[cfe-dev] avoid loops in the exploded graph?

Whisperity via cfe-dev cfe-dev at lists.llvm.org
Tue Oct 9 05:50:41 PDT 2018


Use Occam’s razor, there is a very easy explanation for that particular TODO. Namely that no one had the time or effort to fix the checkers mentioned in the comment, and potentially the plethora of others introduced that exploit the early return. 😊

From: Donát Nagy via cfe-dev
Sent: 2018. október 9., kedd 14:03
To: dkszelethus at gmail.com; lewisurn at gmail.com
Cc: cfe-dev at lists.llvm.org
Subject: Re: [cfe-dev] avoid loops in the exploded graph?

Hi,

My intuition is that in some situations it is guaranteed that the state will be modified and then the if (NewState!=State) check is unnecessary. In other situations there is a possibility that the state will remain the same (e.g. the expression was already marked as tainted, so nothing changes) and this could theoretically create to a loop edge (an edge from the current node to itself) in the exploded graph.

A loop edge is obviously undesirable, because it could send graph traversal into infinite loops, so wee need to guarantee that they do not appear in the exploded graph.  As Kristóf wrote, the method CheckerContext::addTransitionImpl() always guarantees this with a defensive check (with a TODO to turn it into an assert); but in some checkers it is also tested by the checker itself.

Donát

On k, 2018-10-09 at 11:10 +0200, Kristóf Umann via cfe-dev wrote:
Hi!
If you look at how addTransition is implemented, you can see a big TODO before a defensive check:
https://github.com/llvm-mirror/clang/blob/7c8e954f273730037b454edf94c8d13123dbedf6/include/clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h#L289
According to that, it would seem that a checker-side check should always be there.
Let's see what others think of this -- to me it would make more sense to have this check within addTransition, and I'm unsure myself why that TODO is there.

Lou Wynn via cfe-dev <cfe-dev at lists.llvm.org> ezt írta (időpont: 2018. okt. 9., K, 3:42):
Hi,
I read the comment "avoid loops in the exploded graph" in the following snippet of code which is on page 32 of the workbook.
LocationContext *LC = C. getLocationContext (); 
ProgramStateRef State = C. getState (); 
const Expr *E = /* Obtain an expression value of which is untrusted */; 
ProgramStateRef NewState = State -> addTaint (E, LC ); 
if ( NewState != State ) // avoid loops in the exploded graph 
  C. addTransition ( NewState );
My question is why the new tainted state requires the if statement to prevent loops, while other new states in the book do not have the if statement when C.addTransition(State) is used? Do other states which are not tainted not need to prevent loops? For example, on page 30, when a new state is added, it reads:
ProgramStateRef State = C. getState (); 
State = modifyState ( State ); // do stuff 
C. addTransition ( State );
There is no if state to prevent a loop.
-- 
Love,
Lou

_______________________________________________
cfe-dev mailing list
cfe-dev at lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
_______________________________________________
cfe-dev mailing list
cfe-dev at lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20181009/479f3228/attachment-0001.html>


More information about the cfe-dev mailing list