Query: translate GroupBy element reads with ROW_NUMBER only

[QHarshil]

Fixes #36380
Fixes #29240

GroupBy(k).Select(g => g.OrderBy(o).First()) translates to a GROUP BY joined back on the key to a ROW_NUMBER() subquery. That reads the source twice, and the = join loses the element of the NULL group: FirstOrDefault() returns null for it, and First() over a projected element throws "Sequence contains no elements" (#29240).

This PR translates it to the ROW_NUMBER() filter alone:

SELECT [o0].[OrderID], [o0].[CustomerID], [o0].[EmployeeID], [o0].[OrderDate]
FROM (
    SELECT [o].[OrderID], [o].[CustomerID], [o].[EmployeeID], [o].[OrderDate], ROW_NUMBER() OVER(PARTITION BY [o].[CustomerID] ORDER BY [o].[OrderID] DESC) AS [row]
    FROM [Orders] AS [o]
) AS [o0]
WHERE [o0].[row] <= 1

It applies when the projection uses the grouping only through g.Key and one First, FirstOrDefault, Last or LastOrDefault over the elements, optionally ordered and projected, so MaxBy and MinBy too. Every group has a first row, so the groups are the same. Operators after the read compose over it, so a Where on the element becomes a plain WHERE instead of a correlated TOP(1) subquery in HAVING.

Today's translation stays for a filter, Skip or ElementAt in the group, aggregates, two element reads, scalar reads like g.First().Name, GroupBy in a subquery or a correlated SelectMany, and elements with included navigations when more operators follow. Those still lose the NULL group.

Implementation

• TryTranslateGroupingElementRead, next to TryTranslateGroupingElementProjection from Stream GroupBy elements instead of joining the source back to itself #38963, matches the shape and translates the orderings before changing the SelectExpression. Any other shape falls through.
• The internal SelectExpression.ApplyFirstRowPerGroup drops the grouping and adds the partitioned ROW_NUMBER() filter, using the fallback orderings AddJoin uses.

Performance

EF's SQL before and after, 5% NULL keys, median of 5. SQL Server 2025 ran in an x64 container on Apple Silicon.

rows	groups	index	query	SQL Server before	after	SQLite before	after
1M	1,000	no	First()	211 ms	114 ms	1,103 ms	795 ms
1M	50,000	yes	First()	174 ms	146 ms	1,232 ms	948 ms
1M	1,000	no	First(), then Where	2,028 ms	140 ms	23,725 ms	839 ms
1M	50,000	yes	First(), then Where	2,590 ms	130 ms	1,991 ms	1,661 ms

Non-NULL groups returned the same rows in every run.

Tests

• 18 new tests: nullable keys (value type, reference, navigation, composite) with First, FirstOrDefault, LastOrDefault, MaxBy, a projected element, an element selector and the result-selector overload; a Where/OrderBy/Take after the read; GroupBy nullable column followed by First() over group returns null record #29240's code; owned types; a precompiled query; and four shapes that keep today's translation, whose baselines show the old SQL.
• 11 of them fail on main, on SQL Server and SQLite, with a wrong NULL group, "Sequence contains no elements" or a NullReferenceException.
• Group_by_json_scalar_Orderby_json_scalar_FirstOrDefault was skipped for JSON/Query: translate groupby followed by orderby json scalar + FirstOrDefault #29287 but passes on main, so I unskipped it.
• SQL baselines change for 28 existing tests, all reads of a group's element.
• SQL Server 2025 functional, 51,149 tests: 3 failed in the batched run and pass when rerun on their own, on this branch and on main. SQLite functional (38,512) and InMemory functional (28,396) pass, as do the unit test projects and EFCore.ApiBaseline.Tests.

---

• I've read the guidelines for contributing and seen the walkthrough
• I've posted a comment on an issue with a detailed description of how I am planning to contribute and got approval from a member of the team
• The code builds and tests pass locally (also verified by our automated build checks)
• Commit messages follow this format
• Tests for the changes have been added (for bug fixes / features)
• Code follows the same patterns and style as existing code in this repo

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[QHarshil]

Reopened to rerun CI. The earlier jobs were cancelled before a runner picked them up.

[Copilot]

🔵 Needs a closer look

It changes mutable core query-translation behavior across many compositional and provider-specific paths, warranting final human review.

0 open findings

🧠 Review effort: Balanced

---

Give feedback about Copilot approvals in this survey to enter a drawing for a $150 gift card.