P2996R13 Reflection for C++26

[jensmaurer]

Feel free to cherry-pick into #8004 once @Eisenwave is done with the library part.

[jensmaurer]

Fixed link, thanks.

[tkoeppe]

We could certainly move all the commits into a single PR and the test and review that again. Let's wait for a first round of reviews on both current PRs to settle, then we'll take it from there.

[Eisenwave]

@tkoeppe by the way, you've put the milestone date for 2025-06 as 06-31. Is that actually correct? I was under the impression that we would want to have the updated working draft in the post-Sofia mailing, if not sooner so that the ISO review process can start ASAP.

[tkoeppe]

@Eisenwave Sorry, no idea, I don't recall making that label or how it was made. Don't worry about it. We're not going to stop the process because of what a label says.

[jensmaurer]

That milestone is by me. I've moved the date to 07-15.

[jensmaurer]

yes, please

[jensmaurer]

Yes, please. All headers (historic accidents notwithstanding) should be name.syn; it simplifies macro usage pointing to them so much.

[jensmaurer]

Sounds good, except that'd like to have a global rename of "meta.reflection*" labels to "meta.refl*" for space economy (without loss of clarity, in my view).

[Eisenwave]

It may be better to do this as a bulk change after motions. There are quite a lot of reflections papers that would all need individual fixups otherwise, and it's making it more complicated than it needs to be.

[Eisenwave]

I'll resolve those comments today. @jensmaurer when you said that types always use code font, does that mean that any use of e.g. some_trait_t<$T$> has to go, basically? The PR contains quite a lot of those currently.

The paper makes these italic, so I assume that's \exposid{T} then?

Also, does every FIXME require its own fixup commit? If yes, then there are like 20 incoming commits that each add one comma ...

[jensmaurer]

Also, does every FIXME require its own fixup commit? If yes, then there are like 20 incoming commits that each add one comma ...

No, one big commit is good enough. This will be squashed anyway.

For T, in general, we use T and U as type variables in upright code font, even if they are sort-of meta-variables. (If something could be a valid C++ snippet, we use upright code font.) Since an expression cannot always be written \tcode{E} (value category and bit-field-ness are not expressible as "E"), we use $E$. In contrast, all properties of a type (now: differentiation vs. type alias makes that claim dubious) are present in a typedef that happens to be called T, so we can write \tcode{T}.

I suggest you leave $T$ for now (this can notionally express differentiation between type and type alias).

[Eisenwave]

Alright, the PR is now in a state where all the little comma FIXME and other minor things are resolved. I'd hold off on major changes like renaming to meta.syn and meta.refl due to all the papers based on reflection being affected by this change.