fix: support ClickHouse C-style ternary operator (? :) (#2436)

[fudianchn]

What

Support the ClickHouse C-style ternary conditional operator cond ? then : else (an alias for if(cond, then, else)), fixing #2436:

SELECT x > 0 ? 'y' : 'n' FROM t
SELECT * FROM t WHERE a OR b ? c : d
SELECT abs(x > 0 ? 1 : 2) FROM t

JSQLParser currently rejects these with ParseException: Encountered: "?".

Why / Root cause

? is already double-booked in the grammar: as a JDBC parameter atom (JdbcParameter) and as the PostgreSQL JSON existence operator (RegularConditionRHS consumes expr ? 'key'). The ternary needs the third reading: ? after a complete boolean expression, closed by a :. Neither existing path can express that, so x > 0 ? 'y' : 'n' fails at the ?.

How

The ternary is integrated into the existing Pratt boolean loop prattExpressionRest, at the same precedence level as OR (C semantics: binds looser than AND/OR, right-associative):

• a pending ? is routed to the ternary via a new isTernaryAhead() scan (same pattern as isUnparenthesizedSelectAhead): it is a ternary when a standalone : closes the then-branch at the same nesting depth before any expression boundary (,, ;, ), EOF, clause keyword). Otherwise the ? keeps its current meaning as the JSON operator — isConditionSuffixAhead()/isComparisonOperatorAhead() decline it only when it is a ternary.
• the : separator collided with the JSON path operator (b : path in PrimaryExpression/JsonExpression); inside a then-branch a top-level : now closes the ternary instead (guarded by a ternaryThenBranchDepth counter, LOOKAHEAD(2, { ... })).

AST: new TernaryExpression(condition, thenExpression, elseExpression) wired into ExpressionVisitor, ExpressionVisitorAdapter, TablesNamesFinder, ExpressionDeParser and ExpressionValidator, following the pattern of RangeExpression.

Scope

Works in every expression context (select items, WHERE, GROUP BY/HAVING, ORDER BY, function arguments, join ON, UPDATE SET, INSERT VALUES, subqueries, nested ternaries in both branches, JDBC parameters as branches). The PostgreSQL JSON operators (?, ?|, ?&, ->, : paths) and JDBC parameters (?, ?5, LIMIT ?) keep parsing unchanged; a JSON ? followed later by a top-level : at depth 0 is inherently ambiguous across dialects and stays a JSON operator only when no : closes it first.

Testing

TernaryExpressionTest (32 tests): the issue case with AST shape assertions, 16 context round-trips (parse + deparse), precedence/associativity AST assertions (a OR b ? c : d groups as (a OR b) ? c : d, a ? b : c ? d : e nests right), and 11 unaffected-syntax guards for JSON operators, JSON paths, JDBC parameters and array ranges.

Local: spotlessApply, checkstyleMain/Test, spotbugsMain, pmdMain clean; full test suite green (4737 tests, 0 failures).

Performance

gradle jmh, JSQLParserBenchmark.parseSQLStatements on performance.sql, version=latest, 10 forks × 10 iterations (100 samples) on a 32-core host:

build	ms/op
master	3.714 ± 0.021
this change	3.730 ± 0.023

The +0.43% delta lies within the 99.9% confidence intervals (the CIs overlap from 3.707 to 3.735): no regression. When no ? is present the only added cost is one string comparison per prattExpressionRest iteration.

Verification of the original issue

SELECT x > 0 ? 'y' : 'n' FROM t now parses and round-trips:

SELECT x > 0 ? 'y' : 'n' FROM t
condition      : x > 0 (GreaterThan)
thenExpression : 'y'
elseExpression : 'n'

[manticore-projects]

Excellent work! Thank your for your contribution!