model-writtenrun 7701 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Heikki Linnakangas proposed removing protocol-2 COPY FROM STDIN support because its byte-by-byte end-marker detection prevented safe prefetching for planned encoding-conversion improvements, attaching an initial removal and lookahead patch in #1.
- Tom Lane argued for removing protocol 2 completely in #2; after Alvaro Herrera established that JDBC had already dropped it in #3, Tom advocated removing both server and libpq support in #4.
- Heikki also suggested moving away from the fast-path function interface in #5, but Tom opposed that expansion because fast-path remained part of protocol 3 and might have other users in #6.
- Heikki posted full protocol removal and a separate COPY lookahead simplification in #8, retaining legacy error handling so old clients and servers could still communicate connection failures while switching early backend errors to protocol 3.
- Alvaro identified an ABI problem caused by removing CONNECTION_SETENV in #11; Tom agreed existing enum values must remain stable in #12, and Heikki accepted preserving the value without a library version bump in #13.
- John Naylor verified that a PostgreSQL 7.3 client received the expected unsupported-protocol error and checked for remaining protocol-2 code in #14.
- Tom rebased and tested the removal patch, preserved the public enum value, and removed additional obsolete client-side branches in #15, explicitly leaving the COPY lookahead patch unreviewed; he then found a stale translation-file reference in #16.
- Heikki committed Tom's revised removal patch as 3174d69fb9, adding the translation-file cleanup and minor formatting changes, and announced the landing in #17.
- The extra newline Tom observed during compatibility testing in #15 led to 85d94c5753, which appends a newline only when absent; Heikki explained the historical cause in #17, and Tom subsequently raised backpatching in #18.
- Tom reported crashes on the MinGW buildfarm member walleye in #19; Heikki investigated and separately identified stale-build linker failures on drongo in #20.
- Andrew Dunstan acknowledged drongo's build problem in #21, and Tom confirmed its next run succeeded while questioning dependency-file regeneration in #22; the thread ends with protocol removal and the newline fix landed, walleye's crashes unresolved, and no recorded landing of the COPY lookahead patch or resolution of the backpatch question.