mumbleskates
commented
3 months ago incidentally, most (all? probably, all) of my contributions to Prost were discovered/developed while forking and almost completely rewriting it. Downstream of here i have changed the nature of the encoding slightly (not dramatically), but i have also achieved huge strides in usability:
- fully trait-based dispatch for encoding and decoding, with derive attributes simple enough to use on existing user types and support for generics and lifetimes on message types
- ignored fields
- useful functionality on the object-safe traits
- oneofs that may bear their own empty variant rather than needing to nest within
Option - oneofs in messages must have known matching tags, checked at compile time
- greatly increased type support, including nesting values in ways that protobuf schemas cannot even express
- novel linear-time encoding via writing the output back-to-front, which can be dramatically faster without caching extra data inside the structs
- guaranteed field ordering (protobuf as an encoding has no real use for this)
- distinguished decoding of canonical data (protobuf can't do this as easily)
- all of this and it's actually faster
if any of these sound useful to upstream it could be a good project for myself or someone to trace back over my steps in achieving all some of these. some of these can probably be upstreamed in a day or two, others could take longer.
caspermeijn
Currently it seems that the "invalid oneof variant" Errs are always discarded because they are being passed directly into
flat_mapfrom a closure, rather than actually propagating the error.This PR also streamlines the checks for duplicated tag numbers and adds some rudimentary tests for the affected code in those proc macros.