[PATCH] templated static_assert

[Richard Smith]

On Mon, Aug 4, 2014 at 11:48 AM, Richard Smith <richard at metafoo.co.uk>
wrote:

> I don't think the diagnostic wording is right: 'template<typename T>' is
> not "a template declaration" in itself. A better model for the diagnostic
> might be this:
>
> <stdin>:1:34: error: a typedef cannot be a template
> template<typename T> typedef int n;
>                                  ^
>
> ... so "a static_assert declaration cannot be a template" (or "[...]
> cannot be templated" or similar).
>
> Also, we should use the same diagnostic for the member-declaration and the
> non-member-declaration cases. (Right now, the non-member case gives a
> useless "C++ requires a type specifier" diagnostic.)
>

Alternative approach: only check for tok::kw_static_assert in member
declarations if we don't have TemplateInfo. Then, when we come to diagnose
the unexpected static_assert keyword (maybe in ParseDeclarationSpecifiers
or ParseImplicitInt), produce a custom diagnostic if we're within a
template declaration.

On Mon, Aug 4, 2014 at 11:23 AM, Aaron Ballman <aaron at aaronballman.com>
> wrote:
>
>> According to [temp]p1, a static_assert cannot be templated (we have a
>> FIXME in the code about this as well). This patch emits a diagnostic
>> turning this into an error instead of silently accepting the template
>> and doing the wrong thing. Eg)
>>
>> struct S {
>>   template <typename Ty = char>
>>   static_assert(sizeof(Ty) != 1, "Not a char");
>> };
>>
>> This code currently compiles without diagnosing, but the static_assert
>> does not result in a diagnostic as you would expect. With this patch,
>> the code now emits a diagnostic on the template declaration.
>>
>> ~Aaron
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140804/c00a6920/attachment.html>