[PATCH] New functionality for the consumed analysis

Christian Wailes chriswailes at google.com
Wed Aug 21 15:53:59 PDT 2013

Aaron, you are correct.  There is not currently a test for that.  I can
either add one to this patch or ensure that one is in the next patch.

DeLesely: That was the assumption that David and I came up with.  This will
produce more errors, but should catch more bugs.  There could be a problem,
however, when a method does change internal state, but not state that is
relevant to the "consumed/unconsumed" status of the object.  This method
couldn't be marked constant, and would therefor transition the object out
of its known state.

- Chris

On Wed, Aug 21, 2013 at 3:16 PM, Delesley Hutchins <delesley at google.com>wrote:

>   LGTM.
> ================
> Comment at: lib/Analysis/Consumed.cpp:446
> @@ -419,1 +445,3 @@
> +          if (!MethodDecl->isConst())
> +            StateMap->setState(PState.getVar(), consumed::CS_Unknown);
>          }
> ----------------
> Is the assumption here that calling a non-const method puts the object
> into an unknown state?  That's a correct, conservative judgement.
> However, an alternative is to assume that unmarked methods do not change
> the object state, which might be more user-friendly in the common case.
> http://llvm-reviews.chandlerc.com/D1468
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130821/0dcdd28d/attachment.html>

More information about the cfe-commits mailing list