bug#69709: `sort` interface improvement and universal ordering predicate

["Mattias Engdegård" <mattias.engdegard@gmail.com>]

17 maj 2024 kl. 14.29 skrev Daniel Mendler <mail@daniel-mendler.de>:

> Just letting you know that I've implemented value< and sort in the
> emacs-30 branch of Compat. It works well so far, but value< is not yet
> completely implemented. See the following commit:
> https://github.com/emacs-compat/compat/commit/8190769d9eb9258dd8361bd322d90228dc586770

An excellent start! I'll post comments on the source site.

> There is one thing I'd like to ask about value<. Would it make sense to
> support comparing mixed types, e.g., numbers and markers or strings and
> symbols?

I went back and forth quite a bit, but decided to start small homogeneous comparisons which would at least give the option to extend to heterogeneous (or in your case, ad-hoc mixed) comparisons later on, without committing too much to any particular design until we have more experience.

When it comes to mixing numbers and markers, I'm still leaning against it. `<` allows mixed comparisons but it doesn't work well with markers from different buffers. `value<` orders markers by grouping them by buffer which feels more useful.

(I've never been a great fan of the way elisp allows comparison and arithmetic on markers and numbers; it probably seemed to be a good idea at the time. That feature seems to have caused more muddled than clear code.)

Same with strings and symbols: I can't remember having seen a collection where I would have wanted them to be sorted together, and `nil` being a symbol means that we would lose a useful error check.

There is a more tempting case for a truly heterogeneous comparisons, like a universal total order. Some other languages have this but I think it is more difficult to construct with so much legacy. I went with unifying all numbers for pragmatic reasons even though -0.0 and NaN are terrible.