Codestin Search App

bdice · 2026-09-13T06:20:27Z

Description

FULL join filtering currently splits each rejected candidate independently. With duplicate equality keys, this can emit an unmatched row even when that row has another successful match, or emit the same unmatched row multiple times.

Track successful matches by row identity, preserve every passing pair, and emit exactly one unmatched entry per row with no passing match. AST and JIT filtering share the corrected materialization, and output-size calculation and precomputed-size handling follow the same semantics. Add native regression and stream coverage, and clarify the input-map contract.

Closes #24145. This native fix is needed by the Java FULL filtering coverage in #24097.

Validation

The original implementation failed six direct FULL regression cases; the corrected implementation passed all 12 cases, including duplicate/null/empty inputs, ownership, sizing, both JIT entry points, and host/conditional-FULL oracles.
JOIN_TEST: 850 passed, 2 existing skips. STREAM_JOIN_TEST: 27 passed.
Those runtime checks were run on the original main-based implementation in a CUDA 13.3 conda devcontainer. The seven changed files apply cleanly to release/26.10 and are identical to that validated implementation; release/26.10 runtime validation is left to this PR's CI.

Checklist

I am familiar with the Contributing Guidelines.
New or existing tests cover these changes.
The documentation is up to date with these changes.

coderabbitai · 2026-09-13T06:27:18Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: d98754f5-b799-4984-8834-c54ff9493395

📥 Commits

Reviewing files that changed from the base of the PR and between 742e4fd and e8f9439.

📒 Files selected for processing (7)

cpp/include/cudf/join/join.hpp
cpp/src/join/filter_join_indices/filter_join_indices.cu
cpp/src/join/filter_join_indices/filter_join_indices_jit.cu
cpp/src/join/filter_join_indices/filter_join_indices_output_size_kernel.cuh
cpp/src/join/filter_join_indices/full_join.hpp
cpp/tests/join/mixed_join_tests.cu
cpp/tests/streams/join_test.cpp

Included review availability: Your plan provides up to 12 included reviews per hour; 10 remain after this review.

📝 Summary

Summary by CodeRabbit

Bug Fixes
- Corrected FULL JOIN filtering to preserve matching pairs while retaining unmatched rows from both inputs.
- Improved handling of duplicate keys, nulls, empty inputs, mixed matches, and many-to-many matches.
- Fixed output-size calculations for filtered FULL JOIN results.
- Ensured unmatched rows bypass predicate evaluation and failed candidates do not produce duplicate unmatched rows.
Tests
- Added coverage for conditional and JIT-based FULL JOIN filtering, including post-filtering and unmatched-row materialization.

Walkthrough

Changes

FULL join filtering

Layer / File(s)	Summary
FULL join semantics and output counts `cpp/include/cudf/join/join.hpp`, `cpp/src/join/filter_join_indices/filter_join_indices_output_size_kernel.cuh`	FULL join filtering now retains passing pairs and adds one unmatched pair only for each row without a passing match. Output counts track left and right rows separately.
FULL join materialization and integration `cpp/src/join/filter_join_indices/full_join.hpp`, `cpp/src/join/filter_join_indices/filter_join_indices.cu`, `cpp/src/join/filter_join_indices/filter_join_indices_jit.cu`	A shared helper tracks successful matches, materializes passing pairs, and appends unmatched rows. Regular and JIT filtering paths use the helper.
FULL join regression coverage `cpp/tests/join/mixed_join_tests.cu`, `cpp/tests/streams/join_test.cpp`	Tests cover duplicate keys, mixed matches, nulls, empty maps, many-to-many matches, output sizing, and regular and JIT post-filtering.

Priority: ➖ Normal

Estimated code review effort: 4 (Complex) | ~45 minutes

Change: Bug fix · Severity of issue fixed: Medium

Suggested reviewers: pointkernel

Merge Risk: ⚪ Minimal · up to e8f94

Empty candidate maps intentionally remain empty across filtering and sizing paths; no actionable merge risk remains.

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Linked Issues check	⚠️ Warning	The implementation correctly tracks successful matches by row identity, preserves passing pairs, reconstructs unmatched rows, and routes AST and JIT FULL_JOIN filtering through the shared helper. The …	Preserve empty-map semantics. When the candidate index maps are empty, return empty filtered index vectors and return output size zero with empty counts. Ensure the size-provided, AST, and JIT paths use the same result.
Docstring Coverage	⚠️ Warning	Docstring coverage is 4.55% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 22 functions across 6 files. (1 skipped: 1…	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely identifies the main change: fixing FULL join filtering for duplicate equality keys.
Description check	✅ Passed	The description directly explains the duplicate-key filtering bug, the row-identity fix, affected filtering paths, output-size behavior, tests, and linked issue.
Out of Scope Changes check	✅ Passed	The changes are limited to FULL_JOIN filtering, output-size calculation, shared materialization, related documentation, and regression and stream tests. These changes directly support issue `#24145` obj…

Full details: Linked Issues check

Explanation

The implementation correctly tracks successful matches by row identity, preserves passing pairs, reconstructs unmatched rows, and routes AST and JIT FULL_JOIN filtering through the shared helper. The new tests cover duplicate keys, nulls, empty tables, sizing, ownership, and streams. However, filter_full_join_indices counts every table row as unmatched when left_indices is empty. filter_join_indices_output_size also allocates left-plus-right counts and converts all zero counts to unmatched entries. The added EmptyMapsWithNonemptyTables test expects no output for empty maps, so this requirement is not met.

Full details: Docstring Coverage

Explanation

Docstring coverage is 4.55% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 22 functions across 6 files. (1 skipped: 1 unsupported.)

Fix all pre-merge checks with AI

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

Create stacked PR
Commit on current branch

🧪 Generate unit tests (beta)

Create PR with unit tests

_{Comment @coderabbitai help to get the list of available commands.}

Fix full join filtering with duplicate equality keys

e8f9439

bdice requested a review from a team as a code owner September 13, 2026 06:20

bdice requested review from lamarrr and vuule September 13, 2026 06:20

github-actions Bot assigned bdice Sep 13, 2026

github-actions Bot added the libcudf Affects libcudf (C++/CUDA) code. label Sep 13, 2026

bdice added a commit to bdice/cudf that referenced this pull request Sep 13, 2026

Move native full join fix to standalone PR NVIDIA#24146

c230c2e

bdice mentioned this pull request Sep 13, 2026

Expose filter_join_indices in Java #24097

Open

3 tasks

Navigation Menu

Uh oh!

Fix full join filtering with duplicate equality keys - #24146

Fix full join filtering with duplicate equality keys#24146
bdice wants to merge 1 commit into
NVIDIA:release/26.10from
bdice:fix/full-join-filter-26.10

bdice commented Sep 13, 2026

Uh oh!

coderabbitai Bot commented Sep 13, 2026

Summary by CodeRabbit

Walkthrough

Changes

❌ Failed checks (2 warnings)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

bdice commented Sep 13, 2026

Description

Validation

Checklist

Uh oh!

coderabbitai Bot commented Sep 13, 2026

Summary by CodeRabbit

Walkthrough

Changes

❌ Failed checks (2 warnings)

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant