fix: add SYSTEM_USER to PostgreSQL unparenthesized function names [FaaFyfxR9WAQrL7FcAgEHJvztd8cVMxvjHRS55rw1nwH]

[waterWang]

Fixes #39384

Changes

Added SYSTEM_USER to PostgreSQLFunctionOption.UNPARENTHESIZED_FUNCTION_NAMES so that the PostgreSQL dialect correctly binds the niladic function SYSTEM_USER instead of misclassifying it as a column reference.

Root Cause

PostgreSQLFunctionOption (line 31-33) lists 11 unparenthesized function names and omits SYSTEM_USER. ColumnSegmentBinder.isUnparenthesizedFunction() reads this set via .contains() to decide function vs column. When the lookup fails, bind() throws ColumnNotFoundException.

MySQLFunctionOption and SQLServerFunctionOption already include SYSTEM_USER.

Verification

• SELECT SYSTEM_USER against a PostgreSQL-typed schema now correctly binds as a niladic function
• No grammar change required (SYSTEM_USER already lexes as an identifier in PostgreSQL)
• openGauss inherits this change via OpenGaussDatabaseMetaData which returns new PostgreSQLFunctionOption()

Signed-off-by: waterWang waterWang@users.noreply.github.com

[terrymanu]

Result

Review Result: Not Mergeable

Feedback Mode: Needs Discussion

Blocking Issues: 1

The patch fixes the narrow no-table reproducer, but does not model PostgreSQL 16’s SYSTEM_USER semantics correctly.

Blocking Issues

1. SYSTEM_USER remains an ordinary column expression

Evidence: PostgreSQL 16 defines SYSTEM_USER as a reserved keyword and its grammar constructs a system_user function call; the official manual documents the unparenthesized expression. At the reviewed head, PostgreSQL’s reservedKeyword and functionExprCommonSubexpr still omit SYSTEM_USER. Exact-head-equivalent parser verification consequently produced ColumnProjectionSegment for SYSTEM_USER, versus ExpressionProjectionSegment for CURRENT_USER.

The added metadata entry is only consulted after column resolution in ColumnSegmentBinder. Therefore, when table metadata contains a quoted system_user column, unquoted SELECT SYSTEM_USER FROM t is bound as that physical column even though PostgreSQL 16 evaluates the reserved function. The added membership assertion cannot detect this path.

Impact: Column-bound processing can treat the function result as table data—for example, MaskMergedResult consumes that bound table/column identity. This leaves issue #39384 incomplete for valid PostgreSQL 16 schemas.

Discussion Needed: Establish whether PostgreSQL ≥16 syntax should be owned by the lexer/parser or handled through a version-aware fallback, including compatibility with older PostgreSQL versions. Also resolve openGauss separately: it inherits PostgreSQLFunctionOption, while its official documentation describes system_user() as parenthesized and its keyword table does not classify SYSTEM_USER as an openGauss keyword. The agreed model needs parser-to-binder behavioral coverage, including same-name column handling.

Coverage

• Reviewed head: 303b4deb795598b5ade8ec47108853409730e1e2; base: c7bf30fc66598831afe237f71f8a0e8e7f93146b.
• Accounted for both authoritative GitHub files and linked requirement #39384; public comments and reviews were empty.
• Reviewed clusters: PostgreSQL function classification and its assertion-only test.
• Traced parser → projection → column binder → result-processing paths; checked PostgreSQL, openGauss, and the cited SQL92 issue #39099.
• Completed root-cause, blast-radius/contracts, tests/runtime, and final convergence passes. No evidence gap remains that could change the blocker set.
• Verification: scoped PostgreSQL parser compilation succeeded; runtime parser comparison confirmed the projection-type mismatch.
• Code-correctness review only. GitHub Actions and CI were not reviewed.