Alternative to #6270.

This commit removes op==(sv,sv) from the note that suggests how to implement the "sufficient additional overloads". I've left the overload in the synopsis untouched, however, under the assumption that its presence there doesn't mean that an implementation must provide it.

I don't think that's a valid assumption ... although users are forbidden from taking the address of that function now, so I don't immediately see how they'd observe whether it's present. Either way, I dislike having declarations in a synopsis that are superfluous or just decorative.

This commit removes op==(sv,sv) from the note that suggests how to implement the "sufficient additional overloads". I've left the overload in the synopsis untouched, however, under the assumption that its presence there doesn't mean that an implementation must provide it.

I don't think that's a valid assumption ... although users are forbidden from taking the address of that function now, so I don't immediately see how they'd observe whether it's present. Either way, I dislike having declarations in a synopsis that are superfluous or just decorative.

I don't know if I'd call it "decorative", I somehow thought that I needed it as an "anchor" to define the semantics of operator==, even if then operator==(sv,sv) isn't actually provided. I can certainly open a LWG issue regarding this, but now I don't know any longer what resolution should I propose. Is there a precedent somewhere I can use as inspiration?

The semantics of equality comparison are defined by [string.view.comparison], not by non-existent declarations in the header synopsis.

If a single operator==(basic_string_view<charT, traits>, type_identity_t<basic_string_view<charT, traits>>) overload is sufficient to fully implement the required semantics, we can remove the prose and just specify precisely that.

So I would:

• Replace the homogeneous declaration in the header synopsis with the one using type_identity_t.
• Remove Example 1 from [string.view.comparisons] because it no longer adds anything of value.
• Change the normative description of operator== to use type_identity_t.

Does the same argument apply for operator<=> as well? Can we remove all mention of "sufficient additional overloads" from the header synopsis and from [string.view.comparison] and just define a single overload? If that works, we can get rid of paragraph 1 and table [tab:string.view.comparison.overloads] entirely, and maybe just turn p1 into a note explaining what the use of type_identity_t does. The table seems redundant, since all those comparisons are automatically supported if we have just == and <=>.

The semantics of equality comparison are defined by [string.view.comparison], not by non-existent declarations in the header synopsis.

If a single operator==(basic_string_view<charT, traits>, type_identity_t<basic_string_view<charT, traits>>) overload is sufficient to fully implement the required semantics, we can remove the prose and just specify precisely that.

OK. I had the impression that the wording deliberately didn't specify operator== like this (i.e. using type_identity_t) in order to give implementations some leeway in how exactly they want to achieve the desidered results. In theory an implementation can also choose some other strategy that doesn't use type_identity_t at all but constraints, SFINAE, etc.

If all of this i just a someho silly pretense and there's no real reason not to specify it with type_identity_t (which, well, is precisely the tool for the job here), then let's go with it.

So I would:

* Replace the homogeneous declaration in the header synopsis with the one using `type_identity_t`.

* Remove _Example 1_ from [string.view.comparisons] because it no longer adds anything of value.

* Change the normative description of `operator==` to use `type_identity_t`.

Does the same argument apply for operator<=> as well?

It does.

Can we remove all mention of "sufficient additional overloads" from the header synopsis and from [string.view.comparison] and just define a single overload? If that works, we can get rid of paragraph 1 and table [tab:string.view.comparison.overloads] entirely, and maybe just turn p1 into a note explaining what the use of type_identity_t does. The table seems redundant, since all those comparisons are automatically supported if we have just == and <=>.

That's the idea. OK, I'll package something like this and send it over as a LWG defect, because now it sounds way bigger than editorial.

type_identity_t didn't exist in the standard when those overloads were added, so using it wasn't an option.

@dangelog Could you have a go at @jwakely's suggestion?

Hello,

This is now https://cplusplus.github.io/LWG/issue3950 , I'm not sure if LWG likes the approach, the reflector thread was a bit inconclusive?

LWG might be able to look at it today.

Ah OK, but if you have an LWG issue, then we don't need this one. Thanks!

I mean, feel free to cherry pick the commit at a later time, if it implements the desired resolution. :-)

A slightly different approach is in #6270 . I guess that can be closed as well for the same reason (not editorial)?

I mean, feel free to cherry pick the commit at a later time, if it implements the desired resolution. :-)

Thanks, that's a nice thought, though it's just a little bit impractical given that we approve issue resolutions as a single, large paper and then draft edits on the back of that. So the important information would be to inform whoever is drafting the edits at that point that this PR already exists. I think for larger changes that could definitely be a nice bit of help (and we need to make sure to make a note on, say, the motions issue), but in this simple case it's probably not going to matter much.

@Dani-Hub, @jwakely: Would it be possible to annoate the LWG issue with a note that wording exists already, so that we remember when the time comes to draft?

done