[cfe-commits] r110317 - /cfe/trunk/lib/Checker/GRExprEngine.cpp

[Jordy Rose]

ProcessAssume should already be correct -- though it doesn't count the
checkers, the early end condition results in a return, not a break.

Looking at the checkers, there are roughly three kinds of callbacks:

1. Full visits. Every checker gets a turn, and they can create multiple
ExplodedNodes. (Ex: _PreVisit, VisitBranchCondition)
2. Evaluations. First checker to claim the statement wins, and it can
create multiple ExplodedNodes. (Ex: EvalCallExpr, EvalNilReceiver)
3. Filters: Every checker gets a turn, but they can't bifurcate the state.
(Ex: ProcessAssume, proposed ProcessRegionChange)

These require three different loops, whether or not we're caching.

The problem is that each callback takes completely different arguments,
and I'm not sure how to parameterize a loop to make calls with different
signatures. Which doesn't make it impossible, just annoying. This
definitely ought to be cleaner.


On Thu, 5 Aug 2010 21:10:43 -0700, Ted Kremenek <kremenek at apple.com>
wrote:
> Looks great Zhongxing.
> 
> We should also update the corresponding logic in ProcessAssume to also
do
> the right checking to make sure that all checkers were evaluated.  Down
the
> line we should also consider if we can merge this logic, instead of
having
> separate copies.
> 
> On Aug 5, 2010, at 5:16 PM, Zhongxing Xu wrote:
> 
>> I see. How about this patch to make the check functional.
>> 
>> Index: lib/Checker/GRExprEngine.cpp
>> ===================================================================
>> --- lib/Checker/GRExprEngine.cpp	(版本 110392)
>> +++ lib/Checker/GRExprEngine.cpp	(工作副本)
>> @@ -206,6 +206,8 @@
>>   unsigned checkersEvaluated = 0;
>> 
>>   for (CheckersOrdered::iterator I=CO->begin(), E=CO->end(); I!=E;
++I){
>> +    if (PrevSet->empty())
>> +      break;
>>     ExplodedNodeSet *CurrSet = 0;
>>     if (I+1 == E)
>>       CurrSet = &Dst;