[llvm-commits] [llvm-gcc-4.2] r64584 - /llvm-gcc-4.2/trunk/gcc/cgraphunit.c

[Bill Wendling]

On Feb 15, 2009, at 1:18 PM, Chris Lattner wrote:

> On Feb 15, 2009, at 12:53 PM, Bill Wendling wrote:
>
>> On Feb 15, 2009, at 10:09 AM, Chris Lattner wrote:
>>
>>> On Feb 15, 2009, at 6:15 AM, Anton Korobeynikov wrote:
>>>
>>>> Hi, Bill
>>>>
>>>>> Fix build failure. Can't embed a directive within a macro  
>>>>> argument.
>>>> This looks rather hacky and definitely something not suitable for
>>>> non-
>>>> llvm build. Why don't just move directive before gcc_assert() and
>>>> have
>>>> two differen gcc_assert() calls?
>>>
>>> Yes, I agree with Anton.  Please don't change the behavior of the
>>> non-
>>> llvm build.
>>>
>> Actually, I was reverting what I put in there...But sure, I'll make
>> the change.
>>
>> As for changing the behavior of non-llvm builds, there are a few
>> places where that's virtually impossible. In particular, we define
>> extra fields in the tree structure. But because these fields are part
>> of a GTY(()) structure, it doesn't seem as if we can #ifdef them out.
>
> The goal isn't to produce exactly the same bits in the cc1 executable,
> the goal is for the behavior to be the same.  Adding another "dead"
> field should be ok.
>
That was only one example. There are others. :-)

-bw