model-writtenrun 2034 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Lukas Fittl opened the thread with #1 in March 2023 after finding that EXPLAIN did not expose the estimated Memoize hit ratio responsible for reducing nested-loop costs, and attached a patch displaying that ratio.
- David Rowley argued in #2 that estimated cache capacity and distinct lookup keys would explain the ratio's inputs more usefully, and Ilia Evdokimov revived the work almost two years later with those two values in #5.
- David's review in #6 prompted clearer names, unsigned capacity formatting, and more compact output across #9 and #13, while his initial concern about larger plan structures eased after he distinguished them from executor state in #11.
- Andrei Lepikhov argued that locally displayed hit ratios were easier to interpret in large plans in #10 and #15, while Tom Lane questioned default-output verbosity and proposed pg_overexplain in #17; Lukas defended ordinary-user access in #18, and Robert Haas opposed requiring a developer-oriented extension in #23.
- Ilia retained capacity and distinct-key estimates in #27, but the unresolved display policy prevented agreement before the PostgreSQL 18 feature freeze, as David explained in #29.
- Ilia tried David's compromise in #30, displaying estimates only during EXPLAIN ANALYZE when cache misses occurred, but Robert rejected requiring execution to obtain planning information and renewed the case for showing the hit ratio in #32.
- Following David's proposal in #34, Ilia returned to the earlier unrestricted display approach and split capacity/distinct-key estimates from the restored hit-ratio feature in #35, including regression-output examples to make the alternatives concrete.
- David proposed a compact Estimates line, percentage formatting, and an explicit lookup count in #36, then clarified that COSTS OFF should suppress the new fields in #39; Ilia implemented those choices as a three-patch series in #41, removing the regression-output churn.
- David identified ambiguous percentage units in structured output and inconsistent field names in #43, leading Ilia to combine the series and adopt explicit Hit Percent labels and est_ field names in #45.
- Andrei's proposal for significant-digit formatting was deferred to a broader discussion in #46 and #47, and his later suggestion to expose comparable IncrementalSort estimates was likewise left for another thread in #49 and #50.
- After Ilia's July rebase in #48, David supplied #51 with naming, comments, Cardinality types, and structure-layout improvements while retaining Ilia's authorship.
- Andrei endorsed the final attachment and compact Buffers-style formatting in #52; a separate structured-output group was considered unnecessary in #53, and Ilia agreed that the existing attachment was appropriate in #55.
- David committed #51 as 4bc62b8684, adding only light typedef and explain.c cleanup described in #56; the result displays capacity, distinct keys, lookups, and hit percentage whenever costs are enabled.
- The thread ended with the feature landed and Lukas endorsing the final posted version and compact output in #57, with broader formatting and other-node enhancements left outside this change.