[BUG] 5.4-SNAPSHOT : ClickHouse : asterisk column modifiers fail or mis-parse (APPLY as alias, docs chain, STRICT, EXCEPT regex)

AI disclosure: this issue was prepared with AI coding agents, revised and verified by me.

Failing SQL Feature:

ClickHouse column modifiers on * / t.*: the documented APPLY modifier is silently mis-parsed as a select-item alias, the documented modifier chain from the SELECT docs fails with a ParseException, and the STRICT variants plus the regex-string form of EXCEPT (implemented in the ClickHouse grammar, covered by its stateless tests) are unsupported.

SQL Example:

-- 1) documented in the ClickHouse SELECT docs, mis-parsed on master:
--    parses as AllColumns with alias "APPLY(sum)"
SELECT * APPLY(sum) FROM t;

-- 2) the modifier chain exactly as shown in the ClickHouse docs, fails:
SELECT * REPLACE(i + 1 AS i) EXCEPT (j) APPLY(sum) FROM columns_transformers;

-- 3) STRICT variants (undocumented, but in the ClickHouse grammar and
--    its stateless tests) and the regex form of EXCEPT, all fail:
SELECT * EXCEPT STRICT (a, b) FROM t;
SELECT * REPLACE STRICT (1 AS a, 2 AS b) FROM t;
SELECT * EXCEPT '^tmp_' FROM t;

Baseline that already parses: SELECT * EXCEPT (a, b) FROM t, SELECT * REPLACE (a AS b) FROM t, SELECT * EXCLUDE (a) FROM t.

Software Information:

JSqlParser version: current master 5.4-SNAPSHOT, commit 02b9d94 (tested 2026-09-14)
Database: ClickHouse

Additional context

ClickHouse grammar accepts STRICT after both EXCEPT and REPLACE, and a string-literal regex pattern for EXCEPT (ExpressionElementParsers.cpp: is_strict, ParserStringLiteral -> ASTColumnsExceptTransformer::setPattern, formatted as EXCEPT STRICT in ASTColumnsTransformers.cpp); covered by the stateless tests 01470_columns_transformers.sql and 02343_analyzer_column_transformers_strict.sql.
The SELECT modifiers documentation lists only APPLY, EXCEPT and REPLACE; the STRICT and regex forms are undocumented.
References: ExpressionElementParsers.cpp, ASTColumnsTransformers.cpp
Related: PR feat: support ClickHouse COLUMNS(...) APPLY/EXCEPT/REPLACE transformers (#2631) #2635 adds the COLUMNS('regexp') context for the documented modifiers; the asterisk position is untouched there, so this issue is independent of it.

I'd like to take this one.

Proposed scope — only the first two items, to keep the patch reviewable:

SELECT * APPLY(sum): parse as AllColumns + ColumnsTransformer instead of falling into the alias path (APPLY is in NonReservedWord(), so it currently becomes the alias "APPLY(sum)").
The modifier chain from the ClickHouse docs (* EXCEPT (a) REPLACE (b AS c)).

I'd leave STRICT and the regex form of EXCEPT for a follow-up — they look like separate grammar work, and I'd rather not bundle four behaviours into one diff.

Approach: extend the AllColumns(boolean) production in JSqlParserCC.jjt (~L7946) to accept a repeated ColumnsTransformer(), reuse the existing ColumnsExpression / ColumnsTransformer classes, then update AllColumns, the ExpressionVisitor / DeParser pair, and add round-trip tests via assertSqlCanBeParsedAndDeparsed().

Heads-up: #2652 also touches AllColumns.java — happy to rebase after it lands, or hold off if you'd prefer.

AI disclosure: I use an AI coding assistant; I review and test everything I submit and take responsibility for it.

Follow-up on the approach — one thing I'd like to get right before writing code.

AllColumns currently models these modifiers itself (exceptColumns / replaceExpressions / exceptKeyword), while ColumnsTransformer (from #2635) models the same keywords for the COLUMNS(...) path. If I give AllColumns a List<ColumnsTransformer>, the same syntax ends up with two representations.

Which would you prefer?

Unify on ColumnsTransformer — AllColumns holds a List<ColumnsTransformer> and the three legacy fields go away. Consistent with ColumnsExpression, and chains like * EXCEPT (a) REPLACE (b AS c) fall out of the repeated-transformer grammar for free. Cost: it's a breaking API change, and AllTableColumns extends AllColumns so its constructors change too. Internal usage looks small — AllTableColumns reads getExceptKeyword(), plus the deparser and one test.
Keep the legacy fields, add transformers only for the new cases — non-breaking, but two representations for EXCEPT / REPLACE.

I lean (1) since 5.4 is still a snapshot, but I don't want to churn public API without a nod.

One more question: should * APPLY(...) and table.* APPLY(...) both be supported? The issue only shows the plain * form.

Navigation Menu

Repository navigation

[BUG] 5.4-SNAPSHOT : ClickHouse : asterisk column modifiers fail or mis-parse (APPLY as alias, docs chain, STRICT, EXCEPT regex) #2636

Failing SQL Feature:

SQL Example:

Software Information:

Additional context

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

[BUG] 5.4-SNAPSHOT : ClickHouse : asterisk column modifiers fail or mis-parse (APPLY as alias, docs chain, STRICT, EXCEPT regex) #2636

Description

Failing SQL Feature:

SQL Example:

Software Information:

Additional context

Activity

kyy-logs commented on Sep 19, 2026

kyy-logs commented on Sep 19, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions