bug#58168: string-lessp glitches and inconsistencies

[Eli Zaretskii <eliz@gnu.org>]

> From: Mattias Engdegård <mattias.engdegard@gmail.com>
> Date: Sat, 1 Oct 2022 21:57:45 +0200
> Cc: 58168@debbugs.gnu.org
> 
> 1 okt. 2022 kl. 07.22 skrev Eli Zaretskii <eliz@gnu.org>:
> 
> > It depends on the use case, but in general I see no problem with
> > signaling errors when we cannot produce reasonably correct results.
> > For example, string-to-unibyte does signal an error in some cases.
> 
> That's fine because that function is documented to do so and always has, but making previously possible comparisons raise errors shouldn't be done lightly.

I didn't say "lightly", nor do I think so.  We need to discuss
specific use cases.

An alternative is to always convert unibyte non-ASCII strings to their
multibyte representation before comparing.

> Comparison between objects is not only useful when someone cares about their order, as in presenting a sorted list to the user. Often what is important is an ability to impose an order, preferably total, for use in building and searching data structures. I came across this bug when implementing a string set.

Always converting to multibyte handles this case, doesn't it?

> >> It's also a matter of performance -- string< has been improved recently but currently we compare text in Latin and Swahili much faster than French and Arabic; it would be nice to close that gap. UTF-8 is designed so that comparing strings by scalar values can be done byte-wise, but the way we encode raw bytes make them sort right between ASCII and Latin-1. Given that the specific order doesn't matter much, we could just run with that.
> > 
> > I see no reason to make comparison of unibyte and multibyte strings
> > perform better.
> 
> Actually I was talking about multibyte-multibyte comparisons.

Then why did you mention raw bytes? their multibyte representation
presents no performance problems, AFAIU.

> You were probably thinking about comparisons between unibyte strings that contain raw bytes and multibyte strings, and those are indeed not very performance-sensitive. However there is no way to detect whether a unibyte string contains non-ASCII chars without looking at every byte, and comparing unibyte ASCII with multibyte is definitely of interest. Strings are still unibyte by default.

You can compare under the assumption that a unibyte string is
pure-ASCII until you bump into the first non-ASCII one.  If that
happens, abandon the comparison, convert the unibyte string to its
multibyte representation, and compare again.