Re: Use an enum for RELKIND_*?

Kyotaro HORIGUCHI <horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp> writes:
> At Tue, 18 Dec 2018 17:13:08 -0800, Andres Freund <andres(at)anarazel(dot)de> wrote in <20181219011308(dot)mopzyvc73nwjzdb6(at)alap3(dot)anarazel(dot)de>
>> Right now there's no easy way to use the compiler to ensure that all
>> places that need to deal with all kinds of relkinds check a new
>> relkind. I think we should make that easier by moving RELKIND_* to an
>> enum, with the existing letters as the value.

> I think we cannot use enums having base-type, so it will work
> unless we forget the cast within switch(). However, I don't think
> it is not a reason not to do so.
>
> switch ((RelationRelkind) rel->rd_rel->relkind)

Hm. This example doesn't seem very compelling. If we put a
"default: elog(ERROR...)" into the switch, compilers will not warn
us about omitted values. On the other hand, if we remove the default:
then we lose robustness against corrupted relkind values coming from disk,
which I think is going to be a net negative. Not to mention that we get
no help at all unless we remember to add the cast to every such switch.

So, while I understand Andres' desire to make this more bulletproof,
I'm not quite sure how this proposal helps in practice.

regards, tom lane