[llvm-commits] [llvm-gcc-4.2] r45052 - in /llvm-gcc-4.2/trunk/gcc/testsuite/gcc.c-torture/execute: 20051012-1.x 920501-1.x 921202-1.x 921208-2.x

Dale Johannesen dalej at apple.com
Fri Dec 14 18:04:47 PST 2007


On Dec 14, 2007, at 5:48 PM, Eric Christopher wrote:

>>
>> No, -w doesn't work to suppress that message (llvm doesn't have a
>> mechanism to
>> suppress it).
>
> Aaah.
>
>> Considering this is undefined behavior, I'm not
>> convinced this is
>> the wrong thing, although your first reaction would be that -w should
>> suppress
>> everything.  I could probably be convinced it should be a hard error
>> though.
>>
>> These tests are incorrect source and I think not running them is OK.
>
> Well, undefined, not incorrect. I think the tests are valid if
> undefined C code that the compiler should accept - but maybe warn  
> about.


I'm having trouble with "valid but undefined"....the standard permits  
a compiler
to insert runtime code to check that parameters match in number and  
type,
for example.  Not that our compilers do that.  (It also permits the  
program to
erase your disk, of course, but what I mentioned is sensible, in fact  
there are
environments that do that.)

I'd prefer to accept the warning and run them, but...




More information about the llvm-commits mailing list