model-writtenrun 7944 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Matthieu Garrigues resumed the existing libpq batching effort with #1, reporting fixes and proposing tests for multiple transactions and transactions spanning batches, while leaving a result-allocation failure concern unresolved.
- Dave Cramer argued that applications should not need a separate PQbatchProcessQueue call in #4; Matthieu agreed to preserve NULL separators between queries in #7 and implemented automatic queue advancement in #9.
- Michael Paquier reported documentation and Windows build failures in #10, prompting Matthieu's attempted fixes in #12.
- Alvaro Herrera took over cleanup in #13, removing an accidental linux.h overwrite and a windows.h inclusion, then expanded result-status documentation in #15 while temporarily disabling a broken single-row test.
- Alvaro requested practical API feedback in #17, and Matthieu reported that an asynchronous HTTP server using batching achieved a two-to-threefold benchmark improvement in #19.
- David G. Johnston reviewed the documentation in #20, questioning blocking-mode support and asking for clearer explanations of execution order, transactions and batching terminology; Matthieu incorporated several suggestions in #22.
- Andres Freund requested restored pgbench support for realistic testing and performance investigation in #21, while Alvaro separately proposed requiring nonblocking connections in #23, a restriction not adopted by the supplied patches.
- Alvaro reorganized documentation and libpq helpers in #24, isolated TAP test cases to prevent cascading failures, repaired single-row testing and removed timing comparisons intended to be handled by pgbench.
- Alvaro also recorded unresolved concerns about asynchronous state transitions in #24, while defending the design that leaves applications responsible for matching results to submitted queries.
- Daniel Vérité identified pgbench variable dependencies as incompatible with unrestricted batching in #25; Alvaro chose simple rejection of result-to-variable commands over dependency tracking in #27, with Andres supporting minimal initial functionality in #28.
- David supplied a separate documentation correction patch in #26, covering memory use, phrasing and error descriptions; those edits were not included in the next posted full patchset #29.
- Daniel added pgbench batch delimiters and rejection of gset/aset within batches in #29, and Alvaro confirmed incorporation into his local copy in #30; this thread ends with integration and review still underway.
- The eventual pipeline-mode commit acb7e4eb6b landed from the later March 2021 submission #472027 in thread #34701, so this thread documents an earlier development phase rather than the final committed patchset.