model-writtenrun 9712 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Mitar proposed temporary materialized views to cache expensive queries for a session and disappear when the client disconnects, attaching an initial implementation in #1.
- Alvaro Herrera supported session-lifetime views, recommended reusing OptTemp and adding documentation and tests, and discouraged ON COMMIT actions in #2; Mitar agreed to leave those actions out in #3, while Alvaro advised leaving the existing GLOBAL warning unchanged in #5.
- Mitar expanded the feature to infer temporary persistence from temporary source relations and added documentation and refresh tests in #6, then added cascading-drop coverage in #8 and deterministic query ordering in #9.
- Andreas Karlsson reported successful builds and tests but questioned removing SECURITY_RESTRICTED_OPERATION in #10; Tom Lane emphasized that privilege changes within a session still matter in #13.
- Andreas identified relation creation inside the receiver startup callback as the obstacle in #14, and Tom rejected delaying security restrictions until that callback because planning and startup also need protection, recommending earlier object creation in #15.
- Andreas's functional testing found missing psql completion in #16, which Mitar added in #20; Andreas subsequently questioned completing unsupported UNLOGGED materialized views in #22.
- Andreas supplied a separate CTAS refactor in #21 that creates the empty relation before query execution while retaining security restrictions, noting that the query snapshot now sees the newly created relation; he rebased it with minor cleanup in #23.
- Michael Paquier recommended a separate thread for the broader CTAS work in #26 and proposed simplifying its receiver state and creation API in #27.
- Andreas explained that EXPLAIN ANALYZE and prepared EXECUTE still run executor setup with WITH NO DATA in #28, then identified inconsistent IF NOT EXISTS handling and questioned EXPLAIN's security-context handling in #29.
- Michael argued for preserving documented behavior and posted independent regression tests for both WITH NO DATA paths in #30; Andreas approved them in #31, and they landed unchanged as 537898bd81, confirmed in #32.
- Michael concluded that the remaining CTAS redesign and EXPLAIN issues made PostgreSQL 12 an unsuitable target in #34, and Andreas agreed in #36.
- Andreas confirmed that temporary materialized views remained blocked on a common creation interface that preserves side-effect restrictions in #38; Mitar agreed to wait in #39, and David Steele marked the feature Returned with Feedback in #40, leaving only the independent regression tests committed.