model-writtenrun 2638 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed pure Bison parsers, reentrant Flex scanners, palloc-based allocation and simpler buffer management in #1, organizing small review patches with the intention of combining them into module-sized commits.
- Peter extended the work to bootstrap, GUC and the PL/pgSQL scanner in #2, and Heikki Linnakangas approved these incremental steps in #3.
- Andreas Karlsson found incorrect replication error-call arguments and a dangling syncrep yyextra pointer, supplied replication fixups and yyextra code, and recommended withholding PL/pgSQL until its remaining scanner state was addressed in #4.
- The cube and seg portions of #2 landed as 802fe923e3 and 1f0de66ea2, with only a seg comment correction.
- C99 buildfarm failures after those landings prompted Peter's typedef-guard delta in #5; Andreas endorsed consistent #ifndef wrappers in #7, and the cube and seg fixes landed as 382092a0cd.
- Andreas's independent blank-line cleanup from #4 landed as 399d0f1e11, while the bootstrap portion of #2 landed with typedef guards as 3e4bacb171.
- Peter initially disputed the yyextra lifetime concern in #8, then recognized the escaping stack pointer and posted the corrected remainder in #10, incorporating Andreas's replication work and retaining stack allocation only where jsonpath's scanner lifetime permitted it.
- Tom Lane reported Flex 2.5.35 prototype warnings in #11; Peter questioned supporting an unobtainable old release in #12, and Tom agreed that the existing warning workarounds could be removed in #13.
- Julien Rouhaud explained that newer i386 coverage already existed in #14, while Tom and Peter favored retaining useful older-platform coverage on a supported Debian release in #15 and #16.
- After Andreas rechecked and approved the remainder in #17, its replication and syncrep portions landed as e4a8fb8fef and db6856c991, with gettext metadata updates and removal of replication debugging text.
- The jsonpath and GUC portions of #10 landed as 2a7425d7ee and d663f150b5, adding a jsonpath gettext adjustment and GUC whitespace cleanup.
- Peter completed the deferred PL/pgSQL parser and scanner-state work in #18 and added a separate switch from YY_EXTRA_TYPE definitions to Flex extra-type options; Heikki approved both and checked syntax compatibility in #19.
- The extra-type patch landed unchanged as b1ef48980d, and the three PL/pgSQL patches landed as 7b27f5fd36 with an additional datatype error-callback scanner fix and warning suppression; Peter declared the main conversion complete in #20.
- Peter posted removal of the old Flex warning workarounds and a tentative minimum-version increase in #21; the workaround-removal portion landed unchanged as 6fdd5d9563.
- Tom recommended eliminating version enforcement instead of raising the minimum in #22, so Peter replaced the proposed version bump with 0869ea43e9 and confirmed completion in #23.
- Peter addressed the final deferred topic in #24 by returning parse results through output parameters in PL/pgSQL, pgbench, replication and syncrep, including passing syncrep error state down to the scanner.
- All four final patches landed together unchanged as 473a575e05, and Peter closed the topic in #25 with no submitted work left pending.