[cfe-dev] <vector> parse but cannot be instanciated

[Douglas Gregor]

On Sep 16, 2009, at 9:13 AM, Cédric Venet wrote:

>
>>
>>> (note that I am using clang on windows with the mingw 3.4.5  
>>> headers and that the define __declspec is needed. perhaps it  
>>> should be added to  the predefined macro? as clang support the  
>>> dllimport attribute)
>>
>> We have some support for parsing __declspec already, but I'm sure  
>> that we're missing it when parsing some C++ constructs. What kinds  
>> of problems are you seeing with __declspec?
>
> It is mingw specific and not a problem on declspec. In _mingw.h,  
> there is:
>
> #ifndef __GNUC__
> # ifndef __MINGW_IMPORT
> #  define __MINGW_IMPORT  __declspec(dllimport)
> # endif
> # ifndef _CRTIMP
> #  define _CRTIMP  __declspec(dllimport)
> # endif
> # define __DECLSPEC_SUPPORTED
> # define __attribute__(x) /* nothing */
> #else /* __GNUC__ */
> # ifdef __declspec
> #  ifndef __MINGW_IMPORT
>  /* Note the extern. This is needed to work around GCC's
>     limitations in handling dllimport attribute.  */
> #   define __MINGW_IMPORT  extern __attribute__ ((__dllimport__))
> #  endif
> #  ifndef _CRTIMP
> #   ifdef __USE_CRTIMP
> #    define _CRTIMP  __attribute__ ((dllimport))
> #   else
> #    define _CRTIMP
> #   endif
> #  endif
> #  define __DECLSPEC_SUPPORTED
> # else /* __declspec */
> #  undef __DECLSPEC_SUPPORTED
> #  undef __MINGW_IMPORT
> #  ifndef _CRTIMP
> #   define _CRTIMP
> #  endif
> # endif /* __declspec */
>
> and if __MINGW_IMPORT is not defined, there is problems latter in  
> the compilation (in time.h for exemple).
> So when you have __GNUC__ defined, you have to have __declspec  
> defined, else it does not work. Anyway putting a define __declspec  
> allowed me to parse this, even if it is a big hack)
> (in fact if __MSVCRT__ can be undefined, __MINGW_IMPORT should not  
> be used...)

Okay, so we need to be defining __declspec on mingw targets. That  
involves creating a new target for MinGW32/64 (in lib/Basic/ 
Targets.cpp) that adds these defines into its getTargetDefines()  
function. The actual spelling of the predefined __declspec (and any  
other mingw-specific macros) will be shown in the output of

	touch foo.cpp
	g++ -dM -E foo.cpp

>>
>>> now, the instanciation of the vector assert:
>>>
>>> Assertion failed: false && "Unable to find declaration for the  
>>> current instantiation", file ..\..\..\..\..\tools\clang\lib\Sema 
>>> \SemaTemplateInstantiateDecl.cpp, line 1411
>>
>> FWIW, this is coming from an attempt to instantiate a default  
>> argument expression in a template (that refers to a typedef), e.g.,
>
> I see, thanks for the explanation. I must say I am eager to have  
> good c++ semantic analyse working to be able to use clang for static  
> analysis and inteligent refactoring, but I know the task is daunting  
> so I am waiting.

We're inching closer :)

	- Doug
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20090916/39bdb59e/attachment.html>