Re: Avoiding rewrite in ALTER TABLE ALTER TYPE

On Wed, Dec 29, 2010 at 11:16:23AM -0500, Tom Lane wrote:
> Noah Misch <noah@leadboat.com> writes:
> > ALTER TABLE ALTER TYPE always rewrites the table heap and its indexes.  In some
> > cases, we can determine that doing so is unhelpful, and that the conversion
> > shall always succeed:
> > I wish to replace table rewrites with table verification scans where possible,
> > then skip those verification scans where possible.
> 
> This has been discussed before; have you read the previous threads?

I cited two threads I had read on the subject.  Were there other important ones?

> I really really dislike the notion of a "verification scan": it's
> basically work that is going to be useless if it fails.  I think your
> argument that it will usually fail quickly is quite unconvincing, and in
> any case the situations where it is useful at all are too thin on the
> ground to be worth the code space to implement it.  It seems sufficient
> to me to skip the rewrite in cases of provable binary compatibility, with
> possibly an extra check for "safe" changes of typmod.  With respect to
> the latter, I agree a type-specific function to compare the typmods
> would be the way to go, although "exemptor" seems a pretty badly chosen
> name for it.

I have attempted to expand on these problems in my reply to Robert.