[clang] [RFC] Initial implementation of P2719 (PR #113510)

[Erich Keane via cfe-commits]

erichkeane wrote:

> I'm getting nitpicky, but there are still unaddressed comments.
> 
> In the interest of landing that soon, we should figure out the following:
> 
> Do we want to keep the document, knowing this is likely to be adopted as a standard feature? I would rather just link to the paper.
> 
>     * We need a changelog entry
> 
>     * Do we want to keep the compiler flags knowing this is likely to be adopted as a language feature? I would prefer checking for c++26 + extension warnings in older language modes
> 
>     * Do we actually want to set the feature test macro now?
> 
>     * Why do we have both a feature test macro and `has_cxx_feature` ?
> 
> 
> I do think landing the PR ahead of Sofia makes perfect sense. This is a large body of work that Apple is keen on seeing upstreamed. It's a great security feature, and the paper is past EWG with strong support.
> 
> The next standard meetings will be well ahead of the clang 21 feature freeze, so we can reassess then if WG21... surprises us.
> 
> @erichkeane @AaronBallman

My opinion is that we implement it as-if it was accepted at plenary.  This isn't controversial as far as I can tell, and the core review doesn't seem to have any deal-breakers as far as I can tell.  

So I agree with your bullet 2.  I think we SHOULD set the feature-test-macro, and otherwise just treat this like it was accepted in plenary.

https://github.com/llvm/llvm-project/pull/113510