I think I would probably remove it from the spec with a note and a pointer to the class that implements it.
Right now we don't have anyone that I'm aware of relying on this serialization format across engines so it isn't a format-level contract. Though we should note that Flink relies on the format to evolve in compatible ways across versions. I think that means that we don't make any guarantees about how it evolves and it can be safely removed since it is not a contract that we are committed to maintaining. Ryan On Wed, Feb 21, 2024 at 1:30 PM Jack Ye <yezhao...@gmail.com> wrote: > Was there any prior discussions on devlist for adding it to the spec? > Could you help link those conversations? > > Thanks, > Jack Ye > > On Wed, Feb 21, 2024 at 1:05 PM Steven Wu <stevenz...@gmail.com> wrote: > >> >> In the recent PR review [1], Ryan and emkornfield has raised a question >> why file scan task JSON serialization was added to the table spec [2]. We >> seems to have a consensus that it *shouldn't* have been added to the >> table spec. >> >> Now the question is what's the process of removing an invalid section >> from table spec? Can we simply remove it? >> >> Thanks, >> Steven >> >> [1] https://github.com/apache/iceberg/pull/9728#discussion_r1492846630 >> [2] https://github.com/apache/iceberg/pull/6934/files >> > -- Ryan Blue Tabular
