fix: support SUBPARTITION TEMPLATE for OceanBase composite partitioning

[daguimu]

Problem

OceanBase CREATE TABLE with SUBPARTITION TEMPLATE fails with:

syntax error, expect OPTIONS, actual IDENTIFIER TEMPLATE

Example SQL (from OceanBase V4.2.2 documentation):

CREATE TABLE t2_m_lr(col1 INT, col2 INT)
PARTITION BY LIST (col1)
SUBPARTITION BY RANGE(col2)
SUBPARTITION TEMPLATE
 (SUBPARTITION mp0 VALUES LESS THAN(100),
  SUBPARTITION mp1 VALUES LESS THAN(200),
  SUBPARTITION mp2 VALUES LESS THAN(300))
 (PARTITION p0 VALUES IN(1,3),
  PARTITION p1 VALUES IN(4,6),
  PARTITION p2 VALUES IN(7,9));

Root cause: The MySQL parser (used by OceanBase) only recognized SUBPARTITION OPTIONS after the subpartition-by clause. When encountering SUBPARTITION TEMPLATE, it tried to match OPTIONS and failed.

Fix

Add SUBPARTITION TEMPLATE parsing to MySqlCreateTableParser. When the parser sees TEMPLATE after SUBPARTITION, it parses the template subpartition list (each containing SUBPARTITION name VALUES ...), storing them in the existing SQLSubPartitionBy.subPartitionTemplate field.

The SUBPARTITION OPTIONS path remains unchanged for backward compatibility.

Tests Added

4 test cases in Issue6160.java:

• test_subpartition_template_list_range — exact SQL from issue (LIST + RANGE template)
• test_subpartition_template_range_hash — RANGE + HASH template
• test_subpartition_template_with_values_in — RANGE + LIST template with VALUES IN
• test_subpartition_options_still_works — regression: SUBPARTITIONS count still works

All 6 existing OceanBase tests pass.

Fixes #6160

[wenshao]

[Suggestion] Template subpartitions are manually constructed with only name + VALUES, skipping per-subpartition properties (TABLESPACE, COMMENT, ENGINE, DATA DIRECTORY, etc.) that this.exprParser.parseSubPartition() handles. The Oracle parser's equivalent code correctly calls parseSubPartition().

Suggested fix: Replace the manual construction with a call to the existing parser method:

acceptIdentifier("SUBPARTITION");
SQLSubPartition subPartition = this.exprParser.parseSubPartition();
SQLPartitionValue values = this.exprParser.parsePartitionValues();
if (values != null) {
    subPartition.setValues(values);
}
subPartition.setParent(subPartitionByClause);
subPartitionByClause.getSubPartitionTemplate().add(subPartition);

[wenshao]

[Suggestion] All 4 test cases assert stmtList.size() == 1 but never verify subPartitionTemplate contents (names, values, count). Neighboring tests (Issue5078, Issue6102) use round-trip verification via SQLParseAssertUtil.assertParseSql().

Suggested fix: Add AST-level assertions (after fixing output visitors for round-trip support):

SQLCreateTableStatement stmt = (SQLCreateTableStatement) stmtList.get(0);
SQLSubPartitionBy spBy = stmt.getPartitionBy().getSubPartitionBy();
List<SQLSubPartition> template = spBy.getSubPartitionTemplate();
assertEquals(3, template.size());
assertEquals("mp0", template.get(0).getName().getSimpleName());

Or use SQLParseAssertUtil.assertParseSql(sql, dbType) for round-trip validation once output visitors are fixed.

[wenshao]

Review: fix: support SUBPARTITION TEMPLATE for OceanBase composite partitioning

Parsing logic is correct and all tests pass (49 OceanBase test classes, 0 failures). The PR successfully adds SUBPARTITION TEMPLATE parsing. However, the parsed template data has limited end-to-end functionality due to gaps in the existing AST infrastructure that this PR exposes.

Findings

1. accept0() in all SQLSubPartitionBy subclasses skips subPartitionTemplate [Suggestion]
The root cause: none of the accept0() implementations in SQLSubPartitionByHash, SQLSubPartitionByRange, SQLSubPartitionByList, MySqlSubPartitionByKey, MySqlSubPartitionByList, or MySqlSubPartitionByValue visit subPartitionTemplate. This makes template subpartitions invisible to ALL visitors — schema analysis, validation, statistics, and custom AST traversal. Fix: add acceptChild(visitor, subPartitionTemplate) to each accept0().

2. Output visitors silently drop SUBPARTITION TEMPLATE for all sub-partition types [Suggestion]
Only SQLASTOutputVisitor.visit(SQLSubPartitionByList) renders the template. visit(SQLSubPartitionByRange), visit(SQLSubPartitionByHash), and the MySQL-specific overrides (MySqlSubPartitionByList, MySqlSubPartitionByKey) all silently drop it. This means round-trip parse→output produces incorrect DDL that is missing the template. Fix: add template output blocks mirroring the existing visit(SQLSubPartitionByList) pattern.

3. Manual SQLSubPartition construction skips properties [Suggestion] (inline comment posted)
The new code manually constructs SQLSubPartition with only name + VALUES, while this.exprParser.parseSubPartition() handles TABLESPACE, COMMENT, ENGINE, etc. The Oracle parser's equivalent code calls parseSubPartition().

4. Tests only verify parse success, not AST structure [Suggestion] (inline comment posted)
All 4 test cases assert stmtList.size() == 1 without checking subPartitionTemplate contents. Neighboring tests use round-trip verification.

5. clone() for SQLSubPartitionByHash/SQLSubPartitionByRange loses template [Possibly, needs human review]
These clone() methods don't call super.cloneTo(), so cloning an AST with template subpartitions silently drops them.

Verdict: Comment

The parsing is correct for the targeted use cases. Findings 1-2 are pre-existing infrastructure gaps exposed by this PR — they should be addressed before or shortly after merge for end-to-end functionality. Finding 3 is a forward-compatibility improvement. Finding 4 strengthens test coverage.

Reviewed by glm-5.1 via Qwen Code /review