Fixes for PR #290

[claude]

This PR contains fixes for the fork PR #290 by @Ch-Abhinav-Chowdary — it carries that PR's commit plus the requested review change.

Original branch: fix/issue-262-color-inversion-expected-preview

Change

Per @shxjames's review ("Remove unnecessary comments in challenge.py"), the 25-line explanatory comment block in reference_impl is replaced with the two concise comments the file originally had.

Net diff vs. main:

 # Reshape to (height, width, 4) for easier processing
 image_reshaped = image.view(height, width, 4)

 # Invert RGB channels (first 3 channels), keep alpha unchanged
-image_reshaped[:, :, :3] = 255 - image_reshaped[:, :, :3]
+inverted_rgb = 255 - image_reshaped[:, :, :3].clone()
+image_reshaped[:, :, :3] = inverted_rgb

Note on issue #262

The remaining .clone() is a no-op. 255 - image_reshaped[:, :, :3] is already an out-of-place op that materialises a fresh tensor before the assignment, so there is no aliasing window and no partially-written buffer for anything to observe. Verified byte-identical output on a 64×64 RGBA image:

outputs identical: True
correct RGB invert: True
alpha preserved: True

So this branch does not actually change reference behaviour, and issue #262 is likely still open — the misleading Expected preview most plausibly lives in the failure renderer (server-side, not in this repo) rather than in challenge.py. Note that in the reported output the head bytes of Expected and Got match and only the final pixel differs, which is not consistent with Expected holding a whole stale pre-inversion buffer; it looks more like a single failing pixel with the two labels swapped. Worth checking the renderer before closing #262.

Happy to drop the .clone() too if you'd prefer this branch be a pure no-op revert.

Verification

• pre-commit run --files challenges/easy/7_color_inversion/challenge.py — all hooks pass
• reference_impl re-checked locally on the example test and all 7 functional cases (RGB inverted, alpha preserved)
• python scripts/run_challenge.py challenges/easy/7_color_inversion --language cuda --action submit with a correct CUDA solution — ✓ All tests passed (functional + performance, T4). Solution file not committed.

🤖 Generated with Claude Code