Fix chained MERGE not seeing sibling MERGE's changes (#1446)

[gregfelice]

Summary

Fixes #1446.

When multiple MERGEs are chained (e.g. MATCH (x) MERGE (x)-[:r]->(:t) MERGE (:C)-[:r]->(:t)), the non-terminal (first) MERGE returned rows one at a time to the parent plan node. The parent MERGE's lateral join materialized its hash table on the first row, before the child MERGE had finished all its iterations. This caused the second MERGE to not see entities created by the first, leading to duplicate nodes.

Root Cause

The execution plan for chained MERGEs nests as:

MERGE2 (terminal)
  └── NestLoop Left Join
        ├── MERGE1 (non-terminal)
        │     └── NestLoop Left Join
        │           ├── child scan (MATCH)
        │           └── lateral check for MERGE1 pattern
        └── lateral check for MERGE2 pattern (Hash Join)

The Hash Join for MERGE2's pattern check is materialized once on the first call. When MERGE1 creates new entities during subsequent iterations, those entities are invisible to the already-materialized hash table.

Fix

Make non-terminal MERGE eager: it processes ALL input rows (creating/matching paths as needed) and buffers the projected results in a List of HeapTuples. Subsequent calls return rows from the buffer one at a time. This ensures all entity creations from the first MERGE are committed before any parent plan node scans the tables.

Changes:

• src/backend/executor/cypher_merge.c: Restructure exec_cypher_merge Case 1 to eagerly buffer results for non-terminal MERGE; terminal MERGE retains existing behavior
• src/include/executor/cypher_utils.h: Add eager_tuples, eager_tuples_index, eager_buffer_filled to cypher_merge_custom_scan_state
• regress/sql/cypher_merge.sql: Add regression tests for First MERGE does not see the second MERGE's changes #1446
• regress/expected/cypher_merge.out: Updated expected output

Reproduction (from issue)

SELECT * FROM create_graph('g');
SELECT * FROM cypher('g', $$ CREATE (:A), (:C) $$) AS (a agtype);
SELECT * FROM cypher('g', $$
    MATCH (x)
    MERGE (x)-[:r]->(:t)
    MERGE (:C)-[:r]->(:t)
    RETURN x
$$) AS (a agtype);

Before fix: Returns 4 rows with duplicate :C nodes created.
After fix: Returns 2 rows (A, C); :C is found and reused by the second MERGE.

Test Plan

• Reporter's exact reproduction case returns correct results
• Verified graph state: A(1), C(1), t(2), 2 edges
• 3-node variant: hub created once and reused across iterations
• All 31 existing regression tests pass

AI Disclosure

AI tools (Claude by Anthropic) were used to assist in developing this fix, including root cause analysis, code changes, and regression tests.

[jrgemignani]

Wrong branch specified

[jrgemignani]

wrong branch

[gregfelice]

Rebased onto master and retarget the PR base branch. Apologies for the wrong branch.

resolved

[Copilot]

Pull request overview

This PR fixes issue #1446 where chained MERGE clauses (e.g., MATCH (x) MERGE ... MERGE ...) failed to see each other's changes, causing duplicate node creation. The root cause was that non-terminal MERGE nodes returned rows one at a time, allowing the parent MERGE's lateral join to materialize its hash table before the child MERGE completed all iterations.

Changes:

• Modified non-terminal MERGE execution to eagerly process all input rows and buffer results before returning any tuples
• Added three new fields to track eager buffering state: eager_tuples, eager_tuples_index, and eager_buffer_filled
• Added comprehensive regression tests reproducing the reported issue and verifying the fix

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
src/include/executor/cypher_utils.h	Added three fields to cypher_merge_custom_scan_state for eager tuple buffering
src/backend/executor/cypher_merge.c	Restructured Case 1 execution logic to eagerly buffer non-terminal MERGE results; added cleanup for eager buffer in end_cypher_merge
regress/sql/cypher_merge.sql	Added regression tests for issue #1446 covering 2-node and 3-node scenarios
regress/expected/cypher_merge.out	Updated expected output with correct results showing nodes are reused properly

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[jrgemignani]

@gregfelice See Copilot review above, plz. Otherwise, looks good.