execAmi.c points out that
* (However, since the only present use of mark/restore is in mergejoin,* there is no need to support mark/restore in
anyplan type that is not* capable of generating ordered output. So the seqscan, tidscan,* and valuesscan support is
actuallyuseless code at present.)
I don't see any prospect that we'd ever adopt mark/restore for any other
purpose than mergejoin, so it seems to me that the code in question is
permanently dead. There's not that much of it, but I'm thinking that
ripping it out and clarifying the commentary in execAmi.c would still
be a net benefit for readability. Any objections?
regards, tom lane