[brawer]

Requested by Sascha · project thread

Follow-up to #928, part of #924. OsmFeatureIndex now keeps the country of each feature's centroid, so later steps needn't decode geometry to find it.

Format. One byte per feature: 0 for none, else 1 + the country's position in a table of two-letter codes. The table is stored in the same file (2 bytes per code), not taken from the running binary, so an index stays readable after the boundary data changes; a code the binary doesn't know reads as no country. The feature storage signature becomes featstg1. OsmFeatureIndex::exists now checks it, so an index in the older format counts as missing and gets built again. If the planet file was deleted, fetch_inputs downloads it again for that.

country-boundaries 1.2.0 knows exactly 255 two-letter codes, which just fits into a byte with 0 for none. countries::tests::test_all fails if a newer version has more; we'd then need two bytes per feature.

Cost. The lookup runs in pass 1 of OsmFeatureIndex::create, from the centroid cell that's already the sort key. In a release build, a lookup takes about 0.16 µs (2 million points over Europe in 0.32 s), so about 30 seconds for the 163 million features of the planet. The file grows by one byte per feature, about 160 MB.

Users.

• collect_wikidata_ids (feat(wikidata): Count OSM features and names per brand #928) takes the country for its per-country brand totals from the index, and no longer decodes geometry.
• Deletion candidates: deletion_reason now takes the country as an argument. A branded unmatched OSM feature gets its geometry decoded only once it turns out to be a candidate; before, every one was decoded to find its country.
• Phone matching doesn't change: it looks up the country at the ATP position (for conflation) or from conflated.parquet (for edits), not from the index.

Testing. New tests: countries stored and read back (Zürich, the Atlantic, no centroid, Taipei), an index in the older format counting as missing, and the code table (sorted, fits a byte, covers every country on a 1° grid). Tests that stood in an empty file for an index now write one with the current signature. cargo test --lib, cargo clippy --all-targets -- -D warnings and cargo fmt pass. CI runs the integration tests, since they need tippecanoe.

---

---

Generated by Claude Code