On Sat, Aug 19, 2023 at 08:35:46AM +0100, Dean Rasheed wrote:
> I note that there are no tests for negative inputs.
I added some in v8.
> Doing a quick test, shows that this changes the current behaviour,
> because all inputs are now treated as 64-bit:
>
> HEAD:
>
> select to_hex((-1234)::int);
> to_hex
> ----------
> fffffb2e
>
> With patch:
>
> select to_hex((-1234)::int);
> to_hex
> ------------------
> fffffffffffffb2e
Good catch. In v8, I fixed this by first casting the input to uint32 for
the 32-bit versions of the functions. This prevents the conversion to
uint64 from setting the rest of the bits. AFAICT this behavior is pretty
well defined in the standard.
> The way that negative inputs are handled really should be documented,
> or at least it should include a couple of examples.
I used your suggestion and noted that the output is the two's complement
representation [0].
[0] https://postgr.es/m/CAEZATCVbkL1ynqpsKiTDpch34%3DSCr5nnau%3DnfNmiy2nM3SJHtw%40mail.gmail.com
--
Nathan Bossart
Amazon Web Services: https://aws.amazon.com
