model-writtenrun 2253 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Steven Niu proposed replacing byteain's exact-size parsing pass with strlen-based upper-bound allocation followed by decoding #1; Kirill Reshke pointed out that strlen still scans the input and requested performance measurements #2, while Steven also defended the simplification of the code #3.
- Stepan Neretin supplied a two-patch series retaining Steven's change and adding a StringInfo-based rewrite with regression tests #4, then corrected author metadata #5 and fixed mixed-declaration compiler warnings #6.
- After Aleksander Alekseev requested a microbenchmark #7, Stepan reported a small but consistent improvement for the revised series using repeated escaped input #8.
- Peter Eisentraut questioned the relationship between the two patches and unrelated code movement, and explained that bytea input tests belonged in src/test/regress/sql/strings.sql rather than contrib/btree_gin #9.
- Tom Lane chose the original strlen-based approach, rejecting StringInfo because its overhead and allocation behavior offered no worthwhile simplification; he requested double-backslash coverage and noted that byteain had moved to bytea.c #10.
- Tom committed the first patch from #6 as 3683af6170, with light cleanups and double-backslash tests, and reported approximately 40% faster execution for long input using Stepan's benchmark #11.
- Stepan accepted the decision and requested access to the final diff #12; David G. Johnston explained the commit archive's gitweb link #15, and Tom supplied the direct commit link #16, leaving the original optimization landed and the StringInfo alternative rejected.