Andres Freund <andres@anarazel.de> writes:
> On 2021-06-11 19:08:57 -0400, Tom Lane wrote:
>> Anyway, I agree that disabling that was a bit of a stopgap hack. This
>> 'nonstring' attribute seems like it would help for ECPG's usage, at
>> least.
> nonstring is supported since gcc 8, which also brought the warnings that
> e71658523 is concerned about. Which makes me think that we should be
> able to get away without a configure test. The one complication is that
> the relevant ecpg code doesn't include c.h.
Ugh. And we *can't* include that there.
> But I think we can just do something like:
> - char sqlstate[5];
> + char sqlstate[5]
> +#if defined(__has_attribute) && __has_attribute(nonstring)
> + __attribute__((nonstring))
> +#endif
> + ;
> };
Hmm. Worth a try, anyway.
> Should we also include a pg_attribute_nonstring definition in c.h?
> Probably not, given that we right now don't have another user?
Yeah, no point till there's another use-case. (I'm not sure
there ever will be, so I'm not excited about adding more
infrastructure than we have to.)
regards, tom lane