Greg Stark <stark@mit.edu> writes:
> On Fri, Mar 16, 2012 at 3:16 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> So I now propose reverting the earlier two patches (but not their
> regression test cases of course) and instead hacking MergeAppend plan
> building as per (2).
> As a wise man once said, "This is tricky stuff". I feel a better that
> I got stuck on this stuff when you're still trying to feel your way
> after this many go-arounds.
Well, looking back on it, I feel this was at bottom a documentation
failure. I think that when I wrote the EquivalenceClass code, I knew
that "child" members did not have similar semantics to regular members.
But I had forgotten that when Teodor reported the MergeAppend bug,
and so misdiagnosed what I was seeing happen as being corruption of
the EC contents, when it wasn't really. I added some documentation
around this point in the patch I committed yesterday...
regards, tom lane
