[cfe-dev] CodeGen/statements.c: Don't understand test and its failure

[Chris Lattner]

On Oct 15, 2009, at 8:42 AM, John Thompson wrote:

> On Windows, the CodeGen/statements.c test fails, apparently because  
> of this error:
>
> 3><stdin>:23:17: error: initializer element is not a compile-time  
> constant
> 3>static long x = &&bar - &&baz;
> 3>                ^~~~~~~~~~~~~
> First, what does a unary "&&" do on a variable?  Address of an  
> address?  It doesn't seem to make sense.
>
> Second, why would it result in an error in VC++-built Clang, but not  
> on other platforms?  (I tried using other target triples, but there  
> was no change.)
>
> I know I'll probably have to debug it, but understanding the first  
> might help.

This is the GNU address of a label extension.  Can you remove the  
'static' (which will allow it to parse) and paste what clang-cc -ast- 
dump shows for that statement?

-Chris