model-writtenrun 8970 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed removing both float passing configure options because version-0 functions were no longer supported, making float4 always pass by value and float8 depend on pointer width; his series also consolidated float8 macros and replaced selected preprocessor conditionals with compiler conditionals #1.
- Tom Lane objected that hardwiring float8 would prevent testing pass-by-reference paths on 64-bit machines and asked to retain a developer option, while accepting the float4 direction #2.
- Robert Haas proposed making Datum eight bytes everywhere and questioned storing typbyval in catalogs #3; Peter Geoghegan supported reducing platform-specific code #6, while Tom Lane warned about instruction and memory costs on 32-bit systems #5.
- Peter Eisentraut separated the float4 removal into its own patch so it could proceed while float8 remained under discussion #12.
- Peter Eisentraut and Tom Lane identified extension compatibility constraints on changing argument-passing conventions #13 #14; Andres Freund sketched centralized conversion macros and an opt-in automatic PASSEDBYVALUE setting #15, while Robert Haas questioned why a float8-named switch controlled other types #17.
- Peter Eisentraut committed the isolated float4 patch from #12 as 2e4db241bf, changing only version numbers beyond rebasing, and announced that he would continue developing the remaining work #18.
- Peter Eisentraut revised the remaining float8 proposal to remove the user-facing configure option while keeping USE_FLOAT8_BYVAL in pg_config_manual.h for developer testing #20; Tom Lane accepted the compromise #21, and Peter committed the attachment unchanged as 4513d8b07b and confirmed completion #22.
- After both option changes landed, Peter Geoghegan renewed concerns about rarely tested 32-bit paths and memory leaks #23, Peter Eisentraut raised continued 32-bit Windows use #24, and Robert Haas again advocated exploring universally eight-byte Datums #25.
- The thread ended with Tom Lane identifying pointer/integer size warnings as a prerequisite problem for wider Datums #26; that broader redesign remained unresolved, and the thread records no landing of the original macro-consolidation and conditional-cleanup patches from #1.