>> I'm speaking primarily of functions as first-class objects, though
>> closures would be nice too. But consider an operation like
>>
>> UPDATE rel SET col1 = MAP ( f OVER col2 )
>>
>> We need to be able to determine whether this is well-typed, just as we
>> do now for any other SQL query. Specifically, we need to check that f
>> is a one argument function whose argument type is that of col2 and
>> whose return type is that of col1. My understanding is that right now
>> types are represented as 32-bit OIDs. I think they'd need to be some
>> sort of more complex structure in order to handle cases like this.
>
> Would above query not be written as
>
> UPDATE rel SET col1 = f(col2);
>
> anyway or am I missing something?
Ah, sorry, I mis-stated it slightly. I was imagining that col2 and
col1 where arrays, and f was a function between the base types, not
the array types.
> imho, having generic tuple tables as we have in INSERT INTO (...)
> VALUES (...),(...),(...)
>
> to be useable in all places like a real table would be helpful in
> many cases.
>
> But this might be completely unrelated :)
Probably. :-)
...Robert