ARROW-7254: [Java] BaseVariableWidthVector#setSafe appears to make value offsets inconsistent

[liyafan82]

See the discussion in ARROW-7254

[github-actions]

https://issues.apache.org/jira/browse/ARROW-7254

[emkornfield]

It setvaluecount happens after set(abcd) is this fix still needed

[emkornfield]

OK, see comment on JIRA, I'm a little bit mixed on if we should patch this change as the pattern goes against what the documented usage should be. That being said, since this is non-obvious it might be better to fix to avoid others getting mixed up with this as well.

[liyafan82]

OK, see comment on JIRA, I'm a little bit mixed on if we should patch this change as the pattern goes against what the documented usage should be. That being said, since this is non-obvious it might be better to fix to avoid others getting mixed up with this as well.

There is no problem if we call setValueCount after manipulating the vector, as the setValueCount will call fillHoles, which brings the vector to a consistent state.

Strictly speaking, this is not consistent with the patten of the life cycle of a vector. This problem only exists for VariableWidthVectors.

[lidavidm]

Thank you @liyafan82 for investigating this!