Michael Paquier <michael@paquier.xyz> writes:
> A couple of years ago, 62e2a8dc has introduced in c.h a set of limits
> (to fix some portability issues from 83ff1618) to make the code more
> system-independent. Those are for example PG_INT32_MIN, etc. The core
> code now mixes the internal PG_ limits with the system ones. Would we
> want to unify a bit the whole thing and replace all the SHRT_MIN/MAX,
> LONG_MIN/MAX and such with the internal limit definitions?
I doubt it's really worth the trouble. I did just make such a change in
commit cbdb8b4c0, but it was mostly (a) so that the different ftoiN/dtoiN
functions would look more alike, and (b) because the relevant variables or
result values were actually declared int16, int32, etc. It would be flat
wrong to replace SHRT_MIN or LONG_MIN in a context where it was used to
check whether a value would fit in a variable declared "short" or "long".
> I suspect that the buildfarm does not have any more members where
> sizeof(int) is 2.
I doubt PG has ever been able to run on two-byte-int hardware. Certainly
not in the buildfarm era.
> I am seeing close to 250 places in the core code,
> most of them for INT_MIN and INT_MAX.
You'd really need to look at the associated variables to see whether any
of those would be better off as INT32_MIN/MAX.
regards, tom lane