[cfe-commits] r52049 - in /cfe/trunk: lib/CodeGen/CGExprConstant.cpp test/CodeGen/struct-init.c

[Nuno Lopes]

> On Fri, Jun 6, 2008 at 9:41 AM, Nuno Lopes <nunoplopes at sapo.pt> wrote:
>> Author: nlopes
>> Date: Fri Jun  6 11:41:08 2008
>> New Revision: 52049
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=52049&view=rev
>> Log:
>> implement constant expr. sub ptr ptr
>
>> --- cfe/trunk/test/CodeGen/struct-init.c (added)
>> +++ cfe/trunk/test/CodeGen/struct-init.c Fri Jun  6 11:41:08 2008
>> @@ -0,0 +1,13 @@
>> +// RUN: clang %s -emit-llvm
>> +
>> +typedef struct _zend_ini_entry zend_ini_entry;
>> +struct _zend_ini_entry {
>> +       void *mh_arg1;
>> +};
>> +
>> +char a;
>> +
>> +const zend_ini_entry ini_entries[] = {
>> +       {  ((char*)&((zend_ini_entry*)0)->mh_arg1 - (char*)(void*)0)},
>> +       {  ((long long*)&a - (long long*)(void*)2)},
>> +};
>
> I have a few things to ask about this testcase.  First off, without
> some hacking on Sema on my end, this testcase doesn't pass; do you
> have a hack in your tree to be looser about constant expressions?

The test passes on my machine. And currently my tree is clean: it has no 
pending patches.


> Secondly, is the "((long long*)&a - (long long*)(void*)2)" case from
> real code?  Supporting the case where the base is a null pointer in
> Sema is a lot easier than that one, so if it isn't from real code, I'm
> not going to bother supporting it.

No, it's not from real code. It was just a help for debugging. But as I 
said, it's working on my laptop..

Nuno