An update here is we have merged the PR (thanks to everyone who reviewed and approved) and the new explanatory page is live on the website[1]
Also, as Micah's suggestion, I have another PR[2] to render the page from the same data that drives the implementation status page, so they are less likely to drift apart. Andrew [1]: https://parquet.apache.org/docs/file-format/versions/ [2]: https://github.com/apache/parquet-site/pull/189 On Wed, Jun 10, 2026 at 5:05 PM Andrew Lamb <[email protected]> wrote: > Thank you -- that is an excellent point. > > I have removed the V1 and V2 columns[1] and applied another suggestion by > Ed[1]to avoid the implication of meaning to V1 and V2. > > Andrew > > [1]: > https://github.com/apache/parquet-site/pull/186/commits/1126b539491040f97627f720f9c593a9705a4377 > [2]: > https://github.com/apache/parquet-site/pull/186/commits/8cb094252a28f8e8b63bd454b4310f50a0e07a61 > > On Wed, Jun 10, 2026 at 8:51 AM Antoine Pitrou <[email protected]> wrote: > >> >> Well, the features table still has columns "V1" and "V2". If we agree >> that 2.0.0 was not special compared to other parquet-format releases (in >> particular, it didn't break forwards compatibility), then why single it >> out? >> >> >> >> Le 10/06/2026 à 13:03, Andrew Lamb a écrit : >> > Thanks for taking a look. >> > >> >> I think this is a nice doc page, except that it's inventing an a >> >> posteriori meaning for "V1" and "V2". >> > >> > I agree that earlier versions of the document did try and invent >> meaning, >> > which Ed and Jorris pointed out[1], and I have tried to remove in >> several >> > updates like [2] (and [3] this morning) I would be happy to remove or >> > reword any additional sections you think are implying such a meaning >> > >> > The current proposed wording is this: >> > >> >> FileMetadata version field >> >> >> >> Each Parquet file has a version field in the thrift FileMetadata. This >> > field has >> >> historically been used inconsistently: writers populate 1 or 2 >> >> without a consistent relationship to the features actually used. See >> the >> >> note in parquet.thrift and this discussion for details. >> >> >> >> parquet-format release versions >> >> >> >> The Thrift definition is released independently of implementations such >> > as >> >> parquet-java or arrow-rs, following the Apache release process. Note >> that >> > release >> >> numbering DOES NOT FOLLOW semantic versioning: >> >> >> >> 1. The major version corresponds to the thrift FileMetadata version >> field. >> >> >> >> 2. Minor releases (e.g. 2.10.0 to 2.11.0) sometimes contain forward >> > incompatible >> >> features. The minor version is not recorded in the file itself. >> > >> > Are there other parts of the document you feel incorrectly imply a >> meaning >> > for V1 and V2? >> > >> > Thanks, >> > Andrew >> > >> > [1]: >> https://github.com/apache/parquet-site/pull/186#discussion_r3380588765 >> > [2]: >> > >> https://github.com/apache/parquet-site/pull/186/commits/89159332dc770c64d88f48fcdeb24be53fc82161 >> > [3]: >> > >> https://github.com/apache/parquet-site/pull/186/commits/0b3a17f0e8cddc39eeccdc3ca2fbd7e2def0b077 >> > >> > >> > On Wed, Jun 10, 2026 at 5:01 AM Antoine Pitrou <[email protected]> >> wrote: >> > >> >> >> >> hi Andrew, >> >> >> >> I think this is a nice doc page, except that it's inventing an a >> >> posteriori meaning for "V1" and "V2". Why is it useful? Why single out >> >> V2 aka. parquet-format 2.0.0? >> >> >> >> Regards >> >> >> >> Antoine. >> >> >> >> >> >> Le 05/06/2026 à 16:18, Andrew Lamb a écrit : >> >>> Dear Parquet Fans, >> >>> >> >>> I have become convinced over the last few discussions that it is more >> >>> important than ever to document clearly what V1 and V2 mean (including >> >> the >> >>> messy reality) >> >>> >> >>> Thus, I spent several hours documenting Parquet features and when each >> >> was >> >>> introduced [1]. I would love any feedback you may have. >> >>> >> >>> Thank you, >> >>> Andrew >> >>> >> >>> [1]: https://github.com/apache/parquet-site/pull/186 >> >>> >> >> >> >> >> >> >> > >> >> >>
