model-writtenrun 2268 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Masahiro Ikeda proposed documenting that adding identity columns or columns with constrained domain types also rewrites the table and its indexes, alongside the existing volatile-default case, in #1.
- Robert Treat retained the introductory tip about fast constant-default additions because users still found rewrites confusing, removed its historical version reference, and polished the reference documentation in #2; Masahiro questioned removing 'However' in #3 and accepted Robert's explanation in #5.
- Álvaro Herrera committed #2 as 11bd831860, with small editorial changes that separated type changes from the enumeration of column-addition cases, and announced the result in #6.
- Jian he reported that the committed explanation omitted the exception for virtual generated columns in #7; Álvaro initially suggested tracking the issue while waiting to see whether related limitations would change in #8.
- After Nathan Bossart asked whether documentation work could proceed in #9, Álvaro recommended documenting current behavior and leaving further code improvements to the next development cycle in #10.
- Peter Eisentraut supplied a follow-up patch in #11 that distinguished identity columns from stored generated columns, explicitly stated that virtual generated columns never require a rewrite, and split the explanation into clearer paragraphs.
- Robert Treat approved the follow-up with wording suggestions in #12, and Peter incorporated them with light adjustments in 06c4f3ae80, avoiding wording that might imply users must rebuild indexes manually.
- Peter announced the follow-up commit and closed the open item in #13; Robert confirmed that the committed wording looked good in #14, leaving both documentation patches landed.