The C/C++ and Rust implementations of Arrow seem to assume (and validate) that the offsets in a variable size array such as Utf8 and Binary are monotonic, even for values that are null (when the validity is not set).
The existing format documentation does not seem to address this matter explicitly, so I propose a change to make it explicit [1]. Since this is related to the overall format and there may be previous discussion on the topic I missed, I wanted to raise it on the mailing list for broader input. I also searched the archives of the mailing list but was not able to find anything previously on this topic Thanks, Andrew [1] https://github.com/apache/arrow/pull/12019