model-writtenrun 8861 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Robert Haas opened the thread with #1, identifying missing DDL restrictions and unnecessarily broad prohibitions on utility commands in parallel mode, and proposed one exhaustive classifier for read-only transactions, parallel execution, and recovery.
- Tom Lane questioned the weakly-read-only flag and ALTER SYSTEM semantics in #2, while Stephen Frost supported explicit permission flags in #6; Robert adopted the flag simplification in #7 while deferring broader behavioral changes.
- Historical investigation established that SQL permits temporary-table writes in read-only transactions, as Tom explained in #10, and Peter Eisentraut proposed changes to pg_dump output as a practical definition of database writes in #14.
- Robert incorporated that explanation into the comments in #15, documenting why maintenance commands and ALTER SYSTEM remain allowed and why COMMIT PREPARED is an exception.
- Discussion continued over configuration changes and backups: Tom challenged the pg_dump criterion in #16, Stephen argued for stronger protection against externally visible changes in #18, and Laurenz Albe raised cross-version restoration concerns in #19; Tom emphasized that read-only transactions and standby restrictions are distinct in #20.
- Robert proposed landing the refactoring independently of ALTER SYSTEM policy in #26, with agreement from Tom in #27 and Stephen in #28; Robert also explained how COMMIT PREPARED can expose previously prepared writes despite a read-only environment in #30, prompting Stephen to favor a separate behavioral change in #31.
- Robert committed #15 as 2eb34ac369 with identical code changes apart from rebase context, and announced the landing in #32; redefining existing read-only semantics remained outside its scope.
- After the commit, Stephen proposed improving user documentation in #33, and Robert invited specifics in #34; the remaining discussion concerned configuration compatibility, ending with Bruce Momjian explaining differing disruption from GUC changes in #37, with no further patch posted for documentation or the unresolved policy questions.