Re: Virtual tuple slots versus TOAST: big problem

Jan Wieck <JanWieck(at)Yahoo(dot)com> writes:
> On 11/20/2005 11:23 AM, Tom Lane wrote:
> Assuming that the saved header values don't need to be recomputed if the
> tuple doesn't need to be toasted at all, I think that toasting is
> expensive enough so that recomputing those values is hardly noticed.

Yeah, probably so. I'll just make the localized API change then.

One side effect of changing it as I suggest is that when control comes
back out of heap_insert/update, the caller will still have the
originally passed tuple and not the toasted version that actually went
to disk. AFAICS this is OK (of course we have to make sure t_self and
other header fields are updated in both copies). The main impact of
the change is that FormIndexDatum will receive the pre-toasting version
of the tuple, but that is actually a *good* thing --- right now, it just
has to fetch back untoasted fields anyway, if there's anything it needs
to do with them.

regards, tom lane