Re: WIP: Faster Expression Processing v4

On 2017-03-23 20:36:32 -0400, Tom Lane wrote:
> So I think that we have got to fix ExecEvalWholeRowVar so that it doesn't
> clobber the state of the slot. Right at the moment, the only way to do
> that seems to be to do this instead of ExecFetchSlotTupleDatum:
>
> tuple = ExecCopySlotTuple(slot);
> dtuple = (HeapTupleHeader)
> DatumGetPointer(heap_copy_tuple_as_datum(tuple,
> slot->tts_tupleDescriptor));
> heap_freetuple(tuple);

Hm. One disadvantage would be that repeated whole-row references to the
same table would be a bit slower, because we'd repeatedly form a tuple
from a virtual one - but I have a hard time coming up with a scenario
where that'd matter. I'd suspect that in the end it'd probably even
have a *positive* performance impact, because right now the next scalar
access will have to deform the whole tuple again, and that seems like a
lot more likely scenario.

Greetings,

Andres Freund