Gregory Stark <stark@enterprisedb.com> writes:
> That does really suck. But I'm not sure what we can do about it. There's no
> SQL which is entirely equivalent to the resulting view.
If we were to do anything about it, I think it would have to be to
forbid the original ALTER. But I don't see any good way to detect
the situation, either.
Consider also that an ALTER ... RENAME could create a similar failure,
if the new column name conflicts with one in some other table that is
joined against someplace.
I think that the described behavior is actually pretty harmless:
you get everything back except the broken view. So I'm not too
unhappy saying that it's going to keep working that way.
regards, tom lane