test: add sqllogictest coverage for DISTINCT / GROUP BY / aggregation on map columns#23406
Merged
Merged
Conversation
Jefffrey
approved these changes
Jul 10, 2026
This was referenced Jul 11, 2026
Contributor
|
thanks @PG1204 |
adriangb
pushed a commit
to pydantic/datafusion
that referenced
this pull request
Jul 15, 2026
…S lists (apache#23521) ## Which issue does this PR close? - Closes apache#23474. ## Rationale for this change A bare `MAP {'k1': NULL}` literal infers a `Null` value type, producing `Map("entries": Struct("key": Utf8, "value": Null))`. Placing it in a VALUES list alongside concretely-typed map rows failed with: Error during planning: Inconsistent data type across values list at row 1 column 0. This is because `type_union_resolution_coercion` (the pairwise worker behind `type_union_resolution`, which the VALUES builder uses to widen row types) had recursion arms for `Dictionary`, `RunEndEncoded`, `Struct`, and `List` (via `list_coercion`), but none for `Map`. Scalar NULL inference across a VALUES list (`VALUES (1), (NULL)`) already worked; the same inference just never reached into a map's value type. This gap was noted in apache#23406, where the tests had to use a `CAST(NULL AS BIGINT)` workaround. ## What changes are included in this PR? - A one-line fix in `datafusion/expr-common/src/type_coercion/binary.rs`: wire the existing `map_coercion` helper (already used by `comparison_coercion` and `type_union_coercion`) into `type_union_resolution_coercion`'s fallback chain. The recursion into the map's key/value types then applies the exact same rules as scalar VALUES coercion (`null_coercion` for NULL inference, `binary_numeric_coercion` for numeric widening), symmetrically in either row order. - A unit test (`test_type_union_resolution_map`) covering Null + Int64 value types in both orders, Int64 + Null + Float64 widening, and Map vs. Struct still refusing to unify. - A new "map NULL value coercion in VALUES" section in `datafusion/sqllogictest/test_files/map.slt` covering: the issue reproducer, the reversed row order, all-NULL values (resulting type `Map("entries": Struct("key": Utf8, "value": Null))`, verified via `arrow_typeof`), multi-key rows with one NULL value, NULL scattered across three rows, numeric widening to Float64, two-level nested maps, NULL round-tripping as NULL through `SELECT`, and `INSERT INTO ... VALUES`. - Incompatible concrete value types still error: `(MAP {'k': 1}), (MAP {'k': 'hello'})` follows the scalar VALUES rule (coerce to the numeric type, then fail with `Cast error: Cannot cast string 'hello' to value of Int64 type`, which is the same behavior as `VALUES (1), ('hello')`). - A note above the apache#23406 tests documenting that the `CAST(NULL AS BIGINT)` workaround is no longer required (those tests are left unchanged). ## Are these changes tested? Yes. New unit test in `datafusion-expr-common` and new sqllogictests in `map.slt` as described above. The full sqllogictest suite (492 files) passes locally with zero regressions, as do `cargo test -p datafusion-expr -p datafusion-expr-common`, `cargo fmt`, and `cargo clippy --workspace --all-targets -- -D warnings`. ## Are there any user-facing changes? `VALUES` lists (and other contexts that use `type_union_resolution`, such as `CASE`/`COALESCE`/array literals) now unify Map types whose key/value types are coercible, including bare NULL map values. No API changes; previously failing queries now succeed, and incompatible-type errors are preserved. --------- Co-authored-by: Jeffrey Vo <jeffrey.vo.australia@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Which issue does this PR close?
Rationale for this change
SELECT DISTINCTon map columns used to fail withArrowError(NotYetImplemented)because arrow-rs'sRowConverterdid not support Map types. That support has since landed upstream (apache/arrow-rs#7879) and is available on DataFusion main via arrow 59.1, so these queries now work. As suggested in #15428 (comment) (maintainer comment), this PR adds sqllogictest coverage to lock in the behavior and close the issue.What changes are included in this PR?
Adds a new test section to
datafusion/sqllogictest/test_files/map.sltcovering:SELECT DISTINCTon a map column, including the exact reproducer from the issue (DISTINCT ... LIMIT), collapsing of duplicate maps and duplicate NULLs, andDISTINCTover map + scalar columns togetherGROUP BYon a map column,GROUP BYmap + scalar keys, andHAVINGwith a map grouping keyCOUNT(map),COUNT(DISTINCT map),SUMgrouped by map, andarray_aggof map valuesDISTINCT(equal to each other, distinct fromNULL), NULL maps, and unsorted maps with the same entries in different order being treated as distinct valuesUNION(distinct) on map columnsDISTINCT/GROUP BYover the existing 209-rowparquet_map.parquetfile for coverage on real Parquet-backed map dataNote: a bare
MAP {'key': NULL}literal infers aNullvalue type and fails type coercion across a VALUES list againstMap(Utf8, Int64)rows, so the test usesCAST(NULL AS BIGINT)for the null map value. That coercion gap may be worth a separate follow-up issue.Are these changes tested?
Yes, this PR is test-only. All tests pass locally via
cargo test -p datafusion-sqllogictest --test sqllogictests -- map.slt.Are there any user-facing changes?
No. Test-only change; no API or behavior changes.