model-writtenrun 1679 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- jian he opened the thread with #1, proposing to expand virtual generated columns, compute stored generated values before COPY WHERE filtering, and reject system-column references except tableoid.
- Kirill Reshke reported successful testing and suggested using CMD_UTILITY in #2; jian he defended retaining CMD_INSERT because it matched the generated-column executor machinery in #3.
- Masahiko Sawada demonstrated that early computation exposed generated values to BEFORE INSERT triggers and questioned both filtering semantics and the tableoid exception in #4; jian he responded with #5, splitting the generated-column patches and using a temporary slot for stored values.
- Peter Eisentraut replaced that approach with a backpatchable prohibition of both generated-column kinds, including whole-row references, in #6, deferring possible support to future releases and leaving system-column handling to separate discussion because other system columns were already rejected and tableoid behavior remained unresolved.
- Masahiko Sawada endorsed the replacement and requested a more appropriate error code and removal of an unrelated whitespace edit in #7; Peter incorporated both corrections from #6 into aa606b9316 and announced completion in #8.
- With the prohibition committed but the commitfest entry still open, Peter Eisentraut argued in #9 against pursuing the earlier support patches because COPY WHERE filters input data while generated columns belong to the target table, and trigger timing would make the semantics complicated.