Re: Error-safe user functions

On Thu, Dec 15, 2022 at 9:03 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> Here are some proposed patches for converting range_in and multirange_in.
>
> 0001 tackles the straightforward part, which is trapping syntax errors
> and called-input-function errors.  The only thing that I think might
> be controversial here is that I chose to change the signatures of
> the exposed functions range_serialize and make_range rather than
> inventing xxx_safe variants.  I think this is all right, because
> AFAIK the only likely reason for extensions to call either of those
> is that custom types' canonical functions would need to call
> range_serialize --- and those will need to be touched anyway,
> see 0002.
>
> What 0001 does not cover is trapping errors occurring in range
> canonicalize functions.  I'd first thought maybe doing that wasn't
> worth the trouble, but it's not really very hard to fix the built-in
> canonicalize functions, as shown in 0002.  Probably extensions would
> not find it much harder, and in any case they're not really required
> to make their errors soft.
>
> Any objections?
>

There are other a bunch of hard errors from get_multirange_io_data(),
get_range_io_data() and its subroutine can hit, shouldn't we care
about those?


Regards,
Amul