Re: [WIP RFC] Add support for keyword-based attributes

[Richard Sandiford <richard.sandiford@arm.com>]

Joseph Myers <joseph@codesourcery.com> writes:
> On Mon, 17 Jul 2023, Michael Matz via Gcc-patches wrote:
>
>> So, essentially you want unignorable attributes, right?  Then implement 
>> exactly that: add one new keyword "__known_attribute__" (invent a better 
>> name, maybe :) ), semantics exactly as with __attribute__ (including using 
>> the same underlying lists in our data structures), with only one single 
>> deviation: instead of the warning you give an error for unhandled 
>> attributes.  Done.
>
> Assuming you also want the better-defined standard rules about how [[]] 
> attributes appertain to particular entities, rather than the different 
> __attribute__ rules, that would suggest something like [[!some::attr]] for 
> the case of attributes that can't be ignored but otherwise are handled 
> like standard [[]] attributes.

Yeah, that would work.  But I'd rather not gate the SME work on getting
an extension like that into C and C++.

As it stands, some clang maintainers pushed back against the use of
attributes for important semantics, and preferred keywords instead.
It's clear from this threads that the GCC maintainers prefer attributes
to keywords.  (And it turns out that some other clang maintainers do too,
though not as strongly.)

So I think the easiest way of keeping both constituencies happy(-ish)
is to provide both standard attributes and "keywords", but allow
the "keywords" to be macros that expand to standard attributes.

Would it be OK to add support for:

  [[__extension__ ...]]

to suppress the pedwarn about using [[]] prior to C2X?  Then we can
predefine __arm_streaming to [[__extension__ arm::streaming]], etc.

Thanks,
Richard