On Sat, Jun 25, 2016 at 8:32 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I think that this is an argument against further proliferation of
>> #define's for this kind of thing, not an argument against adding a way
>> to force tuplesort based hash index builds.
>
> So ... what's your point? This statement doesn't seem to lead to a
> conclusion in favor of either of the alternatives I mentioned.
I'm in favor of just adding a debug option. We should introduce a more
general idea of a #define that enables a variety of debugging options,
because it's silly to have to worry about buildfarm coverage each and
every time this crops up. I'm not entirely sure what level that should
be scoped at. The status quo is impractical.
Incidentally, did you see to it that there is buildfarm coverage for
RAW_EXPRESSION_COVERAGE_TEST?
--
Peter Geoghegan