fix: collect tables from all SELECT and DML trailing clauses in TablesNamesFinder

[fudianchn]

AI disclosure: this change was prepared with AI coding agents, reviewed and revised line by line by me.

What

TablesNamesFinder now traverses all Expression- and Table-bearing clauses of the SELECT family and the DML trailing clauses it previously skipped, so tables located in those clauses are collected instead of silently dropped.

Why

TablesNamesFinder is the public helper for table extraction (SQL analysis, auditing, firewalls); a silently incomplete result is the worst failure mode for those consumers. All shapes in #2516 parse fine today and return incomplete table sets. This continues the family fixed by #2478 / #2479 (piped queries, DML side clauses, analytic clauses).

How

• PlainSelect: DISTINCT ON items, INTO tables, LATERAL VIEW generator functions, PREFERRING, GROUP BY (incl. GROUPING SETS), QUALIFY, WINDOW definitions, PIVOT / UNPIVOT, ORDER BY, LIMIT / LIMIT BY, OFFSET, FETCH, SETTINGS, INTO TEMP target.
• ParenthesedSelect, SetOperationList, Values: the ORDER BY / LIMIT / OFFSET / FETCH clauses they carry themselves.
• Delete, Update: their ORDER BY and LIMIT clauses (MySQL single-table DML).
• Attached PIVOT / UNPIVOT / PIVOT XML (incl. the PIVOT XML ... IN (SELECT ...) subquery) is reached via the newly implemented PivotVisitor, the same mechanism as MergeOperationVisitor / PipeOperationVisitor in fix: complete TablesNamesFinder traversal for piped queries, DML side clauses and analytic functions (#2478) #2479.

All traversals reuse the ExpressionVisitor / FromItemVisitor default helpers (visitOrderBy, visitLimit, visit(GroupByElement), visitUpdateSets, visitPreferringClause, visitTables, visitFromItem) already used by SelectVisitorAdapter and StatementVisitorAdapter. Two SelectVisitorAdapter @todos (windowDefinitions, lateralViews) are covered too, because both can carry subqueries.

Root cause

The finder's visit methods traversed only a subset of the Expression- and Table-bearing children; the adapters already covered most of these clauses, the finder did not.

Testing

• 21 new tests in TablesNamesFinderTest, one per shape from [BUG] JSQLParser 5.4-SNAPSHOT : all RDBMS : TablesNamesFinder silently misses tables in many SELECT and DML clauses (INTO, GROUP BY, QUALIFY, WINDOW, ORDER BY, LIMIT, DISTINCT ON, LATERAL VIEW, PIVOT XML, SETTINGS) #2516: each failed on master before the change and passes with it. Two mutants verified (removing the INTO traversal and the Table pivot traversal makes their tests fail).
• Local run: gradlew test --tests TablesNamesFinderTest -> 106 passed.
• Not covered on purpose: clauses that cannot hold tables (MySQL INTO OUTFILE/DUMPFILE fields, TOP, Oracle hint, procedure analyse, KSQL window, OPTION clause).

Verification of the original issue

TablesNamesFinder.findTables("SELECT a FROM t GROUP BY a, (SELECT b FROM u)");
// before: [t]    after: [t, u]

With this change every shape in the #2516 matrix returns the complete table set.

Fixes #2516

[manticore-projects]

Awesome job!

[fudianchn]

After your comment I pushed two more updates to the branch:

1. Self-review found the same gap in the DELETE / UPDATE ORDER BY + LIMIT clauses (StatementVisitorAdapter already traverses them), now covered with 2 more tests; [BUG] JSQLParser 5.4-SNAPSHOT : all RDBMS : TablesNamesFinder silently misses tables in many SELECT and DML clauses (INTO, GROUP BY, QUALIFY, WINDOW, ORDER BY, LIMIT, DISTINCT ON, LATERAL VIEW, PIVOT XML, SETTINGS) #2516 updated accordingly.
2. Fixed the commit author email: the earlier pushes were attributed to an unrelated GitHub account by mistake, the current head 002ebc6 is correct.

Full suite: 4971 tests, 0 failures.

[manticore-projects]

Thank you for fixing this!