Quaternion(0, 0, 0, 0)    ->  ZeroDivisionError: division by zero

The first one still raises ValueError, just for a different reason two guards later, so a bare pytest.raises(ValueError) would have passed on the wrong refusal.

Gate clean on the example project: ruff check, ruff format --check, ty check, 21 passed.

The fcntl shim you mentioned is correctly not in the diff.