[cfe-commits] [Patch 7 of 7] -verify fixes and enhancement

[Douglas Gregor]

On Jul 3, 2012, at 5:48 PM, Jordan Rose wrote:

> 
> On Jul 3, 2012, at 3:23 PM, Andy Gibbs wrote:
> 
>> On Tuesday, July 03, 2012 5:18 PM, Jordan Rose wrote:
>>> On Jul 2, 2012, at 10:39 PM, Andy Gibbs wrote:
>>> [...snip...]
>>>> In the case above, Module.h is shared across a number of tests.  In some
>>>> tests the include file was parsed correctly and in others it was not.  (I
>>>> made some comments about a net with holes in another post and this
>>>> is one example of where it applied!)  Unfortunately, this incorrect parsing
>>>> coincided with the cases where the diagnostic also not generated (if you
>>>> look at the original implementation you will understand why), so the
>>>> test-case bug was missed.  Since the diagnostic sometimes is and
>>>> sometimes is not generated, hence the "0-1".
>>> 
>>> I see. It doesn't look like the "umbrella header" warning (-Wincomplete-umbrella)
>>> is exercised anywhere else, though. Perhaps it should be put into a test of
>>> its own? (I think it's reasonable to make a separate "Umbrella.framework"
>>> because of the existing expectation in Module.framework.)
>> 
>> Hmm, I'm not sure I understand enough about how the modules part of the compiler
>> works to change a test-case so drastically.  Would it not be better to get this
>> changed by someone better qualified?
> 
> Let's rope Doug in on this part. Doug, what exactly is -Wincomplete-umbrella for?

It checks, at module build time, whether an umbrella header <Foo/Foo.h> actually includes all of the headers <Foo/*.h>.

> Is it okay to just test it in one place, and make the "default" Module.framework a completely clean framework?

Yes, that would be cleaner.

	- Doug