[cfe-commits] r172570 - /cfe/trunk/lib/CodeGen/CGExpr.cpp

[Richard Smith]

On Thu, Jan 17, 2013 at 10:06 AM, Dmitri Gribenko <gribozavr at gmail.com> wrote:
> Hello David,
>
> On Wed, Jan 16, 2013 at 1:13 AM, David Greene <greened at obbligato.org> wrote:
>> Author: greened
>> Date: Tue Jan 15 17:13:47 2013
>> New Revision: 172570
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=172570&view=rev
>> Log:
>> Avoid unsigned Compare to int
>>
>> Cast arithmetic results to avoid comparison of an unsigned to an int.
>>
>> Modified:
>>     cfe/trunk/lib/CodeGen/CGExpr.cpp
>>
>> Modified: cfe/trunk/lib/CodeGen/CGExpr.cpp
>> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGExpr.cpp?rev=172570&r1=172569&r2=172570&view=diff
>> ==============================================================================
>> --- cfe/trunk/lib/CodeGen/CGExpr.cpp (original)
>> +++ cfe/trunk/lib/CodeGen/CGExpr.cpp Tue Jan 15 17:13:47 2013
>> @@ -1191,7 +1191,7 @@
>>    cast<llvm::LoadInst>(Val)->setAlignment(Info.StorageAlignment);
>>
>>    if (Info.IsSigned) {
>> -    assert((Info.Offset + Info.Size) <= Info.StorageSize);
>> +    assert(static_cast<unsigned>(Info.Offset + Info.Size) <= Info.StorageSize);
>>      unsigned HighBits = Info.StorageSize - Info.Offset - Info.Size;
>>      if (HighBits)
>>        Val = Builder.CreateShl(Val, HighBits, "bf.shl");
>
> Which version of gcc do you use?
>
> I tested gcc 4.4, 4.6, 4.7 with -W -Wall -Wextra on
>
> int f(unsigned a, unsigned b, unsigned c) {
>   if (a + b <= c)
>     return 42;
>   return 0;
> }
>
> and I don't see any warnings.

This is due to GCC's misimplementation of bitfields (they implement a
resolution to a WG14 defect report which the C committee decided not
to ratify). They treat the bitwidth as being part of the type, and
then the usual arithmetic conversions for 'unsigned:16 + unsigned:15'
give 'int' as the type of the LHS of the <=.