[PATCH] Sema: Reorganize and improve semantic tests for dllexport/dllimport

[Aaron Ballman]

As was pointed out in IRC, my complaints revolved around situations
where MSVC silently drops the attribute without warning. I'm much less
concerned with that situation.

~Aaron

On Wed, Feb 12, 2014 at 7:55 PM, Aaron Ballman <aaron at aaronballman.com> wrote:
> On Wed, Feb 12, 2014 at 7:52 PM, Nico Rieck <nico.rieck at gmail.com> wrote:
>> On 13.02.2014 01:33, Aaron Ballman wrote:
>>> As a comment in general for the test cases: you can
>>> dllexport/dllimport declarators according to MSDN (in fact, their
>>> examples show exporting a variable). So are you testing existing
>>> behavior in clang, and then someday we will presumably modify our
>>> behavior? If so, it'd be good to have some FIXME comments around this
>>> sort of test case so it's known that we will be modifying it at some
>>> point.
>>
>> This tests existing correct behavior in Clang.
>>
>> I have all these patches ready, and unimplemented or broken stuff is not
>> tested until implemented. So patches usually just add tests, making it
>> easier to review (I hope).
>
> Yes, but your patch is adding test cases for things which behave one
> way in clang, but a different way in MSVC, for an MSVC feature we're
> trying to emulate. This is why I was suggesting adding some FIXMEs to
> those cases, since presumably those cases will be changed some day to
> follow the MSVC behavior.
>
> ~Aaron