On Mon, Sep 14, 2026 at 4:50 PM Ryan Blue <[email protected]> wrote:

> > I support "option 0," which I'll place at the top of the list as:
> > 'Changes in Parquet files are versioned in the current way.'
>
> I think that the "current way" is that we don't make breaking changes. I
> voted against changing `path_in_schema` to optional because I think it
> breaks our guarantees. I think it's important for us to be able to agree on
> how to make breaking changes.
>

Well, you don't make breaking changes :) But enough people seem not to care
that the horse is out of the barn and "what a reader should do" seems
beside the point. How can you enforce such a thing? A reader can do
whatever it wants and it may do so at the convenience of the author or
users.  If you want to make suggested behavior, that seems fine and perhaps
just stick it as a comment in the thrift file for places where there is an
exception on an individual basis -- I don't think you need to formulate
blanket policy.

The thrift idl already has oddities like ColumnChunk::meta_data that is
defined as "optional", but comes with a comment that says "As such, writers
MUST populate this field".

What are you going to do?

-- 
Andrew Bell
[email protected]

Reply via email to