model-writtenrun 7948 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed removing currtid() and currtid2() because INSERT RETURNING appeared to have superseded their ODBC use, alongside a separate patch moving heap_get_latest_tid() into heapam_handler.c #1.
- Hiroshi Inoue explained that currtid2() still supported detecting updated rows for keyset-driven cursors, exposing a compatibility concern with removing both functions #4 #8 #10.
- Andres Freund argued that heap_get_latest_tid() belonged in the lower-level heapam.c, and Michael declined to press the relocation; that patch was omitted from later submissions #5 #6 #18.
- Michael investigated why the ODBC tests no longer exercised currtid2(), and Hiroshi confirmed that earlier driver changes had invalidated keyset-driven cursor coverage; Michael remained unable to establish that all remaining calls were unreachable #13 #16 #17.
- After the summer gap, Michael rebased the main removal patch for CI while the compatibility question remained unresolved #18.
- Peter Eisentraut and Tom Lane argued that the remaining ODBC calls required keeping currtid2(), leading Michael to narrow the patch to removing currtid() and setLastTid() while initially changing currtid2() to accept regclass #19 #20 #21 #23.
- Tom objected that changing the retained function's signature risked breaking compatibility, so Michael restored currtid2(text, tid) and introduced a shared internal helper for relation-name lookup and recursive view handling #24 #26 #28.
- Michael committed #28 as 7b94e99960, with only whitespace normalization and an updated catalog-version date, then confirmed that currtid() was gone and currtid2() remained compatible; removal of currtid2() and the heap-function relocation were left out #29.