r199053 - Clarify warn_cxx98_compat_attribute diagnostic

[Alp Toker]

On 14/01/2014 03:10, Richard Smith wrote:
> On Mon, Jan 13, 2014 at 6:37 PM, Aaron Ballman <aaron at aaronballman.com 
> <mailto:aaron at aaronballman.com>> wrote:
>
>     On Mon, Jan 13, 2014 at 9:24 PM, Richard Smith
>     <richard at metafoo.co.uk <mailto:richard at metafoo.co.uk>> wrote:
>     > On Sun, Jan 12, 2014 at 11:14 AM, Aaron Ballman
>     <aaron at aaronballman.com <mailto:aaron at aaronballman.com>>
>     > wrote:
>     >>
>     >> After doing a bit more research and discussion off-list, I think
>     >> "generalized attribute" is acceptable.  So patch LGTM as-is.
>     >
>     >
>     > Really? I wouldn't expect someone seeing this diagnostic to
>     understand that
>     > "generalized attribute" means C++11 attributes (it's a really
>     weird term,
>     > since they're not a generalization of anything). This isn't an
>     official name
>     > for them, and doesn't distinguish them from the other attribute
>     syntaxes we
>     > support. Given that this is a diagnostic about compatibility
>     with C++98,
>     > "C++11 attributes" seems like the clearest way of expressing this.
>
>     As Alp had pointed out, we document the name as "generalized
>     attribute" in our feature support documentation,
>
>
> You're right, we did. I just fixed that.
>
>     and it's the original name of the feature.
>
>
> [citation needed]
>
> The proposal calls them "General Attributes for C++" (and all previous 
> revisions of it did the same); the word "Generalized" seems to have 
> been accidentally transferred from "Generalized constant expressions" 
> in the GCC list, and we inherited that mistake when we sync'd our list 
> with theirs in r142015. The paper and C++ standard both call them 
> simply "attributes".
>
>     Also, a quick google search of "generalized
>     attribute" yielded more results than "C++11 attribute" did (not saying
>     this was particularly scientific). So that's why I gave the LGTM on
>     the term.
>
>
> Hah, it seems that lots of people copied our C++11 feature list and 
> GCC's, picking up the wrong name =)

It may be a typo, but the C++ community has clearly adopted the name 
"generalized attributes".

I think we should take it and run with it :-)

Reasons to do so:

  * "C++11 attributes" don't work as a feature name when bringing this
    to C11 as an extension or enabling it in proposed next-generation
    OpenMP modes. It'd be bizarre to diagnose with "C++11 attribute ..."
    in C11.
  * It feels old keeping the version of introduction in the name. We
    don't do this with other features that have been published, why
    introduce a special-case "C++11 attributes"?
  * Leaving the name as just "attributes" is ambiguous in this context
    because users have got used to "attributes" referring to the GNU form.


Alp.




>
>     That being said, my original preference was for C++11 attribute
>     instead, and as you point out, this is a C++98 compat diagnostic, so
>     using "C++11" would be clear. Perhaps I should have stuck with my gut
>     instead. ;-)
>
>     ~Aaron
>
>

-- 
http://www.nuanti.com
the browser experts

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140114/04bb3ee3/attachment.html>