refactor plan to consolidate (March 2026)

[mdsumner]

Post-cpp11 conversion: remaining refactoring

Now that vapour is on cpp11 (#xxx), there are some follow-up improvements worth doing. None are urgent — the package is clean and tested — but they'd reduce complexity for future work.

Split gdallibrary.h

At 784 lines it mixes vector layer utilities, CRS functions, driver listing, field allocation, and legacy read functions. Could split into:

• gdal_layer_utils.h — gdal_layer(), with_ogr_layer(), open_ogr_dataset(), force_layer_feature_count(), gdal_layer_geometry_name(), gdal_layer_extent()
• gdal_crs.h — gdal_proj_to_wkt(), gdal_crs_is_lonlat(), gdal_projection_info()
• gdal_drivers.h — gdal_list_drivers(), gdal_driver(), gdal_layer_names(), gdal_vsi_list(), gdal_version(), proj_version(), gdal_set_config_option(), gdal_get_config_option()
• gdal_field_alloc.h — allocate_fields_list()

Deduplicate field-reading switch logic

layer_read_fields_ia and layer_read_fields_fa in gdalgeometry.h have a cut-down field type switch (only OFTInteger/OFTReal/OFTString) while layer_read_fields_ij has the full switch including DateTime, Binary, etc. The ia/fa variants are marked FIXME copy the wider field support from sf. Extract the per-field-type write logic into a shared helper function.

Remove legacy gdal_read_fields / gdal_read_geometry

The gdal_read_fields() and gdal_read_geometry() functions in gdallibrary.h are the old limit_n/skip_n implementations. The gdalgeometry.h layer_read_fields_ij/layer_read_geom_ij functions (called via read_fields_gdal_cpp/read_geometry_gdal_cpp in 00_geometry.cpp) now handle this. Check if the old functions are still called anywhere; if not, remove them.

Add as_cstr() helper

The std::string(x[0]).c_str() pattern appears ~50 times across the headers. A small helper in common_vapour.h would clean this up:

inline const char* as_cstr(cpp11::r_string x) {
  return CHAR(static_cast<SEXP>(x));
}
// usage: GDALOpen(as_cstr(dsn[0]), GA_ReadOnly)