(our mail crossed in the ether...)
> Let's say, theoretically, that in the future we want to allow people to
> change the type of their columns, plus allow them to change the nullability.
Right.
> Should we come up with a syntax for changing nullability that allows for the
> future changing of column type? If so, then a syntaxes like these might be
> the way to go:
Yup.
> If we just allow the full col spec we could one day support this:
> ALTER TABLE blah ALTER COLUMN col text boolean NOT NULL DEFAULT 'f';
> Which would change the column to that definition (if coercion is possible)
> no matter what current definition is...
Right. No point in *precluding* that with a short-sighted choice of
syntax.
> Is this the eventual goal? Will this cause shift/reduce errors? will we
> need to put the word 'SET' in after 'col'?
Probably not, if we can already do this with CREATE TABLE.
And if we head this direction, then choosing a syntax which most closely
mimics the current CREATE TABLE will allow altering two columns at once,
which would be more efficient presumably than doing one column at a
time.
- Thomas