[ivorbosloper]

The JP converter now reads the Fude Polygons from MAFF's download service instead of Pacific Spatial's converted copy, which stopped at 2024. This adds the 2025 and 2026 editions.

• MAFF publishes one zip of municipality GeoJSONs per prefecture. The download URLs only work when signed, and a signature lasts 30 minutes, so the converter asks the API to sign the files just before it downloads them; only MAFF's own URLs are signed.
• MAFF's CDN only answers requests that look like a browser, so the converter sets request_headers (request_headers for downloads; DuckDB converter downloads archives also without a cache vecorel/cli#63). It refuses data-centre addresses outright (403); converting from such a machine needs the zips placed in the cache first.
• Each prefecture zip becomes one GeoParquet in the cache (extracted.<zip>/) before the usual DuckDB conversion, which reads only parquet. Without a cache, vecorel downloads the zips to a temporary folder (request_headers for downloads; DuckDB converter downloads archives also without a cache vecorel/cli#63), so the converter no longer overrides convert.
• The GeoJSONs declare JGD2011 per geometry, which GDAL ignores, so the CRS is set explicitly.
• The attribution follows MAFF's terms (政府標準利用規約 2.0): each edition carries the wording with its year, and the collection carries it without the year.
• The output columns are the same as before. Where they overlap with the old source, the ids, land types and municipality codes are identical, and the geometries match to floating-point precision.

Updates vecorel-cli to 0.4.1, which brings vecorel/cli#62 and #63.

🤖 Generated with Claude Code

[ivorbosloper]

Can only be merged after a new vecorel release....

[m-mohr]

vecorel-cli v0.4.1 has been released

[m-mohr]

CI is failing...

[Copilot]

🔵 Needs a closer look

A cached-input regression remains unresolved, and compatibility across all six source editions still needs validation.

No open issues 🎉

1 previously missed issue

In code that hasn't changed since last review

Preserve existing Parquet inputs when caching is enabled

fiboa_cli/​datasets/​jp.py:111

An existing .parquet or .geoparquet supplied through input_files is read directly by DuckDBBaseConverter.convert without a cache, but adding cache now routes the same file here and raises BadZipFile. Pass existing Parquet inputs through unchanged before opening the archive, so enabling caching does not break an otherwise supported input format.

2 resolved since last review

• Atomically replace cache after successful Parquet write
• Sign each ZIP immediately before downloading

_{Review effort: Balanced}