Etsuro Fujita <fujita.etsuro@lab.ntt.co.jp> writes:
> (2014/08/27 3:27), Tom Lane wrote:
>> I looked this over, and TBH I'm rather disappointed. The patch adds
>> 150 lines of dubiously-correct code in order to save ... uh, well,
> Just for my study, could you tell me why you think that the code is
> "dubiously-correct"?
It might be fine; I did not actually review the new
adjust_appendrel_attr_needed code in any detail. What's scaring me off it
is (1) it's a lot longer and more complicated than I'd thought it would
be, and (2) you already made several bug fixes in it, which is often an
indicator that additional problems lurk.
It's possible there's some other, simpler, way to compute child
attr_needed arrays that would resolve (1) and (2). However, even if we
had a simple and obviously-correct way to do that, it still seems like
there's not very much benefit to be had after all. So my thought that
this would be worth doing seems wrong, and I must apologize to you for
sending you chasing down a dead end :-(
regards, tom lane