Re: check for interrupts in set_rtable_names

Jeff Janes <jeff(dot)janes(at)gmail(dot)com> writes:
> On Fri, Nov 13, 2015 at 3:13 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> There's probably no reason not to do that, but I'd be much more interested
>> in eliminating the slowness to begin with ...

> I was thinking about that as well, but I don't think that would be
> back-patchable, at least not the way I was envisioning it.

> I was thinking of detecting bad cases (had to count to over 10 before
> finding a novel name, more than 10 times) and then switching from an
> object-local count, to a global count, for the numbers to add to the
> name. But that would be a behavior change under some conditions.

I experimented with using a hash table to avoid the O(N^2) behavior.
This seems to work quite well, and I think it doesn't change the results
(leastwise, it does not break the regression tests).

It would be possible to do something similar to dodge the O(N^2) costs
in make_colname_unique/colname_is_unique; but it would be a lot messier,
and the tests I did suggest that it's fairly hard to get into a regime
where those functions are a huge part of the runtime anyway, because
we have limits on the numbers of columns. So I'm inclined to leave that
alone.

regards, tom lane