Re: [CIL users] CIL handles __builtin_object_size incorrectly

[Jesse M D. <jd...@su...>]

Gabriel,
I'm attaching two gzipped tar files, one that includes the makefile, source file, and all files 
generated by the unpatched svn Cil, and another that includes the same files after applying my 
patch. I wasn't exactly sure which ones you wanted, but they were small enough to send everything 
and let you decide.

I agree with you that my patch may not be correct. I wasn't entirely sure how Cil was deciding how 
to modify builtins, and I'm also not sure about the GCC specification. I do think, however, that the 
patch is in the right place, and, as you can see by diffing the output files from the two 
attachments, the extra increment is eliminated, which was my immediate goal.

Jesse

On 02/15/2011 08:00 AM, Gabriel Kerneis wrote:
> On Thu, Feb 03, 2011 at 02:13:12PM -0500, Jesse M Draper wrote:
>> My patch to src/frontc/cabs2cil.ml checks for
>> "__builtin_object_size" and, following the pattern in
>> "__builtin_constant_p", drops the side effects.
> I am not convinced this patch is correct (although current behaviour is
> obviously a bug).  Gcc doc states:
>      If there are any side-effects in them, it returns (size_t) -1 for
>      type 0 or 1 and (size_t) 0 for type 2 or 3.
> which your patch does not perform.
>
> I'll think about it.
>
>> I have not been able to reproduce the problem on my Ubuntu system
>> because its gcc does not generate the call to __builtin_object_size,
>> and I have not figured out how gcc decides whether or not to generate
>> it.
> Could you please send the cilbug.c.i (produced on MacOS X)?  I would
> like to include it in the testsuite.
>
> Best,