model-writtenrun 3868 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Dec 2022 - Jan 2023 · 7 entries
- Joseph Koshakow reopened the proposal for infinite intervals in #1, suggesting reserved extreme month values and noting the compatibility implications for existing finite intervals.
- Ashutosh Bapat supplied an unfinished implementation in #2 that represented infinities with extreme values in all three interval fields and invited Joseph to complete it.
- Joseph expanded the tests and implementation in #4 and #5, choosing IEEE-like arithmetic with errors for indeterminate results and moving shared infinity checks into timestamp.h macros.
- Jian identified incorrect infinity division and requested positive-infinity input support in #8; Joseph added NaN and infinity checks in #11, while Vik Fearing said in #9 that he would handle the input spelling in a separate thread.
- Jian supplied test consolidation and arithmetic cleanups in #13 and #14, which Joseph incorporated in #16 before correcting missing cases, replacing procedural tests with SQL joins, and clarifying extraction behavior in #19.
- Jian's generate_series() examples in #21 prompted a wider audit of interval consumers, leading Joseph to extend infinity handling and introduce make_interval() overflow checks in #23.
- Joseph changed infinity recognition to inspect only months in #24, leaving a compatibility decision that would later be revisited.
Mar - Apr 2023 · 7 entries
- After a gap, Joseph rebased in #26, and Ashutosh's detailed review in #27 led to the three-part series in #28: integer helpers, make_interval() overflow protection, and infinity support.
- Tom Lane's feedback in #29 and #33 led to explicit floating-point comparisons and correctly parenthesized macro conditions in #30 and #34.
- Ashutosh approved the prerequisites and suggested DAYS_PER_WEEK in #35, then recommended float8_mul() and reconsideration of three-field infinity recognition in #41.
- Ashutosh's multiplication fix-ups in #44 and #45 corrected finite-times-infinity and zero-times-infinity behavior; Joseph incorporated them alongside floating-point conversion bounds checks in #46.
- Joseph proposed shared infinity-arithmetic helpers and subtraction through negation in #47, but Tom rejected the added complexity, misleading error contexts, and intermediate-overflow hazard in #48 and #50.
- Joseph retained the earlier arithmetic structure and removed error contexts in #51, then incorporated Ashutosh's omitted multiplication tests from #52 and refined horology coverage through #56.
- Ashutosh marked the work ready for a committer and targeted PostgreSQL 17 in #55; the June refresh in #58 was followed by an assertion correction in #59.
Jun - Jul 2023 · 1 entry
- Tom diagnosed an invalid assumption about isinf() return values in #60, and Joseph fixed the portability issue and resulting compiler warning in #61 and #62.
Aug - Nov 2023 · 16 entries
- After the August refresh in #63, Dean Rasheed found broken infinity input and moving-aggregate behavior in #64, reopening implementation work.
- Ashutosh initially restored removed parser code in #65, but Dean showed in #69 that this still mishandled reserved words and infinity with ago, leading to explicit parser validation in #75.
- Jian introduced an internal aggregate state with separate infinity counts in #67; Dean's reviews in #70 and #72 drove allocation reductions, corrected parallel-combine counts, and shared finite-arithmetic helpers.
- Ashutosh refactored the aggregate implementation in #75, then incorporated Jian's serialization patch from #78 and overflow-safe addition/subtraction in #80.
- Ashutosh corrected deserialization allocation to use the current memory context in #81, and Dean endorsed that choice in #82.
- Dean's substantial review revision in #85 restored three-field infinity recognition to preserve comparison consistency across upgrades, fixed infinite window offsets, and removed the unnecessary interval-to-timestamp selectivity helper.
- Dean also dropped the proposed multiply-add integer helpers in #85, judging their utility and semantics insufficient, and rewrote make_interval() using existing overflow primitives; Ashutosh approved that rewrite in #86.
- Ashutosh added negative window-offset tests and BRIN work in #87; Dean retained the window tests but removed BRIN code being handled in another thread and updated deserialization in #89.
- Dean committed the unchanged make_interval() component of #89 as b2d55447a5, then published the squashed infinity remainder with a btree_gin adjustment in #91.
- Ashutosh reported a complete reread and supplied separate BRIN infinity tests in #92, which Dean folded into the main patch in #94.
- During that final pass, Dean discovered pre-existing interval_time() overflow and missing interval-negation checks, adding a separate conversion fix and restoring negation validation and tests in #94.
- Catalog OID collisions prompted #95, and Windows failures prompted #96; Dean then traced the remaining failure to a separate pg_sub_s64_overflow() bug and opened another thread in #98.
- Ashutosh bundled that external subtraction fix with the two remaining patches in #99, and Dean confirmed successful CI in #100.
- Dean committed the unchanged interval_time() component of #99 as 3850d4dec1 and announced the prerequisite landings in #101, resending only the main feature with helper functions repositioned.
- Dean committed infinity support from #101 as 519fc1bd9e, with only consolidated error checks and a catalog-version bump, and announced the landing in #102.
- The thread ended with thanks from Ashutosh and Joseph in #103 and #104; the main feature and both associated overflow fixes had landed, with no remaining patch awaiting review.