On Mon, Nov 21, 2022 at 1:34 PM Andres Freund <andres@anarazel.de> wrote:
> Hm. But to get to that point we already need to have decided that xmax
> is not a normal xid. Unhelpfully we reuse the 'xid' variable for xmax as
> well:
> xid = HeapTupleHeaderGetRawXmax(tuple);
>
> I don't really know the HEAP_XMAX_INVALID branch is trying to do. For
> one, xid already is set to HeapTupleHeaderGetRawXmax(), why is it
> refetching the value?
Right, that detail is correct, but still weird. And suggests that it
might not have been super well thought through.
> So it looks to me like this path should just test !TransactionIdIsValid(xid)?
Agreed. Plus there should be a comment that reminds you that this is a
normal regular transaction ID (easy to miss, because the initial "if"
block for Multis is rather large).
I will push something like that soon.
--
Peter Geoghegan