Follow-up split from #207 (whose scan chunk-planning fix is in fix/207-nested-struct). With nested struct columns now scanning and decoding to StructArray, CSV export of such a column throws its clear-but-terminal unsupported array type for CSV export: StructArray (CsvExporter.cellValue).
countries-of-the-world (Raincloud corpus, #205) is the pinning dataset: count/select region country_code work after the #207 fix; only exporting the data struct column aborts.
There is no existing precedent in cellValue for composite cells (no VariantArray arm either). A natural shape is a JSON-object cell ({"field":value,...}) — needs a deliberate format decision since CSV consumers must be able to parse it back. Until then the conformance matrix keeps countries-of-the-world as a gap pointing here.
Follow-up split from #207 (whose scan chunk-planning fix is in
fix/207-nested-struct). With nested struct columns now scanning and decoding toStructArray, CSV export of such a column throws its clear-but-terminalunsupported array type for CSV export: StructArray(CsvExporter.cellValue).countries-of-the-world(Raincloud corpus, #205) is the pinning dataset:count/select region country_codework after the #207 fix; only exporting thedatastruct column aborts.There is no existing precedent in
cellValuefor composite cells (no VariantArray arm either). A natural shape is a JSON-object cell ({"field":value,...}) — needs a deliberate format decision since CSV consumers must be able to parse it back. Until then the conformance matrix keepscountries-of-the-worldas a gap pointing here.