Repository navigation
10.x: RangeIntersectAgg returns empty over a sequence holding a null range, where range_intersect_agg skips the NULL; RangeAgg throws (port of #361) #363
Description
Problem
The same defect as #361, on release/10.x (10.1.0). RangeAggregateExtensions.cs and RangeSet.cs are identical to main, and docs/ranges.md#aggregate-a-sequence-of-ranges makes the same promise: the two are "the in-memory versions of PostgreSQL's range_agg and range_intersect_agg". Those skip NULL inputs. The in-memory versions don't. Over a nullable DateRange column holding NULL, [2026-03-01,2026-03-10], [2026-03-05,2026-03-25] and NULL:
RangeIntersectAgg()returnsemptyin memory and[2026-03-05,2026-03-10]in SQL, with no exception.RangeAgg()throwsInvalidOperationExceptionin memory and returns{[2026-03-01,2026-03-25]}in SQL.
Change (10.2.0)
- Cherry-pick RangeIntersectAgg returns empty over a sequence holding a null range, where range_intersect_agg skips the NULL; RangeAgg throws #361's fix with
-x:RangeAggandRangeIntersectAggskip null ranges, over ranges and over sets. Over nothing but nulls, they return what they return over an empty sequence. ### Changed answersrow in[10.2.0]linking this issue and itsdocs/migration.md#changed-answers-in-1020entry, in the root changelog and in the core and NodaTime package changelogs (the aggregates are generic, so the NodaTime ranges get the same answers).RangeAgggoing from a throw to a value goes in the same row.
Done when
On release/10.x, RangeAggregateParityTests over a nullable column holding NULLs among other shapes passes on PostgreSQL 16 and 18 for every family, and fails with the fix reverted. Close by hand after the merge.
Activity
CaffeinatedCoder commentedon Oct 24, 2026
The
mainfix is PR #368 (commit 99a6989). The port cherry-picks it with-x. It brings the seedtest/oracles/seeds/aggregates-fold-a-null-input.patchand claim R48. Its### Breakinglines under[11.0.0-preview.1]become a### Changed answersrow under[10.2.0](core and NodaTime changelogs), and its migration subsection moves under## Changed answers in 10.2.0. Rundotnet run tools/oracles.cs seeds aggon the branch: the three existing aggregate seeds must still apply there.CaffeinatedCoder commentedon Oct 24, 2026
Port PR: #369 against
release/10.x, the-xcherry-pick of 99a6989 with the changelog and migration entries rewritten for[10.2.0]/## Changed answers in 10.2.0. Locally green on PostgreSQL 16 and 18 and on EF Core 10.0.13 / Npgsql 10.0.3. With the fix reverted the four new tests fail, and all four aggregate seeds are caught on the branch.CaffeinatedCoder commentedon Oct 24, 2026
Fixed on
release/10.xby #369, merged as 87b473a:RangeAgg()andRangeIntersectAgg()skip anullrange or set, in every per-type overload (core and NodaTime) and both set overloads, with the### Changed answersrow in[10.2.0]and the subsection under## Changed answers in 10.2.0. Ships in 10.2.0 (#358). Closed by hand, sinceClosesacts only onmain.