Re: Improving the names generated for indexes on expressions

On Tue, Jun 16, 2026 at 9:44 PM Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
>
> On Thu, Jun 4, 2026 at 6:31 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > I spent some time trying to parse out why I don't like that
> > suggestion, and eventually realized that it's this: I don't want
> > the behavior of CREATE INDEX to be dependent on every last detail
> > of ruleutils.c. I'm afraid that that'd introduce undesirable
> > cross-version changes in the names selected for indexes. Now,
> > considering this sort of change at all requires an assumption that
> > we can get away with breaking any applications that are sensitive
> > to that. But we only have to assume that we can get away with that
> > once. If we get ruleutils.c involved then I foresee a steady drip
> > drip drip of edge-case naming changes, and I think that will annoy
> > people.
>
> Hmm, I don't know. I suppose you could go back and see whether making
> these changes in the back-branches would produce different regression
> test changes than making them against the master branch.
>
> > In any event, the cfbot has been nagging me that this patch needs
> > a rebase, so here's v3. The only change from v2 is that there are
> > some new test cases in indexing.sql that need adjustment.
>
> ...but whether you do that research or not, I still think this is a
> very significant improvement. For those not wanting to open up the
> patch:
>
> - An index on ((a + 0)) now gets the name idxpart1_a_0_idx instead of
> idxpart1_expr_idx.
> - An index on ((a + b)) now gets the name idxpart_a_b_idx instead of
> idxpart_expr_idx.
> - An index on (abs(b)) now gets the name idxpart1_abs_b_idx instead of
> idxpart1_abs_idx.
>
> Personal taste certainly enters into the calculus here, but IMHO
> calling everything BLAH_expr_idx because the topmost thing is some
> kind of operator invocation is a really poor user experience, and even
> for cases where the top-level thing is a function invocation,
> including more than just the name of the top-level function in the
> index name seems like a really significant improvement.
>

+1. I haven't followed this thread but the examples along with your
explanation for the same clearly sounds like an improvement to me.

--
With Regards,
Amit Kapila.