[mmporong]

Initialize the temporary FOV map buffer to zero so an out-of-bounds point of view cannot return uninitialized visibility flags.

compute_fov() currently warns about an out-of-bounds point of view and continues. libtcod then returns an error without filling the FOV cells. Since the wrapper allocates those cells with np.empty, the result can contain stale visibility data. Zero-initialization preserves the existing warning behavior and returns an all-false mask in this case.

The regression tests cover all four map boundaries with a poisoned allocation so the failure is deterministic. All four fail before the fix. After the fix, these tests and the existing map tests pass (9 passed). Ruff reports no issues on the changed source and test files.

Validation used the tcod 21.2.2 wheel with only tcod/map.py replaced by the updated file. Before the change, that file was byte-for-byte identical to the current source. A full source build and rendering tests were not run.

[HexDecimal]

The results of compute_fov with an out-of-bounds pov are both undocumented and deprecated. If necessary, the recommend changes are informed by the deprecation warning itself which is to convert its warning into an error. Defining the previously undefined behavior in the way this pull request does ironically adds technical debt to the codebase, but that can be acceptable if the change is more intentional and the documentation is updated to match.

The lack of context awareness around this PR implies a poor use of automation tools. Ideally these would be used to help code match the documentation or fix major bugs, this is neither case. Raising an issue may have been preferable to a pull request.

Non-English commit messages are tricky to review. Most Git tools do not have auto-translations. I'm not quite sure what to do about this yet.