"Guillaume Smet" <guillaume(dot)smet(at)gmail(dot)com> writes:
> So the basic explanation is that it's in both lists due to the partial
> index and only qpqual keeps the condition? I would have expected the
> opposite but it doesn't change anything I suppose?
It gets the right answer, yes. I'm not sure if we could safely put the
condition into the recheck instead of the filter. The particular code
I showed you has to go the direction it does, because a condition in the
filter has to be checked even if the bitmap is not lossy. I seem to
recall concluding that we had to recheck partial-index conditions even
if the bitmap is not lossy, but I can't reconstruct my reasoning at the
moment.
regards, tom lane