Any plans to add narrower ints (Int8/Int16) and unsigned (UInt8/16/32/64)?

[Sober7135]

Hi! Quick question — the spec/API seems to have Int32/Int64 but not Int8/Int16 or any unsigned types. Are there plans to add Int8/Int16 and UInt8/UInt16/UInt32/UInt64?
Thanks!

[yangxk1]

Hi, @Sober7135, thanks for bring that up. The data types currently supported take into account the common support of parquet/orc, etc., and will consider supporting more data types in the future. Currently, unsupported types can be covered using other types (refer to https://parquet.apache.org/docs/file-format/types/)
Could you describe the use case or scenario you wan to use GraphAr which need those types?

[Sober7135]

Thanks for the quick reply! I’m integrating GraphAr with miniGU. miniGU’s type system includes Int8/Int16 and UInt8/UInt16/UInt32/UInt64. Direct support would enable lossless schema parity; upcasting to Int32/Int64 drops the unsigned/narrow semantics.

[yangxk1]

Thanks for your interest in GraphAr.
imo, GAR is based on Apache Arrow. Since Arrow already supports unsigned types and int8/int16, adding support for these types in GraphAr only requires updating types.cc.
Whether they are supported at the physical level depends on the chosen storage format (CSV/Parquet/ORC). For example, int8 and int16 can be stored in ORC (reference).

I noticed that miniGU is a rust repository,the arrow also has a rust version :https://arrow.apache.org/rust/arrow/index.html