model-writtenrun 9768 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund opened the discussion by asking whether pluggable storage should retain the name heap, rename the existing storage method, or rename generic heap_* infrastructure to distinguish it from storage-specific code #1.
- Thomas Munro explained that heap conventionally means unordered tuple storage #4, and Andres noted that efficient index-organized tables would require further work despite the new interface enabling progress toward them #6.
- Alvaro Herrera suggested names preserving the heap distinction #7, but Robert Haas favored retaining the familiar heap name and renaming generic infrastructure instead #10.
- Andres proposed moving relation_open and related functions into separate relation files and renaming heap_open variants to table_open with compatibility redirects #12; Robert supported separating the relation and table interfaces #14.
- Andres posted a three-part implementation that introduced the separate interfaces, narrowed existing includes, and renamed callers, keeping table.h independent of relation.h and retaining compatibility macros #15.
- Haribabu Kommi examined the resulting header usage and questioned generic relation_open calls where the relation type was known #16, while Robert raised the duplicate relation.h filename concern #17 and Tom Lane agreed #18.
- With no better header name forthcoming, Andres proceeded #20 and announced the commits #21: 4b21acf522 landed the interface extraction with comment and include cleanup, 111944c5ee landed the include migration with small additional cleanup, and e0c4ec0728 landed the caller renaming with formatting and rebase adjustments.
- The January series was fully committed, while renaming or splitting nodes/relation.h remained optional follow-up work #21.
- Andres reopened the naming discussion in March for HeapUpdateFailureData, HTSU_Result, and their result codes #22; Robert mildly favored changing the failure-data name but cautioned against replacing already familiar result-code names #23.
- Andres explained the need to distinguish concurrent deletion from update #24, then showed an inline TM_Result enum using Modified terminology and considered shorter prefixes, an OK success value, and removal of compatibility aliases #25.
- Haribabu supported the clearer enum and dropping compatibility macros #26; in the final message, Andres suggested a TM_ prefix to protect the global enum namespace and explained why OK suited tuple locking #27.
- The later 5db6df0c01 adopted TM_Result, TM_FailureData, and TM_Ok within the broader tuple-modification implementation, whose latest matching patchset belongs to thread #38991; this thread supplied naming discussion rather than that implementation.