Hey, Sorry for the delay - now that the enum values are power-of-two I think this is fine for any hypothetical encoding for Flight. In particular gRPC allows binary headers so if we wanted to directly send a dummy schema that would be fine, or we could encode it as a bitfield. (There is some limit on header size but it is fairly generous and this shouldn't take up much data.)
Thanks, David On 6/25/20, Antoine Pitrou <anto...@python.org> wrote: > > I would be mostly interested in feedback by David Li and other Flight > developers, otherwise it's fine to me. > > Regards > > Antoine. > > > Le 25/06/2020 à 05:12, Micah Kornfield a écrit : >> I've updated the PR. More feedback welcome, I'd like to start a vote by >> end-of-week if possible. >> >> On Wed, Jun 24, 2020 at 12:48 PM Micah Kornfield <emkornfi...@gmail.com> >> wrote: >> >>> I agree flight might need to encode this data slightly differently for >>> negotiation purposes. I will update the enum to use power of 2 values >>> so >>> this isn't precluded, but I think for parsing in the schema, it is >>> clearer >>> to model this as a list of enums. >>> >>> Any other thoughts? >>> >>> Thanks, >>> Micah >>> >>> >>> >>> On Tue, Jun 23, 2020 at 7:58 AM Wes McKinney <wesmck...@gmail.com> >>> wrote: >>> >>>> I'm in favor of the list of enums myself -- it seems like it will be >>>> easier to work with and less error prone in general. Storage space >>>> with this should not be an issue. >>>> >>>> On Tue, Jun 23, 2020 at 7:31 AM Antoine Pitrou <anto...@python.org> >>>> wrote: >>>>> >>>>> >>>>> >>>>> Le 23/06/2020 à 06:02, Micah Kornfield a écrit : >>>>>>> >>>>>> A bit-field could work, but it is a little easier to mess-up (e.g. >>>>>> endianness). >>>>> >>>>> I don't think endianness can be an issue. A bitfield would have the >>>>> same representation as any "long" Flatbuffers field. >>>>> >>>>>> I would lean towards a list of enums because I think it is >>>>>> harder to mess up, and we don't need to guess at the number of values >>>>>> necessary, but that is subjective, and I could go either way on this. >>>>> >>>>> The main concern is "make it easy to query features". For example, if >>>>> Flight wants to encode the features as a gRPC header, it's easier to >>>>> do >>>>> if it's a single integer value (though the value is opaque). >>>>> >>>>> Regards >>>>> >>>>> Antoine. >>>> >>> >> >
