[cfe-dev] errors running clang/test testcases

[Bjørn Roald]

Steve Naroff skrev:
> Thanks much for the test case. Commit r45951 should fix this problem.
>
> Please let me know if you have any problems,

The testcase I sendt was solved by r45951 :-).  However clang is still 
not happy with test/CodeGen/mandel.c.  This macro expanded code is nuts, 
so don't ask me what it mean or whether it make sence to support this 
stuff.  Looking at it closer I think the last one I sendt may have been 
a side-step from the original problem, but nevertheless a clang/gcc 
diff.  I now get the following set of interesting tast cases.  The two 
first testcases involve casting boolean value to (void*), which seems to 
cause different behaviour than  gcc. The last 3 simply show that the 
compilers does the same thing if dealing with int directly or if the 
conditional operation is removed.

void f3_2_3_0() { __typeof__(* (__typeof__ (0 ? (double*) 0 : (void *) 
(!(((double) 0.25) && ((double) 0.25 - 1))))) 0 ) a;} // gcc: Ok , clang 
error: variable has incomplete type ...
void f3_2_3_1() { __typeof__(* (__typeof__ (0 ? (double*) 0 : (void *) 
(!(int) 1))) 0 ) a;} // gcc: Ok , clang error: variable has incomplete 
type ...
void f3_2_3_2() { __typeof__(* (__typeof__ (0 ? (double*) 0 : (void *) ( 
(int) 0))) 0 ) a;} // gcc/clang: Ok
void f3_2_3_3() { __typeof__(* (__typeof__ (0 ? (double*) 0 : (void *) ( 
(int) 1))) 0 ) a;} // gcc/clang: error
void f3_2_3_4() { __typeof__(* (__typeof__ (                  (void *) 
(!(int) 1))) 0 ) a;} // gcc/clang: error

regards,
-- 
Bjørn