Tom Lane wrote:
> Jorge Godoy <jgodoy@gmail.com> writes:
>> Em Thursday 18 October 2007 16:37:59 Joe Conway escreveu:
>>> The row is pretty useless without a rowid in this context -- it seems
>>> like the best thing to do would be to skip those rows entirely. Of
>>> course you could argue I suppose that it ought to throw an ERROR and
>>> bail out entirely. Maybe a good compromise would be to skip the row but
>>> throw a NOTICE?
>
>> If I were using it and having this problem I'd rather have an ERROR.
>
> I can think of four reasonably credible alternatives:
>
> 1. Treat NULL rowid as a category in its own right. This would conform
> with the behavior of GROUP BY and DISTINCT, for instance.
>
> 2. Throw an ERROR if NULL rowid is seen.
> Not being a heavy user of crosstab(), I'm not sure which of these is the
> most appropriate, but #1 seems the most defensible from a theoretical
> perspective.
>
> Since the bug has gone undiscovered this long, it seems obvious that
> not too many people actually try to feed null rowids to crosstab; so
> expending a lot of effort to fix it is probably not reasonable.
> If you don't like #1 I'd vote for #2 second.
Hadn't really thought about #1, but now that you mention it, it does
make sense. #1 gets my vote too. I'll pick this up next week if that's OK.
Joe