jdbc-v2: ANTLR4 parser backends report a source table as the INSERT target, writing rows into the wrong table

jdbc-v2/src/main/java/com/clickhouse/jdbc/internal/SqlParserFacade.java から始め、特に ParsedPreparedStatementListener.enterInsertStmt と enterTableExprIdentifier を確認してから、issue にある parser レベルの例を両方の ANTLR4 backend で実行します。source table、CTE、または scalar subquery が続く場合でも INSERT のターゲットが dst または db1.dst のままであり、SELECT 文と table-function の処理が現在の動作を維持していれば完了です。

Description

With either ANTLR4 parser backend (jdbc_sql_parser=ANTLR4 or ANTLR4_PARAMS_PARSER),
ParsedPreparedStatement.getTable() of an INSERT returns the last table identifier that appears anywhere in the
statement instead of the insert target. Any table read by the statement - a CTE name, a FROM table, a table in a
scalar subquery inside the VALUES list - replaces the target. Parsing reports no errors, so the wrong target is
silent. The default JAVACC backend is correct in every case below.

Observed on main (91ec4d3) with server 26.8.2.7:

SQL	JAVACC	ANTLR4 / ANTLR4_PARAMS_PARSER
INSERT INTO dst WITH r AS (SELECT 1 AS n) SELECT n FROM r	dst	r (CTE name)
INSERT INTO dst SELECT * FROM src	dst	src
INSERT INTO db1.dst SELECT * FROM db2.src	db1.dst	db2.src (database also overwritten)
INSERT INTO dst WITH r AS (SELECT 1 AS n) SELECT n FROM r JOIN other USING (n)	dst	other
INSERT INTO dst VALUES (?, (SELECT max(x) FROM src))	dst	src
INSERT INTO dst (a) VALUES (?)	dst	dst (correct - no source table)
INSERT INTO dst SELECT 1	dst	dst (correct - no source table)

The last row of the table is the harmful one. ConnectionImpl#prepareStatement uses getTable() to resolve the
schema for the beta RowBinary writer, and its guard (!isInsertWithSelect() && getAssignValuesGroups() == 1 && !isUseFunction()) does not exclude a VALUES list holding a scalar subquery. So with
beta.row_binary_for_simple_insert=true the writer is built against the source table and the row is inserted
into it. executeUpdate() returns normally and the target table stays empty - silent data loss plus a write into a
table the statement only reads.

Both non-default options are needed for the wrong write (jdbc_sql_parser=ANTLR4* plus the beta writer). The wrong
table name itself is returned by the parser regardless of the writer setting.

This is not #3083 (that one is the default JAVACC backend and is about values being shifted within the correct
target table) and not #3015 / #3027 (table functions and unparsable value expressions).

Steps to reproduce

1. CREATE TABLE src (x Int32) ENGINE=Memory; CREATE TABLE dst (a Int32, b Int32) ENGINE=Memory;
INSERT INTO src VALUES (7),(9);
2. Open a connection with jdbc_sql_parser=ANTLR4 and beta.row_binary_for_simple_insert=true.
3. prepareStatement("INSERT INTO dst VALUES (?, (SELECT max(x) FROM src))"), setInt(1, 42), executeUpdate().
4. SELECT * FROM dst and SELECT * FROM src.

Error Log or Exception StackTrace

No error. executeUpdate() reports success.

### parser=JAVACC
  stmt class = WriterStatementImpl
  EXCEPTION: java.sql.SQLException: java.lang.IllegalArgumentException: An attempt to write null into not nullable column 'b'
### parser=ANTLR4
  stmt class = WriterStatementImpl
  executeUpdate OK
### parser=ANTLR4_PARAMS_PARSER
  stmt class = WriterStatementImpl
  executeUpdate OK

Table contents after the three runs - dst is empty, src holds the two rows that the ANTLR4 runs wrote:

-- dst:
-- src:
7
9
42
42

(The JAVACC line is the separate, already reported #3083 behaviour: the writer is chosen for a values list that is
not placeholders only. It at least targets the correct table.)

Expected Behaviour

getTable() of an INSERT is the insert target, for every backend - dst in all rows of the table above, and
db1.dst for the qualified case. The server accepts all of these statements and writes into the target only, e.g.

$ curl --data-binary "INSERT INTO dst VALUES (1, (SELECT max(x) FROM src))" http://server:8123/
$ curl --data-binary "SELECT * FROM dst FORMAT TSV" http://server:8123/
1	9
$ curl --data-binary "INSERT INTO dst WITH r AS (SELECT 1 AS n) SELECT n, n FROM r" http://server:8123/
$ curl --data-binary "SELECT * FROM dst ORDER BY a FORMAT TSV" http://server:8123/
1	9
1	1

Root cause

jdbc-v2/src/main/java/com/clickhouse/jdbc/internal/SqlParserFacade.java

• ParsedPreparedStatementListener.enterTableExprIdentifier (line 441) calls
  extractAndSetDatabaseAndTable for every tableExprIdentifier in the tree - that rule matches the tables a
  query reads, not the insert target.
• enterInsertStmt (line 448) sets the target correctly, but insertStmt is entered before its nested
  tableExprIdentifier nodes, so each source table seen later overwrites the target through the shared
  parsedStatement.setTable / setDatabase (line 488).
• INSERT INTO dst SELECT 1 and a plain VALUES insert keep the correct name only because no tableExprIdentifier
  follows.

The JavaCC backend keeps the target because it assigns the table from the insert production only.

Suggested fix

Make the insert target win over source tables in the ANTLR4 listener - for example have
enterTableExprIdentifier skip assignment once an insert target has been recorded (or record source tables
separately from table/database), so enterInsertStmt remains authoritative for an INSERT.

Contrast cases that must keep their current behaviour:

• SELECT * FROM src -> src, and WITH r AS (SELECT 1 AS n) SELECT n FROM r -> r. For a statement that is not an
  insert, tableExprIdentifier is the only source of the name and both backends agree today.
• INSERT INTO [TABLE] FUNCTION f(...) must stay useFunction=true and off the writer path (#3015 / #3016).
• INSERT INTO db1.dst SELECT ... must report database db1, not db2.

Separately, ConnectionImpl#prepareStatement's writer guard could reject a VALUES list that contains a subquery;
that part is the same missing-guard family as #3083.

Code Example

Properties p = new Properties();
p.setProperty("jdbc_sql_parser", "ANTLR4");                        // or ANTLR4_PARAMS_PARSER
p.setProperty("beta.row_binary_for_simple_insert", "true");

try (Connection c = DriverManager.getConnection(url, p);
     PreparedStatement ps = c.prepareStatement("INSERT INTO dst VALUES (?, (SELECT max(x) FROM src))")) {
    ps.setInt(1, 42);
    ps.executeUpdate();          // succeeds; 42 lands in src, dst stays empty
}

Parser level, no server needed:

SqlParserFacade parser = SqlParserFacade.getParser("ANTLR4",
        new JdbcConfiguration("jdbc:ch:http://localhost:8123", new Properties()));
ParsedPreparedStatement s = parser.parsePreparedStatement(
        "INSERT INTO dst WITH r AS (SELECT 1 AS n) SELECT n FROM r");
assert s.isInsert();
assert !s.isHasErrors();
assert "dst".equals(s.getTable());   // fails: returns "r"

Configuration

Client Configuration

jdbc_sql_parser = ANTLR4            // or ANTLR4_PARAMS_PARSER; JAVACC (default) is unaffected
beta.row_binary_for_simple_insert = true   // only needed for the wrong write, not for the wrong name

Environment

• Cloud
• Client version: 0.11.0-rc1 (main, 91ec4d3)
• Language version: OpenJDK 17.0.18
• OS: Ubuntu 24.04 (container)

ClickHouse Server

• ClickHouse Server version: 26.8.2.7
• ClickHouse Server non-default settings, if any: none
• CREATE TABLE statements for tables involved:

CREATE TABLE src (x Int32) ENGINE = Memory;
CREATE TABLE dst (a Int32, b Int32) ENGINE = Memory;

• Sample data: INSERT INTO src VALUES (7),(9);

---

Found by automated analysis of this client while working on #3122 / #3128 (WITH RECURSIVE grammar gap), then
verified end to end against a live 26.8.2.7 server rather than by code inspection.