[CALCITE-5575] Incorrect priority of builtinFunctionCallMethods in Parser.jj

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Information Provided
• Affects Version/s: 1.34.0
• Component/s: core
• Assignee: Unassigned
• Reporter: James Turton
• Created: 13/Mar/23 12:32
• Updated: 17/Mar/23 09:00
• Resolved: 17/Mar/23 09:00

Description

CALCITE-5469 and others have introduced new date and time functions. In Parser.jj (line 6152) these new functions are handled above the builtinFunctionCallMethods Freemarker loop that allows applications to add their own function defintions.

        node = DateTimeConstructorCall() { return node; }
    |
        node = DateDiffFunctionCall() { return node; }
    |
        node = DateTruncFunctionCall() { return node; }
    |
        node = TimestampAddFunctionCall() { return node; }
    |
        node = DatetimeDiffFunctionCall() { return node; }
    |
        node = TimestampDiffFunctionCall() { return node; }
    |
        node = TimestampDiff3FunctionCall() { return node; }
    |
        node = TimestampTruncFunctionCall() { return node; }
    |
        node = TimeDiffFunctionCall() { return node; }
    |
        node = TimeTruncFunctionCall() { return node; }
    |
<#list (parser.builtinFunctionCallMethods!default.parser.builtinFunctionCallMethods) as method>
        node = ${method} { return node; }
    |
</#list>

This means (I think) that applications are effectively unable to override the new date and time function definitions with their own definitions. In particular, Drill cannot override the new three-parameter DATE_DIFF function in order to provide the parser method needed by its own two-parameter DATE_DIFF function resulting in a regression in the form of a broken DATE_DIFF function.

If the above is correct then can the builtinFunctionCallMethods Freemarker loop move above these date and time functions? Could it even move above the special syntax functions higher up, for that matter?

Incidentally I also noted in Parser.jj that the new date and time functions mentioned above are also handled seperately in JdbcFunctionCall which implements JDBC function call syntax. This makes me unsure whether application overrides provided in builtinFunctionCallMethods will take priority over these Calcite definitions when they are invoked with JDBC syntax.

 

Issue Links

• is related to: CALCITE-5469 Implement BigQuery DATETIME_ADD/DATETIME_DIFF
• is related to: CALCITE-5579 Allow Time Frames as Valid Expressions