Re: Underscores in numeric literals

On Tue, Dec 27, 2022 at 09:55:32AM -0500, Tom Lane wrote:
> Peter Eisentraut <peter(dot)eisentraut(at)enterprisedb(dot)com> writes:
> > Here is a patch to add support for underscores in numeric literals, for
> > visual grouping, like
>
> > 1_500_000_000
> > 0b10001000_00000000
> > 0o_1_755
> > 0xFFFF_FFFF
> > 1.618_034
>
> > per SQL:202x draft.
>
> > This adds support in the lexer as well as in the integer type input
> > functions.
> > TODO: float/numeric type input support
>
> Hmm ... I'm on board with allowing this in SQL if the committee says
> so.

> I'm not especially on board with accepting it in datatype input
> functions. There's been zero demand for that AFAIR. Moreover,
> I don't think we need the inevitable I/O performance hit, nor the
> increased risk of accepting garbage, nor the certainty of
> inconsistency with other places that don't get converted (because
> they depend on strtoul() or whatever).

+1 to accept underscores only in literals and leave input functions
alone.

(When I realized that python3.6 changed to accept things like
int("3_5"), I felt compelled to write a wrapper to check for embedded
underscores and raise an exception in that case. And I'm sure it
affected performance.)

--
Justin