fix(ddl): ALGORITHM/LOCK clause no longer discards subsequent ALTER operations (#1140)

[minguyen9988]

Problem

MySqlDDLParserListenerImpl.enterAlterTable walks the ALTER clause list and emits ClickHouse DDL per clause. On encountering AlterBySetAlgorithmContext it logged a message and break-ed out of the walk.

ALGORITHM= is a MySQL-only execution hint with no ClickHouse equivalent, so emitting nothing for it is correct — but abandoning the whole walk is not. Every operation positioned after the hint was silently discarded.

The statement from #1140:

ALTER TABLE test_lot ADD COLUMN event_ref_type_id INTEGER, algorithm=instant,
                     ADD COLUMN event_ref_id BIGINT, algorithm=instant;

generated only:

ALTER TABLE db.test_lot ADD COLUMN event_ref_type_id Nullable(Int32)

event_ref_id is never created in ClickHouse.

Why this is worse than a DDL gap

The failure is silent on every side. MySQL applies both columns, the connector raises no error, and the destination table simply lacks the column. Because the INSERT column list is built by iterating the ClickHouse column map, every subsequent row silently drops that column's value. There is no error, no warning, and no signal that the two schemas have diverged — the data loss is discovered later, by a checksum job or by a user noticing a blank column.

LOCK= is the same class of MySQL-only hint. It was falling through to the no-op terminal branch, leaving the separator that preceded it dangling in the generated statement — e.g. ... Nullable(Int32),, ADD COLUMN ..., which ClickHouse rejects outright.

Fix

Both hints now skip their own clause and continue the walk, dropping the now-orphaned separator via a small removeTrailingComma() helper.

A hint in trailing position is handled too: the comma sweep runs once more after the loop, so ..., ALGORITHM=INPLACE, LOCK=NONE leaves no dangling separator.

The existing behaviour for a trailing ALGORITHM clause is unchanged — that case already worked, and its existing tests still pass.

Tests

Three regression cases added to MySqlDDLParserListenerImplTest:

Test	Covers
testAlterAddColumnWithInterleavedAlgorithmClauses	the exact statement from #1140
testAlterAddColumnWithInterleavedLockClause	LOCK= in non-trailing position
testAlterAddColumnWithTrailingHintsLeavesNoDanglingComma	trailing hints emit no dangling separator

Verified failing before the change and passing after — not merely passing after:

WITHOUT the fix:  Tests run: 3, Failures: 2
  testAlterAddColumnWithInterleavedAlgorithmClauses
    expected: <...e_id nullable(int32)[, add column event_ref_id nullable(int64)]>
    but was:  <...e_id nullable(int32)[]>
  testAlterAddColumnWithInterleavedLockClause
    expected: <...col nullable(int32),[] add column second_c...>
    but was:  <...col nullable(int32),[,] add column second_c...>

WITH the fix (full class):  Tests run: 108, Failures: 0, Errors: 0, Skipped: 1

The full MySqlDDLParserListenerImplTest class passes, so no existing parser behaviour regressed.

How this was found

End-to-end verification of the 2.10.0 rollup against a MySQL → connector → ClickHouse test environment, comparing values (not row counts) across a DDL/DML matrix. This defect reproduced identically on 2.8.0 and 2.10.0.

Fixes #1140