On Mon, Aug 1, 2016 at 10:26 AM, Wes McKinney <wesmck...@gmail.com> wrote: > hi Sanjay, > > You may have seen: > > https://github.com/apache/arrow/blob/master/format/Layout.md#byte-order-endianness > > For big-endian systems, they are free to work with integers with the > native byte order, but for the purposes of any memory sharing / IPC / > RPC of Arrow memory layout, my understanding is our intention is for > big-endian to be byte swapped. Adding byte order / endianness to the > Arrow metadata for IPC messaging would add complexity for relatively > little benefit. If a big-endian system is using Arrow internally and > never needs to interact with another system using Arrow, then the need > to byte swap would not be an issue. > > Does this make sense? It's not a question of whether Arrow "supports" > big endian platforms but rather what's considered "native" byte > ordering for the purposes of relocating the data structures. > > If anyone has additional thoughts or if I'm misrepresenting the intent > of the spec documents please chime in. > > Thanks > Wes >
I created ARROW-245 to add some clarifying language around this point. Thanks
