[llvm] r179111 - Revert r176408 and r176408 to address PR15540.

[Nuno Lopes]

Hi,

In summary, I would like to reapply r176407, yes.
The reason is that it is the correct fix for PR15540. This new function, 
getUnderlyingObjectSize(), gives exactly the semantics that BasicAA wants 
(which is to get the size of the object pointed by a given pointer). The 
function getObjectSize(), on the other hand, gives the size of an object 
minus the offset of the given pointer.  Using getObjectSize() is therefore 
wrong (I wrote that code, but later I realized my mistake).

I believe r176407 is the correct fix, and it should be reapplied.

Nuno


-----Original Message----- 
From: Nadav Rotem
Sent: Sunday, April 14, 2013 5:40 AM
Subject: Re: [llvm] r179111 - Revert r176408 and r176408 to address PR15540.


Hi Nuno,

I looked at PR15540 again and I read Shuxin's email, and I am not sure that 
I understand the whole picture. I understand that in r176408 you 
re-committed patches that you previously reverted after a discussion with 
Shuxin and Arnold. It is difficult for me to follow what's going on right 
now.  I understand that you want to re-apply 176408. Can you please send the 
patch for review first ? This will give others a chance to review the code 
and avoid further confusion.

Thanks,
Nadav



On Apr 13, 2013, at 7:55 PM, Nadav Rotem <nrotem at apple.com> wrote:



On Apr 13, 2013, at 6:22 PM, Nuno Lopes <nunoplopes at sapo.pt> wrote:


Hi Nadav,

I'm sorry to say but you reverted too much. In particular:
- You should not remove  getUnderlyingObjectSize(), since it's required for 
correctness of BasicAA.
- BasicAA changes shouldn't be reverted for the same reason.
- Tests 18 and 19 have nothing to do with this, and should stay.

I can perform these changes myself if you prefer.



Hi Nuno,

Thanks for looking at this. Are you going to re-apply r176407 ?

Thanks,
Nadav