Gregory Stark <stark(at)enterprisedb(dot)com> writes:
> Or is it estimating the width based on the belief that only the thousand
> column is actually going to be emitted?
Right. The width is used to estimate how much space would be needed
for, eg, sorting or hashing the plan node's output. In any case where
something like that is actually happening, we *should* be emitting only
the required columns, so I didn't see any particular need to make
use_physical_tlist change the reported width. OTOH this bug shows that
maybe that was hiding useful information ...
regards, tom lane