Standalone Arrow CSV reader exposed as a WASI P2 component. Same
workaround-D pattern as
arrow-ipc-typed
and
arrow-compute-typed:
bytes cross the WIT boundary, the Rust side does the parsing +
type-inference + downcast, the result is serialized into per-type
list<option<T>> columns. No arrow:core resource sharing.
arrow:csv-typed/read:
read-csv(bytes, options) -> list<named-column>— single-batch read. Column types inferred from the first N lines peroptions.max-records-for-inference(0 = whole stream).- Supported column types:
bool,int64,float64,string. Other inferred numerics (i8/u32/f32 etc.) widen to one of the four; richer types (dates, decimals, etc.) follow when a real consumer asks.
bash scripts/build.sh
Produces build/arrow-csv-typed.component.wasm.
Apache-2.0