>
> This just looks like PG missing a feature. plpgsql has much less user
> and developer time spent on it, so I'd expect to find more strangeness
> in darker corners like this.
>
this rule should be simply removed. It's not problem. The people long
time believe so row cannot be null ever. I don't know if this is from
Oracle or somewhere. SQL/PSM allows it. This semantic is little bit
difficult. There is rule so any object is NULL when all fields is NULL
too. I thing, so it's true. There is object, that has zero
information. When You thinking about it, you have to forgot any your
knowledges from languages that's knows pointers. Maybe some people has
problem, because they put in equality NULL from SQL and NULL pointer.
regards
Pavel Stehule