How to style expos-only members [Was: Restyle expos-only member basic-formatting-string::str]

[tkoeppe]

See https://github.com/cplusplus/draft/issues/4680#issuecomment-862806234

We should style the exposition-only member str of basic-formatting-string in regular upright font, as we typically do for expos-only members. We should also rename it to str_, as we do for private expos-only members.

[jensmaurer]

The private ranges members are all styled \exposid.

And, outside of ranges, we have a fair amount of member names not ending in _.

Personally, I think all exposition-only member names should be italics, and we should find a better / more pronounceable way of creating a new namespace, other than appending _. (In ranges, we occasionally have "current" and "current_", and they mean different things.)

[tkoeppe]

Do you have a sense of how many non-ranges private members there are?

[tkoeppe]

Changing the issue to decide the general style first.

[jensmaurer]

Do you have a sense of how many non-ranges private members there are?

Rough estimate, counting the \expos markers (ignoring \exposid):

$ grep -w expos *.tex | grep -v ranges.tex | wc -l 275

In other words, oodles of stuff.

[tkoeppe]

OK, thanks. I feel like we're not going to resolve this for this mailing. Maybe for the next pre-meeting mailing!

[jensmaurer]

Editorial meeting 2021-06-18: Consistent, global rule: All private data members have a _ suffix. All exposition-only names (including private data members) are styled with \exposid{...}.

The presence of the _ suffix is saddening, but we don't have a better idea right now.

[Dani-Hub]

Re @tkoeppe :

Do you have a sense of how many non-ranges private members there are?

In N4885 there are 51 occurrences within Clause [ranges], specifically the following ones: subrange: 2 single_view: 1 iota_view: 2 iota_view<W, Bound>::iterator: 1 iota_view<W, Bound>::sentinel: 1 basic_istream_view: 2 basic_istream_view<Val, CharT, Traits>::iterator: 1 ref_view: 1 filter_view: 2 filter_view<V, Pred>::iterator: 2 filter_view<V, Pred>::sentinel: 1 transform_view: 2 transform_view<V, F>::iterator: 2 transform_view<V, F>::sentinel: 1 take_view: 2 take_view::sentinel: 1 take_while_view: 2 take_while_view<V, Pred>::sentinel: 2 drop_view: 2 drop_while_view: 2 join_view: 2 join_view::iterator: 3 join_view::sentinel: 1 split_view: 3 split_view<V, Pattern>::outer-iterator: 2 split_view<V, Pattern>::outer-iterator ::value_type: 1 split_view<V, Pattern>::inner-iterator: 2 common_view: 1 reverse_view: 1 elements_view: 1 elements_view<V, N>::iterator: 1 elements_view<V, N>::sentinel: 1

[tkoeppe]

Thanks, Daniel! Unfortunately that seems to mean that we have a lot more "old" private members that we now need to update to the new style :-S

[JohelEGP]

Editorial meeting 2021-06-18: Consistent, global rule: All private data members have a _ suffix. All exposition-only names (including private data members) are styled with \exposid{...}.

The presence of the _ suffix is saddening, but we don't have a better idea right now.

Non-static only? https://github.com/cplusplus/draft/blob/a4e40b0ada8a15b601566af3f96bf89314e7ef60/source/ranges.tex#L5256

[JohelEGP]

The following Clauses are known to follow the rule (or not):

• [x] [ranges]

[W-E-Brown]

FWIW, I believe this is an incredibly silly de minimis issue to have spent any time on. However, since we have/are, ...

I find that appending a useless trailing underscore to be an especially egregious waste of time. Moreover, such an emendation sends an entirely unintended message to readers, namely, that the Standard Library recommends the consistent use of trailing underscores for all such exposition-only member names. Since I can find no such policy or recommendation from WG21 (or even from LWG), I (1) object on the grounds that this is not a purely editorial change and (2) request reversion to status quo ante.

[jensmaurer]

For [ranges], nothing was changed; the incoming papers generally had italics private data members with trailing underscores. Worse, sometimes we use current_ and current to mean different things (the latter is a meta-macro).

The most important thing about private data members is that their names are not in any way reserved by the standard library and e.g. a user is allowed to #define a macro with that name. Styling them in an non-utterable way (e.g. italics) seems to help in conveying such a notation. See also our exposition-only concept names.

[tkoeppe]

@W-E-Brown: The status quo ante is not consistent across the standard, and something we want to improve. We will necessarily have to touch a fair number of places in the standard, no matter which way we go. Marking expos-only members up in the source seems helpful, regardless of how we style them, so that part of the work would need to get done no matter what.

Have you got any concrete objections to any particular style? Otherwise, we seem to have a strong preference for italics and kebab-case (like all exposition-only names), and a mild preference for the trailing underscore based on partial existing practice.

I'm inclined to use that throughout unless we have good reasons to prefer something else. But please do note that the ranges library would be ambiguous without the trailing underscore, so alternative proposals would have to provide an alternative disambiguation.

[tkoeppe]

Editorial meeting: we would like to continue the direction we decided at the previous meeting, but would like to achieve consensus among the editors.

[W-E-Brown]

On Nov 19, 2021, at 12:18 PM, Thomas Köppe @.***> wrote: @W-E-Brown https://github.com/W-E-Brown: The status quo ante is not consistent across the standard,

So? You say that like it's a bad thing. and something we want to improve.

I have yet to hear a cogent reason for this. We will necessarily have to touch a fair number of places in the standard, no matter which way we go.

I don't see why we should bother. Marking expos-only members up in the source seems helpful, regardless of how we style them, so that part of the work would need to get done no matter what.

I've not heard of anyone who has been misled or confused by our current style, so I'm unsure who would find any such adjustment to be "helpful." Have you got any concrete objections to any particular style? Otherwise, we seem to have a strong preference for italics and kebab-case (like all exposition-only names), and a mild preference for the trailing underscore based on partial existing practice.

If this is really worth such relatively large effort for what I see as no gain, I can live with italics.

I can even live with kebab-case, if truly necessary.

However, a trailing underscore is simply inconsistent with kebab-case. It's not a practice we should be implicitly recommending by imposing everywhere.

I'm inclined to use that throughout unless we have good reasons to prefer something else.

No, we need good reasons to change the status quo. I've not seen any in this thread. But please do note that the ranges library would be ambiguous without the trailing underscore, so alternative proposals would have to provide an alternative disambiguation.

Then let's leave the names alone as-is, and be content with (at most) italics to achieve the consistency you seem to find desirable.

Or let's find better, underscore-free, names for the ranges library to use. (The only meaningful use of a trailing underscore would have been in names such as kilo_, i.e., in recognized prefixes. But we eschewed trailing underscores there. I just don't understand why we would want to put them in more places where they make no inherent sense.)

Sorry, but I still don't find any of this worth even the effort that's already been expended on it. Having multiple ways of expressing valid C++ code is a strength, not a weakness, of the language. Seeking consistency by establishing a convention for such expression seems a waste of our time; I have yet to see a problem here that needs a solution.

But if it's still strongly felt that some convention about such names is absolutely needed, then italicize to your hearts' content. But if you additionally want kebab-case, then the trailing underscores (like all underscores) have to go away and the italics alone will have to serve as sufficient disambiguation.

-- WEB