[cfe-dev] Interaction of target attribute change with modules.

[Sean Silva]

On Tue, Jun 23, 2015 at 1:35 PM, Richard Smith <richard at metafoo.co.uk>
wrote:

> On Mon, Jun 22, 2015 at 6:05 PM, Sean Silva <chisophugis at gmail.com> wrote:
>
>> Hi All,
>>
>> With the change to use target attributes in the builtin headers, we are
>> now in a weird situation where all of the `requires` inside
>> lib/Headers/module.modulemap are basically useless. They are included
>> regardless of the status of their `requires`. My pending patch
>> http://reviews.llvm.org/D10423 then causes a build failure because we
>> try to import unavailable modules; without it, we just textually include
>> them which causes the declarations to appear to be owned by different
>> modules than expected.
>>
>> Any ideas? It seems like to keep semantic parity with the "all intrinsic
>> headers can be brought in" when textually including, we should remove the
>> `requires`.
>>
>
> We should keep the 'requires x86' on _Builtin_intrinsics.intel, but remove
> all the subtarget 'requires' declarations from its submodules.
>

Yeah, that's what I meant. Obviously we don't want to be pulling in a
completely other ISA. I'll send a patch.

-- Sean Silva
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150624/4e498ce5/attachment.html>