[huxiaoyi-ovo]

Reference issue

Fixes #1318.

What does this implement/fix?

DijkstraPlanner.calc_index() previously subtracted the map's world-coordinate origin (min_x, min_y, in metres) from Node.x and Node.y, even though these node coordinates are already zero-based grid indices (in cells). This mixes units and produces an incorrect nonzero index for the grid-origin cell whenever the world origin is offset.

Use the direct row-major index node.y * self.x_width + node.x. The fix corrects the index definition. It can also change an incorrect result: a valid start cell previously assigned key -1 could collide with the parent-chain sentinel and be omitted from the returned path. The search algorithm and intended path ordering remain unchanged.

A focused regression test checks two maps with different nonzero world origins and non-unit resolutions (0.5 and 2.0 m/cell), including:

• mapping the world origin to grid cell (0, 0) and index 0;
• mapping an interior cell to y * width + x;
• mapping the final grid cell to width * height - 1.

A second end-to-end regression follows Afloat16's reproduction and asserts the complete reversed path [(1.0, 3.0), (1.0, 2.0)] instead of the old goal-only result.

This also addresses the maintainer's earlier request for an offset-origin, non-unit-resolution regression test on #1360 (closed without merge).

Additional information

The original two-commit head was validated against upstream master (cdd0cc888802b584c2d654ca85e3c5460973487d) using fork-only draft PR #2:

• Linux Python 3.14 full CI — passed on the original head.
• Windows Python 3.14 full CI — passed on the original head.
• macOS Python 3.14 full CI — passed on the original head.
• CodeQL code scanning — passed on the original head.

The end-to-end test was added in commit 9e96bf4; the latest CI status should be checked separately.

• Diff: only PathPlanning/Dijkstra/dijkstra.py and tests/test_dijkstra.py, with no new dependencies.

CheckList

• Did you add an unittest for your new example or defect fix?
• Did you add documents for your new example? (Not applicable: corrects an internal grid-index formula; no public API or algorithm documentation changes.)
• All CIs are green? (Fork CI: Linux, Windows, macOS, CodeQL; upstream checks will run separately.)

AI assistance was used to review the issue history and prepare the minimal fix and regression test; the final branch was checked against upstream, and the full fork CI passed.

[Afloat16]

Automated Codex-assisted supporting check; no human review or approval is claimed.

The index normalization also fixes a path-reconstruction failure, so the description's claim that planned paths are unchanged has an exception:

from PathPlanning.Dijkstra import dijkstra

dijkstra.show_animation = False
planner = dijkstra.DijkstraPlanner(
    [1.0, 5.0, 1.0, 5.0], [1.0, 1.0, 5.0, 5.0], 1.0, 0.0
)
rx, ry = planner.planning(1.0, 2.0, 1.0, 3.0)
print(list(zip(rx, ry)))

With Python 3.13.5 and the repository-pinned NumPy 2.5.3 / Matplotlib 3.11.2, both endpoints pass verify_node:

• Upstream 8c392897: [(1.0, 3.0)] (the start is missing).
• This PR's 452164c4: [(1.0, 3.0), (1.0, 2.0)], the complete path in the planner's reverse order.

The old formula assigns start grid cell (0, 1) the key -1, which collides with the parent-chain terminator. The row-major formula gives it key 4.

This would be a useful end-to-end regression alongside the key-value tests. Only this matched case was run against the complete, unmodified module at each immutable revision, not the full suite or CI. Matplotlib emitted Fontconfig cache diagnostics; both runs completed normally.

[huxiaoyi-ovo]

Thanks @Afloat16, good catch. The start cell really can end up with key -1, so this affects path reconstruction, not just the internal index values. I added your example as an end-to-end regression in 9e96bf4 and corrected the PR description. The new CI run is underway.