Re: ERROR: functions in index expression must be marked IMMUTABLE

"Sven R. Kunze" <srkunze@xxxxxxx> · Fri, 3 Mar 2017 12:44:27 +0100



    On 03.03.2017 11:43, Geoff Winkless
      wrote:

    
        One
          alternative would be to make to_date accept all language
          variants of months simultaneously. A quick search of google
          suggests that there aren't any overlaps between languages (ie
          where one language uses "Foo" for March and another uses "Foo"
          for May), although you would have to do some more intense
          research to be sure. As far as I can see there's no other
          reason why to_date would need to be marked as stable/volatile,
          is there?
      
    
    I don't think so. It could be viable.

    
          On
            the down side I imagine it would involve some
            potentially-prohibitively-large lookup tables; it would also
            end up with a technical incompatibility in that what ANSI
            SQL would reject as not-a-date might be parsed as a date.
        
      
    There is another issue: languages change (admittedly very slowly)
    but I wouldn't want PostgreSQL to be incompatible with future
    generations.

    Your performance argument weighs heavier, though.

    
          I'm
            not in a position to judge if either of those would be
            acceptable.

          
    Do you think I should post to pgsql-hackers?

    
    Sven