split() treats the delimiter as a regular expression while replace() treats it literally

[hkjiang26]

AGE Version: extension 1.8.0 on PostgreSQL 18.6
Installation Method: Docker
API: Cypher and direct SQL

Steps to reproduce (paste into psql as-is)

LOAD 'age';
SET search_path = ag_catalog, public;

-- Cypher path (re-run: DROP GRAPH xdb CASCADE first)
SELECT * FROM create_graph('xdb');
SELECT * FROM cypher('xdb', $$ RETURN split('a.b.c', '.') $$) AS (r agtype);

-- SQL path (same C function, no graph needed)
SELECT age_split('a.b.c', '.');        -- -> ["", "", "", "", "", ""]
SELECT age_split('a1b1c', '1');        -- -> ["a", "b", "c"]   (plain delimiter: works)
SELECT age_replace('a.b.c', '.', '-'); -- -> "a-b-c"
SELECT age_split('a.b.c', '\.');      -- -> ["a", "b", "c"]   (only because it is regex-escaped)
SELECT age_split('a|b', '|');          -- -> ["a", "|", "b"]
SELECT age_split('a+b+c', '+');        -- -> ERROR: invalid regular expression: quantifier operand invalid

Expected behavior

openCypher defines the split delimiter as a literal string: split('a.b.c', '.') → ['a','b','c']. Neo4j, Memgraph and FalkorDB all return this without any escaping, and replace('a.b.c', '.', '-') → 'a-b-c' in AGE too — so both functions must interpret '.' identically.

Actual behavior

• split('a.b.c', '.') → ["","","","","",""] (six empty strings): the delimiter is compiled as a regular expression, so '.' matches every character.
• split('a1b1c', '1') → ["a","b","c"]: plain delimiters work, so the regex interpretation only surfaces for metacharacters — exactly where queries silently break.
• split('a.b.c', '\.') → ["a","b","c"]: correct only because the user manually regex-escapes the dot.
• split('a|b', '|') → ["a","|","b"]: the empty-alternation regex splits at every position.
• split('a+b+c', '+') → ERROR: invalid regular expression: quantifier operand invalid.
• replace('a.b.c', '.', '-') → "a-b-c": the same engine treats the same delimiter literally.

The engine is internally inconsistent, deviates from openCypher and all sibling engines, and silently corrupts results for common literal delimiters ('.', '|', '+', ...).

Root cause

age_split() (src/backend/utils/adt/agtype.c) passes the delimiter verbatim to PostgreSQL's regexp_split_to_array(), which compiles it as a regular expression; age_replace() performs a literal replacement. Cypher split()/replace() are rewritten to these same SQL functions, so both APIs are affected.