Re: Improving the names generated for indexes on expressions

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Tue, Sep 16, 2025 at 12:56 AM David G. Johnston
> <david(dot)g(dot)johnston(at)gmail(dot)com> wrote:
>> If there are no function names present, output “expr” in lieu of a function name. Then just output any columns that are present. No operators, no constants.

> In the previous discussion, the user's expression indexes were on
> these expressions:

> jsondata -> 'a' -> 'b'
> jsondata -> 'x' -> 'y'

> So "no operators, no constants" wouldn't really allow us to make any
> useful progress, inasmuch as it would throw away everything that
> matters.

Precisely. It's exactly expression indexes on OpExprs that
FigureColumn is completely useless for; if we don't do something
for those then we've not moved the needle much. I initially
tried leaving out Consts, but that didn't work well on the
indexes in the regression tests, let alone the field example
Robert cites.

One thing I was thinking about after putting up the initial draft
was to suppress the underscores around operator names, so that
instead of

mytab_jsondata_->_a_->_b_idx

the above would net you

mytab_jsondata->a->b_idx

It's less consistent but looks less busy too.

> I am not really sure we want to do what Tom proposes here because, as
> Pavel says, it would result in a lot of indexes containing special
> characters in the name.

Question is, why should we care about that?

regards, tom lane