On Thu, Jul 18, 2024 at 10:05:34AM +0200, Laurenz Albe wrote:
> On Wed, 2024-07-17 at 15:03 -0700, Noah Misch wrote:
> > If I'm counting the votes right, you and Tom have voted that the feature's
> > current state is okay, and I and Laurenz have voted that it's not okay.
>
> Maybe I should expand my position.
>
> I am very much for the built-in CTYPE provider. When I said that I am against
> changes in major versions, I mean changes that are likely to affect real-life
> usage patterns. If there are modifications affecting a code point that was
> previously unassigned, it is *theoretically* possible, but very unlikely, that
> someone has stored it in a database. I would want to deliberate about any change
> affecting such a code point, and if the change seems highly desirable, we can
> consider applying it.
>
> What I am against is routinely updating the built-in provider to adopt any changes
> that Unicode makes.
Given all the messages on this thread, if the feature remains in PostgreSQL, I
advise you to be ready to tolerate PostgreSQL "routinely updating the built-in
provider to adopt any changes that Unicode makes". Maybe someone will change
something in v18 so it's not like that, but don't count on it.
Would you like to change your vote to "okay", keep your vote at "not okay", or
change it to an abstention?
> To make a comparison with Tom's argument upthread: we have slightly changed how
> floating point computations work, even though they are IMMUTABLE. But I'd argue
> that very few people build indexes on the results of floating point arithmetic
> (and those who do are probably doing something wrong), so the risk is acceptable.
> But people index strings all the time.
Agreed.