[cfe-commits] [PATCH] RegionStore

[Zhongxing Xu]

On Mon, Oct 20, 2008 at 2:06 PM, Ted Kremenek <kremenek at apple.com> wrote:

>
> On Oct 19, 2008, at 10:05 PM, Zhongxing Xu wrote:
>
>  As a more direct example, consider code
>>
>> int a[10];
>> a[1] = 3;
>>
>> When we visit a[1], we evaluate a's rvalue. In VisitDeclRefExpr, we first
>> get a's lvalue loc::MemRegionVal(a's region), then do a load on it. We
>> expect to load loc::MemRegionVal(a's region) from itself. Then a
>> ImplicitCast will convert it to a pointer to a's first element.
>>
>
> Part of my confusion is that BasicStore doesn't model arrays.  We cannot
> expect to model array bindings using the simple VarDecl to SVal binding in
> BasicStore.  A "GetSVal" on a VarDecl with an array type should just return
> UnknownVal, since we don't have "element-sensitivity" in BasicStore.


OK. This is reasonable. I removed the special handling of array VarDecl in
the BasicStore. Now both 'a' and '*p' will evaluate to UnknownVal. Because
GetSVal will return UnknownVal now.

On the other hand, do you agree the special handling in RegionStore? My
intention is this: let both rvalue and lvalue of an array variable be
loc::MemRegionVal(its region). Then the ImplicitCast cast this to
loc::MemRegionVal(first element's region).


>
>
> I think I see your point, but I'm not (completely) convinced it is the
> right approach to do this in the store, as it really seems to be an issue of
> syntax and performing the right conversions in GRExprEngine.  VisitLValue is
> called on "a[1]", and "Visit" is called on "a" (returning its rvalue).  The
> AST dump of the assignment looks like this:
>
>  (BinaryOperator 0xd07920 <line:3:3, col:10> 'int' '='
>    (ArraySubscriptExpr 0xd078e0 <col:3, col:6> 'int'
>      (ImplicitCastExpr 0xd078d0 <col:3> 'int *'
>        (DeclRefExpr 0xd07870 <col:3> 'int [10]' Var='a' 0xd07820))
>      (IntegerLiteral 0xd07890 <col:5> 'int' 1))
>    (IntegerLiteral 0xd07900 <col:10> 'int' 3)))
>
> Do you think it seems more reasonable for VisitCast to specially handle the
> case of casting an array type to a pointer?  One possible solution in this
> case would be to have VisitCast invoke VisitLValue on its subexpression when
> performing such a cast.  I'm not certain if this is the right thing to do.
>

We'd better not invode VisitLValue in VisitCast, because the whole
ImplicitCastExpr is used in an rvalue context (as the base of the
ArraySubscriptExpr).

As I said above, we do the usual Visit in VisitCast, get
loc::MemRegionVal(array's region) through EvalLoad(), with the help of
special handling in RegionStore. Then cast this into loc::MemRegionVal(first
element's region).
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20081020/86d4a357/attachment.html>