Hi,

As a short update, I just checked the PR for PARQUET-1419 and although in
its current form it is a breaking change, it can be easily rewritten to
become backwards-compatible so this part of the problem does not apply any
more.

Br,

Zoltan

On Tue, Sep 25, 2018 at 7:10 PM Zoltan Ivanfi <[email protected]> wrote:

> Hi,
>
> On the Parquet sync we discussed that the practice of maintaining a copy
> of parquet.thrift in parquet-cpp is dangerous and that we must take care to
> not release parquet-format changes in parquet-cpp before we officially
> release them in parquet-format. As I got back to my computer and started to
> create a JIRA about this, I noticed that unfortunately this has already
> happened.
>
> The encryption-releated parquet.thrift changes have not only been added to
> only parquet-format, but to parquet-cpp as well, and these changes got
> released in parquet-cpp 1.5.0. This is very unfortunate, because
> PARQUET-1419 would change the encryption in a breaking way, which is only
> acceptable as long as the original is not released. Additionally, it has
> been discussed that a formal voting should take place before incorporating
> the encryption features in the format.
>
> Now that parquet-cpp has already shipped these changes, we must choose the
> lesser evil of the following two options:
>
>    - Release a parquet-cpp 1.6.0 with a breaking change and risk that
>    encrypted data files already written with parquet-cpp 1.5.0 will not be
>    readable any more.
>    - Release the encryption in parquet-format as it is, regardless of
>    voting results and discard PARQUET-1419.
>
> Personally I have a hard time deciding which one I consider lesser evil.
> What are your opinions?
>
> Thanks,
>
> Zoltan
>

Reply via email to