Re: Composite Datums containing toasted fields are a bad idea(?)

Heikki Linnakangas <hlinnakangas(at)vmware(dot)com> writes:
> On 04/25/2014 02:40 AM, Tom Lane wrote:
>> * The patch changes HeapTupleGetDatum from a simple inline macro into
>> a function call. This means that third-party extensions will not get
>> protection against creation of toast-pointer-containing composite Datums
>> until they recompile.

> One consequence of that is that an extension compiled with headers from
> new minor version won't work with binaries from an older minor version.
> Packagers beware.

Yeah, that's a possible issue, though I think we've done such things
before. In any case, alternative approaches to fixing this would likely
also involve extensions needing to call core functions that don't exist
today; though maybe the number of affected extensions would be smaller.

regards, tom lane