On Feb 8, 2012, at 3:37 PM, Anna Zaks <[email protected]> wrote:

> 
> On Feb 8, 2012, at 3:25 PM, Ted Kremenek wrote:
> 
>> On Feb 8, 2012, at 3:16 PM, Anna Zaks <[email protected]> wrote:
>> 
>>> Author: zaks
>>> Date: Wed Feb  8 17:16:56 2012
>>> New Revision: 150112
>>> 
>>> URL: http://llvm.org/viewvc/llvm-project?rev=150112&view=rev
>>> Log:
>>> [analyzer] MallocChecker: implement pessimistic version of the checker,
>>> which allows values to escape through unknown calls.
>>> 
>>> Assumes all calls but the malloc family are unknown.
>> 
>> One nit: we should only do this if the call was not inlined.  Otherwise 
>> we're being overly pessimistic.
> 
> Shouldn't inlining be transparent to the checkers?

It should be transparent, but checkers should not be ignorant of its 
implications.

> Ex: Should the checkers receive the post/pre visit callbacks on a CallExpr if 
> it's being inlined?

Absolutely.  The pre- and post- visit checks are for checking preconditions and 
postconditions, which is orthogonal to how the actual function call gets 
evaluated.

This is probably worth some design discussion.

> Another concern I wanted to raise is that I've renamed the optimistic(old) 
> checker to MallocWithAnnotationsd and named the new checker Malloc. I think 
> the naming makes sense, but not sure if we are going to effect existing users 
> of the old Malloc checker.

Since the old checker behavior wasn't widely published (and the checker was 
"experimental" on top of that) I think this is fine.  Existing users can 
migrate with little difficulty.

_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits

Reply via email to